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,
458 unsigned long page_idx;
459 int order_size = 1 << order;
461 if (unlikely(PageCompound(page)))
462 if (unlikely(destroy_compound_page(page, order)))
465 VM_BUG_ON(migratetype == -1);
467 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
469 VM_BUG_ON(page_idx & (order_size - 1));
470 VM_BUG_ON(bad_range(zone, page));
472 __mod_zone_page_state(zone, NR_FREE_PAGES, order_size);
473 while (order < MAX_ORDER-1) {
474 unsigned long combined_idx;
477 buddy = __page_find_buddy(page, page_idx, order);
478 if (!page_is_buddy(page, buddy, order))
481 /* Our buddy is free, merge with it and move up one order. */
482 list_del(&buddy->lru);
483 zone->free_area[order].nr_free--;
484 rmv_page_order(buddy);
485 combined_idx = __find_combined_index(page_idx, order);
486 page = page + (combined_idx - page_idx);
487 page_idx = combined_idx;
490 set_page_order(page, order);
492 &zone->free_area[order].free_list[migratetype]);
493 zone->free_area[order].nr_free++;
496 static inline int free_pages_check(struct page *page)
498 if (unlikely(page_mapcount(page) |
499 (page->mapping != NULL) |
500 (page_count(page) != 0) |
501 (page->flags & PAGE_FLAGS_CHECK_AT_FREE))) {
505 if (page->flags & PAGE_FLAGS_CHECK_AT_PREP)
506 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
511 * Frees a list of pages.
512 * Assumes all pages on list are in same zone, and of same order.
513 * count is the number of pages to free.
515 * If the zone was previously in an "all pages pinned" state then look to
516 * see if this freeing clears that state.
518 * And clear the zone's pages_scanned counter, to hold off the "all pages are
519 * pinned" detection logic.
521 static void free_pages_bulk(struct zone *zone, int count,
522 struct list_head *list, int order)
524 spin_lock(&zone->lock);
525 zone_clear_flag(zone, ZONE_ALL_UNRECLAIMABLE);
526 zone->pages_scanned = 0;
530 VM_BUG_ON(list_empty(list));
531 page = list_entry(list->prev, struct page, lru);
532 /* have to delete it as __free_one_page list manipulates */
533 list_del(&page->lru);
534 __free_one_page(page, zone, order, page_private(page));
536 spin_unlock(&zone->lock);
539 static void free_one_page(struct zone *zone, struct page *page, int order,
542 spin_lock(&zone->lock);
543 zone_clear_flag(zone, ZONE_ALL_UNRECLAIMABLE);
544 zone->pages_scanned = 0;
545 __free_one_page(page, zone, order, migratetype);
546 spin_unlock(&zone->lock);
549 static void __free_pages_ok(struct page *page, unsigned int order)
554 int clearMlocked = PageMlocked(page);
556 for (i = 0 ; i < (1 << order) ; ++i)
557 bad += free_pages_check(page + i);
561 if (!PageHighMem(page)) {
562 debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order);
563 debug_check_no_obj_freed(page_address(page),
566 arch_free_page(page, order);
567 kernel_map_pages(page, 1 << order, 0);
569 local_irq_save(flags);
570 if (unlikely(clearMlocked))
571 free_page_mlock(page);
572 __count_vm_events(PGFREE, 1 << order);
573 free_one_page(page_zone(page), page, order,
574 get_pageblock_migratetype(page));
575 local_irq_restore(flags);
579 * permit the bootmem allocator to evade page validation on high-order frees
581 void __meminit __free_pages_bootmem(struct page *page, unsigned int order)
584 __ClearPageReserved(page);
585 set_page_count(page, 0);
586 set_page_refcounted(page);
592 for (loop = 0; loop < BITS_PER_LONG; loop++) {
593 struct page *p = &page[loop];
595 if (loop + 1 < BITS_PER_LONG)
597 __ClearPageReserved(p);
598 set_page_count(p, 0);
601 set_page_refcounted(page);
602 __free_pages(page, order);
608 * The order of subdivision here is critical for the IO subsystem.
609 * Please do not alter this order without good reasons and regression
610 * testing. Specifically, as large blocks of memory are subdivided,
611 * the order in which smaller blocks are delivered depends on the order
612 * they're subdivided in this function. This is the primary factor
613 * influencing the order in which pages are delivered to the IO
614 * subsystem according to empirical testing, and this is also justified
615 * by considering the behavior of a buddy system containing a single
616 * large block of memory acted on by a series of small allocations.
617 * This behavior is a critical factor in sglist merging's success.
621 static inline void expand(struct zone *zone, struct page *page,
622 int low, int high, struct free_area *area,
625 unsigned long size = 1 << high;
631 VM_BUG_ON(bad_range(zone, &page[size]));
632 list_add(&page[size].lru, &area->free_list[migratetype]);
634 set_page_order(&page[size], high);
639 * This page is about to be returned from the page allocator
641 static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
643 if (unlikely(page_mapcount(page) |
644 (page->mapping != NULL) |
645 (page_count(page) != 0) |
646 (page->flags & PAGE_FLAGS_CHECK_AT_PREP))) {
651 set_page_private(page, 0);
652 set_page_refcounted(page);
654 arch_alloc_page(page, order);
655 kernel_map_pages(page, 1 << order, 1);
657 if (gfp_flags & __GFP_ZERO)
658 prep_zero_page(page, order, gfp_flags);
660 if (order && (gfp_flags & __GFP_COMP))
661 prep_compound_page(page, order);
667 * Go through the free lists for the given migratetype and remove
668 * the smallest available page from the freelists
671 struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
674 unsigned int current_order;
675 struct free_area * area;
678 /* Find a page of the appropriate size in the preferred list */
679 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
680 area = &(zone->free_area[current_order]);
681 if (list_empty(&area->free_list[migratetype]))
684 page = list_entry(area->free_list[migratetype].next,
686 list_del(&page->lru);
687 rmv_page_order(page);
689 __mod_zone_page_state(zone, NR_FREE_PAGES, - (1UL << order));
690 expand(zone, page, order, current_order, area, migratetype);
699 * This array describes the order lists are fallen back to when
700 * the free lists for the desirable migrate type are depleted
702 static int fallbacks[MIGRATE_TYPES][MIGRATE_TYPES-1] = {
703 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
704 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
705 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
706 [MIGRATE_RESERVE] = { MIGRATE_RESERVE, MIGRATE_RESERVE, MIGRATE_RESERVE }, /* Never used */
710 * Move the free pages in a range to the free lists of the requested type.
711 * Note that start_page and end_pages are not aligned on a pageblock
712 * boundary. If alignment is required, use move_freepages_block()
714 static int move_freepages(struct zone *zone,
715 struct page *start_page, struct page *end_page,
722 #ifndef CONFIG_HOLES_IN_ZONE
724 * page_zone is not safe to call in this context when
725 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
726 * anyway as we check zone boundaries in move_freepages_block().
727 * Remove at a later date when no bug reports exist related to
728 * grouping pages by mobility
730 BUG_ON(page_zone(start_page) != page_zone(end_page));
733 for (page = start_page; page <= end_page;) {
734 /* Make sure we are not inadvertently changing nodes */
735 VM_BUG_ON(page_to_nid(page) != zone_to_nid(zone));
737 if (!pfn_valid_within(page_to_pfn(page))) {
742 if (!PageBuddy(page)) {
747 order = page_order(page);
748 list_del(&page->lru);
750 &zone->free_area[order].free_list[migratetype]);
752 pages_moved += 1 << order;
758 static int move_freepages_block(struct zone *zone, struct page *page,
761 unsigned long start_pfn, end_pfn;
762 struct page *start_page, *end_page;
764 start_pfn = page_to_pfn(page);
765 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
766 start_page = pfn_to_page(start_pfn);
767 end_page = start_page + pageblock_nr_pages - 1;
768 end_pfn = start_pfn + pageblock_nr_pages - 1;
770 /* Do not cross zone boundaries */
771 if (start_pfn < zone->zone_start_pfn)
773 if (end_pfn >= zone->zone_start_pfn + zone->spanned_pages)
776 return move_freepages(zone, start_page, end_page, migratetype);
779 /* Remove an element from the buddy allocator from the fallback list */
780 static inline struct page *
781 __rmqueue_fallback(struct zone *zone, int order, int start_migratetype)
783 struct free_area * area;
788 /* Find the largest possible block of pages in the other list */
789 for (current_order = MAX_ORDER-1; current_order >= order;
791 for (i = 0; i < MIGRATE_TYPES - 1; i++) {
792 migratetype = fallbacks[start_migratetype][i];
794 /* MIGRATE_RESERVE handled later if necessary */
795 if (migratetype == MIGRATE_RESERVE)
798 area = &(zone->free_area[current_order]);
799 if (list_empty(&area->free_list[migratetype]))
802 page = list_entry(area->free_list[migratetype].next,
807 * If breaking a large block of pages, move all free
808 * pages to the preferred allocation list. If falling
809 * back for a reclaimable kernel allocation, be more
810 * agressive about taking ownership of free pages
812 if (unlikely(current_order >= (pageblock_order >> 1)) ||
813 start_migratetype == MIGRATE_RECLAIMABLE) {
815 pages = move_freepages_block(zone, page,
818 /* Claim the whole block if over half of it is free */
819 if (pages >= (1 << (pageblock_order-1)))
820 set_pageblock_migratetype(page,
823 migratetype = start_migratetype;
826 /* Remove the page from the freelists */
827 list_del(&page->lru);
828 rmv_page_order(page);
829 __mod_zone_page_state(zone, NR_FREE_PAGES,
832 if (current_order == pageblock_order)
833 set_pageblock_migratetype(page,
836 expand(zone, page, order, current_order, area, migratetype);
845 * Do the hard work of removing an element from the buddy allocator.
846 * Call me with the zone->lock already held.
848 static struct page *__rmqueue(struct zone *zone, unsigned int order,
854 page = __rmqueue_smallest(zone, order, migratetype);
856 if (unlikely(!page) && migratetype != MIGRATE_RESERVE) {
857 page = __rmqueue_fallback(zone, order, migratetype);
860 * Use MIGRATE_RESERVE rather than fail an allocation. goto
861 * is used because __rmqueue_smallest is an inline function
862 * and we want just one call site
865 migratetype = MIGRATE_RESERVE;
874 * Obtain a specified number of elements from the buddy allocator, all under
875 * a single hold of the lock, for efficiency. Add them to the supplied list.
876 * Returns the number of new pages which were placed at *list.
878 static int rmqueue_bulk(struct zone *zone, unsigned int order,
879 unsigned long count, struct list_head *list,
884 spin_lock(&zone->lock);
885 for (i = 0; i < count; ++i) {
886 struct page *page = __rmqueue(zone, order, migratetype);
887 if (unlikely(page == NULL))
891 * Split buddy pages returned by expand() are received here
892 * in physical page order. The page is added to the callers and
893 * list and the list head then moves forward. From the callers
894 * perspective, the linked list is ordered by page number in
895 * some conditions. This is useful for IO devices that can
896 * merge IO requests if the physical pages are ordered
899 list_add(&page->lru, list);
900 set_page_private(page, migratetype);
903 spin_unlock(&zone->lock);
909 * Called from the vmstat counter updater to drain pagesets of this
910 * currently executing processor on remote nodes after they have
913 * Note that this function must be called with the thread pinned to
914 * a single processor.
916 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
921 local_irq_save(flags);
922 if (pcp->count >= pcp->batch)
923 to_drain = pcp->batch;
925 to_drain = pcp->count;
926 free_pages_bulk(zone, to_drain, &pcp->list, 0);
927 pcp->count -= to_drain;
928 local_irq_restore(flags);
933 * Drain pages of the indicated processor.
935 * The processor must either be the current processor and the
936 * thread pinned to the current processor or a processor that
939 static void drain_pages(unsigned int cpu)
944 for_each_populated_zone(zone) {
945 struct per_cpu_pageset *pset;
946 struct per_cpu_pages *pcp;
948 pset = zone_pcp(zone, cpu);
951 local_irq_save(flags);
952 free_pages_bulk(zone, pcp->count, &pcp->list, 0);
954 local_irq_restore(flags);
959 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
961 void drain_local_pages(void *arg)
963 drain_pages(smp_processor_id());
967 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
969 void drain_all_pages(void)
971 on_each_cpu(drain_local_pages, NULL, 1);
974 #ifdef CONFIG_HIBERNATION
976 void mark_free_pages(struct zone *zone)
978 unsigned long pfn, max_zone_pfn;
981 struct list_head *curr;
983 if (!zone->spanned_pages)
986 spin_lock_irqsave(&zone->lock, flags);
988 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
989 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
990 if (pfn_valid(pfn)) {
991 struct page *page = pfn_to_page(pfn);
993 if (!swsusp_page_is_forbidden(page))
994 swsusp_unset_page_free(page);
997 for_each_migratetype_order(order, t) {
998 list_for_each(curr, &zone->free_area[order].free_list[t]) {
1001 pfn = page_to_pfn(list_entry(curr, struct page, lru));
1002 for (i = 0; i < (1UL << order); i++)
1003 swsusp_set_page_free(pfn_to_page(pfn + i));
1006 spin_unlock_irqrestore(&zone->lock, flags);
1008 #endif /* CONFIG_PM */
1011 * Free a 0-order page
1013 static void free_hot_cold_page(struct page *page, int cold)
1015 struct zone *zone = page_zone(page);
1016 struct per_cpu_pages *pcp;
1017 unsigned long flags;
1018 int clearMlocked = PageMlocked(page);
1021 page->mapping = NULL;
1022 if (free_pages_check(page))
1025 if (!PageHighMem(page)) {
1026 debug_check_no_locks_freed(page_address(page), PAGE_SIZE);
1027 debug_check_no_obj_freed(page_address(page), PAGE_SIZE);
1029 arch_free_page(page, 0);
1030 kernel_map_pages(page, 1, 0);
1032 pcp = &zone_pcp(zone, get_cpu())->pcp;
1033 local_irq_save(flags);
1034 if (unlikely(clearMlocked))
1035 free_page_mlock(page);
1036 __count_vm_event(PGFREE);
1039 list_add_tail(&page->lru, &pcp->list);
1041 list_add(&page->lru, &pcp->list);
1042 set_page_private(page, get_pageblock_migratetype(page));
1044 if (pcp->count >= pcp->high) {
1045 free_pages_bulk(zone, pcp->batch, &pcp->list, 0);
1046 pcp->count -= pcp->batch;
1048 local_irq_restore(flags);
1052 void free_hot_page(struct page *page)
1054 free_hot_cold_page(page, 0);
1057 void free_cold_page(struct page *page)
1059 free_hot_cold_page(page, 1);
1063 * split_page takes a non-compound higher-order page, and splits it into
1064 * n (1<<order) sub-pages: page[0..n]
1065 * Each sub-page must be freed individually.
