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>
50 #include <asm/tlbflush.h>
51 #include <asm/div64.h>
55 * Array of node states.
57 nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
58 [N_POSSIBLE] = NODE_MASK_ALL,
59 [N_ONLINE] = { { [0] = 1UL } },
61 [N_NORMAL_MEMORY] = { { [0] = 1UL } },
63 [N_HIGH_MEMORY] = { { [0] = 1UL } },
65 [N_CPU] = { { [0] = 1UL } },
68 EXPORT_SYMBOL(node_states);
70 unsigned long totalram_pages __read_mostly;
71 unsigned long totalreserve_pages __read_mostly;
72 unsigned long highest_memmap_pfn __read_mostly;
73 int percpu_pagelist_fraction;
75 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
76 int pageblock_order __read_mostly;
79 static void __free_pages_ok(struct page *page, unsigned int order);
82 * results with 256, 32 in the lowmem_reserve sysctl:
83 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
84 * 1G machine -> (16M dma, 784M normal, 224M high)
85 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
86 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
87 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
89 * TBD: should special case ZONE_DMA32 machines here - in those we normally
90 * don't need any ZONE_NORMAL reservation
92 int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
93 #ifdef CONFIG_ZONE_DMA
96 #ifdef CONFIG_ZONE_DMA32
105 EXPORT_SYMBOL(totalram_pages);
107 static char * const zone_names[MAX_NR_ZONES] = {
108 #ifdef CONFIG_ZONE_DMA
111 #ifdef CONFIG_ZONE_DMA32
115 #ifdef CONFIG_HIGHMEM
121 int min_free_kbytes = 1024;
123 unsigned long __meminitdata nr_kernel_pages;
124 unsigned long __meminitdata nr_all_pages;
125 static unsigned long __meminitdata dma_reserve;
127 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
129 * MAX_ACTIVE_REGIONS determines the maximum number of distinct
130 * ranges of memory (RAM) that may be registered with add_active_range().
131 * Ranges passed to add_active_range() will be merged if possible
132 * so the number of times add_active_range() can be called is
133 * related to the number of nodes and the number of holes
135 #ifdef CONFIG_MAX_ACTIVE_REGIONS
136 /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
137 #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
139 #if MAX_NUMNODES >= 32
140 /* If there can be many nodes, allow up to 50 holes per node */
141 #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
143 /* By default, allow up to 256 distinct regions */
144 #define MAX_ACTIVE_REGIONS 256
148 static struct node_active_region __meminitdata early_node_map[MAX_ACTIVE_REGIONS];
149 static int __meminitdata nr_nodemap_entries;
150 static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
151 static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
152 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
153 static unsigned long __meminitdata node_boundary_start_pfn[MAX_NUMNODES];
154 static unsigned long __meminitdata node_boundary_end_pfn[MAX_NUMNODES];
155 #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
156 static unsigned long __initdata required_kernelcore;
157 static unsigned long __initdata required_movablecore;
158 static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
160 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
162 EXPORT_SYMBOL(movable_zone);
163 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
166 int nr_node_ids __read_mostly = MAX_NUMNODES;
167 EXPORT_SYMBOL(nr_node_ids);
170 int page_group_by_mobility_disabled __read_mostly;
172 static void set_pageblock_migratetype(struct page *page, int migratetype)
174 set_pageblock_flags_group(page, (unsigned long)migratetype,
175 PB_migrate, PB_migrate_end);
178 #ifdef CONFIG_DEBUG_VM
179 static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
183 unsigned long pfn = page_to_pfn(page);
186 seq = zone_span_seqbegin(zone);
187 if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
189 else if (pfn < zone->zone_start_pfn)
191 } while (zone_span_seqretry(zone, seq));
196 static int page_is_consistent(struct zone *zone, struct page *page)
198 if (!pfn_valid_within(page_to_pfn(page)))
200 if (zone != page_zone(page))
206 * Temporary debugging check for pages not lying within a given zone.
208 static int bad_range(struct zone *zone, struct page *page)
210 if (page_outside_zone_boundaries(zone, page))
212 if (!page_is_consistent(zone, page))
218 static inline int bad_range(struct zone *zone, struct page *page)
224 static void bad_page(struct page *page)
226 printk(KERN_EMERG "Bad page state in process %s pfn:%05lx\n",
227 current->comm, page_to_pfn(page));
229 "page:%p flags:%p count:%d mapcount:%d mapping:%p index:%lx\n",
230 page, (void *)page->flags, page_count(page),
231 page_mapcount(page), page->mapping, page->index);
232 printk(KERN_EMERG "Trying to fix it up, but a reboot is needed\n");
236 /* Leave bad fields for debug, except PageBuddy could make trouble */
237 __ClearPageBuddy(page);
238 add_taint(TAINT_BAD_PAGE);
242 * Higher-order pages are called "compound pages". They are structured thusly:
244 * The first PAGE_SIZE page is called the "head page".
246 * The remaining PAGE_SIZE pages are called "tail pages".
248 * All pages have PG_compound set. All pages have their ->private pointing at
249 * the head page (even the head page has this).
251 * The first tail page's ->lru.next holds the address of the compound page's
252 * put_page() function. Its ->lru.prev holds the order of allocation.
253 * This usage means that zero-order pages may not be compound.
256 static void free_compound_page(struct page *page)
258 __free_pages_ok(page, compound_order(page));
261 void prep_compound_page(struct page *page, unsigned long order)
264 int nr_pages = 1 << order;
266 set_compound_page_dtor(page, free_compound_page);
267 set_compound_order(page, order);
269 for (i = 1; i < nr_pages; i++) {
270 struct page *p = page + i;
273 p->first_page = page;
277 #ifdef CONFIG_HUGETLBFS
278 void prep_compound_gigantic_page(struct page *page, unsigned long order)
281 int nr_pages = 1 << order;
282 struct page *p = page + 1;
284 set_compound_page_dtor(page, free_compound_page);
285 set_compound_order(page, order);
287 for (i = 1; i < nr_pages; i++, p = mem_map_next(p, page, i)) {
289 p->first_page = page;
294 static int destroy_compound_page(struct page *page, unsigned long order)
297 int nr_pages = 1 << order;
300 if (unlikely(compound_order(page) != order) ||
301 unlikely(!PageHead(page))) {
306 __ClearPageHead(page);
308 for (i = 1; i < nr_pages; i++) {
309 struct page *p = page + i;
311 if (unlikely(!PageTail(p) | (p->first_page != page))) {
321 static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
326 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
327 * and __GFP_HIGHMEM from hard or soft interrupt context.
329 VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
330 for (i = 0; i < (1 << order); i++)
331 clear_highpage(page + i);
334 static inline void set_page_order(struct page *page, int order)
336 set_page_private(page, order);
337 __SetPageBuddy(page);
340 static inline void rmv_page_order(struct page *page)
342 __ClearPageBuddy(page);
343 set_page_private(page, 0);
347 * Locate the struct page for both the matching buddy in our
348 * pair (buddy1) and the combined O(n+1) page they form (page).
350 * 1) Any buddy B1 will have an order O twin B2 which satisfies
351 * the following equation:
353 * For example, if the starting buddy (buddy2) is #8 its order
355 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
357 * 2) Any buddy B will have an order O+1 parent P which
358 * satisfies the following equation:
361 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
363 static inline struct page *
364 __page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order)
366 unsigned long buddy_idx = page_idx ^ (1 << order);
368 return page + (buddy_idx - page_idx);
371 static inline unsigned long
372 __find_combined_index(unsigned long page_idx, unsigned int order)
374 return (page_idx & ~(1 << order));
378 * This function checks whether a page is free && is the buddy
379 * we can do coalesce a page and its buddy if
380 * (a) the buddy is not in a hole &&
381 * (b) the buddy is in the buddy system &&
382 * (c) a page and its buddy have the same order &&
383 * (d) a page and its buddy are in the same zone.
385 * For recording whether a page is in the buddy system, we use PG_buddy.
386 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
388 * For recording page's order, we use page_private(page).
390 static inline int page_is_buddy(struct page *page, struct page *buddy,
393 if (!pfn_valid_within(page_to_pfn(buddy)))
396 if (page_zone_id(page) != page_zone_id(buddy))
399 if (PageBuddy(buddy) && page_order(buddy) == order) {
400 BUG_ON(page_count(buddy) != 0);
407 * Freeing function for a buddy system allocator.
409 * The concept of a buddy system is to maintain direct-mapped table
410 * (containing bit values) for memory blocks of various "orders".
411 * The bottom level table contains the map for the smallest allocatable
412 * units of memory (here, pages), and each level above it describes
413 * pairs of units from the levels below, hence, "buddies".
414 * At a high level, all that happens here is marking the table entry
415 * at the bottom level available, and propagating the changes upward
416 * as necessary, plus some accounting needed to play nicely with other
417 * parts of the VM system.
418 * At each level, we keep a list of pages, which are heads of continuous
419 * free pages of length of (1 << order) and marked with PG_buddy. Page's
420 * order is recorded in page_private(page) field.
421 * So when we are allocating or freeing one, we can derive the state of the
422 * other. That is, if we allocate a small block, and both were
423 * free, the remainder of the region must be split into blocks.
424 * If a block is freed, and its buddy is also free, then this
425 * triggers coalescing into a block of larger size.
430 static inline void __free_one_page(struct page *page,
431 struct zone *zone, unsigned int order)
433 unsigned long page_idx;
434 int order_size = 1 << order;
435 int migratetype = get_pageblock_migratetype(page);
437 if (unlikely(PageCompound(page)))
438 if (unlikely(destroy_compound_page(page, order)))
441 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
443 VM_BUG_ON(page_idx & (order_size - 1));
444 VM_BUG_ON(bad_range(zone, page));
446 __mod_zone_page_state(zone, NR_FREE_PAGES, order_size);
447 while (order < MAX_ORDER-1) {
448 unsigned long combined_idx;
451 buddy = __page_find_buddy(page, page_idx, order);
452 if (!page_is_buddy(page, buddy, order))
455 /* Our buddy is free, merge with it and move up one order. */
456 list_del(&buddy->lru);
457 zone->free_area[order].nr_free--;
458 rmv_page_order(buddy);
459 combined_idx = __find_combined_index(page_idx, order);
460 page = page + (combined_idx - page_idx);
461 page_idx = combined_idx;
464 set_page_order(page, order);
466 &zone->free_area[order].free_list[migratetype]);
467 zone->free_area[order].nr_free++;
470 static inline int free_pages_check(struct page *page)
472 free_page_mlock(page);
473 if (unlikely(page_mapcount(page) |
474 (page->mapping != NULL) |
475 (page_count(page) != 0) |
476 (page->flags & PAGE_FLAGS_CHECK_AT_FREE))) {
480 if (page->flags & PAGE_FLAGS_CHECK_AT_PREP)
481 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
486 * Frees a list of pages.
487 * Assumes all pages on list are in same zone, and of same order.
488 * count is the number of pages to free.
490 * If the zone was previously in an "all pages pinned" state then look to
491 * see if this freeing clears that state.
493 * And clear the zone's pages_scanned counter, to hold off the "all pages are
494 * pinned" detection logic.
496 static void free_pages_bulk(struct zone *zone, int count,
497 struct list_head *list, int order)
499 spin_lock(&zone->lock);
500 zone_clear_flag(zone, ZONE_ALL_UNRECLAIMABLE);
501 zone->pages_scanned = 0;
505 VM_BUG_ON(list_empty(list));
506 page = list_entry(list->prev, struct page, lru);
507 /* have to delete it as __free_one_page list manipulates */
508 list_del(&page->lru);
509 __free_one_page(page, zone, order);
511 spin_unlock(&zone->lock);
514 static void free_one_page(struct zone *zone, struct page *page, int order)
516 spin_lock(&zone->lock);
517 zone_clear_flag(zone, ZONE_ALL_UNRECLAIMABLE);
518 zone->pages_scanned = 0;
519 __free_one_page(page, zone, order);
520 spin_unlock(&zone->lock);
523 static void __free_pages_ok(struct page *page, unsigned int order)
529 for (i = 0 ; i < (1 << order) ; ++i)
530 bad += free_pages_check(page + i);
534 if (!PageHighMem(page)) {
535 debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order);
536 debug_check_no_obj_freed(page_address(page),
539 arch_free_page(page, order);
540 kernel_map_pages(page, 1 << order, 0);
542 local_irq_save(flags);
543 __count_vm_events(PGFREE, 1 << order);
544 free_one_page(page_zone(page), page, order);
545 local_irq_restore(flags);
549 * permit the bootmem allocator to evade page validation on high-order frees
551 void __meminit __free_pages_bootmem(struct page *page, unsigned int order)
554 __ClearPageReserved(page);
555 set_page_count(page, 0);
556 set_page_refcounted(page);
562 for (loop = 0; loop < BITS_PER_LONG; loop++) {
563 struct page *p = &page[loop];
565 if (loop + 1 < BITS_PER_LONG)
567 __ClearPageReserved(p);
568 set_page_count(p, 0);
571 set_page_refcounted(page);
572 __free_pages(page, order);
578 * The order of subdivision here is critical for the IO subsystem.
579 * Please do not alter this order without good reasons and regression
580 * testing. Specifically, as large blocks of memory are subdivided,
581 * the order in which smaller blocks are delivered depends on the order
582 * they're subdivided in this function. This is the primary factor
583 * influencing the order in which pages are delivered to the IO
584 * subsystem according to empirical testing, and this is also justified
585 * by considering the behavior of a buddy system containing a single
586 * large block of memory acted on by a series of small allocations.
587 * This behavior is a critical factor in sglist merging's success.
591 static inline void expand(struct zone *zone, struct page *page,
592 int low, int high, struct free_area *area,
595 unsigned long size = 1 << high;
601 VM_BUG_ON(bad_range(zone, &page[size]));
602 list_add(&page[size].lru, &area->free_list[migratetype]);
604 set_page_order(&page[size], high);
609 * This page is about to be returned from the page allocator
611 static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
613 if (unlikely(page_mapcount(page) |
614 (page->mapping != NULL) |
615 (page_count(page) != 0) |
616 (page->flags & PAGE_FLAGS_CHECK_AT_PREP))) {
621 set_page_private(page, 0);
622 set_page_refcounted(page);
624 arch_alloc_page(page, order);
625 kernel_map_pages(page, 1 << order, 1);
627 if (gfp_flags & __GFP_ZERO)
628 prep_zero_page(page, order, gfp_flags);
630 if (order && (gfp_flags & __GFP_COMP))
631 prep_compound_page(page, order);
637 * Go through the free lists for the given migratetype and remove
638 * the smallest available page from the freelists
640 static struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
643 unsigned int current_order;
644 struct free_area * area;
647 /* Find a page of the appropriate size in the preferred list */
648 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
649 area = &(zone->free_area[current_order]);
650 if (list_empty(&area->free_list[migratetype]))
653 page = list_entry(area->free_list[migratetype].next,
655 list_del(&page->lru);
656 rmv_page_order(page);
658 __mod_zone_page_state(zone, NR_FREE_PAGES, - (1UL << order));
659 expand(zone, page, order, current_order, area, migratetype);
668 * This array describes the order lists are fallen back to when
669 * the free lists for the desirable migrate type are depleted
671 static int fallbacks[MIGRATE_TYPES][MIGRATE_TYPES-1] = {
672 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
673 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
674 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
675 [MIGRATE_RESERVE] = { MIGRATE_RESERVE, MIGRATE_RESERVE, MIGRATE_RESERVE }, /* Never used */
679 * Move the free pages in a range to the free lists of the requested type.
