2 * linux/mm/page_alloc.c
4 * Manages the free list, the system allocates free pages here.
5 * Note that kmalloc() lives in slab.c
7 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
8 * Swap reorganised 29.12.95, Stephen Tweedie
9 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
10 * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999
11 * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
12 * Zone balancing, Kanoj Sarcar, SGI, Jan 2000
13 * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002
14 * (lots of bits borrowed from Ingo Molnar & Andrew Morton)
17 #include <linux/stddef.h>
19 #include <linux/swap.h>
20 #include <linux/interrupt.h>
21 #include <linux/pagemap.h>
22 #include <linux/jiffies.h>
23 #include <linux/bootmem.h>
24 #include <linux/compiler.h>
25 #include <linux/kernel.h>
26 #include <linux/module.h>
27 #include <linux/suspend.h>
28 #include <linux/pagevec.h>
29 #include <linux/blkdev.h>
30 #include <linux/slab.h>
31 #include <linux/oom.h>
32 #include <linux/notifier.h>
33 #include <linux/topology.h>
34 #include <linux/sysctl.h>
35 #include <linux/cpu.h>
36 #include <linux/cpuset.h>
37 #include <linux/memory_hotplug.h>
38 #include <linux/nodemask.h>
39 #include <linux/vmalloc.h>
40 #include <linux/mempolicy.h>
41 #include <linux/stop_machine.h>
42 #include <linux/sort.h>
43 #include <linux/pfn.h>
44 #include <linux/backing-dev.h>
45 #include <linux/fault-inject.h>
46 #include <linux/page-isolation.h>
47 #include <linux/page_cgroup.h>
48 #include <linux/debugobjects.h>
49 #include <linux/kmemleak.h>
51 #include <asm/tlbflush.h>
52 #include <asm/div64.h>
56 * Array of node states.
58 nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
59 [N_POSSIBLE] = NODE_MASK_ALL,
60 [N_ONLINE] = { { [0] = 1UL } },
62 [N_NORMAL_MEMORY] = { { [0] = 1UL } },
64 [N_HIGH_MEMORY] = { { [0] = 1UL } },
66 [N_CPU] = { { [0] = 1UL } },
69 EXPORT_SYMBOL(node_states);
71 unsigned long totalram_pages __read_mostly;
72 unsigned long totalreserve_pages __read_mostly;
73 unsigned long highest_memmap_pfn __read_mostly;
74 int percpu_pagelist_fraction;
76 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
77 int pageblock_order __read_mostly;
80 static void __free_pages_ok(struct page *page, unsigned int order);
83 * results with 256, 32 in the lowmem_reserve sysctl:
84 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
85 * 1G machine -> (16M dma, 784M normal, 224M high)
86 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
87 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
88 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
90 * TBD: should special case ZONE_DMA32 machines here - in those we normally
91 * don't need any ZONE_NORMAL reservation
93 int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
94 #ifdef CONFIG_ZONE_DMA
97 #ifdef CONFIG_ZONE_DMA32
100 #ifdef CONFIG_HIGHMEM
106 EXPORT_SYMBOL(totalram_pages);
108 static char * const zone_names[MAX_NR_ZONES] = {
109 #ifdef CONFIG_ZONE_DMA
112 #ifdef CONFIG_ZONE_DMA32
116 #ifdef CONFIG_HIGHMEM
122 int min_free_kbytes = 1024;
124 unsigned long __meminitdata nr_kernel_pages;
125 unsigned long __meminitdata nr_all_pages;
126 static unsigned long __meminitdata dma_reserve;
128 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
130 * MAX_ACTIVE_REGIONS determines the maximum number of distinct
131 * ranges of memory (RAM) that may be registered with add_active_range().
132 * Ranges passed to add_active_range() will be merged if possible
133 * so the number of times add_active_range() can be called is
134 * related to the number of nodes and the number of holes
136 #ifdef CONFIG_MAX_ACTIVE_REGIONS
137 /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
138 #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
140 #if MAX_NUMNODES >= 32
141 /* If there can be many nodes, allow up to 50 holes per node */
142 #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
144 /* By default, allow up to 256 distinct regions */
145 #define MAX_ACTIVE_REGIONS 256
149 static struct node_active_region __meminitdata early_node_map[MAX_ACTIVE_REGIONS];
150 static int __meminitdata nr_nodemap_entries;
151 static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
152 static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
153 static unsigned long __initdata required_kernelcore;
154 static unsigned long __initdata required_movablecore;
155 static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
157 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
159 EXPORT_SYMBOL(movable_zone);
160 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
163 int nr_node_ids __read_mostly = MAX_NUMNODES;
164 int nr_online_nodes __read_mostly = 1;
165 EXPORT_SYMBOL(nr_node_ids);
166 EXPORT_SYMBOL(nr_online_nodes);
169 int page_group_by_mobility_disabled __read_mostly;
171 static void set_pageblock_migratetype(struct page *page, int migratetype)
174 if (unlikely(page_group_by_mobility_disabled))
175 migratetype = MIGRATE_UNMOVABLE;
177 set_pageblock_flags_group(page, (unsigned long)migratetype,
178 PB_migrate, PB_migrate_end);
181 #ifdef CONFIG_DEBUG_VM
182 static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
186 unsigned long pfn = page_to_pfn(page);
189 seq = zone_span_seqbegin(zone);
190 if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
192 else if (pfn < zone->zone_start_pfn)
194 } while (zone_span_seqretry(zone, seq));
199 static int page_is_consistent(struct zone *zone, struct page *page)
201 if (!pfn_valid_within(page_to_pfn(page)))
203 if (zone != page_zone(page))
209 * Temporary debugging check for pages not lying within a given zone.
211 static int bad_range(struct zone *zone, struct page *page)
213 if (page_outside_zone_boundaries(zone, page))
215 if (!page_is_consistent(zone, page))
221 static inline int bad_range(struct zone *zone, struct page *page)
227 static void bad_page(struct page *page)
229 static unsigned long resume;
230 static unsigned long nr_shown;
231 static unsigned long nr_unshown;
234 * Allow a burst of 60 reports, then keep quiet for that minute;
235 * or allow a steady drip of one report per second.
237 if (nr_shown == 60) {
238 if (time_before(jiffies, resume)) {
244 "BUG: Bad page state: %lu messages suppressed\n",
251 resume = jiffies + 60 * HZ;
253 printk(KERN_ALERT "BUG: Bad page state in process %s pfn:%05lx\n",
254 current->comm, page_to_pfn(page));
256 "page:%p flags:%p count:%d mapcount:%d mapping:%p index:%lx\n",
257 page, (void *)page->flags, page_count(page),
258 page_mapcount(page), page->mapping, page->index);
262 /* Leave bad fields for debug, except PageBuddy could make trouble */
263 __ClearPageBuddy(page);
264 add_taint(TAINT_BAD_PAGE);
268 * Higher-order pages are called "compound pages". They are structured thusly:
270 * The first PAGE_SIZE page is called the "head page".
272 * The remaining PAGE_SIZE pages are called "tail pages".
274 * All pages have PG_compound set. All pages have their ->private pointing at
275 * the head page (even the head page has this).
277 * The first tail page's ->lru.next holds the address of the compound page's
278 * put_page() function. Its ->lru.prev holds the order of allocation.
279 * This usage means that zero-order pages may not be compound.
282 static void free_compound_page(struct page *page)
284 __free_pages_ok(page, compound_order(page));
287 void prep_compound_page(struct page *page, unsigned long order)
290 int nr_pages = 1 << order;
292 set_compound_page_dtor(page, free_compound_page);
293 set_compound_order(page, order);
295 for (i = 1; i < nr_pages; i++) {
296 struct page *p = page + i;
299 p->first_page = page;
303 #ifdef CONFIG_HUGETLBFS
304 void prep_compound_gigantic_page(struct page *page, unsigned long order)
307 int nr_pages = 1 << order;
308 struct page *p = page + 1;
310 set_compound_page_dtor(page, free_compound_page);
311 set_compound_order(page, order);
313 for (i = 1; i < nr_pages; i++, p = mem_map_next(p, page, i)) {
315 p->first_page = page;
320 static int destroy_compound_page(struct page *page, unsigned long order)
323 int nr_pages = 1 << order;
326 if (unlikely(compound_order(page) != order) ||
327 unlikely(!PageHead(page))) {
332 __ClearPageHead(page);
334 for (i = 1; i < nr_pages; i++) {
335 struct page *p = page + i;
337 if (unlikely(!PageTail(p) || (p->first_page != page))) {
347 static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
352 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
353 * and __GFP_HIGHMEM from hard or soft interrupt context.
355 VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
356 for (i = 0; i < (1 << order); i++)
357 clear_highpage(page + i);
360 static inline void set_page_order(struct page *page, int order)
362 set_page_private(page, order);
363 __SetPageBuddy(page);
366 static inline void rmv_page_order(struct page *page)
368 __ClearPageBuddy(page);
369 set_page_private(page, 0);
373 * Locate the struct page for both the matching buddy in our
374 * pair (buddy1) and the combined O(n+1) page they form (page).
376 * 1) Any buddy B1 will have an order O twin B2 which satisfies
377 * the following equation:
379 * For example, if the starting buddy (buddy2) is #8 its order
381 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
383 * 2) Any buddy B will have an order O+1 parent P which
384 * satisfies the following equation:
387 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
389 static inline struct page *
390 __page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order)
392 unsigned long buddy_idx = page_idx ^ (1 << order);
394 return page + (buddy_idx - page_idx);
397 static inline unsigned long
398 __find_combined_index(unsigned long page_idx, unsigned int order)
400 return (page_idx & ~(1 << order));
404 * This function checks whether a page is free && is the buddy
405 * we can do coalesce a page and its buddy if
406 * (a) the buddy is not in a hole &&
407 * (b) the buddy is in the buddy system &&
408 * (c) a page and its buddy have the same order &&
409 * (d) a page and its buddy are in the same zone.
411 * For recording whether a page is in the buddy system, we use PG_buddy.
412 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
414 * For recording page's order, we use page_private(page).
416 static inline int page_is_buddy(struct page *page, struct page *buddy,
419 if (!pfn_valid_within(page_to_pfn(buddy)))
422 if (page_zone_id(page) != page_zone_id(buddy))
425 if (PageBuddy(buddy) && page_order(buddy) == order) {
426 VM_BUG_ON(page_count(buddy) != 0);
433 * Freeing function for a buddy system allocator.
435 * The concept of a buddy system is to maintain direct-mapped table
436 * (containing bit values) for memory blocks of various "orders".
437 * The bottom level table contains the map for the smallest allocatable
438 * units of memory (here, pages), and each level above it describes
439 * pairs of units from the levels below, hence, "buddies".
440 * At a high level, all that happens here is marking the table entry
441 * at the bottom level available, and propagating the changes upward
442 * as necessary, plus some accounting needed to play nicely with other
443 * parts of the VM system.
444 * At each level, we keep a list of pages, which are heads of continuous
445 * free pages of length of (1 << order) and marked with PG_buddy. Page's
446 * order is recorded in page_private(page) field.
447 * So when we are allocating or freeing one, we can derive the state of the
448 * other. That is, if we allocate a small block, and both were
449 * free, the remainder of the region must be split into blocks.
450 * If a block is freed, and its buddy is also free, then this
451 * triggers coalescing into a block of larger size.
456 static inline void __free_one_page(struct page *page,
457 struct zone *zone, unsigned int order,
460 unsigned long page_idx;
462 if (unlikely(PageCompound(page)))
463 if (unlikely(destroy_compound_page(page, order)))
466 VM_BUG_ON(migratetype == -1);
468 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
470 VM_BUG_ON(page_idx & ((1 << order) - 1));
471 VM_BUG_ON(bad_range(zone, page));
473 while (order < MAX_ORDER-1) {
474 unsigned long combined_idx;
477 buddy = __page_find_buddy(page, page_idx, order);
478 if (!page_is_buddy(page, buddy, order))
481 /* Our buddy is free, merge with it and move up one order. */
482 list_del(&buddy->lru);
483 zone->free_area[order].nr_free--;
484 rmv_page_order(buddy);
485 combined_idx = __find_combined_index(page_idx, order);
486 page = page + (combined_idx - page_idx);
487 page_idx = combined_idx;
490 set_page_order(page, order);
492 &zone->free_area[order].free_list[migratetype]);
493 zone->free_area[order].nr_free++;
496 static inline int free_pages_check(struct page *page)
498 if (unlikely(page_mapcount(page) |
499 (page->mapping != NULL) |
500 (atomic_read(&page->_count) != 0) |
501 (page->flags & PAGE_FLAGS_CHECK_AT_FREE))) {
505 if (page->flags & PAGE_FLAGS_CHECK_AT_PREP)
506 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
511 * Frees a list of pages.
512 * Assumes all pages on list are in same zone, and of same order.
513 * count is the number of pages to free.
515 * If the zone was previously in an "all pages pinned" state then look to
516 * see if this freeing clears that state.
518 * And clear the zone's pages_scanned counter, to hold off the "all pages are
519 * pinned" detection logic.
521 static void free_pages_bulk(struct zone *zone, int count,
522 struct list_head *list, int order)
524 spin_lock(&zone->lock);
525 zone_clear_flag(zone, ZONE_ALL_UNRECLAIMABLE);
526 zone->pages_scanned = 0;
528 __mod_zone_page_state(zone, NR_FREE_PAGES, count << order);
532 VM_BUG_ON(list_empty(list));
533 page = list_entry(list->prev, struct page, lru);
534 /* have to delete it as __free_one_page list manipulates */
535 list_del(&page->lru);
536 __free_one_page(page, zone, order, page_private(page));
538 spin_unlock(&zone->lock);
541 static void free_one_page(struct zone *zone, struct page *page, int order,
544 spin_lock(&zone->lock);
545 zone_clear_flag(zone, ZONE_ALL_UNRECLAIMABLE);
546 zone->pages_scanned = 0;
548 __mod_zone_page_state(zone, NR_FREE_PAGES, 1 << order);
549 __free_one_page(page, zone, order, migratetype);
550 spin_unlock(&zone->lock);
553 static void __free_pages_ok(struct page *page, unsigned int order)
558 int clearMlocked = PageMlocked(page);
560 for (i = 0 ; i < (1 << order) ; ++i)
561 bad += free_pages_check(page + i);
565 if (!PageHighMem(page)) {
566 debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order);
567 debug_check_no_obj_freed(page_address(page),
570 arch_free_page(page, order);
571 kernel_map_pages(page, 1 << order, 0);
573 local_irq_save(flags);
574 if (unlikely(clearMlocked))
575 free_page_mlock(page);
576 __count_vm_events(PGFREE, 1 << order);
577 free_one_page(page_zone(page), page, order,
578 get_pageblock_migratetype(page));
579 local_irq_restore(flags);
583 * permit the bootmem allocator to evade page validation on high-order frees
585 void __meminit __free_pages_bootmem(struct page *page, unsigned int order)
588 __ClearPageReserved(page);
589 set_page_count(page, 0);
590 set_page_refcounted(page);
596 for (loop = 0; loop < BITS_PER_LONG; loop++) {
597 struct page *p = &page[loop];
599 if (loop + 1 < BITS_PER_LONG)
601 __ClearPageReserved(p);
602 set_page_count(p, 0);
605 set_page_refcounted(page);
606 __free_pages(page, order);
612 * The order of subdivision here is critical for the IO subsystem.
613 * Please do not alter this order without good reasons and regression
614 * testing. Specifically, as large blocks of memory are subdivided,
615 * the order in which smaller blocks are delivered depends on the order
616 * they're subdivided in this function. This is the primary factor
617 * influencing the order in which pages are delivered to the IO
618 * subsystem according to empirical testing, and this is also justified
619 * by considering the behavior of a buddy system containing a single
620 * large block of memory acted on by a series of small allocations.
621 * This behavior is a critical factor in sglist merging's success.
