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 free_page_mlock(page);
499 if (unlikely(page_mapcount(page) |
500 (page->mapping != NULL) |
501 (page_count(page) != 0) |
502 (page->flags & PAGE_FLAGS_CHECK_AT_FREE))) {
506 if (page->flags & PAGE_FLAGS_CHECK_AT_PREP)
507 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
512 * Frees a list of pages.
513 * Assumes all pages on list are in same zone, and of same order.
514 * count is the number of pages to free.
516 * If the zone was previously in an "all pages pinned" state then look to
517 * see if this freeing clears that state.
519 * And clear the zone's pages_scanned counter, to hold off the "all pages are
520 * pinned" detection logic.
522 static void free_pages_bulk(struct zone *zone, int count,
523 struct list_head *list, int order)
525 spin_lock(&zone->lock);
526 zone_clear_flag(zone, ZONE_ALL_UNRECLAIMABLE);
527 zone->pages_scanned = 0;
531 VM_BUG_ON(list_empty(list));
532 page = list_entry(list->prev, struct page, lru);
533 /* have to delete it as __free_one_page list manipulates */
534 list_del(&page->lru);
535 __free_one_page(page, zone, order, page_private(page));
537 spin_unlock(&zone->lock);
540 static void free_one_page(struct zone *zone, struct page *page, int order,
543 spin_lock(&zone->lock);
544 zone_clear_flag(zone, ZONE_ALL_UNRECLAIMABLE);
545 zone->pages_scanned = 0;
546 __free_one_page(page, zone, order, migratetype);
547 spin_unlock(&zone->lock);
550 static void __free_pages_ok(struct page *page, unsigned int order)
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 __count_vm_events(PGFREE, 1 << order);
571 free_one_page(page_zone(page), page, order,
572 get_pageblock_migratetype(page));
573 local_irq_restore(flags);
577 * permit the bootmem allocator to evade page validation on high-order frees
579 void __meminit __free_pages_bootmem(struct page *page, unsigned int order)
582 __ClearPageReserved(page);
583 set_page_count(page, 0);
584 set_page_refcounted(page);
590 for (loop = 0; loop < BITS_PER_LONG; loop++) {
591 struct page *p = &page[loop];
593 if (loop + 1 < BITS_PER_LONG)
595 __ClearPageReserved(p);
596 set_page_count(p, 0);
599 set_page_refcounted(page);
600 __free_pages(page, order);
606 * The order of subdivision here is critical for the IO subsystem.
607 * Please do not alter this order without good reasons and regression
608 * testing. Specifically, as large blocks of memory are subdivided,
609 * the order in which smaller blocks are delivered depends on the order
610 * they're subdivided in this function. This is the primary factor
611 * influencing the order in which pages are delivered to the IO
612 * subsystem according to empirical testing, and this is also justified
613 * by considering the behavior of a buddy system containing a single
614 * large block of memory acted on by a series of small allocations.
615 * This behavior is a critical factor in sglist merging's success.
619 static inline void expand(struct zone *zone, struct page *page,
620 int low, int high, struct free_area *area,
623 unsigned long size = 1 << high;
629 VM_BUG_ON(bad_range(zone, &page[size]));
630 list_add(&page[size].lru, &area->free_list[migratetype]);
632 set_page_order(&page[size], high);
637 * This page is about to be returned from the page allocator
639 static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
641 if (unlikely(page_mapcount(page) |
642 (page->mapping != NULL) |
643 (page_count(page) != 0) |
644 (page->flags & PAGE_FLAGS_CHECK_AT_PREP))) {
649 set_page_private(page, 0);
650 set_page_refcounted(page);
652 arch_alloc_page(page, order);
653 kernel_map_pages(page, 1 << order, 1);
655 if (gfp_flags & __GFP_ZERO)
656 prep_zero_page(page, order, gfp_flags);
658 if (order && (gfp_flags & __GFP_COMP))
659 prep_compound_page(page, order);
665 * Go through the free lists for the given migratetype and remove
666 * the smallest available page from the freelists
669 struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
672 unsigned int current_order;
673 struct free_area * area;
676 /* Find a page of the appropriate size in the preferred list */
677 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
678 area = &(zone->free_area[current_order]);
679 if (list_empty(&area->free_list[migratetype]))
682 page = list_entry(area->free_list[migratetype].next,
684 list_del(&page->lru);
685 rmv_page_order(page);
687 __mod_zone_page_state(zone, NR_FREE_PAGES, - (1UL << order));
688 expand(zone, page, order, current_order, area, migratetype);
697 * This array describes the order lists are fallen back to when
698 * the free lists for the desirable migrate type are depleted
700 static int fallbacks[MIGRATE_TYPES][MIGRATE_TYPES-1] = {
701 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
702 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
703 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
704 [MIGRATE_RESERVE] = { MIGRATE_RESERVE, MIGRATE_RESERVE, MIGRATE_RESERVE }, /* Never used */
708 * Move the free pages in a range to the free lists of the requested type.
709 * Note that start_page and end_pages are not aligned on a pageblock
710 * boundary. If alignment is required, use move_freepages_block()
712 static int move_freepages(struct zone *zone,
713 struct page *start_page, struct page *end_page,
720 #ifndef CONFIG_HOLES_IN_ZONE
722 * page_zone is not safe to call in this context when
723 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
724 * anyway as we check zone boundaries in move_freepages_block().
725 * Remove at a later date when no bug reports exist related to
726 * grouping pages by mobility
728 BUG_ON(page_zone(start_page) != page_zone(end_page));
731 for (page = start_page; page <= end_page;) {
732 /* Make sure we are not inadvertently changing nodes */
733 VM_BUG_ON(page_to_nid(page) != zone_to_nid(zone));
735 if (!pfn_valid_within(page_to_pfn(page))) {
740 if (!PageBuddy(page)) {
745 order = page_order(page);
746 list_del(&page->lru);
748 &zone->free_area[order].free_list[migratetype]);
750 pages_moved += 1 << order;
756 static int move_freepages_block(struct zone *zone, struct page *page,
759 unsigned long start_pfn, end_pfn;
760 struct page *start_page, *end_page;
762 start_pfn = page_to_pfn(page);
763 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
764 start_page = pfn_to_page(start_pfn);
765 end_page = start_page + pageblock_nr_pages - 1;
766 end_pfn = start_pfn + pageblock_nr_pages - 1;
768 /* Do not cross zone boundaries */
769 if (start_pfn < zone->zone_start_pfn)
771 if (end_pfn >= zone->zone_start_pfn + zone->spanned_pages)
774 return move_freepages(zone, start_page, end_page, migratetype);
777 /* Remove an element from the buddy allocator from the fallback list */
778 static inline struct page *
779 __rmqueue_fallback(struct zone *zone, int order, int start_migratetype)
781 struct free_area * area;
786 /* Find the largest possible block of pages in the other list */
787 for (current_order = MAX_ORDER-1; current_order >= order;
789 for (i = 0; i < MIGRATE_TYPES - 1; i++) {
790 migratetype = fallbacks[start_migratetype][i];
792 /* MIGRATE_RESERVE handled later if necessary */
793 if (migratetype == MIGRATE_RESERVE)
796 area = &(zone->free_area[current_order]);
797 if (list_empty(&area->free_list[migratetype]))
800 page = list_entry(area->free_list[migratetype].next,
805 * If breaking a large block of pages, move all free
806 * pages to the preferred allocation list. If falling
807 * back for a reclaimable kernel allocation, be more
808 * agressive about taking ownership of free pages
810 if (unlikely(current_order >= (pageblock_order >> 1)) ||
811 start_migratetype == MIGRATE_RECLAIMABLE) {
813 pages = move_freepages_block(zone, page,
816 /* Claim the whole block if over half of it is free */
817 if (pages >= (1 << (pageblock_order-1)))
818 set_pageblock_migratetype(page,
821 migratetype = start_migratetype;
824 /* Remove the page from the freelists */
825 list_del(&page->lru);
826 rmv_page_order(page);
827 __mod_zone_page_state(zone, NR_FREE_PAGES,
830 if (current_order == pageblock_order)
831 set_pageblock_migratetype(page,
834 expand(zone, page, order, current_order, area, migratetype);
843 * Do the hard work of removing an element from the buddy allocator.
844 * Call me with the zone->lock already held.
846 static struct page *__rmqueue(struct zone *zone, unsigned int order,
852 page = __rmqueue_smallest(zone, order, migratetype);
854 if (unlikely(!page) && migratetype != MIGRATE_RESERVE) {
855 page = __rmqueue_fallback(zone, order, migratetype);
858 * Use MIGRATE_RESERVE rather than fail an allocation. goto
859 * is used because __rmqueue_smallest is an inline function
860 * and we want just one call site
863 migratetype = MIGRATE_RESERVE;
872 * Obtain a specified number of elements from the buddy allocator, all under
873 * a single hold of the lock, for efficiency. Add them to the supplied list.
874 * Returns the number of new pages which were placed at *list.
876 static int rmqueue_bulk(struct zone *zone, unsigned int order,
877 unsigned long count, struct list_head *list,
882 spin_lock(&zone->lock);
883 for (i = 0; i < count; ++i) {
884 struct page *page = __rmqueue(zone, order, migratetype);
885 if (unlikely(page == NULL))
889 * Split buddy pages returned by expand() are received here
890 * in physical page order. The page is added to the callers and
891 * list and the list head then moves forward. From the callers
892 * perspective, the linked list is ordered by page number in
893 * some conditions. This is useful for IO devices that can
894 * merge IO requests if the physical pages are ordered
897 list_add(&page->lru, list);
898 set_page_private(page, migratetype);
901 spin_unlock(&zone->lock);
907 * Called from the vmstat counter updater to drain pagesets of this
908 * currently executing processor on remote nodes after they have
911 * Note that this function must be called with the thread pinned to
912 * a single processor.
914 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
919 local_irq_save(flags);
920 if (pcp->count >= pcp->batch)
921 to_drain = pcp->batch;
923 to_drain = pcp->count;
924 free_pages_bulk(zone, to_drain, &pcp->list, 0);
925 pcp->count -= to_drain;
926 local_irq_restore(flags);
931 * Drain pages of the indicated processor.
933 * The processor must either be the current processor and the
934 * thread pinned to the current processor or a processor that
937 static void drain_pages(unsigned int cpu)
942 for_each_populated_zone(zone) {
943 struct per_cpu_pageset *pset;
944 struct per_cpu_pages *pcp;
946 pset = zone_pcp(zone, cpu);
949 local_irq_save(flags);
950 free_pages_bulk(zone, pcp->count, &pcp->list, 0);
952 local_irq_restore(flags);
957 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
959 void drain_local_pages(void *arg)
961 drain_pages(smp_processor_id());
965 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
967 void drain_all_pages(void)
969 on_each_cpu(drain_local_pages, NULL, 1);
972 #ifdef CONFIG_HIBERNATION
974 void mark_free_pages(struct zone *zone)
976 unsigned long pfn, max_zone_pfn;
979 struct list_head *curr;
981 if (!zone->spanned_pages)
984 spin_lock_irqsave(&zone->lock, flags);
986 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
987 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
988 if (pfn_valid(pfn)) {
989 struct page *page = pfn_to_page(pfn);
991 if (!swsusp_page_is_forbidden(page))
992 swsusp_unset_page_free(page);
995 for_each_migratetype_order(order, t) {
996 list_for_each(curr, &zone->free_area[order].free_list[t]) {
999 pfn = page_to_pfn(list_entry(curr, struct page, lru));
1000 for (i = 0; i < (1UL << order); i++)
1001 swsusp_set_page_free(pfn_to_page(pfn + i));
1004 spin_unlock_irqrestore(&zone->lock, flags);
1006 #endif /* CONFIG_PM */
1009 * Free a 0-order page
1011 static void free_hot_cold_page(struct page *page, int cold)
1013 struct zone *zone = page_zone(page);
1014 struct per_cpu_pages *pcp;
1015 unsigned long flags;
1018 page->mapping = NULL;
1019 if (free_pages_check(page))
1022 if (!PageHighMem(page)) {
1023 debug_check_no_locks_freed(page_address(page), PAGE_SIZE);
1024 debug_check_no_obj_freed(page_address(page), PAGE_SIZE);
1026 arch_free_page(page, 0);
1027 kernel_map_pages(page, 1, 0);
1029 pcp = &zone_pcp(zone, get_cpu())->pcp;
1030 local_irq_save(flags);
1031 __count_vm_event(PGFREE);
1033 list_add_tail(&page->lru, &pcp->list);
1035 list_add(&page->lru, &pcp->list);
1036 set_page_private(page, get_pageblock_migratetype(page));
1038 if (pcp->count >= pcp->high) {
1039 free_pages_bulk(zone, pcp->batch, &pcp->list, 0);
1040 pcp->count -= pcp->batch;
1042 local_irq_restore(flags);
1046 void free_hot_page(struct page *page)
1048 free_hot_cold_page(page, 0);
1051 void free_cold_page(struct page *page)
1053 free_hot_cold_page(page, 1);
1057 * split_page takes a non-compound higher-order page, and splits it into
1058 * n (1<<order) sub-pages: page[0..n]
1059 * Each sub-page must be freed individually.
1061 * Note: this is probably too low level an operation for use in drivers.
1062 * Please consult with lkml before using this in your driver.
