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
665 struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
668 unsigned int current_order;
669 struct free_area * area;
672 /* Find a page of the appropriate size in the preferred list */
673 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
674 area = &(zone->free_area[current_order]);
675 if (list_empty(&area->free_list[migratetype]))
678 page = list_entry(area->free_list[migratetype].next,
680 list_del(&page->lru);
681 rmv_page_order(page);
683 __mod_zone_page_state(zone, NR_FREE_PAGES, - (1UL << order));
684 expand(zone, page, order, current_order, area, migratetype);
693 * This array describes the order lists are fallen back to when
694 * the free lists for the desirable migrate type are depleted
696 static int fallbacks[MIGRATE_TYPES][MIGRATE_TYPES-1] = {
697 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
698 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
699 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
700 [MIGRATE_RESERVE] = { MIGRATE_RESERVE, MIGRATE_RESERVE, MIGRATE_RESERVE }, /* Never used */
704 * Move the free pages in a range to the free lists of the requested type.
705 * Note that start_page and end_pages are not aligned on a pageblock
706 * boundary. If alignment is required, use move_freepages_block()
708 static int move_freepages(struct zone *zone,
709 struct page *start_page, struct page *end_page,
716 #ifndef CONFIG_HOLES_IN_ZONE
718 * page_zone is not safe to call in this context when
719 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
720 * anyway as we check zone boundaries in move_freepages_block().
721 * Remove at a later date when no bug reports exist related to
722 * grouping pages by mobility
724 BUG_ON(page_zone(start_page) != page_zone(end_page));
727 for (page = start_page; page <= end_page;) {
728 /* Make sure we are not inadvertently changing nodes */
729 VM_BUG_ON(page_to_nid(page) != zone_to_nid(zone));
731 if (!pfn_valid_within(page_to_pfn(page))) {
736 if (!PageBuddy(page)) {
741 order = page_order(page);
742 list_del(&page->lru);
744 &zone->free_area[order].free_list[migratetype]);
746 pages_moved += 1 << order;
752 static int move_freepages_block(struct zone *zone, struct page *page,
755 unsigned long start_pfn, end_pfn;
756 struct page *start_page, *end_page;
758 start_pfn = page_to_pfn(page);
759 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
760 start_page = pfn_to_page(start_pfn);
761 end_page = start_page + pageblock_nr_pages - 1;
762 end_pfn = start_pfn + pageblock_nr_pages - 1;
764 /* Do not cross zone boundaries */
765 if (start_pfn < zone->zone_start_pfn)
767 if (end_pfn >= zone->zone_start_pfn + zone->spanned_pages)
770 return move_freepages(zone, start_page, end_page, migratetype);
773 /* Remove an element from the buddy allocator from the fallback list */
774 static struct page *__rmqueue_fallback(struct zone *zone, int order,
775 int start_migratetype)
777 struct free_area * area;
782 /* Find the largest possible block of pages in the other list */
783 for (current_order = MAX_ORDER-1; current_order >= order;
785 for (i = 0; i < MIGRATE_TYPES - 1; i++) {
786 migratetype = fallbacks[start_migratetype][i];
788 /* MIGRATE_RESERVE handled later if necessary */
789 if (migratetype == MIGRATE_RESERVE)
792 area = &(zone->free_area[current_order]);
793 if (list_empty(&area->free_list[migratetype]))
796 page = list_entry(area->free_list[migratetype].next,
801 * If breaking a large block of pages, move all free
802 * pages to the preferred allocation list. If falling
803 * back for a reclaimable kernel allocation, be more
804 * agressive about taking ownership of free pages
806 if (unlikely(current_order >= (pageblock_order >> 1)) ||
807 start_migratetype == MIGRATE_RECLAIMABLE) {
809 pages = move_freepages_block(zone, page,
812 /* Claim the whole block if over half of it is free */
813 if (pages >= (1 << (pageblock_order-1)))
814 set_pageblock_migratetype(page,
817 migratetype = start_migratetype;
820 /* Remove the page from the freelists */
821 list_del(&page->lru);
822 rmv_page_order(page);
823 __mod_zone_page_state(zone, NR_FREE_PAGES,
826 if (current_order == pageblock_order)
827 set_pageblock_migratetype(page,
830 expand(zone, page, order, current_order, area, migratetype);
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,
848 page = __rmqueue_smallest(zone, order, migratetype);
850 if (unlikely(!page) && migratetype != MIGRATE_RESERVE) {
851 page = __rmqueue_fallback(zone, order, migratetype);
854 * Use MIGRATE_RESERVE rather than fail an allocation. goto
855 * is used because __rmqueue_smallest is an inline function
856 * and we want just one call site
859 migratetype = MIGRATE_RESERVE;
868 * Obtain a specified number of elements from the buddy allocator, all under
869 * a single hold of the lock, for efficiency. Add them to the supplied list.
870 * Returns the number of new pages which were placed at *list.
872 static int rmqueue_bulk(struct zone *zone, unsigned int order,
873 unsigned long count, struct list_head *list,
878 spin_lock(&zone->lock);
879 for (i = 0; i < count; ++i) {
880 struct page *page = __rmqueue(zone, order, migratetype);
881 if (unlikely(page == NULL))
885 * Split buddy pages returned by expand() are received here
886 * in physical page order. The page is added to the callers and
887 * list and the list head then moves forward. From the callers
888 * perspective, the linked list is ordered by page number in
889 * some conditions. This is useful for IO devices that can
890 * merge IO requests if the physical pages are ordered
893 list_add(&page->lru, list);
894 set_page_private(page, migratetype);
897 spin_unlock(&zone->lock);
903 * Called from the vmstat counter updater to drain pagesets of this
904 * currently executing processor on remote nodes after they have
907 * Note that this function must be called with the thread pinned to
908 * a single processor.
910 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
915 local_irq_save(flags);
916 if (pcp->count >= pcp->batch)
917 to_drain = pcp->batch;
919 to_drain = pcp->count;
920 free_pages_bulk(zone, to_drain, &pcp->list, 0);
921 pcp->count -= to_drain;
922 local_irq_restore(flags);
927 * Drain pages of the indicated processor.
929 * The processor must either be the current processor and the
930 * thread pinned to the current processor or a processor that
933 static void drain_pages(unsigned int cpu)
938 for_each_populated_zone(zone) {
939 struct per_cpu_pageset *pset;
940 struct per_cpu_pages *pcp;
942 pset = zone_pcp(zone, cpu);
945 local_irq_save(flags);
946 free_pages_bulk(zone, pcp->count, &pcp->list, 0);
948 local_irq_restore(flags);
953 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
955 void drain_local_pages(void *arg)
957 drain_pages(smp_processor_id());
961 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
963 void drain_all_pages(void)
965 on_each_cpu(drain_local_pages, NULL, 1);
968 #ifdef CONFIG_HIBERNATION
970 void mark_free_pages(struct zone *zone)
972 unsigned long pfn, max_zone_pfn;
975 struct list_head *curr;
977 if (!zone->spanned_pages)
980 spin_lock_irqsave(&zone->lock, flags);
982 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
983 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
984 if (pfn_valid(pfn)) {
985 struct page *page = pfn_to_page(pfn);
987 if (!swsusp_page_is_forbidden(page))
988 swsusp_unset_page_free(page);
991 for_each_migratetype_order(order, t) {
992 list_for_each(curr, &zone->free_area[order].free_list[t]) {
995 pfn = page_to_pfn(list_entry(curr, struct page, lru));
996 for (i = 0; i < (1UL << order); i++)
997 swsusp_set_page_free(pfn_to_page(pfn + i));
1000 spin_unlock_irqrestore(&zone->lock, flags);
1002 #endif /* CONFIG_PM */
1005 * Free a 0-order page
1007 static void free_hot_cold_page(struct page *page, int cold)
1009 struct zone *zone = page_zone(page);
1010 struct per_cpu_pages *pcp;
1011 unsigned long flags;
1014 page->mapping = NULL;
1015 if (free_pages_check(page))
1018 if (!PageHighMem(page)) {
1019 debug_check_no_locks_freed(page_address(page), PAGE_SIZE);
1020 debug_check_no_obj_freed(page_address(page), PAGE_SIZE);
1022 arch_free_page(page, 0);
1023 kernel_map_pages(page, 1, 0);
1025 pcp = &zone_pcp(zone, get_cpu())->pcp;
1026 local_irq_save(flags);
1027 __count_vm_event(PGFREE);
1029 list_add_tail(&page->lru, &pcp->list);
1031 list_add(&page->lru, &pcp->list);
1032 set_page_private(page, get_pageblock_migratetype(page));
1034 if (pcp->count >= pcp->high) {
1035 free_pages_bulk(zone, pcp->batch, &pcp->list, 0);
1036 pcp->count -= pcp->batch;
1038 local_irq_restore(flags);
1042 void free_hot_page(struct page *page)
1044 free_hot_cold_page(page, 0);
1047 void free_cold_page(struct page *page)
1049 free_hot_cold_page(page, 1);
1053 * split_page takes a non-compound higher-order page, and splits it into
1054 * n (1<<order) sub-pages: page[0..n]
1055 * Each sub-page must be freed individually.
1057 * Note: this is probably too low level an operation for use in drivers.
1058 * Please consult with lkml before using this in your driver.
1060 void split_page(struct page *page, unsigned int order)
1064 VM_BUG_ON(PageCompound(page));
1065 VM_BUG_ON(!page_count(page));
1066 for (i = 1; i < (1 << order); i++)
1067 set_page_refcounted(page + i);
1071 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1072 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1075 static struct page *buffered_rmqueue(struct zone *preferred_zone,
1076 struct zone *zone, int order, gfp_t gfp_flags,
1079 unsigned long flags;
1081 int cold = !!(gfp_flags & __GFP_COLD);
1086 if (likely(order == 0)) {
1087 struct per_cpu_pages *pcp;
1089 pcp = &zone_pcp(zone, cpu)->pcp;
1090 local_irq_save(flags);
1092 pcp->count = rmqueue_bulk(zone, 0,
1093 pcp->batch, &pcp->list, migratetype);
1094 if (unlikely(!pcp->count))
1098 /* Find a page of the appropriate migrate type */
1100 list_for_each_entry_reverse(page, &pcp->list, lru)
1101 if (page_private(page) == migratetype)
1104 list_for_each_entry(page, &pcp->list, lru)
1105 if (page_private(page) == migratetype)
1109 /* Allocate more to the pcp list if necessary */
1110 if (unlikely(&page->lru == &pcp->list)) {
1111 pcp->count += rmqueue_bulk(zone, 0,
1112 pcp->batch, &pcp->list, migratetype);
1113 page = list_entry(pcp->list.next, struct page, lru);
1116 list_del(&page->lru);
1119 spin_lock_irqsave(&zone->lock, flags);
1120 page = __rmqueue(zone, order, migratetype);
1121 spin_unlock(&zone->lock);
1126 __count_zone_vm_events(PGALLOC, zone, 1 << order);
1127 zone_statistics(preferred_zone, zone);
1128 local_irq_restore(flags);
1131 VM_BUG_ON(bad_range(zone, page));
1132 if (prep_new_page(page, order, gfp_flags))
1137 local_irq_restore(flags);
1142 #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
1143 #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
1144 #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
1145 #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
1146 #define ALLOC_HARDER 0x10 /* try to alloc harder */
1147 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1148 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
1150 #ifdef CONFIG_FAIL_PAGE_ALLOC
1152 static struct fail_page_alloc_attr {
1153 struct fault_attr attr;
1155 u32 ignore_gfp_highmem;
1156 u32 ignore_gfp_wait;
1159 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1161 struct dentry *ignore_gfp_highmem_file;
1162 struct dentry *ignore_gfp_wait_file;
1163 struct dentry *min_order_file;
1165 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1167 } fail_page_alloc = {
1168 .attr = FAULT_ATTR_INITIALIZER,
1169 .ignore_gfp_wait = 1,
1170 .ignore_gfp_highmem = 1,
1174 static int __init setup_fail_page_alloc(char *str)
1176 return setup_fault_attr(&fail_page_alloc.attr, str);
1178 __setup("fail_page_alloc=", setup_fail_page_alloc);
1180 static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1182 if (order < fail_page_alloc.min_order)
1184 if (gfp_mask & __GFP_NOFAIL)
1186 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
1188 if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
1191 return should_fail(&fail_page_alloc.attr, 1 << order);
1194 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1196 static int __init fail_page_alloc_debugfs(void)
1198 mode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
1202 err = init_fault_attr_dentries(&fail_page_alloc.attr,
1206 dir = fail_page_alloc.attr.dentries.dir;
1208 fail_page_alloc.ignore_gfp_wait_file =
1209 debugfs_create_bool("ignore-gfp-wait", mode, dir,
1210 &fail_page_alloc.ignore_gfp_wait);
1212 fail_page_alloc.ignore_gfp_highmem_file =
1213 debugfs_create_bool("ignore-gfp-highmem", mode, dir,
1214 &fail_page_alloc.ignore_gfp_highmem);
1215 fail_page_alloc.min_order_file =
1216 debugfs_create_u32("min-order", mode, dir,
1217 &fail_page_alloc.min_order);
1219 if (!fail_page_alloc.ignore_gfp_wait_file ||
1220 !fail_page_alloc.ignore_gfp_highmem_file ||
1221 !fail_page_alloc.min_order_file) {
1223 debugfs_remove(fail_page_alloc.ignore_gfp_wait_file);
1224 debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file);
1225 debugfs_remove(fail_page_alloc.min_order_file);
1226 cleanup_fault_attr_dentries(&fail_page_alloc.attr);
1232 late_initcall(fail_page_alloc_debugfs);
1234 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1236 #else /* CONFIG_FAIL_PAGE_ALLOC */
1238 static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1243 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1246 * Return 1 if free pages are above 'mark'. This takes into account the order
1247 * of the allocation.