680 * Note that start_page and end_pages are not aligned on a pageblock
681 * boundary. If alignment is required, use move_freepages_block()
683 static int move_freepages(struct zone *zone,
684 struct page *start_page, struct page *end_page,
691 #ifndef CONFIG_HOLES_IN_ZONE
693 * page_zone is not safe to call in this context when
694 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
695 * anyway as we check zone boundaries in move_freepages_block().
696 * Remove at a later date when no bug reports exist related to
697 * grouping pages by mobility
699 BUG_ON(page_zone(start_page) != page_zone(end_page));
702 for (page = start_page; page <= end_page;) {
703 /* Make sure we are not inadvertently changing nodes */
704 VM_BUG_ON(page_to_nid(page) != zone_to_nid(zone));
706 if (!pfn_valid_within(page_to_pfn(page))) {
711 if (!PageBuddy(page)) {
716 order = page_order(page);
717 list_del(&page->lru);
719 &zone->free_area[order].free_list[migratetype]);
721 pages_moved += 1 << order;
727 static int move_freepages_block(struct zone *zone, struct page *page,
730 unsigned long start_pfn, end_pfn;
731 struct page *start_page, *end_page;
733 start_pfn = page_to_pfn(page);
734 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
735 start_page = pfn_to_page(start_pfn);
736 end_page = start_page + pageblock_nr_pages - 1;
737 end_pfn = start_pfn + pageblock_nr_pages - 1;
739 /* Do not cross zone boundaries */
740 if (start_pfn < zone->zone_start_pfn)
742 if (end_pfn >= zone->zone_start_pfn + zone->spanned_pages)
745 return move_freepages(zone, start_page, end_page, migratetype);
748 /* Remove an element from the buddy allocator from the fallback list */
749 static struct page *__rmqueue_fallback(struct zone *zone, int order,
750 int start_migratetype)
752 struct free_area * area;
757 /* Find the largest possible block of pages in the other list */
758 for (current_order = MAX_ORDER-1; current_order >= order;
760 for (i = 0; i < MIGRATE_TYPES - 1; i++) {
761 migratetype = fallbacks[start_migratetype][i];
763 /* MIGRATE_RESERVE handled later if necessary */
764 if (migratetype == MIGRATE_RESERVE)
767 area = &(zone->free_area[current_order]);
768 if (list_empty(&area->free_list[migratetype]))
771 page = list_entry(area->free_list[migratetype].next,
776 * If breaking a large block of pages, move all free
777 * pages to the preferred allocation list. If falling
778 * back for a reclaimable kernel allocation, be more
779 * agressive about taking ownership of free pages
781 if (unlikely(current_order >= (pageblock_order >> 1)) ||
782 start_migratetype == MIGRATE_RECLAIMABLE) {
784 pages = move_freepages_block(zone, page,
787 /* Claim the whole block if over half of it is free */
788 if (pages >= (1 << (pageblock_order-1)))
789 set_pageblock_migratetype(page,
792 migratetype = start_migratetype;
795 /* Remove the page from the freelists */
796 list_del(&page->lru);
797 rmv_page_order(page);
798 __mod_zone_page_state(zone, NR_FREE_PAGES,
801 if (current_order == pageblock_order)
802 set_pageblock_migratetype(page,
805 expand(zone, page, order, current_order, area, migratetype);
810 /* Use MIGRATE_RESERVE rather than fail an allocation */
811 return __rmqueue_smallest(zone, order, MIGRATE_RESERVE);
815 * Do the hard work of removing an element from the buddy allocator.
816 * Call me with the zone->lock already held.
818 static struct page *__rmqueue(struct zone *zone, unsigned int order,
823 page = __rmqueue_smallest(zone, order, migratetype);
826 page = __rmqueue_fallback(zone, order, migratetype);
832 * Obtain a specified number of elements from the buddy allocator, all under
833 * a single hold of the lock, for efficiency. Add them to the supplied list.
834 * Returns the number of new pages which were placed at *list.
836 static int rmqueue_bulk(struct zone *zone, unsigned int order,
837 unsigned long count, struct list_head *list,
842 spin_lock(&zone->lock);
843 for (i = 0; i < count; ++i) {
844 struct page *page = __rmqueue(zone, order, migratetype);
845 if (unlikely(page == NULL))
849 * Split buddy pages returned by expand() are received here
850 * in physical page order. The page is added to the callers and
851 * list and the list head then moves forward. From the callers
852 * perspective, the linked list is ordered by page number in
853 * some conditions. This is useful for IO devices that can
854 * merge IO requests if the physical pages are ordered
857 list_add(&page->lru, list);
858 set_page_private(page, migratetype);
861 spin_unlock(&zone->lock);
867 * Called from the vmstat counter updater to drain pagesets of this
868 * currently executing processor on remote nodes after they have
871 * Note that this function must be called with the thread pinned to
872 * a single processor.
874 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
879 local_irq_save(flags);
880 if (pcp->count >= pcp->batch)
881 to_drain = pcp->batch;
883 to_drain = pcp->count;
884 free_pages_bulk(zone, to_drain, &pcp->list, 0);
885 pcp->count -= to_drain;
886 local_irq_restore(flags);
891 * Drain pages of the indicated processor.
893 * The processor must either be the current processor and the
894 * thread pinned to the current processor or a processor that
897 static void drain_pages(unsigned int cpu)
902 for_each_zone(zone) {
903 struct per_cpu_pageset *pset;
904 struct per_cpu_pages *pcp;
906 if (!populated_zone(zone))
909 pset = zone_pcp(zone, cpu);
912 local_irq_save(flags);
913 free_pages_bulk(zone, pcp->count, &pcp->list, 0);
915 local_irq_restore(flags);
920 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
922 void drain_local_pages(void *arg)
924 drain_pages(smp_processor_id());
928 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
930 void drain_all_pages(void)
932 on_each_cpu(drain_local_pages, NULL, 1);
935 #ifdef CONFIG_HIBERNATION
937 void mark_free_pages(struct zone *zone)
939 unsigned long pfn, max_zone_pfn;
942 struct list_head *curr;
944 if (!zone->spanned_pages)
947 spin_lock_irqsave(&zone->lock, flags);
949 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
950 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
951 if (pfn_valid(pfn)) {
952 struct page *page = pfn_to_page(pfn);
954 if (!swsusp_page_is_forbidden(page))
955 swsusp_unset_page_free(page);
958 for_each_migratetype_order(order, t) {
959 list_for_each(curr, &zone->free_area[order].free_list[t]) {
962 pfn = page_to_pfn(list_entry(curr, struct page, lru));
963 for (i = 0; i < (1UL << order); i++)
964 swsusp_set_page_free(pfn_to_page(pfn + i));
967 spin_unlock_irqrestore(&zone->lock, flags);
969 #endif /* CONFIG_PM */
972 * Free a 0-order page
974 static void free_hot_cold_page(struct page *page, int cold)
976 struct zone *zone = page_zone(page);
977 struct per_cpu_pages *pcp;
981 page->mapping = NULL;
982 if (free_pages_check(page))
985 if (!PageHighMem(page)) {
986 debug_check_no_locks_freed(page_address(page), PAGE_SIZE);
987 debug_check_no_obj_freed(page_address(page), PAGE_SIZE);
989 arch_free_page(page, 0);
990 kernel_map_pages(page, 1, 0);
992 pcp = &zone_pcp(zone, get_cpu())->pcp;
993 local_irq_save(flags);
994 __count_vm_event(PGFREE);
996 list_add_tail(&page->lru, &pcp->list);
998 list_add(&page->lru, &pcp->list);
999 set_page_private(page, get_pageblock_migratetype(page));
1001 if (pcp->count >= pcp->high) {
1002 free_pages_bulk(zone, pcp->batch, &pcp->list, 0);
1003 pcp->count -= pcp->batch;
1005 local_irq_restore(flags);
1009 void free_hot_page(struct page *page)
1011 free_hot_cold_page(page, 0);
1014 void free_cold_page(struct page *page)
1016 free_hot_cold_page(page, 1);
1020 * split_page takes a non-compound higher-order page, and splits it into
1021 * n (1<<order) sub-pages: page[0..n]
1022 * Each sub-page must be freed individually.
1024 * Note: this is probably too low level an operation for use in drivers.
1025 * Please consult with lkml before using this in your driver.
1027 void split_page(struct page *page, unsigned int order)
1031 VM_BUG_ON(PageCompound(page));
1032 VM_BUG_ON(!page_count(page));
1033 for (i = 1; i < (1 << order); i++)
1034 set_page_refcounted(page + i);
1038 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1039 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1042 static struct page *buffered_rmqueue(struct zone *preferred_zone,
1043 struct zone *zone, int order, gfp_t gfp_flags)
1045 unsigned long flags;
1047 int cold = !!(gfp_flags & __GFP_COLD);
1049 int migratetype = allocflags_to_migratetype(gfp_flags);
1053 if (likely(order == 0)) {
1054 struct per_cpu_pages *pcp;
1056 pcp = &zone_pcp(zone, cpu)->pcp;
1057 local_irq_save(flags);
1059 pcp->count = rmqueue_bulk(zone, 0,
1060 pcp->batch, &pcp->list, migratetype);
1061 if (unlikely(!pcp->count))
1065 /* Find a page of the appropriate migrate type */
1067 list_for_each_entry_reverse(page, &pcp->list, lru)
1068 if (page_private(page) == migratetype)
1071 list_for_each_entry(page, &pcp->list, lru)
1072 if (page_private(page) == migratetype)
1076 /* Allocate more to the pcp list if necessary */
1077 if (unlikely(&page->lru == &pcp->list)) {
1078 pcp->count += rmqueue_bulk(zone, 0,
1079 pcp->batch, &pcp->list, migratetype);
1080 page = list_entry(pcp->list.next, struct page, lru);
1083 list_del(&page->lru);
1086 spin_lock_irqsave(&zone->lock, flags);
1087 page = __rmqueue(zone, order, migratetype);
1088 spin_unlock(&zone->lock);
1093 __count_zone_vm_events(PGALLOC, zone, 1 << order);
1094 zone_statistics(preferred_zone, zone);
1095 local_irq_restore(flags);
1098 VM_BUG_ON(bad_range(zone, page));
1099 if (prep_new_page(page, order, gfp_flags))
1104 local_irq_restore(flags);
1109 #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
1110 #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
1111 #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
1112 #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
1113 #define ALLOC_HARDER 0x10 /* try to alloc harder */
1114 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1115 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
1117 #ifdef CONFIG_FAIL_PAGE_ALLOC
1119 static struct fail_page_alloc_attr {
1120 struct fault_attr attr;
1122 u32 ignore_gfp_highmem;
1123 u32 ignore_gfp_wait;
1126 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1128 struct dentry *ignore_gfp_highmem_file;
1129 struct dentry *ignore_gfp_wait_file;
1130 struct dentry *min_order_file;
1132 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1134 } fail_page_alloc = {
1135 .attr = FAULT_ATTR_INITIALIZER,
1136 .ignore_gfp_wait = 1,
1137 .ignore_gfp_highmem = 1,
1141 static int __init setup_fail_page_alloc(char *str)
1143 return setup_fault_attr(&fail_page_alloc.attr, str);
1145 __setup("fail_page_alloc=", setup_fail_page_alloc);
1147 static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1149 if (order < fail_page_alloc.min_order)
1151 if (gfp_mask & __GFP_NOFAIL)
1153 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
1155 if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
1158 return should_fail(&fail_page_alloc.attr, 1 << order);
1161 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1163 static int __init fail_page_alloc_debugfs(void)
1165 mode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
1169 err = init_fault_attr_dentries(&fail_page_alloc.attr,
1173 dir = fail_page_alloc.attr.dentries.dir;
1175 fail_page_alloc.ignore_gfp_wait_file =
1176 debugfs_create_bool("ignore-gfp-wait", mode, dir,
1177 &fail_page_alloc.ignore_gfp_wait);
1179 fail_page_alloc.ignore_gfp_highmem_file =
1180 debugfs_create_bool("ignore-gfp-highmem", mode, dir,
1181 &fail_page_alloc.ignore_gfp_highmem);
1182 fail_page_alloc.min_order_file =
1183 debugfs_create_u32("min-order", mode, dir,
1184 &fail_page_alloc.min_order);
1186 if (!fail_page_alloc.ignore_gfp_wait_file ||
1187 !fail_page_alloc.ignore_gfp_highmem_file ||
1188 !fail_page_alloc.min_order_file) {
1190 debugfs_remove(fail_page_alloc.ignore_gfp_wait_file);
1191 debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file);
1192 debugfs_remove(fail_page_alloc.min_order_file);
1193 cleanup_fault_attr_dentries(&fail_page_alloc.attr);
1199 late_initcall(fail_page_alloc_debugfs);
1201 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1203 #else /* CONFIG_FAIL_PAGE_ALLOC */
1205 static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1210 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1213 * Return 1 if free pages are above 'mark'. This takes into account the order
1214 * of the allocation.
1216 int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1217 int classzone_idx, int alloc_flags)
1219 /* free_pages my go negative - that's OK */
1221 long free_pages = zone_page_state(z, NR_FREE_PAGES) - (1 << order) + 1;
1224 if (alloc_flags & ALLOC_HIGH)
1226 if (alloc_flags & ALLOC_HARDER)
1229 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
1231 for (o = 0; o < order; o++) {
1232 /* At the next order, this order's pages become unavailable */
1233 free_pages -= z->free_area[o].nr_free << o;
1235 /* Require fewer higher order pages to be free */
1238 if (free_pages <= min)
1246 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1247 * skip over zones that are not allowed by the cpuset, or that have
1248 * been recently (in last second) found to be nearly full. See further
1249 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1250 * that have to skip over a lot of full or unallowed zones.
1252 * If the zonelist cache is present in the passed in zonelist, then
1253 * returns a pointer to the allowed node mask (either the current
1254 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
1256 * If the zonelist cache is not available for this zonelist, does
1257 * nothing and returns NULL.
1259 * If the fullzones BITMAP in the zonelist cache is stale (more than
1260 * a second since last zap'd) then we zap it out (clear its bits.)
1262 * We hold off even calling zlc_setup, until after we've checked the
1263 * first zone in the zonelist, on the theory that most allocations will
1264 * be satisfied from that first zone, so best to examine that zone as
1265 * quickly as we can.
1267 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1269 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1270 nodemask_t *allowednodes; /* zonelist_cache approximation */
1272 zlc = zonelist->zlcache_ptr;
1276 if (time_after(jiffies, zlc->last_full_zap + HZ)) {
1277 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1278 zlc->last_full_zap = jiffies;
1281 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1282 &cpuset_current_mems_allowed :
1283 &node_states[N_HIGH_MEMORY];
1284 return allowednodes;
1288 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1289 * if it is worth looking at further for free memory:
1290 * 1) Check that the zone isn't thought to be full (doesn't have its
1291 * bit set in the zonelist_cache fullzones BITMAP).
1292 * 2) Check that the zones node (obtained from the zonelist_cache
1293 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1294 * Return true (non-zero) if zone is worth looking at further, or
1295 * else return false (zero) if it is not.
1297 * This check -ignores- the distinction between various watermarks,
1298 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1299 * found to be full for any variation of these watermarks, it will
1300 * be considered full for up to one second by all requests, unless
1301 * we are so low on memory on all allowed nodes that we are forced
1302 * into the second scan of the zonelist.