1064 void split_page(struct page *page, unsigned int order)
1068 VM_BUG_ON(PageCompound(page));
1069 VM_BUG_ON(!page_count(page));
1070 for (i = 1; i < (1 << order); i++)
1071 set_page_refcounted(page + i);
1075 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1076 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1080 struct page *buffered_rmqueue(struct zone *preferred_zone,
1081 struct zone *zone, int order, gfp_t gfp_flags,
1084 unsigned long flags;
1086 int cold = !!(gfp_flags & __GFP_COLD);
1091 if (likely(order == 0)) {
1092 struct per_cpu_pages *pcp;
1094 pcp = &zone_pcp(zone, cpu)->pcp;
1095 local_irq_save(flags);
1097 pcp->count = rmqueue_bulk(zone, 0,
1098 pcp->batch, &pcp->list, migratetype);
1099 if (unlikely(!pcp->count))
1103 /* Find a page of the appropriate migrate type */
1105 list_for_each_entry_reverse(page, &pcp->list, lru)
1106 if (page_private(page) == migratetype)
1109 list_for_each_entry(page, &pcp->list, lru)
1110 if (page_private(page) == migratetype)
1114 /* Allocate more to the pcp list if necessary */
1115 if (unlikely(&page->lru == &pcp->list)) {
1116 pcp->count += rmqueue_bulk(zone, 0,
1117 pcp->batch, &pcp->list, migratetype);
1118 page = list_entry(pcp->list.next, struct page, lru);
1121 list_del(&page->lru);
1124 spin_lock_irqsave(&zone->lock, flags);
1125 page = __rmqueue(zone, order, migratetype);
1126 spin_unlock(&zone->lock);
1131 __count_zone_vm_events(PGALLOC, zone, 1 << order);
1132 zone_statistics(preferred_zone, zone);
1133 local_irq_restore(flags);
1136 VM_BUG_ON(bad_range(zone, page));
1137 if (prep_new_page(page, order, gfp_flags))
1142 local_irq_restore(flags);
1147 #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
1148 #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
1149 #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
1150 #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
1151 #define ALLOC_HARDER 0x10 /* try to alloc harder */
1152 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1153 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
1155 #ifdef CONFIG_FAIL_PAGE_ALLOC
1157 static struct fail_page_alloc_attr {
1158 struct fault_attr attr;
1160 u32 ignore_gfp_highmem;
1161 u32 ignore_gfp_wait;
1164 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1166 struct dentry *ignore_gfp_highmem_file;
1167 struct dentry *ignore_gfp_wait_file;
1168 struct dentry *min_order_file;
1170 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1172 } fail_page_alloc = {
1173 .attr = FAULT_ATTR_INITIALIZER,
1174 .ignore_gfp_wait = 1,
1175 .ignore_gfp_highmem = 1,
1179 static int __init setup_fail_page_alloc(char *str)
1181 return setup_fault_attr(&fail_page_alloc.attr, str);
1183 __setup("fail_page_alloc=", setup_fail_page_alloc);
1185 static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1187 if (order < fail_page_alloc.min_order)
1189 if (gfp_mask & __GFP_NOFAIL)
1191 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
1193 if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
1196 return should_fail(&fail_page_alloc.attr, 1 << order);
1199 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1201 static int __init fail_page_alloc_debugfs(void)
1203 mode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
1207 err = init_fault_attr_dentries(&fail_page_alloc.attr,
1211 dir = fail_page_alloc.attr.dentries.dir;
1213 fail_page_alloc.ignore_gfp_wait_file =
1214 debugfs_create_bool("ignore-gfp-wait", mode, dir,
1215 &fail_page_alloc.ignore_gfp_wait);
1217 fail_page_alloc.ignore_gfp_highmem_file =
1218 debugfs_create_bool("ignore-gfp-highmem", mode, dir,
1219 &fail_page_alloc.ignore_gfp_highmem);
1220 fail_page_alloc.min_order_file =
1221 debugfs_create_u32("min-order", mode, dir,
1222 &fail_page_alloc.min_order);
1224 if (!fail_page_alloc.ignore_gfp_wait_file ||
1225 !fail_page_alloc.ignore_gfp_highmem_file ||
1226 !fail_page_alloc.min_order_file) {
1228 debugfs_remove(fail_page_alloc.ignore_gfp_wait_file);
1229 debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file);
1230 debugfs_remove(fail_page_alloc.min_order_file);
1231 cleanup_fault_attr_dentries(&fail_page_alloc.attr);
1237 late_initcall(fail_page_alloc_debugfs);
1239 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1241 #else /* CONFIG_FAIL_PAGE_ALLOC */
1243 static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1248 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1251 * Return 1 if free pages are above 'mark'. This takes into account the order
1252 * of the allocation.
1254 int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1255 int classzone_idx, int alloc_flags)
1257 /* free_pages my go negative - that's OK */
1259 long free_pages = zone_page_state(z, NR_FREE_PAGES) - (1 << order) + 1;
1262 if (alloc_flags & ALLOC_HIGH)
1264 if (alloc_flags & ALLOC_HARDER)
1267 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
1269 for (o = 0; o < order; o++) {
1270 /* At the next order, this order's pages become unavailable */
1271 free_pages -= z->free_area[o].nr_free << o;
1273 /* Require fewer higher order pages to be free */
1276 if (free_pages <= min)
1284 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1285 * skip over zones that are not allowed by the cpuset, or that have
1286 * been recently (in last second) found to be nearly full. See further
1287 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1288 * that have to skip over a lot of full or unallowed zones.
1290 * If the zonelist cache is present in the passed in zonelist, then
1291 * returns a pointer to the allowed node mask (either the current
1292 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
1294 * If the zonelist cache is not available for this zonelist, does
1295 * nothing and returns NULL.
1297 * If the fullzones BITMAP in the zonelist cache is stale (more than
1298 * a second since last zap'd) then we zap it out (clear its bits.)
1300 * We hold off even calling zlc_setup, until after we've checked the
1301 * first zone in the zonelist, on the theory that most allocations will
1302 * be satisfied from that first zone, so best to examine that zone as
1303 * quickly as we can.
1305 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1307 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1308 nodemask_t *allowednodes; /* zonelist_cache approximation */
1310 zlc = zonelist->zlcache_ptr;
1314 if (time_after(jiffies, zlc->last_full_zap + HZ)) {
1315 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1316 zlc->last_full_zap = jiffies;
1319 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1320 &cpuset_current_mems_allowed :
1321 &node_states[N_HIGH_MEMORY];
1322 return allowednodes;
1326 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1327 * if it is worth looking at further for free memory:
1328 * 1) Check that the zone isn't thought to be full (doesn't have its
1329 * bit set in the zonelist_cache fullzones BITMAP).
1330 * 2) Check that the zones node (obtained from the zonelist_cache
1331 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1332 * Return true (non-zero) if zone is worth looking at further, or
1333 * else return false (zero) if it is not.
1335 * This check -ignores- the distinction between various watermarks,
1336 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1337 * found to be full for any variation of these watermarks, it will
1338 * be considered full for up to one second by all requests, unless
1339 * we are so low on memory on all allowed nodes that we are forced
1340 * into the second scan of the zonelist.
1342 * In the second scan we ignore this zonelist cache and exactly
1343 * apply the watermarks to all zones, even it is slower to do so.
1344 * We are low on memory in the second scan, and should leave no stone
1345 * unturned looking for a free page.
1347 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
1348 nodemask_t *allowednodes)
1350 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1351 int i; /* index of *z in zonelist zones */
1352 int n; /* node that zone *z is on */
1354 zlc = zonelist->zlcache_ptr;
1358 i = z - zonelist->_zonerefs;
1361 /* This zone is worth trying if it is allowed but not full */
1362 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1366 * Given 'z' scanning a zonelist, set the corresponding bit in
1367 * zlc->fullzones, so that subsequent attempts to allocate a page
1368 * from that zone don't waste time re-examining it.
1370 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
1372 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1373 int i; /* index of *z in zonelist zones */
1375 zlc = zonelist->zlcache_ptr;
1379 i = z - zonelist->_zonerefs;
1381 set_bit(i, zlc->fullzones);
1384 #else /* CONFIG_NUMA */
1386 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1391 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
1392 nodemask_t *allowednodes)
1397 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
1400 #endif /* CONFIG_NUMA */
1403 * get_page_from_freelist goes through the zonelist trying to allocate
1406 static struct page *
1407 get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order,
1408 struct zonelist *zonelist, int high_zoneidx, int alloc_flags,
1409 struct zone *preferred_zone, int migratetype)
1412 struct page *page = NULL;
1415 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1416 int zlc_active = 0; /* set if using zonelist_cache */
1417 int did_zlc_setup = 0; /* just call zlc_setup() one time */
1419 if (WARN_ON_ONCE(order >= MAX_ORDER))
1422 classzone_idx = zone_idx(preferred_zone);
1425 * Scan zonelist, looking for a zone with enough free.
1426 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1428 for_each_zone_zonelist_nodemask(zone, z, zonelist,
1429 high_zoneidx, nodemask) {
1430 if (NUMA_BUILD && zlc_active &&
1431 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1433 if ((alloc_flags & ALLOC_CPUSET) &&
1434 !cpuset_zone_allowed_softwall(zone, gfp_mask))
1437 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
1439 if (alloc_flags & ALLOC_WMARK_MIN)
1440 mark = zone->pages_min;
1441 else if (alloc_flags & ALLOC_WMARK_LOW)
1442 mark = zone->pages_low;
1444 mark = zone->pages_high;
1445 if (!zone_watermark_ok(zone, order, mark,
1446 classzone_idx, alloc_flags)) {
1447 if (!zone_reclaim_mode ||
1448 !zone_reclaim(zone, gfp_mask, order))
1449 goto this_zone_full;
1453 page = buffered_rmqueue(preferred_zone, zone, order,
1454 gfp_mask, migratetype);
1459 zlc_mark_zone_full(zonelist, z);
1461 if (NUMA_BUILD && !did_zlc_setup) {
1462 /* we do zlc_setup after the first zone is tried */
1463 allowednodes = zlc_setup(zonelist, alloc_flags);
1469 if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) {
1470 /* Disable zlc cache for second zonelist scan */
1478 should_alloc_retry(gfp_t gfp_mask, unsigned int order,
1479 unsigned long pages_reclaimed)
1481 /* Do not loop if specifically requested */
1482 if (gfp_mask & __GFP_NORETRY)
1486 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
1487 * means __GFP_NOFAIL, but that may not be true in other
1490 if (order <= PAGE_ALLOC_COSTLY_ORDER)
1494 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
1495 * specified, then we retry until we no longer reclaim any pages
1496 * (above), or we've reclaimed an order of pages at least as
1497 * large as the allocation's order. In both cases, if the
1498 * allocation still fails, we stop retrying.
1500 if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order))
1504 * Don't let big-order allocations loop unless the caller
1505 * explicitly requests that.
1507 if (gfp_mask & __GFP_NOFAIL)
1513 static inline struct page *
1514 __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
1515 struct zonelist *zonelist, enum zone_type high_zoneidx,
1516 nodemask_t *nodemask, struct zone *preferred_zone,
1521 /* Acquire the OOM killer lock for the zones in zonelist */
1522 if (!try_set_zone_oom(zonelist, gfp_mask)) {
1523 schedule_timeout_uninterruptible(1);
1528 * Go through the zonelist yet one more time, keep very high watermark
1529 * here, this is only to catch a parallel oom killing, we must fail if
1530 * we're still under heavy pressure.
1532 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask,
1533 order, zonelist, high_zoneidx,
1534 ALLOC_WMARK_HIGH|ALLOC_CPUSET,
1535 preferred_zone, migratetype);
1539 /* The OOM killer will not help higher order allocs */
1540 if (order > PAGE_ALLOC_COSTLY_ORDER)
1543 /* Exhausted what can be done so it's blamo time */
1544 out_of_memory(zonelist, gfp_mask, order);
1547 clear_zonelist_oom(zonelist, gfp_mask);
1551 /* The really slow allocator path where we enter direct reclaim */
1552 static inline struct page *
1553 __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
1554 struct zonelist *zonelist, enum zone_type high_zoneidx,
1555 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
1556 int migratetype, unsigned long *did_some_progress)
1558 struct page *page = NULL;
1559 struct reclaim_state reclaim_state;
1560 struct task_struct *p = current;
1564 /* We now go into synchronous reclaim */
1565 cpuset_memory_pressure_bump();
1568 * The task's cpuset might have expanded its set of allowable nodes
1570 p->flags |= PF_MEMALLOC;
1571 lockdep_set_current_reclaim_state(gfp_mask);
1572 reclaim_state.reclaimed_slab = 0;
1573 p->reclaim_state = &reclaim_state;
1575 *did_some_progress = try_to_free_pages(zonelist, order, gfp_mask, nodemask);
1577 p->reclaim_state = NULL;
1578 lockdep_clear_current_reclaim_state();
1579 p->flags &= ~PF_MEMALLOC;
1586 if (likely(*did_some_progress))
1587 page = get_page_from_freelist(gfp_mask, nodemask, order,
1588 zonelist, high_zoneidx,
1589 alloc_flags, preferred_zone,
1595 * This is called in the allocator slow-path if the allocation request is of
1596 * sufficient urgency to ignore watermarks and take other desperate measures
1598 static inline struct page *
1599 __alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order,
1600 struct zonelist *zonelist, enum zone_type high_zoneidx,
1601 nodemask_t *nodemask, struct zone *preferred_zone,
1607 page = get_page_from_freelist(gfp_mask, nodemask, order,
1608 zonelist, high_zoneidx, ALLOC_NO_WATERMARKS,
1609 preferred_zone, migratetype);
1611 if (!page && gfp_mask & __GFP_NOFAIL)
1612 congestion_wait(WRITE, HZ/50);
1613 } while (!page && (gfp_mask & __GFP_NOFAIL));
1619 void wake_all_kswapd(unsigned int order, struct zonelist *zonelist,
1620 enum zone_type high_zoneidx)
1625 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx)
1626 wakeup_kswapd(zone, order);
1630 gfp_to_alloc_flags(gfp_t gfp_mask)
1632 struct task_struct *p = current;
1633 int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
1634 const gfp_t wait = gfp_mask & __GFP_WAIT;
1636 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
1637 BUILD_BUG_ON(__GFP_HIGH != ALLOC_HIGH);
1640 * The caller may dip into page reserves a bit more if the caller
1641 * cannot run direct reclaim, or if the caller has realtime scheduling
1642 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1643 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1645 alloc_flags |= (gfp_mask & __GFP_HIGH);
1648 alloc_flags |= ALLOC_HARDER;
1650 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1651 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1653 alloc_flags &= ~ALLOC_CPUSET;
1654 } else if (unlikely(rt_task(p)))
1655 alloc_flags |= ALLOC_HARDER;
1657 if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) {
1658 if (!in_interrupt() &&
1659 ((p->flags & PF_MEMALLOC) ||
1660 unlikely(test_thread_flag(TIF_MEMDIE))))
1661 alloc_flags |= ALLOC_NO_WATERMARKS;
1667 static inline struct page *
1668 __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
1669 struct zonelist *zonelist, enum zone_type high_zoneidx,
1670 nodemask_t *nodemask, struct zone *preferred_zone,
1673 const gfp_t wait = gfp_mask & __GFP_WAIT;
1674 struct page *page = NULL;
1676 unsigned long pages_reclaimed = 0;
1677 unsigned long did_some_progress;
1678 struct task_struct *p = current;
1681 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1682 * __GFP_NOWARN set) should not cause reclaim since the subsystem
1683 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1684 * using a larger set of nodes after it has established that the
1685 * allowed per node queues are empty and that nodes are
1688 if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
1691 wake_all_kswapd(order, zonelist, high_zoneidx);
1694 * OK, we're below the kswapd watermark and have kicked background
1695 * reclaim. Now things get more complex, so set up alloc_flags according
1696 * to how we want to proceed.
