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/stddef.h>
19 #include <linux/swap.h>
20 #include <linux/interrupt.h>
21 #include <linux/pagemap.h>
22 #include <linux/jiffies.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/oom.h>
32 #include <linux/notifier.h>
33 #include <linux/topology.h>
34 #include <linux/sysctl.h>
35 #include <linux/cpu.h>
36 #include <linux/cpuset.h>
37 #include <linux/memory_hotplug.h>
38 #include <linux/nodemask.h>
39 #include <linux/vmalloc.h>
40 #include <linux/mempolicy.h>
41 #include <linux/stop_machine.h>
42 #include <linux/sort.h>
43 #include <linux/pfn.h>
44 #include <linux/backing-dev.h>
45 #include <linux/fault-inject.h>
46 #include <linux/page-isolation.h>
47 #include <linux/page_cgroup.h>
48 #include <linux/debugobjects.h>
49 #include <linux/kmemleak.h>
51 #include <asm/tlbflush.h>
52 #include <asm/div64.h>
56 * Array of node states.
58 nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
59 [N_POSSIBLE] = NODE_MASK_ALL,
60 [N_ONLINE] = { { [0] = 1UL } },
62 [N_NORMAL_MEMORY] = { { [0] = 1UL } },
64 [N_HIGH_MEMORY] = { { [0] = 1UL } },
66 [N_CPU] = { { [0] = 1UL } },
69 EXPORT_SYMBOL(node_states);
71 unsigned long totalram_pages __read_mostly;
72 unsigned long totalreserve_pages __read_mostly;
73 unsigned long highest_memmap_pfn __read_mostly;
74 int percpu_pagelist_fraction;
76 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
77 int pageblock_order __read_mostly;
80 static void __free_pages_ok(struct page *page, unsigned int order);
83 * results with 256, 32 in the lowmem_reserve sysctl:
84 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
85 * 1G machine -> (16M dma, 784M normal, 224M high)
86 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
87 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
88 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
90 * TBD: should special case ZONE_DMA32 machines here - in those we normally
91 * don't need any ZONE_NORMAL reservation
93 int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
94 #ifdef CONFIG_ZONE_DMA
97 #ifdef CONFIG_ZONE_DMA32
100 #ifdef CONFIG_HIGHMEM
106 EXPORT_SYMBOL(totalram_pages);
108 static char * const zone_names[MAX_NR_ZONES] = {
109 #ifdef CONFIG_ZONE_DMA
112 #ifdef CONFIG_ZONE_DMA32
116 #ifdef CONFIG_HIGHMEM
122 int min_free_kbytes = 1024;
124 unsigned long __meminitdata nr_kernel_pages;
125 unsigned long __meminitdata nr_all_pages;
126 static unsigned long __meminitdata dma_reserve;
128 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
130 * MAX_ACTIVE_REGIONS determines the maximum number of distinct
131 * ranges of memory (RAM) that may be registered with add_active_range().
132 * Ranges passed to add_active_range() will be merged if possible
133 * so the number of times add_active_range() can be called is
134 * related to the number of nodes and the number of holes
136 #ifdef CONFIG_MAX_ACTIVE_REGIONS
137 /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
138 #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
140 #if MAX_NUMNODES >= 32
141 /* If there can be many nodes, allow up to 50 holes per node */
142 #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
144 /* By default, allow up to 256 distinct regions */
145 #define MAX_ACTIVE_REGIONS 256
149 static struct node_active_region __meminitdata early_node_map[MAX_ACTIVE_REGIONS];
150 static int __meminitdata nr_nodemap_entries;
151 static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
152 static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
153 static unsigned long __initdata required_kernelcore;
154 static unsigned long __initdata required_movablecore;
155 static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
157 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
159 EXPORT_SYMBOL(movable_zone);
160 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
163 int nr_node_ids __read_mostly = MAX_NUMNODES;
164 EXPORT_SYMBOL(nr_node_ids);
167 int page_group_by_mobility_disabled __read_mostly;
169 static void set_pageblock_migratetype(struct page *page, int migratetype)
172 if (unlikely(page_group_by_mobility_disabled))
173 migratetype = MIGRATE_UNMOVABLE;
175 set_pageblock_flags_group(page, (unsigned long)migratetype,
176 PB_migrate, PB_migrate_end);
179 #ifdef CONFIG_DEBUG_VM
180 static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
184 unsigned long pfn = page_to_pfn(page);
187 seq = zone_span_seqbegin(zone);
188 if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
190 else if (pfn < zone->zone_start_pfn)
192 } while (zone_span_seqretry(zone, seq));
197 static int page_is_consistent(struct zone *zone, struct page *page)
199 if (!pfn_valid_within(page_to_pfn(page)))
201 if (zone != page_zone(page))
207 * Temporary debugging check for pages not lying within a given zone.
209 static int bad_range(struct zone *zone, struct page *page)
211 if (page_outside_zone_boundaries(zone, page))
213 if (!page_is_consistent(zone, page))
219 static inline int bad_range(struct zone *zone, struct page *page)
225 static void bad_page(struct page *page)
227 static unsigned long resume;
228 static unsigned long nr_shown;
229 static unsigned long nr_unshown;
232 * Allow a burst of 60 reports, then keep quiet for that minute;
233 * or allow a steady drip of one report per second.
235 if (nr_shown == 60) {
236 if (time_before(jiffies, resume)) {
242 "BUG: Bad page state: %lu messages suppressed\n",
249 resume = jiffies + 60 * HZ;
251 printk(KERN_ALERT "BUG: Bad page state in process %s pfn:%05lx\n",
252 current->comm, page_to_pfn(page));
254 "page:%p flags:%p count:%d mapcount:%d mapping:%p index:%lx\n",
255 page, (void *)page->flags, page_count(page),
256 page_mapcount(page), page->mapping, page->index);
260 /* Leave bad fields for debug, except PageBuddy could make trouble */
261 __ClearPageBuddy(page);
262 add_taint(TAINT_BAD_PAGE);
266 * Higher-order pages are called "compound pages". They are structured thusly:
268 * The first PAGE_SIZE page is called the "head page".
270 * The remaining PAGE_SIZE pages are called "tail pages".
272 * All pages have PG_compound set. All pages have their ->private pointing at
273 * the head page (even the head page has this).
275 * The first tail page's ->lru.next holds the address of the compound page's
276 * put_page() function. Its ->lru.prev holds the order of allocation.
277 * This usage means that zero-order pages may not be compound.
280 static void free_compound_page(struct page *page)
282 __free_pages_ok(page, compound_order(page));
285 void prep_compound_page(struct page *page, unsigned long order)
288 int nr_pages = 1 << order;
290 set_compound_page_dtor(page, free_compound_page);
291 set_compound_order(page, order);
293 for (i = 1; i < nr_pages; i++) {
294 struct page *p = page + i;
297 p->first_page = page;
301 #ifdef CONFIG_HUGETLBFS
302 void prep_compound_gigantic_page(struct page *page, unsigned long order)
305 int nr_pages = 1 << order;
306 struct page *p = page + 1;
308 set_compound_page_dtor(page, free_compound_page);
309 set_compound_order(page, order);
311 for (i = 1; i < nr_pages; i++, p = mem_map_next(p, page, i)) {
313 p->first_page = page;
318 static int destroy_compound_page(struct page *page, unsigned long order)
321 int nr_pages = 1 << order;
324 if (unlikely(compound_order(page) != order) ||
325 unlikely(!PageHead(page))) {
330 __ClearPageHead(page);
332 for (i = 1; i < nr_pages; i++) {
333 struct page *p = page + i;
335 if (unlikely(!PageTail(p) || (p->first_page != page))) {
345 static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
350 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
351 * and __GFP_HIGHMEM from hard or soft interrupt context.
353 VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
354 for (i = 0; i < (1 << order); i++)
355 clear_highpage(page + i);
358 static inline void set_page_order(struct page *page, int order)
360 set_page_private(page, order);
361 __SetPageBuddy(page);
364 static inline void rmv_page_order(struct page *page)
366 __ClearPageBuddy(page);
367 set_page_private(page, 0);
371 * Locate the struct page for both the matching buddy in our
372 * pair (buddy1) and the combined O(n+1) page they form (page).
374 * 1) Any buddy B1 will have an order O twin B2 which satisfies
375 * the following equation:
377 * For example, if the starting buddy (buddy2) is #8 its order
379 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
381 * 2) Any buddy B will have an order O+1 parent P which
382 * satisfies the following equation:
385 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
387 static inline struct page *
388 __page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order)
390 unsigned long buddy_idx = page_idx ^ (1 << order);
392 return page + (buddy_idx - page_idx);
395 static inline unsigned long
396 __find_combined_index(unsigned long page_idx, unsigned int order)
398 return (page_idx & ~(1 << order));
402 * This function checks whether a page is free && is the buddy
403 * we can do coalesce a page and its buddy if
404 * (a) the buddy is not in a hole &&
405 * (b) the buddy is in the buddy system &&
406 * (c) a page and its buddy have the same order &&
407 * (d) a page and its buddy are in the same zone.
409 * For recording whether a page is in the buddy system, we use PG_buddy.
410 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
412 * For recording page's order, we use page_private(page).
414 static inline int page_is_buddy(struct page *page, struct page *buddy,
417 if (!pfn_valid_within(page_to_pfn(buddy)))
420 if (page_zone_id(page) != page_zone_id(buddy))
423 if (PageBuddy(buddy) && page_order(buddy) == order) {
424 BUG_ON(page_count(buddy) != 0);
431 * Freeing function for a buddy system allocator.
433 * The concept of a buddy system is to maintain direct-mapped table
434 * (containing bit values) for memory blocks of various "orders".
435 * The bottom level table contains the map for the smallest allocatable
436 * units of memory (here, pages), and each level above it describes
437 * pairs of units from the levels below, hence, "buddies".
438 * At a high level, all that happens here is marking the table entry
439 * at the bottom level available, and propagating the changes upward
440 * as necessary, plus some accounting needed to play nicely with other
441 * parts of the VM system.
442 * At each level, we keep a list of pages, which are heads of continuous
443 * free pages of length of (1 << order) and marked with PG_buddy. Page's
444 * order is recorded in page_private(page) field.
445 * So when we are allocating or freeing one, we can derive the state of the
446 * other. That is, if we allocate a small block, and both were
447 * free, the remainder of the region must be split into blocks.
448 * If a block is freed, and its buddy is also free, then this
449 * triggers coalescing into a block of larger size.
454 static inline void __free_one_page(struct page *page,
455 struct zone *zone, unsigned int order)
457 unsigned long page_idx;
458 int order_size = 1 << order;
459 int migratetype = get_pageblock_migratetype(page);
461 if (unlikely(PageCompound(page)))
462 if (unlikely(destroy_compound_page(page, order)))
465 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
467 VM_BUG_ON(page_idx & (order_size - 1));
468 VM_BUG_ON(bad_range(zone, page));
470 __mod_zone_page_state(zone, NR_FREE_PAGES, order_size);
471 while (order < MAX_ORDER-1) {
472 unsigned long combined_idx;
475 buddy = __page_find_buddy(page, page_idx, order);
476 if (!page_is_buddy(page, buddy, order))
479 /* Our buddy is free, merge with it and move up one order. */
480 list_del(&buddy->lru);
481 zone->free_area[order].nr_free--;
482 rmv_page_order(buddy);
483 combined_idx = __find_combined_index(page_idx, order);
484 page = page + (combined_idx - page_idx);
485 page_idx = combined_idx;
488 set_page_order(page, order);
490 &zone->free_area[order].free_list[migratetype]);
491 zone->free_area[order].nr_free++;
494 static inline int free_pages_check(struct page *page)
496 free_page_mlock(page);
497 if (unlikely(page_mapcount(page) |
498 (page->mapping != NULL) |
499 (page_count(page) != 0) |
500 (page->flags & PAGE_FLAGS_CHECK_AT_FREE))) {
504 if (page->flags & PAGE_FLAGS_CHECK_AT_PREP)
505 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
510 * Frees a list of pages.
511 * Assumes all pages on list are in same zone, and of same order.
512 * count is the number of pages to free.
514 * If the zone was previously in an "all pages pinned" state then look to
515 * see if this freeing clears that state.
517 * And clear the zone's pages_scanned counter, to hold off the "all pages are
518 * pinned" detection logic.
520 static void free_pages_bulk(struct zone *zone, int count,
521 struct list_head *list, int order)
523 spin_lock(&zone->lock);
524 zone_clear_flag(zone, ZONE_ALL_UNRECLAIMABLE);
525 zone->pages_scanned = 0;
529 VM_BUG_ON(list_empty(list));
530 page = list_entry(list->prev, struct page, lru);
531 /* have to delete it as __free_one_page list manipulates */
532 list_del(&page->lru);
533 __free_one_page(page, zone, order);
535 spin_unlock(&zone->lock);
538 static void free_one_page(struct zone *zone, struct page *page, int order)
540 spin_lock(&zone->lock);
541 zone_clear_flag(zone, ZONE_ALL_UNRECLAIMABLE);
542 zone->pages_scanned = 0;
543 __free_one_page(page, zone, order);
544 spin_unlock(&zone->lock);
547 static void __free_pages_ok(struct page *page, unsigned int order)
553 for (i = 0 ; i < (1 << order) ; ++i)
554 bad += free_pages_check(page + i);
558 if (!PageHighMem(page)) {
559 debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order);
560 debug_check_no_obj_freed(page_address(page),
563 arch_free_page(page, order);
564 kernel_map_pages(page, 1 << order, 0);
566 local_irq_save(flags);
567 __count_vm_events(PGFREE, 1 << order);
568 free_one_page(page_zone(page), page, order);
569 local_irq_restore(flags);
573 * permit the bootmem allocator to evade page validation on high-order frees
575 void __meminit __free_pages_bootmem(struct page *page, unsigned int order)
578 __ClearPageReserved(page);
579 set_page_count(page, 0);
580 set_page_refcounted(page);
586 for (loop = 0; loop < BITS_PER_LONG; loop++) {
587 struct page *p = &page[loop];
589 if (loop + 1 < BITS_PER_LONG)
591 __ClearPageReserved(p);
592 set_page_count(p, 0);
595 set_page_refcounted(page);
596 __free_pages(page, order);
602 * The order of subdivision here is critical for the IO subsystem.
603 * Please do not alter this order without good reasons and regression
604 * testing. Specifically, as large blocks of memory are subdivided,
605 * the order in which smaller blocks are delivered depends on the order
606 * they're subdivided in this function. This is the primary factor
607 * influencing the order in which pages are delivered to the IO
608 * subsystem according to empirical testing, and this is also justified
609 * by considering the behavior of a buddy system containing a single
610 * large block of memory acted on by a series of small allocations.
611 * This behavior is a critical factor in sglist merging's success.
