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 static void __free_pages_ok(struct page *page, unsigned int order);
73 * results with 256, 32 in the lowmem_reserve sysctl:
74 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
75 * 1G machine -> (16M dma, 784M normal, 224M high)
76 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
77 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
78 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
80 * TBD: should special case ZONE_DMA32 machines here - in those we normally
81 * don't need any ZONE_NORMAL reservation
83 int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
84 #ifdef CONFIG_ZONE_DMA
87 #ifdef CONFIG_ZONE_DMA32
96 EXPORT_SYMBOL(totalram_pages);
98 static char * const zone_names[MAX_NR_ZONES] = {
99 #ifdef CONFIG_ZONE_DMA
102 #ifdef CONFIG_ZONE_DMA32
106 #ifdef CONFIG_HIGHMEM
112 int min_free_kbytes = 1024;
114 unsigned long __meminitdata nr_kernel_pages;
115 unsigned long __meminitdata nr_all_pages;
116 static unsigned long __meminitdata dma_reserve;
118 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
120 * MAX_ACTIVE_REGIONS determines the maxmimum number of distinct
121 * ranges of memory (RAM) that may be registered with add_active_range().
122 * Ranges passed to add_active_range() will be merged if possible
123 * so the number of times add_active_range() can be called is
124 * related to the number of nodes and the number of holes
126 #ifdef CONFIG_MAX_ACTIVE_REGIONS
127 /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
128 #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
130 #if MAX_NUMNODES >= 32
131 /* If there can be many nodes, allow up to 50 holes per node */
132 #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
134 /* By default, allow up to 256 distinct regions */
135 #define MAX_ACTIVE_REGIONS 256
139 static struct node_active_region __meminitdata early_node_map[MAX_ACTIVE_REGIONS];
140 static int __meminitdata nr_nodemap_entries;
141 static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
142 static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
143 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
144 static unsigned long __meminitdata node_boundary_start_pfn[MAX_NUMNODES];
145 static unsigned long __meminitdata node_boundary_end_pfn[MAX_NUMNODES];
146 #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
147 unsigned long __initdata required_kernelcore;
148 unsigned long __initdata required_movablecore;
149 unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
151 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
153 EXPORT_SYMBOL(movable_zone);
154 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
157 int nr_node_ids __read_mostly = MAX_NUMNODES;
158 EXPORT_SYMBOL(nr_node_ids);
161 #ifdef CONFIG_PAGE_GROUP_BY_MOBILITY
162 static inline int get_pageblock_migratetype(struct page *page)
164 return get_pageblock_flags_group(page, PB_migrate, PB_migrate_end);
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 static inline int allocflags_to_migratetype(gfp_t gfp_flags, int order)
175 WARN_ON((gfp_flags & GFP_MOVABLE_MASK) == GFP_MOVABLE_MASK);
177 /* Cluster high-order atomic allocations together */
178 if (unlikely(order > 0) &&
179 (!(gfp_flags & __GFP_WAIT) || in_interrupt()))
180 return MIGRATE_HIGHATOMIC;
182 /* Cluster based on mobility */
183 return (((gfp_flags & __GFP_MOVABLE) != 0) << 1) |
184 ((gfp_flags & __GFP_RECLAIMABLE) != 0);
188 static inline int get_pageblock_migratetype(struct page *page)
190 return MIGRATE_UNMOVABLE;
193 static void set_pageblock_migratetype(struct page *page, int migratetype)
197 static inline int allocflags_to_migratetype(gfp_t gfp_flags, int order)
199 return MIGRATE_UNMOVABLE;
201 #endif /* CONFIG_PAGE_GROUP_BY_MOBILITY */
203 #ifdef CONFIG_DEBUG_VM
204 static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
208 unsigned long pfn = page_to_pfn(page);
211 seq = zone_span_seqbegin(zone);
212 if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
214 else if (pfn < zone->zone_start_pfn)
216 } while (zone_span_seqretry(zone, seq));
221 static int page_is_consistent(struct zone *zone, struct page *page)
223 if (!pfn_valid_within(page_to_pfn(page)))
225 if (zone != page_zone(page))
231 * Temporary debugging check for pages not lying within a given zone.
233 static int bad_range(struct zone *zone, struct page *page)
235 if (page_outside_zone_boundaries(zone, page))
237 if (!page_is_consistent(zone, page))
243 static inline int bad_range(struct zone *zone, struct page *page)
249 static void bad_page(struct page *page)
251 printk(KERN_EMERG "Bad page state in process '%s'\n"
252 KERN_EMERG "page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n"
253 KERN_EMERG "Trying to fix it up, but a reboot is needed\n"
254 KERN_EMERG "Backtrace:\n",
255 current->comm, page, (int)(2*sizeof(unsigned long)),
256 (unsigned long)page->flags, page->mapping,
257 page_mapcount(page), page_count(page));
259 page->flags &= ~(1 << PG_lru |
269 set_page_count(page, 0);
270 reset_page_mapcount(page);
271 page->mapping = NULL;
272 add_taint(TAINT_BAD_PAGE);
276 * Higher-order pages are called "compound pages". They are structured thusly:
278 * The first PAGE_SIZE page is called the "head page".
280 * The remaining PAGE_SIZE pages are called "tail pages".
282 * All pages have PG_compound set. All pages have their ->private pointing at
283 * the head page (even the head page has this).
285 * The first tail page's ->lru.next holds the address of the compound page's
286 * put_page() function. Its ->lru.prev holds the order of allocation.
287 * This usage means that zero-order pages may not be compound.
290 static void free_compound_page(struct page *page)
292 __free_pages_ok(page, compound_order(page));
295 static void prep_compound_page(struct page *page, unsigned long order)
298 int nr_pages = 1 << order;
300 set_compound_page_dtor(page, free_compound_page);
301 set_compound_order(page, order);
303 for (i = 1; i < nr_pages; i++) {
304 struct page *p = page + i;
307 p->first_page = page;
311 static void destroy_compound_page(struct page *page, unsigned long order)
314 int nr_pages = 1 << order;
316 if (unlikely(compound_order(page) != order))
319 if (unlikely(!PageHead(page)))
321 __ClearPageHead(page);
322 for (i = 1; i < nr_pages; i++) {
323 struct page *p = page + i;
325 if (unlikely(!PageTail(p) |
326 (p->first_page != page)))
332 static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
336 VM_BUG_ON((gfp_flags & (__GFP_WAIT | __GFP_HIGHMEM)) == __GFP_HIGHMEM);
338 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
339 * and __GFP_HIGHMEM from hard or soft interrupt context.
341 VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
342 for (i = 0; i < (1 << order); i++)
343 clear_highpage(page + i);
347 * function for dealing with page's order in buddy system.
348 * zone->lock is already acquired when we use these.
349 * So, we don't need atomic page->flags operations here.
351 static inline unsigned long page_order(struct page *page)
353 return page_private(page);
356 static inline void set_page_order(struct page *page, int order)
358 set_page_private(page, order);
359 __SetPageBuddy(page);
362 static inline void rmv_page_order(struct page *page)
364 __ClearPageBuddy(page);
365 set_page_private(page, 0);
369 * Locate the struct page for both the matching buddy in our
370 * pair (buddy1) and the combined O(n+1) page they form (page).
372 * 1) Any buddy B1 will have an order O twin B2 which satisfies
373 * the following equation:
375 * For example, if the starting buddy (buddy2) is #8 its order
377 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
379 * 2) Any buddy B will have an order O+1 parent P which
380 * satisfies the following equation:
383 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
385 static inline struct page *
386 __page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order)
388 unsigned long buddy_idx = page_idx ^ (1 << order);
390 return page + (buddy_idx - page_idx);
393 static inline unsigned long
394 __find_combined_index(unsigned long page_idx, unsigned int order)
396 return (page_idx & ~(1 << order));
400 * This function checks whether a page is free && is the buddy
401 * we can do coalesce a page and its buddy if
402 * (a) the buddy is not in a hole &&
403 * (b) the buddy is in the buddy system &&
404 * (c) a page and its buddy have the same order &&
405 * (d) a page and its buddy are in the same zone.
407 * For recording whether a page is in the buddy system, we use PG_buddy.
408 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
410 * For recording page's order, we use page_private(page).
412 static inline int page_is_buddy(struct page *page, struct page *buddy,
415 if (!pfn_valid_within(page_to_pfn(buddy)))
418 if (page_zone_id(page) != page_zone_id(buddy))
421 if (PageBuddy(buddy) && page_order(buddy) == order) {
422 BUG_ON(page_count(buddy) != 0);
429 * Freeing function for a buddy system allocator.
431 * The concept of a buddy system is to maintain direct-mapped table
432 * (containing bit values) for memory blocks of various "orders".
433 * The bottom level table contains the map for the smallest allocatable
434 * units of memory (here, pages), and each level above it describes
435 * pairs of units from the levels below, hence, "buddies".
436 * At a high level, all that happens here is marking the table entry
437 * at the bottom level available, and propagating the changes upward
438 * as necessary, plus some accounting needed to play nicely with other
439 * parts of the VM system.
440 * At each level, we keep a list of pages, which are heads of continuous
441 * free pages of length of (1 << order) and marked with PG_buddy. Page's
442 * order is recorded in page_private(page) field.
443 * So when we are allocating or freeing one, we can derive the state of the
444 * other. That is, if we allocate a small block, and both were
445 * free, the remainder of the region must be split into blocks.
446 * If a block is freed, and its buddy is also free, then this
447 * triggers coalescing into a block of larger size.
452 static inline void __free_one_page(struct page *page,
453 struct zone *zone, unsigned int order)
455 unsigned long page_idx;
456 int order_size = 1 << order;
457 int migratetype = get_pageblock_migratetype(page);
459 if (unlikely(PageCompound(page)))
460 destroy_compound_page(page, order);
462 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
464 VM_BUG_ON(page_idx & (order_size - 1));
465 VM_BUG_ON(bad_range(zone, page));
467 __mod_zone_page_state(zone, NR_FREE_PAGES, order_size);
468 while (order < MAX_ORDER-1) {
469 unsigned long combined_idx;
472 buddy = __page_find_buddy(page, page_idx, order);
473 if (!page_is_buddy(page, buddy, order))
474 break; /* Move the buddy up one level. */
476 list_del(&buddy->lru);
477 zone->free_area[order].nr_free--;
478 rmv_page_order(buddy);
479 combined_idx = __find_combined_index(page_idx, order);
480 page = page + (combined_idx - page_idx);
481 page_idx = combined_idx;
484 set_page_order(page, order);
486 &zone->free_area[order].free_list[migratetype]);
487 zone->free_area[order].nr_free++;
490 static inline int free_pages_check(struct page *page)
492 if (unlikely(page_mapcount(page) |
493 (page->mapping != NULL) |
494 (page_count(page) != 0) |
507 __ClearPageDirty(page);
509 * For now, we report if PG_reserved was found set, but do not
510 * clear it, and do not free the page. But we shall soon need
511 * to do more, for when the ZERO_PAGE count wraps negative.
513 return PageReserved(page);
517 * Frees a list of pages.
518 * Assumes all pages on list are in same zone, and of same order.
519 * count is the number of pages to free.
521 * If the zone was previously in an "all pages pinned" state then look to
522 * see if this freeing clears that state.
524 * And clear the zone's pages_scanned counter, to hold off the "all pages are
525 * pinned" detection logic.
527 static void free_pages_bulk(struct zone *zone, int count,
528 struct list_head *list, int order)
530 spin_lock(&zone->lock);
531 zone->all_unreclaimable = 0;
532 zone->pages_scanned = 0;
536 VM_BUG_ON(list_empty(list));
537 page = list_entry(list->prev, struct page, lru);
538 /* have to delete it as __free_one_page list manipulates */
539 list_del(&page->lru);
540 __free_one_page(page, zone, order);
542 spin_unlock(&zone->lock);
545 static void free_one_page(struct zone *zone, struct page *page, int order)
547 spin_lock(&zone->lock);
548 zone->all_unreclaimable = 0;
549 zone->pages_scanned = 0;
550 __free_one_page(page, zone, order);
551 spin_unlock(&zone->lock);
554 static void __free_pages_ok(struct page *page, unsigned int order)
560 for (i = 0 ; i < (1 << order) ; ++i)
561 reserved += free_pages_check(page + i);
565 if (!PageHighMem(page))
566 debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order);
567 arch_free_page(page, order);
568 kernel_map_pages(page, 1 << order, 0);
570 local_irq_save(flags);
571 __count_vm_events(PGFREE, 1 << order);
572 free_one_page(page_zone(page), page, order);
573 local_irq_restore(flags);
577 * permit the bootmem allocator to evade page validation on high-order frees
579 void fastcall __init __free_pages_bootmem(struct page *page, unsigned int order)
582 __ClearPageReserved(page);
583 set_page_count(page, 0);
584 set_page_refcounted(page);
590 for (loop = 0; loop < BITS_PER_LONG; loop++) {
591 struct page *p = &page[loop];
593 if (loop + 1 < BITS_PER_LONG)
595 __ClearPageReserved(p);
596 set_page_count(p, 0);
599 set_page_refcounted(page);
600 __free_pages(page, order);
606 * The order of subdivision here is critical for the IO subsystem.
607 * Please do not alter this order without good reasons and regression
608 * testing. Specifically, as large blocks of memory are subdivided,
609 * the order in which smaller blocks are delivered depends on the order
610 * they're subdivided in this function. This is the primary factor
611 * influencing the order in which pages are delivered to the IO
612 * subsystem according to empirical testing, and this is also justified
613 * by considering the behavior of a buddy system containing a single
614 * large block of memory acted on by a series of small allocations.
615 * This behavior is a critical factor in sglist merging's success.
619 static inline void expand(struct zone *zone, struct page *page,
620 int low, int high, struct free_area *area,
623 unsigned long size = 1 << high;
629 VM_BUG_ON(bad_range(zone, &page[size]));
630 list_add(&page[size].lru, &area->free_list[migratetype]);
632 set_page_order(&page[size], high);
637 * This page is about to be returned from the page allocator
639 static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
641 if (unlikely(page_mapcount(page) |
642 (page->mapping != NULL) |
643 (page_count(page) != 0) |
658 * For now, we report if PG_reserved was found set, but do not
659 * clear it, and do not allocate the page: as a safety net.
