blob: fe14a8c87fc27d8822faeaff71cf990f6f9a95ef [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/mm/page_alloc.c
3 *
4 * Manages the free list, the system allocates free pages here.
5 * Note that kmalloc() lives in slab.c
6 *
7 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
8 * Swap reorganised 29.12.95, Stephen Tweedie
9 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
10 * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999
11 * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
12 * Zone balancing, Kanoj Sarcar, SGI, Jan 2000
13 * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002
14 * (lots of bits borrowed from Ingo Molnar & Andrew Morton)
15 */
16
17#include <linux/config.h>
18#include <linux/stddef.h>
19#include <linux/mm.h>
20#include <linux/swap.h>
21#include <linux/interrupt.h>
22#include <linux/pagemap.h>
23#include <linux/bootmem.h>
24#include <linux/compiler.h>
Randy Dunlap9f158332005-09-13 01:25:16 -070025#include <linux/kernel.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include <linux/module.h>
27#include <linux/suspend.h>
28#include <linux/pagevec.h>
29#include <linux/blkdev.h>
30#include <linux/slab.h>
31#include <linux/notifier.h>
32#include <linux/topology.h>
33#include <linux/sysctl.h>
34#include <linux/cpu.h>
35#include <linux/cpuset.h>
Dave Hansenbdc8cb92005-10-29 18:16:53 -070036#include <linux/memory_hotplug.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070037#include <linux/nodemask.h>
38#include <linux/vmalloc.h>
39
40#include <asm/tlbflush.h>
41#include "internal.h"
42
43/*
44 * MCD - HACK: Find somewhere to initialize this EARLY, or make this
45 * initializer cleaner
46 */
Christoph Lameterc3d8c142005-09-06 15:16:33 -070047nodemask_t node_online_map __read_mostly = { { [0] = 1UL } };
Dean Nelson7223a932005-03-23 19:00:00 -070048EXPORT_SYMBOL(node_online_map);
Christoph Lameterc3d8c142005-09-06 15:16:33 -070049nodemask_t node_possible_map __read_mostly = NODE_MASK_ALL;
Dean Nelson7223a932005-03-23 19:00:00 -070050EXPORT_SYMBOL(node_possible_map);
Christoph Lameterc3d8c142005-09-06 15:16:33 -070051struct pglist_data *pgdat_list __read_mostly;
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -070052unsigned long totalram_pages __read_mostly;
53unsigned long totalhigh_pages __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070054long nr_swap_pages;
55
56/*
57 * results with 256, 32 in the lowmem_reserve sysctl:
58 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
59 * 1G machine -> (16M dma, 784M normal, 224M high)
60 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
61 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
62 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
Andi Kleena2f1b422005-11-05 17:25:53 +010063 *
64 * TBD: should special case ZONE_DMA32 machines here - in those we normally
65 * don't need any ZONE_NORMAL reservation
Linus Torvalds1da177e2005-04-16 15:20:36 -070066 */
Andi Kleena2f1b422005-11-05 17:25:53 +010067int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = { 256, 256, 32 };
Linus Torvalds1da177e2005-04-16 15:20:36 -070068
69EXPORT_SYMBOL(totalram_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -070070
71/*
72 * Used by page_zone() to look up the address of the struct zone whose
73 * id is encoded in the upper bits of page->flags
74 */
Christoph Lameterc3d8c142005-09-06 15:16:33 -070075struct zone *zone_table[1 << ZONETABLE_SHIFT] __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070076EXPORT_SYMBOL(zone_table);
77
Andi Kleena2f1b422005-11-05 17:25:53 +010078static char *zone_names[MAX_NR_ZONES] = { "DMA", "DMA32", "Normal", "HighMem" };
Linus Torvalds1da177e2005-04-16 15:20:36 -070079int min_free_kbytes = 1024;
80
81unsigned long __initdata nr_kernel_pages;
82unsigned long __initdata nr_all_pages;
83
Dave Hansenc6a57e12005-10-29 18:16:52 -070084static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -070085{
Dave Hansenbdc8cb92005-10-29 18:16:53 -070086 int ret = 0;
87 unsigned seq;
88 unsigned long pfn = page_to_pfn(page);
Dave Hansenc6a57e12005-10-29 18:16:52 -070089
Dave Hansenbdc8cb92005-10-29 18:16:53 -070090 do {
91 seq = zone_span_seqbegin(zone);
92 if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
93 ret = 1;
94 else if (pfn < zone->zone_start_pfn)
95 ret = 1;
96 } while (zone_span_seqretry(zone, seq));
97
98 return ret;
Dave Hansenc6a57e12005-10-29 18:16:52 -070099}
100
101static int page_is_consistent(struct zone *zone, struct page *page)
102{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103#ifdef CONFIG_HOLES_IN_ZONE
104 if (!pfn_valid(page_to_pfn(page)))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700105 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106#endif
107 if (zone != page_zone(page))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700108 return 0;
109
110 return 1;
111}
112/*
113 * Temporary debugging check for pages not lying within a given zone.
114 */
115static int bad_range(struct zone *zone, struct page *page)
116{
117 if (page_outside_zone_boundaries(zone, page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118 return 1;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700119 if (!page_is_consistent(zone, page))
120 return 1;
121
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122 return 0;
123}
124
125static void bad_page(const char *function, struct page *page)
126{
127 printk(KERN_EMERG "Bad page state at %s (in process '%s', page %p)\n",
128 function, current->comm, page);
129 printk(KERN_EMERG "flags:0x%0*lx mapping:%p mapcount:%d count:%d\n",
Andi Kleen07808b72005-11-05 17:25:53 +0100130 (int)(2*sizeof(unsigned long)), (unsigned long)page->flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131 page->mapping, page_mapcount(page), page_count(page));
132 printk(KERN_EMERG "Backtrace:\n");
133 dump_stack();
134 printk(KERN_EMERG "Trying to fix it up, but a reboot is needed\n");
Hugh Dickins334795e2005-06-21 17:15:08 -0700135 page->flags &= ~(1 << PG_lru |
136 1 << PG_private |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137 1 << PG_locked |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138 1 << PG_active |
139 1 << PG_dirty |
Hugh Dickins334795e2005-06-21 17:15:08 -0700140 1 << PG_reclaim |
141 1 << PG_slab |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142 1 << PG_swapcache |
Hugh Dickins689bceb2005-11-21 21:32:20 -0800143 1 << PG_writeback );
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144 set_page_count(page, 0);
145 reset_page_mapcount(page);
146 page->mapping = NULL;
Randy Dunlap9f158332005-09-13 01:25:16 -0700147 add_taint(TAINT_BAD_PAGE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148}
149
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150/*
151 * Higher-order pages are called "compound pages". They are structured thusly:
152 *
153 * The first PAGE_SIZE page is called the "head page".
154 *
155 * The remaining PAGE_SIZE pages are called "tail pages".
156 *
157 * All pages have PG_compound set. All pages have their ->private pointing at
158 * the head page (even the head page has this).
159 *
160 * The first tail page's ->mapping, if non-zero, holds the address of the
161 * compound page's put_page() function.
162 *
163 * The order of the allocation is stored in the first tail page's ->index
164 * This is only for debug at present. This usage means that zero-order pages
165 * may not be compound.
166 */
167static void prep_compound_page(struct page *page, unsigned long order)
168{
169 int i;
170 int nr_pages = 1 << order;
171
172 page[1].mapping = NULL;
173 page[1].index = order;
174 for (i = 0; i < nr_pages; i++) {
175 struct page *p = page + i;
176
177 SetPageCompound(p);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700178 set_page_private(p, (unsigned long)page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179 }
180}
181
182static void destroy_compound_page(struct page *page, unsigned long order)
183{
184 int i;
185 int nr_pages = 1 << order;
186
187 if (!PageCompound(page))
188 return;
189
190 if (page[1].index != order)
191 bad_page(__FUNCTION__, page);
192
193 for (i = 0; i < nr_pages; i++) {
194 struct page *p = page + i;
195
196 if (!PageCompound(p))
197 bad_page(__FUNCTION__, page);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700198 if (page_private(p) != (unsigned long)page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199 bad_page(__FUNCTION__, page);
200 ClearPageCompound(p);
201 }
202}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203
204/*
205 * function for dealing with page's order in buddy system.
206 * zone->lock is already acquired when we use these.
207 * So, we don't need atomic page->flags operations here.
208 */
209static inline unsigned long page_order(struct page *page) {
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700210 return page_private(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211}
212
213static inline void set_page_order(struct page *page, int order) {
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700214 set_page_private(page, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 __SetPagePrivate(page);
216}
217
218static inline void rmv_page_order(struct page *page)
219{
220 __ClearPagePrivate(page);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700221 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222}
223
224/*
225 * Locate the struct page for both the matching buddy in our
226 * pair (buddy1) and the combined O(n+1) page they form (page).
227 *
228 * 1) Any buddy B1 will have an order O twin B2 which satisfies
229 * the following equation:
230 * B2 = B1 ^ (1 << O)
231 * For example, if the starting buddy (buddy2) is #8 its order
232 * 1 buddy is #10:
233 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
234 *
235 * 2) Any buddy B will have an order O+1 parent P which
236 * satisfies the following equation:
237 * P = B & ~(1 << O)
238 *
239 * Assumption: *_mem_map is contigious at least up to MAX_ORDER
240 */
241static inline struct page *
242__page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order)
243{
244 unsigned long buddy_idx = page_idx ^ (1 << order);
245
246 return page + (buddy_idx - page_idx);
247}
248
249static inline unsigned long
250__find_combined_index(unsigned long page_idx, unsigned int order)
251{
252 return (page_idx & ~(1 << order));
253}
254
255/*
256 * This function checks whether a page is free && is the buddy
257 * we can do coalesce a page and its buddy if
258 * (a) the buddy is free &&
259 * (b) the buddy is on the buddy system &&
260 * (c) a page and its buddy have the same order.
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700261 * for recording page's order, we use page_private(page) and PG_private.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262 *
263 */
264static inline int page_is_buddy(struct page *page, int order)
265{
266 if (PagePrivate(page) &&
267 (page_order(page) == order) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268 page_count(page) == 0)
269 return 1;
270 return 0;
271}
272
273/*
274 * Freeing function for a buddy system allocator.
275 *
276 * The concept of a buddy system is to maintain direct-mapped table
277 * (containing bit values) for memory blocks of various "orders".
278 * The bottom level table contains the map for the smallest allocatable
279 * units of memory (here, pages), and each level above it describes
280 * pairs of units from the levels below, hence, "buddies".
281 * At a high level, all that happens here is marking the table entry
282 * at the bottom level available, and propagating the changes upward
283 * as necessary, plus some accounting needed to play nicely with other
284 * parts of the VM system.
285 * At each level, we keep a list of pages, which are heads of continuous
286 * free pages of length of (1 << order) and marked with PG_Private.Page's
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700287 * order is recorded in page_private(page) field.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288 * So when we are allocating or freeing one, we can derive the state of the
289 * other. That is, if we allocate a small block, and both were
290 * free, the remainder of the region must be split into blocks.
291 * If a block is freed, and its buddy is also free, then this
292 * triggers coalescing into a block of larger size.
