blob: 234bd4895d14dc9a837473ce5474eee03253b0f1 [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);
Hugh Dickinsd98c7a02006-02-14 13:52:59 -080059static void __free_pages_ok(struct page *page, unsigned int order);
David Howellsa226f6c2006-01-06 00:11:08 -080060
Linus Torvalds1da177e2005-04-16 15:20:36 -070061/*
62 * results with 256, 32 in the lowmem_reserve sysctl:
63 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
64 * 1G machine -> (16M dma, 784M normal, 224M high)
65 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
66 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
67 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
Andi Kleena2f1b422005-11-05 17:25:53 +010068 *
69 * TBD: should special case ZONE_DMA32 machines here - in those we normally
70 * don't need any ZONE_NORMAL reservation
Linus Torvalds1da177e2005-04-16 15:20:36 -070071 */
Andi Kleena2f1b422005-11-05 17:25:53 +010072int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = { 256, 256, 32 };
Linus Torvalds1da177e2005-04-16 15:20:36 -070073
74EXPORT_SYMBOL(totalram_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -070075
76/*
77 * Used by page_zone() to look up the address of the struct zone whose
78 * id is encoded in the upper bits of page->flags
79 */
Christoph Lameterc3d8c142005-09-06 15:16:33 -070080struct zone *zone_table[1 << ZONETABLE_SHIFT] __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070081EXPORT_SYMBOL(zone_table);
82
Andi Kleena2f1b422005-11-05 17:25:53 +010083static char *zone_names[MAX_NR_ZONES] = { "DMA", "DMA32", "Normal", "HighMem" };
Linus Torvalds1da177e2005-04-16 15:20:36 -070084int min_free_kbytes = 1024;
85
86unsigned long __initdata nr_kernel_pages;
87unsigned long __initdata nr_all_pages;
88
Nick Piggin13e74442006-01-06 00:10:58 -080089#ifdef CONFIG_DEBUG_VM
Dave Hansenc6a57e12005-10-29 18:16:52 -070090static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -070091{
Dave Hansenbdc8cb92005-10-29 18:16:53 -070092 int ret = 0;
93 unsigned seq;
94 unsigned long pfn = page_to_pfn(page);
Dave Hansenc6a57e12005-10-29 18:16:52 -070095
Dave Hansenbdc8cb92005-10-29 18:16:53 -070096 do {
97 seq = zone_span_seqbegin(zone);
98 if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
99 ret = 1;
100 else if (pfn < zone->zone_start_pfn)
101 ret = 1;
102 } while (zone_span_seqretry(zone, seq));
103
104 return ret;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700105}
106
107static int page_is_consistent(struct zone *zone, struct page *page)
108{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109#ifdef CONFIG_HOLES_IN_ZONE
110 if (!pfn_valid(page_to_pfn(page)))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700111 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112#endif
113 if (zone != page_zone(page))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700114 return 0;
115
116 return 1;
117}
118/*
119 * Temporary debugging check for pages not lying within a given zone.
120 */
121static int bad_range(struct zone *zone, struct page *page)
122{
123 if (page_outside_zone_boundaries(zone, page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124 return 1;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700125 if (!page_is_consistent(zone, page))
126 return 1;
127
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128 return 0;
129}
130
Nick Piggin13e74442006-01-06 00:10:58 -0800131#else
132static inline int bad_range(struct zone *zone, struct page *page)
133{
134 return 0;
135}
136#endif
137
Nick Piggin224abf92006-01-06 00:11:11 -0800138static void bad_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139{
Nick Piggin224abf92006-01-06 00:11:11 -0800140 printk(KERN_EMERG "Bad page state in process '%s'\n"
Hugh Dickins7365f3d2006-01-11 12:17:18 -0800141 KERN_EMERG "page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n"
142 KERN_EMERG "Trying to fix it up, but a reboot is needed\n"
143 KERN_EMERG "Backtrace:\n",
Nick Piggin224abf92006-01-06 00:11:11 -0800144 current->comm, page, (int)(2*sizeof(unsigned long)),
145 (unsigned long)page->flags, page->mapping,
146 page_mapcount(page), page_count(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147 dump_stack();
Hugh Dickins334795e2005-06-21 17:15:08 -0700148 page->flags &= ~(1 << PG_lru |
149 1 << PG_private |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150 1 << PG_locked |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151 1 << PG_active |
152 1 << PG_dirty |
Hugh Dickins334795e2005-06-21 17:15:08 -0700153 1 << PG_reclaim |
154 1 << PG_slab |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155 1 << PG_swapcache |
Hugh Dickins689bceb2005-11-21 21:32:20 -0800156 1 << PG_writeback );
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157 set_page_count(page, 0);
158 reset_page_mapcount(page);
159 page->mapping = NULL;
Randy Dunlap9f158332005-09-13 01:25:16 -0700160 add_taint(TAINT_BAD_PAGE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161}
162
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163/*
164 * Higher-order pages are called "compound pages". They are structured thusly:
165 *
166 * The first PAGE_SIZE page is called the "head page".
167 *
168 * The remaining PAGE_SIZE pages are called "tail pages".
169 *
170 * All pages have PG_compound set. All pages have their ->private pointing at
171 * the head page (even the head page has this).
172 *
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800173 * The first tail page's ->lru.next holds the address of the compound page's
174 * put_page() function. Its ->lru.prev holds the order of allocation.
175 * This usage means that zero-order pages may not be compound.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176 */
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800177
178static void free_compound_page(struct page *page)
179{
180 __free_pages_ok(page, (unsigned long)page[1].lru.prev);
181}
182
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183static void prep_compound_page(struct page *page, unsigned long order)
184{
185 int i;
186 int nr_pages = 1 << order;
187
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800188 page[1].lru.next = (void *)free_compound_page; /* set dtor */
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800189 page[1].lru.prev = (void *)order;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190 for (i = 0; i < nr_pages; i++) {
191 struct page *p = page + i;
192
193 SetPageCompound(p);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700194 set_page_private(p, (unsigned long)page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195 }
196}
197
198static void destroy_compound_page(struct page *page, unsigned long order)
199{
200 int i;
201 int nr_pages = 1 << order;
202
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800203 if (unlikely((unsigned long)page[1].lru.prev != order))
Nick Piggin224abf92006-01-06 00:11:11 -0800204 bad_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205
206 for (i = 0; i < nr_pages; i++) {
207 struct page *p = page + i;
208
Nick Piggin224abf92006-01-06 00:11:11 -0800209 if (unlikely(!PageCompound(p) |
210 (page_private(p) != (unsigned long)page)))
211 bad_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212 ClearPageCompound(p);
213 }
214}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215
216/*
217 * function for dealing with page's order in buddy system.
218 * zone->lock is already acquired when we use these.
219 * So, we don't need atomic page->flags operations here.
220 */
221static inline unsigned long page_order(struct page *page) {
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700222 return page_private(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223}
224
225static inline void set_page_order(struct page *page, int order) {
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700226 set_page_private(page, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227 __SetPagePrivate(page);
228}
229
230static inline void rmv_page_order(struct page *page)
231{
232 __ClearPagePrivate(page);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700233 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234}
235
236/*
237 * Locate the struct page for both the matching buddy in our
238 * pair (buddy1) and the combined O(n+1) page they form (page).
239 *
240 * 1) Any buddy B1 will have an order O twin B2 which satisfies
241 * the following equation:
242 * B2 = B1 ^ (1 << O)
243 * For example, if the starting buddy (buddy2) is #8 its order
244 * 1 buddy is #10:
245 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
246 *
247 * 2) Any buddy B will have an order O+1 parent P which
248 * satisfies the following equation:
249 * P = B & ~(1 << O)
250 *
251 * Assumption: *_mem_map is contigious at least up to MAX_ORDER
252 */
253static inline struct page *
254__page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order)
255{
256 unsigned long buddy_idx = page_idx ^ (1 << order);
257
258 return page + (buddy_idx - page_idx);
259}
260
261static inline unsigned long
262__find_combined_index(unsigned long page_idx, unsigned int order)
263{
264 return (page_idx & ~(1 << order));
265}
266
267/*
268 * This function checks whether a page is free && is the buddy
269 * we can do coalesce a page and its buddy if
Nick Piggin13e74442006-01-06 00:10:58 -0800270 * (a) the buddy is not in a hole &&
271 * (b) the buddy is free &&
272 * (c) the buddy is on the buddy system &&
273 * (d) a page and its buddy have the same order.
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700274 * for recording page's order, we use page_private(page) and PG_private.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275 *
276 */
277static inline int page_is_buddy(struct page *page, int order)
278{
Nick Piggin13e74442006-01-06 00:10:58 -0800279#ifdef CONFIG_HOLES_IN_ZONE
280 if (!pfn_valid(page_to_pfn(page)))
281 return 0;
282#endif
283
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284 if (PagePrivate(page) &&
285 (page_order(page) == order) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286 page_count(page) == 0)
287 return 1;
288 return 0;
289}
290
291/*
292 * Freeing function for a buddy system allocator.
293 *
294 * The concept of a buddy system is to maintain direct-mapped table
295 * (containing bit values) for memory blocks of various "orders".
296 * The bottom level table contains the map for the smallest allocatable
297 * units of memory (here, pages), and each level above it describes
298 * pairs of units from the levels below, hence, "buddies".
299 * At a high level, all that happens here is marking the table entry
300 * at the bottom level available, and propagating the changes upward
301 * as necessary, plus some accounting needed to play nicely with other
302 * parts of the VM system.
303 * At each level, we keep a list of pages, which are heads of continuous
304 * free pages of length of (1 << order) and marked with PG_Private.Page's
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700305 * order is recorded in page_private(page) field.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306 * So when we are allocating or freeing one, we can derive the state of the
307 * other. That is, if we allocate a small block, and both were
308 * free, the remainder of the region must be split into blocks.
309 * If a block is freed, and its buddy is also free, then this
310 * triggers coalescing into a block of larger size.
311 *
312 * -- wli
313 */
314
Nick Piggin48db57f2006-01-08 01:00:42 -0800315static inline void __free_one_page(struct page *page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316 struct zone *zone, unsigned int order)
317{
318 unsigned long page_idx;
319 int order_size = 1 << order;
320
Nick Piggin224abf92006-01-06 00:11:11 -0800321 if (unlikely(PageCompound(page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322 destroy_compound_page(page, order);
323
324 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
325
326 BUG_ON(page_idx & (order_size - 1));
327 BUG_ON(bad_range(zone, page));
328
329 zone->free_pages += order_size;
330 while (order < MAX_ORDER-1) {
331 unsigned long combined_idx;
332 struct free_area *area;
333 struct page *buddy;
334
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335 buddy = __page_find_buddy(page, page_idx, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336 if (!page_is_buddy(buddy, order))
337 break; /* Move the buddy up one level. */
Nick Piggin13e74442006-01-06 00:10:58 -0800338
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339 list_del(&buddy->lru);
340 area = zone->free_area + order;
341 area->nr_free--;
342 rmv_page_order(buddy);
Nick Piggin13e74442006-01-06 00:10:58 -0800343 combined_idx = __find_combined_index(page_idx, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344 page = page + (combined_idx - page_idx);
345 page_idx = combined_idx;
346 order++;
347 }
348 set_page_order(page, order);
349 list_add(&page->lru, &zone->free_area[order].free_list);
350 zone->free_area[order].nr_free++;
351}
352
Nick Piggin224abf92006-01-06 00:11:11 -0800353static inline int free_pages_check(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354{
Nick Piggin92be2e32006-01-06 00:10:57 -0800355 if (unlikely(page_mapcount(page) |
356 (page->mapping != NULL) |
357 (page_count(page) != 0) |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358 (page->flags & (
359 1 << PG_lru |
360 1 << PG_private |
361 1 << PG_locked |
362 1 << PG_active |
363 1 << PG_reclaim |
364 1 << PG_slab |
365 1 << PG_swapcache |
Nick Pigginb5810032005-10-29 18:16:12 -0700366 1 << PG_writeback |
Nick Piggin92be2e32006-01-06 00:10:57 -0800367 1 << PG_reserved ))))
Nick Piggin224abf92006-01-06 00:11:11 -0800368 bad_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 if (PageDirty(page))
Nick Piggin242e5462005-09-03 15:54:50 -0700370 __ClearPageDirty(page);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800371 /*
372 * For now, we report if PG_reserved was found set, but do not
373 * clear it, and do not free the page. But we shall soon need
374 * to do more, for when the ZERO_PAGE count wraps negative.
375 */
376 return PageReserved(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377}
378
379/*
380 * Frees a list of pages.
381 * Assumes all pages on list are in same zone, and of same order.
Renaud Lienhart207f36e2005-09-10 00:26:59 -0700382 * count is the number of pages to free.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383 *
384 * If the zone was previously in an "all pages pinned" state then look to
385 * see if this freeing clears that state.