661 if (PageReserved(page))
664 page->flags &= ~(1 << PG_uptodate | 1 << PG_error | 1 << PG_readahead |
665 1 << PG_referenced | 1 << PG_arch_1 |
666 1 << PG_owner_priv_1 | 1 << PG_mappedtodisk);
667 set_page_private(page, 0);
668 set_page_refcounted(page);
670 arch_alloc_page(page, order);
671 kernel_map_pages(page, 1 << order, 1);
673 if (gfp_flags & __GFP_ZERO)
674 prep_zero_page(page, order, gfp_flags);
676 if (order && (gfp_flags & __GFP_COMP))
677 prep_compound_page(page, order);
682 #ifdef CONFIG_PAGE_GROUP_BY_MOBILITY
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_HIGHATOMIC },
689 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_HIGHATOMIC },
690 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE,MIGRATE_HIGHATOMIC },
691 [MIGRATE_HIGHATOMIC] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE,MIGRATE_MOVABLE},
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 in a MAX_ORDER_NR_PAGES
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,
705 int blocks_moved = 0;
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 * CONFIG_PAGE_GROUP_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]);
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 & ~(MAX_ORDER_NR_PAGES-1);
747 start_page = pfn_to_page(start_pfn);
748 end_page = start_page + MAX_ORDER_NR_PAGES - 1;
749 end_pfn = start_pfn + MAX_ORDER_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 /* Remove an element from the buddy allocator from the fallback list */
761 static struct page *__rmqueue_fallback(struct zone *zone, int order,
762 int start_migratetype)
764 struct free_area * area;
768 int nonatomic_fallback_atomic = 0;
771 /* Find the largest possible block of pages in the other list */
772 for (current_order = MAX_ORDER-1; current_order >= order;
774 for (i = 0; i < MIGRATE_TYPES - 1; i++) {
775 migratetype = fallbacks[start_migratetype][i];
778 * Make it hard to fallback to blocks used for
779 * high-order atomic allocations
781 if (migratetype == MIGRATE_HIGHATOMIC &&
782 start_migratetype != MIGRATE_UNMOVABLE &&
783 !nonatomic_fallback_atomic)
786 area = &(zone->free_area[current_order]);
787 if (list_empty(&area->free_list[migratetype]))
790 page = list_entry(area->free_list[migratetype].next,
795 * If breaking a large block of pages, move all free
796 * pages to the preferred allocation list
798 if (unlikely(current_order >= MAX_ORDER / 2)) {
799 migratetype = start_migratetype;
800 move_freepages_block(zone, page, migratetype);
803 /* Remove the page from the freelists */
804 list_del(&page->lru);
805 rmv_page_order(page);
806 __mod_zone_page_state(zone, NR_FREE_PAGES,
809 if (current_order == MAX_ORDER - 1)
810 set_pageblock_migratetype(page,
813 expand(zone, page, order, current_order, area, migratetype);
818 /* Allow fallback to high-order atomic blocks if memory is that low */
819 if (!nonatomic_fallback_atomic) {
820 nonatomic_fallback_atomic = 1;
827 static struct page *__rmqueue_fallback(struct zone *zone, int order,
828 int start_migratetype)
832 #endif /* CONFIG_PAGE_GROUP_BY_MOBILITY */
835 * Do the hard work of removing an element from the buddy allocator.
836 * Call me with the zone->lock already held.
838 static struct page *__rmqueue(struct zone *zone, unsigned int order,
841 struct free_area * area;
842 unsigned int current_order;
845 /* Find a page of the appropriate size in the preferred list */
846 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
847 area = &(zone->free_area[current_order]);
848 if (list_empty(&area->free_list[migratetype]))
851 page = list_entry(area->free_list[migratetype].next,
853 list_del(&page->lru);
854 rmv_page_order(page);
856 __mod_zone_page_state(zone, NR_FREE_PAGES, - (1UL << order));
857 expand(zone, page, order, current_order, area, migratetype);
861 page = __rmqueue_fallback(zone, order, migratetype);
869 * Obtain a specified number of elements from the buddy allocator, all under
870 * a single hold of the lock, for efficiency. Add them to the supplied list.
871 * Returns the number of new pages which were placed at *list.
873 static int rmqueue_bulk(struct zone *zone, unsigned int order,
874 unsigned long count, struct list_head *list,
879 spin_lock(&zone->lock);
880 for (i = 0; i < count; ++i) {
881 struct page *page = __rmqueue(zone, order, migratetype);
882 if (unlikely(page == NULL))
884 list_add(&page->lru, list);
885 set_page_private(page, migratetype);
887 spin_unlock(&zone->lock);
893 * Called from the vmstat counter updater to drain pagesets of this
894 * currently executing processor on remote nodes after they have
897 * Note that this function must be called with the thread pinned to
898 * a single processor.
900 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
905 local_irq_save(flags);
906 if (pcp->count >= pcp->batch)
907 to_drain = pcp->batch;
909 to_drain = pcp->count;
910 free_pages_bulk(zone, to_drain, &pcp->list, 0);
911 pcp->count -= to_drain;
912 local_irq_restore(flags);
916 static void __drain_pages(unsigned int cpu)
922 for_each_zone(zone) {
923 struct per_cpu_pageset *pset;
925 if (!populated_zone(zone))
928 pset = zone_pcp(zone, cpu);
929 for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) {
930 struct per_cpu_pages *pcp;
933 local_irq_save(flags);
934 free_pages_bulk(zone, pcp->count, &pcp->list, 0);
936 local_irq_restore(flags);
941 #ifdef CONFIG_HIBERNATION
943 void mark_free_pages(struct zone *zone)
945 unsigned long pfn, max_zone_pfn;
948 struct list_head *curr;
950 if (!zone->spanned_pages)
953 spin_lock_irqsave(&zone->lock, flags);
955 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
956 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
957 if (pfn_valid(pfn)) {
958 struct page *page = pfn_to_page(pfn);
960 if (!swsusp_page_is_forbidden(page))
961 swsusp_unset_page_free(page);
964 for_each_migratetype_order(order, t) {
965 list_for_each(curr, &zone->free_area[order].free_list[t]) {
968 pfn = page_to_pfn(list_entry(curr, struct page, lru));
969 for (i = 0; i < (1UL << order); i++)
970 swsusp_set_page_free(pfn_to_page(pfn + i));
973 spin_unlock_irqrestore(&zone->lock, flags);
975 #endif /* CONFIG_PM */
977 #if defined(CONFIG_HIBERNATION) || defined(CONFIG_PAGE_GROUP_BY_MOBILITY)
979 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
981 void drain_local_pages(void)
985 local_irq_save(flags);
986 __drain_pages(smp_processor_id());
987 local_irq_restore(flags);
990 void smp_drain_local_pages(void *arg)
996 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
998 void drain_all_local_pages(void)
1000 unsigned long flags;
1002 local_irq_save(flags);
1003 __drain_pages(smp_processor_id());
1004 local_irq_restore(flags);
1006 smp_call_function(smp_drain_local_pages, NULL, 0, 1);
1009 void drain_all_local_pages(void) {}
1010 #endif /* CONFIG_HIBERNATION || CONFIG_PAGE_GROUP_BY_MOBILITY */
1013 * Free a 0-order page
1015 static void fastcall free_hot_cold_page(struct page *page, int cold)
1017 struct zone *zone = page_zone(page);
1018 struct per_cpu_pages *pcp;
1019 unsigned long flags;
1022 page->mapping = NULL;
1023 if (free_pages_check(page))
1026 if (!PageHighMem(page))
1027 debug_check_no_locks_freed(page_address(page), PAGE_SIZE);
1028 arch_free_page(page, 0);
1029 kernel_map_pages(page, 1, 0);
1031 pcp = &zone_pcp(zone, get_cpu())->pcp[cold];
1032 local_irq_save(flags);
1033 __count_vm_event(PGFREE);
1034 list_add(&page->lru, &pcp->list);
1035 set_page_private(page, get_pageblock_migratetype(page));
1037 if (pcp->count >= pcp->high) {
1038 free_pages_bulk(zone, pcp->batch, &pcp->list, 0);
1039 pcp->count -= pcp->batch;
1041 local_irq_restore(flags);
1045 void fastcall free_hot_page(struct page *page)
1047 free_hot_cold_page(page, 0);
1050 void fastcall free_cold_page(struct page *page)
1052 free_hot_cold_page(page, 1);
1056 * split_page takes a non-compound higher-order page, and splits it into
1057 * n (1<<order) sub-pages: page[0..n]
1058 * Each sub-page must be freed individually.
1060 * Note: this is probably too low level an operation for use in drivers.
1061 * Please consult with lkml before using this in your driver.
1063 void split_page(struct page *page, unsigned int order)
1067 VM_BUG_ON(PageCompound(page));
1068 VM_BUG_ON(!page_count(page));
1069 for (i = 1; i < (1 << order); i++)
1070 set_page_refcounted(page + i);
1074 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1075 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1078 static struct page *buffered_rmqueue(struct zonelist *zonelist,
1079 struct zone *zone, int order, gfp_t gfp_flags)
1081 unsigned long flags;
1083 int cold = !!(gfp_flags & __GFP_COLD);
1085 int migratetype = allocflags_to_migratetype(gfp_flags, order);
1089 if (likely(order == 0)) {
1090 struct per_cpu_pages *pcp;
1092 pcp = &zone_pcp(zone, cpu)->pcp[cold];
1093 local_irq_save(flags);
1095 pcp->count = rmqueue_bulk(zone, 0,
1096 pcp->batch, &pcp->list, migratetype);
1097 if (unlikely(!pcp->count))
1101 #ifdef CONFIG_PAGE_GROUP_BY_MOBILITY
1102 /* Find a page of the appropriate migrate type */
1103 list_for_each_entry(page, &pcp->list, lru)
1104 if (page_private(page) == migratetype)
1107 /* Allocate more to the pcp list if necessary */
1108 if (unlikely(&page->lru == &pcp->list)) {
1109 pcp->count += rmqueue_bulk(zone, 0,
1110 pcp->batch, &pcp->list, migratetype);
1111 page = list_entry(pcp->list.next, struct page, lru);
1114 page = list_entry(pcp->list.next, struct page, lru);
1115 #endif /* CONFIG_PAGE_GROUP_BY_MOBILITY */
1117 list_del(&page->lru);
1120 spin_lock_irqsave(&zone->lock, flags);
1121 page = __rmqueue(zone, order, migratetype);
1122 spin_unlock(&zone->lock);
1127 __count_zone_vm_events(PGALLOC, zone, 1 << order);
1128 zone_statistics(zonelist, zone);
1129 local_irq_restore(flags);
1132 VM_BUG_ON(bad_range(zone, page));
1133 if (prep_new_page(page, order, gfp_flags))
1138 local_irq_restore(flags);
1143 #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
1144 #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
1145 #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
1146 #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
1147 #define ALLOC_HARDER 0x10 /* try to alloc harder */
1148 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1149 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
1151 #ifdef CONFIG_FAIL_PAGE_ALLOC
1153 static struct fail_page_alloc_attr {
1154 struct fault_attr attr;
1156 u32 ignore_gfp_highmem;
1157 u32 ignore_gfp_wait;
1160 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1162 struct dentry *ignore_gfp_highmem_file;
1163 struct dentry *ignore_gfp_wait_file;
1164 struct dentry *min_order_file;
1166 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1168 } fail_page_alloc = {
1169 .attr = FAULT_ATTR_INITIALIZER,
1170 .ignore_gfp_wait = 1,
1171 .ignore_gfp_highmem = 1,
1175 static int __init setup_fail_page_alloc(char *str)
1177 return setup_fault_attr(&fail_page_alloc.attr, str);
1179 __setup("fail_page_alloc=", setup_fail_page_alloc);
1181 static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1183 if (order < fail_page_alloc.min_order)
1185 if (gfp_mask & __GFP_NOFAIL)
1187 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
1189 if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
1192 return should_fail(&fail_page_alloc.attr, 1 << order);
1195 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1197 static int __init fail_page_alloc_debugfs(void)
1199 mode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
1203 err = init_fault_attr_dentries(&fail_page_alloc.attr,
1207 dir = fail_page_alloc.attr.dentries.dir;
1209 fail_page_alloc.ignore_gfp_wait_file =
1210 debugfs_create_bool("ignore-gfp-wait", mode, dir,
1211 &fail_page_alloc.ignore_gfp_wait);
1213 fail_page_alloc.ignore_gfp_highmem_file =
1214 debugfs_create_bool("ignore-gfp-highmem", mode, dir,
1215 &fail_page_alloc.ignore_gfp_highmem);
1216 fail_page_alloc.min_order_file =
1217 debugfs_create_u32("min-order", mode, dir,
1218 &fail_page_alloc.min_order);
1220 if (!fail_page_alloc.ignore_gfp_wait_file ||
1221 !fail_page_alloc.ignore_gfp_highmem_file ||
1222 !fail_page_alloc.min_order_file) {
1224 debugfs_remove(fail_page_alloc.ignore_gfp_wait_file);
1225 debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file);
1226 debugfs_remove(fail_page_alloc.min_order_file);
1227 cleanup_fault_attr_dentries(&fail_page_alloc.attr);
1233 late_initcall(fail_page_alloc_debugfs);
1235 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1237 #else /* CONFIG_FAIL_PAGE_ALLOC */
1239 static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1244 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1247 * Return 1 if free pages are above 'mark'. This takes into account the order
1248 * of the allocation.