293 *
294 * -- wli
295 */
296
297static inline void __free_pages_bulk (struct page *page,
298 struct zone *zone, unsigned int order)
299{
300 unsigned long page_idx;
301 int order_size = 1 << order;
302
303 if (unlikely(order))
304 destroy_compound_page(page, order);
305
306 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
307
308 BUG_ON(page_idx & (order_size - 1));
309 BUG_ON(bad_range(zone, page));
310
311 zone->free_pages += order_size;
312 while (order < MAX_ORDER-1) {
313 unsigned long combined_idx;
314 struct free_area *area;
315 struct page *buddy;
316
317 combined_idx = __find_combined_index(page_idx, order);
318 buddy = __page_find_buddy(page, page_idx, order);
319
320 if (bad_range(zone, buddy))
321 break;
322 if (!page_is_buddy(buddy, order))
323 break; /* Move the buddy up one level. */
324 list_del(&buddy->lru);
325 area = zone->free_area + order;
326 area->nr_free--;
327 rmv_page_order(buddy);
328 page = page + (combined_idx - page_idx);
329 page_idx = combined_idx;
330 order++;
331 }
332 set_page_order(page, order);
333 list_add(&page->lru, &zone->free_area[order].free_list);
334 zone->free_area[order].nr_free++;
335}
336
Hugh Dickins689bceb2005-11-21 21:32:20 -0800337static inline int free_pages_check(const char *function, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338{
339 if ( page_mapcount(page) ||
340 page->mapping != NULL ||
341 page_count(page) != 0 ||
342 (page->flags & (
343 1 << PG_lru |
344 1 << PG_private |
345 1 << PG_locked |
346 1 << PG_active |
347 1 << PG_reclaim |
348 1 << PG_slab |
349 1 << PG_swapcache |
Nick Pigginb5810032005-10-29 18:16:12 -0700350 1 << PG_writeback |
351 1 << PG_reserved )))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352 bad_page(function, page);
353 if (PageDirty(page))
Nick Piggin242e5462005-09-03 15:54:50 -0700354 __ClearPageDirty(page);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800355 /*
356 * For now, we report if PG_reserved was found set, but do not
357 * clear it, and do not free the page. But we shall soon need
358 * to do more, for when the ZERO_PAGE count wraps negative.
359 */
360 return PageReserved(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361}
362
363/*
364 * Frees a list of pages.
365 * Assumes all pages on list are in same zone, and of same order.
Renaud Lienhart207f36e2005-09-10 00:26:59 -0700366 * count is the number of pages to free.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367 *
368 * If the zone was previously in an "all pages pinned" state then look to
369 * see if this freeing clears that state.
370 *
371 * And clear the zone's pages_scanned counter, to hold off the "all pages are
372 * pinned" detection logic.
373 */
374static int
375free_pages_bulk(struct zone *zone, int count,
376 struct list_head *list, unsigned int order)
377{
378 unsigned long flags;
379 struct page *page = NULL;
380 int ret = 0;
381
382 spin_lock_irqsave(&zone->lock, flags);
383 zone->all_unreclaimable = 0;
384 zone->pages_scanned = 0;
385 while (!list_empty(list) && count--) {
386 page = list_entry(list->prev, struct page, lru);
387 /* have to delete it as __free_pages_bulk list manipulates */
388 list_del(&page->lru);
389 __free_pages_bulk(page, zone, order);
390 ret++;
391 }
392 spin_unlock_irqrestore(&zone->lock, flags);
393 return ret;
394}
395
396void __free_pages_ok(struct page *page, unsigned int order)
397{
398 LIST_HEAD(list);
399 int i;
Hugh Dickins689bceb2005-11-21 21:32:20 -0800400 int reserved = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401
402 arch_free_page(page, order);
403
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404#ifndef CONFIG_MMU
405 if (order > 0)
406 for (i = 1 ; i < (1 << order) ; ++i)
407 __put_page(page + i);
408#endif
409
410 for (i = 0 ; i < (1 << order) ; ++i)
Hugh Dickins689bceb2005-11-21 21:32:20 -0800411 reserved += free_pages_check(__FUNCTION__, page + i);
412 if (reserved)
413 return;
414
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 list_add(&page->lru, &list);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800416 mod_page_state(pgfree, 1 << order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417 kernel_map_pages(page, 1<<order, 0);
418 free_pages_bulk(page_zone(page), 1, &list, order);
419}
420
421
422/*
423 * The order of subdivision here is critical for the IO subsystem.
424 * Please do not alter this order without good reasons and regression
425 * testing. Specifically, as large blocks of memory are subdivided,
426 * the order in which smaller blocks are delivered depends on the order
427 * they're subdivided in this function. This is the primary factor
428 * influencing the order in which pages are delivered to the IO
429 * subsystem according to empirical testing, and this is also justified
430 * by considering the behavior of a buddy system containing a single
431 * large block of memory acted on by a series of small allocations.
432 * This behavior is a critical factor in sglist merging's success.
433 *
434 * -- wli
435 */
436static inline struct page *
437expand(struct zone *zone, struct page *page,
438 int low, int high, struct free_area *area)
439{
440 unsigned long size = 1 << high;
441
442 while (high > low) {
443 area--;
444 high--;
445 size >>= 1;
446 BUG_ON(bad_range(zone, &page[size]));
447 list_add(&page[size].lru, &area->free_list);
448 area->nr_free++;
449 set_page_order(&page[size], high);
450 }
451 return page;
452}
453
454void set_page_refs(struct page *page, int order)
455{
456#ifdef CONFIG_MMU
457 set_page_count(page, 1);
458#else
459 int i;
460
461 /*
462 * We need to reference all the pages for this order, otherwise if
463 * anyone accesses one of the pages with (get/put) it will be freed.
464 * - eg: access_process_vm()
465 */
466 for (i = 0; i < (1 << order); i++)
467 set_page_count(page + i, 1);
468#endif /* CONFIG_MMU */
469}
470
471/*
472 * This page is about to be returned from the page allocator
473 */
Hugh Dickins689bceb2005-11-21 21:32:20 -0800474static int prep_new_page(struct page *page, int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475{
Hugh Dickins334795e2005-06-21 17:15:08 -0700476 if ( page_mapcount(page) ||
477 page->mapping != NULL ||
478 page_count(page) != 0 ||
479 (page->flags & (
480 1 << PG_lru |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481 1 << PG_private |
482 1 << PG_locked |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483 1 << PG_active |
484 1 << PG_dirty |
485 1 << PG_reclaim |
Hugh Dickins334795e2005-06-21 17:15:08 -0700486 1 << PG_slab |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487 1 << PG_swapcache |
Nick Pigginb5810032005-10-29 18:16:12 -0700488 1 << PG_writeback |
489 1 << PG_reserved )))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490 bad_page(__FUNCTION__, page);
491
Hugh Dickins689bceb2005-11-21 21:32:20 -0800492 /*
493 * For now, we report if PG_reserved was found set, but do not
494 * clear it, and do not allocate the page: as a safety net.
495 */
496 if (PageReserved(page))
497 return 1;
498
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 page->flags &= ~(1 << PG_uptodate | 1 << PG_error |
500 1 << PG_referenced | 1 << PG_arch_1 |
501 1 << PG_checked | 1 << PG_mappedtodisk);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700502 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503 set_page_refs(page, order);
504 kernel_map_pages(page, 1 << order, 1);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800505 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506}
507
508/*
509 * Do the hard work of removing an element from the buddy allocator.
510 * Call me with the zone->lock already held.
511 */
512static struct page *__rmqueue(struct zone *zone, unsigned int order)
513{
514 struct free_area * area;
515 unsigned int current_order;
516 struct page *page;
517
518 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
519 area = zone->free_area + current_order;
520 if (list_empty(&area->free_list))
521 continue;
522
523 page = list_entry(area->free_list.next, struct page, lru);
524 list_del(&page->lru);
525 rmv_page_order(page);
526 area->nr_free--;
527 zone->free_pages -= 1UL << order;
528 return expand(zone, page, order, current_order, area);
529 }
530
531 return NULL;
532}
533
534/*
535 * Obtain a specified number of elements from the buddy allocator, all under
536 * a single hold of the lock, for efficiency. Add them to the supplied list.
537 * Returns the number of new pages which were placed at *list.
538 */
539static int rmqueue_bulk(struct zone *zone, unsigned int order,
540 unsigned long count, struct list_head *list)
541{
542 unsigned long flags;
543 int i;
544 int allocated = 0;
545 struct page *page;
546
547 spin_lock_irqsave(&zone->lock, flags);
548 for (i = 0; i < count; ++i) {
549 page = __rmqueue(zone, order);
550 if (page == NULL)
551 break;
552 allocated++;
553 list_add_tail(&page->lru, list);
554 }
555 spin_unlock_irqrestore(&zone->lock, flags);
556 return allocated;
557}
558
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700559#ifdef CONFIG_NUMA
560/* Called from the slab reaper to drain remote pagesets */
561void drain_remote_pages(void)
562{
563 struct zone *zone;
564 int i;
565 unsigned long flags;
566
567 local_irq_save(flags);
568 for_each_zone(zone) {
569 struct per_cpu_pageset *pset;
570
571 /* Do not drain local pagesets */
572 if (zone->zone_pgdat->node_id == numa_node_id())
573 continue;
574
575 pset = zone->pageset[smp_processor_id()];
576 for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) {
577 struct per_cpu_pages *pcp;
578
579 pcp = &pset->pcp[i];
580 if (pcp->count)
581 pcp->count -= free_pages_bulk(zone, pcp->count,
582 &pcp->list, 0);
583 }
584 }
585 local_irq_restore(flags);
586}
587#endif
588
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589#if defined(CONFIG_PM) || defined(CONFIG_HOTPLUG_CPU)
590static void __drain_pages(unsigned int cpu)
591{
592 struct zone *zone;
593 int i;
594
595 for_each_zone(zone) {
596 struct per_cpu_pageset *pset;
597
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700598 pset = zone_pcp(zone, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599 for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) {
600 struct per_cpu_pages *pcp;
601
602 pcp = &pset->pcp[i];
603 pcp->count -= free_pages_bulk(zone, pcp->count,
604 &pcp->list, 0);
605 }
606 }
607}
608#endif /* CONFIG_PM || CONFIG_HOTPLUG_CPU */
609
610#ifdef CONFIG_PM
611
612void mark_free_pages(struct zone *zone)
613{
614 unsigned long zone_pfn, flags;
615 int order;
616 struct list_head *curr;
617
618 if (!zone->spanned_pages)
619 return;
620
621 spin_lock_irqsave(&zone->lock, flags);
622 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn)
623 ClearPageNosaveFree(pfn_to_page(zone_pfn + zone->zone_start_pfn));
624
625 for (order = MAX_ORDER - 1; order >= 0; --order)
626 list_for_each(curr, &zone->free_area[order].free_list) {
627 unsigned long start_pfn, i;
628
629 start_pfn = page_to_pfn(list_entry(curr, struct page, lru));
630
631 for (i=0; i < (1<<order); i++)
632 SetPageNosaveFree(pfn_to_page(start_pfn+i));
633 }
634 spin_unlock_irqrestore(&zone->lock, flags);
635}
636
637/*
638 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
639 */
640void drain_local_pages(void)
641{
642 unsigned long flags;
643
644 local_irq_save(flags);
645 __drain_pages(smp_processor_id());
646 local_irq_restore(flags);
647}
648#endif /* CONFIG_PM */
649
650static void zone_statistics(struct zonelist *zonelist, struct zone *z)
651{
652#ifdef CONFIG_NUMA
653 unsigned long flags;
654 int cpu;
655 pg_data_t *pg = z->zone_pgdat;
656 pg_data_t *orig = zonelist->zones[0]->zone_pgdat;
657 struct per_cpu_pageset *p;
658
659 local_irq_save(flags);
660 cpu = smp_processor_id();
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700661 p = zone_pcp(z,cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662 if (pg == orig) {
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700663 p->numa_hit++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664 } else {
665 p->numa_miss++;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700666 zone_pcp(zonelist->zones[0], cpu)->numa_foreign++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667 }
668 if (pg == NODE_DATA(numa_node_id()))
669 p->local_node++;
670 else
671 p->other_node++;
672 local_irq_restore(flags);
673#endif
674}
675
676/*
677 * Free a 0-order page
678 */
679static void FASTCALL(free_hot_cold_page(struct page *page, int cold));
680static void fastcall free_hot_cold_page(struct page *page, int cold)
681{
682 struct zone *zone = page_zone(page);
683 struct per_cpu_pages *pcp;
684 unsigned long flags;
685
686 arch_free_page(page, 0);
687
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688 if (PageAnon(page))
689 page->mapping = NULL;
Hugh Dickins689bceb2005-11-21 21:32:20 -0800690 if (free_pages_check(__FUNCTION__, page))
691 return;
692
693 inc_page_state(pgfree);
694 kernel_map_pages(page, 1, 0);
695
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700696 pcp = &zone_pcp(zone, get_cpu())->pcp[cold];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 local_irq_save(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698 list_add(&page->lru, &pcp->list);
699 pcp->count++;
Christoph Lameter2caaad42005-06-21 17:15:00 -0700700 if (pcp->count >= pcp->high)
701 pcp->count -= free_pages_bulk(zone, pcp->batch, &pcp->list, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702 local_irq_restore(flags);
703 put_cpu();
704}
705
706void fastcall free_hot_page(struct page *page)
707{
708 free_hot_cold_page(page, 0);
709}
710
711void fastcall free_cold_page(struct page *page)
712{
713 free_hot_cold_page(page, 1);
714}
715
Al Virodd0fc662005-10-07 07:46:04 +0100716static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717{
718 int i;
719
720 BUG_ON((gfp_flags & (__GFP_WAIT | __GFP_HIGHMEM)) == __GFP_HIGHMEM);
721 for(i = 0; i < (1 << order); i++)
722 clear_highpage(page + i);
723}
724
725/*
726 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
727 * we cheat by calling it from here, in the order > 0 path. Saves a branch
728 * or two.
