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/bootmem.h>
23 #include <linux/compiler.h>
24 #include <linux/kernel.h>
25 #include <linux/module.h>
26 #include <linux/suspend.h>
27 #include <linux/pagevec.h>
28 #include <linux/blkdev.h>
29 #include <linux/slab.h>
30 #include <linux/notifier.h>
31 #include <linux/topology.h>
32 #include <linux/sysctl.h>
33 #include <linux/cpu.h>
34 #include <linux/cpuset.h>
35 #include <linux/memory_hotplug.h>
36 #include <linux/nodemask.h>
37 #include <linux/vmalloc.h>
38 #include <linux/mempolicy.h>
39 #include <linux/stop_machine.h>
40 #include <linux/sort.h>
41 #include <linux/pfn.h>
42 #include <linux/backing-dev.h>
43 #include <linux/fault-inject.h>
45 #include <asm/tlbflush.h>
46 #include <asm/div64.h>
50 * Array of node states.
52 nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
53 [N_POSSIBLE] = NODE_MASK_ALL,
54 [N_ONLINE] = { { [0] = 1UL } },
56 [N_NORMAL_MEMORY] = { { [0] = 1UL } },
58 [N_HIGH_MEMORY] = { { [0] = 1UL } },
60 [N_CPU] = { { [0] = 1UL } },
63 EXPORT_SYMBOL(node_states);
65 unsigned long totalram_pages __read_mostly;
66 unsigned long totalreserve_pages __read_mostly;
68 int percpu_pagelist_fraction;
70 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
71 int pageblock_order __read_mostly;
74 static void __free_pages_ok(struct page *page, unsigned int order);
77 * results with 256, 32 in the lowmem_reserve sysctl:
78 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
79 * 1G machine -> (16M dma, 784M normal, 224M high)
80 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
81 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
82 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
84 * TBD: should special case ZONE_DMA32 machines here - in those we normally
85 * don't need any ZONE_NORMAL reservation
87 int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
88 #ifdef CONFIG_ZONE_DMA
91 #ifdef CONFIG_ZONE_DMA32
100 EXPORT_SYMBOL(totalram_pages);
102 static char * const zone_names[MAX_NR_ZONES] = {
103 #ifdef CONFIG_ZONE_DMA
106 #ifdef CONFIG_ZONE_DMA32
110 #ifdef CONFIG_HIGHMEM
116 int min_free_kbytes = 1024;
118 unsigned long __meminitdata nr_kernel_pages;
119 unsigned long __meminitdata nr_all_pages;
120 static unsigned long __meminitdata dma_reserve;
122 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
124 * MAX_ACTIVE_REGIONS determines the maxmimum number of distinct
125 * ranges of memory (RAM) that may be registered with add_active_range().
126 * Ranges passed to add_active_range() will be merged if possible
127 * so the number of times add_active_range() can be called is
128 * related to the number of nodes and the number of holes
130 #ifdef CONFIG_MAX_ACTIVE_REGIONS
131 /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
132 #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
134 #if MAX_NUMNODES >= 32
135 /* If there can be many nodes, allow up to 50 holes per node */
136 #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
138 /* By default, allow up to 256 distinct regions */
139 #define MAX_ACTIVE_REGIONS 256
143 static struct node_active_region __meminitdata early_node_map[MAX_ACTIVE_REGIONS];
144 static int __meminitdata nr_nodemap_entries;
145 static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
146 static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
147 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
148 static unsigned long __meminitdata node_boundary_start_pfn[MAX_NUMNODES];
149 static unsigned long __meminitdata node_boundary_end_pfn[MAX_NUMNODES];
150 #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
151 unsigned long __initdata required_kernelcore;
152 unsigned long __initdata required_movablecore;
153 unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
155 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
157 EXPORT_SYMBOL(movable_zone);
158 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
161 int nr_node_ids __read_mostly = MAX_NUMNODES;
162 EXPORT_SYMBOL(nr_node_ids);
165 int page_group_by_mobility_disabled __read_mostly;
167 static void set_pageblock_migratetype(struct page *page, int migratetype)
169 set_pageblock_flags_group(page, (unsigned long)migratetype,
170 PB_migrate, PB_migrate_end);
173 #ifdef CONFIG_DEBUG_VM
174 static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
178 unsigned long pfn = page_to_pfn(page);
181 seq = zone_span_seqbegin(zone);
182 if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
184 else if (pfn < zone->zone_start_pfn)
186 } while (zone_span_seqretry(zone, seq));
191 static int page_is_consistent(struct zone *zone, struct page *page)
193 if (!pfn_valid_within(page_to_pfn(page)))
195 if (zone != page_zone(page))
201 * Temporary debugging check for pages not lying within a given zone.
203 static int bad_range(struct zone *zone, struct page *page)
205 if (page_outside_zone_boundaries(zone, page))
207 if (!page_is_consistent(zone, page))
213 static inline int bad_range(struct zone *zone, struct page *page)
219 static void bad_page(struct page *page)
221 printk(KERN_EMERG "Bad page state in process '%s'\n"
222 KERN_EMERG "page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n"
223 KERN_EMERG "Trying to fix it up, but a reboot is needed\n"
224 KERN_EMERG "Backtrace:\n",
225 current->comm, page, (int)(2*sizeof(unsigned long)),
226 (unsigned long)page->flags, page->mapping,
227 page_mapcount(page), page_count(page));
229 page->flags &= ~(1 << PG_lru |
239 set_page_count(page, 0);
240 reset_page_mapcount(page);
241 page->mapping = NULL;
242 add_taint(TAINT_BAD_PAGE);
246 * Higher-order pages are called "compound pages". They are structured thusly:
248 * The first PAGE_SIZE page is called the "head page".
250 * The remaining PAGE_SIZE pages are called "tail pages".
252 * All pages have PG_compound set. All pages have their ->private pointing at
253 * the head page (even the head page has this).
255 * The first tail page's ->lru.next holds the address of the compound page's
256 * put_page() function. Its ->lru.prev holds the order of allocation.
257 * This usage means that zero-order pages may not be compound.
260 static void free_compound_page(struct page *page)
262 __free_pages_ok(page, compound_order(page));
265 static void prep_compound_page(struct page *page, unsigned long order)
268 int nr_pages = 1 << order;
270 set_compound_page_dtor(page, free_compound_page);
271 set_compound_order(page, order);
273 for (i = 1; i < nr_pages; i++) {
274 struct page *p = page + i;
277 p->first_page = page;
281 static void destroy_compound_page(struct page *page, unsigned long order)
284 int nr_pages = 1 << order;
286 if (unlikely(compound_order(page) != order))
289 if (unlikely(!PageHead(page)))
291 __ClearPageHead(page);
292 for (i = 1; i < nr_pages; i++) {
293 struct page *p = page + i;
295 if (unlikely(!PageTail(p) |
296 (p->first_page != page)))
302 static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
306 VM_BUG_ON((gfp_flags & (__GFP_WAIT | __GFP_HIGHMEM)) == __GFP_HIGHMEM);
308 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
309 * and __GFP_HIGHMEM from hard or soft interrupt context.
311 VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
312 for (i = 0; i < (1 << order); i++)
313 clear_highpage(page + i);
317 * function for dealing with page's order in buddy system.
318 * zone->lock is already acquired when we use these.
319 * So, we don't need atomic page->flags operations here.
321 static inline unsigned long page_order(struct page *page)
323 return page_private(page);
326 static inline void set_page_order(struct page *page, int order)
328 set_page_private(page, order);
329 __SetPageBuddy(page);
332 static inline void rmv_page_order(struct page *page)
334 __ClearPageBuddy(page);
335 set_page_private(page, 0);
339 * Locate the struct page for both the matching buddy in our
340 * pair (buddy1) and the combined O(n+1) page they form (page).
342 * 1) Any buddy B1 will have an order O twin B2 which satisfies
343 * the following equation:
345 * For example, if the starting buddy (buddy2) is #8 its order
347 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
349 * 2) Any buddy B will have an order O+1 parent P which
350 * satisfies the following equation:
353 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
355 static inline struct page *
356 __page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order)
358 unsigned long buddy_idx = page_idx ^ (1 << order);
360 return page + (buddy_idx - page_idx);
363 static inline unsigned long
364 __find_combined_index(unsigned long page_idx, unsigned int order)
366 return (page_idx & ~(1 << order));
370 * This function checks whether a page is free && is the buddy
371 * we can do coalesce a page and its buddy if
372 * (a) the buddy is not in a hole &&
373 * (b) the buddy is in the buddy system &&
374 * (c) a page and its buddy have the same order &&
375 * (d) a page and its buddy are in the same zone.
377 * For recording whether a page is in the buddy system, we use PG_buddy.
378 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
380 * For recording page's order, we use page_private(page).
382 static inline int page_is_buddy(struct page *page, struct page *buddy,
385 if (!pfn_valid_within(page_to_pfn(buddy)))
388 if (page_zone_id(page) != page_zone_id(buddy))
391 if (PageBuddy(buddy) && page_order(buddy) == order) {
392 BUG_ON(page_count(buddy) != 0);
399 * Freeing function for a buddy system allocator.
401 * The concept of a buddy system is to maintain direct-mapped table
402 * (containing bit values) for memory blocks of various "orders".
403 * The bottom level table contains the map for the smallest allocatable
404 * units of memory (here, pages), and each level above it describes
405 * pairs of units from the levels below, hence, "buddies".
406 * At a high level, all that happens here is marking the table entry
407 * at the bottom level available, and propagating the changes upward
408 * as necessary, plus some accounting needed to play nicely with other
409 * parts of the VM system.
410 * At each level, we keep a list of pages, which are heads of continuous
411 * free pages of length of (1 << order) and marked with PG_buddy. Page's
412 * order is recorded in page_private(page) field.
413 * So when we are allocating or freeing one, we can derive the state of the
414 * other. That is, if we allocate a small block, and both were
415 * free, the remainder of the region must be split into blocks.
416 * If a block is freed, and its buddy is also free, then this
417 * triggers coalescing into a block of larger size.
422 static inline void __free_one_page(struct page *page,
423 struct zone *zone, unsigned int order)
425 unsigned long page_idx;
426 int order_size = 1 << order;
427 int migratetype = get_pageblock_migratetype(page);
429 if (unlikely(PageCompound(page)))
430 destroy_compound_page(page, order);
432 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
434 VM_BUG_ON(page_idx & (order_size - 1));
435 VM_BUG_ON(bad_range(zone, page));
437 __mod_zone_page_state(zone, NR_FREE_PAGES, order_size);
438 while (order < MAX_ORDER-1) {
439 unsigned long combined_idx;
442 buddy = __page_find_buddy(page, page_idx, order);
443 if (!page_is_buddy(page, buddy, order))
444 break; /* Move the buddy up one level. */
446 list_del(&buddy->lru);
447 zone->free_area[order].nr_free--;
448 rmv_page_order(buddy);
449 combined_idx = __find_combined_index(page_idx, order);
450 page = page + (combined_idx - page_idx);
451 page_idx = combined_idx;
454 set_page_order(page, order);
456 &zone->free_area[order].free_list[migratetype]);
457 zone->free_area[order].nr_free++;
460 static inline int free_pages_check(struct page *page)
462 if (unlikely(page_mapcount(page) |
463 (page->mapping != NULL) |
464 (page_count(page) != 0) |
477 __ClearPageDirty(page);
479 * For now, we report if PG_reserved was found set, but do not
480 * clear it, and do not free the page. But we shall soon need
481 * to do more, for when the ZERO_PAGE count wraps negative.
483 return PageReserved(page);
487 * Frees a list of pages.
488 * Assumes all pages on list are in same zone, and of same order.
489 * count is the number of pages to free.
491 * If the zone was previously in an "all pages pinned" state then look to
492 * see if this freeing clears that state.
494 * And clear the zone's pages_scanned counter, to hold off the "all pages are
495 * pinned" detection logic.
497 static void free_pages_bulk(struct zone *zone, int count,
498 struct list_head *list, int order)
500 spin_lock(&zone->lock);
501 zone->all_unreclaimable = 0;
502 zone->pages_scanned = 0;
506 VM_BUG_ON(list_empty(list));
507 page = list_entry(list->prev, struct page, lru);
508 /* have to delete it as __free_one_page list manipulates */
509 list_del(&page->lru);
510 __free_one_page(page, zone, order);
512 spin_unlock(&zone->lock);
515 static void free_one_page(struct zone *zone, struct page *page, int order)
517 spin_lock(&zone->lock);
518 zone->all_unreclaimable = 0;
519 zone->pages_scanned = 0;
520 __free_one_page(page, zone, order);
521 spin_unlock(&zone->lock);
524 static void __free_pages_ok(struct page *page, unsigned int order)
530 for (i = 0 ; i < (1 << order) ; ++i)
531 reserved += free_pages_check(page + i);
535 if (!PageHighMem(page))
536 debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order);
537 arch_free_page(page, order);
538 kernel_map_pages(page, 1 << order, 0);
540 local_irq_save(flags);
541 __count_vm_events(PGFREE, 1 << order);
542 free_one_page(page_zone(page), page, order);
543 local_irq_restore(flags);
547 * permit the bootmem allocator to evade page validation on high-order frees
549 void fastcall __init __free_pages_bootmem(struct page *page, unsigned int order)
552 __ClearPageReserved(page);
553 set_page_count(page, 0);
554 set_page_refcounted(page);
560 for (loop = 0; loop < BITS_PER_LONG; loop++) {
561 struct page *p = &page[loop];
563 if (loop + 1 < BITS_PER_LONG)
565 __ClearPageReserved(p);
566 set_page_count(p, 0);
569 set_page_refcounted(page);
570 __free_pages(page, order);
576 * The order of subdivision here is critical for the IO subsystem.
577 * Please do not alter this order without good reasons and regression
578 * testing. Specifically, as large blocks of memory are subdivided,
579 * the order in which smaller blocks are delivered depends on the order
580 * they're subdivided in this function. This is the primary factor
581 * influencing the order in which pages are delivered to the IO
582 * subsystem according to empirical testing, and this is also justified
583 * by considering the behavior of a buddy system containing a single
584 * large block of memory acted on by a series of small allocations.
585 * This behavior is a critical factor in sglist merging's success.