1698 alloc_flags = gfp_to_alloc_flags(gfp_mask);
1701 /* This is the last chance, in general, before the goto nopage. */
1702 page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist,
1703 high_zoneidx, alloc_flags & ~ALLOC_NO_WATERMARKS,
1704 preferred_zone, migratetype);
1709 /* Allocate without watermarks if the context allows */
1710 if (alloc_flags & ALLOC_NO_WATERMARKS) {
1711 page = __alloc_pages_high_priority(gfp_mask, order,
1712 zonelist, high_zoneidx, nodemask,
1713 preferred_zone, migratetype);
1718 /* Atomic allocations - we can't balance anything */
1722 /* Avoid recursion of direct reclaim */
1723 if (p->flags & PF_MEMALLOC)
1726 /* Try direct reclaim and then allocating */
1727 page = __alloc_pages_direct_reclaim(gfp_mask, order,
1728 zonelist, high_zoneidx,
1730 alloc_flags, preferred_zone,
1731 migratetype, &did_some_progress);
1736 * If we failed to make any progress reclaiming, then we are
1737 * running out of options and have to consider going OOM
1739 if (!did_some_progress) {
1740 if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
1741 page = __alloc_pages_may_oom(gfp_mask, order,
1742 zonelist, high_zoneidx,
1743 nodemask, preferred_zone,
1749 * The OOM killer does not trigger for high-order allocations
1750 * but if no progress is being made, there are no other
1751 * options and retrying is unlikely to help
1753 if (order > PAGE_ALLOC_COSTLY_ORDER)
1760 /* Check if we should retry the allocation */
1761 pages_reclaimed += did_some_progress;
1762 if (should_alloc_retry(gfp_mask, order, pages_reclaimed)) {
1763 /* Wait for some write requests to complete then retry */
1764 congestion_wait(WRITE, HZ/50);
1769 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
1770 printk(KERN_WARNING "%s: page allocation failure."
1771 " order:%d, mode:0x%x\n",
1772 p->comm, order, gfp_mask);
1782 * This is the 'heart' of the zoned buddy allocator.
1785 __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
1786 struct zonelist *zonelist, nodemask_t *nodemask)
1788 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
1789 struct zone *preferred_zone;
1791 int migratetype = allocflags_to_migratetype(gfp_mask);
1793 lockdep_trace_alloc(gfp_mask);
1795 might_sleep_if(gfp_mask & __GFP_WAIT);
1797 if (should_fail_alloc_page(gfp_mask, order))
1801 * Check the zones suitable for the gfp_mask contain at least one
1802 * valid zone. It's possible to have an empty zonelist as a result
1803 * of GFP_THISNODE and a memoryless node
1805 if (unlikely(!zonelist->_zonerefs->zone))
1808 /* The preferred zone is used for statistics later */
1809 first_zones_zonelist(zonelist, high_zoneidx, nodemask, &preferred_zone);
1810 if (!preferred_zone)
1813 /* First allocation attempt */
1814 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order,
1815 zonelist, high_zoneidx, ALLOC_WMARK_LOW|ALLOC_CPUSET,
1816 preferred_zone, migratetype);
1817 if (unlikely(!page))
1818 page = __alloc_pages_slowpath(gfp_mask, order,
1819 zonelist, high_zoneidx, nodemask,
1820 preferred_zone, migratetype);
1824 EXPORT_SYMBOL(__alloc_pages_nodemask);
1827 * Common helper functions.
1829 unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
1832 page = alloc_pages(gfp_mask, order);
1835 return (unsigned long) page_address(page);
1838 EXPORT_SYMBOL(__get_free_pages);
1840 unsigned long get_zeroed_page(gfp_t gfp_mask)
1845 * get_zeroed_page() returns a 32-bit address, which cannot represent
1848 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
1850 page = alloc_pages(gfp_mask | __GFP_ZERO, 0);
1852 return (unsigned long) page_address(page);
1856 EXPORT_SYMBOL(get_zeroed_page);
1858 void __pagevec_free(struct pagevec *pvec)
1860 int i = pagevec_count(pvec);
1863 free_hot_cold_page(pvec->pages[i], pvec->cold);
1866 void __free_pages(struct page *page, unsigned int order)
1868 if (put_page_testzero(page)) {
1870 free_hot_page(page);
1872 __free_pages_ok(page, order);
1876 EXPORT_SYMBOL(__free_pages);
1878 void free_pages(unsigned long addr, unsigned int order)
1881 VM_BUG_ON(!virt_addr_valid((void *)addr));
1882 __free_pages(virt_to_page((void *)addr), order);
1886 EXPORT_SYMBOL(free_pages);
1889 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
1890 * @size: the number of bytes to allocate
1891 * @gfp_mask: GFP flags for the allocation
1893 * This function is similar to alloc_pages(), except that it allocates the
1894 * minimum number of pages to satisfy the request. alloc_pages() can only
1895 * allocate memory in power-of-two pages.
1897 * This function is also limited by MAX_ORDER.
1899 * Memory allocated by this function must be released by free_pages_exact().
1901 void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
1903 unsigned int order = get_order(size);
1906 addr = __get_free_pages(gfp_mask, order);
1908 unsigned long alloc_end = addr + (PAGE_SIZE << order);
1909 unsigned long used = addr + PAGE_ALIGN(size);
1911 split_page(virt_to_page(addr), order);
1912 while (used < alloc_end) {
1918 return (void *)addr;
1920 EXPORT_SYMBOL(alloc_pages_exact);
1923 * free_pages_exact - release memory allocated via alloc_pages_exact()
1924 * @virt: the value returned by alloc_pages_exact.
1925 * @size: size of allocation, same value as passed to alloc_pages_exact().
1927 * Release the memory allocated by a previous call to alloc_pages_exact.
1929 void free_pages_exact(void *virt, size_t size)
1931 unsigned long addr = (unsigned long)virt;
1932 unsigned long end = addr + PAGE_ALIGN(size);
1934 while (addr < end) {
1939 EXPORT_SYMBOL(free_pages_exact);
1941 static unsigned int nr_free_zone_pages(int offset)
1946 /* Just pick one node, since fallback list is circular */
1947 unsigned int sum = 0;
1949 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
1951 for_each_zone_zonelist(zone, z, zonelist, offset) {
1952 unsigned long size = zone->present_pages;
1953 unsigned long high = zone->pages_high;
1962 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1964 unsigned int nr_free_buffer_pages(void)
1966 return nr_free_zone_pages(gfp_zone(GFP_USER));
1968 EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
1971 * Amount of free RAM allocatable within all zones
1973 unsigned int nr_free_pagecache_pages(void)
1975 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
1978 static inline void show_node(struct zone *zone)
1981 printk("Node %d ", zone_to_nid(zone));
1984 void si_meminfo(struct sysinfo *val)
1986 val->totalram = totalram_pages;
1988 val->freeram = global_page_state(NR_FREE_PAGES);
1989 val->bufferram = nr_blockdev_pages();
1990 val->totalhigh = totalhigh_pages;
1991 val->freehigh = nr_free_highpages();
1992 val->mem_unit = PAGE_SIZE;
1995 EXPORT_SYMBOL(si_meminfo);
1998 void si_meminfo_node(struct sysinfo *val, int nid)
2000 pg_data_t *pgdat = NODE_DATA(nid);
2002 val->totalram = pgdat->node_present_pages;
2003 val->freeram = node_page_state(nid, NR_FREE_PAGES);
2004 #ifdef CONFIG_HIGHMEM
2005 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
2006 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
2012 val->mem_unit = PAGE_SIZE;
2016 #define K(x) ((x) << (PAGE_SHIFT-10))
2019 * Show free area list (used inside shift_scroll-lock stuff)
2020 * We also calculate the percentage fragmentation. We do this by counting the
2021 * memory on each free list with the exception of the first item on the list.
2023 void show_free_areas(void)
2028 for_each_populated_zone(zone) {
2030 printk("%s per-cpu:\n", zone->name);
2032 for_each_online_cpu(cpu) {
2033 struct per_cpu_pageset *pageset;
2035 pageset = zone_pcp(zone, cpu);
2037 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
2038 cpu, pageset->pcp.high,
2039 pageset->pcp.batch, pageset->pcp.count);
2043 printk("Active_anon:%lu active_file:%lu inactive_anon:%lu\n"
2044 " inactive_file:%lu"
2045 //TODO: check/adjust line lengths
2046 #ifdef CONFIG_UNEVICTABLE_LRU
2049 " dirty:%lu writeback:%lu unstable:%lu\n"
2050 " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
2051 global_page_state(NR_ACTIVE_ANON),
2052 global_page_state(NR_ACTIVE_FILE),
2053 global_page_state(NR_INACTIVE_ANON),
2054 global_page_state(NR_INACTIVE_FILE),
2055 #ifdef CONFIG_UNEVICTABLE_LRU
2056 global_page_state(NR_UNEVICTABLE),
2058 global_page_state(NR_FILE_DIRTY),
2059 global_page_state(NR_WRITEBACK),
2060 global_page_state(NR_UNSTABLE_NFS),
2061 global_page_state(NR_FREE_PAGES),
2062 global_page_state(NR_SLAB_RECLAIMABLE) +
2063 global_page_state(NR_SLAB_UNRECLAIMABLE),
2064 global_page_state(NR_FILE_MAPPED),
2065 global_page_state(NR_PAGETABLE),
2066 global_page_state(NR_BOUNCE));
2068 for_each_populated_zone(zone) {
2077 " active_anon:%lukB"
2078 " inactive_anon:%lukB"
2079 " active_file:%lukB"
2080 " inactive_file:%lukB"
2081 #ifdef CONFIG_UNEVICTABLE_LRU
2082 " unevictable:%lukB"
2085 " pages_scanned:%lu"
2086 " all_unreclaimable? %s"
2089 K(zone_page_state(zone, NR_FREE_PAGES)),
2092 K(zone->pages_high),
2093 K(zone_page_state(zone, NR_ACTIVE_ANON)),
2094 K(zone_page_state(zone, NR_INACTIVE_ANON)),
2095 K(zone_page_state(zone, NR_ACTIVE_FILE)),
2096 K(zone_page_state(zone, NR_INACTIVE_FILE)),
2097 #ifdef CONFIG_UNEVICTABLE_LRU
2098 K(zone_page_state(zone, NR_UNEVICTABLE)),
2100 K(zone->present_pages),
2101 zone->pages_scanned,
2102 (zone_is_all_unreclaimable(zone) ? "yes" : "no")
2104 printk("lowmem_reserve[]:");
2105 for (i = 0; i < MAX_NR_ZONES; i++)
2106 printk(" %lu", zone->lowmem_reserve[i]);
2110 for_each_populated_zone(zone) {
2111 unsigned long nr[MAX_ORDER], flags, order, total = 0;
2114 printk("%s: ", zone->name);
2116 spin_lock_irqsave(&zone->lock, flags);
2117 for (order = 0; order < MAX_ORDER; order++) {
2118 nr[order] = zone->free_area[order].nr_free;
2119 total += nr[order] << order;
2121 spin_unlock_irqrestore(&zone->lock, flags);
2122 for (order = 0; order < MAX_ORDER; order++)
2123 printk("%lu*%lukB ", nr[order], K(1UL) << order);
2124 printk("= %lukB\n", K(total));
2127 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
2129 show_swap_cache_info();
2132 static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
2134 zoneref->zone = zone;
2135 zoneref->zone_idx = zone_idx(zone);
2139 * Builds allocation fallback zone lists.
2141 * Add all populated zones of a node to the zonelist.
2143 static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
2144 int nr_zones, enum zone_type zone_type)
2148 BUG_ON(zone_type >= MAX_NR_ZONES);
2153 zone = pgdat->node_zones + zone_type;
2154 if (populated_zone(zone)) {
2155 zoneref_set_zone(zone,
2156 &zonelist->_zonerefs[nr_zones++]);
2157 check_highest_zone(zone_type);
2160 } while (zone_type);
2167 * 0 = automatic detection of better ordering.
2168 * 1 = order by ([node] distance, -zonetype)
2169 * 2 = order by (-zonetype, [node] distance)
2171 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
2172 * the same zonelist. So only NUMA can configure this param.
2174 #define ZONELIST_ORDER_DEFAULT 0
2175 #define ZONELIST_ORDER_NODE 1
2176 #define ZONELIST_ORDER_ZONE 2
2178 /* zonelist order in the kernel.
2179 * set_zonelist_order() will set this to NODE or ZONE.
2181 static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
2182 static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
2186 /* The value user specified ....changed by config */
2187 static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2188 /* string for sysctl */
2189 #define NUMA_ZONELIST_ORDER_LEN 16
2190 char numa_zonelist_order[16] = "default";
2193 * interface for configure zonelist ordering.
2194 * command line option "numa_zonelist_order"
2195 * = "[dD]efault - default, automatic configuration.