1250 int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1251 int classzone_idx, int alloc_flags)
1253 /* free_pages my go negative - that's OK */
1255 long free_pages = zone_page_state(z, NR_FREE_PAGES) - (1 << order) + 1;
1258 if (alloc_flags & ALLOC_HIGH)
1260 if (alloc_flags & ALLOC_HARDER)
1263 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
1265 for (o = 0; o < order; o++) {
1266 /* At the next order, this order's pages become unavailable */
1267 free_pages -= z->free_area[o].nr_free << o;
1269 /* Require fewer higher order pages to be free */
1272 if (free_pages <= min)
1280 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1281 * skip over zones that are not allowed by the cpuset, or that have
1282 * been recently (in last second) found to be nearly full. See further
1283 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1284 * that have to skip over alot of full or unallowed zones.
1286 * If the zonelist cache is present in the passed in zonelist, then
1287 * returns a pointer to the allowed node mask (either the current
1288 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
1290 * If the zonelist cache is not available for this zonelist, does
1291 * nothing and returns NULL.
1293 * If the fullzones BITMAP in the zonelist cache is stale (more than
1294 * a second since last zap'd) then we zap it out (clear its bits.)
1296 * We hold off even calling zlc_setup, until after we've checked the
1297 * first zone in the zonelist, on the theory that most allocations will
1298 * be satisfied from that first zone, so best to examine that zone as
1299 * quickly as we can.
1301 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1303 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1304 nodemask_t *allowednodes; /* zonelist_cache approximation */
1306 zlc = zonelist->zlcache_ptr;
1310 if (jiffies - zlc->last_full_zap > 1 * HZ) {
1311 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1312 zlc->last_full_zap = jiffies;
1315 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1316 &cpuset_current_mems_allowed :
1317 &node_states[N_HIGH_MEMORY];
1318 return allowednodes;
1322 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1323 * if it is worth looking at further for free memory:
1324 * 1) Check that the zone isn't thought to be full (doesn't have its
1325 * bit set in the zonelist_cache fullzones BITMAP).
1326 * 2) Check that the zones node (obtained from the zonelist_cache
1327 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1328 * Return true (non-zero) if zone is worth looking at further, or
1329 * else return false (zero) if it is not.
1331 * This check -ignores- the distinction between various watermarks,
1332 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1333 * found to be full for any variation of these watermarks, it will
1334 * be considered full for up to one second by all requests, unless
1335 * we are so low on memory on all allowed nodes that we are forced
1336 * into the second scan of the zonelist.
1338 * In the second scan we ignore this zonelist cache and exactly
1339 * apply the watermarks to all zones, even it is slower to do so.
1340 * We are low on memory in the second scan, and should leave no stone
1341 * unturned looking for a free page.
1343 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zone **z,
1344 nodemask_t *allowednodes)
1346 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1347 int i; /* index of *z in zonelist zones */
1348 int n; /* node that zone *z is on */
1350 zlc = zonelist->zlcache_ptr;
1354 i = z - zonelist->zones;
1357 /* This zone is worth trying if it is allowed but not full */
1358 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1362 * Given 'z' scanning a zonelist, set the corresponding bit in
1363 * zlc->fullzones, so that subsequent attempts to allocate a page
1364 * from that zone don't waste time re-examining it.
1366 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zone **z)
1368 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1369 int i; /* index of *z in zonelist zones */
1371 zlc = zonelist->zlcache_ptr;
1375 i = z - zonelist->zones;
1377 set_bit(i, zlc->fullzones);
1380 #else /* CONFIG_NUMA */
1382 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1387 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zone **z,
1388 nodemask_t *allowednodes)
1393 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zone **z)
1396 #endif /* CONFIG_NUMA */
1399 * get_page_from_freelist goes through the zonelist trying to allocate
1402 static struct page *
1403 get_page_from_freelist(gfp_t gfp_mask, unsigned int order,
1404 struct zonelist *zonelist, int alloc_flags)
1407 struct page *page = NULL;
1408 int classzone_idx = zone_idx(zonelist->zones[0]);
1410 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1411 int zlc_active = 0; /* set if using zonelist_cache */
1412 int did_zlc_setup = 0; /* just call zlc_setup() one time */
1413 enum zone_type highest_zoneidx = -1; /* Gets set for policy zonelists */
1417 * Scan zonelist, looking for a zone with enough free.
1418 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1420 z = zonelist->zones;
1424 * In NUMA, this could be a policy zonelist which contains
1425 * zones that may not be allowed by the current gfp_mask.
1426 * Check the zone is allowed by the current flags
1428 if (unlikely(alloc_should_filter_zonelist(zonelist))) {
1429 if (highest_zoneidx == -1)
1430 highest_zoneidx = gfp_zone(gfp_mask);
1431 if (zone_idx(*z) > highest_zoneidx)
1435 if (NUMA_BUILD && zlc_active &&
1436 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1439 if ((alloc_flags & ALLOC_CPUSET) &&
1440 !cpuset_zone_allowed_softwall(zone, gfp_mask))
1443 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
1445 if (alloc_flags & ALLOC_WMARK_MIN)
1446 mark = zone->pages_min;
1447 else if (alloc_flags & ALLOC_WMARK_LOW)
1448 mark = zone->pages_low;
1450 mark = zone->pages_high;
1451 if (!zone_watermark_ok(zone, order, mark,
1452 classzone_idx, alloc_flags)) {
1453 if (!zone_reclaim_mode ||
1454 !zone_reclaim(zone, gfp_mask, order))
1455 goto this_zone_full;
1459 page = buffered_rmqueue(zonelist, zone, order, gfp_mask);
1464 zlc_mark_zone_full(zonelist, z);
1466 if (NUMA_BUILD && !did_zlc_setup) {
1467 /* we do zlc_setup after the first zone is tried */
1468 allowednodes = zlc_setup(zonelist, alloc_flags);
1472 } while (*(++z) != NULL);
1474 if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) {
1475 /* Disable zlc cache for second zonelist scan */
1483 * This is the 'heart' of the zoned buddy allocator.
1485 struct page * fastcall
1486 __alloc_pages(gfp_t gfp_mask, unsigned int order,
1487 struct zonelist *zonelist)
1489 const gfp_t wait = gfp_mask & __GFP_WAIT;
1492 struct reclaim_state reclaim_state;
1493 struct task_struct *p = current;
1496 int did_some_progress;
1498 might_sleep_if(wait);
1500 if (should_fail_alloc_page(gfp_mask, order))
1504 z = zonelist->zones; /* the list of zones suitable for gfp_mask */
1506 if (unlikely(*z == NULL)) {
1508 * Happens if we have an empty zonelist as a result of
1509 * GFP_THISNODE being used on a memoryless node
1514 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
1515 zonelist, ALLOC_WMARK_LOW|ALLOC_CPUSET);
1520 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1521 * __GFP_NOWARN set) should not cause reclaim since the subsystem
1522 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1523 * using a larger set of nodes after it has established that the
1524 * allowed per node queues are empty and that nodes are
1527 if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
1530 for (z = zonelist->zones; *z; z++)
1531 wakeup_kswapd(*z, order);
1534 * OK, we're below the kswapd watermark and have kicked background
1535 * reclaim. Now things get more complex, so set up alloc_flags according
1536 * to how we want to proceed.
1538 * The caller may dip into page reserves a bit more if the caller
1539 * cannot run direct reclaim, or if the caller has realtime scheduling
1540 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1541 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1543 alloc_flags = ALLOC_WMARK_MIN;
1544 if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait)
1545 alloc_flags |= ALLOC_HARDER;
1546 if (gfp_mask & __GFP_HIGH)
1547 alloc_flags |= ALLOC_HIGH;
1549 alloc_flags |= ALLOC_CPUSET;
1552 * Go through the zonelist again. Let __GFP_HIGH and allocations
1553 * coming from realtime tasks go deeper into reserves.
1555 * This is the last chance, in general, before the goto nopage.
1556 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1557 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1559 page = get_page_from_freelist(gfp_mask, order, zonelist, alloc_flags);
1563 /* This allocation should allow future memory freeing. */
1566 if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE)))
1567 && !in_interrupt()) {
1568 if (!(gfp_mask & __GFP_NOMEMALLOC)) {
1570 /* go through the zonelist yet again, ignoring mins */
1571 page = get_page_from_freelist(gfp_mask, order,
1572 zonelist, ALLOC_NO_WATERMARKS);
1575 if (gfp_mask & __GFP_NOFAIL) {
1576 congestion_wait(WRITE, HZ/50);
1583 /* Atomic allocations - we can't balance anything */
1589 /* We now go into synchronous reclaim */
1590 cpuset_memory_pressure_bump();
1591 p->flags |= PF_MEMALLOC;
1592 reclaim_state.reclaimed_slab = 0;
1593 p->reclaim_state = &reclaim_state;
1595 did_some_progress = try_to_free_pages(zonelist->zones, order, gfp_mask);
1597 p->reclaim_state = NULL;
1598 p->flags &= ~PF_MEMALLOC;
1603 drain_all_local_pages();
1605 if (likely(did_some_progress)) {
1606 page = get_page_from_freelist(gfp_mask, order,
1607 zonelist, alloc_flags);
1610 } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
1612 * Go through the zonelist yet one more time, keep
1613 * very high watermark here, this is only to catch
1614 * a parallel oom killing, we must fail if we're still
1615 * under heavy pressure.
1617 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
1618 zonelist, ALLOC_WMARK_HIGH|ALLOC_CPUSET);
1622 /* The OOM killer will not help higher order allocs so fail */
1623 if (order > PAGE_ALLOC_COSTLY_ORDER)
1626 out_of_memory(zonelist, gfp_mask, order);
1631 * Don't let big-order allocations loop unless the caller explicitly
1632 * requests that. Wait for some write requests to complete then retry.
1634 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
1635 * <= 3, but that may not be true in other implementations.
1638 if (!(gfp_mask & __GFP_NORETRY)) {
1639 if ((order <= PAGE_ALLOC_COSTLY_ORDER) ||
1640 (gfp_mask & __GFP_REPEAT))
1642 if (gfp_mask & __GFP_NOFAIL)
1646 congestion_wait(WRITE, HZ/50);
1651 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
1652 printk(KERN_WARNING "%s: page allocation failure."
1653 " order:%d, mode:0x%x\n",
1654 p->comm, order, gfp_mask);
1662 EXPORT_SYMBOL(__alloc_pages);
1665 * Common helper functions.
1667 fastcall unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
1670 page = alloc_pages(gfp_mask, order);
1673 return (unsigned long) page_address(page);
1676 EXPORT_SYMBOL(__get_free_pages);
1678 fastcall unsigned long get_zeroed_page(gfp_t gfp_mask)
1683 * get_zeroed_page() returns a 32-bit address, which cannot represent
1686 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
1688 page = alloc_pages(gfp_mask | __GFP_ZERO, 0);
1690 return (unsigned long) page_address(page);
1694 EXPORT_SYMBOL(get_zeroed_page);
1696 void __pagevec_free(struct pagevec *pvec)
1698 int i = pagevec_count(pvec);
1701 free_hot_cold_page(pvec->pages[i], pvec->cold);
1704 fastcall void __free_pages(struct page *page, unsigned int order)
1706 if (put_page_testzero(page)) {
1708 free_hot_page(page);
1710 __free_pages_ok(page, order);
1714 EXPORT_SYMBOL(__free_pages);
1716 fastcall void free_pages(unsigned long addr, unsigned int order)
1719 VM_BUG_ON(!virt_addr_valid((void *)addr));
1720 __free_pages(virt_to_page((void *)addr), order);
1724 EXPORT_SYMBOL(free_pages);
1726 static unsigned int nr_free_zone_pages(int offset)
1728 /* Just pick one node, since fallback list is circular */
1729 pg_data_t *pgdat = NODE_DATA(numa_node_id());
1730 unsigned int sum = 0;
1732 struct zonelist *zonelist = pgdat->node_zonelists + offset;
1733 struct zone **zonep = zonelist->zones;
1736 for (zone = *zonep++; zone; zone = *zonep++) {
1737 unsigned long size = zone->present_pages;
1738 unsigned long high = zone->pages_high;
1747 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1749 unsigned int nr_free_buffer_pages(void)
1751 return nr_free_zone_pages(gfp_zone(GFP_USER));
1753 EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
1756 * Amount of free RAM allocatable within all zones
1758 unsigned int nr_free_pagecache_pages(void)
1760 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
1763 static inline void show_node(struct zone *zone)
1766 printk("Node %d ", zone_to_nid(zone));
1769 void si_meminfo(struct sysinfo *val)
1771 val->totalram = totalram_pages;
1773 val->freeram = global_page_state(NR_FREE_PAGES);
1774 val->bufferram = nr_blockdev_pages();
1775 val->totalhigh = totalhigh_pages;
1776 val->freehigh = nr_free_highpages();
1777 val->mem_unit = PAGE_SIZE;
1780 EXPORT_SYMBOL(si_meminfo);
1783 void si_meminfo_node(struct sysinfo *val, int nid)
1785 pg_data_t *pgdat = NODE_DATA(nid);
1787 val->totalram = pgdat->node_present_pages;
1788 val->freeram = node_page_state(nid, NR_FREE_PAGES);
1789 #ifdef CONFIG_HIGHMEM
1790 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1791 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
1797 val->mem_unit = PAGE_SIZE;
1801 #define K(x) ((x) << (PAGE_SHIFT-10))
1804 * Show free area list (used inside shift_scroll-lock stuff)
1805 * We also calculate the percentage fragmentation. We do this by counting the
1806 * memory on each free list with the exception of the first item on the list.