386 *
387 * And clear the zone's pages_scanned counter, to hold off the "all pages are
388 * pinned" detection logic.
389 */
Nick Piggin48db57f2006-01-08 01:00:42 -0800390static void free_pages_bulk(struct zone *zone, int count,
391 struct list_head *list, int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392{
Nick Pigginc54ad302006-01-06 00:10:56 -0800393 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394 zone->all_unreclaimable = 0;
395 zone->pages_scanned = 0;
Nick Piggin48db57f2006-01-08 01:00:42 -0800396 while (count--) {
397 struct page *page;
398
399 BUG_ON(list_empty(list));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400 page = list_entry(list->prev, struct page, lru);
Nick Piggin48db57f2006-01-08 01:00:42 -0800401 /* have to delete it as __free_one_page list manipulates */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402 list_del(&page->lru);
Nick Piggin48db57f2006-01-08 01:00:42 -0800403 __free_one_page(page, zone, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404 }
Nick Pigginc54ad302006-01-06 00:10:56 -0800405 spin_unlock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406}
407
Nick Piggin48db57f2006-01-08 01:00:42 -0800408static void free_one_page(struct zone *zone, struct page *page, int order)
409{
410 LIST_HEAD(list);
411 list_add(&page->lru, &list);
412 free_pages_bulk(zone, 1, &list, order);
413}
414
415static void __free_pages_ok(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416{
Nick Pigginc54ad302006-01-06 00:10:56 -0800417 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418 int i;
Hugh Dickins689bceb2005-11-21 21:32:20 -0800419 int reserved = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420
421 arch_free_page(page, order);
Ingo Molnarde5097c2006-01-09 15:59:21 -0800422 if (!PageHighMem(page))
423 mutex_debug_check_no_locks_freed(page_address(page),
David Woodhousea4fc7ab2006-01-11 14:41:26 +0000424 PAGE_SIZE<<order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426#ifndef CONFIG_MMU
Nick Piggin48db57f2006-01-08 01:00:42 -0800427 for (i = 1 ; i < (1 << order) ; ++i)
428 __put_page(page + i);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429#endif
430
431 for (i = 0 ; i < (1 << order) ; ++i)
Nick Piggin224abf92006-01-06 00:11:11 -0800432 reserved += free_pages_check(page + i);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800433 if (reserved)
434 return;
435
Nick Piggin48db57f2006-01-08 01:00:42 -0800436 kernel_map_pages(page, 1 << order, 0);
Nick Pigginc54ad302006-01-06 00:10:56 -0800437 local_irq_save(flags);
Nick Piggina74609f2006-01-06 00:11:20 -0800438 __mod_page_state(pgfree, 1 << order);
Nick Piggin48db57f2006-01-08 01:00:42 -0800439 free_one_page(page_zone(page), page, order);
Nick Pigginc54ad302006-01-06 00:10:56 -0800440 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441}
442
David Howellsa226f6c2006-01-06 00:11:08 -0800443/*
444 * permit the bootmem allocator to evade page validation on high-order frees
445 */
446void fastcall __init __free_pages_bootmem(struct page *page, unsigned int order)
447{
448 if (order == 0) {
449 __ClearPageReserved(page);
450 set_page_count(page, 0);
451
452 free_hot_cold_page(page, 0);
453 } else {
454 LIST_HEAD(list);
455 int loop;
456
457 for (loop = 0; loop < BITS_PER_LONG; loop++) {
458 struct page *p = &page[loop];
459
460 if (loop + 16 < BITS_PER_LONG)
461 prefetchw(p + 16);
462 __ClearPageReserved(p);
463 set_page_count(p, 0);
464 }
465
466 arch_free_page(page, order);
467
468 mod_page_state(pgfree, 1 << order);
469
470 list_add(&page->lru, &list);
471 kernel_map_pages(page, 1 << order, 0);
472 free_pages_bulk(page_zone(page), 1, &list, order);
473 }
474}
475
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476
477/*
478 * The order of subdivision here is critical for the IO subsystem.
479 * Please do not alter this order without good reasons and regression
480 * testing. Specifically, as large blocks of memory are subdivided,
481 * the order in which smaller blocks are delivered depends on the order
482 * they're subdivided in this function. This is the primary factor
483 * influencing the order in which pages are delivered to the IO
484 * subsystem according to empirical testing, and this is also justified
485 * by considering the behavior of a buddy system containing a single
486 * large block of memory acted on by a series of small allocations.
487 * This behavior is a critical factor in sglist merging's success.
488 *
489 * -- wli
490 */
Nick Piggin085cc7d2006-01-06 00:11:01 -0800491static inline void expand(struct zone *zone, struct page *page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492 int low, int high, struct free_area *area)
493{
494 unsigned long size = 1 << high;
495
496 while (high > low) {
497 area--;
498 high--;
499 size >>= 1;
500 BUG_ON(bad_range(zone, &page[size]));
501 list_add(&page[size].lru, &area->free_list);
502 area->nr_free++;
503 set_page_order(&page[size], high);
504 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505}
506
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507/*
508 * This page is about to be returned from the page allocator
509 */
Hugh Dickins689bceb2005-11-21 21:32:20 -0800510static int prep_new_page(struct page *page, int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511{
Nick Piggin92be2e32006-01-06 00:10:57 -0800512 if (unlikely(page_mapcount(page) |
513 (page->mapping != NULL) |
514 (page_count(page) != 0) |
Hugh Dickins334795e2005-06-21 17:15:08 -0700515 (page->flags & (
516 1 << PG_lru |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517 1 << PG_private |
518 1 << PG_locked |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519 1 << PG_active |
520 1 << PG_dirty |
521 1 << PG_reclaim |
Hugh Dickins334795e2005-06-21 17:15:08 -0700522 1 << PG_slab |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523 1 << PG_swapcache |
Nick Pigginb5810032005-10-29 18:16:12 -0700524 1 << PG_writeback |
Nick Piggin92be2e32006-01-06 00:10:57 -0800525 1 << PG_reserved ))))
Nick Piggin224abf92006-01-06 00:11:11 -0800526 bad_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527
Hugh Dickins689bceb2005-11-21 21:32:20 -0800528 /*
529 * For now, we report if PG_reserved was found set, but do not
530 * clear it, and do not allocate the page: as a safety net.
531 */
532 if (PageReserved(page))
533 return 1;
534
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535 page->flags &= ~(1 << PG_uptodate | 1 << PG_error |
536 1 << PG_referenced | 1 << PG_arch_1 |
537 1 << PG_checked | 1 << PG_mappedtodisk);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700538 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539 set_page_refs(page, order);
540 kernel_map_pages(page, 1 << order, 1);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800541 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542}
543
544/*
545 * Do the hard work of removing an element from the buddy allocator.
546 * Call me with the zone->lock already held.
547 */
548static struct page *__rmqueue(struct zone *zone, unsigned int order)
549{
550 struct free_area * area;
551 unsigned int current_order;
552 struct page *page;
553
554 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
555 area = zone->free_area + current_order;
556 if (list_empty(&area->free_list))
557 continue;
558
559 page = list_entry(area->free_list.next, struct page, lru);
560 list_del(&page->lru);
561 rmv_page_order(page);
562 area->nr_free--;
563 zone->free_pages -= 1UL << order;
Nick Piggin085cc7d2006-01-06 00:11:01 -0800564 expand(zone, page, order, current_order, area);
565 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 }
567
568 return NULL;
569}
570
571/*
572 * Obtain a specified number of elements from the buddy allocator, all under
573 * a single hold of the lock, for efficiency. Add them to the supplied list.
574 * Returns the number of new pages which were placed at *list.
575 */
576static int rmqueue_bulk(struct zone *zone, unsigned int order,
577 unsigned long count, struct list_head *list)
578{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580
Nick Pigginc54ad302006-01-06 00:10:56 -0800581 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 for (i = 0; i < count; ++i) {
Nick Piggin085cc7d2006-01-06 00:11:01 -0800583 struct page *page = __rmqueue(zone, order);
584 if (unlikely(page == NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 list_add_tail(&page->lru, list);
587 }
Nick Pigginc54ad302006-01-06 00:10:56 -0800588 spin_unlock(&zone->lock);
Nick Piggin085cc7d2006-01-06 00:11:01 -0800589 return i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590}
591
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700592#ifdef CONFIG_NUMA
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800593/*
594 * Called from the slab reaper to drain pagesets on a particular node that
595 * belong to the currently executing processor.
596 */
597void drain_node_pages(int nodeid)
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700598{
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800599 int i, z;
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700600 unsigned long flags;
601
602 local_irq_save(flags);
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800603 for (z = 0; z < MAX_NR_ZONES; z++) {
604 struct zone *zone = NODE_DATA(nodeid)->node_zones + z;
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700605 struct per_cpu_pageset *pset;
606
Nick Piggin23316bc2006-01-08 01:00:41 -0800607 pset = zone_pcp(zone, smp_processor_id());
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700608 for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) {
609 struct per_cpu_pages *pcp;
610
611 pcp = &pset->pcp[i];
Nick Piggin48db57f2006-01-08 01:00:42 -0800612 free_pages_bulk(zone, pcp->count, &pcp->list, 0);
613 pcp->count = 0;
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700614 }
615 }
616 local_irq_restore(flags);
617}
618#endif
619
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620#if defined(CONFIG_PM) || defined(CONFIG_HOTPLUG_CPU)
621static void __drain_pages(unsigned int cpu)
622{
Nick Pigginc54ad302006-01-06 00:10:56 -0800623 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 struct zone *zone;
625 int i;
626
627 for_each_zone(zone) {
628 struct per_cpu_pageset *pset;
629
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700630 pset = zone_pcp(zone, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631 for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) {
632 struct per_cpu_pages *pcp;
633
634 pcp = &pset->pcp[i];
Nick Pigginc54ad302006-01-06 00:10:56 -0800635 local_irq_save(flags);
Nick Piggin48db57f2006-01-08 01:00:42 -0800636 free_pages_bulk(zone, pcp->count, &pcp->list, 0);
637 pcp->count = 0;
Nick Pigginc54ad302006-01-06 00:10:56 -0800638 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 }
640 }
641}
642#endif /* CONFIG_PM || CONFIG_HOTPLUG_CPU */
643
644#ifdef CONFIG_PM
645
646void mark_free_pages(struct zone *zone)
647{
648 unsigned long zone_pfn, flags;
649 int order;
650 struct list_head *curr;
651
652 if (!zone->spanned_pages)
653 return;
654
655 spin_lock_irqsave(&zone->lock, flags);
656 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn)
657 ClearPageNosaveFree(pfn_to_page(zone_pfn + zone->zone_start_pfn));
658
659 for (order = MAX_ORDER - 1; order >= 0; --order)
660 list_for_each(curr, &zone->free_area[order].free_list) {
661 unsigned long start_pfn, i;
662
663 start_pfn = page_to_pfn(list_entry(curr, struct page, lru));
664
665 for (i=0; i < (1<<order); i++)
666 SetPageNosaveFree(pfn_to_page(start_pfn+i));
667 }
668 spin_unlock_irqrestore(&zone->lock, flags);
669}
670
671/*
672 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
673 */
674void drain_local_pages(void)
675{
676 unsigned long flags;
677
678 local_irq_save(flags);
679 __drain_pages(smp_processor_id());
680 local_irq_restore(flags);
681}
682#endif /* CONFIG_PM */
683
Nick Piggina74609f2006-01-06 00:11:20 -0800684static void zone_statistics(struct zonelist *zonelist, struct zone *z, int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685{
686#ifdef CONFIG_NUMA
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 pg_data_t *pg = z->zone_pgdat;
688 pg_data_t *orig = zonelist->zones[0]->zone_pgdat;
689 struct per_cpu_pageset *p;
690
Nick Piggina74609f2006-01-06 00:11:20 -0800691 p = zone_pcp(z, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692 if (pg == orig) {
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700693 p->numa_hit++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694 } else {
695 p->numa_miss++;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700696 zone_pcp(zonelist->zones[0], cpu)->numa_foreign++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 }
698 if (pg == NODE_DATA(numa_node_id()))
699 p->local_node++;
700 else
701 p->other_node++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702#endif
703}
704
705/*
706 * Free a 0-order page
707 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708static void fastcall free_hot_cold_page(struct page *page, int cold)
709{
710 struct zone *zone = page_zone(page);
711 struct per_cpu_pages *pcp;
712 unsigned long flags;
713
714 arch_free_page(page, 0);
715
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 if (PageAnon(page))
717 page->mapping = NULL;
Nick Piggin224abf92006-01-06 00:11:11 -0800718 if (free_pages_check(page))
Hugh Dickins689bceb2005-11-21 21:32:20 -0800719 return;
720
Hugh Dickins689bceb2005-11-21 21:32:20 -0800721 kernel_map_pages(page, 1, 0);
722
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700723 pcp = &zone_pcp(zone, get_cpu())->pcp[cold];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724 local_irq_save(flags);
Nick Piggina74609f2006-01-06 00:11:20 -0800725 __inc_page_state(pgfree);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 list_add(&page->lru, &pcp->list);
727 pcp->count++;
Nick Piggin48db57f2006-01-08 01:00:42 -0800728 if (pcp->count >= pcp->high) {
729 free_pages_bulk(zone, pcp->batch, &pcp->list, 0);
730 pcp->count -= pcp->batch;
731 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732 local_irq_restore(flags);
733 put_cpu();
734}
735
736void fastcall free_hot_page(struct page *page)
737{
738 free_hot_cold_page(page, 0);
739}
740
741void fastcall free_cold_page(struct page *page)
742{
743 free_hot_cold_page(page, 1);
744}
745
Al Virodd0fc662005-10-07 07:46:04 +0100746static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747{
748 int i;
749
750 BUG_ON((gfp_flags & (__GFP_WAIT | __GFP_HIGHMEM)) == __GFP_HIGHMEM);
751 for(i = 0; i < (1 << order); i++)
752 clear_highpage(page + i);
753}
754
755/*
756 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
757 * we cheat by calling it from here, in the order > 0 path. Saves a branch
758 * or two.