615 static inline void expand(struct zone *zone, struct page *page,
616 int low, int high, struct free_area *area,
619 unsigned long size = 1 << high;
625 VM_BUG_ON(bad_range(zone, &page[size]));
626 list_add(&page[size].lru, &area->free_list[migratetype]);
628 set_page_order(&page[size], high);
633 * This page is about to be returned from the page allocator
635 static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
637 if (unlikely(page_mapcount(page) |
638 (page->mapping != NULL) |
639 (page_count(page) != 0) |
640 (page->flags & PAGE_FLAGS_CHECK_AT_PREP))) {
645 set_page_private(page, 0);
646 set_page_refcounted(page);
648 arch_alloc_page(page, order);
649 kernel_map_pages(page, 1 << order, 1);
651 if (gfp_flags & __GFP_ZERO)
652 prep_zero_page(page, order, gfp_flags);
654 if (order && (gfp_flags & __GFP_COMP))
655 prep_compound_page(page, order);
661 * Go through the free lists for the given migratetype and remove
662 * the smallest available page from the freelists
664 static struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
667 unsigned int current_order;
668 struct free_area * area;
671 /* Find a page of the appropriate size in the preferred list */
672 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
673 area = &(zone->free_area[current_order]);
674 if (list_empty(&area->free_list[migratetype]))
677 page = list_entry(area->free_list[migratetype].next,
679 list_del(&page->lru);
680 rmv_page_order(page);
682 __mod_zone_page_state(zone, NR_FREE_PAGES, - (1UL << order));
683 expand(zone, page, order, current_order, area, migratetype);
692 * This array describes the order lists are fallen back to when
693 * the free lists for the desirable migrate type are depleted
695 static int fallbacks[MIGRATE_TYPES][MIGRATE_TYPES-1] = {
696 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
697 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
698 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
699 [MIGRATE_RESERVE] = { MIGRATE_RESERVE, MIGRATE_RESERVE, MIGRATE_RESERVE }, /* Never used */
703 * Move the free pages in a range to the free lists of the requested type.
704 * Note that start_page and end_pages are not aligned on a pageblock
705 * boundary. If alignment is required, use move_freepages_block()
707 static int move_freepages(struct zone *zone,
708 struct page *start_page, struct page *end_page,
715 #ifndef CONFIG_HOLES_IN_ZONE
717 * page_zone is not safe to call in this context when
718 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
719 * anyway as we check zone boundaries in move_freepages_block().
720 * Remove at a later date when no bug reports exist related to
721 * grouping pages by mobility
723 BUG_ON(page_zone(start_page) != page_zone(end_page));
726 for (page = start_page; page <= end_page;) {
727 /* Make sure we are not inadvertently changing nodes */
728 VM_BUG_ON(page_to_nid(page) != zone_to_nid(zone));
730 if (!pfn_valid_within(page_to_pfn(page))) {
735 if (!PageBuddy(page)) {
740 order = page_order(page);
741 list_del(&page->lru);
743 &zone->free_area[order].free_list[migratetype]);
745 pages_moved += 1 << order;
751 static int move_freepages_block(struct zone *zone, struct page *page,
754 unsigned long start_pfn, end_pfn;
755 struct page *start_page, *end_page;
757 start_pfn = page_to_pfn(page);
758 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
759 start_page = pfn_to_page(start_pfn);
760 end_page = start_page + pageblock_nr_pages - 1;
761 end_pfn = start_pfn + pageblock_nr_pages - 1;
763 /* Do not cross zone boundaries */
764 if (start_pfn < zone->zone_start_pfn)
766 if (end_pfn >= zone->zone_start_pfn + zone->spanned_pages)
769 return move_freepages(zone, start_page, end_page, migratetype);
772 /* Remove an element from the buddy allocator from the fallback list */
773 static struct page *__rmqueue_fallback(struct zone *zone, int order,
774 int start_migratetype)
776 struct free_area * area;
781 /* Find the largest possible block of pages in the other list */
782 for (current_order = MAX_ORDER-1; current_order >= order;
784 for (i = 0; i < MIGRATE_TYPES - 1; i++) {
785 migratetype = fallbacks[start_migratetype][i];
787 /* MIGRATE_RESERVE handled later if necessary */
788 if (migratetype == MIGRATE_RESERVE)
791 area = &(zone->free_area[current_order]);
792 if (list_empty(&area->free_list[migratetype]))
795 page = list_entry(area->free_list[migratetype].next,
800 * If breaking a large block of pages, move all free
801 * pages to the preferred allocation list. If falling
802 * back for a reclaimable kernel allocation, be more
803 * agressive about taking ownership of free pages
805 if (unlikely(current_order >= (pageblock_order >> 1)) ||
806 start_migratetype == MIGRATE_RECLAIMABLE) {
808 pages = move_freepages_block(zone, page,
811 /* Claim the whole block if over half of it is free */
812 if (pages >= (1 << (pageblock_order-1)))
813 set_pageblock_migratetype(page,
816 migratetype = start_migratetype;
819 /* Remove the page from the freelists */
820 list_del(&page->lru);
821 rmv_page_order(page);
822 __mod_zone_page_state(zone, NR_FREE_PAGES,
825 if (current_order == pageblock_order)
826 set_pageblock_migratetype(page,
829 expand(zone, page, order, current_order, area, migratetype);
834 /* Use MIGRATE_RESERVE rather than fail an allocation */
835 return __rmqueue_smallest(zone, order, MIGRATE_RESERVE);
839 * Do the hard work of removing an element from the buddy allocator.
840 * Call me with the zone->lock already held.
842 static struct page *__rmqueue(struct zone *zone, unsigned int order,
847 page = __rmqueue_smallest(zone, order, migratetype);
850 page = __rmqueue_fallback(zone, order, migratetype);
856 * Obtain a specified number of elements from the buddy allocator, all under
857 * a single hold of the lock, for efficiency. Add them to the supplied list.
858 * Returns the number of new pages which were placed at *list.
860 static int rmqueue_bulk(struct zone *zone, unsigned int order,
861 unsigned long count, struct list_head *list,
866 spin_lock(&zone->lock);
867 for (i = 0; i < count; ++i) {
868 struct page *page = __rmqueue(zone, order, migratetype);
869 if (unlikely(page == NULL))
873 * Split buddy pages returned by expand() are received here
874 * in physical page order. The page is added to the callers and
875 * list and the list head then moves forward. From the callers
876 * perspective, the linked list is ordered by page number in
877 * some conditions. This is useful for IO devices that can
878 * merge IO requests if the physical pages are ordered
881 list_add(&page->lru, list);
882 set_page_private(page, migratetype);
885 spin_unlock(&zone->lock);
891 * Called from the vmstat counter updater to drain pagesets of this
892 * currently executing processor on remote nodes after they have
895 * Note that this function must be called with the thread pinned to
896 * a single processor.
898 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
903 local_irq_save(flags);
904 if (pcp->count >= pcp->batch)
905 to_drain = pcp->batch;
907 to_drain = pcp->count;
908 free_pages_bulk(zone, to_drain, &pcp->list, 0);
909 pcp->count -= to_drain;
910 local_irq_restore(flags);
915 * Drain pages of the indicated processor.
917 * The processor must either be the current processor and the
918 * thread pinned to the current processor or a processor that
921 static void drain_pages(unsigned int cpu)
926 for_each_populated_zone(zone) {
927 struct per_cpu_pageset *pset;
928 struct per_cpu_pages *pcp;
930 pset = zone_pcp(zone, cpu);
933 local_irq_save(flags);
934 free_pages_bulk(zone, pcp->count, &pcp->list, 0);
936 local_irq_restore(flags);
941 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
943 void drain_local_pages(void *arg)
945 drain_pages(smp_processor_id());
949 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
951 void drain_all_pages(void)
953 on_each_cpu(drain_local_pages, NULL, 1);
956 #ifdef CONFIG_HIBERNATION
958 void mark_free_pages(struct zone *zone)
960 unsigned long pfn, max_zone_pfn;
963 struct list_head *curr;
965 if (!zone->spanned_pages)
968 spin_lock_irqsave(&zone->lock, flags);
970 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
971 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
972 if (pfn_valid(pfn)) {
973 struct page *page = pfn_to_page(pfn);
975 if (!swsusp_page_is_forbidden(page))
976 swsusp_unset_page_free(page);
979 for_each_migratetype_order(order, t) {
980 list_for_each(curr, &zone->free_area[order].free_list[t]) {
983 pfn = page_to_pfn(list_entry(curr, struct page, lru));
984 for (i = 0; i < (1UL << order); i++)
985 swsusp_set_page_free(pfn_to_page(pfn + i));
988 spin_unlock_irqrestore(&zone->lock, flags);
990 #endif /* CONFIG_PM */
993 * Free a 0-order page
995 static void free_hot_cold_page(struct page *page, int cold)
997 struct zone *zone = page_zone(page);
998 struct per_cpu_pages *pcp;
1002 page->mapping = NULL;
1003 if (free_pages_check(page))
1006 if (!PageHighMem(page)) {
1007 debug_check_no_locks_freed(page_address(page), PAGE_SIZE);
1008 debug_check_no_obj_freed(page_address(page), PAGE_SIZE);
1010 arch_free_page(page, 0);
1011 kernel_map_pages(page, 1, 0);
1013 pcp = &zone_pcp(zone, get_cpu())->pcp;
1014 local_irq_save(flags);
1015 __count_vm_event(PGFREE);
1017 list_add_tail(&page->lru, &pcp->list);
1019 list_add(&page->lru, &pcp->list);
1020 set_page_private(page, get_pageblock_migratetype(page));
1022 if (pcp->count >= pcp->high) {
1023 free_pages_bulk(zone, pcp->batch, &pcp->list, 0);
1024 pcp->count -= pcp->batch;
1026 local_irq_restore(flags);
1030 void free_hot_page(struct page *page)
1032 free_hot_cold_page(page, 0);
1035 void free_cold_page(struct page *page)
1037 free_hot_cold_page(page, 1);
1041 * split_page takes a non-compound higher-order page, and splits it into
1042 * n (1<<order) sub-pages: page[0..n]
1043 * Each sub-page must be freed individually.
1045 * Note: this is probably too low level an operation for use in drivers.
1046 * Please consult with lkml before using this in your driver.
1048 void split_page(struct page *page, unsigned int order)
1052 VM_BUG_ON(PageCompound(page));
1053 VM_BUG_ON(!page_count(page));
1054 for (i = 1; i < (1 << order); i++)
1055 set_page_refcounted(page + i);
1059 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1060 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1063 static struct page *buffered_rmqueue(struct zone *preferred_zone,
1064 struct zone *zone, int order, gfp_t gfp_flags)
1066 unsigned long flags;
1068 int cold = !!(gfp_flags & __GFP_COLD);
1070 int migratetype = allocflags_to_migratetype(gfp_flags);
1074 if (likely(order == 0)) {
1075 struct per_cpu_pages *pcp;
1077 pcp = &zone_pcp(zone, cpu)->pcp;
1078 local_irq_save(flags);
1080 pcp->count = rmqueue_bulk(zone, 0,
1081 pcp->batch, &pcp->list, migratetype);
1082 if (unlikely(!pcp->count))
1086 /* Find a page of the appropriate migrate type */
1088 list_for_each_entry_reverse(page, &pcp->list, lru)
1089 if (page_private(page) == migratetype)
1092 list_for_each_entry(page, &pcp->list, lru)
1093 if (page_private(page) == migratetype)
1097 /* Allocate more to the pcp list if necessary */
1098 if (unlikely(&page->lru == &pcp->list)) {
1099 pcp->count += rmqueue_bulk(zone, 0,
1100 pcp->batch, &pcp->list, migratetype);
1101 page = list_entry(pcp->list.next, struct page, lru);
1104 list_del(&page->lru);
1107 spin_lock_irqsave(&zone->lock, flags);
1108 page = __rmqueue(zone, order, migratetype);
1109 spin_unlock(&zone->lock);
1114 __count_zone_vm_events(PGALLOC, zone, 1 << order);
1115 zone_statistics(preferred_zone, zone);
1116 local_irq_restore(flags);
1119 VM_BUG_ON(bad_range(zone, page));
1120 if (prep_new_page(page, order, gfp_flags))
1125 local_irq_restore(flags);
1130 #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
1131 #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
1132 #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
1133 #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
1134 #define ALLOC_HARDER 0x10 /* try to alloc harder */
1135 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1136 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
1138 #ifdef CONFIG_FAIL_PAGE_ALLOC
1140 static struct fail_page_alloc_attr {
1141 struct fault_attr attr;
1143 u32 ignore_gfp_highmem;
1144 u32 ignore_gfp_wait;
1147 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1149 struct dentry *ignore_gfp_highmem_file;
1150 struct dentry *ignore_gfp_wait_file;
1151 struct dentry *min_order_file;
1153 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1155 } fail_page_alloc = {
1156 .attr = FAULT_ATTR_INITIALIZER,
1157 .ignore_gfp_wait = 1,
1158 .ignore_gfp_highmem = 1,
1162 static int __init setup_fail_page_alloc(char *str)
1164 return setup_fault_attr(&fail_page_alloc.attr, str);
1166 __setup("fail_page_alloc=", setup_fail_page_alloc);
1168 static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1170 if (order < fail_page_alloc.min_order)
1172 if (gfp_mask & __GFP_NOFAIL)
1174 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
1176 if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
1179 return should_fail(&fail_page_alloc.attr, 1 << order);
1182 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1184 static int __init fail_page_alloc_debugfs(void)
1186 mode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
1190 err = init_fault_attr_dentries(&fail_page_alloc.attr,
1194 dir = fail_page_alloc.attr.dentries.dir;
1196 fail_page_alloc.ignore_gfp_wait_file =
1197 debugfs_create_bool("ignore-gfp-wait", mode, dir,
1198 &fail_page_alloc.ignore_gfp_wait);
1200 fail_page_alloc.ignore_gfp_highmem_file =
1201 debugfs_create_bool("ignore-gfp-highmem", mode, dir,
1202 &fail_page_alloc.ignore_gfp_highmem);
1203 fail_page_alloc.min_order_file =
1204 debugfs_create_u32("min-order", mode, dir,
1205 &fail_page_alloc.min_order);
1207 if (!fail_page_alloc.ignore_gfp_wait_file ||
1208 !fail_page_alloc.ignore_gfp_highmem_file ||
1209 !fail_page_alloc.min_order_file) {
1211 debugfs_remove(fail_page_alloc.ignore_gfp_wait_file);
1212 debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file);
1213 debugfs_remove(fail_page_alloc.min_order_file);
1214 cleanup_fault_attr_dentries(&fail_page_alloc.attr);
1220 late_initcall(fail_page_alloc_debugfs);
1222 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1224 #else /* CONFIG_FAIL_PAGE_ALLOC */
1226 static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1231 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1234 * Return 1 if free pages are above 'mark'. This takes into account the order
1235 * of the allocation.
1237 int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1238 int classzone_idx, int alloc_flags)
1240 /* free_pages my go negative - that's OK */
1242 long free_pages = zone_page_state(z, NR_FREE_PAGES) - (1 << order) + 1;
1245 if (alloc_flags & ALLOC_HIGH)
1247 if (alloc_flags & ALLOC_HARDER)
1250 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
1252 for (o = 0; o < order; o++) {
1253 /* At the next order, this order's pages become unavailable */
1254 free_pages -= z->free_area[o].nr_free << o;
1256 /* Require fewer higher order pages to be free */
1259 if (free_pages <= min)
1267 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1268 * skip over zones that are not allowed by the cpuset, or that have
1269 * been recently (in last second) found to be nearly full. See further
1270 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1271 * that have to skip over a lot of full or unallowed zones.
1273 * If the zonelist cache is present in the passed in zonelist, then
1274 * returns a pointer to the allowed node mask (either the current
1275 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
1277 * If the zonelist cache is not available for this zonelist, does
1278 * nothing and returns NULL.
1280 * If the fullzones BITMAP in the zonelist cache is stale (more than
1281 * a second since last zap'd) then we zap it out (clear its bits.)
1283 * We hold off even calling zlc_setup, until after we've checked the
1284 * first zone in the zonelist, on the theory that most allocations will
1285 * be satisfied from that first zone, so best to examine that zone as
1286 * quickly as we can.
1288 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1290 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1291 nodemask_t *allowednodes; /* zonelist_cache approximation */
1293 zlc = zonelist->zlcache_ptr;
1297 if (time_after(jiffies, zlc->last_full_zap + HZ)) {
1298 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1299 zlc->last_full_zap = jiffies;
1302 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1303 &cpuset_current_mems_allowed :
1304 &node_states[N_HIGH_MEMORY];
1305 return allowednodes;
1309 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1310 * if it is worth looking at further for free memory:
1311 * 1) Check that the zone isn't thought to be full (doesn't have its
1312 * bit set in the zonelist_cache fullzones BITMAP).