1304 * In the second scan we ignore this zonelist cache and exactly
1305 * apply the watermarks to all zones, even it is slower to do so.
1306 * We are low on memory in the second scan, and should leave no stone
1307 * unturned looking for a free page.
1309 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
1310 nodemask_t *allowednodes)
1312 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1313 int i; /* index of *z in zonelist zones */
1314 int n; /* node that zone *z is on */
1316 zlc = zonelist->zlcache_ptr;
1320 i = z - zonelist->_zonerefs;
1323 /* This zone is worth trying if it is allowed but not full */
1324 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1328 * Given 'z' scanning a zonelist, set the corresponding bit in
1329 * zlc->fullzones, so that subsequent attempts to allocate a page
1330 * from that zone don't waste time re-examining it.
1332 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
1334 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1335 int i; /* index of *z in zonelist zones */
1337 zlc = zonelist->zlcache_ptr;
1341 i = z - zonelist->_zonerefs;
1343 set_bit(i, zlc->fullzones);
1346 #else /* CONFIG_NUMA */
1348 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1353 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
1354 nodemask_t *allowednodes)
1359 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
1362 #endif /* CONFIG_NUMA */
1365 * get_page_from_freelist goes through the zonelist trying to allocate
1368 static struct page *
1369 get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order,
1370 struct zonelist *zonelist, int high_zoneidx, int alloc_flags)
1373 struct page *page = NULL;
1375 struct zone *zone, *preferred_zone;
1376 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1377 int zlc_active = 0; /* set if using zonelist_cache */
1378 int did_zlc_setup = 0; /* just call zlc_setup() one time */
1380 (void)first_zones_zonelist(zonelist, high_zoneidx, nodemask,
1382 if (!preferred_zone)
1385 classzone_idx = zone_idx(preferred_zone);
1389 * Scan zonelist, looking for a zone with enough free.
1390 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1392 for_each_zone_zonelist_nodemask(zone, z, zonelist,
1393 high_zoneidx, nodemask) {
1394 if (NUMA_BUILD && zlc_active &&
1395 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1397 if ((alloc_flags & ALLOC_CPUSET) &&
1398 !cpuset_zone_allowed_softwall(zone, gfp_mask))
1401 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
1403 if (alloc_flags & ALLOC_WMARK_MIN)
1404 mark = zone->pages_min;
1405 else if (alloc_flags & ALLOC_WMARK_LOW)
1406 mark = zone->pages_low;
1408 mark = zone->pages_high;
1409 if (!zone_watermark_ok(zone, order, mark,
1410 classzone_idx, alloc_flags)) {
1411 if (!zone_reclaim_mode ||
1412 !zone_reclaim(zone, gfp_mask, order))
1413 goto this_zone_full;
1417 page = buffered_rmqueue(preferred_zone, zone, order, gfp_mask);
1422 zlc_mark_zone_full(zonelist, z);
1424 if (NUMA_BUILD && !did_zlc_setup) {
1425 /* we do zlc_setup after the first zone is tried */
1426 allowednodes = zlc_setup(zonelist, alloc_flags);
1432 if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) {
1433 /* Disable zlc cache for second zonelist scan */
1441 * This is the 'heart' of the zoned buddy allocator.
1444 __alloc_pages_internal(gfp_t gfp_mask, unsigned int order,
1445 struct zonelist *zonelist, nodemask_t *nodemask)
1447 const gfp_t wait = gfp_mask & __GFP_WAIT;
1448 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
1452 struct reclaim_state reclaim_state;
1453 struct task_struct *p = current;
1456 unsigned long did_some_progress;
1457 unsigned long pages_reclaimed = 0;
1459 might_sleep_if(wait);
1461 if (should_fail_alloc_page(gfp_mask, order))
1465 z = zonelist->_zonerefs; /* the list of zones suitable for gfp_mask */
1467 if (unlikely(!z->zone)) {
1469 * Happens if we have an empty zonelist as a result of
1470 * GFP_THISNODE being used on a memoryless node
1475 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order,
1476 zonelist, high_zoneidx, ALLOC_WMARK_LOW|ALLOC_CPUSET);
1481 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1482 * __GFP_NOWARN set) should not cause reclaim since the subsystem
1483 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1484 * using a larger set of nodes after it has established that the
1485 * allowed per node queues are empty and that nodes are
1488 if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
1491 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx)
1492 wakeup_kswapd(zone, order);
1495 * OK, we're below the kswapd watermark and have kicked background
1496 * reclaim. Now things get more complex, so set up alloc_flags according
1497 * to how we want to proceed.
1499 * The caller may dip into page reserves a bit more if the caller
1500 * cannot run direct reclaim, or if the caller has realtime scheduling
1501 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1502 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1504 alloc_flags = ALLOC_WMARK_MIN;
1505 if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait)
1506 alloc_flags |= ALLOC_HARDER;
1507 if (gfp_mask & __GFP_HIGH)
1508 alloc_flags |= ALLOC_HIGH;
1510 alloc_flags |= ALLOC_CPUSET;
1513 * Go through the zonelist again. Let __GFP_HIGH and allocations
1514 * coming from realtime tasks go deeper into reserves.
1516 * This is the last chance, in general, before the goto nopage.
1517 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1518 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1520 page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist,
1521 high_zoneidx, alloc_flags);
1525 /* This allocation should allow future memory freeing. */
1528 if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE)))
1529 && !in_interrupt()) {
1530 if (!(gfp_mask & __GFP_NOMEMALLOC)) {
1532 /* go through the zonelist yet again, ignoring mins */
1533 page = get_page_from_freelist(gfp_mask, nodemask, order,
1534 zonelist, high_zoneidx, ALLOC_NO_WATERMARKS);
1537 if (gfp_mask & __GFP_NOFAIL) {
1538 congestion_wait(WRITE, HZ/50);
1545 /* Atomic allocations - we can't balance anything */
1551 /* We now go into synchronous reclaim */
1552 cpuset_memory_pressure_bump();
1554 * The task's cpuset might have expanded its set of allowable nodes
1556 cpuset_update_task_memory_state();
1557 p->flags |= PF_MEMALLOC;
1558 reclaim_state.reclaimed_slab = 0;
1559 p->reclaim_state = &reclaim_state;
1561 did_some_progress = try_to_free_pages(zonelist, order, gfp_mask);
1563 p->reclaim_state = NULL;
1564 p->flags &= ~PF_MEMALLOC;
1571 if (likely(did_some_progress)) {
1572 page = get_page_from_freelist(gfp_mask, nodemask, order,
1573 zonelist, high_zoneidx, alloc_flags);
1576 } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
1577 if (!try_set_zone_oom(zonelist, gfp_mask)) {
1578 schedule_timeout_uninterruptible(1);
1583 * Go through the zonelist yet one more time, keep
1584 * very high watermark here, this is only to catch
1585 * a parallel oom killing, we must fail if we're still
1586 * under heavy pressure.
1588 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask,
1589 order, zonelist, high_zoneidx,
1590 ALLOC_WMARK_HIGH|ALLOC_CPUSET);
1592 clear_zonelist_oom(zonelist, gfp_mask);
1596 /* The OOM killer will not help higher order allocs so fail */
1597 if (order > PAGE_ALLOC_COSTLY_ORDER) {
1598 clear_zonelist_oom(zonelist, gfp_mask);
1602 out_of_memory(zonelist, gfp_mask, order);
1603 clear_zonelist_oom(zonelist, gfp_mask);
1608 * Don't let big-order allocations loop unless the caller explicitly
1609 * requests that. Wait for some write requests to complete then retry.
1611 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
1612 * means __GFP_NOFAIL, but that may not be true in other
1615 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
1616 * specified, then we retry until we no longer reclaim any pages
1617 * (above), or we've reclaimed an order of pages at least as
1618 * large as the allocation's order. In both cases, if the
1619 * allocation still fails, we stop retrying.
1621 pages_reclaimed += did_some_progress;
1623 if (!(gfp_mask & __GFP_NORETRY)) {
1624 if (order <= PAGE_ALLOC_COSTLY_ORDER) {
1627 if (gfp_mask & __GFP_REPEAT &&
1628 pages_reclaimed < (1 << order))
1631 if (gfp_mask & __GFP_NOFAIL)
1635 congestion_wait(WRITE, HZ/50);
1640 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
1641 printk(KERN_WARNING "%s: page allocation failure."
1642 " order:%d, mode:0x%x\n",
1643 p->comm, order, gfp_mask);
1650 EXPORT_SYMBOL(__alloc_pages_internal);
1653 * Common helper functions.
1655 unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
1658 page = alloc_pages(gfp_mask, order);
1661 return (unsigned long) page_address(page);
1664 EXPORT_SYMBOL(__get_free_pages);
1666 unsigned long get_zeroed_page(gfp_t gfp_mask)
1671 * get_zeroed_page() returns a 32-bit address, which cannot represent
1674 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
1676 page = alloc_pages(gfp_mask | __GFP_ZERO, 0);
1678 return (unsigned long) page_address(page);
1682 EXPORT_SYMBOL(get_zeroed_page);
1684 void __pagevec_free(struct pagevec *pvec)
1686 int i = pagevec_count(pvec);
1689 free_hot_cold_page(pvec->pages[i], pvec->cold);
1692 void __free_pages(struct page *page, unsigned int order)
1694 if (put_page_testzero(page)) {
1696 free_hot_page(page);
1698 __free_pages_ok(page, order);
1702 EXPORT_SYMBOL(__free_pages);
1704 void free_pages(unsigned long addr, unsigned int order)
1707 VM_BUG_ON(!virt_addr_valid((void *)addr));
1708 __free_pages(virt_to_page((void *)addr), order);
1712 EXPORT_SYMBOL(free_pages);
1715 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
1716 * @size: the number of bytes to allocate
1717 * @gfp_mask: GFP flags for the allocation
1719 * This function is similar to alloc_pages(), except that it allocates the
1720 * minimum number of pages to satisfy the request. alloc_pages() can only
1721 * allocate memory in power-of-two pages.
1723 * This function is also limited by MAX_ORDER.
1725 * Memory allocated by this function must be released by free_pages_exact().
1727 void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
1729 unsigned int order = get_order(size);
1732 addr = __get_free_pages(gfp_mask, order);
1734 unsigned long alloc_end = addr + (PAGE_SIZE << order);
1735 unsigned long used = addr + PAGE_ALIGN(size);
1737 split_page(virt_to_page(addr), order);
1738 while (used < alloc_end) {
1744 return (void *)addr;
1746 EXPORT_SYMBOL(alloc_pages_exact);
1749 * free_pages_exact - release memory allocated via alloc_pages_exact()
1750 * @virt: the value returned by alloc_pages_exact.
1751 * @size: size of allocation, same value as passed to alloc_pages_exact().
1753 * Release the memory allocated by a previous call to alloc_pages_exact.
1755 void free_pages_exact(void *virt, size_t size)
1757 unsigned long addr = (unsigned long)virt;
1758 unsigned long end = addr + PAGE_ALIGN(size);
1760 while (addr < end) {
1765 EXPORT_SYMBOL(free_pages_exact);
1767 static unsigned int nr_free_zone_pages(int offset)
1772 /* Just pick one node, since fallback list is circular */
1773 unsigned int sum = 0;
1775 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
1777 for_each_zone_zonelist(zone, z, zonelist, offset) {
1778 unsigned long size = zone->present_pages;
1779 unsigned long high = zone->pages_high;
1788 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1790 unsigned int nr_free_buffer_pages(void)
1792 return nr_free_zone_pages(gfp_zone(GFP_USER));
1794 EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
1797 * Amount of free RAM allocatable within all zones
1799 unsigned int nr_free_pagecache_pages(void)
1801 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
1804 static inline void show_node(struct zone *zone)
1807 printk("Node %d ", zone_to_nid(zone));
1810 void si_meminfo(struct sysinfo *val)
1812 val->totalram = totalram_pages;
1814 val->freeram = global_page_state(NR_FREE_PAGES);
1815 val->bufferram = nr_blockdev_pages();
1816 val->totalhigh = totalhigh_pages;
1817 val->freehigh = nr_free_highpages();
1818 val->mem_unit = PAGE_SIZE;
1821 EXPORT_SYMBOL(si_meminfo);
1824 void si_meminfo_node(struct sysinfo *val, int nid)
1826 pg_data_t *pgdat = NODE_DATA(nid);
1828 val->totalram = pgdat->node_present_pages;
1829 val->freeram = node_page_state(nid, NR_FREE_PAGES);
1830 #ifdef CONFIG_HIGHMEM
1831 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1832 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
1838 val->mem_unit = PAGE_SIZE;
1842 #define K(x) ((x) << (PAGE_SHIFT-10))
1845 * Show free area list (used inside shift_scroll-lock stuff)
1846 * We also calculate the percentage fragmentation. We do this by counting the
1847 * memory on each free list with the exception of the first item on the list.
1849 void show_free_areas(void)
1854 for_each_zone(zone) {
1855 if (!populated_zone(zone))
1859 printk("%s per-cpu:\n", zone->name);
1861 for_each_online_cpu(cpu) {
1862 struct per_cpu_pageset *pageset;
1864 pageset = zone_pcp(zone, cpu);
1866 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
1867 cpu, pageset->pcp.high,
1868 pageset->pcp.batch, pageset->pcp.count);
1872 printk("Active_anon:%lu active_file:%lu inactive_anon:%lu\n"
1873 " inactive_file:%lu"
1874 //TODO: check/adjust line lengths
1875 #ifdef CONFIG_UNEVICTABLE_LRU
1878 " dirty:%lu writeback:%lu unstable:%lu\n"
1879 " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
1880 global_page_state(NR_ACTIVE_ANON),
1881 global_page_state(NR_ACTIVE_FILE),
1882 global_page_state(NR_INACTIVE_ANON),
1883 global_page_state(NR_INACTIVE_FILE),
1884 #ifdef CONFIG_UNEVICTABLE_LRU
1885 global_page_state(NR_UNEVICTABLE),
1887 global_page_state(NR_FILE_DIRTY),
1888 global_page_state(NR_WRITEBACK),
1889 global_page_state(NR_UNSTABLE_NFS),
1890 global_page_state(NR_FREE_PAGES),
1891 global_page_state(NR_SLAB_RECLAIMABLE) +
1892 global_page_state(NR_SLAB_UNRECLAIMABLE),
1893 global_page_state(NR_FILE_MAPPED),
1894 global_page_state(NR_PAGETABLE),
1895 global_page_state(NR_BOUNCE));
1897 for_each_zone(zone) {
1900 if (!populated_zone(zone))
1909 " active_anon:%lukB"
1910 " inactive_anon:%lukB"
1911 " active_file:%lukB"
1912 " inactive_file:%lukB"
1913 #ifdef CONFIG_UNEVICTABLE_LRU
1914 " unevictable:%lukB"
1917 " pages_scanned:%lu"
1918 " all_unreclaimable? %s"
1921 K(zone_page_state(zone, NR_FREE_PAGES)),
1924 K(zone->pages_high),
1925 K(zone_page_state(zone, NR_ACTIVE_ANON)),
1926 K(zone_page_state(zone, NR_INACTIVE_ANON)),
1927 K(zone_page_state(zone, NR_ACTIVE_FILE)),
1928 K(zone_page_state(zone, NR_INACTIVE_FILE)),
1929 #ifdef CONFIG_UNEVICTABLE_LRU
1930 K(zone_page_state(zone, NR_UNEVICTABLE)),
1932 K(zone->present_pages),
1933 zone->pages_scanned,
1934 (zone_is_all_unreclaimable(zone) ? "yes" : "no")
1936 printk("lowmem_reserve[]:");
1937 for (i = 0; i < MAX_NR_ZONES; i++)
1938 printk(" %lu", zone->lowmem_reserve[i]);
1942 for_each_zone(zone) {
1943 unsigned long nr[MAX_ORDER], flags, order, total = 0;
1945 if (!populated_zone(zone))
1949 printk("%s: ", zone->name);
1951 spin_lock_irqsave(&zone->lock, flags);
1952 for (order = 0; order < MAX_ORDER; order++) {
1953 nr[order] = zone->free_area[order].nr_free;
1954 total += nr[order] << order;
1956 spin_unlock_irqrestore(&zone->lock, flags);
1957 for (order = 0; order < MAX_ORDER; order++)
1958 printk("%lu*%lukB ", nr[order], K(1UL) << order);
1959 printk("= %lukB\n", K(total));
1962 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
1964 show_swap_cache_info();
1967 static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
1969 zoneref->zone = zone;
1970 zoneref->zone_idx = zone_idx(zone);
1974 * Builds allocation fallback zone lists.
