2 * linux/mm/page_alloc.c
4 * Manages the free list, the system allocates free pages here.
5 * Note that kmalloc() lives in slab.c
7 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
8 * Swap reorganised 29.12.95, Stephen Tweedie
9 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
10 * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999
11 * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
12 * Zone balancing, Kanoj Sarcar, SGI, Jan 2000
13 * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002
14 * (lots of bits borrowed from Ingo Molnar & Andrew Morton)
17 #include <linux/stddef.h>
19 #include <linux/swap.h>
20 #include <linux/interrupt.h>
21 #include <linux/pagemap.h>
22 #include <linux/bootmem.h>
23 #include <linux/compiler.h>
24 #include <linux/kernel.h>
25 #include <linux/module.h>
26 #include <linux/suspend.h>
27 #include <linux/pagevec.h>
28 #include <linux/blkdev.h>
29 #include <linux/slab.h>
30 #include <linux/notifier.h>
31 #include <linux/topology.h>
32 #include <linux/sysctl.h>
33 #include <linux/cpu.h>
34 #include <linux/cpuset.h>
35 #include <linux/memory_hotplug.h>
36 #include <linux/nodemask.h>
37 #include <linux/vmalloc.h>
38 #include <linux/mempolicy.h>
39 #include <linux/stop_machine.h>
40 #include <linux/sort.h>
41 #include <linux/pfn.h>
42 #include <linux/backing-dev.h>
43 #include <linux/fault-inject.h>
45 #include <asm/tlbflush.h>
46 #include <asm/div64.h>
50 * Array of node states.
52 nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
53 [N_POSSIBLE] = NODE_MASK_ALL,
54 [N_ONLINE] = { { [0] = 1UL } },
56 [N_NORMAL_MEMORY] = { { [0] = 1UL } },
58 [N_HIGH_MEMORY] = { { [0] = 1UL } },
60 [N_CPU] = { { [0] = 1UL } },
63 EXPORT_SYMBOL(node_states);
65 unsigned long totalram_pages __read_mostly;
66 unsigned long totalreserve_pages __read_mostly;
68 int percpu_pagelist_fraction;
70 static void __free_pages_ok(struct page *page, unsigned int order);
73 * results with 256, 32 in the lowmem_reserve sysctl:
74 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
75 * 1G machine -> (16M dma, 784M normal, 224M high)
76 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
77 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
78 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
80 * TBD: should special case ZONE_DMA32 machines here - in those we normally
81 * don't need any ZONE_NORMAL reservation
83 int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
84 #ifdef CONFIG_ZONE_DMA
87 #ifdef CONFIG_ZONE_DMA32
96 EXPORT_SYMBOL(totalram_pages);
98 static char * const zone_names[MAX_NR_ZONES] = {
99 #ifdef CONFIG_ZONE_DMA
102 #ifdef CONFIG_ZONE_DMA32
106 #ifdef CONFIG_HIGHMEM
112 int min_free_kbytes = 1024;
114 unsigned long __meminitdata nr_kernel_pages;
115 unsigned long __meminitdata nr_all_pages;
116 static unsigned long __meminitdata dma_reserve;
118 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
120 * MAX_ACTIVE_REGIONS determines the maxmimum number of distinct
121 * ranges of memory (RAM) that may be registered with add_active_range().
122 * Ranges passed to add_active_range() will be merged if possible
123 * so the number of times add_active_range() can be called is
124 * related to the number of nodes and the number of holes
126 #ifdef CONFIG_MAX_ACTIVE_REGIONS
127 /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
128 #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
130 #if MAX_NUMNODES >= 32
131 /* If there can be many nodes, allow up to 50 holes per node */
132 #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
134 /* By default, allow up to 256 distinct regions */
135 #define MAX_ACTIVE_REGIONS 256
139 static struct node_active_region __meminitdata early_node_map[MAX_ACTIVE_REGIONS];
140 static int __meminitdata nr_nodemap_entries;
141 static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
142 static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
143 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
144 static unsigned long __meminitdata node_boundary_start_pfn[MAX_NUMNODES];
145 static unsigned long __meminitdata node_boundary_end_pfn[MAX_NUMNODES];
146 #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
147 unsigned long __initdata required_kernelcore;
148 unsigned long __initdata required_movablecore;
149 unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
151 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
153 EXPORT_SYMBOL(movable_zone);
154 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
157 int nr_node_ids __read_mostly = MAX_NUMNODES;
158 EXPORT_SYMBOL(nr_node_ids);
161 static inline int get_pageblock_migratetype(struct page *page)
163 return get_pageblock_flags_group(page, PB_migrate, PB_migrate_end);
166 static void set_pageblock_migratetype(struct page *page, int migratetype)
168 set_pageblock_flags_group(page, (unsigned long)migratetype,
169 PB_migrate, PB_migrate_end);
172 static inline int gfpflags_to_migratetype(gfp_t gfp_flags)
174 return ((gfp_flags & __GFP_MOVABLE) != 0);
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 printk(KERN_EMERG "Bad page state in process '%s'\n"
226 KERN_EMERG "page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n"
227 KERN_EMERG "Trying to fix it up, but a reboot is needed\n"
228 KERN_EMERG "Backtrace:\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 page->flags &= ~(1 << PG_lru |
243 set_page_count(page, 0);
244 reset_page_mapcount(page);
245 page->mapping = NULL;
246 add_taint(TAINT_BAD_PAGE);
250 * Higher-order pages are called "compound pages". They are structured thusly:
252 * The first PAGE_SIZE page is called the "head page".
254 * The remaining PAGE_SIZE pages are called "tail pages".
256 * All pages have PG_compound set. All pages have their ->private pointing at
257 * the head page (even the head page has this).
259 * The first tail page's ->lru.next holds the address of the compound page's
260 * put_page() function. Its ->lru.prev holds the order of allocation.
261 * This usage means that zero-order pages may not be compound.
264 static void free_compound_page(struct page *page)
266 __free_pages_ok(page, compound_order(page));
269 static void prep_compound_page(struct page *page, unsigned long order)
272 int nr_pages = 1 << order;
274 set_compound_page_dtor(page, free_compound_page);
275 set_compound_order(page, order);
277 for (i = 1; i < nr_pages; i++) {
278 struct page *p = page + i;
281 p->first_page = page;
285 static void destroy_compound_page(struct page *page, unsigned long order)
288 int nr_pages = 1 << order;
290 if (unlikely(compound_order(page) != order))
293 if (unlikely(!PageHead(page)))
295 __ClearPageHead(page);
296 for (i = 1; i < nr_pages; i++) {
297 struct page *p = page + i;
299 if (unlikely(!PageTail(p) |
300 (p->first_page != page)))
306 static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
310 VM_BUG_ON((gfp_flags & (__GFP_WAIT | __GFP_HIGHMEM)) == __GFP_HIGHMEM);
312 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
313 * and __GFP_HIGHMEM from hard or soft interrupt context.
315 VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
316 for (i = 0; i < (1 << order); i++)
317 clear_highpage(page + i);
321 * function for dealing with page's order in buddy system.
322 * zone->lock is already acquired when we use these.
323 * So, we don't need atomic page->flags operations here.
325 static inline unsigned long page_order(struct page *page)
327 return page_private(page);
330 static inline void set_page_order(struct page *page, int order)
332 set_page_private(page, order);
333 __SetPageBuddy(page);
336 static inline void rmv_page_order(struct page *page)
338 __ClearPageBuddy(page);
339 set_page_private(page, 0);
343 * Locate the struct page for both the matching buddy in our
344 * pair (buddy1) and the combined O(n+1) page they form (page).
346 * 1) Any buddy B1 will have an order O twin B2 which satisfies
347 * the following equation:
349 * For example, if the starting buddy (buddy2) is #8 its order
351 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
353 * 2) Any buddy B will have an order O+1 parent P which
354 * satisfies the following equation:
357 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
359 static inline struct page *
360 __page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order)
362 unsigned long buddy_idx = page_idx ^ (1 << order);
364 return page + (buddy_idx - page_idx);
367 static inline unsigned long
368 __find_combined_index(unsigned long page_idx, unsigned int order)
370 return (page_idx & ~(1 << order));
374 * This function checks whether a page is free && is the buddy
375 * we can do coalesce a page and its buddy if
376 * (a) the buddy is not in a hole &&
377 * (b) the buddy is in the buddy system &&
378 * (c) a page and its buddy have the same order &&
379 * (d) a page and its buddy are in the same zone.
381 * For recording whether a page is in the buddy system, we use PG_buddy.
382 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
384 * For recording page's order, we use page_private(page).
386 static inline int page_is_buddy(struct page *page, struct page *buddy,
389 if (!pfn_valid_within(page_to_pfn(buddy)))
392 if (page_zone_id(page) != page_zone_id(buddy))
395 if (PageBuddy(buddy) && page_order(buddy) == order) {
396 BUG_ON(page_count(buddy) != 0);
403 * Freeing function for a buddy system allocator.
405 * The concept of a buddy system is to maintain direct-mapped table
406 * (containing bit values) for memory blocks of various "orders".
407 * The bottom level table contains the map for the smallest allocatable
408 * units of memory (here, pages), and each level above it describes
409 * pairs of units from the levels below, hence, "buddies".
410 * At a high level, all that happens here is marking the table entry
411 * at the bottom level available, and propagating the changes upward
412 * as necessary, plus some accounting needed to play nicely with other
413 * parts of the VM system.
414 * At each level, we keep a list of pages, which are heads of continuous
415 * free pages of length of (1 << order) and marked with PG_buddy. Page's
416 * order is recorded in page_private(page) field.
417 * So when we are allocating or freeing one, we can derive the state of the
418 * other. That is, if we allocate a small block, and both were
419 * free, the remainder of the region must be split into blocks.
420 * If a block is freed, and its buddy is also free, then this
421 * triggers coalescing into a block of larger size.
426 static inline void __free_one_page(struct page *page,
427 struct zone *zone, unsigned int order)
429 unsigned long page_idx;
430 int order_size = 1 << order;
431 int migratetype = get_pageblock_migratetype(page);
433 if (unlikely(PageCompound(page)))
434 destroy_compound_page(page, order);
436 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
438 VM_BUG_ON(page_idx & (order_size - 1));
439 VM_BUG_ON(bad_range(zone, page));
441 __mod_zone_page_state(zone, NR_FREE_PAGES, order_size);
442 while (order < MAX_ORDER-1) {
443 unsigned long combined_idx;
446 buddy = __page_find_buddy(page, page_idx, order);
447 if (!page_is_buddy(page, buddy, order))
448 break; /* Move the buddy up one level. */
450 list_del(&buddy->lru);
451 zone->free_area[order].nr_free--;
452 rmv_page_order(buddy);
453 combined_idx = __find_combined_index(page_idx, order);
454 page = page + (combined_idx - page_idx);
455 page_idx = combined_idx;
458 set_page_order(page, order);
460 &zone->free_area[order].free_list[migratetype]);
461 zone->free_area[order].nr_free++;
464 static inline int free_pages_check(struct page *page)
466 if (unlikely(page_mapcount(page) |
467 (page->mapping != NULL) |
468 (page_count(page) != 0) |
481 __ClearPageDirty(page);
483 * For now, we report if PG_reserved was found set, but do not
484 * clear it, and do not free the page. But we shall soon need
485 * to do more, for when the ZERO_PAGE count wraps negative.
487 return PageReserved(page);
491 * Frees a list of pages.
492 * Assumes all pages on list are in same zone, and of same order.
493 * count is the number of pages to free.
495 * If the zone was previously in an "all pages pinned" state then look to
496 * see if this freeing clears that state.
498 * And clear the zone's pages_scanned counter, to hold off the "all pages are
499 * pinned" detection logic.
501 static void free_pages_bulk(struct zone *zone, int count,
502 struct list_head *list, int order)
504 spin_lock(&zone->lock);
505 zone->all_unreclaimable = 0;
506 zone->pages_scanned = 0;
510 VM_BUG_ON(list_empty(list));
511 page = list_entry(list->prev, struct page, lru);
512 /* have to delete it as __free_one_page list manipulates */
513 list_del(&page->lru);
514 __free_one_page(page, zone, order);
516 spin_unlock(&zone->lock);
519 static void free_one_page(struct zone *zone, struct page *page, int order)
521 spin_lock(&zone->lock);
522 zone->all_unreclaimable = 0;
523 zone->pages_scanned = 0;
524 __free_one_page(page, zone, order);
525 spin_unlock(&zone->lock);
528 static void __free_pages_ok(struct page *page, unsigned int order)
534 for (i = 0 ; i < (1 << order) ; ++i)
535 reserved += free_pages_check(page + i);
539 if (!PageHighMem(page))
540 debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order);
541 arch_free_page(page, order);
542 kernel_map_pages(page, 1 << order, 0);
544 local_irq_save(flags);
545 __count_vm_events(PGFREE, 1 << order);
546 free_one_page(page_zone(page), page, order);
547 local_irq_restore(flags);
551 * permit the bootmem allocator to evade page validation on high-order frees
553 void fastcall __init __free_pages_bootmem(struct page *page, unsigned int order)
556 __ClearPageReserved(page);
557 set_page_count(page, 0);
558 set_page_refcounted(page);
564 for (loop = 0; loop < BITS_PER_LONG; loop++) {
565 struct page *p = &page[loop];
567 if (loop + 1 < BITS_PER_LONG)
569 __ClearPageReserved(p);
570 set_page_count(p, 0);
573 set_page_refcounted(page);
574 __free_pages(page, order);
580 * The order of subdivision here is critical for the IO subsystem.
581 * Please do not alter this order without good reasons and regression
582 * testing. Specifically, as large blocks of memory are subdivided,
583 * the order in which smaller blocks are delivered depends on the order
584 * they're subdivided in this function. This is the primary factor
585 * influencing the order in which pages are delivered to the IO
586 * subsystem according to empirical testing, and this is also justified
587 * by considering the behavior of a buddy system containing a single
588 * large block of memory acted on by a series of small allocations.
