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 int page_group_by_mobility_disabled __read_mostly;
164 static inline int get_pageblock_migratetype(struct page *page)
166 if (unlikely(page_group_by_mobility_disabled))
167 return MIGRATE_UNMOVABLE;
169 return get_pageblock_flags_group(page, PB_migrate, PB_migrate_end);
172 static void set_pageblock_migratetype(struct page *page, int migratetype)
174 set_pageblock_flags_group(page, (unsigned long)migratetype,
175 PB_migrate, PB_migrate_end);
178 static inline int allocflags_to_migratetype(gfp_t gfp_flags, int order)
180 WARN_ON((gfp_flags & GFP_MOVABLE_MASK) == GFP_MOVABLE_MASK);
182 if (unlikely(page_group_by_mobility_disabled))
183 return MIGRATE_UNMOVABLE;
185 /* Cluster high-order atomic allocations together */
186 if (unlikely(order > 0) &&
187 (!(gfp_flags & __GFP_WAIT) || in_interrupt()))
188 return MIGRATE_HIGHATOMIC;
190 /* Cluster based on mobility */
191 return (((gfp_flags & __GFP_MOVABLE) != 0) << 1) |
192 ((gfp_flags & __GFP_RECLAIMABLE) != 0);
196 static inline int get_pageblock_migratetype(struct page *page)
198 return MIGRATE_UNMOVABLE;
201 static void set_pageblock_migratetype(struct page *page, int migratetype)
205 static inline int allocflags_to_migratetype(gfp_t gfp_flags, int order)
207 return MIGRATE_UNMOVABLE;
209 #endif /* CONFIG_PAGE_GROUP_BY_MOBILITY */
211 #ifdef CONFIG_DEBUG_VM
212 static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
216 unsigned long pfn = page_to_pfn(page);
219 seq = zone_span_seqbegin(zone);
220 if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
222 else if (pfn < zone->zone_start_pfn)
224 } while (zone_span_seqretry(zone, seq));
229 static int page_is_consistent(struct zone *zone, struct page *page)
231 if (!pfn_valid_within(page_to_pfn(page)))
233 if (zone != page_zone(page))
239 * Temporary debugging check for pages not lying within a given zone.
241 static int bad_range(struct zone *zone, struct page *page)
243 if (page_outside_zone_boundaries(zone, page))
245 if (!page_is_consistent(zone, page))
251 static inline int bad_range(struct zone *zone, struct page *page)
257 static void bad_page(struct page *page)
259 printk(KERN_EMERG "Bad page state in process '%s'\n"
260 KERN_EMERG "page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n"
261 KERN_EMERG "Trying to fix it up, but a reboot is needed\n"
262 KERN_EMERG "Backtrace:\n",
263 current->comm, page, (int)(2*sizeof(unsigned long)),
264 (unsigned long)page->flags, page->mapping,
265 page_mapcount(page), page_count(page));
267 page->flags &= ~(1 << PG_lru |
277 set_page_count(page, 0);
278 reset_page_mapcount(page);
279 page->mapping = NULL;
280 add_taint(TAINT_BAD_PAGE);
284 * Higher-order pages are called "compound pages". They are structured thusly:
286 * The first PAGE_SIZE page is called the "head page".
288 * The remaining PAGE_SIZE pages are called "tail pages".
290 * All pages have PG_compound set. All pages have their ->private pointing at
291 * the head page (even the head page has this).
293 * The first tail page's ->lru.next holds the address of the compound page's
294 * put_page() function. Its ->lru.prev holds the order of allocation.
295 * This usage means that zero-order pages may not be compound.
298 static void free_compound_page(struct page *page)
300 __free_pages_ok(page, compound_order(page));
303 static void prep_compound_page(struct page *page, unsigned long order)
306 int nr_pages = 1 << order;
308 set_compound_page_dtor(page, free_compound_page);
309 set_compound_order(page, order);
311 for (i = 1; i < nr_pages; i++) {
312 struct page *p = page + i;
315 p->first_page = page;
319 static void destroy_compound_page(struct page *page, unsigned long order)
322 int nr_pages = 1 << order;
324 if (unlikely(compound_order(page) != order))
327 if (unlikely(!PageHead(page)))
329 __ClearPageHead(page);
330 for (i = 1; i < nr_pages; i++) {
331 struct page *p = page + i;
333 if (unlikely(!PageTail(p) |
334 (p->first_page != page)))
340 static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
344 VM_BUG_ON((gfp_flags & (__GFP_WAIT | __GFP_HIGHMEM)) == __GFP_HIGHMEM);
346 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
347 * and __GFP_HIGHMEM from hard or soft interrupt context.
349 VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
350 for (i = 0; i < (1 << order); i++)
351 clear_highpage(page + i);
355 * function for dealing with page's order in buddy system.
356 * zone->lock is already acquired when we use these.
357 * So, we don't need atomic page->flags operations here.
359 static inline unsigned long page_order(struct page *page)
361 return page_private(page);
364 static inline void set_page_order(struct page *page, int order)
366 set_page_private(page, order);
367 __SetPageBuddy(page);
370 static inline void rmv_page_order(struct page *page)
372 __ClearPageBuddy(page);
373 set_page_private(page, 0);
377 * Locate the struct page for both the matching buddy in our
378 * pair (buddy1) and the combined O(n+1) page they form (page).
380 * 1) Any buddy B1 will have an order O twin B2 which satisfies
381 * the following equation:
383 * For example, if the starting buddy (buddy2) is #8 its order
385 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
387 * 2) Any buddy B will have an order O+1 parent P which
388 * satisfies the following equation:
391 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
393 static inline struct page *
394 __page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order)
396 unsigned long buddy_idx = page_idx ^ (1 << order);
398 return page + (buddy_idx - page_idx);
401 static inline unsigned long
402 __find_combined_index(unsigned long page_idx, unsigned int order)
404 return (page_idx & ~(1 << order));
408 * This function checks whether a page is free && is the buddy
409 * we can do coalesce a page and its buddy if
410 * (a) the buddy is not in a hole &&
411 * (b) the buddy is in the buddy system &&
412 * (c) a page and its buddy have the same order &&
413 * (d) a page and its buddy are in the same zone.
415 * For recording whether a page is in the buddy system, we use PG_buddy.
416 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
418 * For recording page's order, we use page_private(page).
420 static inline int page_is_buddy(struct page *page, struct page *buddy,
423 if (!pfn_valid_within(page_to_pfn(buddy)))
426 if (page_zone_id(page) != page_zone_id(buddy))
429 if (PageBuddy(buddy) && page_order(buddy) == order) {
430 BUG_ON(page_count(buddy) != 0);
437 * Freeing function for a buddy system allocator.
439 * The concept of a buddy system is to maintain direct-mapped table
440 * (containing bit values) for memory blocks of various "orders".
441 * The bottom level table contains the map for the smallest allocatable
442 * units of memory (here, pages), and each level above it describes
443 * pairs of units from the levels below, hence, "buddies".
444 * At a high level, all that happens here is marking the table entry
445 * at the bottom level available, and propagating the changes upward
446 * as necessary, plus some accounting needed to play nicely with other
447 * parts of the VM system.
448 * At each level, we keep a list of pages, which are heads of continuous
449 * free pages of length of (1 << order) and marked with PG_buddy. Page's
450 * order is recorded in page_private(page) field.
451 * So when we are allocating or freeing one, we can derive the state of the
452 * other. That is, if we allocate a small block, and both were
453 * free, the remainder of the region must be split into blocks.
454 * If a block is freed, and its buddy is also free, then this
455 * triggers coalescing into a block of larger size.
460 static inline void __free_one_page(struct page *page,
461 struct zone *zone, unsigned int order)
463 unsigned long page_idx;
464 int order_size = 1 << order;
465 int migratetype = get_pageblock_migratetype(page);
467 if (unlikely(PageCompound(page)))
468 destroy_compound_page(page, order);
470 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
472 VM_BUG_ON(page_idx & (order_size - 1));
473 VM_BUG_ON(bad_range(zone, page));
475 __mod_zone_page_state(zone, NR_FREE_PAGES, order_size);
476 while (order < MAX_ORDER-1) {
477 unsigned long combined_idx;
480 buddy = __page_find_buddy(page, page_idx, order);
481 if (!page_is_buddy(page, buddy, order))
482 break; /* Move the buddy up one level. */
484 list_del(&buddy->lru);
485 zone->free_area[order].nr_free--;
486 rmv_page_order(buddy);
487 combined_idx = __find_combined_index(page_idx, order);
488 page = page + (combined_idx - page_idx);
489 page_idx = combined_idx;
492 set_page_order(page, order);
494 &zone->free_area[order].free_list[migratetype]);
495 zone->free_area[order].nr_free++;
498 static inline int free_pages_check(struct page *page)
500 if (unlikely(page_mapcount(page) |
501 (page->mapping != NULL) |
502 (page_count(page) != 0) |
515 __ClearPageDirty(page);
517 * For now, we report if PG_reserved was found set, but do not
518 * clear it, and do not free the page. But we shall soon need
519 * to do more, for when the ZERO_PAGE count wraps negative.
521 return PageReserved(page);
525 * Frees a list of pages.
526 * Assumes all pages on list are in same zone, and of same order.
527 * count is the number of pages to free.
529 * If the zone was previously in an "all pages pinned" state then look to
530 * see if this freeing clears that state.
532 * And clear the zone's pages_scanned counter, to hold off the "all pages are
533 * pinned" detection logic.
535 static void free_pages_bulk(struct zone *zone, int count,
536 struct list_head *list, int order)
538 spin_lock(&zone->lock);
539 zone->all_unreclaimable = 0;
540 zone->pages_scanned = 0;
544 VM_BUG_ON(list_empty(list));
545 page = list_entry(list->prev, struct page, lru);
546 /* have to delete it as __free_one_page list manipulates */
547 list_del(&page->lru);
548 __free_one_page(page, zone, order);
550 spin_unlock(&zone->lock);
553 static void free_one_page(struct zone *zone, struct page *page, int order)
555 spin_lock(&zone->lock);
556 zone->all_unreclaimable = 0;
557 zone->pages_scanned = 0;
558 __free_one_page(page, zone, order);
559 spin_unlock(&zone->lock);
562 static void __free_pages_ok(struct page *page, unsigned int order)
568 for (i = 0 ; i < (1 << order) ; ++i)
569 reserved += free_pages_check(page + i);
573 if (!PageHighMem(page))
574 debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order);
575 arch_free_page(page, order);
576 kernel_map_pages(page, 1 << order, 0);
578 local_irq_save(flags);
579 __count_vm_events(PGFREE, 1 << order);
580 free_one_page(page_zone(page), page, order);
581 local_irq_restore(flags);
585 * permit the bootmem allocator to evade page validation on high-order frees
587 void fastcall __init __free_pages_bootmem(struct page *page, unsigned int order)
590 __ClearPageReserved(page);
591 set_page_count(page, 0);
592 set_page_refcounted(page);
598 for (loop = 0; loop < BITS_PER_LONG; loop++) {
599 struct page *p = &page[loop];
601 if (loop + 1 < BITS_PER_LONG)
603 __ClearPageReserved(p);
604 set_page_count(p, 0);
607 set_page_refcounted(page);
608 __free_pages(page, order);
614 * The order of subdivision here is critical for the IO subsystem.
615 * Please do not alter this order without good reasons and regression
616 * testing. Specifically, as large blocks of memory are subdivided,
617 * the order in which smaller blocks are delivered depends on the order
618 * they're subdivided in this function. This is the primary factor
619 * influencing the order in which pages are delivered to the IO
620 * subsystem according to empirical testing, and this is also justified
621 * by considering the behavior of a buddy system containing a single
622 * large block of memory acted on by a series of small allocations.
623 * This behavior is a critical factor in sglist merging's success.
627 static inline void expand(struct zone *zone, struct page *page,
628 int low, int high, struct free_area *area,
631 unsigned long size = 1 << high;
637 VM_BUG_ON(bad_range(zone, &page[size]));
638 list_add(&page[size].lru, &area->free_list[migratetype]);
640 set_page_order(&page[size], high);
645 * This page is about to be returned from the page allocator
647 static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
649 if (unlikely(page_mapcount(page) |
650 (page->mapping != NULL) |
651 (page_count(page) != 0) |
666 * For now, we report if PG_reserved was found set, but do not
667 * clear it, and do not allocate the page: as a safety net.
669 if (PageReserved(page))
672 page->flags &= ~(1 << PG_uptodate | 1 << PG_error | 1 << PG_readahead |
673 1 << PG_referenced | 1 << PG_arch_1 |
674 1 << PG_owner_priv_1 | 1 << PG_mappedtodisk);
675 set_page_private(page, 0);
676 set_page_refcounted(page);
678 arch_alloc_page(page, order);
679 kernel_map_pages(page, 1 << order, 1);
681 if (gfp_flags & __GFP_ZERO)
682 prep_zero_page(page, order, gfp_flags);
684 if (order && (gfp_flags & __GFP_COMP))
685 prep_compound_page(page, order);
690 #ifdef CONFIG_PAGE_GROUP_BY_MOBILITY
692 * This array describes the order lists are fallen back to when
693 * the free lists for the desirable migrate type are depleted
695 static int fallbacks[MIGRATE_TYPES][MIGRATE_TYPES-1] = {
696 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_HIGHATOMIC },
697 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_HIGHATOMIC },
698 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE,MIGRATE_HIGHATOMIC },
699 [MIGRATE_HIGHATOMIC] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE,MIGRATE_MOVABLE},
703 * Move the free pages in a range to the free lists of the requested type.
704 * Note that start_page and end_pages are not aligned in a MAX_ORDER_NR_PAGES
705 * boundary. If alignment is required, use move_freepages_block()
707 int move_freepages(struct zone *zone,
708 struct page *start_page, struct page *end_page,
713 int blocks_moved = 0;
715 #ifndef CONFIG_HOLES_IN_ZONE
717 * page_zone is not safe to call in this context when
718 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
719 * anyway as we check zone boundaries in move_freepages_block().