729 */
730static struct page *
Al Virodd0fc662005-10-07 07:46:04 +0100731buffered_rmqueue(struct zone *zone, int order, gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732{
733 unsigned long flags;
Hugh Dickins689bceb2005-11-21 21:32:20 -0800734 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735 int cold = !!(gfp_flags & __GFP_COLD);
736
Hugh Dickins689bceb2005-11-21 21:32:20 -0800737again:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738 if (order == 0) {
739 struct per_cpu_pages *pcp;
740
Hugh Dickins689bceb2005-11-21 21:32:20 -0800741 page = NULL;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700742 pcp = &zone_pcp(zone, get_cpu())->pcp[cold];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743 local_irq_save(flags);
744 if (pcp->count <= pcp->low)
745 pcp->count += rmqueue_bulk(zone, 0,
746 pcp->batch, &pcp->list);
747 if (pcp->count) {
748 page = list_entry(pcp->list.next, struct page, lru);
749 list_del(&page->lru);
750 pcp->count--;
751 }
752 local_irq_restore(flags);
753 put_cpu();
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800754 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 spin_lock_irqsave(&zone->lock, flags);
756 page = __rmqueue(zone, order);
757 spin_unlock_irqrestore(&zone->lock, flags);
758 }
759
760 if (page != NULL) {
761 BUG_ON(bad_range(zone, page));
762 mod_page_state_zone(zone, pgalloc, 1 << order);
Hugh Dickins689bceb2005-11-21 21:32:20 -0800763 if (prep_new_page(page, order))
764 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765
766 if (gfp_flags & __GFP_ZERO)
767 prep_zero_page(page, order, gfp_flags);
768
769 if (order && (gfp_flags & __GFP_COMP))
770 prep_compound_page(page, order);
771 }
772 return page;
773}
774
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800775#define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
Nick Piggin31488902005-11-28 13:44:03 -0800776#define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
777#define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
778#define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
779#define ALLOC_HARDER 0x10 /* try to alloc harder */
780#define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
781#define ALLOC_CPUSET 0x40 /* check for correct cpuset */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800782
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783/*
784 * Return 1 if free pages are above 'mark'. This takes into account the order
785 * of the allocation.
786 */
787int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800788 int classzone_idx, int alloc_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789{
790 /* free_pages my go negative - that's OK */
791 long min = mark, free_pages = z->free_pages - (1 << order) + 1;
792 int o;
793
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800794 if (alloc_flags & ALLOC_HIGH)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 min -= min / 2;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800796 if (alloc_flags & ALLOC_HARDER)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797 min -= min / 4;
798
799 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
800 return 0;
801 for (o = 0; o < order; o++) {
802 /* At the next order, this order's pages become unavailable */
803 free_pages -= z->free_area[o].nr_free << o;
804
805 /* Require fewer higher order pages to be free */
806 min >>= 1;
807
808 if (free_pages <= min)
809 return 0;
810 }
811 return 1;
812}
813
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800814/*
815 * get_page_from_freeliest goes through the zonelist trying to allocate
816 * a page.
817 */
818static struct page *
819get_page_from_freelist(gfp_t gfp_mask, unsigned int order,
820 struct zonelist *zonelist, int alloc_flags)
Martin Hicks753ee722005-06-21 17:14:41 -0700821{
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800822 struct zone **z = zonelist->zones;
823 struct page *page = NULL;
824 int classzone_idx = zone_idx(*z);
825
826 /*
827 * Go through the zonelist once, looking for a zone with enough free.
828 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
829 */
830 do {
831 if ((alloc_flags & ALLOC_CPUSET) &&
832 !cpuset_zone_allowed(*z, gfp_mask))
833 continue;
834
835 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
Nick Piggin31488902005-11-28 13:44:03 -0800836 unsigned long mark;
837 if (alloc_flags & ALLOC_WMARK_MIN)
838 mark = (*z)->pages_min;
839 else if (alloc_flags & ALLOC_WMARK_LOW)
840 mark = (*z)->pages_low;
841 else
842 mark = (*z)->pages_high;
843 if (!zone_watermark_ok(*z, order, mark,
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800844 classzone_idx, alloc_flags))
845 continue;
846 }
847
848 page = buffered_rmqueue(*z, order, gfp_mask);
849 if (page) {
850 zone_statistics(zonelist, *z);
851 break;
852 }
853 } while (*(++z) != NULL);
854 return page;
Martin Hicks753ee722005-06-21 17:14:41 -0700855}
856
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857/*
858 * This is the 'heart' of the zoned buddy allocator.
859 */
860struct page * fastcall
Al Virodd0fc662005-10-07 07:46:04 +0100861__alloc_pages(gfp_t gfp_mask, unsigned int order,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862 struct zonelist *zonelist)
863{
Al Viro260b2362005-10-21 03:22:44 -0400864 const gfp_t wait = gfp_mask & __GFP_WAIT;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800865 struct zone **z;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866 struct page *page;
867 struct reclaim_state reclaim_state;
868 struct task_struct *p = current;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 int do_retry;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800870 int alloc_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871 int did_some_progress;
872
873 might_sleep_if(wait);
874
Jens Axboe6b1de912005-11-17 21:35:02 +0100875restart:
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800876 z = zonelist->zones; /* the list of zones suitable for gfp_mask */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800878 if (unlikely(*z == NULL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879 /* Should this ever happen?? */
880 return NULL;
881 }
Jens Axboe6b1de912005-11-17 21:35:02 +0100882
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800883 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
Nick Piggin31488902005-11-28 13:44:03 -0800884 zonelist, ALLOC_WMARK_LOW|ALLOC_CPUSET);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800885 if (page)
886 goto got_pg;
887
Jens Axboe6b1de912005-11-17 21:35:02 +0100888 do {
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800889 wakeup_kswapd(*z, order);
Jens Axboe6b1de912005-11-17 21:35:02 +0100890 } while (*(++z));
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800891
Paul Jackson9bf22292005-09-06 15:18:12 -0700892 /*
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800893 * OK, we're below the kswapd watermark and have kicked background
894 * reclaim. Now things get more complex, so set up alloc_flags according
895 * to how we want to proceed.
896 *
897 * The caller may dip into page reserves a bit more if the caller
898 * cannot run direct reclaim, or if the caller has realtime scheduling
899 * policy.
Paul Jackson9bf22292005-09-06 15:18:12 -0700900 */
Nick Piggin31488902005-11-28 13:44:03 -0800901 alloc_flags = ALLOC_WMARK_MIN;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800902 if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait)
903 alloc_flags |= ALLOC_HARDER;
904 if (gfp_mask & __GFP_HIGH)
905 alloc_flags |= ALLOC_HIGH;
906 if (wait)
907 alloc_flags |= ALLOC_CPUSET;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908
909 /*
910 * Go through the zonelist again. Let __GFP_HIGH and allocations
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800911 * coming from realtime tasks go deeper into reserves.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912 *
913 * This is the last chance, in general, before the goto nopage.
914 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
Paul Jackson9bf22292005-09-06 15:18:12 -0700915 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916 */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800917 page = get_page_from_freelist(gfp_mask, order, zonelist, alloc_flags);
918 if (page)
919 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920
921 /* This allocation should allow future memory freeing. */
Nick Pigginb84a35b2005-05-01 08:58:36 -0700922
923 if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE)))
924 && !in_interrupt()) {
925 if (!(gfp_mask & __GFP_NOMEMALLOC)) {
Kirill Korotaev885036d2005-11-13 16:06:41 -0800926nofail_alloc:
Nick Pigginb84a35b2005-05-01 08:58:36 -0700927 /* go through the zonelist yet again, ignoring mins */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800928 page = get_page_from_freelist(gfp_mask, order,
929 zonelist, ALLOC_NO_WATERMARKS|ALLOC_CPUSET);
930 if (page)
931 goto got_pg;
Kirill Korotaev885036d2005-11-13 16:06:41 -0800932 if (gfp_mask & __GFP_NOFAIL) {
933 blk_congestion_wait(WRITE, HZ/50);
934 goto nofail_alloc;
935 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936 }
937 goto nopage;
938 }
939
940 /* Atomic allocations - we can't balance anything */
941 if (!wait)
942 goto nopage;
943
944rebalance:
945 cond_resched();
946
947 /* We now go into synchronous reclaim */
948 p->flags |= PF_MEMALLOC;
949 reclaim_state.reclaimed_slab = 0;
950 p->reclaim_state = &reclaim_state;
951
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800952 did_some_progress = try_to_free_pages(zonelist->zones, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953
954 p->reclaim_state = NULL;
955 p->flags &= ~PF_MEMALLOC;
956
957 cond_resched();
958
959 if (likely(did_some_progress)) {
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800960 page = get_page_from_freelist(gfp_mask, order,
961 zonelist, alloc_flags);
962 if (page)
963 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964 } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
965 /*
966 * Go through the zonelist yet one more time, keep
967 * very high watermark here, this is only to catch
968 * a parallel oom killing, we must fail if we're still
969 * under heavy pressure.
970 */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800971 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
Nick Piggin31488902005-11-28 13:44:03 -0800972 zonelist, ALLOC_WMARK_HIGH|ALLOC_CPUSET);
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800973 if (page)
974 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975
Marcelo Tosatti79b9ce32005-07-07 17:56:04 -0700976 out_of_memory(gfp_mask, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977 goto restart;
978 }
979
980 /*
981 * Don't let big-order allocations loop unless the caller explicitly
982 * requests that. Wait for some write requests to complete then retry.
983 *
984 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
985 * <= 3, but that may not be true in other implementations.
986 */
987 do_retry = 0;
988 if (!(gfp_mask & __GFP_NORETRY)) {
989 if ((order <= 3) || (gfp_mask & __GFP_REPEAT))
990 do_retry = 1;
991 if (gfp_mask & __GFP_NOFAIL)
992 do_retry = 1;
993 }
994 if (do_retry) {
995 blk_congestion_wait(WRITE, HZ/50);
996 goto rebalance;
997 }
998
999nopage:
1000 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
1001 printk(KERN_WARNING "%s: page allocation failure."