1067 * Note: this is probably too low level an operation for use in drivers.
1068 * Please consult with lkml before using this in your driver.
1070 void split_page(struct page *page, unsigned int order)
1074 VM_BUG_ON(PageCompound(page));
1075 VM_BUG_ON(!page_count(page));
1076 for (i = 1; i < (1 << order); i++)
1077 set_page_refcounted(page + i);
1081 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1082 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1086 struct page *buffered_rmqueue(struct zone *preferred_zone,
1087 struct zone *zone, int order, gfp_t gfp_flags,
1090 unsigned long flags;
1092 int cold = !!(gfp_flags & __GFP_COLD);
1097 if (likely(order == 0)) {
1098 struct per_cpu_pages *pcp;
1100 pcp = &zone_pcp(zone, cpu)->pcp;
1101 local_irq_save(flags);
1103 pcp->count = rmqueue_bulk(zone, 0,
1104 pcp->batch, &pcp->list, migratetype);
1105 if (unlikely(!pcp->count))
1109 /* Find a page of the appropriate migrate type */
1111 list_for_each_entry_reverse(page, &pcp->list, lru)
1112 if (page_private(page) == migratetype)
1115 list_for_each_entry(page, &pcp->list, lru)
1116 if (page_private(page) == migratetype)
1120 /* Allocate more to the pcp list if necessary */
1121 if (unlikely(&page->lru == &pcp->list)) {
1122 pcp->count += rmqueue_bulk(zone, 0,
1123 pcp->batch, &pcp->list, migratetype);
1124 page = list_entry(pcp->list.next, struct page, lru);
1127 list_del(&page->lru);
1130 spin_lock_irqsave(&zone->lock, flags);
1131 page = __rmqueue(zone, order, migratetype);
1132 spin_unlock(&zone->lock);
1137 __count_zone_vm_events(PGALLOC, zone, 1 << order);
1138 zone_statistics(preferred_zone, zone);
1139 local_irq_restore(flags);
1142 VM_BUG_ON(bad_range(zone, page));
1143 if (prep_new_page(page, order, gfp_flags))
1148 local_irq_restore(flags);
1153 #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
1154 #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
1155 #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
1156 #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
1157 #define ALLOC_HARDER 0x10 /* try to alloc harder */
1158 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1159 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
1161 #ifdef CONFIG_FAIL_PAGE_ALLOC
1163 static struct fail_page_alloc_attr {
1164 struct fault_attr attr;
1166 u32 ignore_gfp_highmem;
1167 u32 ignore_gfp_wait;
1170 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1172 struct dentry *ignore_gfp_highmem_file;
1173 struct dentry *ignore_gfp_wait_file;
1174 struct dentry *min_order_file;
1176 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1178 } fail_page_alloc = {
1179 .attr = FAULT_ATTR_INITIALIZER,
1180 .ignore_gfp_wait = 1,
1181 .ignore_gfp_highmem = 1,
1185 static int __init setup_fail_page_alloc(char *str)
1187 return setup_fault_attr(&fail_page_alloc.attr, str);
1189 __setup("fail_page_alloc=", setup_fail_page_alloc);
1191 static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1193 if (order < fail_page_alloc.min_order)
1195 if (gfp_mask & __GFP_NOFAIL)
1197 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
1199 if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
1202 return should_fail(&fail_page_alloc.attr, 1 << order);
1205 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1207 static int __init fail_page_alloc_debugfs(void)
1209 mode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
1213 err = init_fault_attr_dentries(&fail_page_alloc.attr,
1217 dir = fail_page_alloc.attr.dentries.dir;
1219 fail_page_alloc.ignore_gfp_wait_file =
1220 debugfs_create_bool("ignore-gfp-wait", mode, dir,
1221 &fail_page_alloc.ignore_gfp_wait);
1223 fail_page_alloc.ignore_gfp_highmem_file =
1224 debugfs_create_bool("ignore-gfp-highmem", mode, dir,
1225 &fail_page_alloc.ignore_gfp_highmem);
1226 fail_page_alloc.min_order_file =
1227 debugfs_create_u32("min-order", mode, dir,
1228 &fail_page_alloc.min_order);
1230 if (!fail_page_alloc.ignore_gfp_wait_file ||
1231 !fail_page_alloc.ignore_gfp_highmem_file ||
1232 !fail_page_alloc.min_order_file) {
1234 debugfs_remove(fail_page_alloc.ignore_gfp_wait_file);
1235 debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file);
1236 debugfs_remove(fail_page_alloc.min_order_file);
1237 cleanup_fault_attr_dentries(&fail_page_alloc.attr);
1243 late_initcall(fail_page_alloc_debugfs);
1245 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1247 #else /* CONFIG_FAIL_PAGE_ALLOC */
1249 static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1254 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1257 * Return 1 if free pages are above 'mark'. This takes into account the order
1258 * of the allocation.
1260 int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1261 int classzone_idx, int alloc_flags)
1263 /* free_pages my go negative - that's OK */
1265 long free_pages = zone_page_state(z, NR_FREE_PAGES) - (1 << order) + 1;
1268 if (alloc_flags & ALLOC_HIGH)
1270 if (alloc_flags & ALLOC_HARDER)
1273 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
1275 for (o = 0; o < order; o++) {
1276 /* At the next order, this order's pages become unavailable */
1277 free_pages -= z->free_area[o].nr_free << o;
1279 /* Require fewer higher order pages to be free */
1282 if (free_pages <= min)
1290 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1291 * skip over zones that are not allowed by the cpuset, or that have
1292 * been recently (in last second) found to be nearly full. See further
1293 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1294 * that have to skip over a lot of full or unallowed zones.
1296 * If the zonelist cache is present in the passed in zonelist, then
1297 * returns a pointer to the allowed node mask (either the current
1298 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
1300 * If the zonelist cache is not available for this zonelist, does
1301 * nothing and returns NULL.
1303 * If the fullzones BITMAP in the zonelist cache is stale (more than
1304 * a second since last zap'd) then we zap it out (clear its bits.)
1306 * We hold off even calling zlc_setup, until after we've checked the
1307 * first zone in the zonelist, on the theory that most allocations will
1308 * be satisfied from that first zone, so best to examine that zone as
1309 * quickly as we can.
1311 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1313 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1314 nodemask_t *allowednodes; /* zonelist_cache approximation */
1316 zlc = zonelist->zlcache_ptr;
1320 if (time_after(jiffies, zlc->last_full_zap + HZ)) {
1321 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1322 zlc->last_full_zap = jiffies;
1325 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1326 &cpuset_current_mems_allowed :
1327 &node_states[N_HIGH_MEMORY];
1328 return allowednodes;
1332 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1333 * if it is worth looking at further for free memory:
1334 * 1) Check that the zone isn't thought to be full (doesn't have its
1335 * bit set in the zonelist_cache fullzones BITMAP).
1336 * 2) Check that the zones node (obtained from the zonelist_cache
1337 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1338 * Return true (non-zero) if zone is worth looking at further, or
1339 * else return false (zero) if it is not.
1341 * This check -ignores- the distinction between various watermarks,
1342 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1343 * found to be full for any variation of these watermarks, it will
1344 * be considered full for up to one second by all requests, unless
1345 * we are so low on memory on all allowed nodes that we are forced
1346 * into the second scan of the zonelist.
1348 * In the second scan we ignore this zonelist cache and exactly
1349 * apply the watermarks to all zones, even it is slower to do so.
1350 * We are low on memory in the second scan, and should leave no stone
1351 * unturned looking for a free page.
1353 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
1354 nodemask_t *allowednodes)
1356 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1357 int i; /* index of *z in zonelist zones */
1358 int n; /* node that zone *z is on */
1360 zlc = zonelist->zlcache_ptr;
1364 i = z - zonelist->_zonerefs;
1367 /* This zone is worth trying if it is allowed but not full */
1368 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1372 * Given 'z' scanning a zonelist, set the corresponding bit in
1373 * zlc->fullzones, so that subsequent attempts to allocate a page
1374 * from that zone don't waste time re-examining it.
1376 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
1378 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1379 int i; /* index of *z in zonelist zones */
1381 zlc = zonelist->zlcache_ptr;
1385 i = z - zonelist->_zonerefs;
1387 set_bit(i, zlc->fullzones);
1390 #else /* CONFIG_NUMA */
1392 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1397 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
1398 nodemask_t *allowednodes)
1403 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
1406 #endif /* CONFIG_NUMA */
1409 * get_page_from_freelist goes through the zonelist trying to allocate
1412 static struct page *
1413 get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order,
1414 struct zonelist *zonelist, int high_zoneidx, int alloc_flags,
1415 struct zone *preferred_zone, int migratetype)
1418 struct page *page = NULL;
1421 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1422 int zlc_active = 0; /* set if using zonelist_cache */
1423 int did_zlc_setup = 0; /* just call zlc_setup() one time */
1425 if (WARN_ON_ONCE(order >= MAX_ORDER))
1428 classzone_idx = zone_idx(preferred_zone);
1431 * Scan zonelist, looking for a zone with enough free.
1432 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1434 for_each_zone_zonelist_nodemask(zone, z, zonelist,
1435 high_zoneidx, nodemask) {
1436 if (NUMA_BUILD && zlc_active &&
1437 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1439 if ((alloc_flags & ALLOC_CPUSET) &&
1440 !cpuset_zone_allowed_softwall(zone, gfp_mask))
1443 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
1445 if (alloc_flags & ALLOC_WMARK_MIN)
1446 mark = zone->pages_min;
1447 else if (alloc_flags & ALLOC_WMARK_LOW)
1448 mark = zone->pages_low;
1450 mark = zone->pages_high;
1451 if (!zone_watermark_ok(zone, order, mark,
1452 classzone_idx, alloc_flags)) {
1453 if (!zone_reclaim_mode ||
1454 !zone_reclaim(zone, gfp_mask, order))
1455 goto this_zone_full;
1459 page = buffered_rmqueue(preferred_zone, zone, order,
1460 gfp_mask, migratetype);
1465 zlc_mark_zone_full(zonelist, z);
1467 if (NUMA_BUILD && !did_zlc_setup && num_online_nodes() > 1) {
1469 * we do zlc_setup after the first zone is tried but only
1470 * if there are multiple nodes make it worthwhile
1472 allowednodes = zlc_setup(zonelist, alloc_flags);
1478 if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) {
1479 /* Disable zlc cache for second zonelist scan */
1487 should_alloc_retry(gfp_t gfp_mask, unsigned int order,
1488 unsigned long pages_reclaimed)
1490 /* Do not loop if specifically requested */
1491 if (gfp_mask & __GFP_NORETRY)
1495 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
1496 * means __GFP_NOFAIL, but that may not be true in other
1499 if (order <= PAGE_ALLOC_COSTLY_ORDER)
1503 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
1504 * specified, then we retry until we no longer reclaim any pages
1505 * (above), or we've reclaimed an order of pages at least as
1506 * large as the allocation's order. In both cases, if the
1507 * allocation still fails, we stop retrying.
1509 if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order))
1513 * Don't let big-order allocations loop unless the caller
1514 * explicitly requests that.
1516 if (gfp_mask & __GFP_NOFAIL)
1522 static inline struct page *
1523 __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
1524 struct zonelist *zonelist, enum zone_type high_zoneidx,
1525 nodemask_t *nodemask, struct zone *preferred_zone,
1530 /* Acquire the OOM killer lock for the zones in zonelist */
1531 if (!try_set_zone_oom(zonelist, gfp_mask)) {
1532 schedule_timeout_uninterruptible(1);
1537 * Go through the zonelist yet one more time, keep very high watermark
1538 * here, this is only to catch a parallel oom killing, we must fail if
1539 * we're still under heavy pressure.
1541 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask,
1542 order, zonelist, high_zoneidx,
1543 ALLOC_WMARK_HIGH|ALLOC_CPUSET,
1544 preferred_zone, migratetype);
1548 /* The OOM killer will not help higher order allocs */
1549 if (order > PAGE_ALLOC_COSTLY_ORDER)
1552 /* Exhausted what can be done so it's blamo time */
1553 out_of_memory(zonelist, gfp_mask, order);
1556 clear_zonelist_oom(zonelist, gfp_mask);
1560 /* The really slow allocator path where we enter direct reclaim */
1561 static inline struct page *
1562 __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
1563 struct zonelist *zonelist, enum zone_type high_zoneidx,
1564 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
1565 int migratetype, unsigned long *did_some_progress)
1567 struct page *page = NULL;
1568 struct reclaim_state reclaim_state;
1569 struct task_struct *p = current;
1573 /* We now go into synchronous reclaim */
1574 cpuset_memory_pressure_bump();
1577 * The task's cpuset might have expanded its set of allowable nodes
1579 p->flags |= PF_MEMALLOC;
1580 lockdep_set_current_reclaim_state(gfp_mask);
1581 reclaim_state.reclaimed_slab = 0;
1582 p->reclaim_state = &reclaim_state;
1584 *did_some_progress = try_to_free_pages(zonelist, order, gfp_mask, nodemask);
1586 p->reclaim_state = NULL;
1587 lockdep_clear_current_reclaim_state();
1588 p->flags &= ~PF_MEMALLOC;
1595 if (likely(*did_some_progress))
1596 page = get_page_from_freelist(gfp_mask, nodemask, order,
1597 zonelist, high_zoneidx,
1598 alloc_flags, preferred_zone,
1604 * This is called in the allocator slow-path if the allocation request is of
1605 * sufficient urgency to ignore watermarks and take other desperate measures
1607 static inline struct page *
1608 __alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order,
1609 struct zonelist *zonelist, enum zone_type high_zoneidx,
1610 nodemask_t *nodemask, struct zone *preferred_zone,
1616 page = get_page_from_freelist(gfp_mask, nodemask, order,
1617 zonelist, high_zoneidx, ALLOC_NO_WATERMARKS,
1618 preferred_zone, migratetype);
1620 if (!page && gfp_mask & __GFP_NOFAIL)
1621 congestion_wait(WRITE, HZ/50);
1622 } while (!page && (gfp_mask & __GFP_NOFAIL));
1628 void wake_all_kswapd(unsigned int order, struct zonelist *zonelist,
1629 enum zone_type high_zoneidx)
1634 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx)
1635 wakeup_kswapd(zone, order);
1639 gfp_to_alloc_flags(gfp_t gfp_mask)
1641 struct task_struct *p = current;
1642 int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
1643 const gfp_t wait = gfp_mask & __GFP_WAIT;
1645 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
1646 BUILD_BUG_ON(__GFP_HIGH != ALLOC_HIGH);
1649 * The caller may dip into page reserves a bit more if the caller
1650 * cannot run direct reclaim, or if the caller has realtime scheduling
1651 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1652 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1654 alloc_flags |= (gfp_mask & __GFP_HIGH);
1657 alloc_flags |= ALLOC_HARDER;
1659 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1660 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1662 alloc_flags &= ~ALLOC_CPUSET;
1663 } else if (unlikely(rt_task(p)))
1664 alloc_flags |= ALLOC_HARDER;
1666 if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) {
1667 if (!in_interrupt() &&
1668 ((p->flags & PF_MEMALLOC) ||
1669 unlikely(test_thread_flag(TIF_MEMDIE))))
1670 alloc_flags |= ALLOC_NO_WATERMARKS;
1676 static inline struct page *
1677 __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
1678 struct zonelist *zonelist, enum zone_type high_zoneidx,
1679 nodemask_t *nodemask, struct zone *preferred_zone,
1682 const gfp_t wait = gfp_mask & __GFP_WAIT;
1683 struct page *page = NULL;
1685 unsigned long pages_reclaimed = 0;
1686 unsigned long did_some_progress;
1687 struct task_struct *p = current;
1690 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1691 * __GFP_NOWARN set) should not cause reclaim since the subsystem
1692 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1693 * using a larger set of nodes after it has established that the
1694 * allowed per node queues are empty and that nodes are
1697 if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
1700 wake_all_kswapd(order, zonelist, high_zoneidx);
1703 * OK, we're below the kswapd watermark and have kicked background
1704 * reclaim. Now things get more complex, so set up alloc_flags according
1705 * to how we want to proceed.