720 * Remove at a later date when no bug reports exist related to
721 * CONFIG_PAGE_GROUP_BY_MOBILITY
723 BUG_ON(page_zone(start_page) != page_zone(end_page));
726 for (page = start_page; page <= end_page;) {
727 if (!pfn_valid_within(page_to_pfn(page))) {
732 if (!PageBuddy(page)) {
737 order = page_order(page);
738 list_del(&page->lru);
740 &zone->free_area[order].free_list[migratetype]);
748 int move_freepages_block(struct zone *zone, struct page *page, int migratetype)
750 unsigned long start_pfn, end_pfn;
751 struct page *start_page, *end_page;
753 start_pfn = page_to_pfn(page);
754 start_pfn = start_pfn & ~(MAX_ORDER_NR_PAGES-1);
755 start_page = pfn_to_page(start_pfn);
756 end_page = start_page + MAX_ORDER_NR_PAGES - 1;
757 end_pfn = start_pfn + MAX_ORDER_NR_PAGES - 1;
759 /* Do not cross zone boundaries */
760 if (start_pfn < zone->zone_start_pfn)
762 if (end_pfn >= zone->zone_start_pfn + zone->spanned_pages)
765 return move_freepages(zone, start_page, end_page, migratetype);
768 /* Return the page with the lowest PFN in the list */
769 static struct page *min_page(struct list_head *list)
771 unsigned long min_pfn = -1UL;
772 struct page *min_page = NULL, *page;;
774 list_for_each_entry(page, list, lru) {
775 unsigned long pfn = page_to_pfn(page);
785 /* Remove an element from the buddy allocator from the fallback list */
786 static struct page *__rmqueue_fallback(struct zone *zone, int order,
787 int start_migratetype)
789 struct free_area * area;
793 int nonatomic_fallback_atomic = 0;
796 /* Find the largest possible block of pages in the other list */
797 for (current_order = MAX_ORDER-1; current_order >= order;
799 for (i = 0; i < MIGRATE_TYPES - 1; i++) {
800 migratetype = fallbacks[start_migratetype][i];
803 * Make it hard to fallback to blocks used for
804 * high-order atomic allocations
806 if (migratetype == MIGRATE_HIGHATOMIC &&
807 start_migratetype != MIGRATE_UNMOVABLE &&
808 !nonatomic_fallback_atomic)
811 area = &(zone->free_area[current_order]);
812 if (list_empty(&area->free_list[migratetype]))
815 /* Bias kernel allocations towards low pfns */
816 page = list_entry(area->free_list[migratetype].next,
818 if (unlikely(start_migratetype != MIGRATE_MOVABLE))
819 page = min_page(&area->free_list[migratetype]);
823 * If breaking a large block of pages, move all free
824 * pages to the preferred allocation list
826 if (unlikely(current_order >= MAX_ORDER / 2)) {
827 migratetype = start_migratetype;
828 move_freepages_block(zone, page, migratetype);
831 /* Remove the page from the freelists */
832 list_del(&page->lru);
833 rmv_page_order(page);
834 __mod_zone_page_state(zone, NR_FREE_PAGES,
837 if (current_order == MAX_ORDER - 1)
838 set_pageblock_migratetype(page,
841 expand(zone, page, order, current_order, area, migratetype);
846 /* Allow fallback to high-order atomic blocks if memory is that low */
847 if (!nonatomic_fallback_atomic) {
848 nonatomic_fallback_atomic = 1;
855 static struct page *__rmqueue_fallback(struct zone *zone, int order,
856 int start_migratetype)
860 #endif /* CONFIG_PAGE_GROUP_BY_MOBILITY */
863 * Do the hard work of removing an element from the buddy allocator.
864 * Call me with the zone->lock already held.
866 static struct page *__rmqueue(struct zone *zone, unsigned int order,
869 struct free_area * area;
870 unsigned int current_order;
873 /* Find a page of the appropriate size in the preferred list */
874 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
875 area = &(zone->free_area[current_order]);
876 if (list_empty(&area->free_list[migratetype]))
879 page = list_entry(area->free_list[migratetype].next,
881 list_del(&page->lru);
882 rmv_page_order(page);
884 __mod_zone_page_state(zone, NR_FREE_PAGES, - (1UL << order));
885 expand(zone, page, order, current_order, area, migratetype);
889 page = __rmqueue_fallback(zone, order, migratetype);
897 * Obtain a specified number of elements from the buddy allocator, all under
898 * a single hold of the lock, for efficiency. Add them to the supplied list.
899 * Returns the number of new pages which were placed at *list.
901 static int rmqueue_bulk(struct zone *zone, unsigned int order,
902 unsigned long count, struct list_head *list,
907 spin_lock(&zone->lock);
908 for (i = 0; i < count; ++i) {
909 struct page *page = __rmqueue(zone, order, migratetype);
910 if (unlikely(page == NULL))
912 list_add(&page->lru, list);
913 set_page_private(page, migratetype);
915 spin_unlock(&zone->lock);
921 * Called from the vmstat counter updater to drain pagesets of this
922 * currently executing processor on remote nodes after they have
925 * Note that this function must be called with the thread pinned to
926 * a single processor.
928 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
933 local_irq_save(flags);
934 if (pcp->count >= pcp->batch)
935 to_drain = pcp->batch;
937 to_drain = pcp->count;
938 free_pages_bulk(zone, to_drain, &pcp->list, 0);
939 pcp->count -= to_drain;
940 local_irq_restore(flags);
944 static void __drain_pages(unsigned int cpu)
950 for_each_zone(zone) {
951 struct per_cpu_pageset *pset;
953 if (!populated_zone(zone))
956 pset = zone_pcp(zone, cpu);
957 for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) {
958 struct per_cpu_pages *pcp;
961 local_irq_save(flags);
962 free_pages_bulk(zone, pcp->count, &pcp->list, 0);
964 local_irq_restore(flags);
969 #ifdef CONFIG_HIBERNATION
971 void mark_free_pages(struct zone *zone)
973 unsigned long pfn, max_zone_pfn;
976 struct list_head *curr;
978 if (!zone->spanned_pages)
981 spin_lock_irqsave(&zone->lock, flags);
983 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
984 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
985 if (pfn_valid(pfn)) {
986 struct page *page = pfn_to_page(pfn);
988 if (!swsusp_page_is_forbidden(page))
989 swsusp_unset_page_free(page);
992 for_each_migratetype_order(order, t) {
993 list_for_each(curr, &zone->free_area[order].free_list[t]) {
996 pfn = page_to_pfn(list_entry(curr, struct page, lru));
997 for (i = 0; i < (1UL << order); i++)
998 swsusp_set_page_free(pfn_to_page(pfn + i));
1001 spin_unlock_irqrestore(&zone->lock, flags);
1003 #endif /* CONFIG_PM */
1005 #if defined(CONFIG_HIBERNATION) || defined(CONFIG_PAGE_GROUP_BY_MOBILITY)
1007 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
1009 void drain_local_pages(void)
1011 unsigned long flags;
1013 local_irq_save(flags);
1014 __drain_pages(smp_processor_id());
1015 local_irq_restore(flags);
1018 void smp_drain_local_pages(void *arg)
1020 drain_local_pages();
1024 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
1026 void drain_all_local_pages(void)
1028 unsigned long flags;
1030 local_irq_save(flags);
1031 __drain_pages(smp_processor_id());
1032 local_irq_restore(flags);
1034 smp_call_function(smp_drain_local_pages, NULL, 0, 1);
1037 void drain_all_local_pages(void) {}
1038 #endif /* CONFIG_HIBERNATION || CONFIG_PAGE_GROUP_BY_MOBILITY */
1041 * Free a 0-order page
1043 static void fastcall free_hot_cold_page(struct page *page, int cold)
1045 struct zone *zone = page_zone(page);
1046 struct per_cpu_pages *pcp;
1047 unsigned long flags;
1050 page->mapping = NULL;
1051 if (free_pages_check(page))
1054 if (!PageHighMem(page))
1055 debug_check_no_locks_freed(page_address(page), PAGE_SIZE);
1056 arch_free_page(page, 0);
1057 kernel_map_pages(page, 1, 0);
1059 pcp = &zone_pcp(zone, get_cpu())->pcp[cold];
1060 local_irq_save(flags);
1061 __count_vm_event(PGFREE);
1062 list_add(&page->lru, &pcp->list);
1063 set_page_private(page, get_pageblock_migratetype(page));
1065 if (pcp->count >= pcp->high) {
1066 free_pages_bulk(zone, pcp->batch, &pcp->list, 0);
1067 pcp->count -= pcp->batch;
1069 local_irq_restore(flags);
1073 void fastcall free_hot_page(struct page *page)
1075 free_hot_cold_page(page, 0);
1078 void fastcall free_cold_page(struct page *page)
1080 free_hot_cold_page(page, 1);
1084 * split_page takes a non-compound higher-order page, and splits it into
1085 * n (1<<order) sub-pages: page[0..n]
1086 * Each sub-page must be freed individually.
1088 * Note: this is probably too low level an operation for use in drivers.
1089 * Please consult with lkml before using this in your driver.
1091 void split_page(struct page *page, unsigned int order)
1095 VM_BUG_ON(PageCompound(page));
1096 VM_BUG_ON(!page_count(page));
1097 for (i = 1; i < (1 << order); i++)
1098 set_page_refcounted(page + i);
1102 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1103 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1106 static struct page *buffered_rmqueue(struct zonelist *zonelist,
1107 struct zone *zone, int order, gfp_t gfp_flags)
1109 unsigned long flags;
1111 int cold = !!(gfp_flags & __GFP_COLD);
1113 int migratetype = allocflags_to_migratetype(gfp_flags, order);
1117 if (likely(order == 0)) {
1118 struct per_cpu_pages *pcp;
1120 pcp = &zone_pcp(zone, cpu)->pcp[cold];
1121 local_irq_save(flags);
1123 pcp->count = rmqueue_bulk(zone, 0,
1124 pcp->batch, &pcp->list, migratetype);
1125 if (unlikely(!pcp->count))
1129 #ifdef CONFIG_PAGE_GROUP_BY_MOBILITY
1130 /* Find a page of the appropriate migrate type */
1131 list_for_each_entry(page, &pcp->list, lru)
1132 if (page_private(page) == migratetype)
1135 /* Allocate more to the pcp list if necessary */
1136 if (unlikely(&page->lru == &pcp->list)) {
1137 pcp->count += rmqueue_bulk(zone, 0,
1138 pcp->batch, &pcp->list, migratetype);
1139 page = list_entry(pcp->list.next, struct page, lru);
1142 page = list_entry(pcp->list.next, struct page, lru);
1143 #endif /* CONFIG_PAGE_GROUP_BY_MOBILITY */
1145 list_del(&page->lru);
1148 spin_lock_irqsave(&zone->lock, flags);
1149 page = __rmqueue(zone, order, migratetype);
1150 spin_unlock(&zone->lock);
1155 __count_zone_vm_events(PGALLOC, zone, 1 << order);
1156 zone_statistics(zonelist, zone);
1157 local_irq_restore(flags);
1160 VM_BUG_ON(bad_range(zone, page));
1161 if (prep_new_page(page, order, gfp_flags))
1166 local_irq_restore(flags);
1171 #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
1172 #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
1173 #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
1174 #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
1175 #define ALLOC_HARDER 0x10 /* try to alloc harder */
1176 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1177 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
1179 #ifdef CONFIG_FAIL_PAGE_ALLOC
1181 static struct fail_page_alloc_attr {
1182 struct fault_attr attr;
1184 u32 ignore_gfp_highmem;
1185 u32 ignore_gfp_wait;
1188 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1190 struct dentry *ignore_gfp_highmem_file;
1191 struct dentry *ignore_gfp_wait_file;
1192 struct dentry *min_order_file;
1194 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1196 } fail_page_alloc = {
1197 .attr = FAULT_ATTR_INITIALIZER,
1198 .ignore_gfp_wait = 1,
1199 .ignore_gfp_highmem = 1,
1203 static int __init setup_fail_page_alloc(char *str)
1205 return setup_fault_attr(&fail_page_alloc.attr, str);
1207 __setup("fail_page_alloc=", setup_fail_page_alloc);
1209 static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1211 if (order < fail_page_alloc.min_order)
1213 if (gfp_mask & __GFP_NOFAIL)
1215 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
1217 if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
1220 return should_fail(&fail_page_alloc.attr, 1 << order);
1223 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1225 static int __init fail_page_alloc_debugfs(void)
1227 mode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
1231 err = init_fault_attr_dentries(&fail_page_alloc.attr,
1235 dir = fail_page_alloc.attr.dentries.dir;
1237 fail_page_alloc.ignore_gfp_wait_file =
1238 debugfs_create_bool("ignore-gfp-wait", mode, dir,
1239 &fail_page_alloc.ignore_gfp_wait);
1241 fail_page_alloc.ignore_gfp_highmem_file =
1242 debugfs_create_bool("ignore-gfp-highmem", mode, dir,
1243 &fail_page_alloc.ignore_gfp_highmem);
1244 fail_page_alloc.min_order_file =
1245 debugfs_create_u32("min-order", mode, dir,
1246 &fail_page_alloc.min_order);
1248 if (!fail_page_alloc.ignore_gfp_wait_file ||
1249 !fail_page_alloc.ignore_gfp_highmem_file ||
1250 !fail_page_alloc.min_order_file) {
1252 debugfs_remove(fail_page_alloc.ignore_gfp_wait_file);
1253 debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file);
1254 debugfs_remove(fail_page_alloc.min_order_file);
1255 cleanup_fault_attr_dentries(&fail_page_alloc.attr);
1261 late_initcall(fail_page_alloc_debugfs);
1263 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1265 #else /* CONFIG_FAIL_PAGE_ALLOC */
1267 static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1272 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1275 * Return 1 if free pages are above 'mark'. This takes into account the order
1276 * of the allocation.