759 */
Nick Piggina74609f2006-01-06 00:11:20 -0800760static struct page *buffered_rmqueue(struct zonelist *zonelist,
761 struct zone *zone, int order, gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762{
763 unsigned long flags;
Hugh Dickins689bceb2005-11-21 21:32:20 -0800764 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765 int cold = !!(gfp_flags & __GFP_COLD);
Nick Piggina74609f2006-01-06 00:11:20 -0800766 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767
Hugh Dickins689bceb2005-11-21 21:32:20 -0800768again:
Nick Piggina74609f2006-01-06 00:11:20 -0800769 cpu = get_cpu();
Nick Piggin48db57f2006-01-08 01:00:42 -0800770 if (likely(order == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771 struct per_cpu_pages *pcp;
772
Nick Piggina74609f2006-01-06 00:11:20 -0800773 pcp = &zone_pcp(zone, cpu)->pcp[cold];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774 local_irq_save(flags);
Nick Piggina74609f2006-01-06 00:11:20 -0800775 if (!pcp->count) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 pcp->count += rmqueue_bulk(zone, 0,
777 pcp->batch, &pcp->list);
Nick Piggina74609f2006-01-06 00:11:20 -0800778 if (unlikely(!pcp->count))
779 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780 }
Nick Piggina74609f2006-01-06 00:11:20 -0800781 page = list_entry(pcp->list.next, struct page, lru);
782 list_del(&page->lru);
783 pcp->count--;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800784 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785 spin_lock_irqsave(&zone->lock, flags);
786 page = __rmqueue(zone, order);
Nick Piggina74609f2006-01-06 00:11:20 -0800787 spin_unlock(&zone->lock);
788 if (!page)
789 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790 }
791
Nick Piggina74609f2006-01-06 00:11:20 -0800792 __mod_page_state_zone(zone, pgalloc, 1 << order);
793 zone_statistics(zonelist, zone, cpu);
794 local_irq_restore(flags);
795 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796
Nick Piggina74609f2006-01-06 00:11:20 -0800797 BUG_ON(bad_range(zone, page));
798 if (prep_new_page(page, order))
799 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800
Nick Piggina74609f2006-01-06 00:11:20 -0800801 if (gfp_flags & __GFP_ZERO)
802 prep_zero_page(page, order, gfp_flags);
803
804 if (order && (gfp_flags & __GFP_COMP))
805 prep_compound_page(page, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806 return page;
Nick Piggina74609f2006-01-06 00:11:20 -0800807
808failed:
809 local_irq_restore(flags);
810 put_cpu();
811 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812}
813
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800814#define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
Nick Piggin31488902005-11-28 13:44:03 -0800815#define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
816#define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
817#define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
818#define ALLOC_HARDER 0x10 /* try to alloc harder */
819#define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
820#define ALLOC_CPUSET 0x40 /* check for correct cpuset */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800821
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822/*
823 * Return 1 if free pages are above 'mark'. This takes into account the order
824 * of the allocation.
825 */
826int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800827 int classzone_idx, int alloc_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828{
829 /* free_pages my go negative - that's OK */
830 long min = mark, free_pages = z->free_pages - (1 << order) + 1;
831 int o;
832
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800833 if (alloc_flags & ALLOC_HIGH)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834 min -= min / 2;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800835 if (alloc_flags & ALLOC_HARDER)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836 min -= min / 4;
837
838 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
839 return 0;
840 for (o = 0; o < order; o++) {
841 /* At the next order, this order's pages become unavailable */
842 free_pages -= z->free_area[o].nr_free << o;
843
844 /* Require fewer higher order pages to be free */
845 min >>= 1;
846
847 if (free_pages <= min)
848 return 0;
849 }
850 return 1;
851}
852
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800853/*
854 * get_page_from_freeliest goes through the zonelist trying to allocate
855 * a page.
856 */
857static struct page *
858get_page_from_freelist(gfp_t gfp_mask, unsigned int order,
859 struct zonelist *zonelist, int alloc_flags)
Martin Hicks753ee722005-06-21 17:14:41 -0700860{
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800861 struct zone **z = zonelist->zones;
862 struct page *page = NULL;
863 int classzone_idx = zone_idx(*z);
864
865 /*
866 * Go through the zonelist once, looking for a zone with enough free.
867 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
868 */
869 do {
870 if ((alloc_flags & ALLOC_CPUSET) &&
871 !cpuset_zone_allowed(*z, gfp_mask))
872 continue;
873
874 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
Nick Piggin31488902005-11-28 13:44:03 -0800875 unsigned long mark;
876 if (alloc_flags & ALLOC_WMARK_MIN)
877 mark = (*z)->pages_min;
878 else if (alloc_flags & ALLOC_WMARK_LOW)
879 mark = (*z)->pages_low;
880 else
881 mark = (*z)->pages_high;
882 if (!zone_watermark_ok(*z, order, mark,
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800883 classzone_idx, alloc_flags))
Christoph Lameter9eeff232006-01-18 17:42:31 -0800884 if (!zone_reclaim_mode ||
885 !zone_reclaim(*z, gfp_mask, order))
886 continue;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800887 }
888
Nick Piggina74609f2006-01-06 00:11:20 -0800889 page = buffered_rmqueue(zonelist, *z, order, gfp_mask);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800890 if (page) {
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800891 break;
892 }
893 } while (*(++z) != NULL);
894 return page;
Martin Hicks753ee722005-06-21 17:14:41 -0700895}
896
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897/*
898 * This is the 'heart' of the zoned buddy allocator.
899 */
900struct page * fastcall
Al Virodd0fc662005-10-07 07:46:04 +0100901__alloc_pages(gfp_t gfp_mask, unsigned int order,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 struct zonelist *zonelist)
903{
Al Viro260b2362005-10-21 03:22:44 -0400904 const gfp_t wait = gfp_mask & __GFP_WAIT;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800905 struct zone **z;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906 struct page *page;
907 struct reclaim_state reclaim_state;
908 struct task_struct *p = current;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909 int do_retry;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800910 int alloc_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911 int did_some_progress;
912
913 might_sleep_if(wait);
914
Jens Axboe6b1de912005-11-17 21:35:02 +0100915restart:
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800916 z = zonelist->zones; /* the list of zones suitable for gfp_mask */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800918 if (unlikely(*z == NULL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919 /* Should this ever happen?? */
920 return NULL;
921 }
Jens Axboe6b1de912005-11-17 21:35:02 +0100922
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800923 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
Nick Piggin31488902005-11-28 13:44:03 -0800924 zonelist, ALLOC_WMARK_LOW|ALLOC_CPUSET);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800925 if (page)
926 goto got_pg;
927
Jens Axboe6b1de912005-11-17 21:35:02 +0100928 do {
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800929 wakeup_kswapd(*z, order);
Jens Axboe6b1de912005-11-17 21:35:02 +0100930 } while (*(++z));
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800931
Paul Jackson9bf22292005-09-06 15:18:12 -0700932 /*
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800933 * OK, we're below the kswapd watermark and have kicked background
934 * reclaim. Now things get more complex, so set up alloc_flags according
935 * to how we want to proceed.
936 *
937 * The caller may dip into page reserves a bit more if the caller
938 * cannot run direct reclaim, or if the caller has realtime scheduling
Paul Jackson4eac9152006-01-11 12:17:19 -0800939 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
940 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
Paul Jackson9bf22292005-09-06 15:18:12 -0700941 */
Nick Piggin31488902005-11-28 13:44:03 -0800942 alloc_flags = ALLOC_WMARK_MIN;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800943 if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait)
944 alloc_flags |= ALLOC_HARDER;
945 if (gfp_mask & __GFP_HIGH)
946 alloc_flags |= ALLOC_HIGH;
Paul Jackson47f3a862006-01-06 00:10:32 -0800947 alloc_flags |= ALLOC_CPUSET;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948
949 /*
950 * Go through the zonelist again. Let __GFP_HIGH and allocations
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800951 * coming from realtime tasks go deeper into reserves.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952 *
953 * This is the last chance, in general, before the goto nopage.
954 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
Paul Jackson9bf22292005-09-06 15:18:12 -0700955 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956 */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800957 page = get_page_from_freelist(gfp_mask, order, zonelist, alloc_flags);
958 if (page)
959 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960
961 /* This allocation should allow future memory freeing. */
Nick Pigginb84a35b2005-05-01 08:58:36 -0700962
963 if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE)))
964 && !in_interrupt()) {
965 if (!(gfp_mask & __GFP_NOMEMALLOC)) {
Kirill Korotaev885036d2005-11-13 16:06:41 -0800966nofail_alloc:
Nick Pigginb84a35b2005-05-01 08:58:36 -0700967 /* go through the zonelist yet again, ignoring mins */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800968 page = get_page_from_freelist(gfp_mask, order,
Paul Jackson47f3a862006-01-06 00:10:32 -0800969 zonelist, ALLOC_NO_WATERMARKS);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800970 if (page)
971 goto got_pg;
Kirill Korotaev885036d2005-11-13 16:06:41 -0800972 if (gfp_mask & __GFP_NOFAIL) {
973 blk_congestion_wait(WRITE, HZ/50);
974 goto nofail_alloc;
975 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976 }
977 goto nopage;
978 }
979
980 /* Atomic allocations - we can't balance anything */
981 if (!wait)
982 goto nopage;
983
984rebalance:
985 cond_resched();
986
987 /* We now go into synchronous reclaim */
Paul Jackson3e0d98b2006-01-08 01:01:49 -0800988 cpuset_memory_pressure_bump();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989 p->flags |= PF_MEMALLOC;
990 reclaim_state.reclaimed_slab = 0;
991 p->reclaim_state = &reclaim_state;
992
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800993 did_some_progress = try_to_free_pages(zonelist->zones, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994
995 p->reclaim_state = NULL;
996 p->flags &= ~PF_MEMALLOC;
997
998 cond_resched();
999
1000 if (likely(did_some_progress)) {
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001001 page = get_page_from_freelist(gfp_mask, order,
1002 zonelist, alloc_flags);
1003 if (page)
1004 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005 } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
1006 /*
1007 * Go through the zonelist yet one more time, keep
1008 * very high watermark here, this is only to catch
1009 * a parallel oom killing, we must fail if we're still
1010 * under heavy pressure.
1011 */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001012 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
Nick Piggin31488902005-11-28 13:44:03 -08001013 zonelist, ALLOC_WMARK_HIGH|ALLOC_CPUSET);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001014 if (page)
1015 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016
Christoph Lameter9b0f8b02006-02-20 18:27:52 -08001017 out_of_memory(zonelist, gfp_mask, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 goto restart;
1019 }
1020
1021 /*
1022 * Don't let big-order allocations loop unless the caller explicitly
1023 * requests that. Wait for some write requests to complete then retry.
1024 *
1025 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
1026 * <= 3, but that may not be true in other implementations.
1027 */
1028 do_retry = 0;
1029 if (!(gfp_mask & __GFP_NORETRY)) {
1030 if ((order <= 3) || (gfp_mask & __GFP_REPEAT))
1031 do_retry = 1;
1032 if (gfp_mask & __GFP_NOFAIL)
1033 do_retry = 1;
1034 }
1035 if (do_retry) {
1036 blk_congestion_wait(WRITE, HZ/50);
1037 goto rebalance;
1038 }
1039
1040nopage:
1041 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
1042 printk(KERN_WARNING "%s: page allocation failure."
