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 static int destroy_compound_page(struct page *page, unsigned long order)
306 int nr_pages = 1 << order;
309 if (unlikely(compound_order(page) != order) ||
310 unlikely(!PageHead(page))) {
315 __ClearPageHead(page);
317 for (i = 1; i < nr_pages; i++) {
318 struct page *p = page + i;
320 if (unlikely(!PageTail(p) || (p->first_page != page))) {
330 static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
335 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
336 * and __GFP_HIGHMEM from hard or soft interrupt context.
338 VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
339 for (i = 0; i < (1 << order); i++)
340 clear_highpage(page + i);
343 static inline void set_page_order(struct page *page, int order)
345 set_page_private(page, order);
346 __SetPageBuddy(page);
349 static inline void rmv_page_order(struct page *page)
351 __ClearPageBuddy(page);
352 set_page_private(page, 0);
356 * Locate the struct page for both the matching buddy in our
357 * pair (buddy1) and the combined O(n+1) page they form (page).
359 * 1) Any buddy B1 will have an order O twin B2 which satisfies
360 * the following equation:
362 * For example, if the starting buddy (buddy2) is #8 its order
364 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
366 * 2) Any buddy B will have an order O+1 parent P which
367 * satisfies the following equation:
370 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
372 static inline struct page *
373 __page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order)
375 unsigned long buddy_idx = page_idx ^ (1 << order);
377 return page + (buddy_idx - page_idx);
380 static inline unsigned long
381 __find_combined_index(unsigned long page_idx, unsigned int order)
383 return (page_idx & ~(1 << order));
387 * This function checks whether a page is free && is the buddy
388 * we can do coalesce a page and its buddy if
389 * (a) the buddy is not in a hole &&
390 * (b) the buddy is in the buddy system &&
391 * (c) a page and its buddy have the same order &&
392 * (d) a page and its buddy are in the same zone.
394 * For recording whether a page is in the buddy system, we use PG_buddy.
395 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
397 * For recording page's order, we use page_private(page).
399 static inline int page_is_buddy(struct page *page, struct page *buddy,
402 if (!pfn_valid_within(page_to_pfn(buddy)))
405 if (page_zone_id(page) != page_zone_id(buddy))
408 if (PageBuddy(buddy) && page_order(buddy) == order) {
409 VM_BUG_ON(page_count(buddy) != 0);
416 * Freeing function for a buddy system allocator.
418 * The concept of a buddy system is to maintain direct-mapped table
419 * (containing bit values) for memory blocks of various "orders".
420 * The bottom level table contains the map for the smallest allocatable
421 * units of memory (here, pages), and each level above it describes
422 * pairs of units from the levels below, hence, "buddies".
423 * At a high level, all that happens here is marking the table entry
424 * at the bottom level available, and propagating the changes upward
425 * as necessary, plus some accounting needed to play nicely with other
426 * parts of the VM system.
427 * At each level, we keep a list of pages, which are heads of continuous
428 * free pages of length of (1 << order) and marked with PG_buddy. Page's
429 * order is recorded in page_private(page) field.
430 * So when we are allocating or freeing one, we can derive the state of the
431 * other. That is, if we allocate a small block, and both were
432 * free, the remainder of the region must be split into blocks.
433 * If a block is freed, and its buddy is also free, then this
434 * triggers coalescing into a block of larger size.
439 static inline void __free_one_page(struct page *page,
440 struct zone *zone, unsigned int order,
443 unsigned long page_idx;
445 if (unlikely(PageCompound(page)))
446 if (unlikely(destroy_compound_page(page, order)))
449 VM_BUG_ON(migratetype == -1);
451 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
453 VM_BUG_ON(page_idx & ((1 << order) - 1));
454 VM_BUG_ON(bad_range(zone, page));
456 while (order < MAX_ORDER-1) {
457 unsigned long combined_idx;
460 buddy = __page_find_buddy(page, page_idx, order);
461 if (!page_is_buddy(page, buddy, order))
464 /* Our buddy is free, merge with it and move up one order. */
465 list_del(&buddy->lru);
466 zone->free_area[order].nr_free--;
467 rmv_page_order(buddy);
468 combined_idx = __find_combined_index(page_idx, order);
469 page = page + (combined_idx - page_idx);
470 page_idx = combined_idx;
473 set_page_order(page, order);
475 &zone->free_area[order].free_list[migratetype]);
476 zone->free_area[order].nr_free++;
479 #ifdef CONFIG_HAVE_MLOCKED_PAGE_BIT
481 * free_page_mlock() -- clean up attempts to free and mlocked() page.
482 * Page should not be on lru, so no need to fix that up.
483 * free_pages_check() will verify...
485 static inline void free_page_mlock(struct page *page)
487 __ClearPageMlocked(page);
488 __dec_zone_page_state(page, NR_MLOCK);
489 __count_vm_event(UNEVICTABLE_MLOCKFREED);
492 static void free_page_mlock(struct page *page) { }
495 static inline int free_pages_check(struct page *page)
497 if (unlikely(page_mapcount(page) |
498 (page->mapping != NULL) |
499 (atomic_read(&page->_count) != 0) |
500 (page->flags & PAGE_FLAGS_CHECK_AT_FREE))) {
504 if (page->flags & PAGE_FLAGS_CHECK_AT_PREP)
505 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
510 * Frees a list of pages.
511 * Assumes all pages on list are in same zone, and of same order.
512 * count is the number of pages to free.
514 * If the zone was previously in an "all pages pinned" state then look to
515 * see if this freeing clears that state.
517 * And clear the zone's pages_scanned counter, to hold off the "all pages are
518 * pinned" detection logic.
520 static void free_pages_bulk(struct zone *zone, int count,
521 struct list_head *list, int order)
523 spin_lock(&zone->lock);
524 zone_clear_flag(zone, ZONE_ALL_UNRECLAIMABLE);
525 zone->pages_scanned = 0;
527 __mod_zone_page_state(zone, NR_FREE_PAGES, count << order);
531 VM_BUG_ON(list_empty(list));
532 page = list_entry(list->prev, struct page, lru);
533 /* have to delete it as __free_one_page list manipulates */
534 list_del(&page->lru);
535 __free_one_page(page, zone, order, page_private(page));
537 spin_unlock(&zone->lock);
540 static void free_one_page(struct zone *zone, struct page *page, int order,
543 spin_lock(&zone->lock);
544 zone_clear_flag(zone, ZONE_ALL_UNRECLAIMABLE);
545 zone->pages_scanned = 0;
547 __mod_zone_page_state(zone, NR_FREE_PAGES, 1 << order);
548 __free_one_page(page, zone, order, migratetype);
549 spin_unlock(&zone->lock);
552 static void __free_pages_ok(struct page *page, unsigned int order)
557 int clearMlocked = PageMlocked(page);
559 for (i = 0 ; i < (1 << order) ; ++i)
560 bad += free_pages_check(page + i);
564 if (!PageHighMem(page)) {
565 debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order);
566 debug_check_no_obj_freed(page_address(page),
569 arch_free_page(page, order);
570 kernel_map_pages(page, 1 << order, 0);
572 local_irq_save(flags);
573 if (unlikely(clearMlocked))
574 free_page_mlock(page);
575 __count_vm_events(PGFREE, 1 << order);
576 free_one_page(page_zone(page), page, order,
577 get_pageblock_migratetype(page));
578 local_irq_restore(flags);
582 * permit the bootmem allocator to evade page validation on high-order frees
584 void __meminit __free_pages_bootmem(struct page *page, unsigned int order)
587 __ClearPageReserved(page);
588 set_page_count(page, 0);
589 set_page_refcounted(page);
595 for (loop = 0; loop < BITS_PER_LONG; loop++) {
596 struct page *p = &page[loop];
598 if (loop + 1 < BITS_PER_LONG)
600 __ClearPageReserved(p);
601 set_page_count(p, 0);
604 set_page_refcounted(page);
605 __free_pages(page, order);
611 * The order of subdivision here is critical for the IO subsystem.
612 * Please do not alter this order without good reasons and regression
613 * testing. Specifically, as large blocks of memory are subdivided,
614 * the order in which smaller blocks are delivered depends on the order
615 * they're subdivided in this function. This is the primary factor
616 * influencing the order in which pages are delivered to the IO
617 * subsystem according to empirical testing, and this is also justified
618 * by considering the behavior of a buddy system containing a single
619 * large block of memory acted on by a series of small allocations.
620 * This behavior is a critical factor in sglist merging's success.
624 static inline void expand(struct zone *zone, struct page *page,
625 int low, int high, struct free_area *area,
628 unsigned long size = 1 << high;
634 VM_BUG_ON(bad_range(zone, &page[size]));
635 list_add(&page[size].lru, &area->free_list[migratetype]);
637 set_page_order(&page[size], high);
642 * This page is about to be returned from the page allocator
644 static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
646 if (unlikely(page_mapcount(page) |
647 (page->mapping != NULL) |
648 (atomic_read(&page->_count) != 0) |
649 (page->flags & PAGE_FLAGS_CHECK_AT_PREP))) {
654 set_page_private(page, 0);
655 set_page_refcounted(page);
657 arch_alloc_page(page, order);
658 kernel_map_pages(page, 1 << order, 1);
660 if (gfp_flags & __GFP_ZERO)
661 prep_zero_page(page, order, gfp_flags);
663 if (order && (gfp_flags & __GFP_COMP))
664 prep_compound_page(page, order);
670 * Go through the free lists for the given migratetype and remove
671 * the smallest available page from the freelists
674 struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
677 unsigned int current_order;
678 struct free_area * area;
681 /* Find a page of the appropriate size in the preferred list */
682 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
683 area = &(zone->free_area[current_order]);
684 if (list_empty(&area->free_list[migratetype]))
687 page = list_entry(area->free_list[migratetype].next,
689 list_del(&page->lru);
690 rmv_page_order(page);
692 expand(zone, page, order, current_order, area, migratetype);
701 * This array describes the order lists are fallen back to when
702 * the free lists for the desirable migrate type are depleted
704 static int fallbacks[MIGRATE_TYPES][MIGRATE_TYPES-1] = {
705 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
706 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
707 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
708 [MIGRATE_RESERVE] = { MIGRATE_RESERVE, MIGRATE_RESERVE, MIGRATE_RESERVE }, /* Never used */
712 * Move the free pages in a range to the free lists of the requested type.
713 * Note that start_page and end_pages are not aligned on a pageblock
714 * boundary. If alignment is required, use move_freepages_block()
716 static int move_freepages(struct zone *zone,
717 struct page *start_page, struct page *end_page,
724 #ifndef CONFIG_HOLES_IN_ZONE
726 * page_zone is not safe to call in this context when
727 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
728 * anyway as we check zone boundaries in move_freepages_block().
729 * Remove at a later date when no bug reports exist related to
730 * grouping pages by mobility
732 BUG_ON(page_zone(start_page) != page_zone(end_page));
735 for (page = start_page; page <= end_page;) {
736 /* Make sure we are not inadvertently changing nodes */
737 VM_BUG_ON(page_to_nid(page) != zone_to_nid(zone));
739 if (!pfn_valid_within(page_to_pfn(page))) {
744 if (!PageBuddy(page)) {
749 order = page_order(page);
750 list_del(&page->lru);
752 &zone->free_area[order].free_list[migratetype]);
754 pages_moved += 1 << order;
760 static int move_freepages_block(struct zone *zone, struct page *page,
763 unsigned long start_pfn, end_pfn;
764 struct page *start_page, *end_page;
766 start_pfn = page_to_pfn(page);
767 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
768 start_page = pfn_to_page(start_pfn);
769 end_page = start_page + pageblock_nr_pages - 1;
770 end_pfn = start_pfn + pageblock_nr_pages - 1;
772 /* Do not cross zone boundaries */
773 if (start_pfn < zone->zone_start_pfn)
775 if (end_pfn >= zone->zone_start_pfn + zone->spanned_pages)
778 return move_freepages(zone, start_page, end_page, migratetype);
781 /* Remove an element from the buddy allocator from the fallback list */
782 static inline struct page *
783 __rmqueue_fallback(struct zone *zone, int order, int start_migratetype)
785 struct free_area * area;
790 /* Find the largest possible block of pages in the other list */
791 for (current_order = MAX_ORDER-1; current_order >= order;
793 for (i = 0; i < MIGRATE_TYPES - 1; i++) {
794 migratetype = fallbacks[start_migratetype][i];
796 /* MIGRATE_RESERVE handled later if necessary */
797 if (migratetype == MIGRATE_RESERVE)
800 area = &(zone->free_area[current_order]);
801 if (list_empty(&area->free_list[migratetype]))
804 page = list_entry(area->free_list[migratetype].next,
809 * If breaking a large block of pages, move all free
810 * pages to the preferred allocation list. If falling
811 * back for a reclaimable kernel allocation, be more
812 * agressive about taking ownership of free pages
814 if (unlikely(current_order >= (pageblock_order >> 1)) ||
815 start_migratetype == MIGRATE_RECLAIMABLE) {
817 pages = move_freepages_block(zone, page,
820 /* Claim the whole block if over half of it is free */
821 if (pages >= (1 << (pageblock_order-1)))
822 set_pageblock_migratetype(page,
825 migratetype = start_migratetype;
828 /* Remove the page from the freelists */
829 list_del(&page->lru);
830 rmv_page_order(page);
832 if (current_order == pageblock_order)
833 set_pageblock_migratetype(page,
836 expand(zone, page, order, current_order, area, migratetype);
845 * Do the hard work of removing an element from the buddy allocator.
846 * Call me with the zone->lock already held.
848 static struct page *__rmqueue(struct zone *zone, unsigned int order,
854 page = __rmqueue_smallest(zone, order, migratetype);
856 if (unlikely(!page) && migratetype != MIGRATE_RESERVE) {
857 page = __rmqueue_fallback(zone, order, migratetype);
860 * Use MIGRATE_RESERVE rather than fail an allocation. goto
861 * is used because __rmqueue_smallest is an inline function
862 * and we want just one call site
865 migratetype = MIGRATE_RESERVE;
874 * Obtain a specified number of elements from the buddy allocator, all under
875 * a single hold of the lock, for efficiency. Add them to the supplied list.
876 * Returns the number of new pages which were placed at *list.
878 static int rmqueue_bulk(struct zone *zone, unsigned int order,
879 unsigned long count, struct list_head *list,
884 spin_lock(&zone->lock);
885 for (i = 0; i < count; ++i) {
886 struct page *page = __rmqueue(zone, order, migratetype);
887 if (unlikely(page == NULL))
891 * Split buddy pages returned by expand() are received here
892 * in physical page order. The page is added to the callers and
893 * list and the list head then moves forward. From the callers
894 * perspective, the linked list is ordered by page number in
895 * some conditions. This is useful for IO devices that can
896 * merge IO requests if the physical pages are ordered
899 list_add(&page->lru, list);
900 set_page_private(page, migratetype);
903 __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
904 spin_unlock(&zone->lock);
910 * Called from the vmstat counter updater to drain pagesets of this
911 * currently executing processor on remote nodes after they have
914 * Note that this function must be called with the thread pinned to
915 * a single processor.