589 * This behavior is a critical factor in sglist merging's success.
593 static inline void expand(struct zone *zone, struct page *page,
594 int low, int high, struct free_area *area,
597 unsigned long size = 1 << high;
603 VM_BUG_ON(bad_range(zone, &page[size]));
604 list_add(&page[size].lru, &area->free_list[migratetype]);
606 set_page_order(&page[size], high);
611 * This page is about to be returned from the page allocator
613 static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
615 if (unlikely(page_mapcount(page) |
616 (page->mapping != NULL) |
617 (page_count(page) != 0) |
632 * For now, we report if PG_reserved was found set, but do not
633 * clear it, and do not allocate the page: as a safety net.
635 if (PageReserved(page))
638 page->flags &= ~(1 << PG_uptodate | 1 << PG_error | 1 << PG_readahead |
639 1 << PG_referenced | 1 << PG_arch_1 |
640 1 << PG_owner_priv_1 | 1 << PG_mappedtodisk);
641 set_page_private(page, 0);
642 set_page_refcounted(page);
644 arch_alloc_page(page, order);
645 kernel_map_pages(page, 1 << order, 1);
647 if (gfp_flags & __GFP_ZERO)
648 prep_zero_page(page, order, gfp_flags);
650 if (order && (gfp_flags & __GFP_COMP))
651 prep_compound_page(page, order);
657 * This array describes the order lists are fallen back to when
658 * the free lists for the desirable migrate type are depleted
660 static int fallbacks[MIGRATE_TYPES][MIGRATE_TYPES-1] = {
661 [MIGRATE_UNMOVABLE] = { MIGRATE_MOVABLE },
662 [MIGRATE_MOVABLE] = { MIGRATE_UNMOVABLE },
665 /* Remove an element from the buddy allocator from the fallback list */
666 static struct page *__rmqueue_fallback(struct zone *zone, int order,
667 int start_migratetype)
669 struct free_area * area;
674 /* Find the largest possible block of pages in the other list */
675 for (current_order = MAX_ORDER-1; current_order >= order;
677 for (i = 0; i < MIGRATE_TYPES - 1; i++) {
678 migratetype = fallbacks[start_migratetype][i];
680 area = &(zone->free_area[current_order]);
681 if (list_empty(&area->free_list[migratetype]))
684 page = list_entry(area->free_list[migratetype].next,
689 * If breaking a large block of pages, place the buddies
690 * on the preferred allocation list
692 if (unlikely(current_order >= MAX_ORDER / 2))
693 migratetype = start_migratetype;
695 /* Remove the page from the freelists */
696 list_del(&page->lru);
697 rmv_page_order(page);
698 __mod_zone_page_state(zone, NR_FREE_PAGES,
701 if (current_order == MAX_ORDER - 1)
702 set_pageblock_migratetype(page,
705 expand(zone, page, order, current_order, area, migratetype);
714 * Do the hard work of removing an element from the buddy allocator.
715 * Call me with the zone->lock already held.
717 static struct page *__rmqueue(struct zone *zone, unsigned int order,
720 struct free_area * area;
721 unsigned int current_order;
724 /* Find a page of the appropriate size in the preferred list */
725 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
726 area = &(zone->free_area[current_order]);
727 if (list_empty(&area->free_list[migratetype]))
730 page = list_entry(area->free_list[migratetype].next,
732 list_del(&page->lru);
733 rmv_page_order(page);
735 __mod_zone_page_state(zone, NR_FREE_PAGES, - (1UL << order));
736 expand(zone, page, order, current_order, area, migratetype);
740 page = __rmqueue_fallback(zone, order, migratetype);
748 * Obtain a specified number of elements from the buddy allocator, all under
749 * a single hold of the lock, for efficiency. Add them to the supplied list.
750 * Returns the number of new pages which were placed at *list.
752 static int rmqueue_bulk(struct zone *zone, unsigned int order,
753 unsigned long count, struct list_head *list,
758 spin_lock(&zone->lock);
759 for (i = 0; i < count; ++i) {
760 struct page *page = __rmqueue(zone, order, migratetype);
761 if (unlikely(page == NULL))
763 list_add(&page->lru, list);
764 set_page_private(page, migratetype);
766 spin_unlock(&zone->lock);
772 * Called from the vmstat counter updater to drain pagesets of this
773 * currently executing processor on remote nodes after they have
776 * Note that this function must be called with the thread pinned to
777 * a single processor.
779 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
784 local_irq_save(flags);
785 if (pcp->count >= pcp->batch)
786 to_drain = pcp->batch;
788 to_drain = pcp->count;
789 free_pages_bulk(zone, to_drain, &pcp->list, 0);
790 pcp->count -= to_drain;
791 local_irq_restore(flags);
795 static void __drain_pages(unsigned int cpu)
801 for_each_zone(zone) {
802 struct per_cpu_pageset *pset;
804 if (!populated_zone(zone))
807 pset = zone_pcp(zone, cpu);
808 for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) {
809 struct per_cpu_pages *pcp;
812 local_irq_save(flags);
813 free_pages_bulk(zone, pcp->count, &pcp->list, 0);
815 local_irq_restore(flags);
820 #ifdef CONFIG_HIBERNATION
822 void mark_free_pages(struct zone *zone)
824 unsigned long pfn, max_zone_pfn;
827 struct list_head *curr;
829 if (!zone->spanned_pages)
832 spin_lock_irqsave(&zone->lock, flags);
834 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
835 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
836 if (pfn_valid(pfn)) {
837 struct page *page = pfn_to_page(pfn);
839 if (!swsusp_page_is_forbidden(page))
840 swsusp_unset_page_free(page);
843 for_each_migratetype_order(order, t) {
844 list_for_each(curr, &zone->free_area[order].free_list[t]) {
847 pfn = page_to_pfn(list_entry(curr, struct page, lru));
848 for (i = 0; i < (1UL << order); i++)
849 swsusp_set_page_free(pfn_to_page(pfn + i));
852 spin_unlock_irqrestore(&zone->lock, flags);
856 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
858 void drain_local_pages(void)
862 local_irq_save(flags);
863 __drain_pages(smp_processor_id());
864 local_irq_restore(flags);
866 #endif /* CONFIG_HIBERNATION */
869 * Free a 0-order page
871 static void fastcall free_hot_cold_page(struct page *page, int cold)
873 struct zone *zone = page_zone(page);
874 struct per_cpu_pages *pcp;
878 page->mapping = NULL;
879 if (free_pages_check(page))
882 if (!PageHighMem(page))
883 debug_check_no_locks_freed(page_address(page), PAGE_SIZE);
884 arch_free_page(page, 0);
885 kernel_map_pages(page, 1, 0);
887 pcp = &zone_pcp(zone, get_cpu())->pcp[cold];
888 local_irq_save(flags);
889 __count_vm_event(PGFREE);
890 list_add(&page->lru, &pcp->list);
891 set_page_private(page, get_pageblock_migratetype(page));
893 if (pcp->count >= pcp->high) {
894 free_pages_bulk(zone, pcp->batch, &pcp->list, 0);
895 pcp->count -= pcp->batch;
897 local_irq_restore(flags);
901 void fastcall free_hot_page(struct page *page)
903 free_hot_cold_page(page, 0);
906 void fastcall free_cold_page(struct page *page)
908 free_hot_cold_page(page, 1);
912 * split_page takes a non-compound higher-order page, and splits it into
913 * n (1<<order) sub-pages: page[0..n]
914 * Each sub-page must be freed individually.
916 * Note: this is probably too low level an operation for use in drivers.
917 * Please consult with lkml before using this in your driver.
919 void split_page(struct page *page, unsigned int order)
923 VM_BUG_ON(PageCompound(page));
924 VM_BUG_ON(!page_count(page));
925 for (i = 1; i < (1 << order); i++)
926 set_page_refcounted(page + i);
930 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
931 * we cheat by calling it from here, in the order > 0 path. Saves a branch
934 static struct page *buffered_rmqueue(struct zonelist *zonelist,
935 struct zone *zone, int order, gfp_t gfp_flags)
939 int cold = !!(gfp_flags & __GFP_COLD);
941 int migratetype = gfpflags_to_migratetype(gfp_flags);
945 if (likely(order == 0)) {
946 struct per_cpu_pages *pcp;
948 pcp = &zone_pcp(zone, cpu)->pcp[cold];
949 local_irq_save(flags);
951 pcp->count = rmqueue_bulk(zone, 0,
952 pcp->batch, &pcp->list, migratetype);
953 if (unlikely(!pcp->count))
956 /* Find a page of the appropriate migrate type */
957 list_for_each_entry(page, &pcp->list, lru) {
958 if (page_private(page) == migratetype) {
959 list_del(&page->lru);
966 * Check if a page of the appropriate migrate type
967 * was found. If not, allocate more to the pcp list
969 if (&page->lru == &pcp->list) {
970 pcp->count += rmqueue_bulk(zone, 0,
971 pcp->batch, &pcp->list, migratetype);
972 page = list_entry(pcp->list.next, struct page, lru);
973 VM_BUG_ON(page_private(page) != migratetype);
974 list_del(&page->lru);
978 spin_lock_irqsave(&zone->lock, flags);
979 page = __rmqueue(zone, order, migratetype);
980 spin_unlock(&zone->lock);
985 __count_zone_vm_events(PGALLOC, zone, 1 << order);
986 zone_statistics(zonelist, zone);
987 local_irq_restore(flags);
990 VM_BUG_ON(bad_range(zone, page));
991 if (prep_new_page(page, order, gfp_flags))
996 local_irq_restore(flags);
1001 #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
1002 #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
1003 #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
1004 #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
1005 #define ALLOC_HARDER 0x10 /* try to alloc harder */
1006 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1007 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
1009 #ifdef CONFIG_FAIL_PAGE_ALLOC
1011 static struct fail_page_alloc_attr {
1012 struct fault_attr attr;
1014 u32 ignore_gfp_highmem;
1015 u32 ignore_gfp_wait;
1018 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1020 struct dentry *ignore_gfp_highmem_file;
1021 struct dentry *ignore_gfp_wait_file;
1022 struct dentry *min_order_file;
1024 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1026 } fail_page_alloc = {
1027 .attr = FAULT_ATTR_INITIALIZER,
1028 .ignore_gfp_wait = 1,
1029 .ignore_gfp_highmem = 1,
1033 static int __init setup_fail_page_alloc(char *str)
1035 return setup_fault_attr(&fail_page_alloc.attr, str);
1037 __setup("fail_page_alloc=", setup_fail_page_alloc);
1039 static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1041 if (order < fail_page_alloc.min_order)
1043 if (gfp_mask & __GFP_NOFAIL)
1045 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
1047 if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
1050 return should_fail(&fail_page_alloc.attr, 1 << order);
1053 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1055 static int __init fail_page_alloc_debugfs(void)
1057 mode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
1061 err = init_fault_attr_dentries(&fail_page_alloc.attr,
1065 dir = fail_page_alloc.attr.dentries.dir;
1067 fail_page_alloc.ignore_gfp_wait_file =
1068 debugfs_create_bool("ignore-gfp-wait", mode, dir,
1069 &fail_page_alloc.ignore_gfp_wait);
1071 fail_page_alloc.ignore_gfp_highmem_file =
1072 debugfs_create_bool("ignore-gfp-highmem", mode, dir,
1073 &fail_page_alloc.ignore_gfp_highmem);
1074 fail_page_alloc.min_order_file =
1075 debugfs_create_u32("min-order", mode, dir,
1076 &fail_page_alloc.min_order);
1078 if (!fail_page_alloc.ignore_gfp_wait_file ||
1079 !fail_page_alloc.ignore_gfp_highmem_file ||
1080 !fail_page_alloc.min_order_file) {
1082 debugfs_remove(fail_page_alloc.ignore_gfp_wait_file);
1083 debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file);
1084 debugfs_remove(fail_page_alloc.min_order_file);
1085 cleanup_fault_attr_dentries(&fail_page_alloc.attr);
1091 late_initcall(fail_page_alloc_debugfs);
1093 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1095 #else /* CONFIG_FAIL_PAGE_ALLOC */
1097 static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1102 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1105 * Return 1 if free pages are above 'mark'. This takes into account the order
1106 * of the allocation.
1108 int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1109 int classzone_idx, int alloc_flags)
1111 /* free_pages my go negative - that's OK */
1113 long free_pages = zone_page_state(z, NR_FREE_PAGES) - (1 << order) + 1;
1116 if (alloc_flags & ALLOC_HIGH)
1118 if (alloc_flags & ALLOC_HARDER)
1121 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
1123 for (o = 0; o < order; o++) {
1124 /* At the next order, this order's pages become unavailable */
1125 free_pages -= z->free_area[o].nr_free << o;
1127 /* Require fewer higher order pages to be free */
1130 if (free_pages <= min)
1138 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1139 * skip over zones that are not allowed by the cpuset, or that have
1140 * been recently (in last second) found to be nearly full. See further
1141 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1142 * that have to skip over alot of full or unallowed zones.
1144 * If the zonelist cache is present in the passed in zonelist, then
1145 * returns a pointer to the allowed node mask (either the current
1146 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
1148 * If the zonelist cache is not available for this zonelist, does
1149 * nothing and returns NULL.
1151 * If the fullzones BITMAP in the zonelist cache is stale (more than
1152 * a second since last zap'd) then we zap it out (clear its bits.)
1154 * We hold off even calling zlc_setup, until after we've checked the
1155 * first zone in the zonelist, on the theory that most allocations will
1156 * be satisfied from that first zone, so best to examine that zone as
1157 * quickly as we can.
