blob: 353ce9039a86f477ac13897bb76ab522c718ab87 [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
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/stddef.h>
18#include <linux/mm.h>
19#include <linux/swap.h>
20#include <linux/interrupt.h>
21#include <linux/pagemap.h>
22#include <linux/bootmem.h>
23#include <linux/compiler.h>
Randy Dunlap9f158332005-09-13 01:25:16 -070024#include <linux/kernel.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/module.h>
26#include <linux/suspend.h>
27#include <linux/pagevec.h>
28#include <linux/blkdev.h>
29#include <linux/slab.h>
30#include <linux/notifier.h>
31#include <linux/topology.h>
32#include <linux/sysctl.h>
33#include <linux/cpu.h>
34#include <linux/cpuset.h>
Dave Hansenbdc8cb92005-10-29 18:16:53 -070035#include <linux/memory_hotplug.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include <linux/nodemask.h>
37#include <linux/vmalloc.h>
Christoph Lameter4be38e32006-01-06 00:11:17 -080038#include <linux/mempolicy.h>
Yasunori Goto68113782006-06-23 02:03:11 -070039#include <linux/stop_machine.h>
Mel Gormanc7132162006-09-27 01:49:43 -070040#include <linux/sort.h>
41#include <linux/pfn.h>
Andrew Morton3fcfab12006-10-19 23:28:16 -070042#include <linux/backing-dev.h>
Akinobu Mita933e3122006-12-08 02:39:45 -080043#include <linux/fault-inject.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070044
45#include <asm/tlbflush.h>
Andrew Mortonac924c62006-05-15 09:43:59 -070046#include <asm/div64.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070047#include "internal.h"
48
49/*
50 * MCD - HACK: Find somewhere to initialize this EARLY, or make this
51 * initializer cleaner
52 */
Christoph Lameterc3d8c142005-09-06 15:16:33 -070053nodemask_t node_online_map __read_mostly = { { [0] = 1UL } };
Dean Nelson7223a932005-03-23 19:00:00 -070054EXPORT_SYMBOL(node_online_map);
Christoph Lameterc3d8c142005-09-06 15:16:33 -070055nodemask_t node_possible_map __read_mostly = NODE_MASK_ALL;
Dean Nelson7223a932005-03-23 19:00:00 -070056EXPORT_SYMBOL(node_possible_map);
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -070057unsigned long totalram_pages __read_mostly;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -070058unsigned long totalreserve_pages __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070059long nr_swap_pages;
Rohit Seth8ad4b1f2006-01-08 01:00:40 -080060int percpu_pagelist_fraction;
Linus Torvalds1da177e2005-04-16 15:20:36 -070061
Hugh Dickinsd98c7a02006-02-14 13:52:59 -080062static void __free_pages_ok(struct page *page, unsigned int order);
David Howellsa226f6c2006-01-06 00:11:08 -080063
Linus Torvalds1da177e2005-04-16 15:20:36 -070064/*
65 * results with 256, 32 in the lowmem_reserve sysctl:
66 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
67 * 1G machine -> (16M dma, 784M normal, 224M high)
68 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
69 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
70 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
Andi Kleena2f1b422005-11-05 17:25:53 +010071 *
72 * TBD: should special case ZONE_DMA32 machines here - in those we normally
73 * don't need any ZONE_NORMAL reservation
Linus Torvalds1da177e2005-04-16 15:20:36 -070074 */
Christoph Lameter2f1b6242006-09-25 23:31:13 -070075int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -080076#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -070077 256,
Christoph Lameter4b51d662007-02-10 01:43:10 -080078#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -070079#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -070080 256,
Christoph Lameterfb0e7942006-09-25 23:31:13 -070081#endif
Christoph Lametere53ef382006-09-25 23:31:14 -070082#ifdef CONFIG_HIGHMEM
Christoph Lameter2f1b6242006-09-25 23:31:13 -070083 32
Christoph Lametere53ef382006-09-25 23:31:14 -070084#endif
Christoph Lameter2f1b6242006-09-25 23:31:13 -070085};
Linus Torvalds1da177e2005-04-16 15:20:36 -070086
87EXPORT_SYMBOL(totalram_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -070088
Helge Deller15ad7cd2006-12-06 20:40:36 -080089static char * const zone_names[MAX_NR_ZONES] = {
Christoph Lameter4b51d662007-02-10 01:43:10 -080090#ifdef CONFIG_ZONE_DMA
Christoph Lameter2f1b6242006-09-25 23:31:13 -070091 "DMA",
Christoph Lameter4b51d662007-02-10 01:43:10 -080092#endif
Christoph Lameterfb0e7942006-09-25 23:31:13 -070093#ifdef CONFIG_ZONE_DMA32
Christoph Lameter2f1b6242006-09-25 23:31:13 -070094 "DMA32",
Christoph Lameterfb0e7942006-09-25 23:31:13 -070095#endif
Christoph Lameter2f1b6242006-09-25 23:31:13 -070096 "Normal",
Christoph Lametere53ef382006-09-25 23:31:14 -070097#ifdef CONFIG_HIGHMEM
Christoph Lameter2f1b6242006-09-25 23:31:13 -070098 "HighMem"
Christoph Lametere53ef382006-09-25 23:31:14 -070099#endif
Christoph Lameter2f1b6242006-09-25 23:31:13 -0700100};
101
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102int min_free_kbytes = 1024;
103
Yasunori Goto86356ab2006-06-23 02:03:09 -0700104unsigned long __meminitdata nr_kernel_pages;
105unsigned long __meminitdata nr_all_pages;
Mel Gorman0e0b8642006-09-27 01:49:56 -0700106static unsigned long __initdata dma_reserve;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107
Mel Gormanc7132162006-09-27 01:49:43 -0700108#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
109 /*
110 * MAX_ACTIVE_REGIONS determines the maxmimum number of distinct
111 * ranges of memory (RAM) that may be registered with add_active_range().
112 * Ranges passed to add_active_range() will be merged if possible
113 * so the number of times add_active_range() can be called is
114 * related to the number of nodes and the number of holes
115 */
116 #ifdef CONFIG_MAX_ACTIVE_REGIONS
117 /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
118 #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
119 #else
120 #if MAX_NUMNODES >= 32
121 /* If there can be many nodes, allow up to 50 holes per node */
122 #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
123 #else
124 /* By default, allow up to 256 distinct regions */
125 #define MAX_ACTIVE_REGIONS 256
126 #endif
127 #endif
128
129 struct node_active_region __initdata early_node_map[MAX_ACTIVE_REGIONS];
130 int __initdata nr_nodemap_entries;
131 unsigned long __initdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
132 unsigned long __initdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
Mel Gormanfb014392006-09-27 01:49:59 -0700133#ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
134 unsigned long __initdata node_boundary_start_pfn[MAX_NUMNODES];
135 unsigned long __initdata node_boundary_end_pfn[MAX_NUMNODES];
136#endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
Mel Gormanc7132162006-09-27 01:49:43 -0700137#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
138
Nick Piggin13e74442006-01-06 00:10:58 -0800139#ifdef CONFIG_DEBUG_VM
Dave Hansenc6a57e12005-10-29 18:16:52 -0700140static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141{
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700142 int ret = 0;
143 unsigned seq;
144 unsigned long pfn = page_to_pfn(page);
Dave Hansenc6a57e12005-10-29 18:16:52 -0700145
Dave Hansenbdc8cb92005-10-29 18:16:53 -0700146 do {
147 seq = zone_span_seqbegin(zone);
148 if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
149 ret = 1;
150 else if (pfn < zone->zone_start_pfn)
151 ret = 1;
152 } while (zone_span_seqretry(zone, seq));
153
154 return ret;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700155}
156
157static int page_is_consistent(struct zone *zone, struct page *page)
158{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159#ifdef CONFIG_HOLES_IN_ZONE
160 if (!pfn_valid(page_to_pfn(page)))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700161 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162#endif
163 if (zone != page_zone(page))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700164 return 0;
165
166 return 1;
167}
168/*
169 * Temporary debugging check for pages not lying within a given zone.
170 */
171static int bad_range(struct zone *zone, struct page *page)
172{
173 if (page_outside_zone_boundaries(zone, page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174 return 1;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700175 if (!page_is_consistent(zone, page))
176 return 1;
177
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178 return 0;
179}
Nick Piggin13e74442006-01-06 00:10:58 -0800180#else
181static inline int bad_range(struct zone *zone, struct page *page)
182{
183 return 0;
184}
185#endif
186
Nick Piggin224abf92006-01-06 00:11:11 -0800187static void bad_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188{
Nick Piggin224abf92006-01-06 00:11:11 -0800189 printk(KERN_EMERG "Bad page state in process '%s'\n"
Hugh Dickins7365f3d2006-01-11 12:17:18 -0800190 KERN_EMERG "page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n"
191 KERN_EMERG "Trying to fix it up, but a reboot is needed\n"
192 KERN_EMERG "Backtrace:\n",
Nick Piggin224abf92006-01-06 00:11:11 -0800193 current->comm, page, (int)(2*sizeof(unsigned long)),
194 (unsigned long)page->flags, page->mapping,
195 page_mapcount(page), page_count(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196 dump_stack();
Hugh Dickins334795e2005-06-21 17:15:08 -0700197 page->flags &= ~(1 << PG_lru |
198 1 << PG_private |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199 1 << PG_locked |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200 1 << PG_active |
201 1 << PG_dirty |
Hugh Dickins334795e2005-06-21 17:15:08 -0700202 1 << PG_reclaim |
203 1 << PG_slab |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204 1 << PG_swapcache |
Nick Piggin676165a2006-04-10 11:21:48 +1000205 1 << PG_writeback |
206 1 << PG_buddy );
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 set_page_count(page, 0);
208 reset_page_mapcount(page);
209 page->mapping = NULL;
Randy Dunlap9f158332005-09-13 01:25:16 -0700210 add_taint(TAINT_BAD_PAGE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211}
212
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213/*
214 * Higher-order pages are called "compound pages". They are structured thusly:
215 *
216 * The first PAGE_SIZE page is called the "head page".
217 *
218 * The remaining PAGE_SIZE pages are called "tail pages".
219 *
220 * All pages have PG_compound set. All pages have their ->private pointing at
221 * the head page (even the head page has this).
222 *
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800223 * The first tail page's ->lru.next holds the address of the compound page's
224 * put_page() function. Its ->lru.prev holds the order of allocation.
225 * This usage means that zero-order pages may not be compound.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226 */
Hugh Dickinsd98c7a02006-02-14 13:52:59 -0800227
228static void free_compound_page(struct page *page)
229{
230 __free_pages_ok(page, (unsigned long)page[1].lru.prev);
231}
232
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233static void prep_compound_page(struct page *page, unsigned long order)
234{
235 int i;
236 int nr_pages = 1 << order;
237
Andy Whitcroft33f2ef82006-12-06 20:33:32 -0800238 set_compound_page_dtor(page, free_compound_page);
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800239 page[1].lru.prev = (void *)order;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240 for (i = 0; i < nr_pages; i++) {
241 struct page *p = page + i;
242
Nick Piggin5e9dace2006-03-22 00:08:01 -0800243 __SetPageCompound(p);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700244 set_page_private(p, (unsigned long)page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245 }
246}
247
248static void destroy_compound_page(struct page *page, unsigned long order)
249{
250 int i;
251 int nr_pages = 1 << order;
252
Hugh Dickins41d78ba2006-02-14 13:52:58 -0800253 if (unlikely((unsigned long)page[1].lru.prev != order))
Nick Piggin224abf92006-01-06 00:11:11 -0800254 bad_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255
256 for (i = 0; i < nr_pages; i++) {
257 struct page *p = page + i;
258
Nick Piggin224abf92006-01-06 00:11:11 -0800259 if (unlikely(!PageCompound(p) |
260 (page_private(p) != (unsigned long)page)))
261 bad_page(page);
Nick Piggin5e9dace2006-03-22 00:08:01 -0800262 __ClearPageCompound(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 }
264}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265
Nick Piggin17cf4402006-03-22 00:08:41 -0800266static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
267{
268 int i;
269
Nick Piggin725d7042006-09-25 23:30:55 -0700270 VM_BUG_ON((gfp_flags & (__GFP_WAIT | __GFP_HIGHMEM)) == __GFP_HIGHMEM);
Andrew Morton6626c5d2006-03-22 00:08:42 -0800271 /*
272 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
273 * and __GFP_HIGHMEM from hard or soft interrupt context.
274 */
Nick Piggin725d7042006-09-25 23:30:55 -0700275 VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
Nick Piggin17cf4402006-03-22 00:08:41 -0800276 for (i = 0; i < (1 << order); i++)
277 clear_highpage(page + i);
278}
279
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280/*
281 * function for dealing with page's order in buddy system.
282 * zone->lock is already acquired when we use these.
283 * So, we don't need atomic page->flags operations here.
284 */
Andrew Morton6aa30012006-04-18 22:20:52 -0700285static inline unsigned long page_order(struct page *page)
286{
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700287 return page_private(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288}
289
Andrew Morton6aa30012006-04-18 22:20:52 -0700290static inline void set_page_order(struct page *page, int order)
291{
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700292 set_page_private(page, order);
Nick Piggin676165a2006-04-10 11:21:48 +1000293 __SetPageBuddy(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294}
295
296static inline void rmv_page_order(struct page *page)
297{
Nick Piggin676165a2006-04-10 11:21:48 +1000298 __ClearPageBuddy(page);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700299 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300}
301
302/*
303 * Locate the struct page for both the matching buddy in our
304 * pair (buddy1) and the combined O(n+1) page they form (page).
305 *
306 * 1) Any buddy B1 will have an order O twin B2 which satisfies
307 * the following equation:
308 * B2 = B1 ^ (1 << O)
309 * For example, if the starting buddy (buddy2) is #8 its order
310 * 1 buddy is #10:
311 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
312 *
313 * 2) Any buddy B will have an order O+1 parent P which
314 * satisfies the following equation:
315 * P = B & ~(1 << O)
316 *
Andreas Mohrd6e05ed2006-06-26 18:35:02 +0200317 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318 */
319static inline struct page *
320__page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order)
321{
322 unsigned long buddy_idx = page_idx ^ (1 << order);
323
324 return page + (buddy_idx - page_idx);
325}
326
327static inline unsigned long
328__find_combined_index(unsigned long page_idx, unsigned int order)
329{
330 return (page_idx & ~(1 << order));
331}
332
333/*
334 * This function checks whether a page is free && is the buddy
335 * we can do coalesce a page and its buddy if
Nick Piggin13e74442006-01-06 00:10:58 -0800336 * (a) the buddy is not in a hole &&
Nick Piggin676165a2006-04-10 11:21:48 +1000337 * (b) the buddy is in the buddy system &&
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700338 * (c) a page and its buddy have the same order &&
339 * (d) a page and its buddy are in the same zone.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340 *
Nick Piggin676165a2006-04-10 11:21:48 +1000341 * For recording whether a page is in the buddy system, we use PG_buddy.
342 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
343 *
344 * For recording page's order, we use page_private(page).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345 */
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700346static inline int page_is_buddy(struct page *page, struct page *buddy,
347 int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348{
Nick Piggin13e74442006-01-06 00:10:58 -0800349#ifdef CONFIG_HOLES_IN_ZONE
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700350 if (!pfn_valid(page_to_pfn(buddy)))
Nick Piggin13e74442006-01-06 00:10:58 -0800351 return 0;
352#endif
353
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700354 if (page_zone_id(page) != page_zone_id(buddy))
355 return 0;
356
357 if (PageBuddy(buddy) && page_order(buddy) == order) {
358 BUG_ON(page_count(buddy) != 0);
Andrew Morton6aa30012006-04-18 22:20:52 -0700359 return 1;
Nick Piggin676165a2006-04-10 11:21:48 +1000360 }
Andrew Morton6aa30012006-04-18 22:20:52 -0700361 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362}
363
364/*
365 * Freeing function for a buddy system allocator.
366 *
367 * The concept of a buddy system is to maintain direct-mapped table
368 * (containing bit values) for memory blocks of various "orders".
369 * The bottom level table contains the map for the smallest allocatable
370 * units of memory (here, pages), and each level above it describes
371 * pairs of units from the levels below, hence, "buddies".
372 * At a high level, all that happens here is marking the table entry
373 * at the bottom level available, and propagating the changes upward
374 * as necessary, plus some accounting needed to play nicely with other
375 * parts of the VM system.
376 * At each level, we keep a list of pages, which are heads of continuous
Nick Piggin676165a2006-04-10 11:21:48 +1000377 * free pages of length of (1 << order) and marked with PG_buddy. Page's
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700378 * order is recorded in page_private(page) field.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379 * So when we are allocating or freeing one, we can derive the state of the
380 * other. That is, if we allocate a small block, and both were
381 * free, the remainder of the region must be split into blocks.
382 * If a block is freed, and its buddy is also free, then this
383 * triggers coalescing into a block of larger size.