917 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
922 local_irq_save(flags);
923 if (pcp->count >= pcp->batch)
924 to_drain = pcp->batch;
926 to_drain = pcp->count;
927 free_pages_bulk(zone, to_drain, &pcp->list, 0);
928 pcp->count -= to_drain;
929 local_irq_restore(flags);
934 * Drain pages of the indicated processor.
936 * The processor must either be the current processor and the
937 * thread pinned to the current processor or a processor that
940 static void drain_pages(unsigned int cpu)
945 for_each_populated_zone(zone) {
946 struct per_cpu_pageset *pset;
947 struct per_cpu_pages *pcp;
949 pset = zone_pcp(zone, cpu);
952 local_irq_save(flags);
953 free_pages_bulk(zone, pcp->count, &pcp->list, 0);
955 local_irq_restore(flags);
960 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
962 void drain_local_pages(void *arg)
964 drain_pages(smp_processor_id());
968 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
970 void drain_all_pages(void)
972 on_each_cpu(drain_local_pages, NULL, 1);
975 #ifdef CONFIG_HIBERNATION
977 void mark_free_pages(struct zone *zone)
979 unsigned long pfn, max_zone_pfn;
982 struct list_head *curr;
984 if (!zone->spanned_pages)
987 spin_lock_irqsave(&zone->lock, flags);
989 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
990 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
991 if (pfn_valid(pfn)) {
992 struct page *page = pfn_to_page(pfn);
994 if (!swsusp_page_is_forbidden(page))
995 swsusp_unset_page_free(page);
998 for_each_migratetype_order(order, t) {
999 list_for_each(curr, &zone->free_area[order].free_list[t]) {
1002 pfn = page_to_pfn(list_entry(curr, struct page, lru));
1003 for (i = 0; i < (1UL << order); i++)
1004 swsusp_set_page_free(pfn_to_page(pfn + i));
1007 spin_unlock_irqrestore(&zone->lock, flags);
1009 #endif /* CONFIG_PM */
1012 * Free a 0-order page
1014 static void free_hot_cold_page(struct page *page, int cold)
1016 struct zone *zone = page_zone(page);
1017 struct per_cpu_pages *pcp;
1018 unsigned long flags;
1019 int clearMlocked = PageMlocked(page);
1022 page->mapping = NULL;
1023 if (free_pages_check(page))
1026 if (!PageHighMem(page)) {
1027 debug_check_no_locks_freed(page_address(page), PAGE_SIZE);
1028 debug_check_no_obj_freed(page_address(page), PAGE_SIZE);
1030 arch_free_page(page, 0);
1031 kernel_map_pages(page, 1, 0);
1033 pcp = &zone_pcp(zone, get_cpu())->pcp;
1034 set_page_private(page, get_pageblock_migratetype(page));
1035 local_irq_save(flags);
1036 if (unlikely(clearMlocked))
1037 free_page_mlock(page);
1038 __count_vm_event(PGFREE);
1041 list_add_tail(&page->lru, &pcp->list);
1043 list_add(&page->lru, &pcp->list);
1045 if (pcp->count >= pcp->high) {
1046 free_pages_bulk(zone, pcp->batch, &pcp->list, 0);
1047 pcp->count -= pcp->batch;
1049 local_irq_restore(flags);
1053 void free_hot_page(struct page *page)
1055 free_hot_cold_page(page, 0);
1058 void free_cold_page(struct page *page)
1060 free_hot_cold_page(page, 1);
1064 * split_page takes a non-compound higher-order page, and splits it into
1065 * n (1<<order) sub-pages: page[0..n]
1066 * Each sub-page must be freed individually.
1068 * Note: this is probably too low level an operation for use in drivers.
1069 * Please consult with lkml before using this in your driver.
1071 void split_page(struct page *page, unsigned int order)
1075 VM_BUG_ON(PageCompound(page));
1076 VM_BUG_ON(!page_count(page));
1077 for (i = 1; i < (1 << order); i++)
1078 set_page_refcounted(page + i);
1082 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1083 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1087 struct page *buffered_rmqueue(struct zone *preferred_zone,
1088 struct zone *zone, int order, gfp_t gfp_flags,
1091 unsigned long flags;
1093 int cold = !!(gfp_flags & __GFP_COLD);
1098 if (likely(order == 0)) {
1099 struct per_cpu_pages *pcp;
1101 pcp = &zone_pcp(zone, cpu)->pcp;
1102 local_irq_save(flags);
1104 pcp->count = rmqueue_bulk(zone, 0,
1105 pcp->batch, &pcp->list, migratetype);
1106 if (unlikely(!pcp->count))
1110 /* Find a page of the appropriate migrate type */
1112 list_for_each_entry_reverse(page, &pcp->list, lru)
1113 if (page_private(page) == migratetype)
1116 list_for_each_entry(page, &pcp->list, lru)
1117 if (page_private(page) == migratetype)
1121 /* Allocate more to the pcp list if necessary */
1122 if (unlikely(&page->lru == &pcp->list)) {
1123 pcp->count += rmqueue_bulk(zone, 0,
1124 pcp->batch, &pcp->list, migratetype);
1125 page = list_entry(pcp->list.next, struct page, lru);
1128 list_del(&page->lru);
1131 spin_lock_irqsave(&zone->lock, flags);
1132 page = __rmqueue(zone, order, migratetype);
1133 __mod_zone_page_state(zone, NR_FREE_PAGES, -(1 << order));
1134 spin_unlock(&zone->lock);
1139 __count_zone_vm_events(PGALLOC, zone, 1 << order);
1140 zone_statistics(preferred_zone, zone);
1141 local_irq_restore(flags);
1144 VM_BUG_ON(bad_range(zone, page));
1145 if (prep_new_page(page, order, gfp_flags))
1150 local_irq_restore(flags);
1155 /* The ALLOC_WMARK bits are used as an index to zone->watermark */
1156 #define ALLOC_WMARK_MIN WMARK_MIN
1157 #define ALLOC_WMARK_LOW WMARK_LOW
1158 #define ALLOC_WMARK_HIGH WMARK_HIGH
1159 #define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */
1161 /* Mask to get the watermark bits */
1162 #define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1)
1164 #define ALLOC_HARDER 0x10 /* try to alloc harder */
1165 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1166 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
1168 #ifdef CONFIG_FAIL_PAGE_ALLOC
1170 static struct fail_page_alloc_attr {
1171 struct fault_attr attr;
1173 u32 ignore_gfp_highmem;
1174 u32 ignore_gfp_wait;
1177 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1179 struct dentry *ignore_gfp_highmem_file;
1180 struct dentry *ignore_gfp_wait_file;
1181 struct dentry *min_order_file;
1183 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1185 } fail_page_alloc = {
1186 .attr = FAULT_ATTR_INITIALIZER,
1187 .ignore_gfp_wait = 1,
1188 .ignore_gfp_highmem = 1,
1192 static int __init setup_fail_page_alloc(char *str)
1194 return setup_fault_attr(&fail_page_alloc.attr, str);
1196 __setup("fail_page_alloc=", setup_fail_page_alloc);
1198 static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1200 if (order < fail_page_alloc.min_order)
1202 if (gfp_mask & __GFP_NOFAIL)
1204 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
1206 if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
1209 return should_fail(&fail_page_alloc.attr, 1 << order);
1212 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1214 static int __init fail_page_alloc_debugfs(void)
1216 mode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
1220 err = init_fault_attr_dentries(&fail_page_alloc.attr,
1224 dir = fail_page_alloc.attr.dentries.dir;
1226 fail_page_alloc.ignore_gfp_wait_file =
1227 debugfs_create_bool("ignore-gfp-wait", mode, dir,
1228 &fail_page_alloc.ignore_gfp_wait);
1230 fail_page_alloc.ignore_gfp_highmem_file =
1231 debugfs_create_bool("ignore-gfp-highmem", mode, dir,
1232 &fail_page_alloc.ignore_gfp_highmem);
1233 fail_page_alloc.min_order_file =
1234 debugfs_create_u32("min-order", mode, dir,
1235 &fail_page_alloc.min_order);
1237 if (!fail_page_alloc.ignore_gfp_wait_file ||
1238 !fail_page_alloc.ignore_gfp_highmem_file ||
1239 !fail_page_alloc.min_order_file) {
1241 debugfs_remove(fail_page_alloc.ignore_gfp_wait_file);
1242 debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file);
1243 debugfs_remove(fail_page_alloc.min_order_file);
1244 cleanup_fault_attr_dentries(&fail_page_alloc.attr);
1250 late_initcall(fail_page_alloc_debugfs);
1252 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1254 #else /* CONFIG_FAIL_PAGE_ALLOC */
1256 static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1261 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1264 * Return 1 if free pages are above 'mark'. This takes into account the order
1265 * of the allocation.
1267 int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1268 int classzone_idx, int alloc_flags)
1270 /* free_pages my go negative - that's OK */
1272 long free_pages = zone_page_state(z, NR_FREE_PAGES) - (1 << order) + 1;
1275 if (alloc_flags & ALLOC_HIGH)
1277 if (alloc_flags & ALLOC_HARDER)
1280 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
1282 for (o = 0; o < order; o++) {
1283 /* At the next order, this order's pages become unavailable */
1284 free_pages -= z->free_area[o].nr_free << o;
1286 /* Require fewer higher order pages to be free */
1289 if (free_pages <= min)
1297 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1298 * skip over zones that are not allowed by the cpuset, or that have
1299 * been recently (in last second) found to be nearly full. See further
1300 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1301 * that have to skip over a lot of full or unallowed zones.
1303 * If the zonelist cache is present in the passed in zonelist, then
1304 * returns a pointer to the allowed node mask (either the current
1305 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
1307 * If the zonelist cache is not available for this zonelist, does
1308 * nothing and returns NULL.
1310 * If the fullzones BITMAP in the zonelist cache is stale (more than
1311 * a second since last zap'd) then we zap it out (clear its bits.)
1313 * We hold off even calling zlc_setup, until after we've checked the
1314 * first zone in the zonelist, on the theory that most allocations will
1315 * be satisfied from that first zone, so best to examine that zone as
1316 * quickly as we can.
1318 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1320 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1321 nodemask_t *allowednodes; /* zonelist_cache approximation */
1323 zlc = zonelist->zlcache_ptr;
1327 if (time_after(jiffies, zlc->last_full_zap + HZ)) {
1328 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1329 zlc->last_full_zap = jiffies;
1332 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1333 &cpuset_current_mems_allowed :
1334 &node_states[N_HIGH_MEMORY];
1335 return allowednodes;
1339 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1340 * if it is worth looking at further for free memory:
1341 * 1) Check that the zone isn't thought to be full (doesn't have its
1342 * bit set in the zonelist_cache fullzones BITMAP).
1343 * 2) Check that the zones node (obtained from the zonelist_cache
1344 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1345 * Return true (non-zero) if zone is worth looking at further, or
1346 * else return false (zero) if it is not.
1348 * This check -ignores- the distinction between various watermarks,
1349 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1350 * found to be full for any variation of these watermarks, it will
1351 * be considered full for up to one second by all requests, unless
1352 * we are so low on memory on all allowed nodes that we are forced
1353 * into the second scan of the zonelist.
1355 * In the second scan we ignore this zonelist cache and exactly
1356 * apply the watermarks to all zones, even it is slower to do so.
1357 * We are low on memory in the second scan, and should leave no stone
1358 * unturned looking for a free page.
1360 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
1361 nodemask_t *allowednodes)
1363 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1364 int i; /* index of *z in zonelist zones */
1365 int n; /* node that zone *z is on */
1367 zlc = zonelist->zlcache_ptr;
1371 i = z - zonelist->_zonerefs;
1374 /* This zone is worth trying if it is allowed but not full */
1375 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1379 * Given 'z' scanning a zonelist, set the corresponding bit in
1380 * zlc->fullzones, so that subsequent attempts to allocate a page
1381 * from that zone don't waste time re-examining it.
1383 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
1385 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1386 int i; /* index of *z in zonelist zones */
1388 zlc = zonelist->zlcache_ptr;
1392 i = z - zonelist->_zonerefs;
1394 set_bit(i, zlc->fullzones);
1397 #else /* CONFIG_NUMA */
1399 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1404 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
1405 nodemask_t *allowednodes)
1410 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
1413 #endif /* CONFIG_NUMA */
1416 * get_page_from_freelist goes through the zonelist trying to allocate
1419 static struct page *
1420 get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order,
1421 struct zonelist *zonelist, int high_zoneidx, int alloc_flags,
1422 struct zone *preferred_zone, int migratetype)
1425 struct page *page = NULL;
1428 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1429 int zlc_active = 0; /* set if using zonelist_cache */
1430 int did_zlc_setup = 0; /* just call zlc_setup() one time */
1432 classzone_idx = zone_idx(preferred_zone);
1435 * Scan zonelist, looking for a zone with enough free.
1436 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1438 for_each_zone_zonelist_nodemask(zone, z, zonelist,
1439 high_zoneidx, nodemask) {
1440 if (NUMA_BUILD && zlc_active &&
1441 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1443 if ((alloc_flags & ALLOC_CPUSET) &&
1444 !cpuset_zone_allowed_softwall(zone, gfp_mask))
1447 BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
1448 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
1450 mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
1451 if (!zone_watermark_ok(zone, order, mark,
1452 classzone_idx, alloc_flags)) {
1453 if (!zone_reclaim_mode ||
1454 !zone_reclaim(zone, gfp_mask, order))
1455 goto this_zone_full;
1459 page = buffered_rmqueue(preferred_zone, zone, order,
1460 gfp_mask, migratetype);
1465 zlc_mark_zone_full(zonelist, z);
1467 if (NUMA_BUILD && !did_zlc_setup && nr_online_nodes > 1) {
1469 * we do zlc_setup after the first zone is tried but only
1470 * if there are multiple nodes make it worthwhile
1472 allowednodes = zlc_setup(zonelist, alloc_flags);
1478 if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) {
1479 /* Disable zlc cache for second zonelist scan */
1487 should_alloc_retry(gfp_t gfp_mask, unsigned int order,
1488 unsigned long pages_reclaimed)
1490 /* Do not loop if specifically requested */
1491 if (gfp_mask & __GFP_NORETRY)
1495 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
1496 * means __GFP_NOFAIL, but that may not be true in other
1499 if (order <= PAGE_ALLOC_COSTLY_ORDER)
1503 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
1504 * specified, then we retry until we no longer reclaim any pages
1505 * (above), or we've reclaimed an order of pages at least as
1506 * large as the allocation's order. In both cases, if the
1507 * allocation still fails, we stop retrying.
1509 if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order))
1513 * Don't let big-order allocations loop unless the caller
1514 * explicitly requests that.