1249 int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1250 int classzone_idx, int alloc_flags)
1252 /* free_pages my go negative - that's OK */
1254 long free_pages = zone_page_state(z, NR_FREE_PAGES) - (1 << order) + 1;
1257 if (alloc_flags & ALLOC_HIGH)
1259 if (alloc_flags & ALLOC_HARDER)
1262 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
1264 for (o = 0; o < order; o++) {
1265 /* At the next order, this order's pages become unavailable */
1266 free_pages -= z->free_area[o].nr_free << o;
1268 /* Require fewer higher order pages to be free */
1271 if (free_pages <= min)
1279 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1280 * skip over zones that are not allowed by the cpuset, or that have
1281 * been recently (in last second) found to be nearly full. See further
1282 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1283 * that have to skip over a lot of full or unallowed zones.
1285 * If the zonelist cache is present in the passed in zonelist, then
1286 * returns a pointer to the allowed node mask (either the current
1287 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
1289 * If the zonelist cache is not available for this zonelist, does
1290 * nothing and returns NULL.
1292 * If the fullzones BITMAP in the zonelist cache is stale (more than
1293 * a second since last zap'd) then we zap it out (clear its bits.)
1295 * We hold off even calling zlc_setup, until after we've checked the
1296 * first zone in the zonelist, on the theory that most allocations will
1297 * be satisfied from that first zone, so best to examine that zone as
1298 * quickly as we can.
1300 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1302 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1303 nodemask_t *allowednodes; /* zonelist_cache approximation */
1305 zlc = zonelist->zlcache_ptr;
1309 if (time_after(jiffies, zlc->last_full_zap + HZ)) {
1310 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1311 zlc->last_full_zap = jiffies;
1314 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1315 &cpuset_current_mems_allowed :
1316 &node_states[N_HIGH_MEMORY];
1317 return allowednodes;
1321 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1322 * if it is worth looking at further for free memory:
1323 * 1) Check that the zone isn't thought to be full (doesn't have its
1324 * bit set in the zonelist_cache fullzones BITMAP).
1325 * 2) Check that the zones node (obtained from the zonelist_cache
1326 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1327 * Return true (non-zero) if zone is worth looking at further, or
1328 * else return false (zero) if it is not.
1330 * This check -ignores- the distinction between various watermarks,
1331 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1332 * found to be full for any variation of these watermarks, it will
1333 * be considered full for up to one second by all requests, unless
1334 * we are so low on memory on all allowed nodes that we are forced
1335 * into the second scan of the zonelist.
1337 * In the second scan we ignore this zonelist cache and exactly
1338 * apply the watermarks to all zones, even it is slower to do so.
1339 * We are low on memory in the second scan, and should leave no stone
1340 * unturned looking for a free page.
1342 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
1343 nodemask_t *allowednodes)
1345 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1346 int i; /* index of *z in zonelist zones */
1347 int n; /* node that zone *z is on */
1349 zlc = zonelist->zlcache_ptr;
1353 i = z - zonelist->_zonerefs;
1356 /* This zone is worth trying if it is allowed but not full */
1357 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1361 * Given 'z' scanning a zonelist, set the corresponding bit in
1362 * zlc->fullzones, so that subsequent attempts to allocate a page
1363 * from that zone don't waste time re-examining it.
1365 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
1367 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1368 int i; /* index of *z in zonelist zones */
1370 zlc = zonelist->zlcache_ptr;
1374 i = z - zonelist->_zonerefs;
1376 set_bit(i, zlc->fullzones);
1379 #else /* CONFIG_NUMA */
1381 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1386 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
1387 nodemask_t *allowednodes)
1392 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
1395 #endif /* CONFIG_NUMA */
1398 * get_page_from_freelist goes through the zonelist trying to allocate
1401 static struct page *
1402 get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order,
1403 struct zonelist *zonelist, int high_zoneidx, int alloc_flags,
1404 struct zone *preferred_zone, int migratetype)
1407 struct page *page = NULL;
1410 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1411 int zlc_active = 0; /* set if using zonelist_cache */
1412 int did_zlc_setup = 0; /* just call zlc_setup() one time */
1414 if (WARN_ON_ONCE(order >= MAX_ORDER))
1417 classzone_idx = zone_idx(preferred_zone);
1420 * Scan zonelist, looking for a zone with enough free.
1421 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1423 for_each_zone_zonelist_nodemask(zone, z, zonelist,
1424 high_zoneidx, nodemask) {
1425 if (NUMA_BUILD && zlc_active &&
1426 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1428 if ((alloc_flags & ALLOC_CPUSET) &&
1429 !cpuset_zone_allowed_softwall(zone, gfp_mask))
1432 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
1434 if (alloc_flags & ALLOC_WMARK_MIN)
1435 mark = zone->pages_min;
1436 else if (alloc_flags & ALLOC_WMARK_LOW)
1437 mark = zone->pages_low;
1439 mark = zone->pages_high;
1440 if (!zone_watermark_ok(zone, order, mark,
1441 classzone_idx, alloc_flags)) {
1442 if (!zone_reclaim_mode ||
1443 !zone_reclaim(zone, gfp_mask, order))
1444 goto this_zone_full;
1448 page = buffered_rmqueue(preferred_zone, zone, order,
1449 gfp_mask, migratetype);
1454 zlc_mark_zone_full(zonelist, z);
1456 if (NUMA_BUILD && !did_zlc_setup) {
1457 /* we do zlc_setup after the first zone is tried */
1458 allowednodes = zlc_setup(zonelist, alloc_flags);
1464 if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) {
1465 /* Disable zlc cache for second zonelist scan */
1473 should_alloc_retry(gfp_t gfp_mask, unsigned int order,
1474 unsigned long pages_reclaimed)
1476 /* Do not loop if specifically requested */
1477 if (gfp_mask & __GFP_NORETRY)
1481 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
1482 * means __GFP_NOFAIL, but that may not be true in other
1485 if (order <= PAGE_ALLOC_COSTLY_ORDER)
1489 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
1490 * specified, then we retry until we no longer reclaim any pages
1491 * (above), or we've reclaimed an order of pages at least as
1492 * large as the allocation's order. In both cases, if the
1493 * allocation still fails, we stop retrying.
1495 if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order))
1499 * Don't let big-order allocations loop unless the caller
1500 * explicitly requests that.
1502 if (gfp_mask & __GFP_NOFAIL)
1508 static inline struct page *
1509 __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
1510 struct zonelist *zonelist, enum zone_type high_zoneidx,
1511 nodemask_t *nodemask, struct zone *preferred_zone,
1516 /* Acquire the OOM killer lock for the zones in zonelist */
1517 if (!try_set_zone_oom(zonelist, gfp_mask)) {
1518 schedule_timeout_uninterruptible(1);
1523 * Go through the zonelist yet one more time, keep very high watermark
1524 * here, this is only to catch a parallel oom killing, we must fail if
1525 * we're still under heavy pressure.
1527 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask,
1528 order, zonelist, high_zoneidx,
1529 ALLOC_WMARK_HIGH|ALLOC_CPUSET,
1530 preferred_zone, migratetype);
1534 /* The OOM killer will not help higher order allocs */
1535 if (order > PAGE_ALLOC_COSTLY_ORDER)
1538 /* Exhausted what can be done so it's blamo time */
1539 out_of_memory(zonelist, gfp_mask, order);
1542 clear_zonelist_oom(zonelist, gfp_mask);
1546 /* The really slow allocator path where we enter direct reclaim */
1547 static inline struct page *
1548 __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
1549 struct zonelist *zonelist, enum zone_type high_zoneidx,
1550 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
1551 int migratetype, unsigned long *did_some_progress)
1553 struct page *page = NULL;
1554 struct reclaim_state reclaim_state;
1555 struct task_struct *p = current;
1559 /* We now go into synchronous reclaim */
1560 cpuset_memory_pressure_bump();
1563 * The task's cpuset might have expanded its set of allowable nodes
1565 p->flags |= PF_MEMALLOC;
1566 lockdep_set_current_reclaim_state(gfp_mask);
1567 reclaim_state.reclaimed_slab = 0;
1568 p->reclaim_state = &reclaim_state;
1570 *did_some_progress = try_to_free_pages(zonelist, order, gfp_mask, nodemask);
1572 p->reclaim_state = NULL;
1573 lockdep_clear_current_reclaim_state();
1574 p->flags &= ~PF_MEMALLOC;
1581 if (likely(*did_some_progress))
1582 page = get_page_from_freelist(gfp_mask, nodemask, order,
1583 zonelist, high_zoneidx,
1584 alloc_flags, preferred_zone,
1590 * This is called in the allocator slow-path if the allocation request is of
1591 * sufficient urgency to ignore watermarks and take other desperate measures
1593 static inline struct page *
1594 __alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order,
1595 struct zonelist *zonelist, enum zone_type high_zoneidx,
1596 nodemask_t *nodemask, struct zone *preferred_zone,
1602 page = get_page_from_freelist(gfp_mask, nodemask, order,
1603 zonelist, high_zoneidx, ALLOC_NO_WATERMARKS,
1604 preferred_zone, migratetype);
1606 if (!page && gfp_mask & __GFP_NOFAIL)
1607 congestion_wait(WRITE, HZ/50);
1608 } while (!page && (gfp_mask & __GFP_NOFAIL));
1614 void wake_all_kswapd(unsigned int order, struct zonelist *zonelist,
1615 enum zone_type high_zoneidx)
1620 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx)
1621 wakeup_kswapd(zone, order);
1625 gfp_to_alloc_flags(gfp_t gfp_mask)
1627 struct task_struct *p = current;
1628 int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
1629 const gfp_t wait = gfp_mask & __GFP_WAIT;
1631 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
1632 BUILD_BUG_ON(__GFP_HIGH != ALLOC_HIGH);
1635 * The caller may dip into page reserves a bit more if the caller
1636 * cannot run direct reclaim, or if the caller has realtime scheduling
1637 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1638 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1640 alloc_flags |= (gfp_mask & __GFP_HIGH);
1643 alloc_flags |= ALLOC_HARDER;
1645 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1646 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1648 alloc_flags &= ~ALLOC_CPUSET;
1649 } else if (unlikely(rt_task(p)))
1650 alloc_flags |= ALLOC_HARDER;
1652 if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) {
1653 if (!in_interrupt() &&
1654 ((p->flags & PF_MEMALLOC) ||
1655 unlikely(test_thread_flag(TIF_MEMDIE))))
1656 alloc_flags |= ALLOC_NO_WATERMARKS;
1662 static inline struct page *
1663 __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
1664 struct zonelist *zonelist, enum zone_type high_zoneidx,
1665 nodemask_t *nodemask, struct zone *preferred_zone,
1668 const gfp_t wait = gfp_mask & __GFP_WAIT;
1669 struct page *page = NULL;
1671 unsigned long pages_reclaimed = 0;
1672 unsigned long did_some_progress;
1673 struct task_struct *p = current;
1676 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1677 * __GFP_NOWARN set) should not cause reclaim since the subsystem
1678 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1679 * using a larger set of nodes after it has established that the
1680 * allowed per node queues are empty and that nodes are
1683 if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
1686 wake_all_kswapd(order, zonelist, high_zoneidx);
1689 * OK, we're below the kswapd watermark and have kicked background
1690 * reclaim. Now things get more complex, so set up alloc_flags according
1691 * to how we want to proceed.
1693 alloc_flags = gfp_to_alloc_flags(gfp_mask);
1696 /* This is the last chance, in general, before the goto nopage. */
1697 page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist,
1698 high_zoneidx, alloc_flags & ~ALLOC_NO_WATERMARKS,
1699 preferred_zone, migratetype);
1704 /* Allocate without watermarks if the context allows */
1705 if (alloc_flags & ALLOC_NO_WATERMARKS) {
1706 page = __alloc_pages_high_priority(gfp_mask, order,
1707 zonelist, high_zoneidx, nodemask,
1708 preferred_zone, migratetype);
1713 /* Atomic allocations - we can't balance anything */
1717 /* Avoid recursion of direct reclaim */
1718 if (p->flags & PF_MEMALLOC)
1721 /* Try direct reclaim and then allocating */
1722 page = __alloc_pages_direct_reclaim(gfp_mask, order,
1723 zonelist, high_zoneidx,
1725 alloc_flags, preferred_zone,
1726 migratetype, &did_some_progress);
1731 * If we failed to make any progress reclaiming, then we are
1732 * running out of options and have to consider going OOM
1734 if (!did_some_progress) {
1735 if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
1736 page = __alloc_pages_may_oom(gfp_mask, order,
1737 zonelist, high_zoneidx,
1738 nodemask, preferred_zone,
1744 * The OOM killer does not trigger for high-order allocations
1745 * but if no progress is being made, there are no other
1746 * options and retrying is unlikely to help
1748 if (order > PAGE_ALLOC_COSTLY_ORDER)
1755 /* Check if we should retry the allocation */
1756 pages_reclaimed += did_some_progress;
1757 if (should_alloc_retry(gfp_mask, order, pages_reclaimed)) {
1758 /* Wait for some write requests to complete then retry */
1759 congestion_wait(WRITE, HZ/50);
1764 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
1765 printk(KERN_WARNING "%s: page allocation failure."
1766 " order:%d, mode:0x%x\n",
1767 p->comm, order, gfp_mask);
1777 * This is the 'heart' of the zoned buddy allocator.