2196 * = "[nN]ode - order by node locality, then by zone within node
2197 * = "[zZ]one - order by zone, then by locality within zone
2200 static int __parse_numa_zonelist_order(char *s)
2202 if (*s == 'd' || *s == 'D') {
2203 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2204 } else if (*s == 'n' || *s == 'N') {
2205 user_zonelist_order = ZONELIST_ORDER_NODE;
2206 } else if (*s == 'z' || *s == 'Z') {
2207 user_zonelist_order = ZONELIST_ORDER_ZONE;
2210 "Ignoring invalid numa_zonelist_order value: "
2217 static __init int setup_numa_zonelist_order(char *s)
2220 return __parse_numa_zonelist_order(s);
2223 early_param("numa_zonelist_order", setup_numa_zonelist_order);
2226 * sysctl handler for numa_zonelist_order
2228 int numa_zonelist_order_handler(ctl_table *table, int write,
2229 struct file *file, void __user *buffer, size_t *length,
2232 char saved_string[NUMA_ZONELIST_ORDER_LEN];
2236 strncpy(saved_string, (char*)table->data,
2237 NUMA_ZONELIST_ORDER_LEN);
2238 ret = proc_dostring(table, write, file, buffer, length, ppos);
2242 int oldval = user_zonelist_order;
2243 if (__parse_numa_zonelist_order((char*)table->data)) {
2245 * bogus value. restore saved string
2247 strncpy((char*)table->data, saved_string,
2248 NUMA_ZONELIST_ORDER_LEN);
2249 user_zonelist_order = oldval;
2250 } else if (oldval != user_zonelist_order)
2251 build_all_zonelists();
2257 #define MAX_NODE_LOAD (num_online_nodes())
2258 static int node_load[MAX_NUMNODES];
2261 * find_next_best_node - find the next node that should appear in a given node's fallback list
2262 * @node: node whose fallback list we're appending
2263 * @used_node_mask: nodemask_t of already used nodes
2265 * We use a number of factors to determine which is the next node that should
2266 * appear on a given node's fallback list. The node should not have appeared
2267 * already in @node's fallback list, and it should be the next closest node
2268 * according to the distance array (which contains arbitrary distance values
2269 * from each node to each node in the system), and should also prefer nodes
2270 * with no CPUs, since presumably they'll have very little allocation pressure
2271 * on them otherwise.
2272 * It returns -1 if no node is found.
2274 static int find_next_best_node(int node, nodemask_t *used_node_mask)
2277 int min_val = INT_MAX;
2279 const struct cpumask *tmp = cpumask_of_node(0);
2281 /* Use the local node if we haven't already */
2282 if (!node_isset(node, *used_node_mask)) {
2283 node_set(node, *used_node_mask);
2287 for_each_node_state(n, N_HIGH_MEMORY) {
2289 /* Don't want a node to appear more than once */
2290 if (node_isset(n, *used_node_mask))
2293 /* Use the distance array to find the distance */
2294 val = node_distance(node, n);
2296 /* Penalize nodes under us ("prefer the next node") */
2299 /* Give preference to headless and unused nodes */
2300 tmp = cpumask_of_node(n);
2301 if (!cpumask_empty(tmp))
2302 val += PENALTY_FOR_NODE_WITH_CPUS;
2304 /* Slight preference for less loaded node */
2305 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
2306 val += node_load[n];
2308 if (val < min_val) {
2315 node_set(best_node, *used_node_mask);
2322 * Build zonelists ordered by node and zones within node.
2323 * This results in maximum locality--normal zone overflows into local
2324 * DMA zone, if any--but risks exhausting DMA zone.
2326 static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
2329 struct zonelist *zonelist;
2331 zonelist = &pgdat->node_zonelists[0];
2332 for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
2334 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2336 zonelist->_zonerefs[j].zone = NULL;
2337 zonelist->_zonerefs[j].zone_idx = 0;
2341 * Build gfp_thisnode zonelists
2343 static void build_thisnode_zonelists(pg_data_t *pgdat)
2346 struct zonelist *zonelist;
2348 zonelist = &pgdat->node_zonelists[1];
2349 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
2350 zonelist->_zonerefs[j].zone = NULL;
2351 zonelist->_zonerefs[j].zone_idx = 0;
2355 * Build zonelists ordered by zone and nodes within zones.
2356 * This results in conserving DMA zone[s] until all Normal memory is
2357 * exhausted, but results in overflowing to remote node while memory
2358 * may still exist in local DMA zone.
2360 static int node_order[MAX_NUMNODES];
2362 static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
2365 int zone_type; /* needs to be signed */
2367 struct zonelist *zonelist;
2369 zonelist = &pgdat->node_zonelists[0];
2371 for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
2372 for (j = 0; j < nr_nodes; j++) {
2373 node = node_order[j];
2374 z = &NODE_DATA(node)->node_zones[zone_type];
2375 if (populated_zone(z)) {
2377 &zonelist->_zonerefs[pos++]);
2378 check_highest_zone(zone_type);
2382 zonelist->_zonerefs[pos].zone = NULL;
2383 zonelist->_zonerefs[pos].zone_idx = 0;
2386 static int default_zonelist_order(void)
2389 unsigned long low_kmem_size,total_size;
2393 * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem.
2394 * If they are really small and used heavily, the system can fall
2395 * into OOM very easily.
2396 * This function detect ZONE_DMA/DMA32 size and confgigures zone order.
2398 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2401 for_each_online_node(nid) {
2402 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2403 z = &NODE_DATA(nid)->node_zones[zone_type];
2404 if (populated_zone(z)) {
2405 if (zone_type < ZONE_NORMAL)
2406 low_kmem_size += z->present_pages;
2407 total_size += z->present_pages;
2411 if (!low_kmem_size || /* there are no DMA area. */
2412 low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
2413 return ZONELIST_ORDER_NODE;
2415 * look into each node's config.
2416 * If there is a node whose DMA/DMA32 memory is very big area on
2417 * local memory, NODE_ORDER may be suitable.
2419 average_size = total_size /
2420 (nodes_weight(node_states[N_HIGH_MEMORY]) + 1);
2421 for_each_online_node(nid) {
2424 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2425 z = &NODE_DATA(nid)->node_zones[zone_type];
2426 if (populated_zone(z)) {
2427 if (zone_type < ZONE_NORMAL)
2428 low_kmem_size += z->present_pages;
2429 total_size += z->present_pages;
2432 if (low_kmem_size &&
2433 total_size > average_size && /* ignore small node */
2434 low_kmem_size > total_size * 70/100)
2435 return ZONELIST_ORDER_NODE;
2437 return ZONELIST_ORDER_ZONE;
2440 static void set_zonelist_order(void)
2442 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
2443 current_zonelist_order = default_zonelist_order();
2445 current_zonelist_order = user_zonelist_order;
2448 static void build_zonelists(pg_data_t *pgdat)
2452 nodemask_t used_mask;
2453 int local_node, prev_node;
2454 struct zonelist *zonelist;
2455 int order = current_zonelist_order;
2457 /* initialize zonelists */
2458 for (i = 0; i < MAX_ZONELISTS; i++) {
2459 zonelist = pgdat->node_zonelists + i;
2460 zonelist->_zonerefs[0].zone = NULL;
2461 zonelist->_zonerefs[0].zone_idx = 0;
2464 /* NUMA-aware ordering of nodes */
2465 local_node = pgdat->node_id;
2466 load = num_online_nodes();
2467 prev_node = local_node;
2468 nodes_clear(used_mask);
2470 memset(node_load, 0, sizeof(node_load));
2471 memset(node_order, 0, sizeof(node_order));
2474 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
2475 int distance = node_distance(local_node, node);
2478 * If another node is sufficiently far away then it is better
2479 * to reclaim pages in a zone before going off node.
2481 if (distance > RECLAIM_DISTANCE)
2482 zone_reclaim_mode = 1;
2485 * We don't want to pressure a particular node.
2486 * So adding penalty to the first node in same
2487 * distance group to make it round-robin.
2489 if (distance != node_distance(local_node, prev_node))
2490 node_load[node] = load;
2494 if (order == ZONELIST_ORDER_NODE)
2495 build_zonelists_in_node_order(pgdat, node);
2497 node_order[j++] = node; /* remember order */
2500 if (order == ZONELIST_ORDER_ZONE) {
2501 /* calculate node order -- i.e., DMA last! */
2502 build_zonelists_in_zone_order(pgdat, j);
2505 build_thisnode_zonelists(pgdat);
2508 /* Construct the zonelist performance cache - see further mmzone.h */
2509 static void build_zonelist_cache(pg_data_t *pgdat)
2511 struct zonelist *zonelist;
2512 struct zonelist_cache *zlc;
2515 zonelist = &pgdat->node_zonelists[0];
2516 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
2517 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
2518 for (z = zonelist->_zonerefs; z->zone; z++)
2519 zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z);
2523 #else /* CONFIG_NUMA */
2525 static void set_zonelist_order(void)
2527 current_zonelist_order = ZONELIST_ORDER_ZONE;
2530 static void build_zonelists(pg_data_t *pgdat)
2532 int node, local_node;
2534 struct zonelist *zonelist;
2536 local_node = pgdat->node_id;
2538 zonelist = &pgdat->node_zonelists[0];
2539 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
2542 * Now we build the zonelist so that it contains the zones
2543 * of all the other nodes.
2544 * We don't want to pressure a particular node, so when
2545 * building the zones for node N, we make sure that the
2546 * zones coming right after the local ones are those from
2547 * node N+1 (modulo N)
2549 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
2550 if (!node_online(node))
2552 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2555 for (node = 0; node < local_node; node++) {
2556 if (!node_online(node))
2558 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2562 zonelist->_zonerefs[j].zone = NULL;
2563 zonelist->_zonerefs[j].zone_idx = 0;
2566 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
2567 static void build_zonelist_cache(pg_data_t *pgdat)
2569 pgdat->node_zonelists[0].zlcache_ptr = NULL;
2572 #endif /* CONFIG_NUMA */
2574 /* return values int ....just for stop_machine() */
2575 static int __build_all_zonelists(void *dummy)
2579 for_each_online_node(nid) {
2580 pg_data_t *pgdat = NODE_DATA(nid);
2582 build_zonelists(pgdat);
2583 build_zonelist_cache(pgdat);
2588 void build_all_zonelists(void)
2590 set_zonelist_order();
2592 if (system_state == SYSTEM_BOOTING) {
2593 __build_all_zonelists(NULL);
2594 mminit_verify_zonelist();
2595 cpuset_init_current_mems_allowed();
2597 /* we have to stop all cpus to guarantee there is no user
2599 stop_machine(__build_all_zonelists, NULL, NULL);
2600 /* cpuset refresh routine should be here */
2602 vm_total_pages = nr_free_pagecache_pages();
2604 * Disable grouping by mobility if the number of pages in the
2605 * system is too low to allow the mechanism to work. It would be
2606 * more accurate, but expensive to check per-zone. This check is
2607 * made on memory-hotadd so a system can start with mobility
2608 * disabled and enable it later
2610 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
2611 page_group_by_mobility_disabled = 1;
2613 page_group_by_mobility_disabled = 0;
2615 printk("Built %i zonelists in %s order, mobility grouping %s. "
2616 "Total pages: %ld\n",
2618 zonelist_order_name[current_zonelist_order],
2619 page_group_by_mobility_disabled ? "off" : "on",
2622 printk("Policy zone: %s\n", zone_names[policy_zone]);
2627 * Helper functions to size the waitqueue hash table.
2628 * Essentially these want to choose hash table sizes sufficiently
2629 * large so that collisions trying to wait on pages are rare.
2630 * But in fact, the number of active page waitqueues on typical
2631 * systems is ridiculously low, less than 200. So this is even
2632 * conservative, even though it seems large.
2634 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
2635 * waitqueues, i.e. the size of the waitq table given the number of pages.
2637 #define PAGES_PER_WAITQUEUE 256
2639 #ifndef CONFIG_MEMORY_HOTPLUG
2640 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
2642 unsigned long size = 1;
2644 pages /= PAGES_PER_WAITQUEUE;
2646 while (size < pages)
2650 * Once we have dozens or even hundreds of threads sleeping
2651 * on IO we've got bigger problems than wait queue collision.
2652 * Limit the size of the wait table to a reasonable size.
2654 size = min(size, 4096UL);
2656 return max(size, 4UL);
2660 * A zone's size might be changed by hot-add, so it is not possible to determine
2661 * a suitable size for its wait_table. So we use the maximum size now.
2663 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
2665 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
2666 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
2667 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
2669 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
2670 * or more by the traditional way. (See above). It equals:
2672 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
2673 * ia64(16K page size) : = ( 8G + 4M)byte.
2674 * powerpc (64K page size) : = (32G +16M)byte.
2676 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
2683 * This is an integer logarithm so that shifts can be used later
2684 * to extract the more random high bits from the multiplicative
2685 * hash function before the remainder is taken.
2687 static inline unsigned long wait_table_bits(unsigned long size)
2692 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
2695 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
2696 * of blocks reserved is based on zone->pages_min. The memory within the
2697 * reserve will tend to store contiguous free pages. Setting min_free_kbytes
2698 * higher will lead to a bigger reserve which will get freed as contiguous
2699 * blocks as reclaim kicks in
2701 static void setup_zone_migrate_reserve(struct zone *zone)
2703 unsigned long start_pfn, pfn, end_pfn;
2705 unsigned long reserve, block_migratetype;
2707 /* Get the start pfn, end pfn and the number of blocks to reserve */
2708 start_pfn = zone->zone_start_pfn;
2709 end_pfn = start_pfn + zone->spanned_pages;
2710 reserve = roundup(zone->pages_min, pageblock_nr_pages) >>
2713 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
2714 if (!pfn_valid(pfn))
2716 page = pfn_to_page(pfn);
2718 /* Watch out for overlapping nodes */
2719 if (page_to_nid(page) != zone_to_nid(zone))
2722 /* Blocks with reserved pages will never free, skip them. */
2723 if (PageReserved(page))
2726 block_migratetype = get_pageblock_migratetype(page);
2728 /* If this block is reserved, account for it */
2729 if (reserve > 0 && block_migratetype == MIGRATE_RESERVE) {
2734 /* Suitable for reserving if this block is movable */
2735 if (reserve > 0 && block_migratetype == MIGRATE_MOVABLE) {
2736 set_pageblock_migratetype(page, MIGRATE_RESERVE);
2737 move_freepages_block(zone, page, MIGRATE_RESERVE);
2743 * If the reserve is met and this is a previous reserved block,
2746 if (block_migratetype == MIGRATE_RESERVE) {
2747 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
2748 move_freepages_block(zone, page, MIGRATE_MOVABLE);
2754 * Initially all pages are reserved - free ones are freed
2755 * up by free_all_bootmem() once the early boot process is
2756 * done. Non-atomic initialization, single-pass.
2758 void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
2759 unsigned long start_pfn, enum memmap_context context)
2762 unsigned long end_pfn = start_pfn + size;
2766 if (highest_memmap_pfn < end_pfn - 1)
2767 highest_memmap_pfn = end_pfn - 1;
2769 z = &NODE_DATA(nid)->node_zones[zone];
2770 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
2772 * There can be holes in boot-time mem_map[]s
2773 * handed to this function. They do not
2774 * exist on hotplugged memory.