1707 alloc_flags = gfp_to_alloc_flags(gfp_mask);
1710 /* This is the last chance, in general, before the goto nopage. */
1711 page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist,
1712 high_zoneidx, alloc_flags & ~ALLOC_NO_WATERMARKS,
1713 preferred_zone, migratetype);
1718 /* Allocate without watermarks if the context allows */
1719 if (alloc_flags & ALLOC_NO_WATERMARKS) {
1720 page = __alloc_pages_high_priority(gfp_mask, order,
1721 zonelist, high_zoneidx, nodemask,
1722 preferred_zone, migratetype);
1727 /* Atomic allocations - we can't balance anything */
1731 /* Avoid recursion of direct reclaim */
1732 if (p->flags & PF_MEMALLOC)
1735 /* Try direct reclaim and then allocating */
1736 page = __alloc_pages_direct_reclaim(gfp_mask, order,
1737 zonelist, high_zoneidx,
1739 alloc_flags, preferred_zone,
1740 migratetype, &did_some_progress);
1745 * If we failed to make any progress reclaiming, then we are
1746 * running out of options and have to consider going OOM
1748 if (!did_some_progress) {
1749 if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
1750 page = __alloc_pages_may_oom(gfp_mask, order,
1751 zonelist, high_zoneidx,
1752 nodemask, preferred_zone,
1758 * The OOM killer does not trigger for high-order allocations
1759 * but if no progress is being made, there are no other
1760 * options and retrying is unlikely to help
1762 if (order > PAGE_ALLOC_COSTLY_ORDER)
1769 /* Check if we should retry the allocation */
1770 pages_reclaimed += did_some_progress;
1771 if (should_alloc_retry(gfp_mask, order, pages_reclaimed)) {
1772 /* Wait for some write requests to complete then retry */
1773 congestion_wait(WRITE, HZ/50);
1778 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
1779 printk(KERN_WARNING "%s: page allocation failure."
1780 " order:%d, mode:0x%x\n",
1781 p->comm, order, gfp_mask);
1791 * This is the 'heart' of the zoned buddy allocator.
1794 __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
1795 struct zonelist *zonelist, nodemask_t *nodemask)
1797 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
1798 struct zone *preferred_zone;
1800 int migratetype = allocflags_to_migratetype(gfp_mask);
1802 lockdep_trace_alloc(gfp_mask);
1804 might_sleep_if(gfp_mask & __GFP_WAIT);
1806 if (should_fail_alloc_page(gfp_mask, order))
1810 * Check the zones suitable for the gfp_mask contain at least one
1811 * valid zone. It's possible to have an empty zonelist as a result
1812 * of GFP_THISNODE and a memoryless node
1814 if (unlikely(!zonelist->_zonerefs->zone))
1817 /* The preferred zone is used for statistics later */
1818 first_zones_zonelist(zonelist, high_zoneidx, nodemask, &preferred_zone);
1819 if (!preferred_zone)
1822 /* First allocation attempt */
1823 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order,
1824 zonelist, high_zoneidx, ALLOC_WMARK_LOW|ALLOC_CPUSET,
1825 preferred_zone, migratetype);
1826 if (unlikely(!page))
1827 page = __alloc_pages_slowpath(gfp_mask, order,
1828 zonelist, high_zoneidx, nodemask,
1829 preferred_zone, migratetype);
1833 EXPORT_SYMBOL(__alloc_pages_nodemask);
1836 * Common helper functions.
1838 unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
1841 page = alloc_pages(gfp_mask, order);
1844 return (unsigned long) page_address(page);
1847 EXPORT_SYMBOL(__get_free_pages);
1849 unsigned long get_zeroed_page(gfp_t gfp_mask)
1854 * get_zeroed_page() returns a 32-bit address, which cannot represent
1857 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
1859 page = alloc_pages(gfp_mask | __GFP_ZERO, 0);
1861 return (unsigned long) page_address(page);
1865 EXPORT_SYMBOL(get_zeroed_page);
1867 void __pagevec_free(struct pagevec *pvec)
1869 int i = pagevec_count(pvec);
1872 free_hot_cold_page(pvec->pages[i], pvec->cold);
1875 void __free_pages(struct page *page, unsigned int order)
1877 if (put_page_testzero(page)) {
1879 free_hot_page(page);
1881 __free_pages_ok(page, order);
1885 EXPORT_SYMBOL(__free_pages);
1887 void free_pages(unsigned long addr, unsigned int order)
1890 VM_BUG_ON(!virt_addr_valid((void *)addr));
1891 __free_pages(virt_to_page((void *)addr), order);
1895 EXPORT_SYMBOL(free_pages);
1898 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
1899 * @size: the number of bytes to allocate
1900 * @gfp_mask: GFP flags for the allocation
1902 * This function is similar to alloc_pages(), except that it allocates the
1903 * minimum number of pages to satisfy the request. alloc_pages() can only
1904 * allocate memory in power-of-two pages.
1906 * This function is also limited by MAX_ORDER.
1908 * Memory allocated by this function must be released by free_pages_exact().
1910 void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
1912 unsigned int order = get_order(size);
1915 addr = __get_free_pages(gfp_mask, order);
1917 unsigned long alloc_end = addr + (PAGE_SIZE << order);
1918 unsigned long used = addr + PAGE_ALIGN(size);
1920 split_page(virt_to_page(addr), order);
1921 while (used < alloc_end) {
1927 return (void *)addr;
1929 EXPORT_SYMBOL(alloc_pages_exact);
1932 * free_pages_exact - release memory allocated via alloc_pages_exact()
1933 * @virt: the value returned by alloc_pages_exact.
1934 * @size: size of allocation, same value as passed to alloc_pages_exact().
1936 * Release the memory allocated by a previous call to alloc_pages_exact.
1938 void free_pages_exact(void *virt, size_t size)
1940 unsigned long addr = (unsigned long)virt;
1941 unsigned long end = addr + PAGE_ALIGN(size);
1943 while (addr < end) {
1948 EXPORT_SYMBOL(free_pages_exact);
1950 static unsigned int nr_free_zone_pages(int offset)
1955 /* Just pick one node, since fallback list is circular */
1956 unsigned int sum = 0;
1958 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
1960 for_each_zone_zonelist(zone, z, zonelist, offset) {
1961 unsigned long size = zone->present_pages;
1962 unsigned long high = zone->pages_high;
1971 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1973 unsigned int nr_free_buffer_pages(void)
1975 return nr_free_zone_pages(gfp_zone(GFP_USER));
1977 EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
1980 * Amount of free RAM allocatable within all zones
1982 unsigned int nr_free_pagecache_pages(void)
1984 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
1987 static inline void show_node(struct zone *zone)
1990 printk("Node %d ", zone_to_nid(zone));
1993 void si_meminfo(struct sysinfo *val)
1995 val->totalram = totalram_pages;
1997 val->freeram = global_page_state(NR_FREE_PAGES);
1998 val->bufferram = nr_blockdev_pages();
1999 val->totalhigh = totalhigh_pages;
2000 val->freehigh = nr_free_highpages();
2001 val->mem_unit = PAGE_SIZE;
2004 EXPORT_SYMBOL(si_meminfo);
2007 void si_meminfo_node(struct sysinfo *val, int nid)
2009 pg_data_t *pgdat = NODE_DATA(nid);
2011 val->totalram = pgdat->node_present_pages;
2012 val->freeram = node_page_state(nid, NR_FREE_PAGES);
2013 #ifdef CONFIG_HIGHMEM
2014 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
2015 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
2021 val->mem_unit = PAGE_SIZE;
2025 #define K(x) ((x) << (PAGE_SHIFT-10))
2028 * Show free area list (used inside shift_scroll-lock stuff)
2029 * We also calculate the percentage fragmentation. We do this by counting the
2030 * memory on each free list with the exception of the first item on the list.
2032 void show_free_areas(void)
2037 for_each_populated_zone(zone) {
2039 printk("%s per-cpu:\n", zone->name);
2041 for_each_online_cpu(cpu) {
2042 struct per_cpu_pageset *pageset;
2044 pageset = zone_pcp(zone, cpu);
2046 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
2047 cpu, pageset->pcp.high,
2048 pageset->pcp.batch, pageset->pcp.count);
2052 printk("Active_anon:%lu active_file:%lu inactive_anon:%lu\n"
2053 " inactive_file:%lu"
2054 //TODO: check/adjust line lengths
2055 #ifdef CONFIG_UNEVICTABLE_LRU
2058 " dirty:%lu writeback:%lu unstable:%lu\n"
2059 " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
2060 global_page_state(NR_ACTIVE_ANON),
2061 global_page_state(NR_ACTIVE_FILE),
2062 global_page_state(NR_INACTIVE_ANON),
2063 global_page_state(NR_INACTIVE_FILE),
2064 #ifdef CONFIG_UNEVICTABLE_LRU
2065 global_page_state(NR_UNEVICTABLE),
2067 global_page_state(NR_FILE_DIRTY),
2068 global_page_state(NR_WRITEBACK),
2069 global_page_state(NR_UNSTABLE_NFS),
2070 global_page_state(NR_FREE_PAGES),
2071 global_page_state(NR_SLAB_RECLAIMABLE) +
2072 global_page_state(NR_SLAB_UNRECLAIMABLE),
2073 global_page_state(NR_FILE_MAPPED),
2074 global_page_state(NR_PAGETABLE),
2075 global_page_state(NR_BOUNCE));
2077 for_each_populated_zone(zone) {
2086 " active_anon:%lukB"
2087 " inactive_anon:%lukB"
2088 " active_file:%lukB"
2089 " inactive_file:%lukB"
2090 #ifdef CONFIG_UNEVICTABLE_LRU
2091 " unevictable:%lukB"
2094 " pages_scanned:%lu"
2095 " all_unreclaimable? %s"
2098 K(zone_page_state(zone, NR_FREE_PAGES)),
2101 K(zone->pages_high),
2102 K(zone_page_state(zone, NR_ACTIVE_ANON)),
2103 K(zone_page_state(zone, NR_INACTIVE_ANON)),
2104 K(zone_page_state(zone, NR_ACTIVE_FILE)),
2105 K(zone_page_state(zone, NR_INACTIVE_FILE)),
2106 #ifdef CONFIG_UNEVICTABLE_LRU
2107 K(zone_page_state(zone, NR_UNEVICTABLE)),
2109 K(zone->present_pages),
2110 zone->pages_scanned,
2111 (zone_is_all_unreclaimable(zone) ? "yes" : "no")
2113 printk("lowmem_reserve[]:");
2114 for (i = 0; i < MAX_NR_ZONES; i++)
2115 printk(" %lu", zone->lowmem_reserve[i]);
2119 for_each_populated_zone(zone) {
2120 unsigned long nr[MAX_ORDER], flags, order, total = 0;
2123 printk("%s: ", zone->name);
2125 spin_lock_irqsave(&zone->lock, flags);
2126 for (order = 0; order < MAX_ORDER; order++) {
2127 nr[order] = zone->free_area[order].nr_free;
2128 total += nr[order] << order;
2130 spin_unlock_irqrestore(&zone->lock, flags);
2131 for (order = 0; order < MAX_ORDER; order++)
2132 printk("%lu*%lukB ", nr[order], K(1UL) << order);
2133 printk("= %lukB\n", K(total));
2136 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
2138 show_swap_cache_info();
2141 static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
2143 zoneref->zone = zone;
2144 zoneref->zone_idx = zone_idx(zone);
2148 * Builds allocation fallback zone lists.
2150 * Add all populated zones of a node to the zonelist.
2152 static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
2153 int nr_zones, enum zone_type zone_type)
2157 BUG_ON(zone_type >= MAX_NR_ZONES);
2162 zone = pgdat->node_zones + zone_type;
2163 if (populated_zone(zone)) {
2164 zoneref_set_zone(zone,
2165 &zonelist->_zonerefs[nr_zones++]);
2166 check_highest_zone(zone_type);
2169 } while (zone_type);
2176 * 0 = automatic detection of better ordering.
2177 * 1 = order by ([node] distance, -zonetype)
2178 * 2 = order by (-zonetype, [node] distance)
2180 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
2181 * the same zonelist. So only NUMA can configure this param.
2183 #define ZONELIST_ORDER_DEFAULT 0
2184 #define ZONELIST_ORDER_NODE 1
2185 #define ZONELIST_ORDER_ZONE 2
2187 /* zonelist order in the kernel.
2188 * set_zonelist_order() will set this to NODE or ZONE.
2190 static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
2191 static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
2195 /* The value user specified ....changed by config */
2196 static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2197 /* string for sysctl */
2198 #define NUMA_ZONELIST_ORDER_LEN 16
2199 char numa_zonelist_order[16] = "default";
2202 * interface for configure zonelist ordering.