1808 void show_free_areas(void)
1813 for_each_zone(zone) {
1814 if (!populated_zone(zone))
1818 printk("%s per-cpu:\n", zone->name);
1820 for_each_online_cpu(cpu) {
1821 struct per_cpu_pageset *pageset;
1823 pageset = zone_pcp(zone, cpu);
1825 printk("CPU %4d: Hot: hi:%5d, btch:%4d usd:%4d "
1826 "Cold: hi:%5d, btch:%4d usd:%4d\n",
1827 cpu, pageset->pcp[0].high,
1828 pageset->pcp[0].batch, pageset->pcp[0].count,
1829 pageset->pcp[1].high, pageset->pcp[1].batch,
1830 pageset->pcp[1].count);
1834 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu unstable:%lu\n"
1835 " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
1836 global_page_state(NR_ACTIVE),
1837 global_page_state(NR_INACTIVE),
1838 global_page_state(NR_FILE_DIRTY),
1839 global_page_state(NR_WRITEBACK),
1840 global_page_state(NR_UNSTABLE_NFS),
1841 global_page_state(NR_FREE_PAGES),
1842 global_page_state(NR_SLAB_RECLAIMABLE) +
1843 global_page_state(NR_SLAB_UNRECLAIMABLE),
1844 global_page_state(NR_FILE_MAPPED),
1845 global_page_state(NR_PAGETABLE),
1846 global_page_state(NR_BOUNCE));
1848 for_each_zone(zone) {
1851 if (!populated_zone(zone))
1863 " pages_scanned:%lu"
1864 " all_unreclaimable? %s"
1867 K(zone_page_state(zone, NR_FREE_PAGES)),
1870 K(zone->pages_high),
1871 K(zone_page_state(zone, NR_ACTIVE)),
1872 K(zone_page_state(zone, NR_INACTIVE)),
1873 K(zone->present_pages),
1874 zone->pages_scanned,
1875 (zone->all_unreclaimable ? "yes" : "no")
1877 printk("lowmem_reserve[]:");
1878 for (i = 0; i < MAX_NR_ZONES; i++)
1879 printk(" %lu", zone->lowmem_reserve[i]);
1883 for_each_zone(zone) {
1884 unsigned long nr[MAX_ORDER], flags, order, total = 0;
1886 if (!populated_zone(zone))
1890 printk("%s: ", zone->name);
1892 spin_lock_irqsave(&zone->lock, flags);
1893 for (order = 0; order < MAX_ORDER; order++) {
1894 nr[order] = zone->free_area[order].nr_free;
1895 total += nr[order] << order;
1897 spin_unlock_irqrestore(&zone->lock, flags);
1898 for (order = 0; order < MAX_ORDER; order++)
1899 printk("%lu*%lukB ", nr[order], K(1UL) << order);
1900 printk("= %lukB\n", K(total));
1903 show_swap_cache_info();
1907 * Builds allocation fallback zone lists.
1909 * Add all populated zones of a node to the zonelist.
1911 static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
1912 int nr_zones, enum zone_type zone_type)
1916 BUG_ON(zone_type >= MAX_NR_ZONES);
1921 zone = pgdat->node_zones + zone_type;
1922 if (populated_zone(zone)) {
1923 zonelist->zones[nr_zones++] = zone;
1924 check_highest_zone(zone_type);
1927 } while (zone_type);
1934 * 0 = automatic detection of better ordering.
1935 * 1 = order by ([node] distance, -zonetype)
1936 * 2 = order by (-zonetype, [node] distance)
1938 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
1939 * the same zonelist. So only NUMA can configure this param.
1941 #define ZONELIST_ORDER_DEFAULT 0
1942 #define ZONELIST_ORDER_NODE 1
1943 #define ZONELIST_ORDER_ZONE 2
1945 /* zonelist order in the kernel.
1946 * set_zonelist_order() will set this to NODE or ZONE.
1948 static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
1949 static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
1953 /* The value user specified ....changed by config */
1954 static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
1955 /* string for sysctl */
1956 #define NUMA_ZONELIST_ORDER_LEN 16
1957 char numa_zonelist_order[16] = "default";
1960 * interface for configure zonelist ordering.
1961 * command line option "numa_zonelist_order"
1962 * = "[dD]efault - default, automatic configuration.
1963 * = "[nN]ode - order by node locality, then by zone within node
1964 * = "[zZ]one - order by zone, then by locality within zone
1967 static int __parse_numa_zonelist_order(char *s)
1969 if (*s == 'd' || *s == 'D') {
1970 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
1971 } else if (*s == 'n' || *s == 'N') {
1972 user_zonelist_order = ZONELIST_ORDER_NODE;
1973 } else if (*s == 'z' || *s == 'Z') {
1974 user_zonelist_order = ZONELIST_ORDER_ZONE;
1977 "Ignoring invalid numa_zonelist_order value: "
1984 static __init int setup_numa_zonelist_order(char *s)
1987 return __parse_numa_zonelist_order(s);
1990 early_param("numa_zonelist_order", setup_numa_zonelist_order);
1993 * sysctl handler for numa_zonelist_order
1995 int numa_zonelist_order_handler(ctl_table *table, int write,
1996 struct file *file, void __user *buffer, size_t *length,
1999 char saved_string[NUMA_ZONELIST_ORDER_LEN];
2003 strncpy(saved_string, (char*)table->data,
2004 NUMA_ZONELIST_ORDER_LEN);
2005 ret = proc_dostring(table, write, file, buffer, length, ppos);
2009 int oldval = user_zonelist_order;
2010 if (__parse_numa_zonelist_order((char*)table->data)) {
2012 * bogus value. restore saved string
2014 strncpy((char*)table->data, saved_string,
2015 NUMA_ZONELIST_ORDER_LEN);
2016 user_zonelist_order = oldval;
2017 } else if (oldval != user_zonelist_order)
2018 build_all_zonelists();
2024 #define MAX_NODE_LOAD (num_online_nodes())
2025 static int node_load[MAX_NUMNODES];
2028 * find_next_best_node - find the next node that should appear in a given node's fallback list
2029 * @node: node whose fallback list we're appending
2030 * @used_node_mask: nodemask_t of already used nodes
2032 * We use a number of factors to determine which is the next node that should
2033 * appear on a given node's fallback list. The node should not have appeared
2034 * already in @node's fallback list, and it should be the next closest node
2035 * according to the distance array (which contains arbitrary distance values
2036 * from each node to each node in the system), and should also prefer nodes
2037 * with no CPUs, since presumably they'll have very little allocation pressure
2038 * on them otherwise.
2039 * It returns -1 if no node is found.
2041 static int find_next_best_node(int node, nodemask_t *used_node_mask)
2044 int min_val = INT_MAX;
2047 /* Use the local node if we haven't already */
2048 if (!node_isset(node, *used_node_mask)) {
2049 node_set(node, *used_node_mask);
2053 for_each_node_state(n, N_HIGH_MEMORY) {
2056 /* Don't want a node to appear more than once */
2057 if (node_isset(n, *used_node_mask))
2060 /* Use the distance array to find the distance */
2061 val = node_distance(node, n);
2063 /* Penalize nodes under us ("prefer the next node") */
2066 /* Give preference to headless and unused nodes */
2067 tmp = node_to_cpumask(n);
2068 if (!cpus_empty(tmp))
2069 val += PENALTY_FOR_NODE_WITH_CPUS;
2071 /* Slight preference for less loaded node */
2072 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
2073 val += node_load[n];
2075 if (val < min_val) {
2082 node_set(best_node, *used_node_mask);
2089 * Build zonelists ordered by node and zones within node.
2090 * This results in maximum locality--normal zone overflows into local
2091 * DMA zone, if any--but risks exhausting DMA zone.
2093 static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
2097 struct zonelist *zonelist;
2099 for (i = 0; i < MAX_NR_ZONES; i++) {
2100 zonelist = pgdat->node_zonelists + i;
2101 for (j = 0; zonelist->zones[j] != NULL; j++)
2103 j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
2104 zonelist->zones[j] = NULL;
2109 * Build gfp_thisnode zonelists
2111 static void build_thisnode_zonelists(pg_data_t *pgdat)
2115 struct zonelist *zonelist;
2117 for (i = 0; i < MAX_NR_ZONES; i++) {
2118 zonelist = pgdat->node_zonelists + MAX_NR_ZONES + i;
2119 j = build_zonelists_node(pgdat, zonelist, 0, i);
2120 zonelist->zones[j] = NULL;
2125 * Build zonelists ordered by zone and nodes within zones.
2126 * This results in conserving DMA zone[s] until all Normal memory is
2127 * exhausted, but results in overflowing to remote node while memory
2128 * may still exist in local DMA zone.
2130 static int node_order[MAX_NUMNODES];
2132 static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
2136 int zone_type; /* needs to be signed */
2138 struct zonelist *zonelist;
2140 for (i = 0; i < MAX_NR_ZONES; i++) {
2141 zonelist = pgdat->node_zonelists + i;
2143 for (zone_type = i; zone_type >= 0; zone_type--) {
2144 for (j = 0; j < nr_nodes; j++) {
2145 node = node_order[j];
2146 z = &NODE_DATA(node)->node_zones[zone_type];
2147 if (populated_zone(z)) {
2148 zonelist->zones[pos++] = z;
2149 check_highest_zone(zone_type);
2153 zonelist->zones[pos] = NULL;
2157 static int default_zonelist_order(void)
2160 unsigned long low_kmem_size,total_size;
2164 * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem.
2165 * If they are really small and used heavily, the system can fall
2166 * into OOM very easily.
2167 * This function detect ZONE_DMA/DMA32 size and confgigures zone order.
2169 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2172 for_each_online_node(nid) {
2173 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2174 z = &NODE_DATA(nid)->node_zones[zone_type];
2175 if (populated_zone(z)) {
2176 if (zone_type < ZONE_NORMAL)
2177 low_kmem_size += z->present_pages;
2178 total_size += z->present_pages;
2182 if (!low_kmem_size || /* there are no DMA area. */
2183 low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
2184 return ZONELIST_ORDER_NODE;
2186 * look into each node's config.
2187 * If there is a node whose DMA/DMA32 memory is very big area on
2188 * local memory, NODE_ORDER may be suitable.
2190 average_size = total_size /
2191 (nodes_weight(node_states[N_HIGH_MEMORY]) + 1);
2192 for_each_online_node(nid) {
2195 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2196 z = &NODE_DATA(nid)->node_zones[zone_type];
2197 if (populated_zone(z)) {
2198 if (zone_type < ZONE_NORMAL)
2199 low_kmem_size += z->present_pages;
2200 total_size += z->present_pages;
2203 if (low_kmem_size &&
2204 total_size > average_size && /* ignore small node */
2205 low_kmem_size > total_size * 70/100)
2206 return ZONELIST_ORDER_NODE;
2208 return ZONELIST_ORDER_ZONE;
2211 static void set_zonelist_order(void)
2213 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
2214 current_zonelist_order = default_zonelist_order();
2216 current_zonelist_order = user_zonelist_order;
2219 static void build_zonelists(pg_data_t *pgdat)
2223 nodemask_t used_mask;
2224 int local_node, prev_node;
2225 struct zonelist *zonelist;
2226 int order = current_zonelist_order;
2228 /* initialize zonelists */
2229 for (i = 0; i < MAX_ZONELISTS; i++) {
2230 zonelist = pgdat->node_zonelists + i;
2231 zonelist->zones[0] = NULL;
2234 /* NUMA-aware ordering of nodes */
2235 local_node = pgdat->node_id;
2236 load = num_online_nodes();
2237 prev_node = local_node;
2238 nodes_clear(used_mask);
2240 memset(node_load, 0, sizeof(node_load));
2241 memset(node_order, 0, sizeof(node_order));
2244 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
2245 int distance = node_distance(local_node, node);
2248 * If another node is sufficiently far away then it is better
2249 * to reclaim pages in a zone before going off node.
2251 if (distance > RECLAIM_DISTANCE)
2252 zone_reclaim_mode = 1;
2255 * We don't want to pressure a particular node.
2256 * So adding penalty to the first node in same
2257 * distance group to make it round-robin.
2259 if (distance != node_distance(local_node, prev_node))
2260 node_load[node] = load;
2264 if (order == ZONELIST_ORDER_NODE)
2265 build_zonelists_in_node_order(pgdat, node);
2267 node_order[j++] = node; /* remember order */
2270 if (order == ZONELIST_ORDER_ZONE) {
2271 /* calculate node order -- i.e., DMA last! */
2272 build_zonelists_in_zone_order(pgdat, j);
2275 build_thisnode_zonelists(pgdat);
2278 /* Construct the zonelist performance cache - see further mmzone.h */
2279 static void build_zonelist_cache(pg_data_t *pgdat)
2283 for (i = 0; i < MAX_NR_ZONES; i++) {
2284 struct zonelist *zonelist;
2285 struct zonelist_cache *zlc;
2288 zonelist = pgdat->node_zonelists + i;
2289 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
2290 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
2291 for (z = zonelist->zones; *z; z++)
2292 zlc->z_to_n[z - zonelist->zones] = zone_to_nid(*z);
2297 #else /* CONFIG_NUMA */
2299 static void set_zonelist_order(void)
2301 current_zonelist_order = ZONELIST_ORDER_ZONE;
2304 static void build_zonelists(pg_data_t *pgdat)
2306 int node, local_node;
2309 local_node = pgdat->node_id;
2310 for (i = 0; i < MAX_NR_ZONES; i++) {
2311 struct zonelist *zonelist;
2313 zonelist = pgdat->node_zonelists + i;
2315 j = build_zonelists_node(pgdat, zonelist, 0, i);
2317 * Now we build the zonelist so that it contains the zones
2318 * of all the other nodes.
2319 * We don't want to pressure a particular node, so when
2320 * building the zones for node N, we make sure that the
2321 * zones coming right after the local ones are those from
2322 * node N+1 (modulo N)
2324 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
2325 if (!node_online(node))
2327 j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
2329 for (node = 0; node < local_node; node++) {
2330 if (!node_online(node))
2332 j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
2335 zonelist->zones[j] = NULL;
2339 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
2340 static void build_zonelist_cache(pg_data_t *pgdat)
2344 for (i = 0; i < MAX_NR_ZONES; i++)
2345 pgdat->node_zonelists[i].zlcache_ptr = NULL;
2348 #endif /* CONFIG_NUMA */
2350 /* return values int ....just for stop_machine_run() */
2351 static int __build_all_zonelists(void *dummy)
2355 for_each_online_node(nid) {
2356 pg_data_t *pgdat = NODE_DATA(nid);
2358 build_zonelists(pgdat);
2359 build_zonelist_cache(pgdat);
2364 void build_all_zonelists(void)
2366 set_zonelist_order();
2368 if (system_state == SYSTEM_BOOTING) {
2369 __build_all_zonelists(NULL);
2370 cpuset_init_current_mems_allowed();
2372 /* we have to stop all cpus to guaranntee there is no user
2374 stop_machine_run(__build_all_zonelists, NULL, NR_CPUS);
2375 /* cpuset refresh routine should be here */
2377 vm_total_pages = nr_free_pagecache_pages();
2378 printk("Built %i zonelists in %s order. Total pages: %ld\n",
2380 zonelist_order_name[current_zonelist_order],
2383 printk("Policy zone: %s\n", zone_names[policy_zone]);
2388 * Helper functions to size the waitqueue hash table.