1516 if (gfp_mask & __GFP_NOFAIL)
1522 static inline struct page *
1523 __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
1524 struct zonelist *zonelist, enum zone_type high_zoneidx,
1525 nodemask_t *nodemask, struct zone *preferred_zone,
1530 /* Acquire the OOM killer lock for the zones in zonelist */
1531 if (!try_set_zone_oom(zonelist, gfp_mask)) {
1532 schedule_timeout_uninterruptible(1);
1537 * Go through the zonelist yet one more time, keep very high watermark
1538 * here, this is only to catch a parallel oom killing, we must fail if
1539 * we're still under heavy pressure.
1541 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask,
1542 order, zonelist, high_zoneidx,
1543 ALLOC_WMARK_HIGH|ALLOC_CPUSET,
1544 preferred_zone, migratetype);
1548 /* The OOM killer will not help higher order allocs */
1549 if (order > PAGE_ALLOC_COSTLY_ORDER)
1552 /* Exhausted what can be done so it's blamo time */
1553 out_of_memory(zonelist, gfp_mask, order);
1556 clear_zonelist_oom(zonelist, gfp_mask);
1560 /* The really slow allocator path where we enter direct reclaim */
1561 static inline struct page *
1562 __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
1563 struct zonelist *zonelist, enum zone_type high_zoneidx,
1564 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
1565 int migratetype, unsigned long *did_some_progress)
1567 struct page *page = NULL;
1568 struct reclaim_state reclaim_state;
1569 struct task_struct *p = current;
1573 /* We now go into synchronous reclaim */
1574 cpuset_memory_pressure_bump();
1577 * The task's cpuset might have expanded its set of allowable nodes
1579 p->flags |= PF_MEMALLOC;
1580 lockdep_set_current_reclaim_state(gfp_mask);
1581 reclaim_state.reclaimed_slab = 0;
1582 p->reclaim_state = &reclaim_state;
1584 *did_some_progress = try_to_free_pages(zonelist, order, gfp_mask, nodemask);
1586 p->reclaim_state = NULL;
1587 lockdep_clear_current_reclaim_state();
1588 p->flags &= ~PF_MEMALLOC;
1595 if (likely(*did_some_progress))
1596 page = get_page_from_freelist(gfp_mask, nodemask, order,
1597 zonelist, high_zoneidx,
1598 alloc_flags, preferred_zone,
1604 * This is called in the allocator slow-path if the allocation request is of
1605 * sufficient urgency to ignore watermarks and take other desperate measures
1607 static inline struct page *
1608 __alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order,
1609 struct zonelist *zonelist, enum zone_type high_zoneidx,
1610 nodemask_t *nodemask, struct zone *preferred_zone,
1616 page = get_page_from_freelist(gfp_mask, nodemask, order,
1617 zonelist, high_zoneidx, ALLOC_NO_WATERMARKS,
1618 preferred_zone, migratetype);
1620 if (!page && gfp_mask & __GFP_NOFAIL)
1621 congestion_wait(WRITE, HZ/50);
1622 } while (!page && (gfp_mask & __GFP_NOFAIL));
1628 void wake_all_kswapd(unsigned int order, struct zonelist *zonelist,
1629 enum zone_type high_zoneidx)
1634 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx)
1635 wakeup_kswapd(zone, order);
1639 gfp_to_alloc_flags(gfp_t gfp_mask)
1641 struct task_struct *p = current;
1642 int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
1643 const gfp_t wait = gfp_mask & __GFP_WAIT;
1645 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
1646 BUILD_BUG_ON(__GFP_HIGH != ALLOC_HIGH);
1649 * The caller may dip into page reserves a bit more if the caller
1650 * cannot run direct reclaim, or if the caller has realtime scheduling
1651 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1652 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1654 alloc_flags |= (gfp_mask & __GFP_HIGH);
1657 alloc_flags |= ALLOC_HARDER;
1659 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1660 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1662 alloc_flags &= ~ALLOC_CPUSET;
1663 } else if (unlikely(rt_task(p)))
1664 alloc_flags |= ALLOC_HARDER;
1666 if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) {
1667 if (!in_interrupt() &&
1668 ((p->flags & PF_MEMALLOC) ||
1669 unlikely(test_thread_flag(TIF_MEMDIE))))
1670 alloc_flags |= ALLOC_NO_WATERMARKS;
1676 static inline struct page *
1677 __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
1678 struct zonelist *zonelist, enum zone_type high_zoneidx,
1679 nodemask_t *nodemask, struct zone *preferred_zone,
1682 const gfp_t wait = gfp_mask & __GFP_WAIT;
1683 struct page *page = NULL;
1685 unsigned long pages_reclaimed = 0;
1686 unsigned long did_some_progress;
1687 struct task_struct *p = current;
1690 * In the slowpath, we sanity check order to avoid ever trying to
1691 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
1692 * be using allocators in order of preference for an area that is
1695 if (WARN_ON_ONCE(order >= MAX_ORDER))
1699 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1700 * __GFP_NOWARN set) should not cause reclaim since the subsystem
1701 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1702 * using a larger set of nodes after it has established that the
1703 * allowed per node queues are empty and that nodes are
1706 if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
1709 wake_all_kswapd(order, zonelist, high_zoneidx);
1712 * OK, we're below the kswapd watermark and have kicked background
1713 * reclaim. Now things get more complex, so set up alloc_flags according
1714 * to how we want to proceed.
1716 alloc_flags = gfp_to_alloc_flags(gfp_mask);
1719 /* This is the last chance, in general, before the goto nopage. */
1720 page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist,
1721 high_zoneidx, alloc_flags & ~ALLOC_NO_WATERMARKS,
1722 preferred_zone, migratetype);
1727 /* Allocate without watermarks if the context allows */
1728 if (alloc_flags & ALLOC_NO_WATERMARKS) {
1729 page = __alloc_pages_high_priority(gfp_mask, order,
1730 zonelist, high_zoneidx, nodemask,
1731 preferred_zone, migratetype);
1736 /* Atomic allocations - we can't balance anything */
1740 /* Avoid recursion of direct reclaim */
1741 if (p->flags & PF_MEMALLOC)
1744 /* Try direct reclaim and then allocating */
1745 page = __alloc_pages_direct_reclaim(gfp_mask, order,
1746 zonelist, high_zoneidx,
1748 alloc_flags, preferred_zone,
1749 migratetype, &did_some_progress);
1754 * If we failed to make any progress reclaiming, then we are
1755 * running out of options and have to consider going OOM
1757 if (!did_some_progress) {
1758 if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
1759 page = __alloc_pages_may_oom(gfp_mask, order,
1760 zonelist, high_zoneidx,
1761 nodemask, preferred_zone,
1767 * The OOM killer does not trigger for high-order allocations
1768 * but if no progress is being made, there are no other
1769 * options and retrying is unlikely to help
1771 if (order > PAGE_ALLOC_COSTLY_ORDER)
1778 /* Check if we should retry the allocation */
1779 pages_reclaimed += did_some_progress;
1780 if (should_alloc_retry(gfp_mask, order, pages_reclaimed)) {
1781 /* Wait for some write requests to complete then retry */
1782 congestion_wait(WRITE, HZ/50);
1787 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
1788 printk(KERN_WARNING "%s: page allocation failure."
1789 " order:%d, mode:0x%x\n",
1790 p->comm, order, gfp_mask);
1800 * This is the 'heart' of the zoned buddy allocator.
1803 __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
1804 struct zonelist *zonelist, nodemask_t *nodemask)
1806 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
1807 struct zone *preferred_zone;
1809 int migratetype = allocflags_to_migratetype(gfp_mask);
1811 lockdep_trace_alloc(gfp_mask);
1813 might_sleep_if(gfp_mask & __GFP_WAIT);
1815 if (should_fail_alloc_page(gfp_mask, order))
1819 * Check the zones suitable for the gfp_mask contain at least one
1820 * valid zone. It's possible to have an empty zonelist as a result
1821 * of GFP_THISNODE and a memoryless node
1823 if (unlikely(!zonelist->_zonerefs->zone))
1826 /* The preferred zone is used for statistics later */
1827 first_zones_zonelist(zonelist, high_zoneidx, nodemask, &preferred_zone);
1828 if (!preferred_zone)
1831 /* First allocation attempt */
1832 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order,
1833 zonelist, high_zoneidx, ALLOC_WMARK_LOW|ALLOC_CPUSET,
1834 preferred_zone, migratetype);
1835 if (unlikely(!page))
1836 page = __alloc_pages_slowpath(gfp_mask, order,
1837 zonelist, high_zoneidx, nodemask,
1838 preferred_zone, migratetype);
1842 EXPORT_SYMBOL(__alloc_pages_nodemask);
1845 * Common helper functions.
1847 unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
1850 page = alloc_pages(gfp_mask, order);
1853 return (unsigned long) page_address(page);
1856 EXPORT_SYMBOL(__get_free_pages);
1858 unsigned long get_zeroed_page(gfp_t gfp_mask)
1863 * get_zeroed_page() returns a 32-bit address, which cannot represent
1866 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
1868 page = alloc_pages(gfp_mask | __GFP_ZERO, 0);
1870 return (unsigned long) page_address(page);
1874 EXPORT_SYMBOL(get_zeroed_page);
1876 void __pagevec_free(struct pagevec *pvec)
1878 int i = pagevec_count(pvec);
1881 free_hot_cold_page(pvec->pages[i], pvec->cold);
1884 void __free_pages(struct page *page, unsigned int order)
1886 if (put_page_testzero(page)) {
1888 free_hot_page(page);
1890 __free_pages_ok(page, order);
1894 EXPORT_SYMBOL(__free_pages);
1896 void free_pages(unsigned long addr, unsigned int order)
1899 VM_BUG_ON(!virt_addr_valid((void *)addr));
1900 __free_pages(virt_to_page((void *)addr), order);
1904 EXPORT_SYMBOL(free_pages);
1907 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
1908 * @size: the number of bytes to allocate
1909 * @gfp_mask: GFP flags for the allocation
1911 * This function is similar to alloc_pages(), except that it allocates the
1912 * minimum number of pages to satisfy the request. alloc_pages() can only
1913 * allocate memory in power-of-two pages.
1915 * This function is also limited by MAX_ORDER.
1917 * Memory allocated by this function must be released by free_pages_exact().
1919 void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
1921 unsigned int order = get_order(size);
1924 addr = __get_free_pages(gfp_mask, order);
1926 unsigned long alloc_end = addr + (PAGE_SIZE << order);
1927 unsigned long used = addr + PAGE_ALIGN(size);
1929 split_page(virt_to_page(addr), order);
1930 while (used < alloc_end) {
1936 return (void *)addr;
1938 EXPORT_SYMBOL(alloc_pages_exact);
1941 * free_pages_exact - release memory allocated via alloc_pages_exact()
1942 * @virt: the value returned by alloc_pages_exact.
1943 * @size: size of allocation, same value as passed to alloc_pages_exact().
1945 * Release the memory allocated by a previous call to alloc_pages_exact.
1947 void free_pages_exact(void *virt, size_t size)
1949 unsigned long addr = (unsigned long)virt;
1950 unsigned long end = addr + PAGE_ALIGN(size);
1952 while (addr < end) {
1957 EXPORT_SYMBOL(free_pages_exact);
1959 static unsigned int nr_free_zone_pages(int offset)
1964 /* Just pick one node, since fallback list is circular */
1965 unsigned int sum = 0;
1967 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
1969 for_each_zone_zonelist(zone, z, zonelist, offset) {
1970 unsigned long size = zone->present_pages;
1971 unsigned long high = high_wmark_pages(zone);
1980 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1982 unsigned int nr_free_buffer_pages(void)
1984 return nr_free_zone_pages(gfp_zone(GFP_USER));
1986 EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
1989 * Amount of free RAM allocatable within all zones
1991 unsigned int nr_free_pagecache_pages(void)
1993 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
1996 static inline void show_node(struct zone *zone)
1999 printk("Node %d ", zone_to_nid(zone));
2002 void si_meminfo(struct sysinfo *val)
2004 val->totalram = totalram_pages;
2006 val->freeram = global_page_state(NR_FREE_PAGES);
2007 val->bufferram = nr_blockdev_pages();
2008 val->totalhigh = totalhigh_pages;
2009 val->freehigh = nr_free_highpages();
2010 val->mem_unit = PAGE_SIZE;
2013 EXPORT_SYMBOL(si_meminfo);
2016 void si_meminfo_node(struct sysinfo *val, int nid)
2018 pg_data_t *pgdat = NODE_DATA(nid);
2020 val->totalram = pgdat->node_present_pages;
2021 val->freeram = node_page_state(nid, NR_FREE_PAGES);
2022 #ifdef CONFIG_HIGHMEM
2023 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
2024 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
2030 val->mem_unit = PAGE_SIZE;
2034 #define K(x) ((x) << (PAGE_SHIFT-10))
2037 * Show free area list (used inside shift_scroll-lock stuff)
2038 * We also calculate the percentage fragmentation. We do this by counting the
2039 * memory on each free list with the exception of the first item on the list.
2041 void show_free_areas(void)
2046 for_each_populated_zone(zone) {
2048 printk("%s per-cpu:\n", zone->name);
2050 for_each_online_cpu(cpu) {
2051 struct per_cpu_pageset *pageset;
2053 pageset = zone_pcp(zone, cpu);
2055 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
2056 cpu, pageset->pcp.high,
2057 pageset->pcp.batch, pageset->pcp.count);
2061 printk("Active_anon:%lu active_file:%lu inactive_anon:%lu\n"
2062 " inactive_file:%lu"
2063 //TODO: check/adjust line lengths
2064 #ifdef CONFIG_UNEVICTABLE_LRU
2067 " dirty:%lu writeback:%lu unstable:%lu\n"
2068 " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
2069 global_page_state(NR_ACTIVE_ANON),
2070 global_page_state(NR_ACTIVE_FILE),
2071 global_page_state(NR_INACTIVE_ANON),
2072 global_page_state(NR_INACTIVE_FILE),
2073 #ifdef CONFIG_UNEVICTABLE_LRU
2074 global_page_state(NR_UNEVICTABLE),
2076 global_page_state(NR_FILE_DIRTY),
2077 global_page_state(NR_WRITEBACK),
2078 global_page_state(NR_UNSTABLE_NFS),
2079 global_page_state(NR_FREE_PAGES),
2080 global_page_state(NR_SLAB_RECLAIMABLE) +
2081 global_page_state(NR_SLAB_UNRECLAIMABLE),
2082 global_page_state(NR_FILE_MAPPED),
2083 global_page_state(NR_PAGETABLE),
2084 global_page_state(NR_BOUNCE));
2086 for_each_populated_zone(zone) {
2095 " active_anon:%lukB"
2096 " inactive_anon:%lukB"
2097 " active_file:%lukB"
2098 " inactive_file:%lukB"
2099 #ifdef CONFIG_UNEVICTABLE_LRU
2100 " unevictable:%lukB"
2103 " pages_scanned:%lu"
2104 " all_unreclaimable? %s"
2107 K(zone_page_state(zone, NR_FREE_PAGES)),
2108 K(min_wmark_pages(zone)),
2109 K(low_wmark_pages(zone)),
2110 K(high_wmark_pages(zone)),
2111 K(zone_page_state(zone, NR_ACTIVE_ANON)),
2112 K(zone_page_state(zone, NR_INACTIVE_ANON)),
2113 K(zone_page_state(zone, NR_ACTIVE_FILE)),
2114 K(zone_page_state(zone, NR_INACTIVE_FILE)),
2115 #ifdef CONFIG_UNEVICTABLE_LRU
2116 K(zone_page_state(zone, NR_UNEVICTABLE)),
2118 K(zone->present_pages),
2119 zone->pages_scanned,
2120 (zone_is_all_unreclaimable(zone) ? "yes" : "no")
2122 printk("lowmem_reserve[]:");
2123 for (i = 0; i < MAX_NR_ZONES; i++)
2124 printk(" %lu", zone->lowmem_reserve[i]);
2128 for_each_populated_zone(zone) {
2129 unsigned long nr[MAX_ORDER], flags, order, total = 0;
2132 printk("%s: ", zone->name);
2134 spin_lock_irqsave(&zone->lock, flags);
2135 for (order = 0; order < MAX_ORDER; order++) {
2136 nr[order] = zone->free_area[order].nr_free;
2137 total += nr[order] << order;
2139 spin_unlock_irqrestore(&zone->lock, flags);
2140 for (order = 0; order < MAX_ORDER; order++)
2141 printk("%lu*%lukB ", nr[order], K(1UL) << order);
2142 printk("= %lukB\n", K(total));
2145 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
2147 show_swap_cache_info();
2150 static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
2152 zoneref->zone = zone;
2153 zoneref->zone_idx = zone_idx(zone);
2157 * Builds allocation fallback zone lists.