1278 int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1279 int classzone_idx, int alloc_flags)
1281 /* free_pages my go negative - that's OK */
1283 long free_pages = zone_page_state(z, NR_FREE_PAGES) - (1 << order) + 1;
1286 if (alloc_flags & ALLOC_HIGH)
1288 if (alloc_flags & ALLOC_HARDER)
1291 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
1293 for (o = 0; o < order; o++) {
1294 /* At the next order, this order's pages become unavailable */
1295 free_pages -= z->free_area[o].nr_free << o;
1297 /* Require fewer higher order pages to be free */
1300 if (free_pages <= min)
1308 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1309 * skip over zones that are not allowed by the cpuset, or that have
1310 * been recently (in last second) found to be nearly full. See further
1311 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1312 * that have to skip over alot of full or unallowed zones.
1314 * If the zonelist cache is present in the passed in zonelist, then
1315 * returns a pointer to the allowed node mask (either the current
1316 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
1318 * If the zonelist cache is not available for this zonelist, does
1319 * nothing and returns NULL.
1321 * If the fullzones BITMAP in the zonelist cache is stale (more than
1322 * a second since last zap'd) then we zap it out (clear its bits.)
1324 * We hold off even calling zlc_setup, until after we've checked the
1325 * first zone in the zonelist, on the theory that most allocations will
1326 * be satisfied from that first zone, so best to examine that zone as
1327 * quickly as we can.
1329 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1331 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1332 nodemask_t *allowednodes; /* zonelist_cache approximation */
1334 zlc = zonelist->zlcache_ptr;
1338 if (jiffies - zlc->last_full_zap > 1 * HZ) {
1339 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1340 zlc->last_full_zap = jiffies;
1343 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1344 &cpuset_current_mems_allowed :
1345 &node_states[N_HIGH_MEMORY];
1346 return allowednodes;
1350 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1351 * if it is worth looking at further for free memory:
1352 * 1) Check that the zone isn't thought to be full (doesn't have its
1353 * bit set in the zonelist_cache fullzones BITMAP).
1354 * 2) Check that the zones node (obtained from the zonelist_cache
1355 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1356 * Return true (non-zero) if zone is worth looking at further, or
1357 * else return false (zero) if it is not.
1359 * This check -ignores- the distinction between various watermarks,
1360 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1361 * found to be full for any variation of these watermarks, it will
1362 * be considered full for up to one second by all requests, unless
1363 * we are so low on memory on all allowed nodes that we are forced
1364 * into the second scan of the zonelist.
1366 * In the second scan we ignore this zonelist cache and exactly
1367 * apply the watermarks to all zones, even it is slower to do so.
1368 * We are low on memory in the second scan, and should leave no stone
1369 * unturned looking for a free page.
1371 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zone **z,
1372 nodemask_t *allowednodes)
1374 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1375 int i; /* index of *z in zonelist zones */
1376 int n; /* node that zone *z is on */
1378 zlc = zonelist->zlcache_ptr;
1382 i = z - zonelist->zones;
1385 /* This zone is worth trying if it is allowed but not full */
1386 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1390 * Given 'z' scanning a zonelist, set the corresponding bit in
1391 * zlc->fullzones, so that subsequent attempts to allocate a page
1392 * from that zone don't waste time re-examining it.
1394 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zone **z)
1396 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1397 int i; /* index of *z in zonelist zones */
1399 zlc = zonelist->zlcache_ptr;
1403 i = z - zonelist->zones;
1405 set_bit(i, zlc->fullzones);
1408 #else /* CONFIG_NUMA */
1410 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1415 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zone **z,
1416 nodemask_t *allowednodes)
1421 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zone **z)
1424 #endif /* CONFIG_NUMA */
1427 * get_page_from_freelist goes through the zonelist trying to allocate
1430 static struct page *
1431 get_page_from_freelist(gfp_t gfp_mask, unsigned int order,
1432 struct zonelist *zonelist, int alloc_flags)
1435 struct page *page = NULL;
1436 int classzone_idx = zone_idx(zonelist->zones[0]);
1438 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1439 int zlc_active = 0; /* set if using zonelist_cache */
1440 int did_zlc_setup = 0; /* just call zlc_setup() one time */
1441 enum zone_type highest_zoneidx = -1; /* Gets set for policy zonelists */
1445 * Scan zonelist, looking for a zone with enough free.
1446 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1448 z = zonelist->zones;
1452 * In NUMA, this could be a policy zonelist which contains
1453 * zones that may not be allowed by the current gfp_mask.
1454 * Check the zone is allowed by the current flags
1456 if (unlikely(alloc_should_filter_zonelist(zonelist))) {
1457 if (highest_zoneidx == -1)
1458 highest_zoneidx = gfp_zone(gfp_mask);
1459 if (zone_idx(*z) > highest_zoneidx)
1463 if (NUMA_BUILD && zlc_active &&
1464 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1467 if ((alloc_flags & ALLOC_CPUSET) &&
1468 !cpuset_zone_allowed_softwall(zone, gfp_mask))
1471 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
1473 if (alloc_flags & ALLOC_WMARK_MIN)
1474 mark = zone->pages_min;
1475 else if (alloc_flags & ALLOC_WMARK_LOW)
1476 mark = zone->pages_low;
1478 mark = zone->pages_high;
1479 if (!zone_watermark_ok(zone, order, mark,
1480 classzone_idx, alloc_flags)) {
1481 if (!zone_reclaim_mode ||
1482 !zone_reclaim(zone, gfp_mask, order))
1483 goto this_zone_full;
1487 page = buffered_rmqueue(zonelist, zone, order, gfp_mask);
1492 zlc_mark_zone_full(zonelist, z);
1494 if (NUMA_BUILD && !did_zlc_setup) {
1495 /* we do zlc_setup after the first zone is tried */
1496 allowednodes = zlc_setup(zonelist, alloc_flags);
1500 } while (*(++z) != NULL);
1502 if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) {
1503 /* Disable zlc cache for second zonelist scan */
1511 * This is the 'heart' of the zoned buddy allocator.
1513 struct page * fastcall
1514 __alloc_pages(gfp_t gfp_mask, unsigned int order,
1515 struct zonelist *zonelist)
1517 const gfp_t wait = gfp_mask & __GFP_WAIT;
1520 struct reclaim_state reclaim_state;
1521 struct task_struct *p = current;
1524 int did_some_progress;
1526 might_sleep_if(wait);
1528 if (should_fail_alloc_page(gfp_mask, order))
1532 z = zonelist->zones; /* the list of zones suitable for gfp_mask */
1534 if (unlikely(*z == NULL)) {
1536 * Happens if we have an empty zonelist as a result of
1537 * GFP_THISNODE being used on a memoryless node
1542 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
1543 zonelist, ALLOC_WMARK_LOW|ALLOC_CPUSET);
1548 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1549 * __GFP_NOWARN set) should not cause reclaim since the subsystem
1550 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1551 * using a larger set of nodes after it has established that the
1552 * allowed per node queues are empty and that nodes are
1555 if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
1558 for (z = zonelist->zones; *z; z++)
1559 wakeup_kswapd(*z, order);
1562 * OK, we're below the kswapd watermark and have kicked background
1563 * reclaim. Now things get more complex, so set up alloc_flags according
1564 * to how we want to proceed.
1566 * The caller may dip into page reserves a bit more if the caller
1567 * cannot run direct reclaim, or if the caller has realtime scheduling
1568 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1569 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1571 alloc_flags = ALLOC_WMARK_MIN;
1572 if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait)
1573 alloc_flags |= ALLOC_HARDER;
1574 if (gfp_mask & __GFP_HIGH)
1575 alloc_flags |= ALLOC_HIGH;
1577 alloc_flags |= ALLOC_CPUSET;
1580 * Go through the zonelist again. Let __GFP_HIGH and allocations
1581 * coming from realtime tasks go deeper into reserves.
1583 * This is the last chance, in general, before the goto nopage.
1584 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1585 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1587 page = get_page_from_freelist(gfp_mask, order, zonelist, alloc_flags);
1591 /* This allocation should allow future memory freeing. */
1594 if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE)))
1595 && !in_interrupt()) {
1596 if (!(gfp_mask & __GFP_NOMEMALLOC)) {
1598 /* go through the zonelist yet again, ignoring mins */
1599 page = get_page_from_freelist(gfp_mask, order,
1600 zonelist, ALLOC_NO_WATERMARKS);
1603 if (gfp_mask & __GFP_NOFAIL) {
1604 congestion_wait(WRITE, HZ/50);
1611 /* Atomic allocations - we can't balance anything */
1617 /* We now go into synchronous reclaim */
1618 cpuset_memory_pressure_bump();
1619 p->flags |= PF_MEMALLOC;
1620 reclaim_state.reclaimed_slab = 0;
1621 p->reclaim_state = &reclaim_state;
1623 did_some_progress = try_to_free_pages(zonelist->zones, order, gfp_mask);
1625 p->reclaim_state = NULL;
1626 p->flags &= ~PF_MEMALLOC;
1631 drain_all_local_pages();
1633 if (likely(did_some_progress)) {
1634 page = get_page_from_freelist(gfp_mask, order,
1635 zonelist, alloc_flags);
1638 } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
1640 * Go through the zonelist yet one more time, keep
1641 * very high watermark here, this is only to catch
1642 * a parallel oom killing, we must fail if we're still
1643 * under heavy pressure.
1645 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
1646 zonelist, ALLOC_WMARK_HIGH|ALLOC_CPUSET);
1650 /* The OOM killer will not help higher order allocs so fail */
1651 if (order > PAGE_ALLOC_COSTLY_ORDER)
1654 out_of_memory(zonelist, gfp_mask, order);
1659 * Don't let big-order allocations loop unless the caller explicitly
1660 * requests that. Wait for some write requests to complete then retry.
1662 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
1663 * <= 3, but that may not be true in other implementations.
1666 if (!(gfp_mask & __GFP_NORETRY)) {
1667 if ((order <= PAGE_ALLOC_COSTLY_ORDER) ||
1668 (gfp_mask & __GFP_REPEAT))
1670 if (gfp_mask & __GFP_NOFAIL)
1674 congestion_wait(WRITE, HZ/50);
1679 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
1680 printk(KERN_WARNING "%s: page allocation failure."
1681 " order:%d, mode:0x%x\n",
1682 p->comm, order, gfp_mask);
1690 EXPORT_SYMBOL(__alloc_pages);
1693 * Common helper functions.
1695 fastcall unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
1698 page = alloc_pages(gfp_mask, order);
1701 return (unsigned long) page_address(page);
1704 EXPORT_SYMBOL(__get_free_pages);
1706 fastcall unsigned long get_zeroed_page(gfp_t gfp_mask)
1711 * get_zeroed_page() returns a 32-bit address, which cannot represent
1714 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
1716 page = alloc_pages(gfp_mask | __GFP_ZERO, 0);
1718 return (unsigned long) page_address(page);
1722 EXPORT_SYMBOL(get_zeroed_page);
1724 void __pagevec_free(struct pagevec *pvec)
1726 int i = pagevec_count(pvec);
1729 free_hot_cold_page(pvec->pages[i], pvec->cold);
1732 fastcall void __free_pages(struct page *page, unsigned int order)
1734 if (put_page_testzero(page)) {
1736 free_hot_page(page);
1738 __free_pages_ok(page, order);
1742 EXPORT_SYMBOL(__free_pages);
1744 fastcall void free_pages(unsigned long addr, unsigned int order)
1747 VM_BUG_ON(!virt_addr_valid((void *)addr));
1748 __free_pages(virt_to_page((void *)addr), order);
1752 EXPORT_SYMBOL(free_pages);
1754 static unsigned int nr_free_zone_pages(int offset)
1756 /* Just pick one node, since fallback list is circular */
1757 pg_data_t *pgdat = NODE_DATA(numa_node_id());
1758 unsigned int sum = 0;
1760 struct zonelist *zonelist = pgdat->node_zonelists + offset;
1761 struct zone **zonep = zonelist->zones;
1764 for (zone = *zonep++; zone; zone = *zonep++) {
1765 unsigned long size = zone->present_pages;
1766 unsigned long high = zone->pages_high;
1775 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1777 unsigned int nr_free_buffer_pages(void)
1779 return nr_free_zone_pages(gfp_zone(GFP_USER));
1781 EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
1784 * Amount of free RAM allocatable within all zones
1786 unsigned int nr_free_pagecache_pages(void)
1788 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
1791 static inline void show_node(struct zone *zone)
1794 printk("Node %d ", zone_to_nid(zone));
1797 void si_meminfo(struct sysinfo *val)
1799 val->totalram = totalram_pages;
1801 val->freeram = global_page_state(NR_FREE_PAGES);
1802 val->bufferram = nr_blockdev_pages();
1803 val->totalhigh = totalhigh_pages;
1804 val->freehigh = nr_free_highpages();
1805 val->mem_unit = PAGE_SIZE;
1808 EXPORT_SYMBOL(si_meminfo);
1811 void si_meminfo_node(struct sysinfo *val, int nid)
1813 pg_data_t *pgdat = NODE_DATA(nid);
1815 val->totalram = pgdat->node_present_pages;
1816 val->freeram = node_page_state(nid, NR_FREE_PAGES);
1817 #ifdef CONFIG_HIGHMEM
1818 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1819 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
1825 val->mem_unit = PAGE_SIZE;
1829 #define K(x) ((x) << (PAGE_SHIFT-10))
1832 * Show free area list (used inside shift_scroll-lock stuff)
1833 * We also calculate the percentage fragmentation. We do this by counting the
1834 * memory on each free list with the exception of the first item on the list.