2389 * Essentially these want to choose hash table sizes sufficiently
2390 * large so that collisions trying to wait on pages are rare.
2391 * But in fact, the number of active page waitqueues on typical
2392 * systems is ridiculously low, less than 200. So this is even
2393 * conservative, even though it seems large.
2395 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
2396 * waitqueues, i.e. the size of the waitq table given the number of pages.
2398 #define PAGES_PER_WAITQUEUE 256
2400 #ifndef CONFIG_MEMORY_HOTPLUG
2401 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
2403 unsigned long size = 1;
2405 pages /= PAGES_PER_WAITQUEUE;
2407 while (size < pages)
2411 * Once we have dozens or even hundreds of threads sleeping
2412 * on IO we've got bigger problems than wait queue collision.
2413 * Limit the size of the wait table to a reasonable size.
2415 size = min(size, 4096UL);
2417 return max(size, 4UL);
2421 * A zone's size might be changed by hot-add, so it is not possible to determine
2422 * a suitable size for its wait_table. So we use the maximum size now.
2424 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
2426 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
2427 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
2428 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
2430 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
2431 * or more by the traditional way. (See above). It equals:
2433 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
2434 * ia64(16K page size) : = ( 8G + 4M)byte.
2435 * powerpc (64K page size) : = (32G +16M)byte.
2437 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
2444 * This is an integer logarithm so that shifts can be used later
2445 * to extract the more random high bits from the multiplicative
2446 * hash function before the remainder is taken.
2448 static inline unsigned long wait_table_bits(unsigned long size)
2453 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
2456 * Initially all pages are reserved - free ones are freed
2457 * up by free_all_bootmem() once the early boot process is
2458 * done. Non-atomic initialization, single-pass.
2460 void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
2461 unsigned long start_pfn, enum memmap_context context)
2464 unsigned long end_pfn = start_pfn + size;
2467 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
2469 * There can be holes in boot-time mem_map[]s
2470 * handed to this function. They do not
2471 * exist on hotplugged memory.
2473 if (context == MEMMAP_EARLY) {
2474 if (!early_pfn_valid(pfn))
2476 if (!early_pfn_in_nid(pfn, nid))
2479 page = pfn_to_page(pfn);
2480 set_page_links(page, zone, nid, pfn);
2481 init_page_count(page);
2482 reset_page_mapcount(page);
2483 SetPageReserved(page);
2486 * Mark the block movable so that blocks are reserved for
2487 * movable at startup. This will force kernel allocations
2488 * to reserve their blocks rather than leaking throughout
2489 * the address space during boot when many long-lived
2490 * kernel allocations are made
2492 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
2494 INIT_LIST_HEAD(&page->lru);
2495 #ifdef WANT_PAGE_VIRTUAL
2496 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
2497 if (!is_highmem_idx(zone))
2498 set_page_address(page, __va(pfn << PAGE_SHIFT));
2503 static void __meminit zone_init_free_lists(struct pglist_data *pgdat,
2504 struct zone *zone, unsigned long size)
2507 for_each_migratetype_order(order, t) {
2508 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
2509 zone->free_area[order].nr_free = 0;
2513 #ifndef __HAVE_ARCH_MEMMAP_INIT
2514 #define memmap_init(size, nid, zone, start_pfn) \
2515 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
2518 static int __devinit zone_batchsize(struct zone *zone)
2523 * The per-cpu-pages pools are set to around 1000th of the
2524 * size of the zone. But no more than 1/2 of a meg.
2526 * OK, so we don't know how big the cache is. So guess.
2528 batch = zone->present_pages / 1024;
2529 if (batch * PAGE_SIZE > 512 * 1024)
2530 batch = (512 * 1024) / PAGE_SIZE;
2531 batch /= 4; /* We effectively *= 4 below */
2536 * Clamp the batch to a 2^n - 1 value. Having a power
2537 * of 2 value was found to be more likely to have
2538 * suboptimal cache aliasing properties in some cases.
2540 * For example if 2 tasks are alternately allocating
2541 * batches of pages, one task can end up with a lot
2542 * of pages of one half of the possible page colors
2543 * and the other with pages of the other colors.
2545 batch = (1 << (fls(batch + batch/2)-1)) - 1;
2550 inline void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
2552 struct per_cpu_pages *pcp;
2554 memset(p, 0, sizeof(*p));
2556 pcp = &p->pcp[0]; /* hot */
2558 pcp->high = 6 * batch;
2559 pcp->batch = max(1UL, 1 * batch);
2560 INIT_LIST_HEAD(&pcp->list);
2562 pcp = &p->pcp[1]; /* cold*/
2564 pcp->high = 2 * batch;
2565 pcp->batch = max(1UL, batch/2);
2566 INIT_LIST_HEAD(&pcp->list);
2570 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
2571 * to the value high for the pageset p.
2574 static void setup_pagelist_highmark(struct per_cpu_pageset *p,
2577 struct per_cpu_pages *pcp;
2579 pcp = &p->pcp[0]; /* hot list */
2581 pcp->batch = max(1UL, high/4);
2582 if ((high/4) > (PAGE_SHIFT * 8))
2583 pcp->batch = PAGE_SHIFT * 8;
2589 * Boot pageset table. One per cpu which is going to be used for all
2590 * zones and all nodes. The parameters will be set in such a way
2591 * that an item put on a list will immediately be handed over to
2592 * the buddy list. This is safe since pageset manipulation is done
2593 * with interrupts disabled.
2595 * Some NUMA counter updates may also be caught by the boot pagesets.
2597 * The boot_pagesets must be kept even after bootup is complete for
2598 * unused processors and/or zones. They do play a role for bootstrapping
2599 * hotplugged processors.
2601 * zoneinfo_show() and maybe other functions do
2602 * not check if the processor is online before following the pageset pointer.
2603 * Other parts of the kernel may not check if the zone is available.
2605 static struct per_cpu_pageset boot_pageset[NR_CPUS];
2608 * Dynamically allocate memory for the
2609 * per cpu pageset array in struct zone.
2611 static int __cpuinit process_zones(int cpu)
2613 struct zone *zone, *dzone;
2614 int node = cpu_to_node(cpu);
2616 node_set_state(node, N_CPU); /* this node has a cpu */
2618 for_each_zone(zone) {
2620 if (!populated_zone(zone))
2623 zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset),
2625 if (!zone_pcp(zone, cpu))
2628 setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone));
2630 if (percpu_pagelist_fraction)
2631 setup_pagelist_highmark(zone_pcp(zone, cpu),
2632 (zone->present_pages / percpu_pagelist_fraction));
2637 for_each_zone(dzone) {
2638 if (!populated_zone(dzone))
2642 kfree(zone_pcp(dzone, cpu));
2643 zone_pcp(dzone, cpu) = NULL;
2648 static inline void free_zone_pagesets(int cpu)
2652 for_each_zone(zone) {
2653 struct per_cpu_pageset *pset = zone_pcp(zone, cpu);
2655 /* Free per_cpu_pageset if it is slab allocated */
2656 if (pset != &boot_pageset[cpu])
2658 zone_pcp(zone, cpu) = NULL;
2662 static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb,
2663 unsigned long action,
2666 int cpu = (long)hcpu;
2667 int ret = NOTIFY_OK;
2670 case CPU_UP_PREPARE:
2671 case CPU_UP_PREPARE_FROZEN:
2672 if (process_zones(cpu))
2675 case CPU_UP_CANCELED:
2676 case CPU_UP_CANCELED_FROZEN:
2678 case CPU_DEAD_FROZEN:
2679 free_zone_pagesets(cpu);
2687 static struct notifier_block __cpuinitdata pageset_notifier =
2688 { &pageset_cpuup_callback, NULL, 0 };
2690 void __init setup_per_cpu_pageset(void)
2694 /* Initialize per_cpu_pageset for cpu 0.
2695 * A cpuup callback will do this for every cpu
2696 * as it comes online
2698 err = process_zones(smp_processor_id());
2700 register_cpu_notifier(&pageset_notifier);
2705 static noinline __init_refok
2706 int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
2709 struct pglist_data *pgdat = zone->zone_pgdat;
2713 * The per-page waitqueue mechanism uses hashed waitqueues
2716 zone->wait_table_hash_nr_entries =
2717 wait_table_hash_nr_entries(zone_size_pages);
2718 zone->wait_table_bits =
2719 wait_table_bits(zone->wait_table_hash_nr_entries);
2720 alloc_size = zone->wait_table_hash_nr_entries
2721 * sizeof(wait_queue_head_t);
2723 if (system_state == SYSTEM_BOOTING) {
2724 zone->wait_table = (wait_queue_head_t *)
2725 alloc_bootmem_node(pgdat, alloc_size);
2728 * This case means that a zone whose size was 0 gets new memory
2729 * via memory hot-add.
2730 * But it may be the case that a new node was hot-added. In
2731 * this case vmalloc() will not be able to use this new node's
2732 * memory - this wait_table must be initialized to use this new
2733 * node itself as well.
2734 * To use this new node's memory, further consideration will be
2737 zone->wait_table = vmalloc(alloc_size);
2739 if (!zone->wait_table)
2742 for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
2743 init_waitqueue_head(zone->wait_table + i);
2748 static __meminit void zone_pcp_init(struct zone *zone)
2751 unsigned long batch = zone_batchsize(zone);
2753 for (cpu = 0; cpu < NR_CPUS; cpu++) {
2755 /* Early boot. Slab allocator not functional yet */
2756 zone_pcp(zone, cpu) = &boot_pageset[cpu];
2757 setup_pageset(&boot_pageset[cpu],0);
2759 setup_pageset(zone_pcp(zone,cpu), batch);
2762 if (zone->present_pages)
2763 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n",
2764 zone->name, zone->present_pages, batch);
2767 __meminit int init_currently_empty_zone(struct zone *zone,
2768 unsigned long zone_start_pfn,
2770 enum memmap_context context)
2772 struct pglist_data *pgdat = zone->zone_pgdat;
2774 ret = zone_wait_table_init(zone, size);
2777 pgdat->nr_zones = zone_idx(zone) + 1;
2779 zone->zone_start_pfn = zone_start_pfn;
2781 memmap_init(size, pgdat->node_id, zone_idx(zone), zone_start_pfn);
2783 zone_init_free_lists(pgdat, zone, zone->spanned_pages);
2788 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2790 * Basic iterator support. Return the first range of PFNs for a node
2791 * Note: nid == MAX_NUMNODES returns first region regardless of node
2793 static int __meminit first_active_region_index_in_nid(int nid)
2797 for (i = 0; i < nr_nodemap_entries; i++)
2798 if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
2805 * Basic iterator support. Return the next active range of PFNs for a node
2806 * Note: nid == MAX_NUMNODES returns next region regardles of node
2808 static int __meminit next_active_region_index_in_nid(int index, int nid)
2810 for (index = index + 1; index < nr_nodemap_entries; index++)
2811 if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
2817 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
2819 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
2820 * Architectures may implement their own version but if add_active_range()
2821 * was used and there are no special requirements, this is a convenient
2824 int __meminit early_pfn_to_nid(unsigned long pfn)
2828 for (i = 0; i < nr_nodemap_entries; i++) {
2829 unsigned long start_pfn = early_node_map[i].start_pfn;
2830 unsigned long end_pfn = early_node_map[i].end_pfn;
2832 if (start_pfn <= pfn && pfn < end_pfn)
2833 return early_node_map[i].nid;
2838 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
2840 /* Basic iterator support to walk early_node_map[] */
2841 #define for_each_active_range_index_in_nid(i, nid) \
2842 for (i = first_active_region_index_in_nid(nid); i != -1; \
2843 i = next_active_region_index_in_nid(i, nid))
2846 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
2847 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
2848 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
2850 * If an architecture guarantees that all ranges registered with
2851 * add_active_ranges() contain no holes and may be freed, this
2852 * this function may be used instead of calling free_bootmem() manually.
2854 void __init free_bootmem_with_active_regions(int nid,
2855 unsigned long max_low_pfn)
2859 for_each_active_range_index_in_nid(i, nid) {
2860 unsigned long size_pages = 0;
2861 unsigned long end_pfn = early_node_map[i].end_pfn;
2863 if (early_node_map[i].start_pfn >= max_low_pfn)
2866 if (end_pfn > max_low_pfn)
2867 end_pfn = max_low_pfn;
2869 size_pages = end_pfn - early_node_map[i].start_pfn;
2870 free_bootmem_node(NODE_DATA(early_node_map[i].nid),
2871 PFN_PHYS(early_node_map[i].start_pfn),
2872 size_pages << PAGE_SHIFT);
2877 * sparse_memory_present_with_active_regions - Call memory_present for each active range
2878 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
2880 * If an architecture guarantees that all ranges registered with
2881 * add_active_ranges() contain no holes and may be freed, this
2882 * function may be used instead of calling memory_present() manually.