1313 * 2) Check that the zones node (obtained from the zonelist_cache
1314 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1315 * Return true (non-zero) if zone is worth looking at further, or
1316 * else return false (zero) if it is not.
1318 * This check -ignores- the distinction between various watermarks,
1319 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1320 * found to be full for any variation of these watermarks, it will
1321 * be considered full for up to one second by all requests, unless
1322 * we are so low on memory on all allowed nodes that we are forced
1323 * into the second scan of the zonelist.
1325 * In the second scan we ignore this zonelist cache and exactly
1326 * apply the watermarks to all zones, even it is slower to do so.
1327 * We are low on memory in the second scan, and should leave no stone
1328 * unturned looking for a free page.
1330 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
1331 nodemask_t *allowednodes)
1333 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1334 int i; /* index of *z in zonelist zones */
1335 int n; /* node that zone *z is on */
1337 zlc = zonelist->zlcache_ptr;
1341 i = z - zonelist->_zonerefs;
1344 /* This zone is worth trying if it is allowed but not full */
1345 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1349 * Given 'z' scanning a zonelist, set the corresponding bit in
1350 * zlc->fullzones, so that subsequent attempts to allocate a page
1351 * from that zone don't waste time re-examining it.
1353 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
1355 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1356 int i; /* index of *z in zonelist zones */
1358 zlc = zonelist->zlcache_ptr;
1362 i = z - zonelist->_zonerefs;
1364 set_bit(i, zlc->fullzones);
1367 #else /* CONFIG_NUMA */
1369 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1374 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
1375 nodemask_t *allowednodes)
1380 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
1383 #endif /* CONFIG_NUMA */
1386 * get_page_from_freelist goes through the zonelist trying to allocate
1389 static struct page *
1390 get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order,
1391 struct zonelist *zonelist, int high_zoneidx, int alloc_flags,
1392 struct zone *preferred_zone)
1395 struct page *page = NULL;
1398 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1399 int zlc_active = 0; /* set if using zonelist_cache */
1400 int did_zlc_setup = 0; /* just call zlc_setup() one time */
1402 if (WARN_ON_ONCE(order >= MAX_ORDER))
1405 classzone_idx = zone_idx(preferred_zone);
1408 * Scan zonelist, looking for a zone with enough free.
1409 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1411 for_each_zone_zonelist_nodemask(zone, z, zonelist,
1412 high_zoneidx, nodemask) {
1413 if (NUMA_BUILD && zlc_active &&
1414 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1416 if ((alloc_flags & ALLOC_CPUSET) &&
1417 !cpuset_zone_allowed_softwall(zone, gfp_mask))
1420 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
1422 if (alloc_flags & ALLOC_WMARK_MIN)
1423 mark = zone->pages_min;
1424 else if (alloc_flags & ALLOC_WMARK_LOW)
1425 mark = zone->pages_low;
1427 mark = zone->pages_high;
1428 if (!zone_watermark_ok(zone, order, mark,
1429 classzone_idx, alloc_flags)) {
1430 if (!zone_reclaim_mode ||
1431 !zone_reclaim(zone, gfp_mask, order))
1432 goto this_zone_full;
1436 page = buffered_rmqueue(preferred_zone, zone, order, gfp_mask);
1441 zlc_mark_zone_full(zonelist, z);
1443 if (NUMA_BUILD && !did_zlc_setup) {
1444 /* we do zlc_setup after the first zone is tried */
1445 allowednodes = zlc_setup(zonelist, alloc_flags);
1451 if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) {
1452 /* Disable zlc cache for second zonelist scan */
1460 should_alloc_retry(gfp_t gfp_mask, unsigned int order,
1461 unsigned long pages_reclaimed)
1463 /* Do not loop if specifically requested */
1464 if (gfp_mask & __GFP_NORETRY)
1468 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
1469 * means __GFP_NOFAIL, but that may not be true in other
1472 if (order <= PAGE_ALLOC_COSTLY_ORDER)
1476 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
1477 * specified, then we retry until we no longer reclaim any pages
1478 * (above), or we've reclaimed an order of pages at least as
1479 * large as the allocation's order. In both cases, if the
1480 * allocation still fails, we stop retrying.
1482 if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order))
1486 * Don't let big-order allocations loop unless the caller
1487 * explicitly requests that.
1489 if (gfp_mask & __GFP_NOFAIL)
1495 static inline struct page *
1496 __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
1497 struct zonelist *zonelist, enum zone_type high_zoneidx,
1498 nodemask_t *nodemask, struct zone *preferred_zone)
1502 /* Acquire the OOM killer lock for the zones in zonelist */
1503 if (!try_set_zone_oom(zonelist, gfp_mask)) {
1504 schedule_timeout_uninterruptible(1);
1509 * Go through the zonelist yet one more time, keep very high watermark
1510 * here, this is only to catch a parallel oom killing, we must fail if
1511 * we're still under heavy pressure.
1513 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask,
1514 order, zonelist, high_zoneidx,
1515 ALLOC_WMARK_HIGH|ALLOC_CPUSET,
1520 /* The OOM killer will not help higher order allocs */
1521 if (order > PAGE_ALLOC_COSTLY_ORDER)
1524 /* Exhausted what can be done so it's blamo time */
1525 out_of_memory(zonelist, gfp_mask, order);
1528 clear_zonelist_oom(zonelist, gfp_mask);
1532 /* The really slow allocator path where we enter direct reclaim */
1533 static inline struct page *
1534 __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
1535 struct zonelist *zonelist, enum zone_type high_zoneidx,
1536 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
1537 unsigned long *did_some_progress)
1539 struct page *page = NULL;
1540 struct reclaim_state reclaim_state;
1541 struct task_struct *p = current;
1545 /* We now go into synchronous reclaim */
1546 cpuset_memory_pressure_bump();
1549 * The task's cpuset might have expanded its set of allowable nodes
1551 p->flags |= PF_MEMALLOC;
1552 lockdep_set_current_reclaim_state(gfp_mask);
1553 reclaim_state.reclaimed_slab = 0;
1554 p->reclaim_state = &reclaim_state;
1556 *did_some_progress = try_to_free_pages(zonelist, order, gfp_mask, nodemask);
1558 p->reclaim_state = NULL;
1559 lockdep_clear_current_reclaim_state();
1560 p->flags &= ~PF_MEMALLOC;
1567 if (likely(*did_some_progress))
1568 page = get_page_from_freelist(gfp_mask, nodemask, order,
1569 zonelist, high_zoneidx,
1570 alloc_flags, preferred_zone);
1575 is_allocation_high_priority(struct task_struct *p, gfp_t gfp_mask)
1577 if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE)))
1584 * This is called in the allocator slow-path if the allocation request is of
1585 * sufficient urgency to ignore watermarks and take other desperate measures
1587 static inline struct page *
1588 __alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order,
1589 struct zonelist *zonelist, enum zone_type high_zoneidx,
1590 nodemask_t *nodemask, struct zone *preferred_zone)
1595 page = get_page_from_freelist(gfp_mask, nodemask, order,
1596 zonelist, high_zoneidx, ALLOC_NO_WATERMARKS,
1599 if (!page && gfp_mask & __GFP_NOFAIL)
1600 congestion_wait(WRITE, HZ/50);
1601 } while (!page && (gfp_mask & __GFP_NOFAIL));
1607 void wake_all_kswapd(unsigned int order, struct zonelist *zonelist,
1608 enum zone_type high_zoneidx)
1613 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx)
1614 wakeup_kswapd(zone, order);
1617 static inline struct page *
1618 __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
1619 struct zonelist *zonelist, enum zone_type high_zoneidx,
1620 nodemask_t *nodemask, struct zone *preferred_zone)
1622 const gfp_t wait = gfp_mask & __GFP_WAIT;
1623 struct page *page = NULL;
1625 unsigned long pages_reclaimed = 0;
1626 unsigned long did_some_progress;
1627 struct task_struct *p = current;
1630 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1631 * __GFP_NOWARN set) should not cause reclaim since the subsystem
1632 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1633 * using a larger set of nodes after it has established that the
1634 * allowed per node queues are empty and that nodes are
1637 if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
1640 wake_all_kswapd(order, zonelist, high_zoneidx);
1643 * OK, we're below the kswapd watermark and have kicked background
1644 * reclaim. Now things get more complex, so set up alloc_flags according
1645 * to how we want to proceed.
1647 * The caller may dip into page reserves a bit more if the caller
1648 * cannot run direct reclaim, or if the caller has realtime scheduling
1649 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1650 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1652 alloc_flags = ALLOC_WMARK_MIN;
1653 if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait)
1654 alloc_flags |= ALLOC_HARDER;
1655 if (gfp_mask & __GFP_HIGH)
1656 alloc_flags |= ALLOC_HIGH;
1658 alloc_flags |= ALLOC_CPUSET;
1662 * Go through the zonelist again. Let __GFP_HIGH and allocations
1663 * coming from realtime tasks go deeper into reserves.
1665 * This is the last chance, in general, before the goto nopage.
1666 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1667 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1669 page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist,
1670 high_zoneidx, alloc_flags,
1676 /* Allocate without watermarks if the context allows */
1677 if (is_allocation_high_priority(p, gfp_mask)) {
1678 /* Do not dip into emergency reserves if specified */
1679 if (!(gfp_mask & __GFP_NOMEMALLOC)) {
1680 page = __alloc_pages_high_priority(gfp_mask, order,
1681 zonelist, high_zoneidx, nodemask, preferred_zone);
1686 /* Ensure no recursion into the allocator */
1690 /* Atomic allocations - we can't balance anything */
1694 /* Try direct reclaim and then allocating */
1695 page = __alloc_pages_direct_reclaim(gfp_mask, order,
1696 zonelist, high_zoneidx,
1698 alloc_flags, preferred_zone,
1699 &did_some_progress);
1704 * If we failed to make any progress reclaiming, then we are
1705 * running out of options and have to consider going OOM
1707 if (!did_some_progress) {
1708 if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
1709 page = __alloc_pages_may_oom(gfp_mask, order,
1710 zonelist, high_zoneidx,
1711 nodemask, preferred_zone);
1716 * The OOM killer does not trigger for high-order allocations
1717 * but if no progress is being made, there are no other
1718 * options and retrying is unlikely to help
1720 if (order > PAGE_ALLOC_COSTLY_ORDER)
1727 /* Check if we should retry the allocation */
1728 pages_reclaimed += did_some_progress;
1729 if (should_alloc_retry(gfp_mask, order, pages_reclaimed)) {
1730 /* Wait for some write requests to complete then retry */
1731 congestion_wait(WRITE, HZ/50);
1736 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
1737 printk(KERN_WARNING "%s: page allocation failure."
1738 " order:%d, mode:0x%x\n",
1739 p->comm, order, gfp_mask);
1749 * This is the 'heart' of the zoned buddy allocator.
1752 __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
1753 struct zonelist *zonelist, nodemask_t *nodemask)
1755 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
1756 struct zone *preferred_zone;
1759 lockdep_trace_alloc(gfp_mask);
1761 might_sleep_if(gfp_mask & __GFP_WAIT);
1763 if (should_fail_alloc_page(gfp_mask, order))
1767 * Check the zones suitable for the gfp_mask contain at least one
1768 * valid zone. It's possible to have an empty zonelist as a result
1769 * of GFP_THISNODE and a memoryless node
1771 if (unlikely(!zonelist->_zonerefs->zone))
1774 /* The preferred zone is used for statistics later */
1775 first_zones_zonelist(zonelist, high_zoneidx, nodemask, &preferred_zone);
1776 if (!preferred_zone)
1779 /* First allocation attempt */
1780 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order,
1781 zonelist, high_zoneidx, ALLOC_WMARK_LOW|ALLOC_CPUSET,
1783 if (unlikely(!page))
1784 page = __alloc_pages_slowpath(gfp_mask, order,
1785 zonelist, high_zoneidx, nodemask,
1790 EXPORT_SYMBOL(__alloc_pages_nodemask);
1793 * Common helper functions.
1795 unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
1798 page = alloc_pages(gfp_mask, order);
1801 return (unsigned long) page_address(page);
1804 EXPORT_SYMBOL(__get_free_pages);
1806 unsigned long get_zeroed_page(gfp_t gfp_mask)
1811 * get_zeroed_page() returns a 32-bit address, which cannot represent
1814 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
1816 page = alloc_pages(gfp_mask | __GFP_ZERO, 0);
1818 return (unsigned long) page_address(page);
1822 EXPORT_SYMBOL(get_zeroed_page);
1824 void __pagevec_free(struct pagevec *pvec)
1826 int i = pagevec_count(pvec);
1829 free_hot_cold_page(pvec->pages[i], pvec->cold);
1832 void __free_pages(struct page *page, unsigned int order)
1834 if (put_page_testzero(page)) {
1836 free_hot_page(page);
1838 __free_pages_ok(page, order);
1842 EXPORT_SYMBOL(__free_pages);
1844 void free_pages(unsigned long addr, unsigned int order)
1847 VM_BUG_ON(!virt_addr_valid((void *)addr));
1848 __free_pages(virt_to_page((void *)addr), order);
1852 EXPORT_SYMBOL(free_pages);
1855 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
1856 * @size: the number of bytes to allocate
1857 * @gfp_mask: GFP flags for the allocation
1859 * This function is similar to alloc_pages(), except that it allocates the
1860 * minimum number of pages to satisfy the request. alloc_pages() can only
1861 * allocate memory in power-of-two pages.
1863 * This function is also limited by MAX_ORDER.
1865 * Memory allocated by this function must be released by free_pages_exact().
1867 void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
1869 unsigned int order = get_order(size);
1872 addr = __get_free_pages(gfp_mask, order);
1874 unsigned long alloc_end = addr + (PAGE_SIZE << order);
1875 unsigned long used = addr + PAGE_ALIGN(size);
1877 split_page(virt_to_page(addr), order);
1878 while (used < alloc_end) {
1884 return (void *)addr;
1886 EXPORT_SYMBOL(alloc_pages_exact);
1889 * free_pages_exact - release memory allocated via alloc_pages_exact()
1890 * @virt: the value returned by alloc_pages_exact.
1891 * @size: size of allocation, same value as passed to alloc_pages_exact().
1893 * Release the memory allocated by a previous call to alloc_pages_exact.
1895 void free_pages_exact(void *virt, size_t size)
1897 unsigned long addr = (unsigned long)virt;
1898 unsigned long end = addr + PAGE_ALIGN(size);
1900 while (addr < end) {
1905 EXPORT_SYMBOL(free_pages_exact);
1907 static unsigned int nr_free_zone_pages(int offset)
1912 /* Just pick one node, since fallback list is circular */
1913 unsigned int sum = 0;
1915 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
1917 for_each_zone_zonelist(zone, z, zonelist, offset) {
1918 unsigned long size = zone->present_pages;
1919 unsigned long high = zone->pages_high;
1928 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1930 unsigned int nr_free_buffer_pages(void)
1932 return nr_free_zone_pages(gfp_zone(GFP_USER));
1934 EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
1937 * Amount of free RAM allocatable within all zones
1939 unsigned int nr_free_pagecache_pages(void)
1941 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
1944 static inline void show_node(struct zone *zone)
1947 printk("Node %d ", zone_to_nid(zone));
1950 void si_meminfo(struct sysinfo *val)
1952 val->totalram = totalram_pages;
1954 val->freeram = global_page_state(NR_FREE_PAGES);
1955 val->bufferram = nr_blockdev_pages();
1956 val->totalhigh = totalhigh_pages;
1957 val->freehigh = nr_free_highpages();
1958 val->mem_unit = PAGE_SIZE;
1961 EXPORT_SYMBOL(si_meminfo);
1964 void si_meminfo_node(struct sysinfo *val, int nid)
1966 pg_data_t *pgdat = NODE_DATA(nid);
1968 val->totalram = pgdat->node_present_pages;
1969 val->freeram = node_page_state(nid, NR_FREE_PAGES);
1970 #ifdef CONFIG_HIGHMEM
1971 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1972 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
1978 val->mem_unit = PAGE_SIZE;
1982 #define K(x) ((x) << (PAGE_SHIFT-10))
1985 * Show free area list (used inside shift_scroll-lock stuff)
1986 * We also calculate the percentage fragmentation. We do this by counting the
1987 * memory on each free list with the exception of the first item on the list.
