2 * linux/mm/page_alloc.c
4 * Manages the free list, the system allocates free pages here.
5 * Note that kmalloc() lives in slab.c
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)
17 #include <linux/config.h>
18 #include <linux/stddef.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>
25 #include <linux/kernel.h>
26 #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>
36 #include <linux/memory_hotplug.h>
37 #include <linux/nodemask.h>
38 #include <linux/vmalloc.h>
39 #include <linux/mempolicy.h>
41 #include <asm/tlbflush.h>
45 * MCD - HACK: Find somewhere to initialize this EARLY, or make this
48 nodemask_t node_online_map __read_mostly = { { [0] = 1UL } };
49 EXPORT_SYMBOL(node_online_map);
50 nodemask_t node_possible_map __read_mostly = NODE_MASK_ALL;
51 EXPORT_SYMBOL(node_possible_map);
52 unsigned long totalram_pages __read_mostly;
53 unsigned long totalhigh_pages __read_mostly;
55 int percpu_pagelist_fraction;
57 static void __free_pages_ok(struct page *page, unsigned int order);
60 * results with 256, 32 in the lowmem_reserve sysctl:
61 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
62 * 1G machine -> (16M dma, 784M normal, 224M high)
63 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
64 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
65 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
67 * TBD: should special case ZONE_DMA32 machines here - in those we normally
68 * don't need any ZONE_NORMAL reservation
70 int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = { 256, 256, 32 };
72 EXPORT_SYMBOL(totalram_pages);
75 * Used by page_zone() to look up the address of the struct zone whose
76 * id is encoded in the upper bits of page->flags
78 struct zone *zone_table[1 << ZONETABLE_SHIFT] __read_mostly;
79 EXPORT_SYMBOL(zone_table);
81 static char *zone_names[MAX_NR_ZONES] = { "DMA", "DMA32", "Normal", "HighMem" };
82 int min_free_kbytes = 1024;
84 unsigned long __initdata nr_kernel_pages;
85 unsigned long __initdata nr_all_pages;
87 #ifdef CONFIG_DEBUG_VM
88 static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
92 unsigned long pfn = page_to_pfn(page);
95 seq = zone_span_seqbegin(zone);
96 if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
98 else if (pfn < zone->zone_start_pfn)
100 } while (zone_span_seqretry(zone, seq));
105 static int page_is_consistent(struct zone *zone, struct page *page)
107 #ifdef CONFIG_HOLES_IN_ZONE
108 if (!pfn_valid(page_to_pfn(page)))
111 if (zone != page_zone(page))
117 * Temporary debugging check for pages not lying within a given zone.
119 static int bad_range(struct zone *zone, struct page *page)
121 if (page_outside_zone_boundaries(zone, page))
123 if (!page_is_consistent(zone, page))
130 static inline int bad_range(struct zone *zone, struct page *page)
136 static void bad_page(struct page *page)
138 printk(KERN_EMERG "Bad page state in process '%s'\n"
139 KERN_EMERG "page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n"
140 KERN_EMERG "Trying to fix it up, but a reboot is needed\n"
141 KERN_EMERG "Backtrace:\n",
142 current->comm, page, (int)(2*sizeof(unsigned long)),
143 (unsigned long)page->flags, page->mapping,
144 page_mapcount(page), page_count(page));
146 page->flags &= ~(1 << PG_lru |
156 set_page_count(page, 0);
157 reset_page_mapcount(page);
158 page->mapping = NULL;
159 add_taint(TAINT_BAD_PAGE);
163 * Higher-order pages are called "compound pages". They are structured thusly:
165 * The first PAGE_SIZE page is called the "head page".
167 * The remaining PAGE_SIZE pages are called "tail pages".
169 * All pages have PG_compound set. All pages have their ->private pointing at
170 * the head page (even the head page has this).
172 * The first tail page's ->lru.next holds the address of the compound page's
173 * put_page() function. Its ->lru.prev holds the order of allocation.
174 * This usage means that zero-order pages may not be compound.
177 static void free_compound_page(struct page *page)
179 __free_pages_ok(page, (unsigned long)page[1].lru.prev);
182 static void prep_compound_page(struct page *page, unsigned long order)
185 int nr_pages = 1 << order;
187 page[1].lru.next = (void *)free_compound_page; /* set dtor */
188 page[1].lru.prev = (void *)order;
189 for (i = 0; i < nr_pages; i++) {
190 struct page *p = page + i;
192 __SetPageCompound(p);
193 set_page_private(p, (unsigned long)page);
197 static void destroy_compound_page(struct page *page, unsigned long order)
200 int nr_pages = 1 << order;
202 if (unlikely((unsigned long)page[1].lru.prev != order))
205 for (i = 0; i < nr_pages; i++) {
206 struct page *p = page + i;
208 if (unlikely(!PageCompound(p) |
209 (page_private(p) != (unsigned long)page)))
211 __ClearPageCompound(p);
215 static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
219 BUG_ON((gfp_flags & (__GFP_WAIT | __GFP_HIGHMEM)) == __GFP_HIGHMEM);
221 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
222 * and __GFP_HIGHMEM from hard or soft interrupt context.
224 BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
225 for (i = 0; i < (1 << order); i++)
226 clear_highpage(page + i);
230 * function for dealing with page's order in buddy system.
231 * zone->lock is already acquired when we use these.
232 * So, we don't need atomic page->flags operations here.
234 static inline unsigned long page_order(struct page *page) {
235 return page_private(page);
238 static inline void set_page_order(struct page *page, int order) {
239 set_page_private(page, order);
240 __SetPageBuddy(page);
243 static inline void rmv_page_order(struct page *page)
245 __ClearPageBuddy(page);
246 set_page_private(page, 0);
250 * Locate the struct page for both the matching buddy in our
251 * pair (buddy1) and the combined O(n+1) page they form (page).
253 * 1) Any buddy B1 will have an order O twin B2 which satisfies
254 * the following equation:
256 * For example, if the starting buddy (buddy2) is #8 its order
258 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
260 * 2) Any buddy B will have an order O+1 parent P which
261 * satisfies the following equation:
264 * Assumption: *_mem_map is contigious at least up to MAX_ORDER
266 static inline struct page *
267 __page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order)
269 unsigned long buddy_idx = page_idx ^ (1 << order);
271 return page + (buddy_idx - page_idx);
274 static inline unsigned long
275 __find_combined_index(unsigned long page_idx, unsigned int order)
277 return (page_idx & ~(1 << order));
281 * This function checks whether a page is free && is the buddy
282 * we can do coalesce a page and its buddy if
283 * (a) the buddy is not in a hole &&
284 * (b) the buddy is in the buddy system &&
285 * (c) a page and its buddy have the same order.
287 * For recording whether a page is in the buddy system, we use PG_buddy.
288 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
290 * For recording page's order, we use page_private(page).
292 static inline int page_is_buddy(struct page *page, int order)
294 #ifdef CONFIG_HOLES_IN_ZONE
295 if (!pfn_valid(page_to_pfn(page)))
299 if (PageBuddy(page) && page_order(page) == order) {
300 BUG_ON(page_count(page) != 0);
307 * Freeing function for a buddy system allocator.
309 * The concept of a buddy system is to maintain direct-mapped table
310 * (containing bit values) for memory blocks of various "orders".
311 * The bottom level table contains the map for the smallest allocatable
312 * units of memory (here, pages), and each level above it describes
313 * pairs of units from the levels below, hence, "buddies".
314 * At a high level, all that happens here is marking the table entry
315 * at the bottom level available, and propagating the changes upward
316 * as necessary, plus some accounting needed to play nicely with other
317 * parts of the VM system.
318 * At each level, we keep a list of pages, which are heads of continuous
319 * free pages of length of (1 << order) and marked with PG_buddy. Page's
320 * order is recorded in page_private(page) field.
321 * So when we are allocating or freeing one, we can derive the state of the
322 * other. That is, if we allocate a small block, and both were
323 * free, the remainder of the region must be split into blocks.
324 * If a block is freed, and its buddy is also free, then this
325 * triggers coalescing into a block of larger size.