1976 * Add all populated zones of a node to the zonelist.
1978 static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
1979 int nr_zones, enum zone_type zone_type)
1983 BUG_ON(zone_type >= MAX_NR_ZONES);
1988 zone = pgdat->node_zones + zone_type;
1989 if (populated_zone(zone)) {
1990 zoneref_set_zone(zone,
1991 &zonelist->_zonerefs[nr_zones++]);
1992 check_highest_zone(zone_type);
1995 } while (zone_type);
2002 * 0 = automatic detection of better ordering.
2003 * 1 = order by ([node] distance, -zonetype)
2004 * 2 = order by (-zonetype, [node] distance)
2006 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
2007 * the same zonelist. So only NUMA can configure this param.
2009 #define ZONELIST_ORDER_DEFAULT 0
2010 #define ZONELIST_ORDER_NODE 1
2011 #define ZONELIST_ORDER_ZONE 2
2013 /* zonelist order in the kernel.
2014 * set_zonelist_order() will set this to NODE or ZONE.
2016 static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
2017 static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
2021 /* The value user specified ....changed by config */
2022 static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2023 /* string for sysctl */
2024 #define NUMA_ZONELIST_ORDER_LEN 16
2025 char numa_zonelist_order[16] = "default";
2028 * interface for configure zonelist ordering.
2029 * command line option "numa_zonelist_order"
2030 * = "[dD]efault - default, automatic configuration.
2031 * = "[nN]ode - order by node locality, then by zone within node
2032 * = "[zZ]one - order by zone, then by locality within zone
2035 static int __parse_numa_zonelist_order(char *s)
2037 if (*s == 'd' || *s == 'D') {
2038 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2039 } else if (*s == 'n' || *s == 'N') {
2040 user_zonelist_order = ZONELIST_ORDER_NODE;
2041 } else if (*s == 'z' || *s == 'Z') {
2042 user_zonelist_order = ZONELIST_ORDER_ZONE;
2045 "Ignoring invalid numa_zonelist_order value: "
2052 static __init int setup_numa_zonelist_order(char *s)
2055 return __parse_numa_zonelist_order(s);
2058 early_param("numa_zonelist_order", setup_numa_zonelist_order);
2061 * sysctl handler for numa_zonelist_order
2063 int numa_zonelist_order_handler(ctl_table *table, int write,
2064 struct file *file, void __user *buffer, size_t *length,
2067 char saved_string[NUMA_ZONELIST_ORDER_LEN];
2071 strncpy(saved_string, (char*)table->data,
2072 NUMA_ZONELIST_ORDER_LEN);
2073 ret = proc_dostring(table, write, file, buffer, length, ppos);
2077 int oldval = user_zonelist_order;
2078 if (__parse_numa_zonelist_order((char*)table->data)) {
2080 * bogus value. restore saved string
2082 strncpy((char*)table->data, saved_string,
2083 NUMA_ZONELIST_ORDER_LEN);
2084 user_zonelist_order = oldval;
2085 } else if (oldval != user_zonelist_order)
2086 build_all_zonelists();
2092 #define MAX_NODE_LOAD (num_online_nodes())
2093 static int node_load[MAX_NUMNODES];
2096 * find_next_best_node - find the next node that should appear in a given node's fallback list
2097 * @node: node whose fallback list we're appending
2098 * @used_node_mask: nodemask_t of already used nodes
2100 * We use a number of factors to determine which is the next node that should
2101 * appear on a given node's fallback list. The node should not have appeared
2102 * already in @node's fallback list, and it should be the next closest node
2103 * according to the distance array (which contains arbitrary distance values
2104 * from each node to each node in the system), and should also prefer nodes
2105 * with no CPUs, since presumably they'll have very little allocation pressure
2106 * on them otherwise.
2107 * It returns -1 if no node is found.
2109 static int find_next_best_node(int node, nodemask_t *used_node_mask)
2112 int min_val = INT_MAX;
2114 node_to_cpumask_ptr(tmp, 0);
2116 /* Use the local node if we haven't already */
2117 if (!node_isset(node, *used_node_mask)) {
2118 node_set(node, *used_node_mask);
2122 for_each_node_state(n, N_HIGH_MEMORY) {
2124 /* Don't want a node to appear more than once */
2125 if (node_isset(n, *used_node_mask))
2128 /* Use the distance array to find the distance */
2129 val = node_distance(node, n);
2131 /* Penalize nodes under us ("prefer the next node") */
2134 /* Give preference to headless and unused nodes */
2135 node_to_cpumask_ptr_next(tmp, n);
2136 if (!cpus_empty(*tmp))
2137 val += PENALTY_FOR_NODE_WITH_CPUS;
2139 /* Slight preference for less loaded node */
2140 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
2141 val += node_load[n];
2143 if (val < min_val) {
2150 node_set(best_node, *used_node_mask);
2157 * Build zonelists ordered by node and zones within node.
2158 * This results in maximum locality--normal zone overflows into local
2159 * DMA zone, if any--but risks exhausting DMA zone.
2161 static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
2164 struct zonelist *zonelist;
2166 zonelist = &pgdat->node_zonelists[0];
2167 for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
2169 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2171 zonelist->_zonerefs[j].zone = NULL;
2172 zonelist->_zonerefs[j].zone_idx = 0;
2176 * Build gfp_thisnode zonelists
2178 static void build_thisnode_zonelists(pg_data_t *pgdat)
2181 struct zonelist *zonelist;
2183 zonelist = &pgdat->node_zonelists[1];
2184 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
2185 zonelist->_zonerefs[j].zone = NULL;
2186 zonelist->_zonerefs[j].zone_idx = 0;
2190 * Build zonelists ordered by zone and nodes within zones.
2191 * This results in conserving DMA zone[s] until all Normal memory is
2192 * exhausted, but results in overflowing to remote node while memory
2193 * may still exist in local DMA zone.
2195 static int node_order[MAX_NUMNODES];
2197 static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
2200 int zone_type; /* needs to be signed */
2202 struct zonelist *zonelist;
2204 zonelist = &pgdat->node_zonelists[0];
2206 for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
2207 for (j = 0; j < nr_nodes; j++) {
2208 node = node_order[j];
2209 z = &NODE_DATA(node)->node_zones[zone_type];
2210 if (populated_zone(z)) {
2212 &zonelist->_zonerefs[pos++]);
2213 check_highest_zone(zone_type);
2217 zonelist->_zonerefs[pos].zone = NULL;
2218 zonelist->_zonerefs[pos].zone_idx = 0;
2221 static int default_zonelist_order(void)
2224 unsigned long low_kmem_size,total_size;
2228 * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem.
2229 * If they are really small and used heavily, the system can fall
2230 * into OOM very easily.
2231 * This function detect ZONE_DMA/DMA32 size and confgigures zone order.
2233 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2236 for_each_online_node(nid) {
2237 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2238 z = &NODE_DATA(nid)->node_zones[zone_type];
2239 if (populated_zone(z)) {
2240 if (zone_type < ZONE_NORMAL)
2241 low_kmem_size += z->present_pages;
2242 total_size += z->present_pages;
2246 if (!low_kmem_size || /* there are no DMA area. */
2247 low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
2248 return ZONELIST_ORDER_NODE;
2250 * look into each node's config.
2251 * If there is a node whose DMA/DMA32 memory is very big area on
2252 * local memory, NODE_ORDER may be suitable.
2254 average_size = total_size /
2255 (nodes_weight(node_states[N_HIGH_MEMORY]) + 1);
2256 for_each_online_node(nid) {
2259 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2260 z = &NODE_DATA(nid)->node_zones[zone_type];
2261 if (populated_zone(z)) {
2262 if (zone_type < ZONE_NORMAL)
2263 low_kmem_size += z->present_pages;
2264 total_size += z->present_pages;
2267 if (low_kmem_size &&
2268 total_size > average_size && /* ignore small node */
2269 low_kmem_size > total_size * 70/100)
2270 return ZONELIST_ORDER_NODE;
2272 return ZONELIST_ORDER_ZONE;
2275 static void set_zonelist_order(void)
2277 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
2278 current_zonelist_order = default_zonelist_order();
2280 current_zonelist_order = user_zonelist_order;
2283 static void build_zonelists(pg_data_t *pgdat)
2287 nodemask_t used_mask;
2288 int local_node, prev_node;
2289 struct zonelist *zonelist;
2290 int order = current_zonelist_order;
2292 /* initialize zonelists */
2293 for (i = 0; i < MAX_ZONELISTS; i++) {
2294 zonelist = pgdat->node_zonelists + i;
2295 zonelist->_zonerefs[0].zone = NULL;
2296 zonelist->_zonerefs[0].zone_idx = 0;
2299 /* NUMA-aware ordering of nodes */
2300 local_node = pgdat->node_id;
2301 load = num_online_nodes();
2302 prev_node = local_node;
2303 nodes_clear(used_mask);
2305 memset(node_load, 0, sizeof(node_load));
2306 memset(node_order, 0, sizeof(node_order));
2309 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
2310 int distance = node_distance(local_node, node);
2313 * If another node is sufficiently far away then it is better
2314 * to reclaim pages in a zone before going off node.
2316 if (distance > RECLAIM_DISTANCE)
2317 zone_reclaim_mode = 1;
2320 * We don't want to pressure a particular node.
2321 * So adding penalty to the first node in same
2322 * distance group to make it round-robin.
2324 if (distance != node_distance(local_node, prev_node))
2325 node_load[node] = load;
2329 if (order == ZONELIST_ORDER_NODE)
2330 build_zonelists_in_node_order(pgdat, node);
2332 node_order[j++] = node; /* remember order */
2335 if (order == ZONELIST_ORDER_ZONE) {
2336 /* calculate node order -- i.e., DMA last! */
2337 build_zonelists_in_zone_order(pgdat, j);
2340 build_thisnode_zonelists(pgdat);
2343 /* Construct the zonelist performance cache - see further mmzone.h */
2344 static void build_zonelist_cache(pg_data_t *pgdat)
2346 struct zonelist *zonelist;
2347 struct zonelist_cache *zlc;
2350 zonelist = &pgdat->node_zonelists[0];
2351 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
2352 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
2353 for (z = zonelist->_zonerefs; z->zone; z++)
2354 zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z);
2358 #else /* CONFIG_NUMA */
2360 static void set_zonelist_order(void)
2362 current_zonelist_order = ZONELIST_ORDER_ZONE;
2365 static void build_zonelists(pg_data_t *pgdat)
2367 int node, local_node;
2369 struct zonelist *zonelist;
2371 local_node = pgdat->node_id;
2373 zonelist = &pgdat->node_zonelists[0];
2374 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
2377 * Now we build the zonelist so that it contains the zones
2378 * of all the other nodes.
2379 * We don't want to pressure a particular node, so when
2380 * building the zones for node N, we make sure that the
2381 * zones coming right after the local ones are those from
2382 * node N+1 (modulo N)
2384 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
2385 if (!node_online(node))
2387 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2390 for (node = 0; node < local_node; node++) {
2391 if (!node_online(node))
2393 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2397 zonelist->_zonerefs[j].zone = NULL;
2398 zonelist->_zonerefs[j].zone_idx = 0;
2401 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
2402 static void build_zonelist_cache(pg_data_t *pgdat)
2404 pgdat->node_zonelists[0].zlcache_ptr = NULL;
2407 #endif /* CONFIG_NUMA */
2409 /* return values int ....just for stop_machine() */
2410 static int __build_all_zonelists(void *dummy)
2414 for_each_online_node(nid) {
2415 pg_data_t *pgdat = NODE_DATA(nid);
2417 build_zonelists(pgdat);
2418 build_zonelist_cache(pgdat);
2423 void build_all_zonelists(void)
2425 set_zonelist_order();
2427 if (system_state == SYSTEM_BOOTING) {
2428 __build_all_zonelists(NULL);
2429 mminit_verify_zonelist();
2430 cpuset_init_current_mems_allowed();
2432 /* we have to stop all cpus to guarantee there is no user
2434 stop_machine(__build_all_zonelists, NULL, NULL);
2435 /* cpuset refresh routine should be here */
2437 vm_total_pages = nr_free_pagecache_pages();
2439 * Disable grouping by mobility if the number of pages in the
2440 * system is too low to allow the mechanism to work. It would be
2441 * more accurate, but expensive to check per-zone. This check is
2442 * made on memory-hotadd so a system can start with mobility
2443 * disabled and enable it later
2445 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
2446 page_group_by_mobility_disabled = 1;
2448 page_group_by_mobility_disabled = 0;
2450 printk("Built %i zonelists in %s order, mobility grouping %s. "
2451 "Total pages: %ld\n",
2453 zonelist_order_name[current_zonelist_order],
2454 page_group_by_mobility_disabled ? "off" : "on",
2457 printk("Policy zone: %s\n", zone_names[policy_zone]);
2462 * Helper functions to size the waitqueue hash table.
2463 * Essentially these want to choose hash table sizes sufficiently
2464 * large so that collisions trying to wait on pages are rare.
2465 * But in fact, the number of active page waitqueues on typical
2466 * systems is ridiculously low, less than 200. So this is even
2467 * conservative, even though it seems large.
2469 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
2470 * waitqueues, i.e. the size of the waitq table given the number of pages.
2472 #define PAGES_PER_WAITQUEUE 256
2474 #ifndef CONFIG_MEMORY_HOTPLUG
2475 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
2477 unsigned long size = 1;
2479 pages /= PAGES_PER_WAITQUEUE;
2481 while (size < pages)
2485 * Once we have dozens or even hundreds of threads sleeping
2486 * on IO we've got bigger problems than wait queue collision.
2487 * Limit the size of the wait table to a reasonable size.
2489 size = min(size, 4096UL);
2491 return max(size, 4UL);
2495 * A zone's size might be changed by hot-add, so it is not possible to determine
2496 * a suitable size for its wait_table. So we use the maximum size now.
2498 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
2500 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
2501 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
2502 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
2504 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
2505 * or more by the traditional way. (See above). It equals:
2507 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
2508 * ia64(16K page size) : = ( 8G + 4M)byte.
2509 * powerpc (64K page size) : = (32G +16M)byte.