2203 * command line option "numa_zonelist_order"
2204 * = "[dD]efault - default, automatic configuration.
2205 * = "[nN]ode - order by node locality, then by zone within node
2206 * = "[zZ]one - order by zone, then by locality within zone
2209 static int __parse_numa_zonelist_order(char *s)
2211 if (*s == 'd' || *s == 'D') {
2212 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2213 } else if (*s == 'n' || *s == 'N') {
2214 user_zonelist_order = ZONELIST_ORDER_NODE;
2215 } else if (*s == 'z' || *s == 'Z') {
2216 user_zonelist_order = ZONELIST_ORDER_ZONE;
2219 "Ignoring invalid numa_zonelist_order value: "
2226 static __init int setup_numa_zonelist_order(char *s)
2229 return __parse_numa_zonelist_order(s);
2232 early_param("numa_zonelist_order", setup_numa_zonelist_order);
2235 * sysctl handler for numa_zonelist_order
2237 int numa_zonelist_order_handler(ctl_table *table, int write,
2238 struct file *file, void __user *buffer, size_t *length,
2241 char saved_string[NUMA_ZONELIST_ORDER_LEN];
2245 strncpy(saved_string, (char*)table->data,
2246 NUMA_ZONELIST_ORDER_LEN);
2247 ret = proc_dostring(table, write, file, buffer, length, ppos);
2251 int oldval = user_zonelist_order;
2252 if (__parse_numa_zonelist_order((char*)table->data)) {
2254 * bogus value. restore saved string
2256 strncpy((char*)table->data, saved_string,
2257 NUMA_ZONELIST_ORDER_LEN);
2258 user_zonelist_order = oldval;
2259 } else if (oldval != user_zonelist_order)
2260 build_all_zonelists();
2266 #define MAX_NODE_LOAD (num_online_nodes())
2267 static int node_load[MAX_NUMNODES];
2270 * find_next_best_node - find the next node that should appear in a given node's fallback list
2271 * @node: node whose fallback list we're appending
2272 * @used_node_mask: nodemask_t of already used nodes
2274 * We use a number of factors to determine which is the next node that should
2275 * appear on a given node's fallback list. The node should not have appeared
2276 * already in @node's fallback list, and it should be the next closest node
2277 * according to the distance array (which contains arbitrary distance values
2278 * from each node to each node in the system), and should also prefer nodes
2279 * with no CPUs, since presumably they'll have very little allocation pressure
2280 * on them otherwise.
2281 * It returns -1 if no node is found.
2283 static int find_next_best_node(int node, nodemask_t *used_node_mask)
2286 int min_val = INT_MAX;
2288 const struct cpumask *tmp = cpumask_of_node(0);
2290 /* Use the local node if we haven't already */
2291 if (!node_isset(node, *used_node_mask)) {
2292 node_set(node, *used_node_mask);
2296 for_each_node_state(n, N_HIGH_MEMORY) {
2298 /* Don't want a node to appear more than once */
2299 if (node_isset(n, *used_node_mask))
2302 /* Use the distance array to find the distance */
2303 val = node_distance(node, n);
2305 /* Penalize nodes under us ("prefer the next node") */
2308 /* Give preference to headless and unused nodes */
2309 tmp = cpumask_of_node(n);
2310 if (!cpumask_empty(tmp))
2311 val += PENALTY_FOR_NODE_WITH_CPUS;
2313 /* Slight preference for less loaded node */
2314 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
2315 val += node_load[n];
2317 if (val < min_val) {
2324 node_set(best_node, *used_node_mask);
2331 * Build zonelists ordered by node and zones within node.
2332 * This results in maximum locality--normal zone overflows into local
2333 * DMA zone, if any--but risks exhausting DMA zone.
2335 static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
2338 struct zonelist *zonelist;
2340 zonelist = &pgdat->node_zonelists[0];
2341 for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
2343 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2345 zonelist->_zonerefs[j].zone = NULL;
2346 zonelist->_zonerefs[j].zone_idx = 0;
2350 * Build gfp_thisnode zonelists
2352 static void build_thisnode_zonelists(pg_data_t *pgdat)
2355 struct zonelist *zonelist;
2357 zonelist = &pgdat->node_zonelists[1];
2358 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
2359 zonelist->_zonerefs[j].zone = NULL;
2360 zonelist->_zonerefs[j].zone_idx = 0;
2364 * Build zonelists ordered by zone and nodes within zones.
2365 * This results in conserving DMA zone[s] until all Normal memory is
2366 * exhausted, but results in overflowing to remote node while memory
2367 * may still exist in local DMA zone.
2369 static int node_order[MAX_NUMNODES];
2371 static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
2374 int zone_type; /* needs to be signed */
2376 struct zonelist *zonelist;
2378 zonelist = &pgdat->node_zonelists[0];
2380 for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
2381 for (j = 0; j < nr_nodes; j++) {
2382 node = node_order[j];
2383 z = &NODE_DATA(node)->node_zones[zone_type];
2384 if (populated_zone(z)) {
2386 &zonelist->_zonerefs[pos++]);
2387 check_highest_zone(zone_type);
2391 zonelist->_zonerefs[pos].zone = NULL;
2392 zonelist->_zonerefs[pos].zone_idx = 0;
2395 static int default_zonelist_order(void)
2398 unsigned long low_kmem_size,total_size;
2402 * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem.
2403 * If they are really small and used heavily, the system can fall
2404 * into OOM very easily.
2405 * This function detect ZONE_DMA/DMA32 size and confgigures zone order.
2407 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2410 for_each_online_node(nid) {
2411 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2412 z = &NODE_DATA(nid)->node_zones[zone_type];
2413 if (populated_zone(z)) {
2414 if (zone_type < ZONE_NORMAL)
2415 low_kmem_size += z->present_pages;
2416 total_size += z->present_pages;
2420 if (!low_kmem_size || /* there are no DMA area. */
2421 low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
2422 return ZONELIST_ORDER_NODE;
2424 * look into each node's config.
2425 * If there is a node whose DMA/DMA32 memory is very big area on
2426 * local memory, NODE_ORDER may be suitable.
2428 average_size = total_size /
2429 (nodes_weight(node_states[N_HIGH_MEMORY]) + 1);
2430 for_each_online_node(nid) {
2433 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2434 z = &NODE_DATA(nid)->node_zones[zone_type];
2435 if (populated_zone(z)) {
2436 if (zone_type < ZONE_NORMAL)
2437 low_kmem_size += z->present_pages;
2438 total_size += z->present_pages;
2441 if (low_kmem_size &&
2442 total_size > average_size && /* ignore small node */
2443 low_kmem_size > total_size * 70/100)
2444 return ZONELIST_ORDER_NODE;
2446 return ZONELIST_ORDER_ZONE;
2449 static void set_zonelist_order(void)
2451 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
2452 current_zonelist_order = default_zonelist_order();
2454 current_zonelist_order = user_zonelist_order;
2457 static void build_zonelists(pg_data_t *pgdat)
2461 nodemask_t used_mask;
2462 int local_node, prev_node;
2463 struct zonelist *zonelist;
2464 int order = current_zonelist_order;
2466 /* initialize zonelists */
2467 for (i = 0; i < MAX_ZONELISTS; i++) {
2468 zonelist = pgdat->node_zonelists + i;
2469 zonelist->_zonerefs[0].zone = NULL;
2470 zonelist->_zonerefs[0].zone_idx = 0;
2473 /* NUMA-aware ordering of nodes */
2474 local_node = pgdat->node_id;
2475 load = num_online_nodes();
2476 prev_node = local_node;
2477 nodes_clear(used_mask);
2479 memset(node_load, 0, sizeof(node_load));
2480 memset(node_order, 0, sizeof(node_order));
2483 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
2484 int distance = node_distance(local_node, node);
2487 * If another node is sufficiently far away then it is better
2488 * to reclaim pages in a zone before going off node.
2490 if (distance > RECLAIM_DISTANCE)
2491 zone_reclaim_mode = 1;
2494 * We don't want to pressure a particular node.
2495 * So adding penalty to the first node in same
2496 * distance group to make it round-robin.
2498 if (distance != node_distance(local_node, prev_node))
2499 node_load[node] = load;
2503 if (order == ZONELIST_ORDER_NODE)
2504 build_zonelists_in_node_order(pgdat, node);
2506 node_order[j++] = node; /* remember order */
2509 if (order == ZONELIST_ORDER_ZONE) {
2510 /* calculate node order -- i.e., DMA last! */
2511 build_zonelists_in_zone_order(pgdat, j);
2514 build_thisnode_zonelists(pgdat);
2517 /* Construct the zonelist performance cache - see further mmzone.h */
2518 static void build_zonelist_cache(pg_data_t *pgdat)
2520 struct zonelist *zonelist;
2521 struct zonelist_cache *zlc;
2524 zonelist = &pgdat->node_zonelists[0];
2525 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
2526 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
2527 for (z = zonelist->_zonerefs; z->zone; z++)
2528 zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z);
2532 #else /* CONFIG_NUMA */
2534 static void set_zonelist_order(void)
2536 current_zonelist_order = ZONELIST_ORDER_ZONE;
2539 static void build_zonelists(pg_data_t *pgdat)
2541 int node, local_node;
2543 struct zonelist *zonelist;
2545 local_node = pgdat->node_id;
2547 zonelist = &pgdat->node_zonelists[0];
2548 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
2551 * Now we build the zonelist so that it contains the zones
2552 * of all the other nodes.
2553 * We don't want to pressure a particular node, so when
2554 * building the zones for node N, we make sure that the
2555 * zones coming right after the local ones are those from
2556 * node N+1 (modulo N)
2558 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
2559 if (!node_online(node))
2561 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2564 for (node = 0; node < local_node; node++) {
2565 if (!node_online(node))
2567 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2571 zonelist->_zonerefs[j].zone = NULL;
2572 zonelist->_zonerefs[j].zone_idx = 0;
2575 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
2576 static void build_zonelist_cache(pg_data_t *pgdat)
2578 pgdat->node_zonelists[0].zlcache_ptr = NULL;
2581 #endif /* CONFIG_NUMA */
2583 /* return values int ....just for stop_machine() */
2584 static int __build_all_zonelists(void *dummy)
2588 for_each_online_node(nid) {
2589 pg_data_t *pgdat = NODE_DATA(nid);
2591 build_zonelists(pgdat);
2592 build_zonelist_cache(pgdat);
2597 void build_all_zonelists(void)
2599 set_zonelist_order();
2601 if (system_state == SYSTEM_BOOTING) {
2602 __build_all_zonelists(NULL);
2603 mminit_verify_zonelist();
2604 cpuset_init_current_mems_allowed();
2606 /* we have to stop all cpus to guarantee there is no user
2608 stop_machine(__build_all_zonelists, NULL, NULL);
2609 /* cpuset refresh routine should be here */
2611 vm_total_pages = nr_free_pagecache_pages();
2613 * Disable grouping by mobility if the number of pages in the
2614 * system is too low to allow the mechanism to work. It would be
2615 * more accurate, but expensive to check per-zone. This check is
2616 * made on memory-hotadd so a system can start with mobility
2617 * disabled and enable it later
2619 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
2620 page_group_by_mobility_disabled = 1;
2622 page_group_by_mobility_disabled = 0;
2624 printk("Built %i zonelists in %s order, mobility grouping %s. "
2625 "Total pages: %ld\n",
2627 zonelist_order_name[current_zonelist_order],
2628 page_group_by_mobility_disabled ? "off" : "on",
2631 printk("Policy zone: %s\n", zone_names[policy_zone]);
2636 * Helper functions to size the waitqueue hash table.
2637 * Essentially these want to choose hash table sizes sufficiently
2638 * large so that collisions trying to wait on pages are rare.
2639 * But in fact, the number of active page waitqueues on typical
2640 * systems is ridiculously low, less than 200. So this is even
2641 * conservative, even though it seems large.
2643 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
2644 * waitqueues, i.e. the size of the waitq table given the number of pages.
2646 #define PAGES_PER_WAITQUEUE 256
2648 #ifndef CONFIG_MEMORY_HOTPLUG
2649 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
2651 unsigned long size = 1;
2653 pages /= PAGES_PER_WAITQUEUE;
2655 while (size < pages)
2659 * Once we have dozens or even hundreds of threads sleeping
2660 * on IO we've got bigger problems than wait queue collision.
2661 * Limit the size of the wait table to a reasonable size.
2663 size = min(size, 4096UL);
2665 return max(size, 4UL);
2669 * A zone's size might be changed by hot-add, so it is not possible to determine
2670 * a suitable size for its wait_table. So we use the maximum size now.
2672 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
2674 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
2675 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
2676 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
2678 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
2679 * or more by the traditional way. (See above). It equals:
2681 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
2682 * ia64(16K page size) : = ( 8G + 4M)byte.
2683 * powerpc (64K page size) : = (32G +16M)byte.
2685 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
2692 * This is an integer logarithm so that shifts can be used later
2693 * to extract the more random high bits from the multiplicative
2694 * hash function before the remainder is taken.
2696 static inline unsigned long wait_table_bits(unsigned long size)
2701 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
2704 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
2705 * of blocks reserved is based on zone->pages_min. The memory within the
2706 * reserve will tend to store contiguous free pages. Setting min_free_kbytes
2707 * higher will lead to a bigger reserve which will get freed as contiguous
2708 * blocks as reclaim kicks in
2710 static void setup_zone_migrate_reserve(struct zone *zone)
2712 unsigned long start_pfn, pfn, end_pfn;
2714 unsigned long reserve, block_migratetype;
2716 /* Get the start pfn, end pfn and the number of blocks to reserve */
2717 start_pfn = zone->zone_start_pfn;
2718 end_pfn = start_pfn + zone->spanned_pages;
2719 reserve = roundup(zone->pages_min, pageblock_nr_pages) >>
2722 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
2723 if (!pfn_valid(pfn))
2725 page = pfn_to_page(pfn);
2727 /* Watch out for overlapping nodes */
2728 if (page_to_nid(page) != zone_to_nid(zone))
2731 /* Blocks with reserved pages will never free, skip them. */
2732 if (PageReserved(page))
2735 block_migratetype = get_pageblock_migratetype(page);
2737 /* If this block is reserved, account for it */
2738 if (reserve > 0 && block_migratetype == MIGRATE_RESERVE) {
2743 /* Suitable for reserving if this block is movable */
2744 if (reserve > 0 && block_migratetype == MIGRATE_MOVABLE) {
2745 set_pageblock_migratetype(page, MIGRATE_RESERVE);
2746 move_freepages_block(zone, page, MIGRATE_RESERVE);
2752 * If the reserve is met and this is a previous reserved block,
2755 if (block_migratetype == MIGRATE_RESERVE) {
2756 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
2757 move_freepages_block(zone, page, MIGRATE_MOVABLE);
2763 * Initially all pages are reserved - free ones are freed
2764 * up by free_all_bootmem() once the early boot process is
2765 * done. Non-atomic initialization, single-pass.