1043 " order:%d, mode:0x%x\n",
1044 p->comm, order, gfp_mask);
1045 dump_stack();
Janet Morgan578c2fd2005-06-21 17:14:56 -07001046 show_mem();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048got_pg:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049 return page;
1050}
1051
1052EXPORT_SYMBOL(__alloc_pages);
1053
1054/*
1055 * Common helper functions.
1056 */
Al Virodd0fc662005-10-07 07:46:04 +01001057fastcall unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058{
1059 struct page * page;
1060 page = alloc_pages(gfp_mask, order);
1061 if (!page)
1062 return 0;
1063 return (unsigned long) page_address(page);
1064}
1065
1066EXPORT_SYMBOL(__get_free_pages);
1067
Al Virodd0fc662005-10-07 07:46:04 +01001068fastcall unsigned long get_zeroed_page(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069{
1070 struct page * page;
1071
1072 /*
1073 * get_zeroed_page() returns a 32-bit address, which cannot represent
1074 * a highmem page
1075 */
Al Viro260b2362005-10-21 03:22:44 -04001076 BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077
1078 page = alloc_pages(gfp_mask | __GFP_ZERO, 0);
1079 if (page)
1080 return (unsigned long) page_address(page);
1081 return 0;
1082}
1083
1084EXPORT_SYMBOL(get_zeroed_page);
1085
1086void __pagevec_free(struct pagevec *pvec)
1087{
1088 int i = pagevec_count(pvec);
1089
1090 while (--i >= 0)
1091 free_hot_cold_page(pvec->pages[i], pvec->cold);
1092}
1093
1094fastcall void __free_pages(struct page *page, unsigned int order)
1095{
Nick Pigginb5810032005-10-29 18:16:12 -07001096 if (put_page_testzero(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097 if (order == 0)
1098 free_hot_page(page);
1099 else
1100 __free_pages_ok(page, order);
1101 }
1102}
1103
1104EXPORT_SYMBOL(__free_pages);
1105
1106fastcall void free_pages(unsigned long addr, unsigned int order)
1107{
1108 if (addr != 0) {
1109 BUG_ON(!virt_addr_valid((void *)addr));
1110 __free_pages(virt_to_page((void *)addr), order);
1111 }
1112}
1113
1114EXPORT_SYMBOL(free_pages);
1115
1116/*
1117 * Total amount of free (allocatable) RAM:
1118 */
1119unsigned int nr_free_pages(void)
1120{
1121 unsigned int sum = 0;
1122 struct zone *zone;
1123
1124 for_each_zone(zone)
1125 sum += zone->free_pages;
1126
1127 return sum;
1128}
1129
1130EXPORT_SYMBOL(nr_free_pages);
1131
1132#ifdef CONFIG_NUMA
1133unsigned int nr_free_pages_pgdat(pg_data_t *pgdat)
1134{
1135 unsigned int i, sum = 0;
1136
1137 for (i = 0; i < MAX_NR_ZONES; i++)
1138 sum += pgdat->node_zones[i].free_pages;
1139
1140 return sum;
1141}
1142#endif
1143
1144static unsigned int nr_free_zone_pages(int offset)
1145{
Martin J. Blighe310fd42005-07-29 22:59:18 -07001146 /* Just pick one node, since fallback list is circular */
1147 pg_data_t *pgdat = NODE_DATA(numa_node_id());
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148 unsigned int sum = 0;
1149
Martin J. Blighe310fd42005-07-29 22:59:18 -07001150 struct zonelist *zonelist = pgdat->node_zonelists + offset;
1151 struct zone **zonep = zonelist->zones;
1152 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153
Martin J. Blighe310fd42005-07-29 22:59:18 -07001154 for (zone = *zonep++; zone; zone = *zonep++) {
1155 unsigned long size = zone->present_pages;
1156 unsigned long high = zone->pages_high;
1157 if (size > high)
1158 sum += size - high;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159 }
1160
1161 return sum;
1162}
1163
1164/*
1165 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1166 */
1167unsigned int nr_free_buffer_pages(void)
1168{
Al Viroaf4ca452005-10-21 02:55:38 -04001169 return nr_free_zone_pages(gfp_zone(GFP_USER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170}
1171
1172/*
1173 * Amount of free RAM allocatable within all zones
1174 */
1175unsigned int nr_free_pagecache_pages(void)
1176{
Al Viroaf4ca452005-10-21 02:55:38 -04001177 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178}
1179
1180#ifdef CONFIG_HIGHMEM
1181unsigned int nr_free_highpages (void)
1182{
1183 pg_data_t *pgdat;
1184 unsigned int pages = 0;
1185
1186 for_each_pgdat(pgdat)
1187 pages += pgdat->node_zones[ZONE_HIGHMEM].free_pages;
1188
1189 return pages;
1190}
1191#endif
1192
1193#ifdef CONFIG_NUMA
1194static void show_node(struct zone *zone)
1195{
1196 printk("Node %d ", zone->zone_pgdat->node_id);
1197}
1198#else
1199#define show_node(zone) do { } while (0)
1200#endif
1201
1202/*
1203 * Accumulate the page_state information across all CPUs.
1204 * The result is unavoidably approximate - it can change
1205 * during and after execution of this function.
1206 */
1207static DEFINE_PER_CPU(struct page_state, page_states) = {0};
1208
1209atomic_t nr_pagecache = ATOMIC_INIT(0);
1210EXPORT_SYMBOL(nr_pagecache);
1211#ifdef CONFIG_SMP
1212DEFINE_PER_CPU(long, nr_pagecache_local) = 0;
1213#endif
1214
Nick Piggina86b1f52006-01-06 00:11:00 -08001215static void __get_page_state(struct page_state *ret, int nr, cpumask_t *cpumask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216{
1217 int cpu = 0;
1218
Eric Dumazet88a2a4ac2006-02-04 23:27:36 -08001219 memset(ret, 0, nr * sizeof(unsigned long));
Andrew Morton84c20082006-01-08 01:00:28 -08001220 cpus_and(*cpumask, *cpumask, cpu_online_map);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221
Martin Hicksc07e02d2005-09-03 15:55:11 -07001222 cpu = first_cpu(*cpumask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 while (cpu < NR_CPUS) {
1224 unsigned long *in, *out, off;
1225
Eric Dumazet88a2a4ac2006-02-04 23:27:36 -08001226 if (!cpu_isset(cpu, *cpumask))
1227 continue;
1228
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229 in = (unsigned long *)&per_cpu(page_states, cpu);
1230
Martin Hicksc07e02d2005-09-03 15:55:11 -07001231 cpu = next_cpu(cpu, *cpumask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232
Eric Dumazet88a2a4ac2006-02-04 23:27:36 -08001233 if (likely(cpu < NR_CPUS))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234 prefetch(&per_cpu(page_states, cpu));
1235
1236 out = (unsigned long *)ret;
1237 for (off = 0; off < nr; off++)
1238 *out++ += *in++;
1239 }
1240}
1241
Martin Hicksc07e02d2005-09-03 15:55:11 -07001242void get_page_state_node(struct page_state *ret, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243{
1244 int nr;
Martin Hicksc07e02d2005-09-03 15:55:11 -07001245 cpumask_t mask = node_to_cpumask(node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246
1247 nr = offsetof(struct page_state, GET_PAGE_STATE_LAST);
1248 nr /= sizeof(unsigned long);
1249
Martin Hicksc07e02d2005-09-03 15:55:11 -07001250 __get_page_state(ret, nr+1, &mask);
1251}
1252
1253void get_page_state(struct page_state *ret)
1254{
1255 int nr;
1256 cpumask_t mask = CPU_MASK_ALL;
1257
1258 nr = offsetof(struct page_state, GET_PAGE_STATE_LAST);
1259 nr /= sizeof(unsigned long);
1260
1261 __get_page_state(ret, nr + 1, &mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262}
1263
1264void get_full_page_state(struct page_state *ret)
1265{
Martin Hicksc07e02d2005-09-03 15:55:11 -07001266 cpumask_t mask = CPU_MASK_ALL;
1267
1268 __get_page_state(ret, sizeof(*ret) / sizeof(unsigned long), &mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269}
1270
Nick Piggina74609f2006-01-06 00:11:20 -08001271unsigned long read_page_state_offset(unsigned long offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272{
1273 unsigned long ret = 0;
1274 int cpu;
1275
Andrew Morton84c20082006-01-08 01:00:28 -08001276 for_each_online_cpu(cpu) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277 unsigned long in;
1278
1279 in = (unsigned long)&per_cpu(page_states, cpu) + offset;
1280 ret += *((unsigned long *)in);
1281 }
1282 return ret;
1283}
1284
Nick Piggina74609f2006-01-06 00:11:20 -08001285void __mod_page_state_offset(unsigned long offset, unsigned long delta)
1286{
1287 void *ptr;
1288
1289 ptr = &__get_cpu_var(page_states);
1290 *(unsigned long *)(ptr + offset) += delta;
1291}
1292EXPORT_SYMBOL(__mod_page_state_offset);
1293
1294void mod_page_state_offset(unsigned long offset, unsigned long delta)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295{
1296 unsigned long flags;
Nick Piggina74609f2006-01-06 00:11:20 -08001297 void *ptr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298
1299 local_irq_save(flags);
1300 ptr = &__get_cpu_var(page_states);
Nick Piggina74609f2006-01-06 00:11:20 -08001301 *(unsigned long *)(ptr + offset) += delta;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302 local_irq_restore(flags);
1303}
Nick Piggina74609f2006-01-06 00:11:20 -08001304EXPORT_SYMBOL(mod_page_state_offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305
1306void __get_zone_counts(unsigned long *active, unsigned long *inactive,
1307 unsigned long *free, struct pglist_data *pgdat)
1308{
1309 struct zone *zones = pgdat->node_zones;
1310 int i;
1311
1312 *active = 0;
1313 *inactive = 0;
1314 *free = 0;
1315 for (i = 0; i < MAX_NR_ZONES; i++) {
1316 *active += zones[i].nr_active;
1317 *inactive += zones[i].nr_inactive;
1318 *free += zones[i].free_pages;
1319 }
1320}
1321
1322void get_zone_counts(unsigned long *active,
1323 unsigned long *inactive, unsigned long *free)
1324{
1325 struct pglist_data *pgdat;
1326
1327 *active = 0;
1328 *inactive = 0;
1329 *free = 0;
1330 for_each_pgdat(pgdat) {
1331 unsigned long l, m, n;
1332 __get_zone_counts(&l, &m, &n, pgdat);
1333 *active += l;
1334 *inactive += m;
1335 *free += n;
1336 }
1337}
1338
1339void si_meminfo(struct sysinfo *val)
1340{
1341 val->totalram = totalram_pages;
1342 val->sharedram = 0;
1343 val->freeram = nr_free_pages();
1344 val->bufferram = nr_blockdev_pages();
1345#ifdef CONFIG_HIGHMEM
1346 val->totalhigh = totalhigh_pages;
1347 val->freehigh = nr_free_highpages();
1348#else
1349 val->totalhigh = 0;
1350 val->freehigh = 0;
1351#endif
1352 val->mem_unit = PAGE_SIZE;
1353}
1354
1355EXPORT_SYMBOL(si_meminfo);
1356
1357#ifdef CONFIG_NUMA
1358void si_meminfo_node(struct sysinfo *val, int nid)
1359{
1360 pg_data_t *pgdat = NODE_DATA(nid);
1361
1362 val->totalram = pgdat->node_present_pages;
1363 val->freeram = nr_free_pages_pgdat(pgdat);
1364 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1365 val->freehigh = pgdat->node_zones[ZONE_HIGHMEM].free_pages;
1366 val->mem_unit = PAGE_SIZE;
1367}
1368#endif
1369
1370#define K(x) ((x) << (PAGE_SHIFT-10))
1371
1372/*
1373 * Show free area list (used inside shift_scroll-lock stuff)
1374 * We also calculate the percentage fragmentation. We do this by counting the
1375 * memory on each free list with the exception of the first item on the list.