1780 __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
1781 struct zonelist *zonelist, nodemask_t *nodemask)
1783 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
1784 struct zone *preferred_zone;
1786 int migratetype = allocflags_to_migratetype(gfp_mask);
1788 lockdep_trace_alloc(gfp_mask);
1790 might_sleep_if(gfp_mask & __GFP_WAIT);
1792 if (should_fail_alloc_page(gfp_mask, order))
1796 * Check the zones suitable for the gfp_mask contain at least one
1797 * valid zone. It's possible to have an empty zonelist as a result
1798 * of GFP_THISNODE and a memoryless node
1800 if (unlikely(!zonelist->_zonerefs->zone))
1803 /* The preferred zone is used for statistics later */
1804 first_zones_zonelist(zonelist, high_zoneidx, nodemask, &preferred_zone);
1805 if (!preferred_zone)
1808 /* First allocation attempt */
1809 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order,
1810 zonelist, high_zoneidx, ALLOC_WMARK_LOW|ALLOC_CPUSET,
1811 preferred_zone, migratetype);
1812 if (unlikely(!page))
1813 page = __alloc_pages_slowpath(gfp_mask, order,
1814 zonelist, high_zoneidx, nodemask,
1815 preferred_zone, migratetype);
1819 EXPORT_SYMBOL(__alloc_pages_nodemask);
1822 * Common helper functions.
1824 unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
1827 page = alloc_pages(gfp_mask, order);
1830 return (unsigned long) page_address(page);
1833 EXPORT_SYMBOL(__get_free_pages);
1835 unsigned long get_zeroed_page(gfp_t gfp_mask)
1840 * get_zeroed_page() returns a 32-bit address, which cannot represent
1843 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
1845 page = alloc_pages(gfp_mask | __GFP_ZERO, 0);
1847 return (unsigned long) page_address(page);
1851 EXPORT_SYMBOL(get_zeroed_page);
1853 void __pagevec_free(struct pagevec *pvec)
1855 int i = pagevec_count(pvec);
1858 free_hot_cold_page(pvec->pages[i], pvec->cold);
1861 void __free_pages(struct page *page, unsigned int order)
1863 if (put_page_testzero(page)) {
1865 free_hot_page(page);
1867 __free_pages_ok(page, order);
1871 EXPORT_SYMBOL(__free_pages);
1873 void free_pages(unsigned long addr, unsigned int order)
1876 VM_BUG_ON(!virt_addr_valid((void *)addr));
1877 __free_pages(virt_to_page((void *)addr), order);
1881 EXPORT_SYMBOL(free_pages);
1884 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
1885 * @size: the number of bytes to allocate
1886 * @gfp_mask: GFP flags for the allocation
1888 * This function is similar to alloc_pages(), except that it allocates the
1889 * minimum number of pages to satisfy the request. alloc_pages() can only
1890 * allocate memory in power-of-two pages.
1892 * This function is also limited by MAX_ORDER.
1894 * Memory allocated by this function must be released by free_pages_exact().
1896 void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
1898 unsigned int order = get_order(size);
1901 addr = __get_free_pages(gfp_mask, order);
1903 unsigned long alloc_end = addr + (PAGE_SIZE << order);
1904 unsigned long used = addr + PAGE_ALIGN(size);
1906 split_page(virt_to_page(addr), order);
1907 while (used < alloc_end) {
1913 return (void *)addr;
1915 EXPORT_SYMBOL(alloc_pages_exact);
1918 * free_pages_exact - release memory allocated via alloc_pages_exact()
1919 * @virt: the value returned by alloc_pages_exact.
1920 * @size: size of allocation, same value as passed to alloc_pages_exact().
1922 * Release the memory allocated by a previous call to alloc_pages_exact.
1924 void free_pages_exact(void *virt, size_t size)
1926 unsigned long addr = (unsigned long)virt;
1927 unsigned long end = addr + PAGE_ALIGN(size);
1929 while (addr < end) {
1934 EXPORT_SYMBOL(free_pages_exact);
1936 static unsigned int nr_free_zone_pages(int offset)
1941 /* Just pick one node, since fallback list is circular */
1942 unsigned int sum = 0;
1944 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
1946 for_each_zone_zonelist(zone, z, zonelist, offset) {
1947 unsigned long size = zone->present_pages;
1948 unsigned long high = zone->pages_high;
1957 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1959 unsigned int nr_free_buffer_pages(void)
1961 return nr_free_zone_pages(gfp_zone(GFP_USER));
1963 EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
1966 * Amount of free RAM allocatable within all zones
1968 unsigned int nr_free_pagecache_pages(void)
1970 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
1973 static inline void show_node(struct zone *zone)
1976 printk("Node %d ", zone_to_nid(zone));
1979 void si_meminfo(struct sysinfo *val)
1981 val->totalram = totalram_pages;
1983 val->freeram = global_page_state(NR_FREE_PAGES);
1984 val->bufferram = nr_blockdev_pages();
1985 val->totalhigh = totalhigh_pages;
1986 val->freehigh = nr_free_highpages();
1987 val->mem_unit = PAGE_SIZE;
1990 EXPORT_SYMBOL(si_meminfo);
1993 void si_meminfo_node(struct sysinfo *val, int nid)
1995 pg_data_t *pgdat = NODE_DATA(nid);
1997 val->totalram = pgdat->node_present_pages;
1998 val->freeram = node_page_state(nid, NR_FREE_PAGES);
1999 #ifdef CONFIG_HIGHMEM
2000 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
2001 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
2007 val->mem_unit = PAGE_SIZE;
2011 #define K(x) ((x) << (PAGE_SHIFT-10))
2014 * Show free area list (used inside shift_scroll-lock stuff)
2015 * We also calculate the percentage fragmentation. We do this by counting the
2016 * memory on each free list with the exception of the first item on the list.
2018 void show_free_areas(void)
2023 for_each_populated_zone(zone) {
2025 printk("%s per-cpu:\n", zone->name);
2027 for_each_online_cpu(cpu) {
2028 struct per_cpu_pageset *pageset;
2030 pageset = zone_pcp(zone, cpu);
2032 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
2033 cpu, pageset->pcp.high,
2034 pageset->pcp.batch, pageset->pcp.count);
2038 printk("Active_anon:%lu active_file:%lu inactive_anon:%lu\n"
2039 " inactive_file:%lu"
2040 //TODO: check/adjust line lengths
2041 #ifdef CONFIG_UNEVICTABLE_LRU
2044 " dirty:%lu writeback:%lu unstable:%lu\n"
2045 " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
2046 global_page_state(NR_ACTIVE_ANON),
2047 global_page_state(NR_ACTIVE_FILE),
2048 global_page_state(NR_INACTIVE_ANON),
2049 global_page_state(NR_INACTIVE_FILE),
2050 #ifdef CONFIG_UNEVICTABLE_LRU
2051 global_page_state(NR_UNEVICTABLE),
2053 global_page_state(NR_FILE_DIRTY),
2054 global_page_state(NR_WRITEBACK),
2055 global_page_state(NR_UNSTABLE_NFS),
2056 global_page_state(NR_FREE_PAGES),
2057 global_page_state(NR_SLAB_RECLAIMABLE) +
2058 global_page_state(NR_SLAB_UNRECLAIMABLE),
2059 global_page_state(NR_FILE_MAPPED),
2060 global_page_state(NR_PAGETABLE),
2061 global_page_state(NR_BOUNCE));
2063 for_each_populated_zone(zone) {
2072 " active_anon:%lukB"
2073 " inactive_anon:%lukB"
2074 " active_file:%lukB"
2075 " inactive_file:%lukB"
2076 #ifdef CONFIG_UNEVICTABLE_LRU
2077 " unevictable:%lukB"
2080 " pages_scanned:%lu"
2081 " all_unreclaimable? %s"
2084 K(zone_page_state(zone, NR_FREE_PAGES)),
2087 K(zone->pages_high),
2088 K(zone_page_state(zone, NR_ACTIVE_ANON)),
2089 K(zone_page_state(zone, NR_INACTIVE_ANON)),
2090 K(zone_page_state(zone, NR_ACTIVE_FILE)),
2091 K(zone_page_state(zone, NR_INACTIVE_FILE)),
2092 #ifdef CONFIG_UNEVICTABLE_LRU
2093 K(zone_page_state(zone, NR_UNEVICTABLE)),
2095 K(zone->present_pages),
2096 zone->pages_scanned,
2097 (zone_is_all_unreclaimable(zone) ? "yes" : "no")
2099 printk("lowmem_reserve[]:");
2100 for (i = 0; i < MAX_NR_ZONES; i++)
2101 printk(" %lu", zone->lowmem_reserve[i]);
2105 for_each_populated_zone(zone) {
2106 unsigned long nr[MAX_ORDER], flags, order, total = 0;
2109 printk("%s: ", zone->name);
2111 spin_lock_irqsave(&zone->lock, flags);
2112 for (order = 0; order < MAX_ORDER; order++) {
2113 nr[order] = zone->free_area[order].nr_free;
2114 total += nr[order] << order;
2116 spin_unlock_irqrestore(&zone->lock, flags);
2117 for (order = 0; order < MAX_ORDER; order++)
2118 printk("%lu*%lukB ", nr[order], K(1UL) << order);
2119 printk("= %lukB\n", K(total));
2122 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
2124 show_swap_cache_info();
2127 static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
2129 zoneref->zone = zone;
2130 zoneref->zone_idx = zone_idx(zone);
2134 * Builds allocation fallback zone lists.
2136 * Add all populated zones of a node to the zonelist.
2138 static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
2139 int nr_zones, enum zone_type zone_type)
2143 BUG_ON(zone_type >= MAX_NR_ZONES);
2148 zone = pgdat->node_zones + zone_type;
2149 if (populated_zone(zone)) {
2150 zoneref_set_zone(zone,
2151 &zonelist->_zonerefs[nr_zones++]);
2152 check_highest_zone(zone_type);
2155 } while (zone_type);
2162 * 0 = automatic detection of better ordering.
2163 * 1 = order by ([node] distance, -zonetype)
2164 * 2 = order by (-zonetype, [node] distance)
2166 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
2167 * the same zonelist. So only NUMA can configure this param.
2169 #define ZONELIST_ORDER_DEFAULT 0
2170 #define ZONELIST_ORDER_NODE 1
2171 #define ZONELIST_ORDER_ZONE 2
2173 /* zonelist order in the kernel.
2174 * set_zonelist_order() will set this to NODE or ZONE.
2176 static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
2177 static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
2181 /* The value user specified ....changed by config */
2182 static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2183 /* string for sysctl */
2184 #define NUMA_ZONELIST_ORDER_LEN 16
2185 char numa_zonelist_order[16] = "default";
2188 * interface for configure zonelist ordering.
2189 * command line option "numa_zonelist_order"
2190 * = "[dD]efault - default, automatic configuration.
2191 * = "[nN]ode - order by node locality, then by zone within node
2192 * = "[zZ]one - order by zone, then by locality within zone
2195 static int __parse_numa_zonelist_order(char *s)
2197 if (*s == 'd' || *s == 'D') {
2198 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2199 } else if (*s == 'n' || *s == 'N') {
2200 user_zonelist_order = ZONELIST_ORDER_NODE;
2201 } else if (*s == 'z' || *s == 'Z') {
2202 user_zonelist_order = ZONELIST_ORDER_ZONE;
2205 "Ignoring invalid numa_zonelist_order value: "
2212 static __init int setup_numa_zonelist_order(char *s)
2215 return __parse_numa_zonelist_order(s);
2218 early_param("numa_zonelist_order", setup_numa_zonelist_order);
2221 * sysctl handler for numa_zonelist_order
2223 int numa_zonelist_order_handler(ctl_table *table, int write,
2224 struct file *file, void __user *buffer, size_t *length,
2227 char saved_string[NUMA_ZONELIST_ORDER_LEN];
2231 strncpy(saved_string, (char*)table->data,
2232 NUMA_ZONELIST_ORDER_LEN);
2233 ret = proc_dostring(table, write, file, buffer, length, ppos);
2237 int oldval = user_zonelist_order;
2238 if (__parse_numa_zonelist_order((char*)table->data)) {
2240 * bogus value. restore saved string
2242 strncpy((char*)table->data, saved_string,
2243 NUMA_ZONELIST_ORDER_LEN);
2244 user_zonelist_order = oldval;
2245 } else if (oldval != user_zonelist_order)
2246 build_all_zonelists();
2252 #define MAX_NODE_LOAD (num_online_nodes())
2253 static int node_load[MAX_NUMNODES];
2256 * find_next_best_node - find the next node that should appear in a given node's fallback list
2257 * @node: node whose fallback list we're appending
2258 * @used_node_mask: nodemask_t of already used nodes
2260 * We use a number of factors to determine which is the next node that should
2261 * appear on a given node's fallback list. The node should not have appeared
2262 * already in @node's fallback list, and it should be the next closest node
2263 * according to the distance array (which contains arbitrary distance values
2264 * from each node to each node in the system), and should also prefer nodes
2265 * with no CPUs, since presumably they'll have very little allocation pressure
2266 * on them otherwise.
2267 * It returns -1 if no node is found.
2269 static int find_next_best_node(int node, nodemask_t *used_node_mask)
2272 int min_val = INT_MAX;
2274 const struct cpumask *tmp = cpumask_of_node(0);
2276 /* Use the local node if we haven't already */
2277 if (!node_isset(node, *used_node_mask)) {
2278 node_set(node, *used_node_mask);
2282 for_each_node_state(n, N_HIGH_MEMORY) {
2284 /* Don't want a node to appear more than once */
2285 if (node_isset(n, *used_node_mask))
2288 /* Use the distance array to find the distance */
2289 val = node_distance(node, n);
2291 /* Penalize nodes under us ("prefer the next node") */
2294 /* Give preference to headless and unused nodes */
2295 tmp = cpumask_of_node(n);
2296 if (!cpumask_empty(tmp))
2297 val += PENALTY_FOR_NODE_WITH_CPUS;
2299 /* Slight preference for less loaded node */
2300 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
2301 val += node_load[n];
2303 if (val < min_val) {
2310 node_set(best_node, *used_node_mask);
2317 * Build zonelists ordered by node and zones within node.
2318 * This results in maximum locality--normal zone overflows into local
2319 * DMA zone, if any--but risks exhausting DMA zone.