2767 void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
2768 unsigned long start_pfn, enum memmap_context context)
2771 unsigned long end_pfn = start_pfn + size;
2775 if (highest_memmap_pfn < end_pfn - 1)
2776 highest_memmap_pfn = end_pfn - 1;
2778 z = &NODE_DATA(nid)->node_zones[zone];
2779 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
2781 * There can be holes in boot-time mem_map[]s
2782 * handed to this function. They do not
2783 * exist on hotplugged memory.
2785 if (context == MEMMAP_EARLY) {
2786 if (!early_pfn_valid(pfn))
2788 if (!early_pfn_in_nid(pfn, nid))
2791 page = pfn_to_page(pfn);
2792 set_page_links(page, zone, nid, pfn);
2793 mminit_verify_page_links(page, zone, nid, pfn);
2794 init_page_count(page);
2795 reset_page_mapcount(page);
2796 SetPageReserved(page);
2798 * Mark the block movable so that blocks are reserved for
2799 * movable at startup. This will force kernel allocations
2800 * to reserve their blocks rather than leaking throughout
2801 * the address space during boot when many long-lived
2802 * kernel allocations are made. Later some blocks near
2803 * the start are marked MIGRATE_RESERVE by
2804 * setup_zone_migrate_reserve()
2806 * bitmap is created for zone's valid pfn range. but memmap
2807 * can be created for invalid pages (for alignment)
2808 * check here not to call set_pageblock_migratetype() against
2811 if ((z->zone_start_pfn <= pfn)
2812 && (pfn < z->zone_start_pfn + z->spanned_pages)
2813 && !(pfn & (pageblock_nr_pages - 1)))
2814 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
2816 INIT_LIST_HEAD(&page->lru);
2817 #ifdef WANT_PAGE_VIRTUAL
2818 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
2819 if (!is_highmem_idx(zone))
2820 set_page_address(page, __va(pfn << PAGE_SHIFT));
2825 static void __meminit zone_init_free_lists(struct zone *zone)
2828 for_each_migratetype_order(order, t) {
2829 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
2830 zone->free_area[order].nr_free = 0;
2834 #ifndef __HAVE_ARCH_MEMMAP_INIT
2835 #define memmap_init(size, nid, zone, start_pfn) \
2836 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
2839 static int zone_batchsize(struct zone *zone)
2845 * The per-cpu-pages pools are set to around 1000th of the
2846 * size of the zone. But no more than 1/2 of a meg.
2848 * OK, so we don't know how big the cache is. So guess.
2850 batch = zone->present_pages / 1024;
2851 if (batch * PAGE_SIZE > 512 * 1024)
2852 batch = (512 * 1024) / PAGE_SIZE;
2853 batch /= 4; /* We effectively *= 4 below */
2858 * Clamp the batch to a 2^n - 1 value. Having a power
2859 * of 2 value was found to be more likely to have
2860 * suboptimal cache aliasing properties in some cases.
2862 * For example if 2 tasks are alternately allocating
2863 * batches of pages, one task can end up with a lot
2864 * of pages of one half of the possible page colors
2865 * and the other with pages of the other colors.
2867 batch = rounddown_pow_of_two(batch + batch/2) - 1;
2872 /* The deferral and batching of frees should be suppressed under NOMMU
2875 * The problem is that NOMMU needs to be able to allocate large chunks
2876 * of contiguous memory as there's no hardware page translation to
2877 * assemble apparent contiguous memory from discontiguous pages.
2879 * Queueing large contiguous runs of pages for batching, however,
2880 * causes the pages to actually be freed in smaller chunks. As there
2881 * can be a significant delay between the individual batches being
2882 * recycled, this leads to the once large chunks of space being
2883 * fragmented and becoming unavailable for high-order allocations.
2889 static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
2891 struct per_cpu_pages *pcp;
2893 memset(p, 0, sizeof(*p));
2897 pcp->high = 6 * batch;
2898 pcp->batch = max(1UL, 1 * batch);
2899 INIT_LIST_HEAD(&pcp->list);
2903 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
2904 * to the value high for the pageset p.
2907 static void setup_pagelist_highmark(struct per_cpu_pageset *p,
2910 struct per_cpu_pages *pcp;
2914 pcp->batch = max(1UL, high/4);
2915 if ((high/4) > (PAGE_SHIFT * 8))
2916 pcp->batch = PAGE_SHIFT * 8;
2922 * Boot pageset table. One per cpu which is going to be used for all
2923 * zones and all nodes. The parameters will be set in such a way
2924 * that an item put on a list will immediately be handed over to
2925 * the buddy list. This is safe since pageset manipulation is done
2926 * with interrupts disabled.
2928 * Some NUMA counter updates may also be caught by the boot pagesets.
2930 * The boot_pagesets must be kept even after bootup is complete for
2931 * unused processors and/or zones. They do play a role for bootstrapping
2932 * hotplugged processors.
2934 * zoneinfo_show() and maybe other functions do
2935 * not check if the processor is online before following the pageset pointer.
2936 * Other parts of the kernel may not check if the zone is available.
2938 static struct per_cpu_pageset boot_pageset[NR_CPUS];
2941 * Dynamically allocate memory for the
2942 * per cpu pageset array in struct zone.
2944 static int __cpuinit process_zones(int cpu)
2946 struct zone *zone, *dzone;
2947 int node = cpu_to_node(cpu);
2949 node_set_state(node, N_CPU); /* this node has a cpu */
2951 for_each_populated_zone(zone) {
2952 zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset),
2954 if (!zone_pcp(zone, cpu))
2957 setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone));
2959 if (percpu_pagelist_fraction)
2960 setup_pagelist_highmark(zone_pcp(zone, cpu),
2961 (zone->present_pages / percpu_pagelist_fraction));
2966 for_each_zone(dzone) {
2967 if (!populated_zone(dzone))
2971 kfree(zone_pcp(dzone, cpu));
2972 zone_pcp(dzone, cpu) = NULL;
2977 static inline void free_zone_pagesets(int cpu)
2981 for_each_zone(zone) {
2982 struct per_cpu_pageset *pset = zone_pcp(zone, cpu);
2984 /* Free per_cpu_pageset if it is slab allocated */
2985 if (pset != &boot_pageset[cpu])
2987 zone_pcp(zone, cpu) = NULL;
2991 static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb,
2992 unsigned long action,
2995 int cpu = (long)hcpu;
2996 int ret = NOTIFY_OK;
2999 case CPU_UP_PREPARE:
3000 case CPU_UP_PREPARE_FROZEN:
3001 if (process_zones(cpu))
3004 case CPU_UP_CANCELED:
3005 case CPU_UP_CANCELED_FROZEN:
3007 case CPU_DEAD_FROZEN:
3008 free_zone_pagesets(cpu);
3016 static struct notifier_block __cpuinitdata pageset_notifier =
3017 { &pageset_cpuup_callback, NULL, 0 };
3019 void __init setup_per_cpu_pageset(void)
3023 /* Initialize per_cpu_pageset for cpu 0.
3024 * A cpuup callback will do this for every cpu
3025 * as it comes online
3027 err = process_zones(smp_processor_id());
3029 register_cpu_notifier(&pageset_notifier);
3034 static noinline __init_refok
3035 int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
3038 struct pglist_data *pgdat = zone->zone_pgdat;
3042 * The per-page waitqueue mechanism uses hashed waitqueues
3045 zone->wait_table_hash_nr_entries =
3046 wait_table_hash_nr_entries(zone_size_pages);
3047 zone->wait_table_bits =
3048 wait_table_bits(zone->wait_table_hash_nr_entries);
3049 alloc_size = zone->wait_table_hash_nr_entries
3050 * sizeof(wait_queue_head_t);
3052 if (!slab_is_available()) {
3053 zone->wait_table = (wait_queue_head_t *)
3054 alloc_bootmem_node(pgdat, alloc_size);
3057 * This case means that a zone whose size was 0 gets new memory
3058 * via memory hot-add.
3059 * But it may be the case that a new node was hot-added. In
3060 * this case vmalloc() will not be able to use this new node's
3061 * memory - this wait_table must be initialized to use this new
3062 * node itself as well.
3063 * To use this new node's memory, further consideration will be
3066 zone->wait_table = vmalloc(alloc_size);
3068 if (!zone->wait_table)
3071 for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
3072 init_waitqueue_head(zone->wait_table + i);
3077 static __meminit void zone_pcp_init(struct zone *zone)
3080 unsigned long batch = zone_batchsize(zone);
3082 for (cpu = 0; cpu < NR_CPUS; cpu++) {
3084 /* Early boot. Slab allocator not functional yet */
3085 zone_pcp(zone, cpu) = &boot_pageset[cpu];
3086 setup_pageset(&boot_pageset[cpu],0);
3088 setup_pageset(zone_pcp(zone,cpu), batch);
3091 if (zone->present_pages)
3092 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n",
3093 zone->name, zone->present_pages, batch);
3096 __meminit int init_currently_empty_zone(struct zone *zone,
3097 unsigned long zone_start_pfn,
3099 enum memmap_context context)
3101 struct pglist_data *pgdat = zone->zone_pgdat;
3103 ret = zone_wait_table_init(zone, size);
3106 pgdat->nr_zones = zone_idx(zone) + 1;
3108 zone->zone_start_pfn = zone_start_pfn;
3110 mminit_dprintk(MMINIT_TRACE, "memmap_init",
3111 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
3113 (unsigned long)zone_idx(zone),
3114 zone_start_pfn, (zone_start_pfn + size));
3116 zone_init_free_lists(zone);
3121 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3123 * Basic iterator support. Return the first range of PFNs for a node
3124 * Note: nid == MAX_NUMNODES returns first region regardless of node
3126 static int __meminit first_active_region_index_in_nid(int nid)
3130 for (i = 0; i < nr_nodemap_entries; i++)
3131 if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
3138 * Basic iterator support. Return the next active range of PFNs for a node
3139 * Note: nid == MAX_NUMNODES returns next region regardless of node
3141 static int __meminit next_active_region_index_in_nid(int index, int nid)
3143 for (index = index + 1; index < nr_nodemap_entries; index++)
3144 if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
3150 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
3152 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
3153 * Architectures may implement their own version but if add_active_range()
3154 * was used and there are no special requirements, this is a convenient
3157 int __meminit __early_pfn_to_nid(unsigned long pfn)
3161 for (i = 0; i < nr_nodemap_entries; i++) {
3162 unsigned long start_pfn = early_node_map[i].start_pfn;
3163 unsigned long end_pfn = early_node_map[i].end_pfn;
3165 if (start_pfn <= pfn && pfn < end_pfn)
3166 return early_node_map[i].nid;
3168 /* This is a memory hole */
3171 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
3173 int __meminit early_pfn_to_nid(unsigned long pfn)
3177 nid = __early_pfn_to_nid(pfn);
3180 /* just returns 0 */
3184 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
3185 bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
3189 nid = __early_pfn_to_nid(pfn);
3190 if (nid >= 0 && nid != node)
3196 /* Basic iterator support to walk early_node_map[] */
3197 #define for_each_active_range_index_in_nid(i, nid) \
3198 for (i = first_active_region_index_in_nid(nid); i != -1; \
3199 i = next_active_region_index_in_nid(i, nid))
3202 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
3203 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
3204 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
3206 * If an architecture guarantees that all ranges registered with
3207 * add_active_ranges() contain no holes and may be freed, this
3208 * this function may be used instead of calling free_bootmem() manually.
3210 void __init free_bootmem_with_active_regions(int nid,
3211 unsigned long max_low_pfn)
3215 for_each_active_range_index_in_nid(i, nid) {
3216 unsigned long size_pages = 0;
3217 unsigned long end_pfn = early_node_map[i].end_pfn;
3219 if (early_node_map[i].start_pfn >= max_low_pfn)
3222 if (end_pfn > max_low_pfn)
3223 end_pfn = max_low_pfn;
3225 size_pages = end_pfn - early_node_map[i].start_pfn;
3226 free_bootmem_node(NODE_DATA(early_node_map[i].nid),
3227 PFN_PHYS(early_node_map[i].start_pfn),
3228 size_pages << PAGE_SHIFT);
3232 void __init work_with_active_regions(int nid, work_fn_t work_fn, void *data)
3237 for_each_active_range_index_in_nid(i, nid) {
3238 ret = work_fn(early_node_map[i].start_pfn,
3239 early_node_map[i].end_pfn, data);
3245 * sparse_memory_present_with_active_regions - Call memory_present for each active range
3246 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
3248 * If an architecture guarantees that all ranges registered with
3249 * add_active_ranges() contain no holes and may be freed, this
3250 * function may be used instead of calling memory_present() manually.