1376 */
1377void show_free_areas(void)
1378{
1379 struct page_state ps;
1380 int cpu, temperature;
1381 unsigned long active;
1382 unsigned long inactive;
1383 unsigned long free;
1384 struct zone *zone;
1385
1386 for_each_zone(zone) {
1387 show_node(zone);
1388 printk("%s per-cpu:", zone->name);
1389
Con Kolivasf3fe6512006-01-06 00:11:15 -08001390 if (!populated_zone(zone)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391 printk(" empty\n");
1392 continue;
1393 } else
1394 printk("\n");
1395
Dave Jones6b482c62005-11-10 15:45:56 -05001396 for_each_online_cpu(cpu) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397 struct per_cpu_pageset *pageset;
1398
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001399 pageset = zone_pcp(zone, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400
1401 for (temperature = 0; temperature < 2; temperature++)
Nick Piggin2d92c5c2006-01-06 00:10:59 -08001402 printk("cpu %d %s: high %d, batch %d used:%d\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403 cpu,
1404 temperature ? "cold" : "hot",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405 pageset->pcp[temperature].high,
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001406 pageset->pcp[temperature].batch,
1407 pageset->pcp[temperature].count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408 }
1409 }
1410
1411 get_page_state(&ps);
1412 get_zone_counts(&active, &inactive, &free);
1413
Denis Vlasenkoc0d62212005-06-21 17:15:14 -07001414 printk("Free pages: %11ukB (%ukB HighMem)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415 K(nr_free_pages()),
1416 K(nr_free_highpages()));
1417
1418 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu "
1419 "unstable:%lu free:%u slab:%lu mapped:%lu pagetables:%lu\n",
1420 active,
1421 inactive,
1422 ps.nr_dirty,
1423 ps.nr_writeback,
1424 ps.nr_unstable,
1425 nr_free_pages(),
1426 ps.nr_slab,
1427 ps.nr_mapped,
1428 ps.nr_page_table_pages);
1429
1430 for_each_zone(zone) {
1431 int i;
1432
1433 show_node(zone);
1434 printk("%s"
1435 " free:%lukB"
1436 " min:%lukB"
1437 " low:%lukB"
1438 " high:%lukB"
1439 " active:%lukB"
1440 " inactive:%lukB"
1441 " present:%lukB"
1442 " pages_scanned:%lu"
1443 " all_unreclaimable? %s"
1444 "\n",
1445 zone->name,
1446 K(zone->free_pages),
1447 K(zone->pages_min),
1448 K(zone->pages_low),
1449 K(zone->pages_high),
1450 K(zone->nr_active),
1451 K(zone->nr_inactive),
1452 K(zone->present_pages),
1453 zone->pages_scanned,
1454 (zone->all_unreclaimable ? "yes" : "no")
1455 );
1456 printk("lowmem_reserve[]:");
1457 for (i = 0; i < MAX_NR_ZONES; i++)
1458 printk(" %lu", zone->lowmem_reserve[i]);
1459 printk("\n");
1460 }
1461
1462 for_each_zone(zone) {
1463 unsigned long nr, flags, order, total = 0;
1464
1465 show_node(zone);
1466 printk("%s: ", zone->name);
Con Kolivasf3fe6512006-01-06 00:11:15 -08001467 if (!populated_zone(zone)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468 printk("empty\n");
1469 continue;
1470 }
1471
1472 spin_lock_irqsave(&zone->lock, flags);
1473 for (order = 0; order < MAX_ORDER; order++) {
1474 nr = zone->free_area[order].nr_free;
1475 total += nr << order;
1476 printk("%lu*%lukB ", nr, K(1UL) << order);
1477 }
1478 spin_unlock_irqrestore(&zone->lock, flags);
1479 printk("= %lukB\n", K(total));
1480 }
1481
1482 show_swap_cache_info();
1483}
1484
1485/*
1486 * Builds allocation fallback zone lists.
Christoph Lameter1a932052006-01-06 00:11:16 -08001487 *
1488 * Add all populated zones of a node to the zonelist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489 */
Christoph Lameter1a932052006-01-06 00:11:16 -08001490static int __init build_zonelists_node(pg_data_t *pgdat,
Christoph Lameter070f8032006-01-06 00:11:19 -08001491 struct zonelist *zonelist, int nr_zones, int zone_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492{
Christoph Lameter1a932052006-01-06 00:11:16 -08001493 struct zone *zone;
1494
Christoph Lameter070f8032006-01-06 00:11:19 -08001495 BUG_ON(zone_type > ZONE_HIGHMEM);
Christoph Lameter02a68a52006-01-06 00:11:18 -08001496
1497 do {
Christoph Lameter070f8032006-01-06 00:11:19 -08001498 zone = pgdat->node_zones + zone_type;
Christoph Lameter1a932052006-01-06 00:11:16 -08001499 if (populated_zone(zone)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500#ifndef CONFIG_HIGHMEM
Christoph Lameter070f8032006-01-06 00:11:19 -08001501 BUG_ON(zone_type > ZONE_NORMAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502#endif
Christoph Lameter070f8032006-01-06 00:11:19 -08001503 zonelist->zones[nr_zones++] = zone;
1504 check_highest_zone(zone_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001505 }
Christoph Lameter070f8032006-01-06 00:11:19 -08001506 zone_type--;
Christoph Lameter02a68a52006-01-06 00:11:18 -08001507
Christoph Lameter070f8032006-01-06 00:11:19 -08001508 } while (zone_type >= 0);
1509 return nr_zones;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510}
1511
Al Viro260b2362005-10-21 03:22:44 -04001512static inline int highest_zone(int zone_bits)
1513{
1514 int res = ZONE_NORMAL;
1515 if (zone_bits & (__force int)__GFP_HIGHMEM)
1516 res = ZONE_HIGHMEM;
Andi Kleena2f1b422005-11-05 17:25:53 +01001517 if (zone_bits & (__force int)__GFP_DMA32)
1518 res = ZONE_DMA32;
Al Viro260b2362005-10-21 03:22:44 -04001519 if (zone_bits & (__force int)__GFP_DMA)
1520 res = ZONE_DMA;
1521 return res;
1522}
1523
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524#ifdef CONFIG_NUMA
1525#define MAX_NODE_LOAD (num_online_nodes())
1526static int __initdata node_load[MAX_NUMNODES];
1527/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -07001528 * 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 -07001529 * @node: node whose fallback list we're appending
1530 * @used_node_mask: nodemask_t of already used nodes
1531 *
1532 * We use a number of factors to determine which is the next node that should
1533 * appear on a given node's fallback list. The node should not have appeared
1534 * already in @node's fallback list, and it should be the next closest node
1535 * according to the distance array (which contains arbitrary distance values
1536 * from each node to each node in the system), and should also prefer nodes
1537 * with no CPUs, since presumably they'll have very little allocation pressure
1538 * on them otherwise.
1539 * It returns -1 if no node is found.
1540 */
1541static int __init find_next_best_node(int node, nodemask_t *used_node_mask)
1542{
Linus Torvalds4cf808e2006-02-17 20:38:21 +01001543 int n, val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544 int min_val = INT_MAX;
1545 int best_node = -1;
1546
Linus Torvalds4cf808e2006-02-17 20:38:21 +01001547 /* Use the local node if we haven't already */
1548 if (!node_isset(node, *used_node_mask)) {
1549 node_set(node, *used_node_mask);
1550 return node;
1551 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552
Linus Torvalds4cf808e2006-02-17 20:38:21 +01001553 for_each_online_node(n) {
1554 cpumask_t tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555
1556 /* Don't want a node to appear more than once */
1557 if (node_isset(n, *used_node_mask))
1558 continue;
1559
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560 /* Use the distance array to find the distance */
1561 val = node_distance(node, n);
1562
Linus Torvalds4cf808e2006-02-17 20:38:21 +01001563 /* Penalize nodes under us ("prefer the next node") */
1564 val += (n < node);
1565
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566 /* Give preference to headless and unused nodes */
1567 tmp = node_to_cpumask(n);
1568 if (!cpus_empty(tmp))
1569 val += PENALTY_FOR_NODE_WITH_CPUS;
1570
1571 /* Slight preference for less loaded node */
1572 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
1573 val += node_load[n];
1574
1575 if (val < min_val) {
1576 min_val = val;
1577 best_node = n;
1578 }
1579 }
1580
1581 if (best_node >= 0)
1582 node_set(best_node, *used_node_mask);
1583
1584 return best_node;
1585}
1586
1587static void __init build_zonelists(pg_data_t *pgdat)
1588{
1589 int i, j, k, node, local_node;
1590 int prev_node, load;
1591 struct zonelist *zonelist;
1592 nodemask_t used_mask;
1593
1594 /* initialize zonelists */
1595 for (i = 0; i < GFP_ZONETYPES; i++) {
1596 zonelist = pgdat->node_zonelists + i;
1597 zonelist->zones[0] = NULL;
1598 }
1599
1600 /* NUMA-aware ordering of nodes */
1601 local_node = pgdat->node_id;
1602 load = num_online_nodes();
1603 prev_node = local_node;
1604 nodes_clear(used_mask);
1605 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
Christoph Lameter9eeff232006-01-18 17:42:31 -08001606 int distance = node_distance(local_node, node);
1607
1608 /*
1609 * If another node is sufficiently far away then it is better
1610 * to reclaim pages in a zone before going off node.
1611 */
1612 if (distance > RECLAIM_DISTANCE)
1613 zone_reclaim_mode = 1;
1614
Linus Torvalds1da177e2005-04-16 15:20:36 -07001615 /*
1616 * We don't want to pressure a particular node.
1617 * So adding penalty to the first node in same
1618 * distance group to make it round-robin.
1619 */
Christoph Lameter9eeff232006-01-18 17:42:31 -08001620
1621 if (distance != node_distance(local_node, prev_node))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622 node_load[node] += load;
1623 prev_node = node;
1624 load--;
1625 for (i = 0; i < GFP_ZONETYPES; i++) {
1626 zonelist = pgdat->node_zonelists + i;
1627 for (j = 0; zonelist->zones[j] != NULL; j++);
1628
Al Viro260b2362005-10-21 03:22:44 -04001629 k = highest_zone(i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630
1631 j = build_zonelists_node(NODE_DATA(node), zonelist, j, k);
1632 zonelist->zones[j] = NULL;
1633 }
1634 }
1635}
1636
1637#else /* CONFIG_NUMA */
1638
1639static void __init build_zonelists(pg_data_t *pgdat)
1640{
1641 int i, j, k, node, local_node;
1642
1643 local_node = pgdat->node_id;
1644 for (i = 0; i < GFP_ZONETYPES; i++) {
1645 struct zonelist *zonelist;
1646
1647 zonelist = pgdat->node_zonelists + i;
1648
1649 j = 0;
Al Viro260b2362005-10-21 03:22:44 -04001650 k = highest_zone(i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651 j = build_zonelists_node(pgdat, zonelist, j, k);
1652 /*
1653 * Now we build the zonelist so that it contains the zones
1654 * of all the other nodes.
1655 * We don't want to pressure a particular node, so when
1656 * building the zones for node N, we make sure that the
1657 * zones coming right after the local ones are those from
1658 * node N+1 (modulo N)
1659 */
1660 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
1661 if (!node_online(node))
1662 continue;
1663 j = build_zonelists_node(NODE_DATA(node), zonelist, j, k);
1664 }
1665 for (node = 0; node < local_node; node++) {
1666 if (!node_online(node))
1667 continue;
1668 j = build_zonelists_node(NODE_DATA(node), zonelist, j, k);
1669 }
1670
1671 zonelist->zones[j] = NULL;
1672 }
1673}
1674
1675#endif /* CONFIG_NUMA */
1676
1677void __init build_all_zonelists(void)
1678{
1679 int i;
1680
1681 for_each_online_node(i)
1682 build_zonelists(NODE_DATA(i));
1683 printk("Built %i zonelists\n", num_online_nodes());
1684 cpuset_init_current_mems_allowed();
1685}
1686
1687/*
1688 * Helper functions to size the waitqueue hash table.
1689 * Essentially these want to choose hash table sizes sufficiently
1690 * large so that collisions trying to wait on pages are rare.
1691 * But in fact, the number of active page waitqueues on typical
1692 * systems is ridiculously low, less than 200. So this is even
1693 * conservative, even though it seems large.
1694 *
1695 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
1696 * waitqueues, i.e. the size of the waitq table given the number of pages.
1697 */
1698#define PAGES_PER_WAITQUEUE 256
1699
1700static inline unsigned long wait_table_size(unsigned long pages)
1701{
1702 unsigned long size = 1;
1703
1704 pages /= PAGES_PER_WAITQUEUE;
1705
1706 while (size < pages)
1707 size <<= 1;
1708
1709 /*
1710 * Once we have dozens or even hundreds of threads sleeping
1711 * on IO we've got bigger problems than wait queue collision.
1712 * Limit the size of the wait table to a reasonable size.
1713 */
1714 size = min(size, 4096UL);
1715
1716 return max(size, 4UL);
1717}
1718
1719/*
1720 * This is an integer logarithm so that shifts can be used later
1721 * to extract the more random high bits from the multiplicative
1722 * hash function before the remainder is taken.