2321 static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
2324 struct zonelist *zonelist;
2326 zonelist = &pgdat->node_zonelists[0];
2327 for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
2329 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2331 zonelist->_zonerefs[j].zone = NULL;
2332 zonelist->_zonerefs[j].zone_idx = 0;
2336 * Build gfp_thisnode zonelists
2338 static void build_thisnode_zonelists(pg_data_t *pgdat)
2341 struct zonelist *zonelist;
2343 zonelist = &pgdat->node_zonelists[1];
2344 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
2345 zonelist->_zonerefs[j].zone = NULL;
2346 zonelist->_zonerefs[j].zone_idx = 0;
2350 * Build zonelists ordered by zone and nodes within zones.
2351 * This results in conserving DMA zone[s] until all Normal memory is
2352 * exhausted, but results in overflowing to remote node while memory
2353 * may still exist in local DMA zone.
2355 static int node_order[MAX_NUMNODES];
2357 static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
2360 int zone_type; /* needs to be signed */
2362 struct zonelist *zonelist;
2364 zonelist = &pgdat->node_zonelists[0];
2366 for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
2367 for (j = 0; j < nr_nodes; j++) {
2368 node = node_order[j];
2369 z = &NODE_DATA(node)->node_zones[zone_type];
2370 if (populated_zone(z)) {
2372 &zonelist->_zonerefs[pos++]);
2373 check_highest_zone(zone_type);
2377 zonelist->_zonerefs[pos].zone = NULL;
2378 zonelist->_zonerefs[pos].zone_idx = 0;
2381 static int default_zonelist_order(void)
2384 unsigned long low_kmem_size,total_size;
2388 * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem.
2389 * If they are really small and used heavily, the system can fall
2390 * into OOM very easily.
2391 * This function detect ZONE_DMA/DMA32 size and confgigures zone order.
2393 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2396 for_each_online_node(nid) {
2397 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2398 z = &NODE_DATA(nid)->node_zones[zone_type];
2399 if (populated_zone(z)) {
2400 if (zone_type < ZONE_NORMAL)
2401 low_kmem_size += z->present_pages;
2402 total_size += z->present_pages;
2406 if (!low_kmem_size || /* there are no DMA area. */
2407 low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
2408 return ZONELIST_ORDER_NODE;
2410 * look into each node's config.
2411 * If there is a node whose DMA/DMA32 memory is very big area on
2412 * local memory, NODE_ORDER may be suitable.
2414 average_size = total_size /
2415 (nodes_weight(node_states[N_HIGH_MEMORY]) + 1);
2416 for_each_online_node(nid) {
2419 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2420 z = &NODE_DATA(nid)->node_zones[zone_type];
2421 if (populated_zone(z)) {
2422 if (zone_type < ZONE_NORMAL)
2423 low_kmem_size += z->present_pages;
2424 total_size += z->present_pages;
2427 if (low_kmem_size &&
2428 total_size > average_size && /* ignore small node */
2429 low_kmem_size > total_size * 70/100)
2430 return ZONELIST_ORDER_NODE;
2432 return ZONELIST_ORDER_ZONE;
2435 static void set_zonelist_order(void)
2437 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
2438 current_zonelist_order = default_zonelist_order();
2440 current_zonelist_order = user_zonelist_order;
2443 static void build_zonelists(pg_data_t *pgdat)
2447 nodemask_t used_mask;
2448 int local_node, prev_node;
2449 struct zonelist *zonelist;
2450 int order = current_zonelist_order;
2452 /* initialize zonelists */
2453 for (i = 0; i < MAX_ZONELISTS; i++) {
2454 zonelist = pgdat->node_zonelists + i;
2455 zonelist->_zonerefs[0].zone = NULL;
2456 zonelist->_zonerefs[0].zone_idx = 0;
2459 /* NUMA-aware ordering of nodes */
2460 local_node = pgdat->node_id;
2461 load = num_online_nodes();
2462 prev_node = local_node;
2463 nodes_clear(used_mask);
2465 memset(node_load, 0, sizeof(node_load));
2466 memset(node_order, 0, sizeof(node_order));
2469 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
2470 int distance = node_distance(local_node, node);
2473 * If another node is sufficiently far away then it is better
2474 * to reclaim pages in a zone before going off node.
2476 if (distance > RECLAIM_DISTANCE)
2477 zone_reclaim_mode = 1;
2480 * We don't want to pressure a particular node.
2481 * So adding penalty to the first node in same
2482 * distance group to make it round-robin.
2484 if (distance != node_distance(local_node, prev_node))
2485 node_load[node] = load;
2489 if (order == ZONELIST_ORDER_NODE)
2490 build_zonelists_in_node_order(pgdat, node);
2492 node_order[j++] = node; /* remember order */
2495 if (order == ZONELIST_ORDER_ZONE) {
2496 /* calculate node order -- i.e., DMA last! */
2497 build_zonelists_in_zone_order(pgdat, j);
2500 build_thisnode_zonelists(pgdat);
2503 /* Construct the zonelist performance cache - see further mmzone.h */
2504 static void build_zonelist_cache(pg_data_t *pgdat)
2506 struct zonelist *zonelist;
2507 struct zonelist_cache *zlc;
2510 zonelist = &pgdat->node_zonelists[0];
2511 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
2512 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
2513 for (z = zonelist->_zonerefs; z->zone; z++)
2514 zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z);
2518 #else /* CONFIG_NUMA */
2520 static void set_zonelist_order(void)
2522 current_zonelist_order = ZONELIST_ORDER_ZONE;
2525 static void build_zonelists(pg_data_t *pgdat)
2527 int node, local_node;
2529 struct zonelist *zonelist;
2531 local_node = pgdat->node_id;
2533 zonelist = &pgdat->node_zonelists[0];
2534 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
2537 * Now we build the zonelist so that it contains the zones
2538 * of all the other nodes.
2539 * We don't want to pressure a particular node, so when
2540 * building the zones for node N, we make sure that the
2541 * zones coming right after the local ones are those from
2542 * node N+1 (modulo N)
2544 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
2545 if (!node_online(node))
2547 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2550 for (node = 0; node < local_node; node++) {
2551 if (!node_online(node))
2553 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2557 zonelist->_zonerefs[j].zone = NULL;
2558 zonelist->_zonerefs[j].zone_idx = 0;
2561 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
2562 static void build_zonelist_cache(pg_data_t *pgdat)
2564 pgdat->node_zonelists[0].zlcache_ptr = NULL;
2567 #endif /* CONFIG_NUMA */
2569 /* return values int ....just for stop_machine() */
2570 static int __build_all_zonelists(void *dummy)
2574 for_each_online_node(nid) {
2575 pg_data_t *pgdat = NODE_DATA(nid);
2577 build_zonelists(pgdat);
2578 build_zonelist_cache(pgdat);
2583 void build_all_zonelists(void)
2585 set_zonelist_order();
2587 if (system_state == SYSTEM_BOOTING) {
2588 __build_all_zonelists(NULL);
2589 mminit_verify_zonelist();
2590 cpuset_init_current_mems_allowed();
2592 /* we have to stop all cpus to guarantee there is no user
2594 stop_machine(__build_all_zonelists, NULL, NULL);
2595 /* cpuset refresh routine should be here */
2597 vm_total_pages = nr_free_pagecache_pages();
2599 * Disable grouping by mobility if the number of pages in the
2600 * system is too low to allow the mechanism to work. It would be
2601 * more accurate, but expensive to check per-zone. This check is
2602 * made on memory-hotadd so a system can start with mobility
2603 * disabled and enable it later
2605 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
2606 page_group_by_mobility_disabled = 1;
2608 page_group_by_mobility_disabled = 0;
2610 printk("Built %i zonelists in %s order, mobility grouping %s. "
2611 "Total pages: %ld\n",
2613 zonelist_order_name[current_zonelist_order],
2614 page_group_by_mobility_disabled ? "off" : "on",
2617 printk("Policy zone: %s\n", zone_names[policy_zone]);
2622 * Helper functions to size the waitqueue hash table.
2623 * Essentially these want to choose hash table sizes sufficiently
2624 * large so that collisions trying to wait on pages are rare.
2625 * But in fact, the number of active page waitqueues on typical
2626 * systems is ridiculously low, less than 200. So this is even
2627 * conservative, even though it seems large.
2629 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
2630 * waitqueues, i.e. the size of the waitq table given the number of pages.
2632 #define PAGES_PER_WAITQUEUE 256
2634 #ifndef CONFIG_MEMORY_HOTPLUG
2635 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
2637 unsigned long size = 1;
2639 pages /= PAGES_PER_WAITQUEUE;
2641 while (size < pages)
2645 * Once we have dozens or even hundreds of threads sleeping
2646 * on IO we've got bigger problems than wait queue collision.
2647 * Limit the size of the wait table to a reasonable size.
2649 size = min(size, 4096UL);
2651 return max(size, 4UL);
2655 * A zone's size might be changed by hot-add, so it is not possible to determine
2656 * a suitable size for its wait_table. So we use the maximum size now.
2658 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
2660 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
2661 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
2662 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
2664 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
2665 * or more by the traditional way. (See above). It equals:
2667 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
2668 * ia64(16K page size) : = ( 8G + 4M)byte.
2669 * powerpc (64K page size) : = (32G +16M)byte.
2671 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
2678 * This is an integer logarithm so that shifts can be used later
2679 * to extract the more random high bits from the multiplicative
2680 * hash function before the remainder is taken.
2682 static inline unsigned long wait_table_bits(unsigned long size)
2687 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
2690 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
2691 * of blocks reserved is based on zone->pages_min. The memory within the
2692 * reserve will tend to store contiguous free pages. Setting min_free_kbytes
2693 * higher will lead to a bigger reserve which will get freed as contiguous
2694 * blocks as reclaim kicks in
2696 static void setup_zone_migrate_reserve(struct zone *zone)
2698 unsigned long start_pfn, pfn, end_pfn;
2700 unsigned long reserve, block_migratetype;
2702 /* Get the start pfn, end pfn and the number of blocks to reserve */
2703 start_pfn = zone->zone_start_pfn;
2704 end_pfn = start_pfn + zone->spanned_pages;
2705 reserve = roundup(zone->pages_min, pageblock_nr_pages) >>
2708 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
2709 if (!pfn_valid(pfn))
2711 page = pfn_to_page(pfn);
2713 /* Watch out for overlapping nodes */
2714 if (page_to_nid(page) != zone_to_nid(zone))
2717 /* Blocks with reserved pages will never free, skip them. */
2718 if (PageReserved(page))
2721 block_migratetype = get_pageblock_migratetype(page);
2723 /* If this block is reserved, account for it */
2724 if (reserve > 0 && block_migratetype == MIGRATE_RESERVE) {
2729 /* Suitable for reserving if this block is movable */
2730 if (reserve > 0 && block_migratetype == MIGRATE_MOVABLE) {
2731 set_pageblock_migratetype(page, MIGRATE_RESERVE);
2732 move_freepages_block(zone, page, MIGRATE_RESERVE);
2738 * If the reserve is met and this is a previous reserved block,
2741 if (block_migratetype == MIGRATE_RESERVE) {
2742 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
2743 move_freepages_block(zone, page, MIGRATE_MOVABLE);
2749 * Initially all pages are reserved - free ones are freed
2750 * up by free_all_bootmem() once the early boot process is
2751 * done. Non-atomic initialization, single-pass.
2753 void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
2754 unsigned long start_pfn, enum memmap_context context)
2757 unsigned long end_pfn = start_pfn + size;
2761 if (highest_memmap_pfn < end_pfn - 1)
2762 highest_memmap_pfn = end_pfn - 1;
2764 z = &NODE_DATA(nid)->node_zones[zone];
2765 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
2767 * There can be holes in boot-time mem_map[]s
2768 * handed to this function. They do not
2769 * exist on hotplugged memory.
2771 if (context == MEMMAP_EARLY) {
2772 if (!early_pfn_valid(pfn))
2774 if (!early_pfn_in_nid(pfn, nid))
2777 page = pfn_to_page(pfn);
2778 set_page_links(page, zone, nid, pfn);
2779 mminit_verify_page_links(page, zone, nid, pfn);
2780 init_page_count(page);
2781 reset_page_mapcount(page);
2782 SetPageReserved(page);
2784 * Mark the block movable so that blocks are reserved for
2785 * movable at startup. This will force kernel allocations
2786 * to reserve their blocks rather than leaking throughout
2787 * the address space during boot when many long-lived
2788 * kernel allocations are made. Later some blocks near
2789 * the start are marked MIGRATE_RESERVE by
2790 * setup_zone_migrate_reserve()
2792 * bitmap is created for zone's valid pfn range. but memmap
2793 * can be created for invalid pages (for alignment)
2794 * check here not to call set_pageblock_migratetype() against
2797 if ((z->zone_start_pfn <= pfn)
2798 && (pfn < z->zone_start_pfn + z->spanned_pages)
2799 && !(pfn & (pageblock_nr_pages - 1)))
2800 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
2802 INIT_LIST_HEAD(&page->lru);
2803 #ifdef WANT_PAGE_VIRTUAL
2804 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
2805 if (!is_highmem_idx(zone))
2806 set_page_address(page, __va(pfn << PAGE_SHIFT));
2811 static void __meminit zone_init_free_lists(struct zone *zone)
2814 for_each_migratetype_order(order, t) {
2815 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
2816 zone->free_area[order].nr_free = 0;
2820 #ifndef __HAVE_ARCH_MEMMAP_INIT
2821 #define memmap_init(size, nid, zone, start_pfn) \
2822 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
2825 static int zone_batchsize(struct zone *zone)
2831 * The per-cpu-pages pools are set to around 1000th of the
2832 * size of the zone. But no more than 1/2 of a meg.
2834 * OK, so we don't know how big the cache is. So guess.