330 static inline void __free_one_page(struct page *page,
331 struct zone *zone, unsigned int order)
333 unsigned long page_idx;
334 int order_size = 1 << order;
336 if (unlikely(PageCompound(page)))
337 destroy_compound_page(page, order);
339 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
341 BUG_ON(page_idx & (order_size - 1));
342 BUG_ON(bad_range(zone, page));
344 zone->free_pages += order_size;
345 while (order < MAX_ORDER-1) {
346 unsigned long combined_idx;
347 struct free_area *area;
350 buddy = __page_find_buddy(page, page_idx, order);
351 if (!page_is_buddy(buddy, order))
352 break; /* Move the buddy up one level. */
354 list_del(&buddy->lru);
355 area = zone->free_area + order;
357 rmv_page_order(buddy);
358 combined_idx = __find_combined_index(page_idx, order);
359 page = page + (combined_idx - page_idx);
360 page_idx = combined_idx;
363 set_page_order(page, order);
364 list_add(&page->lru, &zone->free_area[order].free_list);
365 zone->free_area[order].nr_free++;
368 static inline int free_pages_check(struct page *page)
370 if (unlikely(page_mapcount(page) |
371 (page->mapping != NULL) |
372 (page_count(page) != 0) |
386 __ClearPageDirty(page);
388 * For now, we report if PG_reserved was found set, but do not
389 * clear it, and do not free the page. But we shall soon need
390 * to do more, for when the ZERO_PAGE count wraps negative.
392 return PageReserved(page);
396 * Frees a list of pages.
397 * Assumes all pages on list are in same zone, and of same order.
398 * count is the number of pages to free.
400 * If the zone was previously in an "all pages pinned" state then look to
401 * see if this freeing clears that state.
403 * And clear the zone's pages_scanned counter, to hold off the "all pages are
404 * pinned" detection logic.
406 static void free_pages_bulk(struct zone *zone, int count,
407 struct list_head *list, int order)
409 spin_lock(&zone->lock);
410 zone->all_unreclaimable = 0;
411 zone->pages_scanned = 0;
415 BUG_ON(list_empty(list));
416 page = list_entry(list->prev, struct page, lru);
417 /* have to delete it as __free_one_page list manipulates */
418 list_del(&page->lru);
419 __free_one_page(page, zone, order);
421 spin_unlock(&zone->lock);
424 static void free_one_page(struct zone *zone, struct page *page, int order)
427 list_add(&page->lru, &list);
428 free_pages_bulk(zone, 1, &list, order);
431 static void __free_pages_ok(struct page *page, unsigned int order)
437 arch_free_page(page, order);
438 if (!PageHighMem(page))
439 mutex_debug_check_no_locks_freed(page_address(page),
442 for (i = 0 ; i < (1 << order) ; ++i)
443 reserved += free_pages_check(page + i);
447 kernel_map_pages(page, 1 << order, 0);
448 local_irq_save(flags);
449 __mod_page_state(pgfree, 1 << order);
450 free_one_page(page_zone(page), page, order);
451 local_irq_restore(flags);
455 * permit the bootmem allocator to evade page validation on high-order frees
457 void fastcall __init __free_pages_bootmem(struct page *page, unsigned int order)
460 __ClearPageReserved(page);
461 set_page_count(page, 0);
462 set_page_refcounted(page);
468 for (loop = 0; loop < BITS_PER_LONG; loop++) {
469 struct page *p = &page[loop];
471 if (loop + 1 < BITS_PER_LONG)
473 __ClearPageReserved(p);
474 set_page_count(p, 0);
477 set_page_refcounted(page);
478 __free_pages(page, order);
484 * The order of subdivision here is critical for the IO subsystem.
485 * Please do not alter this order without good reasons and regression
486 * testing. Specifically, as large blocks of memory are subdivided,
487 * the order in which smaller blocks are delivered depends on the order
488 * they're subdivided in this function. This is the primary factor
489 * influencing the order in which pages are delivered to the IO
490 * subsystem according to empirical testing, and this is also justified
491 * by considering the behavior of a buddy system containing a single
492 * large block of memory acted on by a series of small allocations.
493 * This behavior is a critical factor in sglist merging's success.
497 static inline void expand(struct zone *zone, struct page *page,
498 int low, int high, struct free_area *area)
500 unsigned long size = 1 << high;
506 BUG_ON(bad_range(zone, &page[size]));
507 list_add(&page[size].lru, &area->free_list);
509 set_page_order(&page[size], high);
514 * This page is about to be returned from the page allocator
516 static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
518 if (unlikely(page_mapcount(page) |
519 (page->mapping != NULL) |
520 (page_count(page) != 0) |
536 * For now, we report if PG_reserved was found set, but do not
537 * clear it, and do not allocate the page: as a safety net.
539 if (PageReserved(page))
542 page->flags &= ~(1 << PG_uptodate | 1 << PG_error |
543 1 << PG_referenced | 1 << PG_arch_1 |
544 1 << PG_checked | 1 << PG_mappedtodisk);
545 set_page_private(page, 0);
546 set_page_refcounted(page);
547 kernel_map_pages(page, 1 << order, 1);
549 if (gfp_flags & __GFP_ZERO)
550 prep_zero_page(page, order, gfp_flags);
552 if (order && (gfp_flags & __GFP_COMP))
553 prep_compound_page(page, order);
559 * Do the hard work of removing an element from the buddy allocator.
560 * Call me with the zone->lock already held.
562 static struct page *__rmqueue(struct zone *zone, unsigned int order)
564 struct free_area * area;
565 unsigned int current_order;
568 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
569 area = zone->free_area + current_order;
570 if (list_empty(&area->free_list))
573 page = list_entry(area->free_list.next, struct page, lru);
574 list_del(&page->lru);
575 rmv_page_order(page);
577 zone->free_pages -= 1UL << order;
578 expand(zone, page, order, current_order, area);
586 * Obtain a specified number of elements from the buddy allocator, all under
587 * a single hold of the lock, for efficiency. Add them to the supplied list.
588 * Returns the number of new pages which were placed at *list.
590 static int rmqueue_bulk(struct zone *zone, unsigned int order,
591 unsigned long count, struct list_head *list)
595 spin_lock(&zone->lock);
596 for (i = 0; i < count; ++i) {
597 struct page *page = __rmqueue(zone, order);
598 if (unlikely(page == NULL))
600 list_add_tail(&page->lru, list);
602 spin_unlock(&zone->lock);
608 * Called from the slab reaper to drain pagesets on a particular node that
609 * belong to the currently executing processor.
610 * Note that this function must be called with the thread pinned to
611 * a single processor.
613 void drain_node_pages(int nodeid)
618 for (z = 0; z < MAX_NR_ZONES; z++) {
619 struct zone *zone = NODE_DATA(nodeid)->node_zones + z;
620 struct per_cpu_pageset *pset;
622 pset = zone_pcp(zone, smp_processor_id());
623 for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) {
624 struct per_cpu_pages *pcp;
628 local_irq_save(flags);
629 free_pages_bulk(zone, pcp->count, &pcp->list, 0);
631 local_irq_restore(flags);
638 #if defined(CONFIG_PM) || defined(CONFIG_HOTPLUG_CPU)
639 static void __drain_pages(unsigned int cpu)
645 for_each_zone(zone) {
646 struct per_cpu_pageset *pset;
648 pset = zone_pcp(zone, cpu);
649 for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) {
650 struct per_cpu_pages *pcp;
653 local_irq_save(flags);
654 free_pages_bulk(zone, pcp->count, &pcp->list, 0);
656 local_irq_restore(flags);
660 #endif /* CONFIG_PM || CONFIG_HOTPLUG_CPU */
664 void mark_free_pages(struct zone *zone)
666 unsigned long zone_pfn, flags;
668 struct list_head *curr;
670 if (!zone->spanned_pages)
673 spin_lock_irqsave(&zone->lock, flags);
674 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn)
675 ClearPageNosaveFree(pfn_to_page(zone_pfn + zone->zone_start_pfn));
677 for (order = MAX_ORDER - 1; order >= 0; --order)
678 list_for_each(curr, &zone->free_area[order].free_list) {
679 unsigned long start_pfn, i;
681 start_pfn = page_to_pfn(list_entry(curr, struct page, lru));
683 for (i=0; i < (1<<order); i++)
684 SetPageNosaveFree(pfn_to_page(start_pfn+i));
686 spin_unlock_irqrestore(&zone->lock, flags);
690 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
692 void drain_local_pages(void)
696 local_irq_save(flags);
697 __drain_pages(smp_processor_id());
698 local_irq_restore(flags);
700 #endif /* CONFIG_PM */
702 static void zone_statistics(struct zonelist *zonelist, struct zone *z, int cpu)
705 pg_data_t *pg = z->zone_pgdat;
706 pg_data_t *orig = zonelist->zones[0]->zone_pgdat;
707 struct per_cpu_pageset *p;
709 p = zone_pcp(z, cpu);
714 zone_pcp(zonelist->zones[0], cpu)->numa_foreign++;
716 if (pg == NODE_DATA(numa_node_id()))
724 * Free a 0-order page
726 static void fastcall free_hot_cold_page(struct page *page, int cold)
728 struct zone *zone = page_zone(page);
729 struct per_cpu_pages *pcp;
732 arch_free_page(page, 0);
735 page->mapping = NULL;
736 if (free_pages_check(page))
739 kernel_map_pages(page, 1, 0);
741 pcp = &zone_pcp(zone, get_cpu())->pcp[cold];
742 local_irq_save(flags);
743 __inc_page_state(pgfree);
744 list_add(&page->lru, &pcp->list);
746 if (pcp->count >= pcp->high) {
747 free_pages_bulk(zone, pcp->batch, &pcp->list, 0);
748 pcp->count -= pcp->batch;
750 local_irq_restore(flags);
754 void fastcall free_hot_page(struct page *page)
756 free_hot_cold_page(page, 0);
759 void fastcall free_cold_page(struct page *page)
761 free_hot_cold_page(page, 1);
765 * split_page takes a non-compound higher-order page, and splits it into
766 * n (1<<order) sub-pages: page[0..n]
767 * Each sub-page must be freed individually.