1723 */
1724static inline unsigned long wait_table_bits(unsigned long size)
1725{
1726 return ffz(~size);
1727}
1728
1729#define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
1730
1731static void __init calculate_zone_totalpages(struct pglist_data *pgdat,
1732 unsigned long *zones_size, unsigned long *zholes_size)
1733{
1734 unsigned long realtotalpages, totalpages = 0;
1735 int i;
1736
1737 for (i = 0; i < MAX_NR_ZONES; i++)
1738 totalpages += zones_size[i];
1739 pgdat->node_spanned_pages = totalpages;
1740
1741 realtotalpages = totalpages;
1742 if (zholes_size)
1743 for (i = 0; i < MAX_NR_ZONES; i++)
1744 realtotalpages -= zholes_size[i];
1745 pgdat->node_present_pages = realtotalpages;
1746 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, realtotalpages);
1747}
1748
1749
1750/*
1751 * Initially all pages are reserved - free ones are freed
1752 * up by free_all_bootmem() once the early boot process is
1753 * done. Non-atomic initialization, single-pass.
1754 */
Matt Tolentinoc09b4242006-01-17 07:03:44 +01001755void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756 unsigned long start_pfn)
1757{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001758 struct page *page;
Andy Whitcroft29751f62005-06-23 00:08:00 -07001759 unsigned long end_pfn = start_pfn + size;
1760 unsigned long pfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761
Greg Ungerercbe8dd42006-01-12 01:05:24 -08001762 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001763 if (!early_pfn_valid(pfn))
1764 continue;
1765 page = pfn_to_page(pfn);
1766 set_page_links(page, zone, nid, pfn);
Nick Pigginb5810032005-10-29 18:16:12 -07001767 set_page_count(page, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768 reset_page_mapcount(page);
1769 SetPageReserved(page);
1770 INIT_LIST_HEAD(&page->lru);
1771#ifdef WANT_PAGE_VIRTUAL
1772 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
1773 if (!is_highmem_idx(zone))
Bob Picco3212c6b2005-06-27 14:36:28 -07001774 set_page_address(page, __va(pfn << PAGE_SHIFT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001776 }
1777}
1778
1779void zone_init_free_lists(struct pglist_data *pgdat, struct zone *zone,
1780 unsigned long size)
1781{
1782 int order;
1783 for (order = 0; order < MAX_ORDER ; order++) {
1784 INIT_LIST_HEAD(&zone->free_area[order].free_list);
1785 zone->free_area[order].nr_free = 0;
1786 }
1787}
1788
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001789#define ZONETABLE_INDEX(x, zone_nr) ((x << ZONES_SHIFT) | zone_nr)
1790void zonetable_add(struct zone *zone, int nid, int zid, unsigned long pfn,
1791 unsigned long size)
1792{
1793 unsigned long snum = pfn_to_section_nr(pfn);
1794 unsigned long end = pfn_to_section_nr(pfn + size);
1795
1796 if (FLAGS_HAS_NODE)
1797 zone_table[ZONETABLE_INDEX(nid, zid)] = zone;
1798 else
1799 for (; snum <= end; snum++)
1800 zone_table[ZONETABLE_INDEX(snum, zid)] = zone;
1801}
1802
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803#ifndef __HAVE_ARCH_MEMMAP_INIT
1804#define memmap_init(size, nid, zone, start_pfn) \
1805 memmap_init_zone((size), (nid), (zone), (start_pfn))
1806#endif
1807
Ashok Raj6292d9a2006-02-01 03:04:44 -08001808static int __cpuinit zone_batchsize(struct zone *zone)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001809{
1810 int batch;
1811
1812 /*
1813 * The per-cpu-pages pools are set to around 1000th of the
Seth, Rohitba56e912005-10-29 18:15:47 -07001814 * size of the zone. But no more than 1/2 of a meg.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001815 *
1816 * OK, so we don't know how big the cache is. So guess.
1817 */
1818 batch = zone->present_pages / 1024;
Seth, Rohitba56e912005-10-29 18:15:47 -07001819 if (batch * PAGE_SIZE > 512 * 1024)
1820 batch = (512 * 1024) / PAGE_SIZE;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001821 batch /= 4; /* We effectively *= 4 below */
1822 if (batch < 1)
1823 batch = 1;
1824
1825 /*
Nick Piggin0ceaacc2005-12-04 13:55:25 +11001826 * Clamp the batch to a 2^n - 1 value. Having a power
1827 * of 2 value was found to be more likely to have
1828 * suboptimal cache aliasing properties in some cases.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001829 *
Nick Piggin0ceaacc2005-12-04 13:55:25 +11001830 * For example if 2 tasks are alternately allocating
1831 * batches of pages, one task can end up with a lot
1832 * of pages of one half of the possible page colors
1833 * and the other with pages of the other colors.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001834 */
Nick Piggin0ceaacc2005-12-04 13:55:25 +11001835 batch = (1 << (fls(batch + batch/2)-1)) - 1;
Seth, Rohitba56e912005-10-29 18:15:47 -07001836
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001837 return batch;
1838}
1839
Christoph Lameter2caaad42005-06-21 17:15:00 -07001840inline void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
1841{
1842 struct per_cpu_pages *pcp;
1843
Magnus Damm1c6fe942005-10-26 01:58:59 -07001844 memset(p, 0, sizeof(*p));
1845
Christoph Lameter2caaad42005-06-21 17:15:00 -07001846 pcp = &p->pcp[0]; /* hot */
1847 pcp->count = 0;
Christoph Lameter2caaad42005-06-21 17:15:00 -07001848 pcp->high = 6 * batch;
1849 pcp->batch = max(1UL, 1 * batch);
1850 INIT_LIST_HEAD(&pcp->list);
1851
1852 pcp = &p->pcp[1]; /* cold*/
1853 pcp->count = 0;
Christoph Lameter2caaad42005-06-21 17:15:00 -07001854 pcp->high = 2 * batch;
Seth, Rohite46a5e22005-10-29 18:15:48 -07001855 pcp->batch = max(1UL, batch/2);
Christoph Lameter2caaad42005-06-21 17:15:00 -07001856 INIT_LIST_HEAD(&pcp->list);
1857}
1858
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08001859/*
1860 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
1861 * to the value high for the pageset p.
1862 */
1863
1864static void setup_pagelist_highmark(struct per_cpu_pageset *p,
1865 unsigned long high)
1866{
1867 struct per_cpu_pages *pcp;
1868
1869 pcp = &p->pcp[0]; /* hot list */
1870 pcp->high = high;
1871 pcp->batch = max(1UL, high/4);
1872 if ((high/4) > (PAGE_SHIFT * 8))
1873 pcp->batch = PAGE_SHIFT * 8;
1874}
1875
1876
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001877#ifdef CONFIG_NUMA
1878/*
Christoph Lameter2caaad42005-06-21 17:15:00 -07001879 * Boot pageset table. One per cpu which is going to be used for all
1880 * zones and all nodes. The parameters will be set in such a way
1881 * that an item put on a list will immediately be handed over to
1882 * the buddy list. This is safe since pageset manipulation is done
1883 * with interrupts disabled.
1884 *
1885 * Some NUMA counter updates may also be caught by the boot pagesets.
Christoph Lameterb7c84c62005-06-22 20:26:07 -07001886 *
1887 * The boot_pagesets must be kept even after bootup is complete for
1888 * unused processors and/or zones. They do play a role for bootstrapping
1889 * hotplugged processors.
1890 *
1891 * zoneinfo_show() and maybe other functions do
1892 * not check if the processor is online before following the pageset pointer.
1893 * Other parts of the kernel may not check if the zone is available.
Christoph Lameter2caaad42005-06-21 17:15:00 -07001894 */
Eric Dumazet88a2a4ac2006-02-04 23:27:36 -08001895static struct per_cpu_pageset boot_pageset[NR_CPUS];
Christoph Lameter2caaad42005-06-21 17:15:00 -07001896
1897/*
1898 * Dynamically allocate memory for the
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001899 * per cpu pageset array in struct zone.
1900 */
Ashok Raj6292d9a2006-02-01 03:04:44 -08001901static int __cpuinit process_zones(int cpu)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001902{
1903 struct zone *zone, *dzone;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001904
1905 for_each_zone(zone) {
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001906
Nick Piggin23316bc2006-01-08 01:00:41 -08001907 zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset),
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001908 GFP_KERNEL, cpu_to_node(cpu));
Nick Piggin23316bc2006-01-08 01:00:41 -08001909 if (!zone_pcp(zone, cpu))
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001910 goto bad;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001911
Nick Piggin23316bc2006-01-08 01:00:41 -08001912 setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone));
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08001913
1914 if (percpu_pagelist_fraction)
1915 setup_pagelist_highmark(zone_pcp(zone, cpu),
1916 (zone->present_pages / percpu_pagelist_fraction));
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001917 }
1918
1919 return 0;
1920bad:
1921 for_each_zone(dzone) {
1922 if (dzone == zone)
1923 break;
Nick Piggin23316bc2006-01-08 01:00:41 -08001924 kfree(zone_pcp(dzone, cpu));
1925 zone_pcp(dzone, cpu) = NULL;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001926 }
1927 return -ENOMEM;
1928}
1929
1930static inline void free_zone_pagesets(int cpu)
1931{
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001932 struct zone *zone;
1933
1934 for_each_zone(zone) {
1935 struct per_cpu_pageset *pset = zone_pcp(zone, cpu);
1936
1937 zone_pcp(zone, cpu) = NULL;
1938 kfree(pset);
1939 }
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001940}
1941
Ashok Raj6292d9a2006-02-01 03:04:44 -08001942static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb,
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001943 unsigned long action,
1944 void *hcpu)
1945{
1946 int cpu = (long)hcpu;
1947 int ret = NOTIFY_OK;
1948
1949 switch (action) {
1950 case CPU_UP_PREPARE:
1951 if (process_zones(cpu))
1952 ret = NOTIFY_BAD;
1953 break;
Andi Kleenb0d41692005-11-05 17:25:53 +01001954 case CPU_UP_CANCELED:
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001955 case CPU_DEAD:
1956 free_zone_pagesets(cpu);
1957 break;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001958 default:
1959 break;
1960 }
1961 return ret;
1962}
1963
1964static struct notifier_block pageset_notifier =
1965 { &pageset_cpuup_callback, NULL, 0 };
1966
Al Viro78d99552005-12-15 09:18:25 +00001967void __init setup_per_cpu_pageset(void)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001968{
1969 int err;
1970
1971 /* Initialize per_cpu_pageset for cpu 0.
1972 * A cpuup callback will do this for every cpu
1973 * as it comes online
1974 */
1975 err = process_zones(smp_processor_id());
1976 BUG_ON(err);
1977 register_cpu_notifier(&pageset_notifier);
1978}
1979
1980#endif
1981
Matt Tolentinoc09b4242006-01-17 07:03:44 +01001982static __meminit
Dave Hansened8ece22005-10-29 18:16:50 -07001983void zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
1984{
1985 int i;
1986 struct pglist_data *pgdat = zone->zone_pgdat;
1987
1988 /*
1989 * The per-page waitqueue mechanism uses hashed waitqueues
1990 * per zone.