1159 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1161 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1162 nodemask_t *allowednodes; /* zonelist_cache approximation */
1164 zlc = zonelist->zlcache_ptr;
1168 if (jiffies - zlc->last_full_zap > 1 * HZ) {
1169 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1170 zlc->last_full_zap = jiffies;
1173 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1174 &cpuset_current_mems_allowed :
1175 &node_states[N_HIGH_MEMORY];
1176 return allowednodes;
1180 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1181 * if it is worth looking at further for free memory:
1182 * 1) Check that the zone isn't thought to be full (doesn't have its
1183 * bit set in the zonelist_cache fullzones BITMAP).
1184 * 2) Check that the zones node (obtained from the zonelist_cache
1185 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1186 * Return true (non-zero) if zone is worth looking at further, or
1187 * else return false (zero) if it is not.
1189 * This check -ignores- the distinction between various watermarks,
1190 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1191 * found to be full for any variation of these watermarks, it will
1192 * be considered full for up to one second by all requests, unless
1193 * we are so low on memory on all allowed nodes that we are forced
1194 * into the second scan of the zonelist.
1196 * In the second scan we ignore this zonelist cache and exactly
1197 * apply the watermarks to all zones, even it is slower to do so.
1198 * We are low on memory in the second scan, and should leave no stone
1199 * unturned looking for a free page.
1201 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zone **z,
1202 nodemask_t *allowednodes)
1204 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1205 int i; /* index of *z in zonelist zones */
1206 int n; /* node that zone *z is on */
1208 zlc = zonelist->zlcache_ptr;
1212 i = z - zonelist->zones;
1215 /* This zone is worth trying if it is allowed but not full */
1216 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1220 * Given 'z' scanning a zonelist, set the corresponding bit in
1221 * zlc->fullzones, so that subsequent attempts to allocate a page
1222 * from that zone don't waste time re-examining it.
1224 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zone **z)
1226 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1227 int i; /* index of *z in zonelist zones */
1229 zlc = zonelist->zlcache_ptr;
1233 i = z - zonelist->zones;
1235 set_bit(i, zlc->fullzones);
1238 #else /* CONFIG_NUMA */
1240 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1245 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zone **z,
1246 nodemask_t *allowednodes)
1251 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zone **z)
1254 #endif /* CONFIG_NUMA */
1257 * get_page_from_freelist goes through the zonelist trying to allocate
1260 static struct page *
1261 get_page_from_freelist(gfp_t gfp_mask, unsigned int order,
1262 struct zonelist *zonelist, int alloc_flags)
1265 struct page *page = NULL;
1266 int classzone_idx = zone_idx(zonelist->zones[0]);
1268 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1269 int zlc_active = 0; /* set if using zonelist_cache */
1270 int did_zlc_setup = 0; /* just call zlc_setup() one time */
1271 enum zone_type highest_zoneidx = -1; /* Gets set for policy zonelists */
1275 * Scan zonelist, looking for a zone with enough free.
1276 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1278 z = zonelist->zones;
1282 * In NUMA, this could be a policy zonelist which contains
1283 * zones that may not be allowed by the current gfp_mask.
1284 * Check the zone is allowed by the current flags
1286 if (unlikely(alloc_should_filter_zonelist(zonelist))) {
1287 if (highest_zoneidx == -1)
1288 highest_zoneidx = gfp_zone(gfp_mask);
1289 if (zone_idx(*z) > highest_zoneidx)
1293 if (NUMA_BUILD && zlc_active &&
1294 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1297 if ((alloc_flags & ALLOC_CPUSET) &&
1298 !cpuset_zone_allowed_softwall(zone, gfp_mask))
1301 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
1303 if (alloc_flags & ALLOC_WMARK_MIN)
1304 mark = zone->pages_min;
1305 else if (alloc_flags & ALLOC_WMARK_LOW)
1306 mark = zone->pages_low;
1308 mark = zone->pages_high;
1309 if (!zone_watermark_ok(zone, order, mark,
1310 classzone_idx, alloc_flags)) {
1311 if (!zone_reclaim_mode ||
1312 !zone_reclaim(zone, gfp_mask, order))
1313 goto this_zone_full;
1317 page = buffered_rmqueue(zonelist, zone, order, gfp_mask);
1322 zlc_mark_zone_full(zonelist, z);
1324 if (NUMA_BUILD && !did_zlc_setup) {
1325 /* we do zlc_setup after the first zone is tried */
1326 allowednodes = zlc_setup(zonelist, alloc_flags);
1330 } while (*(++z) != NULL);
1332 if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) {
1333 /* Disable zlc cache for second zonelist scan */
1341 * This is the 'heart' of the zoned buddy allocator.
1343 struct page * fastcall
1344 __alloc_pages(gfp_t gfp_mask, unsigned int order,
1345 struct zonelist *zonelist)
1347 const gfp_t wait = gfp_mask & __GFP_WAIT;
1350 struct reclaim_state reclaim_state;
1351 struct task_struct *p = current;
1354 int did_some_progress;
1356 might_sleep_if(wait);
1358 if (should_fail_alloc_page(gfp_mask, order))
1362 z = zonelist->zones; /* the list of zones suitable for gfp_mask */
1364 if (unlikely(*z == NULL)) {
1366 * Happens if we have an empty zonelist as a result of
1367 * GFP_THISNODE being used on a memoryless node
1372 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
1373 zonelist, ALLOC_WMARK_LOW|ALLOC_CPUSET);
1378 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1379 * __GFP_NOWARN set) should not cause reclaim since the subsystem
1380 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1381 * using a larger set of nodes after it has established that the
1382 * allowed per node queues are empty and that nodes are
1385 if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
1388 for (z = zonelist->zones; *z; z++)
1389 wakeup_kswapd(*z, order);
1392 * OK, we're below the kswapd watermark and have kicked background
1393 * reclaim. Now things get more complex, so set up alloc_flags according
1394 * to how we want to proceed.
1396 * The caller may dip into page reserves a bit more if the caller
1397 * cannot run direct reclaim, or if the caller has realtime scheduling
1398 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1399 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1401 alloc_flags = ALLOC_WMARK_MIN;
1402 if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait)
1403 alloc_flags |= ALLOC_HARDER;
1404 if (gfp_mask & __GFP_HIGH)
1405 alloc_flags |= ALLOC_HIGH;
1407 alloc_flags |= ALLOC_CPUSET;
1410 * Go through the zonelist again. Let __GFP_HIGH and allocations
1411 * coming from realtime tasks go deeper into reserves.
1413 * This is the last chance, in general, before the goto nopage.
1414 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1415 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1417 page = get_page_from_freelist(gfp_mask, order, zonelist, alloc_flags);
1421 /* This allocation should allow future memory freeing. */
1424 if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE)))
1425 && !in_interrupt()) {
1426 if (!(gfp_mask & __GFP_NOMEMALLOC)) {
1428 /* go through the zonelist yet again, ignoring mins */
1429 page = get_page_from_freelist(gfp_mask, order,
1430 zonelist, ALLOC_NO_WATERMARKS);
1433 if (gfp_mask & __GFP_NOFAIL) {
1434 congestion_wait(WRITE, HZ/50);
1441 /* Atomic allocations - we can't balance anything */
1447 /* We now go into synchronous reclaim */
1448 cpuset_memory_pressure_bump();
1449 p->flags |= PF_MEMALLOC;
1450 reclaim_state.reclaimed_slab = 0;
1451 p->reclaim_state = &reclaim_state;
1453 did_some_progress = try_to_free_pages(zonelist->zones, order, gfp_mask);
1455 p->reclaim_state = NULL;
1456 p->flags &= ~PF_MEMALLOC;
1460 if (likely(did_some_progress)) {
1461 page = get_page_from_freelist(gfp_mask, order,
1462 zonelist, alloc_flags);
1465 } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
1467 * Go through the zonelist yet one more time, keep
1468 * very high watermark here, this is only to catch
1469 * a parallel oom killing, we must fail if we're still
1470 * under heavy pressure.
1472 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
1473 zonelist, ALLOC_WMARK_HIGH|ALLOC_CPUSET);
1477 /* The OOM killer will not help higher order allocs so fail */
1478 if (order > PAGE_ALLOC_COSTLY_ORDER)
1481 out_of_memory(zonelist, gfp_mask, order);
1486 * Don't let big-order allocations loop unless the caller explicitly
1487 * requests that. Wait for some write requests to complete then retry.
1489 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
1490 * <= 3, but that may not be true in other implementations.
1493 if (!(gfp_mask & __GFP_NORETRY)) {
1494 if ((order <= PAGE_ALLOC_COSTLY_ORDER) ||
1495 (gfp_mask & __GFP_REPEAT))
1497 if (gfp_mask & __GFP_NOFAIL)
1501 congestion_wait(WRITE, HZ/50);
1506 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
1507 printk(KERN_WARNING "%s: page allocation failure."
1508 " order:%d, mode:0x%x\n",
1509 p->comm, order, gfp_mask);
1517 EXPORT_SYMBOL(__alloc_pages);
1520 * Common helper functions.
1522 fastcall unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
1525 page = alloc_pages(gfp_mask, order);
1528 return (unsigned long) page_address(page);
1531 EXPORT_SYMBOL(__get_free_pages);
1533 fastcall unsigned long get_zeroed_page(gfp_t gfp_mask)
1538 * get_zeroed_page() returns a 32-bit address, which cannot represent
1541 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
1543 page = alloc_pages(gfp_mask | __GFP_ZERO, 0);
1545 return (unsigned long) page_address(page);
1549 EXPORT_SYMBOL(get_zeroed_page);
1551 void __pagevec_free(struct pagevec *pvec)
1553 int i = pagevec_count(pvec);
1556 free_hot_cold_page(pvec->pages[i], pvec->cold);
1559 fastcall void __free_pages(struct page *page, unsigned int order)
1561 if (put_page_testzero(page)) {
1563 free_hot_page(page);
1565 __free_pages_ok(page, order);
1569 EXPORT_SYMBOL(__free_pages);
1571 fastcall void free_pages(unsigned long addr, unsigned int order)
1574 VM_BUG_ON(!virt_addr_valid((void *)addr));
1575 __free_pages(virt_to_page((void *)addr), order);
1579 EXPORT_SYMBOL(free_pages);
1581 static unsigned int nr_free_zone_pages(int offset)
1583 /* Just pick one node, since fallback list is circular */
1584 pg_data_t *pgdat = NODE_DATA(numa_node_id());
1585 unsigned int sum = 0;
1587 struct zonelist *zonelist = pgdat->node_zonelists + offset;
1588 struct zone **zonep = zonelist->zones;
1591 for (zone = *zonep++; zone; zone = *zonep++) {
1592 unsigned long size = zone->present_pages;
1593 unsigned long high = zone->pages_high;
1602 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1604 unsigned int nr_free_buffer_pages(void)
1606 return nr_free_zone_pages(gfp_zone(GFP_USER));
1608 EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
1611 * Amount of free RAM allocatable within all zones
1613 unsigned int nr_free_pagecache_pages(void)
1615 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
1618 static inline void show_node(struct zone *zone)
1621 printk("Node %d ", zone_to_nid(zone));
1624 void si_meminfo(struct sysinfo *val)
1626 val->totalram = totalram_pages;
1628 val->freeram = global_page_state(NR_FREE_PAGES);
1629 val->bufferram = nr_blockdev_pages();
1630 val->totalhigh = totalhigh_pages;
1631 val->freehigh = nr_free_highpages();
1632 val->mem_unit = PAGE_SIZE;
1635 EXPORT_SYMBOL(si_meminfo);
1638 void si_meminfo_node(struct sysinfo *val, int nid)
1640 pg_data_t *pgdat = NODE_DATA(nid);
1642 val->totalram = pgdat->node_present_pages;
1643 val->freeram = node_page_state(nid, NR_FREE_PAGES);
1644 #ifdef CONFIG_HIGHMEM
1645 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1646 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
1652 val->mem_unit = PAGE_SIZE;
1656 #define K(x) ((x) << (PAGE_SHIFT-10))
1659 * Show free area list (used inside shift_scroll-lock stuff)
1660 * We also calculate the percentage fragmentation. We do this by counting the
1661 * memory on each free list with the exception of the first item on the list.
1663 void show_free_areas(void)
1668 for_each_zone(zone) {
1669 if (!populated_zone(zone))
1673 printk("%s per-cpu:\n", zone->name);
1675 for_each_online_cpu(cpu) {
1676 struct per_cpu_pageset *pageset;
1678 pageset = zone_pcp(zone, cpu);
1680 printk("CPU %4d: Hot: hi:%5d, btch:%4d usd:%4d "
1681 "Cold: hi:%5d, btch:%4d usd:%4d\n",
1682 cpu, pageset->pcp[0].high,
1683 pageset->pcp[0].batch, pageset->pcp[0].count,
1684 pageset->pcp[1].high, pageset->pcp[1].batch,
1685 pageset->pcp[1].count);
1689 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu unstable:%lu\n"
1690 " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
1691 global_page_state(NR_ACTIVE),
1692 global_page_state(NR_INACTIVE),
1693 global_page_state(NR_FILE_DIRTY),
1694 global_page_state(NR_WRITEBACK),
1695 global_page_state(NR_UNSTABLE_NFS),
1696 global_page_state(NR_FREE_PAGES),
1697 global_page_state(NR_SLAB_RECLAIMABLE) +
1698 global_page_state(NR_SLAB_UNRECLAIMABLE),
1699 global_page_state(NR_FILE_MAPPED),
1700 global_page_state(NR_PAGETABLE),
1701 global_page_state(NR_BOUNCE));
1703 for_each_zone(zone) {
1706 if (!populated_zone(zone))
1718 " pages_scanned:%lu"
1719 " all_unreclaimable? %s"
1722 K(zone_page_state(zone, NR_FREE_PAGES)),
1725 K(zone->pages_high),
1726 K(zone_page_state(zone, NR_ACTIVE)),
1727 K(zone_page_state(zone, NR_INACTIVE)),
1728 K(zone->present_pages),
1729 zone->pages_scanned,
1730 (zone->all_unreclaimable ? "yes" : "no")
1732 printk("lowmem_reserve[]:");
1733 for (i = 0; i < MAX_NR_ZONES; i++)
1734 printk(" %lu", zone->lowmem_reserve[i]);
1738 for_each_zone(zone) {
1739 unsigned long nr[MAX_ORDER], flags, order, total = 0;
1741 if (!populated_zone(zone))
1745 printk("%s: ", zone->name);
1747 spin_lock_irqsave(&zone->lock, flags);
1748 for (order = 0; order < MAX_ORDER; order++) {
1749 nr[order] = zone->free_area[order].nr_free;
1750 total += nr[order] << order;
1752 spin_unlock_irqrestore(&zone->lock, flags);
1753 for (order = 0; order < MAX_ORDER; order++)
1754 printk("%lu*%lukB ", nr[order], K(1UL) << order);
1755 printk("= %lukB\n", K(total));
1758 show_swap_cache_info();
1762 * Builds allocation fallback zone lists.