2511 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
2518 * This is an integer logarithm so that shifts can be used later
2519 * to extract the more random high bits from the multiplicative
2520 * hash function before the remainder is taken.
2522 static inline unsigned long wait_table_bits(unsigned long size)
2527 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
2530 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
2531 * of blocks reserved is based on zone->pages_min. The memory within the
2532 * reserve will tend to store contiguous free pages. Setting min_free_kbytes
2533 * higher will lead to a bigger reserve which will get freed as contiguous
2534 * blocks as reclaim kicks in
2536 static void setup_zone_migrate_reserve(struct zone *zone)
2538 unsigned long start_pfn, pfn, end_pfn;
2540 unsigned long reserve, block_migratetype;
2542 /* Get the start pfn, end pfn and the number of blocks to reserve */
2543 start_pfn = zone->zone_start_pfn;
2544 end_pfn = start_pfn + zone->spanned_pages;
2545 reserve = roundup(zone->pages_min, pageblock_nr_pages) >>
2548 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
2549 if (!pfn_valid(pfn))
2551 page = pfn_to_page(pfn);
2553 /* Watch out for overlapping nodes */
2554 if (page_to_nid(page) != zone_to_nid(zone))
2557 /* Blocks with reserved pages will never free, skip them. */
2558 if (PageReserved(page))
2561 block_migratetype = get_pageblock_migratetype(page);
2563 /* If this block is reserved, account for it */
2564 if (reserve > 0 && block_migratetype == MIGRATE_RESERVE) {
2569 /* Suitable for reserving if this block is movable */
2570 if (reserve > 0 && block_migratetype == MIGRATE_MOVABLE) {
2571 set_pageblock_migratetype(page, MIGRATE_RESERVE);
2572 move_freepages_block(zone, page, MIGRATE_RESERVE);
2578 * If the reserve is met and this is a previous reserved block,
2581 if (block_migratetype == MIGRATE_RESERVE) {
2582 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
2583 move_freepages_block(zone, page, MIGRATE_MOVABLE);
2589 * Initially all pages are reserved - free ones are freed
2590 * up by free_all_bootmem() once the early boot process is
2591 * done. Non-atomic initialization, single-pass.
2593 void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
2594 unsigned long start_pfn, enum memmap_context context)
2597 unsigned long end_pfn = start_pfn + size;
2601 if (highest_memmap_pfn < end_pfn - 1)
2602 highest_memmap_pfn = end_pfn - 1;
2604 z = &NODE_DATA(nid)->node_zones[zone];
2605 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
2607 * There can be holes in boot-time mem_map[]s
2608 * handed to this function. They do not
2609 * exist on hotplugged memory.
2611 if (context == MEMMAP_EARLY) {
2612 if (!early_pfn_valid(pfn))
2614 if (!early_pfn_in_nid(pfn, nid))
2617 page = pfn_to_page(pfn);
2618 set_page_links(page, zone, nid, pfn);
2619 mminit_verify_page_links(page, zone, nid, pfn);
2620 init_page_count(page);
2621 reset_page_mapcount(page);
2622 SetPageReserved(page);
2624 * Mark the block movable so that blocks are reserved for
2625 * movable at startup. This will force kernel allocations
2626 * to reserve their blocks rather than leaking throughout
2627 * the address space during boot when many long-lived
2628 * kernel allocations are made. Later some blocks near
2629 * the start are marked MIGRATE_RESERVE by
2630 * setup_zone_migrate_reserve()
2632 * bitmap is created for zone's valid pfn range. but memmap
2633 * can be created for invalid pages (for alignment)
2634 * check here not to call set_pageblock_migratetype() against
2637 if ((z->zone_start_pfn <= pfn)
2638 && (pfn < z->zone_start_pfn + z->spanned_pages)
2639 && !(pfn & (pageblock_nr_pages - 1)))
2640 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
2642 INIT_LIST_HEAD(&page->lru);
2643 #ifdef WANT_PAGE_VIRTUAL
2644 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
2645 if (!is_highmem_idx(zone))
2646 set_page_address(page, __va(pfn << PAGE_SHIFT));
2651 static void __meminit zone_init_free_lists(struct zone *zone)
2654 for_each_migratetype_order(order, t) {
2655 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
2656 zone->free_area[order].nr_free = 0;
2660 #ifndef __HAVE_ARCH_MEMMAP_INIT
2661 #define memmap_init(size, nid, zone, start_pfn) \
2662 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
2665 static int zone_batchsize(struct zone *zone)
2670 * The per-cpu-pages pools are set to around 1000th of the
2671 * size of the zone. But no more than 1/2 of a meg.
2673 * OK, so we don't know how big the cache is. So guess.
2675 batch = zone->present_pages / 1024;
2676 if (batch * PAGE_SIZE > 512 * 1024)
2677 batch = (512 * 1024) / PAGE_SIZE;
2678 batch /= 4; /* We effectively *= 4 below */
2683 * Clamp the batch to a 2^n - 1 value. Having a power
2684 * of 2 value was found to be more likely to have
2685 * suboptimal cache aliasing properties in some cases.
2687 * For example if 2 tasks are alternately allocating
2688 * batches of pages, one task can end up with a lot
2689 * of pages of one half of the possible page colors
2690 * and the other with pages of the other colors.
2692 batch = (1 << (fls(batch + batch/2)-1)) - 1;
2697 static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
2699 struct per_cpu_pages *pcp;
2701 memset(p, 0, sizeof(*p));
2705 pcp->high = 6 * batch;
2706 pcp->batch = max(1UL, 1 * batch);
2707 INIT_LIST_HEAD(&pcp->list);
2711 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
2712 * to the value high for the pageset p.
2715 static void setup_pagelist_highmark(struct per_cpu_pageset *p,
2718 struct per_cpu_pages *pcp;
2722 pcp->batch = max(1UL, high/4);
2723 if ((high/4) > (PAGE_SHIFT * 8))
2724 pcp->batch = PAGE_SHIFT * 8;
2730 * Boot pageset table. One per cpu which is going to be used for all
2731 * zones and all nodes. The parameters will be set in such a way
2732 * that an item put on a list will immediately be handed over to
2733 * the buddy list. This is safe since pageset manipulation is done
2734 * with interrupts disabled.
2736 * Some NUMA counter updates may also be caught by the boot pagesets.
2738 * The boot_pagesets must be kept even after bootup is complete for
2739 * unused processors and/or zones. They do play a role for bootstrapping
2740 * hotplugged processors.
2742 * zoneinfo_show() and maybe other functions do
2743 * not check if the processor is online before following the pageset pointer.
2744 * Other parts of the kernel may not check if the zone is available.
2746 static struct per_cpu_pageset boot_pageset[NR_CPUS];
2749 * Dynamically allocate memory for the
2750 * per cpu pageset array in struct zone.
2752 static int __cpuinit process_zones(int cpu)
2754 struct zone *zone, *dzone;
2755 int node = cpu_to_node(cpu);
2757 node_set_state(node, N_CPU); /* this node has a cpu */
2759 for_each_zone(zone) {
2761 if (!populated_zone(zone))
2764 zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset),
2766 if (!zone_pcp(zone, cpu))
2769 setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone));
2771 if (percpu_pagelist_fraction)
2772 setup_pagelist_highmark(zone_pcp(zone, cpu),
2773 (zone->present_pages / percpu_pagelist_fraction));
2778 for_each_zone(dzone) {
2779 if (!populated_zone(dzone))
2783 kfree(zone_pcp(dzone, cpu));
2784 zone_pcp(dzone, cpu) = NULL;
2789 static inline void free_zone_pagesets(int cpu)
2793 for_each_zone(zone) {
2794 struct per_cpu_pageset *pset = zone_pcp(zone, cpu);
2796 /* Free per_cpu_pageset if it is slab allocated */
2797 if (pset != &boot_pageset[cpu])
2799 zone_pcp(zone, cpu) = NULL;
2803 static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb,
2804 unsigned long action,
2807 int cpu = (long)hcpu;
2808 int ret = NOTIFY_OK;
2811 case CPU_UP_PREPARE:
2812 case CPU_UP_PREPARE_FROZEN:
2813 if (process_zones(cpu))
2816 case CPU_UP_CANCELED:
2817 case CPU_UP_CANCELED_FROZEN:
2819 case CPU_DEAD_FROZEN:
2820 free_zone_pagesets(cpu);
2828 static struct notifier_block __cpuinitdata pageset_notifier =
2829 { &pageset_cpuup_callback, NULL, 0 };
2831 void __init setup_per_cpu_pageset(void)
2835 /* Initialize per_cpu_pageset for cpu 0.
2836 * A cpuup callback will do this for every cpu
2837 * as it comes online
2839 err = process_zones(smp_processor_id());
2841 register_cpu_notifier(&pageset_notifier);
2846 static noinline __init_refok
2847 int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
2850 struct pglist_data *pgdat = zone->zone_pgdat;
2854 * The per-page waitqueue mechanism uses hashed waitqueues
2857 zone->wait_table_hash_nr_entries =
2858 wait_table_hash_nr_entries(zone_size_pages);
2859 zone->wait_table_bits =
2860 wait_table_bits(zone->wait_table_hash_nr_entries);
2861 alloc_size = zone->wait_table_hash_nr_entries
2862 * sizeof(wait_queue_head_t);
2864 if (!slab_is_available()) {
2865 zone->wait_table = (wait_queue_head_t *)
2866 alloc_bootmem_node(pgdat, alloc_size);
2869 * This case means that a zone whose size was 0 gets new memory
2870 * via memory hot-add.
2871 * But it may be the case that a new node was hot-added. In
2872 * this case vmalloc() will not be able to use this new node's
2873 * memory - this wait_table must be initialized to use this new
2874 * node itself as well.
2875 * To use this new node's memory, further consideration will be
2878 zone->wait_table = vmalloc(alloc_size);
2880 if (!zone->wait_table)
2883 for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
2884 init_waitqueue_head(zone->wait_table + i);
2889 static __meminit void zone_pcp_init(struct zone *zone)
2892 unsigned long batch = zone_batchsize(zone);
2894 for (cpu = 0; cpu < NR_CPUS; cpu++) {
2896 /* Early boot. Slab allocator not functional yet */
2897 zone_pcp(zone, cpu) = &boot_pageset[cpu];
2898 setup_pageset(&boot_pageset[cpu],0);
2900 setup_pageset(zone_pcp(zone,cpu), batch);
2903 if (zone->present_pages)
2904 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n",
2905 zone->name, zone->present_pages, batch);
2908 __meminit int init_currently_empty_zone(struct zone *zone,
2909 unsigned long zone_start_pfn,
2911 enum memmap_context context)
2913 struct pglist_data *pgdat = zone->zone_pgdat;
2915 ret = zone_wait_table_init(zone, size);
2918 pgdat->nr_zones = zone_idx(zone) + 1;
2920 zone->zone_start_pfn = zone_start_pfn;
2922 mminit_dprintk(MMINIT_TRACE, "memmap_init",
2923 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
2925 (unsigned long)zone_idx(zone),
2926 zone_start_pfn, (zone_start_pfn + size));
2928 zone_init_free_lists(zone);
2933 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2935 * Basic iterator support. Return the first range of PFNs for a node
2936 * Note: nid == MAX_NUMNODES returns first region regardless of node
2938 static int __meminit first_active_region_index_in_nid(int nid)
2942 for (i = 0; i < nr_nodemap_entries; i++)
2943 if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
2950 * Basic iterator support. Return the next active range of PFNs for a node
2951 * Note: nid == MAX_NUMNODES returns next region regardless of node
2953 static int __meminit next_active_region_index_in_nid(int index, int nid)
2955 for (index = index + 1; index < nr_nodemap_entries; index++)
2956 if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
2962 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
2964 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
2965 * Architectures may implement their own version but if add_active_range()
2966 * was used and there are no special requirements, this is a convenient
2969 int __meminit early_pfn_to_nid(unsigned long pfn)
2973 for (i = 0; i < nr_nodemap_entries; i++) {
2974 unsigned long start_pfn = early_node_map[i].start_pfn;
2975 unsigned long end_pfn = early_node_map[i].end_pfn;
2977 if (start_pfn <= pfn && pfn < end_pfn)
2978 return early_node_map[i].nid;
2983 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
2985 /* Basic iterator support to walk early_node_map[] */
2986 #define for_each_active_range_index_in_nid(i, nid) \
2987 for (i = first_active_region_index_in_nid(nid); i != -1; \
2988 i = next_active_region_index_in_nid(i, nid))
2991 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
2992 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
2993 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
2995 * If an architecture guarantees that all ranges registered with
2996 * add_active_ranges() contain no holes and may be freed, this
2997 * this function may be used instead of calling free_bootmem() manually.
2999 void __init free_bootmem_with_active_regions(int nid,
3000 unsigned long max_low_pfn)
3004 for_each_active_range_index_in_nid(i, nid) {
3005 unsigned long size_pages = 0;
3006 unsigned long end_pfn = early_node_map[i].end_pfn;
3008 if (early_node_map[i].start_pfn >= max_low_pfn)
3011 if (end_pfn > max_low_pfn)
3012 end_pfn = max_low_pfn;
3014 size_pages = end_pfn - early_node_map[i].start_pfn;
3015 free_bootmem_node(NODE_DATA(early_node_map[i].nid),
3016 PFN_PHYS(early_node_map[i].start_pfn),
3017 size_pages << PAGE_SHIFT);
3021 void __init work_with_active_regions(int nid, work_fn_t work_fn, void *data)
3026 for_each_active_range_index_in_nid(i, nid) {
3027 ret = work_fn(early_node_map[i].start_pfn,
3028 early_node_map[i].end_pfn, data);
3034 * sparse_memory_present_with_active_regions - Call memory_present for each active range
3035 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
3037 * If an architecture guarantees that all ranges registered with
3038 * add_active_ranges() contain no holes and may be freed, this
3039 * function may be used instead of calling memory_present() manually.