769 * Note: this is probably too low level an operation for use in drivers.
770 * Please consult with lkml before using this in your driver.
772 void split_page(struct page *page, unsigned int order)
776 BUG_ON(PageCompound(page));
777 BUG_ON(!page_count(page));
778 for (i = 1; i < (1 << order); i++)
779 set_page_refcounted(page + i);
783 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
784 * we cheat by calling it from here, in the order > 0 path. Saves a branch
787 static struct page *buffered_rmqueue(struct zonelist *zonelist,
788 struct zone *zone, int order, gfp_t gfp_flags)
792 int cold = !!(gfp_flags & __GFP_COLD);
797 if (likely(order == 0)) {
798 struct per_cpu_pages *pcp;
800 pcp = &zone_pcp(zone, cpu)->pcp[cold];
801 local_irq_save(flags);
803 pcp->count += rmqueue_bulk(zone, 0,
804 pcp->batch, &pcp->list);
805 if (unlikely(!pcp->count))
808 page = list_entry(pcp->list.next, struct page, lru);
809 list_del(&page->lru);
812 spin_lock_irqsave(&zone->lock, flags);
813 page = __rmqueue(zone, order);
814 spin_unlock(&zone->lock);
819 __mod_page_state_zone(zone, pgalloc, 1 << order);
820 zone_statistics(zonelist, zone, cpu);
821 local_irq_restore(flags);
824 BUG_ON(bad_range(zone, page));
825 if (prep_new_page(page, order, gfp_flags))
830 local_irq_restore(flags);
835 #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
836 #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
837 #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
838 #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
839 #define ALLOC_HARDER 0x10 /* try to alloc harder */
840 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
841 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
844 * Return 1 if free pages are above 'mark'. This takes into account the order
847 int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
848 int classzone_idx, int alloc_flags)
850 /* free_pages my go negative - that's OK */
851 long min = mark, free_pages = z->free_pages - (1 << order) + 1;
854 if (alloc_flags & ALLOC_HIGH)
856 if (alloc_flags & ALLOC_HARDER)
859 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
861 for (o = 0; o < order; o++) {
862 /* At the next order, this order's pages become unavailable */
863 free_pages -= z->free_area[o].nr_free << o;
865 /* Require fewer higher order pages to be free */
868 if (free_pages <= min)
875 * get_page_from_freeliest goes through the zonelist trying to allocate
879 get_page_from_freelist(gfp_t gfp_mask, unsigned int order,
880 struct zonelist *zonelist, int alloc_flags)
882 struct zone **z = zonelist->zones;
883 struct page *page = NULL;
884 int classzone_idx = zone_idx(*z);
887 * Go through the zonelist once, looking for a zone with enough free.
888 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
891 if ((alloc_flags & ALLOC_CPUSET) &&
892 !cpuset_zone_allowed(*z, gfp_mask))
895 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
897 if (alloc_flags & ALLOC_WMARK_MIN)
898 mark = (*z)->pages_min;
899 else if (alloc_flags & ALLOC_WMARK_LOW)
900 mark = (*z)->pages_low;
902 mark = (*z)->pages_high;
903 if (!zone_watermark_ok(*z, order, mark,
904 classzone_idx, alloc_flags))
905 if (!zone_reclaim_mode ||
906 !zone_reclaim(*z, gfp_mask, order))
910 page = buffered_rmqueue(zonelist, *z, order, gfp_mask);
914 } while (*(++z) != NULL);
919 * This is the 'heart' of the zoned buddy allocator.
921 struct page * fastcall
922 __alloc_pages(gfp_t gfp_mask, unsigned int order,
923 struct zonelist *zonelist)
925 const gfp_t wait = gfp_mask & __GFP_WAIT;
928 struct reclaim_state reclaim_state;
929 struct task_struct *p = current;
932 int did_some_progress;
934 might_sleep_if(wait);
937 z = zonelist->zones; /* the list of zones suitable for gfp_mask */
939 if (unlikely(*z == NULL)) {
940 /* Should this ever happen?? */
944 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
945 zonelist, ALLOC_WMARK_LOW|ALLOC_CPUSET);
950 if (cpuset_zone_allowed(*z, gfp_mask))
951 wakeup_kswapd(*z, order);
955 * OK, we're below the kswapd watermark and have kicked background
956 * reclaim. Now things get more complex, so set up alloc_flags according
957 * to how we want to proceed.
959 * The caller may dip into page reserves a bit more if the caller
960 * cannot run direct reclaim, or if the caller has realtime scheduling
961 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
962 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
964 alloc_flags = ALLOC_WMARK_MIN;
965 if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait)
966 alloc_flags |= ALLOC_HARDER;
967 if (gfp_mask & __GFP_HIGH)
968 alloc_flags |= ALLOC_HIGH;
969 alloc_flags |= ALLOC_CPUSET;
972 * Go through the zonelist again. Let __GFP_HIGH and allocations
973 * coming from realtime tasks go deeper into reserves.
975 * This is the last chance, in general, before the goto nopage.
976 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
977 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
979 page = get_page_from_freelist(gfp_mask, order, zonelist, alloc_flags);
983 /* This allocation should allow future memory freeing. */
985 if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE)))
986 && !in_interrupt()) {
987 if (!(gfp_mask & __GFP_NOMEMALLOC)) {
989 /* go through the zonelist yet again, ignoring mins */
990 page = get_page_from_freelist(gfp_mask, order,
991 zonelist, ALLOC_NO_WATERMARKS);
994 if (gfp_mask & __GFP_NOFAIL) {
995 blk_congestion_wait(WRITE, HZ/50);
1002 /* Atomic allocations - we can't balance anything */
1009 /* We now go into synchronous reclaim */
1010 cpuset_memory_pressure_bump();
1011 p->flags |= PF_MEMALLOC;
1012 reclaim_state.reclaimed_slab = 0;
1013 p->reclaim_state = &reclaim_state;
1015 did_some_progress = try_to_free_pages(zonelist->zones, gfp_mask);
1017 p->reclaim_state = NULL;
1018 p->flags &= ~PF_MEMALLOC;
1022 if (likely(did_some_progress)) {
1023 page = get_page_from_freelist(gfp_mask, order,
1024 zonelist, alloc_flags);
1027 } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
1029 * Go through the zonelist yet one more time, keep
1030 * very high watermark here, this is only to catch
1031 * a parallel oom killing, we must fail if we're still
1032 * under heavy pressure.
1034 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
1035 zonelist, ALLOC_WMARK_HIGH|ALLOC_CPUSET);
1039 out_of_memory(zonelist, gfp_mask, order);
1044 * Don't let big-order allocations loop unless the caller explicitly
1045 * requests that. Wait for some write requests to complete then retry.
1047 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
1048 * <= 3, but that may not be true in other implementations.
1051 if (!(gfp_mask & __GFP_NORETRY)) {
1052 if ((order <= 3) || (gfp_mask & __GFP_REPEAT))
1054 if (gfp_mask & __GFP_NOFAIL)
1058 blk_congestion_wait(WRITE, HZ/50);
1063 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
1064 printk(KERN_WARNING "%s: page allocation failure."
1065 " order:%d, mode:0x%x\n",
1066 p->comm, order, gfp_mask);
1074 EXPORT_SYMBOL(__alloc_pages);
1077 * Common helper functions.