1002 " order:%d, mode:0x%x\n",
1003 p->comm, order, gfp_mask);
1004 dump_stack();
Janet Morgan578c2fd2005-06-21 17:14:56 -07001005 show_mem();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007got_pg:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008 return page;
1009}
1010
1011EXPORT_SYMBOL(__alloc_pages);
1012
1013/*
1014 * Common helper functions.
1015 */
Al Virodd0fc662005-10-07 07:46:04 +01001016fastcall unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017{
1018 struct page * page;
1019 page = alloc_pages(gfp_mask, order);
1020 if (!page)
1021 return 0;
1022 return (unsigned long) page_address(page);
1023}
1024
1025EXPORT_SYMBOL(__get_free_pages);
1026
Al Virodd0fc662005-10-07 07:46:04 +01001027fastcall unsigned long get_zeroed_page(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028{
1029 struct page * page;
1030
1031 /*
1032 * get_zeroed_page() returns a 32-bit address, which cannot represent
1033 * a highmem page
1034 */
Al Viro260b2362005-10-21 03:22:44 -04001035 BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036
1037 page = alloc_pages(gfp_mask | __GFP_ZERO, 0);
1038 if (page)
1039 return (unsigned long) page_address(page);
1040 return 0;
1041}
1042
1043EXPORT_SYMBOL(get_zeroed_page);
1044
1045void __pagevec_free(struct pagevec *pvec)
1046{
1047 int i = pagevec_count(pvec);
1048
1049 while (--i >= 0)
1050 free_hot_cold_page(pvec->pages[i], pvec->cold);
1051}
1052
1053fastcall void __free_pages(struct page *page, unsigned int order)
1054{
Nick Pigginb5810032005-10-29 18:16:12 -07001055 if (put_page_testzero(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056 if (order == 0)
1057 free_hot_page(page);
1058 else
1059 __free_pages_ok(page, order);
1060 }
1061}
1062
1063EXPORT_SYMBOL(__free_pages);
1064
1065fastcall void free_pages(unsigned long addr, unsigned int order)
1066{
1067 if (addr != 0) {
1068 BUG_ON(!virt_addr_valid((void *)addr));
1069 __free_pages(virt_to_page((void *)addr), order);
1070 }
1071}
1072
1073EXPORT_SYMBOL(free_pages);
1074
1075/*
1076 * Total amount of free (allocatable) RAM:
1077 */
1078unsigned int nr_free_pages(void)
1079{
1080 unsigned int sum = 0;
1081 struct zone *zone;
1082
1083 for_each_zone(zone)
1084 sum += zone->free_pages;
1085
1086 return sum;
1087}
1088
1089EXPORT_SYMBOL(nr_free_pages);
1090
1091#ifdef CONFIG_NUMA
1092unsigned int nr_free_pages_pgdat(pg_data_t *pgdat)
1093{
1094 unsigned int i, sum = 0;
1095
1096 for (i = 0; i < MAX_NR_ZONES; i++)
1097 sum += pgdat->node_zones[i].free_pages;
1098
1099 return sum;
1100}
1101#endif
1102
1103static unsigned int nr_free_zone_pages(int offset)
1104{
Martin J. Blighe310fd42005-07-29 22:59:18 -07001105 /* Just pick one node, since fallback list is circular */
1106 pg_data_t *pgdat = NODE_DATA(numa_node_id());
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107 unsigned int sum = 0;
1108
Martin J. Blighe310fd42005-07-29 22:59:18 -07001109 struct zonelist *zonelist = pgdat->node_zonelists + offset;
1110 struct zone **zonep = zonelist->zones;
1111 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112
Martin J. Blighe310fd42005-07-29 22:59:18 -07001113 for (zone = *zonep++; zone; zone = *zonep++) {
1114 unsigned long size = zone->present_pages;
1115 unsigned long high = zone->pages_high;
1116 if (size > high)
1117 sum += size - high;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118 }
1119
1120 return sum;
1121}
1122
1123/*
1124 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1125 */
1126unsigned int nr_free_buffer_pages(void)
1127{
Al Viroaf4ca452005-10-21 02:55:38 -04001128 return nr_free_zone_pages(gfp_zone(GFP_USER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129}
1130
1131/*
1132 * Amount of free RAM allocatable within all zones
1133 */
1134unsigned int nr_free_pagecache_pages(void)
1135{
Al Viroaf4ca452005-10-21 02:55:38 -04001136 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137}
1138
1139#ifdef CONFIG_HIGHMEM
1140unsigned int nr_free_highpages (void)
1141{
1142 pg_data_t *pgdat;
1143 unsigned int pages = 0;
1144
1145 for_each_pgdat(pgdat)
1146 pages += pgdat->node_zones[ZONE_HIGHMEM].free_pages;
1147
1148 return pages;
1149}
1150#endif
1151
1152#ifdef CONFIG_NUMA
1153static void show_node(struct zone *zone)
1154{
1155 printk("Node %d ", zone->zone_pgdat->node_id);
1156}
1157#else
1158#define show_node(zone) do { } while (0)
1159#endif
1160
1161/*
1162 * Accumulate the page_state information across all CPUs.
1163 * The result is unavoidably approximate - it can change
1164 * during and after execution of this function.
1165 */
1166static DEFINE_PER_CPU(struct page_state, page_states) = {0};
1167
1168atomic_t nr_pagecache = ATOMIC_INIT(0);
1169EXPORT_SYMBOL(nr_pagecache);
1170#ifdef CONFIG_SMP
1171DEFINE_PER_CPU(long, nr_pagecache_local) = 0;
1172#endif
1173
Martin Hicksc07e02d2005-09-03 15:55:11 -07001174void __get_page_state(struct page_state *ret, int nr, cpumask_t *cpumask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175{
1176 int cpu = 0;
1177
1178 memset(ret, 0, sizeof(*ret));
Martin Hicksc07e02d2005-09-03 15:55:11 -07001179 cpus_and(*cpumask, *cpumask, cpu_online_map);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180
Martin Hicksc07e02d2005-09-03 15:55:11 -07001181 cpu = first_cpu(*cpumask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182 while (cpu < NR_CPUS) {
1183 unsigned long *in, *out, off;
1184
1185 in = (unsigned long *)&per_cpu(page_states, cpu);
1186
Martin Hicksc07e02d2005-09-03 15:55:11 -07001187 cpu = next_cpu(cpu, *cpumask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188
1189 if (cpu < NR_CPUS)
1190 prefetch(&per_cpu(page_states, cpu));
1191
1192 out = (unsigned long *)ret;
1193 for (off = 0; off < nr; off++)
1194 *out++ += *in++;
1195 }
1196}
1197
Martin Hicksc07e02d2005-09-03 15:55:11 -07001198void get_page_state_node(struct page_state *ret, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199{
1200 int nr;
Martin Hicksc07e02d2005-09-03 15:55:11 -07001201 cpumask_t mask = node_to_cpumask(node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202
1203 nr = offsetof(struct page_state, GET_PAGE_STATE_LAST);
1204 nr /= sizeof(unsigned long);
1205
Martin Hicksc07e02d2005-09-03 15:55:11 -07001206 __get_page_state(ret, nr+1, &mask);
1207}
1208
1209void get_page_state(struct page_state *ret)
1210{
1211 int nr;
1212 cpumask_t mask = CPU_MASK_ALL;
1213
1214 nr = offsetof(struct page_state, GET_PAGE_STATE_LAST);
1215 nr /= sizeof(unsigned long);
1216
1217 __get_page_state(ret, nr + 1, &mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218}
1219
1220void get_full_page_state(struct page_state *ret)
1221{
Martin Hicksc07e02d2005-09-03 15:55:11 -07001222 cpumask_t mask = CPU_MASK_ALL;
1223
1224 __get_page_state(ret, sizeof(*ret) / sizeof(unsigned long), &mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225}
1226
Benjamin LaHaisec2f29ea2005-06-21 17:14:55 -07001227unsigned long __read_page_state(unsigned long offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228{
1229 unsigned long ret = 0;
1230 int cpu;
1231
1232 for_each_online_cpu(cpu) {
1233 unsigned long in;
1234
1235 in = (unsigned long)&per_cpu(page_states, cpu) + offset;
1236 ret += *((unsigned long *)in);
1237 }
1238 return ret;
1239}
1240
Benjamin LaHaise83e5d8f2005-06-21 17:14:54 -07001241void __mod_page_state(unsigned long offset, unsigned long delta)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242{
1243 unsigned long flags;
1244 void* ptr;
1245
1246 local_irq_save(flags);
1247 ptr = &__get_cpu_var(page_states);
1248 *(unsigned long*)(ptr + offset) += delta;
1249 local_irq_restore(flags);
1250}
1251
1252EXPORT_SYMBOL(__mod_page_state);
1253
1254void __get_zone_counts(unsigned long *active, unsigned long *inactive,
1255 unsigned long *free, struct pglist_data *pgdat)
1256{
1257 struct zone *zones = pgdat->node_zones;
1258 int i;
1259
1260 *active = 0;
1261 *inactive = 0;
1262 *free = 0;
1263 for (i = 0; i < MAX_NR_ZONES; i++) {
1264 *active += zones[i].nr_active;
1265 *inactive += zones[i].nr_inactive;
1266 *free += zones[i].free_pages;
1267 }
1268}
1269
1270void get_zone_counts(unsigned long *active,
1271 unsigned long *inactive, unsigned long *free)
1272{
1273 struct pglist_data *pgdat;
1274
1275 *active = 0;
1276 *inactive = 0;
1277 *free = 0;
1278 for_each_pgdat(pgdat) {
1279 unsigned long l, m, n;
1280 __get_zone_counts(&l, &m, &n, pgdat);
1281 *active += l;
1282 *inactive += m;
1283 *free += n;
1284 }
1285}
1286
1287void si_meminfo(struct sysinfo *val)
1288{
1289 val->totalram = totalram_pages;
1290 val->sharedram = 0;
1291 val->freeram = nr_free_pages();
1292 val->bufferram = nr_blockdev_pages();
1293#ifdef CONFIG_HIGHMEM
1294 val->totalhigh = totalhigh_pages;
1295 val->freehigh = nr_free_highpages();
1296#else
1297 val->totalhigh = 0;
1298 val->freehigh = 0;
1299#endif
1300 val->mem_unit = PAGE_SIZE;
1301}
1302
1303EXPORT_SYMBOL(si_meminfo);
1304
1305#ifdef CONFIG_NUMA
1306void si_meminfo_node(struct sysinfo *val, int nid)
1307{
1308 pg_data_t *pgdat = NODE_DATA(nid);
1309
1310 val->totalram = pgdat->node_present_pages;
1311 val->freeram = nr_free_pages_pgdat(pgdat);
1312 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1313 val->freehigh = pgdat->node_zones[ZONE_HIGHMEM].free_pages;
1314 val->mem_unit = PAGE_SIZE;
1315}
1316#endif
1317
1318#define K(x) ((x) << (PAGE_SHIFT-10))
1319
1320/*
1321 * Show free area list (used inside shift_scroll-lock stuff)
1322 * We also calculate the percentage fragmentation. We do this by counting the
1323 * memory on each free list with the exception of the first item on the list.