589 static inline void expand(struct zone *zone, struct page *page,
590 int low, int high, struct free_area *area,
593 unsigned long size = 1 << high;
599 VM_BUG_ON(bad_range(zone, &page[size]));
600 list_add(&page[size].lru, &area->free_list[migratetype]);
602 set_page_order(&page[size], high);
607 * This page is about to be returned from the page allocator
609 static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
611 if (unlikely(page_mapcount(page) |
612 (page->mapping != NULL) |
613 (page_count(page) != 0) |
628 * For now, we report if PG_reserved was found set, but do not
629 * clear it, and do not allocate the page: as a safety net.
631 if (PageReserved(page))
634 page->flags &= ~(1 << PG_uptodate | 1 << PG_error | 1 << PG_readahead |
635 1 << PG_referenced | 1 << PG_arch_1 |
636 1 << PG_owner_priv_1 | 1 << PG_mappedtodisk);
637 set_page_private(page, 0);
638 set_page_refcounted(page);
640 arch_alloc_page(page, order);
641 kernel_map_pages(page, 1 << order, 1);
643 if (gfp_flags & __GFP_ZERO)
644 prep_zero_page(page, order, gfp_flags);
646 if (order && (gfp_flags & __GFP_COMP))
647 prep_compound_page(page, order);
653 * Go through the free lists for the given migratetype and remove
654 * the smallest available page from the freelists
656 static struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
659 unsigned int current_order;
660 struct free_area * area;
663 /* Find a page of the appropriate size in the preferred list */
664 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
665 area = &(zone->free_area[current_order]);
666 if (list_empty(&area->free_list[migratetype]))
669 page = list_entry(area->free_list[migratetype].next,
671 list_del(&page->lru);
672 rmv_page_order(page);
674 __mod_zone_page_state(zone, NR_FREE_PAGES, - (1UL << order));
675 expand(zone, page, order, current_order, area, migratetype);
684 * This array describes the order lists are fallen back to when
685 * the free lists for the desirable migrate type are depleted
687 static int fallbacks[MIGRATE_TYPES][MIGRATE_TYPES-1] = {
688 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
689 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
690 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
691 [MIGRATE_RESERVE] = { MIGRATE_RESERVE, MIGRATE_RESERVE, MIGRATE_RESERVE }, /* Never used */
695 * Move the free pages in a range to the free lists of the requested type.
696 * Note that start_page and end_pages are not aligned on a pageblock
697 * boundary. If alignment is required, use move_freepages_block()
699 int move_freepages(struct zone *zone,
700 struct page *start_page, struct page *end_page,
707 #ifndef CONFIG_HOLES_IN_ZONE
709 * page_zone is not safe to call in this context when
710 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
711 * anyway as we check zone boundaries in move_freepages_block().
712 * Remove at a later date when no bug reports exist related to
713 * grouping pages by mobility
715 BUG_ON(page_zone(start_page) != page_zone(end_page));
718 for (page = start_page; page <= end_page;) {
719 if (!pfn_valid_within(page_to_pfn(page))) {
724 if (!PageBuddy(page)) {
729 order = page_order(page);
730 list_del(&page->lru);
732 &zone->free_area[order].free_list[migratetype]);
734 pages_moved += 1 << order;
740 int move_freepages_block(struct zone *zone, struct page *page, int migratetype)
742 unsigned long start_pfn, end_pfn;
743 struct page *start_page, *end_page;
745 start_pfn = page_to_pfn(page);
746 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
747 start_page = pfn_to_page(start_pfn);
748 end_page = start_page + pageblock_nr_pages - 1;
749 end_pfn = start_pfn + pageblock_nr_pages - 1;
751 /* Do not cross zone boundaries */
752 if (start_pfn < zone->zone_start_pfn)
754 if (end_pfn >= zone->zone_start_pfn + zone->spanned_pages)
757 return move_freepages(zone, start_page, end_page, migratetype);
760 /* Return the page with the lowest PFN in the list */
761 static struct page *min_page(struct list_head *list)
763 unsigned long min_pfn = -1UL;
764 struct page *min_page = NULL, *page;;
766 list_for_each_entry(page, list, lru) {
767 unsigned long pfn = page_to_pfn(page);
777 /* Remove an element from the buddy allocator from the fallback list */
778 static struct page *__rmqueue_fallback(struct zone *zone, int order,
779 int start_migratetype)
781 struct free_area * area;
786 /* Find the largest possible block of pages in the other list */
787 for (current_order = MAX_ORDER-1; current_order >= order;
789 for (i = 0; i < MIGRATE_TYPES - 1; i++) {
790 migratetype = fallbacks[start_migratetype][i];
792 /* MIGRATE_RESERVE handled later if necessary */
793 if (migratetype == MIGRATE_RESERVE)
796 area = &(zone->free_area[current_order]);
797 if (list_empty(&area->free_list[migratetype]))
800 /* Bias kernel allocations towards low pfns */
801 page = list_entry(area->free_list[migratetype].next,
803 if (unlikely(start_migratetype != MIGRATE_MOVABLE))
804 page = min_page(&area->free_list[migratetype]);
808 * If breaking a large block of pages, move all free
809 * pages to the preferred allocation list. If falling
810 * back for a reclaimable kernel allocation, be more
811 * agressive about taking ownership of free pages
813 if (unlikely(current_order >= (pageblock_order >> 1)) ||
814 start_migratetype == MIGRATE_RECLAIMABLE) {
816 pages = move_freepages_block(zone, page,
819 /* Claim the whole block if over half of it is free */
820 if (pages >= (1 << (pageblock_order-1)))
821 set_pageblock_migratetype(page,
824 migratetype = start_migratetype;
827 /* Remove the page from the freelists */
828 list_del(&page->lru);
829 rmv_page_order(page);
830 __mod_zone_page_state(zone, NR_FREE_PAGES,
833 if (current_order == pageblock_order)
834 set_pageblock_migratetype(page,
837 expand(zone, page, order, current_order, area, migratetype);
842 /* Use MIGRATE_RESERVE rather than fail an allocation */
843 return __rmqueue_smallest(zone, order, MIGRATE_RESERVE);
847 * Do the hard work of removing an element from the buddy allocator.
848 * Call me with the zone->lock already held.
850 static struct page *__rmqueue(struct zone *zone, unsigned int order,
855 page = __rmqueue_smallest(zone, order, migratetype);
858 page = __rmqueue_fallback(zone, order, migratetype);
864 * Obtain a specified number of elements from the buddy allocator, all under
865 * a single hold of the lock, for efficiency. Add them to the supplied list.
866 * Returns the number of new pages which were placed at *list.
868 static int rmqueue_bulk(struct zone *zone, unsigned int order,
869 unsigned long count, struct list_head *list,
874 spin_lock(&zone->lock);
875 for (i = 0; i < count; ++i) {
876 struct page *page = __rmqueue(zone, order, migratetype);
877 if (unlikely(page == NULL))
879 list_add(&page->lru, list);
880 set_page_private(page, migratetype);
882 spin_unlock(&zone->lock);
888 * Called from the vmstat counter updater to drain pagesets of this
889 * currently executing processor on remote nodes after they have
892 * Note that this function must be called with the thread pinned to
893 * a single processor.
895 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
900 local_irq_save(flags);
901 if (pcp->count >= pcp->batch)
902 to_drain = pcp->batch;
904 to_drain = pcp->count;
905 free_pages_bulk(zone, to_drain, &pcp->list, 0);
906 pcp->count -= to_drain;
907 local_irq_restore(flags);
911 static void __drain_pages(unsigned int cpu)
917 for_each_zone(zone) {
918 struct per_cpu_pageset *pset;
920 if (!populated_zone(zone))
923 pset = zone_pcp(zone, cpu);
924 for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) {
925 struct per_cpu_pages *pcp;
928 local_irq_save(flags);
929 free_pages_bulk(zone, pcp->count, &pcp->list, 0);
931 local_irq_restore(flags);
936 #ifdef CONFIG_HIBERNATION
938 void mark_free_pages(struct zone *zone)
940 unsigned long pfn, max_zone_pfn;
943 struct list_head *curr;
945 if (!zone->spanned_pages)
948 spin_lock_irqsave(&zone->lock, flags);
950 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
951 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
952 if (pfn_valid(pfn)) {
953 struct page *page = pfn_to_page(pfn);
955 if (!swsusp_page_is_forbidden(page))
956 swsusp_unset_page_free(page);
959 for_each_migratetype_order(order, t) {
960 list_for_each(curr, &zone->free_area[order].free_list[t]) {
963 pfn = page_to_pfn(list_entry(curr, struct page, lru));
964 for (i = 0; i < (1UL << order); i++)
965 swsusp_set_page_free(pfn_to_page(pfn + i));
968 spin_unlock_irqrestore(&zone->lock, flags);
970 #endif /* CONFIG_PM */
973 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
975 void drain_local_pages(void)
979 local_irq_save(flags);
980 __drain_pages(smp_processor_id());
981 local_irq_restore(flags);
984 void smp_drain_local_pages(void *arg)
990 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
992 void drain_all_local_pages(void)
996 local_irq_save(flags);
997 __drain_pages(smp_processor_id());
998 local_irq_restore(flags);
1000 smp_call_function(smp_drain_local_pages, NULL, 0, 1);
1004 * Free a 0-order page
1006 static void fastcall free_hot_cold_page(struct page *page, int cold)
1008 struct zone *zone = page_zone(page);
1009 struct per_cpu_pages *pcp;
1010 unsigned long flags;
1013 page->mapping = NULL;
1014 if (free_pages_check(page))
1017 if (!PageHighMem(page))
1018 debug_check_no_locks_freed(page_address(page), PAGE_SIZE);
1019 arch_free_page(page, 0);
1020 kernel_map_pages(page, 1, 0);
1022 pcp = &zone_pcp(zone, get_cpu())->pcp[cold];
1023 local_irq_save(flags);
1024 __count_vm_event(PGFREE);
1025 list_add(&page->lru, &pcp->list);
1026 set_page_private(page, get_pageblock_migratetype(page));
1028 if (pcp->count >= pcp->high) {
1029 free_pages_bulk(zone, pcp->batch, &pcp->list, 0);
1030 pcp->count -= pcp->batch;
1032 local_irq_restore(flags);
1036 void fastcall free_hot_page(struct page *page)
1038 free_hot_cold_page(page, 0);
1041 void fastcall free_cold_page(struct page *page)
1043 free_hot_cold_page(page, 1);
1047 * split_page takes a non-compound higher-order page, and splits it into
1048 * n (1<<order) sub-pages: page[0..n]
1049 * Each sub-page must be freed individually.
1051 * Note: this is probably too low level an operation for use in drivers.
1052 * Please consult with lkml before using this in your driver.
1054 void split_page(struct page *page, unsigned int order)
1058 VM_BUG_ON(PageCompound(page));
1059 VM_BUG_ON(!page_count(page));
1060 for (i = 1; i < (1 << order); i++)
1061 set_page_refcounted(page + i);
1065 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1066 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1069 static struct page *buffered_rmqueue(struct zonelist *zonelist,
1070 struct zone *zone, int order, gfp_t gfp_flags)
1072 unsigned long flags;
1074 int cold = !!(gfp_flags & __GFP_COLD);
1076 int migratetype = allocflags_to_migratetype(gfp_flags);
1080 if (likely(order == 0)) {
1081 struct per_cpu_pages *pcp;
1083 pcp = &zone_pcp(zone, cpu)->pcp[cold];
1084 local_irq_save(flags);
1086 pcp->count = rmqueue_bulk(zone, 0,
1087 pcp->batch, &pcp->list, migratetype);
1088 if (unlikely(!pcp->count))
1092 /* Find a page of the appropriate migrate type */
1093 list_for_each_entry(page, &pcp->list, lru)
1094 if (page_private(page) == migratetype)
1097 /* Allocate more to the pcp list if necessary */
1098 if (unlikely(&page->lru == &pcp->list)) {
1099 pcp->count += rmqueue_bulk(zone, 0,
1100 pcp->batch, &pcp->list, migratetype);
1101 page = list_entry(pcp->list.next, struct page, lru);
1104 list_del(&page->lru);
1107 spin_lock_irqsave(&zone->lock, flags);
1108 page = __rmqueue(zone, order, migratetype);
1109 spin_unlock(&zone->lock);
1114 __count_zone_vm_events(PGALLOC, zone, 1 << order);
1115 zone_statistics(zonelist, zone);
1116 local_irq_restore(flags);
1119 VM_BUG_ON(bad_range(zone, page));
1120 if (prep_new_page(page, order, gfp_flags))
1125 local_irq_restore(flags);
1130 #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
1131 #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
1132 #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
1133 #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
1134 #define ALLOC_HARDER 0x10 /* try to alloc harder */
1135 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1136 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
1138 #ifdef CONFIG_FAIL_PAGE_ALLOC
1140 static struct fail_page_alloc_attr {
1141 struct fault_attr attr;
1143 u32 ignore_gfp_highmem;
1144 u32 ignore_gfp_wait;
1147 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1149 struct dentry *ignore_gfp_highmem_file;
1150 struct dentry *ignore_gfp_wait_file;
1151 struct dentry *min_order_file;
1153 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1155 } fail_page_alloc = {
1156 .attr = FAULT_ATTR_INITIALIZER,
1157 .ignore_gfp_wait = 1,
1158 .ignore_gfp_highmem = 1,
1162 static int __init setup_fail_page_alloc(char *str)
1164 return setup_fault_attr(&fail_page_alloc.attr, str);
1166 __setup("fail_page_alloc=", setup_fail_page_alloc);
1168 static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1170 if (order < fail_page_alloc.min_order)
1172 if (gfp_mask & __GFP_NOFAIL)
1174 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
1176 if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
1179 return should_fail(&fail_page_alloc.attr, 1 << order);
1182 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1184 static int __init fail_page_alloc_debugfs(void)
1186 mode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
1190 err = init_fault_attr_dentries(&fail_page_alloc.attr,
1194 dir = fail_page_alloc.attr.dentries.dir;
1196 fail_page_alloc.ignore_gfp_wait_file =
1197 debugfs_create_bool("ignore-gfp-wait", mode, dir,
1198 &fail_page_alloc.ignore_gfp_wait);
1200 fail_page_alloc.ignore_gfp_highmem_file =
1201 debugfs_create_bool("ignore-gfp-highmem", mode, dir,
1202 &fail_page_alloc.ignore_gfp_highmem);
1203 fail_page_alloc.min_order_file =
1204 debugfs_create_u32("min-order", mode, dir,
1205 &fail_page_alloc.min_order);
1207 if (!fail_page_alloc.ignore_gfp_wait_file ||
1208 !fail_page_alloc.ignore_gfp_highmem_file ||
1209 !fail_page_alloc.min_order_file) {
1211 debugfs_remove(fail_page_alloc.ignore_gfp_wait_file);
1212 debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file);
1213 debugfs_remove(fail_page_alloc.min_order_file);
1214 cleanup_fault_attr_dentries(&fail_page_alloc.attr);
1220 late_initcall(fail_page_alloc_debugfs);
1222 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1224 #else /* CONFIG_FAIL_PAGE_ALLOC */
1226 static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1231 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1234 * Return 1 if free pages are above 'mark'. This takes into account the order
1235 * of the allocation.