3041 void __init sparse_memory_present_with_active_regions(int nid)
3045 for_each_active_range_index_in_nid(i, nid)
3046 memory_present(early_node_map[i].nid,
3047 early_node_map[i].start_pfn,
3048 early_node_map[i].end_pfn);
3052 * push_node_boundaries - Push node boundaries to at least the requested boundary
3053 * @nid: The nid of the node to push the boundary for
3054 * @start_pfn: The start pfn of the node
3055 * @end_pfn: The end pfn of the node
3057 * In reserve-based hot-add, mem_map is allocated that is unused until hotadd
3058 * time. Specifically, on x86_64, SRAT will report ranges that can potentially
3059 * be hotplugged even though no physical memory exists. This function allows
3060 * an arch to push out the node boundaries so mem_map is allocated that can
3063 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
3064 void __init push_node_boundaries(unsigned int nid,
3065 unsigned long start_pfn, unsigned long end_pfn)
3067 mminit_dprintk(MMINIT_TRACE, "zoneboundary",
3068 "Entering push_node_boundaries(%u, %lu, %lu)\n",
3069 nid, start_pfn, end_pfn);
3071 /* Initialise the boundary for this node if necessary */
3072 if (node_boundary_end_pfn[nid] == 0)
3073 node_boundary_start_pfn[nid] = -1UL;
3075 /* Update the boundaries */
3076 if (node_boundary_start_pfn[nid] > start_pfn)
3077 node_boundary_start_pfn[nid] = start_pfn;
3078 if (node_boundary_end_pfn[nid] < end_pfn)
3079 node_boundary_end_pfn[nid] = end_pfn;
3082 /* If necessary, push the node boundary out for reserve hotadd */
3083 static void __meminit account_node_boundary(unsigned int nid,
3084 unsigned long *start_pfn, unsigned long *end_pfn)
3086 mminit_dprintk(MMINIT_TRACE, "zoneboundary",
3087 "Entering account_node_boundary(%u, %lu, %lu)\n",
3088 nid, *start_pfn, *end_pfn);
3090 /* Return if boundary information has not been provided */
3091 if (node_boundary_end_pfn[nid] == 0)
3094 /* Check the boundaries and update if necessary */
3095 if (node_boundary_start_pfn[nid] < *start_pfn)
3096 *start_pfn = node_boundary_start_pfn[nid];
3097 if (node_boundary_end_pfn[nid] > *end_pfn)
3098 *end_pfn = node_boundary_end_pfn[nid];
3101 void __init push_node_boundaries(unsigned int nid,
3102 unsigned long start_pfn, unsigned long end_pfn) {}
3104 static void __meminit account_node_boundary(unsigned int nid,
3105 unsigned long *start_pfn, unsigned long *end_pfn) {}
3110 * get_pfn_range_for_nid - Return the start and end page frames for a node
3111 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
3112 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
3113 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
3115 * It returns the start and end page frame of a node based on information
3116 * provided by an arch calling add_active_range(). If called for a node
3117 * with no available memory, a warning is printed and the start and end
3120 void __meminit get_pfn_range_for_nid(unsigned int nid,
3121 unsigned long *start_pfn, unsigned long *end_pfn)
3127 for_each_active_range_index_in_nid(i, nid) {
3128 *start_pfn = min(*start_pfn, early_node_map[i].start_pfn);
3129 *end_pfn = max(*end_pfn, early_node_map[i].end_pfn);
3132 if (*start_pfn == -1UL)
3135 /* Push the node boundaries out if requested */
3136 account_node_boundary(nid, start_pfn, end_pfn);
3140 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3141 * assumption is made that zones within a node are ordered in monotonic
3142 * increasing memory addresses so that the "highest" populated zone is used
3144 static void __init find_usable_zone_for_movable(void)
3147 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
3148 if (zone_index == ZONE_MOVABLE)
3151 if (arch_zone_highest_possible_pfn[zone_index] >
3152 arch_zone_lowest_possible_pfn[zone_index])
3156 VM_BUG_ON(zone_index == -1);
3157 movable_zone = zone_index;
3161 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
3162 * because it is sized independant of architecture. Unlike the other zones,
3163 * the starting point for ZONE_MOVABLE is not fixed. It may be different
3164 * in each node depending on the size of each node and how evenly kernelcore
3165 * is distributed. This helper function adjusts the zone ranges
3166 * provided by the architecture for a given node by using the end of the
3167 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3168 * zones within a node are in order of monotonic increases memory addresses
3170 static void __meminit adjust_zone_range_for_zone_movable(int nid,
3171 unsigned long zone_type,
3172 unsigned long node_start_pfn,
3173 unsigned long node_end_pfn,
3174 unsigned long *zone_start_pfn,
3175 unsigned long *zone_end_pfn)
3177 /* Only adjust if ZONE_MOVABLE is on this node */
3178 if (zone_movable_pfn[nid]) {
3179 /* Size ZONE_MOVABLE */
3180 if (zone_type == ZONE_MOVABLE) {
3181 *zone_start_pfn = zone_movable_pfn[nid];
3182 *zone_end_pfn = min(node_end_pfn,
3183 arch_zone_highest_possible_pfn[movable_zone]);
3185 /* Adjust for ZONE_MOVABLE starting within this range */
3186 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
3187 *zone_end_pfn > zone_movable_pfn[nid]) {
3188 *zone_end_pfn = zone_movable_pfn[nid];
3190 /* Check if this whole range is within ZONE_MOVABLE */
3191 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
3192 *zone_start_pfn = *zone_end_pfn;
3197 * Return the number of pages a zone spans in a node, including holes
3198 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3200 static unsigned long __meminit zone_spanned_pages_in_node(int nid,
3201 unsigned long zone_type,
3202 unsigned long *ignored)
3204 unsigned long node_start_pfn, node_end_pfn;
3205 unsigned long zone_start_pfn, zone_end_pfn;
3207 /* Get the start and end of the node and zone */
3208 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3209 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
3210 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
3211 adjust_zone_range_for_zone_movable(nid, zone_type,
3212 node_start_pfn, node_end_pfn,
3213 &zone_start_pfn, &zone_end_pfn);
3215 /* Check that this node has pages within the zone's required range */
3216 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
3219 /* Move the zone boundaries inside the node if necessary */
3220 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
3221 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
3223 /* Return the spanned pages */
3224 return zone_end_pfn - zone_start_pfn;
3228 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
3229 * then all holes in the requested range will be accounted for.
3231 static unsigned long __meminit __absent_pages_in_range(int nid,
3232 unsigned long range_start_pfn,
3233 unsigned long range_end_pfn)
3236 unsigned long prev_end_pfn = 0, hole_pages = 0;
3237 unsigned long start_pfn;
3239 /* Find the end_pfn of the first active range of pfns in the node */
3240 i = first_active_region_index_in_nid(nid);
3244 prev_end_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3246 /* Account for ranges before physical memory on this node */
3247 if (early_node_map[i].start_pfn > range_start_pfn)
3248 hole_pages = prev_end_pfn - range_start_pfn;
3250 /* Find all holes for the zone within the node */
3251 for (; i != -1; i = next_active_region_index_in_nid(i, nid)) {
3253 /* No need to continue if prev_end_pfn is outside the zone */
3254 if (prev_end_pfn >= range_end_pfn)
3257 /* Make sure the end of the zone is not within the hole */
3258 start_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3259 prev_end_pfn = max(prev_end_pfn, range_start_pfn);
3261 /* Update the hole size cound and move on */
3262 if (start_pfn > range_start_pfn) {
3263 BUG_ON(prev_end_pfn > start_pfn);
3264 hole_pages += start_pfn - prev_end_pfn;
3266 prev_end_pfn = early_node_map[i].end_pfn;
3269 /* Account for ranges past physical memory on this node */
3270 if (range_end_pfn > prev_end_pfn)
3271 hole_pages += range_end_pfn -
3272 max(range_start_pfn, prev_end_pfn);
3278 * absent_pages_in_range - Return number of page frames in holes within a range
3279 * @start_pfn: The start PFN to start searching for holes
3280 * @end_pfn: The end PFN to stop searching for holes
3282 * It returns the number of pages frames in memory holes within a range.
3284 unsigned long __init absent_pages_in_range(unsigned long start_pfn,
3285 unsigned long end_pfn)
3287 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
3290 /* Return the number of page frames in holes in a zone on a node */
3291 static unsigned long __meminit zone_absent_pages_in_node(int nid,
3292 unsigned long zone_type,
3293 unsigned long *ignored)
3295 unsigned long node_start_pfn, node_end_pfn;
3296 unsigned long zone_start_pfn, zone_end_pfn;
3298 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3299 zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type],
3301 zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type],
3304 adjust_zone_range_for_zone_movable(nid, zone_type,
3305 node_start_pfn, node_end_pfn,
3306 &zone_start_pfn, &zone_end_pfn);
3307 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
3311 static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
3312 unsigned long zone_type,
3313 unsigned long *zones_size)
3315 return zones_size[zone_type];
3318 static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
3319 unsigned long zone_type,
3320 unsigned long *zholes_size)
3325 return zholes_size[zone_type];
3330 static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
3331 unsigned long *zones_size, unsigned long *zholes_size)
3333 unsigned long realtotalpages, totalpages = 0;
3336 for (i = 0; i < MAX_NR_ZONES; i++)
3337 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
3339 pgdat->node_spanned_pages = totalpages;
3341 realtotalpages = totalpages;
3342 for (i = 0; i < MAX_NR_ZONES; i++)
3344 zone_absent_pages_in_node(pgdat->node_id, i,
3346 pgdat->node_present_pages = realtotalpages;
3347 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
3351 #ifndef CONFIG_SPARSEMEM
3353 * Calculate the size of the zone->blockflags rounded to an unsigned long
3354 * Start by making sure zonesize is a multiple of pageblock_order by rounding
3355 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
3356 * round what is now in bits to nearest long in bits, then return it in
3359 static unsigned long __init usemap_size(unsigned long zonesize)
3361 unsigned long usemapsize;
3363 usemapsize = roundup(zonesize, pageblock_nr_pages);
3364 usemapsize = usemapsize >> pageblock_order;
3365 usemapsize *= NR_PAGEBLOCK_BITS;
3366 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
3368 return usemapsize / 8;
3371 static void __init setup_usemap(struct pglist_data *pgdat,
3372 struct zone *zone, unsigned long zonesize)
3374 unsigned long usemapsize = usemap_size(zonesize);
3375 zone->pageblock_flags = NULL;
3377 zone->pageblock_flags = alloc_bootmem_node(pgdat, usemapsize);
3380 static void inline setup_usemap(struct pglist_data *pgdat,
3381 struct zone *zone, unsigned long zonesize) {}
3382 #endif /* CONFIG_SPARSEMEM */
3384 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
3386 /* Return a sensible default order for the pageblock size. */
3387 static inline int pageblock_default_order(void)
3389 if (HPAGE_SHIFT > PAGE_SHIFT)
3390 return HUGETLB_PAGE_ORDER;
3395 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
3396 static inline void __init set_pageblock_order(unsigned int order)
3398 /* Check that pageblock_nr_pages has not already been setup */
3399 if (pageblock_order)
3403 * Assume the largest contiguous order of interest is a huge page.
3404 * This value may be variable depending on boot parameters on IA64
3406 pageblock_order = order;
3408 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3411 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
3412 * and pageblock_default_order() are unused as pageblock_order is set
3413 * at compile-time. See include/linux/pageblock-flags.h for the values of
3414 * pageblock_order based on the kernel config
3416 static inline int pageblock_default_order(unsigned int order)
3420 #define set_pageblock_order(x) do {} while (0)
3422 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3425 * Set up the zone data structures:
3426 * - mark all pages reserved
3427 * - mark all memory queues empty
3428 * - clear the memory bitmaps
3430 static void __paginginit free_area_init_core(struct pglist_data *pgdat,
3431 unsigned long *zones_size, unsigned long *zholes_size)
3434 int nid = pgdat->node_id;
3435 unsigned long zone_start_pfn = pgdat->node_start_pfn;
3438 pgdat_resize_init(pgdat);
3439 pgdat->nr_zones = 0;
3440 init_waitqueue_head(&pgdat->kswapd_wait);
3441 pgdat->kswapd_max_order = 0;
3442 pgdat_page_cgroup_init(pgdat);
3444 for (j = 0; j < MAX_NR_ZONES; j++) {
3445 struct zone *zone = pgdat->node_zones + j;
3446 unsigned long size, realsize, memmap_pages;
3449 size = zone_spanned_pages_in_node(nid, j, zones_size);
3450 realsize = size - zone_absent_pages_in_node(nid, j,
3454 * Adjust realsize so that it accounts for how much memory
3455 * is used by this zone for memmap. This affects the watermark
3456 * and per-cpu initialisations
3459 PAGE_ALIGN(size * sizeof(struct page)) >> PAGE_SHIFT;
3460 if (realsize >= memmap_pages) {
3461 realsize -= memmap_pages;
3464 " %s zone: %lu pages used for memmap\n",
3465 zone_names[j], memmap_pages);
3468 " %s zone: %lu pages exceeds realsize %lu\n",
3469 zone_names[j], memmap_pages, realsize);
3471 /* Account for reserved pages */
3472 if (j == 0 && realsize > dma_reserve) {
3473 realsize -= dma_reserve;
3474 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
3475 zone_names[0], dma_reserve);
3478 if (!is_highmem_idx(j))
3479 nr_kernel_pages += realsize;
3480 nr_all_pages += realsize;
3482 zone->spanned_pages = size;
3483 zone->present_pages = realsize;
3486 zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
3488 zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100;
3490 zone->name = zone_names[j];
3491 spin_lock_init(&zone->lock);
3492 spin_lock_init(&zone->lru_lock);
3493 zone_seqlock_init(zone);
3494 zone->zone_pgdat = pgdat;
3496 zone->prev_priority = DEF_PRIORITY;
3498 zone_pcp_init(zone);
3500 INIT_LIST_HEAD(&zone->lru[l].list);
3501 zone->lru[l].nr_scan = 0;
3503 zone->recent_rotated[0] = 0;
3504 zone->recent_rotated[1] = 0;
3505 zone->recent_scanned[0] = 0;
3506 zone->recent_scanned[1] = 0;
3507 zap_zone_vm_stats(zone);
3512 set_pageblock_order(pageblock_default_order());
3513 setup_usemap(pgdat, zone, size);
3514 ret = init_currently_empty_zone(zone, zone_start_pfn,
3515 size, MEMMAP_EARLY);
3517 memmap_init(size, nid, j, zone_start_pfn);
3518 zone_start_pfn += size;
3522 static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
3524 /* Skip empty nodes */
3525 if (!pgdat->node_spanned_pages)
3528 #ifdef CONFIG_FLAT_NODE_MEM_MAP
3529 /* ia64 gets its own node_mem_map, before this, without bootmem */
3530 if (!pgdat->node_mem_map) {
3531 unsigned long size, start, end;
3535 * The zone's endpoints aren't required to be MAX_ORDER
3536 * aligned but the node_mem_map endpoints must be in order
3537 * for the buddy allocator to function correctly.
3539 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
3540 end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
3541 end = ALIGN(end, MAX_ORDER_NR_PAGES);
3542 size = (end - start) * sizeof(struct page);
3543 map = alloc_remap(pgdat->node_id, size);
3545 map = alloc_bootmem_node(pgdat, size);
3546 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
3548 #ifndef CONFIG_NEED_MULTIPLE_NODES
3550 * With no DISCONTIG, the global mem_map is just set as node 0's
3552 if (pgdat == NODE_DATA(0)) {
3553 mem_map = NODE_DATA(0)->node_mem_map;
3554 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3555 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
3556 mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
3557 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3560 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
3563 void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
3564 unsigned long node_start_pfn, unsigned long *zholes_size)
3566 pg_data_t *pgdat = NODE_DATA(nid);
3568 pgdat->node_id = nid;
3569 pgdat->node_start_pfn = node_start_pfn;
3570 calculate_node_totalpages(pgdat, zones_size, zholes_size);
3572 alloc_node_mem_map(pgdat);
3573 #ifdef CONFIG_FLAT_NODE_MEM_MAP
3574 printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
3575 nid, (unsigned long)pgdat,
3576 (unsigned long)pgdat->node_mem_map);
3579 free_area_init_core(pgdat, zones_size, zholes_size);
3582 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3584 #if MAX_NUMNODES > 1
3586 * Figure out the number of possible node ids.
3588 static void __init setup_nr_node_ids(void)
3591 unsigned int highest = 0;
3593 for_each_node_mask(node, node_possible_map)
3595 nr_node_ids = highest + 1;
3598 static inline void setup_nr_node_ids(void)
3604 * add_active_range - Register a range of PFNs backed by physical memory
3605 * @nid: The node ID the range resides on
3606 * @start_pfn: The start PFN of the available physical memory
3607 * @end_pfn: The end PFN of the available physical memory
3609 * These ranges are stored in an early_node_map[] and later used by
3610 * free_area_init_nodes() to calculate zone sizes and holes. If the
3611 * range spans a memory hole, it is up to the architecture to ensure
3612 * the memory is not freed by the bootmem allocator. If possible
3613 * the range being registered will be merged with existing ranges.