1324 */
1325void show_free_areas(void)
1326{
1327 struct page_state ps;
1328 int cpu, temperature;
1329 unsigned long active;
1330 unsigned long inactive;
1331 unsigned long free;
1332 struct zone *zone;
1333
1334 for_each_zone(zone) {
1335 show_node(zone);
1336 printk("%s per-cpu:", zone->name);
1337
1338 if (!zone->present_pages) {
1339 printk(" empty\n");
1340 continue;
1341 } else
1342 printk("\n");
1343
Dave Jones6b482c62005-11-10 15:45:56 -05001344 for_each_online_cpu(cpu) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345 struct per_cpu_pageset *pageset;
1346
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001347 pageset = zone_pcp(zone, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348
1349 for (temperature = 0; temperature < 2; temperature++)
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001350 printk("cpu %d %s: low %d, high %d, batch %d used:%d\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351 cpu,
1352 temperature ? "cold" : "hot",
1353 pageset->pcp[temperature].low,
1354 pageset->pcp[temperature].high,
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001355 pageset->pcp[temperature].batch,
1356 pageset->pcp[temperature].count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357 }
1358 }
1359
1360 get_page_state(&ps);
1361 get_zone_counts(&active, &inactive, &free);
1362
Denis Vlasenkoc0d62212005-06-21 17:15:14 -07001363 printk("Free pages: %11ukB (%ukB HighMem)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364 K(nr_free_pages()),
1365 K(nr_free_highpages()));
1366
1367 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu "
1368 "unstable:%lu free:%u slab:%lu mapped:%lu pagetables:%lu\n",
1369 active,
1370 inactive,
1371 ps.nr_dirty,
1372 ps.nr_writeback,
1373 ps.nr_unstable,
1374 nr_free_pages(),
1375 ps.nr_slab,
1376 ps.nr_mapped,
1377 ps.nr_page_table_pages);
1378
1379 for_each_zone(zone) {
1380 int i;
1381
1382 show_node(zone);
1383 printk("%s"
1384 " free:%lukB"
1385 " min:%lukB"
1386 " low:%lukB"
1387 " high:%lukB"
1388 " active:%lukB"
1389 " inactive:%lukB"
1390 " present:%lukB"
1391 " pages_scanned:%lu"
1392 " all_unreclaimable? %s"
1393 "\n",
1394 zone->name,
1395 K(zone->free_pages),
1396 K(zone->pages_min),
1397 K(zone->pages_low),
1398 K(zone->pages_high),
1399 K(zone->nr_active),
1400 K(zone->nr_inactive),
1401 K(zone->present_pages),
1402 zone->pages_scanned,
1403 (zone->all_unreclaimable ? "yes" : "no")
1404 );
1405 printk("lowmem_reserve[]:");
1406 for (i = 0; i < MAX_NR_ZONES; i++)
1407 printk(" %lu", zone->lowmem_reserve[i]);
1408 printk("\n");
1409 }
1410
1411 for_each_zone(zone) {
1412 unsigned long nr, flags, order, total = 0;
1413
1414 show_node(zone);
1415 printk("%s: ", zone->name);
1416 if (!zone->present_pages) {
1417 printk("empty\n");
1418 continue;
1419 }
1420
1421 spin_lock_irqsave(&zone->lock, flags);
1422 for (order = 0; order < MAX_ORDER; order++) {
1423 nr = zone->free_area[order].nr_free;
1424 total += nr << order;
1425 printk("%lu*%lukB ", nr, K(1UL) << order);
1426 }
1427 spin_unlock_irqrestore(&zone->lock, flags);
1428 printk("= %lukB\n", K(total));
1429 }
1430
1431 show_swap_cache_info();
1432}
1433
1434/*
1435 * Builds allocation fallback zone lists.
1436 */
1437static int __init build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist, int j, int k)
1438{
1439 switch (k) {
1440 struct zone *zone;
1441 default:
1442 BUG();
1443 case ZONE_HIGHMEM:
1444 zone = pgdat->node_zones + ZONE_HIGHMEM;
1445 if (zone->present_pages) {
1446#ifndef CONFIG_HIGHMEM
1447 BUG();
1448#endif
1449 zonelist->zones[j++] = zone;
1450 }
1451 case ZONE_NORMAL:
1452 zone = pgdat->node_zones + ZONE_NORMAL;
1453 if (zone->present_pages)
1454 zonelist->zones[j++] = zone;
Andi Kleena2f1b422005-11-05 17:25:53 +01001455 case ZONE_DMA32:
1456 zone = pgdat->node_zones + ZONE_DMA32;
1457 if (zone->present_pages)
1458 zonelist->zones[j++] = zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459 case ZONE_DMA:
1460 zone = pgdat->node_zones + ZONE_DMA;
1461 if (zone->present_pages)
1462 zonelist->zones[j++] = zone;
1463 }
1464
1465 return j;
1466}
1467
Al Viro260b2362005-10-21 03:22:44 -04001468static inline int highest_zone(int zone_bits)
1469{
1470 int res = ZONE_NORMAL;
1471 if (zone_bits & (__force int)__GFP_HIGHMEM)
1472 res = ZONE_HIGHMEM;
Andi Kleena2f1b422005-11-05 17:25:53 +01001473 if (zone_bits & (__force int)__GFP_DMA32)
1474 res = ZONE_DMA32;
Al Viro260b2362005-10-21 03:22:44 -04001475 if (zone_bits & (__force int)__GFP_DMA)
1476 res = ZONE_DMA;
1477 return res;
1478}
1479
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480#ifdef CONFIG_NUMA
1481#define MAX_NODE_LOAD (num_online_nodes())
1482static int __initdata node_load[MAX_NUMNODES];
1483/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -07001484 * 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 -07001485 * @node: node whose fallback list we're appending
1486 * @used_node_mask: nodemask_t of already used nodes
1487 *
1488 * We use a number of factors to determine which is the next node that should
1489 * appear on a given node's fallback list. The node should not have appeared
1490 * already in @node's fallback list, and it should be the next closest node
1491 * according to the distance array (which contains arbitrary distance values
1492 * from each node to each node in the system), and should also prefer nodes
1493 * with no CPUs, since presumably they'll have very little allocation pressure
1494 * on them otherwise.
1495 * It returns -1 if no node is found.
1496 */
1497static int __init find_next_best_node(int node, nodemask_t *used_node_mask)
1498{
1499 int i, n, val;
1500 int min_val = INT_MAX;
1501 int best_node = -1;
1502
1503 for_each_online_node(i) {
1504 cpumask_t tmp;
1505
1506 /* Start from local node */
1507 n = (node+i) % num_online_nodes();
1508
1509 /* Don't want a node to appear more than once */
1510 if (node_isset(n, *used_node_mask))
1511 continue;
1512
1513 /* Use the local node if we haven't already */
1514 if (!node_isset(node, *used_node_mask)) {
1515 best_node = node;
1516 break;
1517 }
1518
1519 /* Use the distance array to find the distance */
1520 val = node_distance(node, n);
1521
1522 /* Give preference to headless and unused nodes */
1523 tmp = node_to_cpumask(n);
1524 if (!cpus_empty(tmp))
1525 val += PENALTY_FOR_NODE_WITH_CPUS;
1526
1527 /* Slight preference for less loaded node */
1528 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
1529 val += node_load[n];
1530
1531 if (val < min_val) {
1532 min_val = val;
1533 best_node = n;
1534 }
1535 }
1536
1537 if (best_node >= 0)
1538 node_set(best_node, *used_node_mask);
1539
1540 return best_node;
1541}
1542
1543static void __init build_zonelists(pg_data_t *pgdat)
1544{
1545 int i, j, k, node, local_node;
1546 int prev_node, load;
1547 struct zonelist *zonelist;
1548 nodemask_t used_mask;
1549
1550 /* initialize zonelists */
1551 for (i = 0; i < GFP_ZONETYPES; i++) {
1552 zonelist = pgdat->node_zonelists + i;
1553 zonelist->zones[0] = NULL;
1554 }
1555
1556 /* NUMA-aware ordering of nodes */
1557 local_node = pgdat->node_id;
1558 load = num_online_nodes();
1559 prev_node = local_node;
1560 nodes_clear(used_mask);
1561 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
1562 /*
1563 * We don't want to pressure a particular node.
1564 * So adding penalty to the first node in same
1565 * distance group to make it round-robin.
1566 */
1567 if (node_distance(local_node, node) !=
1568 node_distance(local_node, prev_node))
1569 node_load[node] += load;
1570 prev_node = node;
1571 load--;
1572 for (i = 0; i < GFP_ZONETYPES; i++) {
1573 zonelist = pgdat->node_zonelists + i;
1574 for (j = 0; zonelist->zones[j] != NULL; j++);
1575
Al Viro260b2362005-10-21 03:22:44 -04001576 k = highest_zone(i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577
1578 j = build_zonelists_node(NODE_DATA(node), zonelist, j, k);
1579 zonelist->zones[j] = NULL;
1580 }
1581 }
1582}
1583
1584#else /* CONFIG_NUMA */
1585
1586static void __init build_zonelists(pg_data_t *pgdat)
1587{
1588 int i, j, k, node, local_node;
1589
1590 local_node = pgdat->node_id;
1591 for (i = 0; i < GFP_ZONETYPES; i++) {
1592 struct zonelist *zonelist;
1593
1594 zonelist = pgdat->node_zonelists + i;
1595
1596 j = 0;
Al Viro260b2362005-10-21 03:22:44 -04001597 k = highest_zone(i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598 j = build_zonelists_node(pgdat, zonelist, j, k);
1599 /*
1600 * Now we build the zonelist so that it contains the zones
1601 * of all the other nodes.
1602 * We don't want to pressure a particular node, so when
1603 * building the zones for node N, we make sure that the
1604 * zones coming right after the local ones are those from
1605 * node N+1 (modulo N)
1606 */
1607 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
1608 if (!node_online(node))
1609 continue;
1610 j = build_zonelists_node(NODE_DATA(node), zonelist, j, k);
1611 }
1612 for (node = 0; node < local_node; node++) {
1613 if (!node_online(node))
1614 continue;
1615 j = build_zonelists_node(NODE_DATA(node), zonelist, j, k);
1616 }
1617
1618 zonelist->zones[j] = NULL;
1619 }
1620}
1621
1622#endif /* CONFIG_NUMA */
1623
1624void __init build_all_zonelists(void)
1625{
1626 int i;
1627
1628 for_each_online_node(i)
1629 build_zonelists(NODE_DATA(i));
1630 printk("Built %i zonelists\n", num_online_nodes());
1631 cpuset_init_current_mems_allowed();
1632}
1633
1634/*
1635 * Helper functions to size the waitqueue hash table.
1636 * Essentially these want to choose hash table sizes sufficiently
1637 * large so that collisions trying to wait on pages are rare.
1638 * But in fact, the number of active page waitqueues on typical
1639 * systems is ridiculously low, less than 200. So this is even
1640 * conservative, even though it seems large.
1641 *
1642 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
1643 * waitqueues, i.e. the size of the waitq table given the number of pages.
1644 */
1645#define PAGES_PER_WAITQUEUE 256
1646
1647static inline unsigned long wait_table_size(unsigned long pages)
1648{
1649 unsigned long size = 1;
1650
1651 pages /= PAGES_PER_WAITQUEUE;
1652
1653 while (size < pages)
1654 size <<= 1;
1655
1656 /*
1657 * Once we have dozens or even hundreds of threads sleeping
1658 * on IO we've got bigger problems than wait queue collision.
1659 * Limit the size of the wait table to a reasonable size.
1660 */
1661 size = min(size, 4096UL);
1662
1663 return max(size, 4UL);
1664}
1665
1666/*
1667 * This is an integer logarithm so that shifts can be used later
1668 * to extract the more random high bits from the multiplicative
1669 * hash function before the remainder is taken.
1670 */
1671static inline unsigned long wait_table_bits(unsigned long size)
1672{
1673 return ffz(~size);
1674}
1675
1676#define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
1677
1678static void __init calculate_zone_totalpages(struct pglist_data *pgdat,
1679 unsigned long *zones_size, unsigned long *zholes_size)
1680{
1681 unsigned long realtotalpages, totalpages = 0;
1682 int i;
1683
1684 for (i = 0; i < MAX_NR_ZONES; i++)
1685 totalpages += zones_size[i];
1686 pgdat->node_spanned_pages = totalpages;
1687
1688 realtotalpages = totalpages;
1689 if (zholes_size)
1690 for (i = 0; i < MAX_NR_ZONES; i++)
1691 realtotalpages -= zholes_size[i];
1692 pgdat->node_present_pages = realtotalpages;
1693 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, realtotalpages);
1694}
1695
1696
1697/*
1698 * Initially all pages are reserved - free ones are freed
1699 * up by free_all_bootmem() once the early boot process is
1700 * done. Non-atomic initialization, single-pass.