1836 void show_free_areas(void)
1841 for_each_zone(zone) {
1842 if (!populated_zone(zone))
1846 printk("%s per-cpu:\n", zone->name);
1848 for_each_online_cpu(cpu) {
1849 struct per_cpu_pageset *pageset;
1851 pageset = zone_pcp(zone, cpu);
1853 printk("CPU %4d: Hot: hi:%5d, btch:%4d usd:%4d "
1854 "Cold: hi:%5d, btch:%4d usd:%4d\n",
1855 cpu, pageset->pcp[0].high,
1856 pageset->pcp[0].batch, pageset->pcp[0].count,
1857 pageset->pcp[1].high, pageset->pcp[1].batch,
1858 pageset->pcp[1].count);
1862 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu unstable:%lu\n"
1863 " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
1864 global_page_state(NR_ACTIVE),
1865 global_page_state(NR_INACTIVE),
1866 global_page_state(NR_FILE_DIRTY),
1867 global_page_state(NR_WRITEBACK),
1868 global_page_state(NR_UNSTABLE_NFS),
1869 global_page_state(NR_FREE_PAGES),
1870 global_page_state(NR_SLAB_RECLAIMABLE) +
1871 global_page_state(NR_SLAB_UNRECLAIMABLE),
1872 global_page_state(NR_FILE_MAPPED),
1873 global_page_state(NR_PAGETABLE),
1874 global_page_state(NR_BOUNCE));
1876 for_each_zone(zone) {
1879 if (!populated_zone(zone))
1891 " pages_scanned:%lu"
1892 " all_unreclaimable? %s"
1895 K(zone_page_state(zone, NR_FREE_PAGES)),
1898 K(zone->pages_high),
1899 K(zone_page_state(zone, NR_ACTIVE)),
1900 K(zone_page_state(zone, NR_INACTIVE)),
1901 K(zone->present_pages),
1902 zone->pages_scanned,
1903 (zone->all_unreclaimable ? "yes" : "no")
1905 printk("lowmem_reserve[]:");
1906 for (i = 0; i < MAX_NR_ZONES; i++)
1907 printk(" %lu", zone->lowmem_reserve[i]);
1911 for_each_zone(zone) {
1912 unsigned long nr[MAX_ORDER], flags, order, total = 0;
1914 if (!populated_zone(zone))
1918 printk("%s: ", zone->name);
1920 spin_lock_irqsave(&zone->lock, flags);
1921 for (order = 0; order < MAX_ORDER; order++) {
1922 nr[order] = zone->free_area[order].nr_free;
1923 total += nr[order] << order;
1925 spin_unlock_irqrestore(&zone->lock, flags);
1926 for (order = 0; order < MAX_ORDER; order++)
1927 printk("%lu*%lukB ", nr[order], K(1UL) << order);
1928 printk("= %lukB\n", K(total));
1931 show_swap_cache_info();
1935 * Builds allocation fallback zone lists.
1937 * Add all populated zones of a node to the zonelist.
1939 static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
1940 int nr_zones, enum zone_type zone_type)
1944 BUG_ON(zone_type >= MAX_NR_ZONES);
1949 zone = pgdat->node_zones + zone_type;
1950 if (populated_zone(zone)) {
1951 zonelist->zones[nr_zones++] = zone;
1952 check_highest_zone(zone_type);
1955 } while (zone_type);
1962 * 0 = automatic detection of better ordering.
1963 * 1 = order by ([node] distance, -zonetype)
1964 * 2 = order by (-zonetype, [node] distance)
1966 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
1967 * the same zonelist. So only NUMA can configure this param.
1969 #define ZONELIST_ORDER_DEFAULT 0
1970 #define ZONELIST_ORDER_NODE 1
1971 #define ZONELIST_ORDER_ZONE 2
1973 /* zonelist order in the kernel.
1974 * set_zonelist_order() will set this to NODE or ZONE.
1976 static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
1977 static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
1981 /* The value user specified ....changed by config */
1982 static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
1983 /* string for sysctl */
1984 #define NUMA_ZONELIST_ORDER_LEN 16
1985 char numa_zonelist_order[16] = "default";
1988 * interface for configure zonelist ordering.
1989 * command line option "numa_zonelist_order"
1990 * = "[dD]efault - default, automatic configuration.
1991 * = "[nN]ode - order by node locality, then by zone within node
1992 * = "[zZ]one - order by zone, then by locality within zone
1995 static int __parse_numa_zonelist_order(char *s)
1997 if (*s == 'd' || *s == 'D') {
1998 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
1999 } else if (*s == 'n' || *s == 'N') {
2000 user_zonelist_order = ZONELIST_ORDER_NODE;
2001 } else if (*s == 'z' || *s == 'Z') {
2002 user_zonelist_order = ZONELIST_ORDER_ZONE;
2005 "Ignoring invalid numa_zonelist_order value: "
2012 static __init int setup_numa_zonelist_order(char *s)
2015 return __parse_numa_zonelist_order(s);
2018 early_param("numa_zonelist_order", setup_numa_zonelist_order);
2021 * sysctl handler for numa_zonelist_order
2023 int numa_zonelist_order_handler(ctl_table *table, int write,
2024 struct file *file, void __user *buffer, size_t *length,
2027 char saved_string[NUMA_ZONELIST_ORDER_LEN];
2031 strncpy(saved_string, (char*)table->data,
2032 NUMA_ZONELIST_ORDER_LEN);
2033 ret = proc_dostring(table, write, file, buffer, length, ppos);
2037 int oldval = user_zonelist_order;
2038 if (__parse_numa_zonelist_order((char*)table->data)) {
2040 * bogus value. restore saved string
2042 strncpy((char*)table->data, saved_string,
2043 NUMA_ZONELIST_ORDER_LEN);
2044 user_zonelist_order = oldval;
2045 } else if (oldval != user_zonelist_order)
2046 build_all_zonelists();
2052 #define MAX_NODE_LOAD (num_online_nodes())
2053 static int node_load[MAX_NUMNODES];
2056 * find_next_best_node - find the next node that should appear in a given node's fallback list
2057 * @node: node whose fallback list we're appending
2058 * @used_node_mask: nodemask_t of already used nodes
2060 * We use a number of factors to determine which is the next node that should
2061 * appear on a given node's fallback list. The node should not have appeared
2062 * already in @node's fallback list, and it should be the next closest node
2063 * according to the distance array (which contains arbitrary distance values
2064 * from each node to each node in the system), and should also prefer nodes
2065 * with no CPUs, since presumably they'll have very little allocation pressure
2066 * on them otherwise.
2067 * It returns -1 if no node is found.
2069 static int find_next_best_node(int node, nodemask_t *used_node_mask)
2072 int min_val = INT_MAX;
2075 /* Use the local node if we haven't already */
2076 if (!node_isset(node, *used_node_mask)) {
2077 node_set(node, *used_node_mask);
2081 for_each_node_state(n, N_HIGH_MEMORY) {
2084 /* Don't want a node to appear more than once */
2085 if (node_isset(n, *used_node_mask))
2088 /* Use the distance array to find the distance */
2089 val = node_distance(node, n);
2091 /* Penalize nodes under us ("prefer the next node") */
2094 /* Give preference to headless and unused nodes */
2095 tmp = node_to_cpumask(n);
2096 if (!cpus_empty(tmp))
2097 val += PENALTY_FOR_NODE_WITH_CPUS;
2099 /* Slight preference for less loaded node */
2100 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
2101 val += node_load[n];
2103 if (val < min_val) {
2110 node_set(best_node, *used_node_mask);
2117 * Build zonelists ordered by node and zones within node.
2118 * This results in maximum locality--normal zone overflows into local
2119 * DMA zone, if any--but risks exhausting DMA zone.
2121 static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
2125 struct zonelist *zonelist;
2127 for (i = 0; i < MAX_NR_ZONES; i++) {
2128 zonelist = pgdat->node_zonelists + i;
2129 for (j = 0; zonelist->zones[j] != NULL; j++)
2131 j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
2132 zonelist->zones[j] = NULL;
2137 * Build gfp_thisnode zonelists
2139 static void build_thisnode_zonelists(pg_data_t *pgdat)
2143 struct zonelist *zonelist;
2145 for (i = 0; i < MAX_NR_ZONES; i++) {
2146 zonelist = pgdat->node_zonelists + MAX_NR_ZONES + i;
2147 j = build_zonelists_node(pgdat, zonelist, 0, i);
2148 zonelist->zones[j] = NULL;
2153 * Build zonelists ordered by zone and nodes within zones.
2154 * This results in conserving DMA zone[s] until all Normal memory is
2155 * exhausted, but results in overflowing to remote node while memory
2156 * may still exist in local DMA zone.
2158 static int node_order[MAX_NUMNODES];
2160 static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
2164 int zone_type; /* needs to be signed */
2166 struct zonelist *zonelist;
2168 for (i = 0; i < MAX_NR_ZONES; i++) {
2169 zonelist = pgdat->node_zonelists + i;
2171 for (zone_type = i; zone_type >= 0; zone_type--) {
2172 for (j = 0; j < nr_nodes; j++) {
2173 node = node_order[j];
2174 z = &NODE_DATA(node)->node_zones[zone_type];
2175 if (populated_zone(z)) {
2176 zonelist->zones[pos++] = z;
2177 check_highest_zone(zone_type);
2181 zonelist->zones[pos] = NULL;
2185 static int default_zonelist_order(void)
2188 unsigned long low_kmem_size,total_size;
2192 * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem.
2193 * If they are really small and used heavily, the system can fall
2194 * into OOM very easily.
2195 * This function detect ZONE_DMA/DMA32 size and confgigures zone order.
2197 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2200 for_each_online_node(nid) {
2201 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2202 z = &NODE_DATA(nid)->node_zones[zone_type];
2203 if (populated_zone(z)) {
2204 if (zone_type < ZONE_NORMAL)
2205 low_kmem_size += z->present_pages;
2206 total_size += z->present_pages;
2210 if (!low_kmem_size || /* there are no DMA area. */
2211 low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
2212 return ZONELIST_ORDER_NODE;
2214 * look into each node's config.
2215 * If there is a node whose DMA/DMA32 memory is very big area on
2216 * local memory, NODE_ORDER may be suitable.
2218 average_size = total_size /
2219 (nodes_weight(node_states[N_HIGH_MEMORY]) + 1);
2220 for_each_online_node(nid) {
2223 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2224 z = &NODE_DATA(nid)->node_zones[zone_type];
2225 if (populated_zone(z)) {
2226 if (zone_type < ZONE_NORMAL)
2227 low_kmem_size += z->present_pages;
2228 total_size += z->present_pages;
2231 if (low_kmem_size &&
2232 total_size > average_size && /* ignore small node */
2233 low_kmem_size > total_size * 70/100)
2234 return ZONELIST_ORDER_NODE;
2236 return ZONELIST_ORDER_ZONE;
2239 static void set_zonelist_order(void)
2241 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
2242 current_zonelist_order = default_zonelist_order();
2244 current_zonelist_order = user_zonelist_order;
2247 static void build_zonelists(pg_data_t *pgdat)
2251 nodemask_t used_mask;
2252 int local_node, prev_node;
2253 struct zonelist *zonelist;
2254 int order = current_zonelist_order;
2256 /* initialize zonelists */
2257 for (i = 0; i < MAX_ZONELISTS; i++) {
2258 zonelist = pgdat->node_zonelists + i;
2259 zonelist->zones[0] = NULL;
2262 /* NUMA-aware ordering of nodes */
2263 local_node = pgdat->node_id;
2264 load = num_online_nodes();
2265 prev_node = local_node;
2266 nodes_clear(used_mask);
2268 memset(node_load, 0, sizeof(node_load));
2269 memset(node_order, 0, sizeof(node_order));
2272 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
2273 int distance = node_distance(local_node, node);
2276 * If another node is sufficiently far away then it is better
2277 * to reclaim pages in a zone before going off node.
2279 if (distance > RECLAIM_DISTANCE)
2280 zone_reclaim_mode = 1;
2283 * We don't want to pressure a particular node.
2284 * So adding penalty to the first node in same
2285 * distance group to make it round-robin.
2287 if (distance != node_distance(local_node, prev_node))
2288 node_load[node] = load;
2292 if (order == ZONELIST_ORDER_NODE)
2293 build_zonelists_in_node_order(pgdat, node);
2295 node_order[j++] = node; /* remember order */
2298 if (order == ZONELIST_ORDER_ZONE) {
2299 /* calculate node order -- i.e., DMA last! */
2300 build_zonelists_in_zone_order(pgdat, j);
2303 build_thisnode_zonelists(pgdat);
2306 /* Construct the zonelist performance cache - see further mmzone.h */
2307 static void build_zonelist_cache(pg_data_t *pgdat)
2311 for (i = 0; i < MAX_NR_ZONES; i++) {
2312 struct zonelist *zonelist;
2313 struct zonelist_cache *zlc;
2316 zonelist = pgdat->node_zonelists + i;
2317 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
2318 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
2319 for (z = zonelist->zones; *z; z++)
2320 zlc->z_to_n[z - zonelist->zones] = zone_to_nid(*z);
2325 #else /* CONFIG_NUMA */
2327 static void set_zonelist_order(void)
2329 current_zonelist_order = ZONELIST_ORDER_ZONE;
2332 static void build_zonelists(pg_data_t *pgdat)
2334 int node, local_node;
2337 local_node = pgdat->node_id;
2338 for (i = 0; i < MAX_NR_ZONES; i++) {
2339 struct zonelist *zonelist;
2341 zonelist = pgdat->node_zonelists + i;
2343 j = build_zonelists_node(pgdat, zonelist, 0, i);
2345 * Now we build the zonelist so that it contains the zones
2346 * of all the other nodes.
2347 * We don't want to pressure a particular node, so when
2348 * building the zones for node N, we make sure that the
2349 * zones coming right after the local ones are those from
2350 * node N+1 (modulo N)
2352 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
2353 if (!node_online(node))
2355 j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
2357 for (node = 0; node < local_node; node++) {
2358 if (!node_online(node))
2360 j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
2363 zonelist->zones[j] = NULL;
2367 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
2368 static void build_zonelist_cache(pg_data_t *pgdat)
2372 for (i = 0; i < MAX_NR_ZONES; i++)
2373 pgdat->node_zonelists[i].zlcache_ptr = NULL;
2376 #endif /* CONFIG_NUMA */
2378 /* return values int ....just for stop_machine_run() */
2379 static int __build_all_zonelists(void *dummy)
2383 for_each_online_node(nid) {
2384 pg_data_t *pgdat = NODE_DATA(nid);
2386 build_zonelists(pgdat);
2387 build_zonelist_cache(pgdat);
2392 void build_all_zonelists(void)
2394 set_zonelist_order();
2396 if (system_state == SYSTEM_BOOTING) {
2397 __build_all_zonelists(NULL);
2398 cpuset_init_current_mems_allowed();
2400 /* we have to stop all cpus to guaranntee there is no user
2402 stop_machine_run(__build_all_zonelists, NULL, NR_CPUS);
2403 /* cpuset refresh routine should be here */
2405 vm_total_pages = nr_free_pagecache_pages();
2407 * Disable grouping by mobility if the number of pages in the
2408 * system is too low to allow the mechanism to work. It would be
2409 * more accurate, but expensive to check per-zone. This check is
2410 * made on memory-hotadd so a system can start with mobility
2411 * disabled and enable it later
2413 if (vm_total_pages < (MAX_ORDER_NR_PAGES * MIGRATE_TYPES))
2414 page_group_by_mobility_disabled = 1;
2416 page_group_by_mobility_disabled = 0;
2418 printk("Built %i zonelists in %s order, mobility grouping %s. "
2419 "Total pages: %ld\n",
2421 zonelist_order_name[current_zonelist_order],
2422 page_group_by_mobility_disabled ? "off" : "on",
2425 printk("Policy zone: %s\n", zone_names[policy_zone]);
2430 * Helper functions to size the waitqueue hash table.