2836 batch = zone->present_pages / 1024;
2837 if (batch * PAGE_SIZE > 512 * 1024)
2838 batch = (512 * 1024) / PAGE_SIZE;
2839 batch /= 4; /* We effectively *= 4 below */
2844 * Clamp the batch to a 2^n - 1 value. Having a power
2845 * of 2 value was found to be more likely to have
2846 * suboptimal cache aliasing properties in some cases.
2848 * For example if 2 tasks are alternately allocating
2849 * batches of pages, one task can end up with a lot
2850 * of pages of one half of the possible page colors
2851 * and the other with pages of the other colors.
2853 batch = rounddown_pow_of_two(batch + batch/2) - 1;
2858 /* The deferral and batching of frees should be suppressed under NOMMU
2861 * The problem is that NOMMU needs to be able to allocate large chunks
2862 * of contiguous memory as there's no hardware page translation to
2863 * assemble apparent contiguous memory from discontiguous pages.
2865 * Queueing large contiguous runs of pages for batching, however,
2866 * causes the pages to actually be freed in smaller chunks. As there
2867 * can be a significant delay between the individual batches being
2868 * recycled, this leads to the once large chunks of space being
2869 * fragmented and becoming unavailable for high-order allocations.
2875 static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
2877 struct per_cpu_pages *pcp;
2879 memset(p, 0, sizeof(*p));
2883 pcp->high = 6 * batch;
2884 pcp->batch = max(1UL, 1 * batch);
2885 INIT_LIST_HEAD(&pcp->list);
2889 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
2890 * to the value high for the pageset p.
2893 static void setup_pagelist_highmark(struct per_cpu_pageset *p,
2896 struct per_cpu_pages *pcp;
2900 pcp->batch = max(1UL, high/4);
2901 if ((high/4) > (PAGE_SHIFT * 8))
2902 pcp->batch = PAGE_SHIFT * 8;
2908 * Boot pageset table. One per cpu which is going to be used for all
2909 * zones and all nodes. The parameters will be set in such a way
2910 * that an item put on a list will immediately be handed over to
2911 * the buddy list. This is safe since pageset manipulation is done
2912 * with interrupts disabled.
2914 * Some NUMA counter updates may also be caught by the boot pagesets.
2916 * The boot_pagesets must be kept even after bootup is complete for
2917 * unused processors and/or zones. They do play a role for bootstrapping
2918 * hotplugged processors.
2920 * zoneinfo_show() and maybe other functions do
2921 * not check if the processor is online before following the pageset pointer.
2922 * Other parts of the kernel may not check if the zone is available.
2924 static struct per_cpu_pageset boot_pageset[NR_CPUS];
2927 * Dynamically allocate memory for the
2928 * per cpu pageset array in struct zone.
2930 static int __cpuinit process_zones(int cpu)
2932 struct zone *zone, *dzone;
2933 int node = cpu_to_node(cpu);
2935 node_set_state(node, N_CPU); /* this node has a cpu */
2937 for_each_populated_zone(zone) {
2938 zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset),
2940 if (!zone_pcp(zone, cpu))
2943 setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone));
2945 if (percpu_pagelist_fraction)
2946 setup_pagelist_highmark(zone_pcp(zone, cpu),
2947 (zone->present_pages / percpu_pagelist_fraction));
2952 for_each_zone(dzone) {
2953 if (!populated_zone(dzone))
2957 kfree(zone_pcp(dzone, cpu));
2958 zone_pcp(dzone, cpu) = NULL;
2963 static inline void free_zone_pagesets(int cpu)
2967 for_each_zone(zone) {
2968 struct per_cpu_pageset *pset = zone_pcp(zone, cpu);
2970 /* Free per_cpu_pageset if it is slab allocated */
2971 if (pset != &boot_pageset[cpu])
2973 zone_pcp(zone, cpu) = NULL;
2977 static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb,
2978 unsigned long action,
2981 int cpu = (long)hcpu;
2982 int ret = NOTIFY_OK;
2985 case CPU_UP_PREPARE:
2986 case CPU_UP_PREPARE_FROZEN:
2987 if (process_zones(cpu))
2990 case CPU_UP_CANCELED:
2991 case CPU_UP_CANCELED_FROZEN:
2993 case CPU_DEAD_FROZEN:
2994 free_zone_pagesets(cpu);
3002 static struct notifier_block __cpuinitdata pageset_notifier =
3003 { &pageset_cpuup_callback, NULL, 0 };
3005 void __init setup_per_cpu_pageset(void)
3009 /* Initialize per_cpu_pageset for cpu 0.
3010 * A cpuup callback will do this for every cpu
3011 * as it comes online
3013 err = process_zones(smp_processor_id());
3015 register_cpu_notifier(&pageset_notifier);
3020 static noinline __init_refok
3021 int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
3024 struct pglist_data *pgdat = zone->zone_pgdat;
3028 * The per-page waitqueue mechanism uses hashed waitqueues
3031 zone->wait_table_hash_nr_entries =
3032 wait_table_hash_nr_entries(zone_size_pages);
3033 zone->wait_table_bits =
3034 wait_table_bits(zone->wait_table_hash_nr_entries);
3035 alloc_size = zone->wait_table_hash_nr_entries
3036 * sizeof(wait_queue_head_t);
3038 if (!slab_is_available()) {
3039 zone->wait_table = (wait_queue_head_t *)
3040 alloc_bootmem_node(pgdat, alloc_size);
3043 * This case means that a zone whose size was 0 gets new memory
3044 * via memory hot-add.
3045 * But it may be the case that a new node was hot-added. In
3046 * this case vmalloc() will not be able to use this new node's
3047 * memory - this wait_table must be initialized to use this new
3048 * node itself as well.
3049 * To use this new node's memory, further consideration will be
3052 zone->wait_table = vmalloc(alloc_size);
3054 if (!zone->wait_table)
3057 for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
3058 init_waitqueue_head(zone->wait_table + i);
3063 static __meminit void zone_pcp_init(struct zone *zone)
3066 unsigned long batch = zone_batchsize(zone);
3068 for (cpu = 0; cpu < NR_CPUS; cpu++) {
3070 /* Early boot. Slab allocator not functional yet */
3071 zone_pcp(zone, cpu) = &boot_pageset[cpu];
3072 setup_pageset(&boot_pageset[cpu],0);
3074 setup_pageset(zone_pcp(zone,cpu), batch);
3077 if (zone->present_pages)
3078 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n",
3079 zone->name, zone->present_pages, batch);
3082 __meminit int init_currently_empty_zone(struct zone *zone,
3083 unsigned long zone_start_pfn,
3085 enum memmap_context context)
3087 struct pglist_data *pgdat = zone->zone_pgdat;
3089 ret = zone_wait_table_init(zone, size);
3092 pgdat->nr_zones = zone_idx(zone) + 1;
3094 zone->zone_start_pfn = zone_start_pfn;
3096 mminit_dprintk(MMINIT_TRACE, "memmap_init",
3097 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
3099 (unsigned long)zone_idx(zone),
3100 zone_start_pfn, (zone_start_pfn + size));
3102 zone_init_free_lists(zone);
3107 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3109 * Basic iterator support. Return the first range of PFNs for a node
3110 * Note: nid == MAX_NUMNODES returns first region regardless of node
3112 static int __meminit first_active_region_index_in_nid(int nid)
3116 for (i = 0; i < nr_nodemap_entries; i++)
3117 if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
3124 * Basic iterator support. Return the next active range of PFNs for a node
3125 * Note: nid == MAX_NUMNODES returns next region regardless of node
3127 static int __meminit next_active_region_index_in_nid(int index, int nid)
3129 for (index = index + 1; index < nr_nodemap_entries; index++)
3130 if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
3136 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
3138 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
3139 * Architectures may implement their own version but if add_active_range()
3140 * was used and there are no special requirements, this is a convenient
3143 int __meminit __early_pfn_to_nid(unsigned long pfn)
3147 for (i = 0; i < nr_nodemap_entries; i++) {
3148 unsigned long start_pfn = early_node_map[i].start_pfn;
3149 unsigned long end_pfn = early_node_map[i].end_pfn;
3151 if (start_pfn <= pfn && pfn < end_pfn)
3152 return early_node_map[i].nid;
3154 /* This is a memory hole */
3157 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
3159 int __meminit early_pfn_to_nid(unsigned long pfn)
3163 nid = __early_pfn_to_nid(pfn);
3166 /* just returns 0 */
3170 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
3171 bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
3175 nid = __early_pfn_to_nid(pfn);
3176 if (nid >= 0 && nid != node)
3182 /* Basic iterator support to walk early_node_map[] */
3183 #define for_each_active_range_index_in_nid(i, nid) \
3184 for (i = first_active_region_index_in_nid(nid); i != -1; \
3185 i = next_active_region_index_in_nid(i, nid))
3188 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
3189 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
3190 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
3192 * If an architecture guarantees that all ranges registered with
3193 * add_active_ranges() contain no holes and may be freed, this
3194 * this function may be used instead of calling free_bootmem() manually.
3196 void __init free_bootmem_with_active_regions(int nid,
3197 unsigned long max_low_pfn)
3201 for_each_active_range_index_in_nid(i, nid) {
3202 unsigned long size_pages = 0;
3203 unsigned long end_pfn = early_node_map[i].end_pfn;
3205 if (early_node_map[i].start_pfn >= max_low_pfn)
3208 if (end_pfn > max_low_pfn)
3209 end_pfn = max_low_pfn;
3211 size_pages = end_pfn - early_node_map[i].start_pfn;
3212 free_bootmem_node(NODE_DATA(early_node_map[i].nid),
3213 PFN_PHYS(early_node_map[i].start_pfn),
3214 size_pages << PAGE_SHIFT);
3218 void __init work_with_active_regions(int nid, work_fn_t work_fn, void *data)
3223 for_each_active_range_index_in_nid(i, nid) {
3224 ret = work_fn(early_node_map[i].start_pfn,
3225 early_node_map[i].end_pfn, data);
3231 * sparse_memory_present_with_active_regions - Call memory_present for each active range
3232 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
3234 * If an architecture guarantees that all ranges registered with
3235 * add_active_ranges() contain no holes and may be freed, this
3236 * function may be used instead of calling memory_present() manually.
3238 void __init sparse_memory_present_with_active_regions(int nid)
3242 for_each_active_range_index_in_nid(i, nid)
3243 memory_present(early_node_map[i].nid,
3244 early_node_map[i].start_pfn,
3245 early_node_map[i].end_pfn);
3249 * get_pfn_range_for_nid - Return the start and end page frames for a node
3250 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
3251 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
3252 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
3254 * It returns the start and end page frame of a node based on information
3255 * provided by an arch calling add_active_range(). If called for a node
3256 * with no available memory, a warning is printed and the start and end
3259 void __meminit get_pfn_range_for_nid(unsigned int nid,
3260 unsigned long *start_pfn, unsigned long *end_pfn)
3266 for_each_active_range_index_in_nid(i, nid) {
3267 *start_pfn = min(*start_pfn, early_node_map[i].start_pfn);
3268 *end_pfn = max(*end_pfn, early_node_map[i].end_pfn);
3271 if (*start_pfn == -1UL)
3276 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3277 * assumption is made that zones within a node are ordered in monotonic
3278 * increasing memory addresses so that the "highest" populated zone is used
3280 static void __init find_usable_zone_for_movable(void)
3283 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
3284 if (zone_index == ZONE_MOVABLE)
3287 if (arch_zone_highest_possible_pfn[zone_index] >
3288 arch_zone_lowest_possible_pfn[zone_index])
3292 VM_BUG_ON(zone_index == -1);
3293 movable_zone = zone_index;
3297 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
3298 * because it is sized independant of architecture. Unlike the other zones,
3299 * the starting point for ZONE_MOVABLE is not fixed. It may be different
3300 * in each node depending on the size of each node and how evenly kernelcore
3301 * is distributed. This helper function adjusts the zone ranges
3302 * provided by the architecture for a given node by using the end of the
3303 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3304 * zones within a node are in order of monotonic increases memory addresses
3306 static void __meminit adjust_zone_range_for_zone_movable(int nid,
3307 unsigned long zone_type,
3308 unsigned long node_start_pfn,
3309 unsigned long node_end_pfn,
3310 unsigned long *zone_start_pfn,
3311 unsigned long *zone_end_pfn)
3313 /* Only adjust if ZONE_MOVABLE is on this node */
3314 if (zone_movable_pfn[nid]) {
3315 /* Size ZONE_MOVABLE */
3316 if (zone_type == ZONE_MOVABLE) {
3317 *zone_start_pfn = zone_movable_pfn[nid];
3318 *zone_end_pfn = min(node_end_pfn,
3319 arch_zone_highest_possible_pfn[movable_zone]);
3321 /* Adjust for ZONE_MOVABLE starting within this range */
3322 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
3323 *zone_end_pfn > zone_movable_pfn[nid]) {
3324 *zone_end_pfn = zone_movable_pfn[nid];
3326 /* Check if this whole range is within ZONE_MOVABLE */
3327 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
3328 *zone_start_pfn = *zone_end_pfn;
3333 * Return the number of pages a zone spans in a node, including holes
3334 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3336 static unsigned long __meminit zone_spanned_pages_in_node(int nid,
3337 unsigned long zone_type,
3338 unsigned long *ignored)
3340 unsigned long node_start_pfn, node_end_pfn;
3341 unsigned long zone_start_pfn, zone_end_pfn;
3343 /* Get the start and end of the node and zone */
3344 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3345 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
3346 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
3347 adjust_zone_range_for_zone_movable(nid, zone_type,
3348 node_start_pfn, node_end_pfn,
3349 &zone_start_pfn, &zone_end_pfn);
3351 /* Check that this node has pages within the zone's required range */
3352 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
3355 /* Move the zone boundaries inside the node if necessary */
3356 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
3357 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
3359 /* Return the spanned pages */
3360 return zone_end_pfn - zone_start_pfn;
3364 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
3365 * then all holes in the requested range will be accounted for.