1989 void show_free_areas(void)
1994 for_each_populated_zone(zone) {
1996 printk("%s per-cpu:\n", zone->name);
1998 for_each_online_cpu(cpu) {
1999 struct per_cpu_pageset *pageset;
2001 pageset = zone_pcp(zone, cpu);
2003 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
2004 cpu, pageset->pcp.high,
2005 pageset->pcp.batch, pageset->pcp.count);
2009 printk("Active_anon:%lu active_file:%lu inactive_anon:%lu\n"
2010 " inactive_file:%lu"
2011 //TODO: check/adjust line lengths
2012 #ifdef CONFIG_UNEVICTABLE_LRU
2015 " dirty:%lu writeback:%lu unstable:%lu\n"
2016 " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
2017 global_page_state(NR_ACTIVE_ANON),
2018 global_page_state(NR_ACTIVE_FILE),
2019 global_page_state(NR_INACTIVE_ANON),
2020 global_page_state(NR_INACTIVE_FILE),
2021 #ifdef CONFIG_UNEVICTABLE_LRU
2022 global_page_state(NR_UNEVICTABLE),
2024 global_page_state(NR_FILE_DIRTY),
2025 global_page_state(NR_WRITEBACK),
2026 global_page_state(NR_UNSTABLE_NFS),
2027 global_page_state(NR_FREE_PAGES),
2028 global_page_state(NR_SLAB_RECLAIMABLE) +
2029 global_page_state(NR_SLAB_UNRECLAIMABLE),
2030 global_page_state(NR_FILE_MAPPED),
2031 global_page_state(NR_PAGETABLE),
2032 global_page_state(NR_BOUNCE));
2034 for_each_populated_zone(zone) {
2043 " active_anon:%lukB"
2044 " inactive_anon:%lukB"
2045 " active_file:%lukB"
2046 " inactive_file:%lukB"
2047 #ifdef CONFIG_UNEVICTABLE_LRU
2048 " unevictable:%lukB"
2051 " pages_scanned:%lu"
2052 " all_unreclaimable? %s"
2055 K(zone_page_state(zone, NR_FREE_PAGES)),
2058 K(zone->pages_high),
2059 K(zone_page_state(zone, NR_ACTIVE_ANON)),
2060 K(zone_page_state(zone, NR_INACTIVE_ANON)),
2061 K(zone_page_state(zone, NR_ACTIVE_FILE)),
2062 K(zone_page_state(zone, NR_INACTIVE_FILE)),
2063 #ifdef CONFIG_UNEVICTABLE_LRU
2064 K(zone_page_state(zone, NR_UNEVICTABLE)),
2066 K(zone->present_pages),
2067 zone->pages_scanned,
2068 (zone_is_all_unreclaimable(zone) ? "yes" : "no")
2070 printk("lowmem_reserve[]:");
2071 for (i = 0; i < MAX_NR_ZONES; i++)
2072 printk(" %lu", zone->lowmem_reserve[i]);
2076 for_each_populated_zone(zone) {
2077 unsigned long nr[MAX_ORDER], flags, order, total = 0;
2080 printk("%s: ", zone->name);
2082 spin_lock_irqsave(&zone->lock, flags);
2083 for (order = 0; order < MAX_ORDER; order++) {
2084 nr[order] = zone->free_area[order].nr_free;
2085 total += nr[order] << order;
2087 spin_unlock_irqrestore(&zone->lock, flags);
2088 for (order = 0; order < MAX_ORDER; order++)
2089 printk("%lu*%lukB ", nr[order], K(1UL) << order);
2090 printk("= %lukB\n", K(total));
2093 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
2095 show_swap_cache_info();
2098 static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
2100 zoneref->zone = zone;
2101 zoneref->zone_idx = zone_idx(zone);
2105 * Builds allocation fallback zone lists.
2107 * Add all populated zones of a node to the zonelist.
2109 static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
2110 int nr_zones, enum zone_type zone_type)
2114 BUG_ON(zone_type >= MAX_NR_ZONES);
2119 zone = pgdat->node_zones + zone_type;
2120 if (populated_zone(zone)) {
2121 zoneref_set_zone(zone,
2122 &zonelist->_zonerefs[nr_zones++]);
2123 check_highest_zone(zone_type);
2126 } while (zone_type);
2133 * 0 = automatic detection of better ordering.
2134 * 1 = order by ([node] distance, -zonetype)
2135 * 2 = order by (-zonetype, [node] distance)
2137 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
2138 * the same zonelist. So only NUMA can configure this param.
2140 #define ZONELIST_ORDER_DEFAULT 0
2141 #define ZONELIST_ORDER_NODE 1
2142 #define ZONELIST_ORDER_ZONE 2
2144 /* zonelist order in the kernel.
2145 * set_zonelist_order() will set this to NODE or ZONE.
2147 static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
2148 static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
2152 /* The value user specified ....changed by config */
2153 static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2154 /* string for sysctl */
2155 #define NUMA_ZONELIST_ORDER_LEN 16
2156 char numa_zonelist_order[16] = "default";
2159 * interface for configure zonelist ordering.
2160 * command line option "numa_zonelist_order"
2161 * = "[dD]efault - default, automatic configuration.
2162 * = "[nN]ode - order by node locality, then by zone within node
2163 * = "[zZ]one - order by zone, then by locality within zone
2166 static int __parse_numa_zonelist_order(char *s)
2168 if (*s == 'd' || *s == 'D') {
2169 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2170 } else if (*s == 'n' || *s == 'N') {
2171 user_zonelist_order = ZONELIST_ORDER_NODE;
2172 } else if (*s == 'z' || *s == 'Z') {
2173 user_zonelist_order = ZONELIST_ORDER_ZONE;
2176 "Ignoring invalid numa_zonelist_order value: "
2183 static __init int setup_numa_zonelist_order(char *s)
2186 return __parse_numa_zonelist_order(s);
2189 early_param("numa_zonelist_order", setup_numa_zonelist_order);
2192 * sysctl handler for numa_zonelist_order
2194 int numa_zonelist_order_handler(ctl_table *table, int write,
2195 struct file *file, void __user *buffer, size_t *length,
2198 char saved_string[NUMA_ZONELIST_ORDER_LEN];
2202 strncpy(saved_string, (char*)table->data,
2203 NUMA_ZONELIST_ORDER_LEN);
2204 ret = proc_dostring(table, write, file, buffer, length, ppos);
2208 int oldval = user_zonelist_order;
2209 if (__parse_numa_zonelist_order((char*)table->data)) {
2211 * bogus value. restore saved string
2213 strncpy((char*)table->data, saved_string,
2214 NUMA_ZONELIST_ORDER_LEN);
2215 user_zonelist_order = oldval;
2216 } else if (oldval != user_zonelist_order)
2217 build_all_zonelists();
2223 #define MAX_NODE_LOAD (num_online_nodes())
2224 static int node_load[MAX_NUMNODES];
2227 * find_next_best_node - find the next node that should appear in a given node's fallback list
2228 * @node: node whose fallback list we're appending
2229 * @used_node_mask: nodemask_t of already used nodes
2231 * We use a number of factors to determine which is the next node that should
2232 * appear on a given node's fallback list. The node should not have appeared
2233 * already in @node's fallback list, and it should be the next closest node
2234 * according to the distance array (which contains arbitrary distance values
2235 * from each node to each node in the system), and should also prefer nodes
2236 * with no CPUs, since presumably they'll have very little allocation pressure
2237 * on them otherwise.
2238 * It returns -1 if no node is found.
2240 static int find_next_best_node(int node, nodemask_t *used_node_mask)
2243 int min_val = INT_MAX;
2245 const struct cpumask *tmp = cpumask_of_node(0);
2247 /* Use the local node if we haven't already */
2248 if (!node_isset(node, *used_node_mask)) {
2249 node_set(node, *used_node_mask);
2253 for_each_node_state(n, N_HIGH_MEMORY) {
2255 /* Don't want a node to appear more than once */
2256 if (node_isset(n, *used_node_mask))
2259 /* Use the distance array to find the distance */
2260 val = node_distance(node, n);
2262 /* Penalize nodes under us ("prefer the next node") */
2265 /* Give preference to headless and unused nodes */
2266 tmp = cpumask_of_node(n);
2267 if (!cpumask_empty(tmp))
2268 val += PENALTY_FOR_NODE_WITH_CPUS;
2270 /* Slight preference for less loaded node */
2271 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
2272 val += node_load[n];
2274 if (val < min_val) {
2281 node_set(best_node, *used_node_mask);
2288 * Build zonelists ordered by node and zones within node.
2289 * This results in maximum locality--normal zone overflows into local
2290 * DMA zone, if any--but risks exhausting DMA zone.
2292 static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
2295 struct zonelist *zonelist;
2297 zonelist = &pgdat->node_zonelists[0];
2298 for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
2300 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2302 zonelist->_zonerefs[j].zone = NULL;
2303 zonelist->_zonerefs[j].zone_idx = 0;
2307 * Build gfp_thisnode zonelists
2309 static void build_thisnode_zonelists(pg_data_t *pgdat)
2312 struct zonelist *zonelist;
2314 zonelist = &pgdat->node_zonelists[1];
2315 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
2316 zonelist->_zonerefs[j].zone = NULL;
2317 zonelist->_zonerefs[j].zone_idx = 0;
2321 * Build zonelists ordered by zone and nodes within zones.
2322 * This results in conserving DMA zone[s] until all Normal memory is
2323 * exhausted, but results in overflowing to remote node while memory
2324 * may still exist in local DMA zone.
2326 static int node_order[MAX_NUMNODES];
2328 static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
2331 int zone_type; /* needs to be signed */
2333 struct zonelist *zonelist;
2335 zonelist = &pgdat->node_zonelists[0];
2337 for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
2338 for (j = 0; j < nr_nodes; j++) {
2339 node = node_order[j];
2340 z = &NODE_DATA(node)->node_zones[zone_type];
2341 if (populated_zone(z)) {
2343 &zonelist->_zonerefs[pos++]);
2344 check_highest_zone(zone_type);
2348 zonelist->_zonerefs[pos].zone = NULL;
2349 zonelist->_zonerefs[pos].zone_idx = 0;
2352 static int default_zonelist_order(void)
2355 unsigned long low_kmem_size,total_size;
2359 * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem.
2360 * If they are really small and used heavily, the system can fall
2361 * into OOM very easily.
2362 * This function detect ZONE_DMA/DMA32 size and confgigures zone order.
2364 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2367 for_each_online_node(nid) {
2368 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2369 z = &NODE_DATA(nid)->node_zones[zone_type];
2370 if (populated_zone(z)) {
2371 if (zone_type < ZONE_NORMAL)
2372 low_kmem_size += z->present_pages;
2373 total_size += z->present_pages;
2377 if (!low_kmem_size || /* there are no DMA area. */
2378 low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
2379 return ZONELIST_ORDER_NODE;
2381 * look into each node's config.
2382 * If there is a node whose DMA/DMA32 memory is very big area on
2383 * local memory, NODE_ORDER may be suitable.
2385 average_size = total_size /
2386 (nodes_weight(node_states[N_HIGH_MEMORY]) + 1);
2387 for_each_online_node(nid) {
2390 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2391 z = &NODE_DATA(nid)->node_zones[zone_type];
2392 if (populated_zone(z)) {
2393 if (zone_type < ZONE_NORMAL)
2394 low_kmem_size += z->present_pages;
2395 total_size += z->present_pages;
2398 if (low_kmem_size &&
2399 total_size > average_size && /* ignore small node */
2400 low_kmem_size > total_size * 70/100)
2401 return ZONELIST_ORDER_NODE;
2403 return ZONELIST_ORDER_ZONE;
2406 static void set_zonelist_order(void)
2408 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
2409 current_zonelist_order = default_zonelist_order();
2411 current_zonelist_order = user_zonelist_order;
2414 static void build_zonelists(pg_data_t *pgdat)
2418 nodemask_t used_mask;
2419 int local_node, prev_node;
2420 struct zonelist *zonelist;
2421 int order = current_zonelist_order;
2423 /* initialize zonelists */
2424 for (i = 0; i < MAX_ZONELISTS; i++) {
2425 zonelist = pgdat->node_zonelists + i;
2426 zonelist->_zonerefs[0].zone = NULL;
2427 zonelist->_zonerefs[0].zone_idx = 0;
2430 /* NUMA-aware ordering of nodes */
2431 local_node = pgdat->node_id;
2432 load = num_online_nodes();
2433 prev_node = local_node;
2434 nodes_clear(used_mask);
2436 memset(node_load, 0, sizeof(node_load));
2437 memset(node_order, 0, sizeof(node_order));
2440 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
2441 int distance = node_distance(local_node, node);
2444 * If another node is sufficiently far away then it is better
2445 * to reclaim pages in a zone before going off node.
2447 if (distance > RECLAIM_DISTANCE)
2448 zone_reclaim_mode = 1;
2451 * We don't want to pressure a particular node.
2452 * So adding penalty to the first node in same
2453 * distance group to make it round-robin.
2455 if (distance != node_distance(local_node, prev_node))
2456 node_load[node] = load;
2460 if (order == ZONELIST_ORDER_NODE)
2461 build_zonelists_in_node_order(pgdat, node);
2463 node_order[j++] = node; /* remember order */
2466 if (order == ZONELIST_ORDER_ZONE) {
2467 /* calculate node order -- i.e., DMA last! */
2468 build_zonelists_in_zone_order(pgdat, j);
2471 build_thisnode_zonelists(pgdat);
2474 /* Construct the zonelist performance cache - see further mmzone.h */
2475 static void build_zonelist_cache(pg_data_t *pgdat)
2477 struct zonelist *zonelist;
2478 struct zonelist_cache *zlc;
2481 zonelist = &pgdat->node_zonelists[0];
2482 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
2483 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
2484 for (z = zonelist->_zonerefs; z->zone; z++)
2485 zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z);
2489 #else /* CONFIG_NUMA */
2491 static void set_zonelist_order(void)
2493 current_zonelist_order = ZONELIST_ORDER_ZONE;
2496 static void build_zonelists(pg_data_t *pgdat)
2498 int node, local_node;
2500 struct zonelist *zonelist;
2502 local_node = pgdat->node_id;
2504 zonelist = &pgdat->node_zonelists[0];
2505 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
2508 * Now we build the zonelist so that it contains the zones
2509 * of all the other nodes.