2884 void __init sparse_memory_present_with_active_regions(int nid)
2888 for_each_active_range_index_in_nid(i, nid)
2889 memory_present(early_node_map[i].nid,
2890 early_node_map[i].start_pfn,
2891 early_node_map[i].end_pfn);
2895 * push_node_boundaries - Push node boundaries to at least the requested boundary
2896 * @nid: The nid of the node to push the boundary for
2897 * @start_pfn: The start pfn of the node
2898 * @end_pfn: The end pfn of the node
2900 * In reserve-based hot-add, mem_map is allocated that is unused until hotadd
2901 * time. Specifically, on x86_64, SRAT will report ranges that can potentially
2902 * be hotplugged even though no physical memory exists. This function allows
2903 * an arch to push out the node boundaries so mem_map is allocated that can
2906 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
2907 void __init push_node_boundaries(unsigned int nid,
2908 unsigned long start_pfn, unsigned long end_pfn)
2910 printk(KERN_DEBUG "Entering push_node_boundaries(%u, %lu, %lu)\n",
2911 nid, start_pfn, end_pfn);
2913 /* Initialise the boundary for this node if necessary */
2914 if (node_boundary_end_pfn[nid] == 0)
2915 node_boundary_start_pfn[nid] = -1UL;
2917 /* Update the boundaries */
2918 if (node_boundary_start_pfn[nid] > start_pfn)
2919 node_boundary_start_pfn[nid] = start_pfn;
2920 if (node_boundary_end_pfn[nid] < end_pfn)
2921 node_boundary_end_pfn[nid] = end_pfn;
2924 /* If necessary, push the node boundary out for reserve hotadd */
2925 static void __meminit account_node_boundary(unsigned int nid,
2926 unsigned long *start_pfn, unsigned long *end_pfn)
2928 printk(KERN_DEBUG "Entering account_node_boundary(%u, %lu, %lu)\n",
2929 nid, *start_pfn, *end_pfn);
2931 /* Return if boundary information has not been provided */
2932 if (node_boundary_end_pfn[nid] == 0)
2935 /* Check the boundaries and update if necessary */
2936 if (node_boundary_start_pfn[nid] < *start_pfn)
2937 *start_pfn = node_boundary_start_pfn[nid];
2938 if (node_boundary_end_pfn[nid] > *end_pfn)
2939 *end_pfn = node_boundary_end_pfn[nid];
2942 void __init push_node_boundaries(unsigned int nid,
2943 unsigned long start_pfn, unsigned long end_pfn) {}
2945 static void __meminit account_node_boundary(unsigned int nid,
2946 unsigned long *start_pfn, unsigned long *end_pfn) {}
2951 * get_pfn_range_for_nid - Return the start and end page frames for a node
2952 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
2953 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
2954 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
2956 * It returns the start and end page frame of a node based on information
2957 * provided by an arch calling add_active_range(). If called for a node
2958 * with no available memory, a warning is printed and the start and end
2961 void __meminit get_pfn_range_for_nid(unsigned int nid,
2962 unsigned long *start_pfn, unsigned long *end_pfn)
2968 for_each_active_range_index_in_nid(i, nid) {
2969 *start_pfn = min(*start_pfn, early_node_map[i].start_pfn);
2970 *end_pfn = max(*end_pfn, early_node_map[i].end_pfn);
2973 if (*start_pfn == -1UL)
2976 /* Push the node boundaries out if requested */
2977 account_node_boundary(nid, start_pfn, end_pfn);
2981 * This finds a zone that can be used for ZONE_MOVABLE pages. The
2982 * assumption is made that zones within a node are ordered in monotonic
2983 * increasing memory addresses so that the "highest" populated zone is used
2985 void __init find_usable_zone_for_movable(void)
2988 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
2989 if (zone_index == ZONE_MOVABLE)
2992 if (arch_zone_highest_possible_pfn[zone_index] >
2993 arch_zone_lowest_possible_pfn[zone_index])
2997 VM_BUG_ON(zone_index == -1);
2998 movable_zone = zone_index;
3002 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
3003 * because it is sized independant of architecture. Unlike the other zones,
3004 * the starting point for ZONE_MOVABLE is not fixed. It may be different
3005 * in each node depending on the size of each node and how evenly kernelcore
3006 * is distributed. This helper function adjusts the zone ranges
3007 * provided by the architecture for a given node by using the end of the
3008 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3009 * zones within a node are in order of monotonic increases memory addresses
3011 void __meminit adjust_zone_range_for_zone_movable(int nid,
3012 unsigned long zone_type,
3013 unsigned long node_start_pfn,
3014 unsigned long node_end_pfn,
3015 unsigned long *zone_start_pfn,
3016 unsigned long *zone_end_pfn)
3018 /* Only adjust if ZONE_MOVABLE is on this node */
3019 if (zone_movable_pfn[nid]) {
3020 /* Size ZONE_MOVABLE */
3021 if (zone_type == ZONE_MOVABLE) {
3022 *zone_start_pfn = zone_movable_pfn[nid];
3023 *zone_end_pfn = min(node_end_pfn,
3024 arch_zone_highest_possible_pfn[movable_zone]);
3026 /* Adjust for ZONE_MOVABLE starting within this range */
3027 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
3028 *zone_end_pfn > zone_movable_pfn[nid]) {
3029 *zone_end_pfn = zone_movable_pfn[nid];
3031 /* Check if this whole range is within ZONE_MOVABLE */
3032 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
3033 *zone_start_pfn = *zone_end_pfn;
3038 * Return the number of pages a zone spans in a node, including holes
3039 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3041 static unsigned long __meminit zone_spanned_pages_in_node(int nid,
3042 unsigned long zone_type,
3043 unsigned long *ignored)
3045 unsigned long node_start_pfn, node_end_pfn;
3046 unsigned long zone_start_pfn, zone_end_pfn;
3048 /* Get the start and end of the node and zone */
3049 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3050 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
3051 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
3052 adjust_zone_range_for_zone_movable(nid, zone_type,
3053 node_start_pfn, node_end_pfn,
3054 &zone_start_pfn, &zone_end_pfn);
3056 /* Check that this node has pages within the zone's required range */
3057 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
3060 /* Move the zone boundaries inside the node if necessary */
3061 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
3062 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
3064 /* Return the spanned pages */
3065 return zone_end_pfn - zone_start_pfn;
3069 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
3070 * then all holes in the requested range will be accounted for.
3072 unsigned long __meminit __absent_pages_in_range(int nid,
3073 unsigned long range_start_pfn,
3074 unsigned long range_end_pfn)
3077 unsigned long prev_end_pfn = 0, hole_pages = 0;
3078 unsigned long start_pfn;
3080 /* Find the end_pfn of the first active range of pfns in the node */
3081 i = first_active_region_index_in_nid(nid);
3085 prev_end_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3087 /* Account for ranges before physical memory on this node */
3088 if (early_node_map[i].start_pfn > range_start_pfn)
3089 hole_pages = prev_end_pfn - range_start_pfn;
3091 /* Find all holes for the zone within the node */
3092 for (; i != -1; i = next_active_region_index_in_nid(i, nid)) {
3094 /* No need to continue if prev_end_pfn is outside the zone */
3095 if (prev_end_pfn >= range_end_pfn)
3098 /* Make sure the end of the zone is not within the hole */
3099 start_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3100 prev_end_pfn = max(prev_end_pfn, range_start_pfn);
3102 /* Update the hole size cound and move on */
3103 if (start_pfn > range_start_pfn) {
3104 BUG_ON(prev_end_pfn > start_pfn);
3105 hole_pages += start_pfn - prev_end_pfn;
3107 prev_end_pfn = early_node_map[i].end_pfn;
3110 /* Account for ranges past physical memory on this node */
3111 if (range_end_pfn > prev_end_pfn)
3112 hole_pages += range_end_pfn -
3113 max(range_start_pfn, prev_end_pfn);
3119 * absent_pages_in_range - Return number of page frames in holes within a range
3120 * @start_pfn: The start PFN to start searching for holes
3121 * @end_pfn: The end PFN to stop searching for holes
3123 * It returns the number of pages frames in memory holes within a range.
3125 unsigned long __init absent_pages_in_range(unsigned long start_pfn,
3126 unsigned long end_pfn)
3128 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
3131 /* Return the number of page frames in holes in a zone on a node */
3132 static unsigned long __meminit zone_absent_pages_in_node(int nid,
3133 unsigned long zone_type,
3134 unsigned long *ignored)
3136 unsigned long node_start_pfn, node_end_pfn;
3137 unsigned long zone_start_pfn, zone_end_pfn;
3139 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3140 zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type],
3142 zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type],
3145 adjust_zone_range_for_zone_movable(nid, zone_type,
3146 node_start_pfn, node_end_pfn,
3147 &zone_start_pfn, &zone_end_pfn);
3148 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
3152 static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
3153 unsigned long zone_type,
3154 unsigned long *zones_size)
3156 return zones_size[zone_type];
3159 static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
3160 unsigned long zone_type,
3161 unsigned long *zholes_size)
3166 return zholes_size[zone_type];
3171 static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
3172 unsigned long *zones_size, unsigned long *zholes_size)
3174 unsigned long realtotalpages, totalpages = 0;
3177 for (i = 0; i < MAX_NR_ZONES; i++)
3178 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
3180 pgdat->node_spanned_pages = totalpages;
3182 realtotalpages = totalpages;
3183 for (i = 0; i < MAX_NR_ZONES; i++)
3185 zone_absent_pages_in_node(pgdat->node_id, i,
3187 pgdat->node_present_pages = realtotalpages;
3188 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
3192 #ifndef CONFIG_SPARSEMEM
3194 * Calculate the size of the zone->blockflags rounded to an unsigned long
3195 * Start by making sure zonesize is a multiple of MAX_ORDER-1 by rounding up
3196 * Then figure 1 NR_PAGEBLOCK_BITS worth of bits per MAX_ORDER-1, finally
3197 * round what is now in bits to nearest long in bits, then return it in
3200 static unsigned long __init usemap_size(unsigned long zonesize)
3202 unsigned long usemapsize;
3204 usemapsize = roundup(zonesize, MAX_ORDER_NR_PAGES);
3205 usemapsize = usemapsize >> (MAX_ORDER-1);
3206 usemapsize *= NR_PAGEBLOCK_BITS;
3207 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
3209 return usemapsize / 8;
3212 static void __init setup_usemap(struct pglist_data *pgdat,
3213 struct zone *zone, unsigned long zonesize)
3215 unsigned long usemapsize = usemap_size(zonesize);
3216 zone->pageblock_flags = NULL;
3218 zone->pageblock_flags = alloc_bootmem_node(pgdat, usemapsize);
3219 memset(zone->pageblock_flags, 0, usemapsize);
3223 static void inline setup_usemap(struct pglist_data *pgdat,
3224 struct zone *zone, unsigned long zonesize) {}
3225 #endif /* CONFIG_SPARSEMEM */
3228 * Set up the zone data structures:
3229 * - mark all pages reserved
3230 * - mark all memory queues empty
3231 * - clear the memory bitmaps
3233 static void __meminit free_area_init_core(struct pglist_data *pgdat,
3234 unsigned long *zones_size, unsigned long *zholes_size)
3237 int nid = pgdat->node_id;
3238 unsigned long zone_start_pfn = pgdat->node_start_pfn;
3241 pgdat_resize_init(pgdat);
3242 pgdat->nr_zones = 0;
3243 init_waitqueue_head(&pgdat->kswapd_wait);
3244 pgdat->kswapd_max_order = 0;
3246 for (j = 0; j < MAX_NR_ZONES; j++) {
3247 struct zone *zone = pgdat->node_zones + j;
3248 unsigned long size, realsize, memmap_pages;
3250 size = zone_spanned_pages_in_node(nid, j, zones_size);
3251 realsize = size - zone_absent_pages_in_node(nid, j,
3255 * Adjust realsize so that it accounts for how much memory
3256 * is used by this zone for memmap. This affects the watermark
3257 * and per-cpu initialisations
3259 memmap_pages = (size * sizeof(struct page)) >> PAGE_SHIFT;
3260 if (realsize >= memmap_pages) {
3261 realsize -= memmap_pages;
3263 " %s zone: %lu pages used for memmap\n",
3264 zone_names[j], memmap_pages);
3267 " %s zone: %lu pages exceeds realsize %lu\n",
3268 zone_names[j], memmap_pages, realsize);
3270 /* Account for reserved pages */
3271 if (j == 0 && realsize > dma_reserve) {
3272 realsize -= dma_reserve;
3273 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
3274 zone_names[0], dma_reserve);
3277 if (!is_highmem_idx(j))
3278 nr_kernel_pages += realsize;
3279 nr_all_pages += realsize;
3281 zone->spanned_pages = size;
3282 zone->present_pages = realsize;
3285 zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
3287 zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100;
3289 zone->name = zone_names[j];
3290 spin_lock_init(&zone->lock);
3291 spin_lock_init(&zone->lru_lock);
3292 zone_seqlock_init(zone);
3293 zone->zone_pgdat = pgdat;
3295 zone->prev_priority = DEF_PRIORITY;
3297 zone_pcp_init(zone);
3298 INIT_LIST_HEAD(&zone->active_list);
3299 INIT_LIST_HEAD(&zone->inactive_list);
3300 zone->nr_scan_active = 0;
3301 zone->nr_scan_inactive = 0;
3302 zap_zone_vm_stats(zone);
3303 atomic_set(&zone->reclaim_in_progress, 0);
3307 setup_usemap(pgdat, zone, size);
3308 ret = init_currently_empty_zone(zone, zone_start_pfn,
3309 size, MEMMAP_EARLY);
3311 zone_start_pfn += size;
3315 static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
3317 /* Skip empty nodes */
3318 if (!pgdat->node_spanned_pages)
3321 #ifdef CONFIG_FLAT_NODE_MEM_MAP
3322 /* ia64 gets its own node_mem_map, before this, without bootmem */
3323 if (!pgdat->node_mem_map) {
3324 unsigned long size, start, end;
3328 * The zone's endpoints aren't required to be MAX_ORDER
3329 * aligned but the node_mem_map endpoints must be in order
3330 * for the buddy allocator to function correctly.