2159 * Add all populated zones of a node to the zonelist.
2161 static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
2162 int nr_zones, enum zone_type zone_type)
2166 BUG_ON(zone_type >= MAX_NR_ZONES);
2171 zone = pgdat->node_zones + zone_type;
2172 if (populated_zone(zone)) {
2173 zoneref_set_zone(zone,
2174 &zonelist->_zonerefs[nr_zones++]);
2175 check_highest_zone(zone_type);
2178 } while (zone_type);
2185 * 0 = automatic detection of better ordering.
2186 * 1 = order by ([node] distance, -zonetype)
2187 * 2 = order by (-zonetype, [node] distance)
2189 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
2190 * the same zonelist. So only NUMA can configure this param.
2192 #define ZONELIST_ORDER_DEFAULT 0
2193 #define ZONELIST_ORDER_NODE 1
2194 #define ZONELIST_ORDER_ZONE 2
2196 /* zonelist order in the kernel.
2197 * set_zonelist_order() will set this to NODE or ZONE.
2199 static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
2200 static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
2204 /* The value user specified ....changed by config */
2205 static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2206 /* string for sysctl */
2207 #define NUMA_ZONELIST_ORDER_LEN 16
2208 char numa_zonelist_order[16] = "default";
2211 * interface for configure zonelist ordering.
2212 * command line option "numa_zonelist_order"
2213 * = "[dD]efault - default, automatic configuration.
2214 * = "[nN]ode - order by node locality, then by zone within node
2215 * = "[zZ]one - order by zone, then by locality within zone
2218 static int __parse_numa_zonelist_order(char *s)
2220 if (*s == 'd' || *s == 'D') {
2221 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2222 } else if (*s == 'n' || *s == 'N') {
2223 user_zonelist_order = ZONELIST_ORDER_NODE;
2224 } else if (*s == 'z' || *s == 'Z') {
2225 user_zonelist_order = ZONELIST_ORDER_ZONE;
2228 "Ignoring invalid numa_zonelist_order value: "
2235 static __init int setup_numa_zonelist_order(char *s)
2238 return __parse_numa_zonelist_order(s);
2241 early_param("numa_zonelist_order", setup_numa_zonelist_order);
2244 * sysctl handler for numa_zonelist_order
2246 int numa_zonelist_order_handler(ctl_table *table, int write,
2247 struct file *file, void __user *buffer, size_t *length,
2250 char saved_string[NUMA_ZONELIST_ORDER_LEN];
2254 strncpy(saved_string, (char*)table->data,
2255 NUMA_ZONELIST_ORDER_LEN);
2256 ret = proc_dostring(table, write, file, buffer, length, ppos);
2260 int oldval = user_zonelist_order;
2261 if (__parse_numa_zonelist_order((char*)table->data)) {
2263 * bogus value. restore saved string
2265 strncpy((char*)table->data, saved_string,
2266 NUMA_ZONELIST_ORDER_LEN);
2267 user_zonelist_order = oldval;
2268 } else if (oldval != user_zonelist_order)
2269 build_all_zonelists();
2275 #define MAX_NODE_LOAD (nr_online_nodes)
2276 static int node_load[MAX_NUMNODES];
2279 * find_next_best_node - find the next node that should appear in a given node's fallback list
2280 * @node: node whose fallback list we're appending
2281 * @used_node_mask: nodemask_t of already used nodes
2283 * We use a number of factors to determine which is the next node that should
2284 * appear on a given node's fallback list. The node should not have appeared
2285 * already in @node's fallback list, and it should be the next closest node
2286 * according to the distance array (which contains arbitrary distance values
2287 * from each node to each node in the system), and should also prefer nodes
2288 * with no CPUs, since presumably they'll have very little allocation pressure
2289 * on them otherwise.
2290 * It returns -1 if no node is found.
2292 static int find_next_best_node(int node, nodemask_t *used_node_mask)
2295 int min_val = INT_MAX;
2297 const struct cpumask *tmp = cpumask_of_node(0);
2299 /* Use the local node if we haven't already */
2300 if (!node_isset(node, *used_node_mask)) {
2301 node_set(node, *used_node_mask);
2305 for_each_node_state(n, N_HIGH_MEMORY) {
2307 /* Don't want a node to appear more than once */
2308 if (node_isset(n, *used_node_mask))
2311 /* Use the distance array to find the distance */
2312 val = node_distance(node, n);
2314 /* Penalize nodes under us ("prefer the next node") */
2317 /* Give preference to headless and unused nodes */
2318 tmp = cpumask_of_node(n);
2319 if (!cpumask_empty(tmp))
2320 val += PENALTY_FOR_NODE_WITH_CPUS;
2322 /* Slight preference for less loaded node */
2323 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
2324 val += node_load[n];
2326 if (val < min_val) {
2333 node_set(best_node, *used_node_mask);
2340 * Build zonelists ordered by node and zones within node.
2341 * This results in maximum locality--normal zone overflows into local
2342 * DMA zone, if any--but risks exhausting DMA zone.
2344 static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
2347 struct zonelist *zonelist;
2349 zonelist = &pgdat->node_zonelists[0];
2350 for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
2352 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2354 zonelist->_zonerefs[j].zone = NULL;
2355 zonelist->_zonerefs[j].zone_idx = 0;
2359 * Build gfp_thisnode zonelists
2361 static void build_thisnode_zonelists(pg_data_t *pgdat)
2364 struct zonelist *zonelist;
2366 zonelist = &pgdat->node_zonelists[1];
2367 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
2368 zonelist->_zonerefs[j].zone = NULL;
2369 zonelist->_zonerefs[j].zone_idx = 0;
2373 * Build zonelists ordered by zone and nodes within zones.
2374 * This results in conserving DMA zone[s] until all Normal memory is
2375 * exhausted, but results in overflowing to remote node while memory
2376 * may still exist in local DMA zone.
2378 static int node_order[MAX_NUMNODES];
2380 static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
2383 int zone_type; /* needs to be signed */
2385 struct zonelist *zonelist;
2387 zonelist = &pgdat->node_zonelists[0];
2389 for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
2390 for (j = 0; j < nr_nodes; j++) {
2391 node = node_order[j];
2392 z = &NODE_DATA(node)->node_zones[zone_type];
2393 if (populated_zone(z)) {
2395 &zonelist->_zonerefs[pos++]);
2396 check_highest_zone(zone_type);
2400 zonelist->_zonerefs[pos].zone = NULL;
2401 zonelist->_zonerefs[pos].zone_idx = 0;
2404 static int default_zonelist_order(void)
2407 unsigned long low_kmem_size,total_size;
2411 * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem.
2412 * If they are really small and used heavily, the system can fall
2413 * into OOM very easily.
2414 * This function detect ZONE_DMA/DMA32 size and confgigures zone order.
2416 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2419 for_each_online_node(nid) {
2420 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2421 z = &NODE_DATA(nid)->node_zones[zone_type];
2422 if (populated_zone(z)) {
2423 if (zone_type < ZONE_NORMAL)
2424 low_kmem_size += z->present_pages;
2425 total_size += z->present_pages;
2429 if (!low_kmem_size || /* there are no DMA area. */
2430 low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
2431 return ZONELIST_ORDER_NODE;
2433 * look into each node's config.
2434 * If there is a node whose DMA/DMA32 memory is very big area on
2435 * local memory, NODE_ORDER may be suitable.
2437 average_size = total_size /
2438 (nodes_weight(node_states[N_HIGH_MEMORY]) + 1);
2439 for_each_online_node(nid) {
2442 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2443 z = &NODE_DATA(nid)->node_zones[zone_type];
2444 if (populated_zone(z)) {
2445 if (zone_type < ZONE_NORMAL)
2446 low_kmem_size += z->present_pages;
2447 total_size += z->present_pages;
2450 if (low_kmem_size &&
2451 total_size > average_size && /* ignore small node */
2452 low_kmem_size > total_size * 70/100)
2453 return ZONELIST_ORDER_NODE;
2455 return ZONELIST_ORDER_ZONE;
2458 static void set_zonelist_order(void)
2460 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
2461 current_zonelist_order = default_zonelist_order();
2463 current_zonelist_order = user_zonelist_order;
2466 static void build_zonelists(pg_data_t *pgdat)
2470 nodemask_t used_mask;
2471 int local_node, prev_node;
2472 struct zonelist *zonelist;
2473 int order = current_zonelist_order;
2475 /* initialize zonelists */
2476 for (i = 0; i < MAX_ZONELISTS; i++) {
2477 zonelist = pgdat->node_zonelists + i;
2478 zonelist->_zonerefs[0].zone = NULL;
2479 zonelist->_zonerefs[0].zone_idx = 0;
2482 /* NUMA-aware ordering of nodes */
2483 local_node = pgdat->node_id;
2484 load = nr_online_nodes;
2485 prev_node = local_node;
2486 nodes_clear(used_mask);
2488 memset(node_load, 0, sizeof(node_load));
2489 memset(node_order, 0, sizeof(node_order));
2492 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
2493 int distance = node_distance(local_node, node);
2496 * If another node is sufficiently far away then it is better
2497 * to reclaim pages in a zone before going off node.
2499 if (distance > RECLAIM_DISTANCE)
2500 zone_reclaim_mode = 1;
2503 * We don't want to pressure a particular node.
2504 * So adding penalty to the first node in same
2505 * distance group to make it round-robin.
2507 if (distance != node_distance(local_node, prev_node))
2508 node_load[node] = load;
2512 if (order == ZONELIST_ORDER_NODE)
2513 build_zonelists_in_node_order(pgdat, node);
2515 node_order[j++] = node; /* remember order */
2518 if (order == ZONELIST_ORDER_ZONE) {
2519 /* calculate node order -- i.e., DMA last! */
2520 build_zonelists_in_zone_order(pgdat, j);
2523 build_thisnode_zonelists(pgdat);
2526 /* Construct the zonelist performance cache - see further mmzone.h */
2527 static void build_zonelist_cache(pg_data_t *pgdat)
2529 struct zonelist *zonelist;
2530 struct zonelist_cache *zlc;
2533 zonelist = &pgdat->node_zonelists[0];
2534 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
2535 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
2536 for (z = zonelist->_zonerefs; z->zone; z++)
2537 zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z);
2541 #else /* CONFIG_NUMA */
2543 static void set_zonelist_order(void)
2545 current_zonelist_order = ZONELIST_ORDER_ZONE;
2548 static void build_zonelists(pg_data_t *pgdat)
2550 int node, local_node;
2552 struct zonelist *zonelist;
2554 local_node = pgdat->node_id;
2556 zonelist = &pgdat->node_zonelists[0];
2557 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
2560 * Now we build the zonelist so that it contains the zones
2561 * of all the other nodes.
2562 * We don't want to pressure a particular node, so when
2563 * building the zones for node N, we make sure that the
2564 * zones coming right after the local ones are those from
2565 * node N+1 (modulo N)
2567 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
2568 if (!node_online(node))
2570 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2573 for (node = 0; node < local_node; node++) {
2574 if (!node_online(node))
2576 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2580 zonelist->_zonerefs[j].zone = NULL;
2581 zonelist->_zonerefs[j].zone_idx = 0;
2584 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
2585 static void build_zonelist_cache(pg_data_t *pgdat)
2587 pgdat->node_zonelists[0].zlcache_ptr = NULL;
2590 #endif /* CONFIG_NUMA */
2592 /* return values int ....just for stop_machine() */
2593 static int __build_all_zonelists(void *dummy)
2597 for_each_online_node(nid) {
2598 pg_data_t *pgdat = NODE_DATA(nid);
2600 build_zonelists(pgdat);
2601 build_zonelist_cache(pgdat);
2606 void build_all_zonelists(void)
2608 set_zonelist_order();
2610 if (system_state == SYSTEM_BOOTING) {
2611 __build_all_zonelists(NULL);
2612 mminit_verify_zonelist();
2613 cpuset_init_current_mems_allowed();
2615 /* we have to stop all cpus to guarantee there is no user
2617 stop_machine(__build_all_zonelists, NULL, NULL);
2618 /* cpuset refresh routine should be here */
2620 vm_total_pages = nr_free_pagecache_pages();
2622 * Disable grouping by mobility if the number of pages in the
2623 * system is too low to allow the mechanism to work. It would be
2624 * more accurate, but expensive to check per-zone. This check is
2625 * made on memory-hotadd so a system can start with mobility
2626 * disabled and enable it later
2628 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
2629 page_group_by_mobility_disabled = 1;
2631 page_group_by_mobility_disabled = 0;
2633 printk("Built %i zonelists in %s order, mobility grouping %s. "
2634 "Total pages: %ld\n",
2636 zonelist_order_name[current_zonelist_order],
2637 page_group_by_mobility_disabled ? "off" : "on",
2640 printk("Policy zone: %s\n", zone_names[policy_zone]);
2645 * Helper functions to size the waitqueue hash table.
2646 * Essentially these want to choose hash table sizes sufficiently
2647 * large so that collisions trying to wait on pages are rare.
2648 * But in fact, the number of active page waitqueues on typical
2649 * systems is ridiculously low, less than 200. So this is even
2650 * conservative, even though it seems large.
2652 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
2653 * waitqueues, i.e. the size of the waitq table given the number of pages.
2655 #define PAGES_PER_WAITQUEUE 256
2657 #ifndef CONFIG_MEMORY_HOTPLUG
2658 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
2660 unsigned long size = 1;
2662 pages /= PAGES_PER_WAITQUEUE;
2664 while (size < pages)
2668 * Once we have dozens or even hundreds of threads sleeping
2669 * on IO we've got bigger problems than wait queue collision.