1237 int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1238 int classzone_idx, int alloc_flags)
1240 /* free_pages my go negative - that's OK */
1242 long free_pages = zone_page_state(z, NR_FREE_PAGES) - (1 << order) + 1;
1245 if (alloc_flags & ALLOC_HIGH)
1247 if (alloc_flags & ALLOC_HARDER)
1250 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
1252 for (o = 0; o < order; o++) {
1253 /* At the next order, this order's pages become unavailable */
1254 free_pages -= z->free_area[o].nr_free << o;
1256 /* Require fewer higher order pages to be free */
1259 if (free_pages <= min)
1267 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1268 * skip over zones that are not allowed by the cpuset, or that have
1269 * been recently (in last second) found to be nearly full. See further
1270 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1271 * that have to skip over alot of full or unallowed zones.
1273 * If the zonelist cache is present in the passed in zonelist, then
1274 * returns a pointer to the allowed node mask (either the current
1275 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
1277 * If the zonelist cache is not available for this zonelist, does
1278 * nothing and returns NULL.
1280 * If the fullzones BITMAP in the zonelist cache is stale (more than
1281 * a second since last zap'd) then we zap it out (clear its bits.)
1283 * We hold off even calling zlc_setup, until after we've checked the
1284 * first zone in the zonelist, on the theory that most allocations will
1285 * be satisfied from that first zone, so best to examine that zone as
1286 * quickly as we can.
1288 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1290 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1291 nodemask_t *allowednodes; /* zonelist_cache approximation */
1293 zlc = zonelist->zlcache_ptr;
1297 if (jiffies - zlc->last_full_zap > 1 * HZ) {
1298 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1299 zlc->last_full_zap = jiffies;
1302 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1303 &cpuset_current_mems_allowed :
1304 &node_states[N_HIGH_MEMORY];
1305 return allowednodes;
1309 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1310 * if it is worth looking at further for free memory:
1311 * 1) Check that the zone isn't thought to be full (doesn't have its
1312 * bit set in the zonelist_cache fullzones BITMAP).
1313 * 2) Check that the zones node (obtained from the zonelist_cache
1314 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1315 * Return true (non-zero) if zone is worth looking at further, or
1316 * else return false (zero) if it is not.
1318 * This check -ignores- the distinction between various watermarks,
1319 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1320 * found to be full for any variation of these watermarks, it will
1321 * be considered full for up to one second by all requests, unless
1322 * we are so low on memory on all allowed nodes that we are forced
1323 * into the second scan of the zonelist.
1325 * In the second scan we ignore this zonelist cache and exactly
1326 * apply the watermarks to all zones, even it is slower to do so.
1327 * We are low on memory in the second scan, and should leave no stone
1328 * unturned looking for a free page.
1330 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zone **z,
1331 nodemask_t *allowednodes)
1333 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1334 int i; /* index of *z in zonelist zones */
1335 int n; /* node that zone *z is on */
1337 zlc = zonelist->zlcache_ptr;
1341 i = z - zonelist->zones;
1344 /* This zone is worth trying if it is allowed but not full */
1345 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1349 * Given 'z' scanning a zonelist, set the corresponding bit in
1350 * zlc->fullzones, so that subsequent attempts to allocate a page
1351 * from that zone don't waste time re-examining it.
1353 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zone **z)
1355 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1356 int i; /* index of *z in zonelist zones */
1358 zlc = zonelist->zlcache_ptr;
1362 i = z - zonelist->zones;
1364 set_bit(i, zlc->fullzones);
1367 #else /* CONFIG_NUMA */
1369 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1374 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zone **z,
1375 nodemask_t *allowednodes)
1380 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zone **z)
1383 #endif /* CONFIG_NUMA */
1386 * get_page_from_freelist goes through the zonelist trying to allocate
1389 static struct page *
1390 get_page_from_freelist(gfp_t gfp_mask, unsigned int order,
1391 struct zonelist *zonelist, int alloc_flags)
1394 struct page *page = NULL;
1395 int classzone_idx = zone_idx(zonelist->zones[0]);
1397 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1398 int zlc_active = 0; /* set if using zonelist_cache */
1399 int did_zlc_setup = 0; /* just call zlc_setup() one time */
1400 enum zone_type highest_zoneidx = -1; /* Gets set for policy zonelists */
1404 * Scan zonelist, looking for a zone with enough free.
1405 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1407 z = zonelist->zones;
1411 * In NUMA, this could be a policy zonelist which contains
1412 * zones that may not be allowed by the current gfp_mask.
1413 * Check the zone is allowed by the current flags
1415 if (unlikely(alloc_should_filter_zonelist(zonelist))) {
1416 if (highest_zoneidx == -1)
1417 highest_zoneidx = gfp_zone(gfp_mask);
1418 if (zone_idx(*z) > highest_zoneidx)
1422 if (NUMA_BUILD && zlc_active &&
1423 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1426 if ((alloc_flags & ALLOC_CPUSET) &&
1427 !cpuset_zone_allowed_softwall(zone, gfp_mask))
1430 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
1432 if (alloc_flags & ALLOC_WMARK_MIN)
1433 mark = zone->pages_min;
1434 else if (alloc_flags & ALLOC_WMARK_LOW)
1435 mark = zone->pages_low;
1437 mark = zone->pages_high;
1438 if (!zone_watermark_ok(zone, order, mark,
1439 classzone_idx, alloc_flags)) {
1440 if (!zone_reclaim_mode ||
1441 !zone_reclaim(zone, gfp_mask, order))
1442 goto this_zone_full;
1446 page = buffered_rmqueue(zonelist, zone, order, gfp_mask);
1451 zlc_mark_zone_full(zonelist, z);
1453 if (NUMA_BUILD && !did_zlc_setup) {
1454 /* we do zlc_setup after the first zone is tried */
1455 allowednodes = zlc_setup(zonelist, alloc_flags);
1459 } while (*(++z) != NULL);
1461 if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) {
1462 /* Disable zlc cache for second zonelist scan */
1470 * This is the 'heart' of the zoned buddy allocator.
1472 struct page * fastcall
1473 __alloc_pages(gfp_t gfp_mask, unsigned int order,
1474 struct zonelist *zonelist)
1476 const gfp_t wait = gfp_mask & __GFP_WAIT;
1479 struct reclaim_state reclaim_state;
1480 struct task_struct *p = current;
1483 int did_some_progress;
1485 might_sleep_if(wait);
1487 if (should_fail_alloc_page(gfp_mask, order))
1491 z = zonelist->zones; /* the list of zones suitable for gfp_mask */
1493 if (unlikely(*z == NULL)) {
1495 * Happens if we have an empty zonelist as a result of
1496 * GFP_THISNODE being used on a memoryless node
1501 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
1502 zonelist, ALLOC_WMARK_LOW|ALLOC_CPUSET);
1507 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1508 * __GFP_NOWARN set) should not cause reclaim since the subsystem
1509 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1510 * using a larger set of nodes after it has established that the
1511 * allowed per node queues are empty and that nodes are
1514 if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
1517 for (z = zonelist->zones; *z; z++)
1518 wakeup_kswapd(*z, order);
1521 * OK, we're below the kswapd watermark and have kicked background
1522 * reclaim. Now things get more complex, so set up alloc_flags according
1523 * to how we want to proceed.
1525 * The caller may dip into page reserves a bit more if the caller
1526 * cannot run direct reclaim, or if the caller has realtime scheduling
1527 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1528 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1530 alloc_flags = ALLOC_WMARK_MIN;
1531 if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait)
1532 alloc_flags |= ALLOC_HARDER;
1533 if (gfp_mask & __GFP_HIGH)
1534 alloc_flags |= ALLOC_HIGH;
1536 alloc_flags |= ALLOC_CPUSET;
1539 * Go through the zonelist again. Let __GFP_HIGH and allocations
1540 * coming from realtime tasks go deeper into reserves.
1542 * This is the last chance, in general, before the goto nopage.
1543 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1544 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1546 page = get_page_from_freelist(gfp_mask, order, zonelist, alloc_flags);
1550 /* This allocation should allow future memory freeing. */
1553 if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE)))
1554 && !in_interrupt()) {
1555 if (!(gfp_mask & __GFP_NOMEMALLOC)) {
1557 /* go through the zonelist yet again, ignoring mins */
1558 page = get_page_from_freelist(gfp_mask, order,
1559 zonelist, ALLOC_NO_WATERMARKS);
1562 if (gfp_mask & __GFP_NOFAIL) {
1563 congestion_wait(WRITE, HZ/50);
1570 /* Atomic allocations - we can't balance anything */
1576 /* We now go into synchronous reclaim */
1577 cpuset_memory_pressure_bump();
1578 p->flags |= PF_MEMALLOC;
1579 reclaim_state.reclaimed_slab = 0;
1580 p->reclaim_state = &reclaim_state;
1582 did_some_progress = try_to_free_pages(zonelist->zones, order, gfp_mask);
1584 p->reclaim_state = NULL;
1585 p->flags &= ~PF_MEMALLOC;
1590 drain_all_local_pages();
1592 if (likely(did_some_progress)) {
1593 page = get_page_from_freelist(gfp_mask, order,
1594 zonelist, alloc_flags);
1597 } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
1599 * Go through the zonelist yet one more time, keep
1600 * very high watermark here, this is only to catch
1601 * a parallel oom killing, we must fail if we're still
1602 * under heavy pressure.
1604 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
1605 zonelist, ALLOC_WMARK_HIGH|ALLOC_CPUSET);
1609 /* The OOM killer will not help higher order allocs so fail */
1610 if (order > PAGE_ALLOC_COSTLY_ORDER)
1613 out_of_memory(zonelist, gfp_mask, order);
1618 * Don't let big-order allocations loop unless the caller explicitly
1619 * requests that. Wait for some write requests to complete then retry.
1621 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
1622 * <= 3, but that may not be true in other implementations.
1625 if (!(gfp_mask & __GFP_NORETRY)) {
1626 if ((order <= PAGE_ALLOC_COSTLY_ORDER) ||
1627 (gfp_mask & __GFP_REPEAT))
1629 if (gfp_mask & __GFP_NOFAIL)
1633 congestion_wait(WRITE, HZ/50);
1638 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
1639 printk(KERN_WARNING "%s: page allocation failure."
1640 " order:%d, mode:0x%x\n",
1641 p->comm, order, gfp_mask);
1649 EXPORT_SYMBOL(__alloc_pages);
1652 * Common helper functions.
1654 fastcall unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
1657 page = alloc_pages(gfp_mask, order);
1660 return (unsigned long) page_address(page);
1663 EXPORT_SYMBOL(__get_free_pages);
1665 fastcall unsigned long get_zeroed_page(gfp_t gfp_mask)
1670 * get_zeroed_page() returns a 32-bit address, which cannot represent
1673 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
1675 page = alloc_pages(gfp_mask | __GFP_ZERO, 0);
1677 return (unsigned long) page_address(page);
1681 EXPORT_SYMBOL(get_zeroed_page);
1683 void __pagevec_free(struct pagevec *pvec)
1685 int i = pagevec_count(pvec);
1688 free_hot_cold_page(pvec->pages[i], pvec->cold);
1691 fastcall void __free_pages(struct page *page, unsigned int order)
1693 if (put_page_testzero(page)) {
1695 free_hot_page(page);
1697 __free_pages_ok(page, order);
1701 EXPORT_SYMBOL(__free_pages);
1703 fastcall void free_pages(unsigned long addr, unsigned int order)
1706 VM_BUG_ON(!virt_addr_valid((void *)addr));
1707 __free_pages(virt_to_page((void *)addr), order);
1711 EXPORT_SYMBOL(free_pages);
1713 static unsigned int nr_free_zone_pages(int offset)
1715 /* Just pick one node, since fallback list is circular */
1716 pg_data_t *pgdat = NODE_DATA(numa_node_id());
1717 unsigned int sum = 0;
1719 struct zonelist *zonelist = pgdat->node_zonelists + offset;
1720 struct zone **zonep = zonelist->zones;
1723 for (zone = *zonep++; zone; zone = *zonep++) {
1724 unsigned long size = zone->present_pages;
1725 unsigned long high = zone->pages_high;
1734 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1736 unsigned int nr_free_buffer_pages(void)
1738 return nr_free_zone_pages(gfp_zone(GFP_USER));
1740 EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
1743 * Amount of free RAM allocatable within all zones
1745 unsigned int nr_free_pagecache_pages(void)
1747 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
1750 static inline void show_node(struct zone *zone)
1753 printk("Node %d ", zone_to_nid(zone));
1756 void si_meminfo(struct sysinfo *val)
1758 val->totalram = totalram_pages;
1760 val->freeram = global_page_state(NR_FREE_PAGES);
1761 val->bufferram = nr_blockdev_pages();
1762 val->totalhigh = totalhigh_pages;
1763 val->freehigh = nr_free_highpages();
1764 val->mem_unit = PAGE_SIZE;
1767 EXPORT_SYMBOL(si_meminfo);
1770 void si_meminfo_node(struct sysinfo *val, int nid)
1772 pg_data_t *pgdat = NODE_DATA(nid);
1774 val->totalram = pgdat->node_present_pages;
1775 val->freeram = node_page_state(nid, NR_FREE_PAGES);
1776 #ifdef CONFIG_HIGHMEM
1777 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1778 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
1784 val->mem_unit = PAGE_SIZE;
1788 #define K(x) ((x) << (PAGE_SHIFT-10))
1791 * Show free area list (used inside shift_scroll-lock stuff)
1792 * We also calculate the percentage fragmentation. We do this by counting the
1793 * memory on each free list with the exception of the first item on the list.