3615 void __init add_active_range(unsigned int nid, unsigned long start_pfn,
3616 unsigned long end_pfn)
3620 mminit_dprintk(MMINIT_TRACE, "memory_register",
3621 "Entering add_active_range(%d, %#lx, %#lx) "
3622 "%d entries of %d used\n",
3623 nid, start_pfn, end_pfn,
3624 nr_nodemap_entries, MAX_ACTIVE_REGIONS);
3626 mminit_validate_memmodel_limits(&start_pfn, &end_pfn);
3628 /* Merge with existing active regions if possible */
3629 for (i = 0; i < nr_nodemap_entries; i++) {
3630 if (early_node_map[i].nid != nid)
3633 /* Skip if an existing region covers this new one */
3634 if (start_pfn >= early_node_map[i].start_pfn &&
3635 end_pfn <= early_node_map[i].end_pfn)
3638 /* Merge forward if suitable */
3639 if (start_pfn <= early_node_map[i].end_pfn &&
3640 end_pfn > early_node_map[i].end_pfn) {
3641 early_node_map[i].end_pfn = end_pfn;
3645 /* Merge backward if suitable */
3646 if (start_pfn < early_node_map[i].end_pfn &&
3647 end_pfn >= early_node_map[i].start_pfn) {
3648 early_node_map[i].start_pfn = start_pfn;
3653 /* Check that early_node_map is large enough */
3654 if (i >= MAX_ACTIVE_REGIONS) {
3655 printk(KERN_CRIT "More than %d memory regions, truncating\n",
3656 MAX_ACTIVE_REGIONS);
3660 early_node_map[i].nid = nid;
3661 early_node_map[i].start_pfn = start_pfn;
3662 early_node_map[i].end_pfn = end_pfn;
3663 nr_nodemap_entries = i + 1;
3667 * remove_active_range - Shrink an existing registered range of PFNs
3668 * @nid: The node id the range is on that should be shrunk
3669 * @start_pfn: The new PFN of the range
3670 * @end_pfn: The new PFN of the range
3672 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
3673 * The map is kept near the end physical page range that has already been
3674 * registered. This function allows an arch to shrink an existing registered
3677 void __init remove_active_range(unsigned int nid, unsigned long start_pfn,
3678 unsigned long end_pfn)
3683 printk(KERN_DEBUG "remove_active_range (%d, %lu, %lu)\n",
3684 nid, start_pfn, end_pfn);
3686 /* Find the old active region end and shrink */
3687 for_each_active_range_index_in_nid(i, nid) {
3688 if (early_node_map[i].start_pfn >= start_pfn &&
3689 early_node_map[i].end_pfn <= end_pfn) {
3691 early_node_map[i].start_pfn = 0;
3692 early_node_map[i].end_pfn = 0;
3696 if (early_node_map[i].start_pfn < start_pfn &&
3697 early_node_map[i].end_pfn > start_pfn) {
3698 unsigned long temp_end_pfn = early_node_map[i].end_pfn;
3699 early_node_map[i].end_pfn = start_pfn;
3700 if (temp_end_pfn > end_pfn)
3701 add_active_range(nid, end_pfn, temp_end_pfn);
3704 if (early_node_map[i].start_pfn >= start_pfn &&
3705 early_node_map[i].end_pfn > end_pfn &&
3706 early_node_map[i].start_pfn < end_pfn) {
3707 early_node_map[i].start_pfn = end_pfn;
3715 /* remove the blank ones */
3716 for (i = nr_nodemap_entries - 1; i > 0; i--) {
3717 if (early_node_map[i].nid != nid)
3719 if (early_node_map[i].end_pfn)
3721 /* we found it, get rid of it */
3722 for (j = i; j < nr_nodemap_entries - 1; j++)
3723 memcpy(&early_node_map[j], &early_node_map[j+1],
3724 sizeof(early_node_map[j]));
3725 j = nr_nodemap_entries - 1;
3726 memset(&early_node_map[j], 0, sizeof(early_node_map[j]));
3727 nr_nodemap_entries--;
3732 * remove_all_active_ranges - Remove all currently registered regions
3734 * During discovery, it may be found that a table like SRAT is invalid
3735 * and an alternative discovery method must be used. This function removes
3736 * all currently registered regions.
3738 void __init remove_all_active_ranges(void)
3740 memset(early_node_map, 0, sizeof(early_node_map));
3741 nr_nodemap_entries = 0;
3742 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
3743 memset(node_boundary_start_pfn, 0, sizeof(node_boundary_start_pfn));
3744 memset(node_boundary_end_pfn, 0, sizeof(node_boundary_end_pfn));
3745 #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
3748 /* Compare two active node_active_regions */
3749 static int __init cmp_node_active_region(const void *a, const void *b)
3751 struct node_active_region *arange = (struct node_active_region *)a;
3752 struct node_active_region *brange = (struct node_active_region *)b;
3754 /* Done this way to avoid overflows */
3755 if (arange->start_pfn > brange->start_pfn)
3757 if (arange->start_pfn < brange->start_pfn)
3763 /* sort the node_map by start_pfn */
3764 static void __init sort_node_map(void)
3766 sort(early_node_map, (size_t)nr_nodemap_entries,
3767 sizeof(struct node_active_region),
3768 cmp_node_active_region, NULL);
3771 /* Find the lowest pfn for a node */
3772 static unsigned long __init find_min_pfn_for_node(int nid)
3775 unsigned long min_pfn = ULONG_MAX;
3777 /* Assuming a sorted map, the first range found has the starting pfn */
3778 for_each_active_range_index_in_nid(i, nid)
3779 min_pfn = min(min_pfn, early_node_map[i].start_pfn);
3781 if (min_pfn == ULONG_MAX) {
3783 "Could not find start_pfn for node %d\n", nid);
3791 * find_min_pfn_with_active_regions - Find the minimum PFN registered
3793 * It returns the minimum PFN based on information provided via
3794 * add_active_range().
3796 unsigned long __init find_min_pfn_with_active_regions(void)
3798 return find_min_pfn_for_node(MAX_NUMNODES);
3802 * early_calculate_totalpages()
3803 * Sum pages in active regions for movable zone.
3804 * Populate N_HIGH_MEMORY for calculating usable_nodes.
3806 static unsigned long __init early_calculate_totalpages(void)
3809 unsigned long totalpages = 0;
3811 for (i = 0; i < nr_nodemap_entries; i++) {
3812 unsigned long pages = early_node_map[i].end_pfn -
3813 early_node_map[i].start_pfn;
3814 totalpages += pages;
3816 node_set_state(early_node_map[i].nid, N_HIGH_MEMORY);
3822 * Find the PFN the Movable zone begins in each node. Kernel memory
3823 * is spread evenly between nodes as long as the nodes have enough
3824 * memory. When they don't, some nodes will have more kernelcore than
3827 static void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn)
3830 unsigned long usable_startpfn;
3831 unsigned long kernelcore_node, kernelcore_remaining;
3832 unsigned long totalpages = early_calculate_totalpages();
3833 int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]);
3836 * If movablecore was specified, calculate what size of
3837 * kernelcore that corresponds so that memory usable for
3838 * any allocation type is evenly spread. If both kernelcore
3839 * and movablecore are specified, then the value of kernelcore
3840 * will be used for required_kernelcore if it's greater than
3841 * what movablecore would have allowed.
3843 if (required_movablecore) {
3844 unsigned long corepages;
3847 * Round-up so that ZONE_MOVABLE is at least as large as what
3848 * was requested by the user
3850 required_movablecore =
3851 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
3852 corepages = totalpages - required_movablecore;
3854 required_kernelcore = max(required_kernelcore, corepages);
3857 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
3858 if (!required_kernelcore)
3861 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
3862 find_usable_zone_for_movable();
3863 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
3866 /* Spread kernelcore memory as evenly as possible throughout nodes */
3867 kernelcore_node = required_kernelcore / usable_nodes;
3868 for_each_node_state(nid, N_HIGH_MEMORY) {
3870 * Recalculate kernelcore_node if the division per node
3871 * now exceeds what is necessary to satisfy the requested
3872 * amount of memory for the kernel
3874 if (required_kernelcore < kernelcore_node)
3875 kernelcore_node = required_kernelcore / usable_nodes;
3878 * As the map is walked, we track how much memory is usable
3879 * by the kernel using kernelcore_remaining. When it is
3880 * 0, the rest of the node is usable by ZONE_MOVABLE
3882 kernelcore_remaining = kernelcore_node;
3884 /* Go through each range of PFNs within this node */
3885 for_each_active_range_index_in_nid(i, nid) {
3886 unsigned long start_pfn, end_pfn;
3887 unsigned long size_pages;
3889 start_pfn = max(early_node_map[i].start_pfn,
3890 zone_movable_pfn[nid]);
3891 end_pfn = early_node_map[i].end_pfn;
3892 if (start_pfn >= end_pfn)
3895 /* Account for what is only usable for kernelcore */
3896 if (start_pfn < usable_startpfn) {
3897 unsigned long kernel_pages;
3898 kernel_pages = min(end_pfn, usable_startpfn)
3901 kernelcore_remaining -= min(kernel_pages,
3902 kernelcore_remaining);
3903 required_kernelcore -= min(kernel_pages,
3904 required_kernelcore);
3906 /* Continue if range is now fully accounted */
3907 if (end_pfn <= usable_startpfn) {
3910 * Push zone_movable_pfn to the end so
3911 * that if we have to rebalance
3912 * kernelcore across nodes, we will
3913 * not double account here
3915 zone_movable_pfn[nid] = end_pfn;
3918 start_pfn = usable_startpfn;
3922 * The usable PFN range for ZONE_MOVABLE is from
3923 * start_pfn->end_pfn. Calculate size_pages as the
3924 * number of pages used as kernelcore
3926 size_pages = end_pfn - start_pfn;
3927 if (size_pages > kernelcore_remaining)
3928 size_pages = kernelcore_remaining;
3929 zone_movable_pfn[nid] = start_pfn + size_pages;
3932 * Some kernelcore has been met, update counts and
3933 * break if the kernelcore for this node has been
3936 required_kernelcore -= min(required_kernelcore,
3938 kernelcore_remaining -= size_pages;
3939 if (!kernelcore_remaining)
3945 * If there is still required_kernelcore, we do another pass with one
3946 * less node in the count. This will push zone_movable_pfn[nid] further
3947 * along on the nodes that still have memory until kernelcore is
3951 if (usable_nodes && required_kernelcore > usable_nodes)
3954 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
3955 for (nid = 0; nid < MAX_NUMNODES; nid++)
3956 zone_movable_pfn[nid] =
3957 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
3960 /* Any regular memory on that node ? */
3961 static void check_for_regular_memory(pg_data_t *pgdat)
3963 #ifdef CONFIG_HIGHMEM
3964 enum zone_type zone_type;
3966 for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) {
3967 struct zone *zone = &pgdat->node_zones[zone_type];
3968 if (zone->present_pages)
3969 node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY);
3975 * free_area_init_nodes - Initialise all pg_data_t and zone data
3976 * @max_zone_pfn: an array of max PFNs for each zone
3978 * This will call free_area_init_node() for each active node in the system.
3979 * Using the page ranges provided by add_active_range(), the size of each
3980 * zone in each node and their holes is calculated. If the maximum PFN
3981 * between two adjacent zones match, it is assumed that the zone is empty.
3982 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
3983 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
3984 * starts where the previous one ended. For example, ZONE_DMA32 starts
3985 * at arch_max_dma_pfn.
3987 void __init free_area_init_nodes(unsigned long *max_zone_pfn)
3992 /* Sort early_node_map as initialisation assumes it is sorted */
3995 /* Record where the zone boundaries are */
3996 memset(arch_zone_lowest_possible_pfn, 0,
3997 sizeof(arch_zone_lowest_possible_pfn));
3998 memset(arch_zone_highest_possible_pfn, 0,
3999 sizeof(arch_zone_highest_possible_pfn));
4000 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
4001 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
4002 for (i = 1; i < MAX_NR_ZONES; i++) {
4003 if (i == ZONE_MOVABLE)
4005 arch_zone_lowest_possible_pfn[i] =
4006 arch_zone_highest_possible_pfn[i-1];
4007 arch_zone_highest_possible_pfn[i] =
4008 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
4010 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
4011 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
4013 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
4014 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
4015 find_zone_movable_pfns_for_nodes(zone_movable_pfn);
4017 /* Print out the zone ranges */
4018 printk("Zone PFN ranges:\n");
4019 for (i = 0; i < MAX_NR_ZONES; i++) {
4020 if (i == ZONE_MOVABLE)
4022 printk(" %-8s %0#10lx -> %0#10lx\n",
4024 arch_zone_lowest_possible_pfn[i],
4025 arch_zone_highest_possible_pfn[i]);
4028 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
4029 printk("Movable zone start PFN for each node\n");
4030 for (i = 0; i < MAX_NUMNODES; i++) {
4031 if (zone_movable_pfn[i])
4032 printk(" Node %d: %lu\n", i, zone_movable_pfn[i]);
4035 /* Print out the early_node_map[] */
4036 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries);
4037 for (i = 0; i < nr_nodemap_entries; i++)
4038 printk(" %3d: %0#10lx -> %0#10lx\n", early_node_map[i].nid,
4039 early_node_map[i].start_pfn,
4040 early_node_map[i].end_pfn);
4042 /* Initialise every node */
4043 mminit_verify_pageflags_layout();
4044 setup_nr_node_ids();
4045 for_each_online_node(nid) {
4046 pg_data_t *pgdat = NODE_DATA(nid);
4047 free_area_init_node(nid, NULL,
4048 find_min_pfn_for_node(nid), NULL);
4050 /* Any memory on that node */
4051 if (pgdat->node_present_pages)
4052 node_set_state(nid, N_HIGH_MEMORY);
4053 check_for_regular_memory(pgdat);
4057 static int __init cmdline_parse_core(char *p, unsigned long *core)
4059 unsigned long long coremem;
4063 coremem = memparse(p, &p);
4064 *core = coremem >> PAGE_SHIFT;
4066 /* Paranoid check that UL is enough for the coremem value */
4067 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
4073 * kernelcore=size sets the amount of memory for use for allocations that
4074 * cannot be reclaimed or migrated.
4076 static int __init cmdline_parse_kernelcore(char *p)
4078 return cmdline_parse_core(p, &required_kernelcore);
4082 * movablecore=size sets the amount of memory for use for allocations that
4083 * can be reclaimed or migrated.
4085 static int __init cmdline_parse_movablecore(char *p)
4087 return cmdline_parse_core(p, &required_movablecore);
4090 early_param("kernelcore", cmdline_parse_kernelcore);
4091 early_param("movablecore", cmdline_parse_movablecore);
4093 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4096 * set_dma_reserve - set the specified number of pages reserved in the first zone
4097 * @new_dma_reserve: The number of pages to mark reserved
4099 * The per-cpu batchsize and zone watermarks are determined by present_pages.
4100 * In the DMA zone, a significant percentage may be consumed by kernel image
4101 * and other unfreeable allocations which can skew the watermarks badly. This
4102 * function may optionally be used to account for unfreeable pages in the
4103 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
4104 * smaller per-cpu batchsize.