2510 * We don't want to pressure a particular node, so when
2511 * building the zones for node N, we make sure that the
2512 * zones coming right after the local ones are those from
2513 * node N+1 (modulo N)
2515 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
2516 if (!node_online(node))
2518 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2521 for (node = 0; node < local_node; node++) {
2522 if (!node_online(node))
2524 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2528 zonelist->_zonerefs[j].zone = NULL;
2529 zonelist->_zonerefs[j].zone_idx = 0;
2532 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
2533 static void build_zonelist_cache(pg_data_t *pgdat)
2535 pgdat->node_zonelists[0].zlcache_ptr = NULL;
2538 #endif /* CONFIG_NUMA */
2540 /* return values int ....just for stop_machine() */
2541 static int __build_all_zonelists(void *dummy)
2545 for_each_online_node(nid) {
2546 pg_data_t *pgdat = NODE_DATA(nid);
2548 build_zonelists(pgdat);
2549 build_zonelist_cache(pgdat);
2554 void build_all_zonelists(void)
2556 set_zonelist_order();
2558 if (system_state == SYSTEM_BOOTING) {
2559 __build_all_zonelists(NULL);
2560 mminit_verify_zonelist();
2561 cpuset_init_current_mems_allowed();
2563 /* we have to stop all cpus to guarantee there is no user
2565 stop_machine(__build_all_zonelists, NULL, NULL);
2566 /* cpuset refresh routine should be here */
2568 vm_total_pages = nr_free_pagecache_pages();
2570 * Disable grouping by mobility if the number of pages in the
2571 * system is too low to allow the mechanism to work. It would be
2572 * more accurate, but expensive to check per-zone. This check is
2573 * made on memory-hotadd so a system can start with mobility
2574 * disabled and enable it later
2576 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
2577 page_group_by_mobility_disabled = 1;
2579 page_group_by_mobility_disabled = 0;
2581 printk("Built %i zonelists in %s order, mobility grouping %s. "
2582 "Total pages: %ld\n",
2584 zonelist_order_name[current_zonelist_order],
2585 page_group_by_mobility_disabled ? "off" : "on",
2588 printk("Policy zone: %s\n", zone_names[policy_zone]);
2593 * Helper functions to size the waitqueue hash table.
2594 * Essentially these want to choose hash table sizes sufficiently
2595 * large so that collisions trying to wait on pages are rare.
2596 * But in fact, the number of active page waitqueues on typical
2597 * systems is ridiculously low, less than 200. So this is even
2598 * conservative, even though it seems large.
2600 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
2601 * waitqueues, i.e. the size of the waitq table given the number of pages.
2603 #define PAGES_PER_WAITQUEUE 256
2605 #ifndef CONFIG_MEMORY_HOTPLUG
2606 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
2608 unsigned long size = 1;
2610 pages /= PAGES_PER_WAITQUEUE;
2612 while (size < pages)
2616 * Once we have dozens or even hundreds of threads sleeping
2617 * on IO we've got bigger problems than wait queue collision.
2618 * Limit the size of the wait table to a reasonable size.
2620 size = min(size, 4096UL);
2622 return max(size, 4UL);
2626 * A zone's size might be changed by hot-add, so it is not possible to determine
2627 * a suitable size for its wait_table. So we use the maximum size now.
2629 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
2631 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
2632 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
2633 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
2635 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
2636 * or more by the traditional way. (See above). It equals:
2638 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
2639 * ia64(16K page size) : = ( 8G + 4M)byte.
2640 * powerpc (64K page size) : = (32G +16M)byte.
2642 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
2649 * This is an integer logarithm so that shifts can be used later
2650 * to extract the more random high bits from the multiplicative
2651 * hash function before the remainder is taken.
2653 static inline unsigned long wait_table_bits(unsigned long size)
2658 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
2661 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
2662 * of blocks reserved is based on zone->pages_min. The memory within the
2663 * reserve will tend to store contiguous free pages. Setting min_free_kbytes
2664 * higher will lead to a bigger reserve which will get freed as contiguous
2665 * blocks as reclaim kicks in
2667 static void setup_zone_migrate_reserve(struct zone *zone)
2669 unsigned long start_pfn, pfn, end_pfn;
2671 unsigned long reserve, block_migratetype;
2673 /* Get the start pfn, end pfn and the number of blocks to reserve */
2674 start_pfn = zone->zone_start_pfn;
2675 end_pfn = start_pfn + zone->spanned_pages;
2676 reserve = roundup(zone->pages_min, pageblock_nr_pages) >>
2679 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
2680 if (!pfn_valid(pfn))
2682 page = pfn_to_page(pfn);
2684 /* Watch out for overlapping nodes */
2685 if (page_to_nid(page) != zone_to_nid(zone))
2688 /* Blocks with reserved pages will never free, skip them. */
2689 if (PageReserved(page))
2692 block_migratetype = get_pageblock_migratetype(page);
2694 /* If this block is reserved, account for it */
2695 if (reserve > 0 && block_migratetype == MIGRATE_RESERVE) {
2700 /* Suitable for reserving if this block is movable */
2701 if (reserve > 0 && block_migratetype == MIGRATE_MOVABLE) {
2702 set_pageblock_migratetype(page, MIGRATE_RESERVE);
2703 move_freepages_block(zone, page, MIGRATE_RESERVE);
2709 * If the reserve is met and this is a previous reserved block,
2712 if (block_migratetype == MIGRATE_RESERVE) {
2713 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
2714 move_freepages_block(zone, page, MIGRATE_MOVABLE);
2720 * Initially all pages are reserved - free ones are freed
2721 * up by free_all_bootmem() once the early boot process is
2722 * done. Non-atomic initialization, single-pass.
2724 void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
2725 unsigned long start_pfn, enum memmap_context context)
2728 unsigned long end_pfn = start_pfn + size;
2732 if (highest_memmap_pfn < end_pfn - 1)
2733 highest_memmap_pfn = end_pfn - 1;
2735 z = &NODE_DATA(nid)->node_zones[zone];
2736 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
2738 * There can be holes in boot-time mem_map[]s
2739 * handed to this function. They do not
2740 * exist on hotplugged memory.
2742 if (context == MEMMAP_EARLY) {
2743 if (!early_pfn_valid(pfn))
2745 if (!early_pfn_in_nid(pfn, nid))
2748 page = pfn_to_page(pfn);
2749 set_page_links(page, zone, nid, pfn);
2750 mminit_verify_page_links(page, zone, nid, pfn);
2751 init_page_count(page);
2752 reset_page_mapcount(page);
2753 SetPageReserved(page);
2755 * Mark the block movable so that blocks are reserved for
2756 * movable at startup. This will force kernel allocations
2757 * to reserve their blocks rather than leaking throughout
2758 * the address space during boot when many long-lived
2759 * kernel allocations are made. Later some blocks near
2760 * the start are marked MIGRATE_RESERVE by
2761 * setup_zone_migrate_reserve()
2763 * bitmap is created for zone's valid pfn range. but memmap
2764 * can be created for invalid pages (for alignment)
2765 * check here not to call set_pageblock_migratetype() against
2768 if ((z->zone_start_pfn <= pfn)
2769 && (pfn < z->zone_start_pfn + z->spanned_pages)
2770 && !(pfn & (pageblock_nr_pages - 1)))
2771 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
2773 INIT_LIST_HEAD(&page->lru);
2774 #ifdef WANT_PAGE_VIRTUAL
2775 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
2776 if (!is_highmem_idx(zone))
2777 set_page_address(page, __va(pfn << PAGE_SHIFT));
2782 static void __meminit zone_init_free_lists(struct zone *zone)
2785 for_each_migratetype_order(order, t) {
2786 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
2787 zone->free_area[order].nr_free = 0;
2791 #ifndef __HAVE_ARCH_MEMMAP_INIT
2792 #define memmap_init(size, nid, zone, start_pfn) \
2793 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
2796 static int zone_batchsize(struct zone *zone)
2802 * The per-cpu-pages pools are set to around 1000th of the
2803 * size of the zone. But no more than 1/2 of a meg.
2805 * OK, so we don't know how big the cache is. So guess.
2807 batch = zone->present_pages / 1024;
2808 if (batch * PAGE_SIZE > 512 * 1024)
2809 batch = (512 * 1024) / PAGE_SIZE;
2810 batch /= 4; /* We effectively *= 4 below */
2815 * Clamp the batch to a 2^n - 1 value. Having a power
2816 * of 2 value was found to be more likely to have
2817 * suboptimal cache aliasing properties in some cases.
2819 * For example if 2 tasks are alternately allocating
2820 * batches of pages, one task can end up with a lot
2821 * of pages of one half of the possible page colors
2822 * and the other with pages of the other colors.
2824 batch = rounddown_pow_of_two(batch + batch/2) - 1;
2829 /* The deferral and batching of frees should be suppressed under NOMMU
2832 * The problem is that NOMMU needs to be able to allocate large chunks
2833 * of contiguous memory as there's no hardware page translation to
2834 * assemble apparent contiguous memory from discontiguous pages.
2836 * Queueing large contiguous runs of pages for batching, however,
2837 * causes the pages to actually be freed in smaller chunks. As there
2838 * can be a significant delay between the individual batches being
2839 * recycled, this leads to the once large chunks of space being
2840 * fragmented and becoming unavailable for high-order allocations.
2846 static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
2848 struct per_cpu_pages *pcp;
2850 memset(p, 0, sizeof(*p));
2854 pcp->high = 6 * batch;
2855 pcp->batch = max(1UL, 1 * batch);
2856 INIT_LIST_HEAD(&pcp->list);
2860 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
2861 * to the value high for the pageset p.
2864 static void setup_pagelist_highmark(struct per_cpu_pageset *p,
2867 struct per_cpu_pages *pcp;
2871 pcp->batch = max(1UL, high/4);
2872 if ((high/4) > (PAGE_SHIFT * 8))
2873 pcp->batch = PAGE_SHIFT * 8;
2879 * Boot pageset table. One per cpu which is going to be used for all
2880 * zones and all nodes. The parameters will be set in such a way
2881 * that an item put on a list will immediately be handed over to
2882 * the buddy list. This is safe since pageset manipulation is done
2883 * with interrupts disabled.
2885 * Some NUMA counter updates may also be caught by the boot pagesets.
2887 * The boot_pagesets must be kept even after bootup is complete for
2888 * unused processors and/or zones. They do play a role for bootstrapping
2889 * hotplugged processors.
2891 * zoneinfo_show() and maybe other functions do
2892 * not check if the processor is online before following the pageset pointer.
2893 * Other parts of the kernel may not check if the zone is available.
2895 static struct per_cpu_pageset boot_pageset[NR_CPUS];
2898 * Dynamically allocate memory for the
2899 * per cpu pageset array in struct zone.
2901 static int __cpuinit process_zones(int cpu)
2903 struct zone *zone, *dzone;
2904 int node = cpu_to_node(cpu);
2906 node_set_state(node, N_CPU); /* this node has a cpu */
2908 for_each_populated_zone(zone) {
2909 zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset),
2911 if (!zone_pcp(zone, cpu))
2914 setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone));
2916 if (percpu_pagelist_fraction)
2917 setup_pagelist_highmark(zone_pcp(zone, cpu),
2918 (zone->present_pages / percpu_pagelist_fraction));
2923 for_each_zone(dzone) {
2924 if (!populated_zone(dzone))
2928 kfree(zone_pcp(dzone, cpu));
2929 zone_pcp(dzone, cpu) = NULL;
2934 static inline void free_zone_pagesets(int cpu)
2938 for_each_zone(zone) {
2939 struct per_cpu_pageset *pset = zone_pcp(zone, cpu);
2941 /* Free per_cpu_pageset if it is slab allocated */
2942 if (pset != &boot_pageset[cpu])
2944 zone_pcp(zone, cpu) = NULL;
2948 static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb,
2949 unsigned long action,
2952 int cpu = (long)hcpu;
2953 int ret = NOTIFY_OK;
2956 case CPU_UP_PREPARE:
2957 case CPU_UP_PREPARE_FROZEN:
2958 if (process_zones(cpu))
2961 case CPU_UP_CANCELED:
2962 case CPU_UP_CANCELED_FROZEN:
2964 case CPU_DEAD_FROZEN:
2965 free_zone_pagesets(cpu);
2973 static struct notifier_block __cpuinitdata pageset_notifier =
2974 { &pageset_cpuup_callback, NULL, 0 };
2976 void __init setup_per_cpu_pageset(void)
2980 /* Initialize per_cpu_pageset for cpu 0.
2981 * A cpuup callback will do this for every cpu
2982 * as it comes online
2984 err = process_zones(smp_processor_id());
2986 register_cpu_notifier(&pageset_notifier);
2991 static noinline __init_refok
2992 int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
2995 struct pglist_data *pgdat = zone->zone_pgdat;
2999 * The per-page waitqueue mechanism uses hashed waitqueues
3002 zone->wait_table_hash_nr_entries =
3003 wait_table_hash_nr_entries(zone_size_pages);
3004 zone->wait_table_bits =
3005 wait_table_bits(zone->wait_table_hash_nr_entries);
3006 alloc_size = zone->wait_table_hash_nr_entries
3007 * sizeof(wait_queue_head_t);
3009 if (!slab_is_available()) {
3010 zone->wait_table = (wait_queue_head_t *)
3011 alloc_bootmem_node(pgdat, alloc_size);
3014 * This case means that a zone whose size was 0 gets new memory
3015 * via memory hot-add.
3016 * But it may be the case that a new node was hot-added. In
3017 * this case vmalloc() will not be able to use this new node's
3018 * memory - this wait_table must be initialized to use this new
3019 * node itself as well.
3020 * To use this new node's memory, further consideration will be
3023 zone->wait_table = vmalloc(alloc_size);
3025 if (!zone->wait_table)
3028 for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
3029 init_waitqueue_head(zone->wait_table + i);
3034 static __meminit void zone_pcp_init(struct zone *zone)
3037 unsigned long batch = zone_batchsize(zone);
3039 for (cpu = 0; cpu < NR_CPUS; cpu++) {
3041 /* Early boot. Slab allocator not functional yet */
3042 zone_pcp(zone, cpu) = &boot_pageset[cpu];
3043 setup_pageset(&boot_pageset[cpu],0);
3045 setup_pageset(zone_pcp(zone,cpu), batch);
3048 if (zone->present_pages)
3049 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n",
3050 zone->name, zone->present_pages, batch);
3053 __meminit int init_currently_empty_zone(struct zone *zone,
3054 unsigned long zone_start_pfn,
3056 enum memmap_context context)
3058 struct pglist_data *pgdat = zone->zone_pgdat;
3060 ret = zone_wait_table_init(zone, size);
3063 pgdat->nr_zones = zone_idx(zone) + 1;
3065 zone->zone_start_pfn = zone_start_pfn;
3067 mminit_dprintk(MMINIT_TRACE, "memmap_init",
3068 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
3070 (unsigned long)zone_idx(zone),
3071 zone_start_pfn, (zone_start_pfn + size));
3073 zone_init_free_lists(zone);
3078 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3080 * Basic iterator support. Return the first range of PFNs for a node
3081 * Note: nid == MAX_NUMNODES returns first region regardless of node
3083 static int __meminit first_active_region_index_in_nid(int nid)
3087 for (i = 0; i < nr_nodemap_entries; i++)
3088 if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
3095 * Basic iterator support. Return the next active range of PFNs for a node
3096 * Note: nid == MAX_NUMNODES returns next region regardless of node
3098 static int __meminit next_active_region_index_in_nid(int index, int nid)
3100 for (index = index + 1; index < nr_nodemap_entries; index++)
3101 if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
3107 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
3109 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
3110 * Architectures may implement their own version but if add_active_range()
3111 * was used and there are no special requirements, this is a convenient
3114 int __meminit __early_pfn_to_nid(unsigned long pfn)
3118 for (i = 0; i < nr_nodemap_entries; i++) {
3119 unsigned long start_pfn = early_node_map[i].start_pfn;
3120 unsigned long end_pfn = early_node_map[i].end_pfn;
3122 if (start_pfn <= pfn && pfn < end_pfn)
3123 return early_node_map[i].nid;
3125 /* This is a memory hole */
3128 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
3130 int __meminit early_pfn_to_nid(unsigned long pfn)
3134 nid = __early_pfn_to_nid(pfn);
3137 /* just returns 0 */
3141 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
3142 bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
3146 nid = __early_pfn_to_nid(pfn);
3147 if (nid >= 0 && nid != node)
3153 /* Basic iterator support to walk early_node_map[] */
3154 #define for_each_active_range_index_in_nid(i, nid) \
3155 for (i = first_active_region_index_in_nid(nid); i != -1; \
3156 i = next_active_region_index_in_nid(i, nid))
3159 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
3160 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
3161 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
3163 * If an architecture guarantees that all ranges registered with
3164 * add_active_ranges() contain no holes and may be freed, this
3165 * this function may be used instead of calling free_bootmem() manually.