384 *
385 * -- wli
386 */
387
Nick Piggin48db57f2006-01-08 01:00:42 -0800388static inline void __free_one_page(struct page *page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 struct zone *zone, unsigned int order)
390{
391 unsigned long page_idx;
392 int order_size = 1 << order;
393
Nick Piggin224abf92006-01-06 00:11:11 -0800394 if (unlikely(PageCompound(page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395 destroy_compound_page(page, order);
396
397 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
398
Nick Piggin725d7042006-09-25 23:30:55 -0700399 VM_BUG_ON(page_idx & (order_size - 1));
400 VM_BUG_ON(bad_range(zone, page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401
Christoph Lameterd23ad422007-02-10 01:43:02 -0800402 __mod_zone_page_state(zone, NR_FREE_PAGES, order_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403 while (order < MAX_ORDER-1) {
404 unsigned long combined_idx;
405 struct free_area *area;
406 struct page *buddy;
407
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 buddy = __page_find_buddy(page, page_idx, order);
Andy Whitcroftcb2b95e2006-06-23 02:03:01 -0700409 if (!page_is_buddy(page, buddy, order))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410 break; /* Move the buddy up one level. */
Nick Piggin13e74442006-01-06 00:10:58 -0800411
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412 list_del(&buddy->lru);
413 area = zone->free_area + order;
414 area->nr_free--;
415 rmv_page_order(buddy);
Nick Piggin13e74442006-01-06 00:10:58 -0800416 combined_idx = __find_combined_index(page_idx, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417 page = page + (combined_idx - page_idx);
418 page_idx = combined_idx;
419 order++;
420 }
421 set_page_order(page, order);
422 list_add(&page->lru, &zone->free_area[order].free_list);
423 zone->free_area[order].nr_free++;
424}
425
Nick Piggin224abf92006-01-06 00:11:11 -0800426static inline int free_pages_check(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427{
Nick Piggin92be2e32006-01-06 00:10:57 -0800428 if (unlikely(page_mapcount(page) |
429 (page->mapping != NULL) |
430 (page_count(page) != 0) |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431 (page->flags & (
432 1 << PG_lru |
433 1 << PG_private |
434 1 << PG_locked |
435 1 << PG_active |
436 1 << PG_reclaim |
437 1 << PG_slab |
438 1 << PG_swapcache |
Nick Pigginb5810032005-10-29 18:16:12 -0700439 1 << PG_writeback |
Nick Piggin676165a2006-04-10 11:21:48 +1000440 1 << PG_reserved |
441 1 << PG_buddy ))))
Nick Piggin224abf92006-01-06 00:11:11 -0800442 bad_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443 if (PageDirty(page))
Nick Piggin242e5462005-09-03 15:54:50 -0700444 __ClearPageDirty(page);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800445 /*
446 * For now, we report if PG_reserved was found set, but do not
447 * clear it, and do not free the page. But we shall soon need
448 * to do more, for when the ZERO_PAGE count wraps negative.
449 */
450 return PageReserved(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451}
452
453/*
454 * Frees a list of pages.
455 * Assumes all pages on list are in same zone, and of same order.
Renaud Lienhart207f36e2005-09-10 00:26:59 -0700456 * count is the number of pages to free.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457 *
458 * If the zone was previously in an "all pages pinned" state then look to
459 * see if this freeing clears that state.
460 *
461 * And clear the zone's pages_scanned counter, to hold off the "all pages are
462 * pinned" detection logic.
463 */
Nick Piggin48db57f2006-01-08 01:00:42 -0800464static void free_pages_bulk(struct zone *zone, int count,
465 struct list_head *list, int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466{
Nick Pigginc54ad302006-01-06 00:10:56 -0800467 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 zone->all_unreclaimable = 0;
469 zone->pages_scanned = 0;
Nick Piggin48db57f2006-01-08 01:00:42 -0800470 while (count--) {
471 struct page *page;
472
Nick Piggin725d7042006-09-25 23:30:55 -0700473 VM_BUG_ON(list_empty(list));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474 page = list_entry(list->prev, struct page, lru);
Nick Piggin48db57f2006-01-08 01:00:42 -0800475 /* have to delete it as __free_one_page list manipulates */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476 list_del(&page->lru);
Nick Piggin48db57f2006-01-08 01:00:42 -0800477 __free_one_page(page, zone, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 }
Nick Pigginc54ad302006-01-06 00:10:56 -0800479 spin_unlock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480}
481
Nick Piggin48db57f2006-01-08 01:00:42 -0800482static void free_one_page(struct zone *zone, struct page *page, int order)
483{
Christoph Lameter006d22d2006-09-25 23:31:48 -0700484 spin_lock(&zone->lock);
485 zone->all_unreclaimable = 0;
486 zone->pages_scanned = 0;
Paul Jackson0798e512006-12-06 20:31:38 -0800487 __free_one_page(page, zone, order);
Christoph Lameter006d22d2006-09-25 23:31:48 -0700488 spin_unlock(&zone->lock);
Nick Piggin48db57f2006-01-08 01:00:42 -0800489}
490
491static void __free_pages_ok(struct page *page, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492{
Nick Pigginc54ad302006-01-06 00:10:56 -0800493 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494 int i;
Hugh Dickins689bceb2005-11-21 21:32:20 -0800495 int reserved = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497 for (i = 0 ; i < (1 << order) ; ++i)
Nick Piggin224abf92006-01-06 00:11:11 -0800498 reserved += free_pages_check(page + i);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800499 if (reserved)
500 return;
501
Nick Piggin9858db52006-10-11 01:21:30 -0700502 if (!PageHighMem(page))
503 debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order);
Nick Piggindafb1362006-10-11 01:21:30 -0700504 arch_free_page(page, order);
Nick Piggin48db57f2006-01-08 01:00:42 -0800505 kernel_map_pages(page, 1 << order, 0);
Nick Piggindafb1362006-10-11 01:21:30 -0700506
Nick Pigginc54ad302006-01-06 00:10:56 -0800507 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700508 __count_vm_events(PGFREE, 1 << order);
Nick Piggin48db57f2006-01-08 01:00:42 -0800509 free_one_page(page_zone(page), page, order);
Nick Pigginc54ad302006-01-06 00:10:56 -0800510 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511}
512
David Howellsa226f6c2006-01-06 00:11:08 -0800513/*
514 * permit the bootmem allocator to evade page validation on high-order frees
515 */
516void fastcall __init __free_pages_bootmem(struct page *page, unsigned int order)
517{
518 if (order == 0) {
519 __ClearPageReserved(page);
520 set_page_count(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800521 set_page_refcounted(page);
Nick Piggin545b1ea2006-03-22 00:08:07 -0800522 __free_page(page);
David Howellsa226f6c2006-01-06 00:11:08 -0800523 } else {
David Howellsa226f6c2006-01-06 00:11:08 -0800524 int loop;
525
Nick Piggin545b1ea2006-03-22 00:08:07 -0800526 prefetchw(page);
David Howellsa226f6c2006-01-06 00:11:08 -0800527 for (loop = 0; loop < BITS_PER_LONG; loop++) {
528 struct page *p = &page[loop];
529
Nick Piggin545b1ea2006-03-22 00:08:07 -0800530 if (loop + 1 < BITS_PER_LONG)
531 prefetchw(p + 1);
David Howellsa226f6c2006-01-06 00:11:08 -0800532 __ClearPageReserved(p);
533 set_page_count(p, 0);
534 }
535
Nick Piggin7835e982006-03-22 00:08:40 -0800536 set_page_refcounted(page);
Nick Piggin545b1ea2006-03-22 00:08:07 -0800537 __free_pages(page, order);
David Howellsa226f6c2006-01-06 00:11:08 -0800538 }
539}
540
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541
542/*
543 * The order of subdivision here is critical for the IO subsystem.
544 * Please do not alter this order without good reasons and regression
545 * testing. Specifically, as large blocks of memory are subdivided,
546 * the order in which smaller blocks are delivered depends on the order
547 * they're subdivided in this function. This is the primary factor
548 * influencing the order in which pages are delivered to the IO
549 * subsystem according to empirical testing, and this is also justified
550 * by considering the behavior of a buddy system containing a single
551 * large block of memory acted on by a series of small allocations.
552 * This behavior is a critical factor in sglist merging's success.
553 *
554 * -- wli
555 */
Nick Piggin085cc7d2006-01-06 00:11:01 -0800556static inline void expand(struct zone *zone, struct page *page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557 int low, int high, struct free_area *area)
558{
559 unsigned long size = 1 << high;
560
561 while (high > low) {
562 area--;
563 high--;
564 size >>= 1;
Nick Piggin725d7042006-09-25 23:30:55 -0700565 VM_BUG_ON(bad_range(zone, &page[size]));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 list_add(&page[size].lru, &area->free_list);
567 area->nr_free++;
568 set_page_order(&page[size], high);
569 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570}
571
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572/*
573 * This page is about to be returned from the page allocator
574 */
Nick Piggin17cf4402006-03-22 00:08:41 -0800575static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576{
Nick Piggin92be2e32006-01-06 00:10:57 -0800577 if (unlikely(page_mapcount(page) |
578 (page->mapping != NULL) |
579 (page_count(page) != 0) |
Hugh Dickins334795e2005-06-21 17:15:08 -0700580 (page->flags & (
581 1 << PG_lru |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 1 << PG_private |
583 1 << PG_locked |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 1 << PG_active |
585 1 << PG_dirty |
586 1 << PG_reclaim |
Hugh Dickins334795e2005-06-21 17:15:08 -0700587 1 << PG_slab |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 1 << PG_swapcache |
Nick Pigginb5810032005-10-29 18:16:12 -0700589 1 << PG_writeback |
Nick Piggin676165a2006-04-10 11:21:48 +1000590 1 << PG_reserved |
591 1 << PG_buddy ))))
Nick Piggin224abf92006-01-06 00:11:11 -0800592 bad_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593
Hugh Dickins689bceb2005-11-21 21:32:20 -0800594 /*
595 * For now, we report if PG_reserved was found set, but do not
596 * clear it, and do not allocate the page: as a safety net.
597 */
598 if (PageReserved(page))
599 return 1;
600
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 page->flags &= ~(1 << PG_uptodate | 1 << PG_error |
602 1 << PG_referenced | 1 << PG_arch_1 |
Nick Piggin5409bae2007-02-28 20:12:27 -0800603 1 << PG_owner_priv_1 | 1 << PG_mappedtodisk);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700604 set_page_private(page, 0);
Nick Piggin7835e982006-03-22 00:08:40 -0800605 set_page_refcounted(page);
Nick Piggincc102502006-12-06 20:32:00 -0800606
607 arch_alloc_page(page, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 kernel_map_pages(page, 1 << order, 1);
Nick Piggin17cf4402006-03-22 00:08:41 -0800609
610 if (gfp_flags & __GFP_ZERO)
611 prep_zero_page(page, order, gfp_flags);
612
613 if (order && (gfp_flags & __GFP_COMP))
614 prep_compound_page(page, order);
615
Hugh Dickins689bceb2005-11-21 21:32:20 -0800616 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617}
618
619/*
620 * Do the hard work of removing an element from the buddy allocator.
621 * Call me with the zone->lock already held.
622 */
623static struct page *__rmqueue(struct zone *zone, unsigned int order)
624{
625 struct free_area * area;
626 unsigned int current_order;
627 struct page *page;
628
629 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
630 area = zone->free_area + current_order;
631 if (list_empty(&area->free_list))
632 continue;
633
634 page = list_entry(area->free_list.next, struct page, lru);
635 list_del(&page->lru);
636 rmv_page_order(page);
637 area->nr_free--;
Christoph Lameterd23ad422007-02-10 01:43:02 -0800638 __mod_zone_page_state(zone, NR_FREE_PAGES, - (1UL << order));
Nick Piggin085cc7d2006-01-06 00:11:01 -0800639 expand(zone, page, order, current_order, area);
640 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 }
642
643 return NULL;
644}
645
646/*
647 * Obtain a specified number of elements from the buddy allocator, all under
648 * a single hold of the lock, for efficiency. Add them to the supplied list.
649 * Returns the number of new pages which were placed at *list.
650 */
651static int rmqueue_bulk(struct zone *zone, unsigned int order,
652 unsigned long count, struct list_head *list)
653{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655
Nick Pigginc54ad302006-01-06 00:10:56 -0800656 spin_lock(&zone->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 for (i = 0; i < count; ++i) {
Nick Piggin085cc7d2006-01-06 00:11:01 -0800658 struct page *page = __rmqueue(zone, order);
659 if (unlikely(page == NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661 list_add_tail(&page->lru, list);
662 }
Nick Pigginc54ad302006-01-06 00:10:56 -0800663 spin_unlock(&zone->lock);
Nick Piggin085cc7d2006-01-06 00:11:01 -0800664 return i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665}
666
Christoph Lameter74c7aa82007-02-20 13:57:51 -0800667#if MAX_NUMNODES > 1
668int nr_node_ids __read_mostly;
669EXPORT_SYMBOL(nr_node_ids);
670
671/*
672 * Figure out the number of possible node ids.
673 */
674static void __init setup_nr_node_ids(void)
675{
676 unsigned int node;
677 unsigned int highest = 0;
678
679 for_each_node_mask(node, node_possible_map)
680 highest = node;
681 nr_node_ids = highest + 1;
682}
683#else
684static void __init setup_nr_node_ids(void) {}
685#endif
686
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700687#ifdef CONFIG_NUMA
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800688/*
689 * Called from the slab reaper to drain pagesets on a particular node that
Christoph Lameter39bbcb82006-09-25 23:31:49 -0700690 * belongs to the currently executing processor.
Christoph Lameter879336c2006-03-22 00:09:08 -0800691 * Note that this function must be called with the thread pinned to
692 * a single processor.
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800693 */
694void drain_node_pages(int nodeid)
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700695{
Christoph Lameter2f6726e2006-09-25 23:31:18 -0700696 int i;
697 enum zone_type z;
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700698 unsigned long flags;
699
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800700 for (z = 0; z < MAX_NR_ZONES; z++) {
701 struct zone *zone = NODE_DATA(nodeid)->node_zones + z;
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700702 struct per_cpu_pageset *pset;
703
Christoph Lameter39bbcb82006-09-25 23:31:49 -0700704 if (!populated_zone(zone))
705 continue;
706
Nick Piggin23316bc2006-01-08 01:00:41 -0800707 pset = zone_pcp(zone, smp_processor_id());
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700708 for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) {
709 struct per_cpu_pages *pcp;
710
711 pcp = &pset->pcp[i];
Christoph Lameter879336c2006-03-22 00:09:08 -0800712 if (pcp->count) {
Christoph Lameterbc4ba392006-12-06 20:33:02 -0800713 int to_drain;
714
Christoph Lameter879336c2006-03-22 00:09:08 -0800715 local_irq_save(flags);
Christoph Lameterbc4ba392006-12-06 20:33:02 -0800716 if (pcp->count >= pcp->batch)
717 to_drain = pcp->batch;
718 else
719 to_drain = pcp->count;
720 free_pages_bulk(zone, to_drain, &pcp->list, 0);
721 pcp->count -= to_drain;
Christoph Lameter879336c2006-03-22 00:09:08 -0800722 local_irq_restore(flags);
723 }
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700724 }
725 }
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700726}
727#endif
728
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729static void __drain_pages(unsigned int cpu)
730{
Nick Pigginc54ad302006-01-06 00:10:56 -0800731 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732 struct zone *zone;
733 int i;
734
735 for_each_zone(zone) {
736 struct per_cpu_pageset *pset;
737
Christoph Lameterf2e12bb2007-01-05 16:37:02 -0800738 if (!populated_zone(zone))
739 continue;
740
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700741 pset = zone_pcp(zone, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742 for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) {
743 struct per_cpu_pages *pcp;
744
745 pcp = &pset->pcp[i];
Nick Pigginc54ad302006-01-06 00:10:56 -0800746 local_irq_save(flags);
Nick Piggin48db57f2006-01-08 01:00:42 -0800747 free_pages_bulk(zone, pcp->count, &pcp->list, 0);
748 pcp->count = 0;
Nick Pigginc54ad302006-01-06 00:10:56 -0800749 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 }
751 }
752}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753
754#ifdef CONFIG_PM
755
756void mark_free_pages(struct zone *zone)
757{
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700758 unsigned long pfn, max_zone_pfn;
759 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760 int order;
761 struct list_head *curr;
762
763 if (!zone->spanned_pages)
764 return;
765
766 spin_lock_irqsave(&zone->lock, flags);
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700767
768 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
769 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
770 if (pfn_valid(pfn)) {
771 struct page *page = pfn_to_page(pfn);
772
773 if (!PageNosave(page))
774 ClearPageNosaveFree(page);
775 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776
777 for (order = MAX_ORDER - 1; order >= 0; --order)
778 list_for_each(curr, &zone->free_area[order].free_list) {
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700779 unsigned long i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780
Rafael J. Wysockif623f0d2006-09-25 23:32:49 -0700781 pfn = page_to_pfn(list_entry(curr, struct page, lru));
782 for (i = 0; i < (1UL << order); i++)
783 SetPageNosaveFree(pfn_to_page(pfn + i));
784 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 spin_unlock_irqrestore(&zone->lock, flags);
787}
788
789/*
790 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
791 */
792void drain_local_pages(void)
793{
794 unsigned long flags;
795
796 local_irq_save(flags);
797 __drain_pages(smp_processor_id());
798 local_irq_restore(flags);
799}
800#endif /* CONFIG_PM */
801
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802/*
803 * Free a 0-order page
804 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805static void fastcall free_hot_cold_page(struct page *page, int cold)
806{
807 struct zone *zone = page_zone(page);
808 struct per_cpu_pages *pcp;
809 unsigned long flags;
810
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811 if (PageAnon(page))
812 page->mapping = NULL;
Nick Piggin224abf92006-01-06 00:11:11 -0800813 if (free_pages_check(page))
Hugh Dickins689bceb2005-11-21 21:32:20 -0800814 return;
815
Nick Piggin9858db52006-10-11 01:21:30 -0700816 if (!PageHighMem(page))
817 debug_check_no_locks_freed(page_address(page), PAGE_SIZE);
Nick Piggindafb1362006-10-11 01:21:30 -0700818 arch_free_page(page, 0);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800819 kernel_map_pages(page, 1, 0);
820
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700821 pcp = &zone_pcp(zone, get_cpu())->pcp[cold];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822 local_irq_save(flags);
Christoph Lameterf8891e52006-06-30 01:55:45 -0700823 __count_vm_event(PGFREE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824 list_add(&page->lru, &pcp->list);
825 pcp->count++;
Nick Piggin48db57f2006-01-08 01:00:42 -0800826 if (pcp->count >= pcp->high) {
827 free_pages_bulk(zone, pcp->batch, &pcp->list, 0);
828 pcp->count -= pcp->batch;
829 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830 local_irq_restore(flags);
831 put_cpu();
832}
833
834void fastcall free_hot_page(struct page *page)
835{
836 free_hot_cold_page(page, 0);
837}
838
839void fastcall free_cold_page(struct page *page)
840{
841 free_hot_cold_page(page, 1);
842}
843
Nick Piggin8dfcc9b2006-03-22 00:08:05 -0800844/*
845 * split_page takes a non-compound higher-order page, and splits it into
846 * n (1<<order) sub-pages: page[0..n]
847 * Each sub-page must be freed individually.