1764 * Add all populated zones of a node to the zonelist.
1766 static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
1767 int nr_zones, enum zone_type zone_type)
1771 BUG_ON(zone_type >= MAX_NR_ZONES);
1776 zone = pgdat->node_zones + zone_type;
1777 if (populated_zone(zone)) {
1778 zonelist->zones[nr_zones++] = zone;
1779 check_highest_zone(zone_type);
1782 } while (zone_type);
1789 * 0 = automatic detection of better ordering.
1790 * 1 = order by ([node] distance, -zonetype)
1791 * 2 = order by (-zonetype, [node] distance)
1793 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
1794 * the same zonelist. So only NUMA can configure this param.
1796 #define ZONELIST_ORDER_DEFAULT 0
1797 #define ZONELIST_ORDER_NODE 1
1798 #define ZONELIST_ORDER_ZONE 2
1800 /* zonelist order in the kernel.
1801 * set_zonelist_order() will set this to NODE or ZONE.
1803 static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
1804 static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
1808 /* The value user specified ....changed by config */
1809 static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
1810 /* string for sysctl */
1811 #define NUMA_ZONELIST_ORDER_LEN 16
1812 char numa_zonelist_order[16] = "default";
1815 * interface for configure zonelist ordering.
1816 * command line option "numa_zonelist_order"
1817 * = "[dD]efault - default, automatic configuration.
1818 * = "[nN]ode - order by node locality, then by zone within node
1819 * = "[zZ]one - order by zone, then by locality within zone
1822 static int __parse_numa_zonelist_order(char *s)
1824 if (*s == 'd' || *s == 'D') {
1825 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
1826 } else if (*s == 'n' || *s == 'N') {
1827 user_zonelist_order = ZONELIST_ORDER_NODE;
1828 } else if (*s == 'z' || *s == 'Z') {
1829 user_zonelist_order = ZONELIST_ORDER_ZONE;
1832 "Ignoring invalid numa_zonelist_order value: "
1839 static __init int setup_numa_zonelist_order(char *s)
1842 return __parse_numa_zonelist_order(s);
1845 early_param("numa_zonelist_order", setup_numa_zonelist_order);
1848 * sysctl handler for numa_zonelist_order
1850 int numa_zonelist_order_handler(ctl_table *table, int write,
1851 struct file *file, void __user *buffer, size_t *length,
1854 char saved_string[NUMA_ZONELIST_ORDER_LEN];
1858 strncpy(saved_string, (char*)table->data,
1859 NUMA_ZONELIST_ORDER_LEN);
1860 ret = proc_dostring(table, write, file, buffer, length, ppos);
1864 int oldval = user_zonelist_order;
1865 if (__parse_numa_zonelist_order((char*)table->data)) {
1867 * bogus value. restore saved string
1869 strncpy((char*)table->data, saved_string,
1870 NUMA_ZONELIST_ORDER_LEN);
1871 user_zonelist_order = oldval;
1872 } else if (oldval != user_zonelist_order)
1873 build_all_zonelists();
1879 #define MAX_NODE_LOAD (num_online_nodes())
1880 static int node_load[MAX_NUMNODES];
1883 * find_next_best_node - find the next node that should appear in a given node's fallback list
1884 * @node: node whose fallback list we're appending
1885 * @used_node_mask: nodemask_t of already used nodes
1887 * We use a number of factors to determine which is the next node that should
1888 * appear on a given node's fallback list. The node should not have appeared
1889 * already in @node's fallback list, and it should be the next closest node
1890 * according to the distance array (which contains arbitrary distance values
1891 * from each node to each node in the system), and should also prefer nodes
1892 * with no CPUs, since presumably they'll have very little allocation pressure
1893 * on them otherwise.
1894 * It returns -1 if no node is found.
1896 static int find_next_best_node(int node, nodemask_t *used_node_mask)
1899 int min_val = INT_MAX;
1902 /* Use the local node if we haven't already */
1903 if (!node_isset(node, *used_node_mask)) {
1904 node_set(node, *used_node_mask);
1908 for_each_node_state(n, N_HIGH_MEMORY) {
1911 /* Don't want a node to appear more than once */
1912 if (node_isset(n, *used_node_mask))
1915 /* Use the distance array to find the distance */
1916 val = node_distance(node, n);
1918 /* Penalize nodes under us ("prefer the next node") */
1921 /* Give preference to headless and unused nodes */
1922 tmp = node_to_cpumask(n);
1923 if (!cpus_empty(tmp))
1924 val += PENALTY_FOR_NODE_WITH_CPUS;
1926 /* Slight preference for less loaded node */
1927 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
1928 val += node_load[n];
1930 if (val < min_val) {
1937 node_set(best_node, *used_node_mask);
1944 * Build zonelists ordered by node and zones within node.
1945 * This results in maximum locality--normal zone overflows into local
1946 * DMA zone, if any--but risks exhausting DMA zone.
1948 static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
1952 struct zonelist *zonelist;
1954 for (i = 0; i < MAX_NR_ZONES; i++) {
1955 zonelist = pgdat->node_zonelists + i;
1956 for (j = 0; zonelist->zones[j] != NULL; j++)
1958 j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
1959 zonelist->zones[j] = NULL;
1964 * Build gfp_thisnode zonelists
1966 static void build_thisnode_zonelists(pg_data_t *pgdat)
1970 struct zonelist *zonelist;
1972 for (i = 0; i < MAX_NR_ZONES; i++) {
1973 zonelist = pgdat->node_zonelists + MAX_NR_ZONES + i;
1974 j = build_zonelists_node(pgdat, zonelist, 0, i);
1975 zonelist->zones[j] = NULL;
1980 * Build zonelists ordered by zone and nodes within zones.
1981 * This results in conserving DMA zone[s] until all Normal memory is
1982 * exhausted, but results in overflowing to remote node while memory
1983 * may still exist in local DMA zone.
1985 static int node_order[MAX_NUMNODES];
1987 static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
1991 int zone_type; /* needs to be signed */
1993 struct zonelist *zonelist;
1995 for (i = 0; i < MAX_NR_ZONES; i++) {
1996 zonelist = pgdat->node_zonelists + i;
1998 for (zone_type = i; zone_type >= 0; zone_type--) {
1999 for (j = 0; j < nr_nodes; j++) {
2000 node = node_order[j];
2001 z = &NODE_DATA(node)->node_zones[zone_type];
2002 if (populated_zone(z)) {
2003 zonelist->zones[pos++] = z;
2004 check_highest_zone(zone_type);
2008 zonelist->zones[pos] = NULL;
2012 static int default_zonelist_order(void)
2015 unsigned long low_kmem_size,total_size;
2019 * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem.
2020 * If they are really small and used heavily, the system can fall
2021 * into OOM very easily.
2022 * This function detect ZONE_DMA/DMA32 size and confgigures zone order.
2024 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2027 for_each_online_node(nid) {
2028 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2029 z = &NODE_DATA(nid)->node_zones[zone_type];
2030 if (populated_zone(z)) {
2031 if (zone_type < ZONE_NORMAL)
2032 low_kmem_size += z->present_pages;
2033 total_size += z->present_pages;
2037 if (!low_kmem_size || /* there are no DMA area. */
2038 low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
2039 return ZONELIST_ORDER_NODE;
2041 * look into each node's config.
2042 * If there is a node whose DMA/DMA32 memory is very big area on
2043 * local memory, NODE_ORDER may be suitable.
2045 average_size = total_size /
2046 (nodes_weight(node_states[N_HIGH_MEMORY]) + 1);
2047 for_each_online_node(nid) {
2050 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2051 z = &NODE_DATA(nid)->node_zones[zone_type];
2052 if (populated_zone(z)) {
2053 if (zone_type < ZONE_NORMAL)
2054 low_kmem_size += z->present_pages;
2055 total_size += z->present_pages;
2058 if (low_kmem_size &&
2059 total_size > average_size && /* ignore small node */
2060 low_kmem_size > total_size * 70/100)
2061 return ZONELIST_ORDER_NODE;
2063 return ZONELIST_ORDER_ZONE;
2066 static void set_zonelist_order(void)
2068 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
2069 current_zonelist_order = default_zonelist_order();
2071 current_zonelist_order = user_zonelist_order;
2074 static void build_zonelists(pg_data_t *pgdat)
2078 nodemask_t used_mask;
2079 int local_node, prev_node;
2080 struct zonelist *zonelist;
2081 int order = current_zonelist_order;
2083 /* initialize zonelists */
2084 for (i = 0; i < MAX_ZONELISTS; i++) {
2085 zonelist = pgdat->node_zonelists + i;
2086 zonelist->zones[0] = NULL;
2089 /* NUMA-aware ordering of nodes */
2090 local_node = pgdat->node_id;
2091 load = num_online_nodes();
2092 prev_node = local_node;
2093 nodes_clear(used_mask);
2095 memset(node_load, 0, sizeof(node_load));
2096 memset(node_order, 0, sizeof(node_order));
2099 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
2100 int distance = node_distance(local_node, node);
2103 * If another node is sufficiently far away then it is better
2104 * to reclaim pages in a zone before going off node.
2106 if (distance > RECLAIM_DISTANCE)
2107 zone_reclaim_mode = 1;
2110 * We don't want to pressure a particular node.
2111 * So adding penalty to the first node in same
2112 * distance group to make it round-robin.
2114 if (distance != node_distance(local_node, prev_node))
2115 node_load[node] = load;
2119 if (order == ZONELIST_ORDER_NODE)
2120 build_zonelists_in_node_order(pgdat, node);
2122 node_order[j++] = node; /* remember order */
2125 if (order == ZONELIST_ORDER_ZONE) {
2126 /* calculate node order -- i.e., DMA last! */
2127 build_zonelists_in_zone_order(pgdat, j);
2130 build_thisnode_zonelists(pgdat);
2133 /* Construct the zonelist performance cache - see further mmzone.h */
2134 static void build_zonelist_cache(pg_data_t *pgdat)
2138 for (i = 0; i < MAX_NR_ZONES; i++) {
2139 struct zonelist *zonelist;
2140 struct zonelist_cache *zlc;
2143 zonelist = pgdat->node_zonelists + i;
2144 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
2145 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
2146 for (z = zonelist->zones; *z; z++)
2147 zlc->z_to_n[z - zonelist->zones] = zone_to_nid(*z);
2152 #else /* CONFIG_NUMA */
2154 static void set_zonelist_order(void)
2156 current_zonelist_order = ZONELIST_ORDER_ZONE;
2159 static void build_zonelists(pg_data_t *pgdat)
2161 int node, local_node;
2164 local_node = pgdat->node_id;
2165 for (i = 0; i < MAX_NR_ZONES; i++) {
2166 struct zonelist *zonelist;
2168 zonelist = pgdat->node_zonelists + i;
2170 j = build_zonelists_node(pgdat, zonelist, 0, i);
2172 * Now we build the zonelist so that it contains the zones
2173 * of all the other nodes.
2174 * We don't want to pressure a particular node, so when
2175 * building the zones for node N, we make sure that the
2176 * zones coming right after the local ones are those from
2177 * node N+1 (modulo N)
2179 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
2180 if (!node_online(node))
2182 j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
2184 for (node = 0; node < local_node; node++) {
2185 if (!node_online(node))
2187 j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
2190 zonelist->zones[j] = NULL;
2194 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
2195 static void build_zonelist_cache(pg_data_t *pgdat)
2199 for (i = 0; i < MAX_NR_ZONES; i++)
2200 pgdat->node_zonelists[i].zlcache_ptr = NULL;
2203 #endif /* CONFIG_NUMA */
2205 /* return values int ....just for stop_machine_run() */
2206 static int __build_all_zonelists(void *dummy)
2210 for_each_online_node(nid) {
2211 pg_data_t *pgdat = NODE_DATA(nid);
2213 build_zonelists(pgdat);
2214 build_zonelist_cache(pgdat);
2219 void build_all_zonelists(void)
2221 set_zonelist_order();
2223 if (system_state == SYSTEM_BOOTING) {
2224 __build_all_zonelists(NULL);
2225 cpuset_init_current_mems_allowed();
2227 /* we have to stop all cpus to guaranntee there is no user
2229 stop_machine_run(__build_all_zonelists, NULL, NR_CPUS);
2230 /* cpuset refresh routine should be here */
2232 vm_total_pages = nr_free_pagecache_pages();
2233 printk("Built %i zonelists in %s order. Total pages: %ld\n",
2235 zonelist_order_name[current_zonelist_order],
2238 printk("Policy zone: %s\n", zone_names[policy_zone]);
2243 * Helper functions to size the waitqueue hash table.
2244 * Essentially these want to choose hash table sizes sufficiently
2245 * large so that collisions trying to wait on pages are rare.