1701 */
Dave Hansen3947be12005-10-29 18:16:54 -07001702void __devinit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703 unsigned long start_pfn)
1704{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705 struct page *page;
Andy Whitcroft29751f62005-06-23 00:08:00 -07001706 unsigned long end_pfn = start_pfn + size;
1707 unsigned long pfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001709 for (pfn = start_pfn; pfn < end_pfn; pfn++, page++) {
1710 if (!early_pfn_valid(pfn))
1711 continue;
Andy Whitcroft641c7672005-06-23 00:07:59 -07001712 if (!early_pfn_in_nid(pfn, nid))
1713 continue;
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001714 page = pfn_to_page(pfn);
1715 set_page_links(page, zone, nid, pfn);
Nick Pigginb5810032005-10-29 18:16:12 -07001716 set_page_count(page, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717 reset_page_mapcount(page);
1718 SetPageReserved(page);
1719 INIT_LIST_HEAD(&page->lru);
1720#ifdef WANT_PAGE_VIRTUAL
1721 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
1722 if (!is_highmem_idx(zone))
Bob Picco3212c6b2005-06-27 14:36:28 -07001723 set_page_address(page, __va(pfn << PAGE_SHIFT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725 }
1726}
1727
1728void zone_init_free_lists(struct pglist_data *pgdat, struct zone *zone,
1729 unsigned long size)
1730{
1731 int order;
1732 for (order = 0; order < MAX_ORDER ; order++) {
1733 INIT_LIST_HEAD(&zone->free_area[order].free_list);
1734 zone->free_area[order].nr_free = 0;
1735 }
1736}
1737
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001738#define ZONETABLE_INDEX(x, zone_nr) ((x << ZONES_SHIFT) | zone_nr)
1739void zonetable_add(struct zone *zone, int nid, int zid, unsigned long pfn,
1740 unsigned long size)
1741{
1742 unsigned long snum = pfn_to_section_nr(pfn);
1743 unsigned long end = pfn_to_section_nr(pfn + size);
1744
1745 if (FLAGS_HAS_NODE)
1746 zone_table[ZONETABLE_INDEX(nid, zid)] = zone;
1747 else
1748 for (; snum <= end; snum++)
1749 zone_table[ZONETABLE_INDEX(snum, zid)] = zone;
1750}
1751
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752#ifndef __HAVE_ARCH_MEMMAP_INIT
1753#define memmap_init(size, nid, zone, start_pfn) \
1754 memmap_init_zone((size), (nid), (zone), (start_pfn))
1755#endif
1756
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001757static int __devinit zone_batchsize(struct zone *zone)
1758{
1759 int batch;
1760
1761 /*
1762 * The per-cpu-pages pools are set to around 1000th of the
Seth, Rohitba56e912005-10-29 18:15:47 -07001763 * size of the zone. But no more than 1/2 of a meg.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001764 *
1765 * OK, so we don't know how big the cache is. So guess.
1766 */
1767 batch = zone->present_pages / 1024;
Seth, Rohitba56e912005-10-29 18:15:47 -07001768 if (batch * PAGE_SIZE > 512 * 1024)
1769 batch = (512 * 1024) / PAGE_SIZE;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001770 batch /= 4; /* We effectively *= 4 below */
1771 if (batch < 1)
1772 batch = 1;
1773
1774 /*
Nick Piggin0ceaacc2005-12-04 13:55:25 +11001775 * Clamp the batch to a 2^n - 1 value. Having a power
1776 * of 2 value was found to be more likely to have
1777 * suboptimal cache aliasing properties in some cases.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001778 *
Nick Piggin0ceaacc2005-12-04 13:55:25 +11001779 * For example if 2 tasks are alternately allocating
1780 * batches of pages, one task can end up with a lot
1781 * of pages of one half of the possible page colors
1782 * and the other with pages of the other colors.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001783 */
Nick Piggin0ceaacc2005-12-04 13:55:25 +11001784 batch = (1 << (fls(batch + batch/2)-1)) - 1;
Seth, Rohitba56e912005-10-29 18:15:47 -07001785
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001786 return batch;
1787}
1788
Christoph Lameter2caaad42005-06-21 17:15:00 -07001789inline void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
1790{
1791 struct per_cpu_pages *pcp;
1792
Magnus Damm1c6fe942005-10-26 01:58:59 -07001793 memset(p, 0, sizeof(*p));
1794
Christoph Lameter2caaad42005-06-21 17:15:00 -07001795 pcp = &p->pcp[0]; /* hot */
1796 pcp->count = 0;
Seth, Rohite46a5e22005-10-29 18:15:48 -07001797 pcp->low = 0;
Christoph Lameter2caaad42005-06-21 17:15:00 -07001798 pcp->high = 6 * batch;
1799 pcp->batch = max(1UL, 1 * batch);
1800 INIT_LIST_HEAD(&pcp->list);
1801
1802 pcp = &p->pcp[1]; /* cold*/
1803 pcp->count = 0;
1804 pcp->low = 0;
1805 pcp->high = 2 * batch;
Seth, Rohite46a5e22005-10-29 18:15:48 -07001806 pcp->batch = max(1UL, batch/2);
Christoph Lameter2caaad42005-06-21 17:15:00 -07001807 INIT_LIST_HEAD(&pcp->list);
1808}
1809
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001810#ifdef CONFIG_NUMA
1811/*
Christoph Lameter2caaad42005-06-21 17:15:00 -07001812 * Boot pageset table. One per cpu which is going to be used for all
1813 * zones and all nodes. The parameters will be set in such a way
1814 * that an item put on a list will immediately be handed over to
1815 * the buddy list. This is safe since pageset manipulation is done
1816 * with interrupts disabled.
1817 *
1818 * Some NUMA counter updates may also be caught by the boot pagesets.
Christoph Lameterb7c84c62005-06-22 20:26:07 -07001819 *
1820 * The boot_pagesets must be kept even after bootup is complete for
1821 * unused processors and/or zones. They do play a role for bootstrapping
1822 * hotplugged processors.
1823 *
1824 * zoneinfo_show() and maybe other functions do
1825 * not check if the processor is online before following the pageset pointer.
1826 * Other parts of the kernel may not check if the zone is available.
Christoph Lameter2caaad42005-06-21 17:15:00 -07001827 */
1828static struct per_cpu_pageset
Christoph Lameterb7c84c62005-06-22 20:26:07 -07001829 boot_pageset[NR_CPUS];
Christoph Lameter2caaad42005-06-21 17:15:00 -07001830
1831/*
1832 * Dynamically allocate memory for the
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001833 * per cpu pageset array in struct zone.
1834 */
1835static int __devinit process_zones(int cpu)
1836{
1837 struct zone *zone, *dzone;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001838
1839 for_each_zone(zone) {
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001840
Christoph Lameter2caaad42005-06-21 17:15:00 -07001841 zone->pageset[cpu] = kmalloc_node(sizeof(struct per_cpu_pageset),
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001842 GFP_KERNEL, cpu_to_node(cpu));
Christoph Lameter2caaad42005-06-21 17:15:00 -07001843 if (!zone->pageset[cpu])
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001844 goto bad;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001845
Christoph Lameter2caaad42005-06-21 17:15:00 -07001846 setup_pageset(zone->pageset[cpu], zone_batchsize(zone));
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001847 }
1848
1849 return 0;
1850bad:
1851 for_each_zone(dzone) {
1852 if (dzone == zone)
1853 break;
1854 kfree(dzone->pageset[cpu]);
1855 dzone->pageset[cpu] = NULL;
1856 }
1857 return -ENOMEM;
1858}
1859
1860static inline void free_zone_pagesets(int cpu)
1861{
1862#ifdef CONFIG_NUMA
1863 struct zone *zone;
1864
1865 for_each_zone(zone) {
1866 struct per_cpu_pageset *pset = zone_pcp(zone, cpu);
1867
1868 zone_pcp(zone, cpu) = NULL;
1869 kfree(pset);
1870 }
1871#endif
1872}
1873
1874static int __devinit pageset_cpuup_callback(struct notifier_block *nfb,
1875 unsigned long action,
1876 void *hcpu)
1877{
1878 int cpu = (long)hcpu;
1879 int ret = NOTIFY_OK;
1880
1881 switch (action) {
1882 case CPU_UP_PREPARE:
1883 if (process_zones(cpu))
1884 ret = NOTIFY_BAD;
1885 break;
Andi Kleenb0d41692005-11-05 17:25:53 +01001886 case CPU_UP_CANCELED:
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001887 case CPU_DEAD:
1888 free_zone_pagesets(cpu);
1889 break;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001890 default:
1891 break;
1892 }
1893 return ret;
1894}
1895
1896static struct notifier_block pageset_notifier =
1897 { &pageset_cpuup_callback, NULL, 0 };
1898
Al Viro78d99552005-12-15 09:18:25 +00001899void __init setup_per_cpu_pageset(void)
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001900{
1901 int err;
1902
1903 /* Initialize per_cpu_pageset for cpu 0.
1904 * A cpuup callback will do this for every cpu
1905 * as it comes online
1906 */
1907 err = process_zones(smp_processor_id());
1908 BUG_ON(err);
1909 register_cpu_notifier(&pageset_notifier);
1910}
1911
1912#endif
1913
Dave Hansened8ece22005-10-29 18:16:50 -07001914static __devinit
1915void zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
1916{
1917 int i;
1918 struct pglist_data *pgdat = zone->zone_pgdat;
1919
1920 /*
1921 * The per-page waitqueue mechanism uses hashed waitqueues
1922 * per zone.