1991 */
1992 zone->wait_table_size = wait_table_size(zone_size_pages);
1993 zone->wait_table_bits = wait_table_bits(zone->wait_table_size);
1994 zone->wait_table = (wait_queue_head_t *)
1995 alloc_bootmem_node(pgdat, zone->wait_table_size
1996 * sizeof(wait_queue_head_t));
1997
1998 for(i = 0; i < zone->wait_table_size; ++i)
1999 init_waitqueue_head(zone->wait_table + i);
2000}
2001
Matt Tolentinoc09b4242006-01-17 07:03:44 +01002002static __meminit void zone_pcp_init(struct zone *zone)
Dave Hansened8ece22005-10-29 18:16:50 -07002003{
2004 int cpu;
2005 unsigned long batch = zone_batchsize(zone);
2006
2007 for (cpu = 0; cpu < NR_CPUS; cpu++) {
2008#ifdef CONFIG_NUMA
2009 /* Early boot. Slab allocator not functional yet */
Nick Piggin23316bc2006-01-08 01:00:41 -08002010 zone_pcp(zone, cpu) = &boot_pageset[cpu];
Dave Hansened8ece22005-10-29 18:16:50 -07002011 setup_pageset(&boot_pageset[cpu],0);
2012#else
2013 setup_pageset(zone_pcp(zone,cpu), batch);
2014#endif
2015 }
2016 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n",
2017 zone->name, zone->present_pages, batch);
2018}
2019
Matt Tolentinoc09b4242006-01-17 07:03:44 +01002020static __meminit void init_currently_empty_zone(struct zone *zone,
Dave Hansened8ece22005-10-29 18:16:50 -07002021 unsigned long zone_start_pfn, unsigned long size)
2022{
2023 struct pglist_data *pgdat = zone->zone_pgdat;
2024
2025 zone_wait_table_init(zone, size);
2026 pgdat->nr_zones = zone_idx(zone) + 1;
2027
2028 zone->zone_mem_map = pfn_to_page(zone_start_pfn);
2029 zone->zone_start_pfn = zone_start_pfn;
2030
2031 memmap_init(size, pgdat->node_id, zone_idx(zone), zone_start_pfn);
2032
2033 zone_init_free_lists(pgdat, zone, zone->spanned_pages);
2034}
2035
Linus Torvalds1da177e2005-04-16 15:20:36 -07002036/*
2037 * Set up the zone data structures:
2038 * - mark all pages reserved
2039 * - mark all memory queues empty
2040 * - clear the memory bitmaps
2041 */
2042static void __init free_area_init_core(struct pglist_data *pgdat,
2043 unsigned long *zones_size, unsigned long *zholes_size)
2044{
Dave Hansened8ece22005-10-29 18:16:50 -07002045 unsigned long j;
2046 int nid = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047 unsigned long zone_start_pfn = pgdat->node_start_pfn;
2048
Dave Hansen208d54e2005-10-29 18:16:52 -07002049 pgdat_resize_init(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002050 pgdat->nr_zones = 0;
2051 init_waitqueue_head(&pgdat->kswapd_wait);
2052 pgdat->kswapd_max_order = 0;
2053
2054 for (j = 0; j < MAX_NR_ZONES; j++) {
2055 struct zone *zone = pgdat->node_zones + j;
2056 unsigned long size, realsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002057
Linus Torvalds1da177e2005-04-16 15:20:36 -07002058 realsize = size = zones_size[j];
2059 if (zholes_size)
2060 realsize -= zholes_size[j];
2061
Andi Kleena2f1b422005-11-05 17:25:53 +01002062 if (j < ZONE_HIGHMEM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002063 nr_kernel_pages += realsize;
2064 nr_all_pages += realsize;
2065
2066 zone->spanned_pages = size;
2067 zone->present_pages = realsize;
2068 zone->name = zone_names[j];
2069 spin_lock_init(&zone->lock);
2070 spin_lock_init(&zone->lru_lock);
Dave Hansenbdc8cb92005-10-29 18:16:53 -07002071 zone_seqlock_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072 zone->zone_pgdat = pgdat;
2073 zone->free_pages = 0;
2074
2075 zone->temp_priority = zone->prev_priority = DEF_PRIORITY;
2076
Dave Hansened8ece22005-10-29 18:16:50 -07002077 zone_pcp_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002078 INIT_LIST_HEAD(&zone->active_list);
2079 INIT_LIST_HEAD(&zone->inactive_list);
2080 zone->nr_scan_active = 0;
2081 zone->nr_scan_inactive = 0;
2082 zone->nr_active = 0;
2083 zone->nr_inactive = 0;
Martin Hicks53e9a612005-09-03 15:54:51 -07002084 atomic_set(&zone->reclaim_in_progress, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002085 if (!size)
2086 continue;
2087
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002088 zonetable_add(zone, nid, j, zone_start_pfn, size);
Dave Hansened8ece22005-10-29 18:16:50 -07002089 init_currently_empty_zone(zone, zone_start_pfn, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002090 zone_start_pfn += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002091 }
2092}
2093
2094static void __init alloc_node_mem_map(struct pglist_data *pgdat)
2095{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002096 /* Skip empty nodes */
2097 if (!pgdat->node_spanned_pages)
2098 return;
2099
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002100#ifdef CONFIG_FLAT_NODE_MEM_MAP
Linus Torvalds1da177e2005-04-16 15:20:36 -07002101 /* ia64 gets its own node_mem_map, before this, without bootmem */
2102 if (!pgdat->node_mem_map) {
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002103 unsigned long size;
2104 struct page *map;
2105
Linus Torvalds1da177e2005-04-16 15:20:36 -07002106 size = (pgdat->node_spanned_pages + 1) * sizeof(struct page);
Dave Hansen6f167ec2005-06-23 00:07:39 -07002107 map = alloc_remap(pgdat->node_id, size);
2108 if (!map)
2109 map = alloc_bootmem_node(pgdat, size);
2110 pgdat->node_mem_map = map;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002111 }
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002112#ifdef CONFIG_FLATMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002113 /*
2114 * With no DISCONTIG, the global mem_map is just set as node 0's
2115 */
2116 if (pgdat == NODE_DATA(0))
2117 mem_map = NODE_DATA(0)->node_mem_map;
2118#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002119#endif /* CONFIG_FLAT_NODE_MEM_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002120}
2121
2122void __init free_area_init_node(int nid, struct pglist_data *pgdat,
2123 unsigned long *zones_size, unsigned long node_start_pfn,
2124 unsigned long *zholes_size)
2125{
2126 pgdat->node_id = nid;
2127 pgdat->node_start_pfn = node_start_pfn;
2128 calculate_zone_totalpages(pgdat, zones_size, zholes_size);
2129
2130 alloc_node_mem_map(pgdat);
2131
2132 free_area_init_core(pgdat, zones_size, zholes_size);
2133}
2134
Dave Hansen93b75042005-06-23 00:07:47 -07002135#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07002136static bootmem_data_t contig_bootmem_data;
2137struct pglist_data contig_page_data = { .bdata = &contig_bootmem_data };
2138
2139EXPORT_SYMBOL(contig_page_data);
Dave Hansen93b75042005-06-23 00:07:47 -07002140#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002141
2142void __init free_area_init(unsigned long *zones_size)
2143{
Dave Hansen93b75042005-06-23 00:07:47 -07002144 free_area_init_node(0, NODE_DATA(0), zones_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002145 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
2146}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002147
2148#ifdef CONFIG_PROC_FS
2149
2150#include <linux/seq_file.h>
2151
2152static void *frag_start(struct seq_file *m, loff_t *pos)
2153{
2154 pg_data_t *pgdat;
2155 loff_t node = *pos;
2156
2157 for (pgdat = pgdat_list; pgdat && node; pgdat = pgdat->pgdat_next)
2158 --node;
2159
2160 return pgdat;
2161}
2162
2163static void *frag_next(struct seq_file *m, void *arg, loff_t *pos)
2164{
2165 pg_data_t *pgdat = (pg_data_t *)arg;
2166
2167 (*pos)++;
2168 return pgdat->pgdat_next;
2169}
2170
2171static void frag_stop(struct seq_file *m, void *arg)
2172{
2173}
2174
2175/*
2176 * This walks the free areas for each zone.
2177 */
2178static int frag_show(struct seq_file *m, void *arg)
2179{
2180 pg_data_t *pgdat = (pg_data_t *)arg;
2181 struct zone *zone;
2182 struct zone *node_zones = pgdat->node_zones;
2183 unsigned long flags;
2184 int order;
2185
2186 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
Con Kolivasf3fe6512006-01-06 00:11:15 -08002187 if (!populated_zone(zone))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002188 continue;
2189
2190 spin_lock_irqsave(&zone->lock, flags);
2191 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
2192 for (order = 0; order < MAX_ORDER; ++order)
2193 seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
2194 spin_unlock_irqrestore(&zone->lock, flags);
2195 seq_putc(m, '\n');
2196 }
2197 return 0;
2198}
2199
2200struct seq_operations fragmentation_op = {
2201 .start = frag_start,
2202 .next = frag_next,
2203 .stop = frag_stop,
2204 .show = frag_show,
2205};
2206
Nikita Danilov295ab932005-06-21 17:14:38 -07002207/*
2208 * Output information about zones in @pgdat.
2209 */
2210static int zoneinfo_show(struct seq_file *m, void *arg)
2211{
2212 pg_data_t *pgdat = arg;
2213 struct zone *zone;
2214 struct zone *node_zones = pgdat->node_zones;
2215 unsigned long flags;
2216
2217 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; zone++) {
2218 int i;
2219
Con Kolivasf3fe6512006-01-06 00:11:15 -08002220 if (!populated_zone(zone))
Nikita Danilov295ab932005-06-21 17:14:38 -07002221 continue;
2222
2223 spin_lock_irqsave(&zone->lock, flags);
2224 seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name);
2225 seq_printf(m,
2226 "\n pages free %lu"
2227 "\n min %lu"
2228 "\n low %lu"
2229 "\n high %lu"
2230 "\n active %lu"
2231 "\n inactive %lu"
2232 "\n scanned %lu (a: %lu i: %lu)"
2233 "\n spanned %lu"
2234 "\n present %lu",
2235 zone->free_pages,
2236 zone->pages_min,
2237 zone->pages_low,
2238 zone->pages_high,
2239 zone->nr_active,
2240 zone->nr_inactive,
2241 zone->pages_scanned,
2242 zone->nr_scan_active, zone->nr_scan_inactive,
2243 zone->spanned_pages,
2244 zone->present_pages);
2245 seq_printf(m,
2246 "\n protection: (%lu",
2247 zone->lowmem_reserve[0]);
2248 for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
2249 seq_printf(m, ", %lu", zone->lowmem_reserve[i]);
2250 seq_printf(m,
2251 ")"
2252 "\n pagesets");
Nick Piggin23316bc2006-01-08 01:00:41 -08002253 for_each_online_cpu(i) {
Nikita Danilov295ab932005-06-21 17:14:38 -07002254 struct per_cpu_pageset *pageset;
2255 int j;
2256
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002257 pageset = zone_pcp(zone, i);
Nikita Danilov295ab932005-06-21 17:14:38 -07002258 for (j = 0; j < ARRAY_SIZE(pageset->pcp); j++) {
2259 if (pageset->pcp[j].count)
2260 break;
2261 }
2262 if (j == ARRAY_SIZE(pageset->pcp))
2263 continue;
2264 for (j = 0; j < ARRAY_SIZE(pageset->pcp); j++) {
2265 seq_printf(m,
2266 "\n cpu: %i pcp: %i"
2267 "\n count: %i"
Nikita Danilov295ab932005-06-21 17:14:38 -07002268 "\n high: %i"
2269 "\n batch: %i",
2270 i, j,
2271 pageset->pcp[j].count,
Nikita Danilov295ab932005-06-21 17:14:38 -07002272 pageset->pcp[j].high,
2273 pageset->pcp[j].batch);
2274 }
2275#ifdef CONFIG_NUMA
2276 seq_printf(m,
2277 "\n numa_hit: %lu"
2278 "\n numa_miss: %lu"
2279 "\n numa_foreign: %lu"
2280 "\n interleave_hit: %lu"
2281 "\n local_node: %lu"
2282 "\n other_node: %lu",
2283 pageset->numa_hit,
2284 pageset->numa_miss,
2285 pageset->numa_foreign,
2286 pageset->interleave_hit,
2287 pageset->local_node,
2288 pageset->other_node);
2289#endif
2290 }
2291 seq_printf(m,
2292 "\n all_unreclaimable: %u"
2293 "\n prev_priority: %i"
2294 "\n temp_priority: %i"
2295 "\n start_pfn: %lu",
2296 zone->all_unreclaimable,
2297 zone->prev_priority,
2298 zone->temp_priority,
2299 zone->zone_start_pfn);
2300 spin_unlock_irqrestore(&zone->lock, flags);
2301 seq_putc(m, '\n');
2302 }
2303 return 0;
2304}
2305
2306struct seq_operations zoneinfo_op = {
2307 .start = frag_start, /* iterate over all zones. The same as in
2308 * fragmentation. */
2309 .next = frag_next,
2310 .stop = frag_stop,
2311 .show = zoneinfo_show,
2312};
2313
Linus Torvalds1da177e2005-04-16 15:20:36 -07002314static char *vmstat_text[] = {
2315 "nr_dirty",
2316 "nr_writeback",
2317 "nr_unstable",
2318 "nr_page_table_pages",
2319 "nr_mapped",
2320 "nr_slab",
2321
2322 "pgpgin",
2323 "pgpgout",
2324 "pswpin",
2325 "pswpout",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002326
Nick Piggin9328b8f2006-01-06 00:11:10 -08002327 "pgalloc_high",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002328 "pgalloc_normal",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002329 "pgalloc_dma32",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002330 "pgalloc_dma",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002331
Linus Torvalds1da177e2005-04-16 15:20:36 -07002332 "pgfree",
2333 "pgactivate",
2334 "pgdeactivate",
2335
2336 "pgfault",
2337 "pgmajfault",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002338
Linus Torvalds1da177e2005-04-16 15:20:36 -07002339 "pgrefill_high",
2340 "pgrefill_normal",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002341 "pgrefill_dma32",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002342 "pgrefill_dma",
2343
2344 "pgsteal_high",
2345 "pgsteal_normal",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002346 "pgsteal_dma32",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002347 "pgsteal_dma",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002348
Linus Torvalds1da177e2005-04-16 15:20:36 -07002349 "pgscan_kswapd_high",
2350 "pgscan_kswapd_normal",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002351 "pgscan_kswapd_dma32",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002352 "pgscan_kswapd_dma",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002353
Linus Torvalds1da177e2005-04-16 15:20:36 -07002354 "pgscan_direct_high",
2355 "pgscan_direct_normal",
Nick Piggin9328b8f2006-01-06 00:11:10 -08002356 "pgscan_direct_dma32",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002357 "pgscan_direct_dma",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002358
Nick Piggin9328b8f2006-01-06 00:11:10 -08002359 "pginodesteal",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002360 "slabs_scanned",
2361 "kswapd_steal",
2362 "kswapd_inodesteal",
2363 "pageoutrun",
2364 "allocstall",
2365
2366 "pgrotated",
KAMEZAWA Hiroyukiedfbe2b2005-05-01 08:58:37 -07002367 "nr_bounce",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002368};
2369
2370static void *vmstat_start(struct seq_file *m, loff_t *pos)
2371{
2372 struct page_state *ps;
2373
2374 if (*pos >= ARRAY_SIZE(vmstat_text))
2375 return NULL;
2376
2377 ps = kmalloc(sizeof(*ps), GFP_KERNEL);
2378 m->private = ps;
2379 if (!ps)
2380 return ERR_PTR(-ENOMEM);
2381 get_full_page_state(ps);
2382 ps->pgpgin /= 2; /* sectors -> kbytes */
2383 ps->pgpgout /= 2;
2384 return (unsigned long *)ps + *pos;
2385}
2386
2387static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
2388{
2389 (*pos)++;
2390 if (*pos >= ARRAY_SIZE(vmstat_text))
2391 return NULL;
2392 return (unsigned long *)m->private + *pos;
2393}
2394
2395static int vmstat_show(struct seq_file *m, void *arg)
2396{
2397 unsigned long *l = arg;
2398 unsigned long off = l - (unsigned long *)m->private;
2399
2400 seq_printf(m, "%s %lu\n", vmstat_text[off], *l);
2401 return 0;
2402}
2403
2404static void vmstat_stop(struct seq_file *m, void *arg)
2405{
2406 kfree(m->private);
2407 m->private = NULL;
2408}
2409
2410struct seq_operations vmstat_op = {
2411 .start = vmstat_start,
2412 .next = vmstat_next,
2413 .stop = vmstat_stop,
2414 .show = vmstat_show,
2415};
2416
2417#endif /* CONFIG_PROC_FS */
2418
2419#ifdef CONFIG_HOTPLUG_CPU
2420static int page_alloc_cpu_notify(struct notifier_block *self,
2421 unsigned long action, void *hcpu)
2422{
2423 int cpu = (unsigned long)hcpu;
2424 long *count;
2425 unsigned long *src, *dest;
2426
2427 if (action == CPU_DEAD) {
2428 int i;
2429
2430 /* Drain local pagecache count. */
2431 count = &per_cpu(nr_pagecache_local, cpu);
2432 atomic_add(*count, &nr_pagecache);
2433 *count = 0;
2434 local_irq_disable();
2435 __drain_pages(cpu);
2436
2437 /* Add dead cpu's page_states to our own. */
2438 dest = (unsigned long *)&__get_cpu_var(page_states);
2439 src = (unsigned long *)&per_cpu(page_states, cpu);
2440
2441 for (i = 0; i < sizeof(struct page_state)/sizeof(unsigned long);
2442 i++) {
2443 dest[i] += src[i];
2444 src[i] = 0;
2445 }
2446
2447 local_irq_enable();
2448 }
2449 return NOTIFY_OK;
2450}
2451#endif /* CONFIG_HOTPLUG_CPU */
2452
2453void __init page_alloc_init(void)
2454{
2455 hotcpu_notifier(page_alloc_cpu_notify, 0);
2456}
2457
2458/*
2459 * setup_per_zone_lowmem_reserve - called whenever
2460 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
2461 * has a correct pages reserved value, so an adequate number of
2462 * pages are left in the zone after a successful __alloc_pages().