3167 void __init free_bootmem_with_active_regions(int nid,
3168 unsigned long max_low_pfn)
3172 for_each_active_range_index_in_nid(i, nid) {
3173 unsigned long size_pages = 0;
3174 unsigned long end_pfn = early_node_map[i].end_pfn;
3176 if (early_node_map[i].start_pfn >= max_low_pfn)
3179 if (end_pfn > max_low_pfn)
3180 end_pfn = max_low_pfn;
3182 size_pages = end_pfn - early_node_map[i].start_pfn;
3183 free_bootmem_node(NODE_DATA(early_node_map[i].nid),
3184 PFN_PHYS(early_node_map[i].start_pfn),
3185 size_pages << PAGE_SHIFT);
3189 void __init work_with_active_regions(int nid, work_fn_t work_fn, void *data)
3194 for_each_active_range_index_in_nid(i, nid) {
3195 ret = work_fn(early_node_map[i].start_pfn,
3196 early_node_map[i].end_pfn, data);
3202 * sparse_memory_present_with_active_regions - Call memory_present for each active range
3203 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
3205 * If an architecture guarantees that all ranges registered with
3206 * add_active_ranges() contain no holes and may be freed, this
3207 * function may be used instead of calling memory_present() manually.
3209 void __init sparse_memory_present_with_active_regions(int nid)
3213 for_each_active_range_index_in_nid(i, nid)
3214 memory_present(early_node_map[i].nid,
3215 early_node_map[i].start_pfn,
3216 early_node_map[i].end_pfn);
3220 * get_pfn_range_for_nid - Return the start and end page frames for a node
3221 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
3222 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
3223 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
3225 * It returns the start and end page frame of a node based on information
3226 * provided by an arch calling add_active_range(). If called for a node
3227 * with no available memory, a warning is printed and the start and end
3230 void __meminit get_pfn_range_for_nid(unsigned int nid,
3231 unsigned long *start_pfn, unsigned long *end_pfn)
3237 for_each_active_range_index_in_nid(i, nid) {
3238 *start_pfn = min(*start_pfn, early_node_map[i].start_pfn);
3239 *end_pfn = max(*end_pfn, early_node_map[i].end_pfn);
3242 if (*start_pfn == -1UL)
3247 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3248 * assumption is made that zones within a node are ordered in monotonic
3249 * increasing memory addresses so that the "highest" populated zone is used
3251 static void __init find_usable_zone_for_movable(void)
3254 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
3255 if (zone_index == ZONE_MOVABLE)
3258 if (arch_zone_highest_possible_pfn[zone_index] >
3259 arch_zone_lowest_possible_pfn[zone_index])
3263 VM_BUG_ON(zone_index == -1);
3264 movable_zone = zone_index;
3268 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
3269 * because it is sized independant of architecture. Unlike the other zones,
3270 * the starting point for ZONE_MOVABLE is not fixed. It may be different
3271 * in each node depending on the size of each node and how evenly kernelcore
3272 * is distributed. This helper function adjusts the zone ranges
3273 * provided by the architecture for a given node by using the end of the
3274 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3275 * zones within a node are in order of monotonic increases memory addresses
3277 static void __meminit adjust_zone_range_for_zone_movable(int nid,
3278 unsigned long zone_type,
3279 unsigned long node_start_pfn,
3280 unsigned long node_end_pfn,
3281 unsigned long *zone_start_pfn,
3282 unsigned long *zone_end_pfn)
3284 /* Only adjust if ZONE_MOVABLE is on this node */
3285 if (zone_movable_pfn[nid]) {
3286 /* Size ZONE_MOVABLE */
3287 if (zone_type == ZONE_MOVABLE) {
3288 *zone_start_pfn = zone_movable_pfn[nid];
3289 *zone_end_pfn = min(node_end_pfn,
3290 arch_zone_highest_possible_pfn[movable_zone]);
3292 /* Adjust for ZONE_MOVABLE starting within this range */
3293 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
3294 *zone_end_pfn > zone_movable_pfn[nid]) {
3295 *zone_end_pfn = zone_movable_pfn[nid];
3297 /* Check if this whole range is within ZONE_MOVABLE */
3298 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
3299 *zone_start_pfn = *zone_end_pfn;
3304 * Return the number of pages a zone spans in a node, including holes
3305 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3307 static unsigned long __meminit zone_spanned_pages_in_node(int nid,
3308 unsigned long zone_type,
3309 unsigned long *ignored)
3311 unsigned long node_start_pfn, node_end_pfn;
3312 unsigned long zone_start_pfn, zone_end_pfn;
3314 /* Get the start and end of the node and zone */
3315 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3316 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
3317 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
3318 adjust_zone_range_for_zone_movable(nid, zone_type,
3319 node_start_pfn, node_end_pfn,
3320 &zone_start_pfn, &zone_end_pfn);
3322 /* Check that this node has pages within the zone's required range */
3323 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
3326 /* Move the zone boundaries inside the node if necessary */
3327 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
3328 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
3330 /* Return the spanned pages */
3331 return zone_end_pfn - zone_start_pfn;
3335 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
3336 * then all holes in the requested range will be accounted for.
3338 static unsigned long __meminit __absent_pages_in_range(int nid,
3339 unsigned long range_start_pfn,
3340 unsigned long range_end_pfn)
3343 unsigned long prev_end_pfn = 0, hole_pages = 0;
3344 unsigned long start_pfn;
3346 /* Find the end_pfn of the first active range of pfns in the node */
3347 i = first_active_region_index_in_nid(nid);
3351 prev_end_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3353 /* Account for ranges before physical memory on this node */
3354 if (early_node_map[i].start_pfn > range_start_pfn)
3355 hole_pages = prev_end_pfn - range_start_pfn;
3357 /* Find all holes for the zone within the node */
3358 for (; i != -1; i = next_active_region_index_in_nid(i, nid)) {
3360 /* No need to continue if prev_end_pfn is outside the zone */
3361 if (prev_end_pfn >= range_end_pfn)
3364 /* Make sure the end of the zone is not within the hole */
3365 start_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3366 prev_end_pfn = max(prev_end_pfn, range_start_pfn);
3368 /* Update the hole size cound and move on */
3369 if (start_pfn > range_start_pfn) {
3370 BUG_ON(prev_end_pfn > start_pfn);
3371 hole_pages += start_pfn - prev_end_pfn;
3373 prev_end_pfn = early_node_map[i].end_pfn;
3376 /* Account for ranges past physical memory on this node */
3377 if (range_end_pfn > prev_end_pfn)
3378 hole_pages += range_end_pfn -
3379 max(range_start_pfn, prev_end_pfn);
3385 * absent_pages_in_range - Return number of page frames in holes within a range
3386 * @start_pfn: The start PFN to start searching for holes
3387 * @end_pfn: The end PFN to stop searching for holes
3389 * It returns the number of pages frames in memory holes within a range.
3391 unsigned long __init absent_pages_in_range(unsigned long start_pfn,
3392 unsigned long end_pfn)
3394 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
3397 /* Return the number of page frames in holes in a zone on a node */
3398 static unsigned long __meminit zone_absent_pages_in_node(int nid,
3399 unsigned long zone_type,
3400 unsigned long *ignored)
3402 unsigned long node_start_pfn, node_end_pfn;
3403 unsigned long zone_start_pfn, zone_end_pfn;
3405 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3406 zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type],
3408 zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type],
3411 adjust_zone_range_for_zone_movable(nid, zone_type,
3412 node_start_pfn, node_end_pfn,
3413 &zone_start_pfn, &zone_end_pfn);
3414 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
3418 static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
3419 unsigned long zone_type,
3420 unsigned long *zones_size)
3422 return zones_size[zone_type];
3425 static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
3426 unsigned long zone_type,
3427 unsigned long *zholes_size)
3432 return zholes_size[zone_type];
3437 static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
3438 unsigned long *zones_size, unsigned long *zholes_size)
3440 unsigned long realtotalpages, totalpages = 0;
3443 for (i = 0; i < MAX_NR_ZONES; i++)
3444 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
3446 pgdat->node_spanned_pages = totalpages;
3448 realtotalpages = totalpages;
3449 for (i = 0; i < MAX_NR_ZONES; i++)
3451 zone_absent_pages_in_node(pgdat->node_id, i,
3453 pgdat->node_present_pages = realtotalpages;
3454 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
3458 #ifndef CONFIG_SPARSEMEM
3460 * Calculate the size of the zone->blockflags rounded to an unsigned long
3461 * Start by making sure zonesize is a multiple of pageblock_order by rounding
3462 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
3463 * round what is now in bits to nearest long in bits, then return it in
3466 static unsigned long __init usemap_size(unsigned long zonesize)
3468 unsigned long usemapsize;
3470 usemapsize = roundup(zonesize, pageblock_nr_pages);
3471 usemapsize = usemapsize >> pageblock_order;
3472 usemapsize *= NR_PAGEBLOCK_BITS;
3473 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
3475 return usemapsize / 8;
3478 static void __init setup_usemap(struct pglist_data *pgdat,
3479 struct zone *zone, unsigned long zonesize)
3481 unsigned long usemapsize = usemap_size(zonesize);
3482 zone->pageblock_flags = NULL;
3484 zone->pageblock_flags = alloc_bootmem_node(pgdat, usemapsize);
3487 static void inline setup_usemap(struct pglist_data *pgdat,
3488 struct zone *zone, unsigned long zonesize) {}
3489 #endif /* CONFIG_SPARSEMEM */
3491 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
3493 /* Return a sensible default order for the pageblock size. */
3494 static inline int pageblock_default_order(void)
3496 if (HPAGE_SHIFT > PAGE_SHIFT)
3497 return HUGETLB_PAGE_ORDER;
3502 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
3503 static inline void __init set_pageblock_order(unsigned int order)
3505 /* Check that pageblock_nr_pages has not already been setup */
3506 if (pageblock_order)
3510 * Assume the largest contiguous order of interest is a huge page.
3511 * This value may be variable depending on boot parameters on IA64
3513 pageblock_order = order;
3515 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3518 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
3519 * and pageblock_default_order() are unused as pageblock_order is set
3520 * at compile-time. See include/linux/pageblock-flags.h for the values of
3521 * pageblock_order based on the kernel config
3523 static inline int pageblock_default_order(unsigned int order)
3527 #define set_pageblock_order(x) do {} while (0)
3529 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3532 * Set up the zone data structures:
3533 * - mark all pages reserved
3534 * - mark all memory queues empty
3535 * - clear the memory bitmaps
3537 static void __paginginit free_area_init_core(struct pglist_data *pgdat,
3538 unsigned long *zones_size, unsigned long *zholes_size)
3541 int nid = pgdat->node_id;
3542 unsigned long zone_start_pfn = pgdat->node_start_pfn;
3545 pgdat_resize_init(pgdat);
3546 pgdat->nr_zones = 0;
3547 init_waitqueue_head(&pgdat->kswapd_wait);
3548 pgdat->kswapd_max_order = 0;
3549 pgdat_page_cgroup_init(pgdat);
3551 for (j = 0; j < MAX_NR_ZONES; j++) {
3552 struct zone *zone = pgdat->node_zones + j;
3553 unsigned long size, realsize, memmap_pages;
3556 size = zone_spanned_pages_in_node(nid, j, zones_size);
3557 realsize = size - zone_absent_pages_in_node(nid, j,
3561 * Adjust realsize so that it accounts for how much memory
3562 * is used by this zone for memmap. This affects the watermark
3563 * and per-cpu initialisations
3566 PAGE_ALIGN(size * sizeof(struct page)) >> PAGE_SHIFT;
3567 if (realsize >= memmap_pages) {
3568 realsize -= memmap_pages;
3571 " %s zone: %lu pages used for memmap\n",
3572 zone_names[j], memmap_pages);
3575 " %s zone: %lu pages exceeds realsize %lu\n",
3576 zone_names[j], memmap_pages, realsize);
3578 /* Account for reserved pages */
3579 if (j == 0 && realsize > dma_reserve) {
3580 realsize -= dma_reserve;
3581 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
3582 zone_names[0], dma_reserve);
3585 if (!is_highmem_idx(j))
3586 nr_kernel_pages += realsize;
3587 nr_all_pages += realsize;
3589 zone->spanned_pages = size;
3590 zone->present_pages = realsize;
3593 zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
3595 zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100;
3597 zone->name = zone_names[j];
3598 spin_lock_init(&zone->lock);
3599 spin_lock_init(&zone->lru_lock);
3600 zone_seqlock_init(zone);
3601 zone->zone_pgdat = pgdat;
3603 zone->prev_priority = DEF_PRIORITY;
3605 zone_pcp_init(zone);
3607 INIT_LIST_HEAD(&zone->lru[l].list);
3608 zone->lru[l].nr_scan = 0;
3610 zone->reclaim_stat.recent_rotated[0] = 0;
3611 zone->reclaim_stat.recent_rotated[1] = 0;
3612 zone->reclaim_stat.recent_scanned[0] = 0;
3613 zone->reclaim_stat.recent_scanned[1] = 0;
3614 zap_zone_vm_stats(zone);
3619 set_pageblock_order(pageblock_default_order());
3620 setup_usemap(pgdat, zone, size);
3621 ret = init_currently_empty_zone(zone, zone_start_pfn,
3622 size, MEMMAP_EARLY);
3624 memmap_init(size, nid, j, zone_start_pfn);
3625 zone_start_pfn += size;
3629 static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
3631 /* Skip empty nodes */
3632 if (!pgdat->node_spanned_pages)
3635 #ifdef CONFIG_FLAT_NODE_MEM_MAP
3636 /* ia64 gets its own node_mem_map, before this, without bootmem */
3637 if (!pgdat->node_mem_map) {
3638 unsigned long size, start, end;
3642 * The zone's endpoints aren't required to be MAX_ORDER
3643 * aligned but the node_mem_map endpoints must be in order
3644 * for the buddy allocator to function correctly.
3646 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
3647 end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
3648 end = ALIGN(end, MAX_ORDER_NR_PAGES);
3649 size = (end - start) * sizeof(struct page);
3650 map = alloc_remap(pgdat->node_id, size);
3652 map = alloc_bootmem_node(pgdat, size);
3653 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
3655 #ifndef CONFIG_NEED_MULTIPLE_NODES
3657 * With no DISCONTIG, the global mem_map is just set as node 0's
3659 if (pgdat == NODE_DATA(0)) {
3660 mem_map = NODE_DATA(0)->node_mem_map;
3661 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3662 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
3663 mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
3664 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3667 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
3670 void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
3671 unsigned long node_start_pfn, unsigned long *zholes_size)
3673 pg_data_t *pgdat = NODE_DATA(nid);
3675 pgdat->node_id = nid;
3676 pgdat->node_start_pfn = node_start_pfn;
3677 calculate_node_totalpages(pgdat, zones_size, zholes_size);
3679 alloc_node_mem_map(pgdat);
3680 #ifdef CONFIG_FLAT_NODE_MEM_MAP
3681 printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
3682 nid, (unsigned long)pgdat,
3683 (unsigned long)pgdat->node_mem_map);
3686 free_area_init_core(pgdat, zones_size, zholes_size);
3689 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3691 #if MAX_NUMNODES > 1
3693 * Figure out the number of possible node ids.