2776 if (context == MEMMAP_EARLY) {
2777 if (!early_pfn_valid(pfn))
2779 if (!early_pfn_in_nid(pfn, nid))
2782 page = pfn_to_page(pfn);
2783 set_page_links(page, zone, nid, pfn);
2784 mminit_verify_page_links(page, zone, nid, pfn);
2785 init_page_count(page);
2786 reset_page_mapcount(page);
2787 SetPageReserved(page);
2789 * Mark the block movable so that blocks are reserved for
2790 * movable at startup. This will force kernel allocations
2791 * to reserve their blocks rather than leaking throughout
2792 * the address space during boot when many long-lived
2793 * kernel allocations are made. Later some blocks near
2794 * the start are marked MIGRATE_RESERVE by
2795 * setup_zone_migrate_reserve()
2797 * bitmap is created for zone's valid pfn range. but memmap
2798 * can be created for invalid pages (for alignment)
2799 * check here not to call set_pageblock_migratetype() against
2802 if ((z->zone_start_pfn <= pfn)
2803 && (pfn < z->zone_start_pfn + z->spanned_pages)
2804 && !(pfn & (pageblock_nr_pages - 1)))
2805 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
2807 INIT_LIST_HEAD(&page->lru);
2808 #ifdef WANT_PAGE_VIRTUAL
2809 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
2810 if (!is_highmem_idx(zone))
2811 set_page_address(page, __va(pfn << PAGE_SHIFT));
2816 static void __meminit zone_init_free_lists(struct zone *zone)
2819 for_each_migratetype_order(order, t) {
2820 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
2821 zone->free_area[order].nr_free = 0;
2825 #ifndef __HAVE_ARCH_MEMMAP_INIT
2826 #define memmap_init(size, nid, zone, start_pfn) \
2827 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
2830 static int zone_batchsize(struct zone *zone)
2836 * The per-cpu-pages pools are set to around 1000th of the
2837 * size of the zone. But no more than 1/2 of a meg.
2839 * OK, so we don't know how big the cache is. So guess.
2841 batch = zone->present_pages / 1024;
2842 if (batch * PAGE_SIZE > 512 * 1024)
2843 batch = (512 * 1024) / PAGE_SIZE;
2844 batch /= 4; /* We effectively *= 4 below */
2849 * Clamp the batch to a 2^n - 1 value. Having a power
2850 * of 2 value was found to be more likely to have
2851 * suboptimal cache aliasing properties in some cases.
2853 * For example if 2 tasks are alternately allocating
2854 * batches of pages, one task can end up with a lot
2855 * of pages of one half of the possible page colors
2856 * and the other with pages of the other colors.
2858 batch = rounddown_pow_of_two(batch + batch/2) - 1;
2863 /* The deferral and batching of frees should be suppressed under NOMMU
2866 * The problem is that NOMMU needs to be able to allocate large chunks
2867 * of contiguous memory as there's no hardware page translation to
2868 * assemble apparent contiguous memory from discontiguous pages.
2870 * Queueing large contiguous runs of pages for batching, however,
2871 * causes the pages to actually be freed in smaller chunks. As there
2872 * can be a significant delay between the individual batches being
2873 * recycled, this leads to the once large chunks of space being
2874 * fragmented and becoming unavailable for high-order allocations.
2880 static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
2882 struct per_cpu_pages *pcp;
2884 memset(p, 0, sizeof(*p));
2888 pcp->high = 6 * batch;
2889 pcp->batch = max(1UL, 1 * batch);
2890 INIT_LIST_HEAD(&pcp->list);
2894 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
2895 * to the value high for the pageset p.
2898 static void setup_pagelist_highmark(struct per_cpu_pageset *p,
2901 struct per_cpu_pages *pcp;
2905 pcp->batch = max(1UL, high/4);
2906 if ((high/4) > (PAGE_SHIFT * 8))
2907 pcp->batch = PAGE_SHIFT * 8;
2913 * Boot pageset table. One per cpu which is going to be used for all
2914 * zones and all nodes. The parameters will be set in such a way
2915 * that an item put on a list will immediately be handed over to
2916 * the buddy list. This is safe since pageset manipulation is done
2917 * with interrupts disabled.
2919 * Some NUMA counter updates may also be caught by the boot pagesets.
2921 * The boot_pagesets must be kept even after bootup is complete for
2922 * unused processors and/or zones. They do play a role for bootstrapping
2923 * hotplugged processors.
2925 * zoneinfo_show() and maybe other functions do
2926 * not check if the processor is online before following the pageset pointer.
2927 * Other parts of the kernel may not check if the zone is available.
2929 static struct per_cpu_pageset boot_pageset[NR_CPUS];
2932 * Dynamically allocate memory for the
2933 * per cpu pageset array in struct zone.
2935 static int __cpuinit process_zones(int cpu)
2937 struct zone *zone, *dzone;
2938 int node = cpu_to_node(cpu);
2940 node_set_state(node, N_CPU); /* this node has a cpu */
2942 for_each_populated_zone(zone) {
2943 zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset),
2945 if (!zone_pcp(zone, cpu))
2948 setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone));
2950 if (percpu_pagelist_fraction)
2951 setup_pagelist_highmark(zone_pcp(zone, cpu),
2952 (zone->present_pages / percpu_pagelist_fraction));
2957 for_each_zone(dzone) {
2958 if (!populated_zone(dzone))
2962 kfree(zone_pcp(dzone, cpu));
2963 zone_pcp(dzone, cpu) = NULL;
2968 static inline void free_zone_pagesets(int cpu)
2972 for_each_zone(zone) {
2973 struct per_cpu_pageset *pset = zone_pcp(zone, cpu);
2975 /* Free per_cpu_pageset if it is slab allocated */
2976 if (pset != &boot_pageset[cpu])
2978 zone_pcp(zone, cpu) = NULL;
2982 static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb,
2983 unsigned long action,
2986 int cpu = (long)hcpu;
2987 int ret = NOTIFY_OK;
2990 case CPU_UP_PREPARE:
2991 case CPU_UP_PREPARE_FROZEN:
2992 if (process_zones(cpu))
2995 case CPU_UP_CANCELED:
2996 case CPU_UP_CANCELED_FROZEN:
2998 case CPU_DEAD_FROZEN:
2999 free_zone_pagesets(cpu);
3007 static struct notifier_block __cpuinitdata pageset_notifier =
3008 { &pageset_cpuup_callback, NULL, 0 };
3010 void __init setup_per_cpu_pageset(void)
3014 /* Initialize per_cpu_pageset for cpu 0.
3015 * A cpuup callback will do this for every cpu
3016 * as it comes online
3018 err = process_zones(smp_processor_id());
3020 register_cpu_notifier(&pageset_notifier);
3025 static noinline __init_refok
3026 int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
3029 struct pglist_data *pgdat = zone->zone_pgdat;
3033 * The per-page waitqueue mechanism uses hashed waitqueues
3036 zone->wait_table_hash_nr_entries =
3037 wait_table_hash_nr_entries(zone_size_pages);
3038 zone->wait_table_bits =
3039 wait_table_bits(zone->wait_table_hash_nr_entries);
3040 alloc_size = zone->wait_table_hash_nr_entries
3041 * sizeof(wait_queue_head_t);
3043 if (!slab_is_available()) {
3044 zone->wait_table = (wait_queue_head_t *)
3045 alloc_bootmem_node(pgdat, alloc_size);
3048 * This case means that a zone whose size was 0 gets new memory
3049 * via memory hot-add.
3050 * But it may be the case that a new node was hot-added. In
3051 * this case vmalloc() will not be able to use this new node's
3052 * memory - this wait_table must be initialized to use this new
3053 * node itself as well.
3054 * To use this new node's memory, further consideration will be
3057 zone->wait_table = vmalloc(alloc_size);
3059 if (!zone->wait_table)
3062 for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
3063 init_waitqueue_head(zone->wait_table + i);
3068 static __meminit void zone_pcp_init(struct zone *zone)
3071 unsigned long batch = zone_batchsize(zone);
3073 for (cpu = 0; cpu < NR_CPUS; cpu++) {
3075 /* Early boot. Slab allocator not functional yet */
3076 zone_pcp(zone, cpu) = &boot_pageset[cpu];
3077 setup_pageset(&boot_pageset[cpu],0);
3079 setup_pageset(zone_pcp(zone,cpu), batch);
3082 if (zone->present_pages)
3083 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n",
3084 zone->name, zone->present_pages, batch);
3087 __meminit int init_currently_empty_zone(struct zone *zone,
3088 unsigned long zone_start_pfn,
3090 enum memmap_context context)
3092 struct pglist_data *pgdat = zone->zone_pgdat;
3094 ret = zone_wait_table_init(zone, size);
3097 pgdat->nr_zones = zone_idx(zone) + 1;
3099 zone->zone_start_pfn = zone_start_pfn;
3101 mminit_dprintk(MMINIT_TRACE, "memmap_init",
3102 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
3104 (unsigned long)zone_idx(zone),
3105 zone_start_pfn, (zone_start_pfn + size));
3107 zone_init_free_lists(zone);
3112 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3114 * Basic iterator support. Return the first range of PFNs for a node
3115 * Note: nid == MAX_NUMNODES returns first region regardless of node
3117 static int __meminit first_active_region_index_in_nid(int nid)
3121 for (i = 0; i < nr_nodemap_entries; i++)
3122 if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
3129 * Basic iterator support. Return the next active range of PFNs for a node
3130 * Note: nid == MAX_NUMNODES returns next region regardless of node
3132 static int __meminit next_active_region_index_in_nid(int index, int nid)
3134 for (index = index + 1; index < nr_nodemap_entries; index++)
3135 if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
3141 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
3143 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
3144 * Architectures may implement their own version but if add_active_range()
3145 * was used and there are no special requirements, this is a convenient
3148 int __meminit __early_pfn_to_nid(unsigned long pfn)
3152 for (i = 0; i < nr_nodemap_entries; i++) {
3153 unsigned long start_pfn = early_node_map[i].start_pfn;
3154 unsigned long end_pfn = early_node_map[i].end_pfn;
3156 if (start_pfn <= pfn && pfn < end_pfn)
3157 return early_node_map[i].nid;
3159 /* This is a memory hole */
3162 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
3164 int __meminit early_pfn_to_nid(unsigned long pfn)
3168 nid = __early_pfn_to_nid(pfn);
3171 /* just returns 0 */
3175 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
3176 bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
3180 nid = __early_pfn_to_nid(pfn);
3181 if (nid >= 0 && nid != node)
3187 /* Basic iterator support to walk early_node_map[] */
3188 #define for_each_active_range_index_in_nid(i, nid) \
3189 for (i = first_active_region_index_in_nid(nid); i != -1; \
3190 i = next_active_region_index_in_nid(i, nid))
3193 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
3194 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
3195 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
3197 * If an architecture guarantees that all ranges registered with
3198 * add_active_ranges() contain no holes and may be freed, this
3199 * this function may be used instead of calling free_bootmem() manually.
3201 void __init free_bootmem_with_active_regions(int nid,
3202 unsigned long max_low_pfn)
3206 for_each_active_range_index_in_nid(i, nid) {
3207 unsigned long size_pages = 0;
3208 unsigned long end_pfn = early_node_map[i].end_pfn;
3210 if (early_node_map[i].start_pfn >= max_low_pfn)
3213 if (end_pfn > max_low_pfn)
3214 end_pfn = max_low_pfn;
3216 size_pages = end_pfn - early_node_map[i].start_pfn;
3217 free_bootmem_node(NODE_DATA(early_node_map[i].nid),
3218 PFN_PHYS(early_node_map[i].start_pfn),
3219 size_pages << PAGE_SHIFT);
3223 void __init work_with_active_regions(int nid, work_fn_t work_fn, void *data)
3228 for_each_active_range_index_in_nid(i, nid) {
3229 ret = work_fn(early_node_map[i].start_pfn,
3230 early_node_map[i].end_pfn, data);
3236 * sparse_memory_present_with_active_regions - Call memory_present for each active range
3237 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
3239 * If an architecture guarantees that all ranges registered with
3240 * add_active_ranges() contain no holes and may be freed, this
3241 * function may be used instead of calling memory_present() manually.
3243 void __init sparse_memory_present_with_active_regions(int nid)
3247 for_each_active_range_index_in_nid(i, nid)
3248 memory_present(early_node_map[i].nid,
3249 early_node_map[i].start_pfn,
3250 early_node_map[i].end_pfn);
3254 * get_pfn_range_for_nid - Return the start and end page frames for a node
3255 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
3256 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
3257 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
3259 * It returns the start and end page frame of a node based on information
3260 * provided by an arch calling add_active_range(). If called for a node
3261 * with no available memory, a warning is printed and the start and end
3264 void __meminit get_pfn_range_for_nid(unsigned int nid,
3265 unsigned long *start_pfn, unsigned long *end_pfn)
3271 for_each_active_range_index_in_nid(i, nid) {
3272 *start_pfn = min(*start_pfn, early_node_map[i].start_pfn);
3273 *end_pfn = max(*end_pfn, early_node_map[i].end_pfn);
3276 if (*start_pfn == -1UL)
3281 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3282 * assumption is made that zones within a node are ordered in monotonic
3283 * increasing memory addresses so that the "highest" populated zone is used
3285 static void __init find_usable_zone_for_movable(void)
3288 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
3289 if (zone_index == ZONE_MOVABLE)
3292 if (arch_zone_highest_possible_pfn[zone_index] >
3293 arch_zone_lowest_possible_pfn[zone_index])
3297 VM_BUG_ON(zone_index == -1);
3298 movable_zone = zone_index;
3302 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
3303 * because it is sized independant of architecture. Unlike the other zones,
3304 * the starting point for ZONE_MOVABLE is not fixed. It may be different
3305 * in each node depending on the size of each node and how evenly kernelcore
3306 * is distributed. This helper function adjusts the zone ranges
3307 * provided by the architecture for a given node by using the end of the
3308 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3309 * zones within a node are in order of monotonic increases memory addresses
3311 static void __meminit adjust_zone_range_for_zone_movable(int nid,
3312 unsigned long zone_type,
3313 unsigned long node_start_pfn,
3314 unsigned long node_end_pfn,
3315 unsigned long *zone_start_pfn,
3316 unsigned long *zone_end_pfn)
3318 /* Only adjust if ZONE_MOVABLE is on this node */
3319 if (zone_movable_pfn[nid]) {
3320 /* Size ZONE_MOVABLE */
3321 if (zone_type == ZONE_MOVABLE) {
3322 *zone_start_pfn = zone_movable_pfn[nid];
3323 *zone_end_pfn = min(node_end_pfn,
3324 arch_zone_highest_possible_pfn[movable_zone]);
3326 /* Adjust for ZONE_MOVABLE starting within this range */
3327 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
3328 *zone_end_pfn > zone_movable_pfn[nid]) {
3329 *zone_end_pfn = zone_movable_pfn[nid];
3331 /* Check if this whole range is within ZONE_MOVABLE */
3332 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
3333 *zone_start_pfn = *zone_end_pfn;
3338 * Return the number of pages a zone spans in a node, including holes
3339 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3341 static unsigned long __meminit zone_spanned_pages_in_node(int nid,
3342 unsigned long zone_type,
3343 unsigned long *ignored)
3345 unsigned long node_start_pfn, node_end_pfn;
3346 unsigned long zone_start_pfn, zone_end_pfn;
3348 /* Get the start and end of the node and zone */
3349 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3350 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
3351 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
3352 adjust_zone_range_for_zone_movable(nid, zone_type,
3353 node_start_pfn, node_end_pfn,
3354 &zone_start_pfn, &zone_end_pfn);
3356 /* Check that this node has pages within the zone's required range */
3357 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
3360 /* Move the zone boundaries inside the node if necessary */
3361 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
3362 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
3364 /* Return the spanned pages */
3365 return zone_end_pfn - zone_start_pfn;
3369 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
3370 * then all holes in the requested range will be accounted for.