1795 void show_free_areas(void)
1800 for_each_zone(zone) {
1801 if (!populated_zone(zone))
1805 printk("%s per-cpu:\n", zone->name);
1807 for_each_online_cpu(cpu) {
1808 struct per_cpu_pageset *pageset;
1810 pageset = zone_pcp(zone, cpu);
1812 printk("CPU %4d: Hot: hi:%5d, btch:%4d usd:%4d "
1813 "Cold: hi:%5d, btch:%4d usd:%4d\n",
1814 cpu, pageset->pcp[0].high,
1815 pageset->pcp[0].batch, pageset->pcp[0].count,
1816 pageset->pcp[1].high, pageset->pcp[1].batch,
1817 pageset->pcp[1].count);
1821 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu unstable:%lu\n"
1822 " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
1823 global_page_state(NR_ACTIVE),
1824 global_page_state(NR_INACTIVE),
1825 global_page_state(NR_FILE_DIRTY),
1826 global_page_state(NR_WRITEBACK),
1827 global_page_state(NR_UNSTABLE_NFS),
1828 global_page_state(NR_FREE_PAGES),
1829 global_page_state(NR_SLAB_RECLAIMABLE) +
1830 global_page_state(NR_SLAB_UNRECLAIMABLE),
1831 global_page_state(NR_FILE_MAPPED),
1832 global_page_state(NR_PAGETABLE),
1833 global_page_state(NR_BOUNCE));
1835 for_each_zone(zone) {
1838 if (!populated_zone(zone))
1850 " pages_scanned:%lu"
1851 " all_unreclaimable? %s"
1854 K(zone_page_state(zone, NR_FREE_PAGES)),
1857 K(zone->pages_high),
1858 K(zone_page_state(zone, NR_ACTIVE)),
1859 K(zone_page_state(zone, NR_INACTIVE)),
1860 K(zone->present_pages),
1861 zone->pages_scanned,
1862 (zone->all_unreclaimable ? "yes" : "no")
1864 printk("lowmem_reserve[]:");
1865 for (i = 0; i < MAX_NR_ZONES; i++)
1866 printk(" %lu", zone->lowmem_reserve[i]);
1870 for_each_zone(zone) {
1871 unsigned long nr[MAX_ORDER], flags, order, total = 0;
1873 if (!populated_zone(zone))
1877 printk("%s: ", zone->name);
1879 spin_lock_irqsave(&zone->lock, flags);
1880 for (order = 0; order < MAX_ORDER; order++) {
1881 nr[order] = zone->free_area[order].nr_free;
1882 total += nr[order] << order;
1884 spin_unlock_irqrestore(&zone->lock, flags);
1885 for (order = 0; order < MAX_ORDER; order++)
1886 printk("%lu*%lukB ", nr[order], K(1UL) << order);
1887 printk("= %lukB\n", K(total));
1890 show_swap_cache_info();
1894 * Builds allocation fallback zone lists.
1896 * Add all populated zones of a node to the zonelist.
1898 static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
1899 int nr_zones, enum zone_type zone_type)
1903 BUG_ON(zone_type >= MAX_NR_ZONES);
1908 zone = pgdat->node_zones + zone_type;
1909 if (populated_zone(zone)) {
1910 zonelist->zones[nr_zones++] = zone;
1911 check_highest_zone(zone_type);
1914 } while (zone_type);
1921 * 0 = automatic detection of better ordering.
1922 * 1 = order by ([node] distance, -zonetype)
1923 * 2 = order by (-zonetype, [node] distance)
1925 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
1926 * the same zonelist. So only NUMA can configure this param.
1928 #define ZONELIST_ORDER_DEFAULT 0
1929 #define ZONELIST_ORDER_NODE 1
1930 #define ZONELIST_ORDER_ZONE 2
1932 /* zonelist order in the kernel.
1933 * set_zonelist_order() will set this to NODE or ZONE.
1935 static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
1936 static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
1940 /* The value user specified ....changed by config */
1941 static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
1942 /* string for sysctl */
1943 #define NUMA_ZONELIST_ORDER_LEN 16
1944 char numa_zonelist_order[16] = "default";
1947 * interface for configure zonelist ordering.
1948 * command line option "numa_zonelist_order"
1949 * = "[dD]efault - default, automatic configuration.
1950 * = "[nN]ode - order by node locality, then by zone within node
1951 * = "[zZ]one - order by zone, then by locality within zone
1954 static int __parse_numa_zonelist_order(char *s)
1956 if (*s == 'd' || *s == 'D') {
1957 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
1958 } else if (*s == 'n' || *s == 'N') {
1959 user_zonelist_order = ZONELIST_ORDER_NODE;
1960 } else if (*s == 'z' || *s == 'Z') {
1961 user_zonelist_order = ZONELIST_ORDER_ZONE;
1964 "Ignoring invalid numa_zonelist_order value: "
1971 static __init int setup_numa_zonelist_order(char *s)
1974 return __parse_numa_zonelist_order(s);
1977 early_param("numa_zonelist_order", setup_numa_zonelist_order);
1980 * sysctl handler for numa_zonelist_order
1982 int numa_zonelist_order_handler(ctl_table *table, int write,
1983 struct file *file, void __user *buffer, size_t *length,
1986 char saved_string[NUMA_ZONELIST_ORDER_LEN];
1990 strncpy(saved_string, (char*)table->data,
1991 NUMA_ZONELIST_ORDER_LEN);
1992 ret = proc_dostring(table, write, file, buffer, length, ppos);
1996 int oldval = user_zonelist_order;
1997 if (__parse_numa_zonelist_order((char*)table->data)) {
1999 * bogus value. restore saved string
2001 strncpy((char*)table->data, saved_string,
2002 NUMA_ZONELIST_ORDER_LEN);
2003 user_zonelist_order = oldval;
2004 } else if (oldval != user_zonelist_order)
2005 build_all_zonelists();
2011 #define MAX_NODE_LOAD (num_online_nodes())
2012 static int node_load[MAX_NUMNODES];
2015 * find_next_best_node - find the next node that should appear in a given node's fallback list
2016 * @node: node whose fallback list we're appending
2017 * @used_node_mask: nodemask_t of already used nodes
2019 * We use a number of factors to determine which is the next node that should
2020 * appear on a given node's fallback list. The node should not have appeared
2021 * already in @node's fallback list, and it should be the next closest node
2022 * according to the distance array (which contains arbitrary distance values
2023 * from each node to each node in the system), and should also prefer nodes
2024 * with no CPUs, since presumably they'll have very little allocation pressure
2025 * on them otherwise.
2026 * It returns -1 if no node is found.
2028 static int find_next_best_node(int node, nodemask_t *used_node_mask)
2031 int min_val = INT_MAX;
2034 /* Use the local node if we haven't already */
2035 if (!node_isset(node, *used_node_mask)) {
2036 node_set(node, *used_node_mask);
2040 for_each_node_state(n, N_HIGH_MEMORY) {
2043 /* Don't want a node to appear more than once */
2044 if (node_isset(n, *used_node_mask))
2047 /* Use the distance array to find the distance */
2048 val = node_distance(node, n);
2050 /* Penalize nodes under us ("prefer the next node") */
2053 /* Give preference to headless and unused nodes */
2054 tmp = node_to_cpumask(n);
2055 if (!cpus_empty(tmp))
2056 val += PENALTY_FOR_NODE_WITH_CPUS;
2058 /* Slight preference for less loaded node */
2059 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
2060 val += node_load[n];
2062 if (val < min_val) {
2069 node_set(best_node, *used_node_mask);
2076 * Build zonelists ordered by node and zones within node.
2077 * This results in maximum locality--normal zone overflows into local
2078 * DMA zone, if any--but risks exhausting DMA zone.
2080 static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
2084 struct zonelist *zonelist;
2086 for (i = 0; i < MAX_NR_ZONES; i++) {
2087 zonelist = pgdat->node_zonelists + i;
2088 for (j = 0; zonelist->zones[j] != NULL; j++)
2090 j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
2091 zonelist->zones[j] = NULL;
2096 * Build gfp_thisnode zonelists
2098 static void build_thisnode_zonelists(pg_data_t *pgdat)
2102 struct zonelist *zonelist;
2104 for (i = 0; i < MAX_NR_ZONES; i++) {
2105 zonelist = pgdat->node_zonelists + MAX_NR_ZONES + i;
2106 j = build_zonelists_node(pgdat, zonelist, 0, i);
2107 zonelist->zones[j] = NULL;
2112 * Build zonelists ordered by zone and nodes within zones.
2113 * This results in conserving DMA zone[s] until all Normal memory is
2114 * exhausted, but results in overflowing to remote node while memory
2115 * may still exist in local DMA zone.
2117 static int node_order[MAX_NUMNODES];
2119 static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
2123 int zone_type; /* needs to be signed */
2125 struct zonelist *zonelist;
2127 for (i = 0; i < MAX_NR_ZONES; i++) {
2128 zonelist = pgdat->node_zonelists + i;
2130 for (zone_type = i; zone_type >= 0; zone_type--) {
2131 for (j = 0; j < nr_nodes; j++) {
2132 node = node_order[j];
2133 z = &NODE_DATA(node)->node_zones[zone_type];
2134 if (populated_zone(z)) {
2135 zonelist->zones[pos++] = z;
2136 check_highest_zone(zone_type);
2140 zonelist->zones[pos] = NULL;
2144 static int default_zonelist_order(void)
2147 unsigned long low_kmem_size,total_size;
2151 * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem.
2152 * If they are really small and used heavily, the system can fall
2153 * into OOM very easily.
2154 * This function detect ZONE_DMA/DMA32 size and confgigures zone order.
2156 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2159 for_each_online_node(nid) {
2160 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2161 z = &NODE_DATA(nid)->node_zones[zone_type];
2162 if (populated_zone(z)) {
2163 if (zone_type < ZONE_NORMAL)
2164 low_kmem_size += z->present_pages;
2165 total_size += z->present_pages;
2169 if (!low_kmem_size || /* there are no DMA area. */
2170 low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
2171 return ZONELIST_ORDER_NODE;
2173 * look into each node's config.
2174 * If there is a node whose DMA/DMA32 memory is very big area on
2175 * local memory, NODE_ORDER may be suitable.
2177 average_size = total_size /
2178 (nodes_weight(node_states[N_HIGH_MEMORY]) + 1);
2179 for_each_online_node(nid) {
2182 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2183 z = &NODE_DATA(nid)->node_zones[zone_type];
2184 if (populated_zone(z)) {
2185 if (zone_type < ZONE_NORMAL)
2186 low_kmem_size += z->present_pages;
2187 total_size += z->present_pages;
2190 if (low_kmem_size &&
2191 total_size > average_size && /* ignore small node */
2192 low_kmem_size > total_size * 70/100)
2193 return ZONELIST_ORDER_NODE;
2195 return ZONELIST_ORDER_ZONE;
2198 static void set_zonelist_order(void)
2200 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
2201 current_zonelist_order = default_zonelist_order();
2203 current_zonelist_order = user_zonelist_order;
2206 static void build_zonelists(pg_data_t *pgdat)
2210 nodemask_t used_mask;
2211 int local_node, prev_node;
2212 struct zonelist *zonelist;
2213 int order = current_zonelist_order;
2215 /* initialize zonelists */
2216 for (i = 0; i < MAX_ZONELISTS; i++) {
2217 zonelist = pgdat->node_zonelists + i;
2218 zonelist->zones[0] = NULL;
2221 /* NUMA-aware ordering of nodes */
2222 local_node = pgdat->node_id;
2223 load = num_online_nodes();
2224 prev_node = local_node;
2225 nodes_clear(used_mask);
2227 memset(node_load, 0, sizeof(node_load));
2228 memset(node_order, 0, sizeof(node_order));
2231 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
2232 int distance = node_distance(local_node, node);
2235 * If another node is sufficiently far away then it is better
2236 * to reclaim pages in a zone before going off node.
2238 if (distance > RECLAIM_DISTANCE)
2239 zone_reclaim_mode = 1;
2242 * We don't want to pressure a particular node.
2243 * So adding penalty to the first node in same
2244 * distance group to make it round-robin.
2246 if (distance != node_distance(local_node, prev_node))
2247 node_load[node] = load;
2251 if (order == ZONELIST_ORDER_NODE)
2252 build_zonelists_in_node_order(pgdat, node);
2254 node_order[j++] = node; /* remember order */
2257 if (order == ZONELIST_ORDER_ZONE) {
2258 /* calculate node order -- i.e., DMA last! */
2259 build_zonelists_in_zone_order(pgdat, j);
2262 build_thisnode_zonelists(pgdat);
2265 /* Construct the zonelist performance cache - see further mmzone.h */
2266 static void build_zonelist_cache(pg_data_t *pgdat)
2270 for (i = 0; i < MAX_NR_ZONES; i++) {
2271 struct zonelist *zonelist;
2272 struct zonelist_cache *zlc;
2275 zonelist = pgdat->node_zonelists + i;
2276 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
2277 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
2278 for (z = zonelist->zones; *z; z++)
2279 zlc->z_to_n[z - zonelist->zones] = zone_to_nid(*z);
2284 #else /* CONFIG_NUMA */
2286 static void set_zonelist_order(void)
2288 current_zonelist_order = ZONELIST_ORDER_ZONE;
2291 static void build_zonelists(pg_data_t *pgdat)
2293 int node, local_node;
2296 local_node = pgdat->node_id;
2297 for (i = 0; i < MAX_NR_ZONES; i++) {
2298 struct zonelist *zonelist;
2300 zonelist = pgdat->node_zonelists + i;
2302 j = build_zonelists_node(pgdat, zonelist, 0, i);
2304 * Now we build the zonelist so that it contains the zones
2305 * of all the other nodes.
2306 * We don't want to pressure a particular node, so when
2307 * building the zones for node N, we make sure that the
2308 * zones coming right after the local ones are those from
2309 * node N+1 (modulo N)
2311 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
2312 if (!node_online(node))
2314 j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
2316 for (node = 0; node < local_node; node++) {
2317 if (!node_online(node))
2319 j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
2322 zonelist->zones[j] = NULL;
2326 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
2327 static void build_zonelist_cache(pg_data_t *pgdat)
2331 for (i = 0; i < MAX_NR_ZONES; i++)
2332 pgdat->node_zonelists[i].zlcache_ptr = NULL;
2335 #endif /* CONFIG_NUMA */
2337 /* return values int ....just for stop_machine_run() */
2338 static int __build_all_zonelists(void *dummy)
2342 for_each_online_node(nid) {
2343 pg_data_t *pgdat = NODE_DATA(nid);
2345 build_zonelists(pgdat);
2346 build_zonelist_cache(pgdat);
2351 void build_all_zonelists(void)
2353 set_zonelist_order();
2355 if (system_state == SYSTEM_BOOTING) {
2356 __build_all_zonelists(NULL);
2357 cpuset_init_current_mems_allowed();
2359 /* we have to stop all cpus to guaranntee there is no user
2361 stop_machine_run(__build_all_zonelists, NULL, NR_CPUS);
2362 /* cpuset refresh routine should be here */
2364 vm_total_pages = nr_free_pagecache_pages();
2366 * Disable grouping by mobility if the number of pages in the
2367 * system is too low to allow the mechanism to work. It would be
2368 * more accurate, but expensive to check per-zone. This check is
2369 * made on memory-hotadd so a system can start with mobility
2370 * disabled and enable it later
2372 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
2373 page_group_by_mobility_disabled = 1;
2375 page_group_by_mobility_disabled = 0;
2377 printk("Built %i zonelists in %s order, mobility grouping %s. "
2378 "Total pages: %ld\n",
2380 zonelist_order_name[current_zonelist_order],
2381 page_group_by_mobility_disabled ? "off" : "on",
2384 printk("Policy zone: %s\n", zone_names[policy_zone]);
2389 * Helper functions to size the waitqueue hash table.