1079 fastcall unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
1082 page = alloc_pages(gfp_mask, order);
1085 return (unsigned long) page_address(page);
1088 EXPORT_SYMBOL(__get_free_pages);
1090 fastcall unsigned long get_zeroed_page(gfp_t gfp_mask)
1095 * get_zeroed_page() returns a 32-bit address, which cannot represent
1098 BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
1100 page = alloc_pages(gfp_mask | __GFP_ZERO, 0);
1102 return (unsigned long) page_address(page);
1106 EXPORT_SYMBOL(get_zeroed_page);
1108 void __pagevec_free(struct pagevec *pvec)
1110 int i = pagevec_count(pvec);
1113 free_hot_cold_page(pvec->pages[i], pvec->cold);
1116 fastcall void __free_pages(struct page *page, unsigned int order)
1118 if (put_page_testzero(page)) {
1120 free_hot_page(page);
1122 __free_pages_ok(page, order);
1126 EXPORT_SYMBOL(__free_pages);
1128 fastcall void free_pages(unsigned long addr, unsigned int order)
1131 BUG_ON(!virt_addr_valid((void *)addr));
1132 __free_pages(virt_to_page((void *)addr), order);
1136 EXPORT_SYMBOL(free_pages);
1139 * Total amount of free (allocatable) RAM:
1141 unsigned int nr_free_pages(void)
1143 unsigned int sum = 0;
1147 sum += zone->free_pages;
1152 EXPORT_SYMBOL(nr_free_pages);
1155 unsigned int nr_free_pages_pgdat(pg_data_t *pgdat)
1157 unsigned int i, sum = 0;
1159 for (i = 0; i < MAX_NR_ZONES; i++)
1160 sum += pgdat->node_zones[i].free_pages;
1166 static unsigned int nr_free_zone_pages(int offset)
1168 /* Just pick one node, since fallback list is circular */
1169 pg_data_t *pgdat = NODE_DATA(numa_node_id());
1170 unsigned int sum = 0;
1172 struct zonelist *zonelist = pgdat->node_zonelists + offset;
1173 struct zone **zonep = zonelist->zones;
1176 for (zone = *zonep++; zone; zone = *zonep++) {
1177 unsigned long size = zone->present_pages;
1178 unsigned long high = zone->pages_high;
1187 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1189 unsigned int nr_free_buffer_pages(void)
1191 return nr_free_zone_pages(gfp_zone(GFP_USER));
1195 * Amount of free RAM allocatable within all zones
1197 unsigned int nr_free_pagecache_pages(void)
1199 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER));
1202 #ifdef CONFIG_HIGHMEM
1203 unsigned int nr_free_highpages (void)
1206 unsigned int pages = 0;
1208 for_each_online_pgdat(pgdat)
1209 pages += pgdat->node_zones[ZONE_HIGHMEM].free_pages;
1216 static void show_node(struct zone *zone)
1218 printk("Node %d ", zone->zone_pgdat->node_id);
1221 #define show_node(zone) do { } while (0)
1225 * Accumulate the page_state information across all CPUs.
1226 * The result is unavoidably approximate - it can change
1227 * during and after execution of this function.
1229 static DEFINE_PER_CPU(struct page_state, page_states) = {0};
1231 atomic_t nr_pagecache = ATOMIC_INIT(0);
1232 EXPORT_SYMBOL(nr_pagecache);
1234 DEFINE_PER_CPU(long, nr_pagecache_local) = 0;
1237 static void __get_page_state(struct page_state *ret, int nr, cpumask_t *cpumask)
1241 memset(ret, 0, nr * sizeof(unsigned long));
1242 cpus_and(*cpumask, *cpumask, cpu_online_map);
1244 for_each_cpu_mask(cpu, *cpumask) {
1250 in = (unsigned long *)&per_cpu(page_states, cpu);
1252 next_cpu = next_cpu(cpu, *cpumask);
1253 if (likely(next_cpu < NR_CPUS))
1254 prefetch(&per_cpu(page_states, next_cpu));
1256 out = (unsigned long *)ret;
1257 for (off = 0; off < nr; off++)
1262 void get_page_state_node(struct page_state *ret, int node)
1265 cpumask_t mask = node_to_cpumask(node);
1267 nr = offsetof(struct page_state, GET_PAGE_STATE_LAST);
1268 nr /= sizeof(unsigned long);
1270 __get_page_state(ret, nr+1, &mask);
1273 void get_page_state(struct page_state *ret)
1276 cpumask_t mask = CPU_MASK_ALL;
1278 nr = offsetof(struct page_state, GET_PAGE_STATE_LAST);
1279 nr /= sizeof(unsigned long);
1281 __get_page_state(ret, nr + 1, &mask);
1284 void get_full_page_state(struct page_state *ret)
1286 cpumask_t mask = CPU_MASK_ALL;
1288 __get_page_state(ret, sizeof(*ret) / sizeof(unsigned long), &mask);
1291 unsigned long read_page_state_offset(unsigned long offset)
1293 unsigned long ret = 0;
1296 for_each_online_cpu(cpu) {
1299 in = (unsigned long)&per_cpu(page_states, cpu) + offset;
1300 ret += *((unsigned long *)in);
1305 void __mod_page_state_offset(unsigned long offset, unsigned long delta)
1309 ptr = &__get_cpu_var(page_states);
1310 *(unsigned long *)(ptr + offset) += delta;
1312 EXPORT_SYMBOL(__mod_page_state_offset);
1314 void mod_page_state_offset(unsigned long offset, unsigned long delta)
1316 unsigned long flags;
1319 local_irq_save(flags);
1320 ptr = &__get_cpu_var(page_states);
1321 *(unsigned long *)(ptr + offset) += delta;
1322 local_irq_restore(flags);
1324 EXPORT_SYMBOL(mod_page_state_offset);
1326 void __get_zone_counts(unsigned long *active, unsigned long *inactive,
1327 unsigned long *free, struct pglist_data *pgdat)
1329 struct zone *zones = pgdat->node_zones;
1335 for (i = 0; i < MAX_NR_ZONES; i++) {
1336 *active += zones[i].nr_active;
1337 *inactive += zones[i].nr_inactive;
1338 *free += zones[i].free_pages;
1342 void get_zone_counts(unsigned long *active,
1343 unsigned long *inactive, unsigned long *free)
1345 struct pglist_data *pgdat;
1350 for_each_online_pgdat(pgdat) {
1351 unsigned long l, m, n;
1352 __get_zone_counts(&l, &m, &n, pgdat);
1359 void si_meminfo(struct sysinfo *val)
1361 val->totalram = totalram_pages;
1363 val->freeram = nr_free_pages();
1364 val->bufferram = nr_blockdev_pages();
1365 #ifdef CONFIG_HIGHMEM
1366 val->totalhigh = totalhigh_pages;
1367 val->freehigh = nr_free_highpages();
1372 val->mem_unit = PAGE_SIZE;
1375 EXPORT_SYMBOL(si_meminfo);
1378 void si_meminfo_node(struct sysinfo *val, int nid)
1380 pg_data_t *pgdat = NODE_DATA(nid);
1382 val->totalram = pgdat->node_present_pages;
1383 val->freeram = nr_free_pages_pgdat(pgdat);
1384 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1385 val->freehigh = pgdat->node_zones[ZONE_HIGHMEM].free_pages;
1386 val->mem_unit = PAGE_SIZE;
1390 #define K(x) ((x) << (PAGE_SHIFT-10))
1393 * Show free area list (used inside shift_scroll-lock stuff)
1394 * We also calculate the percentage fragmentation. We do this by counting the
1395 * memory on each free list with the exception of the first item on the list.
1397 void show_free_areas(void)
1399 struct page_state ps;
1400 int cpu, temperature;
1401 unsigned long active;
1402 unsigned long inactive;
1406 for_each_zone(zone) {
1408 printk("%s per-cpu:", zone->name);
1410 if (!populated_zone(zone)) {
1416 for_each_online_cpu(cpu) {
1417 struct per_cpu_pageset *pageset;
1419 pageset = zone_pcp(zone, cpu);
1421 for (temperature = 0; temperature < 2; temperature++)
1422 printk("cpu %d %s: high %d, batch %d used:%d\n",
1424 temperature ? "cold" : "hot",
1425 pageset->pcp[temperature].high,
1426 pageset->pcp[temperature].batch,
1427 pageset->pcp[temperature].count);
1431 get_page_state(&ps);
1432 get_zone_counts(&active, &inactive, &free);
1434 printk("Free pages: %11ukB (%ukB HighMem)\n",
1436 K(nr_free_highpages()));
1438 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu "
1439 "unstable:%lu free:%u slab:%lu mapped:%lu pagetables:%lu\n",
1448 ps.nr_page_table_pages);
1450 for_each_zone(zone) {
1462 " pages_scanned:%lu"
1463 " all_unreclaimable? %s"
1466 K(zone->free_pages),
1469 K(zone->pages_high),
1471 K(zone->nr_inactive),
1472 K(zone->present_pages),
1473 zone->pages_scanned,
1474 (zone->all_unreclaimable ? "yes" : "no")
1476 printk("lowmem_reserve[]:");
1477 for (i = 0; i < MAX_NR_ZONES; i++)
1478 printk(" %lu", zone->lowmem_reserve[i]);
1482 for_each_zone(zone) {
1483 unsigned long nr, flags, order, total = 0;
1486 printk("%s: ", zone->name);
1487 if (!populated_zone(zone)) {
1492 spin_lock_irqsave(&zone->lock, flags);
1493 for (order = 0; order < MAX_ORDER; order++) {
1494 nr = zone->free_area[order].nr_free;
1495 total += nr << order;
1496 printk("%lu*%lukB ", nr, K(1UL) << order);
1498 spin_unlock_irqrestore(&zone->lock, flags);
1499 printk("= %lukB\n", K(total));
1502 show_swap_cache_info();
1506 * Builds allocation fallback zone lists.