3695 static void __init setup_nr_node_ids(void)
3698 unsigned int highest = 0;
3700 for_each_node_mask(node, node_possible_map)
3702 nr_node_ids = highest + 1;
3705 static inline void setup_nr_node_ids(void)
3711 * add_active_range - Register a range of PFNs backed by physical memory
3712 * @nid: The node ID the range resides on
3713 * @start_pfn: The start PFN of the available physical memory
3714 * @end_pfn: The end PFN of the available physical memory
3716 * These ranges are stored in an early_node_map[] and later used by
3717 * free_area_init_nodes() to calculate zone sizes and holes. If the
3718 * range spans a memory hole, it is up to the architecture to ensure
3719 * the memory is not freed by the bootmem allocator. If possible
3720 * the range being registered will be merged with existing ranges.
3722 void __init add_active_range(unsigned int nid, unsigned long start_pfn,
3723 unsigned long end_pfn)
3727 mminit_dprintk(MMINIT_TRACE, "memory_register",
3728 "Entering add_active_range(%d, %#lx, %#lx) "
3729 "%d entries of %d used\n",
3730 nid, start_pfn, end_pfn,
3731 nr_nodemap_entries, MAX_ACTIVE_REGIONS);
3733 mminit_validate_memmodel_limits(&start_pfn, &end_pfn);
3735 /* Merge with existing active regions if possible */
3736 for (i = 0; i < nr_nodemap_entries; i++) {
3737 if (early_node_map[i].nid != nid)
3740 /* Skip if an existing region covers this new one */
3741 if (start_pfn >= early_node_map[i].start_pfn &&
3742 end_pfn <= early_node_map[i].end_pfn)
3745 /* Merge forward if suitable */
3746 if (start_pfn <= early_node_map[i].end_pfn &&
3747 end_pfn > early_node_map[i].end_pfn) {
3748 early_node_map[i].end_pfn = end_pfn;
3752 /* Merge backward if suitable */
3753 if (start_pfn < early_node_map[i].end_pfn &&
3754 end_pfn >= early_node_map[i].start_pfn) {
3755 early_node_map[i].start_pfn = start_pfn;
3760 /* Check that early_node_map is large enough */
3761 if (i >= MAX_ACTIVE_REGIONS) {
3762 printk(KERN_CRIT "More than %d memory regions, truncating\n",
3763 MAX_ACTIVE_REGIONS);
3767 early_node_map[i].nid = nid;
3768 early_node_map[i].start_pfn = start_pfn;
3769 early_node_map[i].end_pfn = end_pfn;
3770 nr_nodemap_entries = i + 1;
3774 * remove_active_range - Shrink an existing registered range of PFNs
3775 * @nid: The node id the range is on that should be shrunk
3776 * @start_pfn: The new PFN of the range
3777 * @end_pfn: The new PFN of the range
3779 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
3780 * The map is kept near the end physical page range that has already been
3781 * registered. This function allows an arch to shrink an existing registered
3784 void __init remove_active_range(unsigned int nid, unsigned long start_pfn,
3785 unsigned long end_pfn)
3790 printk(KERN_DEBUG "remove_active_range (%d, %lu, %lu)\n",
3791 nid, start_pfn, end_pfn);
3793 /* Find the old active region end and shrink */
3794 for_each_active_range_index_in_nid(i, nid) {
3795 if (early_node_map[i].start_pfn >= start_pfn &&
3796 early_node_map[i].end_pfn <= end_pfn) {
3798 early_node_map[i].start_pfn = 0;
3799 early_node_map[i].end_pfn = 0;
3803 if (early_node_map[i].start_pfn < start_pfn &&
3804 early_node_map[i].end_pfn > start_pfn) {
3805 unsigned long temp_end_pfn = early_node_map[i].end_pfn;
3806 early_node_map[i].end_pfn = start_pfn;
3807 if (temp_end_pfn > end_pfn)
3808 add_active_range(nid, end_pfn, temp_end_pfn);
3811 if (early_node_map[i].start_pfn >= start_pfn &&
3812 early_node_map[i].end_pfn > end_pfn &&
3813 early_node_map[i].start_pfn < end_pfn) {
3814 early_node_map[i].start_pfn = end_pfn;
3822 /* remove the blank ones */
3823 for (i = nr_nodemap_entries - 1; i > 0; i--) {
3824 if (early_node_map[i].nid != nid)
3826 if (early_node_map[i].end_pfn)
3828 /* we found it, get rid of it */
3829 for (j = i; j < nr_nodemap_entries - 1; j++)
3830 memcpy(&early_node_map[j], &early_node_map[j+1],
3831 sizeof(early_node_map[j]));
3832 j = nr_nodemap_entries - 1;
3833 memset(&early_node_map[j], 0, sizeof(early_node_map[j]));
3834 nr_nodemap_entries--;
3839 * remove_all_active_ranges - Remove all currently registered regions
3841 * During discovery, it may be found that a table like SRAT is invalid
3842 * and an alternative discovery method must be used. This function removes
3843 * all currently registered regions.
3845 void __init remove_all_active_ranges(void)
3847 memset(early_node_map, 0, sizeof(early_node_map));
3848 nr_nodemap_entries = 0;
3851 /* Compare two active node_active_regions */
3852 static int __init cmp_node_active_region(const void *a, const void *b)
3854 struct node_active_region *arange = (struct node_active_region *)a;
3855 struct node_active_region *brange = (struct node_active_region *)b;
3857 /* Done this way to avoid overflows */
3858 if (arange->start_pfn > brange->start_pfn)
3860 if (arange->start_pfn < brange->start_pfn)
3866 /* sort the node_map by start_pfn */
3867 static void __init sort_node_map(void)
3869 sort(early_node_map, (size_t)nr_nodemap_entries,
3870 sizeof(struct node_active_region),
3871 cmp_node_active_region, NULL);
3874 /* Find the lowest pfn for a node */
3875 static unsigned long __init find_min_pfn_for_node(int nid)
3878 unsigned long min_pfn = ULONG_MAX;
3880 /* Assuming a sorted map, the first range found has the starting pfn */
3881 for_each_active_range_index_in_nid(i, nid)
3882 min_pfn = min(min_pfn, early_node_map[i].start_pfn);
3884 if (min_pfn == ULONG_MAX) {
3886 "Could not find start_pfn for node %d\n", nid);
3894 * find_min_pfn_with_active_regions - Find the minimum PFN registered
3896 * It returns the minimum PFN based on information provided via
3897 * add_active_range().
3899 unsigned long __init find_min_pfn_with_active_regions(void)
3901 return find_min_pfn_for_node(MAX_NUMNODES);
3905 * early_calculate_totalpages()
3906 * Sum pages in active regions for movable zone.
3907 * Populate N_HIGH_MEMORY for calculating usable_nodes.
3909 static unsigned long __init early_calculate_totalpages(void)
3912 unsigned long totalpages = 0;
3914 for (i = 0; i < nr_nodemap_entries; i++) {
3915 unsigned long pages = early_node_map[i].end_pfn -
3916 early_node_map[i].start_pfn;
3917 totalpages += pages;
3919 node_set_state(early_node_map[i].nid, N_HIGH_MEMORY);
3925 * Find the PFN the Movable zone begins in each node. Kernel memory
3926 * is spread evenly between nodes as long as the nodes have enough
3927 * memory. When they don't, some nodes will have more kernelcore than
3930 static void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn)
3933 unsigned long usable_startpfn;
3934 unsigned long kernelcore_node, kernelcore_remaining;
3935 unsigned long totalpages = early_calculate_totalpages();
3936 int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]);
3939 * If movablecore was specified, calculate what size of
3940 * kernelcore that corresponds so that memory usable for
3941 * any allocation type is evenly spread. If both kernelcore
3942 * and movablecore are specified, then the value of kernelcore
3943 * will be used for required_kernelcore if it's greater than
3944 * what movablecore would have allowed.
3946 if (required_movablecore) {
3947 unsigned long corepages;
3950 * Round-up so that ZONE_MOVABLE is at least as large as what
3951 * was requested by the user
3953 required_movablecore =
3954 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
3955 corepages = totalpages - required_movablecore;
3957 required_kernelcore = max(required_kernelcore, corepages);
3960 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
3961 if (!required_kernelcore)
3964 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
3965 find_usable_zone_for_movable();
3966 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
3969 /* Spread kernelcore memory as evenly as possible throughout nodes */
3970 kernelcore_node = required_kernelcore / usable_nodes;
3971 for_each_node_state(nid, N_HIGH_MEMORY) {
3973 * Recalculate kernelcore_node if the division per node
3974 * now exceeds what is necessary to satisfy the requested
3975 * amount of memory for the kernel
3977 if (required_kernelcore < kernelcore_node)
3978 kernelcore_node = required_kernelcore / usable_nodes;
3981 * As the map is walked, we track how much memory is usable
3982 * by the kernel using kernelcore_remaining. When it is
3983 * 0, the rest of the node is usable by ZONE_MOVABLE
3985 kernelcore_remaining = kernelcore_node;
3987 /* Go through each range of PFNs within this node */
3988 for_each_active_range_index_in_nid(i, nid) {
3989 unsigned long start_pfn, end_pfn;
3990 unsigned long size_pages;
3992 start_pfn = max(early_node_map[i].start_pfn,
3993 zone_movable_pfn[nid]);
3994 end_pfn = early_node_map[i].end_pfn;
3995 if (start_pfn >= end_pfn)
3998 /* Account for what is only usable for kernelcore */
3999 if (start_pfn < usable_startpfn) {
4000 unsigned long kernel_pages;
4001 kernel_pages = min(end_pfn, usable_startpfn)
4004 kernelcore_remaining -= min(kernel_pages,
4005 kernelcore_remaining);
4006 required_kernelcore -= min(kernel_pages,
4007 required_kernelcore);
4009 /* Continue if range is now fully accounted */
4010 if (end_pfn <= usable_startpfn) {
4013 * Push zone_movable_pfn to the end so
4014 * that if we have to rebalance
4015 * kernelcore across nodes, we will
4016 * not double account here
4018 zone_movable_pfn[nid] = end_pfn;
4021 start_pfn = usable_startpfn;
4025 * The usable PFN range for ZONE_MOVABLE is from
4026 * start_pfn->end_pfn. Calculate size_pages as the
4027 * number of pages used as kernelcore
4029 size_pages = end_pfn - start_pfn;
4030 if (size_pages > kernelcore_remaining)
4031 size_pages = kernelcore_remaining;
4032 zone_movable_pfn[nid] = start_pfn + size_pages;
4035 * Some kernelcore has been met, update counts and
4036 * break if the kernelcore for this node has been
4039 required_kernelcore -= min(required_kernelcore,
4041 kernelcore_remaining -= size_pages;
4042 if (!kernelcore_remaining)
4048 * If there is still required_kernelcore, we do another pass with one
4049 * less node in the count. This will push zone_movable_pfn[nid] further
4050 * along on the nodes that still have memory until kernelcore is
4054 if (usable_nodes && required_kernelcore > usable_nodes)
4057 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
4058 for (nid = 0; nid < MAX_NUMNODES; nid++)
4059 zone_movable_pfn[nid] =
4060 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
4063 /* Any regular memory on that node ? */
4064 static void check_for_regular_memory(pg_data_t *pgdat)
4066 #ifdef CONFIG_HIGHMEM
4067 enum zone_type zone_type;
4069 for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) {
4070 struct zone *zone = &pgdat->node_zones[zone_type];
4071 if (zone->present_pages)
4072 node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY);
4078 * free_area_init_nodes - Initialise all pg_data_t and zone data
4079 * @max_zone_pfn: an array of max PFNs for each zone
4081 * This will call free_area_init_node() for each active node in the system.
4082 * Using the page ranges provided by add_active_range(), the size of each
4083 * zone in each node and their holes is calculated. If the maximum PFN
4084 * between two adjacent zones match, it is assumed that the zone is empty.
4085 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
4086 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
4087 * starts where the previous one ended. For example, ZONE_DMA32 starts
4088 * at arch_max_dma_pfn.
4090 void __init free_area_init_nodes(unsigned long *max_zone_pfn)
4095 /* Sort early_node_map as initialisation assumes it is sorted */
4098 /* Record where the zone boundaries are */
4099 memset(arch_zone_lowest_possible_pfn, 0,
4100 sizeof(arch_zone_lowest_possible_pfn));
4101 memset(arch_zone_highest_possible_pfn, 0,
4102 sizeof(arch_zone_highest_possible_pfn));
4103 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
4104 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
4105 for (i = 1; i < MAX_NR_ZONES; i++) {
4106 if (i == ZONE_MOVABLE)
4108 arch_zone_lowest_possible_pfn[i] =
4109 arch_zone_highest_possible_pfn[i-1];
4110 arch_zone_highest_possible_pfn[i] =
4111 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
4113 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
4114 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
4116 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
4117 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
4118 find_zone_movable_pfns_for_nodes(zone_movable_pfn);
4120 /* Print out the zone ranges */
4121 printk("Zone PFN ranges:\n");
4122 for (i = 0; i < MAX_NR_ZONES; i++) {
4123 if (i == ZONE_MOVABLE)
4125 printk(" %-8s %0#10lx -> %0#10lx\n",
4127 arch_zone_lowest_possible_pfn[i],
4128 arch_zone_highest_possible_pfn[i]);
4131 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
4132 printk("Movable zone start PFN for each node\n");
4133 for (i = 0; i < MAX_NUMNODES; i++) {
4134 if (zone_movable_pfn[i])
4135 printk(" Node %d: %lu\n", i, zone_movable_pfn[i]);
4138 /* Print out the early_node_map[] */
4139 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries);
4140 for (i = 0; i < nr_nodemap_entries; i++)
4141 printk(" %3d: %0#10lx -> %0#10lx\n", early_node_map[i].nid,
4142 early_node_map[i].start_pfn,
4143 early_node_map[i].end_pfn);
4145 /* Initialise every node */
4146 mminit_verify_pageflags_layout();
4147 setup_nr_node_ids();
4148 for_each_online_node(nid) {
4149 pg_data_t *pgdat = NODE_DATA(nid);
4150 free_area_init_node(nid, NULL,
4151 find_min_pfn_for_node(nid), NULL);
4153 /* Any memory on that node */
4154 if (pgdat->node_present_pages)
4155 node_set_state(nid, N_HIGH_MEMORY);
4156 check_for_regular_memory(pgdat);
4160 static int __init cmdline_parse_core(char *p, unsigned long *core)
4162 unsigned long long coremem;
4166 coremem = memparse(p, &p);
4167 *core = coremem >> PAGE_SHIFT;
4169 /* Paranoid check that UL is enough for the coremem value */
4170 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
4176 * kernelcore=size sets the amount of memory for use for allocations that
4177 * cannot be reclaimed or migrated.
4179 static int __init cmdline_parse_kernelcore(char *p)
4181 return cmdline_parse_core(p, &required_kernelcore);
4185 * movablecore=size sets the amount of memory for use for allocations that
4186 * can be reclaimed or migrated.
4188 static int __init cmdline_parse_movablecore(char *p)
4190 return cmdline_parse_core(p, &required_movablecore);
4193 early_param("kernelcore", cmdline_parse_kernelcore);
4194 early_param("movablecore", cmdline_parse_movablecore);
4196 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4199 * set_dma_reserve - set the specified number of pages reserved in the first zone
4200 * @new_dma_reserve: The number of pages to mark reserved
4202 * The per-cpu batchsize and zone watermarks are determined by present_pages.