2670 * Limit the size of the wait table to a reasonable size.
2672 size = min(size, 4096UL);
2674 return max(size, 4UL);
2678 * A zone's size might be changed by hot-add, so it is not possible to determine
2679 * a suitable size for its wait_table. So we use the maximum size now.
2681 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
2683 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
2684 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
2685 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
2687 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
2688 * or more by the traditional way. (See above). It equals:
2690 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
2691 * ia64(16K page size) : = ( 8G + 4M)byte.
2692 * powerpc (64K page size) : = (32G +16M)byte.
2694 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
2701 * This is an integer logarithm so that shifts can be used later
2702 * to extract the more random high bits from the multiplicative
2703 * hash function before the remainder is taken.
2705 static inline unsigned long wait_table_bits(unsigned long size)
2710 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
2713 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
2714 * of blocks reserved is based on min_wmark_pages(zone). The memory within
2715 * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
2716 * higher will lead to a bigger reserve which will get freed as contiguous
2717 * blocks as reclaim kicks in
2719 static void setup_zone_migrate_reserve(struct zone *zone)
2721 unsigned long start_pfn, pfn, end_pfn;
2723 unsigned long reserve, block_migratetype;
2725 /* Get the start pfn, end pfn and the number of blocks to reserve */
2726 start_pfn = zone->zone_start_pfn;
2727 end_pfn = start_pfn + zone->spanned_pages;
2728 reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >>
2731 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
2732 if (!pfn_valid(pfn))
2734 page = pfn_to_page(pfn);
2736 /* Watch out for overlapping nodes */
2737 if (page_to_nid(page) != zone_to_nid(zone))
2740 /* Blocks with reserved pages will never free, skip them. */
2741 if (PageReserved(page))
2744 block_migratetype = get_pageblock_migratetype(page);
2746 /* If this block is reserved, account for it */
2747 if (reserve > 0 && block_migratetype == MIGRATE_RESERVE) {
2752 /* Suitable for reserving if this block is movable */
2753 if (reserve > 0 && block_migratetype == MIGRATE_MOVABLE) {
2754 set_pageblock_migratetype(page, MIGRATE_RESERVE);
2755 move_freepages_block(zone, page, MIGRATE_RESERVE);
2761 * If the reserve is met and this is a previous reserved block,
2764 if (block_migratetype == MIGRATE_RESERVE) {
2765 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
2766 move_freepages_block(zone, page, MIGRATE_MOVABLE);
2772 * Initially all pages are reserved - free ones are freed
2773 * up by free_all_bootmem() once the early boot process is
2774 * done. Non-atomic initialization, single-pass.
2776 void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
2777 unsigned long start_pfn, enum memmap_context context)
2780 unsigned long end_pfn = start_pfn + size;
2784 if (highest_memmap_pfn < end_pfn - 1)
2785 highest_memmap_pfn = end_pfn - 1;
2787 z = &NODE_DATA(nid)->node_zones[zone];
2788 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
2790 * There can be holes in boot-time mem_map[]s
2791 * handed to this function. They do not
2792 * exist on hotplugged memory.
2794 if (context == MEMMAP_EARLY) {
2795 if (!early_pfn_valid(pfn))
2797 if (!early_pfn_in_nid(pfn, nid))
2800 page = pfn_to_page(pfn);
2801 set_page_links(page, zone, nid, pfn);
2802 mminit_verify_page_links(page, zone, nid, pfn);
2803 init_page_count(page);
2804 reset_page_mapcount(page);
2805 SetPageReserved(page);
2807 * Mark the block movable so that blocks are reserved for
2808 * movable at startup. This will force kernel allocations
2809 * to reserve their blocks rather than leaking throughout
2810 * the address space during boot when many long-lived
2811 * kernel allocations are made. Later some blocks near
2812 * the start are marked MIGRATE_RESERVE by
2813 * setup_zone_migrate_reserve()
2815 * bitmap is created for zone's valid pfn range. but memmap
2816 * can be created for invalid pages (for alignment)
2817 * check here not to call set_pageblock_migratetype() against
2820 if ((z->zone_start_pfn <= pfn)
2821 && (pfn < z->zone_start_pfn + z->spanned_pages)
2822 && !(pfn & (pageblock_nr_pages - 1)))
2823 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
2825 INIT_LIST_HEAD(&page->lru);
2826 #ifdef WANT_PAGE_VIRTUAL
2827 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
2828 if (!is_highmem_idx(zone))
2829 set_page_address(page, __va(pfn << PAGE_SHIFT));
2834 static void __meminit zone_init_free_lists(struct zone *zone)
2837 for_each_migratetype_order(order, t) {
2838 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
2839 zone->free_area[order].nr_free = 0;
2843 #ifndef __HAVE_ARCH_MEMMAP_INIT
2844 #define memmap_init(size, nid, zone, start_pfn) \
2845 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
2848 static int zone_batchsize(struct zone *zone)
2854 * The per-cpu-pages pools are set to around 1000th of the
2855 * size of the zone. But no more than 1/2 of a meg.
2857 * OK, so we don't know how big the cache is. So guess.
2859 batch = zone->present_pages / 1024;
2860 if (batch * PAGE_SIZE > 512 * 1024)
2861 batch = (512 * 1024) / PAGE_SIZE;
2862 batch /= 4; /* We effectively *= 4 below */
2867 * Clamp the batch to a 2^n - 1 value. Having a power
2868 * of 2 value was found to be more likely to have
2869 * suboptimal cache aliasing properties in some cases.
2871 * For example if 2 tasks are alternately allocating
2872 * batches of pages, one task can end up with a lot
2873 * of pages of one half of the possible page colors
2874 * and the other with pages of the other colors.
2876 batch = rounddown_pow_of_two(batch + batch/2) - 1;
2881 /* The deferral and batching of frees should be suppressed under NOMMU
2884 * The problem is that NOMMU needs to be able to allocate large chunks
2885 * of contiguous memory as there's no hardware page translation to
2886 * assemble apparent contiguous memory from discontiguous pages.
2888 * Queueing large contiguous runs of pages for batching, however,
2889 * causes the pages to actually be freed in smaller chunks. As there
2890 * can be a significant delay between the individual batches being
2891 * recycled, this leads to the once large chunks of space being
2892 * fragmented and becoming unavailable for high-order allocations.
2898 static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
2900 struct per_cpu_pages *pcp;
2902 memset(p, 0, sizeof(*p));
2906 pcp->high = 6 * batch;
2907 pcp->batch = max(1UL, 1 * batch);
2908 INIT_LIST_HEAD(&pcp->list);
2912 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
2913 * to the value high for the pageset p.
2916 static void setup_pagelist_highmark(struct per_cpu_pageset *p,
2919 struct per_cpu_pages *pcp;
2923 pcp->batch = max(1UL, high/4);
2924 if ((high/4) > (PAGE_SHIFT * 8))
2925 pcp->batch = PAGE_SHIFT * 8;
2931 * Boot pageset table. One per cpu which is going to be used for all
2932 * zones and all nodes. The parameters will be set in such a way
2933 * that an item put on a list will immediately be handed over to
2934 * the buddy list. This is safe since pageset manipulation is done
2935 * with interrupts disabled.
2937 * Some NUMA counter updates may also be caught by the boot pagesets.
2939 * The boot_pagesets must be kept even after bootup is complete for
2940 * unused processors and/or zones. They do play a role for bootstrapping
2941 * hotplugged processors.
2943 * zoneinfo_show() and maybe other functions do
2944 * not check if the processor is online before following the pageset pointer.
2945 * Other parts of the kernel may not check if the zone is available.
2947 static struct per_cpu_pageset boot_pageset[NR_CPUS];
2950 * Dynamically allocate memory for the
2951 * per cpu pageset array in struct zone.
2953 static int __cpuinit process_zones(int cpu)
2955 struct zone *zone, *dzone;
2956 int node = cpu_to_node(cpu);
2958 node_set_state(node, N_CPU); /* this node has a cpu */
2960 for_each_populated_zone(zone) {
2961 zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset),
2963 if (!zone_pcp(zone, cpu))
2966 setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone));
2968 if (percpu_pagelist_fraction)
2969 setup_pagelist_highmark(zone_pcp(zone, cpu),
2970 (zone->present_pages / percpu_pagelist_fraction));
2975 for_each_zone(dzone) {
2976 if (!populated_zone(dzone))
2980 kfree(zone_pcp(dzone, cpu));
2981 zone_pcp(dzone, cpu) = NULL;
2986 static inline void free_zone_pagesets(int cpu)
2990 for_each_zone(zone) {
2991 struct per_cpu_pageset *pset = zone_pcp(zone, cpu);
2993 /* Free per_cpu_pageset if it is slab allocated */
2994 if (pset != &boot_pageset[cpu])
2996 zone_pcp(zone, cpu) = NULL;
3000 static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb,
3001 unsigned long action,
3004 int cpu = (long)hcpu;
3005 int ret = NOTIFY_OK;
3008 case CPU_UP_PREPARE:
3009 case CPU_UP_PREPARE_FROZEN:
3010 if (process_zones(cpu))
3013 case CPU_UP_CANCELED:
3014 case CPU_UP_CANCELED_FROZEN:
3016 case CPU_DEAD_FROZEN:
3017 free_zone_pagesets(cpu);
3025 static struct notifier_block __cpuinitdata pageset_notifier =
3026 { &pageset_cpuup_callback, NULL, 0 };
3028 void __init setup_per_cpu_pageset(void)
3032 /* Initialize per_cpu_pageset for cpu 0.
3033 * A cpuup callback will do this for every cpu
3034 * as it comes online
3036 err = process_zones(smp_processor_id());
3038 register_cpu_notifier(&pageset_notifier);
3043 static noinline __init_refok
3044 int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
3047 struct pglist_data *pgdat = zone->zone_pgdat;
3051 * The per-page waitqueue mechanism uses hashed waitqueues
3054 zone->wait_table_hash_nr_entries =
3055 wait_table_hash_nr_entries(zone_size_pages);
3056 zone->wait_table_bits =
3057 wait_table_bits(zone->wait_table_hash_nr_entries);
3058 alloc_size = zone->wait_table_hash_nr_entries
3059 * sizeof(wait_queue_head_t);
3061 if (!slab_is_available()) {
3062 zone->wait_table = (wait_queue_head_t *)
3063 alloc_bootmem_node(pgdat, alloc_size);
3066 * This case means that a zone whose size was 0 gets new memory
3067 * via memory hot-add.
3068 * But it may be the case that a new node was hot-added. In
3069 * this case vmalloc() will not be able to use this new node's
3070 * memory - this wait_table must be initialized to use this new
3071 * node itself as well.
3072 * To use this new node's memory, further consideration will be
3075 zone->wait_table = vmalloc(alloc_size);
3077 if (!zone->wait_table)
3080 for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
3081 init_waitqueue_head(zone->wait_table + i);
3086 static __meminit void zone_pcp_init(struct zone *zone)
3089 unsigned long batch = zone_batchsize(zone);
3091 for (cpu = 0; cpu < NR_CPUS; cpu++) {
3093 /* Early boot. Slab allocator not functional yet */
3094 zone_pcp(zone, cpu) = &boot_pageset[cpu];
3095 setup_pageset(&boot_pageset[cpu],0);
3097 setup_pageset(zone_pcp(zone,cpu), batch);
3100 if (zone->present_pages)
3101 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n",
3102 zone->name, zone->present_pages, batch);
3105 __meminit int init_currently_empty_zone(struct zone *zone,
3106 unsigned long zone_start_pfn,
3108 enum memmap_context context)
3110 struct pglist_data *pgdat = zone->zone_pgdat;
3112 ret = zone_wait_table_init(zone, size);
3115 pgdat->nr_zones = zone_idx(zone) + 1;
3117 zone->zone_start_pfn = zone_start_pfn;
3119 mminit_dprintk(MMINIT_TRACE, "memmap_init",
3120 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
3122 (unsigned long)zone_idx(zone),
3123 zone_start_pfn, (zone_start_pfn + size));
3125 zone_init_free_lists(zone);
3130 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3132 * Basic iterator support. Return the first range of PFNs for a node
3133 * Note: nid == MAX_NUMNODES returns first region regardless of node
3135 static int __meminit first_active_region_index_in_nid(int nid)
3139 for (i = 0; i < nr_nodemap_entries; i++)
3140 if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
3147 * Basic iterator support. Return the next active range of PFNs for a node
3148 * Note: nid == MAX_NUMNODES returns next region regardless of node
3150 static int __meminit next_active_region_index_in_nid(int index, int nid)
3152 for (index = index + 1; index < nr_nodemap_entries; index++)
3153 if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
3159 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
3161 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
3162 * Architectures may implement their own version but if add_active_range()
3163 * was used and there are no special requirements, this is a convenient
3166 int __meminit __early_pfn_to_nid(unsigned long pfn)
3170 for (i = 0; i < nr_nodemap_entries; i++) {
3171 unsigned long start_pfn = early_node_map[i].start_pfn;
3172 unsigned long end_pfn = early_node_map[i].end_pfn;
3174 if (start_pfn <= pfn && pfn < end_pfn)
3175 return early_node_map[i].nid;
3177 /* This is a memory hole */
3180 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
3182 int __meminit early_pfn_to_nid(unsigned long pfn)
3186 nid = __early_pfn_to_nid(pfn);
3189 /* just returns 0 */
3193 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
3194 bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
3198 nid = __early_pfn_to_nid(pfn);
3199 if (nid >= 0 && nid != node)
3205 /* Basic iterator support to walk early_node_map[] */
3206 #define for_each_active_range_index_in_nid(i, nid) \
3207 for (i = first_active_region_index_in_nid(nid); i != -1; \
3208 i = next_active_region_index_in_nid(i, nid))
3211 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
3212 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
3213 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
3215 * If an architecture guarantees that all ranges registered with
3216 * add_active_ranges() contain no holes and may be freed, this
3217 * this function may be used instead of calling free_bootmem() manually.
3219 void __init free_bootmem_with_active_regions(int nid,
3220 unsigned long max_low_pfn)
3224 for_each_active_range_index_in_nid(i, nid) {
3225 unsigned long size_pages = 0;
3226 unsigned long end_pfn = early_node_map[i].end_pfn;
3228 if (early_node_map[i].start_pfn >= max_low_pfn)
3231 if (end_pfn > max_low_pfn)
3232 end_pfn = max_low_pfn;
3234 size_pages = end_pfn - early_node_map[i].start_pfn;
3235 free_bootmem_node(NODE_DATA(early_node_map[i].nid),
3236 PFN_PHYS(early_node_map[i].start_pfn),
3237 size_pages << PAGE_SHIFT);
3241 void __init work_with_active_regions(int nid, work_fn_t work_fn, void *data)
3246 for_each_active_range_index_in_nid(i, nid) {
3247 ret = work_fn(early_node_map[i].start_pfn,
3248 early_node_map[i].end_pfn, data);
3254 * sparse_memory_present_with_active_regions - Call memory_present for each active range
3255 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
3257 * If an architecture guarantees that all ranges registered with
3258 * add_active_ranges() contain no holes and may be freed, this
3259 * function may be used instead of calling memory_present() manually.