3367 static unsigned long __meminit __absent_pages_in_range(int nid,
3368 unsigned long range_start_pfn,
3369 unsigned long range_end_pfn)
3372 unsigned long prev_end_pfn = 0, hole_pages = 0;
3373 unsigned long start_pfn;
3375 /* Find the end_pfn of the first active range of pfns in the node */
3376 i = first_active_region_index_in_nid(nid);
3380 prev_end_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3382 /* Account for ranges before physical memory on this node */
3383 if (early_node_map[i].start_pfn > range_start_pfn)
3384 hole_pages = prev_end_pfn - range_start_pfn;
3386 /* Find all holes for the zone within the node */
3387 for (; i != -1; i = next_active_region_index_in_nid(i, nid)) {
3389 /* No need to continue if prev_end_pfn is outside the zone */
3390 if (prev_end_pfn >= range_end_pfn)
3393 /* Make sure the end of the zone is not within the hole */
3394 start_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3395 prev_end_pfn = max(prev_end_pfn, range_start_pfn);
3397 /* Update the hole size cound and move on */
3398 if (start_pfn > range_start_pfn) {
3399 BUG_ON(prev_end_pfn > start_pfn);
3400 hole_pages += start_pfn - prev_end_pfn;
3402 prev_end_pfn = early_node_map[i].end_pfn;
3405 /* Account for ranges past physical memory on this node */
3406 if (range_end_pfn > prev_end_pfn)
3407 hole_pages += range_end_pfn -
3408 max(range_start_pfn, prev_end_pfn);
3414 * absent_pages_in_range - Return number of page frames in holes within a range
3415 * @start_pfn: The start PFN to start searching for holes
3416 * @end_pfn: The end PFN to stop searching for holes
3418 * It returns the number of pages frames in memory holes within a range.
3420 unsigned long __init absent_pages_in_range(unsigned long start_pfn,
3421 unsigned long end_pfn)
3423 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
3426 /* Return the number of page frames in holes in a zone on a node */
3427 static unsigned long __meminit zone_absent_pages_in_node(int nid,
3428 unsigned long zone_type,
3429 unsigned long *ignored)
3431 unsigned long node_start_pfn, node_end_pfn;
3432 unsigned long zone_start_pfn, zone_end_pfn;
3434 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3435 zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type],
3437 zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type],
3440 adjust_zone_range_for_zone_movable(nid, zone_type,
3441 node_start_pfn, node_end_pfn,
3442 &zone_start_pfn, &zone_end_pfn);
3443 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
3447 static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
3448 unsigned long zone_type,
3449 unsigned long *zones_size)
3451 return zones_size[zone_type];
3454 static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
3455 unsigned long zone_type,
3456 unsigned long *zholes_size)
3461 return zholes_size[zone_type];
3466 static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
3467 unsigned long *zones_size, unsigned long *zholes_size)
3469 unsigned long realtotalpages, totalpages = 0;
3472 for (i = 0; i < MAX_NR_ZONES; i++)
3473 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
3475 pgdat->node_spanned_pages = totalpages;
3477 realtotalpages = totalpages;
3478 for (i = 0; i < MAX_NR_ZONES; i++)
3480 zone_absent_pages_in_node(pgdat->node_id, i,
3482 pgdat->node_present_pages = realtotalpages;
3483 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
3487 #ifndef CONFIG_SPARSEMEM
3489 * Calculate the size of the zone->blockflags rounded to an unsigned long
3490 * Start by making sure zonesize is a multiple of pageblock_order by rounding
3491 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
3492 * round what is now in bits to nearest long in bits, then return it in
3495 static unsigned long __init usemap_size(unsigned long zonesize)
3497 unsigned long usemapsize;
3499 usemapsize = roundup(zonesize, pageblock_nr_pages);
3500 usemapsize = usemapsize >> pageblock_order;
3501 usemapsize *= NR_PAGEBLOCK_BITS;
3502 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
3504 return usemapsize / 8;
3507 static void __init setup_usemap(struct pglist_data *pgdat,
3508 struct zone *zone, unsigned long zonesize)
3510 unsigned long usemapsize = usemap_size(zonesize);
3511 zone->pageblock_flags = NULL;
3513 zone->pageblock_flags = alloc_bootmem_node(pgdat, usemapsize);
3516 static void inline setup_usemap(struct pglist_data *pgdat,
3517 struct zone *zone, unsigned long zonesize) {}
3518 #endif /* CONFIG_SPARSEMEM */
3520 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
3522 /* Return a sensible default order for the pageblock size. */
3523 static inline int pageblock_default_order(void)
3525 if (HPAGE_SHIFT > PAGE_SHIFT)
3526 return HUGETLB_PAGE_ORDER;
3531 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
3532 static inline void __init set_pageblock_order(unsigned int order)
3534 /* Check that pageblock_nr_pages has not already been setup */
3535 if (pageblock_order)
3539 * Assume the largest contiguous order of interest is a huge page.
3540 * This value may be variable depending on boot parameters on IA64
3542 pageblock_order = order;
3544 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3547 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
3548 * and pageblock_default_order() are unused as pageblock_order is set
3549 * at compile-time. See include/linux/pageblock-flags.h for the values of
3550 * pageblock_order based on the kernel config
3552 static inline int pageblock_default_order(unsigned int order)
3556 #define set_pageblock_order(x) do {} while (0)
3558 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3561 * Set up the zone data structures:
3562 * - mark all pages reserved
3563 * - mark all memory queues empty
3564 * - clear the memory bitmaps
3566 static void __paginginit free_area_init_core(struct pglist_data *pgdat,
3567 unsigned long *zones_size, unsigned long *zholes_size)
3570 int nid = pgdat->node_id;
3571 unsigned long zone_start_pfn = pgdat->node_start_pfn;
3574 pgdat_resize_init(pgdat);
3575 pgdat->nr_zones = 0;
3576 init_waitqueue_head(&pgdat->kswapd_wait);
3577 pgdat->kswapd_max_order = 0;
3578 pgdat_page_cgroup_init(pgdat);
3580 for (j = 0; j < MAX_NR_ZONES; j++) {
3581 struct zone *zone = pgdat->node_zones + j;
3582 unsigned long size, realsize, memmap_pages;
3585 size = zone_spanned_pages_in_node(nid, j, zones_size);
3586 realsize = size - zone_absent_pages_in_node(nid, j,
3590 * Adjust realsize so that it accounts for how much memory
3591 * is used by this zone for memmap. This affects the watermark
3592 * and per-cpu initialisations
3595 PAGE_ALIGN(size * sizeof(struct page)) >> PAGE_SHIFT;
3596 if (realsize >= memmap_pages) {
3597 realsize -= memmap_pages;
3600 " %s zone: %lu pages used for memmap\n",
3601 zone_names[j], memmap_pages);
3604 " %s zone: %lu pages exceeds realsize %lu\n",
3605 zone_names[j], memmap_pages, realsize);
3607 /* Account for reserved pages */
3608 if (j == 0 && realsize > dma_reserve) {
3609 realsize -= dma_reserve;
3610 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
3611 zone_names[0], dma_reserve);
3614 if (!is_highmem_idx(j))
3615 nr_kernel_pages += realsize;
3616 nr_all_pages += realsize;
3618 zone->spanned_pages = size;
3619 zone->present_pages = realsize;
3622 zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
3624 zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100;
3626 zone->name = zone_names[j];
3627 spin_lock_init(&zone->lock);
3628 spin_lock_init(&zone->lru_lock);
3629 zone_seqlock_init(zone);
3630 zone->zone_pgdat = pgdat;
3632 zone->prev_priority = DEF_PRIORITY;
3634 zone_pcp_init(zone);
3636 INIT_LIST_HEAD(&zone->lru[l].list);
3637 zone->lru[l].nr_scan = 0;
3639 zone->reclaim_stat.recent_rotated[0] = 0;
3640 zone->reclaim_stat.recent_rotated[1] = 0;
3641 zone->reclaim_stat.recent_scanned[0] = 0;
3642 zone->reclaim_stat.recent_scanned[1] = 0;
3643 zap_zone_vm_stats(zone);
3648 set_pageblock_order(pageblock_default_order());
3649 setup_usemap(pgdat, zone, size);
3650 ret = init_currently_empty_zone(zone, zone_start_pfn,
3651 size, MEMMAP_EARLY);
3653 memmap_init(size, nid, j, zone_start_pfn);
3654 zone_start_pfn += size;
3658 static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
3660 /* Skip empty nodes */
3661 if (!pgdat->node_spanned_pages)
3664 #ifdef CONFIG_FLAT_NODE_MEM_MAP
3665 /* ia64 gets its own node_mem_map, before this, without bootmem */
3666 if (!pgdat->node_mem_map) {
3667 unsigned long size, start, end;
3671 * The zone's endpoints aren't required to be MAX_ORDER
3672 * aligned but the node_mem_map endpoints must be in order
3673 * for the buddy allocator to function correctly.
3675 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
3676 end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
3677 end = ALIGN(end, MAX_ORDER_NR_PAGES);
3678 size = (end - start) * sizeof(struct page);
3679 map = alloc_remap(pgdat->node_id, size);
3681 map = alloc_bootmem_node(pgdat, size);
3682 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
3684 #ifndef CONFIG_NEED_MULTIPLE_NODES
3686 * With no DISCONTIG, the global mem_map is just set as node 0's
3688 if (pgdat == NODE_DATA(0)) {
3689 mem_map = NODE_DATA(0)->node_mem_map;
3690 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3691 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
3692 mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
3693 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3696 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
3699 void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
3700 unsigned long node_start_pfn, unsigned long *zholes_size)
3702 pg_data_t *pgdat = NODE_DATA(nid);
3704 pgdat->node_id = nid;
3705 pgdat->node_start_pfn = node_start_pfn;
3706 calculate_node_totalpages(pgdat, zones_size, zholes_size);
3708 alloc_node_mem_map(pgdat);
3709 #ifdef CONFIG_FLAT_NODE_MEM_MAP
3710 printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
3711 nid, (unsigned long)pgdat,
3712 (unsigned long)pgdat->node_mem_map);
3715 free_area_init_core(pgdat, zones_size, zholes_size);
3718 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3720 #if MAX_NUMNODES > 1
3722 * Figure out the number of possible node ids.
3724 static void __init setup_nr_node_ids(void)
3727 unsigned int highest = 0;
3729 for_each_node_mask(node, node_possible_map)
3731 nr_node_ids = highest + 1;
3734 static inline void setup_nr_node_ids(void)
3740 * add_active_range - Register a range of PFNs backed by physical memory
3741 * @nid: The node ID the range resides on
3742 * @start_pfn: The start PFN of the available physical memory
3743 * @end_pfn: The end PFN of the available physical memory
3745 * These ranges are stored in an early_node_map[] and later used by
3746 * free_area_init_nodes() to calculate zone sizes and holes. If the
3747 * range spans a memory hole, it is up to the architecture to ensure
3748 * the memory is not freed by the bootmem allocator. If possible
3749 * the range being registered will be merged with existing ranges.
3751 void __init add_active_range(unsigned int nid, unsigned long start_pfn,
3752 unsigned long end_pfn)
3756 mminit_dprintk(MMINIT_TRACE, "memory_register",
3757 "Entering add_active_range(%d, %#lx, %#lx) "
3758 "%d entries of %d used\n",
3759 nid, start_pfn, end_pfn,
3760 nr_nodemap_entries, MAX_ACTIVE_REGIONS);
3762 mminit_validate_memmodel_limits(&start_pfn, &end_pfn);
3764 /* Merge with existing active regions if possible */
3765 for (i = 0; i < nr_nodemap_entries; i++) {
3766 if (early_node_map[i].nid != nid)
3769 /* Skip if an existing region covers this new one */
3770 if (start_pfn >= early_node_map[i].start_pfn &&
3771 end_pfn <= early_node_map[i].end_pfn)
3774 /* Merge forward if suitable */
3775 if (start_pfn <= early_node_map[i].end_pfn &&
3776 end_pfn > early_node_map[i].end_pfn) {
3777 early_node_map[i].end_pfn = end_pfn;
3781 /* Merge backward if suitable */
3782 if (start_pfn < early_node_map[i].end_pfn &&
3783 end_pfn >= early_node_map[i].start_pfn) {
3784 early_node_map[i].start_pfn = start_pfn;
3789 /* Check that early_node_map is large enough */
3790 if (i >= MAX_ACTIVE_REGIONS) {
3791 printk(KERN_CRIT "More than %d memory regions, truncating\n",
3792 MAX_ACTIVE_REGIONS);
3796 early_node_map[i].nid = nid;
3797 early_node_map[i].start_pfn = start_pfn;
3798 early_node_map[i].end_pfn = end_pfn;
3799 nr_nodemap_entries = i + 1;
3803 * remove_active_range - Shrink an existing registered range of PFNs
3804 * @nid: The node id the range is on that should be shrunk
3805 * @start_pfn: The new PFN of the range
3806 * @end_pfn: The new PFN of the range
3808 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
3809 * The map is kept near the end physical page range that has already been
3810 * registered. This function allows an arch to shrink an existing registered
3813 void __init remove_active_range(unsigned int nid, unsigned long start_pfn,
3814 unsigned long end_pfn)
3819 printk(KERN_DEBUG "remove_active_range (%d, %lu, %lu)\n",
3820 nid, start_pfn, end_pfn);
3822 /* Find the old active region end and shrink */
3823 for_each_active_range_index_in_nid(i, nid) {
3824 if (early_node_map[i].start_pfn >= start_pfn &&
3825 early_node_map[i].end_pfn <= end_pfn) {
3827 early_node_map[i].start_pfn = 0;
3828 early_node_map[i].end_pfn = 0;
3832 if (early_node_map[i].start_pfn < start_pfn &&
3833 early_node_map[i].end_pfn > start_pfn) {
3834 unsigned long temp_end_pfn = early_node_map[i].end_pfn;
3835 early_node_map[i].end_pfn = start_pfn;
3836 if (temp_end_pfn > end_pfn)
3837 add_active_range(nid, end_pfn, temp_end_pfn);
3840 if (early_node_map[i].start_pfn >= start_pfn &&
3841 early_node_map[i].end_pfn > end_pfn &&
3842 early_node_map[i].start_pfn < end_pfn) {
3843 early_node_map[i].start_pfn = end_pfn;
3851 /* remove the blank ones */
3852 for (i = nr_nodemap_entries - 1; i > 0; i--) {
3853 if (early_node_map[i].nid != nid)
3855 if (early_node_map[i].end_pfn)
3857 /* we found it, get rid of it */
3858 for (j = i; j < nr_nodemap_entries - 1; j++)
3859 memcpy(&early_node_map[j], &early_node_map[j+1],
3860 sizeof(early_node_map[j]));
3861 j = nr_nodemap_entries - 1;
3862 memset(&early_node_map[j], 0, sizeof(early_node_map[j]));
3863 nr_nodemap_entries--;
3868 * remove_all_active_ranges - Remove all currently registered regions
3870 * During discovery, it may be found that a table like SRAT is invalid
3871 * and an alternative discovery method must be used. This function removes
3872 * all currently registered regions.