1508 * Add all populated zones of a node to the zonelist.
1510 static int __init build_zonelists_node(pg_data_t *pgdat,
1511 struct zonelist *zonelist, int nr_zones, int zone_type)
1515 BUG_ON(zone_type > ZONE_HIGHMEM);
1518 zone = pgdat->node_zones + zone_type;
1519 if (populated_zone(zone)) {
1520 #ifndef CONFIG_HIGHMEM
1521 BUG_ON(zone_type > ZONE_NORMAL);
1523 zonelist->zones[nr_zones++] = zone;
1524 check_highest_zone(zone_type);
1528 } while (zone_type >= 0);
1532 static inline int highest_zone(int zone_bits)
1534 int res = ZONE_NORMAL;
1535 if (zone_bits & (__force int)__GFP_HIGHMEM)
1537 if (zone_bits & (__force int)__GFP_DMA32)
1539 if (zone_bits & (__force int)__GFP_DMA)
1545 #define MAX_NODE_LOAD (num_online_nodes())
1546 static int __initdata node_load[MAX_NUMNODES];
1548 * find_next_best_node - find the next node that should appear in a given node's fallback list
1549 * @node: node whose fallback list we're appending
1550 * @used_node_mask: nodemask_t of already used nodes
1552 * We use a number of factors to determine which is the next node that should
1553 * appear on a given node's fallback list. The node should not have appeared
1554 * already in @node's fallback list, and it should be the next closest node
1555 * according to the distance array (which contains arbitrary distance values
1556 * from each node to each node in the system), and should also prefer nodes
1557 * with no CPUs, since presumably they'll have very little allocation pressure
1558 * on them otherwise.
1559 * It returns -1 if no node is found.
1561 static int __init find_next_best_node(int node, nodemask_t *used_node_mask)
1564 int min_val = INT_MAX;
1567 /* Use the local node if we haven't already */
1568 if (!node_isset(node, *used_node_mask)) {
1569 node_set(node, *used_node_mask);
1573 for_each_online_node(n) {
1576 /* Don't want a node to appear more than once */
1577 if (node_isset(n, *used_node_mask))
1580 /* Use the distance array to find the distance */
1581 val = node_distance(node, n);
1583 /* Penalize nodes under us ("prefer the next node") */
1586 /* Give preference to headless and unused nodes */
1587 tmp = node_to_cpumask(n);
1588 if (!cpus_empty(tmp))
1589 val += PENALTY_FOR_NODE_WITH_CPUS;
1591 /* Slight preference for less loaded node */
1592 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
1593 val += node_load[n];
1595 if (val < min_val) {
1602 node_set(best_node, *used_node_mask);
1607 static void __init build_zonelists(pg_data_t *pgdat)
1609 int i, j, k, node, local_node;
1610 int prev_node, load;
1611 struct zonelist *zonelist;
1612 nodemask_t used_mask;
1614 /* initialize zonelists */
1615 for (i = 0; i < GFP_ZONETYPES; i++) {
1616 zonelist = pgdat->node_zonelists + i;
1617 zonelist->zones[0] = NULL;
1620 /* NUMA-aware ordering of nodes */
1621 local_node = pgdat->node_id;
1622 load = num_online_nodes();
1623 prev_node = local_node;
1624 nodes_clear(used_mask);
1625 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
1626 int distance = node_distance(local_node, node);
1629 * If another node is sufficiently far away then it is better
1630 * to reclaim pages in a zone before going off node.
1632 if (distance > RECLAIM_DISTANCE)
1633 zone_reclaim_mode = 1;
1636 * We don't want to pressure a particular node.
1637 * So adding penalty to the first node in same
1638 * distance group to make it round-robin.
1641 if (distance != node_distance(local_node, prev_node))
1642 node_load[node] += load;
1645 for (i = 0; i < GFP_ZONETYPES; i++) {
1646 zonelist = pgdat->node_zonelists + i;
1647 for (j = 0; zonelist->zones[j] != NULL; j++);
1649 k = highest_zone(i);
1651 j = build_zonelists_node(NODE_DATA(node), zonelist, j, k);
1652 zonelist->zones[j] = NULL;
1657 #else /* CONFIG_NUMA */
1659 static void __init build_zonelists(pg_data_t *pgdat)
1661 int i, j, k, node, local_node;
1663 local_node = pgdat->node_id;
1664 for (i = 0; i < GFP_ZONETYPES; i++) {
1665 struct zonelist *zonelist;
1667 zonelist = pgdat->node_zonelists + i;
1670 k = highest_zone(i);
1671 j = build_zonelists_node(pgdat, zonelist, j, k);
1673 * Now we build the zonelist so that it contains the zones
1674 * of all the other nodes.
1675 * We don't want to pressure a particular node, so when
1676 * building the zones for node N, we make sure that the
1677 * zones coming right after the local ones are those from
1678 * node N+1 (modulo N)
1680 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
1681 if (!node_online(node))
1683 j = build_zonelists_node(NODE_DATA(node), zonelist, j, k);
1685 for (node = 0; node < local_node; node++) {
1686 if (!node_online(node))
1688 j = build_zonelists_node(NODE_DATA(node), zonelist, j, k);
1691 zonelist->zones[j] = NULL;
1695 #endif /* CONFIG_NUMA */
1697 void __init build_all_zonelists(void)
1701 for_each_online_node(i)
1702 build_zonelists(NODE_DATA(i));
1703 printk("Built %i zonelists\n", num_online_nodes());
1704 cpuset_init_current_mems_allowed();
1708 * Helper functions to size the waitqueue hash table.
1709 * Essentially these want to choose hash table sizes sufficiently
1710 * large so that collisions trying to wait on pages are rare.
1711 * But in fact, the number of active page waitqueues on typical
1712 * systems is ridiculously low, less than 200. So this is even
1713 * conservative, even though it seems large.
1715 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
1716 * waitqueues, i.e. the size of the waitq table given the number of pages.
1718 #define PAGES_PER_WAITQUEUE 256
1720 static inline unsigned long wait_table_size(unsigned long pages)
1722 unsigned long size = 1;
1724 pages /= PAGES_PER_WAITQUEUE;
1726 while (size < pages)
1730 * Once we have dozens or even hundreds of threads sleeping
1731 * on IO we've got bigger problems than wait queue collision.
1732 * Limit the size of the wait table to a reasonable size.
1734 size = min(size, 4096UL);
1736 return max(size, 4UL);
1740 * This is an integer logarithm so that shifts can be used later
1741 * to extract the more random high bits from the multiplicative
1742 * hash function before the remainder is taken.
1744 static inline unsigned long wait_table_bits(unsigned long size)
1749 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
1751 static void __init calculate_zone_totalpages(struct pglist_data *pgdat,
1752 unsigned long *zones_size, unsigned long *zholes_size)
1754 unsigned long realtotalpages, totalpages = 0;
1757 for (i = 0; i < MAX_NR_ZONES; i++)
1758 totalpages += zones_size[i];
1759 pgdat->node_spanned_pages = totalpages;
1761 realtotalpages = totalpages;
1763 for (i = 0; i < MAX_NR_ZONES; i++)
1764 realtotalpages -= zholes_size[i];
1765 pgdat->node_present_pages = realtotalpages;
1766 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, realtotalpages);
1771 * Initially all pages are reserved - free ones are freed
1772 * up by free_all_bootmem() once the early boot process is
1773 * done. Non-atomic initialization, single-pass.