3332 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
3333 end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
3334 end = ALIGN(end, MAX_ORDER_NR_PAGES);
3335 size = (end - start) * sizeof(struct page);
3336 map = alloc_remap(pgdat->node_id, size);
3338 map = alloc_bootmem_node(pgdat, size);
3339 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
3341 #ifndef CONFIG_NEED_MULTIPLE_NODES
3343 * With no DISCONTIG, the global mem_map is just set as node 0's
3345 if (pgdat == NODE_DATA(0)) {
3346 mem_map = NODE_DATA(0)->node_mem_map;
3347 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3348 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
3349 mem_map -= pgdat->node_start_pfn;
3350 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3353 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
3356 void __meminit free_area_init_node(int nid, struct pglist_data *pgdat,
3357 unsigned long *zones_size, unsigned long node_start_pfn,
3358 unsigned long *zholes_size)
3360 pgdat->node_id = nid;
3361 pgdat->node_start_pfn = node_start_pfn;
3362 calculate_node_totalpages(pgdat, zones_size, zholes_size);
3364 alloc_node_mem_map(pgdat);
3366 free_area_init_core(pgdat, zones_size, zholes_size);
3369 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3371 #if MAX_NUMNODES > 1
3373 * Figure out the number of possible node ids.
3375 static void __init setup_nr_node_ids(void)
3378 unsigned int highest = 0;
3380 for_each_node_mask(node, node_possible_map)
3382 nr_node_ids = highest + 1;
3385 static inline void setup_nr_node_ids(void)
3391 * add_active_range - Register a range of PFNs backed by physical memory
3392 * @nid: The node ID the range resides on
3393 * @start_pfn: The start PFN of the available physical memory
3394 * @end_pfn: The end PFN of the available physical memory
3396 * These ranges are stored in an early_node_map[] and later used by
3397 * free_area_init_nodes() to calculate zone sizes and holes. If the
3398 * range spans a memory hole, it is up to the architecture to ensure
3399 * the memory is not freed by the bootmem allocator. If possible
3400 * the range being registered will be merged with existing ranges.
3402 void __init add_active_range(unsigned int nid, unsigned long start_pfn,
3403 unsigned long end_pfn)
3407 printk(KERN_DEBUG "Entering add_active_range(%d, %lu, %lu) "
3408 "%d entries of %d used\n",
3409 nid, start_pfn, end_pfn,
3410 nr_nodemap_entries, MAX_ACTIVE_REGIONS);
3412 /* Merge with existing active regions if possible */
3413 for (i = 0; i < nr_nodemap_entries; i++) {
3414 if (early_node_map[i].nid != nid)
3417 /* Skip if an existing region covers this new one */
3418 if (start_pfn >= early_node_map[i].start_pfn &&
3419 end_pfn <= early_node_map[i].end_pfn)
3422 /* Merge forward if suitable */
3423 if (start_pfn <= early_node_map[i].end_pfn &&
3424 end_pfn > early_node_map[i].end_pfn) {
3425 early_node_map[i].end_pfn = end_pfn;
3429 /* Merge backward if suitable */
3430 if (start_pfn < early_node_map[i].end_pfn &&
3431 end_pfn >= early_node_map[i].start_pfn) {
3432 early_node_map[i].start_pfn = start_pfn;
3437 /* Check that early_node_map is large enough */
3438 if (i >= MAX_ACTIVE_REGIONS) {
3439 printk(KERN_CRIT "More than %d memory regions, truncating\n",
3440 MAX_ACTIVE_REGIONS);
3444 early_node_map[i].nid = nid;
3445 early_node_map[i].start_pfn = start_pfn;
3446 early_node_map[i].end_pfn = end_pfn;
3447 nr_nodemap_entries = i + 1;
3451 * shrink_active_range - Shrink an existing registered range of PFNs
3452 * @nid: The node id the range is on that should be shrunk
3453 * @old_end_pfn: The old end PFN of the range
3454 * @new_end_pfn: The new PFN of the range
3456 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
3457 * The map is kept at the end physical page range that has already been
3458 * registered with add_active_range(). This function allows an arch to shrink
3459 * an existing registered range.
3461 void __init shrink_active_range(unsigned int nid, unsigned long old_end_pfn,
3462 unsigned long new_end_pfn)
3466 /* Find the old active region end and shrink */
3467 for_each_active_range_index_in_nid(i, nid)
3468 if (early_node_map[i].end_pfn == old_end_pfn) {
3469 early_node_map[i].end_pfn = new_end_pfn;
3475 * remove_all_active_ranges - Remove all currently registered regions
3477 * During discovery, it may be found that a table like SRAT is invalid
3478 * and an alternative discovery method must be used. This function removes
3479 * all currently registered regions.
3481 void __init remove_all_active_ranges(void)
3483 memset(early_node_map, 0, sizeof(early_node_map));
3484 nr_nodemap_entries = 0;
3485 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
3486 memset(node_boundary_start_pfn, 0, sizeof(node_boundary_start_pfn));
3487 memset(node_boundary_end_pfn, 0, sizeof(node_boundary_end_pfn));
3488 #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
3491 /* Compare two active node_active_regions */
3492 static int __init cmp_node_active_region(const void *a, const void *b)
3494 struct node_active_region *arange = (struct node_active_region *)a;
3495 struct node_active_region *brange = (struct node_active_region *)b;
3497 /* Done this way to avoid overflows */
3498 if (arange->start_pfn > brange->start_pfn)
3500 if (arange->start_pfn < brange->start_pfn)
3506 /* sort the node_map by start_pfn */
3507 static void __init sort_node_map(void)
3509 sort(early_node_map, (size_t)nr_nodemap_entries,
3510 sizeof(struct node_active_region),
3511 cmp_node_active_region, NULL);
3514 /* Find the lowest pfn for a node */
3515 unsigned long __init find_min_pfn_for_node(unsigned long nid)
3518 unsigned long min_pfn = ULONG_MAX;
3520 /* Assuming a sorted map, the first range found has the starting pfn */
3521 for_each_active_range_index_in_nid(i, nid)
3522 min_pfn = min(min_pfn, early_node_map[i].start_pfn);
3524 if (min_pfn == ULONG_MAX) {
3526 "Could not find start_pfn for node %lu\n", nid);
3534 * find_min_pfn_with_active_regions - Find the minimum PFN registered
3536 * It returns the minimum PFN based on information provided via
3537 * add_active_range().
3539 unsigned long __init find_min_pfn_with_active_regions(void)
3541 return find_min_pfn_for_node(MAX_NUMNODES);
3545 * find_max_pfn_with_active_regions - Find the maximum PFN registered
3547 * It returns the maximum PFN based on information provided via
3548 * add_active_range().
3550 unsigned long __init find_max_pfn_with_active_regions(void)
3553 unsigned long max_pfn = 0;
3555 for (i = 0; i < nr_nodemap_entries; i++)
3556 max_pfn = max(max_pfn, early_node_map[i].end_pfn);
3562 * early_calculate_totalpages()
3563 * Sum pages in active regions for movable zone.
3564 * Populate N_HIGH_MEMORY for calculating usable_nodes.
3566 unsigned long __init early_calculate_totalpages(void)
3569 unsigned long totalpages = 0;
3571 for (i = 0; i < nr_nodemap_entries; i++) {
3572 unsigned long pages = early_node_map[i].end_pfn -
3573 early_node_map[i].start_pfn;
3574 totalpages += pages;
3576 node_set_state(early_node_map[i].nid, N_HIGH_MEMORY);
3582 * Find the PFN the Movable zone begins in each node. Kernel memory
3583 * is spread evenly between nodes as long as the nodes have enough
3584 * memory. When they don't, some nodes will have more kernelcore than
3587 void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn)
3590 unsigned long usable_startpfn;
3591 unsigned long kernelcore_node, kernelcore_remaining;
3592 unsigned long totalpages = early_calculate_totalpages();
3593 int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]);
3596 * If movablecore was specified, calculate what size of
3597 * kernelcore that corresponds so that memory usable for
3598 * any allocation type is evenly spread. If both kernelcore
3599 * and movablecore are specified, then the value of kernelcore
3600 * will be used for required_kernelcore if it's greater than
3601 * what movablecore would have allowed.
3603 if (required_movablecore) {
3604 unsigned long corepages;
3607 * Round-up so that ZONE_MOVABLE is at least as large as what
3608 * was requested by the user
3610 required_movablecore =
3611 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
3612 corepages = totalpages - required_movablecore;
3614 required_kernelcore = max(required_kernelcore, corepages);
3617 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
3618 if (!required_kernelcore)
3621 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
3622 find_usable_zone_for_movable();
3623 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
3626 /* Spread kernelcore memory as evenly as possible throughout nodes */
3627 kernelcore_node = required_kernelcore / usable_nodes;
3628 for_each_node_state(nid, N_HIGH_MEMORY) {
3630 * Recalculate kernelcore_node if the division per node
3631 * now exceeds what is necessary to satisfy the requested
3632 * amount of memory for the kernel
3634 if (required_kernelcore < kernelcore_node)
3635 kernelcore_node = required_kernelcore / usable_nodes;
3638 * As the map is walked, we track how much memory is usable
3639 * by the kernel using kernelcore_remaining. When it is
3640 * 0, the rest of the node is usable by ZONE_MOVABLE
3642 kernelcore_remaining = kernelcore_node;
3644 /* Go through each range of PFNs within this node */
3645 for_each_active_range_index_in_nid(i, nid) {
3646 unsigned long start_pfn, end_pfn;
3647 unsigned long size_pages;
3649 start_pfn = max(early_node_map[i].start_pfn,
3650 zone_movable_pfn[nid]);
3651 end_pfn = early_node_map[i].end_pfn;
3652 if (start_pfn >= end_pfn)
3655 /* Account for what is only usable for kernelcore */
3656 if (start_pfn < usable_startpfn) {
3657 unsigned long kernel_pages;
3658 kernel_pages = min(end_pfn, usable_startpfn)
3661 kernelcore_remaining -= min(kernel_pages,
3662 kernelcore_remaining);
3663 required_kernelcore -= min(kernel_pages,
3664 required_kernelcore);
3666 /* Continue if range is now fully accounted */
3667 if (end_pfn <= usable_startpfn) {
3670 * Push zone_movable_pfn to the end so
3671 * that if we have to rebalance
3672 * kernelcore across nodes, we will
3673 * not double account here
3675 zone_movable_pfn[nid] = end_pfn;
3678 start_pfn = usable_startpfn;
3682 * The usable PFN range for ZONE_MOVABLE is from
3683 * start_pfn->end_pfn. Calculate size_pages as the
3684 * number of pages used as kernelcore
3686 size_pages = end_pfn - start_pfn;
3687 if (size_pages > kernelcore_remaining)
3688 size_pages = kernelcore_remaining;
3689 zone_movable_pfn[nid] = start_pfn + size_pages;
3692 * Some kernelcore has been met, update counts and
3693 * break if the kernelcore for this node has been
3696 required_kernelcore -= min(required_kernelcore,
3698 kernelcore_remaining -= size_pages;
3699 if (!kernelcore_remaining)
3705 * If there is still required_kernelcore, we do another pass with one
3706 * less node in the count. This will push zone_movable_pfn[nid] further
3707 * along on the nodes that still have memory until kernelcore is
3711 if (usable_nodes && required_kernelcore > usable_nodes)
3714 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
3715 for (nid = 0; nid < MAX_NUMNODES; nid++)
3716 zone_movable_pfn[nid] =
3717 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
3720 /* Any regular memory on that node ? */
3721 static void check_for_regular_memory(pg_data_t *pgdat)
3723 #ifdef CONFIG_HIGHMEM
3724 enum zone_type zone_type;
3726 for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) {
3727 struct zone *zone = &pgdat->node_zones[zone_type];
3728 if (zone->present_pages)
3729 node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY);
3735 * free_area_init_nodes - Initialise all pg_data_t and zone data
3736 * @max_zone_pfn: an array of max PFNs for each zone
3738 * This will call free_area_init_node() for each active node in the system.
3739 * Using the page ranges provided by add_active_range(), the size of each
3740 * zone in each node and their holes is calculated. If the maximum PFN
3741 * between two adjacent zones match, it is assumed that the zone is empty.
3742 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
3743 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
3744 * starts where the previous one ended. For example, ZONE_DMA32 starts
3745 * at arch_max_dma_pfn.
3747 void __init free_area_init_nodes(unsigned long *max_zone_pfn)
3752 /* Sort early_node_map as initialisation assumes it is sorted */
3755 /* Record where the zone boundaries are */
3756 memset(arch_zone_lowest_possible_pfn, 0,
3757 sizeof(arch_zone_lowest_possible_pfn));
3758 memset(arch_zone_highest_possible_pfn, 0,
3759 sizeof(arch_zone_highest_possible_pfn));
3760 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
3761 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
3762 for (i = 1; i < MAX_NR_ZONES; i++) {
3763 if (i == ZONE_MOVABLE)
3765 arch_zone_lowest_possible_pfn[i] =
3766 arch_zone_highest_possible_pfn[i-1];
3767 arch_zone_highest_possible_pfn[i] =
3768 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
3770 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
3771 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
3773 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
3774 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
3775 find_zone_movable_pfns_for_nodes(zone_movable_pfn);
3777 /* Print out the zone ranges */
3778 printk("Zone PFN ranges:\n");
3779 for (i = 0; i < MAX_NR_ZONES; i++) {
3780 if (i == ZONE_MOVABLE)
3782 printk(" %-8s %8lu -> %8lu\n",
3784 arch_zone_lowest_possible_pfn[i],
3785 arch_zone_highest_possible_pfn[i]);
3788 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
3789 printk("Movable zone start PFN for each node\n");
3790 for (i = 0; i < MAX_NUMNODES; i++) {
3791 if (zone_movable_pfn[i])
3792 printk(" Node %d: %lu\n", i, zone_movable_pfn[i]);
3795 /* Print out the early_node_map[] */
3796 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries);
3797 for (i = 0; i < nr_nodemap_entries; i++)
3798 printk(" %3d: %8lu -> %8lu\n", early_node_map[i].nid,
3799 early_node_map[i].start_pfn,
3800 early_node_map[i].end_pfn);
3802 /* Initialise every node */
3803 setup_nr_node_ids();
3804 for_each_online_node(nid) {
3805 pg_data_t *pgdat = NODE_DATA(nid);
3806 free_area_init_node(nid, pgdat, NULL,
3807 find_min_pfn_for_node(nid), NULL);
3809 /* Any memory on that node */
3810 if (pgdat->node_present_pages)
3811 node_set_state(nid, N_HIGH_MEMORY);
3812 check_for_regular_memory(pgdat);
3816 static int __init cmdline_parse_core(char *p, unsigned long *core)
3818 unsigned long long coremem;
3822 coremem = memparse(p, &p);
3823 *core = coremem >> PAGE_SHIFT;
3825 /* Paranoid check that UL is enough for the coremem value */
3826 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
3832 * kernelcore=size sets the amount of memory for use for allocations that
3833 * cannot be reclaimed or migrated.