3261 void __init sparse_memory_present_with_active_regions(int nid)
3265 for_each_active_range_index_in_nid(i, nid)
3266 memory_present(early_node_map[i].nid,
3267 early_node_map[i].start_pfn,
3268 early_node_map[i].end_pfn);
3272 * get_pfn_range_for_nid - Return the start and end page frames for a node
3273 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
3274 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
3275 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
3277 * It returns the start and end page frame of a node based on information
3278 * provided by an arch calling add_active_range(). If called for a node
3279 * with no available memory, a warning is printed and the start and end
3282 void __meminit get_pfn_range_for_nid(unsigned int nid,
3283 unsigned long *start_pfn, unsigned long *end_pfn)
3289 for_each_active_range_index_in_nid(i, nid) {
3290 *start_pfn = min(*start_pfn, early_node_map[i].start_pfn);
3291 *end_pfn = max(*end_pfn, early_node_map[i].end_pfn);
3294 if (*start_pfn == -1UL)
3299 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3300 * assumption is made that zones within a node are ordered in monotonic
3301 * increasing memory addresses so that the "highest" populated zone is used
3303 static void __init find_usable_zone_for_movable(void)
3306 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
3307 if (zone_index == ZONE_MOVABLE)
3310 if (arch_zone_highest_possible_pfn[zone_index] >
3311 arch_zone_lowest_possible_pfn[zone_index])
3315 VM_BUG_ON(zone_index == -1);
3316 movable_zone = zone_index;
3320 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
3321 * because it is sized independant of architecture. Unlike the other zones,
3322 * the starting point for ZONE_MOVABLE is not fixed. It may be different
3323 * in each node depending on the size of each node and how evenly kernelcore
3324 * is distributed. This helper function adjusts the zone ranges
3325 * provided by the architecture for a given node by using the end of the
3326 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3327 * zones within a node are in order of monotonic increases memory addresses
3329 static void __meminit adjust_zone_range_for_zone_movable(int nid,
3330 unsigned long zone_type,
3331 unsigned long node_start_pfn,
3332 unsigned long node_end_pfn,
3333 unsigned long *zone_start_pfn,
3334 unsigned long *zone_end_pfn)
3336 /* Only adjust if ZONE_MOVABLE is on this node */
3337 if (zone_movable_pfn[nid]) {
3338 /* Size ZONE_MOVABLE */
3339 if (zone_type == ZONE_MOVABLE) {
3340 *zone_start_pfn = zone_movable_pfn[nid];
3341 *zone_end_pfn = min(node_end_pfn,
3342 arch_zone_highest_possible_pfn[movable_zone]);
3344 /* Adjust for ZONE_MOVABLE starting within this range */
3345 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
3346 *zone_end_pfn > zone_movable_pfn[nid]) {
3347 *zone_end_pfn = zone_movable_pfn[nid];
3349 /* Check if this whole range is within ZONE_MOVABLE */
3350 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
3351 *zone_start_pfn = *zone_end_pfn;
3356 * Return the number of pages a zone spans in a node, including holes
3357 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3359 static unsigned long __meminit zone_spanned_pages_in_node(int nid,
3360 unsigned long zone_type,
3361 unsigned long *ignored)
3363 unsigned long node_start_pfn, node_end_pfn;
3364 unsigned long zone_start_pfn, zone_end_pfn;
3366 /* Get the start and end of the node and zone */
3367 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3368 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
3369 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
3370 adjust_zone_range_for_zone_movable(nid, zone_type,
3371 node_start_pfn, node_end_pfn,
3372 &zone_start_pfn, &zone_end_pfn);
3374 /* Check that this node has pages within the zone's required range */
3375 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
3378 /* Move the zone boundaries inside the node if necessary */
3379 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
3380 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
3382 /* Return the spanned pages */
3383 return zone_end_pfn - zone_start_pfn;
3387 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
3388 * then all holes in the requested range will be accounted for.
3390 static unsigned long __meminit __absent_pages_in_range(int nid,
3391 unsigned long range_start_pfn,
3392 unsigned long range_end_pfn)
3395 unsigned long prev_end_pfn = 0, hole_pages = 0;
3396 unsigned long start_pfn;
3398 /* Find the end_pfn of the first active range of pfns in the node */
3399 i = first_active_region_index_in_nid(nid);
3403 prev_end_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3405 /* Account for ranges before physical memory on this node */
3406 if (early_node_map[i].start_pfn > range_start_pfn)
3407 hole_pages = prev_end_pfn - range_start_pfn;
3409 /* Find all holes for the zone within the node */
3410 for (; i != -1; i = next_active_region_index_in_nid(i, nid)) {
3412 /* No need to continue if prev_end_pfn is outside the zone */
3413 if (prev_end_pfn >= range_end_pfn)
3416 /* Make sure the end of the zone is not within the hole */
3417 start_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3418 prev_end_pfn = max(prev_end_pfn, range_start_pfn);
3420 /* Update the hole size cound and move on */
3421 if (start_pfn > range_start_pfn) {
3422 BUG_ON(prev_end_pfn > start_pfn);
3423 hole_pages += start_pfn - prev_end_pfn;
3425 prev_end_pfn = early_node_map[i].end_pfn;
3428 /* Account for ranges past physical memory on this node */
3429 if (range_end_pfn > prev_end_pfn)
3430 hole_pages += range_end_pfn -
3431 max(range_start_pfn, prev_end_pfn);
3437 * absent_pages_in_range - Return number of page frames in holes within a range
3438 * @start_pfn: The start PFN to start searching for holes
3439 * @end_pfn: The end PFN to stop searching for holes
3441 * It returns the number of pages frames in memory holes within a range.
3443 unsigned long __init absent_pages_in_range(unsigned long start_pfn,
3444 unsigned long end_pfn)
3446 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
3449 /* Return the number of page frames in holes in a zone on a node */
3450 static unsigned long __meminit zone_absent_pages_in_node(int nid,
3451 unsigned long zone_type,
3452 unsigned long *ignored)
3454 unsigned long node_start_pfn, node_end_pfn;
3455 unsigned long zone_start_pfn, zone_end_pfn;
3457 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3458 zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type],
3460 zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type],
3463 adjust_zone_range_for_zone_movable(nid, zone_type,
3464 node_start_pfn, node_end_pfn,
3465 &zone_start_pfn, &zone_end_pfn);
3466 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
3470 static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
3471 unsigned long zone_type,
3472 unsigned long *zones_size)
3474 return zones_size[zone_type];
3477 static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
3478 unsigned long zone_type,
3479 unsigned long *zholes_size)
3484 return zholes_size[zone_type];
3489 static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
3490 unsigned long *zones_size, unsigned long *zholes_size)
3492 unsigned long realtotalpages, totalpages = 0;
3495 for (i = 0; i < MAX_NR_ZONES; i++)
3496 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
3498 pgdat->node_spanned_pages = totalpages;
3500 realtotalpages = totalpages;
3501 for (i = 0; i < MAX_NR_ZONES; i++)
3503 zone_absent_pages_in_node(pgdat->node_id, i,
3505 pgdat->node_present_pages = realtotalpages;
3506 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
3510 #ifndef CONFIG_SPARSEMEM
3512 * Calculate the size of the zone->blockflags rounded to an unsigned long
3513 * Start by making sure zonesize is a multiple of pageblock_order by rounding
3514 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
3515 * round what is now in bits to nearest long in bits, then return it in
3518 static unsigned long __init usemap_size(unsigned long zonesize)
3520 unsigned long usemapsize;
3522 usemapsize = roundup(zonesize, pageblock_nr_pages);
3523 usemapsize = usemapsize >> pageblock_order;
3524 usemapsize *= NR_PAGEBLOCK_BITS;
3525 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
3527 return usemapsize / 8;
3530 static void __init setup_usemap(struct pglist_data *pgdat,
3531 struct zone *zone, unsigned long zonesize)
3533 unsigned long usemapsize = usemap_size(zonesize);
3534 zone->pageblock_flags = NULL;
3536 zone->pageblock_flags = alloc_bootmem_node(pgdat, usemapsize);
3539 static void inline setup_usemap(struct pglist_data *pgdat,
3540 struct zone *zone, unsigned long zonesize) {}
3541 #endif /* CONFIG_SPARSEMEM */
3543 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
3545 /* Return a sensible default order for the pageblock size. */
3546 static inline int pageblock_default_order(void)
3548 if (HPAGE_SHIFT > PAGE_SHIFT)
3549 return HUGETLB_PAGE_ORDER;
3554 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
3555 static inline void __init set_pageblock_order(unsigned int order)
3557 /* Check that pageblock_nr_pages has not already been setup */
3558 if (pageblock_order)
3562 * Assume the largest contiguous order of interest is a huge page.
3563 * This value may be variable depending on boot parameters on IA64
3565 pageblock_order = order;
3567 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3570 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
3571 * and pageblock_default_order() are unused as pageblock_order is set
3572 * at compile-time. See include/linux/pageblock-flags.h for the values of
3573 * pageblock_order based on the kernel config
3575 static inline int pageblock_default_order(unsigned int order)
3579 #define set_pageblock_order(x) do {} while (0)
3581 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3584 * Set up the zone data structures:
3585 * - mark all pages reserved
3586 * - mark all memory queues empty
3587 * - clear the memory bitmaps
3589 static void __paginginit free_area_init_core(struct pglist_data *pgdat,
3590 unsigned long *zones_size, unsigned long *zholes_size)
3593 int nid = pgdat->node_id;
3594 unsigned long zone_start_pfn = pgdat->node_start_pfn;
3597 pgdat_resize_init(pgdat);
3598 pgdat->nr_zones = 0;
3599 init_waitqueue_head(&pgdat->kswapd_wait);
3600 pgdat->kswapd_max_order = 0;
3601 pgdat_page_cgroup_init(pgdat);
3603 for (j = 0; j < MAX_NR_ZONES; j++) {
3604 struct zone *zone = pgdat->node_zones + j;
3605 unsigned long size, realsize, memmap_pages;
3608 size = zone_spanned_pages_in_node(nid, j, zones_size);
3609 realsize = size - zone_absent_pages_in_node(nid, j,
3613 * Adjust realsize so that it accounts for how much memory
3614 * is used by this zone for memmap. This affects the watermark
3615 * and per-cpu initialisations
3618 PAGE_ALIGN(size * sizeof(struct page)) >> PAGE_SHIFT;
3619 if (realsize >= memmap_pages) {
3620 realsize -= memmap_pages;
3623 " %s zone: %lu pages used for memmap\n",
3624 zone_names[j], memmap_pages);
3627 " %s zone: %lu pages exceeds realsize %lu\n",
3628 zone_names[j], memmap_pages, realsize);
3630 /* Account for reserved pages */
3631 if (j == 0 && realsize > dma_reserve) {
3632 realsize -= dma_reserve;
3633 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
3634 zone_names[0], dma_reserve);
3637 if (!is_highmem_idx(j))
3638 nr_kernel_pages += realsize;
3639 nr_all_pages += realsize;
3641 zone->spanned_pages = size;
3642 zone->present_pages = realsize;
3645 zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
3647 zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100;
3649 zone->name = zone_names[j];
3650 spin_lock_init(&zone->lock);
3651 spin_lock_init(&zone->lru_lock);
3652 zone_seqlock_init(zone);
3653 zone->zone_pgdat = pgdat;
3655 zone->prev_priority = DEF_PRIORITY;
3657 zone_pcp_init(zone);
3659 INIT_LIST_HEAD(&zone->lru[l].list);
3660 zone->lru[l].nr_saved_scan = 0;
3662 zone->reclaim_stat.recent_rotated[0] = 0;
3663 zone->reclaim_stat.recent_rotated[1] = 0;
3664 zone->reclaim_stat.recent_scanned[0] = 0;
3665 zone->reclaim_stat.recent_scanned[1] = 0;
3666 zap_zone_vm_stats(zone);
3671 set_pageblock_order(pageblock_default_order());
3672 setup_usemap(pgdat, zone, size);
3673 ret = init_currently_empty_zone(zone, zone_start_pfn,
3674 size, MEMMAP_EARLY);
3676 memmap_init(size, nid, j, zone_start_pfn);
3677 zone_start_pfn += size;
3681 static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
3683 /* Skip empty nodes */
3684 if (!pgdat->node_spanned_pages)
3687 #ifdef CONFIG_FLAT_NODE_MEM_MAP
3688 /* ia64 gets its own node_mem_map, before this, without bootmem */
3689 if (!pgdat->node_mem_map) {
3690 unsigned long size, start, end;
3694 * The zone's endpoints aren't required to be MAX_ORDER
3695 * aligned but the node_mem_map endpoints must be in order
3696 * for the buddy allocator to function correctly.
3698 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
3699 end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
3700 end = ALIGN(end, MAX_ORDER_NR_PAGES);
3701 size = (end - start) * sizeof(struct page);
3702 map = alloc_remap(pgdat->node_id, size);
3704 map = alloc_bootmem_node(pgdat, size);
3705 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
3707 #ifndef CONFIG_NEED_MULTIPLE_NODES
3709 * With no DISCONTIG, the global mem_map is just set as node 0's
3711 if (pgdat == NODE_DATA(0)) {
3712 mem_map = NODE_DATA(0)->node_mem_map;
3713 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3714 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
3715 mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
3716 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3719 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
3722 void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
3723 unsigned long node_start_pfn, unsigned long *zholes_size)
3725 pg_data_t *pgdat = NODE_DATA(nid);
3727 pgdat->node_id = nid;
3728 pgdat->node_start_pfn = node_start_pfn;
3729 calculate_node_totalpages(pgdat, zones_size, zholes_size);
3731 alloc_node_mem_map(pgdat);
3732 #ifdef CONFIG_FLAT_NODE_MEM_MAP
3733 printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
3734 nid, (unsigned long)pgdat,
3735 (unsigned long)pgdat->node_mem_map);
3738 free_area_init_core(pgdat, zones_size, zholes_size);
3741 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3743 #if MAX_NUMNODES > 1
3745 * Figure out the number of possible node ids.
3747 static void __init setup_nr_node_ids(void)
3750 unsigned int highest = 0;
3752 for_each_node_mask(node, node_possible_map)
3754 nr_node_ids = highest + 1;
3757 static inline void setup_nr_node_ids(void)
3763 * add_active_range - Register a range of PFNs backed by physical memory
3764 * @nid: The node ID the range resides on
3765 * @start_pfn: The start PFN of the available physical memory
3766 * @end_pfn: The end PFN of the available physical memory
3768 * These ranges are stored in an early_node_map[] and later used by
3769 * free_area_init_nodes() to calculate zone sizes and holes. If the
3770 * range spans a memory hole, it is up to the architecture to ensure
3771 * the memory is not freed by the bootmem allocator. If possible
3772 * the range being registered will be merged with existing ranges.