625 static inline void expand(struct zone *zone, struct page *page,
626 int low, int high, struct free_area *area,
629 unsigned long size = 1 << high;
635 VM_BUG_ON(bad_range(zone, &page[size]));
636 list_add(&page[size].lru, &area->free_list[migratetype]);
638 set_page_order(&page[size], high);
643 * This page is about to be returned from the page allocator
645 static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
647 if (unlikely(page_mapcount(page) |
648 (page->mapping != NULL) |
649 (atomic_read(&page->_count) != 0) |
650 (page->flags & PAGE_FLAGS_CHECK_AT_PREP))) {
655 set_page_private(page, 0);
656 set_page_refcounted(page);
658 arch_alloc_page(page, order);
659 kernel_map_pages(page, 1 << order, 1);
661 if (gfp_flags & __GFP_ZERO)
662 prep_zero_page(page, order, gfp_flags);
664 if (order && (gfp_flags & __GFP_COMP))
665 prep_compound_page(page, order);
671 * Go through the free lists for the given migratetype and remove
672 * the smallest available page from the freelists
675 struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
678 unsigned int current_order;
679 struct free_area * area;
682 /* Find a page of the appropriate size in the preferred list */
683 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
684 area = &(zone->free_area[current_order]);
685 if (list_empty(&area->free_list[migratetype]))
688 page = list_entry(area->free_list[migratetype].next,
690 list_del(&page->lru);
691 rmv_page_order(page);
693 expand(zone, page, order, current_order, area, migratetype);
702 * This array describes the order lists are fallen back to when
703 * the free lists for the desirable migrate type are depleted
705 static int fallbacks[MIGRATE_TYPES][MIGRATE_TYPES-1] = {
706 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
707 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
708 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
709 [MIGRATE_RESERVE] = { MIGRATE_RESERVE, MIGRATE_RESERVE, MIGRATE_RESERVE }, /* Never used */
713 * Move the free pages in a range to the free lists of the requested type.
714 * Note that start_page and end_pages are not aligned on a pageblock
715 * boundary. If alignment is required, use move_freepages_block()
717 static int move_freepages(struct zone *zone,
718 struct page *start_page, struct page *end_page,
725 #ifndef CONFIG_HOLES_IN_ZONE
727 * page_zone is not safe to call in this context when
728 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
729 * anyway as we check zone boundaries in move_freepages_block().
730 * Remove at a later date when no bug reports exist related to
731 * grouping pages by mobility
733 BUG_ON(page_zone(start_page) != page_zone(end_page));
736 for (page = start_page; page <= end_page;) {
737 /* Make sure we are not inadvertently changing nodes */
738 VM_BUG_ON(page_to_nid(page) != zone_to_nid(zone));
740 if (!pfn_valid_within(page_to_pfn(page))) {
745 if (!PageBuddy(page)) {
750 order = page_order(page);
751 list_del(&page->lru);
753 &zone->free_area[order].free_list[migratetype]);
755 pages_moved += 1 << order;
761 static int move_freepages_block(struct zone *zone, struct page *page,
764 unsigned long start_pfn, end_pfn;
765 struct page *start_page, *end_page;
767 start_pfn = page_to_pfn(page);
768 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
769 start_page = pfn_to_page(start_pfn);
770 end_page = start_page + pageblock_nr_pages - 1;
771 end_pfn = start_pfn + pageblock_nr_pages - 1;
773 /* Do not cross zone boundaries */
774 if (start_pfn < zone->zone_start_pfn)
776 if (end_pfn >= zone->zone_start_pfn + zone->spanned_pages)
779 return move_freepages(zone, start_page, end_page, migratetype);
782 /* Remove an element from the buddy allocator from the fallback list */
783 static inline struct page *
784 __rmqueue_fallback(struct zone *zone, int order, int start_migratetype)
786 struct free_area * area;
791 /* Find the largest possible block of pages in the other list */
792 for (current_order = MAX_ORDER-1; current_order >= order;
794 for (i = 0; i < MIGRATE_TYPES - 1; i++) {
795 migratetype = fallbacks[start_migratetype][i];
797 /* MIGRATE_RESERVE handled later if necessary */
798 if (migratetype == MIGRATE_RESERVE)
801 area = &(zone->free_area[current_order]);
802 if (list_empty(&area->free_list[migratetype]))
805 page = list_entry(area->free_list[migratetype].next,
810 * If breaking a large block of pages, move all free
811 * pages to the preferred allocation list. If falling
812 * back for a reclaimable kernel allocation, be more
813 * agressive about taking ownership of free pages
815 if (unlikely(current_order >= (pageblock_order >> 1)) ||
816 start_migratetype == MIGRATE_RECLAIMABLE) {
818 pages = move_freepages_block(zone, page,
821 /* Claim the whole block if over half of it is free */
822 if (pages >= (1 << (pageblock_order-1)))
823 set_pageblock_migratetype(page,
826 migratetype = start_migratetype;
829 /* Remove the page from the freelists */
830 list_del(&page->lru);
831 rmv_page_order(page);
833 if (current_order == pageblock_order)
834 set_pageblock_migratetype(page,
837 expand(zone, page, order, current_order, area, migratetype);
846 * Do the hard work of removing an element from the buddy allocator.
847 * Call me with the zone->lock already held.
849 static struct page *__rmqueue(struct zone *zone, unsigned int order,
855 page = __rmqueue_smallest(zone, order, migratetype);
857 if (unlikely(!page) && migratetype != MIGRATE_RESERVE) {
858 page = __rmqueue_fallback(zone, order, migratetype);
861 * Use MIGRATE_RESERVE rather than fail an allocation. goto
862 * is used because __rmqueue_smallest is an inline function
863 * and we want just one call site
866 migratetype = MIGRATE_RESERVE;
875 * Obtain a specified number of elements from the buddy allocator, all under
876 * a single hold of the lock, for efficiency. Add them to the supplied list.
877 * Returns the number of new pages which were placed at *list.
879 static int rmqueue_bulk(struct zone *zone, unsigned int order,
880 unsigned long count, struct list_head *list,
885 spin_lock(&zone->lock);
886 for (i = 0; i < count; ++i) {
887 struct page *page = __rmqueue(zone, order, migratetype);
888 if (unlikely(page == NULL))
892 * Split buddy pages returned by expand() are received here
893 * in physical page order. The page is added to the callers and
894 * list and the list head then moves forward. From the callers
895 * perspective, the linked list is ordered by page number in
896 * some conditions. This is useful for IO devices that can
897 * merge IO requests if the physical pages are ordered
900 list_add(&page->lru, list);
901 set_page_private(page, migratetype);
904 __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
905 spin_unlock(&zone->lock);
911 * Called from the vmstat counter updater to drain pagesets of this
912 * currently executing processor on remote nodes after they have
915 * Note that this function must be called with the thread pinned to
916 * a single processor.
918 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
923 local_irq_save(flags);
924 if (pcp->count >= pcp->batch)
925 to_drain = pcp->batch;
927 to_drain = pcp->count;
928 free_pages_bulk(zone, to_drain, &pcp->list, 0);
929 pcp->count -= to_drain;
930 local_irq_restore(flags);
935 * Drain pages of the indicated processor.
937 * The processor must either be the current processor and the
938 * thread pinned to the current processor or a processor that
941 static void drain_pages(unsigned int cpu)
946 for_each_populated_zone(zone) {
947 struct per_cpu_pageset *pset;
948 struct per_cpu_pages *pcp;
950 pset = zone_pcp(zone, cpu);
953 local_irq_save(flags);
954 free_pages_bulk(zone, pcp->count, &pcp->list, 0);
956 local_irq_restore(flags);
961 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
963 void drain_local_pages(void *arg)
965 drain_pages(smp_processor_id());
969 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
971 void drain_all_pages(void)
973 on_each_cpu(drain_local_pages, NULL, 1);
976 #ifdef CONFIG_HIBERNATION
978 void mark_free_pages(struct zone *zone)
980 unsigned long pfn, max_zone_pfn;
983 struct list_head *curr;
985 if (!zone->spanned_pages)
988 spin_lock_irqsave(&zone->lock, flags);
990 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
991 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
992 if (pfn_valid(pfn)) {
993 struct page *page = pfn_to_page(pfn);
995 if (!swsusp_page_is_forbidden(page))
996 swsusp_unset_page_free(page);
999 for_each_migratetype_order(order, t) {
1000 list_for_each(curr, &zone->free_area[order].free_list[t]) {
1003 pfn = page_to_pfn(list_entry(curr, struct page, lru));
1004 for (i = 0; i < (1UL << order); i++)
1005 swsusp_set_page_free(pfn_to_page(pfn + i));
1008 spin_unlock_irqrestore(&zone->lock, flags);
1010 #endif /* CONFIG_PM */
1013 * Free a 0-order page
1015 static void free_hot_cold_page(struct page *page, int cold)
1017 struct zone *zone = page_zone(page);
1018 struct per_cpu_pages *pcp;
1019 unsigned long flags;
1020 int clearMlocked = PageMlocked(page);
1023 page->mapping = NULL;
1024 if (free_pages_check(page))
1027 if (!PageHighMem(page)) {
1028 debug_check_no_locks_freed(page_address(page), PAGE_SIZE);
1029 debug_check_no_obj_freed(page_address(page), PAGE_SIZE);
1031 arch_free_page(page, 0);
1032 kernel_map_pages(page, 1, 0);
1034 pcp = &zone_pcp(zone, get_cpu())->pcp;
1035 set_page_private(page, get_pageblock_migratetype(page));
1036 local_irq_save(flags);
1037 if (unlikely(clearMlocked))
1038 free_page_mlock(page);
1039 __count_vm_event(PGFREE);
1042 list_add_tail(&page->lru, &pcp->list);
1044 list_add(&page->lru, &pcp->list);
1046 if (pcp->count >= pcp->high) {
1047 free_pages_bulk(zone, pcp->batch, &pcp->list, 0);
1048 pcp->count -= pcp->batch;
1050 local_irq_restore(flags);
1054 void free_hot_page(struct page *page)
1056 free_hot_cold_page(page, 0);
1059 void free_cold_page(struct page *page)
1061 free_hot_cold_page(page, 1);
1065 * split_page takes a non-compound higher-order page, and splits it into
1066 * n (1<<order) sub-pages: page[0..n]
1067 * Each sub-page must be freed individually.
1069 * Note: this is probably too low level an operation for use in drivers.
1070 * Please consult with lkml before using this in your driver.
1072 void split_page(struct page *page, unsigned int order)
1076 VM_BUG_ON(PageCompound(page));
1077 VM_BUG_ON(!page_count(page));
1078 for (i = 1; i < (1 << order); i++)
1079 set_page_refcounted(page + i);
1083 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1084 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1088 struct page *buffered_rmqueue(struct zone *preferred_zone,
1089 struct zone *zone, int order, gfp_t gfp_flags,
1092 unsigned long flags;
1094 int cold = !!(gfp_flags & __GFP_COLD);
1099 if (likely(order == 0)) {
1100 struct per_cpu_pages *pcp;
1102 pcp = &zone_pcp(zone, cpu)->pcp;
1103 local_irq_save(flags);
1105 pcp->count = rmqueue_bulk(zone, 0,
1106 pcp->batch, &pcp->list, migratetype);
1107 if (unlikely(!pcp->count))
1111 /* Find a page of the appropriate migrate type */
1113 list_for_each_entry_reverse(page, &pcp->list, lru)
1114 if (page_private(page) == migratetype)
1117 list_for_each_entry(page, &pcp->list, lru)
1118 if (page_private(page) == migratetype)
1122 /* Allocate more to the pcp list if necessary */
1123 if (unlikely(&page->lru == &pcp->list)) {
1124 pcp->count += rmqueue_bulk(zone, 0,
1125 pcp->batch, &pcp->list, migratetype);
1126 page = list_entry(pcp->list.next, struct page, lru);
1129 list_del(&page->lru);
1132 spin_lock_irqsave(&zone->lock, flags);
1133 page = __rmqueue(zone, order, migratetype);
1134 __mod_zone_page_state(zone, NR_FREE_PAGES, -(1 << order));
1135 spin_unlock(&zone->lock);
1140 __count_zone_vm_events(PGALLOC, zone, 1 << order);
1141 zone_statistics(preferred_zone, zone);
1142 local_irq_restore(flags);
1145 VM_BUG_ON(bad_range(zone, page));
1146 if (prep_new_page(page, order, gfp_flags))
1151 local_irq_restore(flags);
1156 /* The ALLOC_WMARK bits are used as an index to zone->watermark */
1157 #define ALLOC_WMARK_MIN WMARK_MIN
1158 #define ALLOC_WMARK_LOW WMARK_LOW
1159 #define ALLOC_WMARK_HIGH WMARK_HIGH
1160 #define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */
1162 /* Mask to get the watermark bits */
1163 #define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1)
1165 #define ALLOC_HARDER 0x10 /* try to alloc harder */
1166 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1167 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
1169 #ifdef CONFIG_FAIL_PAGE_ALLOC
1171 static struct fail_page_alloc_attr {
1172 struct fault_attr attr;
1174 u32 ignore_gfp_highmem;
1175 u32 ignore_gfp_wait;
1178 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1180 struct dentry *ignore_gfp_highmem_file;
1181 struct dentry *ignore_gfp_wait_file;
1182 struct dentry *min_order_file;
1184 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1186 } fail_page_alloc = {
1187 .attr = FAULT_ATTR_INITIALIZER,
1188 .ignore_gfp_wait = 1,
1189 .ignore_gfp_highmem = 1,
1193 static int __init setup_fail_page_alloc(char *str)
1195 return setup_fault_attr(&fail_page_alloc.attr, str);
1197 __setup("fail_page_alloc=", setup_fail_page_alloc);
1199 static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1201 if (order < fail_page_alloc.min_order)
1203 if (gfp_mask & __GFP_NOFAIL)
1205 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
1207 if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
1210 return should_fail(&fail_page_alloc.attr, 1 << order);
1213 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1215 static int __init fail_page_alloc_debugfs(void)
1217 mode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
1221 err = init_fault_attr_dentries(&fail_page_alloc.attr,
1225 dir = fail_page_alloc.attr.dentries.dir;
1227 fail_page_alloc.ignore_gfp_wait_file =
1228 debugfs_create_bool("ignore-gfp-wait", mode, dir,
1229 &fail_page_alloc.ignore_gfp_wait);
1231 fail_page_alloc.ignore_gfp_highmem_file =
1232 debugfs_create_bool("ignore-gfp-highmem", mode, dir,
1233 &fail_page_alloc.ignore_gfp_highmem);
1234 fail_page_alloc.min_order_file =
1235 debugfs_create_u32("min-order", mode, dir,
1236 &fail_page_alloc.min_order);
1238 if (!fail_page_alloc.ignore_gfp_wait_file ||
1239 !fail_page_alloc.ignore_gfp_highmem_file ||
1240 !fail_page_alloc.min_order_file) {
1242 debugfs_remove(fail_page_alloc.ignore_gfp_wait_file);
1243 debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file);
1244 debugfs_remove(fail_page_alloc.min_order_file);
1245 cleanup_fault_attr_dentries(&fail_page_alloc.attr);
1251 late_initcall(fail_page_alloc_debugfs);
1253 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1255 #else /* CONFIG_FAIL_PAGE_ALLOC */
1257 static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1262 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1265 * Return 1 if free pages are above 'mark'. This takes into account the order
1266 * of the allocation.
1268 int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1269 int classzone_idx, int alloc_flags)
1271 /* free_pages my go negative - that's OK */
1273 long free_pages = zone_page_state(z, NR_FREE_PAGES) - (1 << order) + 1;
1276 if (alloc_flags & ALLOC_HIGH)
1278 if (alloc_flags & ALLOC_HARDER)
1281 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
1283 for (o = 0; o < order; o++) {
1284 /* At the next order, this order's pages become unavailable */
1285 free_pages -= z->free_area[o].nr_free << o;
1287 /* Require fewer higher order pages to be free */
1290 if (free_pages <= min)
1298 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1299 * skip over zones that are not allowed by the cpuset, or that have
1300 * been recently (in last second) found to be nearly full. See further
1301 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1302 * that have to skip over a lot of full or unallowed zones.