1775 void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
1776 unsigned long start_pfn)
1779 unsigned long end_pfn = start_pfn + size;
1782 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1783 if (!early_pfn_valid(pfn))
1785 page = pfn_to_page(pfn);
1786 set_page_links(page, zone, nid, pfn);
1787 init_page_count(page);
1788 reset_page_mapcount(page);
1789 SetPageReserved(page);
1790 INIT_LIST_HEAD(&page->lru);
1791 #ifdef WANT_PAGE_VIRTUAL
1792 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
1793 if (!is_highmem_idx(zone))
1794 set_page_address(page, __va(pfn << PAGE_SHIFT));
1799 void zone_init_free_lists(struct pglist_data *pgdat, struct zone *zone,
1803 for (order = 0; order < MAX_ORDER ; order++) {
1804 INIT_LIST_HEAD(&zone->free_area[order].free_list);
1805 zone->free_area[order].nr_free = 0;
1809 #define ZONETABLE_INDEX(x, zone_nr) ((x << ZONES_SHIFT) | zone_nr)
1810 void zonetable_add(struct zone *zone, int nid, int zid, unsigned long pfn,
1813 unsigned long snum = pfn_to_section_nr(pfn);
1814 unsigned long end = pfn_to_section_nr(pfn + size);
1817 zone_table[ZONETABLE_INDEX(nid, zid)] = zone;
1819 for (; snum <= end; snum++)
1820 zone_table[ZONETABLE_INDEX(snum, zid)] = zone;
1823 #ifndef __HAVE_ARCH_MEMMAP_INIT
1824 #define memmap_init(size, nid, zone, start_pfn) \
1825 memmap_init_zone((size), (nid), (zone), (start_pfn))
1828 static int __cpuinit zone_batchsize(struct zone *zone)
1833 * The per-cpu-pages pools are set to around 1000th of the
1834 * size of the zone. But no more than 1/2 of a meg.
1836 * OK, so we don't know how big the cache is. So guess.
1838 batch = zone->present_pages / 1024;
1839 if (batch * PAGE_SIZE > 512 * 1024)
1840 batch = (512 * 1024) / PAGE_SIZE;
1841 batch /= 4; /* We effectively *= 4 below */
1846 * Clamp the batch to a 2^n - 1 value. Having a power
1847 * of 2 value was found to be more likely to have
1848 * suboptimal cache aliasing properties in some cases.
1850 * For example if 2 tasks are alternately allocating
1851 * batches of pages, one task can end up with a lot
1852 * of pages of one half of the possible page colors
1853 * and the other with pages of the other colors.
1855 batch = (1 << (fls(batch + batch/2)-1)) - 1;
1860 inline void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
1862 struct per_cpu_pages *pcp;
1864 memset(p, 0, sizeof(*p));
1866 pcp = &p->pcp[0]; /* hot */
1868 pcp->high = 6 * batch;
1869 pcp->batch = max(1UL, 1 * batch);
1870 INIT_LIST_HEAD(&pcp->list);
1872 pcp = &p->pcp[1]; /* cold*/
1874 pcp->high = 2 * batch;
1875 pcp->batch = max(1UL, batch/2);
1876 INIT_LIST_HEAD(&pcp->list);
1880 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
1881 * to the value high for the pageset p.
1884 static void setup_pagelist_highmark(struct per_cpu_pageset *p,
1887 struct per_cpu_pages *pcp;
1889 pcp = &p->pcp[0]; /* hot list */
1891 pcp->batch = max(1UL, high/4);
1892 if ((high/4) > (PAGE_SHIFT * 8))
1893 pcp->batch = PAGE_SHIFT * 8;
1899 * Boot pageset table. One per cpu which is going to be used for all
1900 * zones and all nodes. The parameters will be set in such a way
1901 * that an item put on a list will immediately be handed over to
1902 * the buddy list. This is safe since pageset manipulation is done
1903 * with interrupts disabled.
1905 * Some NUMA counter updates may also be caught by the boot pagesets.
1907 * The boot_pagesets must be kept even after bootup is complete for
1908 * unused processors and/or zones. They do play a role for bootstrapping
1909 * hotplugged processors.
1911 * zoneinfo_show() and maybe other functions do
1912 * not check if the processor is online before following the pageset pointer.
1913 * Other parts of the kernel may not check if the zone is available.
1915 static struct per_cpu_pageset boot_pageset[NR_CPUS];
1918 * Dynamically allocate memory for the
1919 * per cpu pageset array in struct zone.
1921 static int __cpuinit process_zones(int cpu)
1923 struct zone *zone, *dzone;
1925 for_each_zone(zone) {
1927 zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset),
1928 GFP_KERNEL, cpu_to_node(cpu));
1929 if (!zone_pcp(zone, cpu))
1932 setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone));
1934 if (percpu_pagelist_fraction)
1935 setup_pagelist_highmark(zone_pcp(zone, cpu),
1936 (zone->present_pages / percpu_pagelist_fraction));
1941 for_each_zone(dzone) {
1944 kfree(zone_pcp(dzone, cpu));
1945 zone_pcp(dzone, cpu) = NULL;
1950 static inline void free_zone_pagesets(int cpu)
1954 for_each_zone(zone) {
1955 struct per_cpu_pageset *pset = zone_pcp(zone, cpu);
1957 zone_pcp(zone, cpu) = NULL;
1962 static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb,
1963 unsigned long action,
1966 int cpu = (long)hcpu;
1967 int ret = NOTIFY_OK;
1970 case CPU_UP_PREPARE:
1971 if (process_zones(cpu))
1974 case CPU_UP_CANCELED:
1976 free_zone_pagesets(cpu);
1984 static struct notifier_block pageset_notifier =
1985 { &pageset_cpuup_callback, NULL, 0 };
1987 void __init setup_per_cpu_pageset(void)
1991 /* Initialize per_cpu_pageset for cpu 0.
1992 * A cpuup callback will do this for every cpu
1993 * as it comes online
1995 err = process_zones(smp_processor_id());
1997 register_cpu_notifier(&pageset_notifier);
2003 void zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
2006 struct pglist_data *pgdat = zone->zone_pgdat;
2009 * The per-page waitqueue mechanism uses hashed waitqueues
2012 zone->wait_table_size = wait_table_size(zone_size_pages);
2013 zone->wait_table_bits = wait_table_bits(zone->wait_table_size);
2014 zone->wait_table = (wait_queue_head_t *)
2015 alloc_bootmem_node(pgdat, zone->wait_table_size
2016 * sizeof(wait_queue_head_t));
2018 for(i = 0; i < zone->wait_table_size; ++i)
2019 init_waitqueue_head(zone->wait_table + i);
2022 static __meminit void zone_pcp_init(struct zone *zone)
2025 unsigned long batch = zone_batchsize(zone);
2027 for (cpu = 0; cpu < NR_CPUS; cpu++) {
2029 /* Early boot. Slab allocator not functional yet */
2030 zone_pcp(zone, cpu) = &boot_pageset[cpu];
2031 setup_pageset(&boot_pageset[cpu],0);
2033 setup_pageset(zone_pcp(zone,cpu), batch);
2036 if (zone->present_pages)
2037 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n",
2038 zone->name, zone->present_pages, batch);
2041 static __meminit void init_currently_empty_zone(struct zone *zone,
2042 unsigned long zone_start_pfn, unsigned long size)
2044 struct pglist_data *pgdat = zone->zone_pgdat;
2046 zone_wait_table_init(zone, size);
2047 pgdat->nr_zones = zone_idx(zone) + 1;
2049 zone->zone_start_pfn = zone_start_pfn;
2051 memmap_init(size, pgdat->node_id, zone_idx(zone), zone_start_pfn);
2053 zone_init_free_lists(pgdat, zone, zone->spanned_pages);
2057 * Set up the zone data structures:
2058 * - mark all pages reserved
2059 * - mark all memory queues empty
2060 * - clear the memory bitmaps
2062 static void __init free_area_init_core(struct pglist_data *pgdat,
2063 unsigned long *zones_size, unsigned long *zholes_size)
2066 int nid = pgdat->node_id;
2067 unsigned long zone_start_pfn = pgdat->node_start_pfn;
2069 pgdat_resize_init(pgdat);
2070 pgdat->nr_zones = 0;
2071 init_waitqueue_head(&pgdat->kswapd_wait);
2072 pgdat->kswapd_max_order = 0;
2074 for (j = 0; j < MAX_NR_ZONES; j++) {
2075 struct zone *zone = pgdat->node_zones + j;
2076 unsigned long size, realsize;
2078 realsize = size = zones_size[j];
2080 realsize -= zholes_size[j];