2246 * But in fact, the number of active page waitqueues on typical
2247 * systems is ridiculously low, less than 200. So this is even
2248 * conservative, even though it seems large.
2250 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
2251 * waitqueues, i.e. the size of the waitq table given the number of pages.
2253 #define PAGES_PER_WAITQUEUE 256
2255 #ifndef CONFIG_MEMORY_HOTPLUG
2256 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
2258 unsigned long size = 1;
2260 pages /= PAGES_PER_WAITQUEUE;
2262 while (size < pages)
2266 * Once we have dozens or even hundreds of threads sleeping
2267 * on IO we've got bigger problems than wait queue collision.
2268 * Limit the size of the wait table to a reasonable size.
2270 size = min(size, 4096UL);
2272 return max(size, 4UL);
2276 * A zone's size might be changed by hot-add, so it is not possible to determine
2277 * a suitable size for its wait_table. So we use the maximum size now.
2279 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
2281 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
2282 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
2283 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
2285 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
2286 * or more by the traditional way. (See above). It equals:
2288 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
2289 * ia64(16K page size) : = ( 8G + 4M)byte.
2290 * powerpc (64K page size) : = (32G +16M)byte.
2292 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
2299 * This is an integer logarithm so that shifts can be used later
2300 * to extract the more random high bits from the multiplicative
2301 * hash function before the remainder is taken.
2303 static inline unsigned long wait_table_bits(unsigned long size)
2308 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
2311 * Initially all pages are reserved - free ones are freed
2312 * up by free_all_bootmem() once the early boot process is
2313 * done. Non-atomic initialization, single-pass.
2315 void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
2316 unsigned long start_pfn, enum memmap_context context)
2319 unsigned long end_pfn = start_pfn + size;
2322 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
2324 * There can be holes in boot-time mem_map[]s
2325 * handed to this function. They do not
2326 * exist on hotplugged memory.
2328 if (context == MEMMAP_EARLY) {
2329 if (!early_pfn_valid(pfn))
2331 if (!early_pfn_in_nid(pfn, nid))
2334 page = pfn_to_page(pfn);
2335 set_page_links(page, zone, nid, pfn);
2336 init_page_count(page);
2337 reset_page_mapcount(page);
2338 SetPageReserved(page);
2341 * Mark the block movable so that blocks are reserved for
2342 * movable at startup. This will force kernel allocations
2343 * to reserve their blocks rather than leaking throughout
2344 * the address space during boot when many long-lived
2345 * kernel allocations are made
2347 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
2349 INIT_LIST_HEAD(&page->lru);
2350 #ifdef WANT_PAGE_VIRTUAL
2351 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
2352 if (!is_highmem_idx(zone))
2353 set_page_address(page, __va(pfn << PAGE_SHIFT));
2358 static void __meminit zone_init_free_lists(struct pglist_data *pgdat,
2359 struct zone *zone, unsigned long size)
2362 for_each_migratetype_order(order, t) {
2363 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
2364 zone->free_area[order].nr_free = 0;
2368 #ifndef __HAVE_ARCH_MEMMAP_INIT
2369 #define memmap_init(size, nid, zone, start_pfn) \
2370 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
2373 static int __devinit zone_batchsize(struct zone *zone)
2378 * The per-cpu-pages pools are set to around 1000th of the
2379 * size of the zone. But no more than 1/2 of a meg.
2381 * OK, so we don't know how big the cache is. So guess.
2383 batch = zone->present_pages / 1024;
2384 if (batch * PAGE_SIZE > 512 * 1024)
2385 batch = (512 * 1024) / PAGE_SIZE;
2386 batch /= 4; /* We effectively *= 4 below */
2391 * Clamp the batch to a 2^n - 1 value. Having a power
2392 * of 2 value was found to be more likely to have
2393 * suboptimal cache aliasing properties in some cases.
2395 * For example if 2 tasks are alternately allocating
2396 * batches of pages, one task can end up with a lot
2397 * of pages of one half of the possible page colors
2398 * and the other with pages of the other colors.
2400 batch = (1 << (fls(batch + batch/2)-1)) - 1;
2405 inline void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
2407 struct per_cpu_pages *pcp;
2409 memset(p, 0, sizeof(*p));
2411 pcp = &p->pcp[0]; /* hot */
2413 pcp->high = 6 * batch;
2414 pcp->batch = max(1UL, 1 * batch);
2415 INIT_LIST_HEAD(&pcp->list);
2417 pcp = &p->pcp[1]; /* cold*/
2419 pcp->high = 2 * batch;
2420 pcp->batch = max(1UL, batch/2);
2421 INIT_LIST_HEAD(&pcp->list);
2425 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
2426 * to the value high for the pageset p.
2429 static void setup_pagelist_highmark(struct per_cpu_pageset *p,
2432 struct per_cpu_pages *pcp;
2434 pcp = &p->pcp[0]; /* hot list */
2436 pcp->batch = max(1UL, high/4);
2437 if ((high/4) > (PAGE_SHIFT * 8))
2438 pcp->batch = PAGE_SHIFT * 8;
2444 * Boot pageset table. One per cpu which is going to be used for all
2445 * zones and all nodes. The parameters will be set in such a way
2446 * that an item put on a list will immediately be handed over to
2447 * the buddy list. This is safe since pageset manipulation is done
2448 * with interrupts disabled.
2450 * Some NUMA counter updates may also be caught by the boot pagesets.
2452 * The boot_pagesets must be kept even after bootup is complete for
2453 * unused processors and/or zones. They do play a role for bootstrapping
2454 * hotplugged processors.
2456 * zoneinfo_show() and maybe other functions do
2457 * not check if the processor is online before following the pageset pointer.
2458 * Other parts of the kernel may not check if the zone is available.
2460 static struct per_cpu_pageset boot_pageset[NR_CPUS];
2463 * Dynamically allocate memory for the
2464 * per cpu pageset array in struct zone.
2466 static int __cpuinit process_zones(int cpu)
2468 struct zone *zone, *dzone;
2469 int node = cpu_to_node(cpu);
2471 node_set_state(node, N_CPU); /* this node has a cpu */
2473 for_each_zone(zone) {
2475 if (!populated_zone(zone))
2478 zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset),
2480 if (!zone_pcp(zone, cpu))
2483 setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone));
2485 if (percpu_pagelist_fraction)
2486 setup_pagelist_highmark(zone_pcp(zone, cpu),
2487 (zone->present_pages / percpu_pagelist_fraction));
2492 for_each_zone(dzone) {
2493 if (!populated_zone(dzone))
2497 kfree(zone_pcp(dzone, cpu));
2498 zone_pcp(dzone, cpu) = NULL;
2503 static inline void free_zone_pagesets(int cpu)
2507 for_each_zone(zone) {
2508 struct per_cpu_pageset *pset = zone_pcp(zone, cpu);
2510 /* Free per_cpu_pageset if it is slab allocated */
2511 if (pset != &boot_pageset[cpu])
2513 zone_pcp(zone, cpu) = NULL;
2517 static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb,
2518 unsigned long action,
2521 int cpu = (long)hcpu;
2522 int ret = NOTIFY_OK;
2525 case CPU_UP_PREPARE:
2526 case CPU_UP_PREPARE_FROZEN:
2527 if (process_zones(cpu))
2530 case CPU_UP_CANCELED:
2531 case CPU_UP_CANCELED_FROZEN:
2533 case CPU_DEAD_FROZEN:
2534 free_zone_pagesets(cpu);
2542 static struct notifier_block __cpuinitdata pageset_notifier =
2543 { &pageset_cpuup_callback, NULL, 0 };
2545 void __init setup_per_cpu_pageset(void)
2549 /* Initialize per_cpu_pageset for cpu 0.
2550 * A cpuup callback will do this for every cpu
2551 * as it comes online
2553 err = process_zones(smp_processor_id());
2555 register_cpu_notifier(&pageset_notifier);
2560 static noinline __init_refok
2561 int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
2564 struct pglist_data *pgdat = zone->zone_pgdat;
2568 * The per-page waitqueue mechanism uses hashed waitqueues
2571 zone->wait_table_hash_nr_entries =
2572 wait_table_hash_nr_entries(zone_size_pages);
2573 zone->wait_table_bits =
2574 wait_table_bits(zone->wait_table_hash_nr_entries);
2575 alloc_size = zone->wait_table_hash_nr_entries
2576 * sizeof(wait_queue_head_t);
2578 if (system_state == SYSTEM_BOOTING) {
2579 zone->wait_table = (wait_queue_head_t *)
2580 alloc_bootmem_node(pgdat, alloc_size);
2583 * This case means that a zone whose size was 0 gets new memory
2584 * via memory hot-add.
2585 * But it may be the case that a new node was hot-added. In
2586 * this case vmalloc() will not be able to use this new node's
2587 * memory - this wait_table must be initialized to use this new
2588 * node itself as well.
2589 * To use this new node's memory, further consideration will be
2592 zone->wait_table = vmalloc(alloc_size);
2594 if (!zone->wait_table)
2597 for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
2598 init_waitqueue_head(zone->wait_table + i);
2603 static __meminit void zone_pcp_init(struct zone *zone)
2606 unsigned long batch = zone_batchsize(zone);
2608 for (cpu = 0; cpu < NR_CPUS; cpu++) {
2610 /* Early boot. Slab allocator not functional yet */
2611 zone_pcp(zone, cpu) = &boot_pageset[cpu];
2612 setup_pageset(&boot_pageset[cpu],0);
2614 setup_pageset(zone_pcp(zone,cpu), batch);
2617 if (zone->present_pages)
2618 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n",
2619 zone->name, zone->present_pages, batch);
2622 __meminit int init_currently_empty_zone(struct zone *zone,
2623 unsigned long zone_start_pfn,
2625 enum memmap_context context)
2627 struct pglist_data *pgdat = zone->zone_pgdat;
2629 ret = zone_wait_table_init(zone, size);
2632 pgdat->nr_zones = zone_idx(zone) + 1;
2634 zone->zone_start_pfn = zone_start_pfn;
2636 memmap_init(size, pgdat->node_id, zone_idx(zone), zone_start_pfn);
2638 zone_init_free_lists(pgdat, zone, zone->spanned_pages);
2643 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2645 * Basic iterator support. Return the first range of PFNs for a node
2646 * Note: nid == MAX_NUMNODES returns first region regardless of node
2648 static int __meminit first_active_region_index_in_nid(int nid)
2652 for (i = 0; i < nr_nodemap_entries; i++)
2653 if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
2660 * Basic iterator support. Return the next active range of PFNs for a node
2661 * Note: nid == MAX_NUMNODES returns next region regardles of node
2663 static int __meminit next_active_region_index_in_nid(int index, int nid)
2665 for (index = index + 1; index < nr_nodemap_entries; index++)
2666 if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
2672 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
2674 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
2675 * Architectures may implement their own version but if add_active_range()
2676 * was used and there are no special requirements, this is a convenient
2679 int __meminit early_pfn_to_nid(unsigned long pfn)
2683 for (i = 0; i < nr_nodemap_entries; i++) {
2684 unsigned long start_pfn = early_node_map[i].start_pfn;
2685 unsigned long end_pfn = early_node_map[i].end_pfn;
2687 if (start_pfn <= pfn && pfn < end_pfn)
2688 return early_node_map[i].nid;
2693 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
2695 /* Basic iterator support to walk early_node_map[] */
2696 #define for_each_active_range_index_in_nid(i, nid) \
2697 for (i = first_active_region_index_in_nid(nid); i != -1; \
2698 i = next_active_region_index_in_nid(i, nid))
2701 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
2702 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
2703 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
2705 * If an architecture guarantees that all ranges registered with
2706 * add_active_ranges() contain no holes and may be freed, this
2707 * this function may be used instead of calling free_bootmem() manually.
2709 void __init free_bootmem_with_active_regions(int nid,
2710 unsigned long max_low_pfn)
2714 for_each_active_range_index_in_nid(i, nid) {
2715 unsigned long size_pages = 0;
2716 unsigned long end_pfn = early_node_map[i].end_pfn;
2718 if (early_node_map[i].start_pfn >= max_low_pfn)
2721 if (end_pfn > max_low_pfn)
2722 end_pfn = max_low_pfn;
2724 size_pages = end_pfn - early_node_map[i].start_pfn;
2725 free_bootmem_node(NODE_DATA(early_node_map[i].nid),
2726 PFN_PHYS(early_node_map[i].start_pfn),
2727 size_pages << PAGE_SHIFT);
2732 * sparse_memory_present_with_active_regions - Call memory_present for each active range
2733 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
2735 * If an architecture guarantees that all ranges registered with
2736 * add_active_ranges() contain no holes and may be freed, this
2737 * function may be used instead of calling memory_present() manually.