1304 * If the zonelist cache is present in the passed in zonelist, then
1305 * returns a pointer to the allowed node mask (either the current
1306 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
1308 * If the zonelist cache is not available for this zonelist, does
1309 * nothing and returns NULL.
1311 * If the fullzones BITMAP in the zonelist cache is stale (more than
1312 * a second since last zap'd) then we zap it out (clear its bits.)
1314 * We hold off even calling zlc_setup, until after we've checked the
1315 * first zone in the zonelist, on the theory that most allocations will
1316 * be satisfied from that first zone, so best to examine that zone as
1317 * quickly as we can.
1319 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1321 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1322 nodemask_t *allowednodes; /* zonelist_cache approximation */
1324 zlc = zonelist->zlcache_ptr;
1328 if (time_after(jiffies, zlc->last_full_zap + HZ)) {
1329 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1330 zlc->last_full_zap = jiffies;
1333 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1334 &cpuset_current_mems_allowed :
1335 &node_states[N_HIGH_MEMORY];
1336 return allowednodes;
1340 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1341 * if it is worth looking at further for free memory:
1342 * 1) Check that the zone isn't thought to be full (doesn't have its
1343 * bit set in the zonelist_cache fullzones BITMAP).
1344 * 2) Check that the zones node (obtained from the zonelist_cache
1345 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1346 * Return true (non-zero) if zone is worth looking at further, or
1347 * else return false (zero) if it is not.
1349 * This check -ignores- the distinction between various watermarks,
1350 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1351 * found to be full for any variation of these watermarks, it will
1352 * be considered full for up to one second by all requests, unless
1353 * we are so low on memory on all allowed nodes that we are forced
1354 * into the second scan of the zonelist.
1356 * In the second scan we ignore this zonelist cache and exactly
1357 * apply the watermarks to all zones, even it is slower to do so.
1358 * We are low on memory in the second scan, and should leave no stone
1359 * unturned looking for a free page.
1361 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
1362 nodemask_t *allowednodes)
1364 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1365 int i; /* index of *z in zonelist zones */
1366 int n; /* node that zone *z is on */
1368 zlc = zonelist->zlcache_ptr;
1372 i = z - zonelist->_zonerefs;
1375 /* This zone is worth trying if it is allowed but not full */
1376 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1380 * Given 'z' scanning a zonelist, set the corresponding bit in
1381 * zlc->fullzones, so that subsequent attempts to allocate a page
1382 * from that zone don't waste time re-examining it.
1384 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
1386 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1387 int i; /* index of *z in zonelist zones */
1389 zlc = zonelist->zlcache_ptr;
1393 i = z - zonelist->_zonerefs;
1395 set_bit(i, zlc->fullzones);
1398 #else /* CONFIG_NUMA */
1400 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1405 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
1406 nodemask_t *allowednodes)
1411 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
1414 #endif /* CONFIG_NUMA */
1417 * get_page_from_freelist goes through the zonelist trying to allocate
1420 static struct page *
1421 get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order,
1422 struct zonelist *zonelist, int high_zoneidx, int alloc_flags,
1423 struct zone *preferred_zone, int migratetype)
1426 struct page *page = NULL;
1429 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1430 int zlc_active = 0; /* set if using zonelist_cache */
1431 int did_zlc_setup = 0; /* just call zlc_setup() one time */
1433 if (WARN_ON_ONCE(order >= MAX_ORDER))
1436 classzone_idx = zone_idx(preferred_zone);
1439 * Scan zonelist, looking for a zone with enough free.
1440 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1442 for_each_zone_zonelist_nodemask(zone, z, zonelist,
1443 high_zoneidx, nodemask) {
1444 if (NUMA_BUILD && zlc_active &&
1445 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1447 if ((alloc_flags & ALLOC_CPUSET) &&
1448 !cpuset_zone_allowed_softwall(zone, gfp_mask))
1451 BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
1452 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
1454 mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
1455 if (!zone_watermark_ok(zone, order, mark,
1456 classzone_idx, alloc_flags)) {
1457 if (!zone_reclaim_mode ||
1458 !zone_reclaim(zone, gfp_mask, order))
1459 goto this_zone_full;
1463 page = buffered_rmqueue(preferred_zone, zone, order,
1464 gfp_mask, migratetype);
1469 zlc_mark_zone_full(zonelist, z);
1471 if (NUMA_BUILD && !did_zlc_setup && nr_online_nodes > 1) {
1473 * we do zlc_setup after the first zone is tried but only
1474 * if there are multiple nodes make it worthwhile
1476 allowednodes = zlc_setup(zonelist, alloc_flags);
1482 if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) {
1483 /* Disable zlc cache for second zonelist scan */
1491 should_alloc_retry(gfp_t gfp_mask, unsigned int order,
1492 unsigned long pages_reclaimed)
1494 /* Do not loop if specifically requested */
1495 if (gfp_mask & __GFP_NORETRY)
1499 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
1500 * means __GFP_NOFAIL, but that may not be true in other
1503 if (order <= PAGE_ALLOC_COSTLY_ORDER)
1507 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
1508 * specified, then we retry until we no longer reclaim any pages
1509 * (above), or we've reclaimed an order of pages at least as
1510 * large as the allocation's order. In both cases, if the
1511 * allocation still fails, we stop retrying.
1513 if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order))
1517 * Don't let big-order allocations loop unless the caller
1518 * explicitly requests that.
1520 if (gfp_mask & __GFP_NOFAIL)
1526 static inline struct page *
1527 __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
1528 struct zonelist *zonelist, enum zone_type high_zoneidx,
1529 nodemask_t *nodemask, struct zone *preferred_zone,
1534 /* Acquire the OOM killer lock for the zones in zonelist */
1535 if (!try_set_zone_oom(zonelist, gfp_mask)) {
1536 schedule_timeout_uninterruptible(1);
1541 * Go through the zonelist yet one more time, keep very high watermark
1542 * here, this is only to catch a parallel oom killing, we must fail if
1543 * we're still under heavy pressure.
1545 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask,
1546 order, zonelist, high_zoneidx,
1547 ALLOC_WMARK_HIGH|ALLOC_CPUSET,
1548 preferred_zone, migratetype);
1552 /* The OOM killer will not help higher order allocs */
1553 if (order > PAGE_ALLOC_COSTLY_ORDER)
1556 /* Exhausted what can be done so it's blamo time */
1557 out_of_memory(zonelist, gfp_mask, order);
1560 clear_zonelist_oom(zonelist, gfp_mask);
1564 /* The really slow allocator path where we enter direct reclaim */
1565 static inline struct page *
1566 __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
1567 struct zonelist *zonelist, enum zone_type high_zoneidx,
1568 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
1569 int migratetype, unsigned long *did_some_progress)
1571 struct page *page = NULL;
1572 struct reclaim_state reclaim_state;
1573 struct task_struct *p = current;
1577 /* We now go into synchronous reclaim */
1578 cpuset_memory_pressure_bump();
1581 * The task's cpuset might have expanded its set of allowable nodes
1583 p->flags |= PF_MEMALLOC;
1584 lockdep_set_current_reclaim_state(gfp_mask);
1585 reclaim_state.reclaimed_slab = 0;
1586 p->reclaim_state = &reclaim_state;
1588 *did_some_progress = try_to_free_pages(zonelist, order, gfp_mask, nodemask);
1590 p->reclaim_state = NULL;
1591 lockdep_clear_current_reclaim_state();
1592 p->flags &= ~PF_MEMALLOC;
1599 if (likely(*did_some_progress))
1600 page = get_page_from_freelist(gfp_mask, nodemask, order,
1601 zonelist, high_zoneidx,
1602 alloc_flags, preferred_zone,
1608 * This is called in the allocator slow-path if the allocation request is of
1609 * sufficient urgency to ignore watermarks and take other desperate measures
1611 static inline struct page *
1612 __alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order,
1613 struct zonelist *zonelist, enum zone_type high_zoneidx,
1614 nodemask_t *nodemask, struct zone *preferred_zone,
1620 page = get_page_from_freelist(gfp_mask, nodemask, order,
1621 zonelist, high_zoneidx, ALLOC_NO_WATERMARKS,
1622 preferred_zone, migratetype);
1624 if (!page && gfp_mask & __GFP_NOFAIL)
1625 congestion_wait(WRITE, HZ/50);
1626 } while (!page && (gfp_mask & __GFP_NOFAIL));
1632 void wake_all_kswapd(unsigned int order, struct zonelist *zonelist,
1633 enum zone_type high_zoneidx)
1638 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx)
1639 wakeup_kswapd(zone, order);
1643 gfp_to_alloc_flags(gfp_t gfp_mask)
1645 struct task_struct *p = current;
1646 int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
1647 const gfp_t wait = gfp_mask & __GFP_WAIT;
1649 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
1650 BUILD_BUG_ON(__GFP_HIGH != ALLOC_HIGH);
1653 * The caller may dip into page reserves a bit more if the caller
1654 * cannot run direct reclaim, or if the caller has realtime scheduling
1655 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1656 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1658 alloc_flags |= (gfp_mask & __GFP_HIGH);
1661 alloc_flags |= ALLOC_HARDER;
1663 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1664 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1666 alloc_flags &= ~ALLOC_CPUSET;
1667 } else if (unlikely(rt_task(p)))
1668 alloc_flags |= ALLOC_HARDER;
1670 if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) {
1671 if (!in_interrupt() &&
1672 ((p->flags & PF_MEMALLOC) ||
1673 unlikely(test_thread_flag(TIF_MEMDIE))))
1674 alloc_flags |= ALLOC_NO_WATERMARKS;
1680 static inline struct page *
1681 __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
1682 struct zonelist *zonelist, enum zone_type high_zoneidx,
1683 nodemask_t *nodemask, struct zone *preferred_zone,
1686 const gfp_t wait = gfp_mask & __GFP_WAIT;
1687 struct page *page = NULL;
1689 unsigned long pages_reclaimed = 0;
1690 unsigned long did_some_progress;
1691 struct task_struct *p = current;
1694 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1695 * __GFP_NOWARN set) should not cause reclaim since the subsystem
1696 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1697 * using a larger set of nodes after it has established that the
1698 * allowed per node queues are empty and that nodes are
1701 if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
1704 wake_all_kswapd(order, zonelist, high_zoneidx);
1707 * OK, we're below the kswapd watermark and have kicked background
1708 * reclaim. Now things get more complex, so set up alloc_flags according
1709 * to how we want to proceed.
1711 alloc_flags = gfp_to_alloc_flags(gfp_mask);
1714 /* This is the last chance, in general, before the goto nopage. */
1715 page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist,
1716 high_zoneidx, alloc_flags & ~ALLOC_NO_WATERMARKS,
1717 preferred_zone, migratetype);
1722 /* Allocate without watermarks if the context allows */
1723 if (alloc_flags & ALLOC_NO_WATERMARKS) {
1724 page = __alloc_pages_high_priority(gfp_mask, order,
1725 zonelist, high_zoneidx, nodemask,
1726 preferred_zone, migratetype);
1731 /* Atomic allocations - we can't balance anything */
1735 /* Avoid recursion of direct reclaim */
1736 if (p->flags & PF_MEMALLOC)
1739 /* Try direct reclaim and then allocating */
1740 page = __alloc_pages_direct_reclaim(gfp_mask, order,
1741 zonelist, high_zoneidx,
1743 alloc_flags, preferred_zone,
1744 migratetype, &did_some_progress);
1749 * If we failed to make any progress reclaiming, then we are
1750 * running out of options and have to consider going OOM
1752 if (!did_some_progress) {
1753 if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
1754 page = __alloc_pages_may_oom(gfp_mask, order,
1755 zonelist, high_zoneidx,
1756 nodemask, preferred_zone,
1762 * The OOM killer does not trigger for high-order allocations
1763 * but if no progress is being made, there are no other
1764 * options and retrying is unlikely to help
1766 if (order > PAGE_ALLOC_COSTLY_ORDER)
1773 /* Check if we should retry the allocation */
1774 pages_reclaimed += did_some_progress;
1775 if (should_alloc_retry(gfp_mask, order, pages_reclaimed)) {
1776 /* Wait for some write requests to complete then retry */
1777 congestion_wait(WRITE, HZ/50);
1782 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
1783 printk(KERN_WARNING "%s: page allocation failure."
1784 " order:%d, mode:0x%x\n",
1785 p->comm, order, gfp_mask);
1795 * This is the 'heart' of the zoned buddy allocator.
1798 __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
1799 struct zonelist *zonelist, nodemask_t *nodemask)
1801 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
1802 struct zone *preferred_zone;
1804 int migratetype = allocflags_to_migratetype(gfp_mask);
1806 lockdep_trace_alloc(gfp_mask);
1808 might_sleep_if(gfp_mask & __GFP_WAIT);
1810 if (should_fail_alloc_page(gfp_mask, order))
1814 * Check the zones suitable for the gfp_mask contain at least one
1815 * valid zone. It's possible to have an empty zonelist as a result
1816 * of GFP_THISNODE and a memoryless node
1818 if (unlikely(!zonelist->_zonerefs->zone))
1821 /* The preferred zone is used for statistics later */
1822 first_zones_zonelist(zonelist, high_zoneidx, nodemask, &preferred_zone);
1823 if (!preferred_zone)
1826 /* First allocation attempt */
1827 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order,
1828 zonelist, high_zoneidx, ALLOC_WMARK_LOW|ALLOC_CPUSET,
1829 preferred_zone, migratetype);
1830 if (unlikely(!page))
1831 page = __alloc_pages_slowpath(gfp_mask, order,
1832 zonelist, high_zoneidx, nodemask,
1833 preferred_zone, migratetype);
1837 EXPORT_SYMBOL(__alloc_pages_nodemask);
1840 * Common helper functions.
1842 unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
1845 page = alloc_pages(gfp_mask, order);
1848 return (unsigned long) page_address(page);
1851 EXPORT_SYMBOL(__get_free_pages);
1853 unsigned long get_zeroed_page(gfp_t gfp_mask)
1858 * get_zeroed_page() returns a 32-bit address, which cannot represent
1861 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
1863 page = alloc_pages(gfp_mask | __GFP_ZERO, 0);
1865 return (unsigned long) page_address(page);
1869 EXPORT_SYMBOL(get_zeroed_page);
1871 void __pagevec_free(struct pagevec *pvec)
1873 int i = pagevec_count(pvec);
1876 free_hot_cold_page(pvec->pages[i], pvec->cold);
1879 void __free_pages(struct page *page, unsigned int order)
1881 if (put_page_testzero(page)) {
1883 free_hot_page(page);
1885 __free_pages_ok(page, order);
1889 EXPORT_SYMBOL(__free_pages);
1891 void free_pages(unsigned long addr, unsigned int order)
1894 VM_BUG_ON(!virt_addr_valid((void *)addr));
1895 __free_pages(virt_to_page((void *)addr), order);
1899 EXPORT_SYMBOL(free_pages);
1902 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
1903 * @size: the number of bytes to allocate
1904 * @gfp_mask: GFP flags for the allocation
1906 * This function is similar to alloc_pages(), except that it allocates the
1907 * minimum number of pages to satisfy the request. alloc_pages() can only
1908 * allocate memory in power-of-two pages.
1910 * This function is also limited by MAX_ORDER.
1912 * Memory allocated by this function must be released by free_pages_exact().
1914 void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
1916 unsigned int order = get_order(size);
1919 addr = __get_free_pages(gfp_mask, order);
1921 unsigned long alloc_end = addr + (PAGE_SIZE << order);
1922 unsigned long used = addr + PAGE_ALIGN(size);
1924 split_page(virt_to_page(addr), order);
1925 while (used < alloc_end) {
1931 return (void *)addr;
1933 EXPORT_SYMBOL(alloc_pages_exact);
1936 * free_pages_exact - release memory allocated via alloc_pages_exact()
1937 * @virt: the value returned by alloc_pages_exact.