3372 static unsigned long __meminit __absent_pages_in_range(int nid,
3373 unsigned long range_start_pfn,
3374 unsigned long range_end_pfn)
3377 unsigned long prev_end_pfn = 0, hole_pages = 0;
3378 unsigned long start_pfn;
3380 /* Find the end_pfn of the first active range of pfns in the node */
3381 i = first_active_region_index_in_nid(nid);
3385 prev_end_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3387 /* Account for ranges before physical memory on this node */
3388 if (early_node_map[i].start_pfn > range_start_pfn)
3389 hole_pages = prev_end_pfn - range_start_pfn;
3391 /* Find all holes for the zone within the node */
3392 for (; i != -1; i = next_active_region_index_in_nid(i, nid)) {
3394 /* No need to continue if prev_end_pfn is outside the zone */
3395 if (prev_end_pfn >= range_end_pfn)
3398 /* Make sure the end of the zone is not within the hole */
3399 start_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3400 prev_end_pfn = max(prev_end_pfn, range_start_pfn);
3402 /* Update the hole size cound and move on */
3403 if (start_pfn > range_start_pfn) {
3404 BUG_ON(prev_end_pfn > start_pfn);
3405 hole_pages += start_pfn - prev_end_pfn;
3407 prev_end_pfn = early_node_map[i].end_pfn;
3410 /* Account for ranges past physical memory on this node */
3411 if (range_end_pfn > prev_end_pfn)
3412 hole_pages += range_end_pfn -
3413 max(range_start_pfn, prev_end_pfn);
3419 * absent_pages_in_range - Return number of page frames in holes within a range
3420 * @start_pfn: The start PFN to start searching for holes
3421 * @end_pfn: The end PFN to stop searching for holes
3423 * It returns the number of pages frames in memory holes within a range.
3425 unsigned long __init absent_pages_in_range(unsigned long start_pfn,
3426 unsigned long end_pfn)
3428 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
3431 /* Return the number of page frames in holes in a zone on a node */
3432 static unsigned long __meminit zone_absent_pages_in_node(int nid,
3433 unsigned long zone_type,
3434 unsigned long *ignored)
3436 unsigned long node_start_pfn, node_end_pfn;
3437 unsigned long zone_start_pfn, zone_end_pfn;
3439 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3440 zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type],
3442 zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type],
3445 adjust_zone_range_for_zone_movable(nid, zone_type,
3446 node_start_pfn, node_end_pfn,
3447 &zone_start_pfn, &zone_end_pfn);
3448 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
3452 static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
3453 unsigned long zone_type,
3454 unsigned long *zones_size)
3456 return zones_size[zone_type];
3459 static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
3460 unsigned long zone_type,
3461 unsigned long *zholes_size)
3466 return zholes_size[zone_type];
3471 static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
3472 unsigned long *zones_size, unsigned long *zholes_size)
3474 unsigned long realtotalpages, totalpages = 0;
3477 for (i = 0; i < MAX_NR_ZONES; i++)
3478 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
3480 pgdat->node_spanned_pages = totalpages;
3482 realtotalpages = totalpages;
3483 for (i = 0; i < MAX_NR_ZONES; i++)
3485 zone_absent_pages_in_node(pgdat->node_id, i,
3487 pgdat->node_present_pages = realtotalpages;
3488 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
3492 #ifndef CONFIG_SPARSEMEM
3494 * Calculate the size of the zone->blockflags rounded to an unsigned long
3495 * Start by making sure zonesize is a multiple of pageblock_order by rounding
3496 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
3497 * round what is now in bits to nearest long in bits, then return it in
3500 static unsigned long __init usemap_size(unsigned long zonesize)
3502 unsigned long usemapsize;
3504 usemapsize = roundup(zonesize, pageblock_nr_pages);
3505 usemapsize = usemapsize >> pageblock_order;
3506 usemapsize *= NR_PAGEBLOCK_BITS;
3507 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
3509 return usemapsize / 8;
3512 static void __init setup_usemap(struct pglist_data *pgdat,
3513 struct zone *zone, unsigned long zonesize)
3515 unsigned long usemapsize = usemap_size(zonesize);
3516 zone->pageblock_flags = NULL;
3518 zone->pageblock_flags = alloc_bootmem_node(pgdat, usemapsize);
3521 static void inline setup_usemap(struct pglist_data *pgdat,
3522 struct zone *zone, unsigned long zonesize) {}
3523 #endif /* CONFIG_SPARSEMEM */
3525 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
3527 /* Return a sensible default order for the pageblock size. */
3528 static inline int pageblock_default_order(void)
3530 if (HPAGE_SHIFT > PAGE_SHIFT)
3531 return HUGETLB_PAGE_ORDER;
3536 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
3537 static inline void __init set_pageblock_order(unsigned int order)
3539 /* Check that pageblock_nr_pages has not already been setup */
3540 if (pageblock_order)
3544 * Assume the largest contiguous order of interest is a huge page.
3545 * This value may be variable depending on boot parameters on IA64
3547 pageblock_order = order;
3549 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3552 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
3553 * and pageblock_default_order() are unused as pageblock_order is set
3554 * at compile-time. See include/linux/pageblock-flags.h for the values of
3555 * pageblock_order based on the kernel config
3557 static inline int pageblock_default_order(unsigned int order)
3561 #define set_pageblock_order(x) do {} while (0)
3563 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3566 * Set up the zone data structures:
3567 * - mark all pages reserved
3568 * - mark all memory queues empty
3569 * - clear the memory bitmaps
3571 static void __paginginit free_area_init_core(struct pglist_data *pgdat,
3572 unsigned long *zones_size, unsigned long *zholes_size)
3575 int nid = pgdat->node_id;
3576 unsigned long zone_start_pfn = pgdat->node_start_pfn;
3579 pgdat_resize_init(pgdat);
3580 pgdat->nr_zones = 0;
3581 init_waitqueue_head(&pgdat->kswapd_wait);
3582 pgdat->kswapd_max_order = 0;
3583 pgdat_page_cgroup_init(pgdat);
3585 for (j = 0; j < MAX_NR_ZONES; j++) {
3586 struct zone *zone = pgdat->node_zones + j;
3587 unsigned long size, realsize, memmap_pages;
3590 size = zone_spanned_pages_in_node(nid, j, zones_size);
3591 realsize = size - zone_absent_pages_in_node(nid, j,
3595 * Adjust realsize so that it accounts for how much memory
3596 * is used by this zone for memmap. This affects the watermark
3597 * and per-cpu initialisations
3600 PAGE_ALIGN(size * sizeof(struct page)) >> PAGE_SHIFT;
3601 if (realsize >= memmap_pages) {
3602 realsize -= memmap_pages;
3605 " %s zone: %lu pages used for memmap\n",
3606 zone_names[j], memmap_pages);
3609 " %s zone: %lu pages exceeds realsize %lu\n",
3610 zone_names[j], memmap_pages, realsize);
3612 /* Account for reserved pages */
3613 if (j == 0 && realsize > dma_reserve) {
3614 realsize -= dma_reserve;
3615 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
3616 zone_names[0], dma_reserve);
3619 if (!is_highmem_idx(j))
3620 nr_kernel_pages += realsize;
3621 nr_all_pages += realsize;
3623 zone->spanned_pages = size;
3624 zone->present_pages = realsize;
3627 zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
3629 zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100;
3631 zone->name = zone_names[j];
3632 spin_lock_init(&zone->lock);
3633 spin_lock_init(&zone->lru_lock);
3634 zone_seqlock_init(zone);
3635 zone->zone_pgdat = pgdat;
3637 zone->prev_priority = DEF_PRIORITY;
3639 zone_pcp_init(zone);
3641 INIT_LIST_HEAD(&zone->lru[l].list);
3642 zone->lru[l].nr_scan = 0;
3644 zone->reclaim_stat.recent_rotated[0] = 0;
3645 zone->reclaim_stat.recent_rotated[1] = 0;
3646 zone->reclaim_stat.recent_scanned[0] = 0;
3647 zone->reclaim_stat.recent_scanned[1] = 0;
3648 zap_zone_vm_stats(zone);
3653 set_pageblock_order(pageblock_default_order());
3654 setup_usemap(pgdat, zone, size);
3655 ret = init_currently_empty_zone(zone, zone_start_pfn,
3656 size, MEMMAP_EARLY);
3658 memmap_init(size, nid, j, zone_start_pfn);
3659 zone_start_pfn += size;
3663 static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
3665 /* Skip empty nodes */
3666 if (!pgdat->node_spanned_pages)
3669 #ifdef CONFIG_FLAT_NODE_MEM_MAP
3670 /* ia64 gets its own node_mem_map, before this, without bootmem */
3671 if (!pgdat->node_mem_map) {
3672 unsigned long size, start, end;
3676 * The zone's endpoints aren't required to be MAX_ORDER
3677 * aligned but the node_mem_map endpoints must be in order
3678 * for the buddy allocator to function correctly.
3680 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
3681 end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
3682 end = ALIGN(end, MAX_ORDER_NR_PAGES);
3683 size = (end - start) * sizeof(struct page);
3684 map = alloc_remap(pgdat->node_id, size);
3686 map = alloc_bootmem_node(pgdat, size);
3687 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
3689 #ifndef CONFIG_NEED_MULTIPLE_NODES
3691 * With no DISCONTIG, the global mem_map is just set as node 0's
3693 if (pgdat == NODE_DATA(0)) {
3694 mem_map = NODE_DATA(0)->node_mem_map;
3695 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3696 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
3697 mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
3698 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3701 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
3704 void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
3705 unsigned long node_start_pfn, unsigned long *zholes_size)
3707 pg_data_t *pgdat = NODE_DATA(nid);
3709 pgdat->node_id = nid;
3710 pgdat->node_start_pfn = node_start_pfn;
3711 calculate_node_totalpages(pgdat, zones_size, zholes_size);
3713 alloc_node_mem_map(pgdat);
3714 #ifdef CONFIG_FLAT_NODE_MEM_MAP
3715 printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
3716 nid, (unsigned long)pgdat,
3717 (unsigned long)pgdat->node_mem_map);
3720 free_area_init_core(pgdat, zones_size, zholes_size);
3723 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3725 #if MAX_NUMNODES > 1
3727 * Figure out the number of possible node ids.
3729 static void __init setup_nr_node_ids(void)
3732 unsigned int highest = 0;
3734 for_each_node_mask(node, node_possible_map)
3736 nr_node_ids = highest + 1;
3739 static inline void setup_nr_node_ids(void)
3745 * add_active_range - Register a range of PFNs backed by physical memory
3746 * @nid: The node ID the range resides on
3747 * @start_pfn: The start PFN of the available physical memory
3748 * @end_pfn: The end PFN of the available physical memory
3750 * These ranges are stored in an early_node_map[] and later used by
3751 * free_area_init_nodes() to calculate zone sizes and holes. If the
3752 * range spans a memory hole, it is up to the architecture to ensure
3753 * the memory is not freed by the bootmem allocator. If possible
3754 * the range being registered will be merged with existing ranges.