3252 void __init sparse_memory_present_with_active_regions(int nid)
3256 for_each_active_range_index_in_nid(i, nid)
3257 memory_present(early_node_map[i].nid,
3258 early_node_map[i].start_pfn,
3259 early_node_map[i].end_pfn);
3263 * get_pfn_range_for_nid - Return the start and end page frames for a node
3264 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
3265 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
3266 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
3268 * It returns the start and end page frame of a node based on information
3269 * provided by an arch calling add_active_range(). If called for a node
3270 * with no available memory, a warning is printed and the start and end
3273 void __meminit get_pfn_range_for_nid(unsigned int nid,
3274 unsigned long *start_pfn, unsigned long *end_pfn)
3280 for_each_active_range_index_in_nid(i, nid) {
3281 *start_pfn = min(*start_pfn, early_node_map[i].start_pfn);
3282 *end_pfn = max(*end_pfn, early_node_map[i].end_pfn);
3285 if (*start_pfn == -1UL)
3290 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3291 * assumption is made that zones within a node are ordered in monotonic
3292 * increasing memory addresses so that the "highest" populated zone is used
3294 static void __init find_usable_zone_for_movable(void)
3297 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
3298 if (zone_index == ZONE_MOVABLE)
3301 if (arch_zone_highest_possible_pfn[zone_index] >
3302 arch_zone_lowest_possible_pfn[zone_index])
3306 VM_BUG_ON(zone_index == -1);
3307 movable_zone = zone_index;
3311 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
3312 * because it is sized independant of architecture. Unlike the other zones,
3313 * the starting point for ZONE_MOVABLE is not fixed. It may be different
3314 * in each node depending on the size of each node and how evenly kernelcore
3315 * is distributed. This helper function adjusts the zone ranges
3316 * provided by the architecture for a given node by using the end of the
3317 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3318 * zones within a node are in order of monotonic increases memory addresses
3320 static void __meminit adjust_zone_range_for_zone_movable(int nid,
3321 unsigned long zone_type,
3322 unsigned long node_start_pfn,
3323 unsigned long node_end_pfn,
3324 unsigned long *zone_start_pfn,
3325 unsigned long *zone_end_pfn)
3327 /* Only adjust if ZONE_MOVABLE is on this node */
3328 if (zone_movable_pfn[nid]) {
3329 /* Size ZONE_MOVABLE */
3330 if (zone_type == ZONE_MOVABLE) {
3331 *zone_start_pfn = zone_movable_pfn[nid];
3332 *zone_end_pfn = min(node_end_pfn,
3333 arch_zone_highest_possible_pfn[movable_zone]);
3335 /* Adjust for ZONE_MOVABLE starting within this range */
3336 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
3337 *zone_end_pfn > zone_movable_pfn[nid]) {
3338 *zone_end_pfn = zone_movable_pfn[nid];
3340 /* Check if this whole range is within ZONE_MOVABLE */
3341 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
3342 *zone_start_pfn = *zone_end_pfn;
3347 * Return the number of pages a zone spans in a node, including holes
3348 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3350 static unsigned long __meminit zone_spanned_pages_in_node(int nid,
3351 unsigned long zone_type,
3352 unsigned long *ignored)
3354 unsigned long node_start_pfn, node_end_pfn;
3355 unsigned long zone_start_pfn, zone_end_pfn;
3357 /* Get the start and end of the node and zone */
3358 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3359 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
3360 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
3361 adjust_zone_range_for_zone_movable(nid, zone_type,
3362 node_start_pfn, node_end_pfn,
3363 &zone_start_pfn, &zone_end_pfn);
3365 /* Check that this node has pages within the zone's required range */
3366 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
3369 /* Move the zone boundaries inside the node if necessary */
3370 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
3371 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
3373 /* Return the spanned pages */
3374 return zone_end_pfn - zone_start_pfn;
3378 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
3379 * then all holes in the requested range will be accounted for.
3381 static unsigned long __meminit __absent_pages_in_range(int nid,
3382 unsigned long range_start_pfn,
3383 unsigned long range_end_pfn)
3386 unsigned long prev_end_pfn = 0, hole_pages = 0;
3387 unsigned long start_pfn;
3389 /* Find the end_pfn of the first active range of pfns in the node */
3390 i = first_active_region_index_in_nid(nid);
3394 prev_end_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3396 /* Account for ranges before physical memory on this node */
3397 if (early_node_map[i].start_pfn > range_start_pfn)
3398 hole_pages = prev_end_pfn - range_start_pfn;
3400 /* Find all holes for the zone within the node */
3401 for (; i != -1; i = next_active_region_index_in_nid(i, nid)) {
3403 /* No need to continue if prev_end_pfn is outside the zone */
3404 if (prev_end_pfn >= range_end_pfn)
3407 /* Make sure the end of the zone is not within the hole */
3408 start_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3409 prev_end_pfn = max(prev_end_pfn, range_start_pfn);
3411 /* Update the hole size cound and move on */
3412 if (start_pfn > range_start_pfn) {
3413 BUG_ON(prev_end_pfn > start_pfn);
3414 hole_pages += start_pfn - prev_end_pfn;
3416 prev_end_pfn = early_node_map[i].end_pfn;
3419 /* Account for ranges past physical memory on this node */
3420 if (range_end_pfn > prev_end_pfn)
3421 hole_pages += range_end_pfn -
3422 max(range_start_pfn, prev_end_pfn);
3428 * absent_pages_in_range - Return number of page frames in holes within a range
3429 * @start_pfn: The start PFN to start searching for holes
3430 * @end_pfn: The end PFN to stop searching for holes
3432 * It returns the number of pages frames in memory holes within a range.
3434 unsigned long __init absent_pages_in_range(unsigned long start_pfn,
3435 unsigned long end_pfn)
3437 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
3440 /* Return the number of page frames in holes in a zone on a node */
3441 static unsigned long __meminit zone_absent_pages_in_node(int nid,
3442 unsigned long zone_type,
3443 unsigned long *ignored)
3445 unsigned long node_start_pfn, node_end_pfn;
3446 unsigned long zone_start_pfn, zone_end_pfn;
3448 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3449 zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type],
3451 zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type],
3454 adjust_zone_range_for_zone_movable(nid, zone_type,
3455 node_start_pfn, node_end_pfn,
3456 &zone_start_pfn, &zone_end_pfn);
3457 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
3461 static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
3462 unsigned long zone_type,
3463 unsigned long *zones_size)
3465 return zones_size[zone_type];
3468 static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
3469 unsigned long zone_type,
3470 unsigned long *zholes_size)
3475 return zholes_size[zone_type];
3480 static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
3481 unsigned long *zones_size, unsigned long *zholes_size)
3483 unsigned long realtotalpages, totalpages = 0;
3486 for (i = 0; i < MAX_NR_ZONES; i++)
3487 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
3489 pgdat->node_spanned_pages = totalpages;
3491 realtotalpages = totalpages;
3492 for (i = 0; i < MAX_NR_ZONES; i++)
3494 zone_absent_pages_in_node(pgdat->node_id, i,
3496 pgdat->node_present_pages = realtotalpages;
3497 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
3501 #ifndef CONFIG_SPARSEMEM
3503 * Calculate the size of the zone->blockflags rounded to an unsigned long
3504 * Start by making sure zonesize is a multiple of pageblock_order by rounding
3505 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
3506 * round what is now in bits to nearest long in bits, then return it in
3509 static unsigned long __init usemap_size(unsigned long zonesize)
3511 unsigned long usemapsize;
3513 usemapsize = roundup(zonesize, pageblock_nr_pages);
3514 usemapsize = usemapsize >> pageblock_order;
3515 usemapsize *= NR_PAGEBLOCK_BITS;
3516 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
3518 return usemapsize / 8;
3521 static void __init setup_usemap(struct pglist_data *pgdat,
3522 struct zone *zone, unsigned long zonesize)
3524 unsigned long usemapsize = usemap_size(zonesize);
3525 zone->pageblock_flags = NULL;
3527 zone->pageblock_flags = alloc_bootmem_node(pgdat, usemapsize);
3530 static void inline setup_usemap(struct pglist_data *pgdat,
3531 struct zone *zone, unsigned long zonesize) {}
3532 #endif /* CONFIG_SPARSEMEM */
3534 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
3536 /* Return a sensible default order for the pageblock size. */
3537 static inline int pageblock_default_order(void)
3539 if (HPAGE_SHIFT > PAGE_SHIFT)
3540 return HUGETLB_PAGE_ORDER;
3545 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
3546 static inline void __init set_pageblock_order(unsigned int order)
3548 /* Check that pageblock_nr_pages has not already been setup */
3549 if (pageblock_order)
3553 * Assume the largest contiguous order of interest is a huge page.
3554 * This value may be variable depending on boot parameters on IA64
3556 pageblock_order = order;
3558 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3561 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
3562 * and pageblock_default_order() are unused as pageblock_order is set
3563 * at compile-time. See include/linux/pageblock-flags.h for the values of
3564 * pageblock_order based on the kernel config
3566 static inline int pageblock_default_order(unsigned int order)
3570 #define set_pageblock_order(x) do {} while (0)
3572 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3575 * Set up the zone data structures:
3576 * - mark all pages reserved
3577 * - mark all memory queues empty
3578 * - clear the memory bitmaps
3580 static void __paginginit free_area_init_core(struct pglist_data *pgdat,
3581 unsigned long *zones_size, unsigned long *zholes_size)
3584 int nid = pgdat->node_id;
3585 unsigned long zone_start_pfn = pgdat->node_start_pfn;
3588 pgdat_resize_init(pgdat);
3589 pgdat->nr_zones = 0;
3590 init_waitqueue_head(&pgdat->kswapd_wait);
3591 pgdat->kswapd_max_order = 0;
3592 pgdat_page_cgroup_init(pgdat);
3594 for (j = 0; j < MAX_NR_ZONES; j++) {
3595 struct zone *zone = pgdat->node_zones + j;
3596 unsigned long size, realsize, memmap_pages;
3599 size = zone_spanned_pages_in_node(nid, j, zones_size);
3600 realsize = size - zone_absent_pages_in_node(nid, j,
3604 * Adjust realsize so that it accounts for how much memory
3605 * is used by this zone for memmap. This affects the watermark
3606 * and per-cpu initialisations
3609 PAGE_ALIGN(size * sizeof(struct page)) >> PAGE_SHIFT;
3610 if (realsize >= memmap_pages) {
3611 realsize -= memmap_pages;
3614 " %s zone: %lu pages used for memmap\n",
3615 zone_names[j], memmap_pages);
3618 " %s zone: %lu pages exceeds realsize %lu\n",
3619 zone_names[j], memmap_pages, realsize);
3621 /* Account for reserved pages */
3622 if (j == 0 && realsize > dma_reserve) {
3623 realsize -= dma_reserve;
3624 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
3625 zone_names[0], dma_reserve);
3628 if (!is_highmem_idx(j))
3629 nr_kernel_pages += realsize;
3630 nr_all_pages += realsize;
3632 zone->spanned_pages = size;
3633 zone->present_pages = realsize;
3636 zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
3638 zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100;
3640 zone->name = zone_names[j];
3641 spin_lock_init(&zone->lock);
3642 spin_lock_init(&zone->lru_lock);
3643 zone_seqlock_init(zone);
3644 zone->zone_pgdat = pgdat;
3646 zone->prev_priority = DEF_PRIORITY;
3648 zone_pcp_init(zone);
3650 INIT_LIST_HEAD(&zone->lru[l].list);
3651 zone->lru[l].nr_scan = 0;
3653 zone->reclaim_stat.recent_rotated[0] = 0;
3654 zone->reclaim_stat.recent_rotated[1] = 0;
3655 zone->reclaim_stat.recent_scanned[0] = 0;
3656 zone->reclaim_stat.recent_scanned[1] = 0;
3657 zap_zone_vm_stats(zone);
3662 set_pageblock_order(pageblock_default_order());
3663 setup_usemap(pgdat, zone, size);
3664 ret = init_currently_empty_zone(zone, zone_start_pfn,
3665 size, MEMMAP_EARLY);
3667 memmap_init(size, nid, j, zone_start_pfn);
3668 zone_start_pfn += size;
3672 static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
3674 /* Skip empty nodes */
3675 if (!pgdat->node_spanned_pages)
3678 #ifdef CONFIG_FLAT_NODE_MEM_MAP
3679 /* ia64 gets its own node_mem_map, before this, without bootmem */
3680 if (!pgdat->node_mem_map) {
3681 unsigned long size, start, end;
3685 * The zone's endpoints aren't required to be MAX_ORDER
3686 * aligned but the node_mem_map endpoints must be in order
3687 * for the buddy allocator to function correctly.
3689 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
3690 end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
3691 end = ALIGN(end, MAX_ORDER_NR_PAGES);
3692 size = (end - start) * sizeof(struct page);
3693 map = alloc_remap(pgdat->node_id, size);
3695 map = alloc_bootmem_node(pgdat, size);
3696 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
3698 #ifndef CONFIG_NEED_MULTIPLE_NODES
3700 * With no DISCONTIG, the global mem_map is just set as node 0's
3702 if (pgdat == NODE_DATA(0)) {
3703 mem_map = NODE_DATA(0)->node_mem_map;
3704 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3705 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
3706 mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
3707 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3710 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
3713 void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
3714 unsigned long node_start_pfn, unsigned long *zholes_size)
3716 pg_data_t *pgdat = NODE_DATA(nid);
3718 pgdat->node_id = nid;
3719 pgdat->node_start_pfn = node_start_pfn;
3720 calculate_node_totalpages(pgdat, zones_size, zholes_size);
3722 alloc_node_mem_map(pgdat);
3723 #ifdef CONFIG_FLAT_NODE_MEM_MAP
3724 printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
3725 nid, (unsigned long)pgdat,
3726 (unsigned long)pgdat->node_mem_map);
3729 free_area_init_core(pgdat, zones_size, zholes_size);
3732 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3734 #if MAX_NUMNODES > 1
3736 * Figure out the number of possible node ids.
3738 static void __init setup_nr_node_ids(void)
3741 unsigned int highest = 0;
3743 for_each_node_mask(node, node_possible_map)
3745 nr_node_ids = highest + 1;
3748 static inline void setup_nr_node_ids(void)
3754 * add_active_range - Register a range of PFNs backed by physical memory
3755 * @nid: The node ID the range resides on
3756 * @start_pfn: The start PFN of the available physical memory
3757 * @end_pfn: The end PFN of the available physical memory
3759 * These ranges are stored in an early_node_map[] and later used by
3760 * free_area_init_nodes() to calculate zone sizes and holes. If the
3761 * range spans a memory hole, it is up to the architecture to ensure
3762 * the memory is not freed by the bootmem allocator. If possible
3763 * the range being registered will be merged with existing ranges.