3774 void __init add_active_range(unsigned int nid, unsigned long start_pfn,
3775 unsigned long end_pfn)
3779 mminit_dprintk(MMINIT_TRACE, "memory_register",
3780 "Entering add_active_range(%d, %#lx, %#lx) "
3781 "%d entries of %d used\n",
3782 nid, start_pfn, end_pfn,
3783 nr_nodemap_entries, MAX_ACTIVE_REGIONS);
3785 mminit_validate_memmodel_limits(&start_pfn, &end_pfn);
3787 /* Merge with existing active regions if possible */
3788 for (i = 0; i < nr_nodemap_entries; i++) {
3789 if (early_node_map[i].nid != nid)
3792 /* Skip if an existing region covers this new one */
3793 if (start_pfn >= early_node_map[i].start_pfn &&
3794 end_pfn <= early_node_map[i].end_pfn)
3797 /* Merge forward if suitable */
3798 if (start_pfn <= early_node_map[i].end_pfn &&
3799 end_pfn > early_node_map[i].end_pfn) {
3800 early_node_map[i].end_pfn = end_pfn;
3804 /* Merge backward if suitable */
3805 if (start_pfn < early_node_map[i].end_pfn &&
3806 end_pfn >= early_node_map[i].start_pfn) {
3807 early_node_map[i].start_pfn = start_pfn;
3812 /* Check that early_node_map is large enough */
3813 if (i >= MAX_ACTIVE_REGIONS) {
3814 printk(KERN_CRIT "More than %d memory regions, truncating\n",
3815 MAX_ACTIVE_REGIONS);
3819 early_node_map[i].nid = nid;
3820 early_node_map[i].start_pfn = start_pfn;
3821 early_node_map[i].end_pfn = end_pfn;
3822 nr_nodemap_entries = i + 1;
3826 * remove_active_range - Shrink an existing registered range of PFNs
3827 * @nid: The node id the range is on that should be shrunk
3828 * @start_pfn: The new PFN of the range
3829 * @end_pfn: The new PFN of the range
3831 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
3832 * The map is kept near the end physical page range that has already been
3833 * registered. This function allows an arch to shrink an existing registered
3836 void __init remove_active_range(unsigned int nid, unsigned long start_pfn,
3837 unsigned long end_pfn)
3842 printk(KERN_DEBUG "remove_active_range (%d, %lu, %lu)\n",
3843 nid, start_pfn, end_pfn);
3845 /* Find the old active region end and shrink */
3846 for_each_active_range_index_in_nid(i, nid) {
3847 if (early_node_map[i].start_pfn >= start_pfn &&
3848 early_node_map[i].end_pfn <= end_pfn) {
3850 early_node_map[i].start_pfn = 0;
3851 early_node_map[i].end_pfn = 0;
3855 if (early_node_map[i].start_pfn < start_pfn &&
3856 early_node_map[i].end_pfn > start_pfn) {
3857 unsigned long temp_end_pfn = early_node_map[i].end_pfn;
3858 early_node_map[i].end_pfn = start_pfn;
3859 if (temp_end_pfn > end_pfn)
3860 add_active_range(nid, end_pfn, temp_end_pfn);
3863 if (early_node_map[i].start_pfn >= start_pfn &&
3864 early_node_map[i].end_pfn > end_pfn &&
3865 early_node_map[i].start_pfn < end_pfn) {
3866 early_node_map[i].start_pfn = end_pfn;
3874 /* remove the blank ones */
3875 for (i = nr_nodemap_entries - 1; i > 0; i--) {
3876 if (early_node_map[i].nid != nid)
3878 if (early_node_map[i].end_pfn)
3880 /* we found it, get rid of it */
3881 for (j = i; j < nr_nodemap_entries - 1; j++)
3882 memcpy(&early_node_map[j], &early_node_map[j+1],
3883 sizeof(early_node_map[j]));
3884 j = nr_nodemap_entries - 1;
3885 memset(&early_node_map[j], 0, sizeof(early_node_map[j]));
3886 nr_nodemap_entries--;
3891 * remove_all_active_ranges - Remove all currently registered regions
3893 * During discovery, it may be found that a table like SRAT is invalid
3894 * and an alternative discovery method must be used. This function removes
3895 * all currently registered regions.
3897 void __init remove_all_active_ranges(void)
3899 memset(early_node_map, 0, sizeof(early_node_map));
3900 nr_nodemap_entries = 0;
3903 /* Compare two active node_active_regions */
3904 static int __init cmp_node_active_region(const void *a, const void *b)
3906 struct node_active_region *arange = (struct node_active_region *)a;
3907 struct node_active_region *brange = (struct node_active_region *)b;
3909 /* Done this way to avoid overflows */
3910 if (arange->start_pfn > brange->start_pfn)
3912 if (arange->start_pfn < brange->start_pfn)
3918 /* sort the node_map by start_pfn */
3919 static void __init sort_node_map(void)
3921 sort(early_node_map, (size_t)nr_nodemap_entries,
3922 sizeof(struct node_active_region),
3923 cmp_node_active_region, NULL);
3926 /* Find the lowest pfn for a node */
3927 static unsigned long __init find_min_pfn_for_node(int nid)
3930 unsigned long min_pfn = ULONG_MAX;
3932 /* Assuming a sorted map, the first range found has the starting pfn */
3933 for_each_active_range_index_in_nid(i, nid)
3934 min_pfn = min(min_pfn, early_node_map[i].start_pfn);
3936 if (min_pfn == ULONG_MAX) {
3938 "Could not find start_pfn for node %d\n", nid);
3946 * find_min_pfn_with_active_regions - Find the minimum PFN registered
3948 * It returns the minimum PFN based on information provided via
3949 * add_active_range().
3951 unsigned long __init find_min_pfn_with_active_regions(void)
3953 return find_min_pfn_for_node(MAX_NUMNODES);
3957 * early_calculate_totalpages()
3958 * Sum pages in active regions for movable zone.
3959 * Populate N_HIGH_MEMORY for calculating usable_nodes.
3961 static unsigned long __init early_calculate_totalpages(void)
3964 unsigned long totalpages = 0;
3966 for (i = 0; i < nr_nodemap_entries; i++) {
3967 unsigned long pages = early_node_map[i].end_pfn -
3968 early_node_map[i].start_pfn;
3969 totalpages += pages;
3971 node_set_state(early_node_map[i].nid, N_HIGH_MEMORY);
3977 * Find the PFN the Movable zone begins in each node. Kernel memory
3978 * is spread evenly between nodes as long as the nodes have enough
3979 * memory. When they don't, some nodes will have more kernelcore than
3982 static void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn)
3985 unsigned long usable_startpfn;
3986 unsigned long kernelcore_node, kernelcore_remaining;
3987 unsigned long totalpages = early_calculate_totalpages();
3988 int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]);
3991 * If movablecore was specified, calculate what size of
3992 * kernelcore that corresponds so that memory usable for
3993 * any allocation type is evenly spread. If both kernelcore
3994 * and movablecore are specified, then the value of kernelcore
3995 * will be used for required_kernelcore if it's greater than
3996 * what movablecore would have allowed.
3998 if (required_movablecore) {
3999 unsigned long corepages;
4002 * Round-up so that ZONE_MOVABLE is at least as large as what
4003 * was requested by the user
4005 required_movablecore =
4006 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
4007 corepages = totalpages - required_movablecore;
4009 required_kernelcore = max(required_kernelcore, corepages);
4012 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
4013 if (!required_kernelcore)
4016 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
4017 find_usable_zone_for_movable();
4018 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
4021 /* Spread kernelcore memory as evenly as possible throughout nodes */
4022 kernelcore_node = required_kernelcore / usable_nodes;
4023 for_each_node_state(nid, N_HIGH_MEMORY) {
4025 * Recalculate kernelcore_node if the division per node
4026 * now exceeds what is necessary to satisfy the requested
4027 * amount of memory for the kernel
4029 if (required_kernelcore < kernelcore_node)
4030 kernelcore_node = required_kernelcore / usable_nodes;
4033 * As the map is walked, we track how much memory is usable
4034 * by the kernel using kernelcore_remaining. When it is
4035 * 0, the rest of the node is usable by ZONE_MOVABLE
4037 kernelcore_remaining = kernelcore_node;
4039 /* Go through each range of PFNs within this node */
4040 for_each_active_range_index_in_nid(i, nid) {
4041 unsigned long start_pfn, end_pfn;
4042 unsigned long size_pages;
4044 start_pfn = max(early_node_map[i].start_pfn,
4045 zone_movable_pfn[nid]);
4046 end_pfn = early_node_map[i].end_pfn;
4047 if (start_pfn >= end_pfn)
4050 /* Account for what is only usable for kernelcore */
4051 if (start_pfn < usable_startpfn) {
4052 unsigned long kernel_pages;
4053 kernel_pages = min(end_pfn, usable_startpfn)
4056 kernelcore_remaining -= min(kernel_pages,
4057 kernelcore_remaining);
4058 required_kernelcore -= min(kernel_pages,
4059 required_kernelcore);
4061 /* Continue if range is now fully accounted */
4062 if (end_pfn <= usable_startpfn) {
4065 * Push zone_movable_pfn to the end so
4066 * that if we have to rebalance
4067 * kernelcore across nodes, we will
4068 * not double account here
4070 zone_movable_pfn[nid] = end_pfn;
4073 start_pfn = usable_startpfn;
4077 * The usable PFN range for ZONE_MOVABLE is from
4078 * start_pfn->end_pfn. Calculate size_pages as the
4079 * number of pages used as kernelcore
4081 size_pages = end_pfn - start_pfn;
4082 if (size_pages > kernelcore_remaining)
4083 size_pages = kernelcore_remaining;
4084 zone_movable_pfn[nid] = start_pfn + size_pages;
4087 * Some kernelcore has been met, update counts and
4088 * break if the kernelcore for this node has been
4091 required_kernelcore -= min(required_kernelcore,
4093 kernelcore_remaining -= size_pages;
4094 if (!kernelcore_remaining)
4100 * If there is still required_kernelcore, we do another pass with one
4101 * less node in the count. This will push zone_movable_pfn[nid] further
4102 * along on the nodes that still have memory until kernelcore is
4106 if (usable_nodes && required_kernelcore > usable_nodes)
4109 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
4110 for (nid = 0; nid < MAX_NUMNODES; nid++)
4111 zone_movable_pfn[nid] =
4112 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
4115 /* Any regular memory on that node ? */
4116 static void check_for_regular_memory(pg_data_t *pgdat)
4118 #ifdef CONFIG_HIGHMEM
4119 enum zone_type zone_type;
4121 for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) {
4122 struct zone *zone = &pgdat->node_zones[zone_type];
4123 if (zone->present_pages)
4124 node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY);
4130 * free_area_init_nodes - Initialise all pg_data_t and zone data
4131 * @max_zone_pfn: an array of max PFNs for each zone
4133 * This will call free_area_init_node() for each active node in the system.
4134 * Using the page ranges provided by add_active_range(), the size of each
4135 * zone in each node and their holes is calculated. If the maximum PFN
4136 * between two adjacent zones match, it is assumed that the zone is empty.
4137 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
4138 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
4139 * starts where the previous one ended. For example, ZONE_DMA32 starts
4140 * at arch_max_dma_pfn.
4142 void __init free_area_init_nodes(unsigned long *max_zone_pfn)
4147 /* Sort early_node_map as initialisation assumes it is sorted */
4150 /* Record where the zone boundaries are */
4151 memset(arch_zone_lowest_possible_pfn, 0,
4152 sizeof(arch_zone_lowest_possible_pfn));
4153 memset(arch_zone_highest_possible_pfn, 0,
4154 sizeof(arch_zone_highest_possible_pfn));
4155 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
4156 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
4157 for (i = 1; i < MAX_NR_ZONES; i++) {
4158 if (i == ZONE_MOVABLE)
4160 arch_zone_lowest_possible_pfn[i] =
4161 arch_zone_highest_possible_pfn[i-1];
4162 arch_zone_highest_possible_pfn[i] =
4163 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
4165 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
4166 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
4168 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
4169 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
4170 find_zone_movable_pfns_for_nodes(zone_movable_pfn);
4172 /* Print out the zone ranges */
4173 printk("Zone PFN ranges:\n");
4174 for (i = 0; i < MAX_NR_ZONES; i++) {
4175 if (i == ZONE_MOVABLE)
4177 printk(" %-8s %0#10lx -> %0#10lx\n",
4179 arch_zone_lowest_possible_pfn[i],
4180 arch_zone_highest_possible_pfn[i]);
4183 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
4184 printk("Movable zone start PFN for each node\n");
4185 for (i = 0; i < MAX_NUMNODES; i++) {
4186 if (zone_movable_pfn[i])
4187 printk(" Node %d: %lu\n", i, zone_movable_pfn[i]);
4190 /* Print out the early_node_map[] */
4191 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries);
4192 for (i = 0; i < nr_nodemap_entries; i++)
4193 printk(" %3d: %0#10lx -> %0#10lx\n", early_node_map[i].nid,
4194 early_node_map[i].start_pfn,
4195 early_node_map[i].end_pfn);
4197 /* Initialise every node */
4198 mminit_verify_pageflags_layout();
4199 setup_nr_node_ids();
4200 for_each_online_node(nid) {
4201 pg_data_t *pgdat = NODE_DATA(nid);
4202 free_area_init_node(nid, NULL,
4203 find_min_pfn_for_node(nid), NULL);
4205 /* Any memory on that node */
4206 if (pgdat->node_present_pages)
4207 node_set_state(nid, N_HIGH_MEMORY);
4208 check_for_regular_memory(pgdat);
4212 static int __init cmdline_parse_core(char *p, unsigned long *core)
4214 unsigned long long coremem;
4218 coremem = memparse(p, &p);
4219 *core = coremem >> PAGE_SHIFT;
4221 /* Paranoid check that UL is enough for the coremem value */
4222 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
4228 * kernelcore=size sets the amount of memory for use for allocations that
4229 * cannot be reclaimed or migrated.
4231 static int __init cmdline_parse_kernelcore(char *p)
4233 return cmdline_parse_core(p, &required_kernelcore);
4237 * movablecore=size sets the amount of memory for use for allocations that
4238 * can be reclaimed or migrated.
4240 static int __init cmdline_parse_movablecore(char *p)
4242 return cmdline_parse_core(p, &required_movablecore);
4245 early_param("kernelcore", cmdline_parse_kernelcore);
4246 early_param("movablecore", cmdline_parse_movablecore);
4248 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4251 * set_dma_reserve - set the specified number of pages reserved in the first zone
4252 * @new_dma_reserve: The number of pages to mark reserved
4254 * The per-cpu batchsize and zone watermarks are determined by present_pages.
4255 * In the DMA zone, a significant percentage may be consumed by kernel image
4256 * and other unfreeable allocations which can skew the watermarks badly. This
4257 * function may optionally be used to account for unfreeable pages in the
4258 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
4259 * smaller per-cpu batchsize.