3756 void __init add_active_range(unsigned int nid, unsigned long start_pfn,
3757 unsigned long end_pfn)
3761 mminit_dprintk(MMINIT_TRACE, "memory_register",
3762 "Entering add_active_range(%d, %#lx, %#lx) "
3763 "%d entries of %d used\n",
3764 nid, start_pfn, end_pfn,
3765 nr_nodemap_entries, MAX_ACTIVE_REGIONS);
3767 mminit_validate_memmodel_limits(&start_pfn, &end_pfn);
3769 /* Merge with existing active regions if possible */
3770 for (i = 0; i < nr_nodemap_entries; i++) {
3771 if (early_node_map[i].nid != nid)
3774 /* Skip if an existing region covers this new one */
3775 if (start_pfn >= early_node_map[i].start_pfn &&
3776 end_pfn <= early_node_map[i].end_pfn)
3779 /* Merge forward if suitable */
3780 if (start_pfn <= early_node_map[i].end_pfn &&
3781 end_pfn > early_node_map[i].end_pfn) {
3782 early_node_map[i].end_pfn = end_pfn;
3786 /* Merge backward if suitable */
3787 if (start_pfn < early_node_map[i].end_pfn &&
3788 end_pfn >= early_node_map[i].start_pfn) {
3789 early_node_map[i].start_pfn = start_pfn;
3794 /* Check that early_node_map is large enough */
3795 if (i >= MAX_ACTIVE_REGIONS) {
3796 printk(KERN_CRIT "More than %d memory regions, truncating\n",
3797 MAX_ACTIVE_REGIONS);
3801 early_node_map[i].nid = nid;
3802 early_node_map[i].start_pfn = start_pfn;
3803 early_node_map[i].end_pfn = end_pfn;
3804 nr_nodemap_entries = i + 1;
3808 * remove_active_range - Shrink an existing registered range of PFNs
3809 * @nid: The node id the range is on that should be shrunk
3810 * @start_pfn: The new PFN of the range
3811 * @end_pfn: The new PFN of the range
3813 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
3814 * The map is kept near the end physical page range that has already been
3815 * registered. This function allows an arch to shrink an existing registered
3818 void __init remove_active_range(unsigned int nid, unsigned long start_pfn,
3819 unsigned long end_pfn)
3824 printk(KERN_DEBUG "remove_active_range (%d, %lu, %lu)\n",
3825 nid, start_pfn, end_pfn);
3827 /* Find the old active region end and shrink */
3828 for_each_active_range_index_in_nid(i, nid) {
3829 if (early_node_map[i].start_pfn >= start_pfn &&
3830 early_node_map[i].end_pfn <= end_pfn) {
3832 early_node_map[i].start_pfn = 0;
3833 early_node_map[i].end_pfn = 0;
3837 if (early_node_map[i].start_pfn < start_pfn &&
3838 early_node_map[i].end_pfn > start_pfn) {
3839 unsigned long temp_end_pfn = early_node_map[i].end_pfn;
3840 early_node_map[i].end_pfn = start_pfn;
3841 if (temp_end_pfn > end_pfn)
3842 add_active_range(nid, end_pfn, temp_end_pfn);
3845 if (early_node_map[i].start_pfn >= start_pfn &&
3846 early_node_map[i].end_pfn > end_pfn &&
3847 early_node_map[i].start_pfn < end_pfn) {
3848 early_node_map[i].start_pfn = end_pfn;
3856 /* remove the blank ones */
3857 for (i = nr_nodemap_entries - 1; i > 0; i--) {
3858 if (early_node_map[i].nid != nid)
3860 if (early_node_map[i].end_pfn)
3862 /* we found it, get rid of it */
3863 for (j = i; j < nr_nodemap_entries - 1; j++)
3864 memcpy(&early_node_map[j], &early_node_map[j+1],
3865 sizeof(early_node_map[j]));
3866 j = nr_nodemap_entries - 1;
3867 memset(&early_node_map[j], 0, sizeof(early_node_map[j]));
3868 nr_nodemap_entries--;
3873 * remove_all_active_ranges - Remove all currently registered regions
3875 * During discovery, it may be found that a table like SRAT is invalid
3876 * and an alternative discovery method must be used. This function removes
3877 * all currently registered regions.
3879 void __init remove_all_active_ranges(void)
3881 memset(early_node_map, 0, sizeof(early_node_map));
3882 nr_nodemap_entries = 0;
3885 /* Compare two active node_active_regions */
3886 static int __init cmp_node_active_region(const void *a, const void *b)
3888 struct node_active_region *arange = (struct node_active_region *)a;
3889 struct node_active_region *brange = (struct node_active_region *)b;
3891 /* Done this way to avoid overflows */
3892 if (arange->start_pfn > brange->start_pfn)
3894 if (arange->start_pfn < brange->start_pfn)
3900 /* sort the node_map by start_pfn */
3901 static void __init sort_node_map(void)
3903 sort(early_node_map, (size_t)nr_nodemap_entries,
3904 sizeof(struct node_active_region),
3905 cmp_node_active_region, NULL);
3908 /* Find the lowest pfn for a node */
3909 static unsigned long __init find_min_pfn_for_node(int nid)
3912 unsigned long min_pfn = ULONG_MAX;
3914 /* Assuming a sorted map, the first range found has the starting pfn */
3915 for_each_active_range_index_in_nid(i, nid)
3916 min_pfn = min(min_pfn, early_node_map[i].start_pfn);
3918 if (min_pfn == ULONG_MAX) {
3920 "Could not find start_pfn for node %d\n", nid);
3928 * find_min_pfn_with_active_regions - Find the minimum PFN registered
3930 * It returns the minimum PFN based on information provided via
3931 * add_active_range().
3933 unsigned long __init find_min_pfn_with_active_regions(void)
3935 return find_min_pfn_for_node(MAX_NUMNODES);
3939 * early_calculate_totalpages()
3940 * Sum pages in active regions for movable zone.
3941 * Populate N_HIGH_MEMORY for calculating usable_nodes.
3943 static unsigned long __init early_calculate_totalpages(void)
3946 unsigned long totalpages = 0;
3948 for (i = 0; i < nr_nodemap_entries; i++) {
3949 unsigned long pages = early_node_map[i].end_pfn -
3950 early_node_map[i].start_pfn;
3951 totalpages += pages;
3953 node_set_state(early_node_map[i].nid, N_HIGH_MEMORY);
3959 * Find the PFN the Movable zone begins in each node. Kernel memory
3960 * is spread evenly between nodes as long as the nodes have enough
3961 * memory. When they don't, some nodes will have more kernelcore than
3964 static void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn)
3967 unsigned long usable_startpfn;
3968 unsigned long kernelcore_node, kernelcore_remaining;
3969 unsigned long totalpages = early_calculate_totalpages();
3970 int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]);
3973 * If movablecore was specified, calculate what size of
3974 * kernelcore that corresponds so that memory usable for
3975 * any allocation type is evenly spread. If both kernelcore
3976 * and movablecore are specified, then the value of kernelcore
3977 * will be used for required_kernelcore if it's greater than
3978 * what movablecore would have allowed.
3980 if (required_movablecore) {
3981 unsigned long corepages;
3984 * Round-up so that ZONE_MOVABLE is at least as large as what
3985 * was requested by the user
3987 required_movablecore =
3988 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
3989 corepages = totalpages - required_movablecore;
3991 required_kernelcore = max(required_kernelcore, corepages);
3994 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
3995 if (!required_kernelcore)
3998 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
3999 find_usable_zone_for_movable();
4000 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
4003 /* Spread kernelcore memory as evenly as possible throughout nodes */
4004 kernelcore_node = required_kernelcore / usable_nodes;
4005 for_each_node_state(nid, N_HIGH_MEMORY) {
4007 * Recalculate kernelcore_node if the division per node
4008 * now exceeds what is necessary to satisfy the requested
4009 * amount of memory for the kernel
4011 if (required_kernelcore < kernelcore_node)
4012 kernelcore_node = required_kernelcore / usable_nodes;
4015 * As the map is walked, we track how much memory is usable
4016 * by the kernel using kernelcore_remaining. When it is
4017 * 0, the rest of the node is usable by ZONE_MOVABLE
4019 kernelcore_remaining = kernelcore_node;
4021 /* Go through each range of PFNs within this node */
4022 for_each_active_range_index_in_nid(i, nid) {
4023 unsigned long start_pfn, end_pfn;
4024 unsigned long size_pages;
4026 start_pfn = max(early_node_map[i].start_pfn,
4027 zone_movable_pfn[nid]);
4028 end_pfn = early_node_map[i].end_pfn;
4029 if (start_pfn >= end_pfn)
4032 /* Account for what is only usable for kernelcore */
4033 if (start_pfn < usable_startpfn) {
4034 unsigned long kernel_pages;
4035 kernel_pages = min(end_pfn, usable_startpfn)
4038 kernelcore_remaining -= min(kernel_pages,
4039 kernelcore_remaining);
4040 required_kernelcore -= min(kernel_pages,
4041 required_kernelcore);
4043 /* Continue if range is now fully accounted */
4044 if (end_pfn <= usable_startpfn) {
4047 * Push zone_movable_pfn to the end so
4048 * that if we have to rebalance
4049 * kernelcore across nodes, we will
4050 * not double account here
4052 zone_movable_pfn[nid] = end_pfn;
4055 start_pfn = usable_startpfn;
4059 * The usable PFN range for ZONE_MOVABLE is from
4060 * start_pfn->end_pfn. Calculate size_pages as the
4061 * number of pages used as kernelcore
4063 size_pages = end_pfn - start_pfn;
4064 if (size_pages > kernelcore_remaining)
4065 size_pages = kernelcore_remaining;
4066 zone_movable_pfn[nid] = start_pfn + size_pages;
4069 * Some kernelcore has been met, update counts and
4070 * break if the kernelcore for this node has been
4073 required_kernelcore -= min(required_kernelcore,
4075 kernelcore_remaining -= size_pages;
4076 if (!kernelcore_remaining)
4082 * If there is still required_kernelcore, we do another pass with one
4083 * less node in the count. This will push zone_movable_pfn[nid] further
4084 * along on the nodes that still have memory until kernelcore is
4088 if (usable_nodes && required_kernelcore > usable_nodes)
4091 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
4092 for (nid = 0; nid < MAX_NUMNODES; nid++)
4093 zone_movable_pfn[nid] =
4094 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
4097 /* Any regular memory on that node ? */
4098 static void check_for_regular_memory(pg_data_t *pgdat)
4100 #ifdef CONFIG_HIGHMEM
4101 enum zone_type zone_type;
4103 for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) {
4104 struct zone *zone = &pgdat->node_zones[zone_type];
4105 if (zone->present_pages)
4106 node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY);
4112 * free_area_init_nodes - Initialise all pg_data_t and zone data
4113 * @max_zone_pfn: an array of max PFNs for each zone
4115 * This will call free_area_init_node() for each active node in the system.
4116 * Using the page ranges provided by add_active_range(), the size of each
4117 * zone in each node and their holes is calculated. If the maximum PFN
4118 * between two adjacent zones match, it is assumed that the zone is empty.
4119 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
4120 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
4121 * starts where the previous one ended. For example, ZONE_DMA32 starts
4122 * at arch_max_dma_pfn.
4124 void __init free_area_init_nodes(unsigned long *max_zone_pfn)
4129 /* Sort early_node_map as initialisation assumes it is sorted */
4132 /* Record where the zone boundaries are */
4133 memset(arch_zone_lowest_possible_pfn, 0,
4134 sizeof(arch_zone_lowest_possible_pfn));
4135 memset(arch_zone_highest_possible_pfn, 0,
4136 sizeof(arch_zone_highest_possible_pfn));
4137 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
4138 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
4139 for (i = 1; i < MAX_NR_ZONES; i++) {
4140 if (i == ZONE_MOVABLE)
4142 arch_zone_lowest_possible_pfn[i] =
4143 arch_zone_highest_possible_pfn[i-1];
4144 arch_zone_highest_possible_pfn[i] =
4145 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
4147 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
4148 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
4150 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
4151 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
4152 find_zone_movable_pfns_for_nodes(zone_movable_pfn);
4154 /* Print out the zone ranges */
4155 printk("Zone PFN ranges:\n");
4156 for (i = 0; i < MAX_NR_ZONES; i++) {
4157 if (i == ZONE_MOVABLE)
4159 printk(" %-8s %0#10lx -> %0#10lx\n",
4161 arch_zone_lowest_possible_pfn[i],
4162 arch_zone_highest_possible_pfn[i]);
4165 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
4166 printk("Movable zone start PFN for each node\n");
4167 for (i = 0; i < MAX_NUMNODES; i++) {
4168 if (zone_movable_pfn[i])
4169 printk(" Node %d: %lu\n", i, zone_movable_pfn[i]);
4172 /* Print out the early_node_map[] */
4173 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries);
4174 for (i = 0; i < nr_nodemap_entries; i++)
4175 printk(" %3d: %0#10lx -> %0#10lx\n", early_node_map[i].nid,
4176 early_node_map[i].start_pfn,
4177 early_node_map[i].end_pfn);
4179 /* Initialise every node */
4180 mminit_verify_pageflags_layout();
4181 setup_nr_node_ids();
4182 for_each_online_node(nid) {
4183 pg_data_t *pgdat = NODE_DATA(nid);
4184 free_area_init_node(nid, NULL,
4185 find_min_pfn_for_node(nid), NULL);
4187 /* Any memory on that node */
4188 if (pgdat->node_present_pages)
4189 node_set_state(nid, N_HIGH_MEMORY);
4190 check_for_regular_memory(pgdat);
4194 static int __init cmdline_parse_core(char *p, unsigned long *core)
4196 unsigned long long coremem;
4200 coremem = memparse(p, &p);
4201 *core = coremem >> PAGE_SHIFT;
4203 /* Paranoid check that UL is enough for the coremem value */
4204 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
4210 * kernelcore=size sets the amount of memory for use for allocations that
4211 * cannot be reclaimed or migrated.
4213 static int __init cmdline_parse_kernelcore(char *p)
4215 return cmdline_parse_core(p, &required_kernelcore);
4219 * movablecore=size sets the amount of memory for use for allocations that
4220 * can be reclaimed or migrated.
4222 static int __init cmdline_parse_movablecore(char *p)
4224 return cmdline_parse_core(p, &required_movablecore);
4227 early_param("kernelcore", cmdline_parse_kernelcore);
4228 early_param("movablecore", cmdline_parse_movablecore);
4230 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4233 * set_dma_reserve - set the specified number of pages reserved in the first zone
4234 * @new_dma_reserve: The number of pages to mark reserved
4236 * The per-cpu batchsize and zone watermarks are determined by present_pages.
4237 * In the DMA zone, a significant percentage may be consumed by kernel image
4238 * and other unfreeable allocations which can skew the watermarks badly. This
4239 * function may optionally be used to account for unfreeable pages in the
4240 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
4241 * smaller per-cpu batchsize.
4243 void __init set_dma_reserve(unsigned long new_dma_reserve)
4245 dma_reserve = new_dma_reserve;
4248 #ifndef CONFIG_NEED_MULTIPLE_NODES
4249 struct pglist_data __refdata contig_page_data = { .bdata = &bootmem_node_data[0] };
4250 EXPORT_SYMBOL(contig_page_data);
4253 void __init free_area_init(unsigned long *zones_size)
4255 free_area_init_node(0, zones_size,
4256 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
4259 static int page_alloc_cpu_notify(struct notifier_block *self,
4260 unsigned long action, void *hcpu)
4262 int cpu = (unsigned long)hcpu;
4264 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
4268 * Spill the event counters of the dead processor
4269 * into the current processors event counters.
4270 * This artificially elevates the count of the current
4273 vm_events_fold_cpu(cpu);
4276 * Zero the differential counters of the dead processor
4277 * so that the vm statistics are consistent.
4279 * This is only okay since the processor is dead and cannot
4280 * race with what we are doing.
4282 refresh_cpu_vm_stats(cpu);
4287 void __init page_alloc_init(void)
4289 hotcpu_notifier(page_alloc_cpu_notify, 0);
4293 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
4294 * or min_free_kbytes changes.