3765 void __init add_active_range(unsigned int nid, unsigned long start_pfn,
3766 unsigned long end_pfn)
3770 mminit_dprintk(MMINIT_TRACE, "memory_register",
3771 "Entering add_active_range(%d, %#lx, %#lx) "
3772 "%d entries of %d used\n",
3773 nid, start_pfn, end_pfn,
3774 nr_nodemap_entries, MAX_ACTIVE_REGIONS);
3776 mminit_validate_memmodel_limits(&start_pfn, &end_pfn);
3778 /* Merge with existing active regions if possible */
3779 for (i = 0; i < nr_nodemap_entries; i++) {
3780 if (early_node_map[i].nid != nid)
3783 /* Skip if an existing region covers this new one */
3784 if (start_pfn >= early_node_map[i].start_pfn &&
3785 end_pfn <= early_node_map[i].end_pfn)
3788 /* Merge forward if suitable */
3789 if (start_pfn <= early_node_map[i].end_pfn &&
3790 end_pfn > early_node_map[i].end_pfn) {
3791 early_node_map[i].end_pfn = end_pfn;
3795 /* Merge backward if suitable */
3796 if (start_pfn < early_node_map[i].end_pfn &&
3797 end_pfn >= early_node_map[i].start_pfn) {
3798 early_node_map[i].start_pfn = start_pfn;
3803 /* Check that early_node_map is large enough */
3804 if (i >= MAX_ACTIVE_REGIONS) {
3805 printk(KERN_CRIT "More than %d memory regions, truncating\n",
3806 MAX_ACTIVE_REGIONS);
3810 early_node_map[i].nid = nid;
3811 early_node_map[i].start_pfn = start_pfn;
3812 early_node_map[i].end_pfn = end_pfn;
3813 nr_nodemap_entries = i + 1;
3817 * remove_active_range - Shrink an existing registered range of PFNs
3818 * @nid: The node id the range is on that should be shrunk
3819 * @start_pfn: The new PFN of the range
3820 * @end_pfn: The new PFN of the range
3822 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
3823 * The map is kept near the end physical page range that has already been
3824 * registered. This function allows an arch to shrink an existing registered
3827 void __init remove_active_range(unsigned int nid, unsigned long start_pfn,
3828 unsigned long end_pfn)
3833 printk(KERN_DEBUG "remove_active_range (%d, %lu, %lu)\n",
3834 nid, start_pfn, end_pfn);
3836 /* Find the old active region end and shrink */
3837 for_each_active_range_index_in_nid(i, nid) {
3838 if (early_node_map[i].start_pfn >= start_pfn &&
3839 early_node_map[i].end_pfn <= end_pfn) {
3841 early_node_map[i].start_pfn = 0;
3842 early_node_map[i].end_pfn = 0;
3846 if (early_node_map[i].start_pfn < start_pfn &&
3847 early_node_map[i].end_pfn > start_pfn) {
3848 unsigned long temp_end_pfn = early_node_map[i].end_pfn;
3849 early_node_map[i].end_pfn = start_pfn;
3850 if (temp_end_pfn > end_pfn)
3851 add_active_range(nid, end_pfn, temp_end_pfn);
3854 if (early_node_map[i].start_pfn >= start_pfn &&
3855 early_node_map[i].end_pfn > end_pfn &&
3856 early_node_map[i].start_pfn < end_pfn) {
3857 early_node_map[i].start_pfn = end_pfn;
3865 /* remove the blank ones */
3866 for (i = nr_nodemap_entries - 1; i > 0; i--) {
3867 if (early_node_map[i].nid != nid)
3869 if (early_node_map[i].end_pfn)
3871 /* we found it, get rid of it */
3872 for (j = i; j < nr_nodemap_entries - 1; j++)
3873 memcpy(&early_node_map[j], &early_node_map[j+1],
3874 sizeof(early_node_map[j]));
3875 j = nr_nodemap_entries - 1;
3876 memset(&early_node_map[j], 0, sizeof(early_node_map[j]));
3877 nr_nodemap_entries--;
3882 * remove_all_active_ranges - Remove all currently registered regions
3884 * During discovery, it may be found that a table like SRAT is invalid
3885 * and an alternative discovery method must be used. This function removes
3886 * all currently registered regions.
3888 void __init remove_all_active_ranges(void)
3890 memset(early_node_map, 0, sizeof(early_node_map));
3891 nr_nodemap_entries = 0;
3894 /* Compare two active node_active_regions */
3895 static int __init cmp_node_active_region(const void *a, const void *b)
3897 struct node_active_region *arange = (struct node_active_region *)a;
3898 struct node_active_region *brange = (struct node_active_region *)b;
3900 /* Done this way to avoid overflows */
3901 if (arange->start_pfn > brange->start_pfn)
3903 if (arange->start_pfn < brange->start_pfn)
3909 /* sort the node_map by start_pfn */
3910 static void __init sort_node_map(void)
3912 sort(early_node_map, (size_t)nr_nodemap_entries,
3913 sizeof(struct node_active_region),
3914 cmp_node_active_region, NULL);
3917 /* Find the lowest pfn for a node */
3918 static unsigned long __init find_min_pfn_for_node(int nid)
3921 unsigned long min_pfn = ULONG_MAX;
3923 /* Assuming a sorted map, the first range found has the starting pfn */
3924 for_each_active_range_index_in_nid(i, nid)
3925 min_pfn = min(min_pfn, early_node_map[i].start_pfn);
3927 if (min_pfn == ULONG_MAX) {
3929 "Could not find start_pfn for node %d\n", nid);
3937 * find_min_pfn_with_active_regions - Find the minimum PFN registered
3939 * It returns the minimum PFN based on information provided via
3940 * add_active_range().
3942 unsigned long __init find_min_pfn_with_active_regions(void)
3944 return find_min_pfn_for_node(MAX_NUMNODES);
3948 * early_calculate_totalpages()
3949 * Sum pages in active regions for movable zone.
3950 * Populate N_HIGH_MEMORY for calculating usable_nodes.
3952 static unsigned long __init early_calculate_totalpages(void)
3955 unsigned long totalpages = 0;
3957 for (i = 0; i < nr_nodemap_entries; i++) {
3958 unsigned long pages = early_node_map[i].end_pfn -
3959 early_node_map[i].start_pfn;
3960 totalpages += pages;
3962 node_set_state(early_node_map[i].nid, N_HIGH_MEMORY);
3968 * Find the PFN the Movable zone begins in each node. Kernel memory
3969 * is spread evenly between nodes as long as the nodes have enough
3970 * memory. When they don't, some nodes will have more kernelcore than
3973 static void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn)
3976 unsigned long usable_startpfn;
3977 unsigned long kernelcore_node, kernelcore_remaining;
3978 unsigned long totalpages = early_calculate_totalpages();
3979 int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]);
3982 * If movablecore was specified, calculate what size of
3983 * kernelcore that corresponds so that memory usable for
3984 * any allocation type is evenly spread. If both kernelcore
3985 * and movablecore are specified, then the value of kernelcore
3986 * will be used for required_kernelcore if it's greater than
3987 * what movablecore would have allowed.
3989 if (required_movablecore) {
3990 unsigned long corepages;
3993 * Round-up so that ZONE_MOVABLE is at least as large as what
3994 * was requested by the user
3996 required_movablecore =
3997 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
3998 corepages = totalpages - required_movablecore;
4000 required_kernelcore = max(required_kernelcore, corepages);
4003 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
4004 if (!required_kernelcore)
4007 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
4008 find_usable_zone_for_movable();
4009 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
4012 /* Spread kernelcore memory as evenly as possible throughout nodes */
4013 kernelcore_node = required_kernelcore / usable_nodes;
4014 for_each_node_state(nid, N_HIGH_MEMORY) {
4016 * Recalculate kernelcore_node if the division per node
4017 * now exceeds what is necessary to satisfy the requested
4018 * amount of memory for the kernel
4020 if (required_kernelcore < kernelcore_node)
4021 kernelcore_node = required_kernelcore / usable_nodes;
4024 * As the map is walked, we track how much memory is usable
4025 * by the kernel using kernelcore_remaining. When it is
4026 * 0, the rest of the node is usable by ZONE_MOVABLE
4028 kernelcore_remaining = kernelcore_node;
4030 /* Go through each range of PFNs within this node */
4031 for_each_active_range_index_in_nid(i, nid) {
4032 unsigned long start_pfn, end_pfn;
4033 unsigned long size_pages;
4035 start_pfn = max(early_node_map[i].start_pfn,
4036 zone_movable_pfn[nid]);
4037 end_pfn = early_node_map[i].end_pfn;
4038 if (start_pfn >= end_pfn)
4041 /* Account for what is only usable for kernelcore */
4042 if (start_pfn < usable_startpfn) {
4043 unsigned long kernel_pages;
4044 kernel_pages = min(end_pfn, usable_startpfn)
4047 kernelcore_remaining -= min(kernel_pages,
4048 kernelcore_remaining);
4049 required_kernelcore -= min(kernel_pages,
4050 required_kernelcore);
4052 /* Continue if range is now fully accounted */
4053 if (end_pfn <= usable_startpfn) {
4056 * Push zone_movable_pfn to the end so
4057 * that if we have to rebalance
4058 * kernelcore across nodes, we will
4059 * not double account here
4061 zone_movable_pfn[nid] = end_pfn;
4064 start_pfn = usable_startpfn;
4068 * The usable PFN range for ZONE_MOVABLE is from
4069 * start_pfn->end_pfn. Calculate size_pages as the
4070 * number of pages used as kernelcore
4072 size_pages = end_pfn - start_pfn;
4073 if (size_pages > kernelcore_remaining)
4074 size_pages = kernelcore_remaining;
4075 zone_movable_pfn[nid] = start_pfn + size_pages;
4078 * Some kernelcore has been met, update counts and
4079 * break if the kernelcore for this node has been
4082 required_kernelcore -= min(required_kernelcore,
4084 kernelcore_remaining -= size_pages;
4085 if (!kernelcore_remaining)
4091 * If there is still required_kernelcore, we do another pass with one
4092 * less node in the count. This will push zone_movable_pfn[nid] further
4093 * along on the nodes that still have memory until kernelcore is
4097 if (usable_nodes && required_kernelcore > usable_nodes)
4100 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
4101 for (nid = 0; nid < MAX_NUMNODES; nid++)
4102 zone_movable_pfn[nid] =
4103 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
4106 /* Any regular memory on that node ? */
4107 static void check_for_regular_memory(pg_data_t *pgdat)
4109 #ifdef CONFIG_HIGHMEM
4110 enum zone_type zone_type;
4112 for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) {
4113 struct zone *zone = &pgdat->node_zones[zone_type];
4114 if (zone->present_pages)
4115 node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY);
4121 * free_area_init_nodes - Initialise all pg_data_t and zone data
4122 * @max_zone_pfn: an array of max PFNs for each zone
4124 * This will call free_area_init_node() for each active node in the system.
4125 * Using the page ranges provided by add_active_range(), the size of each
4126 * zone in each node and their holes is calculated. If the maximum PFN
4127 * between two adjacent zones match, it is assumed that the zone is empty.
4128 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
4129 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
4130 * starts where the previous one ended. For example, ZONE_DMA32 starts
4131 * at arch_max_dma_pfn.
4133 void __init free_area_init_nodes(unsigned long *max_zone_pfn)
4138 /* Sort early_node_map as initialisation assumes it is sorted */
4141 /* Record where the zone boundaries are */
4142 memset(arch_zone_lowest_possible_pfn, 0,
4143 sizeof(arch_zone_lowest_possible_pfn));
4144 memset(arch_zone_highest_possible_pfn, 0,
4145 sizeof(arch_zone_highest_possible_pfn));
4146 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
4147 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
4148 for (i = 1; i < MAX_NR_ZONES; i++) {
4149 if (i == ZONE_MOVABLE)
4151 arch_zone_lowest_possible_pfn[i] =
4152 arch_zone_highest_possible_pfn[i-1];
4153 arch_zone_highest_possible_pfn[i] =
4154 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
4156 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
4157 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
4159 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
4160 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
4161 find_zone_movable_pfns_for_nodes(zone_movable_pfn);
4163 /* Print out the zone ranges */
4164 printk("Zone PFN ranges:\n");
4165 for (i = 0; i < MAX_NR_ZONES; i++) {
4166 if (i == ZONE_MOVABLE)
4168 printk(" %-8s %0#10lx -> %0#10lx\n",
4170 arch_zone_lowest_possible_pfn[i],
4171 arch_zone_highest_possible_pfn[i]);
4174 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
4175 printk("Movable zone start PFN for each node\n");
4176 for (i = 0; i < MAX_NUMNODES; i++) {
4177 if (zone_movable_pfn[i])
4178 printk(" Node %d: %lu\n", i, zone_movable_pfn[i]);
4181 /* Print out the early_node_map[] */
4182 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries);
4183 for (i = 0; i < nr_nodemap_entries; i++)
4184 printk(" %3d: %0#10lx -> %0#10lx\n", early_node_map[i].nid,
4185 early_node_map[i].start_pfn,
4186 early_node_map[i].end_pfn);
4188 /* Initialise every node */
4189 mminit_verify_pageflags_layout();
4190 setup_nr_node_ids();
4191 for_each_online_node(nid) {
4192 pg_data_t *pgdat = NODE_DATA(nid);
4193 free_area_init_node(nid, NULL,
4194 find_min_pfn_for_node(nid), NULL);
4196 /* Any memory on that node */
4197 if (pgdat->node_present_pages)
4198 node_set_state(nid, N_HIGH_MEMORY);
4199 check_for_regular_memory(pgdat);
4203 static int __init cmdline_parse_core(char *p, unsigned long *core)
4205 unsigned long long coremem;
4209 coremem = memparse(p, &p);
4210 *core = coremem >> PAGE_SHIFT;
4212 /* Paranoid check that UL is enough for the coremem value */
4213 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
4219 * kernelcore=size sets the amount of memory for use for allocations that
4220 * cannot be reclaimed or migrated.
4222 static int __init cmdline_parse_kernelcore(char *p)
4224 return cmdline_parse_core(p, &required_kernelcore);
4228 * movablecore=size sets the amount of memory for use for allocations that
4229 * can be reclaimed or migrated.
4231 static int __init cmdline_parse_movablecore(char *p)
4233 return cmdline_parse_core(p, &required_movablecore);
4236 early_param("kernelcore", cmdline_parse_kernelcore);
4237 early_param("movablecore", cmdline_parse_movablecore);
4239 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4242 * set_dma_reserve - set the specified number of pages reserved in the first zone
4243 * @new_dma_reserve: The number of pages to mark reserved
4245 * The per-cpu batchsize and zone watermarks are determined by present_pages.
4246 * In the DMA zone, a significant percentage may be consumed by kernel image
4247 * and other unfreeable allocations which can skew the watermarks badly. This
4248 * function may optionally be used to account for unfreeable pages in the
4249 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
4250 * smaller per-cpu batchsize.
4252 void __init set_dma_reserve(unsigned long new_dma_reserve)
4254 dma_reserve = new_dma_reserve;
4257 #ifndef CONFIG_NEED_MULTIPLE_NODES
4258 struct pglist_data __refdata contig_page_data = { .bdata = &bootmem_node_data[0] };
4259 EXPORT_SYMBOL(contig_page_data);
4262 void __init free_area_init(unsigned long *zones_size)
4264 free_area_init_node(0, zones_size,
4265 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
4268 static int page_alloc_cpu_notify(struct notifier_block *self,
4269 unsigned long action, void *hcpu)
4271 int cpu = (unsigned long)hcpu;
4273 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
4277 * Spill the event counters of the dead processor
4278 * into the current processors event counters.
4279 * This artificially elevates the count of the current
4282 vm_events_fold_cpu(cpu);
4285 * Zero the differential counters of the dead processor
4286 * so that the vm statistics are consistent.