2082 if (j < ZONE_HIGHMEM)
2083 nr_kernel_pages += realsize;
2084 nr_all_pages += realsize;
2086 zone->spanned_pages = size;
2087 zone->present_pages = realsize;
2088 zone->name = zone_names[j];
2089 spin_lock_init(&zone->lock);
2090 spin_lock_init(&zone->lru_lock);
2091 zone_seqlock_init(zone);
2092 zone->zone_pgdat = pgdat;
2093 zone->free_pages = 0;
2095 zone->temp_priority = zone->prev_priority = DEF_PRIORITY;
2097 zone_pcp_init(zone);
2098 INIT_LIST_HEAD(&zone->active_list);
2099 INIT_LIST_HEAD(&zone->inactive_list);
2100 zone->nr_scan_active = 0;
2101 zone->nr_scan_inactive = 0;
2102 zone->nr_active = 0;
2103 zone->nr_inactive = 0;
2104 atomic_set(&zone->reclaim_in_progress, 0);
2108 zonetable_add(zone, nid, j, zone_start_pfn, size);
2109 init_currently_empty_zone(zone, zone_start_pfn, size);
2110 zone_start_pfn += size;
2114 static void __init alloc_node_mem_map(struct pglist_data *pgdat)
2116 /* Skip empty nodes */
2117 if (!pgdat->node_spanned_pages)
2120 #ifdef CONFIG_FLAT_NODE_MEM_MAP
2121 /* ia64 gets its own node_mem_map, before this, without bootmem */
2122 if (!pgdat->node_mem_map) {
2126 size = (pgdat->node_spanned_pages + 1) * sizeof(struct page);
2127 map = alloc_remap(pgdat->node_id, size);
2129 map = alloc_bootmem_node(pgdat, size);
2130 pgdat->node_mem_map = map;
2132 #ifdef CONFIG_FLATMEM
2134 * With no DISCONTIG, the global mem_map is just set as node 0's
2136 if (pgdat == NODE_DATA(0))
2137 mem_map = NODE_DATA(0)->node_mem_map;
2139 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
2142 void __init free_area_init_node(int nid, struct pglist_data *pgdat,
2143 unsigned long *zones_size, unsigned long node_start_pfn,
2144 unsigned long *zholes_size)
2146 pgdat->node_id = nid;
2147 pgdat->node_start_pfn = node_start_pfn;
2148 calculate_zone_totalpages(pgdat, zones_size, zholes_size);
2150 alloc_node_mem_map(pgdat);
2152 free_area_init_core(pgdat, zones_size, zholes_size);
2155 #ifndef CONFIG_NEED_MULTIPLE_NODES
2156 static bootmem_data_t contig_bootmem_data;
2157 struct pglist_data contig_page_data = { .bdata = &contig_bootmem_data };
2159 EXPORT_SYMBOL(contig_page_data);
2162 void __init free_area_init(unsigned long *zones_size)
2164 free_area_init_node(0, NODE_DATA(0), zones_size,
2165 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
2168 #ifdef CONFIG_PROC_FS
2170 #include <linux/seq_file.h>
2172 static void *frag_start(struct seq_file *m, loff_t *pos)
2176 for (pgdat = first_online_pgdat();
2178 pgdat = next_online_pgdat(pgdat))
2184 static void *frag_next(struct seq_file *m, void *arg, loff_t *pos)
2186 pg_data_t *pgdat = (pg_data_t *)arg;
2189 return next_online_pgdat(pgdat);
2192 static void frag_stop(struct seq_file *m, void *arg)
2197 * This walks the free areas for each zone.
2199 static int frag_show(struct seq_file *m, void *arg)
2201 pg_data_t *pgdat = (pg_data_t *)arg;
2203 struct zone *node_zones = pgdat->node_zones;
2204 unsigned long flags;
2207 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
2208 if (!populated_zone(zone))
2211 spin_lock_irqsave(&zone->lock, flags);
2212 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
2213 for (order = 0; order < MAX_ORDER; ++order)
2214 seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
2215 spin_unlock_irqrestore(&zone->lock, flags);
2221 struct seq_operations fragmentation_op = {
2222 .start = frag_start,
2229 * Output information about zones in @pgdat.
2231 static int zoneinfo_show(struct seq_file *m, void *arg)
2233 pg_data_t *pgdat = arg;
2235 struct zone *node_zones = pgdat->node_zones;
2236 unsigned long flags;
2238 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; zone++) {
2241 if (!populated_zone(zone))
2244 spin_lock_irqsave(&zone->lock, flags);
2245 seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name);
2253 "\n scanned %lu (a: %lu i: %lu)"
2262 zone->pages_scanned,
2263 zone->nr_scan_active, zone->nr_scan_inactive,
2264 zone->spanned_pages,
2265 zone->present_pages);
2267 "\n protection: (%lu",
2268 zone->lowmem_reserve[0]);
2269 for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
2270 seq_printf(m, ", %lu", zone->lowmem_reserve[i]);
2274 for_each_online_cpu(i) {
2275 struct per_cpu_pageset *pageset;
2278 pageset = zone_pcp(zone, i);
2279 for (j = 0; j < ARRAY_SIZE(pageset->pcp); j++) {
2280 if (pageset->pcp[j].count)
2283 if (j == ARRAY_SIZE(pageset->pcp))
2285 for (j = 0; j < ARRAY_SIZE(pageset->pcp); j++) {
2287 "\n cpu: %i pcp: %i"
2292 pageset->pcp[j].count,
2293 pageset->pcp[j].high,
2294 pageset->pcp[j].batch);
2300 "\n numa_foreign: %lu"
2301 "\n interleave_hit: %lu"
2302 "\n local_node: %lu"
2303 "\n other_node: %lu",
2306 pageset->numa_foreign,
2307 pageset->interleave_hit,
2308 pageset->local_node,
2309 pageset->other_node);
2313 "\n all_unreclaimable: %u"
2314 "\n prev_priority: %i"
2315 "\n temp_priority: %i"
2316 "\n start_pfn: %lu",
2317 zone->all_unreclaimable,
2318 zone->prev_priority,
2319 zone->temp_priority,
2320 zone->zone_start_pfn);
2321 spin_unlock_irqrestore(&zone->lock, flags);
2327 struct seq_operations zoneinfo_op = {
2328 .start = frag_start, /* iterate over all zones. The same as in
2332 .show = zoneinfo_show,
2335 static char *vmstat_text[] = {
2339 "nr_page_table_pages",
2370 "pgscan_kswapd_high",
2371 "pgscan_kswapd_normal",
2372 "pgscan_kswapd_dma32",
2373 "pgscan_kswapd_dma",
2375 "pgscan_direct_high",
2376 "pgscan_direct_normal",
2377 "pgscan_direct_dma32",
2378 "pgscan_direct_dma",
2383 "kswapd_inodesteal",
2391 static void *vmstat_start(struct seq_file *m, loff_t *pos)
2393 struct page_state *ps;
2395 if (*pos >= ARRAY_SIZE(vmstat_text))
2398 ps = kmalloc(sizeof(*ps), GFP_KERNEL);
2401 return ERR_PTR(-ENOMEM);
2402 get_full_page_state(ps);
2403 ps->pgpgin /= 2; /* sectors -> kbytes */
2405 return (unsigned long *)ps + *pos;
2408 static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
2411 if (*pos >= ARRAY_SIZE(vmstat_text))
2413 return (unsigned long *)m->private + *pos;
2416 static int vmstat_show(struct seq_file *m, void *arg)
2418 unsigned long *l = arg;
2419 unsigned long off = l - (unsigned long *)m->private;
2421 seq_printf(m, "%s %lu\n", vmstat_text[off], *l);
2425 static void vmstat_stop(struct seq_file *m, void *arg)
2431 struct seq_operations vmstat_op = {
2432 .start = vmstat_start,
2433 .next = vmstat_next,
2434 .stop = vmstat_stop,
2435 .show = vmstat_show,
2438 #endif /* CONFIG_PROC_FS */
2440 #ifdef CONFIG_HOTPLUG_CPU
2441 static int page_alloc_cpu_notify(struct notifier_block *self,
2442 unsigned long action, void *hcpu)
2444 int cpu = (unsigned long)hcpu;
2446 unsigned long *src, *dest;
2448 if (action == CPU_DEAD) {
2451 /* Drain local pagecache count. */
2452 count = &per_cpu(nr_pagecache_local, cpu);
2453 atomic_add(*count, &nr_pagecache);
2455 local_irq_disable();
2458 /* Add dead cpu's page_states to our own. */
2459 dest = (unsigned long *)&__get_cpu_var(page_states);
2460 src = (unsigned long *)&per_cpu(page_states, cpu);
2462 for (i = 0; i < sizeof(struct page_state)/sizeof(unsigned long);
2472 #endif /* CONFIG_HOTPLUG_CPU */
2474 void __init page_alloc_init(void)
2476 hotcpu_notifier(page_alloc_cpu_notify, 0);
2480 * setup_per_zone_lowmem_reserve - called whenever
2481 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
2482 * has a correct pages reserved value, so an adequate number of
2483 * pages are left in the zone after a successful __alloc_pages().