848 *
849 * Note: this is probably too low level an operation for use in drivers.
850 * Please consult with lkml before using this in your driver.
851 */
852void split_page(struct page *page, unsigned int order)
853{
854 int i;
855
Nick Piggin725d7042006-09-25 23:30:55 -0700856 VM_BUG_ON(PageCompound(page));
857 VM_BUG_ON(!page_count(page));
Nick Piggin7835e982006-03-22 00:08:40 -0800858 for (i = 1; i < (1 << order); i++)
859 set_page_refcounted(page + i);
Nick Piggin8dfcc9b2006-03-22 00:08:05 -0800860}
Nick Piggin8dfcc9b2006-03-22 00:08:05 -0800861
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862/*
863 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
864 * we cheat by calling it from here, in the order > 0 path. Saves a branch
865 * or two.
866 */
Nick Piggina74609f2006-01-06 00:11:20 -0800867static struct page *buffered_rmqueue(struct zonelist *zonelist,
868 struct zone *zone, int order, gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869{
870 unsigned long flags;
Hugh Dickins689bceb2005-11-21 21:32:20 -0800871 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872 int cold = !!(gfp_flags & __GFP_COLD);
Nick Piggina74609f2006-01-06 00:11:20 -0800873 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874
Hugh Dickins689bceb2005-11-21 21:32:20 -0800875again:
Nick Piggina74609f2006-01-06 00:11:20 -0800876 cpu = get_cpu();
Nick Piggin48db57f2006-01-08 01:00:42 -0800877 if (likely(order == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878 struct per_cpu_pages *pcp;
879
Nick Piggina74609f2006-01-06 00:11:20 -0800880 pcp = &zone_pcp(zone, cpu)->pcp[cold];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881 local_irq_save(flags);
Nick Piggina74609f2006-01-06 00:11:20 -0800882 if (!pcp->count) {
nkalmala941c7102006-11-02 22:07:04 -0800883 pcp->count = rmqueue_bulk(zone, 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884 pcp->batch, &pcp->list);
Nick Piggina74609f2006-01-06 00:11:20 -0800885 if (unlikely(!pcp->count))
886 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887 }
Nick Piggina74609f2006-01-06 00:11:20 -0800888 page = list_entry(pcp->list.next, struct page, lru);
889 list_del(&page->lru);
890 pcp->count--;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800891 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892 spin_lock_irqsave(&zone->lock, flags);
893 page = __rmqueue(zone, order);
Nick Piggina74609f2006-01-06 00:11:20 -0800894 spin_unlock(&zone->lock);
895 if (!page)
896 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897 }
898
Christoph Lameterf8891e52006-06-30 01:55:45 -0700899 __count_zone_vm_events(PGALLOC, zone, 1 << order);
Christoph Lameterca889e62006-06-30 01:55:44 -0700900 zone_statistics(zonelist, zone);
Nick Piggina74609f2006-01-06 00:11:20 -0800901 local_irq_restore(flags);
902 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903
Nick Piggin725d7042006-09-25 23:30:55 -0700904 VM_BUG_ON(bad_range(zone, page));
Nick Piggin17cf4402006-03-22 00:08:41 -0800905 if (prep_new_page(page, order, gfp_flags))
Nick Piggina74609f2006-01-06 00:11:20 -0800906 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907 return page;
Nick Piggina74609f2006-01-06 00:11:20 -0800908
909failed:
910 local_irq_restore(flags);
911 put_cpu();
912 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913}
914
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800915#define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
Nick Piggin31488902005-11-28 13:44:03 -0800916#define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
917#define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
918#define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
919#define ALLOC_HARDER 0x10 /* try to alloc harder */
920#define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
921#define ALLOC_CPUSET 0x40 /* check for correct cpuset */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800922
Akinobu Mita933e3122006-12-08 02:39:45 -0800923#ifdef CONFIG_FAIL_PAGE_ALLOC
924
925static struct fail_page_alloc_attr {
926 struct fault_attr attr;
927
928 u32 ignore_gfp_highmem;
929 u32 ignore_gfp_wait;
930
931#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
932
933 struct dentry *ignore_gfp_highmem_file;
934 struct dentry *ignore_gfp_wait_file;
935
936#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
937
938} fail_page_alloc = {
939 .attr = FAULT_ATTR_INITIALIZER,
Don Mullis6b1b60f2006-12-08 02:39:53 -0800940 .ignore_gfp_wait = 1,
941 .ignore_gfp_highmem = 1,
Akinobu Mita933e3122006-12-08 02:39:45 -0800942};
943
944static int __init setup_fail_page_alloc(char *str)
945{
946 return setup_fault_attr(&fail_page_alloc.attr, str);
947}
948__setup("fail_page_alloc=", setup_fail_page_alloc);
949
950static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
951{
952 if (gfp_mask & __GFP_NOFAIL)
953 return 0;
954 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
955 return 0;
956 if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
957 return 0;
958
959 return should_fail(&fail_page_alloc.attr, 1 << order);
960}
961
962#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
963
964static int __init fail_page_alloc_debugfs(void)
965{
966 mode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
967 struct dentry *dir;
968 int err;
969
970 err = init_fault_attr_dentries(&fail_page_alloc.attr,
971 "fail_page_alloc");
972 if (err)
973 return err;
974 dir = fail_page_alloc.attr.dentries.dir;
975
976 fail_page_alloc.ignore_gfp_wait_file =
977 debugfs_create_bool("ignore-gfp-wait", mode, dir,
978 &fail_page_alloc.ignore_gfp_wait);
979
980 fail_page_alloc.ignore_gfp_highmem_file =
981 debugfs_create_bool("ignore-gfp-highmem", mode, dir,
982 &fail_page_alloc.ignore_gfp_highmem);
983
984 if (!fail_page_alloc.ignore_gfp_wait_file ||
985 !fail_page_alloc.ignore_gfp_highmem_file) {
986 err = -ENOMEM;
987 debugfs_remove(fail_page_alloc.ignore_gfp_wait_file);
988 debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file);
989 cleanup_fault_attr_dentries(&fail_page_alloc.attr);
990 }
991
992 return err;
993}
994
995late_initcall(fail_page_alloc_debugfs);
996
997#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
998
999#else /* CONFIG_FAIL_PAGE_ALLOC */
1000
1001static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1002{
1003 return 0;
1004}
1005
1006#endif /* CONFIG_FAIL_PAGE_ALLOC */
1007
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008/*
1009 * Return 1 if free pages are above 'mark'. This takes into account the order
1010 * of the allocation.
1011 */
1012int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001013 int classzone_idx, int alloc_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014{
1015 /* free_pages my go negative - that's OK */
Christoph Lameterd23ad422007-02-10 01:43:02 -08001016 long min = mark;
1017 long free_pages = zone_page_state(z, NR_FREE_PAGES) - (1 << order) + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 int o;
1019
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001020 if (alloc_flags & ALLOC_HIGH)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021 min -= min / 2;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001022 if (alloc_flags & ALLOC_HARDER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023 min -= min / 4;
1024
1025 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
1026 return 0;
1027 for (o = 0; o < order; o++) {
1028 /* At the next order, this order's pages become unavailable */
1029 free_pages -= z->free_area[o].nr_free << o;
1030
1031 /* Require fewer higher order pages to be free */
1032 min >>= 1;
1033
1034 if (free_pages <= min)
1035 return 0;
1036 }
1037 return 1;
1038}
1039
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001040#ifdef CONFIG_NUMA
1041/*
1042 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1043 * skip over zones that are not allowed by the cpuset, or that have
1044 * been recently (in last second) found to be nearly full. See further
1045 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1046 * that have to skip over alot of full or unallowed zones.
1047 *
1048 * If the zonelist cache is present in the passed in zonelist, then
1049 * returns a pointer to the allowed node mask (either the current
1050 * tasks mems_allowed, or node_online_map.)
1051 *
1052 * If the zonelist cache is not available for this zonelist, does
1053 * nothing and returns NULL.
1054 *
1055 * If the fullzones BITMAP in the zonelist cache is stale (more than
1056 * a second since last zap'd) then we zap it out (clear its bits.)
1057 *
1058 * We hold off even calling zlc_setup, until after we've checked the
1059 * first zone in the zonelist, on the theory that most allocations will
1060 * be satisfied from that first zone, so best to examine that zone as
1061 * quickly as we can.
1062 */
1063static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1064{
1065 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1066 nodemask_t *allowednodes; /* zonelist_cache approximation */
1067
1068 zlc = zonelist->zlcache_ptr;
1069 if (!zlc)
1070 return NULL;
1071
1072 if (jiffies - zlc->last_full_zap > 1 * HZ) {
1073 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1074 zlc->last_full_zap = jiffies;
1075 }
1076
1077 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1078 &cpuset_current_mems_allowed :
1079 &node_online_map;
1080 return allowednodes;
1081}
1082
1083/*
1084 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1085 * if it is worth looking at further for free memory:
1086 * 1) Check that the zone isn't thought to be full (doesn't have its
1087 * bit set in the zonelist_cache fullzones BITMAP).
1088 * 2) Check that the zones node (obtained from the zonelist_cache
1089 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1090 * Return true (non-zero) if zone is worth looking at further, or
1091 * else return false (zero) if it is not.
1092 *
1093 * This check -ignores- the distinction between various watermarks,
1094 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1095 * found to be full for any variation of these watermarks, it will
1096 * be considered full for up to one second by all requests, unless
1097 * we are so low on memory on all allowed nodes that we are forced
1098 * into the second scan of the zonelist.
1099 *
1100 * In the second scan we ignore this zonelist cache and exactly
1101 * apply the watermarks to all zones, even it is slower to do so.
1102 * We are low on memory in the second scan, and should leave no stone
1103 * unturned looking for a free page.
1104 */
1105static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zone **z,
1106 nodemask_t *allowednodes)
1107{
1108 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1109 int i; /* index of *z in zonelist zones */
1110 int n; /* node that zone *z is on */
1111
1112 zlc = zonelist->zlcache_ptr;
1113 if (!zlc)
1114 return 1;
1115
1116 i = z - zonelist->zones;
1117 n = zlc->z_to_n[i];
1118
1119 /* This zone is worth trying if it is allowed but not full */
1120 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1121}
1122
1123/*
1124 * Given 'z' scanning a zonelist, set the corresponding bit in
1125 * zlc->fullzones, so that subsequent attempts to allocate a page
1126 * from that zone don't waste time re-examining it.
1127 */
1128static void zlc_mark_zone_full(struct zonelist *zonelist, struct zone **z)
1129{
1130 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1131 int i; /* index of *z in zonelist zones */
1132
1133 zlc = zonelist->zlcache_ptr;
1134 if (!zlc)
1135 return;
1136
1137 i = z - zonelist->zones;
1138
1139 set_bit(i, zlc->fullzones);
1140}
1141
1142#else /* CONFIG_NUMA */
1143
1144static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1145{
1146 return NULL;
1147}
1148
1149static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zone **z,
1150 nodemask_t *allowednodes)
1151{
1152 return 1;
1153}
1154
1155static void zlc_mark_zone_full(struct zonelist *zonelist, struct zone **z)
1156{
1157}
1158#endif /* CONFIG_NUMA */
1159
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001160/*
Paul Jackson0798e512006-12-06 20:31:38 -08001161 * get_page_from_freelist goes through the zonelist trying to allocate
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001162 * a page.
1163 */
1164static struct page *
1165get_page_from_freelist(gfp_t gfp_mask, unsigned int order,
1166 struct zonelist *zonelist, int alloc_flags)
Martin Hicks753ee722005-06-21 17:14:41 -07001167{
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001168 struct zone **z;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001169 struct page *page = NULL;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001170 int classzone_idx = zone_idx(zonelist->zones[0]);
Christoph Lameter1192d522006-09-25 23:31:45 -07001171 struct zone *zone;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001172 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1173 int zlc_active = 0; /* set if using zonelist_cache */
1174 int did_zlc_setup = 0; /* just call zlc_setup() one time */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001175
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001176zonelist_scan:
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001177 /*
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001178 * Scan zonelist, looking for a zone with enough free.
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001179 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1180 */
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001181 z = zonelist->zones;
1182
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001183 do {
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001184 if (NUMA_BUILD && zlc_active &&
1185 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1186 continue;
Christoph Lameter1192d522006-09-25 23:31:45 -07001187 zone = *z;
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07001188 if (unlikely(NUMA_BUILD && (gfp_mask & __GFP_THISNODE) &&
Christoph Lameter1192d522006-09-25 23:31:45 -07001189 zone->zone_pgdat != zonelist->zones[0]->zone_pgdat))
Christoph Lameter9b819d22006-09-25 23:31:40 -07001190 break;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001191 if ((alloc_flags & ALLOC_CPUSET) &&
Paul Jackson02a0e532006-12-13 00:34:25 -08001192 !cpuset_zone_allowed_softwall(zone, gfp_mask))
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001193 goto try_next_zone;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001194
1195 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
Nick Piggin31488902005-11-28 13:44:03 -08001196 unsigned long mark;
1197 if (alloc_flags & ALLOC_WMARK_MIN)
Christoph Lameter1192d522006-09-25 23:31:45 -07001198 mark = zone->pages_min;
Nick Piggin31488902005-11-28 13:44:03 -08001199 else if (alloc_flags & ALLOC_WMARK_LOW)
Christoph Lameter1192d522006-09-25 23:31:45 -07001200 mark = zone->pages_low;
Nick Piggin31488902005-11-28 13:44:03 -08001201 else
Christoph Lameter1192d522006-09-25 23:31:45 -07001202 mark = zone->pages_high;
Paul Jackson0798e512006-12-06 20:31:38 -08001203 if (!zone_watermark_ok(zone, order, mark,
1204 classzone_idx, alloc_flags)) {
Christoph Lameter9eeff232006-01-18 17:42:31 -08001205 if (!zone_reclaim_mode ||
Christoph Lameter1192d522006-09-25 23:31:45 -07001206 !zone_reclaim(zone, gfp_mask, order))
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001207 goto this_zone_full;
Paul Jackson0798e512006-12-06 20:31:38 -08001208 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001209 }
1210
Christoph Lameter1192d522006-09-25 23:31:45 -07001211 page = buffered_rmqueue(zonelist, zone, order, gfp_mask);
Paul Jackson0798e512006-12-06 20:31:38 -08001212 if (page)
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001213 break;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001214this_zone_full:
1215 if (NUMA_BUILD)
1216 zlc_mark_zone_full(zonelist, z);
1217try_next_zone:
1218 if (NUMA_BUILD && !did_zlc_setup) {
1219 /* we do zlc_setup after the first zone is tried */
1220 allowednodes = zlc_setup(zonelist, alloc_flags);
1221 zlc_active = 1;
1222 did_zlc_setup = 1;
1223 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001224 } while (*(++z) != NULL);
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001225
1226 if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) {
1227 /* Disable zlc cache for second zonelist scan */
1228 zlc_active = 0;
1229 goto zonelist_scan;
1230 }
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001231 return page;
Martin Hicks753ee722005-06-21 17:14:41 -07001232}
1233
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234/*
1235 * This is the 'heart' of the zoned buddy allocator.
1236 */
1237struct page * fastcall
Al Virodd0fc662005-10-07 07:46:04 +01001238__alloc_pages(gfp_t gfp_mask, unsigned int order,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239 struct zonelist *zonelist)
1240{
Al Viro260b2362005-10-21 03:22:44 -04001241 const gfp_t wait = gfp_mask & __GFP_WAIT;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001242 struct zone **z;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243 struct page *page;
1244 struct reclaim_state reclaim_state;
1245 struct task_struct *p = current;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246 int do_retry;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001247 int alloc_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248 int did_some_progress;
1249
1250 might_sleep_if(wait);
1251
Akinobu Mita933e3122006-12-08 02:39:45 -08001252 if (should_fail_alloc_page(gfp_mask, order))
1253 return NULL;
1254
Jens Axboe6b1de912005-11-17 21:35:02 +01001255restart:
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001256 z = zonelist->zones; /* the list of zones suitable for gfp_mask */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001258 if (unlikely(*z == NULL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259 /* Should this ever happen?? */
1260 return NULL;
1261 }
Jens Axboe6b1de912005-11-17 21:35:02 +01001262
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001263 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
Nick Piggin31488902005-11-28 13:44:03 -08001264 zonelist, ALLOC_WMARK_LOW|ALLOC_CPUSET);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001265 if (page)
1266 goto got_pg;
1267
Christoph Lameter952f3b52006-12-06 20:33:26 -08001268 /*
1269 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1270 * __GFP_NOWARN set) should not cause reclaim since the subsystem
1271 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1272 * using a larger set of nodes after it has established that the
1273 * allowed per node queues are empty and that nodes are
1274 * over allocated.
1275 */
1276 if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
1277 goto nopage;
1278
Paul Jackson0798e512006-12-06 20:31:38 -08001279 for (z = zonelist->zones; *z; z++)
Chris Wright43b0bc02006-06-25 05:47:55 -07001280 wakeup_kswapd(*z, order);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001281
Paul Jackson9bf22292005-09-06 15:18:12 -07001282 /*
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001283 * OK, we're below the kswapd watermark and have kicked background
1284 * reclaim. Now things get more complex, so set up alloc_flags according
1285 * to how we want to proceed.
1286 *
1287 * The caller may dip into page reserves a bit more if the caller
1288 * cannot run direct reclaim, or if the caller has realtime scheduling
Paul Jackson4eac9152006-01-11 12:17:19 -08001289 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1290 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
Paul Jackson9bf22292005-09-06 15:18:12 -07001291 */
Nick Piggin31488902005-11-28 13:44:03 -08001292 alloc_flags = ALLOC_WMARK_MIN;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001293 if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait)
1294 alloc_flags |= ALLOC_HARDER;
1295 if (gfp_mask & __GFP_HIGH)
1296 alloc_flags |= ALLOC_HIGH;
Paul Jacksonbdd804f2006-05-20 15:00:09 -07001297 if (wait)
1298 alloc_flags |= ALLOC_CPUSET;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299
1300 /*
1301 * Go through the zonelist again. Let __GFP_HIGH and allocations
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001302 * coming from realtime tasks go deeper into reserves.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303 *
1304 * This is the last chance, in general, before the goto nopage.
1305 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
Paul Jackson9bf22292005-09-06 15:18:12 -07001306 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307 */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001308 page = get_page_from_freelist(gfp_mask, order, zonelist, alloc_flags);
1309 if (page)
1310 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311
1312 /* This allocation should allow future memory freeing. */
Nick Pigginb84a35b2005-05-01 08:58:36 -07001313
Kirill Korotaevb43a57b2006-12-06 20:32:27 -08001314rebalance:
Nick Pigginb84a35b2005-05-01 08:58:36 -07001315 if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE)))
1316 && !in_interrupt()) {
1317 if (!(gfp_mask & __GFP_NOMEMALLOC)) {
Kirill Korotaev885036d2005-11-13 16:06:41 -08001318nofail_alloc:
Nick Pigginb84a35b2005-05-01 08:58:36 -07001319 /* go through the zonelist yet again, ignoring mins */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001320 page = get_page_from_freelist(gfp_mask, order,
Paul Jackson47f3a862006-01-06 00:10:32 -08001321 zonelist, ALLOC_NO_WATERMARKS);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001322 if (page)
1323 goto got_pg;
Kirill Korotaev885036d2005-11-13 16:06:41 -08001324 if (gfp_mask & __GFP_NOFAIL) {
Andrew Morton3fcfab12006-10-19 23:28:16 -07001325 congestion_wait(WRITE, HZ/50);
Kirill Korotaev885036d2005-11-13 16:06:41 -08001326 goto nofail_alloc;
1327 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328 }
1329 goto nopage;
1330 }
1331
1332 /* Atomic allocations - we can't balance anything */
1333 if (!wait)
1334 goto nopage;
1335
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336 cond_resched();
1337
1338 /* We now go into synchronous reclaim */
Paul Jackson3e0d98b2006-01-08 01:01:49 -08001339 cpuset_memory_pressure_bump();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340 p->flags |= PF_MEMALLOC;
1341 reclaim_state.reclaimed_slab = 0;
1342 p->reclaim_state = &reclaim_state;
1343
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001344 did_some_progress = try_to_free_pages(zonelist->zones, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345
1346 p->reclaim_state = NULL;
1347 p->flags &= ~PF_MEMALLOC;
1348
1349 cond_resched();
1350
1351 if (likely(did_some_progress)) {
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001352 page = get_page_from_freelist(gfp_mask, order,
1353 zonelist, alloc_flags);
1354 if (page)
1355 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
1357 /*
1358 * Go through the zonelist yet one more time, keep
1359 * very high watermark here, this is only to catch
1360 * a parallel oom killing, we must fail if we're still
1361 * under heavy pressure.
1362 */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001363 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
Nick Piggin31488902005-11-28 13:44:03 -08001364 zonelist, ALLOC_WMARK_HIGH|ALLOC_CPUSET);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -08001365 if (page)
1366 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367
Christoph Lameter9b0f8b02006-02-20 18:27:52 -08001368 out_of_memory(zonelist, gfp_mask, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369 goto restart;
1370 }
1371
1372 /*
1373 * Don't let big-order allocations loop unless the caller explicitly
1374 * requests that. Wait for some write requests to complete then retry.
1375 *
1376 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
1377 * <= 3, but that may not be true in other implementations.
1378 */
1379 do_retry = 0;
1380 if (!(gfp_mask & __GFP_NORETRY)) {
1381 if ((order <= 3) || (gfp_mask & __GFP_REPEAT))
1382 do_retry = 1;
1383 if (gfp_mask & __GFP_NOFAIL)
1384 do_retry = 1;
1385 }
1386 if (do_retry) {
Andrew Morton3fcfab12006-10-19 23:28:16 -07001387 congestion_wait(WRITE, HZ/50);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388 goto rebalance;
1389 }
1390
1391nopage:
1392 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
1393 printk(KERN_WARNING "%s: page allocation failure."
1394 " order:%d, mode:0x%x\n",
1395 p->comm, order, gfp_mask);
1396 dump_stack();
Janet Morgan578c2fd2005-06-21 17:14:56 -07001397 show_mem();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399got_pg:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400 return page;
1401}
1402
1403EXPORT_SYMBOL(__alloc_pages);
1404
1405/*
1406 * Common helper functions.
1407 */
Al Virodd0fc662005-10-07 07:46:04 +01001408fastcall unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409{
1410 struct page * page;
1411 page = alloc_pages(gfp_mask, order);
1412 if (!page)
1413 return 0;
1414 return (unsigned long) page_address(page);
1415}
1416
1417EXPORT_SYMBOL(__get_free_pages);
1418
Al Virodd0fc662005-10-07 07:46:04 +01001419fastcall unsigned long get_zeroed_page(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420{
1421 struct page * page;
1422
1423 /*
1424 * get_zeroed_page() returns a 32-bit address, which cannot represent
1425 * a highmem page
1426 */
Nick Piggin725d7042006-09-25 23:30:55 -07001427 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428
1429 page = alloc_pages(gfp_mask | __GFP_ZERO, 0);
1430 if (page)
1431 return (unsigned long) page_address(page);
1432 return 0;
1433}
1434
1435EXPORT_SYMBOL(get_zeroed_page);
1436
1437void __pagevec_free(struct pagevec *pvec)
1438{
1439 int i = pagevec_count(pvec);
1440
1441 while (--i >= 0)
1442 free_hot_cold_page(pvec->pages[i], pvec->cold);
1443}
1444
1445fastcall void __free_pages(struct page *page, unsigned int order)
1446{
Nick Pigginb5810032005-10-29 18:16:12 -07001447 if (put_page_testzero(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448 if (order == 0)
1449 free_hot_page(page);
1450 else
1451 __free_pages_ok(page, order);
1452 }
1453}
1454
1455EXPORT_SYMBOL(__free_pages);
1456
1457fastcall void free_pages(unsigned long addr, unsigned int order)
1458{
1459 if (addr != 0) {
Nick Piggin725d7042006-09-25 23:30:55 -07001460 VM_BUG_ON(!virt_addr_valid((void *)addr));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461 __free_pages(virt_to_page((void *)addr), order);
1462 }
1463}
1464
1465EXPORT_SYMBOL(free_pages);
1466
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467static unsigned int nr_free_zone_pages(int offset)
1468{
Martin J. Blighe310fd42005-07-29 22:59:18 -07001469 /* Just pick one node, since fallback list is circular */
1470 pg_data_t *pgdat = NODE_DATA(numa_node_id());
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471 unsigned int sum = 0;
1472
Martin J. Blighe310fd42005-07-29 22:59:18 -07001473 struct zonelist *zonelist = pgdat->node_zonelists + offset;
1474 struct zone **zonep = zonelist->zones;
1475 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476
Martin J. Blighe310fd42005-07-29 22:59:18 -07001477 for (zone = *zonep++; zone; zone = *zonep++) {
1478 unsigned long size = zone->present_pages;
1479 unsigned long high = zone->pages_high;
1480 if (size > high)
1481 sum += size - high;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482 }
1483
1484 return sum;
1485}
1486
1487/*
1488 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1489 */
1490unsigned int nr_free_buffer_pages(void)
1491{
Al Viroaf4ca452005-10-21 02:55:38 -04001492 return nr_free_zone_pages(gfp_zone(GFP_USER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493}
1494
1495/*
1496 * Amount of free RAM allocatable within all zones
1497 */
1498unsigned int nr_free_pagecache_pages(void)
1499{
Al Viroaf4ca452005-10-21 02:55:38 -04001500 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501}
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07001502
1503static inline void show_node(struct zone *zone)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504{
Christoph Lameter08e0f6a2006-09-27 01:50:06 -07001505 if (NUMA_BUILD)
Andy Whitcroft25ba77c2006-12-06 20:33:03 -08001506 printk("Node %d ", zone_to_nid(zone));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509void si_meminfo(struct sysinfo *val)
1510{
1511 val->totalram = totalram_pages;
1512 val->sharedram = 0;
Christoph Lameterd23ad422007-02-10 01:43:02 -08001513 val->freeram = global_page_state(NR_FREE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514 val->bufferram = nr_blockdev_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515 val->totalhigh = totalhigh_pages;
1516 val->freehigh = nr_free_highpages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001517 val->mem_unit = PAGE_SIZE;
1518}
1519
1520EXPORT_SYMBOL(si_meminfo);
1521
1522#ifdef CONFIG_NUMA
1523void si_meminfo_node(struct sysinfo *val, int nid)
1524{
1525 pg_data_t *pgdat = NODE_DATA(nid);
1526
1527 val->totalram = pgdat->node_present_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08001528 val->freeram = node_page_state(nid, NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07001529#ifdef CONFIG_HIGHMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
Christoph Lameterd23ad422007-02-10 01:43:02 -08001531 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
1532 NR_FREE_PAGES);
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07001533#else
1534 val->totalhigh = 0;
1535 val->freehigh = 0;
1536#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537 val->mem_unit = PAGE_SIZE;
1538}
1539#endif
1540
1541#define K(x) ((x) << (PAGE_SHIFT-10))
1542
1543/*
1544 * Show free area list (used inside shift_scroll-lock stuff)
1545 * We also calculate the percentage fragmentation. We do this by counting the
1546 * memory on each free list with the exception of the first item on the list.
1547 */
1548void show_free_areas(void)
1549{
Jes Sorensenc7241912006-09-27 01:50:05 -07001550 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551 struct zone *zone;
1552
1553 for_each_zone(zone) {
Jes Sorensenc7241912006-09-27 01:50:05 -07001554 if (!populated_zone(zone))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555 continue;
Jes Sorensenc7241912006-09-27 01:50:05 -07001556
1557 show_node(zone);
1558 printk("%s per-cpu:\n", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559
Dave Jones6b482c62005-11-10 15:45:56 -05001560 for_each_online_cpu(cpu) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561 struct per_cpu_pageset *pageset;
1562
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001563 pageset = zone_pcp(zone, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564
Jes Sorensenc7241912006-09-27 01:50:05 -07001565 printk("CPU %4d: Hot: hi:%5d, btch:%4d usd:%4d "
1566 "Cold: hi:%5d, btch:%4d usd:%4d\n",
1567 cpu, pageset->pcp[0].high,
1568 pageset->pcp[0].batch, pageset->pcp[0].count,
1569 pageset->pcp[1].high, pageset->pcp[1].batch,
1570 pageset->pcp[1].count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571 }
1572 }
1573
Andrew Mortona25700a2007-02-08 14:20:40 -08001574 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu unstable:%lu\n"
Christoph Lameterd23ad422007-02-10 01:43:02 -08001575 " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
Christoph Lameter65e458d42007-02-10 01:43:05 -08001576 global_page_state(NR_ACTIVE),
1577 global_page_state(NR_INACTIVE),
Christoph Lameterb1e7a8f2006-06-30 01:55:39 -07001578 global_page_state(NR_FILE_DIRTY),
Christoph Lameterce866b32006-06-30 01:55:40 -07001579 global_page_state(NR_WRITEBACK),
Christoph Lameterfd39fc82006-06-30 01:55:40 -07001580 global_page_state(NR_UNSTABLE_NFS),
Christoph Lameterd23ad422007-02-10 01:43:02 -08001581 global_page_state(NR_FREE_PAGES),
Christoph Lameter972d1a72006-09-25 23:31:51 -07001582 global_page_state(NR_SLAB_RECLAIMABLE) +
1583 global_page_state(NR_SLAB_UNRECLAIMABLE),
Christoph Lameter65ba55f2006-06-30 01:55:34 -07001584 global_page_state(NR_FILE_MAPPED),
Andrew Mortona25700a2007-02-08 14:20:40 -08001585 global_page_state(NR_PAGETABLE),
1586 global_page_state(NR_BOUNCE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587
1588 for_each_zone(zone) {
1589 int i;
1590
Jes Sorensenc7241912006-09-27 01:50:05 -07001591 if (!populated_zone(zone))
1592 continue;
1593
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594 show_node(zone);
1595 printk("%s"
1596 " free:%lukB"
1597 " min:%lukB"
1598 " low:%lukB"
1599 " high:%lukB"
1600 " active:%lukB"
1601 " inactive:%lukB"
1602 " present:%lukB"
1603 " pages_scanned:%lu"
1604 " all_unreclaimable? %s"
1605 "\n",
1606 zone->name,
Christoph Lameterd23ad422007-02-10 01:43:02 -08001607 K(zone_page_state(zone, NR_FREE_PAGES)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608 K(zone->pages_min),
1609 K(zone->pages_low),
1610 K(zone->pages_high),
Christoph Lameterc8785382007-02-10 01:43:01 -08001611 K(zone_page_state(zone, NR_ACTIVE)),
1612 K(zone_page_state(zone, NR_INACTIVE)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613 K(zone->present_pages),
1614 zone->pages_scanned,
1615 (zone->all_unreclaimable ? "yes" : "no")
1616 );
1617 printk("lowmem_reserve[]:");
1618 for (i = 0; i < MAX_NR_ZONES; i++)
1619 printk(" %lu", zone->lowmem_reserve[i]);
1620 printk("\n");
1621 }
1622
1623 for_each_zone(zone) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07001624 unsigned long nr[MAX_ORDER], flags, order, total = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625
Jes Sorensenc7241912006-09-27 01:50:05 -07001626 if (!populated_zone(zone))
1627 continue;
1628
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629 show_node(zone);
1630 printk("%s: ", zone->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001631
1632 spin_lock_irqsave(&zone->lock, flags);
1633 for (order = 0; order < MAX_ORDER; order++) {
Kirill Korotaev8f9de512006-06-23 02:03:50 -07001634 nr[order] = zone->free_area[order].nr_free;
1635 total += nr[order] << order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636 }
1637 spin_unlock_irqrestore(&zone->lock, flags);
Kirill Korotaev8f9de512006-06-23 02:03:50 -07001638 for (order = 0; order < MAX_ORDER; order++)
1639 printk("%lu*%lukB ", nr[order], K(1UL) << order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640 printk("= %lukB\n", K(total));
1641 }
1642
1643 show_swap_cache_info();
1644}
1645
1646/*
1647 * Builds allocation fallback zone lists.
Christoph Lameter1a932052006-01-06 00:11:16 -08001648 *
1649 * Add all populated zones of a node to the zonelist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650 */
Yasunori Goto86356ab2006-06-23 02:03:09 -07001651static int __meminit build_zonelists_node(pg_data_t *pgdat,
Christoph Lameter2f6726e2006-09-25 23:31:18 -07001652 struct zonelist *zonelist, int nr_zones, enum zone_type zone_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653{
Christoph Lameter1a932052006-01-06 00:11:16 -08001654 struct zone *zone;
1655
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07001656 BUG_ON(zone_type >= MAX_NR_ZONES);
Christoph Lameter2f6726e2006-09-25 23:31:18 -07001657 zone_type++;
Christoph Lameter02a68a52006-01-06 00:11:18 -08001658
1659 do {
Christoph Lameter2f6726e2006-09-25 23:31:18 -07001660 zone_type--;
Christoph Lameter070f8032006-01-06 00:11:19 -08001661 zone = pgdat->node_zones + zone_type;
Christoph Lameter1a932052006-01-06 00:11:16 -08001662 if (populated_zone(zone)) {
Christoph Lameter070f8032006-01-06 00:11:19 -08001663 zonelist->zones[nr_zones++] = zone;
1664 check_highest_zone(zone_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665 }
Christoph Lameter02a68a52006-01-06 00:11:18 -08001666
Christoph Lameter2f6726e2006-09-25 23:31:18 -07001667 } while (zone_type);
Christoph Lameter070f8032006-01-06 00:11:19 -08001668 return nr_zones;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001669}
1670
1671#ifdef CONFIG_NUMA
1672#define MAX_NODE_LOAD (num_online_nodes())
Yasunori Goto86356ab2006-06-23 02:03:09 -07001673static int __meminitdata node_load[MAX_NUMNODES];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -07001675 * 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 -07001676 * @node: node whose fallback list we're appending
1677 * @used_node_mask: nodemask_t of already used nodes
1678 *
1679 * We use a number of factors to determine which is the next node that should
1680 * appear on a given node's fallback list. The node should not have appeared
1681 * already in @node's fallback list, and it should be the next closest node
1682 * according to the distance array (which contains arbitrary distance values
1683 * from each node to each node in the system), and should also prefer nodes
1684 * with no CPUs, since presumably they'll have very little allocation pressure
1685 * on them otherwise.
1686 * It returns -1 if no node is found.
1687 */
Yasunori Goto86356ab2006-06-23 02:03:09 -07001688static int __meminit find_next_best_node(int node, nodemask_t *used_node_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689{
Linus Torvalds4cf808e2006-02-17 20:38:21 +01001690 int n, val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001691 int min_val = INT_MAX;
1692 int best_node = -1;
1693
Linus Torvalds4cf808e2006-02-17 20:38:21 +01001694 /* Use the local node if we haven't already */
1695 if (!node_isset(node, *used_node_mask)) {
1696 node_set(node, *used_node_mask);
1697 return node;
1698 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699
Linus Torvalds4cf808e2006-02-17 20:38:21 +01001700 for_each_online_node(n) {
1701 cpumask_t tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702
1703 /* Don't want a node to appear more than once */
1704 if (node_isset(n, *used_node_mask))
1705 continue;
1706
Linus Torvalds1da177e2005-04-16 15:20:36 -07001707 /* Use the distance array to find the distance */
1708 val = node_distance(node, n);
1709
Linus Torvalds4cf808e2006-02-17 20:38:21 +01001710 /* Penalize nodes under us ("prefer the next node") */
1711 val += (n < node);
1712
Linus Torvalds1da177e2005-04-16 15:20:36 -07001713 /* Give preference to headless and unused nodes */
1714 tmp = node_to_cpumask(n);
1715 if (!cpus_empty(tmp))
1716 val += PENALTY_FOR_NODE_WITH_CPUS;
1717
1718 /* Slight preference for less loaded node */
1719 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
1720 val += node_load[n];
1721
1722 if (val < min_val) {
1723 min_val = val;
1724 best_node = n;
1725 }
1726 }
1727
1728 if (best_node >= 0)
1729 node_set(best_node, *used_node_mask);
1730
1731 return best_node;
1732}
1733
Yasunori Goto86356ab2006-06-23 02:03:09 -07001734static void __meminit build_zonelists(pg_data_t *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735{
Christoph Lameter19655d32006-09-25 23:31:19 -07001736 int j, node, local_node;
1737 enum zone_type i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001738 int prev_node, load;
1739 struct zonelist *zonelist;
1740 nodemask_t used_mask;
1741
1742 /* initialize zonelists */
Christoph Lameter19655d32006-09-25 23:31:19 -07001743 for (i = 0; i < MAX_NR_ZONES; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744 zonelist = pgdat->node_zonelists + i;
1745 zonelist->zones[0] = NULL;
1746 }
1747
1748 /* NUMA-aware ordering of nodes */
1749 local_node = pgdat->node_id;
1750 load = num_online_nodes();
1751 prev_node = local_node;
1752 nodes_clear(used_mask);
1753 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
Christoph Lameter9eeff232006-01-18 17:42:31 -08001754 int distance = node_distance(local_node, node);
1755
1756 /*
1757 * If another node is sufficiently far away then it is better
1758 * to reclaim pages in a zone before going off node.
1759 */
1760 if (distance > RECLAIM_DISTANCE)
1761 zone_reclaim_mode = 1;
1762
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763 /*
1764 * We don't want to pressure a particular node.
1765 * So adding penalty to the first node in same
1766 * distance group to make it round-robin.
1767 */
Christoph Lameter9eeff232006-01-18 17:42:31 -08001768
1769 if (distance != node_distance(local_node, prev_node))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001770 node_load[node] += load;
1771 prev_node = node;
1772 load--;
Christoph Lameter19655d32006-09-25 23:31:19 -07001773 for (i = 0; i < MAX_NR_ZONES; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774 zonelist = pgdat->node_zonelists + i;
1775 for (j = 0; zonelist->zones[j] != NULL; j++);
1776
Christoph Lameter19655d32006-09-25 23:31:19 -07001777 j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001778 zonelist->zones[j] = NULL;
1779 }
1780 }
1781}
1782
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001783/* Construct the zonelist performance cache - see further mmzone.h */
1784static void __meminit build_zonelist_cache(pg_data_t *pgdat)
1785{
1786 int i;
1787
1788 for (i = 0; i < MAX_NR_ZONES; i++) {
1789 struct zonelist *zonelist;
1790 struct zonelist_cache *zlc;
1791 struct zone **z;
1792
1793 zonelist = pgdat->node_zonelists + i;
1794 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
1795 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1796 for (z = zonelist->zones; *z; z++)
1797 zlc->z_to_n[z - zonelist->zones] = zone_to_nid(*z);
1798 }
1799}
1800
Linus Torvalds1da177e2005-04-16 15:20:36 -07001801#else /* CONFIG_NUMA */
1802
Yasunori Goto86356ab2006-06-23 02:03:09 -07001803static void __meminit build_zonelists(pg_data_t *pgdat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001804{
Christoph Lameter19655d32006-09-25 23:31:19 -07001805 int node, local_node;
1806 enum zone_type i,j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001807
1808 local_node = pgdat->node_id;
Christoph Lameter19655d32006-09-25 23:31:19 -07001809 for (i = 0; i < MAX_NR_ZONES; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001810 struct zonelist *zonelist;
1811
1812 zonelist = pgdat->node_zonelists + i;
1813
Christoph Lameter19655d32006-09-25 23:31:19 -07001814 j = build_zonelists_node(pgdat, zonelist, 0, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815 /*
1816 * Now we build the zonelist so that it contains the zones
1817 * of all the other nodes.
1818 * We don't want to pressure a particular node, so when
1819 * building the zones for node N, we make sure that the
1820 * zones coming right after the local ones are those from
1821 * node N+1 (modulo N)
1822 */
1823 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
1824 if (!node_online(node))
1825 continue;
Christoph Lameter19655d32006-09-25 23:31:19 -07001826 j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001827 }
1828 for (node = 0; node < local_node; node++) {
1829 if (!node_online(node))
1830 continue;
Christoph Lameter19655d32006-09-25 23:31:19 -07001831 j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832 }
1833
1834 zonelist->zones[j] = NULL;
1835 }
1836}
1837
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001838/* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
1839static void __meminit build_zonelist_cache(pg_data_t *pgdat)
1840{
1841 int i;
1842
1843 for (i = 0; i < MAX_NR_ZONES; i++)
1844 pgdat->node_zonelists[i].zlcache_ptr = NULL;
1845}
1846
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847#endif /* CONFIG_NUMA */
1848
Yasunori Goto68113782006-06-23 02:03:11 -07001849/* return values int ....just for stop_machine_run() */
1850static int __meminit __build_all_zonelists(void *dummy)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001851{
Yasunori Goto68113782006-06-23 02:03:11 -07001852 int nid;
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001853
1854 for_each_online_node(nid) {
Yasunori Goto68113782006-06-23 02:03:11 -07001855 build_zonelists(NODE_DATA(nid));
Paul Jackson9276b1bc2006-12-06 20:31:48 -08001856 build_zonelist_cache(NODE_DATA(nid));
1857 }
Yasunori Goto68113782006-06-23 02:03:11 -07001858 return 0;
1859}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001860
Yasunori Goto68113782006-06-23 02:03:11 -07001861void __meminit build_all_zonelists(void)
1862{
1863 if (system_state == SYSTEM_BOOTING) {
Randy Dunlap423b41d2006-09-27 01:50:12 -07001864 __build_all_zonelists(NULL);
Yasunori Goto68113782006-06-23 02:03:11 -07001865 cpuset_init_current_mems_allowed();
1866 } else {
1867 /* we have to stop all cpus to guaranntee there is no user
1868 of zonelist */
1869 stop_machine_run(__build_all_zonelists, NULL, NR_CPUS);
1870 /* cpuset refresh routine should be here */
1871 }
Andrew Mortonbd1e22b2006-06-23 02:03:47 -07001872 vm_total_pages = nr_free_pagecache_pages();
1873 printk("Built %i zonelists. Total pages: %ld\n",
1874 num_online_nodes(), vm_total_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001875}
1876
1877/*
1878 * Helper functions to size the waitqueue hash table.
1879 * Essentially these want to choose hash table sizes sufficiently
1880 * large so that collisions trying to wait on pages are rare.
1881 * But in fact, the number of active page waitqueues on typical
1882 * systems is ridiculously low, less than 200. So this is even
1883 * conservative, even though it seems large.
1884 *
1885 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
1886 * waitqueues, i.e. the size of the waitq table given the number of pages.
1887 */
1888#define PAGES_PER_WAITQUEUE 256
1889
Yasunori Gotocca448f2006-06-23 02:03:10 -07001890#ifndef CONFIG_MEMORY_HOTPLUG
Yasunori Goto02b694d2006-06-23 02:03:08 -07001891static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892{
1893 unsigned long size = 1;
1894
1895 pages /= PAGES_PER_WAITQUEUE;
1896
1897 while (size < pages)
1898 size <<= 1;
1899
1900 /*
1901 * Once we have dozens or even hundreds of threads sleeping
1902 * on IO we've got bigger problems than wait queue collision.
1903 * Limit the size of the wait table to a reasonable size.
1904 */
1905 size = min(size, 4096UL);
1906
1907 return max(size, 4UL);
1908}
Yasunori Gotocca448f2006-06-23 02:03:10 -07001909#else
1910/*
1911 * A zone's size might be changed by hot-add, so it is not possible to determine
1912 * a suitable size for its wait_table. So we use the maximum size now.
1913 *
1914 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
1915 *
1916 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
1917 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
1918 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
1919 *
1920 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
1921 * or more by the traditional way. (See above). It equals:
1922 *
1923 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
1924 * ia64(16K page size) : = ( 8G + 4M)byte.
1925 * powerpc (64K page size) : = (32G +16M)byte.
1926 */
1927static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
1928{
1929 return 4096UL;
1930}
1931#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001932
1933/*
1934 * This is an integer logarithm so that shifts can be used later
1935 * to extract the more random high bits from the multiplicative
1936 * hash function before the remainder is taken.
1937 */
1938static inline unsigned long wait_table_bits(unsigned long size)
1939{
1940 return ffz(~size);
1941}
1942
1943#define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
1944
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945/*
1946 * Initially all pages are reserved - free ones are freed
1947 * up by free_all_bootmem() once the early boot process is
1948 * done. Non-atomic initialization, single-pass.
1949 */
Matt Tolentinoc09b4242006-01-17 07:03:44 +01001950void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
Dave Hansena2f3aa02007-01-10 23:15:30 -08001951 unsigned long start_pfn, enum memmap_context context)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953 struct page *page;
Andy Whitcroft29751f62005-06-23 00:08:00 -07001954 unsigned long end_pfn = start_pfn + size;
1955 unsigned long pfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001956
Greg Ungerercbe8dd42006-01-12 01:05:24 -08001957 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
Dave Hansena2f3aa02007-01-10 23:15:30 -08001958 /*
1959 * There can be holes in boot-time mem_map[]s
1960 * handed to this function. They do not
1961 * exist on hotplugged memory.
1962 */
1963 if (context == MEMMAP_EARLY) {
1964 if (!early_pfn_valid(pfn))
1965 continue;
1966 if (!early_pfn_in_nid(pfn, nid))
1967 continue;
1968 }
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001969 page = pfn_to_page(pfn);
1970 set_page_links(page, zone, nid, pfn);
Nick Piggin7835e982006-03-22 00:08:40 -08001971 init_page_count(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972 reset_page_mapcount(page);
1973 SetPageReserved(page);
1974 INIT_LIST_HEAD(&page->lru);
1975#ifdef WANT_PAGE_VIRTUAL
1976 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
1977 if (!is_highmem_idx(zone))
Bob Picco3212c6b2005-06-27 14:36:28 -07001978 set_page_address(page, __va(pfn << PAGE_SHIFT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001980 }
1981}
1982
1983void zone_init_free_lists(struct pglist_data *pgdat, struct zone *zone,
1984 unsigned long size)
1985{
1986 int order;
1987 for (order = 0; order < MAX_ORDER ; order++) {
1988 INIT_LIST_HEAD(&zone->free_area[order].free_list);
1989 zone->free_area[order].nr_free = 0;
1990 }
1991}
1992
1993#ifndef __HAVE_ARCH_MEMMAP_INIT
1994#define memmap_init(size, nid, zone, start_pfn) \
Dave Hansena2f3aa02007-01-10 23:15:30 -08001995 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001996#endif
1997
Ashok Raj6292d9a2006-02-01 03:04:44 -08001998static int __cpuinit zone_batchsize(struct zone *zone)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001999{
2000 int batch;
2001
2002 /*
2003 * The per-cpu-pages pools are set to around 1000th of the
Seth, Rohitba56e912005-10-29 18:15:47 -07002004 * size of the zone. But no more than 1/2 of a meg.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002005 *
2006 * OK, so we don't know how big the cache is. So guess.
2007 */
2008 batch = zone->present_pages / 1024;
Seth, Rohitba56e912005-10-29 18:15:47 -07002009 if (batch * PAGE_SIZE > 512 * 1024)
2010 batch = (512 * 1024) / PAGE_SIZE;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002011 batch /= 4; /* We effectively *= 4 below */
2012 if (batch < 1)
2013 batch = 1;
2014
2015 /*
Nick Piggin0ceaacc2005-12-04 13:55:25 +11002016 * Clamp the batch to a 2^n - 1 value. Having a power
2017 * of 2 value was found to be more likely to have
2018 * suboptimal cache aliasing properties in some cases.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002019 *
Nick Piggin0ceaacc2005-12-04 13:55:25 +11002020 * For example if 2 tasks are alternately allocating
2021 * batches of pages, one task can end up with a lot
2022 * of pages of one half of the possible page colors
2023 * and the other with pages of the other colors.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002024 */
Nick Piggin0ceaacc2005-12-04 13:55:25 +11002025 batch = (1 << (fls(batch + batch/2)-1)) - 1;
Seth, Rohitba56e912005-10-29 18:15:47 -07002026
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002027 return batch;
2028}
2029
Christoph Lameter2caaad42005-06-21 17:15:00 -07002030inline void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
2031{
2032 struct per_cpu_pages *pcp;
2033
Magnus Damm1c6fe942005-10-26 01:58:59 -07002034 memset(p, 0, sizeof(*p));
2035
Christoph Lameter2caaad42005-06-21 17:15:00 -07002036 pcp = &p->pcp[0]; /* hot */
2037 pcp->count = 0;
Christoph Lameter2caaad42005-06-21 17:15:00 -07002038 pcp->high = 6 * batch;
2039 pcp->batch = max(1UL, 1 * batch);
2040 INIT_LIST_HEAD(&pcp->list);
2041
2042 pcp = &p->pcp[1]; /* cold*/
2043 pcp->count = 0;
Christoph Lameter2caaad42005-06-21 17:15:00 -07002044 pcp->high = 2 * batch;
Seth, Rohite46a5e22005-10-29 18:15:48 -07002045 pcp->batch = max(1UL, batch/2);
Christoph Lameter2caaad42005-06-21 17:15:00 -07002046 INIT_LIST_HEAD(&pcp->list);
2047}
2048
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08002049/*
2050 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
2051 * to the value high for the pageset p.
2052 */
2053
2054static void setup_pagelist_highmark(struct per_cpu_pageset *p,
2055 unsigned long high)
2056{
2057 struct per_cpu_pages *pcp;
2058
2059 pcp = &p->pcp[0]; /* hot list */
2060 pcp->high = high;
2061 pcp->batch = max(1UL, high/4);
2062 if ((high/4) > (PAGE_SHIFT * 8))
2063 pcp->batch = PAGE_SHIFT * 8;
2064}
2065
2066
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002067#ifdef CONFIG_NUMA
2068/*
Christoph Lameter2caaad42005-06-21 17:15:00 -07002069 * Boot pageset table. One per cpu which is going to be used for all
2070 * zones and all nodes. The parameters will be set in such a way
2071 * that an item put on a list will immediately be handed over to
2072 * the buddy list. This is safe since pageset manipulation is done
2073 * with interrupts disabled.
2074 *
2075 * Some NUMA counter updates may also be caught by the boot pagesets.
Christoph Lameterb7c84c62005-06-22 20:26:07 -07002076 *
2077 * The boot_pagesets must be kept even after bootup is complete for
2078 * unused processors and/or zones. They do play a role for bootstrapping
2079 * hotplugged processors.
2080 *
2081 * zoneinfo_show() and maybe other functions do
2082 * not check if the processor is online before following the pageset pointer.
2083 * Other parts of the kernel may not check if the zone is available.
Christoph Lameter2caaad42005-06-21 17:15:00 -07002084 */
Eric Dumazet88a2a4ac2006-02-04 23:27:36 -08002085static struct per_cpu_pageset boot_pageset[NR_CPUS];
Christoph Lameter2caaad42005-06-21 17:15:00 -07002086
2087/*
2088 * Dynamically allocate memory for the
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002089 * per cpu pageset array in struct zone.
2090 */
Ashok Raj6292d9a2006-02-01 03:04:44 -08002091static int __cpuinit process_zones(int cpu)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002092{
2093 struct zone *zone, *dzone;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002094
2095 for_each_zone(zone) {
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002096
Christoph Lameter66a55032006-09-27 01:50:09 -07002097 if (!populated_zone(zone))
2098 continue;
2099
Nick Piggin23316bc2006-01-08 01:00:41 -08002100 zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset),
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002101 GFP_KERNEL, cpu_to_node(cpu));
Nick Piggin23316bc2006-01-08 01:00:41 -08002102 if (!zone_pcp(zone, cpu))
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002103 goto bad;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002104
Nick Piggin23316bc2006-01-08 01:00:41 -08002105 setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone));
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08002106
2107 if (percpu_pagelist_fraction)
2108 setup_pagelist_highmark(zone_pcp(zone, cpu),
2109 (zone->present_pages / percpu_pagelist_fraction));
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002110 }
2111
2112 return 0;
2113bad:
2114 for_each_zone(dzone) {
2115 if (dzone == zone)
2116 break;
Nick Piggin23316bc2006-01-08 01:00:41 -08002117 kfree(zone_pcp(dzone, cpu));
2118 zone_pcp(dzone, cpu) = NULL;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002119 }
2120 return -ENOMEM;
2121}
2122
2123static inline void free_zone_pagesets(int cpu)
2124{
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002125 struct zone *zone;
2126
2127 for_each_zone(zone) {
2128 struct per_cpu_pageset *pset = zone_pcp(zone, cpu);
2129
David Rientjesf3ef9ea2006-09-25 16:24:57 -07002130 /* Free per_cpu_pageset if it is slab allocated */
2131 if (pset != &boot_pageset[cpu])
2132 kfree(pset);
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002133 zone_pcp(zone, cpu) = NULL;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002134 }
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002135}
2136
Chandra Seetharaman9c7b2162006-06-27 02:54:07 -07002137static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb,
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002138 unsigned long action,
2139 void *hcpu)
2140{
2141 int cpu = (long)hcpu;
2142 int ret = NOTIFY_OK;
2143
2144 switch (action) {
Andy Whitcroftce421c72006-12-06 20:33:08 -08002145 case CPU_UP_PREPARE:
2146 if (process_zones(cpu))
2147 ret = NOTIFY_BAD;
2148 break;
2149 case CPU_UP_CANCELED:
2150 case CPU_DEAD:
2151 free_zone_pagesets(cpu);
2152 break;
2153 default:
2154 break;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002155 }
2156 return ret;
2157}
2158
Chandra Seetharaman74b85f32006-06-27 02:54:09 -07002159static struct notifier_block __cpuinitdata pageset_notifier =
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002160 { &pageset_cpuup_callback, NULL, 0 };
2161
Al Viro78d99552005-12-15 09:18:25 +00002162void __init setup_per_cpu_pageset(void)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002163{
2164 int err;
2165
2166 /* Initialize per_cpu_pageset for cpu 0.
2167 * A cpuup callback will do this for every cpu
2168 * as it comes online
2169 */
2170 err = process_zones(smp_processor_id());
2171 BUG_ON(err);
2172 register_cpu_notifier(&pageset_notifier);
2173}
2174
2175#endif
2176
Matt Tolentinoc09b4242006-01-17 07:03:44 +01002177static __meminit
Yasunori Gotocca448f2006-06-23 02:03:10 -07002178int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
Dave Hansened8ece22005-10-29 18:16:50 -07002179{
2180 int i;
2181 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07002182 size_t alloc_size;
Dave Hansened8ece22005-10-29 18:16:50 -07002183
2184 /*
2185 * The per-page waitqueue mechanism uses hashed waitqueues
2186 * per zone.
2187 */
Yasunori Goto02b694d2006-06-23 02:03:08 -07002188 zone->wait_table_hash_nr_entries =
2189 wait_table_hash_nr_entries(zone_size_pages);
2190 zone->wait_table_bits =
2191 wait_table_bits(zone->wait_table_hash_nr_entries);
Yasunori Gotocca448f2006-06-23 02:03:10 -07002192 alloc_size = zone->wait_table_hash_nr_entries
2193 * sizeof(wait_queue_head_t);
2194
2195 if (system_state == SYSTEM_BOOTING) {
2196 zone->wait_table = (wait_queue_head_t *)
2197 alloc_bootmem_node(pgdat, alloc_size);
2198 } else {
2199 /*
2200 * This case means that a zone whose size was 0 gets new memory
2201 * via memory hot-add.
2202 * But it may be the case that a new node was hot-added. In
2203 * this case vmalloc() will not be able to use this new node's
2204 * memory - this wait_table must be initialized to use this new
2205 * node itself as well.
2206 * To use this new node's memory, further consideration will be
2207 * necessary.
2208 */
2209 zone->wait_table = (wait_queue_head_t *)vmalloc(alloc_size);
2210 }
2211 if (!zone->wait_table)
2212 return -ENOMEM;
Dave Hansened8ece22005-10-29 18:16:50 -07002213
Yasunori Goto02b694d2006-06-23 02:03:08 -07002214 for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
Dave Hansened8ece22005-10-29 18:16:50 -07002215 init_waitqueue_head(zone->wait_table + i);
Yasunori Gotocca448f2006-06-23 02:03:10 -07002216
2217 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07002218}
2219
Matt Tolentinoc09b4242006-01-17 07:03:44 +01002220static __meminit void zone_pcp_init(struct zone *zone)
Dave Hansened8ece22005-10-29 18:16:50 -07002221{
2222 int cpu;
2223 unsigned long batch = zone_batchsize(zone);
2224
2225 for (cpu = 0; cpu < NR_CPUS; cpu++) {
2226#ifdef CONFIG_NUMA
2227 /* Early boot. Slab allocator not functional yet */
Nick Piggin23316bc2006-01-08 01:00:41 -08002228 zone_pcp(zone, cpu) = &boot_pageset[cpu];
Dave Hansened8ece22005-10-29 18:16:50 -07002229 setup_pageset(&boot_pageset[cpu],0);
2230#else
2231 setup_pageset(zone_pcp(zone,cpu), batch);
2232#endif
2233 }
Anton Blanchardf5335c02006-03-25 03:06:49 -08002234 if (zone->present_pages)
2235 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n",
2236 zone->name, zone->present_pages, batch);
Dave Hansened8ece22005-10-29 18:16:50 -07002237}
2238
Yasunori Goto718127c2006-06-23 02:03:10 -07002239__meminit int init_currently_empty_zone(struct zone *zone,
2240 unsigned long zone_start_pfn,
Dave Hansena2f3aa02007-01-10 23:15:30 -08002241 unsigned long size,
2242 enum memmap_context context)
Dave Hansened8ece22005-10-29 18:16:50 -07002243{
2244 struct pglist_data *pgdat = zone->zone_pgdat;
Yasunori Gotocca448f2006-06-23 02:03:10 -07002245 int ret;
2246 ret = zone_wait_table_init(zone, size);
2247 if (ret)
2248 return ret;
Dave Hansened8ece22005-10-29 18:16:50 -07002249 pgdat->nr_zones = zone_idx(zone) + 1;
2250
Dave Hansened8ece22005-10-29 18:16:50 -07002251 zone->zone_start_pfn = zone_start_pfn;
2252
2253 memmap_init(size, pgdat->node_id, zone_idx(zone), zone_start_pfn);
2254
2255 zone_init_free_lists(pgdat, zone, zone->spanned_pages);
Yasunori Goto718127c2006-06-23 02:03:10 -07002256
2257 return 0;
Dave Hansened8ece22005-10-29 18:16:50 -07002258}
2259
Mel Gormanc7132162006-09-27 01:49:43 -07002260#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2261/*
2262 * Basic iterator support. Return the first range of PFNs for a node
2263 * Note: nid == MAX_NUMNODES returns first region regardless of node
2264 */
2265static int __init first_active_region_index_in_nid(int nid)
2266{
2267 int i;
2268
2269 for (i = 0; i < nr_nodemap_entries; i++)
2270 if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
2271 return i;
2272
2273 return -1;
2274}
2275
2276/*
2277 * Basic iterator support. Return the next active range of PFNs for a node
2278 * Note: nid == MAX_NUMNODES returns next region regardles of node
2279 */
2280static int __init next_active_region_index_in_nid(int index, int nid)
2281{
2282 for (index = index + 1; index < nr_nodemap_entries; index++)
2283 if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
2284 return index;
2285
2286 return -1;
2287}
2288
2289#ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
2290/*
2291 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
2292 * Architectures may implement their own version but if add_active_range()
2293 * was used and there are no special requirements, this is a convenient
2294 * alternative
2295 */
2296int __init early_pfn_to_nid(unsigned long pfn)
2297{
2298 int i;
2299
2300 for (i = 0; i < nr_nodemap_entries; i++) {
2301 unsigned long start_pfn = early_node_map[i].start_pfn;
2302 unsigned long end_pfn = early_node_map[i].end_pfn;
2303
2304 if (start_pfn <= pfn && pfn < end_pfn)
2305 return early_node_map[i].nid;
2306 }
2307
2308 return 0;
2309}
2310#endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
2311
2312/* Basic iterator support to walk early_node_map[] */
2313#define for_each_active_range_index_in_nid(i, nid) \
2314 for (i = first_active_region_index_in_nid(nid); i != -1; \
2315 i = next_active_region_index_in_nid(i, nid))
2316
2317/**
2318 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002319 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
2320 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
Mel Gormanc7132162006-09-27 01:49:43 -07002321 *
2322 * If an architecture guarantees that all ranges registered with
2323 * add_active_ranges() contain no holes and may be freed, this
2324 * this function may be used instead of calling free_bootmem() manually.
2325 */
2326void __init free_bootmem_with_active_regions(int nid,
2327 unsigned long max_low_pfn)
2328{
2329 int i;
2330
2331 for_each_active_range_index_in_nid(i, nid) {
2332 unsigned long size_pages = 0;
2333 unsigned long end_pfn = early_node_map[i].end_pfn;
2334
2335 if (early_node_map[i].start_pfn >= max_low_pfn)
2336 continue;
2337
2338 if (end_pfn > max_low_pfn)
2339 end_pfn = max_low_pfn;
2340
2341 size_pages = end_pfn - early_node_map[i].start_pfn;
2342 free_bootmem_node(NODE_DATA(early_node_map[i].nid),
2343 PFN_PHYS(early_node_map[i].start_pfn),
2344 size_pages << PAGE_SHIFT);
2345 }
2346}
2347
2348/**
2349 * sparse_memory_present_with_active_regions - Call memory_present for each active range
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002350 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
Mel Gormanc7132162006-09-27 01:49:43 -07002351 *
2352 * If an architecture guarantees that all ranges registered with
2353 * add_active_ranges() contain no holes and may be freed, this
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002354 * function may be used instead of calling memory_present() manually.
Mel Gormanc7132162006-09-27 01:49:43 -07002355 */
2356void __init sparse_memory_present_with_active_regions(int nid)
2357{
2358 int i;
2359
2360 for_each_active_range_index_in_nid(i, nid)
2361 memory_present(early_node_map[i].nid,
2362 early_node_map[i].start_pfn,
2363 early_node_map[i].end_pfn);
2364}
2365
2366/**
Mel Gormanfb014392006-09-27 01:49:59 -07002367 * push_node_boundaries - Push node boundaries to at least the requested boundary
2368 * @nid: The nid of the node to push the boundary for
2369 * @start_pfn: The start pfn of the node
2370 * @end_pfn: The end pfn of the node
2371 *
2372 * In reserve-based hot-add, mem_map is allocated that is unused until hotadd
2373 * time. Specifically, on x86_64, SRAT will report ranges that can potentially
2374 * be hotplugged even though no physical memory exists. This function allows
2375 * an arch to push out the node boundaries so mem_map is allocated that can
2376 * be used later.
2377 */
2378#ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
2379void __init push_node_boundaries(unsigned int nid,
2380 unsigned long start_pfn, unsigned long end_pfn)
2381{
2382 printk(KERN_DEBUG "Entering push_node_boundaries(%u, %lu, %lu)\n",
2383 nid, start_pfn, end_pfn);
2384
2385 /* Initialise the boundary for this node if necessary */
2386 if (node_boundary_end_pfn[nid] == 0)
2387 node_boundary_start_pfn[nid] = -1UL;
2388
2389 /* Update the boundaries */
2390 if (node_boundary_start_pfn[nid] > start_pfn)
2391 node_boundary_start_pfn[nid] = start_pfn;
2392 if (node_boundary_end_pfn[nid] < end_pfn)
2393 node_boundary_end_pfn[nid] = end_pfn;
2394}
2395
2396/* If necessary, push the node boundary out for reserve hotadd */
2397static void __init account_node_boundary(unsigned int nid,
2398 unsigned long *start_pfn, unsigned long *end_pfn)
2399{
2400 printk(KERN_DEBUG "Entering account_node_boundary(%u, %lu, %lu)\n",
2401 nid, *start_pfn, *end_pfn);
2402
2403 /* Return if boundary information has not been provided */
2404 if (node_boundary_end_pfn[nid] == 0)
2405 return;
2406
2407 /* Check the boundaries and update if necessary */
2408 if (node_boundary_start_pfn[nid] < *start_pfn)
2409 *start_pfn = node_boundary_start_pfn[nid];
2410 if (node_boundary_end_pfn[nid] > *end_pfn)
2411 *end_pfn = node_boundary_end_pfn[nid];
2412}
2413#else
2414void __init push_node_boundaries(unsigned int nid,
2415 unsigned long start_pfn, unsigned long end_pfn) {}
2416
2417static void __init account_node_boundary(unsigned int nid,
2418 unsigned long *start_pfn, unsigned long *end_pfn) {}
2419#endif
2420
2421
2422/**
Mel Gormanc7132162006-09-27 01:49:43 -07002423 * get_pfn_range_for_nid - Return the start and end page frames for a node
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002424 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
2425 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
2426 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
Mel Gormanc7132162006-09-27 01:49:43 -07002427 *
2428 * It returns the start and end page frame of a node based on information
2429 * provided by an arch calling add_active_range(). If called for a node
2430 * with no available memory, a warning is printed and the start and end
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002431 * PFNs will be 0.
Mel Gormanc7132162006-09-27 01:49:43 -07002432 */
2433void __init get_pfn_range_for_nid(unsigned int nid,
2434 unsigned long *start_pfn, unsigned long *end_pfn)
2435{
2436 int i;
2437 *start_pfn = -1UL;
2438 *end_pfn = 0;
2439
2440 for_each_active_range_index_in_nid(i, nid) {
2441 *start_pfn = min(*start_pfn, early_node_map[i].start_pfn);
2442 *end_pfn = max(*end_pfn, early_node_map[i].end_pfn);
2443 }
2444
2445 if (*start_pfn == -1UL) {
2446 printk(KERN_WARNING "Node %u active with no memory\n", nid);
2447 *start_pfn = 0;
2448 }
Mel Gormanfb014392006-09-27 01:49:59 -07002449
2450 /* Push the node boundaries out if requested */
2451 account_node_boundary(nid, start_pfn, end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07002452}
2453
2454/*
2455 * Return the number of pages a zone spans in a node, including holes
2456 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
2457 */
2458unsigned long __init zone_spanned_pages_in_node(int nid,
2459 unsigned long zone_type,
2460 unsigned long *ignored)
2461{
2462 unsigned long node_start_pfn, node_end_pfn;
2463 unsigned long zone_start_pfn, zone_end_pfn;
2464
2465 /* Get the start and end of the node and zone */
2466 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
2467 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
2468 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
2469
2470 /* Check that this node has pages within the zone's required range */
2471 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
2472 return 0;
2473
2474 /* Move the zone boundaries inside the node if necessary */
2475 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
2476 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
2477
2478 /* Return the spanned pages */
2479 return zone_end_pfn - zone_start_pfn;
2480}
2481
2482/*
2483 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002484 * then all holes in the requested range will be accounted for.
Mel Gormanc7132162006-09-27 01:49:43 -07002485 */
2486unsigned long __init __absent_pages_in_range(int nid,
2487 unsigned long range_start_pfn,
2488 unsigned long range_end_pfn)
2489{
2490 int i = 0;
2491 unsigned long prev_end_pfn = 0, hole_pages = 0;
2492 unsigned long start_pfn;
2493
2494 /* Find the end_pfn of the first active range of pfns in the node */
2495 i = first_active_region_index_in_nid(nid);
2496 if (i == -1)
2497 return 0;
2498
Mel Gorman9c7cd682006-09-27 01:49:58 -07002499 /* Account for ranges before physical memory on this node */
2500 if (early_node_map[i].start_pfn > range_start_pfn)
2501 hole_pages = early_node_map[i].start_pfn - range_start_pfn;
2502
Mel Gormanc7132162006-09-27 01:49:43 -07002503 prev_end_pfn = early_node_map[i].start_pfn;
2504
2505 /* Find all holes for the zone within the node */
2506 for (; i != -1; i = next_active_region_index_in_nid(i, nid)) {
2507
2508 /* No need to continue if prev_end_pfn is outside the zone */
2509 if (prev_end_pfn >= range_end_pfn)
2510 break;
2511
2512 /* Make sure the end of the zone is not within the hole */
2513 start_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
2514 prev_end_pfn = max(prev_end_pfn, range_start_pfn);
2515
2516 /* Update the hole size cound and move on */
2517 if (start_pfn > range_start_pfn) {
2518 BUG_ON(prev_end_pfn > start_pfn);
2519 hole_pages += start_pfn - prev_end_pfn;
2520 }
2521 prev_end_pfn = early_node_map[i].end_pfn;
2522 }
2523
Mel Gorman9c7cd682006-09-27 01:49:58 -07002524 /* Account for ranges past physical memory on this node */
2525 if (range_end_pfn > prev_end_pfn)
Mel Gorman0c6cb972006-10-28 10:38:59 -07002526 hole_pages += range_end_pfn -
Mel Gorman9c7cd682006-09-27 01:49:58 -07002527 max(range_start_pfn, prev_end_pfn);
2528
Mel Gormanc7132162006-09-27 01:49:43 -07002529 return hole_pages;
2530}
2531
2532/**
2533 * absent_pages_in_range - Return number of page frames in holes within a range
2534 * @start_pfn: The start PFN to start searching for holes
2535 * @end_pfn: The end PFN to stop searching for holes
2536 *
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002537 * It returns the number of pages frames in memory holes within a range.