4203 * In the DMA zone, a significant percentage may be consumed by kernel image
4204 * and other unfreeable allocations which can skew the watermarks badly. This
4205 * function may optionally be used to account for unfreeable pages in the
4206 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
4207 * smaller per-cpu batchsize.
4209 void __init set_dma_reserve(unsigned long new_dma_reserve)
4211 dma_reserve = new_dma_reserve;
4214 #ifndef CONFIG_NEED_MULTIPLE_NODES
4215 struct pglist_data __refdata contig_page_data = { .bdata = &bootmem_node_data[0] };
4216 EXPORT_SYMBOL(contig_page_data);
4219 void __init free_area_init(unsigned long *zones_size)
4221 free_area_init_node(0, zones_size,
4222 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
4225 static int page_alloc_cpu_notify(struct notifier_block *self,
4226 unsigned long action, void *hcpu)
4228 int cpu = (unsigned long)hcpu;
4230 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
4234 * Spill the event counters of the dead processor
4235 * into the current processors event counters.
4236 * This artificially elevates the count of the current
4239 vm_events_fold_cpu(cpu);
4242 * Zero the differential counters of the dead processor
4243 * so that the vm statistics are consistent.
4245 * This is only okay since the processor is dead and cannot
4246 * race with what we are doing.
4248 refresh_cpu_vm_stats(cpu);
4253 void __init page_alloc_init(void)
4255 hotcpu_notifier(page_alloc_cpu_notify, 0);
4259 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
4260 * or min_free_kbytes changes.
4262 static void calculate_totalreserve_pages(void)
4264 struct pglist_data *pgdat;
4265 unsigned long reserve_pages = 0;
4266 enum zone_type i, j;
4268 for_each_online_pgdat(pgdat) {
4269 for (i = 0; i < MAX_NR_ZONES; i++) {
4270 struct zone *zone = pgdat->node_zones + i;
4271 unsigned long max = 0;
4273 /* Find valid and maximum lowmem_reserve in the zone */
4274 for (j = i; j < MAX_NR_ZONES; j++) {
4275 if (zone->lowmem_reserve[j] > max)
4276 max = zone->lowmem_reserve[j];
4279 /* we treat pages_high as reserved pages. */
4280 max += zone->pages_high;
4282 if (max > zone->present_pages)
4283 max = zone->present_pages;
4284 reserve_pages += max;
4287 totalreserve_pages = reserve_pages;
4291 * setup_per_zone_lowmem_reserve - called whenever
4292 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
4293 * has a correct pages reserved value, so an adequate number of
4294 * pages are left in the zone after a successful __alloc_pages().
4296 static void setup_per_zone_lowmem_reserve(void)
4298 struct pglist_data *pgdat;
4299 enum zone_type j, idx;
4301 for_each_online_pgdat(pgdat) {
4302 for (j = 0; j < MAX_NR_ZONES; j++) {
4303 struct zone *zone = pgdat->node_zones + j;
4304 unsigned long present_pages = zone->present_pages;
4306 zone->lowmem_reserve[j] = 0;
4310 struct zone *lower_zone;
4314 if (sysctl_lowmem_reserve_ratio[idx] < 1)
4315 sysctl_lowmem_reserve_ratio[idx] = 1;
4317 lower_zone = pgdat->node_zones + idx;
4318 lower_zone->lowmem_reserve[j] = present_pages /
4319 sysctl_lowmem_reserve_ratio[idx];
4320 present_pages += lower_zone->present_pages;
4325 /* update totalreserve_pages */
4326 calculate_totalreserve_pages();
4330 * setup_per_zone_pages_min - called when min_free_kbytes changes.
4332 * Ensures that the pages_{min,low,high} values for each zone are set correctly
4333 * with respect to min_free_kbytes.
4335 void setup_per_zone_pages_min(void)
4337 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
4338 unsigned long lowmem_pages = 0;
4340 unsigned long flags;
4342 /* Calculate total number of !ZONE_HIGHMEM pages */
4343 for_each_zone(zone) {
4344 if (!is_highmem(zone))
4345 lowmem_pages += zone->present_pages;
4348 for_each_zone(zone) {
4351 spin_lock_irqsave(&zone->lock, flags);
4352 tmp = (u64)pages_min * zone->present_pages;
4353 do_div(tmp, lowmem_pages);
4354 if (is_highmem(zone)) {
4356 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
4357 * need highmem pages, so cap pages_min to a small
4360 * The (pages_high-pages_low) and (pages_low-pages_min)
4361 * deltas controls asynch page reclaim, and so should
4362 * not be capped for highmem.
4366 min_pages = zone->present_pages / 1024;
4367 if (min_pages < SWAP_CLUSTER_MAX)
4368 min_pages = SWAP_CLUSTER_MAX;
4369 if (min_pages > 128)
4371 zone->pages_min = min_pages;
4374 * If it's a lowmem zone, reserve a number of pages
4375 * proportionate to the zone's size.
4377 zone->pages_min = tmp;
4380 zone->pages_low = zone->pages_min + (tmp >> 2);
4381 zone->pages_high = zone->pages_min + (tmp >> 1);
4382 setup_zone_migrate_reserve(zone);
4383 spin_unlock_irqrestore(&zone->lock, flags);
4386 /* update totalreserve_pages */
4387 calculate_totalreserve_pages();
4391 * setup_per_zone_inactive_ratio - called when min_free_kbytes changes.
4393 * The inactive anon list should be small enough that the VM never has to
4394 * do too much work, but large enough that each inactive page has a chance
4395 * to be referenced again before it is swapped out.
4397 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
4398 * INACTIVE_ANON pages on this zone's LRU, maintained by the
4399 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
4400 * the anonymous pages are kept on the inactive list.
4403 * memory ratio inactive anon
4404 * -------------------------------------
4413 static void setup_per_zone_inactive_ratio(void)
4417 for_each_zone(zone) {
4418 unsigned int gb, ratio;
4420 /* Zone size in gigabytes */
4421 gb = zone->present_pages >> (30 - PAGE_SHIFT);
4422 ratio = int_sqrt(10 * gb);
4426 zone->inactive_ratio = ratio;
4431 * Initialise min_free_kbytes.
4433 * For small machines we want it small (128k min). For large machines
4434 * we want it large (64MB max). But it is not linear, because network
4435 * bandwidth does not increase linearly with machine size. We use
4437 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
4438 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
4454 static int __init init_per_zone_pages_min(void)
4456 unsigned long lowmem_kbytes;
4458 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
4460 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
4461 if (min_free_kbytes < 128)
4462 min_free_kbytes = 128;
4463 if (min_free_kbytes > 65536)
4464 min_free_kbytes = 65536;
4465 setup_per_zone_pages_min();
4466 setup_per_zone_lowmem_reserve();
4467 setup_per_zone_inactive_ratio();
4470 module_init(init_per_zone_pages_min)
4473 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
4474 * that we can call two helper functions whenever min_free_kbytes
4477 int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
4478 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4480 proc_dointvec(table, write, file, buffer, length, ppos);
4482 setup_per_zone_pages_min();
4487 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
4488 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4493 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4498 zone->min_unmapped_pages = (zone->present_pages *
4499 sysctl_min_unmapped_ratio) / 100;
4503 int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
4504 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4509 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4514 zone->min_slab_pages = (zone->present_pages *
4515 sysctl_min_slab_ratio) / 100;
4521 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
4522 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
4523 * whenever sysctl_lowmem_reserve_ratio changes.
4525 * The reserve ratio obviously has absolutely no relation with the
4526 * pages_min watermarks. The lowmem reserve ratio can only make sense
4527 * if in function of the boot time zone sizes.
4529 int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
4530 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4532 proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4533 setup_per_zone_lowmem_reserve();
4538 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
4539 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
4540 * can have before it gets flushed back to buddy allocator.
4543 int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
4544 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4550 ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4551 if (!write || (ret == -EINVAL))
4553 for_each_zone(zone) {
4554 for_each_online_cpu(cpu) {
4556 high = zone->present_pages / percpu_pagelist_fraction;
4557 setup_pagelist_highmark(zone_pcp(zone, cpu), high);
4563 int hashdist = HASHDIST_DEFAULT;
4566 static int __init set_hashdist(char *str)
4570 hashdist = simple_strtoul(str, &str, 0);
4573 __setup("hashdist=", set_hashdist);
4577 * allocate a large system hash table from bootmem
4578 * - it is assumed that the hash table must contain an exact power-of-2
4579 * quantity of entries
4580 * - limit is the number of hash buckets, not the total allocation size
4582 void *__init alloc_large_system_hash(const char *tablename,
4583 unsigned long bucketsize,
4584 unsigned long numentries,
4587 unsigned int *_hash_shift,
4588 unsigned int *_hash_mask,
4589 unsigned long limit)
4591 unsigned long long max = limit;
4592 unsigned long log2qty, size;
4595 /* allow the kernel cmdline to have a say */
4597 /* round applicable memory size up to nearest megabyte */
4598 numentries = nr_kernel_pages;
4599 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
4600 numentries >>= 20 - PAGE_SHIFT;
4601 numentries <<= 20 - PAGE_SHIFT;
4603 /* limit to 1 bucket per 2^scale bytes of low memory */
4604 if (scale > PAGE_SHIFT)
4605 numentries >>= (scale - PAGE_SHIFT);
4607 numentries <<= (PAGE_SHIFT - scale);
4609 /* Make sure we've got at least a 0-order allocation.. */
4610 if (unlikely((numentries * bucketsize) < PAGE_SIZE))
4611 numentries = PAGE_SIZE / bucketsize;
4613 numentries = roundup_pow_of_two(numentries);
4615 /* limit allocation size to 1/16 total memory by default */
4617 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
4618 do_div(max, bucketsize);
4621 if (numentries > max)
4624 log2qty = ilog2(numentries);
4627 size = bucketsize << log2qty;
4628 if (flags & HASH_EARLY)
4629 table = alloc_bootmem_nopanic(size);
4631 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
4633 unsigned long order = get_order(size);
4635 if (order < MAX_ORDER)
4636 table = (void *)__get_free_pages(GFP_ATOMIC,
4639 * If bucketsize is not a power-of-two, we may free
4640 * some pages at the end of hash table.
4643 unsigned long alloc_end = (unsigned long)table +
4644 (PAGE_SIZE << order);
4645 unsigned long used = (unsigned long)table +
4647 split_page(virt_to_page(table), order);
4648 while (used < alloc_end) {
4654 } while (!table && size > PAGE_SIZE && --log2qty);
4657 panic("Failed to allocate %s hash table\n", tablename);
4659 printk(KERN_INFO "%s hash table entries: %d (order: %d, %lu bytes)\n",
4662 ilog2(size) - PAGE_SHIFT,
4666 *_hash_shift = log2qty;
4668 *_hash_mask = (1 << log2qty) - 1;
4671 * If hashdist is set, the table allocation is done with __vmalloc()
4672 * which invokes the kmemleak_alloc() callback. This function may also
4673 * be called before the slab and kmemleak are initialised when
4674 * kmemleak simply buffers the request to be executed later
4675 * (GFP_ATOMIC flag ignored in this case).
4678 kmemleak_alloc(table, size, 1, GFP_ATOMIC);
4683 /* Return a pointer to the bitmap storing bits affecting a block of pages */
4684 static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
4687 #ifdef CONFIG_SPARSEMEM
4688 return __pfn_to_section(pfn)->pageblock_flags;
4690 return zone->pageblock_flags;
4691 #endif /* CONFIG_SPARSEMEM */
4694 static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
4696 #ifdef CONFIG_SPARSEMEM
4697 pfn &= (PAGES_PER_SECTION-1);
4698 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
4700 pfn = pfn - zone->zone_start_pfn;
4701 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
4702 #endif /* CONFIG_SPARSEMEM */
4706 * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
4707 * @page: The page within the block of interest
4708 * @start_bitidx: The first bit of interest to retrieve
4709 * @end_bitidx: The last bit of interest
4710 * returns pageblock_bits flags
4712 unsigned long get_pageblock_flags_group(struct page *page,
4713 int start_bitidx, int end_bitidx)
4716 unsigned long *bitmap;
4717 unsigned long pfn, bitidx;
4718 unsigned long flags = 0;
4719 unsigned long value = 1;
4721 zone = page_zone(page);
4722 pfn = page_to_pfn(page);
4723 bitmap = get_pageblock_bitmap(zone, pfn);
4724 bitidx = pfn_to_bitidx(zone, pfn);
4726 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
4727 if (test_bit(bitidx + start_bitidx, bitmap))
4734 * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
4735 * @page: The page within the block of interest
4736 * @start_bitidx: The first bit of interest
4737 * @end_bitidx: The last bit of interest
4738 * @flags: The flags to set
4740 void set_pageblock_flags_group(struct page *page, unsigned long flags,
4741 int start_bitidx, int end_bitidx)
4744 unsigned long *bitmap;
4745 unsigned long pfn, bitidx;
4746 unsigned long value = 1;
4748 zone = page_zone(page);
4749 pfn = page_to_pfn(page);
4750 bitmap = get_pageblock_bitmap(zone, pfn);
4751 bitidx = pfn_to_bitidx(zone, pfn);
4752 VM_BUG_ON(pfn < zone->zone_start_pfn);
4753 VM_BUG_ON(pfn >= zone->zone_start_pfn + zone->spanned_pages);
4755 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
4757 __set_bit(bitidx + start_bitidx, bitmap);
4759 __clear_bit(bitidx + start_bitidx, bitmap);
4763 * This is designed as sub function...plz see page_isolation.c also.
4764 * set/clear page block's type to be ISOLATE.
4765 * page allocater never alloc memory from ISOLATE block.
4768 int set_migratetype_isolate(struct page *page)
4771 unsigned long flags;
4774 zone = page_zone(page);
4775 spin_lock_irqsave(&zone->lock, flags);
4777 * In future, more migrate types will be able to be isolation target.
4779 if (get_pageblock_migratetype(page) != MIGRATE_MOVABLE)
4781 set_pageblock_migratetype(page, MIGRATE_ISOLATE);
4782 move_freepages_block(zone, page, MIGRATE_ISOLATE);
4785 spin_unlock_irqrestore(&zone->lock, flags);
4791 void unset_migratetype_isolate(struct page *page)
4794 unsigned long flags;
4795 zone = page_zone(page);
4796 spin_lock_irqsave(&zone->lock, flags);
4797 if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE)
4799 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
4800 move_freepages_block(zone, page, MIGRATE_MOVABLE);
4802 spin_unlock_irqrestore(&zone->lock, flags);
4805 #ifdef CONFIG_MEMORY_HOTREMOVE
4807 * All pages in the range must be isolated before calling this.
4810 __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
4816 unsigned long flags;
4817 /* find the first valid pfn */
4818 for (pfn = start_pfn; pfn < end_pfn; pfn++)
4823 zone = page_zone(pfn_to_page(pfn));
4824 spin_lock_irqsave(&zone->lock, flags);
4826 while (pfn < end_pfn) {
4827 if (!pfn_valid(pfn)) {
4831 page = pfn_to_page(pfn);
4832 BUG_ON(page_count(page));
4833 BUG_ON(!PageBuddy(page));
4834 order = page_order(page);
4835 #ifdef CONFIG_DEBUG_VM
4836 printk(KERN_INFO "remove from free list %lx %d %lx\n",
4837 pfn, 1 << order, end_pfn);
4839 list_del(&page->lru);
4840 rmv_page_order(page);
4841 zone->free_area[order].nr_free--;
4842 __mod_zone_page_state(zone, NR_FREE_PAGES,
4844 for (i = 0; i < (1 << order); i++)
4845 SetPageReserved((page+i));
4846 pfn += (1 << order);
4848 spin_unlock_irqrestore(&zone->lock, flags);