2431 * Essentially these want to choose hash table sizes sufficiently
2432 * large so that collisions trying to wait on pages are rare.
2433 * But in fact, the number of active page waitqueues on typical
2434 * systems is ridiculously low, less than 200. So this is even
2435 * conservative, even though it seems large.
2437 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
2438 * waitqueues, i.e. the size of the waitq table given the number of pages.
2440 #define PAGES_PER_WAITQUEUE 256
2442 #ifndef CONFIG_MEMORY_HOTPLUG
2443 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
2445 unsigned long size = 1;
2447 pages /= PAGES_PER_WAITQUEUE;
2449 while (size < pages)
2453 * Once we have dozens or even hundreds of threads sleeping
2454 * on IO we've got bigger problems than wait queue collision.
2455 * Limit the size of the wait table to a reasonable size.
2457 size = min(size, 4096UL);
2459 return max(size, 4UL);
2463 * A zone's size might be changed by hot-add, so it is not possible to determine
2464 * a suitable size for its wait_table. So we use the maximum size now.
2466 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
2468 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
2469 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
2470 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
2472 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
2473 * or more by the traditional way. (See above). It equals:
2475 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
2476 * ia64(16K page size) : = ( 8G + 4M)byte.
2477 * powerpc (64K page size) : = (32G +16M)byte.
2479 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
2486 * This is an integer logarithm so that shifts can be used later
2487 * to extract the more random high bits from the multiplicative
2488 * hash function before the remainder is taken.
2490 static inline unsigned long wait_table_bits(unsigned long size)
2495 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
2498 * Initially all pages are reserved - free ones are freed
2499 * up by free_all_bootmem() once the early boot process is
2500 * done. Non-atomic initialization, single-pass.
2502 void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
2503 unsigned long start_pfn, enum memmap_context context)
2506 unsigned long end_pfn = start_pfn + size;
2509 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
2511 * There can be holes in boot-time mem_map[]s
2512 * handed to this function. They do not
2513 * exist on hotplugged memory.
2515 if (context == MEMMAP_EARLY) {
2516 if (!early_pfn_valid(pfn))
2518 if (!early_pfn_in_nid(pfn, nid))
2521 page = pfn_to_page(pfn);
2522 set_page_links(page, zone, nid, pfn);
2523 init_page_count(page);
2524 reset_page_mapcount(page);
2525 SetPageReserved(page);
2528 * Mark the block movable so that blocks are reserved for
2529 * movable at startup. This will force kernel allocations
2530 * to reserve their blocks rather than leaking throughout
2531 * the address space during boot when many long-lived
2532 * kernel allocations are made
2534 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
2536 INIT_LIST_HEAD(&page->lru);
2537 #ifdef WANT_PAGE_VIRTUAL
2538 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
2539 if (!is_highmem_idx(zone))
2540 set_page_address(page, __va(pfn << PAGE_SHIFT));
2545 static void __meminit zone_init_free_lists(struct pglist_data *pgdat,
2546 struct zone *zone, unsigned long size)
2549 for_each_migratetype_order(order, t) {
2550 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
2551 zone->free_area[order].nr_free = 0;
2555 #ifndef __HAVE_ARCH_MEMMAP_INIT
2556 #define memmap_init(size, nid, zone, start_pfn) \
2557 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
2560 static int __devinit zone_batchsize(struct zone *zone)
2565 * The per-cpu-pages pools are set to around 1000th of the
2566 * size of the zone. But no more than 1/2 of a meg.
2568 * OK, so we don't know how big the cache is. So guess.
2570 batch = zone->present_pages / 1024;
2571 if (batch * PAGE_SIZE > 512 * 1024)
2572 batch = (512 * 1024) / PAGE_SIZE;
2573 batch /= 4; /* We effectively *= 4 below */
2578 * Clamp the batch to a 2^n - 1 value. Having a power
2579 * of 2 value was found to be more likely to have
2580 * suboptimal cache aliasing properties in some cases.
2582 * For example if 2 tasks are alternately allocating
2583 * batches of pages, one task can end up with a lot
2584 * of pages of one half of the possible page colors
2585 * and the other with pages of the other colors.
2587 batch = (1 << (fls(batch + batch/2)-1)) - 1;
2592 inline void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
2594 struct per_cpu_pages *pcp;
2596 memset(p, 0, sizeof(*p));
2598 pcp = &p->pcp[0]; /* hot */
2600 pcp->high = 6 * batch;
2601 pcp->batch = max(1UL, 1 * batch);
2602 INIT_LIST_HEAD(&pcp->list);
2604 pcp = &p->pcp[1]; /* cold*/
2606 pcp->high = 2 * batch;
2607 pcp->batch = max(1UL, batch/2);
2608 INIT_LIST_HEAD(&pcp->list);
2612 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
2613 * to the value high for the pageset p.
2616 static void setup_pagelist_highmark(struct per_cpu_pageset *p,
2619 struct per_cpu_pages *pcp;
2621 pcp = &p->pcp[0]; /* hot list */
2623 pcp->batch = max(1UL, high/4);
2624 if ((high/4) > (PAGE_SHIFT * 8))
2625 pcp->batch = PAGE_SHIFT * 8;
2631 * Boot pageset table. One per cpu which is going to be used for all
2632 * zones and all nodes. The parameters will be set in such a way
2633 * that an item put on a list will immediately be handed over to
2634 * the buddy list. This is safe since pageset manipulation is done
2635 * with interrupts disabled.
2637 * Some NUMA counter updates may also be caught by the boot pagesets.
2639 * The boot_pagesets must be kept even after bootup is complete for
2640 * unused processors and/or zones. They do play a role for bootstrapping
2641 * hotplugged processors.
2643 * zoneinfo_show() and maybe other functions do
2644 * not check if the processor is online before following the pageset pointer.
2645 * Other parts of the kernel may not check if the zone is available.
2647 static struct per_cpu_pageset boot_pageset[NR_CPUS];
2650 * Dynamically allocate memory for the
2651 * per cpu pageset array in struct zone.
2653 static int __cpuinit process_zones(int cpu)
2655 struct zone *zone, *dzone;
2656 int node = cpu_to_node(cpu);
2658 node_set_state(node, N_CPU); /* this node has a cpu */
2660 for_each_zone(zone) {
2662 if (!populated_zone(zone))
2665 zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset),
2667 if (!zone_pcp(zone, cpu))
2670 setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone));
2672 if (percpu_pagelist_fraction)
2673 setup_pagelist_highmark(zone_pcp(zone, cpu),
2674 (zone->present_pages / percpu_pagelist_fraction));
2679 for_each_zone(dzone) {
2680 if (!populated_zone(dzone))
2684 kfree(zone_pcp(dzone, cpu));
2685 zone_pcp(dzone, cpu) = NULL;
2690 static inline void free_zone_pagesets(int cpu)
2694 for_each_zone(zone) {
2695 struct per_cpu_pageset *pset = zone_pcp(zone, cpu);
2697 /* Free per_cpu_pageset if it is slab allocated */
2698 if (pset != &boot_pageset[cpu])
2700 zone_pcp(zone, cpu) = NULL;
2704 static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb,
2705 unsigned long action,
2708 int cpu = (long)hcpu;
2709 int ret = NOTIFY_OK;
2712 case CPU_UP_PREPARE:
2713 case CPU_UP_PREPARE_FROZEN:
2714 if (process_zones(cpu))
2717 case CPU_UP_CANCELED:
2718 case CPU_UP_CANCELED_FROZEN:
2720 case CPU_DEAD_FROZEN:
2721 free_zone_pagesets(cpu);
2729 static struct notifier_block __cpuinitdata pageset_notifier =
2730 { &pageset_cpuup_callback, NULL, 0 };
2732 void __init setup_per_cpu_pageset(void)
2736 /* Initialize per_cpu_pageset for cpu 0.
2737 * A cpuup callback will do this for every cpu
2738 * as it comes online
2740 err = process_zones(smp_processor_id());
2742 register_cpu_notifier(&pageset_notifier);
2747 static noinline __init_refok
2748 int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
2751 struct pglist_data *pgdat = zone->zone_pgdat;
2755 * The per-page waitqueue mechanism uses hashed waitqueues
2758 zone->wait_table_hash_nr_entries =
2759 wait_table_hash_nr_entries(zone_size_pages);
2760 zone->wait_table_bits =
2761 wait_table_bits(zone->wait_table_hash_nr_entries);
2762 alloc_size = zone->wait_table_hash_nr_entries
2763 * sizeof(wait_queue_head_t);
2765 if (system_state == SYSTEM_BOOTING) {
2766 zone->wait_table = (wait_queue_head_t *)
2767 alloc_bootmem_node(pgdat, alloc_size);
2770 * This case means that a zone whose size was 0 gets new memory
2771 * via memory hot-add.
2772 * But it may be the case that a new node was hot-added. In
2773 * this case vmalloc() will not be able to use this new node's
2774 * memory - this wait_table must be initialized to use this new
2775 * node itself as well.
2776 * To use this new node's memory, further consideration will be
2779 zone->wait_table = vmalloc(alloc_size);
2781 if (!zone->wait_table)
2784 for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
2785 init_waitqueue_head(zone->wait_table + i);
2790 static __meminit void zone_pcp_init(struct zone *zone)
2793 unsigned long batch = zone_batchsize(zone);
2795 for (cpu = 0; cpu < NR_CPUS; cpu++) {
2797 /* Early boot. Slab allocator not functional yet */
2798 zone_pcp(zone, cpu) = &boot_pageset[cpu];
2799 setup_pageset(&boot_pageset[cpu],0);
2801 setup_pageset(zone_pcp(zone,cpu), batch);
2804 if (zone->present_pages)
2805 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n",
2806 zone->name, zone->present_pages, batch);
2809 __meminit int init_currently_empty_zone(struct zone *zone,
2810 unsigned long zone_start_pfn,
2812 enum memmap_context context)
2814 struct pglist_data *pgdat = zone->zone_pgdat;
2816 ret = zone_wait_table_init(zone, size);
2819 pgdat->nr_zones = zone_idx(zone) + 1;
2821 zone->zone_start_pfn = zone_start_pfn;
2823 memmap_init(size, pgdat->node_id, zone_idx(zone), zone_start_pfn);
2825 zone_init_free_lists(pgdat, zone, zone->spanned_pages);
2830 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2832 * Basic iterator support. Return the first range of PFNs for a node
2833 * Note: nid == MAX_NUMNODES returns first region regardless of node
2835 static int __meminit first_active_region_index_in_nid(int nid)
2839 for (i = 0; i < nr_nodemap_entries; i++)
2840 if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
2847 * Basic iterator support. Return the next active range of PFNs for a node
2848 * Note: nid == MAX_NUMNODES returns next region regardles of node
2850 static int __meminit next_active_region_index_in_nid(int index, int nid)
2852 for (index = index + 1; index < nr_nodemap_entries; index++)
2853 if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
2859 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
2861 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
2862 * Architectures may implement their own version but if add_active_range()
2863 * was used and there are no special requirements, this is a convenient
2866 int __meminit early_pfn_to_nid(unsigned long pfn)
2870 for (i = 0; i < nr_nodemap_entries; i++) {
2871 unsigned long start_pfn = early_node_map[i].start_pfn;
2872 unsigned long end_pfn = early_node_map[i].end_pfn;
2874 if (start_pfn <= pfn && pfn < end_pfn)
2875 return early_node_map[i].nid;
2880 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
2882 /* Basic iterator support to walk early_node_map[] */
2883 #define for_each_active_range_index_in_nid(i, nid) \
2884 for (i = first_active_region_index_in_nid(nid); i != -1; \
2885 i = next_active_region_index_in_nid(i, nid))
2888 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
2889 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
2890 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
2892 * If an architecture guarantees that all ranges registered with
2893 * add_active_ranges() contain no holes and may be freed, this
2894 * this function may be used instead of calling free_bootmem() manually.
2896 void __init free_bootmem_with_active_regions(int nid,
2897 unsigned long max_low_pfn)
2901 for_each_active_range_index_in_nid(i, nid) {
2902 unsigned long size_pages = 0;
2903 unsigned long end_pfn = early_node_map[i].end_pfn;
2905 if (early_node_map[i].start_pfn >= max_low_pfn)
2908 if (end_pfn > max_low_pfn)
2909 end_pfn = max_low_pfn;
2911 size_pages = end_pfn - early_node_map[i].start_pfn;
2912 free_bootmem_node(NODE_DATA(early_node_map[i].nid),
2913 PFN_PHYS(early_node_map[i].start_pfn),
2914 size_pages << PAGE_SHIFT);
2919 * sparse_memory_present_with_active_regions - Call memory_present for each active range
2920 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
2922 * If an architecture guarantees that all ranges registered with
2923 * add_active_ranges() contain no holes and may be freed, this
2924 * function may be used instead of calling memory_present() manually.