2739 void __init sparse_memory_present_with_active_regions(int nid)
2743 for_each_active_range_index_in_nid(i, nid)
2744 memory_present(early_node_map[i].nid,
2745 early_node_map[i].start_pfn,
2746 early_node_map[i].end_pfn);
2750 * push_node_boundaries - Push node boundaries to at least the requested boundary
2751 * @nid: The nid of the node to push the boundary for
2752 * @start_pfn: The start pfn of the node
2753 * @end_pfn: The end pfn of the node
2755 * In reserve-based hot-add, mem_map is allocated that is unused until hotadd
2756 * time. Specifically, on x86_64, SRAT will report ranges that can potentially
2757 * be hotplugged even though no physical memory exists. This function allows
2758 * an arch to push out the node boundaries so mem_map is allocated that can
2761 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
2762 void __init push_node_boundaries(unsigned int nid,
2763 unsigned long start_pfn, unsigned long end_pfn)
2765 printk(KERN_DEBUG "Entering push_node_boundaries(%u, %lu, %lu)\n",
2766 nid, start_pfn, end_pfn);
2768 /* Initialise the boundary for this node if necessary */
2769 if (node_boundary_end_pfn[nid] == 0)
2770 node_boundary_start_pfn[nid] = -1UL;
2772 /* Update the boundaries */
2773 if (node_boundary_start_pfn[nid] > start_pfn)
2774 node_boundary_start_pfn[nid] = start_pfn;
2775 if (node_boundary_end_pfn[nid] < end_pfn)
2776 node_boundary_end_pfn[nid] = end_pfn;
2779 /* If necessary, push the node boundary out for reserve hotadd */
2780 static void __meminit account_node_boundary(unsigned int nid,
2781 unsigned long *start_pfn, unsigned long *end_pfn)
2783 printk(KERN_DEBUG "Entering account_node_boundary(%u, %lu, %lu)\n",
2784 nid, *start_pfn, *end_pfn);
2786 /* Return if boundary information has not been provided */
2787 if (node_boundary_end_pfn[nid] == 0)
2790 /* Check the boundaries and update if necessary */
2791 if (node_boundary_start_pfn[nid] < *start_pfn)
2792 *start_pfn = node_boundary_start_pfn[nid];
2793 if (node_boundary_end_pfn[nid] > *end_pfn)
2794 *end_pfn = node_boundary_end_pfn[nid];
2797 void __init push_node_boundaries(unsigned int nid,
2798 unsigned long start_pfn, unsigned long end_pfn) {}
2800 static void __meminit account_node_boundary(unsigned int nid,
2801 unsigned long *start_pfn, unsigned long *end_pfn) {}
2806 * get_pfn_range_for_nid - Return the start and end page frames for a node
2807 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
2808 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
2809 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
2811 * It returns the start and end page frame of a node based on information
2812 * provided by an arch calling add_active_range(). If called for a node
2813 * with no available memory, a warning is printed and the start and end
2816 void __meminit get_pfn_range_for_nid(unsigned int nid,
2817 unsigned long *start_pfn, unsigned long *end_pfn)
2823 for_each_active_range_index_in_nid(i, nid) {
2824 *start_pfn = min(*start_pfn, early_node_map[i].start_pfn);
2825 *end_pfn = max(*end_pfn, early_node_map[i].end_pfn);
2828 if (*start_pfn == -1UL)
2831 /* Push the node boundaries out if requested */
2832 account_node_boundary(nid, start_pfn, end_pfn);
2836 * This finds a zone that can be used for ZONE_MOVABLE pages. The
2837 * assumption is made that zones within a node are ordered in monotonic
2838 * increasing memory addresses so that the "highest" populated zone is used
2840 void __init find_usable_zone_for_movable(void)
2843 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
2844 if (zone_index == ZONE_MOVABLE)
2847 if (arch_zone_highest_possible_pfn[zone_index] >
2848 arch_zone_lowest_possible_pfn[zone_index])
2852 VM_BUG_ON(zone_index == -1);
2853 movable_zone = zone_index;
2857 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
2858 * because it is sized independant of architecture. Unlike the other zones,
2859 * the starting point for ZONE_MOVABLE is not fixed. It may be different
2860 * in each node depending on the size of each node and how evenly kernelcore
2861 * is distributed. This helper function adjusts the zone ranges
2862 * provided by the architecture for a given node by using the end of the
2863 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
2864 * zones within a node are in order of monotonic increases memory addresses
2866 void __meminit adjust_zone_range_for_zone_movable(int nid,
2867 unsigned long zone_type,
2868 unsigned long node_start_pfn,
2869 unsigned long node_end_pfn,
2870 unsigned long *zone_start_pfn,
2871 unsigned long *zone_end_pfn)
2873 /* Only adjust if ZONE_MOVABLE is on this node */
2874 if (zone_movable_pfn[nid]) {
2875 /* Size ZONE_MOVABLE */
2876 if (zone_type == ZONE_MOVABLE) {
2877 *zone_start_pfn = zone_movable_pfn[nid];
2878 *zone_end_pfn = min(node_end_pfn,
2879 arch_zone_highest_possible_pfn[movable_zone]);
2881 /* Adjust for ZONE_MOVABLE starting within this range */
2882 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
2883 *zone_end_pfn > zone_movable_pfn[nid]) {
2884 *zone_end_pfn = zone_movable_pfn[nid];
2886 /* Check if this whole range is within ZONE_MOVABLE */
2887 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
2888 *zone_start_pfn = *zone_end_pfn;
2893 * Return the number of pages a zone spans in a node, including holes
2894 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
2896 static unsigned long __meminit zone_spanned_pages_in_node(int nid,
2897 unsigned long zone_type,
2898 unsigned long *ignored)
2900 unsigned long node_start_pfn, node_end_pfn;
2901 unsigned long zone_start_pfn, zone_end_pfn;
2903 /* Get the start and end of the node and zone */
2904 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
2905 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
2906 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
2907 adjust_zone_range_for_zone_movable(nid, zone_type,
2908 node_start_pfn, node_end_pfn,
2909 &zone_start_pfn, &zone_end_pfn);
2911 /* Check that this node has pages within the zone's required range */
2912 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
2915 /* Move the zone boundaries inside the node if necessary */
2916 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
2917 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
2919 /* Return the spanned pages */
2920 return zone_end_pfn - zone_start_pfn;
2924 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
2925 * then all holes in the requested range will be accounted for.
2927 unsigned long __meminit __absent_pages_in_range(int nid,
2928 unsigned long range_start_pfn,
2929 unsigned long range_end_pfn)
2932 unsigned long prev_end_pfn = 0, hole_pages = 0;
2933 unsigned long start_pfn;
2935 /* Find the end_pfn of the first active range of pfns in the node */
2936 i = first_active_region_index_in_nid(nid);
2940 prev_end_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
2942 /* Account for ranges before physical memory on this node */
2943 if (early_node_map[i].start_pfn > range_start_pfn)
2944 hole_pages = prev_end_pfn - range_start_pfn;
2946 /* Find all holes for the zone within the node */
2947 for (; i != -1; i = next_active_region_index_in_nid(i, nid)) {
2949 /* No need to continue if prev_end_pfn is outside the zone */
2950 if (prev_end_pfn >= range_end_pfn)
2953 /* Make sure the end of the zone is not within the hole */
2954 start_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
2955 prev_end_pfn = max(prev_end_pfn, range_start_pfn);
2957 /* Update the hole size cound and move on */
2958 if (start_pfn > range_start_pfn) {
2959 BUG_ON(prev_end_pfn > start_pfn);
2960 hole_pages += start_pfn - prev_end_pfn;
2962 prev_end_pfn = early_node_map[i].end_pfn;
2965 /* Account for ranges past physical memory on this node */
2966 if (range_end_pfn > prev_end_pfn)
2967 hole_pages += range_end_pfn -
2968 max(range_start_pfn, prev_end_pfn);
2974 * absent_pages_in_range - Return number of page frames in holes within a range
2975 * @start_pfn: The start PFN to start searching for holes
2976 * @end_pfn: The end PFN to stop searching for holes
2978 * It returns the number of pages frames in memory holes within a range.
2980 unsigned long __init absent_pages_in_range(unsigned long start_pfn,
2981 unsigned long end_pfn)
2983 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
2986 /* Return the number of page frames in holes in a zone on a node */
2987 static unsigned long __meminit zone_absent_pages_in_node(int nid,
2988 unsigned long zone_type,
2989 unsigned long *ignored)
2991 unsigned long node_start_pfn, node_end_pfn;
2992 unsigned long zone_start_pfn, zone_end_pfn;
2994 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
2995 zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type],
2997 zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type],
3000 adjust_zone_range_for_zone_movable(nid, zone_type,
3001 node_start_pfn, node_end_pfn,
3002 &zone_start_pfn, &zone_end_pfn);
3003 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
3007 static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
3008 unsigned long zone_type,
3009 unsigned long *zones_size)
3011 return zones_size[zone_type];
3014 static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
3015 unsigned long zone_type,
3016 unsigned long *zholes_size)
3021 return zholes_size[zone_type];
3026 static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
3027 unsigned long *zones_size, unsigned long *zholes_size)
3029 unsigned long realtotalpages, totalpages = 0;
3032 for (i = 0; i < MAX_NR_ZONES; i++)
3033 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
3035 pgdat->node_spanned_pages = totalpages;
3037 realtotalpages = totalpages;
3038 for (i = 0; i < MAX_NR_ZONES; i++)
3040 zone_absent_pages_in_node(pgdat->node_id, i,
3042 pgdat->node_present_pages = realtotalpages;
3043 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
3047 #ifndef CONFIG_SPARSEMEM
3049 * Calculate the size of the zone->blockflags rounded to an unsigned long
3050 * Start by making sure zonesize is a multiple of MAX_ORDER-1 by rounding up
3051 * Then figure 1 NR_PAGEBLOCK_BITS worth of bits per MAX_ORDER-1, finally
3052 * round what is now in bits to nearest long in bits, then return it in
3055 static unsigned long __init usemap_size(unsigned long zonesize)
3057 unsigned long usemapsize;
3059 usemapsize = roundup(zonesize, MAX_ORDER_NR_PAGES);
3060 usemapsize = usemapsize >> (MAX_ORDER-1);
3061 usemapsize *= NR_PAGEBLOCK_BITS;
3062 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
3064 return usemapsize / 8;
3067 static void __init setup_usemap(struct pglist_data *pgdat,
3068 struct zone *zone, unsigned long zonesize)
3070 unsigned long usemapsize = usemap_size(zonesize);
3071 zone->pageblock_flags = NULL;
3073 zone->pageblock_flags = alloc_bootmem_node(pgdat, usemapsize);
3074 memset(zone->pageblock_flags, 0, usemapsize);
3078 static void inline setup_usemap(struct pglist_data *pgdat,
3079 struct zone *zone, unsigned long zonesize) {}
3080 #endif /* CONFIG_SPARSEMEM */
3083 * Set up the zone data structures:
3084 * - mark all pages reserved
3085 * - mark all memory queues empty
3086 * - clear the memory bitmaps
3088 static void __meminit free_area_init_core(struct pglist_data *pgdat,
3089 unsigned long *zones_size, unsigned long *zholes_size)
3092 int nid = pgdat->node_id;
3093 unsigned long zone_start_pfn = pgdat->node_start_pfn;
3096 pgdat_resize_init(pgdat);
3097 pgdat->nr_zones = 0;
3098 init_waitqueue_head(&pgdat->kswapd_wait);
3099 pgdat->kswapd_max_order = 0;
3101 for (j = 0; j < MAX_NR_ZONES; j++) {
3102 struct zone *zone = pgdat->node_zones + j;
3103 unsigned long size, realsize, memmap_pages;
3105 size = zone_spanned_pages_in_node(nid, j, zones_size);
3106 realsize = size - zone_absent_pages_in_node(nid, j,
3110 * Adjust realsize so that it accounts for how much memory
3111 * is used by this zone for memmap. This affects the watermark
3112 * and per-cpu initialisations
3114 memmap_pages = (size * sizeof(struct page)) >> PAGE_SHIFT;
3115 if (realsize >= memmap_pages) {
3116 realsize -= memmap_pages;
3118 " %s zone: %lu pages used for memmap\n",
3119 zone_names[j], memmap_pages);
3122 " %s zone: %lu pages exceeds realsize %lu\n",
3123 zone_names[j], memmap_pages, realsize);
3125 /* Account for reserved pages */
3126 if (j == 0 && realsize > dma_reserve) {
3127 realsize -= dma_reserve;
3128 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
3129 zone_names[0], dma_reserve);
3132 if (!is_highmem_idx(j))
3133 nr_kernel_pages += realsize;
3134 nr_all_pages += realsize;
3136 zone->spanned_pages = size;
3137 zone->present_pages = realsize;
3140 zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
3142 zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100;
3144 zone->name = zone_names[j];
3145 spin_lock_init(&zone->lock);
3146 spin_lock_init(&zone->lru_lock);
3147 zone_seqlock_init(zone);
3148 zone->zone_pgdat = pgdat;
3150 zone->prev_priority = DEF_PRIORITY;
3152 zone_pcp_init(zone);
3153 INIT_LIST_HEAD(&zone->active_list);
3154 INIT_LIST_HEAD(&zone->inactive_list);
3155 zone->nr_scan_active = 0;
3156 zone->nr_scan_inactive = 0;
3157 zap_zone_vm_stats(zone);
3158 atomic_set(&zone->reclaim_in_progress, 0);
3162 setup_usemap(pgdat, zone, size);
3163 ret = init_currently_empty_zone(zone, zone_start_pfn,
3164 size, MEMMAP_EARLY);
3166 zone_start_pfn += size;
3170 static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
3172 /* Skip empty nodes */
3173 if (!pgdat->node_spanned_pages)
3176 #ifdef CONFIG_FLAT_NODE_MEM_MAP
3177 /* ia64 gets its own node_mem_map, before this, without bootmem */
3178 if (!pgdat->node_mem_map) {
3179 unsigned long size, start, end;
3183 * The zone's endpoints aren't required to be MAX_ORDER
3184 * aligned but the node_mem_map endpoints must be in order
3185 * for the buddy allocator to function correctly.
3187 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
3188 end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
3189 end = ALIGN(end, MAX_ORDER_NR_PAGES);
3190 size = (end - start) * sizeof(struct page);
3191 map = alloc_remap(pgdat->node_id, size);
3193 map = alloc_bootmem_node(pgdat, size);
3194 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
3196 #ifndef CONFIG_NEED_MULTIPLE_NODES
3198 * With no DISCONTIG, the global mem_map is just set as node 0's
3200 if (pgdat == NODE_DATA(0)) {
3201 mem_map = NODE_DATA(0)->node_mem_map;
3202 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3203 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
3204 mem_map -= pgdat->node_start_pfn;
3205 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3208 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
3211 void __meminit free_area_init_node(int nid, struct pglist_data *pgdat,
3212 unsigned long *zones_size, unsigned long node_start_pfn,
3213 unsigned long *zholes_size)
3215 pgdat->node_id = nid;
3216 pgdat->node_start_pfn = node_start_pfn;
3217 calculate_node_totalpages(pgdat, zones_size, zholes_size);
3219 alloc_node_mem_map(pgdat);
3221 free_area_init_core(pgdat, zones_size, zholes_size);
3224 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3226 #if MAX_NUMNODES > 1
3228 * Figure out the number of possible node ids.
3230 static void __init setup_nr_node_ids(void)
3233 unsigned int highest = 0;
3235 for_each_node_mask(node, node_possible_map)
3237 nr_node_ids = highest + 1;
3240 static inline void setup_nr_node_ids(void)
3246 * add_active_range - Register a range of PFNs backed by physical memory
3247 * @nid: The node ID the range resides on
3248 * @start_pfn: The start PFN of the available physical memory
3249 * @end_pfn: The end PFN of the available physical memory
3251 * These ranges are stored in an early_node_map[] and later used by
3252 * free_area_init_nodes() to calculate zone sizes and holes. If the
3253 * range spans a memory hole, it is up to the architecture to ensure
3254 * the memory is not freed by the bootmem allocator. If possible
3255 * the range being registered will be merged with existing ranges.
3257 void __init add_active_range(unsigned int nid, unsigned long start_pfn,
3258 unsigned long end_pfn)
3262 printk(KERN_DEBUG "Entering add_active_range(%d, %lu, %lu) "
3263 "%d entries of %d used\n",
3264 nid, start_pfn, end_pfn,
3265 nr_nodemap_entries, MAX_ACTIVE_REGIONS);
3267 /* Merge with existing active regions if possible */
3268 for (i = 0; i < nr_nodemap_entries; i++) {
3269 if (early_node_map[i].nid != nid)
3272 /* Skip if an existing region covers this new one */
3273 if (start_pfn >= early_node_map[i].start_pfn &&
3274 end_pfn <= early_node_map[i].end_pfn)
3277 /* Merge forward if suitable */
3278 if (start_pfn <= early_node_map[i].end_pfn &&
3279 end_pfn > early_node_map[i].end_pfn) {
3280 early_node_map[i].end_pfn = end_pfn;
3284 /* Merge backward if suitable */
3285 if (start_pfn < early_node_map[i].end_pfn &&
3286 end_pfn >= early_node_map[i].start_pfn) {
3287 early_node_map[i].start_pfn = start_pfn;
3292 /* Check that early_node_map is large enough */
3293 if (i >= MAX_ACTIVE_REGIONS) {
3294 printk(KERN_CRIT "More than %d memory regions, truncating\n",
3295 MAX_ACTIVE_REGIONS);
3299 early_node_map[i].nid = nid;
3300 early_node_map[i].start_pfn = start_pfn;
3301 early_node_map[i].end_pfn = end_pfn;
3302 nr_nodemap_entries = i + 1;
3306 * shrink_active_range - Shrink an existing registered range of PFNs
3307 * @nid: The node id the range is on that should be shrunk
3308 * @old_end_pfn: The old end PFN of the range
3309 * @new_end_pfn: The new PFN of the range
3311 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
3312 * The map is kept at the end physical page range that has already been
3313 * registered with add_active_range(). This function allows an arch to shrink
3314 * an existing registered range.
3316 void __init shrink_active_range(unsigned int nid, unsigned long old_end_pfn,
3317 unsigned long new_end_pfn)
3321 /* Find the old active region end and shrink */
3322 for_each_active_range_index_in_nid(i, nid)
3323 if (early_node_map[i].end_pfn == old_end_pfn) {
3324 early_node_map[i].end_pfn = new_end_pfn;
3330 * remove_all_active_ranges - Remove all currently registered regions
3332 * During discovery, it may be found that a table like SRAT is invalid
3333 * and an alternative discovery method must be used. This function removes
3334 * all currently registered regions.
3336 void __init remove_all_active_ranges(void)
3338 memset(early_node_map, 0, sizeof(early_node_map));
3339 nr_nodemap_entries = 0;
3340 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
3341 memset(node_boundary_start_pfn, 0, sizeof(node_boundary_start_pfn));
3342 memset(node_boundary_end_pfn, 0, sizeof(node_boundary_end_pfn));
3343 #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
3346 /* Compare two active node_active_regions */
3347 static int __init cmp_node_active_region(const void *a, const void *b)
3349 struct node_active_region *arange = (struct node_active_region *)a;
3350 struct node_active_region *brange = (struct node_active_region *)b;
3352 /* Done this way to avoid overflows */
3353 if (arange->start_pfn > brange->start_pfn)
3355 if (arange->start_pfn < brange->start_pfn)
3361 /* sort the node_map by start_pfn */
3362 static void __init sort_node_map(void)
3364 sort(early_node_map, (size_t)nr_nodemap_entries,
3365 sizeof(struct node_active_region),
3366 cmp_node_active_region, NULL);
3369 /* Find the lowest pfn for a node */
3370 unsigned long __init find_min_pfn_for_node(unsigned long nid)
3373 unsigned long min_pfn = ULONG_MAX;
3375 /* Assuming a sorted map, the first range found has the starting pfn */
3376 for_each_active_range_index_in_nid(i, nid)
3377 min_pfn = min(min_pfn, early_node_map[i].start_pfn);
3379 if (min_pfn == ULONG_MAX) {
3381 "Could not find start_pfn for node %lu\n", nid);
3389 * find_min_pfn_with_active_regions - Find the minimum PFN registered
3391 * It returns the minimum PFN based on information provided via
3392 * add_active_range().
3394 unsigned long __init find_min_pfn_with_active_regions(void)
3396 return find_min_pfn_for_node(MAX_NUMNODES);
3400 * find_max_pfn_with_active_regions - Find the maximum PFN registered
3402 * It returns the maximum PFN based on information provided via
3403 * add_active_range().
3405 unsigned long __init find_max_pfn_with_active_regions(void)
3408 unsigned long max_pfn = 0;
3410 for (i = 0; i < nr_nodemap_entries; i++)
3411 max_pfn = max(max_pfn, early_node_map[i].end_pfn);
3417 * early_calculate_totalpages()
3418 * Sum pages in active regions for movable zone.
3419 * Populate N_HIGH_MEMORY for calculating usable_nodes.
3421 unsigned long __init early_calculate_totalpages(void)
3424 unsigned long totalpages = 0;
3426 for (i = 0; i < nr_nodemap_entries; i++) {
3427 unsigned long pages = early_node_map[i].end_pfn -
3428 early_node_map[i].start_pfn;
3429 totalpages += pages;
3431 node_set_state(early_node_map[i].nid, N_HIGH_MEMORY);
3437 * Find the PFN the Movable zone begins in each node. Kernel memory
3438 * is spread evenly between nodes as long as the nodes have enough
3439 * memory. When they don't, some nodes will have more kernelcore than
3442 void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn)
3445 unsigned long usable_startpfn;
3446 unsigned long kernelcore_node, kernelcore_remaining;
3447 unsigned long totalpages = early_calculate_totalpages();
3448 int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]);
3451 * If movablecore was specified, calculate what size of
3452 * kernelcore that corresponds so that memory usable for
3453 * any allocation type is evenly spread. If both kernelcore
3454 * and movablecore are specified, then the value of kernelcore
3455 * will be used for required_kernelcore if it's greater than
3456 * what movablecore would have allowed.
3458 if (required_movablecore) {
3459 unsigned long corepages;
3462 * Round-up so that ZONE_MOVABLE is at least as large as what
3463 * was requested by the user
3465 required_movablecore =
3466 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
3467 corepages = totalpages - required_movablecore;
3469 required_kernelcore = max(required_kernelcore, corepages);
3472 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
3473 if (!required_kernelcore)
3476 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
3477 find_usable_zone_for_movable();
3478 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
3481 /* Spread kernelcore memory as evenly as possible throughout nodes */
3482 kernelcore_node = required_kernelcore / usable_nodes;
3483 for_each_node_state(nid, N_HIGH_MEMORY) {
3485 * Recalculate kernelcore_node if the division per node
3486 * now exceeds what is necessary to satisfy the requested
3487 * amount of memory for the kernel
3489 if (required_kernelcore < kernelcore_node)
3490 kernelcore_node = required_kernelcore / usable_nodes;
3493 * As the map is walked, we track how much memory is usable
3494 * by the kernel using kernelcore_remaining. When it is
3495 * 0, the rest of the node is usable by ZONE_MOVABLE
3497 kernelcore_remaining = kernelcore_node;
3499 /* Go through each range of PFNs within this node */
3500 for_each_active_range_index_in_nid(i, nid) {
3501 unsigned long start_pfn, end_pfn;
3502 unsigned long size_pages;
3504 start_pfn = max(early_node_map[i].start_pfn,
3505 zone_movable_pfn[nid]);
3506 end_pfn = early_node_map[i].end_pfn;
3507 if (start_pfn >= end_pfn)
3510 /* Account for what is only usable for kernelcore */
3511 if (start_pfn < usable_startpfn) {
3512 unsigned long kernel_pages;
3513 kernel_pages = min(end_pfn, usable_startpfn)
3516 kernelcore_remaining -= min(kernel_pages,
3517 kernelcore_remaining);
3518 required_kernelcore -= min(kernel_pages,
3519 required_kernelcore);
3521 /* Continue if range is now fully accounted */
3522 if (end_pfn <= usable_startpfn) {
3525 * Push zone_movable_pfn to the end so
3526 * that if we have to rebalance
3527 * kernelcore across nodes, we will
3528 * not double account here
3530 zone_movable_pfn[nid] = end_pfn;
3533 start_pfn = usable_startpfn;
3537 * The usable PFN range for ZONE_MOVABLE is from
3538 * start_pfn->end_pfn. Calculate size_pages as the
3539 * number of pages used as kernelcore
3541 size_pages = end_pfn - start_pfn;
3542 if (size_pages > kernelcore_remaining)
3543 size_pages = kernelcore_remaining;
3544 zone_movable_pfn[nid] = start_pfn + size_pages;
3547 * Some kernelcore has been met, update counts and
3548 * break if the kernelcore for this node has been
3551 required_kernelcore -= min(required_kernelcore,
3553 kernelcore_remaining -= size_pages;
3554 if (!kernelcore_remaining)
3560 * If there is still required_kernelcore, we do another pass with one
3561 * less node in the count. This will push zone_movable_pfn[nid] further
3562 * along on the nodes that still have memory until kernelcore is
3566 if (usable_nodes && required_kernelcore > usable_nodes)
3569 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
3570 for (nid = 0; nid < MAX_NUMNODES; nid++)
3571 zone_movable_pfn[nid] =
3572 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
3575 /* Any regular memory on that node ? */
3576 static void check_for_regular_memory(pg_data_t *pgdat)
3578 #ifdef CONFIG_HIGHMEM
3579 enum zone_type zone_type;
3581 for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) {
3582 struct zone *zone = &pgdat->node_zones[zone_type];
3583 if (zone->present_pages)
3584 node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY);
3590 * free_area_init_nodes - Initialise all pg_data_t and zone data
3591 * @max_zone_pfn: an array of max PFNs for each zone
3593 * This will call free_area_init_node() for each active node in the system.
3594 * Using the page ranges provided by add_active_range(), the size of each
3595 * zone in each node and their holes is calculated. If the maximum PFN
3596 * between two adjacent zones match, it is assumed that the zone is empty.
3597 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
3598 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
3599 * starts where the previous one ended. For example, ZONE_DMA32 starts
3600 * at arch_max_dma_pfn.
3602 void __init free_area_init_nodes(unsigned long *max_zone_pfn)
3607 /* Sort early_node_map as initialisation assumes it is sorted */
3610 /* Record where the zone boundaries are */
3611 memset(arch_zone_lowest_possible_pfn, 0,
3612 sizeof(arch_zone_lowest_possible_pfn));
3613 memset(arch_zone_highest_possible_pfn, 0,
3614 sizeof(arch_zone_highest_possible_pfn));
3615 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
3616 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
3617 for (i = 1; i < MAX_NR_ZONES; i++) {
3618 if (i == ZONE_MOVABLE)
3620 arch_zone_lowest_possible_pfn[i] =
3621 arch_zone_highest_possible_pfn[i-1];
3622 arch_zone_highest_possible_pfn[i] =
3623 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
3625 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
3626 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
3628 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
3629 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
3630 find_zone_movable_pfns_for_nodes(zone_movable_pfn);
3632 /* Print out the zone ranges */
3633 printk("Zone PFN ranges:\n");
3634 for (i = 0; i < MAX_NR_ZONES; i++) {
3635 if (i == ZONE_MOVABLE)
3637 printk(" %-8s %8lu -> %8lu\n",
3639 arch_zone_lowest_possible_pfn[i],
3640 arch_zone_highest_possible_pfn[i]);
3643 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
3644 printk("Movable zone start PFN for each node\n");
3645 for (i = 0; i < MAX_NUMNODES; i++) {
3646 if (zone_movable_pfn[i])
3647 printk(" Node %d: %lu\n", i, zone_movable_pfn[i]);
3650 /* Print out the early_node_map[] */
3651 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries);
3652 for (i = 0; i < nr_nodemap_entries; i++)
3653 printk(" %3d: %8lu -> %8lu\n", early_node_map[i].nid,
3654 early_node_map[i].start_pfn,
3655 early_node_map[i].end_pfn);
3657 /* Initialise every node */
3658 setup_nr_node_ids();
3659 for_each_online_node(nid) {
3660 pg_data_t *pgdat = NODE_DATA(nid);
3661 free_area_init_node(nid, pgdat, NULL,
3662 find_min_pfn_for_node(nid), NULL);
3664 /* Any memory on that node */
3665 if (pgdat->node_present_pages)
3666 node_set_state(nid, N_HIGH_MEMORY);
3667 check_for_regular_memory(pgdat);
3671 static int __init cmdline_parse_core(char *p, unsigned long *core)
3673 unsigned long long coremem;
3677 coremem = memparse(p, &p);
3678 *core = coremem >> PAGE_SHIFT;
3680 /* Paranoid check that UL is enough for the coremem value */
3681 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
3687 * kernelcore=size sets the amount of memory for use for allocations that
3688 * cannot be reclaimed or migrated.
3690 static int __init cmdline_parse_kernelcore(char *p)
3692 return cmdline_parse_core(p, &required_kernelcore);
3696 * movablecore=size sets the amount of memory for use for allocations that
3697 * can be reclaimed or migrated.
3699 static int __init cmdline_parse_movablecore(char *p)
3701 return cmdline_parse_core(p, &required_movablecore);
3704 early_param("kernelcore", cmdline_parse_kernelcore);
3705 early_param("movablecore", cmdline_parse_movablecore);
3707 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3710 * set_dma_reserve - set the specified number of pages reserved in the first zone
3711 * @new_dma_reserve: The number of pages to mark reserved
3713 * The per-cpu batchsize and zone watermarks are determined by present_pages.
3714 * In the DMA zone, a significant percentage may be consumed by kernel image
3715 * and other unfreeable allocations which can skew the watermarks badly. This
3716 * function may optionally be used to account for unfreeable pages in the
3717 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
3718 * smaller per-cpu batchsize.
3720 void __init set_dma_reserve(unsigned long new_dma_reserve)
3722 dma_reserve = new_dma_reserve;
3725 #ifndef CONFIG_NEED_MULTIPLE_NODES
3726 static bootmem_data_t contig_bootmem_data;
3727 struct pglist_data contig_page_data = { .bdata = &contig_bootmem_data };
3729 EXPORT_SYMBOL(contig_page_data);
3732 void __init free_area_init(unsigned long *zones_size)
3734 free_area_init_node(0, NODE_DATA(0), zones_size,
3735 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
3738 static int page_alloc_cpu_notify(struct notifier_block *self,
3739 unsigned long action, void *hcpu)
3741 int cpu = (unsigned long)hcpu;
3743 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
3744 local_irq_disable();
3746 vm_events_fold_cpu(cpu);
3748 refresh_cpu_vm_stats(cpu);
3753 void __init page_alloc_init(void)
3755 hotcpu_notifier(page_alloc_cpu_notify, 0);
3759 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
3760 * or min_free_kbytes changes.
3762 static void calculate_totalreserve_pages(void)
3764 struct pglist_data *pgdat;
3765 unsigned long reserve_pages = 0;
3766 enum zone_type i, j;
3768 for_each_online_pgdat(pgdat) {
3769 for (i = 0; i < MAX_NR_ZONES; i++) {
3770 struct zone *zone = pgdat->node_zones + i;
3771 unsigned long max = 0;
3773 /* Find valid and maximum lowmem_reserve in the zone */
3774 for (j = i; j < MAX_NR_ZONES; j++) {
3775 if (zone->lowmem_reserve[j] > max)
3776 max = zone->lowmem_reserve[j];
3779 /* we treat pages_high as reserved pages. */
3780 max += zone->pages_high;
3782 if (max > zone->present_pages)
3783 max = zone->present_pages;
3784 reserve_pages += max;
3787 totalreserve_pages = reserve_pages;
3791 * setup_per_zone_lowmem_reserve - called whenever
3792 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
3793 * has a correct pages reserved value, so an adequate number of
3794 * pages are left in the zone after a successful __alloc_pages().
3796 static void setup_per_zone_lowmem_reserve(void)
3798 struct pglist_data *pgdat;
3799 enum zone_type j, idx;
3801 for_each_online_pgdat(pgdat) {
3802 for (j = 0; j < MAX_NR_ZONES; j++) {
3803 struct zone *zone = pgdat->node_zones + j;
3804 unsigned long present_pages = zone->present_pages;
3806 zone->lowmem_reserve[j] = 0;
3810 struct zone *lower_zone;
3814 if (sysctl_lowmem_reserve_ratio[idx] < 1)
3815 sysctl_lowmem_reserve_ratio[idx] = 1;
3817 lower_zone = pgdat->node_zones + idx;
3818 lower_zone->lowmem_reserve[j] = present_pages /
3819 sysctl_lowmem_reserve_ratio[idx];
3820 present_pages += lower_zone->present_pages;
3825 /* update totalreserve_pages */
3826 calculate_totalreserve_pages();
3830 * setup_per_zone_pages_min - called when min_free_kbytes changes.
3832 * Ensures that the pages_{min,low,high} values for each zone are set correctly
3833 * with respect to min_free_kbytes.
3835 void setup_per_zone_pages_min(void)
3837 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
3838 unsigned long lowmem_pages = 0;
3840 unsigned long flags;
3842 /* Calculate total number of !ZONE_HIGHMEM pages */
3843 for_each_zone(zone) {
3844 if (!is_highmem(zone))
3845 lowmem_pages += zone->present_pages;
3848 for_each_zone(zone) {
3851 spin_lock_irqsave(&zone->lru_lock, flags);
3852 tmp = (u64)pages_min * zone->present_pages;
3853 do_div(tmp, lowmem_pages);
3854 if (is_highmem(zone)) {
3856 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
3857 * need highmem pages, so cap pages_min to a small
3860 * The (pages_high-pages_low) and (pages_low-pages_min)
3861 * deltas controls asynch page reclaim, and so should
3862 * not be capped for highmem.
3866 min_pages = zone->present_pages / 1024;
3867 if (min_pages < SWAP_CLUSTER_MAX)
3868 min_pages = SWAP_CLUSTER_MAX;
3869 if (min_pages > 128)
3871 zone->pages_min = min_pages;
3874 * If it's a lowmem zone, reserve a number of pages
3875 * proportionate to the zone's size.
3877 zone->pages_min = tmp;
3880 zone->pages_low = zone->pages_min + (tmp >> 2);
3881 zone->pages_high = zone->pages_min + (tmp >> 1);
3882 spin_unlock_irqrestore(&zone->lru_lock, flags);
3885 /* update totalreserve_pages */
3886 calculate_totalreserve_pages();
3890 * Initialise min_free_kbytes.
3892 * For small machines we want it small (128k min). For large machines
3893 * we want it large (64MB max). But it is not linear, because network
3894 * bandwidth does not increase linearly with machine size. We use
3896 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
3897 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
3913 static int __init init_per_zone_pages_min(void)
3915 unsigned long lowmem_kbytes;
3917 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
3919 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
3920 if (min_free_kbytes < 128)
3921 min_free_kbytes = 128;
3922 if (min_free_kbytes > 65536)
3923 min_free_kbytes = 65536;
3924 setup_per_zone_pages_min();
3925 setup_per_zone_lowmem_reserve();
3928 module_init(init_per_zone_pages_min)
3931 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
3932 * that we can call two helper functions whenever min_free_kbytes
3935 int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
3936 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
3938 proc_dointvec(table, write, file, buffer, length, ppos);
3940 setup_per_zone_pages_min();
3945 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
3946 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
3951 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
3956 zone->min_unmapped_pages = (zone->present_pages *
3957 sysctl_min_unmapped_ratio) / 100;
3961 int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
3962 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
3967 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
3972 zone->min_slab_pages = (zone->present_pages *
3973 sysctl_min_slab_ratio) / 100;
3979 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
3980 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
3981 * whenever sysctl_lowmem_reserve_ratio changes.
3983 * The reserve ratio obviously has absolutely no relation with the
3984 * pages_min watermarks. The lowmem reserve ratio can only make sense
3985 * if in function of the boot time zone sizes.
3987 int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
3988 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
3990 proc_dointvec_minmax(table, write, file, buffer, length, ppos);
3991 setup_per_zone_lowmem_reserve();
3996 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
3997 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
3998 * can have before it gets flushed back to buddy allocator.
4001 int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
4002 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4008 ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4009 if (!write || (ret == -EINVAL))
4011 for_each_zone(zone) {
4012 for_each_online_cpu(cpu) {
4014 high = zone->present_pages / percpu_pagelist_fraction;
4015 setup_pagelist_highmark(zone_pcp(zone, cpu), high);
4021 int hashdist = HASHDIST_DEFAULT;
4024 static int __init set_hashdist(char *str)
4028 hashdist = simple_strtoul(str, &str, 0);
4031 __setup("hashdist=", set_hashdist);
4035 * allocate a large system hash table from bootmem
4036 * - it is assumed that the hash table must contain an exact power-of-2
4037 * quantity of entries
4038 * - limit is the number of hash buckets, not the total allocation size
4040 void *__init alloc_large_system_hash(const char *tablename,
4041 unsigned long bucketsize,
4042 unsigned long numentries,
4045 unsigned int *_hash_shift,
4046 unsigned int *_hash_mask,
4047 unsigned long limit)
4049 unsigned long long max = limit;
4050 unsigned long log2qty, size;
4053 /* allow the kernel cmdline to have a say */
4055 /* round applicable memory size up to nearest megabyte */
4056 numentries = nr_kernel_pages;
4057 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
4058 numentries >>= 20 - PAGE_SHIFT;
4059 numentries <<= 20 - PAGE_SHIFT;
4061 /* limit to 1 bucket per 2^scale bytes of low memory */
4062 if (scale > PAGE_SHIFT)
4063 numentries >>= (scale - PAGE_SHIFT);
4065 numentries <<= (PAGE_SHIFT - scale);
4067 /* Make sure we've got at least a 0-order allocation.. */
4068 if (unlikely((numentries * bucketsize) < PAGE_SIZE))
4069 numentries = PAGE_SIZE / bucketsize;
4071 numentries = roundup_pow_of_two(numentries);
4073 /* limit allocation size to 1/16 total memory by default */
4075 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
4076 do_div(max, bucketsize);
4079 if (numentries > max)
4082 log2qty = ilog2(numentries);
4085 size = bucketsize << log2qty;
4086 if (flags & HASH_EARLY)
4087 table = alloc_bootmem(size);
4089 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
4091 unsigned long order;
4092 for (order = 0; ((1UL << order) << PAGE_SHIFT) < size; order++)
4094 table = (void*) __get_free_pages(GFP_ATOMIC, order);
4096 * If bucketsize is not a power-of-two, we may free
4097 * some pages at the end of hash table.
4100 unsigned long alloc_end = (unsigned long)table +
4101 (PAGE_SIZE << order);
4102 unsigned long used = (unsigned long)table +
4104 split_page(virt_to_page(table), order);
4105 while (used < alloc_end) {
4111 } while (!table && size > PAGE_SIZE && --log2qty);
4114 panic("Failed to allocate %s hash table\n", tablename);
4116 printk(KERN_INFO "%s hash table entries: %d (order: %d, %lu bytes)\n",
4119 ilog2(size) - PAGE_SHIFT,
4123 *_hash_shift = log2qty;
4125 *_hash_mask = (1 << log2qty) - 1;
4130 #ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE
4131 struct page *pfn_to_page(unsigned long pfn)
4133 return __pfn_to_page(pfn);
4135 unsigned long page_to_pfn(struct page *page)
4137 return __page_to_pfn(page);
4139 EXPORT_SYMBOL(pfn_to_page);
4140 EXPORT_SYMBOL(page_to_pfn);
4141 #endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */
4143 /* Return a pointer to the bitmap storing bits affecting a block of pages */
4144 static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
4147 #ifdef CONFIG_SPARSEMEM
4148 return __pfn_to_section(pfn)->pageblock_flags;
4150 return zone->pageblock_flags;
4151 #endif /* CONFIG_SPARSEMEM */
4154 static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
4156 #ifdef CONFIG_SPARSEMEM
4157 pfn &= (PAGES_PER_SECTION-1);
4158 return (pfn >> (MAX_ORDER-1)) * NR_PAGEBLOCK_BITS;
4160 pfn = pfn - zone->zone_start_pfn;
4161 return (pfn >> (MAX_ORDER-1)) * NR_PAGEBLOCK_BITS;
4162 #endif /* CONFIG_SPARSEMEM */
4166 * get_pageblock_flags_group - Return the requested group of flags for the MAX_ORDER_NR_PAGES block of pages
4167 * @page: The page within the block of interest
4168 * @start_bitidx: The first bit of interest to retrieve
4169 * @end_bitidx: The last bit of interest
4170 * returns pageblock_bits flags
4172 unsigned long get_pageblock_flags_group(struct page *page,
4173 int start_bitidx, int end_bitidx)
4176 unsigned long *bitmap;
4177 unsigned long pfn, bitidx;
4178 unsigned long flags = 0;
4179 unsigned long value = 1;
4181 zone = page_zone(page);
4182 pfn = page_to_pfn(page);
4183 bitmap = get_pageblock_bitmap(zone, pfn);
4184 bitidx = pfn_to_bitidx(zone, pfn);
4186 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
4187 if (test_bit(bitidx + start_bitidx, bitmap))
4194 * set_pageblock_flags_group - Set the requested group of flags for a MAX_ORDER_NR_PAGES block of pages
4195 * @page: The page within the block of interest
4196 * @start_bitidx: The first bit of interest
4197 * @end_bitidx: The last bit of interest
4198 * @flags: The flags to set
4200 void set_pageblock_flags_group(struct page *page, unsigned long flags,
4201 int start_bitidx, int end_bitidx)
4204 unsigned long *bitmap;
4205 unsigned long pfn, bitidx;
4206 unsigned long value = 1;
4208 zone = page_zone(page);
4209 pfn = page_to_pfn(page);
4210 bitmap = get_pageblock_bitmap(zone, pfn);
4211 bitidx = pfn_to_bitidx(zone, pfn);
4213 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
4215 __set_bit(bitidx + start_bitidx, bitmap);
4217 __clear_bit(bitidx + start_bitidx, bitmap);