4296 static void calculate_totalreserve_pages(void)
4298 struct pglist_data *pgdat;
4299 unsigned long reserve_pages = 0;
4300 enum zone_type i, j;
4302 for_each_online_pgdat(pgdat) {
4303 for (i = 0; i < MAX_NR_ZONES; i++) {
4304 struct zone *zone = pgdat->node_zones + i;
4305 unsigned long max = 0;
4307 /* Find valid and maximum lowmem_reserve in the zone */
4308 for (j = i; j < MAX_NR_ZONES; j++) {
4309 if (zone->lowmem_reserve[j] > max)
4310 max = zone->lowmem_reserve[j];
4313 /* we treat pages_high as reserved pages. */
4314 max += zone->pages_high;
4316 if (max > zone->present_pages)
4317 max = zone->present_pages;
4318 reserve_pages += max;
4321 totalreserve_pages = reserve_pages;
4325 * setup_per_zone_lowmem_reserve - called whenever
4326 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
4327 * has a correct pages reserved value, so an adequate number of
4328 * pages are left in the zone after a successful __alloc_pages().
4330 static void setup_per_zone_lowmem_reserve(void)
4332 struct pglist_data *pgdat;
4333 enum zone_type j, idx;
4335 for_each_online_pgdat(pgdat) {
4336 for (j = 0; j < MAX_NR_ZONES; j++) {
4337 struct zone *zone = pgdat->node_zones + j;
4338 unsigned long present_pages = zone->present_pages;
4340 zone->lowmem_reserve[j] = 0;
4344 struct zone *lower_zone;
4348 if (sysctl_lowmem_reserve_ratio[idx] < 1)
4349 sysctl_lowmem_reserve_ratio[idx] = 1;
4351 lower_zone = pgdat->node_zones + idx;
4352 lower_zone->lowmem_reserve[j] = present_pages /
4353 sysctl_lowmem_reserve_ratio[idx];
4354 present_pages += lower_zone->present_pages;
4359 /* update totalreserve_pages */
4360 calculate_totalreserve_pages();
4364 * setup_per_zone_pages_min - called when min_free_kbytes changes.
4366 * Ensures that the pages_{min,low,high} values for each zone are set correctly
4367 * with respect to min_free_kbytes.
4369 void setup_per_zone_pages_min(void)
4371 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
4372 unsigned long lowmem_pages = 0;
4374 unsigned long flags;
4376 /* Calculate total number of !ZONE_HIGHMEM pages */
4377 for_each_zone(zone) {
4378 if (!is_highmem(zone))
4379 lowmem_pages += zone->present_pages;
4382 for_each_zone(zone) {
4385 spin_lock_irqsave(&zone->lock, flags);
4386 tmp = (u64)pages_min * zone->present_pages;
4387 do_div(tmp, lowmem_pages);
4388 if (is_highmem(zone)) {
4390 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
4391 * need highmem pages, so cap pages_min to a small
4394 * The (pages_high-pages_low) and (pages_low-pages_min)
4395 * deltas controls asynch page reclaim, and so should
4396 * not be capped for highmem.
4400 min_pages = zone->present_pages / 1024;
4401 if (min_pages < SWAP_CLUSTER_MAX)
4402 min_pages = SWAP_CLUSTER_MAX;
4403 if (min_pages > 128)
4405 zone->pages_min = min_pages;
4408 * If it's a lowmem zone, reserve a number of pages
4409 * proportionate to the zone's size.
4411 zone->pages_min = tmp;
4414 zone->pages_low = zone->pages_min + (tmp >> 2);
4415 zone->pages_high = zone->pages_min + (tmp >> 1);
4416 setup_zone_migrate_reserve(zone);
4417 spin_unlock_irqrestore(&zone->lock, flags);
4420 /* update totalreserve_pages */
4421 calculate_totalreserve_pages();
4425 * setup_per_zone_inactive_ratio - called when min_free_kbytes changes.
4427 * The inactive anon list should be small enough that the VM never has to
4428 * do too much work, but large enough that each inactive page has a chance
4429 * to be referenced again before it is swapped out.
4431 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
4432 * INACTIVE_ANON pages on this zone's LRU, maintained by the
4433 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
4434 * the anonymous pages are kept on the inactive list.
4437 * memory ratio inactive anon
4438 * -------------------------------------
4447 static void setup_per_zone_inactive_ratio(void)
4451 for_each_zone(zone) {
4452 unsigned int gb, ratio;
4454 /* Zone size in gigabytes */
4455 gb = zone->present_pages >> (30 - PAGE_SHIFT);
4456 ratio = int_sqrt(10 * gb);
4460 zone->inactive_ratio = ratio;
4465 * Initialise min_free_kbytes.
4467 * For small machines we want it small (128k min). For large machines
4468 * we want it large (64MB max). But it is not linear, because network
4469 * bandwidth does not increase linearly with machine size. We use
4471 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
4472 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
4488 static int __init init_per_zone_pages_min(void)
4490 unsigned long lowmem_kbytes;
4492 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
4494 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
4495 if (min_free_kbytes < 128)
4496 min_free_kbytes = 128;
4497 if (min_free_kbytes > 65536)
4498 min_free_kbytes = 65536;
4499 setup_per_zone_pages_min();
4500 setup_per_zone_lowmem_reserve();
4501 setup_per_zone_inactive_ratio();
4504 module_init(init_per_zone_pages_min)
4507 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
4508 * that we can call two helper functions whenever min_free_kbytes
4511 int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
4512 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4514 proc_dointvec(table, write, file, buffer, length, ppos);
4516 setup_per_zone_pages_min();
4521 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
4522 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4527 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4532 zone->min_unmapped_pages = (zone->present_pages *
4533 sysctl_min_unmapped_ratio) / 100;
4537 int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
4538 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4543 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4548 zone->min_slab_pages = (zone->present_pages *
4549 sysctl_min_slab_ratio) / 100;
4555 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
4556 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
4557 * whenever sysctl_lowmem_reserve_ratio changes.
4559 * The reserve ratio obviously has absolutely no relation with the
4560 * pages_min watermarks. The lowmem reserve ratio can only make sense
4561 * if in function of the boot time zone sizes.
4563 int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
4564 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4566 proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4567 setup_per_zone_lowmem_reserve();
4572 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
4573 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
4574 * can have before it gets flushed back to buddy allocator.
4577 int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
4578 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4584 ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4585 if (!write || (ret == -EINVAL))
4587 for_each_zone(zone) {
4588 for_each_online_cpu(cpu) {
4590 high = zone->present_pages / percpu_pagelist_fraction;
4591 setup_pagelist_highmark(zone_pcp(zone, cpu), high);
4597 int hashdist = HASHDIST_DEFAULT;
4600 static int __init set_hashdist(char *str)
4604 hashdist = simple_strtoul(str, &str, 0);
4607 __setup("hashdist=", set_hashdist);
4611 * allocate a large system hash table from bootmem
4612 * - it is assumed that the hash table must contain an exact power-of-2
4613 * quantity of entries
4614 * - limit is the number of hash buckets, not the total allocation size
4616 void *__init alloc_large_system_hash(const char *tablename,
4617 unsigned long bucketsize,
4618 unsigned long numentries,
4621 unsigned int *_hash_shift,
4622 unsigned int *_hash_mask,
4623 unsigned long limit)
4625 unsigned long long max = limit;
4626 unsigned long log2qty, size;
4629 /* allow the kernel cmdline to have a say */
4631 /* round applicable memory size up to nearest megabyte */
4632 numentries = nr_kernel_pages;
4633 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
4634 numentries >>= 20 - PAGE_SHIFT;
4635 numentries <<= 20 - PAGE_SHIFT;
4637 /* limit to 1 bucket per 2^scale bytes of low memory */
4638 if (scale > PAGE_SHIFT)
4639 numentries >>= (scale - PAGE_SHIFT);
4641 numentries <<= (PAGE_SHIFT - scale);
4643 /* Make sure we've got at least a 0-order allocation.. */
4644 if (unlikely((numentries * bucketsize) < PAGE_SIZE))
4645 numentries = PAGE_SIZE / bucketsize;
4647 numentries = roundup_pow_of_two(numentries);
4649 /* limit allocation size to 1/16 total memory by default */
4651 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
4652 do_div(max, bucketsize);
4655 if (numentries > max)
4658 log2qty = ilog2(numentries);
4661 size = bucketsize << log2qty;
4662 if (flags & HASH_EARLY)
4663 table = alloc_bootmem_nopanic(size);
4665 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
4667 unsigned long order = get_order(size);
4669 if (order < MAX_ORDER)
4670 table = (void *)__get_free_pages(GFP_ATOMIC,
4673 * If bucketsize is not a power-of-two, we may free
4674 * some pages at the end of hash table.
4677 unsigned long alloc_end = (unsigned long)table +
4678 (PAGE_SIZE << order);
4679 unsigned long used = (unsigned long)table +
4681 split_page(virt_to_page(table), order);
4682 while (used < alloc_end) {
4688 } while (!table && size > PAGE_SIZE && --log2qty);
4691 panic("Failed to allocate %s hash table\n", tablename);
4693 printk(KERN_INFO "%s hash table entries: %d (order: %d, %lu bytes)\n",
4696 ilog2(size) - PAGE_SHIFT,
4700 *_hash_shift = log2qty;
4702 *_hash_mask = (1 << log2qty) - 1;
4705 * If hashdist is set, the table allocation is done with __vmalloc()
4706 * which invokes the kmemleak_alloc() callback. This function may also
4707 * be called before the slab and kmemleak are initialised when
4708 * kmemleak simply buffers the request to be executed later
4709 * (GFP_ATOMIC flag ignored in this case).
4712 kmemleak_alloc(table, size, 1, GFP_ATOMIC);
4717 /* Return a pointer to the bitmap storing bits affecting a block of pages */
4718 static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
4721 #ifdef CONFIG_SPARSEMEM
4722 return __pfn_to_section(pfn)->pageblock_flags;
4724 return zone->pageblock_flags;
4725 #endif /* CONFIG_SPARSEMEM */
4728 static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
4730 #ifdef CONFIG_SPARSEMEM
4731 pfn &= (PAGES_PER_SECTION-1);
4732 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
4734 pfn = pfn - zone->zone_start_pfn;
4735 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
4736 #endif /* CONFIG_SPARSEMEM */
4740 * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
4741 * @page: The page within the block of interest
4742 * @start_bitidx: The first bit of interest to retrieve
4743 * @end_bitidx: The last bit of interest
4744 * returns pageblock_bits flags
4746 unsigned long get_pageblock_flags_group(struct page *page,
4747 int start_bitidx, int end_bitidx)
4750 unsigned long *bitmap;
4751 unsigned long pfn, bitidx;
4752 unsigned long flags = 0;
4753 unsigned long value = 1;
4755 zone = page_zone(page);
4756 pfn = page_to_pfn(page);
4757 bitmap = get_pageblock_bitmap(zone, pfn);
4758 bitidx = pfn_to_bitidx(zone, pfn);
4760 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
4761 if (test_bit(bitidx + start_bitidx, bitmap))
4768 * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
4769 * @page: The page within the block of interest
4770 * @start_bitidx: The first bit of interest
4771 * @end_bitidx: The last bit of interest
4772 * @flags: The flags to set
4774 void set_pageblock_flags_group(struct page *page, unsigned long flags,
4775 int start_bitidx, int end_bitidx)
4778 unsigned long *bitmap;
4779 unsigned long pfn, bitidx;
4780 unsigned long value = 1;
4782 zone = page_zone(page);
4783 pfn = page_to_pfn(page);
4784 bitmap = get_pageblock_bitmap(zone, pfn);
4785 bitidx = pfn_to_bitidx(zone, pfn);
4786 VM_BUG_ON(pfn < zone->zone_start_pfn);
4787 VM_BUG_ON(pfn >= zone->zone_start_pfn + zone->spanned_pages);
4789 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
4791 __set_bit(bitidx + start_bitidx, bitmap);
4793 __clear_bit(bitidx + start_bitidx, bitmap);
4797 * This is designed as sub function...plz see page_isolation.c also.
4798 * set/clear page block's type to be ISOLATE.
4799 * page allocater never alloc memory from ISOLATE block.
4802 int set_migratetype_isolate(struct page *page)
4805 unsigned long flags;
4808 zone = page_zone(page);
4809 spin_lock_irqsave(&zone->lock, flags);
4811 * In future, more migrate types will be able to be isolation target.
4813 if (get_pageblock_migratetype(page) != MIGRATE_MOVABLE)
4815 set_pageblock_migratetype(page, MIGRATE_ISOLATE);
4816 move_freepages_block(zone, page, MIGRATE_ISOLATE);
4819 spin_unlock_irqrestore(&zone->lock, flags);
4825 void unset_migratetype_isolate(struct page *page)
4828 unsigned long flags;
4829 zone = page_zone(page);
4830 spin_lock_irqsave(&zone->lock, flags);
4831 if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE)
4833 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
4834 move_freepages_block(zone, page, MIGRATE_MOVABLE);
4836 spin_unlock_irqrestore(&zone->lock, flags);
4839 #ifdef CONFIG_MEMORY_HOTREMOVE
4841 * All pages in the range must be isolated before calling this.
4844 __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
4850 unsigned long flags;
4851 /* find the first valid pfn */
4852 for (pfn = start_pfn; pfn < end_pfn; pfn++)
4857 zone = page_zone(pfn_to_page(pfn));
4858 spin_lock_irqsave(&zone->lock, flags);
4860 while (pfn < end_pfn) {
4861 if (!pfn_valid(pfn)) {
4865 page = pfn_to_page(pfn);
4866 BUG_ON(page_count(page));
4867 BUG_ON(!PageBuddy(page));
4868 order = page_order(page);
4869 #ifdef CONFIG_DEBUG_VM
4870 printk(KERN_INFO "remove from free list %lx %d %lx\n",
4871 pfn, 1 << order, end_pfn);
4873 list_del(&page->lru);
4874 rmv_page_order(page);
4875 zone->free_area[order].nr_free--;
4876 __mod_zone_page_state(zone, NR_FREE_PAGES,
4878 for (i = 0; i < (1 << order); i++)
4879 SetPageReserved((page+i));
4880 pfn += (1 << order);
4882 spin_unlock_irqrestore(&zone->lock, flags);