3835 static int __init cmdline_parse_kernelcore(char *p)
3837 return cmdline_parse_core(p, &required_kernelcore);
3841 * movablecore=size sets the amount of memory for use for allocations that
3842 * can be reclaimed or migrated.
3844 static int __init cmdline_parse_movablecore(char *p)
3846 return cmdline_parse_core(p, &required_movablecore);
3849 early_param("kernelcore", cmdline_parse_kernelcore);
3850 early_param("movablecore", cmdline_parse_movablecore);
3852 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3855 * set_dma_reserve - set the specified number of pages reserved in the first zone
3856 * @new_dma_reserve: The number of pages to mark reserved
3858 * The per-cpu batchsize and zone watermarks are determined by present_pages.
3859 * In the DMA zone, a significant percentage may be consumed by kernel image
3860 * and other unfreeable allocations which can skew the watermarks badly. This
3861 * function may optionally be used to account for unfreeable pages in the
3862 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
3863 * smaller per-cpu batchsize.
3865 void __init set_dma_reserve(unsigned long new_dma_reserve)
3867 dma_reserve = new_dma_reserve;
3870 #ifndef CONFIG_NEED_MULTIPLE_NODES
3871 static bootmem_data_t contig_bootmem_data;
3872 struct pglist_data contig_page_data = { .bdata = &contig_bootmem_data };
3874 EXPORT_SYMBOL(contig_page_data);
3877 void __init free_area_init(unsigned long *zones_size)
3879 free_area_init_node(0, NODE_DATA(0), zones_size,
3880 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
3883 static int page_alloc_cpu_notify(struct notifier_block *self,
3884 unsigned long action, void *hcpu)
3886 int cpu = (unsigned long)hcpu;
3888 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
3889 local_irq_disable();
3891 vm_events_fold_cpu(cpu);
3893 refresh_cpu_vm_stats(cpu);
3898 void __init page_alloc_init(void)
3900 hotcpu_notifier(page_alloc_cpu_notify, 0);
3904 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
3905 * or min_free_kbytes changes.
3907 static void calculate_totalreserve_pages(void)
3909 struct pglist_data *pgdat;
3910 unsigned long reserve_pages = 0;
3911 enum zone_type i, j;
3913 for_each_online_pgdat(pgdat) {
3914 for (i = 0; i < MAX_NR_ZONES; i++) {
3915 struct zone *zone = pgdat->node_zones + i;
3916 unsigned long max = 0;
3918 /* Find valid and maximum lowmem_reserve in the zone */
3919 for (j = i; j < MAX_NR_ZONES; j++) {
3920 if (zone->lowmem_reserve[j] > max)
3921 max = zone->lowmem_reserve[j];
3924 /* we treat pages_high as reserved pages. */
3925 max += zone->pages_high;
3927 if (max > zone->present_pages)
3928 max = zone->present_pages;
3929 reserve_pages += max;
3932 totalreserve_pages = reserve_pages;
3936 * setup_per_zone_lowmem_reserve - called whenever
3937 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
3938 * has a correct pages reserved value, so an adequate number of
3939 * pages are left in the zone after a successful __alloc_pages().
3941 static void setup_per_zone_lowmem_reserve(void)
3943 struct pglist_data *pgdat;
3944 enum zone_type j, idx;
3946 for_each_online_pgdat(pgdat) {
3947 for (j = 0; j < MAX_NR_ZONES; j++) {
3948 struct zone *zone = pgdat->node_zones + j;
3949 unsigned long present_pages = zone->present_pages;
3951 zone->lowmem_reserve[j] = 0;
3955 struct zone *lower_zone;
3959 if (sysctl_lowmem_reserve_ratio[idx] < 1)
3960 sysctl_lowmem_reserve_ratio[idx] = 1;
3962 lower_zone = pgdat->node_zones + idx;
3963 lower_zone->lowmem_reserve[j] = present_pages /
3964 sysctl_lowmem_reserve_ratio[idx];
3965 present_pages += lower_zone->present_pages;
3970 /* update totalreserve_pages */
3971 calculate_totalreserve_pages();
3975 * setup_per_zone_pages_min - called when min_free_kbytes changes.
3977 * Ensures that the pages_{min,low,high} values for each zone are set correctly
3978 * with respect to min_free_kbytes.
3980 void setup_per_zone_pages_min(void)
3982 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
3983 unsigned long lowmem_pages = 0;
3985 unsigned long flags;
3987 /* Calculate total number of !ZONE_HIGHMEM pages */
3988 for_each_zone(zone) {
3989 if (!is_highmem(zone))
3990 lowmem_pages += zone->present_pages;
3993 for_each_zone(zone) {
3996 spin_lock_irqsave(&zone->lru_lock, flags);
3997 tmp = (u64)pages_min * zone->present_pages;
3998 do_div(tmp, lowmem_pages);
3999 if (is_highmem(zone)) {
4001 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
4002 * need highmem pages, so cap pages_min to a small
4005 * The (pages_high-pages_low) and (pages_low-pages_min)
4006 * deltas controls asynch page reclaim, and so should
4007 * not be capped for highmem.
4011 min_pages = zone->present_pages / 1024;
4012 if (min_pages < SWAP_CLUSTER_MAX)
4013 min_pages = SWAP_CLUSTER_MAX;
4014 if (min_pages > 128)
4016 zone->pages_min = min_pages;
4019 * If it's a lowmem zone, reserve a number of pages
4020 * proportionate to the zone's size.
4022 zone->pages_min = tmp;
4025 zone->pages_low = zone->pages_min + (tmp >> 2);
4026 zone->pages_high = zone->pages_min + (tmp >> 1);
4027 spin_unlock_irqrestore(&zone->lru_lock, flags);
4030 /* update totalreserve_pages */
4031 calculate_totalreserve_pages();
4035 * Initialise min_free_kbytes.
4037 * For small machines we want it small (128k min). For large machines
4038 * we want it large (64MB max). But it is not linear, because network
4039 * bandwidth does not increase linearly with machine size. We use
4041 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
4042 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
4058 static int __init init_per_zone_pages_min(void)
4060 unsigned long lowmem_kbytes;
4062 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
4064 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
4065 if (min_free_kbytes < 128)
4066 min_free_kbytes = 128;
4067 if (min_free_kbytes > 65536)
4068 min_free_kbytes = 65536;
4069 setup_per_zone_pages_min();
4070 setup_per_zone_lowmem_reserve();
4073 module_init(init_per_zone_pages_min)
4076 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
4077 * that we can call two helper functions whenever min_free_kbytes
4080 int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
4081 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4083 proc_dointvec(table, write, file, buffer, length, ppos);
4085 setup_per_zone_pages_min();
4090 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
4091 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4096 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4101 zone->min_unmapped_pages = (zone->present_pages *
4102 sysctl_min_unmapped_ratio) / 100;
4106 int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
4107 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4112 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4117 zone->min_slab_pages = (zone->present_pages *
4118 sysctl_min_slab_ratio) / 100;
4124 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
4125 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
4126 * whenever sysctl_lowmem_reserve_ratio changes.
4128 * The reserve ratio obviously has absolutely no relation with the
4129 * pages_min watermarks. The lowmem reserve ratio can only make sense
4130 * if in function of the boot time zone sizes.
4132 int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
4133 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4135 proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4136 setup_per_zone_lowmem_reserve();
4141 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
4142 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
4143 * can have before it gets flushed back to buddy allocator.
4146 int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
4147 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4153 ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4154 if (!write || (ret == -EINVAL))
4156 for_each_zone(zone) {
4157 for_each_online_cpu(cpu) {
4159 high = zone->present_pages / percpu_pagelist_fraction;
4160 setup_pagelist_highmark(zone_pcp(zone, cpu), high);
4166 int hashdist = HASHDIST_DEFAULT;
4169 static int __init set_hashdist(char *str)
4173 hashdist = simple_strtoul(str, &str, 0);
4176 __setup("hashdist=", set_hashdist);
4180 * allocate a large system hash table from bootmem
4181 * - it is assumed that the hash table must contain an exact power-of-2
4182 * quantity of entries
4183 * - limit is the number of hash buckets, not the total allocation size
4185 void *__init alloc_large_system_hash(const char *tablename,
4186 unsigned long bucketsize,
4187 unsigned long numentries,
4190 unsigned int *_hash_shift,
4191 unsigned int *_hash_mask,
4192 unsigned long limit)
4194 unsigned long long max = limit;
4195 unsigned long log2qty, size;
4198 /* allow the kernel cmdline to have a say */
4200 /* round applicable memory size up to nearest megabyte */
4201 numentries = nr_kernel_pages;
4202 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
4203 numentries >>= 20 - PAGE_SHIFT;
4204 numentries <<= 20 - PAGE_SHIFT;
4206 /* limit to 1 bucket per 2^scale bytes of low memory */
4207 if (scale > PAGE_SHIFT)
4208 numentries >>= (scale - PAGE_SHIFT);
4210 numentries <<= (PAGE_SHIFT - scale);
4212 /* Make sure we've got at least a 0-order allocation.. */
4213 if (unlikely((numentries * bucketsize) < PAGE_SIZE))
4214 numentries = PAGE_SIZE / bucketsize;
4216 numentries = roundup_pow_of_two(numentries);
4218 /* limit allocation size to 1/16 total memory by default */
4220 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
4221 do_div(max, bucketsize);
4224 if (numentries > max)
4227 log2qty = ilog2(numentries);
4230 size = bucketsize << log2qty;
4231 if (flags & HASH_EARLY)
4232 table = alloc_bootmem(size);
4234 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
4236 unsigned long order;
4237 for (order = 0; ((1UL << order) << PAGE_SHIFT) < size; order++)
4239 table = (void*) __get_free_pages(GFP_ATOMIC, order);
4241 * If bucketsize is not a power-of-two, we may free
4242 * some pages at the end of hash table.
4245 unsigned long alloc_end = (unsigned long)table +
4246 (PAGE_SIZE << order);
4247 unsigned long used = (unsigned long)table +
4249 split_page(virt_to_page(table), order);
4250 while (used < alloc_end) {
4256 } while (!table && size > PAGE_SIZE && --log2qty);
4259 panic("Failed to allocate %s hash table\n", tablename);
4261 printk(KERN_INFO "%s hash table entries: %d (order: %d, %lu bytes)\n",
4264 ilog2(size) - PAGE_SHIFT,
4268 *_hash_shift = log2qty;
4270 *_hash_mask = (1 << log2qty) - 1;
4275 #ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE
4276 struct page *pfn_to_page(unsigned long pfn)
4278 return __pfn_to_page(pfn);
4280 unsigned long page_to_pfn(struct page *page)
4282 return __page_to_pfn(page);
4284 EXPORT_SYMBOL(pfn_to_page);
4285 EXPORT_SYMBOL(page_to_pfn);
4286 #endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */
4288 /* Return a pointer to the bitmap storing bits affecting a block of pages */
4289 static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
4292 #ifdef CONFIG_SPARSEMEM
4293 return __pfn_to_section(pfn)->pageblock_flags;
4295 return zone->pageblock_flags;
4296 #endif /* CONFIG_SPARSEMEM */
4299 static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
4301 #ifdef CONFIG_SPARSEMEM
4302 pfn &= (PAGES_PER_SECTION-1);
4303 return (pfn >> (MAX_ORDER-1)) * NR_PAGEBLOCK_BITS;
4305 pfn = pfn - zone->zone_start_pfn;
4306 return (pfn >> (MAX_ORDER-1)) * NR_PAGEBLOCK_BITS;
4307 #endif /* CONFIG_SPARSEMEM */
4311 * get_pageblock_flags_group - Return the requested group of flags for the MAX_ORDER_NR_PAGES block of pages
4312 * @page: The page within the block of interest
4313 * @start_bitidx: The first bit of interest to retrieve
4314 * @end_bitidx: The last bit of interest
4315 * returns pageblock_bits flags
4317 unsigned long get_pageblock_flags_group(struct page *page,
4318 int start_bitidx, int end_bitidx)
4321 unsigned long *bitmap;
4322 unsigned long pfn, bitidx;
4323 unsigned long flags = 0;
4324 unsigned long value = 1;
4326 zone = page_zone(page);
4327 pfn = page_to_pfn(page);
4328 bitmap = get_pageblock_bitmap(zone, pfn);
4329 bitidx = pfn_to_bitidx(zone, pfn);
4331 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
4332 if (test_bit(bitidx + start_bitidx, bitmap))
4339 * set_pageblock_flags_group - Set the requested group of flags for a MAX_ORDER_NR_PAGES block of pages
4340 * @page: The page within the block of interest
4341 * @start_bitidx: The first bit of interest
4342 * @end_bitidx: The last bit of interest
4343 * @flags: The flags to set
4345 void set_pageblock_flags_group(struct page *page, unsigned long flags,
4346 int start_bitidx, int end_bitidx)
4349 unsigned long *bitmap;
4350 unsigned long pfn, bitidx;
4351 unsigned long value = 1;
4353 zone = page_zone(page);
4354 pfn = page_to_pfn(page);
4355 bitmap = get_pageblock_bitmap(zone, pfn);
4356 bitidx = pfn_to_bitidx(zone, pfn);
4358 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
4360 __set_bit(bitidx + start_bitidx, bitmap);
4362 __clear_bit(bitidx + start_bitidx, bitmap);