4288 * This is only okay since the processor is dead and cannot
4289 * race with what we are doing.
4291 refresh_cpu_vm_stats(cpu);
4296 void __init page_alloc_init(void)
4298 hotcpu_notifier(page_alloc_cpu_notify, 0);
4302 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
4303 * or min_free_kbytes changes.
4305 static void calculate_totalreserve_pages(void)
4307 struct pglist_data *pgdat;
4308 unsigned long reserve_pages = 0;
4309 enum zone_type i, j;
4311 for_each_online_pgdat(pgdat) {
4312 for (i = 0; i < MAX_NR_ZONES; i++) {
4313 struct zone *zone = pgdat->node_zones + i;
4314 unsigned long max = 0;
4316 /* Find valid and maximum lowmem_reserve in the zone */
4317 for (j = i; j < MAX_NR_ZONES; j++) {
4318 if (zone->lowmem_reserve[j] > max)
4319 max = zone->lowmem_reserve[j];
4322 /* we treat pages_high as reserved pages. */
4323 max += zone->pages_high;
4325 if (max > zone->present_pages)
4326 max = zone->present_pages;
4327 reserve_pages += max;
4330 totalreserve_pages = reserve_pages;
4334 * setup_per_zone_lowmem_reserve - called whenever
4335 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
4336 * has a correct pages reserved value, so an adequate number of
4337 * pages are left in the zone after a successful __alloc_pages().
4339 static void setup_per_zone_lowmem_reserve(void)
4341 struct pglist_data *pgdat;
4342 enum zone_type j, idx;
4344 for_each_online_pgdat(pgdat) {
4345 for (j = 0; j < MAX_NR_ZONES; j++) {
4346 struct zone *zone = pgdat->node_zones + j;
4347 unsigned long present_pages = zone->present_pages;
4349 zone->lowmem_reserve[j] = 0;
4353 struct zone *lower_zone;
4357 if (sysctl_lowmem_reserve_ratio[idx] < 1)
4358 sysctl_lowmem_reserve_ratio[idx] = 1;
4360 lower_zone = pgdat->node_zones + idx;
4361 lower_zone->lowmem_reserve[j] = present_pages /
4362 sysctl_lowmem_reserve_ratio[idx];
4363 present_pages += lower_zone->present_pages;
4368 /* update totalreserve_pages */
4369 calculate_totalreserve_pages();
4373 * setup_per_zone_pages_min - called when min_free_kbytes changes.
4375 * Ensures that the pages_{min,low,high} values for each zone are set correctly
4376 * with respect to min_free_kbytes.
4378 void setup_per_zone_pages_min(void)
4380 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
4381 unsigned long lowmem_pages = 0;
4383 unsigned long flags;
4385 /* Calculate total number of !ZONE_HIGHMEM pages */
4386 for_each_zone(zone) {
4387 if (!is_highmem(zone))
4388 lowmem_pages += zone->present_pages;
4391 for_each_zone(zone) {
4394 spin_lock_irqsave(&zone->lock, flags);
4395 tmp = (u64)pages_min * zone->present_pages;
4396 do_div(tmp, lowmem_pages);
4397 if (is_highmem(zone)) {
4399 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
4400 * need highmem pages, so cap pages_min to a small
4403 * The (pages_high-pages_low) and (pages_low-pages_min)
4404 * deltas controls asynch page reclaim, and so should
4405 * not be capped for highmem.
4409 min_pages = zone->present_pages / 1024;
4410 if (min_pages < SWAP_CLUSTER_MAX)
4411 min_pages = SWAP_CLUSTER_MAX;
4412 if (min_pages > 128)
4414 zone->pages_min = min_pages;
4417 * If it's a lowmem zone, reserve a number of pages
4418 * proportionate to the zone's size.
4420 zone->pages_min = tmp;
4423 zone->pages_low = zone->pages_min + (tmp >> 2);
4424 zone->pages_high = zone->pages_min + (tmp >> 1);
4425 setup_zone_migrate_reserve(zone);
4426 spin_unlock_irqrestore(&zone->lock, flags);
4429 /* update totalreserve_pages */
4430 calculate_totalreserve_pages();
4434 * setup_per_zone_inactive_ratio - called when min_free_kbytes changes.
4436 * The inactive anon list should be small enough that the VM never has to
4437 * do too much work, but large enough that each inactive page has a chance
4438 * to be referenced again before it is swapped out.
4440 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
4441 * INACTIVE_ANON pages on this zone's LRU, maintained by the
4442 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
4443 * the anonymous pages are kept on the inactive list.
4446 * memory ratio inactive anon
4447 * -------------------------------------
4456 static void setup_per_zone_inactive_ratio(void)
4460 for_each_zone(zone) {
4461 unsigned int gb, ratio;
4463 /* Zone size in gigabytes */
4464 gb = zone->present_pages >> (30 - PAGE_SHIFT);
4465 ratio = int_sqrt(10 * gb);
4469 zone->inactive_ratio = ratio;
4474 * Initialise min_free_kbytes.
4476 * For small machines we want it small (128k min). For large machines
4477 * we want it large (64MB max). But it is not linear, because network
4478 * bandwidth does not increase linearly with machine size. We use
4480 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
4481 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
4497 static int __init init_per_zone_pages_min(void)
4499 unsigned long lowmem_kbytes;
4501 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
4503 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
4504 if (min_free_kbytes < 128)
4505 min_free_kbytes = 128;
4506 if (min_free_kbytes > 65536)
4507 min_free_kbytes = 65536;
4508 setup_per_zone_pages_min();
4509 setup_per_zone_lowmem_reserve();
4510 setup_per_zone_inactive_ratio();
4513 module_init(init_per_zone_pages_min)
4516 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
4517 * that we can call two helper functions whenever min_free_kbytes
4520 int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
4521 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4523 proc_dointvec(table, write, file, buffer, length, ppos);
4525 setup_per_zone_pages_min();
4530 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
4531 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4536 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4541 zone->min_unmapped_pages = (zone->present_pages *
4542 sysctl_min_unmapped_ratio) / 100;
4546 int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
4547 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4552 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4557 zone->min_slab_pages = (zone->present_pages *
4558 sysctl_min_slab_ratio) / 100;
4564 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
4565 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
4566 * whenever sysctl_lowmem_reserve_ratio changes.
4568 * The reserve ratio obviously has absolutely no relation with the
4569 * pages_min watermarks. The lowmem reserve ratio can only make sense
4570 * if in function of the boot time zone sizes.
4572 int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
4573 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4575 proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4576 setup_per_zone_lowmem_reserve();
4581 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
4582 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
4583 * can have before it gets flushed back to buddy allocator.
4586 int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
4587 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4593 ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4594 if (!write || (ret == -EINVAL))
4596 for_each_zone(zone) {
4597 for_each_online_cpu(cpu) {
4599 high = zone->present_pages / percpu_pagelist_fraction;
4600 setup_pagelist_highmark(zone_pcp(zone, cpu), high);
4606 int hashdist = HASHDIST_DEFAULT;
4609 static int __init set_hashdist(char *str)
4613 hashdist = simple_strtoul(str, &str, 0);
4616 __setup("hashdist=", set_hashdist);
4620 * allocate a large system hash table from bootmem
4621 * - it is assumed that the hash table must contain an exact power-of-2
4622 * quantity of entries
4623 * - limit is the number of hash buckets, not the total allocation size
4625 void *__init alloc_large_system_hash(const char *tablename,
4626 unsigned long bucketsize,
4627 unsigned long numentries,
4630 unsigned int *_hash_shift,
4631 unsigned int *_hash_mask,
4632 unsigned long limit)
4634 unsigned long long max = limit;
4635 unsigned long log2qty, size;
4638 /* allow the kernel cmdline to have a say */
4640 /* round applicable memory size up to nearest megabyte */
4641 numentries = nr_kernel_pages;
4642 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
4643 numentries >>= 20 - PAGE_SHIFT;
4644 numentries <<= 20 - PAGE_SHIFT;
4646 /* limit to 1 bucket per 2^scale bytes of low memory */
4647 if (scale > PAGE_SHIFT)
4648 numentries >>= (scale - PAGE_SHIFT);
4650 numentries <<= (PAGE_SHIFT - scale);
4652 /* Make sure we've got at least a 0-order allocation.. */
4653 if (unlikely((numentries * bucketsize) < PAGE_SIZE))
4654 numentries = PAGE_SIZE / bucketsize;
4656 numentries = roundup_pow_of_two(numentries);
4658 /* limit allocation size to 1/16 total memory by default */
4660 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
4661 do_div(max, bucketsize);
4664 if (numentries > max)
4667 log2qty = ilog2(numentries);
4670 size = bucketsize << log2qty;
4671 if (flags & HASH_EARLY)
4672 table = alloc_bootmem_nopanic(size);
4674 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
4676 unsigned long order = get_order(size);
4678 if (order < MAX_ORDER)
4679 table = (void *)__get_free_pages(GFP_ATOMIC,
4682 * If bucketsize is not a power-of-two, we may free
4683 * some pages at the end of hash table.
4686 unsigned long alloc_end = (unsigned long)table +
4687 (PAGE_SIZE << order);
4688 unsigned long used = (unsigned long)table +
4690 split_page(virt_to_page(table), order);
4691 while (used < alloc_end) {
4697 } while (!table && size > PAGE_SIZE && --log2qty);
4700 panic("Failed to allocate %s hash table\n", tablename);
4702 printk(KERN_INFO "%s hash table entries: %d (order: %d, %lu bytes)\n",
4705 ilog2(size) - PAGE_SHIFT,
4709 *_hash_shift = log2qty;
4711 *_hash_mask = (1 << log2qty) - 1;
4714 * If hashdist is set, the table allocation is done with __vmalloc()
4715 * which invokes the kmemleak_alloc() callback. This function may also
4716 * be called before the slab and kmemleak are initialised when
4717 * kmemleak simply buffers the request to be executed later
4718 * (GFP_ATOMIC flag ignored in this case).
4721 kmemleak_alloc(table, size, 1, GFP_ATOMIC);
4726 /* Return a pointer to the bitmap storing bits affecting a block of pages */
4727 static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
4730 #ifdef CONFIG_SPARSEMEM
4731 return __pfn_to_section(pfn)->pageblock_flags;
4733 return zone->pageblock_flags;
4734 #endif /* CONFIG_SPARSEMEM */
4737 static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
4739 #ifdef CONFIG_SPARSEMEM
4740 pfn &= (PAGES_PER_SECTION-1);
4741 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
4743 pfn = pfn - zone->zone_start_pfn;
4744 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
4745 #endif /* CONFIG_SPARSEMEM */
4749 * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
4750 * @page: The page within the block of interest
4751 * @start_bitidx: The first bit of interest to retrieve
4752 * @end_bitidx: The last bit of interest
4753 * returns pageblock_bits flags
4755 unsigned long get_pageblock_flags_group(struct page *page,
4756 int start_bitidx, int end_bitidx)
4759 unsigned long *bitmap;
4760 unsigned long pfn, bitidx;
4761 unsigned long flags = 0;
4762 unsigned long value = 1;
4764 zone = page_zone(page);
4765 pfn = page_to_pfn(page);
4766 bitmap = get_pageblock_bitmap(zone, pfn);
4767 bitidx = pfn_to_bitidx(zone, pfn);
4769 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
4770 if (test_bit(bitidx + start_bitidx, bitmap))
4777 * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
4778 * @page: The page within the block of interest
4779 * @start_bitidx: The first bit of interest
4780 * @end_bitidx: The last bit of interest
4781 * @flags: The flags to set
4783 void set_pageblock_flags_group(struct page *page, unsigned long flags,
4784 int start_bitidx, int end_bitidx)
4787 unsigned long *bitmap;
4788 unsigned long pfn, bitidx;
4789 unsigned long value = 1;
4791 zone = page_zone(page);
4792 pfn = page_to_pfn(page);
4793 bitmap = get_pageblock_bitmap(zone, pfn);
4794 bitidx = pfn_to_bitidx(zone, pfn);
4795 VM_BUG_ON(pfn < zone->zone_start_pfn);
4796 VM_BUG_ON(pfn >= zone->zone_start_pfn + zone->spanned_pages);
4798 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
4800 __set_bit(bitidx + start_bitidx, bitmap);
4802 __clear_bit(bitidx + start_bitidx, bitmap);
4806 * This is designed as sub function...plz see page_isolation.c also.
4807 * set/clear page block's type to be ISOLATE.
4808 * page allocater never alloc memory from ISOLATE block.
4811 int set_migratetype_isolate(struct page *page)
4814 unsigned long flags;
4817 zone = page_zone(page);
4818 spin_lock_irqsave(&zone->lock, flags);
4820 * In future, more migrate types will be able to be isolation target.
4822 if (get_pageblock_migratetype(page) != MIGRATE_MOVABLE)
4824 set_pageblock_migratetype(page, MIGRATE_ISOLATE);
4825 move_freepages_block(zone, page, MIGRATE_ISOLATE);
4828 spin_unlock_irqrestore(&zone->lock, flags);
4834 void unset_migratetype_isolate(struct page *page)
4837 unsigned long flags;
4838 zone = page_zone(page);
4839 spin_lock_irqsave(&zone->lock, flags);
4840 if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE)
4842 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
4843 move_freepages_block(zone, page, MIGRATE_MOVABLE);
4845 spin_unlock_irqrestore(&zone->lock, flags);
4848 #ifdef CONFIG_MEMORY_HOTREMOVE
4850 * All pages in the range must be isolated before calling this.
4853 __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
4859 unsigned long flags;
4860 /* find the first valid pfn */
4861 for (pfn = start_pfn; pfn < end_pfn; pfn++)
4866 zone = page_zone(pfn_to_page(pfn));
4867 spin_lock_irqsave(&zone->lock, flags);
4869 while (pfn < end_pfn) {
4870 if (!pfn_valid(pfn)) {
4874 page = pfn_to_page(pfn);
4875 BUG_ON(page_count(page));
4876 BUG_ON(!PageBuddy(page));
4877 order = page_order(page);
4878 #ifdef CONFIG_DEBUG_VM
4879 printk(KERN_INFO "remove from free list %lx %d %lx\n",
4880 pfn, 1 << order, end_pfn);
4882 list_del(&page->lru);
4883 rmv_page_order(page);
4884 zone->free_area[order].nr_free--;
4885 __mod_zone_page_state(zone, NR_FREE_PAGES,
4887 for (i = 0; i < (1 << order); i++)
4888 SetPageReserved((page+i));
4889 pfn += (1 << order);
4891 spin_unlock_irqrestore(&zone->lock, flags);