2390 * Essentially these want to choose hash table sizes sufficiently
2391 * large so that collisions trying to wait on pages are rare.
2392 * But in fact, the number of active page waitqueues on typical
2393 * systems is ridiculously low, less than 200. So this is even
2394 * conservative, even though it seems large.
2396 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
2397 * waitqueues, i.e. the size of the waitq table given the number of pages.
2399 #define PAGES_PER_WAITQUEUE 256
2401 #ifndef CONFIG_MEMORY_HOTPLUG
2402 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
2404 unsigned long size = 1;
2406 pages /= PAGES_PER_WAITQUEUE;
2408 while (size < pages)
2412 * Once we have dozens or even hundreds of threads sleeping
2413 * on IO we've got bigger problems than wait queue collision.
2414 * Limit the size of the wait table to a reasonable size.
2416 size = min(size, 4096UL);
2418 return max(size, 4UL);
2422 * A zone's size might be changed by hot-add, so it is not possible to determine
2423 * a suitable size for its wait_table. So we use the maximum size now.
2425 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
2427 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
2428 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
2429 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
2431 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
2432 * or more by the traditional way. (See above). It equals:
2434 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
2435 * ia64(16K page size) : = ( 8G + 4M)byte.
2436 * powerpc (64K page size) : = (32G +16M)byte.
2438 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
2445 * This is an integer logarithm so that shifts can be used later
2446 * to extract the more random high bits from the multiplicative
2447 * hash function before the remainder is taken.
2449 static inline unsigned long wait_table_bits(unsigned long size)
2454 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
2457 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
2458 * of blocks reserved is based on zone->pages_min. The memory within the
2459 * reserve will tend to store contiguous free pages. Setting min_free_kbytes
2460 * higher will lead to a bigger reserve which will get freed as contiguous
2461 * blocks as reclaim kicks in
2463 static void setup_zone_migrate_reserve(struct zone *zone)
2465 unsigned long start_pfn, pfn, end_pfn;
2467 unsigned long reserve, block_migratetype;
2469 /* Get the start pfn, end pfn and the number of blocks to reserve */
2470 start_pfn = zone->zone_start_pfn;
2471 end_pfn = start_pfn + zone->spanned_pages;
2472 reserve = roundup(zone->pages_min, pageblock_nr_pages) >>
2475 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
2476 if (!pfn_valid(pfn))
2478 page = pfn_to_page(pfn);
2480 /* Blocks with reserved pages will never free, skip them. */
2481 if (PageReserved(page))
2484 block_migratetype = get_pageblock_migratetype(page);
2486 /* If this block is reserved, account for it */
2487 if (reserve > 0 && block_migratetype == MIGRATE_RESERVE) {
2492 /* Suitable for reserving if this block is movable */
2493 if (reserve > 0 && block_migratetype == MIGRATE_MOVABLE) {
2494 set_pageblock_migratetype(page, MIGRATE_RESERVE);
2495 move_freepages_block(zone, page, MIGRATE_RESERVE);
2501 * If the reserve is met and this is a previous reserved block,
2504 if (block_migratetype == MIGRATE_RESERVE) {
2505 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
2506 move_freepages_block(zone, page, MIGRATE_MOVABLE);
2512 * Initially all pages are reserved - free ones are freed
2513 * up by free_all_bootmem() once the early boot process is
2514 * done. Non-atomic initialization, single-pass.
2516 void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
2517 unsigned long start_pfn, enum memmap_context context)
2520 unsigned long end_pfn = start_pfn + size;
2523 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
2525 * There can be holes in boot-time mem_map[]s
2526 * handed to this function. They do not
2527 * exist on hotplugged memory.
2529 if (context == MEMMAP_EARLY) {
2530 if (!early_pfn_valid(pfn))
2532 if (!early_pfn_in_nid(pfn, nid))
2535 page = pfn_to_page(pfn);
2536 set_page_links(page, zone, nid, pfn);
2537 init_page_count(page);
2538 reset_page_mapcount(page);
2539 SetPageReserved(page);
2542 * Mark the block movable so that blocks are reserved for
2543 * movable at startup. This will force kernel allocations
2544 * to reserve their blocks rather than leaking throughout
2545 * the address space during boot when many long-lived
2546 * kernel allocations are made. Later some blocks near
2547 * the start are marked MIGRATE_RESERVE by
2548 * setup_zone_migrate_reserve()
2550 if ((pfn & (pageblock_nr_pages-1)))
2551 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
2553 INIT_LIST_HEAD(&page->lru);
2554 #ifdef WANT_PAGE_VIRTUAL
2555 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
2556 if (!is_highmem_idx(zone))
2557 set_page_address(page, __va(pfn << PAGE_SHIFT));
2562 static void __meminit zone_init_free_lists(struct pglist_data *pgdat,
2563 struct zone *zone, unsigned long size)
2566 for_each_migratetype_order(order, t) {
2567 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
2568 zone->free_area[order].nr_free = 0;
2572 #ifndef __HAVE_ARCH_MEMMAP_INIT
2573 #define memmap_init(size, nid, zone, start_pfn) \
2574 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
2577 static int __devinit zone_batchsize(struct zone *zone)
2582 * The per-cpu-pages pools are set to around 1000th of the
2583 * size of the zone. But no more than 1/2 of a meg.
2585 * OK, so we don't know how big the cache is. So guess.
2587 batch = zone->present_pages / 1024;
2588 if (batch * PAGE_SIZE > 512 * 1024)
2589 batch = (512 * 1024) / PAGE_SIZE;
2590 batch /= 4; /* We effectively *= 4 below */
2595 * Clamp the batch to a 2^n - 1 value. Having a power
2596 * of 2 value was found to be more likely to have
2597 * suboptimal cache aliasing properties in some cases.
2599 * For example if 2 tasks are alternately allocating
2600 * batches of pages, one task can end up with a lot
2601 * of pages of one half of the possible page colors
2602 * and the other with pages of the other colors.
2604 batch = (1 << (fls(batch + batch/2)-1)) - 1;
2609 inline void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
2611 struct per_cpu_pages *pcp;
2613 memset(p, 0, sizeof(*p));
2615 pcp = &p->pcp[0]; /* hot */
2617 pcp->high = 6 * batch;
2618 pcp->batch = max(1UL, 1 * batch);
2619 INIT_LIST_HEAD(&pcp->list);
2621 pcp = &p->pcp[1]; /* cold*/
2623 pcp->high = 2 * batch;
2624 pcp->batch = max(1UL, batch/2);
2625 INIT_LIST_HEAD(&pcp->list);
2629 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
2630 * to the value high for the pageset p.
2633 static void setup_pagelist_highmark(struct per_cpu_pageset *p,
2636 struct per_cpu_pages *pcp;
2638 pcp = &p->pcp[0]; /* hot list */
2640 pcp->batch = max(1UL, high/4);
2641 if ((high/4) > (PAGE_SHIFT * 8))
2642 pcp->batch = PAGE_SHIFT * 8;
2648 * Boot pageset table. One per cpu which is going to be used for all
2649 * zones and all nodes. The parameters will be set in such a way
2650 * that an item put on a list will immediately be handed over to
2651 * the buddy list. This is safe since pageset manipulation is done
2652 * with interrupts disabled.
2654 * Some NUMA counter updates may also be caught by the boot pagesets.
2656 * The boot_pagesets must be kept even after bootup is complete for
2657 * unused processors and/or zones. They do play a role for bootstrapping
2658 * hotplugged processors.
2660 * zoneinfo_show() and maybe other functions do
2661 * not check if the processor is online before following the pageset pointer.
2662 * Other parts of the kernel may not check if the zone is available.
2664 static struct per_cpu_pageset boot_pageset[NR_CPUS];
2667 * Dynamically allocate memory for the
2668 * per cpu pageset array in struct zone.
2670 static int __cpuinit process_zones(int cpu)
2672 struct zone *zone, *dzone;
2673 int node = cpu_to_node(cpu);
2675 node_set_state(node, N_CPU); /* this node has a cpu */
2677 for_each_zone(zone) {
2679 if (!populated_zone(zone))
2682 zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset),
2684 if (!zone_pcp(zone, cpu))
2687 setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone));
2689 if (percpu_pagelist_fraction)
2690 setup_pagelist_highmark(zone_pcp(zone, cpu),
2691 (zone->present_pages / percpu_pagelist_fraction));
2696 for_each_zone(dzone) {
2697 if (!populated_zone(dzone))
2701 kfree(zone_pcp(dzone, cpu));
2702 zone_pcp(dzone, cpu) = NULL;
2707 static inline void free_zone_pagesets(int cpu)
2711 for_each_zone(zone) {
2712 struct per_cpu_pageset *pset = zone_pcp(zone, cpu);
2714 /* Free per_cpu_pageset if it is slab allocated */
2715 if (pset != &boot_pageset[cpu])
2717 zone_pcp(zone, cpu) = NULL;
2721 static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb,
2722 unsigned long action,
2725 int cpu = (long)hcpu;
2726 int ret = NOTIFY_OK;
2729 case CPU_UP_PREPARE:
2730 case CPU_UP_PREPARE_FROZEN:
2731 if (process_zones(cpu))
2734 case CPU_UP_CANCELED:
2735 case CPU_UP_CANCELED_FROZEN:
2737 case CPU_DEAD_FROZEN:
2738 free_zone_pagesets(cpu);
2746 static struct notifier_block __cpuinitdata pageset_notifier =
2747 { &pageset_cpuup_callback, NULL, 0 };
2749 void __init setup_per_cpu_pageset(void)
2753 /* Initialize per_cpu_pageset for cpu 0.
2754 * A cpuup callback will do this for every cpu
2755 * as it comes online
2757 err = process_zones(smp_processor_id());
2759 register_cpu_notifier(&pageset_notifier);
2764 static noinline __init_refok
2765 int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
2768 struct pglist_data *pgdat = zone->zone_pgdat;
2772 * The per-page waitqueue mechanism uses hashed waitqueues
2775 zone->wait_table_hash_nr_entries =
2776 wait_table_hash_nr_entries(zone_size_pages);
2777 zone->wait_table_bits =
2778 wait_table_bits(zone->wait_table_hash_nr_entries);
2779 alloc_size = zone->wait_table_hash_nr_entries
2780 * sizeof(wait_queue_head_t);
2782 if (system_state == SYSTEM_BOOTING) {
2783 zone->wait_table = (wait_queue_head_t *)
2784 alloc_bootmem_node(pgdat, alloc_size);
2787 * This case means that a zone whose size was 0 gets new memory
2788 * via memory hot-add.
2789 * But it may be the case that a new node was hot-added. In
2790 * this case vmalloc() will not be able to use this new node's
2791 * memory - this wait_table must be initialized to use this new
2792 * node itself as well.
2793 * To use this new node's memory, further consideration will be
2796 zone->wait_table = vmalloc(alloc_size);
2798 if (!zone->wait_table)
2801 for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
2802 init_waitqueue_head(zone->wait_table + i);
2807 static __meminit void zone_pcp_init(struct zone *zone)
2810 unsigned long batch = zone_batchsize(zone);
2812 for (cpu = 0; cpu < NR_CPUS; cpu++) {
2814 /* Early boot. Slab allocator not functional yet */
2815 zone_pcp(zone, cpu) = &boot_pageset[cpu];
2816 setup_pageset(&boot_pageset[cpu],0);
2818 setup_pageset(zone_pcp(zone,cpu), batch);
2821 if (zone->present_pages)
2822 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n",
2823 zone->name, zone->present_pages, batch);
2826 __meminit int init_currently_empty_zone(struct zone *zone,
2827 unsigned long zone_start_pfn,
2829 enum memmap_context context)
2831 struct pglist_data *pgdat = zone->zone_pgdat;
2833 ret = zone_wait_table_init(zone, size);
2836 pgdat->nr_zones = zone_idx(zone) + 1;
2838 zone->zone_start_pfn = zone_start_pfn;
2840 memmap_init(size, pgdat->node_id, zone_idx(zone), zone_start_pfn);
2842 zone_init_free_lists(pgdat, zone, zone->spanned_pages);
2847 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2849 * Basic iterator support. Return the first range of PFNs for a node
2850 * Note: nid == MAX_NUMNODES returns first region regardless of node
2852 static int __meminit first_active_region_index_in_nid(int nid)
2856 for (i = 0; i < nr_nodemap_entries; i++)
2857 if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
2864 * Basic iterator support. Return the next active range of PFNs for a node
2865 * Note: nid == MAX_NUMNODES returns next region regardles of node
2867 static int __meminit next_active_region_index_in_nid(int index, int nid)
2869 for (index = index + 1; index < nr_nodemap_entries; index++)
2870 if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
2876 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
2878 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
2879 * Architectures may implement their own version but if add_active_range()
2880 * was used and there are no special requirements, this is a convenient
2883 int __meminit early_pfn_to_nid(unsigned long pfn)
2887 for (i = 0; i < nr_nodemap_entries; i++) {
2888 unsigned long start_pfn = early_node_map[i].start_pfn;
2889 unsigned long end_pfn = early_node_map[i].end_pfn;
2891 if (start_pfn <= pfn && pfn < end_pfn)
2892 return early_node_map[i].nid;
2897 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
2899 /* Basic iterator support to walk early_node_map[] */
2900 #define for_each_active_range_index_in_nid(i, nid) \
2901 for (i = first_active_region_index_in_nid(nid); i != -1; \
2902 i = next_active_region_index_in_nid(i, nid))
2905 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
2906 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
2907 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
2909 * If an architecture guarantees that all ranges registered with
2910 * add_active_ranges() contain no holes and may be freed, this
2911 * this function may be used instead of calling free_bootmem() manually.