1938 * @size: size of allocation, same value as passed to alloc_pages_exact().
1940 * Release the memory allocated by a previous call to alloc_pages_exact.
1942 void free_pages_exact(void *virt, size_t size)
1944 unsigned long addr = (unsigned long)virt;
1945 unsigned long end = addr + PAGE_ALIGN(size);
1947 while (addr < end) {
1952 EXPORT_SYMBOL(free_pages_exact);
1954 static unsigned int nr_free_zone_pages(int offset)
1959 /* Just pick one node, since fallback list is circular */
1960 unsigned int sum = 0;
1962 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
1964 for_each_zone_zonelist(zone, z, zonelist, offset) {
1965 unsigned long size = zone->present_pages;
1966 unsigned long high = high_wmark_pages(zone);
1975 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1977 unsigned int nr_free_buffer_pages(void)
1979 return nr_free_zone_pages(gfp_zone(GFP_USER));
1981 EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
1984 * Amount of free RAM allocatable within all zones
1986 unsigned int nr_free_pagecache_pages(void)
1988 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
1991 static inline void show_node(struct zone *zone)
1994 printk("Node %d ", zone_to_nid(zone));
1997 void si_meminfo(struct sysinfo *val)
1999 val->totalram = totalram_pages;
2001 val->freeram = global_page_state(NR_FREE_PAGES);
2002 val->bufferram = nr_blockdev_pages();
2003 val->totalhigh = totalhigh_pages;
2004 val->freehigh = nr_free_highpages();
2005 val->mem_unit = PAGE_SIZE;
2008 EXPORT_SYMBOL(si_meminfo);
2011 void si_meminfo_node(struct sysinfo *val, int nid)
2013 pg_data_t *pgdat = NODE_DATA(nid);
2015 val->totalram = pgdat->node_present_pages;
2016 val->freeram = node_page_state(nid, NR_FREE_PAGES);
2017 #ifdef CONFIG_HIGHMEM
2018 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
2019 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
2025 val->mem_unit = PAGE_SIZE;
2029 #define K(x) ((x) << (PAGE_SHIFT-10))
2032 * Show free area list (used inside shift_scroll-lock stuff)
2033 * We also calculate the percentage fragmentation. We do this by counting the
2034 * memory on each free list with the exception of the first item on the list.
2036 void show_free_areas(void)
2041 for_each_populated_zone(zone) {
2043 printk("%s per-cpu:\n", zone->name);
2045 for_each_online_cpu(cpu) {
2046 struct per_cpu_pageset *pageset;
2048 pageset = zone_pcp(zone, cpu);
2050 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
2051 cpu, pageset->pcp.high,
2052 pageset->pcp.batch, pageset->pcp.count);
2056 printk("Active_anon:%lu active_file:%lu inactive_anon:%lu\n"
2057 " inactive_file:%lu"
2058 //TODO: check/adjust line lengths
2059 #ifdef CONFIG_UNEVICTABLE_LRU
2062 " dirty:%lu writeback:%lu unstable:%lu\n"
2063 " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
2064 global_page_state(NR_ACTIVE_ANON),
2065 global_page_state(NR_ACTIVE_FILE),
2066 global_page_state(NR_INACTIVE_ANON),
2067 global_page_state(NR_INACTIVE_FILE),
2068 #ifdef CONFIG_UNEVICTABLE_LRU
2069 global_page_state(NR_UNEVICTABLE),
2071 global_page_state(NR_FILE_DIRTY),
2072 global_page_state(NR_WRITEBACK),
2073 global_page_state(NR_UNSTABLE_NFS),
2074 global_page_state(NR_FREE_PAGES),
2075 global_page_state(NR_SLAB_RECLAIMABLE) +
2076 global_page_state(NR_SLAB_UNRECLAIMABLE),
2077 global_page_state(NR_FILE_MAPPED),
2078 global_page_state(NR_PAGETABLE),
2079 global_page_state(NR_BOUNCE));
2081 for_each_populated_zone(zone) {
2090 " active_anon:%lukB"
2091 " inactive_anon:%lukB"
2092 " active_file:%lukB"
2093 " inactive_file:%lukB"
2094 #ifdef CONFIG_UNEVICTABLE_LRU
2095 " unevictable:%lukB"
2098 " pages_scanned:%lu"
2099 " all_unreclaimable? %s"
2102 K(zone_page_state(zone, NR_FREE_PAGES)),
2103 K(min_wmark_pages(zone)),
2104 K(low_wmark_pages(zone)),
2105 K(high_wmark_pages(zone)),
2106 K(zone_page_state(zone, NR_ACTIVE_ANON)),
2107 K(zone_page_state(zone, NR_INACTIVE_ANON)),
2108 K(zone_page_state(zone, NR_ACTIVE_FILE)),
2109 K(zone_page_state(zone, NR_INACTIVE_FILE)),
2110 #ifdef CONFIG_UNEVICTABLE_LRU
2111 K(zone_page_state(zone, NR_UNEVICTABLE)),
2113 K(zone->present_pages),
2114 zone->pages_scanned,
2115 (zone_is_all_unreclaimable(zone) ? "yes" : "no")
2117 printk("lowmem_reserve[]:");
2118 for (i = 0; i < MAX_NR_ZONES; i++)
2119 printk(" %lu", zone->lowmem_reserve[i]);
2123 for_each_populated_zone(zone) {
2124 unsigned long nr[MAX_ORDER], flags, order, total = 0;
2127 printk("%s: ", zone->name);
2129 spin_lock_irqsave(&zone->lock, flags);
2130 for (order = 0; order < MAX_ORDER; order++) {
2131 nr[order] = zone->free_area[order].nr_free;
2132 total += nr[order] << order;
2134 spin_unlock_irqrestore(&zone->lock, flags);
2135 for (order = 0; order < MAX_ORDER; order++)
2136 printk("%lu*%lukB ", nr[order], K(1UL) << order);
2137 printk("= %lukB\n", K(total));
2140 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
2142 show_swap_cache_info();
2145 static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
2147 zoneref->zone = zone;
2148 zoneref->zone_idx = zone_idx(zone);
2152 * Builds allocation fallback zone lists.
2154 * Add all populated zones of a node to the zonelist.
2156 static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
2157 int nr_zones, enum zone_type zone_type)
2161 BUG_ON(zone_type >= MAX_NR_ZONES);
2166 zone = pgdat->node_zones + zone_type;
2167 if (populated_zone(zone)) {
2168 zoneref_set_zone(zone,
2169 &zonelist->_zonerefs[nr_zones++]);
2170 check_highest_zone(zone_type);
2173 } while (zone_type);
2180 * 0 = automatic detection of better ordering.
2181 * 1 = order by ([node] distance, -zonetype)
2182 * 2 = order by (-zonetype, [node] distance)
2184 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
2185 * the same zonelist. So only NUMA can configure this param.
2187 #define ZONELIST_ORDER_DEFAULT 0
2188 #define ZONELIST_ORDER_NODE 1
2189 #define ZONELIST_ORDER_ZONE 2
2191 /* zonelist order in the kernel.
2192 * set_zonelist_order() will set this to NODE or ZONE.
2194 static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
2195 static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
2199 /* The value user specified ....changed by config */
2200 static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2201 /* string for sysctl */
2202 #define NUMA_ZONELIST_ORDER_LEN 16
2203 char numa_zonelist_order[16] = "default";
2206 * interface for configure zonelist ordering.
2207 * command line option "numa_zonelist_order"
2208 * = "[dD]efault - default, automatic configuration.
2209 * = "[nN]ode - order by node locality, then by zone within node
2210 * = "[zZ]one - order by zone, then by locality within zone
2213 static int __parse_numa_zonelist_order(char *s)
2215 if (*s == 'd' || *s == 'D') {
2216 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2217 } else if (*s == 'n' || *s == 'N') {
2218 user_zonelist_order = ZONELIST_ORDER_NODE;
2219 } else if (*s == 'z' || *s == 'Z') {
2220 user_zonelist_order = ZONELIST_ORDER_ZONE;
2223 "Ignoring invalid numa_zonelist_order value: "
2230 static __init int setup_numa_zonelist_order(char *s)
2233 return __parse_numa_zonelist_order(s);
2236 early_param("numa_zonelist_order", setup_numa_zonelist_order);
2239 * sysctl handler for numa_zonelist_order
2241 int numa_zonelist_order_handler(ctl_table *table, int write,
2242 struct file *file, void __user *buffer, size_t *length,
2245 char saved_string[NUMA_ZONELIST_ORDER_LEN];
2249 strncpy(saved_string, (char*)table->data,
2250 NUMA_ZONELIST_ORDER_LEN);
2251 ret = proc_dostring(table, write, file, buffer, length, ppos);
2255 int oldval = user_zonelist_order;
2256 if (__parse_numa_zonelist_order((char*)table->data)) {
2258 * bogus value. restore saved string
2260 strncpy((char*)table->data, saved_string,
2261 NUMA_ZONELIST_ORDER_LEN);
2262 user_zonelist_order = oldval;
2263 } else if (oldval != user_zonelist_order)
2264 build_all_zonelists();
2270 #define MAX_NODE_LOAD (nr_online_nodes)
2271 static int node_load[MAX_NUMNODES];
2274 * find_next_best_node - find the next node that should appear in a given node's fallback list
2275 * @node: node whose fallback list we're appending
2276 * @used_node_mask: nodemask_t of already used nodes
2278 * We use a number of factors to determine which is the next node that should
2279 * appear on a given node's fallback list. The node should not have appeared
2280 * already in @node's fallback list, and it should be the next closest node
2281 * according to the distance array (which contains arbitrary distance values
2282 * from each node to each node in the system), and should also prefer nodes
2283 * with no CPUs, since presumably they'll have very little allocation pressure
2284 * on them otherwise.
2285 * It returns -1 if no node is found.
2287 static int find_next_best_node(int node, nodemask_t *used_node_mask)
2290 int min_val = INT_MAX;
2292 const struct cpumask *tmp = cpumask_of_node(0);
2294 /* Use the local node if we haven't already */
2295 if (!node_isset(node, *used_node_mask)) {
2296 node_set(node, *used_node_mask);
2300 for_each_node_state(n, N_HIGH_MEMORY) {
2302 /* Don't want a node to appear more than once */
2303 if (node_isset(n, *used_node_mask))
2306 /* Use the distance array to find the distance */
2307 val = node_distance(node, n);
2309 /* Penalize nodes under us ("prefer the next node") */
2312 /* Give preference to headless and unused nodes */
2313 tmp = cpumask_of_node(n);
2314 if (!cpumask_empty(tmp))
2315 val += PENALTY_FOR_NODE_WITH_CPUS;
2317 /* Slight preference for less loaded node */
2318 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
2319 val += node_load[n];
2321 if (val < min_val) {
2328 node_set(best_node, *used_node_mask);
2335 * Build zonelists ordered by node and zones within node.
2336 * This results in maximum locality--normal zone overflows into local
2337 * DMA zone, if any--but risks exhausting DMA zone.
2339 static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
2342 struct zonelist *zonelist;
2344 zonelist = &pgdat->node_zonelists[0];
2345 for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
2347 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2349 zonelist->_zonerefs[j].zone = NULL;
2350 zonelist->_zonerefs[j].zone_idx = 0;
2354 * Build gfp_thisnode zonelists
2356 static void build_thisnode_zonelists(pg_data_t *pgdat)
2359 struct zonelist *zonelist;
2361 zonelist = &pgdat->node_zonelists[1];
2362 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
2363 zonelist->_zonerefs[j].zone = NULL;
2364 zonelist->_zonerefs[j].zone_idx = 0;
2368 * Build zonelists ordered by zone and nodes within zones.
2369 * This results in conserving DMA zone[s] until all Normal memory is
2370 * exhausted, but results in overflowing to remote node while memory
2371 * may still exist in local DMA zone.
2373 static int node_order[MAX_NUMNODES];
2375 static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
2378 int zone_type; /* needs to be signed */
2380 struct zonelist *zonelist;
2382 zonelist = &pgdat->node_zonelists[0];
2384 for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
2385 for (j = 0; j < nr_nodes; j++) {
2386 node = node_order[j];
2387 z = &NODE_DATA(node)->node_zones[zone_type];
2388 if (populated_zone(z)) {
2390 &zonelist->_zonerefs[pos++]);
2391 check_highest_zone(zone_type);
2395 zonelist->_zonerefs[pos].zone = NULL;
2396 zonelist->_zonerefs[pos].zone_idx = 0;
2399 static int default_zonelist_order(void)
2402 unsigned long low_kmem_size,total_size;
2406 * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem.
2407 * If they are really small and used heavily, the system can fall
2408 * into OOM very easily.
2409 * This function detect ZONE_DMA/DMA32 size and confgigures zone order.
2411 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2414 for_each_online_node(nid) {
2415 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2416 z = &NODE_DATA(nid)->node_zones[zone_type];
2417 if (populated_zone(z)) {
2418 if (zone_type < ZONE_NORMAL)
2419 low_kmem_size += z->present_pages;
2420 total_size += z->present_pages;
2424 if (!low_kmem_size || /* there are no DMA area. */
2425 low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
2426 return ZONELIST_ORDER_NODE;
2428 * look into each node's config.
2429 * If there is a node whose DMA/DMA32 memory is very big area on
2430 * local memory, NODE_ORDER may be suitable.
2432 average_size = total_size /
2433 (nodes_weight(node_states[N_HIGH_MEMORY]) + 1);
2434 for_each_online_node(nid) {
2437 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2438 z = &NODE_DATA(nid)->node_zones[zone_type];
2439 if (populated_zone(z)) {
2440 if (zone_type < ZONE_NORMAL)
2441 low_kmem_size += z->present_pages;
2442 total_size += z->present_pages;
2445 if (low_kmem_size &&
2446 total_size > average_size && /* ignore small node */
2447 low_kmem_size > total_size * 70/100)
2448 return ZONELIST_ORDER_NODE;
2450 return ZONELIST_ORDER_ZONE;
2453 static void set_zonelist_order(void)
2455 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
2456 current_zonelist_order = default_zonelist_order();
2458 current_zonelist_order = user_zonelist_order;
2461 static void build_zonelists(pg_data_t *pgdat)
2465 nodemask_t used_mask;
2466 int local_node, prev_node;
2467 struct zonelist *zonelist;
2468 int order = current_zonelist_order;
2470 /* initialize zonelists */
2471 for (i = 0; i < MAX_ZONELISTS; i++) {
2472 zonelist = pgdat->node_zonelists + i;
2473 zonelist->_zonerefs[0].zone = NULL;
2474 zonelist->_zonerefs[0].zone_idx = 0;
2477 /* NUMA-aware ordering of nodes */
2478 local_node = pgdat->node_id;
2479 load = nr_online_nodes;
2480 prev_node = local_node;
2481 nodes_clear(used_mask);
2483 memset(node_load, 0, sizeof(node_load));
2484 memset(node_order, 0, sizeof(node_order));
2487 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
2488 int distance = node_distance(local_node, node);
2491 * If another node is sufficiently far away then it is better
2492 * to reclaim pages in a zone before going off node.
2494 if (distance > RECLAIM_DISTANCE)
2495 zone_reclaim_mode = 1;
2498 * We don't want to pressure a particular node.
2499 * So adding penalty to the first node in same
2500 * distance group to make it round-robin.