2463 */
2464static void setup_per_zone_lowmem_reserve(void)
2465{
2466 struct pglist_data *pgdat;
2467 int j, idx;
2468
2469 for_each_pgdat(pgdat) {
2470 for (j = 0; j < MAX_NR_ZONES; j++) {
2471 struct zone *zone = pgdat->node_zones + j;
2472 unsigned long present_pages = zone->present_pages;
2473
2474 zone->lowmem_reserve[j] = 0;
2475
2476 for (idx = j-1; idx >= 0; idx--) {
2477 struct zone *lower_zone;
2478
2479 if (sysctl_lowmem_reserve_ratio[idx] < 1)
2480 sysctl_lowmem_reserve_ratio[idx] = 1;
2481
2482 lower_zone = pgdat->node_zones + idx;
2483 lower_zone->lowmem_reserve[j] = present_pages /
2484 sysctl_lowmem_reserve_ratio[idx];
2485 present_pages += lower_zone->present_pages;
2486 }
2487 }
2488 }
2489}
2490
2491/*
2492 * setup_per_zone_pages_min - called when min_free_kbytes changes. Ensures
2493 * that the pages_{min,low,high} values for each zone are set correctly
2494 * with respect to min_free_kbytes.
2495 */
Dave Hansen3947be12005-10-29 18:16:54 -07002496void setup_per_zone_pages_min(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002497{
2498 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
2499 unsigned long lowmem_pages = 0;
2500 struct zone *zone;
2501 unsigned long flags;
2502
2503 /* Calculate total number of !ZONE_HIGHMEM pages */
2504 for_each_zone(zone) {
2505 if (!is_highmem(zone))
2506 lowmem_pages += zone->present_pages;
2507 }
2508
2509 for_each_zone(zone) {
Nick Piggin669ed172005-11-13 16:06:45 -08002510 unsigned long tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002511 spin_lock_irqsave(&zone->lru_lock, flags);
Nick Piggin669ed172005-11-13 16:06:45 -08002512 tmp = (pages_min * zone->present_pages) / lowmem_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002513 if (is_highmem(zone)) {
2514 /*
Nick Piggin669ed172005-11-13 16:06:45 -08002515 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
2516 * need highmem pages, so cap pages_min to a small
2517 * value here.
2518 *
2519 * The (pages_high-pages_low) and (pages_low-pages_min)
2520 * deltas controls asynch page reclaim, and so should
2521 * not be capped for highmem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002522 */
2523 int min_pages;
2524
2525 min_pages = zone->present_pages / 1024;
2526 if (min_pages < SWAP_CLUSTER_MAX)
2527 min_pages = SWAP_CLUSTER_MAX;
2528 if (min_pages > 128)
2529 min_pages = 128;
2530 zone->pages_min = min_pages;
2531 } else {
Nick Piggin669ed172005-11-13 16:06:45 -08002532 /*
2533 * If it's a lowmem zone, reserve a number of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07002534 * proportionate to the zone's size.
2535 */
Nick Piggin669ed172005-11-13 16:06:45 -08002536 zone->pages_min = tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002537 }
2538
Nick Piggin669ed172005-11-13 16:06:45 -08002539 zone->pages_low = zone->pages_min + tmp / 4;
2540 zone->pages_high = zone->pages_min + tmp / 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002541 spin_unlock_irqrestore(&zone->lru_lock, flags);
2542 }
2543}
2544
2545/*
2546 * Initialise min_free_kbytes.
2547 *
2548 * For small machines we want it small (128k min). For large machines
2549 * we want it large (64MB max). But it is not linear, because network
2550 * bandwidth does not increase linearly with machine size. We use
2551 *
2552 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
2553 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
2554 *
2555 * which yields
2556 *
2557 * 16MB: 512k
2558 * 32MB: 724k
2559 * 64MB: 1024k
2560 * 128MB: 1448k
2561 * 256MB: 2048k
2562 * 512MB: 2896k
2563 * 1024MB: 4096k
2564 * 2048MB: 5792k
2565 * 4096MB: 8192k
2566 * 8192MB: 11584k
2567 * 16384MB: 16384k
2568 */
2569static int __init init_per_zone_pages_min(void)
2570{
2571 unsigned long lowmem_kbytes;
2572
2573 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
2574
2575 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
2576 if (min_free_kbytes < 128)
2577 min_free_kbytes = 128;
2578 if (min_free_kbytes > 65536)
2579 min_free_kbytes = 65536;
2580 setup_per_zone_pages_min();
2581 setup_per_zone_lowmem_reserve();
2582 return 0;
2583}
2584module_init(init_per_zone_pages_min)
2585
2586/*
2587 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
2588 * that we can call two helper functions whenever min_free_kbytes
2589 * changes.
2590 */
2591int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
2592 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2593{
2594 proc_dointvec(table, write, file, buffer, length, ppos);
2595 setup_per_zone_pages_min();
2596 return 0;
2597}
2598
2599/*
2600 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
2601 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
2602 * whenever sysctl_lowmem_reserve_ratio changes.
2603 *
2604 * The reserve ratio obviously has absolutely no relation with the
2605 * pages_min watermarks. The lowmem reserve ratio can only make sense
2606 * if in function of the boot time zone sizes.
2607 */
2608int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
2609 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2610{
2611 proc_dointvec_minmax(table, write, file, buffer, length, ppos);
2612 setup_per_zone_lowmem_reserve();
2613 return 0;
2614}
2615
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08002616/*
2617 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
2618 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
2619 * can have before it gets flushed back to buddy allocator.
2620 */
2621
2622int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
2623 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2624{
2625 struct zone *zone;
2626 unsigned int cpu;
2627 int ret;
2628
2629 ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
2630 if (!write || (ret == -EINVAL))
2631 return ret;
2632 for_each_zone(zone) {
2633 for_each_online_cpu(cpu) {
2634 unsigned long high;
2635 high = zone->present_pages / percpu_pagelist_fraction;
2636 setup_pagelist_highmark(zone_pcp(zone, cpu), high);
2637 }
2638 }
2639 return 0;
2640}
2641
Linus Torvalds1da177e2005-04-16 15:20:36 -07002642__initdata int hashdist = HASHDIST_DEFAULT;
2643
2644#ifdef CONFIG_NUMA
2645static int __init set_hashdist(char *str)
2646{
2647 if (!str)
2648 return 0;
2649 hashdist = simple_strtoul(str, &str, 0);
2650 return 1;
2651}
2652__setup("hashdist=", set_hashdist);
2653#endif
2654
2655/*
2656 * allocate a large system hash table from bootmem
2657 * - it is assumed that the hash table must contain an exact power-of-2
2658 * quantity of entries
2659 * - limit is the number of hash buckets, not the total allocation size
2660 */
2661void *__init alloc_large_system_hash(const char *tablename,
2662 unsigned long bucketsize,
2663 unsigned long numentries,
2664 int scale,
2665 int flags,
2666 unsigned int *_hash_shift,
2667 unsigned int *_hash_mask,
2668 unsigned long limit)
2669{
2670 unsigned long long max = limit;
2671 unsigned long log2qty, size;
2672 void *table = NULL;
2673
2674 /* allow the kernel cmdline to have a say */
2675 if (!numentries) {
2676 /* round applicable memory size up to nearest megabyte */
2677 numentries = (flags & HASH_HIGHMEM) ? nr_all_pages : nr_kernel_pages;
2678 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
2679 numentries >>= 20 - PAGE_SHIFT;
2680 numentries <<= 20 - PAGE_SHIFT;
2681
2682 /* limit to 1 bucket per 2^scale bytes of low memory */
2683 if (scale > PAGE_SHIFT)
2684 numentries >>= (scale - PAGE_SHIFT);
2685 else
2686 numentries <<= (PAGE_SHIFT - scale);
2687 }
2688 /* rounded up to nearest power of 2 in size */
2689 numentries = 1UL << (long_log2(numentries) + 1);
2690
2691 /* limit allocation size to 1/16 total memory by default */
2692 if (max == 0) {
2693 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
2694 do_div(max, bucketsize);
2695 }
2696
2697 if (numentries > max)
2698 numentries = max;
2699
2700 log2qty = long_log2(numentries);
2701
2702 do {
2703 size = bucketsize << log2qty;
2704 if (flags & HASH_EARLY)
2705 table = alloc_bootmem(size);
2706 else if (hashdist)
2707 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
2708 else {
2709 unsigned long order;
2710 for (order = 0; ((1UL << order) << PAGE_SHIFT) < size; order++)
2711 ;
2712 table = (void*) __get_free_pages(GFP_ATOMIC, order);
2713 }
2714 } while (!table && size > PAGE_SIZE && --log2qty);
2715
2716 if (!table)
2717 panic("Failed to allocate %s hash table\n", tablename);
2718
2719 printk("%s hash table entries: %d (order: %d, %lu bytes)\n",
2720 tablename,
2721 (1U << log2qty),
2722 long_log2(size) - PAGE_SHIFT,
2723 size);
2724
2725 if (_hash_shift)
2726 *_hash_shift = log2qty;
2727 if (_hash_mask)
2728 *_hash_mask = (1 << log2qty) - 1;
2729
2730 return table;
2731}