Mel Gormanc7132162006-09-27 01:49:43 -07002538 */
2539unsigned long __init absent_pages_in_range(unsigned long start_pfn,
2540 unsigned long end_pfn)
2541{
2542 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
2543}
2544
2545/* Return the number of page frames in holes in a zone on a node */
2546unsigned long __init zone_absent_pages_in_node(int nid,
2547 unsigned long zone_type,
2548 unsigned long *ignored)
2549{
Mel Gorman9c7cd682006-09-27 01:49:58 -07002550 unsigned long node_start_pfn, node_end_pfn;
2551 unsigned long zone_start_pfn, zone_end_pfn;
2552
2553 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
2554 zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type],
2555 node_start_pfn);
2556 zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type],
2557 node_end_pfn);
2558
2559 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07002560}
Mel Gorman0e0b8642006-09-27 01:49:56 -07002561
Mel Gormanc7132162006-09-27 01:49:43 -07002562#else
2563static inline unsigned long zone_spanned_pages_in_node(int nid,
2564 unsigned long zone_type,
2565 unsigned long *zones_size)
2566{
2567 return zones_size[zone_type];
2568}
2569
2570static inline unsigned long zone_absent_pages_in_node(int nid,
2571 unsigned long zone_type,
2572 unsigned long *zholes_size)
2573{
2574 if (!zholes_size)
2575 return 0;
2576
2577 return zholes_size[zone_type];
2578}
Mel Gorman0e0b8642006-09-27 01:49:56 -07002579
Mel Gormanc7132162006-09-27 01:49:43 -07002580#endif
2581
2582static void __init calculate_node_totalpages(struct pglist_data *pgdat,
2583 unsigned long *zones_size, unsigned long *zholes_size)
2584{
2585 unsigned long realtotalpages, totalpages = 0;
2586 enum zone_type i;
2587
2588 for (i = 0; i < MAX_NR_ZONES; i++)
2589 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
2590 zones_size);
2591 pgdat->node_spanned_pages = totalpages;
2592
2593 realtotalpages = totalpages;
2594 for (i = 0; i < MAX_NR_ZONES; i++)
2595 realtotalpages -=
2596 zone_absent_pages_in_node(pgdat->node_id, i,
2597 zholes_size);
2598 pgdat->node_present_pages = realtotalpages;
2599 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
2600 realtotalpages);
2601}
2602
Linus Torvalds1da177e2005-04-16 15:20:36 -07002603/*
2604 * Set up the zone data structures:
2605 * - mark all pages reserved
2606 * - mark all memory queues empty
2607 * - clear the memory bitmaps
2608 */
Yasunori Goto86356ab2006-06-23 02:03:09 -07002609static void __meminit free_area_init_core(struct pglist_data *pgdat,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002610 unsigned long *zones_size, unsigned long *zholes_size)
2611{
Christoph Lameter2f1b6242006-09-25 23:31:13 -07002612 enum zone_type j;
Dave Hansened8ece22005-10-29 18:16:50 -07002613 int nid = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002614 unsigned long zone_start_pfn = pgdat->node_start_pfn;
Yasunori Goto718127c2006-06-23 02:03:10 -07002615 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002616
Dave Hansen208d54e2005-10-29 18:16:52 -07002617 pgdat_resize_init(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002618 pgdat->nr_zones = 0;
2619 init_waitqueue_head(&pgdat->kswapd_wait);
2620 pgdat->kswapd_max_order = 0;
2621
2622 for (j = 0; j < MAX_NR_ZONES; j++) {
2623 struct zone *zone = pgdat->node_zones + j;
Mel Gorman0e0b8642006-09-27 01:49:56 -07002624 unsigned long size, realsize, memmap_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002625
Mel Gormanc7132162006-09-27 01:49:43 -07002626 size = zone_spanned_pages_in_node(nid, j, zones_size);
2627 realsize = size - zone_absent_pages_in_node(nid, j,
2628 zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002629
Mel Gorman0e0b8642006-09-27 01:49:56 -07002630 /*
2631 * Adjust realsize so that it accounts for how much memory
2632 * is used by this zone for memmap. This affects the watermark
2633 * and per-cpu initialisations
2634 */
2635 memmap_pages = (size * sizeof(struct page)) >> PAGE_SHIFT;
2636 if (realsize >= memmap_pages) {
2637 realsize -= memmap_pages;
2638 printk(KERN_DEBUG
2639 " %s zone: %lu pages used for memmap\n",
2640 zone_names[j], memmap_pages);
2641 } else
2642 printk(KERN_WARNING
2643 " %s zone: %lu pages exceeds realsize %lu\n",
2644 zone_names[j], memmap_pages, realsize);
2645
Christoph Lameter62672762007-02-10 01:43:07 -08002646 /* Account for reserved pages */
2647 if (j == 0 && realsize > dma_reserve) {
Mel Gorman0e0b8642006-09-27 01:49:56 -07002648 realsize -= dma_reserve;
Christoph Lameter62672762007-02-10 01:43:07 -08002649 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
2650 zone_names[0], dma_reserve);
Mel Gorman0e0b8642006-09-27 01:49:56 -07002651 }
2652
Christoph Lameter98d2b0e2006-09-25 23:31:12 -07002653 if (!is_highmem_idx(j))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002654 nr_kernel_pages += realsize;
2655 nr_all_pages += realsize;
2656
2657 zone->spanned_pages = size;
2658 zone->present_pages = realsize;
Christoph Lameter96146342006-07-03 00:24:13 -07002659#ifdef CONFIG_NUMA
Christoph Lameterd5f541e2006-09-27 01:50:08 -07002660 zone->node = nid;
Christoph Lameter8417bba2006-09-25 23:31:51 -07002661 zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
Christoph Lameter96146342006-07-03 00:24:13 -07002662 / 100;
Christoph Lameter0ff38492006-09-25 23:31:52 -07002663 zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100;
Christoph Lameter96146342006-07-03 00:24:13 -07002664#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002665 zone->name = zone_names[j];
2666 spin_lock_init(&zone->lock);
2667 spin_lock_init(&zone->lru_lock);
Dave Hansenbdc8cb92005-10-29 18:16:53 -07002668 zone_seqlock_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002669 zone->zone_pgdat = pgdat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002670
Martin Bligh3bb1a852006-10-28 10:38:24 -07002671 zone->prev_priority = DEF_PRIORITY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002672
Dave Hansened8ece22005-10-29 18:16:50 -07002673 zone_pcp_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002674 INIT_LIST_HEAD(&zone->active_list);
2675 INIT_LIST_HEAD(&zone->inactive_list);
2676 zone->nr_scan_active = 0;
2677 zone->nr_scan_inactive = 0;
Christoph Lameter2244b952006-06-30 01:55:33 -07002678 zap_zone_vm_stats(zone);
Martin Hicks53e9a612005-09-03 15:54:51 -07002679 atomic_set(&zone->reclaim_in_progress, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002680 if (!size)
2681 continue;
2682
Dave Hansena2f3aa02007-01-10 23:15:30 -08002683 ret = init_currently_empty_zone(zone, zone_start_pfn,
2684 size, MEMMAP_EARLY);
Yasunori Goto718127c2006-06-23 02:03:10 -07002685 BUG_ON(ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002686 zone_start_pfn += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002687 }
2688}
2689
2690static void __init alloc_node_mem_map(struct pglist_data *pgdat)
2691{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002692 /* Skip empty nodes */
2693 if (!pgdat->node_spanned_pages)
2694 return;
2695
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002696#ifdef CONFIG_FLAT_NODE_MEM_MAP
Linus Torvalds1da177e2005-04-16 15:20:36 -07002697 /* ia64 gets its own node_mem_map, before this, without bootmem */
2698 if (!pgdat->node_mem_map) {
Bob Piccoe984bb42006-05-20 15:00:31 -07002699 unsigned long size, start, end;
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002700 struct page *map;
2701
Bob Piccoe984bb42006-05-20 15:00:31 -07002702 /*
2703 * The zone's endpoints aren't required to be MAX_ORDER
2704 * aligned but the node_mem_map endpoints must be in order
2705 * for the buddy allocator to function correctly.
2706 */
2707 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
2708 end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
2709 end = ALIGN(end, MAX_ORDER_NR_PAGES);
2710 size = (end - start) * sizeof(struct page);
Dave Hansen6f167ec2005-06-23 00:07:39 -07002711 map = alloc_remap(pgdat->node_id, size);
2712 if (!map)
2713 map = alloc_bootmem_node(pgdat, size);
Bob Piccoe984bb42006-05-20 15:00:31 -07002714 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002715 }
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002716#ifdef CONFIG_FLATMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002717 /*
2718 * With no DISCONTIG, the global mem_map is just set as node 0's
2719 */
Mel Gormanc7132162006-09-27 01:49:43 -07002720 if (pgdat == NODE_DATA(0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002721 mem_map = NODE_DATA(0)->node_mem_map;
Mel Gormanc7132162006-09-27 01:49:43 -07002722#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2723 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
2724 mem_map -= pgdat->node_start_pfn;
2725#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
2726 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002727#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002728#endif /* CONFIG_FLAT_NODE_MEM_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002729}
2730
Yasunori Goto86356ab2006-06-23 02:03:09 -07002731void __meminit free_area_init_node(int nid, struct pglist_data *pgdat,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002732 unsigned long *zones_size, unsigned long node_start_pfn,
2733 unsigned long *zholes_size)
2734{
2735 pgdat->node_id = nid;
2736 pgdat->node_start_pfn = node_start_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07002737 calculate_node_totalpages(pgdat, zones_size, zholes_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002738
2739 alloc_node_mem_map(pgdat);
2740
2741 free_area_init_core(pgdat, zones_size, zholes_size);
2742}
2743
Mel Gormanc7132162006-09-27 01:49:43 -07002744#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2745/**
2746 * add_active_range - Register a range of PFNs backed by physical memory
2747 * @nid: The node ID the range resides on
2748 * @start_pfn: The start PFN of the available physical memory
2749 * @end_pfn: The end PFN of the available physical memory
2750 *
2751 * These ranges are stored in an early_node_map[] and later used by
2752 * free_area_init_nodes() to calculate zone sizes and holes. If the
2753 * range spans a memory hole, it is up to the architecture to ensure
2754 * the memory is not freed by the bootmem allocator. If possible
2755 * the range being registered will be merged with existing ranges.
2756 */
2757void __init add_active_range(unsigned int nid, unsigned long start_pfn,
2758 unsigned long end_pfn)
2759{
2760 int i;
2761
2762 printk(KERN_DEBUG "Entering add_active_range(%d, %lu, %lu) "
2763 "%d entries of %d used\n",
2764 nid, start_pfn, end_pfn,
2765 nr_nodemap_entries, MAX_ACTIVE_REGIONS);
2766
2767 /* Merge with existing active regions if possible */
2768 for (i = 0; i < nr_nodemap_entries; i++) {
2769 if (early_node_map[i].nid != nid)
2770 continue;
2771
2772 /* Skip if an existing region covers this new one */
2773 if (start_pfn >= early_node_map[i].start_pfn &&
2774 end_pfn <= early_node_map[i].end_pfn)
2775 return;
2776
2777 /* Merge forward if suitable */
2778 if (start_pfn <= early_node_map[i].end_pfn &&
2779 end_pfn > early_node_map[i].end_pfn) {
2780 early_node_map[i].end_pfn = end_pfn;
2781 return;
2782 }
2783
2784 /* Merge backward if suitable */
2785 if (start_pfn < early_node_map[i].end_pfn &&
2786 end_pfn >= early_node_map[i].start_pfn) {
2787 early_node_map[i].start_pfn = start_pfn;
2788 return;
2789 }
2790 }
2791
2792 /* Check that early_node_map is large enough */
2793 if (i >= MAX_ACTIVE_REGIONS) {
2794 printk(KERN_CRIT "More than %d memory regions, truncating\n",
2795 MAX_ACTIVE_REGIONS);
2796 return;
2797 }
2798
2799 early_node_map[i].nid = nid;
2800 early_node_map[i].start_pfn = start_pfn;
2801 early_node_map[i].end_pfn = end_pfn;
2802 nr_nodemap_entries = i + 1;
2803}
2804
2805/**
2806 * shrink_active_range - Shrink an existing registered range of PFNs
2807 * @nid: The node id the range is on that should be shrunk
2808 * @old_end_pfn: The old end PFN of the range
2809 * @new_end_pfn: The new PFN of the range
2810 *
2811 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
2812 * The map is kept at the end physical page range that has already been
2813 * registered with add_active_range(). This function allows an arch to shrink
2814 * an existing registered range.
2815 */
2816void __init shrink_active_range(unsigned int nid, unsigned long old_end_pfn,
2817 unsigned long new_end_pfn)
2818{
2819 int i;
2820
2821 /* Find the old active region end and shrink */
2822 for_each_active_range_index_in_nid(i, nid)
2823 if (early_node_map[i].end_pfn == old_end_pfn) {
2824 early_node_map[i].end_pfn = new_end_pfn;
2825 break;
2826 }
2827}
2828
2829/**
2830 * remove_all_active_ranges - Remove all currently registered regions
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002831 *
Mel Gormanc7132162006-09-27 01:49:43 -07002832 * During discovery, it may be found that a table like SRAT is invalid
2833 * and an alternative discovery method must be used. This function removes
2834 * all currently registered regions.
2835 */
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002836void __init remove_all_active_ranges(void)
Mel Gormanc7132162006-09-27 01:49:43 -07002837{
2838 memset(early_node_map, 0, sizeof(early_node_map));
2839 nr_nodemap_entries = 0;
Mel Gormanfb014392006-09-27 01:49:59 -07002840#ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
2841 memset(node_boundary_start_pfn, 0, sizeof(node_boundary_start_pfn));
2842 memset(node_boundary_end_pfn, 0, sizeof(node_boundary_end_pfn));
2843#endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
Mel Gormanc7132162006-09-27 01:49:43 -07002844}
2845
2846/* Compare two active node_active_regions */
2847static int __init cmp_node_active_region(const void *a, const void *b)
2848{
2849 struct node_active_region *arange = (struct node_active_region *)a;
2850 struct node_active_region *brange = (struct node_active_region *)b;
2851
2852 /* Done this way to avoid overflows */
2853 if (arange->start_pfn > brange->start_pfn)
2854 return 1;
2855 if (arange->start_pfn < brange->start_pfn)
2856 return -1;
2857
2858 return 0;
2859}
2860
2861/* sort the node_map by start_pfn */
2862static void __init sort_node_map(void)
2863{
2864 sort(early_node_map, (size_t)nr_nodemap_entries,
2865 sizeof(struct node_active_region),
2866 cmp_node_active_region, NULL);
2867}
2868
Mel Gormana6af2bc2007-02-10 01:42:57 -08002869/* Find the lowest pfn for a node */
Mel Gormanc7132162006-09-27 01:49:43 -07002870unsigned long __init find_min_pfn_for_node(unsigned long nid)
2871{
2872 int i;
Mel Gormana6af2bc2007-02-10 01:42:57 -08002873 unsigned long min_pfn = ULONG_MAX;
Mel Gorman1abbfb42006-11-23 12:01:41 +00002874
Mel Gormanc7132162006-09-27 01:49:43 -07002875 /* Assuming a sorted map, the first range found has the starting pfn */
2876 for_each_active_range_index_in_nid(i, nid)
Mel Gormana6af2bc2007-02-10 01:42:57 -08002877 min_pfn = min(min_pfn, early_node_map[i].start_pfn);
Mel Gormanc7132162006-09-27 01:49:43 -07002878
Mel Gormana6af2bc2007-02-10 01:42:57 -08002879 if (min_pfn == ULONG_MAX) {
2880 printk(KERN_WARNING
2881 "Could not find start_pfn for node %lu\n", nid);
2882 return 0;
2883 }
2884
2885 return min_pfn;
Mel Gormanc7132162006-09-27 01:49:43 -07002886}
2887
2888/**
2889 * find_min_pfn_with_active_regions - Find the minimum PFN registered
2890 *
2891 * It returns the minimum PFN based on information provided via
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002892 * add_active_range().
Mel Gormanc7132162006-09-27 01:49:43 -07002893 */
2894unsigned long __init find_min_pfn_with_active_regions(void)
2895{
2896 return find_min_pfn_for_node(MAX_NUMNODES);
2897}
2898
2899/**
2900 * find_max_pfn_with_active_regions - Find the maximum PFN registered
2901 *
2902 * It returns the maximum PFN based on information provided via
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002903 * add_active_range().
Mel Gormanc7132162006-09-27 01:49:43 -07002904 */
2905unsigned long __init find_max_pfn_with_active_regions(void)
2906{
2907 int i;
2908 unsigned long max_pfn = 0;
2909
2910 for (i = 0; i < nr_nodemap_entries; i++)
2911 max_pfn = max(max_pfn, early_node_map[i].end_pfn);
2912
2913 return max_pfn;
2914}
2915
2916/**
2917 * free_area_init_nodes - Initialise all pg_data_t and zone data
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002918 * @max_zone_pfn: an array of max PFNs for each zone
Mel Gormanc7132162006-09-27 01:49:43 -07002919 *
2920 * This will call free_area_init_node() for each active node in the system.
2921 * Using the page ranges provided by add_active_range(), the size of each
2922 * zone in each node and their holes is calculated. If the maximum PFN
2923 * between two adjacent zones match, it is assumed that the zone is empty.
2924 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
2925 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
2926 * starts where the previous one ended. For example, ZONE_DMA32 starts
2927 * at arch_max_dma_pfn.