2502 if (distance != node_distance(local_node, prev_node))
2503 node_load[node] = load;
2507 if (order == ZONELIST_ORDER_NODE)
2508 build_zonelists_in_node_order(pgdat, node);
2510 node_order[j++] = node; /* remember order */
2513 if (order == ZONELIST_ORDER_ZONE) {
2514 /* calculate node order -- i.e., DMA last! */
2515 build_zonelists_in_zone_order(pgdat, j);
2518 build_thisnode_zonelists(pgdat);
2521 /* Construct the zonelist performance cache - see further mmzone.h */
2522 static void build_zonelist_cache(pg_data_t *pgdat)
2524 struct zonelist *zonelist;
2525 struct zonelist_cache *zlc;
2528 zonelist = &pgdat->node_zonelists[0];
2529 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
2530 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
2531 for (z = zonelist->_zonerefs; z->zone; z++)
2532 zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z);
2536 #else /* CONFIG_NUMA */
2538 static void set_zonelist_order(void)
2540 current_zonelist_order = ZONELIST_ORDER_ZONE;
2543 static void build_zonelists(pg_data_t *pgdat)
2545 int node, local_node;
2547 struct zonelist *zonelist;
2549 local_node = pgdat->node_id;
2551 zonelist = &pgdat->node_zonelists[0];
2552 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
2555 * Now we build the zonelist so that it contains the zones
2556 * of all the other nodes.
2557 * We don't want to pressure a particular node, so when
2558 * building the zones for node N, we make sure that the
2559 * zones coming right after the local ones are those from
2560 * node N+1 (modulo N)
2562 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
2563 if (!node_online(node))
2565 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2568 for (node = 0; node < local_node; node++) {
2569 if (!node_online(node))
2571 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2575 zonelist->_zonerefs[j].zone = NULL;
2576 zonelist->_zonerefs[j].zone_idx = 0;
2579 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
2580 static void build_zonelist_cache(pg_data_t *pgdat)
2582 pgdat->node_zonelists[0].zlcache_ptr = NULL;
2585 #endif /* CONFIG_NUMA */
2587 /* return values int ....just for stop_machine() */
2588 static int __build_all_zonelists(void *dummy)
2592 for_each_online_node(nid) {
2593 pg_data_t *pgdat = NODE_DATA(nid);
2595 build_zonelists(pgdat);
2596 build_zonelist_cache(pgdat);
2601 void build_all_zonelists(void)
2603 set_zonelist_order();
2605 if (system_state == SYSTEM_BOOTING) {
2606 __build_all_zonelists(NULL);
2607 mminit_verify_zonelist();
2608 cpuset_init_current_mems_allowed();
2610 /* we have to stop all cpus to guarantee there is no user
2612 stop_machine(__build_all_zonelists, NULL, NULL);
2613 /* cpuset refresh routine should be here */
2615 vm_total_pages = nr_free_pagecache_pages();
2617 * Disable grouping by mobility if the number of pages in the
2618 * system is too low to allow the mechanism to work. It would be
2619 * more accurate, but expensive to check per-zone. This check is
2620 * made on memory-hotadd so a system can start with mobility
2621 * disabled and enable it later
2623 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
2624 page_group_by_mobility_disabled = 1;
2626 page_group_by_mobility_disabled = 0;
2628 printk("Built %i zonelists in %s order, mobility grouping %s. "
2629 "Total pages: %ld\n",
2631 zonelist_order_name[current_zonelist_order],
2632 page_group_by_mobility_disabled ? "off" : "on",
2635 printk("Policy zone: %s\n", zone_names[policy_zone]);
2640 * Helper functions to size the waitqueue hash table.
2641 * Essentially these want to choose hash table sizes sufficiently
2642 * large so that collisions trying to wait on pages are rare.
2643 * But in fact, the number of active page waitqueues on typical
2644 * systems is ridiculously low, less than 200. So this is even
2645 * conservative, even though it seems large.
2647 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
2648 * waitqueues, i.e. the size of the waitq table given the number of pages.
2650 #define PAGES_PER_WAITQUEUE 256
2652 #ifndef CONFIG_MEMORY_HOTPLUG
2653 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
2655 unsigned long size = 1;
2657 pages /= PAGES_PER_WAITQUEUE;
2659 while (size < pages)
2663 * Once we have dozens or even hundreds of threads sleeping
2664 * on IO we've got bigger problems than wait queue collision.
2665 * Limit the size of the wait table to a reasonable size.
2667 size = min(size, 4096UL);
2669 return max(size, 4UL);
2673 * A zone's size might be changed by hot-add, so it is not possible to determine
2674 * a suitable size for its wait_table. So we use the maximum size now.
2676 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
2678 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
2679 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
2680 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
2682 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
2683 * or more by the traditional way. (See above). It equals:
2685 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
2686 * ia64(16K page size) : = ( 8G + 4M)byte.
2687 * powerpc (64K page size) : = (32G +16M)byte.
2689 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
2696 * This is an integer logarithm so that shifts can be used later
2697 * to extract the more random high bits from the multiplicative
2698 * hash function before the remainder is taken.
2700 static inline unsigned long wait_table_bits(unsigned long size)
2705 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
2708 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
2709 * of blocks reserved is based on min_wmark_pages(zone). The memory within
2710 * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
2711 * higher will lead to a bigger reserve which will get freed as contiguous
2712 * blocks as reclaim kicks in
2714 static void setup_zone_migrate_reserve(struct zone *zone)
2716 unsigned long start_pfn, pfn, end_pfn;
2718 unsigned long reserve, block_migratetype;
2720 /* Get the start pfn, end pfn and the number of blocks to reserve */
2721 start_pfn = zone->zone_start_pfn;
2722 end_pfn = start_pfn + zone->spanned_pages;
2723 reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >>
2726 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
2727 if (!pfn_valid(pfn))
2729 page = pfn_to_page(pfn);
2731 /* Watch out for overlapping nodes */
2732 if (page_to_nid(page) != zone_to_nid(zone))
2735 /* Blocks with reserved pages will never free, skip them. */
2736 if (PageReserved(page))
2739 block_migratetype = get_pageblock_migratetype(page);
2741 /* If this block is reserved, account for it */
2742 if (reserve > 0 && block_migratetype == MIGRATE_RESERVE) {
2747 /* Suitable for reserving if this block is movable */
2748 if (reserve > 0 && block_migratetype == MIGRATE_MOVABLE) {
2749 set_pageblock_migratetype(page, MIGRATE_RESERVE);
2750 move_freepages_block(zone, page, MIGRATE_RESERVE);
2756 * If the reserve is met and this is a previous reserved block,
2759 if (block_migratetype == MIGRATE_RESERVE) {
2760 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
2761 move_freepages_block(zone, page, MIGRATE_MOVABLE);
2767 * Initially all pages are reserved - free ones are freed
2768 * up by free_all_bootmem() once the early boot process is
2769 * done. Non-atomic initialization, single-pass.
2771 void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
2772 unsigned long start_pfn, enum memmap_context context)
2775 unsigned long end_pfn = start_pfn + size;
2779 if (highest_memmap_pfn < end_pfn - 1)
2780 highest_memmap_pfn = end_pfn - 1;
2782 z = &NODE_DATA(nid)->node_zones[zone];
2783 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
2785 * There can be holes in boot-time mem_map[]s
2786 * handed to this function. They do not
2787 * exist on hotplugged memory.
2789 if (context == MEMMAP_EARLY) {
2790 if (!early_pfn_valid(pfn))
2792 if (!early_pfn_in_nid(pfn, nid))
2795 page = pfn_to_page(pfn);
2796 set_page_links(page, zone, nid, pfn);
2797 mminit_verify_page_links(page, zone, nid, pfn);
2798 init_page_count(page);
2799 reset_page_mapcount(page);
2800 SetPageReserved(page);
2802 * Mark the block movable so that blocks are reserved for
2803 * movable at startup. This will force kernel allocations
2804 * to reserve their blocks rather than leaking throughout
2805 * the address space during boot when many long-lived
2806 * kernel allocations are made. Later some blocks near
2807 * the start are marked MIGRATE_RESERVE by
2808 * setup_zone_migrate_reserve()
2810 * bitmap is created for zone's valid pfn range. but memmap
2811 * can be created for invalid pages (for alignment)
2812 * check here not to call set_pageblock_migratetype() against
2815 if ((z->zone_start_pfn <= pfn)
2816 && (pfn < z->zone_start_pfn + z->spanned_pages)
2817 && !(pfn & (pageblock_nr_pages - 1)))
2818 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
2820 INIT_LIST_HEAD(&page->lru);
2821 #ifdef WANT_PAGE_VIRTUAL
2822 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
2823 if (!is_highmem_idx(zone))
2824 set_page_address(page, __va(pfn << PAGE_SHIFT));
2829 static void __meminit zone_init_free_lists(struct zone *zone)
2832 for_each_migratetype_order(order, t) {
2833 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
2834 zone->free_area[order].nr_free = 0;
2838 #ifndef __HAVE_ARCH_MEMMAP_INIT
2839 #define memmap_init(size, nid, zone, start_pfn) \
2840 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
2843 static int zone_batchsize(struct zone *zone)
2849 * The per-cpu-pages pools are set to around 1000th of the
2850 * size of the zone. But no more than 1/2 of a meg.
2852 * OK, so we don't know how big the cache is. So guess.
2854 batch = zone->present_pages / 1024;
2855 if (batch * PAGE_SIZE > 512 * 1024)
2856 batch = (512 * 1024) / PAGE_SIZE;
2857 batch /= 4; /* We effectively *= 4 below */
2862 * Clamp the batch to a 2^n - 1 value. Having a power
2863 * of 2 value was found to be more likely to have
2864 * suboptimal cache aliasing properties in some cases.
2866 * For example if 2 tasks are alternately allocating
2867 * batches of pages, one task can end up with a lot
2868 * of pages of one half of the possible page colors
2869 * and the other with pages of the other colors.
2871 batch = rounddown_pow_of_two(batch + batch/2) - 1;
2876 /* The deferral and batching of frees should be suppressed under NOMMU
2879 * The problem is that NOMMU needs to be able to allocate large chunks
2880 * of contiguous memory as there's no hardware page translation to
2881 * assemble apparent contiguous memory from discontiguous pages.
2883 * Queueing large contiguous runs of pages for batching, however,
2884 * causes the pages to actually be freed in smaller chunks. As there
2885 * can be a significant delay between the individual batches being
2886 * recycled, this leads to the once large chunks of space being
2887 * fragmented and becoming unavailable for high-order allocations.
2893 static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
2895 struct per_cpu_pages *pcp;
2897 memset(p, 0, sizeof(*p));
2901 pcp->high = 6 * batch;
2902 pcp->batch = max(1UL, 1 * batch);
2903 INIT_LIST_HEAD(&pcp->list);
2907 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
2908 * to the value high for the pageset p.
2911 static void setup_pagelist_highmark(struct per_cpu_pageset *p,
2914 struct per_cpu_pages *pcp;
2918 pcp->batch = max(1UL, high/4);
2919 if ((high/4) > (PAGE_SHIFT * 8))
2920 pcp->batch = PAGE_SHIFT * 8;
2926 * Boot pageset table. One per cpu which is going to be used for all
2927 * zones and all nodes. The parameters will be set in such a way
2928 * that an item put on a list will immediately be handed over to
2929 * the buddy list. This is safe since pageset manipulation is done
2930 * with interrupts disabled.
2932 * Some NUMA counter updates may also be caught by the boot pagesets.
2934 * The boot_pagesets must be kept even after bootup is complete for
2935 * unused processors and/or zones. They do play a role for bootstrapping
2936 * hotplugged processors.
2938 * zoneinfo_show() and maybe other functions do
2939 * not check if the processor is online before following the pageset pointer.
2940 * Other parts of the kernel may not check if the zone is available.
2942 static struct per_cpu_pageset boot_pageset[NR_CPUS];
2945 * Dynamically allocate memory for the
2946 * per cpu pageset array in struct zone.
2948 static int __cpuinit process_zones(int cpu)
2950 struct zone *zone, *dzone;
2951 int node = cpu_to_node(cpu);
2953 node_set_state(node, N_CPU); /* this node has a cpu */
2955 for_each_populated_zone(zone) {
2956 zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset),
2958 if (!zone_pcp(zone, cpu))
2961 setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone));
2963 if (percpu_pagelist_fraction)
2964 setup_pagelist_highmark(zone_pcp(zone, cpu),
2965 (zone->present_pages / percpu_pagelist_fraction));
2970 for_each_zone(dzone) {
2971 if (!populated_zone(dzone))
2975 kfree(zone_pcp(dzone, cpu));
2976 zone_pcp(dzone, cpu) = NULL;
2981 static inline void free_zone_pagesets(int cpu)
2985 for_each_zone(zone) {
2986 struct per_cpu_pageset *pset = zone_pcp(zone, cpu);
2988 /* Free per_cpu_pageset if it is slab allocated */
2989 if (pset != &boot_pageset[cpu])
2991 zone_pcp(zone, cpu) = NULL;
2995 static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb,
2996 unsigned long action,
2999 int cpu = (long)hcpu;
3000 int ret = NOTIFY_OK;
3003 case CPU_UP_PREPARE:
3004 case CPU_UP_PREPARE_FROZEN:
3005 if (process_zones(cpu))
3008 case CPU_UP_CANCELED:
3009 case CPU_UP_CANCELED_FROZEN:
3011 case CPU_DEAD_FROZEN:
3012 free_zone_pagesets(cpu);
3020 static struct notifier_block __cpuinitdata pageset_notifier =
3021 { &pageset_cpuup_callback, NULL, 0 };
3023 void __init setup_per_cpu_pageset(void)
3027 /* Initialize per_cpu_pageset for cpu 0.
3028 * A cpuup callback will do this for every cpu
3029 * as it comes online
3031 err = process_zones(smp_processor_id());
3033 register_cpu_notifier(&pageset_notifier);
3038 static noinline __init_refok
3039 int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
3042 struct pglist_data *pgdat = zone->zone_pgdat;
3046 * The per-page waitqueue mechanism uses hashed waitqueues
3049 zone->wait_table_hash_nr_entries =
3050 wait_table_hash_nr_entries(zone_size_pages);
3051 zone->wait_table_bits =
3052 wait_table_bits(zone->wait_table_hash_nr_entries);
3053 alloc_size = zone->wait_table_hash_nr_entries
3054 * sizeof(wait_queue_head_t);
3056 if (!slab_is_available()) {
3057 zone->wait_table = (wait_queue_head_t *)
3058 alloc_bootmem_node(pgdat, alloc_size);
3061 * This case means that a zone whose size was 0 gets new memory
3062 * via memory hot-add.
3063 * But it may be the case that a new node was hot-added. In
3064 * this case vmalloc() will not be able to use this new node's
3065 * memory - this wait_table must be initialized to use this new
3066 * node itself as well.
3067 * To use this new node's memory, further consideration will be
3070 zone->wait_table = vmalloc(alloc_size);
3072 if (!zone->wait_table)
3075 for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
3076 init_waitqueue_head(zone->wait_table + i);
3081 static __meminit void zone_pcp_init(struct zone *zone)
3084 unsigned long batch = zone_batchsize(zone);
3086 for (cpu = 0; cpu < NR_CPUS; cpu++) {
3088 /* Early boot. Slab allocator not functional yet */
3089 zone_pcp(zone, cpu) = &boot_pageset[cpu];
3090 setup_pageset(&boot_pageset[cpu],0);
3092 setup_pageset(zone_pcp(zone,cpu), batch);
3095 if (zone->present_pages)
3096 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n",
3097 zone->name, zone->present_pages, batch);
3100 __meminit int init_currently_empty_zone(struct zone *zone,
3101 unsigned long zone_start_pfn,
3103 enum memmap_context context)
3105 struct pglist_data *pgdat = zone->zone_pgdat;
3107 ret = zone_wait_table_init(zone, size);
3110 pgdat->nr_zones = zone_idx(zone) + 1;
3112 zone->zone_start_pfn = zone_start_pfn;
3114 mminit_dprintk(MMINIT_TRACE, "memmap_init",
3115 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
3117 (unsigned long)zone_idx(zone),
3118 zone_start_pfn, (zone_start_pfn + size));
3120 zone_init_free_lists(zone);
3125 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3127 * Basic iterator support. Return the first range of PFNs for a node
3128 * Note: nid == MAX_NUMNODES returns first region regardless of node
3130 static int __meminit first_active_region_index_in_nid(int nid)
3134 for (i = 0; i < nr_nodemap_entries; i++)
3135 if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
3142 * Basic iterator support. Return the next active range of PFNs for a node
3143 * Note: nid == MAX_NUMNODES returns next region regardless of node
3145 static int __meminit next_active_region_index_in_nid(int index, int nid)
3147 for (index = index + 1; index < nr_nodemap_entries; index++)
3148 if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
3154 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
3156 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
3157 * Architectures may implement their own version but if add_active_range()
3158 * was used and there are no special requirements, this is a convenient
3161 int __meminit __early_pfn_to_nid(unsigned long pfn)
3165 for (i = 0; i < nr_nodemap_entries; i++) {
3166 unsigned long start_pfn = early_node_map[i].start_pfn;
3167 unsigned long end_pfn = early_node_map[i].end_pfn;
3169 if (start_pfn <= pfn && pfn < end_pfn)
3170 return early_node_map[i].nid;
3172 /* This is a memory hole */
3175 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
3177 int __meminit early_pfn_to_nid(unsigned long pfn)
3181 nid = __early_pfn_to_nid(pfn);
3184 /* just returns 0 */
3188 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
3189 bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
3193 nid = __early_pfn_to_nid(pfn);
3194 if (nid >= 0 && nid != node)
3200 /* Basic iterator support to walk early_node_map[] */
3201 #define for_each_active_range_index_in_nid(i, nid) \
3202 for (i = first_active_region_index_in_nid(nid); i != -1; \
3203 i = next_active_region_index_in_nid(i, nid))
3206 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
3207 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
3208 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
3210 * If an architecture guarantees that all ranges registered with
3211 * add_active_ranges() contain no holes and may be freed, this
3212 * this function may be used instead of calling free_bootmem() manually.