2485 static void setup_per_zone_lowmem_reserve(void)
2487 struct pglist_data *pgdat;
2490 for_each_online_pgdat(pgdat) {
2491 for (j = 0; j < MAX_NR_ZONES; j++) {
2492 struct zone *zone = pgdat->node_zones + j;
2493 unsigned long present_pages = zone->present_pages;
2495 zone->lowmem_reserve[j] = 0;
2497 for (idx = j-1; idx >= 0; idx--) {
2498 struct zone *lower_zone;
2500 if (sysctl_lowmem_reserve_ratio[idx] < 1)
2501 sysctl_lowmem_reserve_ratio[idx] = 1;
2503 lower_zone = pgdat->node_zones + idx;
2504 lower_zone->lowmem_reserve[j] = present_pages /
2505 sysctl_lowmem_reserve_ratio[idx];
2506 present_pages += lower_zone->present_pages;
2513 * setup_per_zone_pages_min - called when min_free_kbytes changes. Ensures
2514 * that the pages_{min,low,high} values for each zone are set correctly
2515 * with respect to min_free_kbytes.
2517 void setup_per_zone_pages_min(void)
2519 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
2520 unsigned long lowmem_pages = 0;
2522 unsigned long flags;
2524 /* Calculate total number of !ZONE_HIGHMEM pages */
2525 for_each_zone(zone) {
2526 if (!is_highmem(zone))
2527 lowmem_pages += zone->present_pages;
2530 for_each_zone(zone) {
2532 spin_lock_irqsave(&zone->lru_lock, flags);
2533 tmp = (pages_min * zone->present_pages) / lowmem_pages;
2534 if (is_highmem(zone)) {
2536 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
2537 * need highmem pages, so cap pages_min to a small
2540 * The (pages_high-pages_low) and (pages_low-pages_min)
2541 * deltas controls asynch page reclaim, and so should
2542 * not be capped for highmem.
2546 min_pages = zone->present_pages / 1024;
2547 if (min_pages < SWAP_CLUSTER_MAX)
2548 min_pages = SWAP_CLUSTER_MAX;
2549 if (min_pages > 128)
2551 zone->pages_min = min_pages;
2554 * If it's a lowmem zone, reserve a number of pages
2555 * proportionate to the zone's size.
2557 zone->pages_min = tmp;
2560 zone->pages_low = zone->pages_min + tmp / 4;
2561 zone->pages_high = zone->pages_min + tmp / 2;
2562 spin_unlock_irqrestore(&zone->lru_lock, flags);
2567 * Initialise min_free_kbytes.
2569 * For small machines we want it small (128k min). For large machines
2570 * we want it large (64MB max). But it is not linear, because network
2571 * bandwidth does not increase linearly with machine size. We use
2573 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
2574 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
2590 static int __init init_per_zone_pages_min(void)
2592 unsigned long lowmem_kbytes;
2594 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
2596 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
2597 if (min_free_kbytes < 128)
2598 min_free_kbytes = 128;
2599 if (min_free_kbytes > 65536)
2600 min_free_kbytes = 65536;
2601 setup_per_zone_pages_min();
2602 setup_per_zone_lowmem_reserve();
2605 module_init(init_per_zone_pages_min)
2608 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
2609 * that we can call two helper functions whenever min_free_kbytes
2612 int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
2613 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2615 proc_dointvec(table, write, file, buffer, length, ppos);
2616 setup_per_zone_pages_min();
2621 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
2622 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
2623 * whenever sysctl_lowmem_reserve_ratio changes.
2625 * The reserve ratio obviously has absolutely no relation with the
2626 * pages_min watermarks. The lowmem reserve ratio can only make sense
2627 * if in function of the boot time zone sizes.
2629 int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
2630 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2632 proc_dointvec_minmax(table, write, file, buffer, length, ppos);
2633 setup_per_zone_lowmem_reserve();
2638 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
2639 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
2640 * can have before it gets flushed back to buddy allocator.
2643 int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
2644 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2650 ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
2651 if (!write || (ret == -EINVAL))
2653 for_each_zone(zone) {
2654 for_each_online_cpu(cpu) {
2656 high = zone->present_pages / percpu_pagelist_fraction;
2657 setup_pagelist_highmark(zone_pcp(zone, cpu), high);
2663 __initdata int hashdist = HASHDIST_DEFAULT;
2666 static int __init set_hashdist(char *str)
2670 hashdist = simple_strtoul(str, &str, 0);
2673 __setup("hashdist=", set_hashdist);
2677 * allocate a large system hash table from bootmem
2678 * - it is assumed that the hash table must contain an exact power-of-2
2679 * quantity of entries
2680 * - limit is the number of hash buckets, not the total allocation size
2682 void *__init alloc_large_system_hash(const char *tablename,
2683 unsigned long bucketsize,
2684 unsigned long numentries,
2687 unsigned int *_hash_shift,
2688 unsigned int *_hash_mask,
2689 unsigned long limit)
2691 unsigned long long max = limit;
2692 unsigned long log2qty, size;
2695 /* allow the kernel cmdline to have a say */
2697 /* round applicable memory size up to nearest megabyte */
2698 numentries = (flags & HASH_HIGHMEM) ? nr_all_pages : nr_kernel_pages;
2699 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
2700 numentries >>= 20 - PAGE_SHIFT;
2701 numentries <<= 20 - PAGE_SHIFT;
2703 /* limit to 1 bucket per 2^scale bytes of low memory */
2704 if (scale > PAGE_SHIFT)
2705 numentries >>= (scale - PAGE_SHIFT);
2707 numentries <<= (PAGE_SHIFT - scale);
2709 numentries = roundup_pow_of_two(numentries);
2711 /* limit allocation size to 1/16 total memory by default */
2713 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
2714 do_div(max, bucketsize);
2717 if (numentries > max)
2720 log2qty = long_log2(numentries);
2723 size = bucketsize << log2qty;
2724 if (flags & HASH_EARLY)
2725 table = alloc_bootmem(size);
2727 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
2729 unsigned long order;
2730 for (order = 0; ((1UL << order) << PAGE_SHIFT) < size; order++)
2732 table = (void*) __get_free_pages(GFP_ATOMIC, order);
2734 } while (!table && size > PAGE_SIZE && --log2qty);
2737 panic("Failed to allocate %s hash table\n", tablename);
2739 printk("%s hash table entries: %d (order: %d, %lu bytes)\n",
2742 long_log2(size) - PAGE_SHIFT,
2746 *_hash_shift = log2qty;
2748 *_hash_mask = (1 << log2qty) - 1;
2753 #ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE
2755 * pfn <-> page translation. out-of-line version.
2756 * (see asm-generic/memory_model.h)
2758 #if defined(CONFIG_FLATMEM)
2759 struct page *pfn_to_page(unsigned long pfn)
2761 return mem_map + (pfn - ARCH_PFN_OFFSET);
2763 unsigned long page_to_pfn(struct page *page)
2765 return (page - mem_map) + ARCH_PFN_OFFSET;
2767 #elif defined(CONFIG_DISCONTIGMEM)
2768 struct page *pfn_to_page(unsigned long pfn)
2770 int nid = arch_pfn_to_nid(pfn);
2771 return NODE_DATA(nid)->node_mem_map + arch_local_page_offset(pfn,nid);
2773 unsigned long page_to_pfn(struct page *page)
2775 struct pglist_data *pgdat = NODE_DATA(page_to_nid(page));
2776 return (page - pgdat->node_mem_map) + pgdat->node_start_pfn;
2778 #elif defined(CONFIG_SPARSEMEM)
2779 struct page *pfn_to_page(unsigned long pfn)
2781 return __section_mem_map_addr(__pfn_to_section(pfn)) + pfn;
2784 unsigned long page_to_pfn(struct page *page)
2786 long section_id = page_to_section(page);
2787 return page - __section_mem_map_addr(__nr_to_section(section_id));
2789 #endif /* CONFIG_FLATMEM/DISCONTIGMME/SPARSEMEM */
2790 EXPORT_SYMBOL(pfn_to_page);
2791 EXPORT_SYMBOL(page_to_pfn);
2792 #endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */