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/memcontrol.h>
49 #include <asm/tlbflush.h>
50 #include <asm/div64.h>
54 * Array of node states.
56 nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
57 [N_POSSIBLE] = NODE_MASK_ALL,
58 [N_ONLINE] = { { [0] = 1UL } },
60 [N_NORMAL_MEMORY] = { { [0] = 1UL } },
62 [N_HIGH_MEMORY] = { { [0] = 1UL } },
64 [N_CPU] = { { [0] = 1UL } },
67 EXPORT_SYMBOL(node_states);
69 unsigned long totalram_pages __read_mostly;
70 unsigned long totalreserve_pages __read_mostly;
72 int percpu_pagelist_fraction;
74 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
75 int pageblock_order __read_mostly;
78 static void __free_pages_ok(struct page *page, unsigned int order);
81 * results with 256, 32 in the lowmem_reserve sysctl:
82 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
83 * 1G machine -> (16M dma, 784M normal, 224M high)
84 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
85 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
86 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
88 * TBD: should special case ZONE_DMA32 machines here - in those we normally
89 * don't need any ZONE_NORMAL reservation
91 int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
92 #ifdef CONFIG_ZONE_DMA
95 #ifdef CONFIG_ZONE_DMA32
104 EXPORT_SYMBOL(totalram_pages);
106 static char * const zone_names[MAX_NR_ZONES] = {
107 #ifdef CONFIG_ZONE_DMA
110 #ifdef CONFIG_ZONE_DMA32
114 #ifdef CONFIG_HIGHMEM
120 int min_free_kbytes = 1024;
122 unsigned long __meminitdata nr_kernel_pages;
123 unsigned long __meminitdata nr_all_pages;
124 static unsigned long __meminitdata dma_reserve;
126 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
128 * MAX_ACTIVE_REGIONS determines the maximum number of distinct
129 * ranges of memory (RAM) that may be registered with add_active_range().
130 * Ranges passed to add_active_range() will be merged if possible
131 * so the number of times add_active_range() can be called is
132 * related to the number of nodes and the number of holes
134 #ifdef CONFIG_MAX_ACTIVE_REGIONS
135 /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
136 #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
138 #if MAX_NUMNODES >= 32
139 /* If there can be many nodes, allow up to 50 holes per node */
140 #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
142 /* By default, allow up to 256 distinct regions */
143 #define MAX_ACTIVE_REGIONS 256
147 static struct node_active_region __meminitdata early_node_map[MAX_ACTIVE_REGIONS];
148 static int __meminitdata nr_nodemap_entries;
149 static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
150 static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
151 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
152 static unsigned long __meminitdata node_boundary_start_pfn[MAX_NUMNODES];
153 static unsigned long __meminitdata node_boundary_end_pfn[MAX_NUMNODES];
154 #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
155 unsigned long __initdata required_kernelcore;
156 static unsigned long __initdata required_movablecore;
157 unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
159 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
161 EXPORT_SYMBOL(movable_zone);
162 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
165 int nr_node_ids __read_mostly = MAX_NUMNODES;
166 EXPORT_SYMBOL(nr_node_ids);
169 int page_group_by_mobility_disabled __read_mostly;
171 static void set_pageblock_migratetype(struct page *page, int migratetype)
173 set_pageblock_flags_group(page, (unsigned long)migratetype,
174 PB_migrate, PB_migrate_end);
177 #ifdef CONFIG_DEBUG_VM
178 static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
182 unsigned long pfn = page_to_pfn(page);
185 seq = zone_span_seqbegin(zone);
186 if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
188 else if (pfn < zone->zone_start_pfn)
190 } while (zone_span_seqretry(zone, seq));
195 static int page_is_consistent(struct zone *zone, struct page *page)
197 if (!pfn_valid_within(page_to_pfn(page)))
199 if (zone != page_zone(page))
205 * Temporary debugging check for pages not lying within a given zone.
207 static int bad_range(struct zone *zone, struct page *page)
209 if (page_outside_zone_boundaries(zone, page))
211 if (!page_is_consistent(zone, page))
217 static inline int bad_range(struct zone *zone, struct page *page)
223 static void bad_page(struct page *page)
225 void *pc = page_get_page_cgroup(page);
227 printk(KERN_EMERG "Bad page state in process '%s'\n" KERN_EMERG
228 "page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n",
229 current->comm, page, (int)(2*sizeof(unsigned long)),
230 (unsigned long)page->flags, page->mapping,
231 page_mapcount(page), page_count(page));
233 printk(KERN_EMERG "cgroup:%p\n", pc);
234 page_reset_bad_cgroup(page);
236 printk(KERN_EMERG "Trying to fix it up, but a reboot is needed\n"
237 KERN_EMERG "Backtrace:\n");
239 page->flags &= ~(1 << PG_lru |
249 set_page_count(page, 0);
250 reset_page_mapcount(page);
251 page->mapping = NULL;
252 add_taint(TAINT_BAD_PAGE);
256 * Higher-order pages are called "compound pages". They are structured thusly:
258 * The first PAGE_SIZE page is called the "head page".
260 * The remaining PAGE_SIZE pages are called "tail pages".
262 * All pages have PG_compound set. All pages have their ->private pointing at
263 * the head page (even the head page has this).
265 * The first tail page's ->lru.next holds the address of the compound page's
266 * put_page() function. Its ->lru.prev holds the order of allocation.
267 * This usage means that zero-order pages may not be compound.
270 static void free_compound_page(struct page *page)
272 __free_pages_ok(page, compound_order(page));
275 static void prep_compound_page(struct page *page, unsigned long order)
278 int nr_pages = 1 << order;
280 set_compound_page_dtor(page, free_compound_page);
281 set_compound_order(page, order);
283 for (i = 1; i < nr_pages; i++) {
284 struct page *p = page + i;
287 p->first_page = page;
291 static void destroy_compound_page(struct page *page, unsigned long order)
294 int nr_pages = 1 << order;
296 if (unlikely(compound_order(page) != order))
299 if (unlikely(!PageHead(page)))
301 __ClearPageHead(page);
302 for (i = 1; i < nr_pages; i++) {
303 struct page *p = page + i;
305 if (unlikely(!PageTail(p) |
306 (p->first_page != page)))
312 static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
317 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
318 * and __GFP_HIGHMEM from hard or soft interrupt context.
320 VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
321 for (i = 0; i < (1 << order); i++)
322 clear_highpage(page + i);
325 static inline void set_page_order(struct page *page, int order)
327 set_page_private(page, order);
328 __SetPageBuddy(page);
331 static inline void rmv_page_order(struct page *page)
333 __ClearPageBuddy(page);
334 set_page_private(page, 0);
338 * Locate the struct page for both the matching buddy in our
339 * pair (buddy1) and the combined O(n+1) page they form (page).
341 * 1) Any buddy B1 will have an order O twin B2 which satisfies
342 * the following equation:
344 * For example, if the starting buddy (buddy2) is #8 its order
346 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
348 * 2) Any buddy B will have an order O+1 parent P which
349 * satisfies the following equation:
352 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
354 static inline struct page *
355 __page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order)
357 unsigned long buddy_idx = page_idx ^ (1 << order);
359 return page + (buddy_idx - page_idx);
362 static inline unsigned long
363 __find_combined_index(unsigned long page_idx, unsigned int order)
365 return (page_idx & ~(1 << order));
369 * This function checks whether a page is free && is the buddy
370 * we can do coalesce a page and its buddy if
371 * (a) the buddy is not in a hole &&
372 * (b) the buddy is in the buddy system &&
373 * (c) a page and its buddy have the same order &&
374 * (d) a page and its buddy are in the same zone.
376 * For recording whether a page is in the buddy system, we use PG_buddy.
377 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
379 * For recording page's order, we use page_private(page).
381 static inline int page_is_buddy(struct page *page, struct page *buddy,
384 if (!pfn_valid_within(page_to_pfn(buddy)))
387 if (page_zone_id(page) != page_zone_id(buddy))
390 if (PageBuddy(buddy) && page_order(buddy) == order) {
391 BUG_ON(page_count(buddy) != 0);
398 * Freeing function for a buddy system allocator.
400 * The concept of a buddy system is to maintain direct-mapped table
401 * (containing bit values) for memory blocks of various "orders".
402 * The bottom level table contains the map for the smallest allocatable
403 * units of memory (here, pages), and each level above it describes
404 * pairs of units from the levels below, hence, "buddies".
405 * At a high level, all that happens here is marking the table entry
406 * at the bottom level available, and propagating the changes upward
407 * as necessary, plus some accounting needed to play nicely with other
408 * parts of the VM system.
409 * At each level, we keep a list of pages, which are heads of continuous
410 * free pages of length of (1 << order) and marked with PG_buddy. Page's
411 * order is recorded in page_private(page) field.
412 * So when we are allocating or freeing one, we can derive the state of the
413 * other. That is, if we allocate a small block, and both were
414 * free, the remainder of the region must be split into blocks.
415 * If a block is freed, and its buddy is also free, then this
416 * triggers coalescing into a block of larger size.
421 static inline void __free_one_page(struct page *page,
422 struct zone *zone, unsigned int order)
424 unsigned long page_idx;
425 int order_size = 1 << order;
426 int migratetype = get_pageblock_migratetype(page);
428 if (unlikely(PageCompound(page)))
429 destroy_compound_page(page, order);
431 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
433 VM_BUG_ON(page_idx & (order_size - 1));
434 VM_BUG_ON(bad_range(zone, page));
436 __mod_zone_page_state(zone, NR_FREE_PAGES, order_size);
437 while (order < MAX_ORDER-1) {
438 unsigned long combined_idx;
441 buddy = __page_find_buddy(page, page_idx, order);
442 if (!page_is_buddy(page, buddy, order))
443 break; /* Move the buddy up one level. */
445 list_del(&buddy->lru);
446 zone->free_area[order].nr_free--;
447 rmv_page_order(buddy);
448 combined_idx = __find_combined_index(page_idx, order);
449 page = page + (combined_idx - page_idx);
450 page_idx = combined_idx;
453 set_page_order(page, order);
455 &zone->free_area[order].free_list[migratetype]);
456 zone->free_area[order].nr_free++;
459 static inline int free_pages_check(struct page *page)
461 if (unlikely(page_mapcount(page) |
462 (page->mapping != NULL) |
463 (page_get_page_cgroup(page) != NULL) |
464 (page_count(page) != 0) |
477 __ClearPageDirty(page);
479 * For now, we report if PG_reserved was found set, but do not
480 * clear it, and do not free the page. But we shall soon need
481 * to do more, for when the ZERO_PAGE count wraps negative.
483 return PageReserved(page);
487 * Frees a list of pages.
488 * Assumes all pages on list are in same zone, and of same order.
489 * count is the number of pages to free.
491 * If the zone was previously in an "all pages pinned" state then look to
492 * see if this freeing clears that state.
494 * And clear the zone's pages_scanned counter, to hold off the "all pages are
495 * pinned" detection logic.
497 static void free_pages_bulk(struct zone *zone, int count,
498 struct list_head *list, int order)
500 spin_lock(&zone->lock);
501 zone_clear_flag(zone, ZONE_ALL_UNRECLAIMABLE);
502 zone->pages_scanned = 0;
506 VM_BUG_ON(list_empty(list));
507 page = list_entry(list->prev, struct page, lru);
508 /* have to delete it as __free_one_page list manipulates */
509 list_del(&page->lru);
510 __free_one_page(page, zone, order);
512 spin_unlock(&zone->lock);
515 static void free_one_page(struct zone *zone, struct page *page, int order)
517 spin_lock(&zone->lock);
518 zone_clear_flag(zone, ZONE_ALL_UNRECLAIMABLE);
519 zone->pages_scanned = 0;
520 __free_one_page(page, zone, order);
521 spin_unlock(&zone->lock);
524 static void __free_pages_ok(struct page *page, unsigned int order)
530 for (i = 0 ; i < (1 << order) ; ++i)
531 reserved += free_pages_check(page + i);
535 if (!PageHighMem(page))
536 debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order);
537 arch_free_page(page, order);
538 kernel_map_pages(page, 1 << order, 0);
540 local_irq_save(flags);
541 __count_vm_events(PGFREE, 1 << order);
542 free_one_page(page_zone(page), page, order);
543 local_irq_restore(flags);
547 * permit the bootmem allocator to evade page validation on high-order frees
549 void __init __free_pages_bootmem(struct page *page, unsigned int order)
552 __ClearPageReserved(page);
553 set_page_count(page, 0);
554 set_page_refcounted(page);
560 for (loop = 0; loop < BITS_PER_LONG; loop++) {
561 struct page *p = &page[loop];
563 if (loop + 1 < BITS_PER_LONG)
565 __ClearPageReserved(p);
566 set_page_count(p, 0);
569 set_page_refcounted(page);
570 __free_pages(page, order);
576 * The order of subdivision here is critical for the IO subsystem.
577 * Please do not alter this order without good reasons and regression
578 * testing. Specifically, as large blocks of memory are subdivided,
579 * the order in which smaller blocks are delivered depends on the order
580 * they're subdivided in this function. This is the primary factor
581 * influencing the order in which pages are delivered to the IO
582 * subsystem according to empirical testing, and this is also justified
583 * by considering the behavior of a buddy system containing a single
584 * large block of memory acted on by a series of small allocations.
585 * This behavior is a critical factor in sglist merging's success.
589 static inline void expand(struct zone *zone, struct page *page,
590 int low, int high, struct free_area *area,
593 unsigned long size = 1 << high;
599 VM_BUG_ON(bad_range(zone, &page[size]));
600 list_add(&page[size].lru, &area->free_list[migratetype]);
602 set_page_order(&page[size], high);
607 * This page is about to be returned from the page allocator
609 static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
611 if (unlikely(page_mapcount(page) |
612 (page->mapping != NULL) |
613 (page_get_page_cgroup(page) != NULL) |
614 (page_count(page) != 0) |
629 * For now, we report if PG_reserved was found set, but do not
630 * clear it, and do not allocate the page: as a safety net.
632 if (PageReserved(page))
635 page->flags &= ~(1 << PG_uptodate | 1 << PG_error | 1 << PG_readahead |
636 1 << PG_referenced | 1 << PG_arch_1 |
637 1 << PG_owner_priv_1 | 1 << PG_mappedtodisk);
638 set_page_private(page, 0);
639 set_page_refcounted(page);
641 arch_alloc_page(page, order);
642 kernel_map_pages(page, 1 << order, 1);
644 if (gfp_flags & __GFP_ZERO)
645 prep_zero_page(page, order, gfp_flags);
647 if (order && (gfp_flags & __GFP_COMP))
648 prep_compound_page(page, order);
654 * Go through the free lists for the given migratetype and remove
655 * the smallest available page from the freelists
657 static struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
660 unsigned int current_order;
661 struct free_area * area;
664 /* Find a page of the appropriate size in the preferred list */
665 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
666 area = &(zone->free_area[current_order]);
667 if (list_empty(&area->free_list[migratetype]))
670 page = list_entry(area->free_list[migratetype].next,
672 list_del(&page->lru);
673 rmv_page_order(page);
675 __mod_zone_page_state(zone, NR_FREE_PAGES, - (1UL << order));
676 expand(zone, page, order, current_order, area, migratetype);
685 * This array describes the order lists are fallen back to when
686 * the free lists for the desirable migrate type are depleted
688 static int fallbacks[MIGRATE_TYPES][MIGRATE_TYPES-1] = {
689 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
690 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
691 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
692 [MIGRATE_RESERVE] = { MIGRATE_RESERVE, MIGRATE_RESERVE, MIGRATE_RESERVE }, /* Never used */
696 * Move the free pages in a range to the free lists of the requested type.
697 * Note that start_page and end_pages are not aligned on a pageblock
698 * boundary. If alignment is required, use move_freepages_block()
700 int move_freepages(struct zone *zone,
701 struct page *start_page, struct page *end_page,
708 #ifndef CONFIG_HOLES_IN_ZONE
710 * page_zone is not safe to call in this context when
711 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
712 * anyway as we check zone boundaries in move_freepages_block().
713 * Remove at a later date when no bug reports exist related to
714 * grouping pages by mobility
716 BUG_ON(page_zone(start_page) != page_zone(end_page));
719 for (page = start_page; page <= end_page;) {
720 if (!pfn_valid_within(page_to_pfn(page))) {
725 if (!PageBuddy(page)) {
730 order = page_order(page);
731 list_del(&page->lru);
733 &zone->free_area[order].free_list[migratetype]);
735 pages_moved += 1 << order;
741 int move_freepages_block(struct zone *zone, struct page *page, int migratetype)
743 unsigned long start_pfn, end_pfn;
744 struct page *start_page, *end_page;
746 start_pfn = page_to_pfn(page);
747 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
748 start_page = pfn_to_page(start_pfn);
749 end_page = start_page + pageblock_nr_pages - 1;
750 end_pfn = start_pfn + pageblock_nr_pages - 1;
752 /* Do not cross zone boundaries */
753 if (start_pfn < zone->zone_start_pfn)
755 if (end_pfn >= zone->zone_start_pfn + zone->spanned_pages)
758 return move_freepages(zone, start_page, end_page, migratetype);
761 /* Remove an element from the buddy allocator from the fallback list */
762 static struct page *__rmqueue_fallback(struct zone *zone, int order,
763 int start_migratetype)
765 struct free_area * area;
770 /* Find the largest possible block of pages in the other list */
771 for (current_order = MAX_ORDER-1; current_order >= order;
773 for (i = 0; i < MIGRATE_TYPES - 1; i++) {
774 migratetype = fallbacks[start_migratetype][i];
776 /* MIGRATE_RESERVE handled later if necessary */
777 if (migratetype == MIGRATE_RESERVE)
780 area = &(zone->free_area[current_order]);
781 if (list_empty(&area->free_list[migratetype]))
784 page = list_entry(area->free_list[migratetype].next,
789 * If breaking a large block of pages, move all free
790 * pages to the preferred allocation list. If falling
791 * back for a reclaimable kernel allocation, be more
792 * agressive about taking ownership of free pages
794 if (unlikely(current_order >= (pageblock_order >> 1)) ||
795 start_migratetype == MIGRATE_RECLAIMABLE) {
797 pages = move_freepages_block(zone, page,
800 /* Claim the whole block if over half of it is free */
801 if (pages >= (1 << (pageblock_order-1)))
802 set_pageblock_migratetype(page,
805 migratetype = start_migratetype;
808 /* Remove the page from the freelists */
809 list_del(&page->lru);
810 rmv_page_order(page);
811 __mod_zone_page_state(zone, NR_FREE_PAGES,
814 if (current_order == pageblock_order)
815 set_pageblock_migratetype(page,
818 expand(zone, page, order, current_order, area, migratetype);
823 /* Use MIGRATE_RESERVE rather than fail an allocation */
824 return __rmqueue_smallest(zone, order, MIGRATE_RESERVE);
828 * Do the hard work of removing an element from the buddy allocator.
829 * Call me with the zone->lock already held.
831 static struct page *__rmqueue(struct zone *zone, unsigned int order,
836 page = __rmqueue_smallest(zone, order, migratetype);
839 page = __rmqueue_fallback(zone, order, migratetype);
845 * Obtain a specified number of elements from the buddy allocator, all under
846 * a single hold of the lock, for efficiency. Add them to the supplied list.
847 * Returns the number of new pages which were placed at *list.
849 static int rmqueue_bulk(struct zone *zone, unsigned int order,
850 unsigned long count, struct list_head *list,
855 spin_lock(&zone->lock);
856 for (i = 0; i < count; ++i) {
857 struct page *page = __rmqueue(zone, order, migratetype);
858 if (unlikely(page == NULL))
862 * Split buddy pages returned by expand() are received here
863 * in physical page order. The page is added to the callers and
864 * list and the list head then moves forward. From the callers
865 * perspective, the linked list is ordered by page number in
866 * some conditions. This is useful for IO devices that can
867 * merge IO requests if the physical pages are ordered
870 list_add(&page->lru, list);
871 set_page_private(page, migratetype);
874 spin_unlock(&zone->lock);
880 * Called from the vmstat counter updater to drain pagesets of this
881 * currently executing processor on remote nodes after they have
884 * Note that this function must be called with the thread pinned to
885 * a single processor.
887 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
892 local_irq_save(flags);
893 if (pcp->count >= pcp->batch)
894 to_drain = pcp->batch;
896 to_drain = pcp->count;
897 free_pages_bulk(zone, to_drain, &pcp->list, 0);
898 pcp->count -= to_drain;
899 local_irq_restore(flags);
904 * Drain pages of the indicated processor.
906 * The processor must either be the current processor and the
907 * thread pinned to the current processor or a processor that
910 static void drain_pages(unsigned int cpu)
915 for_each_zone(zone) {
916 struct per_cpu_pageset *pset;
917 struct per_cpu_pages *pcp;
919 if (!populated_zone(zone))
922 pset = zone_pcp(zone, cpu);
925 local_irq_save(flags);
926 free_pages_bulk(zone, pcp->count, &pcp->list, 0);
928 local_irq_restore(flags);
933 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
935 void drain_local_pages(void *arg)
937 drain_pages(smp_processor_id());
941 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
943 void drain_all_pages(void)
945 on_each_cpu(drain_local_pages, NULL, 0, 1);
948 #ifdef CONFIG_HIBERNATION
950 void mark_free_pages(struct zone *zone)
952 unsigned long pfn, max_zone_pfn;
955 struct list_head *curr;
957 if (!zone->spanned_pages)
960 spin_lock_irqsave(&zone->lock, flags);
962 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
963 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
964 if (pfn_valid(pfn)) {
965 struct page *page = pfn_to_page(pfn);
967 if (!swsusp_page_is_forbidden(page))
968 swsusp_unset_page_free(page);
971 for_each_migratetype_order(order, t) {
972 list_for_each(curr, &zone->free_area[order].free_list[t]) {
975 pfn = page_to_pfn(list_entry(curr, struct page, lru));
976 for (i = 0; i < (1UL << order); i++)
977 swsusp_set_page_free(pfn_to_page(pfn + i));
980 spin_unlock_irqrestore(&zone->lock, flags);
982 #endif /* CONFIG_PM */
985 * Free a 0-order page
987 static void free_hot_cold_page(struct page *page, int cold)
989 struct zone *zone = page_zone(page);
990 struct per_cpu_pages *pcp;
994 page->mapping = NULL;
995 if (free_pages_check(page))
998 if (!PageHighMem(page))
999 debug_check_no_locks_freed(page_address(page), PAGE_SIZE);
1000 arch_free_page(page, 0);
1001 kernel_map_pages(page, 1, 0);
1003 pcp = &zone_pcp(zone, get_cpu())->pcp;
1004 local_irq_save(flags);
1005 __count_vm_event(PGFREE);
1007 list_add_tail(&page->lru, &pcp->list);
1009 list_add(&page->lru, &pcp->list);
1010 set_page_private(page, get_pageblock_migratetype(page));
1012 if (pcp->count >= pcp->high) {
1013 free_pages_bulk(zone, pcp->batch, &pcp->list, 0);
1014 pcp->count -= pcp->batch;
1016 local_irq_restore(flags);
1020 void free_hot_page(struct page *page)
1022 free_hot_cold_page(page, 0);
1025 void free_cold_page(struct page *page)
1027 free_hot_cold_page(page, 1);
1031 * split_page takes a non-compound higher-order page, and splits it into
1032 * n (1<<order) sub-pages: page[0..n]
1033 * Each sub-page must be freed individually.
1035 * Note: this is probably too low level an operation for use in drivers.
1036 * Please consult with lkml before using this in your driver.
1038 void split_page(struct page *page, unsigned int order)
1042 VM_BUG_ON(PageCompound(page));
1043 VM_BUG_ON(!page_count(page));
1044 for (i = 1; i < (1 << order); i++)
1045 set_page_refcounted(page + i);
1049 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1050 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1053 static struct page *buffered_rmqueue(struct zonelist *zonelist,
1054 struct zone *zone, int order, gfp_t gfp_flags)
1056 unsigned long flags;
1058 int cold = !!(gfp_flags & __GFP_COLD);
1060 int migratetype = allocflags_to_migratetype(gfp_flags);
1064 if (likely(order == 0)) {
1065 struct per_cpu_pages *pcp;
1067 pcp = &zone_pcp(zone, cpu)->pcp;
1068 local_irq_save(flags);
1070 pcp->count = rmqueue_bulk(zone, 0,
1071 pcp->batch, &pcp->list, migratetype);
1072 if (unlikely(!pcp->count))
1076 /* Find a page of the appropriate migrate type */
1078 list_for_each_entry_reverse(page, &pcp->list, lru)
1079 if (page_private(page) == migratetype)
1082 list_for_each_entry(page, &pcp->list, lru)
1083 if (page_private(page) == migratetype)
1087 /* Allocate more to the pcp list if necessary */
1088 if (unlikely(&page->lru == &pcp->list)) {
1089 pcp->count += rmqueue_bulk(zone, 0,
1090 pcp->batch, &pcp->list, migratetype);
1091 page = list_entry(pcp->list.next, struct page, lru);
1094 list_del(&page->lru);
1097 spin_lock_irqsave(&zone->lock, flags);
1098 page = __rmqueue(zone, order, migratetype);
1099 spin_unlock(&zone->lock);
1104 __count_zone_vm_events(PGALLOC, zone, 1 << order);
1105 zone_statistics(zonelist, zone);
1106 local_irq_restore(flags);
1109 VM_BUG_ON(bad_range(zone, page));
1110 if (prep_new_page(page, order, gfp_flags))
1115 local_irq_restore(flags);
1120 #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
1121 #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
1122 #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
1123 #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
1124 #define ALLOC_HARDER 0x10 /* try to alloc harder */
1125 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1126 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
1128 #ifdef CONFIG_FAIL_PAGE_ALLOC
1130 static struct fail_page_alloc_attr {
1131 struct fault_attr attr;
1133 u32 ignore_gfp_highmem;
1134 u32 ignore_gfp_wait;
1137 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1139 struct dentry *ignore_gfp_highmem_file;
1140 struct dentry *ignore_gfp_wait_file;
1141 struct dentry *min_order_file;
1143 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1145 } fail_page_alloc = {
1146 .attr = FAULT_ATTR_INITIALIZER,
1147 .ignore_gfp_wait = 1,
1148 .ignore_gfp_highmem = 1,
1152 static int __init setup_fail_page_alloc(char *str)
1154 return setup_fault_attr(&fail_page_alloc.attr, str);
1156 __setup("fail_page_alloc=", setup_fail_page_alloc);
1158 static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1160 if (order < fail_page_alloc.min_order)
1162 if (gfp_mask & __GFP_NOFAIL)
1164 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
1166 if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
1169 return should_fail(&fail_page_alloc.attr, 1 << order);
1172 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1174 static int __init fail_page_alloc_debugfs(void)
1176 mode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
1180 err = init_fault_attr_dentries(&fail_page_alloc.attr,
1184 dir = fail_page_alloc.attr.dentries.dir;
1186 fail_page_alloc.ignore_gfp_wait_file =
1187 debugfs_create_bool("ignore-gfp-wait", mode, dir,
1188 &fail_page_alloc.ignore_gfp_wait);
1190 fail_page_alloc.ignore_gfp_highmem_file =
1191 debugfs_create_bool("ignore-gfp-highmem", mode, dir,
1192 &fail_page_alloc.ignore_gfp_highmem);
1193 fail_page_alloc.min_order_file =
1194 debugfs_create_u32("min-order", mode, dir,
1195 &fail_page_alloc.min_order);
1197 if (!fail_page_alloc.ignore_gfp_wait_file ||
1198 !fail_page_alloc.ignore_gfp_highmem_file ||
1199 !fail_page_alloc.min_order_file) {
1201 debugfs_remove(fail_page_alloc.ignore_gfp_wait_file);
1202 debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file);
1203 debugfs_remove(fail_page_alloc.min_order_file);
1204 cleanup_fault_attr_dentries(&fail_page_alloc.attr);
1210 late_initcall(fail_page_alloc_debugfs);
1212 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1214 #else /* CONFIG_FAIL_PAGE_ALLOC */
1216 static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1221 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1224 * Return 1 if free pages are above 'mark'. This takes into account the order
1225 * of the allocation.
1227 int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1228 int classzone_idx, int alloc_flags)
1230 /* free_pages my go negative - that's OK */
1232 long free_pages = zone_page_state(z, NR_FREE_PAGES) - (1 << order) + 1;
1235 if (alloc_flags & ALLOC_HIGH)
1237 if (alloc_flags & ALLOC_HARDER)
1240 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
1242 for (o = 0; o < order; o++) {
1243 /* At the next order, this order's pages become unavailable */
1244 free_pages -= z->free_area[o].nr_free << o;
1246 /* Require fewer higher order pages to be free */
1249 if (free_pages <= min)
1257 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1258 * skip over zones that are not allowed by the cpuset, or that have
1259 * been recently (in last second) found to be nearly full. See further
1260 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1261 * that have to skip over a lot of full or unallowed zones.
1263 * If the zonelist cache is present in the passed in zonelist, then
1264 * returns a pointer to the allowed node mask (either the current
1265 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
1267 * If the zonelist cache is not available for this zonelist, does
1268 * nothing and returns NULL.
1270 * If the fullzones BITMAP in the zonelist cache is stale (more than
1271 * a second since last zap'd) then we zap it out (clear its bits.)
1273 * We hold off even calling zlc_setup, until after we've checked the
1274 * first zone in the zonelist, on the theory that most allocations will
1275 * be satisfied from that first zone, so best to examine that zone as
1276 * quickly as we can.
1278 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1280 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1281 nodemask_t *allowednodes; /* zonelist_cache approximation */
1283 zlc = zonelist->zlcache_ptr;
1287 if (time_after(jiffies, zlc->last_full_zap + HZ)) {
1288 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1289 zlc->last_full_zap = jiffies;
1292 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1293 &cpuset_current_mems_allowed :
1294 &node_states[N_HIGH_MEMORY];
1295 return allowednodes;
1299 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1300 * if it is worth looking at further for free memory:
1301 * 1) Check that the zone isn't thought to be full (doesn't have its
1302 * bit set in the zonelist_cache fullzones BITMAP).
1303 * 2) Check that the zones node (obtained from the zonelist_cache
1304 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1305 * Return true (non-zero) if zone is worth looking at further, or
1306 * else return false (zero) if it is not.
1308 * This check -ignores- the distinction between various watermarks,
1309 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1310 * found to be full for any variation of these watermarks, it will
1311 * be considered full for up to one second by all requests, unless
1312 * we are so low on memory on all allowed nodes that we are forced
1313 * into the second scan of the zonelist.
1315 * In the second scan we ignore this zonelist cache and exactly
1316 * apply the watermarks to all zones, even it is slower to do so.
1317 * We are low on memory in the second scan, and should leave no stone
1318 * unturned looking for a free page.
1320 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zone **z,
1321 nodemask_t *allowednodes)
1323 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1324 int i; /* index of *z in zonelist zones */
1325 int n; /* node that zone *z is on */
1327 zlc = zonelist->zlcache_ptr;
1331 i = z - zonelist->zones;
1334 /* This zone is worth trying if it is allowed but not full */
1335 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1339 * Given 'z' scanning a zonelist, set the corresponding bit in
1340 * zlc->fullzones, so that subsequent attempts to allocate a page
1341 * from that zone don't waste time re-examining it.
1343 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zone **z)
1345 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1346 int i; /* index of *z in zonelist zones */
1348 zlc = zonelist->zlcache_ptr;
1352 i = z - zonelist->zones;
1354 set_bit(i, zlc->fullzones);
1357 #else /* CONFIG_NUMA */
1359 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1364 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zone **z,
1365 nodemask_t *allowednodes)
1370 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zone **z)
1373 #endif /* CONFIG_NUMA */
1376 * get_page_from_freelist goes through the zonelist trying to allocate
1379 static struct page *
1380 get_page_from_freelist(gfp_t gfp_mask, unsigned int order,
1381 struct zonelist *zonelist, int alloc_flags)
1384 struct page *page = NULL;
1385 int classzone_idx = zone_idx(zonelist->zones[0]);
1387 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1388 int zlc_active = 0; /* set if using zonelist_cache */
1389 int did_zlc_setup = 0; /* just call zlc_setup() one time */
1390 enum zone_type highest_zoneidx = -1; /* Gets set for policy zonelists */
1394 * Scan zonelist, looking for a zone with enough free.
1395 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1397 z = zonelist->zones;
1401 * In NUMA, this could be a policy zonelist which contains
1402 * zones that may not be allowed by the current gfp_mask.
1403 * Check the zone is allowed by the current flags
1405 if (unlikely(alloc_should_filter_zonelist(zonelist))) {
1406 if (highest_zoneidx == -1)
1407 highest_zoneidx = gfp_zone(gfp_mask);
1408 if (zone_idx(*z) > highest_zoneidx)
1412 if (NUMA_BUILD && zlc_active &&
1413 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1416 if ((alloc_flags & ALLOC_CPUSET) &&
1417 !cpuset_zone_allowed_softwall(zone, gfp_mask))
1420 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
1422 if (alloc_flags & ALLOC_WMARK_MIN)
1423 mark = zone->pages_min;
1424 else if (alloc_flags & ALLOC_WMARK_LOW)
1425 mark = zone->pages_low;
1427 mark = zone->pages_high;
1428 if (!zone_watermark_ok(zone, order, mark,
1429 classzone_idx, alloc_flags)) {
1430 if (!zone_reclaim_mode ||
1431 !zone_reclaim(zone, gfp_mask, order))
1432 goto this_zone_full;
1436 page = buffered_rmqueue(zonelist, zone, order, gfp_mask);
1441 zlc_mark_zone_full(zonelist, z);
1443 if (NUMA_BUILD && !did_zlc_setup) {
1444 /* we do zlc_setup after the first zone is tried */
1445 allowednodes = zlc_setup(zonelist, alloc_flags);
1449 } while (*(++z) != NULL);
1451 if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) {
1452 /* Disable zlc cache for second zonelist scan */
1460 * This is the 'heart' of the zoned buddy allocator.
1463 __alloc_pages(gfp_t gfp_mask, unsigned int order,
1464 struct zonelist *zonelist)
1466 const gfp_t wait = gfp_mask & __GFP_WAIT;
1469 struct reclaim_state reclaim_state;
1470 struct task_struct *p = current;
1473 int did_some_progress;
1475 might_sleep_if(wait);
1477 if (should_fail_alloc_page(gfp_mask, order))
1481 z = zonelist->zones; /* the list of zones suitable for gfp_mask */
1483 if (unlikely(*z == NULL)) {
1485 * Happens if we have an empty zonelist as a result of
1486 * GFP_THISNODE being used on a memoryless node
1491 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
1492 zonelist, ALLOC_WMARK_LOW|ALLOC_CPUSET);
1497 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1498 * __GFP_NOWARN set) should not cause reclaim since the subsystem
1499 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1500 * using a larger set of nodes after it has established that the
1501 * allowed per node queues are empty and that nodes are
1504 if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
1507 for (z = zonelist->zones; *z; z++)
1508 wakeup_kswapd(*z, order);
1511 * OK, we're below the kswapd watermark and have kicked background
1512 * reclaim. Now things get more complex, so set up alloc_flags according
1513 * to how we want to proceed.
1515 * The caller may dip into page reserves a bit more if the caller
1516 * cannot run direct reclaim, or if the caller has realtime scheduling
1517 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1518 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1520 alloc_flags = ALLOC_WMARK_MIN;
1521 if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait)
1522 alloc_flags |= ALLOC_HARDER;
1523 if (gfp_mask & __GFP_HIGH)
1524 alloc_flags |= ALLOC_HIGH;
1526 alloc_flags |= ALLOC_CPUSET;
1529 * Go through the zonelist again. Let __GFP_HIGH and allocations
1530 * coming from realtime tasks go deeper into reserves.
1532 * This is the last chance, in general, before the goto nopage.
1533 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1534 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1536 page = get_page_from_freelist(gfp_mask, order, zonelist, alloc_flags);
1540 /* This allocation should allow future memory freeing. */
1543 if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE)))
1544 && !in_interrupt()) {
1545 if (!(gfp_mask & __GFP_NOMEMALLOC)) {
1547 /* go through the zonelist yet again, ignoring mins */
1548 page = get_page_from_freelist(gfp_mask, order,
1549 zonelist, ALLOC_NO_WATERMARKS);
1552 if (gfp_mask & __GFP_NOFAIL) {
1553 congestion_wait(WRITE, HZ/50);
1560 /* Atomic allocations - we can't balance anything */
1566 /* We now go into synchronous reclaim */
1567 cpuset_memory_pressure_bump();
1568 p->flags |= PF_MEMALLOC;
1569 reclaim_state.reclaimed_slab = 0;
1570 p->reclaim_state = &reclaim_state;
1572 did_some_progress = try_to_free_pages(zonelist, order, gfp_mask);
1574 p->reclaim_state = NULL;
1575 p->flags &= ~PF_MEMALLOC;
1582 if (likely(did_some_progress)) {
1583 page = get_page_from_freelist(gfp_mask, order,
1584 zonelist, alloc_flags);
1587 } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
1588 if (!try_set_zone_oom(zonelist)) {
1589 schedule_timeout_uninterruptible(1);
1594 * Go through the zonelist yet one more time, keep
1595 * very high watermark here, this is only to catch
1596 * a parallel oom killing, we must fail if we're still
1597 * under heavy pressure.
1599 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
1600 zonelist, ALLOC_WMARK_HIGH|ALLOC_CPUSET);
1602 clear_zonelist_oom(zonelist);
1606 /* The OOM killer will not help higher order allocs so fail */
1607 if (order > PAGE_ALLOC_COSTLY_ORDER) {
1608 clear_zonelist_oom(zonelist);
1612 out_of_memory(zonelist, gfp_mask, order);
1613 clear_zonelist_oom(zonelist);
1618 * Don't let big-order allocations loop unless the caller explicitly
1619 * requests that. Wait for some write requests to complete then retry.
1621 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
1622 * <= 3, but that may not be true in other implementations.
1625 if (!(gfp_mask & __GFP_NORETRY)) {
1626 if ((order <= PAGE_ALLOC_COSTLY_ORDER) ||
1627 (gfp_mask & __GFP_REPEAT))
1629 if (gfp_mask & __GFP_NOFAIL)
1633 congestion_wait(WRITE, HZ/50);
1638 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
1639 printk(KERN_WARNING "%s: page allocation failure."
1640 " order:%d, mode:0x%x\n",
1641 p->comm, order, gfp_mask);
1649 EXPORT_SYMBOL(__alloc_pages);
1652 * Common helper functions.
1654 unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
1657 page = alloc_pages(gfp_mask, order);
1660 return (unsigned long) page_address(page);
1663 EXPORT_SYMBOL(__get_free_pages);
1665 unsigned long get_zeroed_page(gfp_t gfp_mask)
1670 * get_zeroed_page() returns a 32-bit address, which cannot represent
1673 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
1675 page = alloc_pages(gfp_mask | __GFP_ZERO, 0);
1677 return (unsigned long) page_address(page);
1681 EXPORT_SYMBOL(get_zeroed_page);
1683 void __pagevec_free(struct pagevec *pvec)
1685 int i = pagevec_count(pvec);
1688 free_hot_cold_page(pvec->pages[i], pvec->cold);
1691 void __free_pages(struct page *page, unsigned int order)
1693 if (put_page_testzero(page)) {
1695 free_hot_page(page);
1697 __free_pages_ok(page, order);
1701 EXPORT_SYMBOL(__free_pages);
1703 void free_pages(unsigned long addr, unsigned int order)
1706 VM_BUG_ON(!virt_addr_valid((void *)addr));
1707 __free_pages(virt_to_page((void *)addr), order);
1711 EXPORT_SYMBOL(free_pages);
1713 static unsigned int nr_free_zone_pages(int offset)
1715 /* Just pick one node, since fallback list is circular */
1716 unsigned int sum = 0;
1718 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
1719 struct zone **zonep = zonelist->zones;
1722 for (zone = *zonep++; zone; zone = *zonep++) {
1723 unsigned long size = zone->present_pages;
1724 unsigned long high = zone->pages_high;
1733 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1735 unsigned int nr_free_buffer_pages(void)
1737 return nr_free_zone_pages(gfp_zone(GFP_USER));
1739 EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
1742 * Amount of free RAM allocatable within all zones
1744 unsigned int nr_free_pagecache_pages(void)
1746 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
1749 static inline void show_node(struct zone *zone)
1752 printk("Node %d ", zone_to_nid(zone));
1755 void si_meminfo(struct sysinfo *val)
1757 val->totalram = totalram_pages;
1759 val->freeram = global_page_state(NR_FREE_PAGES);
1760 val->bufferram = nr_blockdev_pages();
1761 val->totalhigh = totalhigh_pages;
1762 val->freehigh = nr_free_highpages();
1763 val->mem_unit = PAGE_SIZE;
1766 EXPORT_SYMBOL(si_meminfo);
1769 void si_meminfo_node(struct sysinfo *val, int nid)
1771 pg_data_t *pgdat = NODE_DATA(nid);
1773 val->totalram = pgdat->node_present_pages;
1774 val->freeram = node_page_state(nid, NR_FREE_PAGES);
1775 #ifdef CONFIG_HIGHMEM
1776 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1777 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
1783 val->mem_unit = PAGE_SIZE;
1787 #define K(x) ((x) << (PAGE_SHIFT-10))
1790 * Show free area list (used inside shift_scroll-lock stuff)
1791 * We also calculate the percentage fragmentation. We do this by counting the
1792 * memory on each free list with the exception of the first item on the list.
1794 void show_free_areas(void)
1799 for_each_zone(zone) {
1800 if (!populated_zone(zone))
1804 printk("%s per-cpu:\n", zone->name);
1806 for_each_online_cpu(cpu) {
1807 struct per_cpu_pageset *pageset;
1809 pageset = zone_pcp(zone, cpu);
1811 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
1812 cpu, pageset->pcp.high,
1813 pageset->pcp.batch, pageset->pcp.count);
1817 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu unstable:%lu\n"
1818 " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
1819 global_page_state(NR_ACTIVE),
1820 global_page_state(NR_INACTIVE),
1821 global_page_state(NR_FILE_DIRTY),
1822 global_page_state(NR_WRITEBACK),
1823 global_page_state(NR_UNSTABLE_NFS),
1824 global_page_state(NR_FREE_PAGES),
1825 global_page_state(NR_SLAB_RECLAIMABLE) +
1826 global_page_state(NR_SLAB_UNRECLAIMABLE),
1827 global_page_state(NR_FILE_MAPPED),
1828 global_page_state(NR_PAGETABLE),
1829 global_page_state(NR_BOUNCE));
1831 for_each_zone(zone) {
1834 if (!populated_zone(zone))
1846 " pages_scanned:%lu"
1847 " all_unreclaimable? %s"
1850 K(zone_page_state(zone, NR_FREE_PAGES)),
1853 K(zone->pages_high),
1854 K(zone_page_state(zone, NR_ACTIVE)),
1855 K(zone_page_state(zone, NR_INACTIVE)),
1856 K(zone->present_pages),
1857 zone->pages_scanned,
1858 (zone_is_all_unreclaimable(zone) ? "yes" : "no")
1860 printk("lowmem_reserve[]:");
1861 for (i = 0; i < MAX_NR_ZONES; i++)
1862 printk(" %lu", zone->lowmem_reserve[i]);
1866 for_each_zone(zone) {
1867 unsigned long nr[MAX_ORDER], flags, order, total = 0;
1869 if (!populated_zone(zone))
1873 printk("%s: ", zone->name);
1875 spin_lock_irqsave(&zone->lock, flags);
1876 for (order = 0; order < MAX_ORDER; order++) {
1877 nr[order] = zone->free_area[order].nr_free;
1878 total += nr[order] << order;
1880 spin_unlock_irqrestore(&zone->lock, flags);
1881 for (order = 0; order < MAX_ORDER; order++)
1882 printk("%lu*%lukB ", nr[order], K(1UL) << order);
1883 printk("= %lukB\n", K(total));
1886 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
1888 show_swap_cache_info();
1892 * Builds allocation fallback zone lists.
1894 * Add all populated zones of a node to the zonelist.
1896 static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
1897 int nr_zones, enum zone_type zone_type)
1901 BUG_ON(zone_type >= MAX_NR_ZONES);
1906 zone = pgdat->node_zones + zone_type;
1907 if (populated_zone(zone)) {
1908 zonelist->zones[nr_zones++] = zone;
1909 check_highest_zone(zone_type);
1912 } while (zone_type);
1919 * 0 = automatic detection of better ordering.
1920 * 1 = order by ([node] distance, -zonetype)
1921 * 2 = order by (-zonetype, [node] distance)
1923 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
1924 * the same zonelist. So only NUMA can configure this param.
1926 #define ZONELIST_ORDER_DEFAULT 0
1927 #define ZONELIST_ORDER_NODE 1
1928 #define ZONELIST_ORDER_ZONE 2
1930 /* zonelist order in the kernel.
1931 * set_zonelist_order() will set this to NODE or ZONE.
1933 static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
1934 static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
1938 /* The value user specified ....changed by config */
1939 static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
1940 /* string for sysctl */
1941 #define NUMA_ZONELIST_ORDER_LEN 16
1942 char numa_zonelist_order[16] = "default";
1945 * interface for configure zonelist ordering.
1946 * command line option "numa_zonelist_order"
1947 * = "[dD]efault - default, automatic configuration.
1948 * = "[nN]ode - order by node locality, then by zone within node
1949 * = "[zZ]one - order by zone, then by locality within zone
1952 static int __parse_numa_zonelist_order(char *s)
1954 if (*s == 'd' || *s == 'D') {
1955 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
1956 } else if (*s == 'n' || *s == 'N') {
1957 user_zonelist_order = ZONELIST_ORDER_NODE;
1958 } else if (*s == 'z' || *s == 'Z') {
1959 user_zonelist_order = ZONELIST_ORDER_ZONE;
1962 "Ignoring invalid numa_zonelist_order value: "
1969 static __init int setup_numa_zonelist_order(char *s)
1972 return __parse_numa_zonelist_order(s);
1975 early_param("numa_zonelist_order", setup_numa_zonelist_order);
1978 * sysctl handler for numa_zonelist_order
1980 int numa_zonelist_order_handler(ctl_table *table, int write,
1981 struct file *file, void __user *buffer, size_t *length,
1984 char saved_string[NUMA_ZONELIST_ORDER_LEN];
1988 strncpy(saved_string, (char*)table->data,
1989 NUMA_ZONELIST_ORDER_LEN);
1990 ret = proc_dostring(table, write, file, buffer, length, ppos);
1994 int oldval = user_zonelist_order;
1995 if (__parse_numa_zonelist_order((char*)table->data)) {
1997 * bogus value. restore saved string
1999 strncpy((char*)table->data, saved_string,
2000 NUMA_ZONELIST_ORDER_LEN);
2001 user_zonelist_order = oldval;
2002 } else if (oldval != user_zonelist_order)
2003 build_all_zonelists();
2009 #define MAX_NODE_LOAD (num_online_nodes())
2010 static int node_load[MAX_NUMNODES];
2013 * find_next_best_node - find the next node that should appear in a given node's fallback list
2014 * @node: node whose fallback list we're appending
2015 * @used_node_mask: nodemask_t of already used nodes
2017 * We use a number of factors to determine which is the next node that should
2018 * appear on a given node's fallback list. The node should not have appeared
2019 * already in @node's fallback list, and it should be the next closest node
2020 * according to the distance array (which contains arbitrary distance values
2021 * from each node to each node in the system), and should also prefer nodes
2022 * with no CPUs, since presumably they'll have very little allocation pressure
2023 * on them otherwise.
2024 * It returns -1 if no node is found.
2026 static int find_next_best_node(int node, nodemask_t *used_node_mask)
2029 int min_val = INT_MAX;
2031 node_to_cpumask_ptr(tmp, 0);
2033 /* Use the local node if we haven't already */
2034 if (!node_isset(node, *used_node_mask)) {
2035 node_set(node, *used_node_mask);
2039 for_each_node_state(n, N_HIGH_MEMORY) {
2041 /* Don't want a node to appear more than once */
2042 if (node_isset(n, *used_node_mask))
2045 /* Use the distance array to find the distance */
2046 val = node_distance(node, n);
2048 /* Penalize nodes under us ("prefer the next node") */
2051 /* Give preference to headless and unused nodes */
2052 node_to_cpumask_ptr_next(tmp, n);
2053 if (!cpus_empty(*tmp))
2054 val += PENALTY_FOR_NODE_WITH_CPUS;
2056 /* Slight preference for less loaded node */
2057 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
2058 val += node_load[n];
2060 if (val < min_val) {
2067 node_set(best_node, *used_node_mask);
2074 * Build zonelists ordered by node and zones within node.
2075 * This results in maximum locality--normal zone overflows into local
2076 * DMA zone, if any--but risks exhausting DMA zone.
2078 static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
2082 struct zonelist *zonelist;
2084 for (i = 0; i < MAX_NR_ZONES; i++) {
2085 zonelist = pgdat->node_zonelists + i;
2086 for (j = 0; zonelist->zones[j] != NULL; j++)
2088 j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
2089 zonelist->zones[j] = NULL;
2094 * Build gfp_thisnode zonelists
2096 static void build_thisnode_zonelists(pg_data_t *pgdat)
2100 struct zonelist *zonelist;
2102 for (i = 0; i < MAX_NR_ZONES; i++) {
2103 zonelist = pgdat->node_zonelists + MAX_NR_ZONES + i;
2104 j = build_zonelists_node(pgdat, zonelist, 0, i);
2105 zonelist->zones[j] = NULL;
2110 * Build zonelists ordered by zone and nodes within zones.
2111 * This results in conserving DMA zone[s] until all Normal memory is
2112 * exhausted, but results in overflowing to remote node while memory
2113 * may still exist in local DMA zone.
2115 static int node_order[MAX_NUMNODES];
2117 static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
2121 int zone_type; /* needs to be signed */
2123 struct zonelist *zonelist;
2125 for (i = 0; i < MAX_NR_ZONES; i++) {
2126 zonelist = pgdat->node_zonelists + i;
2128 for (zone_type = i; zone_type >= 0; zone_type--) {
2129 for (j = 0; j < nr_nodes; j++) {
2130 node = node_order[j];
2131 z = &NODE_DATA(node)->node_zones[zone_type];
2132 if (populated_zone(z)) {
2133 zonelist->zones[pos++] = z;
2134 check_highest_zone(zone_type);
2138 zonelist->zones[pos] = NULL;
2142 static int default_zonelist_order(void)
2145 unsigned long low_kmem_size,total_size;
2149 * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem.
2150 * If they are really small and used heavily, the system can fall
2151 * into OOM very easily.
2152 * This function detect ZONE_DMA/DMA32 size and confgigures zone order.
2154 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2157 for_each_online_node(nid) {
2158 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2159 z = &NODE_DATA(nid)->node_zones[zone_type];
2160 if (populated_zone(z)) {
2161 if (zone_type < ZONE_NORMAL)
2162 low_kmem_size += z->present_pages;
2163 total_size += z->present_pages;
2167 if (!low_kmem_size || /* there are no DMA area. */
2168 low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
2169 return ZONELIST_ORDER_NODE;
2171 * look into each node's config.
2172 * If there is a node whose DMA/DMA32 memory is very big area on
2173 * local memory, NODE_ORDER may be suitable.
2175 average_size = total_size /
2176 (nodes_weight(node_states[N_HIGH_MEMORY]) + 1);
2177 for_each_online_node(nid) {
2180 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2181 z = &NODE_DATA(nid)->node_zones[zone_type];
2182 if (populated_zone(z)) {
2183 if (zone_type < ZONE_NORMAL)
2184 low_kmem_size += z->present_pages;
2185 total_size += z->present_pages;
2188 if (low_kmem_size &&
2189 total_size > average_size && /* ignore small node */
2190 low_kmem_size > total_size * 70/100)
2191 return ZONELIST_ORDER_NODE;
2193 return ZONELIST_ORDER_ZONE;
2196 static void set_zonelist_order(void)
2198 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
2199 current_zonelist_order = default_zonelist_order();
2201 current_zonelist_order = user_zonelist_order;
2204 static void build_zonelists(pg_data_t *pgdat)
2208 nodemask_t used_mask;
2209 int local_node, prev_node;
2210 struct zonelist *zonelist;
2211 int order = current_zonelist_order;
2213 /* initialize zonelists */
2214 for (i = 0; i < MAX_ZONELISTS; i++) {
2215 zonelist = pgdat->node_zonelists + i;
2216 zonelist->zones[0] = NULL;
2219 /* NUMA-aware ordering of nodes */
2220 local_node = pgdat->node_id;
2221 load = num_online_nodes();
2222 prev_node = local_node;
2223 nodes_clear(used_mask);
2225 memset(node_load, 0, sizeof(node_load));
2226 memset(node_order, 0, sizeof(node_order));
2229 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
2230 int distance = node_distance(local_node, node);
2233 * If another node is sufficiently far away then it is better
2234 * to reclaim pages in a zone before going off node.
2236 if (distance > RECLAIM_DISTANCE)
2237 zone_reclaim_mode = 1;
2240 * We don't want to pressure a particular node.
2241 * So adding penalty to the first node in same
2242 * distance group to make it round-robin.
2244 if (distance != node_distance(local_node, prev_node))
2245 node_load[node] = load;
2249 if (order == ZONELIST_ORDER_NODE)
2250 build_zonelists_in_node_order(pgdat, node);
2252 node_order[j++] = node; /* remember order */
2255 if (order == ZONELIST_ORDER_ZONE) {
2256 /* calculate node order -- i.e., DMA last! */
2257 build_zonelists_in_zone_order(pgdat, j);
2260 build_thisnode_zonelists(pgdat);
2263 /* Construct the zonelist performance cache - see further mmzone.h */
2264 static void build_zonelist_cache(pg_data_t *pgdat)
2268 for (i = 0; i < MAX_NR_ZONES; i++) {
2269 struct zonelist *zonelist;
2270 struct zonelist_cache *zlc;
2273 zonelist = pgdat->node_zonelists + i;
2274 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
2275 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
2276 for (z = zonelist->zones; *z; z++)
2277 zlc->z_to_n[z - zonelist->zones] = zone_to_nid(*z);
2282 #else /* CONFIG_NUMA */
2284 static void set_zonelist_order(void)
2286 current_zonelist_order = ZONELIST_ORDER_ZONE;
2289 static void build_zonelists(pg_data_t *pgdat)
2291 int node, local_node;
2294 local_node = pgdat->node_id;
2295 for (i = 0; i < MAX_NR_ZONES; i++) {
2296 struct zonelist *zonelist;
2298 zonelist = pgdat->node_zonelists + i;
2300 j = build_zonelists_node(pgdat, zonelist, 0, i);
2302 * Now we build the zonelist so that it contains the zones
2303 * of all the other nodes.
2304 * We don't want to pressure a particular node, so when
2305 * building the zones for node N, we make sure that the
2306 * zones coming right after the local ones are those from
2307 * node N+1 (modulo N)
2309 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
2310 if (!node_online(node))
2312 j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
2314 for (node = 0; node < local_node; node++) {
2315 if (!node_online(node))
2317 j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
2320 zonelist->zones[j] = NULL;
2324 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
2325 static void build_zonelist_cache(pg_data_t *pgdat)
2329 for (i = 0; i < MAX_NR_ZONES; i++)
2330 pgdat->node_zonelists[i].zlcache_ptr = NULL;
2333 #endif /* CONFIG_NUMA */
2335 /* return values int ....just for stop_machine_run() */
2336 static int __build_all_zonelists(void *dummy)
2340 for_each_online_node(nid) {
2341 pg_data_t *pgdat = NODE_DATA(nid);
2343 build_zonelists(pgdat);
2344 build_zonelist_cache(pgdat);
2349 void build_all_zonelists(void)
2351 set_zonelist_order();
2353 if (system_state == SYSTEM_BOOTING) {
2354 __build_all_zonelists(NULL);
2355 cpuset_init_current_mems_allowed();
2357 /* we have to stop all cpus to guarantee there is no user
2359 stop_machine_run(__build_all_zonelists, NULL, NR_CPUS);
2360 /* cpuset refresh routine should be here */
2362 vm_total_pages = nr_free_pagecache_pages();
2364 * Disable grouping by mobility if the number of pages in the
2365 * system is too low to allow the mechanism to work. It would be
2366 * more accurate, but expensive to check per-zone. This check is
2367 * made on memory-hotadd so a system can start with mobility
2368 * disabled and enable it later
2370 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
2371 page_group_by_mobility_disabled = 1;
2373 page_group_by_mobility_disabled = 0;
2375 printk("Built %i zonelists in %s order, mobility grouping %s. "
2376 "Total pages: %ld\n",
2378 zonelist_order_name[current_zonelist_order],
2379 page_group_by_mobility_disabled ? "off" : "on",
2382 printk("Policy zone: %s\n", zone_names[policy_zone]);
2387 * Helper functions to size the waitqueue hash table.
2388 * Essentially these want to choose hash table sizes sufficiently
2389 * large so that collisions trying to wait on pages are rare.
2390 * But in fact, the number of active page waitqueues on typical
2391 * systems is ridiculously low, less than 200. So this is even
2392 * conservative, even though it seems large.
2394 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
2395 * waitqueues, i.e. the size of the waitq table given the number of pages.
2397 #define PAGES_PER_WAITQUEUE 256
2399 #ifndef CONFIG_MEMORY_HOTPLUG
2400 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
2402 unsigned long size = 1;
2404 pages /= PAGES_PER_WAITQUEUE;
2406 while (size < pages)
2410 * Once we have dozens or even hundreds of threads sleeping
2411 * on IO we've got bigger problems than wait queue collision.
2412 * Limit the size of the wait table to a reasonable size.
2414 size = min(size, 4096UL);
2416 return max(size, 4UL);
2420 * A zone's size might be changed by hot-add, so it is not possible to determine
2421 * a suitable size for its wait_table. So we use the maximum size now.
2423 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
2425 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
2426 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
2427 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
2429 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
2430 * or more by the traditional way. (See above). It equals:
2432 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
2433 * ia64(16K page size) : = ( 8G + 4M)byte.
2434 * powerpc (64K page size) : = (32G +16M)byte.
2436 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
2443 * This is an integer logarithm so that shifts can be used later
2444 * to extract the more random high bits from the multiplicative
2445 * hash function before the remainder is taken.
2447 static inline unsigned long wait_table_bits(unsigned long size)
2452 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
2455 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
2456 * of blocks reserved is based on zone->pages_min. The memory within the
2457 * reserve will tend to store contiguous free pages. Setting min_free_kbytes
2458 * higher will lead to a bigger reserve which will get freed as contiguous
2459 * blocks as reclaim kicks in
2461 static void setup_zone_migrate_reserve(struct zone *zone)
2463 unsigned long start_pfn, pfn, end_pfn;
2465 unsigned long reserve, block_migratetype;
2467 /* Get the start pfn, end pfn and the number of blocks to reserve */
2468 start_pfn = zone->zone_start_pfn;
2469 end_pfn = start_pfn + zone->spanned_pages;
2470 reserve = roundup(zone->pages_min, pageblock_nr_pages) >>
2473 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
2474 if (!pfn_valid(pfn))
2476 page = pfn_to_page(pfn);
2478 /* Blocks with reserved pages will never free, skip them. */
2479 if (PageReserved(page))
2482 block_migratetype = get_pageblock_migratetype(page);
2484 /* If this block is reserved, account for it */
2485 if (reserve > 0 && block_migratetype == MIGRATE_RESERVE) {
2490 /* Suitable for reserving if this block is movable */
2491 if (reserve > 0 && block_migratetype == MIGRATE_MOVABLE) {
2492 set_pageblock_migratetype(page, MIGRATE_RESERVE);
2493 move_freepages_block(zone, page, MIGRATE_RESERVE);
2499 * If the reserve is met and this is a previous reserved block,
2502 if (block_migratetype == MIGRATE_RESERVE) {
2503 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
2504 move_freepages_block(zone, page, MIGRATE_MOVABLE);
2510 * Initially all pages are reserved - free ones are freed
2511 * up by free_all_bootmem() once the early boot process is
2512 * done. Non-atomic initialization, single-pass.
2514 void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
2515 unsigned long start_pfn, enum memmap_context context)
2518 unsigned long end_pfn = start_pfn + size;
2521 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
2523 * There can be holes in boot-time mem_map[]s
2524 * handed to this function. They do not
2525 * exist on hotplugged memory.
2527 if (context == MEMMAP_EARLY) {
2528 if (!early_pfn_valid(pfn))
2530 if (!early_pfn_in_nid(pfn, nid))
2533 page = pfn_to_page(pfn);
2534 set_page_links(page, zone, nid, pfn);
2535 init_page_count(page);
2536 reset_page_mapcount(page);
2537 SetPageReserved(page);
2540 * Mark the block movable so that blocks are reserved for
2541 * movable at startup. This will force kernel allocations
2542 * to reserve their blocks rather than leaking throughout
2543 * the address space during boot when many long-lived
2544 * kernel allocations are made. Later some blocks near
2545 * the start are marked MIGRATE_RESERVE by
2546 * setup_zone_migrate_reserve()
2548 if ((pfn & (pageblock_nr_pages-1)))
2549 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
2551 INIT_LIST_HEAD(&page->lru);
2552 #ifdef WANT_PAGE_VIRTUAL
2553 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
2554 if (!is_highmem_idx(zone))
2555 set_page_address(page, __va(pfn << PAGE_SHIFT));
2560 static void __meminit zone_init_free_lists(struct zone *zone)
2563 for_each_migratetype_order(order, t) {
2564 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
2565 zone->free_area[order].nr_free = 0;
2569 #ifndef __HAVE_ARCH_MEMMAP_INIT
2570 #define memmap_init(size, nid, zone, start_pfn) \
2571 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
2574 static int zone_batchsize(struct zone *zone)
2579 * The per-cpu-pages pools are set to around 1000th of the
2580 * size of the zone. But no more than 1/2 of a meg.
2582 * OK, so we don't know how big the cache is. So guess.
2584 batch = zone->present_pages / 1024;
2585 if (batch * PAGE_SIZE > 512 * 1024)
2586 batch = (512 * 1024) / PAGE_SIZE;
2587 batch /= 4; /* We effectively *= 4 below */
2592 * Clamp the batch to a 2^n - 1 value. Having a power
2593 * of 2 value was found to be more likely to have
2594 * suboptimal cache aliasing properties in some cases.
2596 * For example if 2 tasks are alternately allocating
2597 * batches of pages, one task can end up with a lot
2598 * of pages of one half of the possible page colors
2599 * and the other with pages of the other colors.
2601 batch = (1 << (fls(batch + batch/2)-1)) - 1;
2606 inline void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
2608 struct per_cpu_pages *pcp;
2610 memset(p, 0, sizeof(*p));
2614 pcp->high = 6 * batch;
2615 pcp->batch = max(1UL, 1 * batch);
2616 INIT_LIST_HEAD(&pcp->list);
2620 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
2621 * to the value high for the pageset p.
2624 static void setup_pagelist_highmark(struct per_cpu_pageset *p,
2627 struct per_cpu_pages *pcp;
2631 pcp->batch = max(1UL, high/4);
2632 if ((high/4) > (PAGE_SHIFT * 8))
2633 pcp->batch = PAGE_SHIFT * 8;
2639 * Boot pageset table. One per cpu which is going to be used for all
2640 * zones and all nodes. The parameters will be set in such a way
2641 * that an item put on a list will immediately be handed over to
2642 * the buddy list. This is safe since pageset manipulation is done
2643 * with interrupts disabled.
2645 * Some NUMA counter updates may also be caught by the boot pagesets.
2647 * The boot_pagesets must be kept even after bootup is complete for
2648 * unused processors and/or zones. They do play a role for bootstrapping
2649 * hotplugged processors.
2651 * zoneinfo_show() and maybe other functions do
2652 * not check if the processor is online before following the pageset pointer.
2653 * Other parts of the kernel may not check if the zone is available.
2655 static struct per_cpu_pageset boot_pageset[NR_CPUS];
2658 * Dynamically allocate memory for the
2659 * per cpu pageset array in struct zone.
2661 static int __cpuinit process_zones(int cpu)
2663 struct zone *zone, *dzone;
2664 int node = cpu_to_node(cpu);
2666 node_set_state(node, N_CPU); /* this node has a cpu */
2668 for_each_zone(zone) {
2670 if (!populated_zone(zone))
2673 zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset),
2675 if (!zone_pcp(zone, cpu))
2678 setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone));
2680 if (percpu_pagelist_fraction)
2681 setup_pagelist_highmark(zone_pcp(zone, cpu),
2682 (zone->present_pages / percpu_pagelist_fraction));
2687 for_each_zone(dzone) {
2688 if (!populated_zone(dzone))
2692 kfree(zone_pcp(dzone, cpu));
2693 zone_pcp(dzone, cpu) = NULL;
2698 static inline void free_zone_pagesets(int cpu)
2702 for_each_zone(zone) {
2703 struct per_cpu_pageset *pset = zone_pcp(zone, cpu);
2705 /* Free per_cpu_pageset if it is slab allocated */
2706 if (pset != &boot_pageset[cpu])
2708 zone_pcp(zone, cpu) = NULL;
2712 static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb,
2713 unsigned long action,
2716 int cpu = (long)hcpu;
2717 int ret = NOTIFY_OK;
2720 case CPU_UP_PREPARE:
2721 case CPU_UP_PREPARE_FROZEN:
2722 if (process_zones(cpu))
2725 case CPU_UP_CANCELED:
2726 case CPU_UP_CANCELED_FROZEN:
2728 case CPU_DEAD_FROZEN:
2729 free_zone_pagesets(cpu);
2737 static struct notifier_block __cpuinitdata pageset_notifier =
2738 { &pageset_cpuup_callback, NULL, 0 };
2740 void __init setup_per_cpu_pageset(void)
2744 /* Initialize per_cpu_pageset for cpu 0.
2745 * A cpuup callback will do this for every cpu
2746 * as it comes online
2748 err = process_zones(smp_processor_id());
2750 register_cpu_notifier(&pageset_notifier);
2755 static noinline __init_refok
2756 int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
2759 struct pglist_data *pgdat = zone->zone_pgdat;
2763 * The per-page waitqueue mechanism uses hashed waitqueues
2766 zone->wait_table_hash_nr_entries =
2767 wait_table_hash_nr_entries(zone_size_pages);
2768 zone->wait_table_bits =
2769 wait_table_bits(zone->wait_table_hash_nr_entries);
2770 alloc_size = zone->wait_table_hash_nr_entries
2771 * sizeof(wait_queue_head_t);
2773 if (system_state == SYSTEM_BOOTING) {
2774 zone->wait_table = (wait_queue_head_t *)
2775 alloc_bootmem_node(pgdat, alloc_size);
2778 * This case means that a zone whose size was 0 gets new memory
2779 * via memory hot-add.
2780 * But it may be the case that a new node was hot-added. In
2781 * this case vmalloc() will not be able to use this new node's
2782 * memory - this wait_table must be initialized to use this new
2783 * node itself as well.
2784 * To use this new node's memory, further consideration will be
2787 zone->wait_table = vmalloc(alloc_size);
2789 if (!zone->wait_table)
2792 for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
2793 init_waitqueue_head(zone->wait_table + i);
2798 static __meminit void zone_pcp_init(struct zone *zone)
2801 unsigned long batch = zone_batchsize(zone);
2803 for (cpu = 0; cpu < NR_CPUS; cpu++) {
2805 /* Early boot. Slab allocator not functional yet */
2806 zone_pcp(zone, cpu) = &boot_pageset[cpu];
2807 setup_pageset(&boot_pageset[cpu],0);
2809 setup_pageset(zone_pcp(zone,cpu), batch);
2812 if (zone->present_pages)
2813 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n",
2814 zone->name, zone->present_pages, batch);
2817 __meminit int init_currently_empty_zone(struct zone *zone,
2818 unsigned long zone_start_pfn,
2820 enum memmap_context context)
2822 struct pglist_data *pgdat = zone->zone_pgdat;
2824 ret = zone_wait_table_init(zone, size);
2827 pgdat->nr_zones = zone_idx(zone) + 1;
2829 zone->zone_start_pfn = zone_start_pfn;
2831 memmap_init(size, pgdat->node_id, zone_idx(zone), zone_start_pfn);
2833 zone_init_free_lists(zone);
2838 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2840 * Basic iterator support. Return the first range of PFNs for a node
2841 * Note: nid == MAX_NUMNODES returns first region regardless of node
2843 static int __meminit first_active_region_index_in_nid(int nid)
2847 for (i = 0; i < nr_nodemap_entries; i++)
2848 if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
2855 * Basic iterator support. Return the next active range of PFNs for a node
2856 * Note: nid == MAX_NUMNODES returns next region regardless of node
2858 static int __meminit next_active_region_index_in_nid(int index, int nid)
2860 for (index = index + 1; index < nr_nodemap_entries; index++)
2861 if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
2867 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
2869 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
2870 * Architectures may implement their own version but if add_active_range()
2871 * was used and there are no special requirements, this is a convenient
2874 int __meminit early_pfn_to_nid(unsigned long pfn)
2878 for (i = 0; i < nr_nodemap_entries; i++) {
2879 unsigned long start_pfn = early_node_map[i].start_pfn;
2880 unsigned long end_pfn = early_node_map[i].end_pfn;
2882 if (start_pfn <= pfn && pfn < end_pfn)
2883 return early_node_map[i].nid;
2888 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
2890 /* Basic iterator support to walk early_node_map[] */
2891 #define for_each_active_range_index_in_nid(i, nid) \
2892 for (i = first_active_region_index_in_nid(nid); i != -1; \
2893 i = next_active_region_index_in_nid(i, nid))
2896 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
2897 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
2898 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
2900 * If an architecture guarantees that all ranges registered with
2901 * add_active_ranges() contain no holes and may be freed, this
2902 * this function may be used instead of calling free_bootmem() manually.
2904 void __init free_bootmem_with_active_regions(int nid,
2905 unsigned long max_low_pfn)
2909 for_each_active_range_index_in_nid(i, nid) {
2910 unsigned long size_pages = 0;
2911 unsigned long end_pfn = early_node_map[i].end_pfn;
2913 if (early_node_map[i].start_pfn >= max_low_pfn)
2916 if (end_pfn > max_low_pfn)
2917 end_pfn = max_low_pfn;
2919 size_pages = end_pfn - early_node_map[i].start_pfn;
2920 free_bootmem_node(NODE_DATA(early_node_map[i].nid),
2921 PFN_PHYS(early_node_map[i].start_pfn),
2922 size_pages << PAGE_SHIFT);
2927 * sparse_memory_present_with_active_regions - Call memory_present for each active range
2928 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
2930 * If an architecture guarantees that all ranges registered with
2931 * add_active_ranges() contain no holes and may be freed, this
2932 * function may be used instead of calling memory_present() manually.
2934 void __init sparse_memory_present_with_active_regions(int nid)
2938 for_each_active_range_index_in_nid(i, nid)
2939 memory_present(early_node_map[i].nid,
2940 early_node_map[i].start_pfn,
2941 early_node_map[i].end_pfn);
2945 * push_node_boundaries - Push node boundaries to at least the requested boundary
2946 * @nid: The nid of the node to push the boundary for
2947 * @start_pfn: The start pfn of the node
2948 * @end_pfn: The end pfn of the node
2950 * In reserve-based hot-add, mem_map is allocated that is unused until hotadd
2951 * time. Specifically, on x86_64, SRAT will report ranges that can potentially
2952 * be hotplugged even though no physical memory exists. This function allows
2953 * an arch to push out the node boundaries so mem_map is allocated that can
2956 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
2957 void __init push_node_boundaries(unsigned int nid,
2958 unsigned long start_pfn, unsigned long end_pfn)
2960 printk(KERN_DEBUG "Entering push_node_boundaries(%u, %lu, %lu)\n",
2961 nid, start_pfn, end_pfn);
2963 /* Initialise the boundary for this node if necessary */
2964 if (node_boundary_end_pfn[nid] == 0)
2965 node_boundary_start_pfn[nid] = -1UL;
2967 /* Update the boundaries */
2968 if (node_boundary_start_pfn[nid] > start_pfn)
2969 node_boundary_start_pfn[nid] = start_pfn;
2970 if (node_boundary_end_pfn[nid] < end_pfn)
2971 node_boundary_end_pfn[nid] = end_pfn;
2974 /* If necessary, push the node boundary out for reserve hotadd */
2975 static void __meminit account_node_boundary(unsigned int nid,
2976 unsigned long *start_pfn, unsigned long *end_pfn)
2978 printk(KERN_DEBUG "Entering account_node_boundary(%u, %lu, %lu)\n",
2979 nid, *start_pfn, *end_pfn);
2981 /* Return if boundary information has not been provided */
2982 if (node_boundary_end_pfn[nid] == 0)
2985 /* Check the boundaries and update if necessary */
2986 if (node_boundary_start_pfn[nid] < *start_pfn)
2987 *start_pfn = node_boundary_start_pfn[nid];
2988 if (node_boundary_end_pfn[nid] > *end_pfn)
2989 *end_pfn = node_boundary_end_pfn[nid];
2992 void __init push_node_boundaries(unsigned int nid,
2993 unsigned long start_pfn, unsigned long end_pfn) {}
2995 static void __meminit account_node_boundary(unsigned int nid,
2996 unsigned long *start_pfn, unsigned long *end_pfn) {}
3001 * get_pfn_range_for_nid - Return the start and end page frames for a node
3002 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
3003 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
3004 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
3006 * It returns the start and end page frame of a node based on information
3007 * provided by an arch calling add_active_range(). If called for a node
3008 * with no available memory, a warning is printed and the start and end
3011 void __meminit get_pfn_range_for_nid(unsigned int nid,
3012 unsigned long *start_pfn, unsigned long *end_pfn)
3018 for_each_active_range_index_in_nid(i, nid) {
3019 *start_pfn = min(*start_pfn, early_node_map[i].start_pfn);
3020 *end_pfn = max(*end_pfn, early_node_map[i].end_pfn);
3023 if (*start_pfn == -1UL)
3026 /* Push the node boundaries out if requested */
3027 account_node_boundary(nid, start_pfn, end_pfn);
3031 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3032 * assumption is made that zones within a node are ordered in monotonic
3033 * increasing memory addresses so that the "highest" populated zone is used
3035 void __init find_usable_zone_for_movable(void)
3038 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
3039 if (zone_index == ZONE_MOVABLE)
3042 if (arch_zone_highest_possible_pfn[zone_index] >
3043 arch_zone_lowest_possible_pfn[zone_index])
3047 VM_BUG_ON(zone_index == -1);
3048 movable_zone = zone_index;
3052 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
3053 * because it is sized independant of architecture. Unlike the other zones,
3054 * the starting point for ZONE_MOVABLE is not fixed. It may be different
3055 * in each node depending on the size of each node and how evenly kernelcore
3056 * is distributed. This helper function adjusts the zone ranges
3057 * provided by the architecture for a given node by using the end of the
3058 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3059 * zones within a node are in order of monotonic increases memory addresses
3061 void __meminit adjust_zone_range_for_zone_movable(int nid,
3062 unsigned long zone_type,
3063 unsigned long node_start_pfn,
3064 unsigned long node_end_pfn,
3065 unsigned long *zone_start_pfn,
3066 unsigned long *zone_end_pfn)
3068 /* Only adjust if ZONE_MOVABLE is on this node */
3069 if (zone_movable_pfn[nid]) {
3070 /* Size ZONE_MOVABLE */
3071 if (zone_type == ZONE_MOVABLE) {
3072 *zone_start_pfn = zone_movable_pfn[nid];
3073 *zone_end_pfn = min(node_end_pfn,
3074 arch_zone_highest_possible_pfn[movable_zone]);
3076 /* Adjust for ZONE_MOVABLE starting within this range */
3077 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
3078 *zone_end_pfn > zone_movable_pfn[nid]) {
3079 *zone_end_pfn = zone_movable_pfn[nid];
3081 /* Check if this whole range is within ZONE_MOVABLE */
3082 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
3083 *zone_start_pfn = *zone_end_pfn;
3088 * Return the number of pages a zone spans in a node, including holes
3089 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3091 static unsigned long __meminit zone_spanned_pages_in_node(int nid,
3092 unsigned long zone_type,
3093 unsigned long *ignored)
3095 unsigned long node_start_pfn, node_end_pfn;
3096 unsigned long zone_start_pfn, zone_end_pfn;
3098 /* Get the start and end of the node and zone */
3099 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3100 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
3101 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
3102 adjust_zone_range_for_zone_movable(nid, zone_type,
3103 node_start_pfn, node_end_pfn,
3104 &zone_start_pfn, &zone_end_pfn);
3106 /* Check that this node has pages within the zone's required range */
3107 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
3110 /* Move the zone boundaries inside the node if necessary */
3111 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
3112 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
3114 /* Return the spanned pages */
3115 return zone_end_pfn - zone_start_pfn;
3119 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
3120 * then all holes in the requested range will be accounted for.
3122 unsigned long __meminit __absent_pages_in_range(int nid,
3123 unsigned long range_start_pfn,
3124 unsigned long range_end_pfn)
3127 unsigned long prev_end_pfn = 0, hole_pages = 0;
3128 unsigned long start_pfn;
3130 /* Find the end_pfn of the first active range of pfns in the node */
3131 i = first_active_region_index_in_nid(nid);
3135 prev_end_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3137 /* Account for ranges before physical memory on this node */
3138 if (early_node_map[i].start_pfn > range_start_pfn)
3139 hole_pages = prev_end_pfn - range_start_pfn;
3141 /* Find all holes for the zone within the node */
3142 for (; i != -1; i = next_active_region_index_in_nid(i, nid)) {
3144 /* No need to continue if prev_end_pfn is outside the zone */
3145 if (prev_end_pfn >= range_end_pfn)
3148 /* Make sure the end of the zone is not within the hole */
3149 start_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3150 prev_end_pfn = max(prev_end_pfn, range_start_pfn);
3152 /* Update the hole size cound and move on */
3153 if (start_pfn > range_start_pfn) {
3154 BUG_ON(prev_end_pfn > start_pfn);
3155 hole_pages += start_pfn - prev_end_pfn;
3157 prev_end_pfn = early_node_map[i].end_pfn;
3160 /* Account for ranges past physical memory on this node */
3161 if (range_end_pfn > prev_end_pfn)
3162 hole_pages += range_end_pfn -
3163 max(range_start_pfn, prev_end_pfn);
3169 * absent_pages_in_range - Return number of page frames in holes within a range
3170 * @start_pfn: The start PFN to start searching for holes
3171 * @end_pfn: The end PFN to stop searching for holes
3173 * It returns the number of pages frames in memory holes within a range.
3175 unsigned long __init absent_pages_in_range(unsigned long start_pfn,
3176 unsigned long end_pfn)
3178 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
3181 /* Return the number of page frames in holes in a zone on a node */
3182 static unsigned long __meminit zone_absent_pages_in_node(int nid,
3183 unsigned long zone_type,
3184 unsigned long *ignored)
3186 unsigned long node_start_pfn, node_end_pfn;
3187 unsigned long zone_start_pfn, zone_end_pfn;
3189 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3190 zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type],
3192 zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type],
3195 adjust_zone_range_for_zone_movable(nid, zone_type,
3196 node_start_pfn, node_end_pfn,
3197 &zone_start_pfn, &zone_end_pfn);
3198 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
3202 static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
3203 unsigned long zone_type,
3204 unsigned long *zones_size)
3206 return zones_size[zone_type];
3209 static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
3210 unsigned long zone_type,
3211 unsigned long *zholes_size)
3216 return zholes_size[zone_type];
3221 static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
3222 unsigned long *zones_size, unsigned long *zholes_size)
3224 unsigned long realtotalpages, totalpages = 0;
3227 for (i = 0; i < MAX_NR_ZONES; i++)
3228 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
3230 pgdat->node_spanned_pages = totalpages;
3232 realtotalpages = totalpages;
3233 for (i = 0; i < MAX_NR_ZONES; i++)
3235 zone_absent_pages_in_node(pgdat->node_id, i,
3237 pgdat->node_present_pages = realtotalpages;
3238 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
3242 #ifndef CONFIG_SPARSEMEM
3244 * Calculate the size of the zone->blockflags rounded to an unsigned long
3245 * Start by making sure zonesize is a multiple of pageblock_order by rounding
3246 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
3247 * round what is now in bits to nearest long in bits, then return it in
3250 static unsigned long __init usemap_size(unsigned long zonesize)
3252 unsigned long usemapsize;
3254 usemapsize = roundup(zonesize, pageblock_nr_pages);
3255 usemapsize = usemapsize >> pageblock_order;
3256 usemapsize *= NR_PAGEBLOCK_BITS;
3257 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
3259 return usemapsize / 8;
3262 static void __init setup_usemap(struct pglist_data *pgdat,
3263 struct zone *zone, unsigned long zonesize)
3265 unsigned long usemapsize = usemap_size(zonesize);
3266 zone->pageblock_flags = NULL;
3268 zone->pageblock_flags = alloc_bootmem_node(pgdat, usemapsize);
3269 memset(zone->pageblock_flags, 0, usemapsize);
3273 static void inline setup_usemap(struct pglist_data *pgdat,
3274 struct zone *zone, unsigned long zonesize) {}
3275 #endif /* CONFIG_SPARSEMEM */
3277 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
3279 /* Return a sensible default order for the pageblock size. */
3280 static inline int pageblock_default_order(void)
3282 if (HPAGE_SHIFT > PAGE_SHIFT)
3283 return HUGETLB_PAGE_ORDER;
3288 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
3289 static inline void __init set_pageblock_order(unsigned int order)
3291 /* Check that pageblock_nr_pages has not already been setup */
3292 if (pageblock_order)
3296 * Assume the largest contiguous order of interest is a huge page.
3297 * This value may be variable depending on boot parameters on IA64
3299 pageblock_order = order;
3301 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3304 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
3305 * and pageblock_default_order() are unused as pageblock_order is set
3306 * at compile-time. See include/linux/pageblock-flags.h for the values of
3307 * pageblock_order based on the kernel config
3309 static inline int pageblock_default_order(unsigned int order)
3313 #define set_pageblock_order(x) do {} while (0)
3315 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3318 * Set up the zone data structures:
3319 * - mark all pages reserved
3320 * - mark all memory queues empty
3321 * - clear the memory bitmaps
3323 static void __paginginit free_area_init_core(struct pglist_data *pgdat,
3324 unsigned long *zones_size, unsigned long *zholes_size)
3327 int nid = pgdat->node_id;
3328 unsigned long zone_start_pfn = pgdat->node_start_pfn;
3331 pgdat_resize_init(pgdat);
3332 pgdat->nr_zones = 0;
3333 init_waitqueue_head(&pgdat->kswapd_wait);
3334 pgdat->kswapd_max_order = 0;
3336 for (j = 0; j < MAX_NR_ZONES; j++) {
3337 struct zone *zone = pgdat->node_zones + j;
3338 unsigned long size, realsize, memmap_pages;
3340 size = zone_spanned_pages_in_node(nid, j, zones_size);
3341 realsize = size - zone_absent_pages_in_node(nid, j,
3345 * Adjust realsize so that it accounts for how much memory
3346 * is used by this zone for memmap. This affects the watermark
3347 * and per-cpu initialisations
3349 memmap_pages = (size * sizeof(struct page)) >> PAGE_SHIFT;
3350 if (realsize >= memmap_pages) {
3351 realsize -= memmap_pages;
3353 " %s zone: %lu pages used for memmap\n",
3354 zone_names[j], memmap_pages);
3357 " %s zone: %lu pages exceeds realsize %lu\n",
3358 zone_names[j], memmap_pages, realsize);
3360 /* Account for reserved pages */
3361 if (j == 0 && realsize > dma_reserve) {
3362 realsize -= dma_reserve;
3363 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
3364 zone_names[0], dma_reserve);
3367 if (!is_highmem_idx(j))
3368 nr_kernel_pages += realsize;
3369 nr_all_pages += realsize;
3371 zone->spanned_pages = size;
3372 zone->present_pages = realsize;
3375 zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
3377 zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100;
3379 zone->name = zone_names[j];
3380 spin_lock_init(&zone->lock);
3381 spin_lock_init(&zone->lru_lock);
3382 zone_seqlock_init(zone);
3383 zone->zone_pgdat = pgdat;
3385 zone->prev_priority = DEF_PRIORITY;
3387 zone_pcp_init(zone);
3388 INIT_LIST_HEAD(&zone->active_list);
3389 INIT_LIST_HEAD(&zone->inactive_list);
3390 zone->nr_scan_active = 0;
3391 zone->nr_scan_inactive = 0;
3392 zap_zone_vm_stats(zone);
3397 set_pageblock_order(pageblock_default_order());
3398 setup_usemap(pgdat, zone, size);
3399 ret = init_currently_empty_zone(zone, zone_start_pfn,
3400 size, MEMMAP_EARLY);
3402 zone_start_pfn += size;
3406 static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
3408 /* Skip empty nodes */
3409 if (!pgdat->node_spanned_pages)
3412 #ifdef CONFIG_FLAT_NODE_MEM_MAP
3413 /* ia64 gets its own node_mem_map, before this, without bootmem */
3414 if (!pgdat->node_mem_map) {
3415 unsigned long size, start, end;
3419 * The zone's endpoints aren't required to be MAX_ORDER
3420 * aligned but the node_mem_map endpoints must be in order
3421 * for the buddy allocator to function correctly.
3423 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
3424 end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
3425 end = ALIGN(end, MAX_ORDER_NR_PAGES);
3426 size = (end - start) * sizeof(struct page);
3427 map = alloc_remap(pgdat->node_id, size);
3429 map = alloc_bootmem_node(pgdat, size);
3430 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
3432 #ifndef CONFIG_NEED_MULTIPLE_NODES
3434 * With no DISCONTIG, the global mem_map is just set as node 0's
3436 if (pgdat == NODE_DATA(0)) {
3437 mem_map = NODE_DATA(0)->node_mem_map;
3438 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3439 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
3440 mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
3441 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3444 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
3447 void __paginginit free_area_init_node(int nid, struct pglist_data *pgdat,
3448 unsigned long *zones_size, unsigned long node_start_pfn,
3449 unsigned long *zholes_size)
3451 pgdat->node_id = nid;
3452 pgdat->node_start_pfn = node_start_pfn;
3453 calculate_node_totalpages(pgdat, zones_size, zholes_size);
3455 alloc_node_mem_map(pgdat);
3457 free_area_init_core(pgdat, zones_size, zholes_size);
3460 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3462 #if MAX_NUMNODES > 1
3464 * Figure out the number of possible node ids.
3466 static void __init setup_nr_node_ids(void)
3469 unsigned int highest = 0;
3471 for_each_node_mask(node, node_possible_map)
3473 nr_node_ids = highest + 1;
3476 static inline void setup_nr_node_ids(void)
3482 * add_active_range - Register a range of PFNs backed by physical memory
3483 * @nid: The node ID the range resides on
3484 * @start_pfn: The start PFN of the available physical memory
3485 * @end_pfn: The end PFN of the available physical memory
3487 * These ranges are stored in an early_node_map[] and later used by
3488 * free_area_init_nodes() to calculate zone sizes and holes. If the
3489 * range spans a memory hole, it is up to the architecture to ensure
3490 * the memory is not freed by the bootmem allocator. If possible
3491 * the range being registered will be merged with existing ranges.
3493 void __init add_active_range(unsigned int nid, unsigned long start_pfn,
3494 unsigned long end_pfn)
3498 printk(KERN_DEBUG "Entering add_active_range(%d, %lu, %lu) "
3499 "%d entries of %d used\n",
3500 nid, start_pfn, end_pfn,
3501 nr_nodemap_entries, MAX_ACTIVE_REGIONS);
3503 /* Merge with existing active regions if possible */
3504 for (i = 0; i < nr_nodemap_entries; i++) {
3505 if (early_node_map[i].nid != nid)
3508 /* Skip if an existing region covers this new one */
3509 if (start_pfn >= early_node_map[i].start_pfn &&
3510 end_pfn <= early_node_map[i].end_pfn)
3513 /* Merge forward if suitable */
3514 if (start_pfn <= early_node_map[i].end_pfn &&
3515 end_pfn > early_node_map[i].end_pfn) {
3516 early_node_map[i].end_pfn = end_pfn;
3520 /* Merge backward if suitable */
3521 if (start_pfn < early_node_map[i].end_pfn &&
3522 end_pfn >= early_node_map[i].start_pfn) {
3523 early_node_map[i].start_pfn = start_pfn;
3528 /* Check that early_node_map is large enough */
3529 if (i >= MAX_ACTIVE_REGIONS) {
3530 printk(KERN_CRIT "More than %d memory regions, truncating\n",
3531 MAX_ACTIVE_REGIONS);
3535 early_node_map[i].nid = nid;
3536 early_node_map[i].start_pfn = start_pfn;
3537 early_node_map[i].end_pfn = end_pfn;
3538 nr_nodemap_entries = i + 1;
3542 * shrink_active_range - Shrink an existing registered range of PFNs
3543 * @nid: The node id the range is on that should be shrunk
3544 * @old_end_pfn: The old end PFN of the range
3545 * @new_end_pfn: The new PFN of the range
3547 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
3548 * The map is kept at the end physical page range that has already been
3549 * registered with add_active_range(). This function allows an arch to shrink
3550 * an existing registered range.
3552 void __init shrink_active_range(unsigned int nid, unsigned long old_end_pfn,
3553 unsigned long new_end_pfn)
3557 /* Find the old active region end and shrink */
3558 for_each_active_range_index_in_nid(i, nid)
3559 if (early_node_map[i].end_pfn == old_end_pfn) {
3560 early_node_map[i].end_pfn = new_end_pfn;
3566 * remove_all_active_ranges - Remove all currently registered regions
3568 * During discovery, it may be found that a table like SRAT is invalid
3569 * and an alternative discovery method must be used. This function removes
3570 * all currently registered regions.
3572 void __init remove_all_active_ranges(void)
3574 memset(early_node_map, 0, sizeof(early_node_map));
3575 nr_nodemap_entries = 0;
3576 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
3577 memset(node_boundary_start_pfn, 0, sizeof(node_boundary_start_pfn));
3578 memset(node_boundary_end_pfn, 0, sizeof(node_boundary_end_pfn));
3579 #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
3582 /* Compare two active node_active_regions */
3583 static int __init cmp_node_active_region(const void *a, const void *b)
3585 struct node_active_region *arange = (struct node_active_region *)a;
3586 struct node_active_region *brange = (struct node_active_region *)b;
3588 /* Done this way to avoid overflows */
3589 if (arange->start_pfn > brange->start_pfn)
3591 if (arange->start_pfn < brange->start_pfn)
3597 /* sort the node_map by start_pfn */
3598 static void __init sort_node_map(void)
3600 sort(early_node_map, (size_t)nr_nodemap_entries,
3601 sizeof(struct node_active_region),
3602 cmp_node_active_region, NULL);
3605 /* Find the lowest pfn for a node */
3606 unsigned long __init find_min_pfn_for_node(unsigned long nid)
3609 unsigned long min_pfn = ULONG_MAX;
3611 /* Assuming a sorted map, the first range found has the starting pfn */
3612 for_each_active_range_index_in_nid(i, nid)
3613 min_pfn = min(min_pfn, early_node_map[i].start_pfn);
3615 if (min_pfn == ULONG_MAX) {
3617 "Could not find start_pfn for node %lu\n", nid);
3625 * find_min_pfn_with_active_regions - Find the minimum PFN registered
3627 * It returns the minimum PFN based on information provided via
3628 * add_active_range().
3630 unsigned long __init find_min_pfn_with_active_regions(void)
3632 return find_min_pfn_for_node(MAX_NUMNODES);
3636 * find_max_pfn_with_active_regions - Find the maximum PFN registered
3638 * It returns the maximum PFN based on information provided via
3639 * add_active_range().
3641 unsigned long __init find_max_pfn_with_active_regions(void)
3644 unsigned long max_pfn = 0;
3646 for (i = 0; i < nr_nodemap_entries; i++)
3647 max_pfn = max(max_pfn, early_node_map[i].end_pfn);
3653 * early_calculate_totalpages()
3654 * Sum pages in active regions for movable zone.
3655 * Populate N_HIGH_MEMORY for calculating usable_nodes.
3657 static unsigned long __init early_calculate_totalpages(void)
3660 unsigned long totalpages = 0;
3662 for (i = 0; i < nr_nodemap_entries; i++) {
3663 unsigned long pages = early_node_map[i].end_pfn -
3664 early_node_map[i].start_pfn;
3665 totalpages += pages;
3667 node_set_state(early_node_map[i].nid, N_HIGH_MEMORY);
3673 * Find the PFN the Movable zone begins in each node. Kernel memory
3674 * is spread evenly between nodes as long as the nodes have enough
3675 * memory. When they don't, some nodes will have more kernelcore than
3678 void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn)
3681 unsigned long usable_startpfn;
3682 unsigned long kernelcore_node, kernelcore_remaining;
3683 unsigned long totalpages = early_calculate_totalpages();
3684 int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]);
3687 * If movablecore was specified, calculate what size of
3688 * kernelcore that corresponds so that memory usable for
3689 * any allocation type is evenly spread. If both kernelcore
3690 * and movablecore are specified, then the value of kernelcore
3691 * will be used for required_kernelcore if it's greater than
3692 * what movablecore would have allowed.
3694 if (required_movablecore) {
3695 unsigned long corepages;
3698 * Round-up so that ZONE_MOVABLE is at least as large as what
3699 * was requested by the user
3701 required_movablecore =
3702 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
3703 corepages = totalpages - required_movablecore;
3705 required_kernelcore = max(required_kernelcore, corepages);
3708 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
3709 if (!required_kernelcore)
3712 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
3713 find_usable_zone_for_movable();
3714 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
3717 /* Spread kernelcore memory as evenly as possible throughout nodes */
3718 kernelcore_node = required_kernelcore / usable_nodes;
3719 for_each_node_state(nid, N_HIGH_MEMORY) {
3721 * Recalculate kernelcore_node if the division per node
3722 * now exceeds what is necessary to satisfy the requested
3723 * amount of memory for the kernel
3725 if (required_kernelcore < kernelcore_node)
3726 kernelcore_node = required_kernelcore / usable_nodes;
3729 * As the map is walked, we track how much memory is usable
3730 * by the kernel using kernelcore_remaining. When it is
3731 * 0, the rest of the node is usable by ZONE_MOVABLE
3733 kernelcore_remaining = kernelcore_node;
3735 /* Go through each range of PFNs within this node */
3736 for_each_active_range_index_in_nid(i, nid) {
3737 unsigned long start_pfn, end_pfn;
3738 unsigned long size_pages;
3740 start_pfn = max(early_node_map[i].start_pfn,
3741 zone_movable_pfn[nid]);
3742 end_pfn = early_node_map[i].end_pfn;
3743 if (start_pfn >= end_pfn)
3746 /* Account for what is only usable for kernelcore */
3747 if (start_pfn < usable_startpfn) {
3748 unsigned long kernel_pages;
3749 kernel_pages = min(end_pfn, usable_startpfn)
3752 kernelcore_remaining -= min(kernel_pages,
3753 kernelcore_remaining);
3754 required_kernelcore -= min(kernel_pages,
3755 required_kernelcore);
3757 /* Continue if range is now fully accounted */
3758 if (end_pfn <= usable_startpfn) {
3761 * Push zone_movable_pfn to the end so
3762 * that if we have to rebalance
3763 * kernelcore across nodes, we will
3764 * not double account here
3766 zone_movable_pfn[nid] = end_pfn;
3769 start_pfn = usable_startpfn;
3773 * The usable PFN range for ZONE_MOVABLE is from
3774 * start_pfn->end_pfn. Calculate size_pages as the
3775 * number of pages used as kernelcore
3777 size_pages = end_pfn - start_pfn;
3778 if (size_pages > kernelcore_remaining)
3779 size_pages = kernelcore_remaining;
3780 zone_movable_pfn[nid] = start_pfn + size_pages;
3783 * Some kernelcore has been met, update counts and
3784 * break if the kernelcore for this node has been
3787 required_kernelcore -= min(required_kernelcore,
3789 kernelcore_remaining -= size_pages;
3790 if (!kernelcore_remaining)
3796 * If there is still required_kernelcore, we do another pass with one
3797 * less node in the count. This will push zone_movable_pfn[nid] further
3798 * along on the nodes that still have memory until kernelcore is
3802 if (usable_nodes && required_kernelcore > usable_nodes)
3805 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
3806 for (nid = 0; nid < MAX_NUMNODES; nid++)
3807 zone_movable_pfn[nid] =
3808 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
3811 /* Any regular memory on that node ? */
3812 static void check_for_regular_memory(pg_data_t *pgdat)
3814 #ifdef CONFIG_HIGHMEM
3815 enum zone_type zone_type;
3817 for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) {
3818 struct zone *zone = &pgdat->node_zones[zone_type];
3819 if (zone->present_pages)
3820 node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY);
3826 * free_area_init_nodes - Initialise all pg_data_t and zone data
3827 * @max_zone_pfn: an array of max PFNs for each zone
3829 * This will call free_area_init_node() for each active node in the system.
3830 * Using the page ranges provided by add_active_range(), the size of each
3831 * zone in each node and their holes is calculated. If the maximum PFN
3832 * between two adjacent zones match, it is assumed that the zone is empty.
3833 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
3834 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
3835 * starts where the previous one ended. For example, ZONE_DMA32 starts
3836 * at arch_max_dma_pfn.
3838 void __init free_area_init_nodes(unsigned long *max_zone_pfn)
3843 /* Sort early_node_map as initialisation assumes it is sorted */
3846 /* Record where the zone boundaries are */
3847 memset(arch_zone_lowest_possible_pfn, 0,
3848 sizeof(arch_zone_lowest_possible_pfn));
3849 memset(arch_zone_highest_possible_pfn, 0,
3850 sizeof(arch_zone_highest_possible_pfn));
3851 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
3852 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
3853 for (i = 1; i < MAX_NR_ZONES; i++) {
3854 if (i == ZONE_MOVABLE)
3856 arch_zone_lowest_possible_pfn[i] =
3857 arch_zone_highest_possible_pfn[i-1];
3858 arch_zone_highest_possible_pfn[i] =
3859 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
3861 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
3862 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
3864 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
3865 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
3866 find_zone_movable_pfns_for_nodes(zone_movable_pfn);
3868 /* Print out the zone ranges */
3869 printk("Zone PFN ranges:\n");
3870 for (i = 0; i < MAX_NR_ZONES; i++) {
3871 if (i == ZONE_MOVABLE)
3873 printk(" %-8s %8lu -> %8lu\n",
3875 arch_zone_lowest_possible_pfn[i],
3876 arch_zone_highest_possible_pfn[i]);
3879 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
3880 printk("Movable zone start PFN for each node\n");
3881 for (i = 0; i < MAX_NUMNODES; i++) {
3882 if (zone_movable_pfn[i])
3883 printk(" Node %d: %lu\n", i, zone_movable_pfn[i]);
3886 /* Print out the early_node_map[] */
3887 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries);
3888 for (i = 0; i < nr_nodemap_entries; i++)
3889 printk(" %3d: %8lu -> %8lu\n", early_node_map[i].nid,
3890 early_node_map[i].start_pfn,
3891 early_node_map[i].end_pfn);
3893 /* Initialise every node */
3894 setup_nr_node_ids();
3895 for_each_online_node(nid) {
3896 pg_data_t *pgdat = NODE_DATA(nid);
3897 free_area_init_node(nid, pgdat, NULL,
3898 find_min_pfn_for_node(nid), NULL);
3900 /* Any memory on that node */
3901 if (pgdat->node_present_pages)
3902 node_set_state(nid, N_HIGH_MEMORY);
3903 check_for_regular_memory(pgdat);
3907 static int __init cmdline_parse_core(char *p, unsigned long *core)
3909 unsigned long long coremem;
3913 coremem = memparse(p, &p);
3914 *core = coremem >> PAGE_SHIFT;
3916 /* Paranoid check that UL is enough for the coremem value */
3917 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
3923 * kernelcore=size sets the amount of memory for use for allocations that
3924 * cannot be reclaimed or migrated.
3926 static int __init cmdline_parse_kernelcore(char *p)
3928 return cmdline_parse_core(p, &required_kernelcore);
3932 * movablecore=size sets the amount of memory for use for allocations that
3933 * can be reclaimed or migrated.
3935 static int __init cmdline_parse_movablecore(char *p)
3937 return cmdline_parse_core(p, &required_movablecore);
3940 early_param("kernelcore", cmdline_parse_kernelcore);
3941 early_param("movablecore", cmdline_parse_movablecore);
3943 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3946 * set_dma_reserve - set the specified number of pages reserved in the first zone
3947 * @new_dma_reserve: The number of pages to mark reserved
3949 * The per-cpu batchsize and zone watermarks are determined by present_pages.
3950 * In the DMA zone, a significant percentage may be consumed by kernel image
3951 * and other unfreeable allocations which can skew the watermarks badly. This
3952 * function may optionally be used to account for unfreeable pages in the
3953 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
3954 * smaller per-cpu batchsize.
3956 void __init set_dma_reserve(unsigned long new_dma_reserve)
3958 dma_reserve = new_dma_reserve;
3961 #ifndef CONFIG_NEED_MULTIPLE_NODES
3962 static bootmem_data_t contig_bootmem_data;
3963 struct pglist_data contig_page_data = { .bdata = &contig_bootmem_data };
3965 EXPORT_SYMBOL(contig_page_data);
3968 void __init free_area_init(unsigned long *zones_size)
3970 free_area_init_node(0, NODE_DATA(0), zones_size,
3971 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
3974 static int page_alloc_cpu_notify(struct notifier_block *self,
3975 unsigned long action, void *hcpu)
3977 int cpu = (unsigned long)hcpu;
3979 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
3983 * Spill the event counters of the dead processor
3984 * into the current processors event counters.
3985 * This artificially elevates the count of the current
3988 vm_events_fold_cpu(cpu);
3991 * Zero the differential counters of the dead processor
3992 * so that the vm statistics are consistent.
3994 * This is only okay since the processor is dead and cannot
3995 * race with what we are doing.
3997 refresh_cpu_vm_stats(cpu);
4002 void __init page_alloc_init(void)
4004 hotcpu_notifier(page_alloc_cpu_notify, 0);
4008 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
4009 * or min_free_kbytes changes.
4011 static void calculate_totalreserve_pages(void)
4013 struct pglist_data *pgdat;
4014 unsigned long reserve_pages = 0;
4015 enum zone_type i, j;
4017 for_each_online_pgdat(pgdat) {
4018 for (i = 0; i < MAX_NR_ZONES; i++) {
4019 struct zone *zone = pgdat->node_zones + i;
4020 unsigned long max = 0;
4022 /* Find valid and maximum lowmem_reserve in the zone */
4023 for (j = i; j < MAX_NR_ZONES; j++) {
4024 if (zone->lowmem_reserve[j] > max)
4025 max = zone->lowmem_reserve[j];
4028 /* we treat pages_high as reserved pages. */
4029 max += zone->pages_high;
4031 if (max > zone->present_pages)
4032 max = zone->present_pages;
4033 reserve_pages += max;
4036 totalreserve_pages = reserve_pages;
4040 * setup_per_zone_lowmem_reserve - called whenever
4041 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
4042 * has a correct pages reserved value, so an adequate number of
4043 * pages are left in the zone after a successful __alloc_pages().
4045 static void setup_per_zone_lowmem_reserve(void)
4047 struct pglist_data *pgdat;
4048 enum zone_type j, idx;
4050 for_each_online_pgdat(pgdat) {
4051 for (j = 0; j < MAX_NR_ZONES; j++) {
4052 struct zone *zone = pgdat->node_zones + j;
4053 unsigned long present_pages = zone->present_pages;
4055 zone->lowmem_reserve[j] = 0;
4059 struct zone *lower_zone;
4063 if (sysctl_lowmem_reserve_ratio[idx] < 1)
4064 sysctl_lowmem_reserve_ratio[idx] = 1;
4066 lower_zone = pgdat->node_zones + idx;
4067 lower_zone->lowmem_reserve[j] = present_pages /
4068 sysctl_lowmem_reserve_ratio[idx];
4069 present_pages += lower_zone->present_pages;
4074 /* update totalreserve_pages */
4075 calculate_totalreserve_pages();
4079 * setup_per_zone_pages_min - called when min_free_kbytes changes.
4081 * Ensures that the pages_{min,low,high} values for each zone are set correctly
4082 * with respect to min_free_kbytes.
4084 void setup_per_zone_pages_min(void)
4086 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
4087 unsigned long lowmem_pages = 0;
4089 unsigned long flags;
4091 /* Calculate total number of !ZONE_HIGHMEM pages */
4092 for_each_zone(zone) {
4093 if (!is_highmem(zone))
4094 lowmem_pages += zone->present_pages;
4097 for_each_zone(zone) {
4100 spin_lock_irqsave(&zone->lru_lock, flags);
4101 tmp = (u64)pages_min * zone->present_pages;
4102 do_div(tmp, lowmem_pages);
4103 if (is_highmem(zone)) {
4105 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
4106 * need highmem pages, so cap pages_min to a small
4109 * The (pages_high-pages_low) and (pages_low-pages_min)
4110 * deltas controls asynch page reclaim, and so should
4111 * not be capped for highmem.
4115 min_pages = zone->present_pages / 1024;
4116 if (min_pages < SWAP_CLUSTER_MAX)
4117 min_pages = SWAP_CLUSTER_MAX;
4118 if (min_pages > 128)
4120 zone->pages_min = min_pages;
4123 * If it's a lowmem zone, reserve a number of pages
4124 * proportionate to the zone's size.
4126 zone->pages_min = tmp;
4129 zone->pages_low = zone->pages_min + (tmp >> 2);
4130 zone->pages_high = zone->pages_min + (tmp >> 1);
4131 setup_zone_migrate_reserve(zone);
4132 spin_unlock_irqrestore(&zone->lru_lock, flags);
4135 /* update totalreserve_pages */
4136 calculate_totalreserve_pages();
4140 * Initialise min_free_kbytes.
4142 * For small machines we want it small (128k min). For large machines
4143 * we want it large (64MB max). But it is not linear, because network
4144 * bandwidth does not increase linearly with machine size. We use
4146 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
4147 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
4163 static int __init init_per_zone_pages_min(void)
4165 unsigned long lowmem_kbytes;
4167 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
4169 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
4170 if (min_free_kbytes < 128)
4171 min_free_kbytes = 128;
4172 if (min_free_kbytes > 65536)
4173 min_free_kbytes = 65536;
4174 setup_per_zone_pages_min();
4175 setup_per_zone_lowmem_reserve();
4178 module_init(init_per_zone_pages_min)
4181 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
4182 * that we can call two helper functions whenever min_free_kbytes
4185 int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
4186 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4188 proc_dointvec(table, write, file, buffer, length, ppos);
4190 setup_per_zone_pages_min();
4195 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
4196 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4201 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4206 zone->min_unmapped_pages = (zone->present_pages *
4207 sysctl_min_unmapped_ratio) / 100;
4211 int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
4212 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4217 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4222 zone->min_slab_pages = (zone->present_pages *
4223 sysctl_min_slab_ratio) / 100;
4229 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
4230 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
4231 * whenever sysctl_lowmem_reserve_ratio changes.
4233 * The reserve ratio obviously has absolutely no relation with the
4234 * pages_min watermarks. The lowmem reserve ratio can only make sense
4235 * if in function of the boot time zone sizes.
4237 int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
4238 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4240 proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4241 setup_per_zone_lowmem_reserve();
4246 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
4247 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
4248 * can have before it gets flushed back to buddy allocator.
4251 int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
4252 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4258 ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4259 if (!write || (ret == -EINVAL))
4261 for_each_zone(zone) {
4262 for_each_online_cpu(cpu) {
4264 high = zone->present_pages / percpu_pagelist_fraction;
4265 setup_pagelist_highmark(zone_pcp(zone, cpu), high);
4271 int hashdist = HASHDIST_DEFAULT;
4274 static int __init set_hashdist(char *str)
4278 hashdist = simple_strtoul(str, &str, 0);
4281 __setup("hashdist=", set_hashdist);
4285 * allocate a large system hash table from bootmem
4286 * - it is assumed that the hash table must contain an exact power-of-2
4287 * quantity of entries
4288 * - limit is the number of hash buckets, not the total allocation size
4290 void *__init alloc_large_system_hash(const char *tablename,
4291 unsigned long bucketsize,
4292 unsigned long numentries,
4295 unsigned int *_hash_shift,
4296 unsigned int *_hash_mask,
4297 unsigned long limit)
4299 unsigned long long max = limit;
4300 unsigned long log2qty, size;
4303 /* allow the kernel cmdline to have a say */
4305 /* round applicable memory size up to nearest megabyte */
4306 numentries = nr_kernel_pages;
4307 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
4308 numentries >>= 20 - PAGE_SHIFT;
4309 numentries <<= 20 - PAGE_SHIFT;
4311 /* limit to 1 bucket per 2^scale bytes of low memory */
4312 if (scale > PAGE_SHIFT)
4313 numentries >>= (scale - PAGE_SHIFT);
4315 numentries <<= (PAGE_SHIFT - scale);
4317 /* Make sure we've got at least a 0-order allocation.. */
4318 if (unlikely((numentries * bucketsize) < PAGE_SIZE))
4319 numentries = PAGE_SIZE / bucketsize;
4321 numentries = roundup_pow_of_two(numentries);
4323 /* limit allocation size to 1/16 total memory by default */
4325 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
4326 do_div(max, bucketsize);
4329 if (numentries > max)
4332 log2qty = ilog2(numentries);
4335 size = bucketsize << log2qty;
4336 if (flags & HASH_EARLY)
4337 table = alloc_bootmem(size);
4339 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
4341 unsigned long order;
4342 for (order = 0; ((1UL << order) << PAGE_SHIFT) < size; order++)
4344 table = (void*) __get_free_pages(GFP_ATOMIC, order);
4346 * If bucketsize is not a power-of-two, we may free
4347 * some pages at the end of hash table.
4350 unsigned long alloc_end = (unsigned long)table +
4351 (PAGE_SIZE << order);
4352 unsigned long used = (unsigned long)table +
4354 split_page(virt_to_page(table), order);
4355 while (used < alloc_end) {
4361 } while (!table && size > PAGE_SIZE && --log2qty);
4364 panic("Failed to allocate %s hash table\n", tablename);
4366 printk(KERN_INFO "%s hash table entries: %d (order: %d, %lu bytes)\n",
4369 ilog2(size) - PAGE_SHIFT,
4373 *_hash_shift = log2qty;
4375 *_hash_mask = (1 << log2qty) - 1;
4380 #ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE
4381 struct page *pfn_to_page(unsigned long pfn)
4383 return __pfn_to_page(pfn);
4385 unsigned long page_to_pfn(struct page *page)
4387 return __page_to_pfn(page);
4389 EXPORT_SYMBOL(pfn_to_page);
4390 EXPORT_SYMBOL(page_to_pfn);
4391 #endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */
4393 /* Return a pointer to the bitmap storing bits affecting a block of pages */
4394 static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
4397 #ifdef CONFIG_SPARSEMEM
4398 return __pfn_to_section(pfn)->pageblock_flags;
4400 return zone->pageblock_flags;
4401 #endif /* CONFIG_SPARSEMEM */
4404 static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
4406 #ifdef CONFIG_SPARSEMEM
4407 pfn &= (PAGES_PER_SECTION-1);
4408 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
4410 pfn = pfn - zone->zone_start_pfn;
4411 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
4412 #endif /* CONFIG_SPARSEMEM */
4416 * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
4417 * @page: The page within the block of interest
4418 * @start_bitidx: The first bit of interest to retrieve
4419 * @end_bitidx: The last bit of interest
4420 * returns pageblock_bits flags
4422 unsigned long get_pageblock_flags_group(struct page *page,
4423 int start_bitidx, int end_bitidx)
4426 unsigned long *bitmap;
4427 unsigned long pfn, bitidx;
4428 unsigned long flags = 0;
4429 unsigned long value = 1;
4431 zone = page_zone(page);
4432 pfn = page_to_pfn(page);
4433 bitmap = get_pageblock_bitmap(zone, pfn);
4434 bitidx = pfn_to_bitidx(zone, pfn);
4436 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
4437 if (test_bit(bitidx + start_bitidx, bitmap))
4444 * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
4445 * @page: The page within the block of interest
4446 * @start_bitidx: The first bit of interest
4447 * @end_bitidx: The last bit of interest
4448 * @flags: The flags to set
4450 void set_pageblock_flags_group(struct page *page, unsigned long flags,
4451 int start_bitidx, int end_bitidx)
4454 unsigned long *bitmap;
4455 unsigned long pfn, bitidx;
4456 unsigned long value = 1;
4458 zone = page_zone(page);
4459 pfn = page_to_pfn(page);
4460 bitmap = get_pageblock_bitmap(zone, pfn);
4461 bitidx = pfn_to_bitidx(zone, pfn);
4463 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
4465 __set_bit(bitidx + start_bitidx, bitmap);
4467 __clear_bit(bitidx + start_bitidx, bitmap);
4471 * This is designed as sub function...plz see page_isolation.c also.
4472 * set/clear page block's type to be ISOLATE.
4473 * page allocater never alloc memory from ISOLATE block.
4476 int set_migratetype_isolate(struct page *page)
4479 unsigned long flags;
4482 zone = page_zone(page);
4483 spin_lock_irqsave(&zone->lock, flags);
4485 * In future, more migrate types will be able to be isolation target.
4487 if (get_pageblock_migratetype(page) != MIGRATE_MOVABLE)
4489 set_pageblock_migratetype(page, MIGRATE_ISOLATE);
4490 move_freepages_block(zone, page, MIGRATE_ISOLATE);
4493 spin_unlock_irqrestore(&zone->lock, flags);
4499 void unset_migratetype_isolate(struct page *page)
4502 unsigned long flags;
4503 zone = page_zone(page);
4504 spin_lock_irqsave(&zone->lock, flags);
4505 if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE)
4507 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
4508 move_freepages_block(zone, page, MIGRATE_MOVABLE);
4510 spin_unlock_irqrestore(&zone->lock, flags);
4513 #ifdef CONFIG_MEMORY_HOTREMOVE
4515 * All pages in the range must be isolated before calling this.
4518 __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
4524 unsigned long flags;
4525 /* find the first valid pfn */
4526 for (pfn = start_pfn; pfn < end_pfn; pfn++)
4531 zone = page_zone(pfn_to_page(pfn));
4532 spin_lock_irqsave(&zone->lock, flags);
4534 while (pfn < end_pfn) {
4535 if (!pfn_valid(pfn)) {
4539 page = pfn_to_page(pfn);
4540 BUG_ON(page_count(page));
4541 BUG_ON(!PageBuddy(page));
4542 order = page_order(page);
4543 #ifdef CONFIG_DEBUG_VM
4544 printk(KERN_INFO "remove from free list %lx %d %lx\n",
4545 pfn, 1 << order, end_pfn);
4547 list_del(&page->lru);
4548 rmv_page_order(page);
4549 zone->free_area[order].nr_free--;
4550 __mod_zone_page_state(zone, NR_FREE_PAGES,
4552 for (i = 0; i < (1 << order); i++)
4553 SetPageReserved((page+i));
4554 pfn += (1 << order);
4556 spin_unlock_irqrestore(&zone->lock, flags);