1923 */
1924 zone->wait_table_size = wait_table_size(zone_size_pages);
1925 zone->wait_table_bits = wait_table_bits(zone->wait_table_size);
1926 zone->wait_table = (wait_queue_head_t *)
1927 alloc_bootmem_node(pgdat, zone->wait_table_size
1928 * sizeof(wait_queue_head_t));
1929
1930 for(i = 0; i < zone->wait_table_size; ++i)
1931 init_waitqueue_head(zone->wait_table + i);
1932}
1933
1934static __devinit void zone_pcp_init(struct zone *zone)
1935{
1936 int cpu;
1937 unsigned long batch = zone_batchsize(zone);
1938
1939 for (cpu = 0; cpu < NR_CPUS; cpu++) {
1940#ifdef CONFIG_NUMA
1941 /* Early boot. Slab allocator not functional yet */
1942 zone->pageset[cpu] = &boot_pageset[cpu];
1943 setup_pageset(&boot_pageset[cpu],0);
1944#else
1945 setup_pageset(zone_pcp(zone,cpu), batch);
1946#endif
1947 }
1948 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n",
1949 zone->name, zone->present_pages, batch);
1950}
1951
1952static __devinit void init_currently_empty_zone(struct zone *zone,
1953 unsigned long zone_start_pfn, unsigned long size)
1954{
1955 struct pglist_data *pgdat = zone->zone_pgdat;
1956
1957 zone_wait_table_init(zone, size);
1958 pgdat->nr_zones = zone_idx(zone) + 1;
1959
1960 zone->zone_mem_map = pfn_to_page(zone_start_pfn);
1961 zone->zone_start_pfn = zone_start_pfn;
1962
1963 memmap_init(size, pgdat->node_id, zone_idx(zone), zone_start_pfn);
1964
1965 zone_init_free_lists(pgdat, zone, zone->spanned_pages);
1966}
1967
Linus Torvalds1da177e2005-04-16 15:20:36 -07001968/*
1969 * Set up the zone data structures:
1970 * - mark all pages reserved
1971 * - mark all memory queues empty
1972 * - clear the memory bitmaps
1973 */
1974static void __init free_area_init_core(struct pglist_data *pgdat,
1975 unsigned long *zones_size, unsigned long *zholes_size)
1976{
Dave Hansened8ece22005-10-29 18:16:50 -07001977 unsigned long j;
1978 int nid = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979 unsigned long zone_start_pfn = pgdat->node_start_pfn;
1980
Dave Hansen208d54e2005-10-29 18:16:52 -07001981 pgdat_resize_init(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982 pgdat->nr_zones = 0;
1983 init_waitqueue_head(&pgdat->kswapd_wait);
1984 pgdat->kswapd_max_order = 0;
1985
1986 for (j = 0; j < MAX_NR_ZONES; j++) {
1987 struct zone *zone = pgdat->node_zones + j;
1988 unsigned long size, realsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001989
Linus Torvalds1da177e2005-04-16 15:20:36 -07001990 realsize = size = zones_size[j];
1991 if (zholes_size)
1992 realsize -= zholes_size[j];
1993
Andi Kleena2f1b422005-11-05 17:25:53 +01001994 if (j < ZONE_HIGHMEM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001995 nr_kernel_pages += realsize;
1996 nr_all_pages += realsize;
1997
1998 zone->spanned_pages = size;
1999 zone->present_pages = realsize;
2000 zone->name = zone_names[j];
2001 spin_lock_init(&zone->lock);
2002 spin_lock_init(&zone->lru_lock);
Dave Hansenbdc8cb92005-10-29 18:16:53 -07002003 zone_seqlock_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002004 zone->zone_pgdat = pgdat;
2005 zone->free_pages = 0;
2006
2007 zone->temp_priority = zone->prev_priority = DEF_PRIORITY;
2008
Dave Hansened8ece22005-10-29 18:16:50 -07002009 zone_pcp_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002010 INIT_LIST_HEAD(&zone->active_list);
2011 INIT_LIST_HEAD(&zone->inactive_list);
2012 zone->nr_scan_active = 0;
2013 zone->nr_scan_inactive = 0;
2014 zone->nr_active = 0;
2015 zone->nr_inactive = 0;
Martin Hicks53e9a612005-09-03 15:54:51 -07002016 atomic_set(&zone->reclaim_in_progress, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002017 if (!size)
2018 continue;
2019
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002020 zonetable_add(zone, nid, j, zone_start_pfn, size);
Dave Hansened8ece22005-10-29 18:16:50 -07002021 init_currently_empty_zone(zone, zone_start_pfn, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002022 zone_start_pfn += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002023 }
2024}
2025
2026static void __init alloc_node_mem_map(struct pglist_data *pgdat)
2027{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002028 /* Skip empty nodes */
2029 if (!pgdat->node_spanned_pages)
2030 return;
2031
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002032#ifdef CONFIG_FLAT_NODE_MEM_MAP
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033 /* ia64 gets its own node_mem_map, before this, without bootmem */
2034 if (!pgdat->node_mem_map) {
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002035 unsigned long size;
2036 struct page *map;
2037
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038 size = (pgdat->node_spanned_pages + 1) * sizeof(struct page);
Dave Hansen6f167ec2005-06-23 00:07:39 -07002039 map = alloc_remap(pgdat->node_id, size);
2040 if (!map)
2041 map = alloc_bootmem_node(pgdat, size);
2042 pgdat->node_mem_map = map;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043 }
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002044#ifdef CONFIG_FLATMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045 /*
2046 * With no DISCONTIG, the global mem_map is just set as node 0's
2047 */
2048 if (pgdat == NODE_DATA(0))
2049 mem_map = NODE_DATA(0)->node_mem_map;
2050#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002051#endif /* CONFIG_FLAT_NODE_MEM_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052}
2053
2054void __init free_area_init_node(int nid, struct pglist_data *pgdat,
2055 unsigned long *zones_size, unsigned long node_start_pfn,
2056 unsigned long *zholes_size)
2057{
2058 pgdat->node_id = nid;
2059 pgdat->node_start_pfn = node_start_pfn;
2060 calculate_zone_totalpages(pgdat, zones_size, zholes_size);
2061
2062 alloc_node_mem_map(pgdat);
2063
2064 free_area_init_core(pgdat, zones_size, zholes_size);
2065}
2066
Dave Hansen93b75042005-06-23 00:07:47 -07002067#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07002068static bootmem_data_t contig_bootmem_data;
2069struct pglist_data contig_page_data = { .bdata = &contig_bootmem_data };
2070
2071EXPORT_SYMBOL(contig_page_data);
Dave Hansen93b75042005-06-23 00:07:47 -07002072#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002073
2074void __init free_area_init(unsigned long *zones_size)
2075{
Dave Hansen93b75042005-06-23 00:07:47 -07002076 free_area_init_node(0, NODE_DATA(0), zones_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002077 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
2078}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002079
2080#ifdef CONFIG_PROC_FS
2081
2082#include <linux/seq_file.h>
2083
2084static void *frag_start(struct seq_file *m, loff_t *pos)
2085{
2086 pg_data_t *pgdat;
2087 loff_t node = *pos;
2088
2089 for (pgdat = pgdat_list; pgdat && node; pgdat = pgdat->pgdat_next)
2090 --node;
2091
2092 return pgdat;
2093}
2094
2095static void *frag_next(struct seq_file *m, void *arg, loff_t *pos)
2096{
2097 pg_data_t *pgdat = (pg_data_t *)arg;
2098
2099 (*pos)++;
2100 return pgdat->pgdat_next;
2101}
2102
2103static void frag_stop(struct seq_file *m, void *arg)
2104{
2105}
2106
2107/*
2108 * This walks the free areas for each zone.
2109 */
2110static int frag_show(struct seq_file *m, void *arg)
2111{
2112 pg_data_t *pgdat = (pg_data_t *)arg;
2113 struct zone *zone;
2114 struct zone *node_zones = pgdat->node_zones;
2115 unsigned long flags;
2116 int order;
2117
2118 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
2119 if (!zone->present_pages)
2120 continue;
2121
2122 spin_lock_irqsave(&zone->lock, flags);
2123 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
2124 for (order = 0; order < MAX_ORDER; ++order)
2125 seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
2126 spin_unlock_irqrestore(&zone->lock, flags);
2127 seq_putc(m, '\n');
2128 }
2129 return 0;
2130}
2131
2132struct seq_operations fragmentation_op = {
2133 .start = frag_start,
2134 .next = frag_next,
2135 .stop = frag_stop,
2136 .show = frag_show,
2137};
2138
Nikita Danilov295ab932005-06-21 17:14:38 -07002139/*
2140 * Output information about zones in @pgdat.
2141 */
2142static int zoneinfo_show(struct seq_file *m, void *arg)
2143{
2144 pg_data_t *pgdat = arg;
2145 struct zone *zone;
2146 struct zone *node_zones = pgdat->node_zones;
2147 unsigned long flags;
2148
2149 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; zone++) {
2150 int i;
2151
2152 if (!zone->present_pages)
2153 continue;
2154
2155 spin_lock_irqsave(&zone->lock, flags);
2156 seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name);
2157 seq_printf(m,
2158 "\n pages free %lu"
2159 "\n min %lu"
2160 "\n low %lu"
2161 "\n high %lu"
2162 "\n active %lu"
2163 "\n inactive %lu"
2164 "\n scanned %lu (a: %lu i: %lu)"
2165 "\n spanned %lu"
2166 "\n present %lu",
2167 zone->free_pages,
2168 zone->pages_min,
2169 zone->pages_low,
2170 zone->pages_high,
2171 zone->nr_active,
2172 zone->nr_inactive,
2173 zone->pages_scanned,
2174 zone->nr_scan_active, zone->nr_scan_inactive,
2175 zone->spanned_pages,
2176 zone->present_pages);
2177 seq_printf(m,
2178 "\n protection: (%lu",
2179 zone->lowmem_reserve[0]);
2180 for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
2181 seq_printf(m, ", %lu", zone->lowmem_reserve[i]);
2182 seq_printf(m,
2183 ")"
2184 "\n pagesets");
2185 for (i = 0; i < ARRAY_SIZE(zone->pageset); i++) {
2186 struct per_cpu_pageset *pageset;
2187 int j;
2188
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002189 pageset = zone_pcp(zone, i);
Nikita Danilov295ab932005-06-21 17:14:38 -07002190 for (j = 0; j < ARRAY_SIZE(pageset->pcp); j++) {
2191 if (pageset->pcp[j].count)
2192 break;
2193 }
2194 if (j == ARRAY_SIZE(pageset->pcp))
2195 continue;
2196 for (j = 0; j < ARRAY_SIZE(pageset->pcp); j++) {
2197 seq_printf(m,
2198 "\n cpu: %i pcp: %i"
2199 "\n count: %i"
2200 "\n low: %i"
2201 "\n high: %i"
2202 "\n batch: %i",
2203 i, j,
2204 pageset->pcp[j].count,
2205 pageset->pcp[j].low,
2206 pageset->pcp[j].high,
2207 pageset->pcp[j].batch);
2208 }
2209#ifdef CONFIG_NUMA
2210 seq_printf(m,
2211 "\n numa_hit: %lu"
2212 "\n numa_miss: %lu"
2213 "\n numa_foreign: %lu"
2214 "\n interleave_hit: %lu"
2215 "\n local_node: %lu"
2216 "\n other_node: %lu",
2217 pageset->numa_hit,
2218 pageset->numa_miss,
2219 pageset->numa_foreign,
2220 pageset->interleave_hit,
2221 pageset->local_node,
2222 pageset->other_node);
2223#endif
2224 }
2225 seq_printf(m,
2226 "\n all_unreclaimable: %u"
2227 "\n prev_priority: %i"
2228 "\n temp_priority: %i"
2229 "\n start_pfn: %lu",
2230 zone->all_unreclaimable,
2231 zone->prev_priority,
2232 zone->temp_priority,
2233 zone->zone_start_pfn);
2234 spin_unlock_irqrestore(&zone->lock, flags);
2235 seq_putc(m, '\n');
2236 }
2237 return 0;
2238}
2239
2240struct seq_operations zoneinfo_op = {
2241 .start = frag_start, /* iterate over all zones. The same as in
2242 * fragmentation. */
2243 .next = frag_next,
2244 .stop = frag_stop,
2245 .show = zoneinfo_show,
2246};
2247
Linus Torvalds1da177e2005-04-16 15:20:36 -07002248static char *vmstat_text[] = {
2249 "nr_dirty",
2250 "nr_writeback",
2251 "nr_unstable",
2252 "nr_page_table_pages",
2253 "nr_mapped",
2254 "nr_slab",
2255
2256 "pgpgin",
2257 "pgpgout",
2258 "pswpin",
2259 "pswpout",
2260 "pgalloc_high",
2261
2262 "pgalloc_normal",
2263 "pgalloc_dma",
2264 "pgfree",
2265 "pgactivate",
2266 "pgdeactivate",
2267
2268 "pgfault",
2269 "pgmajfault",
2270 "pgrefill_high",
2271 "pgrefill_normal",
2272 "pgrefill_dma",
2273
2274 "pgsteal_high",
2275 "pgsteal_normal",
2276 "pgsteal_dma",
2277 "pgscan_kswapd_high",
2278 "pgscan_kswapd_normal",
2279
2280 "pgscan_kswapd_dma",
2281 "pgscan_direct_high",
2282 "pgscan_direct_normal",
2283 "pgscan_direct_dma",
2284 "pginodesteal",
2285
2286 "slabs_scanned",
2287 "kswapd_steal",
2288 "kswapd_inodesteal",
2289 "pageoutrun",
2290 "allocstall",
2291
2292 "pgrotated",
KAMEZAWA Hiroyukiedfbe2b2005-05-01 08:58:37 -07002293 "nr_bounce",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002294};
2295
2296static void *vmstat_start(struct seq_file *m, loff_t *pos)
2297{
2298 struct page_state *ps;
2299
2300 if (*pos >= ARRAY_SIZE(vmstat_text))
2301 return NULL;
2302
2303 ps = kmalloc(sizeof(*ps), GFP_KERNEL);
2304 m->private = ps;
2305 if (!ps)
2306 return ERR_PTR(-ENOMEM);
2307 get_full_page_state(ps);
2308 ps->pgpgin /= 2; /* sectors -> kbytes */
2309 ps->pgpgout /= 2;
2310 return (unsigned long *)ps + *pos;
2311}
2312
2313static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
2314{
2315 (*pos)++;
2316 if (*pos >= ARRAY_SIZE(vmstat_text))
2317 return NULL;
2318 return (unsigned long *)m->private + *pos;
2319}
2320
2321static int vmstat_show(struct seq_file *m, void *arg)
2322{
2323 unsigned long *l = arg;
2324 unsigned long off = l - (unsigned long *)m->private;
2325
2326 seq_printf(m, "%s %lu\n", vmstat_text[off], *l);
2327 return 0;
2328}
2329
2330static void vmstat_stop(struct seq_file *m, void *arg)
2331{
2332 kfree(m->private);
2333 m->private = NULL;
2334}
2335
2336struct seq_operations vmstat_op = {
2337 .start = vmstat_start,
2338 .next = vmstat_next,
2339 .stop = vmstat_stop,
2340 .show = vmstat_show,
2341};
2342
2343#endif /* CONFIG_PROC_FS */
2344
2345#ifdef CONFIG_HOTPLUG_CPU
2346static int page_alloc_cpu_notify(struct notifier_block *self,
2347 unsigned long action, void *hcpu)
2348{
2349 int cpu = (unsigned long)hcpu;
2350 long *count;
2351 unsigned long *src, *dest;
2352
2353 if (action == CPU_DEAD) {
2354 int i;
2355
2356 /* Drain local pagecache count. */
2357 count = &per_cpu(nr_pagecache_local, cpu);
2358 atomic_add(*count, &nr_pagecache);
2359 *count = 0;
2360 local_irq_disable();
2361 __drain_pages(cpu);
2362
2363 /* Add dead cpu's page_states to our own. */
2364 dest = (unsigned long *)&__get_cpu_var(page_states);
2365 src = (unsigned long *)&per_cpu(page_states, cpu);
2366
2367 for (i = 0; i < sizeof(struct page_state)/sizeof(unsigned long);
2368 i++) {
2369 dest[i] += src[i];
2370 src[i] = 0;
2371 }
2372
2373 local_irq_enable();
2374 }
2375 return NOTIFY_OK;
2376}
2377#endif /* CONFIG_HOTPLUG_CPU */
2378
2379void __init page_alloc_init(void)
2380{
2381 hotcpu_notifier(page_alloc_cpu_notify, 0);
2382}
2383
2384/*
2385 * setup_per_zone_lowmem_reserve - called whenever
2386 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
2387 * has a correct pages reserved value, so an adequate number of
2388 * pages are left in the zone after a successful __alloc_pages().