2926 void __init sparse_memory_present_with_active_regions(int nid)
2930 for_each_active_range_index_in_nid(i, nid)
2931 memory_present(early_node_map[i].nid,
2932 early_node_map[i].start_pfn,
2933 early_node_map[i].end_pfn);
2937 * push_node_boundaries - Push node boundaries to at least the requested boundary
2938 * @nid: The nid of the node to push the boundary for
2939 * @start_pfn: The start pfn of the node
2940 * @end_pfn: The end pfn of the node
2942 * In reserve-based hot-add, mem_map is allocated that is unused until hotadd
2943 * time. Specifically, on x86_64, SRAT will report ranges that can potentially
2944 * be hotplugged even though no physical memory exists. This function allows
2945 * an arch to push out the node boundaries so mem_map is allocated that can
2948 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
2949 void __init push_node_boundaries(unsigned int nid,
2950 unsigned long start_pfn, unsigned long end_pfn)
2952 printk(KERN_DEBUG "Entering push_node_boundaries(%u, %lu, %lu)\n",
2953 nid, start_pfn, end_pfn);
2955 /* Initialise the boundary for this node if necessary */
2956 if (node_boundary_end_pfn[nid] == 0)
2957 node_boundary_start_pfn[nid] = -1UL;
2959 /* Update the boundaries */
2960 if (node_boundary_start_pfn[nid] > start_pfn)
2961 node_boundary_start_pfn[nid] = start_pfn;
2962 if (node_boundary_end_pfn[nid] < end_pfn)
2963 node_boundary_end_pfn[nid] = end_pfn;
2966 /* If necessary, push the node boundary out for reserve hotadd */
2967 static void __meminit account_node_boundary(unsigned int nid,
2968 unsigned long *start_pfn, unsigned long *end_pfn)
2970 printk(KERN_DEBUG "Entering account_node_boundary(%u, %lu, %lu)\n",
2971 nid, *start_pfn, *end_pfn);
2973 /* Return if boundary information has not been provided */
2974 if (node_boundary_end_pfn[nid] == 0)
2977 /* Check the boundaries and update if necessary */
2978 if (node_boundary_start_pfn[nid] < *start_pfn)
2979 *start_pfn = node_boundary_start_pfn[nid];
2980 if (node_boundary_end_pfn[nid] > *end_pfn)
2981 *end_pfn = node_boundary_end_pfn[nid];
2984 void __init push_node_boundaries(unsigned int nid,
2985 unsigned long start_pfn, unsigned long end_pfn) {}
2987 static void __meminit account_node_boundary(unsigned int nid,
2988 unsigned long *start_pfn, unsigned long *end_pfn) {}
2993 * get_pfn_range_for_nid - Return the start and end page frames for a node
2994 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
2995 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
2996 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
2998 * It returns the start and end page frame of a node based on information
2999 * provided by an arch calling add_active_range(). If called for a node
3000 * with no available memory, a warning is printed and the start and end
3003 void __meminit get_pfn_range_for_nid(unsigned int nid,
3004 unsigned long *start_pfn, unsigned long *end_pfn)
3010 for_each_active_range_index_in_nid(i, nid) {
3011 *start_pfn = min(*start_pfn, early_node_map[i].start_pfn);
3012 *end_pfn = max(*end_pfn, early_node_map[i].end_pfn);
3015 if (*start_pfn == -1UL)
3018 /* Push the node boundaries out if requested */
3019 account_node_boundary(nid, start_pfn, end_pfn);
3023 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3024 * assumption is made that zones within a node are ordered in monotonic
3025 * increasing memory addresses so that the "highest" populated zone is used
3027 void __init find_usable_zone_for_movable(void)
3030 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
3031 if (zone_index == ZONE_MOVABLE)
3034 if (arch_zone_highest_possible_pfn[zone_index] >
3035 arch_zone_lowest_possible_pfn[zone_index])
3039 VM_BUG_ON(zone_index == -1);
3040 movable_zone = zone_index;
3044 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
3045 * because it is sized independant of architecture. Unlike the other zones,
3046 * the starting point for ZONE_MOVABLE is not fixed. It may be different
3047 * in each node depending on the size of each node and how evenly kernelcore
3048 * is distributed. This helper function adjusts the zone ranges
3049 * provided by the architecture for a given node by using the end of the
3050 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3051 * zones within a node are in order of monotonic increases memory addresses
3053 void __meminit adjust_zone_range_for_zone_movable(int nid,
3054 unsigned long zone_type,
3055 unsigned long node_start_pfn,
3056 unsigned long node_end_pfn,
3057 unsigned long *zone_start_pfn,
3058 unsigned long *zone_end_pfn)
3060 /* Only adjust if ZONE_MOVABLE is on this node */
3061 if (zone_movable_pfn[nid]) {
3062 /* Size ZONE_MOVABLE */
3063 if (zone_type == ZONE_MOVABLE) {
3064 *zone_start_pfn = zone_movable_pfn[nid];
3065 *zone_end_pfn = min(node_end_pfn,
3066 arch_zone_highest_possible_pfn[movable_zone]);
3068 /* Adjust for ZONE_MOVABLE starting within this range */
3069 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
3070 *zone_end_pfn > zone_movable_pfn[nid]) {
3071 *zone_end_pfn = zone_movable_pfn[nid];
3073 /* Check if this whole range is within ZONE_MOVABLE */
3074 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
3075 *zone_start_pfn = *zone_end_pfn;
3080 * Return the number of pages a zone spans in a node, including holes
3081 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3083 static unsigned long __meminit zone_spanned_pages_in_node(int nid,
3084 unsigned long zone_type,
3085 unsigned long *ignored)
3087 unsigned long node_start_pfn, node_end_pfn;
3088 unsigned long zone_start_pfn, zone_end_pfn;
3090 /* Get the start and end of the node and zone */
3091 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3092 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
3093 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
3094 adjust_zone_range_for_zone_movable(nid, zone_type,
3095 node_start_pfn, node_end_pfn,
3096 &zone_start_pfn, &zone_end_pfn);
3098 /* Check that this node has pages within the zone's required range */
3099 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
3102 /* Move the zone boundaries inside the node if necessary */
3103 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
3104 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
3106 /* Return the spanned pages */
3107 return zone_end_pfn - zone_start_pfn;
3111 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
3112 * then all holes in the requested range will be accounted for.
3114 unsigned long __meminit __absent_pages_in_range(int nid,
3115 unsigned long range_start_pfn,
3116 unsigned long range_end_pfn)
3119 unsigned long prev_end_pfn = 0, hole_pages = 0;
3120 unsigned long start_pfn;
3122 /* Find the end_pfn of the first active range of pfns in the node */
3123 i = first_active_region_index_in_nid(nid);
3127 prev_end_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3129 /* Account for ranges before physical memory on this node */
3130 if (early_node_map[i].start_pfn > range_start_pfn)
3131 hole_pages = prev_end_pfn - range_start_pfn;
3133 /* Find all holes for the zone within the node */
3134 for (; i != -1; i = next_active_region_index_in_nid(i, nid)) {
3136 /* No need to continue if prev_end_pfn is outside the zone */
3137 if (prev_end_pfn >= range_end_pfn)
3140 /* Make sure the end of the zone is not within the hole */
3141 start_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3142 prev_end_pfn = max(prev_end_pfn, range_start_pfn);
3144 /* Update the hole size cound and move on */
3145 if (start_pfn > range_start_pfn) {
3146 BUG_ON(prev_end_pfn > start_pfn);
3147 hole_pages += start_pfn - prev_end_pfn;
3149 prev_end_pfn = early_node_map[i].end_pfn;
3152 /* Account for ranges past physical memory on this node */
3153 if (range_end_pfn > prev_end_pfn)
3154 hole_pages += range_end_pfn -
3155 max(range_start_pfn, prev_end_pfn);
3161 * absent_pages_in_range - Return number of page frames in holes within a range
3162 * @start_pfn: The start PFN to start searching for holes
3163 * @end_pfn: The end PFN to stop searching for holes
3165 * It returns the number of pages frames in memory holes within a range.
3167 unsigned long __init absent_pages_in_range(unsigned long start_pfn,
3168 unsigned long end_pfn)
3170 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
3173 /* Return the number of page frames in holes in a zone on a node */
3174 static unsigned long __meminit zone_absent_pages_in_node(int nid,
3175 unsigned long zone_type,
3176 unsigned long *ignored)
3178 unsigned long node_start_pfn, node_end_pfn;
3179 unsigned long zone_start_pfn, zone_end_pfn;
3181 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3182 zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type],
3184 zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type],
3187 adjust_zone_range_for_zone_movable(nid, zone_type,
3188 node_start_pfn, node_end_pfn,
3189 &zone_start_pfn, &zone_end_pfn);
3190 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
3194 static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
3195 unsigned long zone_type,
3196 unsigned long *zones_size)
3198 return zones_size[zone_type];
3201 static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
3202 unsigned long zone_type,
3203 unsigned long *zholes_size)
3208 return zholes_size[zone_type];
3213 static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
3214 unsigned long *zones_size, unsigned long *zholes_size)
3216 unsigned long realtotalpages, totalpages = 0;
3219 for (i = 0; i < MAX_NR_ZONES; i++)
3220 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
3222 pgdat->node_spanned_pages = totalpages;
3224 realtotalpages = totalpages;
3225 for (i = 0; i < MAX_NR_ZONES; i++)
3227 zone_absent_pages_in_node(pgdat->node_id, i,
3229 pgdat->node_present_pages = realtotalpages;
3230 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
3234 #ifndef CONFIG_SPARSEMEM
3236 * Calculate the size of the zone->blockflags rounded to an unsigned long
3237 * Start by making sure zonesize is a multiple of MAX_ORDER-1 by rounding up
3238 * Then figure 1 NR_PAGEBLOCK_BITS worth of bits per MAX_ORDER-1, finally
3239 * round what is now in bits to nearest long in bits, then return it in
3242 static unsigned long __init usemap_size(unsigned long zonesize)
3244 unsigned long usemapsize;
3246 usemapsize = roundup(zonesize, MAX_ORDER_NR_PAGES);
3247 usemapsize = usemapsize >> (MAX_ORDER-1);
3248 usemapsize *= NR_PAGEBLOCK_BITS;
3249 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
3251 return usemapsize / 8;
3254 static void __init setup_usemap(struct pglist_data *pgdat,
3255 struct zone *zone, unsigned long zonesize)
3257 unsigned long usemapsize = usemap_size(zonesize);
3258 zone->pageblock_flags = NULL;
3260 zone->pageblock_flags = alloc_bootmem_node(pgdat, usemapsize);
3261 memset(zone->pageblock_flags, 0, usemapsize);
3265 static void inline setup_usemap(struct pglist_data *pgdat,
3266 struct zone *zone, unsigned long zonesize) {}
3267 #endif /* CONFIG_SPARSEMEM */
3270 * Set up the zone data structures:
3271 * - mark all pages reserved
3272 * - mark all memory queues empty
3273 * - clear the memory bitmaps
3275 static void __meminit free_area_init_core(struct pglist_data *pgdat,
3276 unsigned long *zones_size, unsigned long *zholes_size)
3279 int nid = pgdat->node_id;
3280 unsigned long zone_start_pfn = pgdat->node_start_pfn;
3283 pgdat_resize_init(pgdat);
3284 pgdat->nr_zones = 0;
3285 init_waitqueue_head(&pgdat->kswapd_wait);
3286 pgdat->kswapd_max_order = 0;
3288 for (j = 0; j < MAX_NR_ZONES; j++) {
3289 struct zone *zone = pgdat->node_zones + j;
3290 unsigned long size, realsize, memmap_pages;
3292 size = zone_spanned_pages_in_node(nid, j, zones_size);
3293 realsize = size - zone_absent_pages_in_node(nid, j,
3297 * Adjust realsize so that it accounts for how much memory
3298 * is used by this zone for memmap. This affects the watermark
3299 * and per-cpu initialisations
3301 memmap_pages = (size * sizeof(struct page)) >> PAGE_SHIFT;
3302 if (realsize >= memmap_pages) {
3303 realsize -= memmap_pages;
3305 " %s zone: %lu pages used for memmap\n",
3306 zone_names[j], memmap_pages);
3309 " %s zone: %lu pages exceeds realsize %lu\n",
3310 zone_names[j], memmap_pages, realsize);
3312 /* Account for reserved pages */
3313 if (j == 0 && realsize > dma_reserve) {
3314 realsize -= dma_reserve;
3315 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
3316 zone_names[0], dma_reserve);
3319 if (!is_highmem_idx(j))
3320 nr_kernel_pages += realsize;
3321 nr_all_pages += realsize;
3323 zone->spanned_pages = size;
3324 zone->present_pages = realsize;
3327 zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
3329 zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100;
3331 zone->name = zone_names[j];
3332 spin_lock_init(&zone->lock);
3333 spin_lock_init(&zone->lru_lock);
3334 zone_seqlock_init(zone);
3335 zone->zone_pgdat = pgdat;
3337 zone->prev_priority = DEF_PRIORITY;
3339 zone_pcp_init(zone);
3340 INIT_LIST_HEAD(&zone->active_list);
3341 INIT_LIST_HEAD(&zone->inactive_list);
3342 zone->nr_scan_active = 0;
3343 zone->nr_scan_inactive = 0;
3344 zap_zone_vm_stats(zone);
3345 atomic_set(&zone->reclaim_in_progress, 0);
3349 setup_usemap(pgdat, zone, size);
3350 ret = init_currently_empty_zone(zone, zone_start_pfn,
3351 size, MEMMAP_EARLY);
3353 zone_start_pfn += size;
3357 static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
3359 /* Skip empty nodes */
3360 if (!pgdat->node_spanned_pages)
3363 #ifdef CONFIG_FLAT_NODE_MEM_MAP
3364 /* ia64 gets its own node_mem_map, before this, without bootmem */
3365 if (!pgdat->node_mem_map) {
3366 unsigned long size, start, end;
3370 * The zone's endpoints aren't required to be MAX_ORDER
3371 * aligned but the node_mem_map endpoints must be in order
3372 * for the buddy allocator to function correctly.