3214 void __init free_bootmem_with_active_regions(int nid,
3215 unsigned long max_low_pfn)
3219 for_each_active_range_index_in_nid(i, nid) {
3220 unsigned long size_pages = 0;
3221 unsigned long end_pfn = early_node_map[i].end_pfn;
3223 if (early_node_map[i].start_pfn >= max_low_pfn)
3226 if (end_pfn > max_low_pfn)
3227 end_pfn = max_low_pfn;
3229 size_pages = end_pfn - early_node_map[i].start_pfn;
3230 free_bootmem_node(NODE_DATA(early_node_map[i].nid),
3231 PFN_PHYS(early_node_map[i].start_pfn),
3232 size_pages << PAGE_SHIFT);
3236 void __init work_with_active_regions(int nid, work_fn_t work_fn, void *data)
3241 for_each_active_range_index_in_nid(i, nid) {
3242 ret = work_fn(early_node_map[i].start_pfn,
3243 early_node_map[i].end_pfn, data);
3249 * sparse_memory_present_with_active_regions - Call memory_present for each active range
3250 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
3252 * If an architecture guarantees that all ranges registered with
3253 * add_active_ranges() contain no holes and may be freed, this
3254 * function may be used instead of calling memory_present() manually.
3256 void __init sparse_memory_present_with_active_regions(int nid)
3260 for_each_active_range_index_in_nid(i, nid)
3261 memory_present(early_node_map[i].nid,
3262 early_node_map[i].start_pfn,
3263 early_node_map[i].end_pfn);
3267 * get_pfn_range_for_nid - Return the start and end page frames for a node
3268 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
3269 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
3270 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
3272 * It returns the start and end page frame of a node based on information
3273 * provided by an arch calling add_active_range(). If called for a node
3274 * with no available memory, a warning is printed and the start and end
3277 void __meminit get_pfn_range_for_nid(unsigned int nid,
3278 unsigned long *start_pfn, unsigned long *end_pfn)
3284 for_each_active_range_index_in_nid(i, nid) {
3285 *start_pfn = min(*start_pfn, early_node_map[i].start_pfn);
3286 *end_pfn = max(*end_pfn, early_node_map[i].end_pfn);
3289 if (*start_pfn == -1UL)
3294 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3295 * assumption is made that zones within a node are ordered in monotonic
3296 * increasing memory addresses so that the "highest" populated zone is used
3298 static void __init find_usable_zone_for_movable(void)
3301 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
3302 if (zone_index == ZONE_MOVABLE)
3305 if (arch_zone_highest_possible_pfn[zone_index] >
3306 arch_zone_lowest_possible_pfn[zone_index])
3310 VM_BUG_ON(zone_index == -1);
3311 movable_zone = zone_index;
3315 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
3316 * because it is sized independant of architecture. Unlike the other zones,
3317 * the starting point for ZONE_MOVABLE is not fixed. It may be different
3318 * in each node depending on the size of each node and how evenly kernelcore
3319 * is distributed. This helper function adjusts the zone ranges
3320 * provided by the architecture for a given node by using the end of the
3321 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3322 * zones within a node are in order of monotonic increases memory addresses
3324 static void __meminit adjust_zone_range_for_zone_movable(int nid,
3325 unsigned long zone_type,
3326 unsigned long node_start_pfn,
3327 unsigned long node_end_pfn,
3328 unsigned long *zone_start_pfn,
3329 unsigned long *zone_end_pfn)
3331 /* Only adjust if ZONE_MOVABLE is on this node */
3332 if (zone_movable_pfn[nid]) {
3333 /* Size ZONE_MOVABLE */
3334 if (zone_type == ZONE_MOVABLE) {
3335 *zone_start_pfn = zone_movable_pfn[nid];
3336 *zone_end_pfn = min(node_end_pfn,
3337 arch_zone_highest_possible_pfn[movable_zone]);
3339 /* Adjust for ZONE_MOVABLE starting within this range */
3340 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
3341 *zone_end_pfn > zone_movable_pfn[nid]) {
3342 *zone_end_pfn = zone_movable_pfn[nid];
3344 /* Check if this whole range is within ZONE_MOVABLE */
3345 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
3346 *zone_start_pfn = *zone_end_pfn;
3351 * Return the number of pages a zone spans in a node, including holes
3352 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3354 static unsigned long __meminit zone_spanned_pages_in_node(int nid,
3355 unsigned long zone_type,
3356 unsigned long *ignored)
3358 unsigned long node_start_pfn, node_end_pfn;
3359 unsigned long zone_start_pfn, zone_end_pfn;
3361 /* Get the start and end of the node and zone */
3362 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3363 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
3364 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
3365 adjust_zone_range_for_zone_movable(nid, zone_type,
3366 node_start_pfn, node_end_pfn,
3367 &zone_start_pfn, &zone_end_pfn);
3369 /* Check that this node has pages within the zone's required range */
3370 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
3373 /* Move the zone boundaries inside the node if necessary */
3374 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
3375 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
3377 /* Return the spanned pages */
3378 return zone_end_pfn - zone_start_pfn;
3382 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
3383 * then all holes in the requested range will be accounted for.
3385 static unsigned long __meminit __absent_pages_in_range(int nid,
3386 unsigned long range_start_pfn,
3387 unsigned long range_end_pfn)
3390 unsigned long prev_end_pfn = 0, hole_pages = 0;
3391 unsigned long start_pfn;
3393 /* Find the end_pfn of the first active range of pfns in the node */
3394 i = first_active_region_index_in_nid(nid);
3398 prev_end_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3400 /* Account for ranges before physical memory on this node */
3401 if (early_node_map[i].start_pfn > range_start_pfn)
3402 hole_pages = prev_end_pfn - range_start_pfn;
3404 /* Find all holes for the zone within the node */
3405 for (; i != -1; i = next_active_region_index_in_nid(i, nid)) {
3407 /* No need to continue if prev_end_pfn is outside the zone */
3408 if (prev_end_pfn >= range_end_pfn)
3411 /* Make sure the end of the zone is not within the hole */
3412 start_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3413 prev_end_pfn = max(prev_end_pfn, range_start_pfn);
3415 /* Update the hole size cound and move on */
3416 if (start_pfn > range_start_pfn) {
3417 BUG_ON(prev_end_pfn > start_pfn);
3418 hole_pages += start_pfn - prev_end_pfn;
3420 prev_end_pfn = early_node_map[i].end_pfn;
3423 /* Account for ranges past physical memory on this node */
3424 if (range_end_pfn > prev_end_pfn)
3425 hole_pages += range_end_pfn -
3426 max(range_start_pfn, prev_end_pfn);
3432 * absent_pages_in_range - Return number of page frames in holes within a range
3433 * @start_pfn: The start PFN to start searching for holes
3434 * @end_pfn: The end PFN to stop searching for holes
3436 * It returns the number of pages frames in memory holes within a range.
3438 unsigned long __init absent_pages_in_range(unsigned long start_pfn,
3439 unsigned long end_pfn)
3441 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
3444 /* Return the number of page frames in holes in a zone on a node */
3445 static unsigned long __meminit zone_absent_pages_in_node(int nid,
3446 unsigned long zone_type,
3447 unsigned long *ignored)
3449 unsigned long node_start_pfn, node_end_pfn;
3450 unsigned long zone_start_pfn, zone_end_pfn;
3452 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3453 zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type],
3455 zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type],
3458 adjust_zone_range_for_zone_movable(nid, zone_type,
3459 node_start_pfn, node_end_pfn,
3460 &zone_start_pfn, &zone_end_pfn);
3461 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
3465 static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
3466 unsigned long zone_type,
3467 unsigned long *zones_size)
3469 return zones_size[zone_type];
3472 static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
3473 unsigned long zone_type,
3474 unsigned long *zholes_size)
3479 return zholes_size[zone_type];
3484 static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
3485 unsigned long *zones_size, unsigned long *zholes_size)
3487 unsigned long realtotalpages, totalpages = 0;
3490 for (i = 0; i < MAX_NR_ZONES; i++)
3491 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
3493 pgdat->node_spanned_pages = totalpages;
3495 realtotalpages = totalpages;
3496 for (i = 0; i < MAX_NR_ZONES; i++)
3498 zone_absent_pages_in_node(pgdat->node_id, i,
3500 pgdat->node_present_pages = realtotalpages;
3501 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
3505 #ifndef CONFIG_SPARSEMEM
3507 * Calculate the size of the zone->blockflags rounded to an unsigned long
3508 * Start by making sure zonesize is a multiple of pageblock_order by rounding
3509 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
3510 * round what is now in bits to nearest long in bits, then return it in
3513 static unsigned long __init usemap_size(unsigned long zonesize)
3515 unsigned long usemapsize;
3517 usemapsize = roundup(zonesize, pageblock_nr_pages);
3518 usemapsize = usemapsize >> pageblock_order;
3519 usemapsize *= NR_PAGEBLOCK_BITS;
3520 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
3522 return usemapsize / 8;
3525 static void __init setup_usemap(struct pglist_data *pgdat,
3526 struct zone *zone, unsigned long zonesize)
3528 unsigned long usemapsize = usemap_size(zonesize);
3529 zone->pageblock_flags = NULL;
3531 zone->pageblock_flags = alloc_bootmem_node(pgdat, usemapsize);
3534 static void inline setup_usemap(struct pglist_data *pgdat,
3535 struct zone *zone, unsigned long zonesize) {}
3536 #endif /* CONFIG_SPARSEMEM */
3538 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
3540 /* Return a sensible default order for the pageblock size. */
3541 static inline int pageblock_default_order(void)
3543 if (HPAGE_SHIFT > PAGE_SHIFT)
3544 return HUGETLB_PAGE_ORDER;
3549 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
3550 static inline void __init set_pageblock_order(unsigned int order)
3552 /* Check that pageblock_nr_pages has not already been setup */
3553 if (pageblock_order)
3557 * Assume the largest contiguous order of interest is a huge page.
3558 * This value may be variable depending on boot parameters on IA64
3560 pageblock_order = order;
3562 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3565 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
3566 * and pageblock_default_order() are unused as pageblock_order is set
3567 * at compile-time. See include/linux/pageblock-flags.h for the values of
3568 * pageblock_order based on the kernel config
3570 static inline int pageblock_default_order(unsigned int order)
3574 #define set_pageblock_order(x) do {} while (0)
3576 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3579 * Set up the zone data structures:
3580 * - mark all pages reserved
3581 * - mark all memory queues empty
3582 * - clear the memory bitmaps
3584 static void __paginginit free_area_init_core(struct pglist_data *pgdat,
3585 unsigned long *zones_size, unsigned long *zholes_size)
3588 int nid = pgdat->node_id;
3589 unsigned long zone_start_pfn = pgdat->node_start_pfn;
3592 pgdat_resize_init(pgdat);
3593 pgdat->nr_zones = 0;
3594 init_waitqueue_head(&pgdat->kswapd_wait);
3595 pgdat->kswapd_max_order = 0;
3596 pgdat_page_cgroup_init(pgdat);
3598 for (j = 0; j < MAX_NR_ZONES; j++) {
3599 struct zone *zone = pgdat->node_zones + j;
3600 unsigned long size, realsize, memmap_pages;
3603 size = zone_spanned_pages_in_node(nid, j, zones_size);
3604 realsize = size - zone_absent_pages_in_node(nid, j,
3608 * Adjust realsize so that it accounts for how much memory
3609 * is used by this zone for memmap. This affects the watermark
3610 * and per-cpu initialisations
3613 PAGE_ALIGN(size * sizeof(struct page)) >> PAGE_SHIFT;
3614 if (realsize >= memmap_pages) {
3615 realsize -= memmap_pages;
3618 " %s zone: %lu pages used for memmap\n",
3619 zone_names[j], memmap_pages);
3622 " %s zone: %lu pages exceeds realsize %lu\n",
3623 zone_names[j], memmap_pages, realsize);
3625 /* Account for reserved pages */
3626 if (j == 0 && realsize > dma_reserve) {
3627 realsize -= dma_reserve;
3628 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
3629 zone_names[0], dma_reserve);
3632 if (!is_highmem_idx(j))
3633 nr_kernel_pages += realsize;
3634 nr_all_pages += realsize;
3636 zone->spanned_pages = size;
3637 zone->present_pages = realsize;
3640 zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
3642 zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100;
3644 zone->name = zone_names[j];
3645 spin_lock_init(&zone->lock);
3646 spin_lock_init(&zone->lru_lock);
3647 zone_seqlock_init(zone);
3648 zone->zone_pgdat = pgdat;
3650 zone->prev_priority = DEF_PRIORITY;
3652 zone_pcp_init(zone);
3654 INIT_LIST_HEAD(&zone->lru[l].list);
3655 zone->lru[l].nr_scan = 0;
3657 zone->reclaim_stat.recent_rotated[0] = 0;
3658 zone->reclaim_stat.recent_rotated[1] = 0;
3659 zone->reclaim_stat.recent_scanned[0] = 0;
3660 zone->reclaim_stat.recent_scanned[1] = 0;
3661 zap_zone_vm_stats(zone);
3666 set_pageblock_order(pageblock_default_order());
3667 setup_usemap(pgdat, zone, size);
3668 ret = init_currently_empty_zone(zone, zone_start_pfn,
3669 size, MEMMAP_EARLY);
3671 memmap_init(size, nid, j, zone_start_pfn);
3672 zone_start_pfn += size;
3676 static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
3678 /* Skip empty nodes */
3679 if (!pgdat->node_spanned_pages)
3682 #ifdef CONFIG_FLAT_NODE_MEM_MAP
3683 /* ia64 gets its own node_mem_map, before this, without bootmem */
3684 if (!pgdat->node_mem_map) {
3685 unsigned long size, start, end;
3689 * The zone's endpoints aren't required to be MAX_ORDER
3690 * aligned but the node_mem_map endpoints must be in order
3691 * for the buddy allocator to function correctly.