2913 void __init free_bootmem_with_active_regions(int nid,
2914 unsigned long max_low_pfn)
2918 for_each_active_range_index_in_nid(i, nid) {
2919 unsigned long size_pages = 0;
2920 unsigned long end_pfn = early_node_map[i].end_pfn;
2922 if (early_node_map[i].start_pfn >= max_low_pfn)
2925 if (end_pfn > max_low_pfn)
2926 end_pfn = max_low_pfn;
2928 size_pages = end_pfn - early_node_map[i].start_pfn;
2929 free_bootmem_node(NODE_DATA(early_node_map[i].nid),
2930 PFN_PHYS(early_node_map[i].start_pfn),
2931 size_pages << PAGE_SHIFT);
2936 * sparse_memory_present_with_active_regions - Call memory_present for each active range
2937 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
2939 * If an architecture guarantees that all ranges registered with
2940 * add_active_ranges() contain no holes and may be freed, this
2941 * function may be used instead of calling memory_present() manually.
2943 void __init sparse_memory_present_with_active_regions(int nid)
2947 for_each_active_range_index_in_nid(i, nid)
2948 memory_present(early_node_map[i].nid,
2949 early_node_map[i].start_pfn,
2950 early_node_map[i].end_pfn);
2954 * push_node_boundaries - Push node boundaries to at least the requested boundary
2955 * @nid: The nid of the node to push the boundary for
2956 * @start_pfn: The start pfn of the node
2957 * @end_pfn: The end pfn of the node
2959 * In reserve-based hot-add, mem_map is allocated that is unused until hotadd
2960 * time. Specifically, on x86_64, SRAT will report ranges that can potentially
2961 * be hotplugged even though no physical memory exists. This function allows
2962 * an arch to push out the node boundaries so mem_map is allocated that can
2965 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
2966 void __init push_node_boundaries(unsigned int nid,
2967 unsigned long start_pfn, unsigned long end_pfn)
2969 printk(KERN_DEBUG "Entering push_node_boundaries(%u, %lu, %lu)\n",
2970 nid, start_pfn, end_pfn);
2972 /* Initialise the boundary for this node if necessary */
2973 if (node_boundary_end_pfn[nid] == 0)
2974 node_boundary_start_pfn[nid] = -1UL;
2976 /* Update the boundaries */
2977 if (node_boundary_start_pfn[nid] > start_pfn)
2978 node_boundary_start_pfn[nid] = start_pfn;
2979 if (node_boundary_end_pfn[nid] < end_pfn)
2980 node_boundary_end_pfn[nid] = end_pfn;
2983 /* If necessary, push the node boundary out for reserve hotadd */
2984 static void __meminit account_node_boundary(unsigned int nid,
2985 unsigned long *start_pfn, unsigned long *end_pfn)
2987 printk(KERN_DEBUG "Entering account_node_boundary(%u, %lu, %lu)\n",
2988 nid, *start_pfn, *end_pfn);
2990 /* Return if boundary information has not been provided */
2991 if (node_boundary_end_pfn[nid] == 0)
2994 /* Check the boundaries and update if necessary */
2995 if (node_boundary_start_pfn[nid] < *start_pfn)
2996 *start_pfn = node_boundary_start_pfn[nid];
2997 if (node_boundary_end_pfn[nid] > *end_pfn)
2998 *end_pfn = node_boundary_end_pfn[nid];
3001 void __init push_node_boundaries(unsigned int nid,
3002 unsigned long start_pfn, unsigned long end_pfn) {}
3004 static void __meminit account_node_boundary(unsigned int nid,
3005 unsigned long *start_pfn, unsigned long *end_pfn) {}
3010 * get_pfn_range_for_nid - Return the start and end page frames for a node
3011 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
3012 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
3013 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
3015 * It returns the start and end page frame of a node based on information
3016 * provided by an arch calling add_active_range(). If called for a node
3017 * with no available memory, a warning is printed and the start and end
3020 void __meminit get_pfn_range_for_nid(unsigned int nid,
3021 unsigned long *start_pfn, unsigned long *end_pfn)
3027 for_each_active_range_index_in_nid(i, nid) {
3028 *start_pfn = min(*start_pfn, early_node_map[i].start_pfn);
3029 *end_pfn = max(*end_pfn, early_node_map[i].end_pfn);
3032 if (*start_pfn == -1UL)
3035 /* Push the node boundaries out if requested */
3036 account_node_boundary(nid, start_pfn, end_pfn);
3040 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3041 * assumption is made that zones within a node are ordered in monotonic
3042 * increasing memory addresses so that the "highest" populated zone is used
3044 void __init find_usable_zone_for_movable(void)
3047 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
3048 if (zone_index == ZONE_MOVABLE)
3051 if (arch_zone_highest_possible_pfn[zone_index] >
3052 arch_zone_lowest_possible_pfn[zone_index])
3056 VM_BUG_ON(zone_index == -1);
3057 movable_zone = zone_index;
3061 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
3062 * because it is sized independant of architecture. Unlike the other zones,
3063 * the starting point for ZONE_MOVABLE is not fixed. It may be different
3064 * in each node depending on the size of each node and how evenly kernelcore
3065 * is distributed. This helper function adjusts the zone ranges
3066 * provided by the architecture for a given node by using the end of the
3067 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3068 * zones within a node are in order of monotonic increases memory addresses
3070 void __meminit adjust_zone_range_for_zone_movable(int nid,
3071 unsigned long zone_type,
3072 unsigned long node_start_pfn,
3073 unsigned long node_end_pfn,
3074 unsigned long *zone_start_pfn,
3075 unsigned long *zone_end_pfn)
3077 /* Only adjust if ZONE_MOVABLE is on this node */
3078 if (zone_movable_pfn[nid]) {
3079 /* Size ZONE_MOVABLE */
3080 if (zone_type == ZONE_MOVABLE) {
3081 *zone_start_pfn = zone_movable_pfn[nid];
3082 *zone_end_pfn = min(node_end_pfn,
3083 arch_zone_highest_possible_pfn[movable_zone]);
3085 /* Adjust for ZONE_MOVABLE starting within this range */
3086 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
3087 *zone_end_pfn > zone_movable_pfn[nid]) {
3088 *zone_end_pfn = zone_movable_pfn[nid];
3090 /* Check if this whole range is within ZONE_MOVABLE */
3091 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
3092 *zone_start_pfn = *zone_end_pfn;
3097 * Return the number of pages a zone spans in a node, including holes
3098 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3100 static unsigned long __meminit zone_spanned_pages_in_node(int nid,
3101 unsigned long zone_type,
3102 unsigned long *ignored)
3104 unsigned long node_start_pfn, node_end_pfn;
3105 unsigned long zone_start_pfn, zone_end_pfn;
3107 /* Get the start and end of the node and zone */
3108 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3109 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
3110 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
3111 adjust_zone_range_for_zone_movable(nid, zone_type,
3112 node_start_pfn, node_end_pfn,
3113 &zone_start_pfn, &zone_end_pfn);
3115 /* Check that this node has pages within the zone's required range */
3116 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
3119 /* Move the zone boundaries inside the node if necessary */
3120 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
3121 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
3123 /* Return the spanned pages */
3124 return zone_end_pfn - zone_start_pfn;
3128 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
3129 * then all holes in the requested range will be accounted for.
3131 unsigned long __meminit __absent_pages_in_range(int nid,
3132 unsigned long range_start_pfn,
3133 unsigned long range_end_pfn)
3136 unsigned long prev_end_pfn = 0, hole_pages = 0;
3137 unsigned long start_pfn;
3139 /* Find the end_pfn of the first active range of pfns in the node */
3140 i = first_active_region_index_in_nid(nid);
3144 prev_end_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3146 /* Account for ranges before physical memory on this node */
3147 if (early_node_map[i].start_pfn > range_start_pfn)
3148 hole_pages = prev_end_pfn - range_start_pfn;
3150 /* Find all holes for the zone within the node */
3151 for (; i != -1; i = next_active_region_index_in_nid(i, nid)) {
3153 /* No need to continue if prev_end_pfn is outside the zone */
3154 if (prev_end_pfn >= range_end_pfn)
3157 /* Make sure the end of the zone is not within the hole */
3158 start_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3159 prev_end_pfn = max(prev_end_pfn, range_start_pfn);
3161 /* Update the hole size cound and move on */
3162 if (start_pfn > range_start_pfn) {
3163 BUG_ON(prev_end_pfn > start_pfn);
3164 hole_pages += start_pfn - prev_end_pfn;
3166 prev_end_pfn = early_node_map[i].end_pfn;
3169 /* Account for ranges past physical memory on this node */
3170 if (range_end_pfn > prev_end_pfn)
3171 hole_pages += range_end_pfn -
3172 max(range_start_pfn, prev_end_pfn);
3178 * absent_pages_in_range - Return number of page frames in holes within a range
3179 * @start_pfn: The start PFN to start searching for holes
3180 * @end_pfn: The end PFN to stop searching for holes
3182 * It returns the number of pages frames in memory holes within a range.
3184 unsigned long __init absent_pages_in_range(unsigned long start_pfn,
3185 unsigned long end_pfn)
3187 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
3190 /* Return the number of page frames in holes in a zone on a node */
3191 static unsigned long __meminit zone_absent_pages_in_node(int nid,
3192 unsigned long zone_type,
3193 unsigned long *ignored)
3195 unsigned long node_start_pfn, node_end_pfn;
3196 unsigned long zone_start_pfn, zone_end_pfn;
3198 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3199 zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type],
3201 zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type],
3204 adjust_zone_range_for_zone_movable(nid, zone_type,
3205 node_start_pfn, node_end_pfn,
3206 &zone_start_pfn, &zone_end_pfn);
3207 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
3211 static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
3212 unsigned long zone_type,
3213 unsigned long *zones_size)
3215 return zones_size[zone_type];
3218 static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
3219 unsigned long zone_type,
3220 unsigned long *zholes_size)
3225 return zholes_size[zone_type];
3230 static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
3231 unsigned long *zones_size, unsigned long *zholes_size)
3233 unsigned long realtotalpages, totalpages = 0;
3236 for (i = 0; i < MAX_NR_ZONES; i++)
3237 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
3239 pgdat->node_spanned_pages = totalpages;
3241 realtotalpages = totalpages;
3242 for (i = 0; i < MAX_NR_ZONES; i++)
3244 zone_absent_pages_in_node(pgdat->node_id, i,
3246 pgdat->node_present_pages = realtotalpages;
3247 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
3251 #ifndef CONFIG_SPARSEMEM
3253 * Calculate the size of the zone->blockflags rounded to an unsigned long
3254 * Start by making sure zonesize is a multiple of pageblock_order by rounding
3255 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
3256 * round what is now in bits to nearest long in bits, then return it in
3259 static unsigned long __init usemap_size(unsigned long zonesize)
3261 unsigned long usemapsize;
3263 usemapsize = roundup(zonesize, pageblock_nr_pages);
3264 usemapsize = usemapsize >> pageblock_order;
3265 usemapsize *= NR_PAGEBLOCK_BITS;
3266 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
3268 return usemapsize / 8;
3271 static void __init setup_usemap(struct pglist_data *pgdat,
3272 struct zone *zone, unsigned long zonesize)
3274 unsigned long usemapsize = usemap_size(zonesize);
3275 zone->pageblock_flags = NULL;
3277 zone->pageblock_flags = alloc_bootmem_node(pgdat, usemapsize);
3278 memset(zone->pageblock_flags, 0, usemapsize);
3282 static void inline setup_usemap(struct pglist_data *pgdat,
3283 struct zone *zone, unsigned long zonesize) {}
3284 #endif /* CONFIG_SPARSEMEM */
3286 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
3287 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
3288 static inline void __init set_pageblock_order(unsigned int order)
3290 /* Check that pageblock_nr_pages has not already been setup */
3291 if (pageblock_order)
3295 * Assume the largest contiguous order of interest is a huge page.
3296 * This value may be variable depending on boot parameters on IA64
3298 pageblock_order = order;
3300 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3302 /* Defined this way to avoid accidently referencing HUGETLB_PAGE_ORDER */
3303 #define set_pageblock_order(x) do {} while (0)
3305 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3308 * Set up the zone data structures:
3309 * - mark all pages reserved
3310 * - mark all memory queues empty
3311 * - clear the memory bitmaps
3313 static void __meminit free_area_init_core(struct pglist_data *pgdat,
3314 unsigned long *zones_size, unsigned long *zholes_size)
3317 int nid = pgdat->node_id;
3318 unsigned long zone_start_pfn = pgdat->node_start_pfn;
3321 pgdat_resize_init(pgdat);
3322 pgdat->nr_zones = 0;
3323 init_waitqueue_head(&pgdat->kswapd_wait);
3324 pgdat->kswapd_max_order = 0;
3326 for (j = 0; j < MAX_NR_ZONES; j++) {
3327 struct zone *zone = pgdat->node_zones + j;
3328 unsigned long size, realsize, memmap_pages;
3330 size = zone_spanned_pages_in_node(nid, j, zones_size);
3331 realsize = size - zone_absent_pages_in_node(nid, j,
3335 * Adjust realsize so that it accounts for how much memory
3336 * is used by this zone for memmap. This affects the watermark
3337 * and per-cpu initialisations
3339 memmap_pages = (size * sizeof(struct page)) >> PAGE_SHIFT;
3340 if (realsize >= memmap_pages) {
3341 realsize -= memmap_pages;
3343 " %s zone: %lu pages used for memmap\n",
3344 zone_names[j], memmap_pages);
3347 " %s zone: %lu pages exceeds realsize %lu\n",
3348 zone_names[j], memmap_pages, realsize);
3350 /* Account for reserved pages */
3351 if (j == 0 && realsize > dma_reserve) {
3352 realsize -= dma_reserve;
3353 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
3354 zone_names[0], dma_reserve);
3357 if (!is_highmem_idx(j))
3358 nr_kernel_pages += realsize;
3359 nr_all_pages += realsize;
3361 zone->spanned_pages = size;
3362 zone->present_pages = realsize;
3365 zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
3367 zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100;
3369 zone->name = zone_names[j];
3370 spin_lock_init(&zone->lock);
3371 spin_lock_init(&zone->lru_lock);
3372 zone_seqlock_init(zone);
3373 zone->zone_pgdat = pgdat;
3375 zone->prev_priority = DEF_PRIORITY;
3377 zone_pcp_init(zone);
3378 INIT_LIST_HEAD(&zone->active_list);
3379 INIT_LIST_HEAD(&zone->inactive_list);
3380 zone->nr_scan_active = 0;
3381 zone->nr_scan_inactive = 0;
3382 zap_zone_vm_stats(zone);
3383 atomic_set(&zone->reclaim_in_progress, 0);
3387 set_pageblock_order(HUGETLB_PAGE_ORDER);
3388 setup_usemap(pgdat, zone, size);
3389 ret = init_currently_empty_zone(zone, zone_start_pfn,
3390 size, MEMMAP_EARLY);
3392 zone_start_pfn += size;
3396 static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
3398 /* Skip empty nodes */
3399 if (!pgdat->node_spanned_pages)
3402 #ifdef CONFIG_FLAT_NODE_MEM_MAP
3403 /* ia64 gets its own node_mem_map, before this, without bootmem */
3404 if (!pgdat->node_mem_map) {
3405 unsigned long size, start, end;
3409 * The zone's endpoints aren't required to be MAX_ORDER
3410 * aligned but the node_mem_map endpoints must be in order
3411 * for the buddy allocator to function correctly.