3374 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
3375 end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
3376 end = ALIGN(end, MAX_ORDER_NR_PAGES);
3377 size = (end - start) * sizeof(struct page);
3378 map = alloc_remap(pgdat->node_id, size);
3380 map = alloc_bootmem_node(pgdat, size);
3381 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
3383 #ifndef CONFIG_NEED_MULTIPLE_NODES
3385 * With no DISCONTIG, the global mem_map is just set as node 0's
3387 if (pgdat == NODE_DATA(0)) {
3388 mem_map = NODE_DATA(0)->node_mem_map;
3389 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3390 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
3391 mem_map -= pgdat->node_start_pfn;
3392 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3395 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
3398 void __meminit free_area_init_node(int nid, struct pglist_data *pgdat,
3399 unsigned long *zones_size, unsigned long node_start_pfn,
3400 unsigned long *zholes_size)
3402 pgdat->node_id = nid;
3403 pgdat->node_start_pfn = node_start_pfn;
3404 calculate_node_totalpages(pgdat, zones_size, zholes_size);
3406 alloc_node_mem_map(pgdat);
3408 free_area_init_core(pgdat, zones_size, zholes_size);
3411 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3413 #if MAX_NUMNODES > 1
3415 * Figure out the number of possible node ids.
3417 static void __init setup_nr_node_ids(void)
3420 unsigned int highest = 0;
3422 for_each_node_mask(node, node_possible_map)
3424 nr_node_ids = highest + 1;
3427 static inline void setup_nr_node_ids(void)
3433 * add_active_range - Register a range of PFNs backed by physical memory
3434 * @nid: The node ID the range resides on
3435 * @start_pfn: The start PFN of the available physical memory
3436 * @end_pfn: The end PFN of the available physical memory
3438 * These ranges are stored in an early_node_map[] and later used by
3439 * free_area_init_nodes() to calculate zone sizes and holes. If the
3440 * range spans a memory hole, it is up to the architecture to ensure
3441 * the memory is not freed by the bootmem allocator. If possible
3442 * the range being registered will be merged with existing ranges.
3444 void __init add_active_range(unsigned int nid, unsigned long start_pfn,
3445 unsigned long end_pfn)
3449 printk(KERN_DEBUG "Entering add_active_range(%d, %lu, %lu) "
3450 "%d entries of %d used\n",
3451 nid, start_pfn, end_pfn,
3452 nr_nodemap_entries, MAX_ACTIVE_REGIONS);
3454 /* Merge with existing active regions if possible */
3455 for (i = 0; i < nr_nodemap_entries; i++) {
3456 if (early_node_map[i].nid != nid)
3459 /* Skip if an existing region covers this new one */
3460 if (start_pfn >= early_node_map[i].start_pfn &&
3461 end_pfn <= early_node_map[i].end_pfn)
3464 /* Merge forward if suitable */
3465 if (start_pfn <= early_node_map[i].end_pfn &&
3466 end_pfn > early_node_map[i].end_pfn) {
3467 early_node_map[i].end_pfn = end_pfn;
3471 /* Merge backward if suitable */
3472 if (start_pfn < early_node_map[i].end_pfn &&
3473 end_pfn >= early_node_map[i].start_pfn) {
3474 early_node_map[i].start_pfn = start_pfn;
3479 /* Check that early_node_map is large enough */
3480 if (i >= MAX_ACTIVE_REGIONS) {
3481 printk(KERN_CRIT "More than %d memory regions, truncating\n",
3482 MAX_ACTIVE_REGIONS);
3486 early_node_map[i].nid = nid;
3487 early_node_map[i].start_pfn = start_pfn;
3488 early_node_map[i].end_pfn = end_pfn;
3489 nr_nodemap_entries = i + 1;
3493 * shrink_active_range - Shrink an existing registered range of PFNs
3494 * @nid: The node id the range is on that should be shrunk
3495 * @old_end_pfn: The old end PFN of the range
3496 * @new_end_pfn: The new PFN of the range
3498 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
3499 * The map is kept at the end physical page range that has already been
3500 * registered with add_active_range(). This function allows an arch to shrink
3501 * an existing registered range.
3503 void __init shrink_active_range(unsigned int nid, unsigned long old_end_pfn,
3504 unsigned long new_end_pfn)
3508 /* Find the old active region end and shrink */
3509 for_each_active_range_index_in_nid(i, nid)
3510 if (early_node_map[i].end_pfn == old_end_pfn) {
3511 early_node_map[i].end_pfn = new_end_pfn;
3517 * remove_all_active_ranges - Remove all currently registered regions
3519 * During discovery, it may be found that a table like SRAT is invalid
3520 * and an alternative discovery method must be used. This function removes
3521 * all currently registered regions.
3523 void __init remove_all_active_ranges(void)
3525 memset(early_node_map, 0, sizeof(early_node_map));
3526 nr_nodemap_entries = 0;
3527 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
3528 memset(node_boundary_start_pfn, 0, sizeof(node_boundary_start_pfn));
3529 memset(node_boundary_end_pfn, 0, sizeof(node_boundary_end_pfn));
3530 #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
3533 /* Compare two active node_active_regions */
3534 static int __init cmp_node_active_region(const void *a, const void *b)
3536 struct node_active_region *arange = (struct node_active_region *)a;
3537 struct node_active_region *brange = (struct node_active_region *)b;
3539 /* Done this way to avoid overflows */
3540 if (arange->start_pfn > brange->start_pfn)
3542 if (arange->start_pfn < brange->start_pfn)
3548 /* sort the node_map by start_pfn */
3549 static void __init sort_node_map(void)
3551 sort(early_node_map, (size_t)nr_nodemap_entries,
3552 sizeof(struct node_active_region),
3553 cmp_node_active_region, NULL);
3556 /* Find the lowest pfn for a node */
3557 unsigned long __init find_min_pfn_for_node(unsigned long nid)
3560 unsigned long min_pfn = ULONG_MAX;
3562 /* Assuming a sorted map, the first range found has the starting pfn */
3563 for_each_active_range_index_in_nid(i, nid)
3564 min_pfn = min(min_pfn, early_node_map[i].start_pfn);
3566 if (min_pfn == ULONG_MAX) {
3568 "Could not find start_pfn for node %lu\n", nid);
3576 * find_min_pfn_with_active_regions - Find the minimum PFN registered
3578 * It returns the minimum PFN based on information provided via
3579 * add_active_range().
3581 unsigned long __init find_min_pfn_with_active_regions(void)
3583 return find_min_pfn_for_node(MAX_NUMNODES);
3587 * find_max_pfn_with_active_regions - Find the maximum PFN registered
3589 * It returns the maximum PFN based on information provided via
3590 * add_active_range().
3592 unsigned long __init find_max_pfn_with_active_regions(void)
3595 unsigned long max_pfn = 0;
3597 for (i = 0; i < nr_nodemap_entries; i++)
3598 max_pfn = max(max_pfn, early_node_map[i].end_pfn);
3604 * early_calculate_totalpages()
3605 * Sum pages in active regions for movable zone.
3606 * Populate N_HIGH_MEMORY for calculating usable_nodes.
3608 unsigned long __init early_calculate_totalpages(void)
3611 unsigned long totalpages = 0;
3613 for (i = 0; i < nr_nodemap_entries; i++) {
3614 unsigned long pages = early_node_map[i].end_pfn -
3615 early_node_map[i].start_pfn;
3616 totalpages += pages;
3618 node_set_state(early_node_map[i].nid, N_HIGH_MEMORY);
3624 * Find the PFN the Movable zone begins in each node. Kernel memory
3625 * is spread evenly between nodes as long as the nodes have enough
3626 * memory. When they don't, some nodes will have more kernelcore than
3629 void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn)
3632 unsigned long usable_startpfn;
3633 unsigned long kernelcore_node, kernelcore_remaining;
3634 unsigned long totalpages = early_calculate_totalpages();
3635 int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]);
3638 * If movablecore was specified, calculate what size of
3639 * kernelcore that corresponds so that memory usable for
3640 * any allocation type is evenly spread. If both kernelcore
3641 * and movablecore are specified, then the value of kernelcore
3642 * will be used for required_kernelcore if it's greater than
3643 * what movablecore would have allowed.
3645 if (required_movablecore) {
3646 unsigned long corepages;
3649 * Round-up so that ZONE_MOVABLE is at least as large as what
3650 * was requested by the user
3652 required_movablecore =
3653 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
3654 corepages = totalpages - required_movablecore;
3656 required_kernelcore = max(required_kernelcore, corepages);
3659 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
3660 if (!required_kernelcore)
3663 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
3664 find_usable_zone_for_movable();
3665 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
3668 /* Spread kernelcore memory as evenly as possible throughout nodes */
3669 kernelcore_node = required_kernelcore / usable_nodes;
3670 for_each_node_state(nid, N_HIGH_MEMORY) {
3672 * Recalculate kernelcore_node if the division per node
3673 * now exceeds what is necessary to satisfy the requested
3674 * amount of memory for the kernel
3676 if (required_kernelcore < kernelcore_node)
3677 kernelcore_node = required_kernelcore / usable_nodes;
3680 * As the map is walked, we track how much memory is usable
3681 * by the kernel using kernelcore_remaining. When it is
3682 * 0, the rest of the node is usable by ZONE_MOVABLE
3684 kernelcore_remaining = kernelcore_node;
3686 /* Go through each range of PFNs within this node */
3687 for_each_active_range_index_in_nid(i, nid) {
3688 unsigned long start_pfn, end_pfn;
3689 unsigned long size_pages;
3691 start_pfn = max(early_node_map[i].start_pfn,
3692 zone_movable_pfn[nid]);
3693 end_pfn = early_node_map[i].end_pfn;
3694 if (start_pfn >= end_pfn)
3697 /* Account for what is only usable for kernelcore */
3698 if (start_pfn < usable_startpfn) {
3699 unsigned long kernel_pages;
3700 kernel_pages = min(end_pfn, usable_startpfn)
3703 kernelcore_remaining -= min(kernel_pages,
3704 kernelcore_remaining);
3705 required_kernelcore -= min(kernel_pages,
3706 required_kernelcore);
3708 /* Continue if range is now fully accounted */
3709 if (end_pfn <= usable_startpfn) {
3712 * Push zone_movable_pfn to the end so
3713 * that if we have to rebalance
3714 * kernelcore across nodes, we will
3715 * not double account here
3717 zone_movable_pfn[nid] = end_pfn;
3720 start_pfn = usable_startpfn;
3724 * The usable PFN range for ZONE_MOVABLE is from
3725 * start_pfn->end_pfn. Calculate size_pages as the
3726 * number of pages used as kernelcore
3728 size_pages = end_pfn - start_pfn;
3729 if (size_pages > kernelcore_remaining)
3730 size_pages = kernelcore_remaining;
3731 zone_movable_pfn[nid] = start_pfn + size_pages;
3734 * Some kernelcore has been met, update counts and
3735 * break if the kernelcore for this node has been
3738 required_kernelcore -= min(required_kernelcore,
3740 kernelcore_remaining -= size_pages;
3741 if (!kernelcore_remaining)
3747 * If there is still required_kernelcore, we do another pass with one
3748 * less node in the count. This will push zone_movable_pfn[nid] further
3749 * along on the nodes that still have memory until kernelcore is
3753 if (usable_nodes && required_kernelcore > usable_nodes)
3756 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
3757 for (nid = 0; nid < MAX_NUMNODES; nid++)
3758 zone_movable_pfn[nid] =
3759 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
3762 /* Any regular memory on that node ? */
3763 static void check_for_regular_memory(pg_data_t *pgdat)
3765 #ifdef CONFIG_HIGHMEM
3766 enum zone_type zone_type;
3768 for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) {
3769 struct zone *zone = &pgdat->node_zones[zone_type];
3770 if (zone->present_pages)
3771 node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY);
3777 * free_area_init_nodes - Initialise all pg_data_t and zone data
3778 * @max_zone_pfn: an array of max PFNs for each zone
3780 * This will call free_area_init_node() for each active node in the system.
3781 * Using the page ranges provided by add_active_range(), the size of each
3782 * zone in each node and their holes is calculated. If the maximum PFN
3783 * between two adjacent zones match, it is assumed that the zone is empty.
3784 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
3785 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
3786 * starts where the previous one ended. For example, ZONE_DMA32 starts
3787 * at arch_max_dma_pfn.
3789 void __init free_area_init_nodes(unsigned long *max_zone_pfn)
3794 /* Sort early_node_map as initialisation assumes it is sorted */
3797 /* Record where the zone boundaries are */
3798 memset(arch_zone_lowest_possible_pfn, 0,
3799 sizeof(arch_zone_lowest_possible_pfn));
3800 memset(arch_zone_highest_possible_pfn, 0,
3801 sizeof(arch_zone_highest_possible_pfn));
3802 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
3803 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
3804 for (i = 1; i < MAX_NR_ZONES; i++) {
3805 if (i == ZONE_MOVABLE)
3807 arch_zone_lowest_possible_pfn[i] =
3808 arch_zone_highest_possible_pfn[i-1];
3809 arch_zone_highest_possible_pfn[i] =
3810 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
3812 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
3813 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
3815 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
3816 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
3817 find_zone_movable_pfns_for_nodes(zone_movable_pfn);
3819 /* Print out the zone ranges */
3820 printk("Zone PFN ranges:\n");
3821 for (i = 0; i < MAX_NR_ZONES; i++) {
3822 if (i == ZONE_MOVABLE)
3824 printk(" %-8s %8lu -> %8lu\n",
3826 arch_zone_lowest_possible_pfn[i],
3827 arch_zone_highest_possible_pfn[i]);
3830 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
3831 printk("Movable zone start PFN for each node\n");
3832 for (i = 0; i < MAX_NUMNODES; i++) {
3833 if (zone_movable_pfn[i])
3834 printk(" Node %d: %lu\n", i, zone_movable_pfn[i]);
3837 /* Print out the early_node_map[] */
3838 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries);
3839 for (i = 0; i < nr_nodemap_entries; i++)
3840 printk(" %3d: %8lu -> %8lu\n", early_node_map[i].nid,
3841 early_node_map[i].start_pfn,
3842 early_node_map[i].end_pfn);
3844 /* Initialise every node */
3845 setup_nr_node_ids();
3846 for_each_online_node(nid) {
3847 pg_data_t *pgdat = NODE_DATA(nid);
3848 free_area_init_node(nid, pgdat, NULL,
3849 find_min_pfn_for_node(nid), NULL);
3851 /* Any memory on that node */
3852 if (pgdat->node_present_pages)
3853 node_set_state(nid, N_HIGH_MEMORY);
3854 check_for_regular_memory(pgdat);
3858 static int __init cmdline_parse_core(char *p, unsigned long *core)
3860 unsigned long long coremem;
3864 coremem = memparse(p, &p);
3865 *core = coremem >> PAGE_SHIFT;
3867 /* Paranoid check that UL is enough for the coremem value */
3868 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
3874 * kernelcore=size sets the amount of memory for use for allocations that
3875 * cannot be reclaimed or migrated.