3874 void __init remove_all_active_ranges(void)
3876 memset(early_node_map, 0, sizeof(early_node_map));
3877 nr_nodemap_entries = 0;
3880 /* Compare two active node_active_regions */
3881 static int __init cmp_node_active_region(const void *a, const void *b)
3883 struct node_active_region *arange = (struct node_active_region *)a;
3884 struct node_active_region *brange = (struct node_active_region *)b;
3886 /* Done this way to avoid overflows */
3887 if (arange->start_pfn > brange->start_pfn)
3889 if (arange->start_pfn < brange->start_pfn)
3895 /* sort the node_map by start_pfn */
3896 static void __init sort_node_map(void)
3898 sort(early_node_map, (size_t)nr_nodemap_entries,
3899 sizeof(struct node_active_region),
3900 cmp_node_active_region, NULL);
3903 /* Find the lowest pfn for a node */
3904 static unsigned long __init find_min_pfn_for_node(int nid)
3907 unsigned long min_pfn = ULONG_MAX;
3909 /* Assuming a sorted map, the first range found has the starting pfn */
3910 for_each_active_range_index_in_nid(i, nid)
3911 min_pfn = min(min_pfn, early_node_map[i].start_pfn);
3913 if (min_pfn == ULONG_MAX) {
3915 "Could not find start_pfn for node %d\n", nid);
3923 * find_min_pfn_with_active_regions - Find the minimum PFN registered
3925 * It returns the minimum PFN based on information provided via
3926 * add_active_range().
3928 unsigned long __init find_min_pfn_with_active_regions(void)
3930 return find_min_pfn_for_node(MAX_NUMNODES);
3934 * early_calculate_totalpages()
3935 * Sum pages in active regions for movable zone.
3936 * Populate N_HIGH_MEMORY for calculating usable_nodes.
3938 static unsigned long __init early_calculate_totalpages(void)
3941 unsigned long totalpages = 0;
3943 for (i = 0; i < nr_nodemap_entries; i++) {
3944 unsigned long pages = early_node_map[i].end_pfn -
3945 early_node_map[i].start_pfn;
3946 totalpages += pages;
3948 node_set_state(early_node_map[i].nid, N_HIGH_MEMORY);
3954 * Find the PFN the Movable zone begins in each node. Kernel memory
3955 * is spread evenly between nodes as long as the nodes have enough
3956 * memory. When they don't, some nodes will have more kernelcore than
3959 static void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn)
3962 unsigned long usable_startpfn;
3963 unsigned long kernelcore_node, kernelcore_remaining;
3964 unsigned long totalpages = early_calculate_totalpages();
3965 int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]);
3968 * If movablecore was specified, calculate what size of
3969 * kernelcore that corresponds so that memory usable for
3970 * any allocation type is evenly spread. If both kernelcore
3971 * and movablecore are specified, then the value of kernelcore
3972 * will be used for required_kernelcore if it's greater than
3973 * what movablecore would have allowed.
3975 if (required_movablecore) {
3976 unsigned long corepages;
3979 * Round-up so that ZONE_MOVABLE is at least as large as what
3980 * was requested by the user
3982 required_movablecore =
3983 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
3984 corepages = totalpages - required_movablecore;
3986 required_kernelcore = max(required_kernelcore, corepages);
3989 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
3990 if (!required_kernelcore)
3993 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
3994 find_usable_zone_for_movable();
3995 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
3998 /* Spread kernelcore memory as evenly as possible throughout nodes */
3999 kernelcore_node = required_kernelcore / usable_nodes;
4000 for_each_node_state(nid, N_HIGH_MEMORY) {
4002 * Recalculate kernelcore_node if the division per node
4003 * now exceeds what is necessary to satisfy the requested
4004 * amount of memory for the kernel
4006 if (required_kernelcore < kernelcore_node)
4007 kernelcore_node = required_kernelcore / usable_nodes;
4010 * As the map is walked, we track how much memory is usable
4011 * by the kernel using kernelcore_remaining. When it is
4012 * 0, the rest of the node is usable by ZONE_MOVABLE
4014 kernelcore_remaining = kernelcore_node;
4016 /* Go through each range of PFNs within this node */
4017 for_each_active_range_index_in_nid(i, nid) {
4018 unsigned long start_pfn, end_pfn;
4019 unsigned long size_pages;
4021 start_pfn = max(early_node_map[i].start_pfn,
4022 zone_movable_pfn[nid]);
4023 end_pfn = early_node_map[i].end_pfn;
4024 if (start_pfn >= end_pfn)
4027 /* Account for what is only usable for kernelcore */
4028 if (start_pfn < usable_startpfn) {
4029 unsigned long kernel_pages;
4030 kernel_pages = min(end_pfn, usable_startpfn)
4033 kernelcore_remaining -= min(kernel_pages,
4034 kernelcore_remaining);
4035 required_kernelcore -= min(kernel_pages,
4036 required_kernelcore);
4038 /* Continue if range is now fully accounted */
4039 if (end_pfn <= usable_startpfn) {
4042 * Push zone_movable_pfn to the end so
4043 * that if we have to rebalance
4044 * kernelcore across nodes, we will
4045 * not double account here
4047 zone_movable_pfn[nid] = end_pfn;
4050 start_pfn = usable_startpfn;
4054 * The usable PFN range for ZONE_MOVABLE is from
4055 * start_pfn->end_pfn. Calculate size_pages as the
4056 * number of pages used as kernelcore
4058 size_pages = end_pfn - start_pfn;
4059 if (size_pages > kernelcore_remaining)
4060 size_pages = kernelcore_remaining;
4061 zone_movable_pfn[nid] = start_pfn + size_pages;
4064 * Some kernelcore has been met, update counts and
4065 * break if the kernelcore for this node has been
4068 required_kernelcore -= min(required_kernelcore,
4070 kernelcore_remaining -= size_pages;
4071 if (!kernelcore_remaining)
4077 * If there is still required_kernelcore, we do another pass with one
4078 * less node in the count. This will push zone_movable_pfn[nid] further
4079 * along on the nodes that still have memory until kernelcore is
4083 if (usable_nodes && required_kernelcore > usable_nodes)
4086 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
4087 for (nid = 0; nid < MAX_NUMNODES; nid++)
4088 zone_movable_pfn[nid] =
4089 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
4092 /* Any regular memory on that node ? */
4093 static void check_for_regular_memory(pg_data_t *pgdat)
4095 #ifdef CONFIG_HIGHMEM
4096 enum zone_type zone_type;
4098 for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) {
4099 struct zone *zone = &pgdat->node_zones[zone_type];
4100 if (zone->present_pages)
4101 node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY);
4107 * free_area_init_nodes - Initialise all pg_data_t and zone data
4108 * @max_zone_pfn: an array of max PFNs for each zone
4110 * This will call free_area_init_node() for each active node in the system.
4111 * Using the page ranges provided by add_active_range(), the size of each
4112 * zone in each node and their holes is calculated. If the maximum PFN
4113 * between two adjacent zones match, it is assumed that the zone is empty.
4114 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
4115 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
4116 * starts where the previous one ended. For example, ZONE_DMA32 starts
4117 * at arch_max_dma_pfn.
4119 void __init free_area_init_nodes(unsigned long *max_zone_pfn)
4124 /* Sort early_node_map as initialisation assumes it is sorted */
4127 /* Record where the zone boundaries are */
4128 memset(arch_zone_lowest_possible_pfn, 0,
4129 sizeof(arch_zone_lowest_possible_pfn));
4130 memset(arch_zone_highest_possible_pfn, 0,
4131 sizeof(arch_zone_highest_possible_pfn));
4132 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
4133 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
4134 for (i = 1; i < MAX_NR_ZONES; i++) {
4135 if (i == ZONE_MOVABLE)
4137 arch_zone_lowest_possible_pfn[i] =
4138 arch_zone_highest_possible_pfn[i-1];
4139 arch_zone_highest_possible_pfn[i] =
4140 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
4142 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
4143 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
4145 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
4146 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
4147 find_zone_movable_pfns_for_nodes(zone_movable_pfn);
4149 /* Print out the zone ranges */
4150 printk("Zone PFN ranges:\n");
4151 for (i = 0; i < MAX_NR_ZONES; i++) {
4152 if (i == ZONE_MOVABLE)
4154 printk(" %-8s %0#10lx -> %0#10lx\n",
4156 arch_zone_lowest_possible_pfn[i],
4157 arch_zone_highest_possible_pfn[i]);
4160 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
4161 printk("Movable zone start PFN for each node\n");
4162 for (i = 0; i < MAX_NUMNODES; i++) {
4163 if (zone_movable_pfn[i])
4164 printk(" Node %d: %lu\n", i, zone_movable_pfn[i]);
4167 /* Print out the early_node_map[] */
4168 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries);
4169 for (i = 0; i < nr_nodemap_entries; i++)
4170 printk(" %3d: %0#10lx -> %0#10lx\n", early_node_map[i].nid,
4171 early_node_map[i].start_pfn,
4172 early_node_map[i].end_pfn);
4174 /* Initialise every node */
4175 mminit_verify_pageflags_layout();
4176 setup_nr_node_ids();
4177 for_each_online_node(nid) {
4178 pg_data_t *pgdat = NODE_DATA(nid);
4179 free_area_init_node(nid, NULL,
4180 find_min_pfn_for_node(nid), NULL);
4182 /* Any memory on that node */
4183 if (pgdat->node_present_pages)
4184 node_set_state(nid, N_HIGH_MEMORY);
4185 check_for_regular_memory(pgdat);
4189 static int __init cmdline_parse_core(char *p, unsigned long *core)
4191 unsigned long long coremem;
4195 coremem = memparse(p, &p);
4196 *core = coremem >> PAGE_SHIFT;
4198 /* Paranoid check that UL is enough for the coremem value */
4199 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
4205 * kernelcore=size sets the amount of memory for use for allocations that
4206 * cannot be reclaimed or migrated.
4208 static int __init cmdline_parse_kernelcore(char *p)
4210 return cmdline_parse_core(p, &required_kernelcore);
4214 * movablecore=size sets the amount of memory for use for allocations that
4215 * can be reclaimed or migrated.
4217 static int __init cmdline_parse_movablecore(char *p)
4219 return cmdline_parse_core(p, &required_movablecore);
4222 early_param("kernelcore", cmdline_parse_kernelcore);
4223 early_param("movablecore", cmdline_parse_movablecore);
4225 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4228 * set_dma_reserve - set the specified number of pages reserved in the first zone
4229 * @new_dma_reserve: The number of pages to mark reserved
4231 * The per-cpu batchsize and zone watermarks are determined by present_pages.
4232 * In the DMA zone, a significant percentage may be consumed by kernel image
4233 * and other unfreeable allocations which can skew the watermarks badly. This
4234 * function may optionally be used to account for unfreeable pages in the
4235 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
4236 * smaller per-cpu batchsize.
4238 void __init set_dma_reserve(unsigned long new_dma_reserve)
4240 dma_reserve = new_dma_reserve;
4243 #ifndef CONFIG_NEED_MULTIPLE_NODES
4244 struct pglist_data __refdata contig_page_data = { .bdata = &bootmem_node_data[0] };
4245 EXPORT_SYMBOL(contig_page_data);
4248 void __init free_area_init(unsigned long *zones_size)
4250 free_area_init_node(0, zones_size,
4251 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
4254 static int page_alloc_cpu_notify(struct notifier_block *self,
4255 unsigned long action, void *hcpu)
4257 int cpu = (unsigned long)hcpu;
4259 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
4263 * Spill the event counters of the dead processor
4264 * into the current processors event counters.
4265 * This artificially elevates the count of the current
4268 vm_events_fold_cpu(cpu);
4271 * Zero the differential counters of the dead processor
4272 * so that the vm statistics are consistent.
4274 * This is only okay since the processor is dead and cannot
4275 * race with what we are doing.
4277 refresh_cpu_vm_stats(cpu);
4282 void __init page_alloc_init(void)
4284 hotcpu_notifier(page_alloc_cpu_notify, 0);
4288 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
4289 * or min_free_kbytes changes.