3693 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
3694 end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
3695 end = ALIGN(end, MAX_ORDER_NR_PAGES);
3696 size = (end - start) * sizeof(struct page);
3697 map = alloc_remap(pgdat->node_id, size);
3699 map = alloc_bootmem_node(pgdat, size);
3700 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
3702 #ifndef CONFIG_NEED_MULTIPLE_NODES
3704 * With no DISCONTIG, the global mem_map is just set as node 0's
3706 if (pgdat == NODE_DATA(0)) {
3707 mem_map = NODE_DATA(0)->node_mem_map;
3708 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3709 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
3710 mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
3711 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3714 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
3717 void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
3718 unsigned long node_start_pfn, unsigned long *zholes_size)
3720 pg_data_t *pgdat = NODE_DATA(nid);
3722 pgdat->node_id = nid;
3723 pgdat->node_start_pfn = node_start_pfn;
3724 calculate_node_totalpages(pgdat, zones_size, zholes_size);
3726 alloc_node_mem_map(pgdat);
3727 #ifdef CONFIG_FLAT_NODE_MEM_MAP
3728 printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
3729 nid, (unsigned long)pgdat,
3730 (unsigned long)pgdat->node_mem_map);
3733 free_area_init_core(pgdat, zones_size, zholes_size);
3736 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3738 #if MAX_NUMNODES > 1
3740 * Figure out the number of possible node ids.
3742 static void __init setup_nr_node_ids(void)
3745 unsigned int highest = 0;
3747 for_each_node_mask(node, node_possible_map)
3749 nr_node_ids = highest + 1;
3752 static inline void setup_nr_node_ids(void)
3758 * add_active_range - Register a range of PFNs backed by physical memory
3759 * @nid: The node ID the range resides on
3760 * @start_pfn: The start PFN of the available physical memory
3761 * @end_pfn: The end PFN of the available physical memory
3763 * These ranges are stored in an early_node_map[] and later used by
3764 * free_area_init_nodes() to calculate zone sizes and holes. If the
3765 * range spans a memory hole, it is up to the architecture to ensure
3766 * the memory is not freed by the bootmem allocator. If possible
3767 * the range being registered will be merged with existing ranges.
3769 void __init add_active_range(unsigned int nid, unsigned long start_pfn,
3770 unsigned long end_pfn)
3774 mminit_dprintk(MMINIT_TRACE, "memory_register",
3775 "Entering add_active_range(%d, %#lx, %#lx) "
3776 "%d entries of %d used\n",
3777 nid, start_pfn, end_pfn,
3778 nr_nodemap_entries, MAX_ACTIVE_REGIONS);
3780 mminit_validate_memmodel_limits(&start_pfn, &end_pfn);
3782 /* Merge with existing active regions if possible */
3783 for (i = 0; i < nr_nodemap_entries; i++) {
3784 if (early_node_map[i].nid != nid)
3787 /* Skip if an existing region covers this new one */
3788 if (start_pfn >= early_node_map[i].start_pfn &&
3789 end_pfn <= early_node_map[i].end_pfn)
3792 /* Merge forward if suitable */
3793 if (start_pfn <= early_node_map[i].end_pfn &&
3794 end_pfn > early_node_map[i].end_pfn) {
3795 early_node_map[i].end_pfn = end_pfn;
3799 /* Merge backward if suitable */
3800 if (start_pfn < early_node_map[i].end_pfn &&
3801 end_pfn >= early_node_map[i].start_pfn) {
3802 early_node_map[i].start_pfn = start_pfn;
3807 /* Check that early_node_map is large enough */
3808 if (i >= MAX_ACTIVE_REGIONS) {
3809 printk(KERN_CRIT "More than %d memory regions, truncating\n",
3810 MAX_ACTIVE_REGIONS);
3814 early_node_map[i].nid = nid;
3815 early_node_map[i].start_pfn = start_pfn;
3816 early_node_map[i].end_pfn = end_pfn;
3817 nr_nodemap_entries = i + 1;
3821 * remove_active_range - Shrink an existing registered range of PFNs
3822 * @nid: The node id the range is on that should be shrunk
3823 * @start_pfn: The new PFN of the range
3824 * @end_pfn: The new PFN of the range
3826 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
3827 * The map is kept near the end physical page range that has already been
3828 * registered. This function allows an arch to shrink an existing registered
3831 void __init remove_active_range(unsigned int nid, unsigned long start_pfn,
3832 unsigned long end_pfn)
3837 printk(KERN_DEBUG "remove_active_range (%d, %lu, %lu)\n",
3838 nid, start_pfn, end_pfn);
3840 /* Find the old active region end and shrink */
3841 for_each_active_range_index_in_nid(i, nid) {
3842 if (early_node_map[i].start_pfn >= start_pfn &&
3843 early_node_map[i].end_pfn <= end_pfn) {
3845 early_node_map[i].start_pfn = 0;
3846 early_node_map[i].end_pfn = 0;
3850 if (early_node_map[i].start_pfn < start_pfn &&
3851 early_node_map[i].end_pfn > start_pfn) {
3852 unsigned long temp_end_pfn = early_node_map[i].end_pfn;
3853 early_node_map[i].end_pfn = start_pfn;
3854 if (temp_end_pfn > end_pfn)
3855 add_active_range(nid, end_pfn, temp_end_pfn);
3858 if (early_node_map[i].start_pfn >= start_pfn &&
3859 early_node_map[i].end_pfn > end_pfn &&
3860 early_node_map[i].start_pfn < end_pfn) {
3861 early_node_map[i].start_pfn = end_pfn;
3869 /* remove the blank ones */
3870 for (i = nr_nodemap_entries - 1; i > 0; i--) {
3871 if (early_node_map[i].nid != nid)
3873 if (early_node_map[i].end_pfn)
3875 /* we found it, get rid of it */
3876 for (j = i; j < nr_nodemap_entries - 1; j++)
3877 memcpy(&early_node_map[j], &early_node_map[j+1],
3878 sizeof(early_node_map[j]));
3879 j = nr_nodemap_entries - 1;
3880 memset(&early_node_map[j], 0, sizeof(early_node_map[j]));
3881 nr_nodemap_entries--;
3886 * remove_all_active_ranges - Remove all currently registered regions
3888 * During discovery, it may be found that a table like SRAT is invalid
3889 * and an alternative discovery method must be used. This function removes
3890 * all currently registered regions.
3892 void __init remove_all_active_ranges(void)
3894 memset(early_node_map, 0, sizeof(early_node_map));
3895 nr_nodemap_entries = 0;
3898 /* Compare two active node_active_regions */
3899 static int __init cmp_node_active_region(const void *a, const void *b)
3901 struct node_active_region *arange = (struct node_active_region *)a;
3902 struct node_active_region *brange = (struct node_active_region *)b;
3904 /* Done this way to avoid overflows */
3905 if (arange->start_pfn > brange->start_pfn)
3907 if (arange->start_pfn < brange->start_pfn)
3913 /* sort the node_map by start_pfn */
3914 static void __init sort_node_map(void)
3916 sort(early_node_map, (size_t)nr_nodemap_entries,
3917 sizeof(struct node_active_region),
3918 cmp_node_active_region, NULL);
3921 /* Find the lowest pfn for a node */
3922 static unsigned long __init find_min_pfn_for_node(int nid)
3925 unsigned long min_pfn = ULONG_MAX;
3927 /* Assuming a sorted map, the first range found has the starting pfn */
3928 for_each_active_range_index_in_nid(i, nid)
3929 min_pfn = min(min_pfn, early_node_map[i].start_pfn);
3931 if (min_pfn == ULONG_MAX) {
3933 "Could not find start_pfn for node %d\n", nid);
3941 * find_min_pfn_with_active_regions - Find the minimum PFN registered
3943 * It returns the minimum PFN based on information provided via
3944 * add_active_range().
3946 unsigned long __init find_min_pfn_with_active_regions(void)
3948 return find_min_pfn_for_node(MAX_NUMNODES);
3952 * early_calculate_totalpages()
3953 * Sum pages in active regions for movable zone.
3954 * Populate N_HIGH_MEMORY for calculating usable_nodes.
3956 static unsigned long __init early_calculate_totalpages(void)
3959 unsigned long totalpages = 0;
3961 for (i = 0; i < nr_nodemap_entries; i++) {
3962 unsigned long pages = early_node_map[i].end_pfn -
3963 early_node_map[i].start_pfn;
3964 totalpages += pages;
3966 node_set_state(early_node_map[i].nid, N_HIGH_MEMORY);
3972 * Find the PFN the Movable zone begins in each node. Kernel memory
3973 * is spread evenly between nodes as long as the nodes have enough
3974 * memory. When they don't, some nodes will have more kernelcore than
3977 static void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn)
3980 unsigned long usable_startpfn;
3981 unsigned long kernelcore_node, kernelcore_remaining;
3982 unsigned long totalpages = early_calculate_totalpages();
3983 int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]);
3986 * If movablecore was specified, calculate what size of
3987 * kernelcore that corresponds so that memory usable for
3988 * any allocation type is evenly spread. If both kernelcore
3989 * and movablecore are specified, then the value of kernelcore
3990 * will be used for required_kernelcore if it's greater than
3991 * what movablecore would have allowed.
3993 if (required_movablecore) {
3994 unsigned long corepages;
3997 * Round-up so that ZONE_MOVABLE is at least as large as what
3998 * was requested by the user
4000 required_movablecore =
4001 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
4002 corepages = totalpages - required_movablecore;
4004 required_kernelcore = max(required_kernelcore, corepages);
4007 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
4008 if (!required_kernelcore)
4011 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
4012 find_usable_zone_for_movable();
4013 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
4016 /* Spread kernelcore memory as evenly as possible throughout nodes */
4017 kernelcore_node = required_kernelcore / usable_nodes;
4018 for_each_node_state(nid, N_HIGH_MEMORY) {
4020 * Recalculate kernelcore_node if the division per node
4021 * now exceeds what is necessary to satisfy the requested
4022 * amount of memory for the kernel
4024 if (required_kernelcore < kernelcore_node)
4025 kernelcore_node = required_kernelcore / usable_nodes;
4028 * As the map is walked, we track how much memory is usable
4029 * by the kernel using kernelcore_remaining. When it is
4030 * 0, the rest of the node is usable by ZONE_MOVABLE
4032 kernelcore_remaining = kernelcore_node;
4034 /* Go through each range of PFNs within this node */
4035 for_each_active_range_index_in_nid(i, nid) {
4036 unsigned long start_pfn, end_pfn;
4037 unsigned long size_pages;
4039 start_pfn = max(early_node_map[i].start_pfn,
4040 zone_movable_pfn[nid]);
4041 end_pfn = early_node_map[i].end_pfn;
4042 if (start_pfn >= end_pfn)
4045 /* Account for what is only usable for kernelcore */
4046 if (start_pfn < usable_startpfn) {
4047 unsigned long kernel_pages;
4048 kernel_pages = min(end_pfn, usable_startpfn)
4051 kernelcore_remaining -= min(kernel_pages,
4052 kernelcore_remaining);
4053 required_kernelcore -= min(kernel_pages,
4054 required_kernelcore);
4056 /* Continue if range is now fully accounted */
4057 if (end_pfn <= usable_startpfn) {
4060 * Push zone_movable_pfn to the end so
4061 * that if we have to rebalance
4062 * kernelcore across nodes, we will
4063 * not double account here
4065 zone_movable_pfn[nid] = end_pfn;
4068 start_pfn = usable_startpfn;
4072 * The usable PFN range for ZONE_MOVABLE is from
4073 * start_pfn->end_pfn. Calculate size_pages as the
4074 * number of pages used as kernelcore
4076 size_pages = end_pfn - start_pfn;
4077 if (size_pages > kernelcore_remaining)
4078 size_pages = kernelcore_remaining;
4079 zone_movable_pfn[nid] = start_pfn + size_pages;
4082 * Some kernelcore has been met, update counts and
4083 * break if the kernelcore for this node has been
4086 required_kernelcore -= min(required_kernelcore,
4088 kernelcore_remaining -= size_pages;
4089 if (!kernelcore_remaining)
4095 * If there is still required_kernelcore, we do another pass with one
4096 * less node in the count. This will push zone_movable_pfn[nid] further
4097 * along on the nodes that still have memory until kernelcore is
4101 if (usable_nodes && required_kernelcore > usable_nodes)
4104 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
4105 for (nid = 0; nid < MAX_NUMNODES; nid++)
4106 zone_movable_pfn[nid] =
4107 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
4110 /* Any regular memory on that node ? */
4111 static void check_for_regular_memory(pg_data_t *pgdat)
4113 #ifdef CONFIG_HIGHMEM
4114 enum zone_type zone_type;
4116 for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) {
4117 struct zone *zone = &pgdat->node_zones[zone_type];
4118 if (zone->present_pages)
4119 node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY);
4125 * free_area_init_nodes - Initialise all pg_data_t and zone data
4126 * @max_zone_pfn: an array of max PFNs for each zone
4128 * This will call free_area_init_node() for each active node in the system.
4129 * Using the page ranges provided by add_active_range(), the size of each
4130 * zone in each node and their holes is calculated. If the maximum PFN
4131 * between two adjacent zones match, it is assumed that the zone is empty.
4132 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
4133 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
4134 * starts where the previous one ended. For example, ZONE_DMA32 starts
4135 * at arch_max_dma_pfn.
4137 void __init free_area_init_nodes(unsigned long *max_zone_pfn)
4142 /* Sort early_node_map as initialisation assumes it is sorted */
4145 /* Record where the zone boundaries are */
4146 memset(arch_zone_lowest_possible_pfn, 0,
4147 sizeof(arch_zone_lowest_possible_pfn));
4148 memset(arch_zone_highest_possible_pfn, 0,
4149 sizeof(arch_zone_highest_possible_pfn));
4150 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
4151 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
4152 for (i = 1; i < MAX_NR_ZONES; i++) {
4153 if (i == ZONE_MOVABLE)
4155 arch_zone_lowest_possible_pfn[i] =
4156 arch_zone_highest_possible_pfn[i-1];
4157 arch_zone_highest_possible_pfn[i] =
4158 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
4160 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
4161 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
4163 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
4164 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
4165 find_zone_movable_pfns_for_nodes(zone_movable_pfn);
4167 /* Print out the zone ranges */
4168 printk("Zone PFN ranges:\n");
4169 for (i = 0; i < MAX_NR_ZONES; i++) {
4170 if (i == ZONE_MOVABLE)
4172 printk(" %-8s %0#10lx -> %0#10lx\n",
4174 arch_zone_lowest_possible_pfn[i],
4175 arch_zone_highest_possible_pfn[i]);
4178 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
4179 printk("Movable zone start PFN for each node\n");
4180 for (i = 0; i < MAX_NUMNODES; i++) {
4181 if (zone_movable_pfn[i])
4182 printk(" Node %d: %lu\n", i, zone_movable_pfn[i]);
4185 /* Print out the early_node_map[] */
4186 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries);
4187 for (i = 0; i < nr_nodemap_entries; i++)
4188 printk(" %3d: %0#10lx -> %0#10lx\n", early_node_map[i].nid,
4189 early_node_map[i].start_pfn,
4190 early_node_map[i].end_pfn);
4192 /* Initialise every node */
4193 mminit_verify_pageflags_layout();
4194 setup_nr_node_ids();
4195 for_each_online_node(nid) {
4196 pg_data_t *pgdat = NODE_DATA(nid);
4197 free_area_init_node(nid, NULL,
4198 find_min_pfn_for_node(nid), NULL);
4200 /* Any memory on that node */
4201 if (pgdat->node_present_pages)
4202 node_set_state(nid, N_HIGH_MEMORY);
4203 check_for_regular_memory(pgdat);
4207 static int __init cmdline_parse_core(char *p, unsigned long *core)
4209 unsigned long long coremem;
4213 coremem = memparse(p, &p);
4214 *core = coremem >> PAGE_SHIFT;
4216 /* Paranoid check that UL is enough for the coremem value */
4217 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
4223 * kernelcore=size sets the amount of memory for use for allocations that
4224 * cannot be reclaimed or migrated.
4226 static int __init cmdline_parse_kernelcore(char *p)
4228 return cmdline_parse_core(p, &required_kernelcore);
4232 * movablecore=size sets the amount of memory for use for allocations that
4233 * can be reclaimed or migrated.