4106 void __init set_dma_reserve(unsigned long new_dma_reserve)
4108 dma_reserve = new_dma_reserve;
4111 #ifndef CONFIG_NEED_MULTIPLE_NODES
4112 struct pglist_data __refdata contig_page_data = { .bdata = &bootmem_node_data[0] };
4113 EXPORT_SYMBOL(contig_page_data);
4116 void __init free_area_init(unsigned long *zones_size)
4118 free_area_init_node(0, zones_size,
4119 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
4122 static int page_alloc_cpu_notify(struct notifier_block *self,
4123 unsigned long action, void *hcpu)
4125 int cpu = (unsigned long)hcpu;
4127 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
4131 * Spill the event counters of the dead processor
4132 * into the current processors event counters.
4133 * This artificially elevates the count of the current
4136 vm_events_fold_cpu(cpu);
4139 * Zero the differential counters of the dead processor
4140 * so that the vm statistics are consistent.
4142 * This is only okay since the processor is dead and cannot
4143 * race with what we are doing.
4145 refresh_cpu_vm_stats(cpu);
4150 void __init page_alloc_init(void)
4152 hotcpu_notifier(page_alloc_cpu_notify, 0);
4156 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
4157 * or min_free_kbytes changes.
4159 static void calculate_totalreserve_pages(void)
4161 struct pglist_data *pgdat;
4162 unsigned long reserve_pages = 0;
4163 enum zone_type i, j;
4165 for_each_online_pgdat(pgdat) {
4166 for (i = 0; i < MAX_NR_ZONES; i++) {
4167 struct zone *zone = pgdat->node_zones + i;
4168 unsigned long max = 0;
4170 /* Find valid and maximum lowmem_reserve in the zone */
4171 for (j = i; j < MAX_NR_ZONES; j++) {
4172 if (zone->lowmem_reserve[j] > max)
4173 max = zone->lowmem_reserve[j];
4176 /* we treat pages_high as reserved pages. */
4177 max += zone->pages_high;
4179 if (max > zone->present_pages)
4180 max = zone->present_pages;
4181 reserve_pages += max;
4184 totalreserve_pages = reserve_pages;
4188 * setup_per_zone_lowmem_reserve - called whenever
4189 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
4190 * has a correct pages reserved value, so an adequate number of
4191 * pages are left in the zone after a successful __alloc_pages().
4193 static void setup_per_zone_lowmem_reserve(void)
4195 struct pglist_data *pgdat;
4196 enum zone_type j, idx;
4198 for_each_online_pgdat(pgdat) {
4199 for (j = 0; j < MAX_NR_ZONES; j++) {
4200 struct zone *zone = pgdat->node_zones + j;
4201 unsigned long present_pages = zone->present_pages;
4203 zone->lowmem_reserve[j] = 0;
4207 struct zone *lower_zone;
4211 if (sysctl_lowmem_reserve_ratio[idx] < 1)
4212 sysctl_lowmem_reserve_ratio[idx] = 1;
4214 lower_zone = pgdat->node_zones + idx;
4215 lower_zone->lowmem_reserve[j] = present_pages /
4216 sysctl_lowmem_reserve_ratio[idx];
4217 present_pages += lower_zone->present_pages;
4222 /* update totalreserve_pages */
4223 calculate_totalreserve_pages();
4227 * setup_per_zone_pages_min - called when min_free_kbytes changes.
4229 * Ensures that the pages_{min,low,high} values for each zone are set correctly
4230 * with respect to min_free_kbytes.
4232 void setup_per_zone_pages_min(void)
4234 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
4235 unsigned long lowmem_pages = 0;
4237 unsigned long flags;
4239 /* Calculate total number of !ZONE_HIGHMEM pages */
4240 for_each_zone(zone) {
4241 if (!is_highmem(zone))
4242 lowmem_pages += zone->present_pages;
4245 for_each_zone(zone) {
4248 spin_lock_irqsave(&zone->lock, flags);
4249 tmp = (u64)pages_min * zone->present_pages;
4250 do_div(tmp, lowmem_pages);
4251 if (is_highmem(zone)) {
4253 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
4254 * need highmem pages, so cap pages_min to a small
4257 * The (pages_high-pages_low) and (pages_low-pages_min)
4258 * deltas controls asynch page reclaim, and so should
4259 * not be capped for highmem.
4263 min_pages = zone->present_pages / 1024;
4264 if (min_pages < SWAP_CLUSTER_MAX)
4265 min_pages = SWAP_CLUSTER_MAX;
4266 if (min_pages > 128)
4268 zone->pages_min = min_pages;
4271 * If it's a lowmem zone, reserve a number of pages
4272 * proportionate to the zone's size.
4274 zone->pages_min = tmp;
4277 zone->pages_low = zone->pages_min + (tmp >> 2);
4278 zone->pages_high = zone->pages_min + (tmp >> 1);
4279 setup_zone_migrate_reserve(zone);
4280 spin_unlock_irqrestore(&zone->lock, flags);
4283 /* update totalreserve_pages */
4284 calculate_totalreserve_pages();
4288 * setup_per_zone_inactive_ratio - called when min_free_kbytes changes.
4290 * The inactive anon list should be small enough that the VM never has to
4291 * do too much work, but large enough that each inactive page has a chance
4292 * to be referenced again before it is swapped out.
4294 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
4295 * INACTIVE_ANON pages on this zone's LRU, maintained by the
4296 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
4297 * the anonymous pages are kept on the inactive list.
4300 * memory ratio inactive anon
4301 * -------------------------------------
4310 static void setup_per_zone_inactive_ratio(void)
4314 for_each_zone(zone) {
4315 unsigned int gb, ratio;
4317 /* Zone size in gigabytes */
4318 gb = zone->present_pages >> (30 - PAGE_SHIFT);
4319 ratio = int_sqrt(10 * gb);
4323 zone->inactive_ratio = ratio;
4328 * Initialise min_free_kbytes.
4330 * For small machines we want it small (128k min). For large machines
4331 * we want it large (64MB max). But it is not linear, because network
4332 * bandwidth does not increase linearly with machine size. We use
4334 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
4335 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
4351 static int __init init_per_zone_pages_min(void)
4353 unsigned long lowmem_kbytes;
4355 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
4357 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
4358 if (min_free_kbytes < 128)
4359 min_free_kbytes = 128;
4360 if (min_free_kbytes > 65536)
4361 min_free_kbytes = 65536;
4362 setup_per_zone_pages_min();
4363 setup_per_zone_lowmem_reserve();
4364 setup_per_zone_inactive_ratio();
4367 module_init(init_per_zone_pages_min)
4370 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
4371 * that we can call two helper functions whenever min_free_kbytes
4374 int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
4375 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4377 proc_dointvec(table, write, file, buffer, length, ppos);
4379 setup_per_zone_pages_min();
4384 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
4385 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4390 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4395 zone->min_unmapped_pages = (zone->present_pages *
4396 sysctl_min_unmapped_ratio) / 100;
4400 int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
4401 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4406 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4411 zone->min_slab_pages = (zone->present_pages *
4412 sysctl_min_slab_ratio) / 100;
4418 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
4419 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
4420 * whenever sysctl_lowmem_reserve_ratio changes.
4422 * The reserve ratio obviously has absolutely no relation with the
4423 * pages_min watermarks. The lowmem reserve ratio can only make sense
4424 * if in function of the boot time zone sizes.
4426 int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
4427 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4429 proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4430 setup_per_zone_lowmem_reserve();
4435 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
4436 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
4437 * can have before it gets flushed back to buddy allocator.
4440 int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
4441 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4447 ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4448 if (!write || (ret == -EINVAL))
4450 for_each_zone(zone) {
4451 for_each_online_cpu(cpu) {
4453 high = zone->present_pages / percpu_pagelist_fraction;
4454 setup_pagelist_highmark(zone_pcp(zone, cpu), high);
4460 int hashdist = HASHDIST_DEFAULT;
4463 static int __init set_hashdist(char *str)
4467 hashdist = simple_strtoul(str, &str, 0);
4470 __setup("hashdist=", set_hashdist);
4474 * allocate a large system hash table from bootmem
4475 * - it is assumed that the hash table must contain an exact power-of-2
4476 * quantity of entries
4477 * - limit is the number of hash buckets, not the total allocation size
4479 void *__init alloc_large_system_hash(const char *tablename,
4480 unsigned long bucketsize,
4481 unsigned long numentries,
4484 unsigned int *_hash_shift,
4485 unsigned int *_hash_mask,
4486 unsigned long limit)
4488 unsigned long long max = limit;
4489 unsigned long log2qty, size;
4492 /* allow the kernel cmdline to have a say */
4494 /* round applicable memory size up to nearest megabyte */
4495 numentries = nr_kernel_pages;
4496 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
4497 numentries >>= 20 - PAGE_SHIFT;
4498 numentries <<= 20 - PAGE_SHIFT;
4500 /* limit to 1 bucket per 2^scale bytes of low memory */
4501 if (scale > PAGE_SHIFT)
4502 numentries >>= (scale - PAGE_SHIFT);
4504 numentries <<= (PAGE_SHIFT - scale);
4506 /* Make sure we've got at least a 0-order allocation.. */
4507 if (unlikely((numentries * bucketsize) < PAGE_SIZE))
4508 numentries = PAGE_SIZE / bucketsize;
4510 numentries = roundup_pow_of_two(numentries);
4512 /* limit allocation size to 1/16 total memory by default */
4514 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
4515 do_div(max, bucketsize);
4518 if (numentries > max)
4521 log2qty = ilog2(numentries);
4524 size = bucketsize << log2qty;
4525 if (flags & HASH_EARLY)
4526 table = alloc_bootmem_nopanic(size);
4528 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
4530 unsigned long order = get_order(size);
4531 table = (void*) __get_free_pages(GFP_ATOMIC, order);
4533 * If bucketsize is not a power-of-two, we may free
4534 * some pages at the end of hash table.
4537 unsigned long alloc_end = (unsigned long)table +
4538 (PAGE_SIZE << order);
4539 unsigned long used = (unsigned long)table +
4541 split_page(virt_to_page(table), order);
4542 while (used < alloc_end) {
4548 } while (!table && size > PAGE_SIZE && --log2qty);
4551 panic("Failed to allocate %s hash table\n", tablename);
4553 printk(KERN_INFO "%s hash table entries: %d (order: %d, %lu bytes)\n",
4556 ilog2(size) - PAGE_SHIFT,
4560 *_hash_shift = log2qty;
4562 *_hash_mask = (1 << log2qty) - 1;
4567 #ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE
4568 struct page *pfn_to_page(unsigned long pfn)
4570 return __pfn_to_page(pfn);
4572 unsigned long page_to_pfn(struct page *page)
4574 return __page_to_pfn(page);
4576 EXPORT_SYMBOL(pfn_to_page);
4577 EXPORT_SYMBOL(page_to_pfn);
4578 #endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */
4580 /* Return a pointer to the bitmap storing bits affecting a block of pages */
4581 static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
4584 #ifdef CONFIG_SPARSEMEM
4585 return __pfn_to_section(pfn)->pageblock_flags;
4587 return zone->pageblock_flags;
4588 #endif /* CONFIG_SPARSEMEM */
4591 static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
4593 #ifdef CONFIG_SPARSEMEM
4594 pfn &= (PAGES_PER_SECTION-1);
4595 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
4597 pfn = pfn - zone->zone_start_pfn;
4598 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
4599 #endif /* CONFIG_SPARSEMEM */
4603 * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
4604 * @page: The page within the block of interest
4605 * @start_bitidx: The first bit of interest to retrieve
4606 * @end_bitidx: The last bit of interest
4607 * returns pageblock_bits flags
4609 unsigned long get_pageblock_flags_group(struct page *page,
4610 int start_bitidx, int end_bitidx)
4613 unsigned long *bitmap;
4614 unsigned long pfn, bitidx;
4615 unsigned long flags = 0;
4616 unsigned long value = 1;
4618 zone = page_zone(page);
4619 pfn = page_to_pfn(page);
4620 bitmap = get_pageblock_bitmap(zone, pfn);
4621 bitidx = pfn_to_bitidx(zone, pfn);
4623 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
4624 if (test_bit(bitidx + start_bitidx, bitmap))
4631 * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
4632 * @page: The page within the block of interest
4633 * @start_bitidx: The first bit of interest
4634 * @end_bitidx: The last bit of interest
4635 * @flags: The flags to set
4637 void set_pageblock_flags_group(struct page *page, unsigned long flags,
4638 int start_bitidx, int end_bitidx)
4641 unsigned long *bitmap;
4642 unsigned long pfn, bitidx;
4643 unsigned long value = 1;
4645 zone = page_zone(page);
4646 pfn = page_to_pfn(page);
4647 bitmap = get_pageblock_bitmap(zone, pfn);
4648 bitidx = pfn_to_bitidx(zone, pfn);
4649 VM_BUG_ON(pfn < zone->zone_start_pfn);
4650 VM_BUG_ON(pfn >= zone->zone_start_pfn + zone->spanned_pages);
4652 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
4654 __set_bit(bitidx + start_bitidx, bitmap);
4656 __clear_bit(bitidx + start_bitidx, bitmap);
4660 * This is designed as sub function...plz see page_isolation.c also.
4661 * set/clear page block's type to be ISOLATE.
4662 * page allocater never alloc memory from ISOLATE block.
4665 int set_migratetype_isolate(struct page *page)
4668 unsigned long flags;
4671 zone = page_zone(page);
4672 spin_lock_irqsave(&zone->lock, flags);
4674 * In future, more migrate types will be able to be isolation target.
4676 if (get_pageblock_migratetype(page) != MIGRATE_MOVABLE)
4678 set_pageblock_migratetype(page, MIGRATE_ISOLATE);
4679 move_freepages_block(zone, page, MIGRATE_ISOLATE);
4682 spin_unlock_irqrestore(&zone->lock, flags);
4688 void unset_migratetype_isolate(struct page *page)
4691 unsigned long flags;
4692 zone = page_zone(page);
4693 spin_lock_irqsave(&zone->lock, flags);
4694 if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE)
4696 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
4697 move_freepages_block(zone, page, MIGRATE_MOVABLE);
4699 spin_unlock_irqrestore(&zone->lock, flags);
4702 #ifdef CONFIG_MEMORY_HOTREMOVE
4704 * All pages in the range must be isolated before calling this.
4707 __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
4713 unsigned long flags;
4714 /* find the first valid pfn */
4715 for (pfn = start_pfn; pfn < end_pfn; pfn++)
4720 zone = page_zone(pfn_to_page(pfn));
4721 spin_lock_irqsave(&zone->lock, flags);
4723 while (pfn < end_pfn) {
4724 if (!pfn_valid(pfn)) {
4728 page = pfn_to_page(pfn);
4729 BUG_ON(page_count(page));
4730 BUG_ON(!PageBuddy(page));
4731 order = page_order(page);
4732 #ifdef CONFIG_DEBUG_VM
4733 printk(KERN_INFO "remove from free list %lx %d %lx\n",
4734 pfn, 1 << order, end_pfn);
4736 list_del(&page->lru);
4737 rmv_page_order(page);
4738 zone->free_area[order].nr_free--;
4739 __mod_zone_page_state(zone, NR_FREE_PAGES,
4741 for (i = 0; i < (1 << order); i++)
4742 SetPageReserved((page+i));
4743 pfn += (1 << order);
4745 spin_unlock_irqrestore(&zone->lock, flags);