3413 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
3414 end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
3415 end = ALIGN(end, MAX_ORDER_NR_PAGES);
3416 size = (end - start) * sizeof(struct page);
3417 map = alloc_remap(pgdat->node_id, size);
3419 map = alloc_bootmem_node(pgdat, size);
3420 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
3422 #ifndef CONFIG_NEED_MULTIPLE_NODES
3424 * With no DISCONTIG, the global mem_map is just set as node 0's
3426 if (pgdat == NODE_DATA(0)) {
3427 mem_map = NODE_DATA(0)->node_mem_map;
3428 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3429 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
3430 mem_map -= pgdat->node_start_pfn;
3431 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3434 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
3437 void __meminit free_area_init_node(int nid, struct pglist_data *pgdat,
3438 unsigned long *zones_size, unsigned long node_start_pfn,
3439 unsigned long *zholes_size)
3441 pgdat->node_id = nid;
3442 pgdat->node_start_pfn = node_start_pfn;
3443 calculate_node_totalpages(pgdat, zones_size, zholes_size);
3445 alloc_node_mem_map(pgdat);
3447 free_area_init_core(pgdat, zones_size, zholes_size);
3450 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3452 #if MAX_NUMNODES > 1
3454 * Figure out the number of possible node ids.
3456 static void __init setup_nr_node_ids(void)
3459 unsigned int highest = 0;
3461 for_each_node_mask(node, node_possible_map)
3463 nr_node_ids = highest + 1;
3466 static inline void setup_nr_node_ids(void)
3472 * add_active_range - Register a range of PFNs backed by physical memory
3473 * @nid: The node ID the range resides on
3474 * @start_pfn: The start PFN of the available physical memory
3475 * @end_pfn: The end PFN of the available physical memory
3477 * These ranges are stored in an early_node_map[] and later used by
3478 * free_area_init_nodes() to calculate zone sizes and holes. If the
3479 * range spans a memory hole, it is up to the architecture to ensure
3480 * the memory is not freed by the bootmem allocator. If possible
3481 * the range being registered will be merged with existing ranges.
3483 void __init add_active_range(unsigned int nid, unsigned long start_pfn,
3484 unsigned long end_pfn)
3488 printk(KERN_DEBUG "Entering add_active_range(%d, %lu, %lu) "
3489 "%d entries of %d used\n",
3490 nid, start_pfn, end_pfn,
3491 nr_nodemap_entries, MAX_ACTIVE_REGIONS);
3493 /* Merge with existing active regions if possible */
3494 for (i = 0; i < nr_nodemap_entries; i++) {
3495 if (early_node_map[i].nid != nid)
3498 /* Skip if an existing region covers this new one */
3499 if (start_pfn >= early_node_map[i].start_pfn &&
3500 end_pfn <= early_node_map[i].end_pfn)
3503 /* Merge forward if suitable */
3504 if (start_pfn <= early_node_map[i].end_pfn &&
3505 end_pfn > early_node_map[i].end_pfn) {
3506 early_node_map[i].end_pfn = end_pfn;
3510 /* Merge backward if suitable */
3511 if (start_pfn < early_node_map[i].end_pfn &&
3512 end_pfn >= early_node_map[i].start_pfn) {
3513 early_node_map[i].start_pfn = start_pfn;
3518 /* Check that early_node_map is large enough */
3519 if (i >= MAX_ACTIVE_REGIONS) {
3520 printk(KERN_CRIT "More than %d memory regions, truncating\n",
3521 MAX_ACTIVE_REGIONS);
3525 early_node_map[i].nid = nid;
3526 early_node_map[i].start_pfn = start_pfn;
3527 early_node_map[i].end_pfn = end_pfn;
3528 nr_nodemap_entries = i + 1;
3532 * shrink_active_range - Shrink an existing registered range of PFNs
3533 * @nid: The node id the range is on that should be shrunk
3534 * @old_end_pfn: The old end PFN of the range
3535 * @new_end_pfn: The new PFN of the range
3537 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
3538 * The map is kept at the end physical page range that has already been
3539 * registered with add_active_range(). This function allows an arch to shrink
3540 * an existing registered range.
3542 void __init shrink_active_range(unsigned int nid, unsigned long old_end_pfn,
3543 unsigned long new_end_pfn)
3547 /* Find the old active region end and shrink */
3548 for_each_active_range_index_in_nid(i, nid)
3549 if (early_node_map[i].end_pfn == old_end_pfn) {
3550 early_node_map[i].end_pfn = new_end_pfn;
3556 * remove_all_active_ranges - Remove all currently registered regions
3558 * During discovery, it may be found that a table like SRAT is invalid
3559 * and an alternative discovery method must be used. This function removes
3560 * all currently registered regions.
3562 void __init remove_all_active_ranges(void)
3564 memset(early_node_map, 0, sizeof(early_node_map));
3565 nr_nodemap_entries = 0;
3566 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
3567 memset(node_boundary_start_pfn, 0, sizeof(node_boundary_start_pfn));
3568 memset(node_boundary_end_pfn, 0, sizeof(node_boundary_end_pfn));
3569 #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
3572 /* Compare two active node_active_regions */
3573 static int __init cmp_node_active_region(const void *a, const void *b)
3575 struct node_active_region *arange = (struct node_active_region *)a;
3576 struct node_active_region *brange = (struct node_active_region *)b;
3578 /* Done this way to avoid overflows */
3579 if (arange->start_pfn > brange->start_pfn)
3581 if (arange->start_pfn < brange->start_pfn)
3587 /* sort the node_map by start_pfn */
3588 static void __init sort_node_map(void)
3590 sort(early_node_map, (size_t)nr_nodemap_entries,
3591 sizeof(struct node_active_region),
3592 cmp_node_active_region, NULL);
3595 /* Find the lowest pfn for a node */
3596 unsigned long __init find_min_pfn_for_node(unsigned long nid)
3599 unsigned long min_pfn = ULONG_MAX;
3601 /* Assuming a sorted map, the first range found has the starting pfn */
3602 for_each_active_range_index_in_nid(i, nid)
3603 min_pfn = min(min_pfn, early_node_map[i].start_pfn);
3605 if (min_pfn == ULONG_MAX) {
3607 "Could not find start_pfn for node %lu\n", nid);
3615 * find_min_pfn_with_active_regions - Find the minimum PFN registered
3617 * It returns the minimum PFN based on information provided via
3618 * add_active_range().
3620 unsigned long __init find_min_pfn_with_active_regions(void)
3622 return find_min_pfn_for_node(MAX_NUMNODES);
3626 * find_max_pfn_with_active_regions - Find the maximum PFN registered
3628 * It returns the maximum PFN based on information provided via
3629 * add_active_range().
3631 unsigned long __init find_max_pfn_with_active_regions(void)
3634 unsigned long max_pfn = 0;
3636 for (i = 0; i < nr_nodemap_entries; i++)
3637 max_pfn = max(max_pfn, early_node_map[i].end_pfn);
3643 * early_calculate_totalpages()
3644 * Sum pages in active regions for movable zone.
3645 * Populate N_HIGH_MEMORY for calculating usable_nodes.
3647 unsigned long __init early_calculate_totalpages(void)
3650 unsigned long totalpages = 0;
3652 for (i = 0; i < nr_nodemap_entries; i++) {
3653 unsigned long pages = early_node_map[i].end_pfn -
3654 early_node_map[i].start_pfn;
3655 totalpages += pages;
3657 node_set_state(early_node_map[i].nid, N_HIGH_MEMORY);
3663 * Find the PFN the Movable zone begins in each node. Kernel memory
3664 * is spread evenly between nodes as long as the nodes have enough
3665 * memory. When they don't, some nodes will have more kernelcore than
3668 void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn)
3671 unsigned long usable_startpfn;
3672 unsigned long kernelcore_node, kernelcore_remaining;
3673 unsigned long totalpages = early_calculate_totalpages();
3674 int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]);
3677 * If movablecore was specified, calculate what size of
3678 * kernelcore that corresponds so that memory usable for
3679 * any allocation type is evenly spread. If both kernelcore
3680 * and movablecore are specified, then the value of kernelcore
3681 * will be used for required_kernelcore if it's greater than
3682 * what movablecore would have allowed.
3684 if (required_movablecore) {
3685 unsigned long corepages;
3688 * Round-up so that ZONE_MOVABLE is at least as large as what
3689 * was requested by the user
3691 required_movablecore =
3692 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
3693 corepages = totalpages - required_movablecore;
3695 required_kernelcore = max(required_kernelcore, corepages);
3698 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
3699 if (!required_kernelcore)
3702 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
3703 find_usable_zone_for_movable();
3704 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
3707 /* Spread kernelcore memory as evenly as possible throughout nodes */
3708 kernelcore_node = required_kernelcore / usable_nodes;
3709 for_each_node_state(nid, N_HIGH_MEMORY) {
3711 * Recalculate kernelcore_node if the division per node
3712 * now exceeds what is necessary to satisfy the requested
3713 * amount of memory for the kernel
3715 if (required_kernelcore < kernelcore_node)
3716 kernelcore_node = required_kernelcore / usable_nodes;
3719 * As the map is walked, we track how much memory is usable
3720 * by the kernel using kernelcore_remaining. When it is
3721 * 0, the rest of the node is usable by ZONE_MOVABLE
3723 kernelcore_remaining = kernelcore_node;
3725 /* Go through each range of PFNs within this node */
3726 for_each_active_range_index_in_nid(i, nid) {
3727 unsigned long start_pfn, end_pfn;
3728 unsigned long size_pages;
3730 start_pfn = max(early_node_map[i].start_pfn,
3731 zone_movable_pfn[nid]);
3732 end_pfn = early_node_map[i].end_pfn;
3733 if (start_pfn >= end_pfn)
3736 /* Account for what is only usable for kernelcore */
3737 if (start_pfn < usable_startpfn) {
3738 unsigned long kernel_pages;
3739 kernel_pages = min(end_pfn, usable_startpfn)
3742 kernelcore_remaining -= min(kernel_pages,
3743 kernelcore_remaining);
3744 required_kernelcore -= min(kernel_pages,
3745 required_kernelcore);
3747 /* Continue if range is now fully accounted */
3748 if (end_pfn <= usable_startpfn) {
3751 * Push zone_movable_pfn to the end so
3752 * that if we have to rebalance
3753 * kernelcore across nodes, we will
3754 * not double account here
3756 zone_movable_pfn[nid] = end_pfn;
3759 start_pfn = usable_startpfn;
3763 * The usable PFN range for ZONE_MOVABLE is from
3764 * start_pfn->end_pfn. Calculate size_pages as the
3765 * number of pages used as kernelcore
3767 size_pages = end_pfn - start_pfn;
3768 if (size_pages > kernelcore_remaining)
3769 size_pages = kernelcore_remaining;
3770 zone_movable_pfn[nid] = start_pfn + size_pages;
3773 * Some kernelcore has been met, update counts and
3774 * break if the kernelcore for this node has been
3777 required_kernelcore -= min(required_kernelcore,
3779 kernelcore_remaining -= size_pages;
3780 if (!kernelcore_remaining)
3786 * If there is still required_kernelcore, we do another pass with one
3787 * less node in the count. This will push zone_movable_pfn[nid] further
3788 * along on the nodes that still have memory until kernelcore is
3792 if (usable_nodes && required_kernelcore > usable_nodes)
3795 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
3796 for (nid = 0; nid < MAX_NUMNODES; nid++)
3797 zone_movable_pfn[nid] =
3798 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
3801 /* Any regular memory on that node ? */
3802 static void check_for_regular_memory(pg_data_t *pgdat)
3804 #ifdef CONFIG_HIGHMEM
3805 enum zone_type zone_type;
3807 for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) {
3808 struct zone *zone = &pgdat->node_zones[zone_type];
3809 if (zone->present_pages)
3810 node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY);
3816 * free_area_init_nodes - Initialise all pg_data_t and zone data
3817 * @max_zone_pfn: an array of max PFNs for each zone
3819 * This will call free_area_init_node() for each active node in the system.
3820 * Using the page ranges provided by add_active_range(), the size of each
3821 * zone in each node and their holes is calculated. If the maximum PFN
3822 * between two adjacent zones match, it is assumed that the zone is empty.
3823 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
3824 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
3825 * starts where the previous one ended. For example, ZONE_DMA32 starts
3826 * at arch_max_dma_pfn.