2389 */
2390static void setup_per_zone_lowmem_reserve(void)
2391{
2392 struct pglist_data *pgdat;
2393 int j, idx;
2394
2395 for_each_pgdat(pgdat) {
2396 for (j = 0; j < MAX_NR_ZONES; j++) {
2397 struct zone *zone = pgdat->node_zones + j;
2398 unsigned long present_pages = zone->present_pages;
2399
2400 zone->lowmem_reserve[j] = 0;
2401
2402 for (idx = j-1; idx >= 0; idx--) {
2403 struct zone *lower_zone;
2404
2405 if (sysctl_lowmem_reserve_ratio[idx] < 1)
2406 sysctl_lowmem_reserve_ratio[idx] = 1;
2407
2408 lower_zone = pgdat->node_zones + idx;
2409 lower_zone->lowmem_reserve[j] = present_pages /
2410 sysctl_lowmem_reserve_ratio[idx];
2411 present_pages += lower_zone->present_pages;
2412 }
2413 }
2414 }
2415}
2416
2417/*
2418 * setup_per_zone_pages_min - called when min_free_kbytes changes. Ensures
2419 * that the pages_{min,low,high} values for each zone are set correctly
2420 * with respect to min_free_kbytes.
2421 */
Dave Hansen3947be12005-10-29 18:16:54 -07002422void setup_per_zone_pages_min(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002423{
2424 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
2425 unsigned long lowmem_pages = 0;
2426 struct zone *zone;
2427 unsigned long flags;
2428
2429 /* Calculate total number of !ZONE_HIGHMEM pages */
2430 for_each_zone(zone) {
2431 if (!is_highmem(zone))
2432 lowmem_pages += zone->present_pages;
2433 }
2434
2435 for_each_zone(zone) {
Nick Piggin669ed172005-11-13 16:06:45 -08002436 unsigned long tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002437 spin_lock_irqsave(&zone->lru_lock, flags);
Nick Piggin669ed172005-11-13 16:06:45 -08002438 tmp = (pages_min * zone->present_pages) / lowmem_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002439 if (is_highmem(zone)) {
2440 /*
Nick Piggin669ed172005-11-13 16:06:45 -08002441 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
2442 * need highmem pages, so cap pages_min to a small
2443 * value here.
2444 *
2445 * The (pages_high-pages_low) and (pages_low-pages_min)
2446 * deltas controls asynch page reclaim, and so should
2447 * not be capped for highmem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002448 */
2449 int min_pages;
2450
2451 min_pages = zone->present_pages / 1024;
2452 if (min_pages < SWAP_CLUSTER_MAX)
2453 min_pages = SWAP_CLUSTER_MAX;
2454 if (min_pages > 128)
2455 min_pages = 128;
2456 zone->pages_min = min_pages;
2457 } else {
Nick Piggin669ed172005-11-13 16:06:45 -08002458 /*
2459 * If it's a lowmem zone, reserve a number of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07002460 * proportionate to the zone's size.
2461 */
Nick Piggin669ed172005-11-13 16:06:45 -08002462 zone->pages_min = tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002463 }
2464
Nick Piggin669ed172005-11-13 16:06:45 -08002465 zone->pages_low = zone->pages_min + tmp / 4;
2466 zone->pages_high = zone->pages_min + tmp / 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002467 spin_unlock_irqrestore(&zone->lru_lock, flags);
2468 }
2469}
2470
2471/*
2472 * Initialise min_free_kbytes.
2473 *
2474 * For small machines we want it small (128k min). For large machines
2475 * we want it large (64MB max). But it is not linear, because network
2476 * bandwidth does not increase linearly with machine size. We use
2477 *
2478 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
2479 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
2480 *
2481 * which yields
2482 *
2483 * 16MB: 512k
2484 * 32MB: 724k
2485 * 64MB: 1024k
2486 * 128MB: 1448k
2487 * 256MB: 2048k
2488 * 512MB: 2896k
2489 * 1024MB: 4096k
2490 * 2048MB: 5792k
2491 * 4096MB: 8192k
2492 * 8192MB: 11584k
2493 * 16384MB: 16384k
2494 */
2495static int __init init_per_zone_pages_min(void)
2496{
2497 unsigned long lowmem_kbytes;
2498
2499 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
2500
2501 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
2502 if (min_free_kbytes < 128)
2503 min_free_kbytes = 128;
2504 if (min_free_kbytes > 65536)
2505 min_free_kbytes = 65536;
2506 setup_per_zone_pages_min();
2507 setup_per_zone_lowmem_reserve();
2508 return 0;
2509}
2510module_init(init_per_zone_pages_min)
2511
2512/*
2513 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
2514 * that we can call two helper functions whenever min_free_kbytes
2515 * changes.
2516 */
2517int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
2518 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2519{
2520 proc_dointvec(table, write, file, buffer, length, ppos);
2521 setup_per_zone_pages_min();
2522 return 0;
2523}
2524
2525/*
2526 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
2527 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
2528 * whenever sysctl_lowmem_reserve_ratio changes.
2529 *
2530 * The reserve ratio obviously has absolutely no relation with the
2531 * pages_min watermarks. The lowmem reserve ratio can only make sense
2532 * if in function of the boot time zone sizes.
2533 */
2534int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
2535 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2536{
2537 proc_dointvec_minmax(table, write, file, buffer, length, ppos);
2538 setup_per_zone_lowmem_reserve();
2539 return 0;
2540}
2541
2542__initdata int hashdist = HASHDIST_DEFAULT;
2543
2544#ifdef CONFIG_NUMA
2545static int __init set_hashdist(char *str)
2546{
2547 if (!str)
2548 return 0;
2549 hashdist = simple_strtoul(str, &str, 0);
2550 return 1;
2551}
2552__setup("hashdist=", set_hashdist);
2553#endif
2554
2555/*
2556 * allocate a large system hash table from bootmem
2557 * - it is assumed that the hash table must contain an exact power-of-2
2558 * quantity of entries
2559 * - limit is the number of hash buckets, not the total allocation size
2560 */
2561void *__init alloc_large_system_hash(const char *tablename,
2562 unsigned long bucketsize,
2563 unsigned long numentries,
2564 int scale,
2565 int flags,
2566 unsigned int *_hash_shift,
2567 unsigned int *_hash_mask,
2568 unsigned long limit)
2569{
2570 unsigned long long max = limit;
2571 unsigned long log2qty, size;
2572 void *table = NULL;
2573
2574 /* allow the kernel cmdline to have a say */
2575 if (!numentries) {
2576 /* round applicable memory size up to nearest megabyte */
2577 numentries = (flags & HASH_HIGHMEM) ? nr_all_pages : nr_kernel_pages;
2578 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
2579 numentries >>= 20 - PAGE_SHIFT;
2580 numentries <<= 20 - PAGE_SHIFT;
2581
2582 /* limit to 1 bucket per 2^scale bytes of low memory */
2583 if (scale > PAGE_SHIFT)
2584 numentries >>= (scale - PAGE_SHIFT);
2585 else
2586 numentries <<= (PAGE_SHIFT - scale);
2587 }
2588 /* rounded up to nearest power of 2 in size */
2589 numentries = 1UL << (long_log2(numentries) + 1);
2590
2591 /* limit allocation size to 1/16 total memory by default */
2592 if (max == 0) {
2593 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
2594 do_div(max, bucketsize);
2595 }
2596
2597 if (numentries > max)
2598 numentries = max;
2599
2600 log2qty = long_log2(numentries);
2601
2602 do {
2603 size = bucketsize << log2qty;
2604 if (flags & HASH_EARLY)
2605 table = alloc_bootmem(size);
2606 else if (hashdist)
2607 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
2608 else {
2609 unsigned long order;
2610 for (order = 0; ((1UL << order) << PAGE_SHIFT) < size; order++)
2611 ;
2612 table = (void*) __get_free_pages(GFP_ATOMIC, order);
2613 }
2614 } while (!table && size > PAGE_SIZE && --log2qty);
2615
2616 if (!table)
2617 panic("Failed to allocate %s hash table\n", tablename);
2618
2619 printk("%s hash table entries: %d (order: %d, %lu bytes)\n",
2620 tablename,
2621 (1U << log2qty),
2622 long_log2(size) - PAGE_SHIFT,
2623 size);
2624
2625 if (_hash_shift)
2626 *_hash_shift = log2qty;
2627 if (_hash_mask)
2628 *_hash_mask = (1 << log2qty) - 1;
2629
2630 return table;
2631}