2928 */
2929void __init free_area_init_nodes(unsigned long *max_zone_pfn)
2930{
2931 unsigned long nid;
2932 enum zone_type i;
2933
Mel Gormana6af2bc2007-02-10 01:42:57 -08002934 /* Sort early_node_map as initialisation assumes it is sorted */
2935 sort_node_map();
2936
Mel Gormanc7132162006-09-27 01:49:43 -07002937 /* Record where the zone boundaries are */
2938 memset(arch_zone_lowest_possible_pfn, 0,
2939 sizeof(arch_zone_lowest_possible_pfn));
2940 memset(arch_zone_highest_possible_pfn, 0,
2941 sizeof(arch_zone_highest_possible_pfn));
2942 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
2943 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
2944 for (i = 1; i < MAX_NR_ZONES; i++) {
2945 arch_zone_lowest_possible_pfn[i] =
2946 arch_zone_highest_possible_pfn[i-1];
2947 arch_zone_highest_possible_pfn[i] =
2948 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
2949 }
2950
Mel Gormanc7132162006-09-27 01:49:43 -07002951 /* Print out the zone ranges */
2952 printk("Zone PFN ranges:\n");
2953 for (i = 0; i < MAX_NR_ZONES; i++)
2954 printk(" %-8s %8lu -> %8lu\n",
2955 zone_names[i],
2956 arch_zone_lowest_possible_pfn[i],
2957 arch_zone_highest_possible_pfn[i]);
2958
2959 /* Print out the early_node_map[] */
2960 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries);
2961 for (i = 0; i < nr_nodemap_entries; i++)
2962 printk(" %3d: %8lu -> %8lu\n", early_node_map[i].nid,
2963 early_node_map[i].start_pfn,
2964 early_node_map[i].end_pfn);
2965
2966 /* Initialise every node */
Christoph Lameter8ef82862007-02-20 13:57:52 -08002967 setup_nr_node_ids();
Mel Gormanc7132162006-09-27 01:49:43 -07002968 for_each_online_node(nid) {
2969 pg_data_t *pgdat = NODE_DATA(nid);
2970 free_area_init_node(nid, pgdat, NULL,
2971 find_min_pfn_for_node(nid), NULL);
2972 }
2973}
2974#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
2975
Mel Gorman0e0b8642006-09-27 01:49:56 -07002976/**
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002977 * set_dma_reserve - set the specified number of pages reserved in the first zone
2978 * @new_dma_reserve: The number of pages to mark reserved
Mel Gorman0e0b8642006-09-27 01:49:56 -07002979 *
2980 * The per-cpu batchsize and zone watermarks are determined by present_pages.
2981 * In the DMA zone, a significant percentage may be consumed by kernel image
2982 * and other unfreeable allocations which can skew the watermarks badly. This
Randy Dunlap88ca3b92006-10-04 02:15:25 -07002983 * function may optionally be used to account for unfreeable pages in the
2984 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
2985 * smaller per-cpu batchsize.
Mel Gorman0e0b8642006-09-27 01:49:56 -07002986 */
2987void __init set_dma_reserve(unsigned long new_dma_reserve)
2988{
2989 dma_reserve = new_dma_reserve;
2990}
2991
Dave Hansen93b75042005-06-23 00:07:47 -07002992#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07002993static bootmem_data_t contig_bootmem_data;
2994struct pglist_data contig_page_data = { .bdata = &contig_bootmem_data };
2995
2996EXPORT_SYMBOL(contig_page_data);
Dave Hansen93b75042005-06-23 00:07:47 -07002997#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002998
2999void __init free_area_init(unsigned long *zones_size)
3000{
Dave Hansen93b75042005-06-23 00:07:47 -07003001 free_area_init_node(0, NODE_DATA(0), zones_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003002 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
3003}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003004
Linus Torvalds1da177e2005-04-16 15:20:36 -07003005static int page_alloc_cpu_notify(struct notifier_block *self,
3006 unsigned long action, void *hcpu)
3007{
3008 int cpu = (unsigned long)hcpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003009
3010 if (action == CPU_DEAD) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003011 local_irq_disable();
3012 __drain_pages(cpu);
Christoph Lameterf8891e52006-06-30 01:55:45 -07003013 vm_events_fold_cpu(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003014 local_irq_enable();
Christoph Lameter2244b952006-06-30 01:55:33 -07003015 refresh_cpu_vm_stats(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003016 }
3017 return NOTIFY_OK;
3018}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003019
3020void __init page_alloc_init(void)
3021{
3022 hotcpu_notifier(page_alloc_cpu_notify, 0);
3023}
3024
3025/*
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07003026 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
3027 * or min_free_kbytes changes.
3028 */
3029static void calculate_totalreserve_pages(void)
3030{
3031 struct pglist_data *pgdat;
3032 unsigned long reserve_pages = 0;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07003033 enum zone_type i, j;
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07003034
3035 for_each_online_pgdat(pgdat) {
3036 for (i = 0; i < MAX_NR_ZONES; i++) {
3037 struct zone *zone = pgdat->node_zones + i;
3038 unsigned long max = 0;
3039
3040 /* Find valid and maximum lowmem_reserve in the zone */
3041 for (j = i; j < MAX_NR_ZONES; j++) {
3042 if (zone->lowmem_reserve[j] > max)
3043 max = zone->lowmem_reserve[j];
3044 }
3045
3046 /* we treat pages_high as reserved pages. */
3047 max += zone->pages_high;
3048
3049 if (max > zone->present_pages)
3050 max = zone->present_pages;
3051 reserve_pages += max;
3052 }
3053 }
3054 totalreserve_pages = reserve_pages;
3055}
3056
3057/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003058 * setup_per_zone_lowmem_reserve - called whenever
3059 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
3060 * has a correct pages reserved value, so an adequate number of
3061 * pages are left in the zone after a successful __alloc_pages().
3062 */
3063static void setup_per_zone_lowmem_reserve(void)
3064{
3065 struct pglist_data *pgdat;
Christoph Lameter2f6726e2006-09-25 23:31:18 -07003066 enum zone_type j, idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003067
KAMEZAWA Hiroyukiec936fc2006-03-27 01:15:59 -08003068 for_each_online_pgdat(pgdat) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003069 for (j = 0; j < MAX_NR_ZONES; j++) {
3070 struct zone *zone = pgdat->node_zones + j;
3071 unsigned long present_pages = zone->present_pages;
3072
3073 zone->lowmem_reserve[j] = 0;
3074
Christoph Lameter2f6726e2006-09-25 23:31:18 -07003075 idx = j;
3076 while (idx) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003077 struct zone *lower_zone;
3078
Christoph Lameter2f6726e2006-09-25 23:31:18 -07003079 idx--;
3080
Linus Torvalds1da177e2005-04-16 15:20:36 -07003081 if (sysctl_lowmem_reserve_ratio[idx] < 1)
3082 sysctl_lowmem_reserve_ratio[idx] = 1;
3083
3084 lower_zone = pgdat->node_zones + idx;
3085 lower_zone->lowmem_reserve[j] = present_pages /
3086 sysctl_lowmem_reserve_ratio[idx];
3087 present_pages += lower_zone->present_pages;
3088 }
3089 }
3090 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07003091
3092 /* update totalreserve_pages */
3093 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003094}
3095
Randy Dunlap88ca3b92006-10-04 02:15:25 -07003096/**
3097 * setup_per_zone_pages_min - called when min_free_kbytes changes.
3098 *
3099 * Ensures that the pages_{min,low,high} values for each zone are set correctly
3100 * with respect to min_free_kbytes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003101 */
Dave Hansen3947be12005-10-29 18:16:54 -07003102void setup_per_zone_pages_min(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003103{
3104 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
3105 unsigned long lowmem_pages = 0;
3106 struct zone *zone;
3107 unsigned long flags;
3108
3109 /* Calculate total number of !ZONE_HIGHMEM pages */
3110 for_each_zone(zone) {
3111 if (!is_highmem(zone))
3112 lowmem_pages += zone->present_pages;
3113 }
3114
3115 for_each_zone(zone) {
Andrew Mortonac924c62006-05-15 09:43:59 -07003116 u64 tmp;
3117
Linus Torvalds1da177e2005-04-16 15:20:36 -07003118 spin_lock_irqsave(&zone->lru_lock, flags);
Andrew Mortonac924c62006-05-15 09:43:59 -07003119 tmp = (u64)pages_min * zone->present_pages;
3120 do_div(tmp, lowmem_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003121 if (is_highmem(zone)) {
3122 /*
Nick Piggin669ed172005-11-13 16:06:45 -08003123 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
3124 * need highmem pages, so cap pages_min to a small
3125 * value here.
3126 *
3127 * The (pages_high-pages_low) and (pages_low-pages_min)
3128 * deltas controls asynch page reclaim, and so should
3129 * not be capped for highmem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003130 */
3131 int min_pages;
3132
3133 min_pages = zone->present_pages / 1024;
3134 if (min_pages < SWAP_CLUSTER_MAX)
3135 min_pages = SWAP_CLUSTER_MAX;
3136 if (min_pages > 128)
3137 min_pages = 128;
3138 zone->pages_min = min_pages;
3139 } else {
Nick Piggin669ed172005-11-13 16:06:45 -08003140 /*
3141 * If it's a lowmem zone, reserve a number of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07003142 * proportionate to the zone's size.
3143 */
Nick Piggin669ed172005-11-13 16:06:45 -08003144 zone->pages_min = tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003145 }
3146
Andrew Mortonac924c62006-05-15 09:43:59 -07003147 zone->pages_low = zone->pages_min + (tmp >> 2);
3148 zone->pages_high = zone->pages_min + (tmp >> 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003149 spin_unlock_irqrestore(&zone->lru_lock, flags);
3150 }
Hideo AOKIcb45b0e2006-04-10 22:52:59 -07003151
3152 /* update totalreserve_pages */
3153 calculate_totalreserve_pages();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003154}
3155
3156/*
3157 * Initialise min_free_kbytes.
3158 *
3159 * For small machines we want it small (128k min). For large machines
3160 * we want it large (64MB max). But it is not linear, because network
3161 * bandwidth does not increase linearly with machine size. We use
3162 *
3163 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
3164 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
3165 *
3166 * which yields
3167 *
3168 * 16MB: 512k
3169 * 32MB: 724k
3170 * 64MB: 1024k
3171 * 128MB: 1448k
3172 * 256MB: 2048k
3173 * 512MB: 2896k
3174 * 1024MB: 4096k
3175 * 2048MB: 5792k
3176 * 4096MB: 8192k
3177 * 8192MB: 11584k
3178 * 16384MB: 16384k
3179 */
3180static int __init init_per_zone_pages_min(void)
3181{
3182 unsigned long lowmem_kbytes;
3183
3184 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
3185
3186 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
3187 if (min_free_kbytes < 128)
3188 min_free_kbytes = 128;
3189 if (min_free_kbytes > 65536)
3190 min_free_kbytes = 65536;
3191 setup_per_zone_pages_min();
3192 setup_per_zone_lowmem_reserve();
3193 return 0;
3194}
3195module_init(init_per_zone_pages_min)
3196
3197/*
3198 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
3199 * that we can call two helper functions whenever min_free_kbytes
3200 * changes.
3201 */
3202int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
3203 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
3204{
3205 proc_dointvec(table, write, file, buffer, length, ppos);
3206 setup_per_zone_pages_min();
3207 return 0;
3208}
3209
Christoph Lameter96146342006-07-03 00:24:13 -07003210#ifdef CONFIG_NUMA
3211int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
3212 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
3213{
3214 struct zone *zone;
3215 int rc;
3216
3217 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
3218 if (rc)
3219 return rc;
3220
3221 for_each_zone(zone)
Christoph Lameter8417bba2006-09-25 23:31:51 -07003222 zone->min_unmapped_pages = (zone->present_pages *
Christoph Lameter96146342006-07-03 00:24:13 -07003223 sysctl_min_unmapped_ratio) / 100;
3224 return 0;
3225}
Christoph Lameter0ff38492006-09-25 23:31:52 -07003226
3227int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
3228 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
3229{
3230 struct zone *zone;
3231 int rc;
3232
3233 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
3234 if (rc)
3235 return rc;
3236
3237 for_each_zone(zone)
3238 zone->min_slab_pages = (zone->present_pages *
3239 sysctl_min_slab_ratio) / 100;
3240 return 0;
3241}
Christoph Lameter96146342006-07-03 00:24:13 -07003242#endif
3243
Linus Torvalds1da177e2005-04-16 15:20:36 -07003244/*
3245 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
3246 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
3247 * whenever sysctl_lowmem_reserve_ratio changes.
3248 *
3249 * The reserve ratio obviously has absolutely no relation with the
3250 * pages_min watermarks. The lowmem reserve ratio can only make sense
3251 * if in function of the boot time zone sizes.
3252 */
3253int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
3254 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
3255{
3256 proc_dointvec_minmax(table, write, file, buffer, length, ppos);
3257 setup_per_zone_lowmem_reserve();
3258 return 0;
3259}
3260
Rohit Seth8ad4b1f2006-01-08 01:00:40 -08003261/*
3262 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
3263 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
3264 * can have before it gets flushed back to buddy allocator.
3265 */
3266
3267int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
3268 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
3269{
3270 struct zone *zone;
3271 unsigned int cpu;
3272 int ret;
3273
3274 ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
3275 if (!write || (ret == -EINVAL))
3276 return ret;
3277 for_each_zone(zone) {
3278 for_each_online_cpu(cpu) {
3279 unsigned long high;
3280 high = zone->present_pages / percpu_pagelist_fraction;
3281 setup_pagelist_highmark(zone_pcp(zone, cpu), high);
3282 }
3283 }
3284 return 0;
3285}
3286
David S. Millerf034b5d2006-08-24 03:08:07 -07003287int hashdist = HASHDIST_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003288
3289#ifdef CONFIG_NUMA
3290static int __init set_hashdist(char *str)
3291{
3292 if (!str)
3293 return 0;
3294 hashdist = simple_strtoul(str, &str, 0);
3295 return 1;
3296}
3297__setup("hashdist=", set_hashdist);
3298#endif
3299
3300/*
3301 * allocate a large system hash table from bootmem
3302 * - it is assumed that the hash table must contain an exact power-of-2
3303 * quantity of entries
3304 * - limit is the number of hash buckets, not the total allocation size
3305 */
3306void *__init alloc_large_system_hash(const char *tablename,
3307 unsigned long bucketsize,
3308 unsigned long numentries,
3309 int scale,
3310 int flags,
3311 unsigned int *_hash_shift,
3312 unsigned int *_hash_mask,
3313 unsigned long limit)
3314{
3315 unsigned long long max = limit;
3316 unsigned long log2qty, size;
3317 void *table = NULL;
3318
3319 /* allow the kernel cmdline to have a say */
3320 if (!numentries) {
3321 /* round applicable memory size up to nearest megabyte */
Andrew Morton04903662006-12-06 20:37:33 -08003322 numentries = nr_kernel_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003323 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
3324 numentries >>= 20 - PAGE_SHIFT;
3325 numentries <<= 20 - PAGE_SHIFT;
3326
3327 /* limit to 1 bucket per 2^scale bytes of low memory */
3328 if (scale > PAGE_SHIFT)
3329 numentries >>= (scale - PAGE_SHIFT);
3330 else
3331 numentries <<= (PAGE_SHIFT - scale);
Paul Mundt9ab37b82007-01-05 16:36:30 -08003332
3333 /* Make sure we've got at least a 0-order allocation.. */
3334 if (unlikely((numentries * bucketsize) < PAGE_SIZE))
3335 numentries = PAGE_SIZE / bucketsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003336 }
John Hawkes6e692ed2006-03-25 03:08:02 -08003337 numentries = roundup_pow_of_two(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003338
3339 /* limit allocation size to 1/16 total memory by default */
3340 if (max == 0) {
3341 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
3342 do_div(max, bucketsize);
3343 }
3344
3345 if (numentries > max)
3346 numentries = max;
3347
David Howellsf0d1b0b2006-12-08 02:37:49 -08003348 log2qty = ilog2(numentries);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003349
3350 do {
3351 size = bucketsize << log2qty;
3352 if (flags & HASH_EARLY)
3353 table = alloc_bootmem(size);
3354 else if (hashdist)
3355 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
3356 else {
3357 unsigned long order;
3358 for (order = 0; ((1UL << order) << PAGE_SHIFT) < size; order++)
3359 ;
3360 table = (void*) __get_free_pages(GFP_ATOMIC, order);
3361 }
3362 } while (!table && size > PAGE_SIZE && --log2qty);
3363
3364 if (!table)
3365 panic("Failed to allocate %s hash table\n", tablename);
3366
3367 printk("%s hash table entries: %d (order: %d, %lu bytes)\n",
3368 tablename,
3369 (1U << log2qty),
David Howellsf0d1b0b2006-12-08 02:37:49 -08003370 ilog2(size) - PAGE_SHIFT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003371 size);
3372
3373 if (_hash_shift)
3374 *_hash_shift = log2qty;
3375 if (_hash_mask)
3376 *_hash_mask = (1 << log2qty) - 1;
3377
3378 return table;
3379}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08003380
3381#ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08003382struct page *pfn_to_page(unsigned long pfn)
3383{
Andy Whitcroft67de6482006-06-23 02:03:12 -07003384 return __pfn_to_page(pfn);
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08003385}
3386unsigned long page_to_pfn(struct page *page)
3387{
Andy Whitcroft67de6482006-06-23 02:03:12 -07003388 return __page_to_pfn(page);
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08003389}
KAMEZAWA Hiroyukia117e662006-03-27 01:15:25 -08003390EXPORT_SYMBOL(pfn_to_page);
3391EXPORT_SYMBOL(page_to_pfn);
3392#endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */
Andrew Morton6220ec72006-10-19 23:29:05 -07003393
Andrew Morton6220ec72006-10-19 23:29:05 -07003394