3877 static int __init cmdline_parse_kernelcore(char *p)
3879 return cmdline_parse_core(p, &required_kernelcore);
3883 * movablecore=size sets the amount of memory for use for allocations that
3884 * can be reclaimed or migrated.
3886 static int __init cmdline_parse_movablecore(char *p)
3888 return cmdline_parse_core(p, &required_movablecore);
3891 early_param("kernelcore", cmdline_parse_kernelcore);
3892 early_param("movablecore", cmdline_parse_movablecore);
3894 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3897 * set_dma_reserve - set the specified number of pages reserved in the first zone
3898 * @new_dma_reserve: The number of pages to mark reserved
3900 * The per-cpu batchsize and zone watermarks are determined by present_pages.
3901 * In the DMA zone, a significant percentage may be consumed by kernel image
3902 * and other unfreeable allocations which can skew the watermarks badly. This
3903 * function may optionally be used to account for unfreeable pages in the
3904 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
3905 * smaller per-cpu batchsize.
3907 void __init set_dma_reserve(unsigned long new_dma_reserve)
3909 dma_reserve = new_dma_reserve;
3912 #ifndef CONFIG_NEED_MULTIPLE_NODES
3913 static bootmem_data_t contig_bootmem_data;
3914 struct pglist_data contig_page_data = { .bdata = &contig_bootmem_data };
3916 EXPORT_SYMBOL(contig_page_data);
3919 void __init free_area_init(unsigned long *zones_size)
3921 free_area_init_node(0, NODE_DATA(0), zones_size,
3922 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
3925 static int page_alloc_cpu_notify(struct notifier_block *self,
3926 unsigned long action, void *hcpu)
3928 int cpu = (unsigned long)hcpu;
3930 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
3931 local_irq_disable();
3933 vm_events_fold_cpu(cpu);
3935 refresh_cpu_vm_stats(cpu);
3940 void __init page_alloc_init(void)
3942 hotcpu_notifier(page_alloc_cpu_notify, 0);
3946 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
3947 * or min_free_kbytes changes.
3949 static void calculate_totalreserve_pages(void)
3951 struct pglist_data *pgdat;
3952 unsigned long reserve_pages = 0;
3953 enum zone_type i, j;
3955 for_each_online_pgdat(pgdat) {
3956 for (i = 0; i < MAX_NR_ZONES; i++) {
3957 struct zone *zone = pgdat->node_zones + i;
3958 unsigned long max = 0;
3960 /* Find valid and maximum lowmem_reserve in the zone */
3961 for (j = i; j < MAX_NR_ZONES; j++) {
3962 if (zone->lowmem_reserve[j] > max)
3963 max = zone->lowmem_reserve[j];
3966 /* we treat pages_high as reserved pages. */
3967 max += zone->pages_high;
3969 if (max > zone->present_pages)
3970 max = zone->present_pages;
3971 reserve_pages += max;
3974 totalreserve_pages = reserve_pages;
3978 * setup_per_zone_lowmem_reserve - called whenever
3979 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
3980 * has a correct pages reserved value, so an adequate number of
3981 * pages are left in the zone after a successful __alloc_pages().
3983 static void setup_per_zone_lowmem_reserve(void)
3985 struct pglist_data *pgdat;
3986 enum zone_type j, idx;
3988 for_each_online_pgdat(pgdat) {
3989 for (j = 0; j < MAX_NR_ZONES; j++) {
3990 struct zone *zone = pgdat->node_zones + j;
3991 unsigned long present_pages = zone->present_pages;
3993 zone->lowmem_reserve[j] = 0;
3997 struct zone *lower_zone;
4001 if (sysctl_lowmem_reserve_ratio[idx] < 1)
4002 sysctl_lowmem_reserve_ratio[idx] = 1;
4004 lower_zone = pgdat->node_zones + idx;
4005 lower_zone->lowmem_reserve[j] = present_pages /
4006 sysctl_lowmem_reserve_ratio[idx];
4007 present_pages += lower_zone->present_pages;
4012 /* update totalreserve_pages */
4013 calculate_totalreserve_pages();
4017 * setup_per_zone_pages_min - called when min_free_kbytes changes.
4019 * Ensures that the pages_{min,low,high} values for each zone are set correctly
4020 * with respect to min_free_kbytes.
4022 void setup_per_zone_pages_min(void)
4024 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
4025 unsigned long lowmem_pages = 0;
4027 unsigned long flags;
4029 /* Calculate total number of !ZONE_HIGHMEM pages */
4030 for_each_zone(zone) {
4031 if (!is_highmem(zone))
4032 lowmem_pages += zone->present_pages;
4035 for_each_zone(zone) {
4038 spin_lock_irqsave(&zone->lru_lock, flags);
4039 tmp = (u64)pages_min * zone->present_pages;
4040 do_div(tmp, lowmem_pages);
4041 if (is_highmem(zone)) {
4043 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
4044 * need highmem pages, so cap pages_min to a small
4047 * The (pages_high-pages_low) and (pages_low-pages_min)
4048 * deltas controls asynch page reclaim, and so should
4049 * not be capped for highmem.
4053 min_pages = zone->present_pages / 1024;
4054 if (min_pages < SWAP_CLUSTER_MAX)
4055 min_pages = SWAP_CLUSTER_MAX;
4056 if (min_pages > 128)
4058 zone->pages_min = min_pages;
4061 * If it's a lowmem zone, reserve a number of pages
4062 * proportionate to the zone's size.
4064 zone->pages_min = tmp;
4067 zone->pages_low = zone->pages_min + (tmp >> 2);
4068 zone->pages_high = zone->pages_min + (tmp >> 1);
4069 spin_unlock_irqrestore(&zone->lru_lock, flags);
4072 /* update totalreserve_pages */
4073 calculate_totalreserve_pages();
4077 * Initialise min_free_kbytes.
4079 * For small machines we want it small (128k min). For large machines
4080 * we want it large (64MB max). But it is not linear, because network
4081 * bandwidth does not increase linearly with machine size. We use
4083 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
4084 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
4100 static int __init init_per_zone_pages_min(void)
4102 unsigned long lowmem_kbytes;
4104 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
4106 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
4107 if (min_free_kbytes < 128)
4108 min_free_kbytes = 128;
4109 if (min_free_kbytes > 65536)
4110 min_free_kbytes = 65536;
4111 setup_per_zone_pages_min();
4112 setup_per_zone_lowmem_reserve();
4115 module_init(init_per_zone_pages_min)
4118 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
4119 * that we can call two helper functions whenever min_free_kbytes
4122 int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
4123 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4125 proc_dointvec(table, write, file, buffer, length, ppos);
4127 setup_per_zone_pages_min();
4132 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
4133 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4138 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4143 zone->min_unmapped_pages = (zone->present_pages *
4144 sysctl_min_unmapped_ratio) / 100;
4148 int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
4149 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4154 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4159 zone->min_slab_pages = (zone->present_pages *
4160 sysctl_min_slab_ratio) / 100;
4166 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
4167 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
4168 * whenever sysctl_lowmem_reserve_ratio changes.
4170 * The reserve ratio obviously has absolutely no relation with the
4171 * pages_min watermarks. The lowmem reserve ratio can only make sense
4172 * if in function of the boot time zone sizes.
4174 int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
4175 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4177 proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4178 setup_per_zone_lowmem_reserve();
4183 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
4184 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
4185 * can have before it gets flushed back to buddy allocator.
4188 int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
4189 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4195 ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4196 if (!write || (ret == -EINVAL))
4198 for_each_zone(zone) {
4199 for_each_online_cpu(cpu) {
4201 high = zone->present_pages / percpu_pagelist_fraction;
4202 setup_pagelist_highmark(zone_pcp(zone, cpu), high);
4208 int hashdist = HASHDIST_DEFAULT;
4211 static int __init set_hashdist(char *str)
4215 hashdist = simple_strtoul(str, &str, 0);
4218 __setup("hashdist=", set_hashdist);
4222 * allocate a large system hash table from bootmem
4223 * - it is assumed that the hash table must contain an exact power-of-2
4224 * quantity of entries
4225 * - limit is the number of hash buckets, not the total allocation size
4227 void *__init alloc_large_system_hash(const char *tablename,
4228 unsigned long bucketsize,
4229 unsigned long numentries,
4232 unsigned int *_hash_shift,
4233 unsigned int *_hash_mask,
4234 unsigned long limit)
4236 unsigned long long max = limit;
4237 unsigned long log2qty, size;
4240 /* allow the kernel cmdline to have a say */
4242 /* round applicable memory size up to nearest megabyte */
4243 numentries = nr_kernel_pages;
4244 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
4245 numentries >>= 20 - PAGE_SHIFT;
4246 numentries <<= 20 - PAGE_SHIFT;
4248 /* limit to 1 bucket per 2^scale bytes of low memory */
4249 if (scale > PAGE_SHIFT)
4250 numentries >>= (scale - PAGE_SHIFT);
4252 numentries <<= (PAGE_SHIFT - scale);
4254 /* Make sure we've got at least a 0-order allocation.. */
4255 if (unlikely((numentries * bucketsize) < PAGE_SIZE))
4256 numentries = PAGE_SIZE / bucketsize;
4258 numentries = roundup_pow_of_two(numentries);
4260 /* limit allocation size to 1/16 total memory by default */
4262 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
4263 do_div(max, bucketsize);
4266 if (numentries > max)
4269 log2qty = ilog2(numentries);
4272 size = bucketsize << log2qty;
4273 if (flags & HASH_EARLY)
4274 table = alloc_bootmem(size);
4276 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
4278 unsigned long order;
4279 for (order = 0; ((1UL << order) << PAGE_SHIFT) < size; order++)
4281 table = (void*) __get_free_pages(GFP_ATOMIC, order);
4283 * If bucketsize is not a power-of-two, we may free
4284 * some pages at the end of hash table.
4287 unsigned long alloc_end = (unsigned long)table +
4288 (PAGE_SIZE << order);
4289 unsigned long used = (unsigned long)table +
4291 split_page(virt_to_page(table), order);
4292 while (used < alloc_end) {
4298 } while (!table && size > PAGE_SIZE && --log2qty);
4301 panic("Failed to allocate %s hash table\n", tablename);
4303 printk(KERN_INFO "%s hash table entries: %d (order: %d, %lu bytes)\n",
4306 ilog2(size) - PAGE_SHIFT,
4310 *_hash_shift = log2qty;
4312 *_hash_mask = (1 << log2qty) - 1;
4317 #ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE
4318 struct page *pfn_to_page(unsigned long pfn)
4320 return __pfn_to_page(pfn);
4322 unsigned long page_to_pfn(struct page *page)
4324 return __page_to_pfn(page);
4326 EXPORT_SYMBOL(pfn_to_page);
4327 EXPORT_SYMBOL(page_to_pfn);
4328 #endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */
4330 /* Return a pointer to the bitmap storing bits affecting a block of pages */
4331 static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
4334 #ifdef CONFIG_SPARSEMEM
4335 return __pfn_to_section(pfn)->pageblock_flags;
4337 return zone->pageblock_flags;
4338 #endif /* CONFIG_SPARSEMEM */
4341 static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
4343 #ifdef CONFIG_SPARSEMEM
4344 pfn &= (PAGES_PER_SECTION-1);
4345 return (pfn >> (MAX_ORDER-1)) * NR_PAGEBLOCK_BITS;
4347 pfn = pfn - zone->zone_start_pfn;
4348 return (pfn >> (MAX_ORDER-1)) * NR_PAGEBLOCK_BITS;
4349 #endif /* CONFIG_SPARSEMEM */
4353 * get_pageblock_flags_group - Return the requested group of flags for the MAX_ORDER_NR_PAGES block of pages
4354 * @page: The page within the block of interest
4355 * @start_bitidx: The first bit of interest to retrieve
4356 * @end_bitidx: The last bit of interest
4357 * returns pageblock_bits flags
4359 unsigned long get_pageblock_flags_group(struct page *page,
4360 int start_bitidx, int end_bitidx)
4363 unsigned long *bitmap;
4364 unsigned long pfn, bitidx;
4365 unsigned long flags = 0;
4366 unsigned long value = 1;
4368 zone = page_zone(page);
4369 pfn = page_to_pfn(page);
4370 bitmap = get_pageblock_bitmap(zone, pfn);
4371 bitidx = pfn_to_bitidx(zone, pfn);
4373 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
4374 if (test_bit(bitidx + start_bitidx, bitmap))
4381 * set_pageblock_flags_group - Set the requested group of flags for a MAX_ORDER_NR_PAGES block of pages
4382 * @page: The page within the block of interest
4383 * @start_bitidx: The first bit of interest
4384 * @end_bitidx: The last bit of interest
4385 * @flags: The flags to set
4387 void set_pageblock_flags_group(struct page *page, unsigned long flags,
4388 int start_bitidx, int end_bitidx)
4391 unsigned long *bitmap;
4392 unsigned long pfn, bitidx;
4393 unsigned long value = 1;
4395 zone = page_zone(page);
4396 pfn = page_to_pfn(page);
4397 bitmap = get_pageblock_bitmap(zone, pfn);
4398 bitidx = pfn_to_bitidx(zone, pfn);
4400 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
4402 __set_bit(bitidx + start_bitidx, bitmap);
4404 __clear_bit(bitidx + start_bitidx, bitmap);