4235 static int __init cmdline_parse_movablecore(char *p)
4237 return cmdline_parse_core(p, &required_movablecore);
4240 early_param("kernelcore", cmdline_parse_kernelcore);
4241 early_param("movablecore", cmdline_parse_movablecore);
4243 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4246 * set_dma_reserve - set the specified number of pages reserved in the first zone
4247 * @new_dma_reserve: The number of pages to mark reserved
4249 * The per-cpu batchsize and zone watermarks are determined by present_pages.
4250 * In the DMA zone, a significant percentage may be consumed by kernel image
4251 * and other unfreeable allocations which can skew the watermarks badly. This
4252 * function may optionally be used to account for unfreeable pages in the
4253 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
4254 * smaller per-cpu batchsize.
4256 void __init set_dma_reserve(unsigned long new_dma_reserve)
4258 dma_reserve = new_dma_reserve;
4261 #ifndef CONFIG_NEED_MULTIPLE_NODES
4262 struct pglist_data __refdata contig_page_data = { .bdata = &bootmem_node_data[0] };
4263 EXPORT_SYMBOL(contig_page_data);
4266 void __init free_area_init(unsigned long *zones_size)
4268 free_area_init_node(0, zones_size,
4269 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
4272 static int page_alloc_cpu_notify(struct notifier_block *self,
4273 unsigned long action, void *hcpu)
4275 int cpu = (unsigned long)hcpu;
4277 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
4281 * Spill the event counters of the dead processor
4282 * into the current processors event counters.
4283 * This artificially elevates the count of the current
4286 vm_events_fold_cpu(cpu);
4289 * Zero the differential counters of the dead processor
4290 * so that the vm statistics are consistent.
4292 * This is only okay since the processor is dead and cannot
4293 * race with what we are doing.
4295 refresh_cpu_vm_stats(cpu);
4300 void __init page_alloc_init(void)
4302 hotcpu_notifier(page_alloc_cpu_notify, 0);
4306 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
4307 * or min_free_kbytes changes.
4309 static void calculate_totalreserve_pages(void)
4311 struct pglist_data *pgdat;
4312 unsigned long reserve_pages = 0;
4313 enum zone_type i, j;
4315 for_each_online_pgdat(pgdat) {
4316 for (i = 0; i < MAX_NR_ZONES; i++) {
4317 struct zone *zone = pgdat->node_zones + i;
4318 unsigned long max = 0;
4320 /* Find valid and maximum lowmem_reserve in the zone */
4321 for (j = i; j < MAX_NR_ZONES; j++) {
4322 if (zone->lowmem_reserve[j] > max)
4323 max = zone->lowmem_reserve[j];
4326 /* we treat the high watermark as reserved pages. */
4327 max += high_wmark_pages(zone);
4329 if (max > zone->present_pages)
4330 max = zone->present_pages;
4331 reserve_pages += max;
4334 totalreserve_pages = reserve_pages;
4338 * setup_per_zone_lowmem_reserve - called whenever
4339 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
4340 * has a correct pages reserved value, so an adequate number of
4341 * pages are left in the zone after a successful __alloc_pages().
4343 static void setup_per_zone_lowmem_reserve(void)
4345 struct pglist_data *pgdat;
4346 enum zone_type j, idx;
4348 for_each_online_pgdat(pgdat) {
4349 for (j = 0; j < MAX_NR_ZONES; j++) {
4350 struct zone *zone = pgdat->node_zones + j;
4351 unsigned long present_pages = zone->present_pages;
4353 zone->lowmem_reserve[j] = 0;
4357 struct zone *lower_zone;
4361 if (sysctl_lowmem_reserve_ratio[idx] < 1)
4362 sysctl_lowmem_reserve_ratio[idx] = 1;
4364 lower_zone = pgdat->node_zones + idx;
4365 lower_zone->lowmem_reserve[j] = present_pages /
4366 sysctl_lowmem_reserve_ratio[idx];
4367 present_pages += lower_zone->present_pages;
4372 /* update totalreserve_pages */
4373 calculate_totalreserve_pages();
4377 * setup_per_zone_pages_min - called when min_free_kbytes changes.
4379 * Ensures that the pages_{min,low,high} values for each zone are set correctly
4380 * with respect to min_free_kbytes.
4382 void setup_per_zone_pages_min(void)
4384 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
4385 unsigned long lowmem_pages = 0;
4387 unsigned long flags;
4389 /* Calculate total number of !ZONE_HIGHMEM pages */
4390 for_each_zone(zone) {
4391 if (!is_highmem(zone))
4392 lowmem_pages += zone->present_pages;
4395 for_each_zone(zone) {
4398 spin_lock_irqsave(&zone->lock, flags);
4399 tmp = (u64)pages_min * zone->present_pages;
4400 do_div(tmp, lowmem_pages);
4401 if (is_highmem(zone)) {
4403 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
4404 * need highmem pages, so cap pages_min to a small
4407 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
4408 * deltas controls asynch page reclaim, and so should
4409 * not be capped for highmem.
4413 min_pages = zone->present_pages / 1024;
4414 if (min_pages < SWAP_CLUSTER_MAX)
4415 min_pages = SWAP_CLUSTER_MAX;
4416 if (min_pages > 128)
4418 zone->watermark[WMARK_MIN] = min_pages;
4421 * If it's a lowmem zone, reserve a number of pages
4422 * proportionate to the zone's size.
4424 zone->watermark[WMARK_MIN] = tmp;
4427 zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + (tmp >> 2);
4428 zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1);
4429 setup_zone_migrate_reserve(zone);
4430 spin_unlock_irqrestore(&zone->lock, flags);
4433 /* update totalreserve_pages */
4434 calculate_totalreserve_pages();
4438 * setup_per_zone_inactive_ratio - called when min_free_kbytes changes.
4440 * The inactive anon list should be small enough that the VM never has to
4441 * do too much work, but large enough that each inactive page has a chance
4442 * to be referenced again before it is swapped out.
4444 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
4445 * INACTIVE_ANON pages on this zone's LRU, maintained by the
4446 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
4447 * the anonymous pages are kept on the inactive list.
4450 * memory ratio inactive anon
4451 * -------------------------------------
4460 static void setup_per_zone_inactive_ratio(void)
4464 for_each_zone(zone) {
4465 unsigned int gb, ratio;
4467 /* Zone size in gigabytes */
4468 gb = zone->present_pages >> (30 - PAGE_SHIFT);
4469 ratio = int_sqrt(10 * gb);
4473 zone->inactive_ratio = ratio;
4478 * Initialise min_free_kbytes.
4480 * For small machines we want it small (128k min). For large machines
4481 * we want it large (64MB max). But it is not linear, because network
4482 * bandwidth does not increase linearly with machine size. We use
4484 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
4485 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
4501 static int __init init_per_zone_pages_min(void)
4503 unsigned long lowmem_kbytes;
4505 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
4507 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
4508 if (min_free_kbytes < 128)
4509 min_free_kbytes = 128;
4510 if (min_free_kbytes > 65536)
4511 min_free_kbytes = 65536;
4512 setup_per_zone_pages_min();
4513 setup_per_zone_lowmem_reserve();
4514 setup_per_zone_inactive_ratio();
4517 module_init(init_per_zone_pages_min)
4520 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
4521 * that we can call two helper functions whenever min_free_kbytes
4524 int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
4525 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4527 proc_dointvec(table, write, file, buffer, length, ppos);
4529 setup_per_zone_pages_min();
4534 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
4535 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4540 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4545 zone->min_unmapped_pages = (zone->present_pages *
4546 sysctl_min_unmapped_ratio) / 100;
4550 int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
4551 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4556 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4561 zone->min_slab_pages = (zone->present_pages *
4562 sysctl_min_slab_ratio) / 100;
4568 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
4569 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
4570 * whenever sysctl_lowmem_reserve_ratio changes.
4572 * The reserve ratio obviously has absolutely no relation with the
4573 * minimum watermarks. The lowmem reserve ratio can only make sense
4574 * if in function of the boot time zone sizes.
4576 int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
4577 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4579 proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4580 setup_per_zone_lowmem_reserve();
4585 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
4586 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
4587 * can have before it gets flushed back to buddy allocator.
4590 int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
4591 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4597 ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4598 if (!write || (ret == -EINVAL))
4600 for_each_zone(zone) {
4601 for_each_online_cpu(cpu) {
4603 high = zone->present_pages / percpu_pagelist_fraction;
4604 setup_pagelist_highmark(zone_pcp(zone, cpu), high);
4610 int hashdist = HASHDIST_DEFAULT;
4613 static int __init set_hashdist(char *str)
4617 hashdist = simple_strtoul(str, &str, 0);
4620 __setup("hashdist=", set_hashdist);
4624 * allocate a large system hash table from bootmem
4625 * - it is assumed that the hash table must contain an exact power-of-2
4626 * quantity of entries
4627 * - limit is the number of hash buckets, not the total allocation size
4629 void *__init alloc_large_system_hash(const char *tablename,
4630 unsigned long bucketsize,
4631 unsigned long numentries,
4634 unsigned int *_hash_shift,
4635 unsigned int *_hash_mask,
4636 unsigned long limit)
4638 unsigned long long max = limit;
4639 unsigned long log2qty, size;
4642 /* allow the kernel cmdline to have a say */
4644 /* round applicable memory size up to nearest megabyte */
4645 numentries = nr_kernel_pages;
4646 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
4647 numentries >>= 20 - PAGE_SHIFT;
4648 numentries <<= 20 - PAGE_SHIFT;
4650 /* limit to 1 bucket per 2^scale bytes of low memory */
4651 if (scale > PAGE_SHIFT)
4652 numentries >>= (scale - PAGE_SHIFT);
4654 numentries <<= (PAGE_SHIFT - scale);
4656 /* Make sure we've got at least a 0-order allocation.. */
4657 if (unlikely((numentries * bucketsize) < PAGE_SIZE))
4658 numentries = PAGE_SIZE / bucketsize;
4660 numentries = roundup_pow_of_two(numentries);
4662 /* limit allocation size to 1/16 total memory by default */
4664 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
4665 do_div(max, bucketsize);
4668 if (numentries > max)
4671 log2qty = ilog2(numentries);
4674 size = bucketsize << log2qty;
4675 if (flags & HASH_EARLY)
4676 table = alloc_bootmem_nopanic(size);
4678 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
4680 unsigned long order = get_order(size);
4682 if (order < MAX_ORDER)
4683 table = (void *)__get_free_pages(GFP_ATOMIC,
4686 * If bucketsize is not a power-of-two, we may free
4687 * some pages at the end of hash table.
4690 unsigned long alloc_end = (unsigned long)table +
4691 (PAGE_SIZE << order);
4692 unsigned long used = (unsigned long)table +
4694 split_page(virt_to_page(table), order);
4695 while (used < alloc_end) {
4701 } while (!table && size > PAGE_SIZE && --log2qty);
4704 panic("Failed to allocate %s hash table\n", tablename);
4706 printk(KERN_INFO "%s hash table entries: %d (order: %d, %lu bytes)\n",
4709 ilog2(size) - PAGE_SHIFT,
4713 *_hash_shift = log2qty;
4715 *_hash_mask = (1 << log2qty) - 1;
4718 * If hashdist is set, the table allocation is done with __vmalloc()
4719 * which invokes the kmemleak_alloc() callback. This function may also
4720 * be called before the slab and kmemleak are initialised when
4721 * kmemleak simply buffers the request to be executed later
4722 * (GFP_ATOMIC flag ignored in this case).
4725 kmemleak_alloc(table, size, 1, GFP_ATOMIC);
4730 /* Return a pointer to the bitmap storing bits affecting a block of pages */
4731 static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
4734 #ifdef CONFIG_SPARSEMEM
4735 return __pfn_to_section(pfn)->pageblock_flags;
4737 return zone->pageblock_flags;
4738 #endif /* CONFIG_SPARSEMEM */
4741 static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
4743 #ifdef CONFIG_SPARSEMEM
4744 pfn &= (PAGES_PER_SECTION-1);
4745 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
4747 pfn = pfn - zone->zone_start_pfn;
4748 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
4749 #endif /* CONFIG_SPARSEMEM */
4753 * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
4754 * @page: The page within the block of interest
4755 * @start_bitidx: The first bit of interest to retrieve
4756 * @end_bitidx: The last bit of interest
4757 * returns pageblock_bits flags
4759 unsigned long get_pageblock_flags_group(struct page *page,
4760 int start_bitidx, int end_bitidx)
4763 unsigned long *bitmap;
4764 unsigned long pfn, bitidx;
4765 unsigned long flags = 0;
4766 unsigned long value = 1;
4768 zone = page_zone(page);
4769 pfn = page_to_pfn(page);
4770 bitmap = get_pageblock_bitmap(zone, pfn);
4771 bitidx = pfn_to_bitidx(zone, pfn);
4773 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
4774 if (test_bit(bitidx + start_bitidx, bitmap))
4781 * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
4782 * @page: The page within the block of interest
4783 * @start_bitidx: The first bit of interest
4784 * @end_bitidx: The last bit of interest
4785 * @flags: The flags to set
4787 void set_pageblock_flags_group(struct page *page, unsigned long flags,
4788 int start_bitidx, int end_bitidx)
4791 unsigned long *bitmap;
4792 unsigned long pfn, bitidx;
4793 unsigned long value = 1;
4795 zone = page_zone(page);
4796 pfn = page_to_pfn(page);
4797 bitmap = get_pageblock_bitmap(zone, pfn);
4798 bitidx = pfn_to_bitidx(zone, pfn);
4799 VM_BUG_ON(pfn < zone->zone_start_pfn);
4800 VM_BUG_ON(pfn >= zone->zone_start_pfn + zone->spanned_pages);
4802 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
4804 __set_bit(bitidx + start_bitidx, bitmap);
4806 __clear_bit(bitidx + start_bitidx, bitmap);
4810 * This is designed as sub function...plz see page_isolation.c also.
4811 * set/clear page block's type to be ISOLATE.
4812 * page allocater never alloc memory from ISOLATE block.
4815 int set_migratetype_isolate(struct page *page)
4818 unsigned long flags;
4821 zone = page_zone(page);
4822 spin_lock_irqsave(&zone->lock, flags);
4824 * In future, more migrate types will be able to be isolation target.
4826 if (get_pageblock_migratetype(page) != MIGRATE_MOVABLE)
4828 set_pageblock_migratetype(page, MIGRATE_ISOLATE);
4829 move_freepages_block(zone, page, MIGRATE_ISOLATE);
4832 spin_unlock_irqrestore(&zone->lock, flags);
4838 void unset_migratetype_isolate(struct page *page)
4841 unsigned long flags;
4842 zone = page_zone(page);
4843 spin_lock_irqsave(&zone->lock, flags);
4844 if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE)
4846 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
4847 move_freepages_block(zone, page, MIGRATE_MOVABLE);
4849 spin_unlock_irqrestore(&zone->lock, flags);
4852 #ifdef CONFIG_MEMORY_HOTREMOVE
4854 * All pages in the range must be isolated before calling this.
4857 __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
4863 unsigned long flags;
4864 /* find the first valid pfn */
4865 for (pfn = start_pfn; pfn < end_pfn; pfn++)
4870 zone = page_zone(pfn_to_page(pfn));
4871 spin_lock_irqsave(&zone->lock, flags);
4873 while (pfn < end_pfn) {
4874 if (!pfn_valid(pfn)) {
4878 page = pfn_to_page(pfn);
4879 BUG_ON(page_count(page));
4880 BUG_ON(!PageBuddy(page));
4881 order = page_order(page);
4882 #ifdef CONFIG_DEBUG_VM
4883 printk(KERN_INFO "remove from free list %lx %d %lx\n",
4884 pfn, 1 << order, end_pfn);
4886 list_del(&page->lru);
4887 rmv_page_order(page);
4888 zone->free_area[order].nr_free--;
4889 __mod_zone_page_state(zone, NR_FREE_PAGES,
4891 for (i = 0; i < (1 << order); i++)
4892 SetPageReserved((page+i));
4893 pfn += (1 << order);
4895 spin_unlock_irqrestore(&zone->lock, flags);