4261 void __init set_dma_reserve(unsigned long new_dma_reserve)
4263 dma_reserve = new_dma_reserve;
4266 #ifndef CONFIG_NEED_MULTIPLE_NODES
4267 struct pglist_data __refdata contig_page_data = { .bdata = &bootmem_node_data[0] };
4268 EXPORT_SYMBOL(contig_page_data);
4271 void __init free_area_init(unsigned long *zones_size)
4273 free_area_init_node(0, zones_size,
4274 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
4277 static int page_alloc_cpu_notify(struct notifier_block *self,
4278 unsigned long action, void *hcpu)
4280 int cpu = (unsigned long)hcpu;
4282 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
4286 * Spill the event counters of the dead processor
4287 * into the current processors event counters.
4288 * This artificially elevates the count of the current
4291 vm_events_fold_cpu(cpu);
4294 * Zero the differential counters of the dead processor
4295 * so that the vm statistics are consistent.
4297 * This is only okay since the processor is dead and cannot
4298 * race with what we are doing.
4300 refresh_cpu_vm_stats(cpu);
4305 void __init page_alloc_init(void)
4307 hotcpu_notifier(page_alloc_cpu_notify, 0);
4311 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
4312 * or min_free_kbytes changes.
4314 static void calculate_totalreserve_pages(void)
4316 struct pglist_data *pgdat;
4317 unsigned long reserve_pages = 0;
4318 enum zone_type i, j;
4320 for_each_online_pgdat(pgdat) {
4321 for (i = 0; i < MAX_NR_ZONES; i++) {
4322 struct zone *zone = pgdat->node_zones + i;
4323 unsigned long max = 0;
4325 /* Find valid and maximum lowmem_reserve in the zone */
4326 for (j = i; j < MAX_NR_ZONES; j++) {
4327 if (zone->lowmem_reserve[j] > max)
4328 max = zone->lowmem_reserve[j];
4331 /* we treat the high watermark as reserved pages. */
4332 max += high_wmark_pages(zone);
4334 if (max > zone->present_pages)
4335 max = zone->present_pages;
4336 reserve_pages += max;
4339 totalreserve_pages = reserve_pages;
4343 * setup_per_zone_lowmem_reserve - called whenever
4344 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
4345 * has a correct pages reserved value, so an adequate number of
4346 * pages are left in the zone after a successful __alloc_pages().
4348 static void setup_per_zone_lowmem_reserve(void)
4350 struct pglist_data *pgdat;
4351 enum zone_type j, idx;
4353 for_each_online_pgdat(pgdat) {
4354 for (j = 0; j < MAX_NR_ZONES; j++) {
4355 struct zone *zone = pgdat->node_zones + j;
4356 unsigned long present_pages = zone->present_pages;
4358 zone->lowmem_reserve[j] = 0;
4362 struct zone *lower_zone;
4366 if (sysctl_lowmem_reserve_ratio[idx] < 1)
4367 sysctl_lowmem_reserve_ratio[idx] = 1;
4369 lower_zone = pgdat->node_zones + idx;
4370 lower_zone->lowmem_reserve[j] = present_pages /
4371 sysctl_lowmem_reserve_ratio[idx];
4372 present_pages += lower_zone->present_pages;
4377 /* update totalreserve_pages */
4378 calculate_totalreserve_pages();
4382 * setup_per_zone_pages_min - called when min_free_kbytes changes.
4384 * Ensures that the pages_{min,low,high} values for each zone are set correctly
4385 * with respect to min_free_kbytes.
4387 void setup_per_zone_pages_min(void)
4389 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
4390 unsigned long lowmem_pages = 0;
4392 unsigned long flags;
4394 /* Calculate total number of !ZONE_HIGHMEM pages */
4395 for_each_zone(zone) {
4396 if (!is_highmem(zone))
4397 lowmem_pages += zone->present_pages;
4400 for_each_zone(zone) {
4403 spin_lock_irqsave(&zone->lock, flags);
4404 tmp = (u64)pages_min * zone->present_pages;
4405 do_div(tmp, lowmem_pages);
4406 if (is_highmem(zone)) {
4408 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
4409 * need highmem pages, so cap pages_min to a small
4412 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
4413 * deltas controls asynch page reclaim, and so should
4414 * not be capped for highmem.
4418 min_pages = zone->present_pages / 1024;
4419 if (min_pages < SWAP_CLUSTER_MAX)
4420 min_pages = SWAP_CLUSTER_MAX;
4421 if (min_pages > 128)
4423 zone->watermark[WMARK_MIN] = min_pages;
4426 * If it's a lowmem zone, reserve a number of pages
4427 * proportionate to the zone's size.
4429 zone->watermark[WMARK_MIN] = tmp;
4432 zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + (tmp >> 2);
4433 zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1);
4434 setup_zone_migrate_reserve(zone);
4435 spin_unlock_irqrestore(&zone->lock, flags);
4438 /* update totalreserve_pages */
4439 calculate_totalreserve_pages();
4443 * setup_per_zone_inactive_ratio - called when min_free_kbytes changes.
4445 * The inactive anon list should be small enough that the VM never has to
4446 * do too much work, but large enough that each inactive page has a chance
4447 * to be referenced again before it is swapped out.
4449 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
4450 * INACTIVE_ANON pages on this zone's LRU, maintained by the
4451 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
4452 * the anonymous pages are kept on the inactive list.
4455 * memory ratio inactive anon
4456 * -------------------------------------
4465 static void setup_per_zone_inactive_ratio(void)
4469 for_each_zone(zone) {
4470 unsigned int gb, ratio;
4472 /* Zone size in gigabytes */
4473 gb = zone->present_pages >> (30 - PAGE_SHIFT);
4474 ratio = int_sqrt(10 * gb);
4478 zone->inactive_ratio = ratio;
4483 * Initialise min_free_kbytes.
4485 * For small machines we want it small (128k min). For large machines
4486 * we want it large (64MB max). But it is not linear, because network
4487 * bandwidth does not increase linearly with machine size. We use
4489 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
4490 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
4506 static int __init init_per_zone_pages_min(void)
4508 unsigned long lowmem_kbytes;
4510 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
4512 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
4513 if (min_free_kbytes < 128)
4514 min_free_kbytes = 128;
4515 if (min_free_kbytes > 65536)
4516 min_free_kbytes = 65536;
4517 setup_per_zone_pages_min();
4518 setup_per_zone_lowmem_reserve();
4519 setup_per_zone_inactive_ratio();
4522 module_init(init_per_zone_pages_min)
4525 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
4526 * that we can call two helper functions whenever min_free_kbytes
4529 int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
4530 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4532 proc_dointvec(table, write, file, buffer, length, ppos);
4534 setup_per_zone_pages_min();
4539 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
4540 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4545 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4550 zone->min_unmapped_pages = (zone->present_pages *
4551 sysctl_min_unmapped_ratio) / 100;
4555 int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
4556 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4561 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4566 zone->min_slab_pages = (zone->present_pages *
4567 sysctl_min_slab_ratio) / 100;
4573 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
4574 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
4575 * whenever sysctl_lowmem_reserve_ratio changes.
4577 * The reserve ratio obviously has absolutely no relation with the
4578 * minimum watermarks. The lowmem reserve ratio can only make sense
4579 * if in function of the boot time zone sizes.
4581 int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
4582 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4584 proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4585 setup_per_zone_lowmem_reserve();
4590 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
4591 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
4592 * can have before it gets flushed back to buddy allocator.
4595 int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
4596 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4602 ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4603 if (!write || (ret == -EINVAL))
4605 for_each_zone(zone) {
4606 for_each_online_cpu(cpu) {
4608 high = zone->present_pages / percpu_pagelist_fraction;
4609 setup_pagelist_highmark(zone_pcp(zone, cpu), high);
4615 int hashdist = HASHDIST_DEFAULT;
4618 static int __init set_hashdist(char *str)
4622 hashdist = simple_strtoul(str, &str, 0);
4625 __setup("hashdist=", set_hashdist);
4629 * allocate a large system hash table from bootmem
4630 * - it is assumed that the hash table must contain an exact power-of-2
4631 * quantity of entries
4632 * - limit is the number of hash buckets, not the total allocation size
4634 void *__init alloc_large_system_hash(const char *tablename,
4635 unsigned long bucketsize,
4636 unsigned long numentries,
4639 unsigned int *_hash_shift,
4640 unsigned int *_hash_mask,
4641 unsigned long limit)
4643 unsigned long long max = limit;
4644 unsigned long log2qty, size;
4647 /* allow the kernel cmdline to have a say */
4649 /* round applicable memory size up to nearest megabyte */
4650 numentries = nr_kernel_pages;
4651 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
4652 numentries >>= 20 - PAGE_SHIFT;
4653 numentries <<= 20 - PAGE_SHIFT;
4655 /* limit to 1 bucket per 2^scale bytes of low memory */
4656 if (scale > PAGE_SHIFT)
4657 numentries >>= (scale - PAGE_SHIFT);
4659 numentries <<= (PAGE_SHIFT - scale);
4661 /* Make sure we've got at least a 0-order allocation.. */
4662 if (unlikely((numentries * bucketsize) < PAGE_SIZE))
4663 numentries = PAGE_SIZE / bucketsize;
4665 numentries = roundup_pow_of_two(numentries);
4667 /* limit allocation size to 1/16 total memory by default */
4669 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
4670 do_div(max, bucketsize);
4673 if (numentries > max)
4676 log2qty = ilog2(numentries);
4679 size = bucketsize << log2qty;
4680 if (flags & HASH_EARLY)
4681 table = alloc_bootmem_nopanic(size);
4683 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
4686 * If bucketsize is not a power-of-two, we may free
4687 * some pages at the end of hash table which
4688 * alloc_pages_exact() automatically does
4690 if (get_order(size) < MAX_ORDER)
4691 table = alloc_pages_exact(size, GFP_ATOMIC);
4693 } while (!table && size > PAGE_SIZE && --log2qty);
4696 panic("Failed to allocate %s hash table\n", tablename);
4698 printk(KERN_INFO "%s hash table entries: %d (order: %d, %lu bytes)\n",
4701 ilog2(size) - PAGE_SHIFT,
4705 *_hash_shift = log2qty;
4707 *_hash_mask = (1 << log2qty) - 1;
4710 * If hashdist is set, the table allocation is done with __vmalloc()
4711 * which invokes the kmemleak_alloc() callback. This function may also
4712 * be called before the slab and kmemleak are initialised when
4713 * kmemleak simply buffers the request to be executed later
4714 * (GFP_ATOMIC flag ignored in this case).
4717 kmemleak_alloc(table, size, 1, GFP_ATOMIC);
4722 /* Return a pointer to the bitmap storing bits affecting a block of pages */
4723 static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
4726 #ifdef CONFIG_SPARSEMEM
4727 return __pfn_to_section(pfn)->pageblock_flags;
4729 return zone->pageblock_flags;
4730 #endif /* CONFIG_SPARSEMEM */
4733 static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
4735 #ifdef CONFIG_SPARSEMEM
4736 pfn &= (PAGES_PER_SECTION-1);
4737 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
4739 pfn = pfn - zone->zone_start_pfn;
4740 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
4741 #endif /* CONFIG_SPARSEMEM */
4745 * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
4746 * @page: The page within the block of interest
4747 * @start_bitidx: The first bit of interest to retrieve
4748 * @end_bitidx: The last bit of interest
4749 * returns pageblock_bits flags
4751 unsigned long get_pageblock_flags_group(struct page *page,
4752 int start_bitidx, int end_bitidx)
4755 unsigned long *bitmap;
4756 unsigned long pfn, bitidx;
4757 unsigned long flags = 0;
4758 unsigned long value = 1;
4760 zone = page_zone(page);
4761 pfn = page_to_pfn(page);
4762 bitmap = get_pageblock_bitmap(zone, pfn);
4763 bitidx = pfn_to_bitidx(zone, pfn);
4765 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
4766 if (test_bit(bitidx + start_bitidx, bitmap))
4773 * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
4774 * @page: The page within the block of interest
4775 * @start_bitidx: The first bit of interest
4776 * @end_bitidx: The last bit of interest
4777 * @flags: The flags to set
4779 void set_pageblock_flags_group(struct page *page, unsigned long flags,
4780 int start_bitidx, int end_bitidx)
4783 unsigned long *bitmap;
4784 unsigned long pfn, bitidx;
4785 unsigned long value = 1;
4787 zone = page_zone(page);
4788 pfn = page_to_pfn(page);
4789 bitmap = get_pageblock_bitmap(zone, pfn);
4790 bitidx = pfn_to_bitidx(zone, pfn);
4791 VM_BUG_ON(pfn < zone->zone_start_pfn);
4792 VM_BUG_ON(pfn >= zone->zone_start_pfn + zone->spanned_pages);
4794 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
4796 __set_bit(bitidx + start_bitidx, bitmap);
4798 __clear_bit(bitidx + start_bitidx, bitmap);
4802 * This is designed as sub function...plz see page_isolation.c also.
4803 * set/clear page block's type to be ISOLATE.
4804 * page allocater never alloc memory from ISOLATE block.
4807 int set_migratetype_isolate(struct page *page)
4810 unsigned long flags;
4813 zone = page_zone(page);
4814 spin_lock_irqsave(&zone->lock, flags);
4816 * In future, more migrate types will be able to be isolation target.
4818 if (get_pageblock_migratetype(page) != MIGRATE_MOVABLE)
4820 set_pageblock_migratetype(page, MIGRATE_ISOLATE);
4821 move_freepages_block(zone, page, MIGRATE_ISOLATE);
4824 spin_unlock_irqrestore(&zone->lock, flags);
4830 void unset_migratetype_isolate(struct page *page)
4833 unsigned long flags;
4834 zone = page_zone(page);
4835 spin_lock_irqsave(&zone->lock, flags);
4836 if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE)
4838 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
4839 move_freepages_block(zone, page, MIGRATE_MOVABLE);
4841 spin_unlock_irqrestore(&zone->lock, flags);
4844 #ifdef CONFIG_MEMORY_HOTREMOVE
4846 * All pages in the range must be isolated before calling this.
4849 __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
4855 unsigned long flags;
4856 /* find the first valid pfn */
4857 for (pfn = start_pfn; pfn < end_pfn; pfn++)
4862 zone = page_zone(pfn_to_page(pfn));
4863 spin_lock_irqsave(&zone->lock, flags);
4865 while (pfn < end_pfn) {
4866 if (!pfn_valid(pfn)) {
4870 page = pfn_to_page(pfn);
4871 BUG_ON(page_count(page));
4872 BUG_ON(!PageBuddy(page));
4873 order = page_order(page);
4874 #ifdef CONFIG_DEBUG_VM
4875 printk(KERN_INFO "remove from free list %lx %d %lx\n",
4876 pfn, 1 << order, end_pfn);
4878 list_del(&page->lru);
4879 rmv_page_order(page);
4880 zone->free_area[order].nr_free--;
4881 __mod_zone_page_state(zone, NR_FREE_PAGES,
4883 for (i = 0; i < (1 << order); i++)
4884 SetPageReserved((page+i));
4885 pfn += (1 << order);
4887 spin_unlock_irqrestore(&zone->lock, flags);