3828 void __init free_area_init_nodes(unsigned long *max_zone_pfn)
3833 /* Sort early_node_map as initialisation assumes it is sorted */
3836 /* Record where the zone boundaries are */
3837 memset(arch_zone_lowest_possible_pfn, 0,
3838 sizeof(arch_zone_lowest_possible_pfn));
3839 memset(arch_zone_highest_possible_pfn, 0,
3840 sizeof(arch_zone_highest_possible_pfn));
3841 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
3842 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
3843 for (i = 1; i < MAX_NR_ZONES; i++) {
3844 if (i == ZONE_MOVABLE)
3846 arch_zone_lowest_possible_pfn[i] =
3847 arch_zone_highest_possible_pfn[i-1];
3848 arch_zone_highest_possible_pfn[i] =
3849 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
3851 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
3852 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
3854 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
3855 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
3856 find_zone_movable_pfns_for_nodes(zone_movable_pfn);
3858 /* Print out the zone ranges */
3859 printk("Zone PFN ranges:\n");
3860 for (i = 0; i < MAX_NR_ZONES; i++) {
3861 if (i == ZONE_MOVABLE)
3863 printk(" %-8s %8lu -> %8lu\n",
3865 arch_zone_lowest_possible_pfn[i],
3866 arch_zone_highest_possible_pfn[i]);
3869 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
3870 printk("Movable zone start PFN for each node\n");
3871 for (i = 0; i < MAX_NUMNODES; i++) {
3872 if (zone_movable_pfn[i])
3873 printk(" Node %d: %lu\n", i, zone_movable_pfn[i]);
3876 /* Print out the early_node_map[] */
3877 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries);
3878 for (i = 0; i < nr_nodemap_entries; i++)
3879 printk(" %3d: %8lu -> %8lu\n", early_node_map[i].nid,
3880 early_node_map[i].start_pfn,
3881 early_node_map[i].end_pfn);
3883 /* Initialise every node */
3884 setup_nr_node_ids();
3885 for_each_online_node(nid) {
3886 pg_data_t *pgdat = NODE_DATA(nid);
3887 free_area_init_node(nid, pgdat, NULL,
3888 find_min_pfn_for_node(nid), NULL);
3890 /* Any memory on that node */
3891 if (pgdat->node_present_pages)
3892 node_set_state(nid, N_HIGH_MEMORY);
3893 check_for_regular_memory(pgdat);
3897 static int __init cmdline_parse_core(char *p, unsigned long *core)
3899 unsigned long long coremem;
3903 coremem = memparse(p, &p);
3904 *core = coremem >> PAGE_SHIFT;
3906 /* Paranoid check that UL is enough for the coremem value */
3907 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
3913 * kernelcore=size sets the amount of memory for use for allocations that
3914 * cannot be reclaimed or migrated.
3916 static int __init cmdline_parse_kernelcore(char *p)
3918 return cmdline_parse_core(p, &required_kernelcore);
3922 * movablecore=size sets the amount of memory for use for allocations that
3923 * can be reclaimed or migrated.
3925 static int __init cmdline_parse_movablecore(char *p)
3927 return cmdline_parse_core(p, &required_movablecore);
3930 early_param("kernelcore", cmdline_parse_kernelcore);
3931 early_param("movablecore", cmdline_parse_movablecore);
3933 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3936 * set_dma_reserve - set the specified number of pages reserved in the first zone
3937 * @new_dma_reserve: The number of pages to mark reserved
3939 * The per-cpu batchsize and zone watermarks are determined by present_pages.
3940 * In the DMA zone, a significant percentage may be consumed by kernel image
3941 * and other unfreeable allocations which can skew the watermarks badly. This
3942 * function may optionally be used to account for unfreeable pages in the
3943 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
3944 * smaller per-cpu batchsize.
3946 void __init set_dma_reserve(unsigned long new_dma_reserve)
3948 dma_reserve = new_dma_reserve;
3951 #ifndef CONFIG_NEED_MULTIPLE_NODES
3952 static bootmem_data_t contig_bootmem_data;
3953 struct pglist_data contig_page_data = { .bdata = &contig_bootmem_data };
3955 EXPORT_SYMBOL(contig_page_data);
3958 void __init free_area_init(unsigned long *zones_size)
3960 free_area_init_node(0, NODE_DATA(0), zones_size,
3961 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
3964 static int page_alloc_cpu_notify(struct notifier_block *self,
3965 unsigned long action, void *hcpu)
3967 int cpu = (unsigned long)hcpu;
3969 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
3970 local_irq_disable();
3972 vm_events_fold_cpu(cpu);
3974 refresh_cpu_vm_stats(cpu);
3979 void __init page_alloc_init(void)
3981 hotcpu_notifier(page_alloc_cpu_notify, 0);
3985 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
3986 * or min_free_kbytes changes.
3988 static void calculate_totalreserve_pages(void)
3990 struct pglist_data *pgdat;
3991 unsigned long reserve_pages = 0;
3992 enum zone_type i, j;
3994 for_each_online_pgdat(pgdat) {
3995 for (i = 0; i < MAX_NR_ZONES; i++) {
3996 struct zone *zone = pgdat->node_zones + i;
3997 unsigned long max = 0;
3999 /* Find valid and maximum lowmem_reserve in the zone */
4000 for (j = i; j < MAX_NR_ZONES; j++) {
4001 if (zone->lowmem_reserve[j] > max)
4002 max = zone->lowmem_reserve[j];
4005 /* we treat pages_high as reserved pages. */
4006 max += zone->pages_high;
4008 if (max > zone->present_pages)
4009 max = zone->present_pages;
4010 reserve_pages += max;
4013 totalreserve_pages = reserve_pages;
4017 * setup_per_zone_lowmem_reserve - called whenever
4018 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
4019 * has a correct pages reserved value, so an adequate number of
4020 * pages are left in the zone after a successful __alloc_pages().
4022 static void setup_per_zone_lowmem_reserve(void)
4024 struct pglist_data *pgdat;
4025 enum zone_type j, idx;
4027 for_each_online_pgdat(pgdat) {
4028 for (j = 0; j < MAX_NR_ZONES; j++) {
4029 struct zone *zone = pgdat->node_zones + j;
4030 unsigned long present_pages = zone->present_pages;
4032 zone->lowmem_reserve[j] = 0;
4036 struct zone *lower_zone;
4040 if (sysctl_lowmem_reserve_ratio[idx] < 1)
4041 sysctl_lowmem_reserve_ratio[idx] = 1;
4043 lower_zone = pgdat->node_zones + idx;
4044 lower_zone->lowmem_reserve[j] = present_pages /
4045 sysctl_lowmem_reserve_ratio[idx];
4046 present_pages += lower_zone->present_pages;
4051 /* update totalreserve_pages */
4052 calculate_totalreserve_pages();
4056 * setup_per_zone_pages_min - called when min_free_kbytes changes.
4058 * Ensures that the pages_{min,low,high} values for each zone are set correctly
4059 * with respect to min_free_kbytes.
4061 void setup_per_zone_pages_min(void)
4063 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
4064 unsigned long lowmem_pages = 0;
4066 unsigned long flags;
4068 /* Calculate total number of !ZONE_HIGHMEM pages */
4069 for_each_zone(zone) {
4070 if (!is_highmem(zone))
4071 lowmem_pages += zone->present_pages;
4074 for_each_zone(zone) {
4077 spin_lock_irqsave(&zone->lru_lock, flags);
4078 tmp = (u64)pages_min * zone->present_pages;
4079 do_div(tmp, lowmem_pages);
4080 if (is_highmem(zone)) {
4082 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
4083 * need highmem pages, so cap pages_min to a small
4086 * The (pages_high-pages_low) and (pages_low-pages_min)
4087 * deltas controls asynch page reclaim, and so should
4088 * not be capped for highmem.
4092 min_pages = zone->present_pages / 1024;
4093 if (min_pages < SWAP_CLUSTER_MAX)
4094 min_pages = SWAP_CLUSTER_MAX;
4095 if (min_pages > 128)
4097 zone->pages_min = min_pages;
4100 * If it's a lowmem zone, reserve a number of pages
4101 * proportionate to the zone's size.
4103 zone->pages_min = tmp;
4106 zone->pages_low = zone->pages_min + (tmp >> 2);
4107 zone->pages_high = zone->pages_min + (tmp >> 1);
4108 setup_zone_migrate_reserve(zone);
4109 spin_unlock_irqrestore(&zone->lru_lock, flags);
4112 /* update totalreserve_pages */
4113 calculate_totalreserve_pages();
4117 * Initialise min_free_kbytes.
4119 * For small machines we want it small (128k min). For large machines
4120 * we want it large (64MB max). But it is not linear, because network
4121 * bandwidth does not increase linearly with machine size. We use
4123 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
4124 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
4140 static int __init init_per_zone_pages_min(void)
4142 unsigned long lowmem_kbytes;
4144 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
4146 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
4147 if (min_free_kbytes < 128)
4148 min_free_kbytes = 128;
4149 if (min_free_kbytes > 65536)
4150 min_free_kbytes = 65536;
4151 setup_per_zone_pages_min();
4152 setup_per_zone_lowmem_reserve();
4155 module_init(init_per_zone_pages_min)
4158 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
4159 * that we can call two helper functions whenever min_free_kbytes
4162 int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
4163 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4165 proc_dointvec(table, write, file, buffer, length, ppos);
4167 setup_per_zone_pages_min();
4172 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
4173 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4178 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4183 zone->min_unmapped_pages = (zone->present_pages *
4184 sysctl_min_unmapped_ratio) / 100;
4188 int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
4189 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4194 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4199 zone->min_slab_pages = (zone->present_pages *
4200 sysctl_min_slab_ratio) / 100;
4206 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
4207 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
4208 * whenever sysctl_lowmem_reserve_ratio changes.
4210 * The reserve ratio obviously has absolutely no relation with the
4211 * pages_min watermarks. The lowmem reserve ratio can only make sense
4212 * if in function of the boot time zone sizes.
4214 int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
4215 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4217 proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4218 setup_per_zone_lowmem_reserve();
4223 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
4224 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
4225 * can have before it gets flushed back to buddy allocator.
4228 int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
4229 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4235 ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4236 if (!write || (ret == -EINVAL))
4238 for_each_zone(zone) {
4239 for_each_online_cpu(cpu) {
4241 high = zone->present_pages / percpu_pagelist_fraction;
4242 setup_pagelist_highmark(zone_pcp(zone, cpu), high);
4248 int hashdist = HASHDIST_DEFAULT;
4251 static int __init set_hashdist(char *str)
4255 hashdist = simple_strtoul(str, &str, 0);
4258 __setup("hashdist=", set_hashdist);
4262 * allocate a large system hash table from bootmem
4263 * - it is assumed that the hash table must contain an exact power-of-2
4264 * quantity of entries
4265 * - limit is the number of hash buckets, not the total allocation size
4267 void *__init alloc_large_system_hash(const char *tablename,
4268 unsigned long bucketsize,
4269 unsigned long numentries,
4272 unsigned int *_hash_shift,
4273 unsigned int *_hash_mask,
4274 unsigned long limit)
4276 unsigned long long max = limit;
4277 unsigned long log2qty, size;
4280 /* allow the kernel cmdline to have a say */
4282 /* round applicable memory size up to nearest megabyte */
4283 numentries = nr_kernel_pages;
4284 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
4285 numentries >>= 20 - PAGE_SHIFT;
4286 numentries <<= 20 - PAGE_SHIFT;
4288 /* limit to 1 bucket per 2^scale bytes of low memory */
4289 if (scale > PAGE_SHIFT)
4290 numentries >>= (scale - PAGE_SHIFT);
4292 numentries <<= (PAGE_SHIFT - scale);
4294 /* Make sure we've got at least a 0-order allocation.. */
4295 if (unlikely((numentries * bucketsize) < PAGE_SIZE))
4296 numentries = PAGE_SIZE / bucketsize;
4298 numentries = roundup_pow_of_two(numentries);
4300 /* limit allocation size to 1/16 total memory by default */
4302 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
4303 do_div(max, bucketsize);
4306 if (numentries > max)
4309 log2qty = ilog2(numentries);
4312 size = bucketsize << log2qty;
4313 if (flags & HASH_EARLY)
4314 table = alloc_bootmem(size);
4316 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
4318 unsigned long order;
4319 for (order = 0; ((1UL << order) << PAGE_SHIFT) < size; order++)
4321 table = (void*) __get_free_pages(GFP_ATOMIC, order);
4323 * If bucketsize is not a power-of-two, we may free
4324 * some pages at the end of hash table.
4327 unsigned long alloc_end = (unsigned long)table +
4328 (PAGE_SIZE << order);
4329 unsigned long used = (unsigned long)table +
4331 split_page(virt_to_page(table), order);
4332 while (used < alloc_end) {
4338 } while (!table && size > PAGE_SIZE && --log2qty);
4341 panic("Failed to allocate %s hash table\n", tablename);
4343 printk(KERN_INFO "%s hash table entries: %d (order: %d, %lu bytes)\n",
4346 ilog2(size) - PAGE_SHIFT,
4350 *_hash_shift = log2qty;
4352 *_hash_mask = (1 << log2qty) - 1;
4357 #ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE
4358 struct page *pfn_to_page(unsigned long pfn)
4360 return __pfn_to_page(pfn);
4362 unsigned long page_to_pfn(struct page *page)
4364 return __page_to_pfn(page);
4366 EXPORT_SYMBOL(pfn_to_page);
4367 EXPORT_SYMBOL(page_to_pfn);
4368 #endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */
4370 /* Return a pointer to the bitmap storing bits affecting a block of pages */
4371 static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
4374 #ifdef CONFIG_SPARSEMEM
4375 return __pfn_to_section(pfn)->pageblock_flags;
4377 return zone->pageblock_flags;
4378 #endif /* CONFIG_SPARSEMEM */
4381 static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
4383 #ifdef CONFIG_SPARSEMEM
4384 pfn &= (PAGES_PER_SECTION-1);
4385 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
4387 pfn = pfn - zone->zone_start_pfn;
4388 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
4389 #endif /* CONFIG_SPARSEMEM */
4393 * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
4394 * @page: The page within the block of interest
4395 * @start_bitidx: The first bit of interest to retrieve
4396 * @end_bitidx: The last bit of interest
4397 * returns pageblock_bits flags
4399 unsigned long get_pageblock_flags_group(struct page *page,
4400 int start_bitidx, int end_bitidx)
4403 unsigned long *bitmap;
4404 unsigned long pfn, bitidx;
4405 unsigned long flags = 0;
4406 unsigned long value = 1;
4408 zone = page_zone(page);
4409 pfn = page_to_pfn(page);
4410 bitmap = get_pageblock_bitmap(zone, pfn);
4411 bitidx = pfn_to_bitidx(zone, pfn);
4413 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
4414 if (test_bit(bitidx + start_bitidx, bitmap))
4421 * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
4422 * @page: The page within the block of interest
4423 * @start_bitidx: The first bit of interest
4424 * @end_bitidx: The last bit of interest
4425 * @flags: The flags to set
4427 void set_pageblock_flags_group(struct page *page, unsigned long flags,
4428 int start_bitidx, int end_bitidx)
4431 unsigned long *bitmap;
4432 unsigned long pfn, bitidx;
4433 unsigned long value = 1;
4435 zone = page_zone(page);
4436 pfn = page_to_pfn(page);
4437 bitmap = get_pageblock_bitmap(zone, pfn);
4438 bitidx = pfn_to_bitidx(zone, pfn);
4440 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
4442 __set_bit(bitidx + start_bitidx, bitmap);
4444 __clear_bit(bitidx + start_bitidx, bitmap);