4291 static void calculate_totalreserve_pages(void)
4293 struct pglist_data *pgdat;
4294 unsigned long reserve_pages = 0;
4295 enum zone_type i, j;
4297 for_each_online_pgdat(pgdat) {
4298 for (i = 0; i < MAX_NR_ZONES; i++) {
4299 struct zone *zone = pgdat->node_zones + i;
4300 unsigned long max = 0;
4302 /* Find valid and maximum lowmem_reserve in the zone */
4303 for (j = i; j < MAX_NR_ZONES; j++) {
4304 if (zone->lowmem_reserve[j] > max)
4305 max = zone->lowmem_reserve[j];
4308 /* we treat pages_high as reserved pages. */
4309 max += zone->pages_high;
4311 if (max > zone->present_pages)
4312 max = zone->present_pages;
4313 reserve_pages += max;
4316 totalreserve_pages = reserve_pages;
4320 * setup_per_zone_lowmem_reserve - called whenever
4321 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
4322 * has a correct pages reserved value, so an adequate number of
4323 * pages are left in the zone after a successful __alloc_pages().
4325 static void setup_per_zone_lowmem_reserve(void)
4327 struct pglist_data *pgdat;
4328 enum zone_type j, idx;
4330 for_each_online_pgdat(pgdat) {
4331 for (j = 0; j < MAX_NR_ZONES; j++) {
4332 struct zone *zone = pgdat->node_zones + j;
4333 unsigned long present_pages = zone->present_pages;
4335 zone->lowmem_reserve[j] = 0;
4339 struct zone *lower_zone;
4343 if (sysctl_lowmem_reserve_ratio[idx] < 1)
4344 sysctl_lowmem_reserve_ratio[idx] = 1;
4346 lower_zone = pgdat->node_zones + idx;
4347 lower_zone->lowmem_reserve[j] = present_pages /
4348 sysctl_lowmem_reserve_ratio[idx];
4349 present_pages += lower_zone->present_pages;
4354 /* update totalreserve_pages */
4355 calculate_totalreserve_pages();
4359 * setup_per_zone_pages_min - called when min_free_kbytes changes.
4361 * Ensures that the pages_{min,low,high} values for each zone are set correctly
4362 * with respect to min_free_kbytes.
4364 void setup_per_zone_pages_min(void)
4366 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
4367 unsigned long lowmem_pages = 0;
4369 unsigned long flags;
4371 /* Calculate total number of !ZONE_HIGHMEM pages */
4372 for_each_zone(zone) {
4373 if (!is_highmem(zone))
4374 lowmem_pages += zone->present_pages;
4377 for_each_zone(zone) {
4380 spin_lock_irqsave(&zone->lock, flags);
4381 tmp = (u64)pages_min * zone->present_pages;
4382 do_div(tmp, lowmem_pages);
4383 if (is_highmem(zone)) {
4385 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
4386 * need highmem pages, so cap pages_min to a small
4389 * The (pages_high-pages_low) and (pages_low-pages_min)
4390 * deltas controls asynch page reclaim, and so should
4391 * not be capped for highmem.
4395 min_pages = zone->present_pages / 1024;
4396 if (min_pages < SWAP_CLUSTER_MAX)
4397 min_pages = SWAP_CLUSTER_MAX;
4398 if (min_pages > 128)
4400 zone->pages_min = min_pages;
4403 * If it's a lowmem zone, reserve a number of pages
4404 * proportionate to the zone's size.
4406 zone->pages_min = tmp;
4409 zone->pages_low = zone->pages_min + (tmp >> 2);
4410 zone->pages_high = zone->pages_min + (tmp >> 1);
4411 setup_zone_migrate_reserve(zone);
4412 spin_unlock_irqrestore(&zone->lock, flags);
4415 /* update totalreserve_pages */
4416 calculate_totalreserve_pages();
4420 * setup_per_zone_inactive_ratio - called when min_free_kbytes changes.
4422 * The inactive anon list should be small enough that the VM never has to
4423 * do too much work, but large enough that each inactive page has a chance
4424 * to be referenced again before it is swapped out.
4426 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
4427 * INACTIVE_ANON pages on this zone's LRU, maintained by the
4428 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
4429 * the anonymous pages are kept on the inactive list.
4432 * memory ratio inactive anon
4433 * -------------------------------------
4442 static void setup_per_zone_inactive_ratio(void)
4446 for_each_zone(zone) {
4447 unsigned int gb, ratio;
4449 /* Zone size in gigabytes */
4450 gb = zone->present_pages >> (30 - PAGE_SHIFT);
4451 ratio = int_sqrt(10 * gb);
4455 zone->inactive_ratio = ratio;
4460 * Initialise min_free_kbytes.
4462 * For small machines we want it small (128k min). For large machines
4463 * we want it large (64MB max). But it is not linear, because network
4464 * bandwidth does not increase linearly with machine size. We use
4466 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
4467 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
4483 static int __init init_per_zone_pages_min(void)
4485 unsigned long lowmem_kbytes;
4487 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
4489 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
4490 if (min_free_kbytes < 128)
4491 min_free_kbytes = 128;
4492 if (min_free_kbytes > 65536)
4493 min_free_kbytes = 65536;
4494 setup_per_zone_pages_min();
4495 setup_per_zone_lowmem_reserve();
4496 setup_per_zone_inactive_ratio();
4499 module_init(init_per_zone_pages_min)
4502 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
4503 * that we can call two helper functions whenever min_free_kbytes
4506 int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
4507 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4509 proc_dointvec(table, write, file, buffer, length, ppos);
4511 setup_per_zone_pages_min();
4516 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
4517 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4522 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4527 zone->min_unmapped_pages = (zone->present_pages *
4528 sysctl_min_unmapped_ratio) / 100;
4532 int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
4533 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4538 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4543 zone->min_slab_pages = (zone->present_pages *
4544 sysctl_min_slab_ratio) / 100;
4550 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
4551 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
4552 * whenever sysctl_lowmem_reserve_ratio changes.
4554 * The reserve ratio obviously has absolutely no relation with the
4555 * pages_min watermarks. The lowmem reserve ratio can only make sense
4556 * if in function of the boot time zone sizes.
4558 int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
4559 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4561 proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4562 setup_per_zone_lowmem_reserve();
4567 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
4568 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
4569 * can have before it gets flushed back to buddy allocator.
4572 int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
4573 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4579 ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4580 if (!write || (ret == -EINVAL))
4582 for_each_zone(zone) {
4583 for_each_online_cpu(cpu) {
4585 high = zone->present_pages / percpu_pagelist_fraction;
4586 setup_pagelist_highmark(zone_pcp(zone, cpu), high);
4592 int hashdist = HASHDIST_DEFAULT;
4595 static int __init set_hashdist(char *str)
4599 hashdist = simple_strtoul(str, &str, 0);
4602 __setup("hashdist=", set_hashdist);
4606 * allocate a large system hash table from bootmem
4607 * - it is assumed that the hash table must contain an exact power-of-2
4608 * quantity of entries
4609 * - limit is the number of hash buckets, not the total allocation size
4611 void *__init alloc_large_system_hash(const char *tablename,
4612 unsigned long bucketsize,
4613 unsigned long numentries,
4616 unsigned int *_hash_shift,
4617 unsigned int *_hash_mask,
4618 unsigned long limit)
4620 unsigned long long max = limit;
4621 unsigned long log2qty, size;
4624 /* allow the kernel cmdline to have a say */
4626 /* round applicable memory size up to nearest megabyte */
4627 numentries = nr_kernel_pages;
4628 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
4629 numentries >>= 20 - PAGE_SHIFT;
4630 numentries <<= 20 - PAGE_SHIFT;
4632 /* limit to 1 bucket per 2^scale bytes of low memory */
4633 if (scale > PAGE_SHIFT)
4634 numentries >>= (scale - PAGE_SHIFT);
4636 numentries <<= (PAGE_SHIFT - scale);
4638 /* Make sure we've got at least a 0-order allocation.. */
4639 if (unlikely((numentries * bucketsize) < PAGE_SIZE))
4640 numentries = PAGE_SIZE / bucketsize;
4642 numentries = roundup_pow_of_two(numentries);
4644 /* limit allocation size to 1/16 total memory by default */
4646 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
4647 do_div(max, bucketsize);
4650 if (numentries > max)
4653 log2qty = ilog2(numentries);
4656 size = bucketsize << log2qty;
4657 if (flags & HASH_EARLY)
4658 table = alloc_bootmem_nopanic(size);
4660 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
4662 unsigned long order = get_order(size);
4664 if (order < MAX_ORDER)
4665 table = (void *)__get_free_pages(GFP_ATOMIC,
4668 * If bucketsize is not a power-of-two, we may free
4669 * some pages at the end of hash table.
4672 unsigned long alloc_end = (unsigned long)table +
4673 (PAGE_SIZE << order);
4674 unsigned long used = (unsigned long)table +
4676 split_page(virt_to_page(table), order);
4677 while (used < alloc_end) {
4683 } while (!table && size > PAGE_SIZE && --log2qty);
4686 panic("Failed to allocate %s hash table\n", tablename);
4688 printk(KERN_INFO "%s hash table entries: %d (order: %d, %lu bytes)\n",
4691 ilog2(size) - PAGE_SHIFT,
4695 *_hash_shift = log2qty;
4697 *_hash_mask = (1 << log2qty) - 1;
4700 * If hashdist is set, the table allocation is done with __vmalloc()
4701 * which invokes the kmemleak_alloc() callback. This function may also
4702 * be called before the slab and kmemleak are initialised when
4703 * kmemleak simply buffers the request to be executed later
4704 * (GFP_ATOMIC flag ignored in this case).
4707 kmemleak_alloc(table, size, 1, GFP_ATOMIC);
4712 /* Return a pointer to the bitmap storing bits affecting a block of pages */
4713 static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
4716 #ifdef CONFIG_SPARSEMEM
4717 return __pfn_to_section(pfn)->pageblock_flags;
4719 return zone->pageblock_flags;
4720 #endif /* CONFIG_SPARSEMEM */
4723 static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
4725 #ifdef CONFIG_SPARSEMEM
4726 pfn &= (PAGES_PER_SECTION-1);
4727 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
4729 pfn = pfn - zone->zone_start_pfn;
4730 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
4731 #endif /* CONFIG_SPARSEMEM */
4735 * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
4736 * @page: The page within the block of interest
4737 * @start_bitidx: The first bit of interest to retrieve
4738 * @end_bitidx: The last bit of interest
4739 * returns pageblock_bits flags
4741 unsigned long get_pageblock_flags_group(struct page *page,
4742 int start_bitidx, int end_bitidx)
4745 unsigned long *bitmap;
4746 unsigned long pfn, bitidx;
4747 unsigned long flags = 0;
4748 unsigned long value = 1;
4750 zone = page_zone(page);
4751 pfn = page_to_pfn(page);
4752 bitmap = get_pageblock_bitmap(zone, pfn);
4753 bitidx = pfn_to_bitidx(zone, pfn);
4755 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
4756 if (test_bit(bitidx + start_bitidx, bitmap))
4763 * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
4764 * @page: The page within the block of interest
4765 * @start_bitidx: The first bit of interest
4766 * @end_bitidx: The last bit of interest
4767 * @flags: The flags to set
4769 void set_pageblock_flags_group(struct page *page, unsigned long flags,
4770 int start_bitidx, int end_bitidx)
4773 unsigned long *bitmap;
4774 unsigned long pfn, bitidx;
4775 unsigned long value = 1;
4777 zone = page_zone(page);
4778 pfn = page_to_pfn(page);
4779 bitmap = get_pageblock_bitmap(zone, pfn);
4780 bitidx = pfn_to_bitidx(zone, pfn);
4781 VM_BUG_ON(pfn < zone->zone_start_pfn);
4782 VM_BUG_ON(pfn >= zone->zone_start_pfn + zone->spanned_pages);
4784 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
4786 __set_bit(bitidx + start_bitidx, bitmap);
4788 __clear_bit(bitidx + start_bitidx, bitmap);
4792 * This is designed as sub function...plz see page_isolation.c also.
4793 * set/clear page block's type to be ISOLATE.
4794 * page allocater never alloc memory from ISOLATE block.
4797 int set_migratetype_isolate(struct page *page)
4800 unsigned long flags;
4803 zone = page_zone(page);
4804 spin_lock_irqsave(&zone->lock, flags);
4806 * In future, more migrate types will be able to be isolation target.
4808 if (get_pageblock_migratetype(page) != MIGRATE_MOVABLE)
4810 set_pageblock_migratetype(page, MIGRATE_ISOLATE);
4811 move_freepages_block(zone, page, MIGRATE_ISOLATE);
4814 spin_unlock_irqrestore(&zone->lock, flags);
4820 void unset_migratetype_isolate(struct page *page)
4823 unsigned long flags;
4824 zone = page_zone(page);
4825 spin_lock_irqsave(&zone->lock, flags);
4826 if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE)
4828 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
4829 move_freepages_block(zone, page, MIGRATE_MOVABLE);
4831 spin_unlock_irqrestore(&zone->lock, flags);
4834 #ifdef CONFIG_MEMORY_HOTREMOVE
4836 * All pages in the range must be isolated before calling this.
4839 __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
4845 unsigned long flags;
4846 /* find the first valid pfn */
4847 for (pfn = start_pfn; pfn < end_pfn; pfn++)
4852 zone = page_zone(pfn_to_page(pfn));
4853 spin_lock_irqsave(&zone->lock, flags);
4855 while (pfn < end_pfn) {
4856 if (!pfn_valid(pfn)) {
4860 page = pfn_to_page(pfn);
4861 BUG_ON(page_count(page));
4862 BUG_ON(!PageBuddy(page));
4863 order = page_order(page);
4864 #ifdef CONFIG_DEBUG_VM
4865 printk(KERN_INFO "remove from free list %lx %d %lx\n",
4866 pfn, 1 << order, end_pfn);
4868 list_del(&page->lru);
4869 rmv_page_order(page);
4870 zone->free_area[order].nr_free--;
4871 __mod_zone_page_state(zone, NR_FREE_PAGES,
4873 for (i = 0; i < (1 << order); i++)
4874 SetPageReserved((page+i));
4875 pfn += (1 << order);
4877 spin_unlock_irqrestore(&zone->lock, flags);