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>
43 #include <asm/tlbflush.h>
44 #include <asm/div64.h>
48 * MCD - HACK: Find somewhere to initialize this EARLY, or make this
51 nodemask_t node_online_map __read_mostly = { { [0] = 1UL } };
52 EXPORT_SYMBOL(node_online_map);
53 nodemask_t node_possible_map __read_mostly = NODE_MASK_ALL;
54 EXPORT_SYMBOL(node_possible_map);
55 unsigned long totalram_pages __read_mostly;
56 unsigned long totalreserve_pages __read_mostly;
58 int percpu_pagelist_fraction;
60 static void __free_pages_ok(struct page *page, unsigned int order);
63 * results with 256, 32 in the lowmem_reserve sysctl:
64 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
65 * 1G machine -> (16M dma, 784M normal, 224M high)
66 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
67 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
68 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
70 * TBD: should special case ZONE_DMA32 machines here - in those we normally
71 * don't need any ZONE_NORMAL reservation
73 int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
75 #ifdef CONFIG_ZONE_DMA32
83 EXPORT_SYMBOL(totalram_pages);
86 * Used by page_zone() to look up the address of the struct zone whose
87 * id is encoded in the upper bits of page->flags
89 struct zone *zone_table[1 << ZONETABLE_SHIFT] __read_mostly;
90 EXPORT_SYMBOL(zone_table);
92 static char *zone_names[MAX_NR_ZONES] = {
94 #ifdef CONFIG_ZONE_DMA32
103 int min_free_kbytes = 1024;
105 unsigned long __meminitdata nr_kernel_pages;
106 unsigned long __meminitdata nr_all_pages;
108 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
110 * MAX_ACTIVE_REGIONS determines the maxmimum number of distinct
111 * ranges of memory (RAM) that may be registered with add_active_range().
112 * Ranges passed to add_active_range() will be merged if possible
113 * so the number of times add_active_range() can be called is
114 * related to the number of nodes and the number of holes
116 #ifdef CONFIG_MAX_ACTIVE_REGIONS
117 /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
118 #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
120 #if MAX_NUMNODES >= 32
121 /* If there can be many nodes, allow up to 50 holes per node */
122 #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
124 /* By default, allow up to 256 distinct regions */
125 #define MAX_ACTIVE_REGIONS 256
129 struct node_active_region __initdata early_node_map[MAX_ACTIVE_REGIONS];
130 int __initdata nr_nodemap_entries;
131 unsigned long __initdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
132 unsigned long __initdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
133 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
135 #ifdef CONFIG_DEBUG_VM
136 static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
140 unsigned long pfn = page_to_pfn(page);
143 seq = zone_span_seqbegin(zone);
144 if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
146 else if (pfn < zone->zone_start_pfn)
148 } while (zone_span_seqretry(zone, seq));
153 static int page_is_consistent(struct zone *zone, struct page *page)
155 #ifdef CONFIG_HOLES_IN_ZONE
156 if (!pfn_valid(page_to_pfn(page)))
159 if (zone != page_zone(page))
165 * Temporary debugging check for pages not lying within a given zone.
167 static int bad_range(struct zone *zone, struct page *page)
169 if (page_outside_zone_boundaries(zone, page))
171 if (!page_is_consistent(zone, page))
177 static inline int bad_range(struct zone *zone, struct page *page)
183 static void bad_page(struct page *page)
185 printk(KERN_EMERG "Bad page state in process '%s'\n"
186 KERN_EMERG "page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n"
187 KERN_EMERG "Trying to fix it up, but a reboot is needed\n"
188 KERN_EMERG "Backtrace:\n",
189 current->comm, page, (int)(2*sizeof(unsigned long)),
190 (unsigned long)page->flags, page->mapping,
191 page_mapcount(page), page_count(page));
193 page->flags &= ~(1 << PG_lru |
203 set_page_count(page, 0);
204 reset_page_mapcount(page);
205 page->mapping = NULL;
206 add_taint(TAINT_BAD_PAGE);
210 * Higher-order pages are called "compound pages". They are structured thusly:
212 * The first PAGE_SIZE page is called the "head page".
214 * The remaining PAGE_SIZE pages are called "tail pages".
216 * All pages have PG_compound set. All pages have their ->private pointing at
217 * the head page (even the head page has this).
219 * The first tail page's ->lru.next holds the address of the compound page's
220 * put_page() function. Its ->lru.prev holds the order of allocation.
221 * This usage means that zero-order pages may not be compound.
224 static void free_compound_page(struct page *page)
226 __free_pages_ok(page, (unsigned long)page[1].lru.prev);
229 static void prep_compound_page(struct page *page, unsigned long order)
232 int nr_pages = 1 << order;
234 page[1].lru.next = (void *)free_compound_page; /* set dtor */
235 page[1].lru.prev = (void *)order;
236 for (i = 0; i < nr_pages; i++) {
237 struct page *p = page + i;
239 __SetPageCompound(p);
240 set_page_private(p, (unsigned long)page);
244 static void destroy_compound_page(struct page *page, unsigned long order)
247 int nr_pages = 1 << order;
249 if (unlikely((unsigned long)page[1].lru.prev != order))
252 for (i = 0; i < nr_pages; i++) {
253 struct page *p = page + i;
255 if (unlikely(!PageCompound(p) |
256 (page_private(p) != (unsigned long)page)))
258 __ClearPageCompound(p);
262 static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
266 VM_BUG_ON((gfp_flags & (__GFP_WAIT | __GFP_HIGHMEM)) == __GFP_HIGHMEM);
268 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
269 * and __GFP_HIGHMEM from hard or soft interrupt context.
271 VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
272 for (i = 0; i < (1 << order); i++)
273 clear_highpage(page + i);
277 * function for dealing with page's order in buddy system.
278 * zone->lock is already acquired when we use these.
279 * So, we don't need atomic page->flags operations here.
281 static inline unsigned long page_order(struct page *page)
283 return page_private(page);
286 static inline void set_page_order(struct page *page, int order)
288 set_page_private(page, order);
289 __SetPageBuddy(page);
292 static inline void rmv_page_order(struct page *page)
294 __ClearPageBuddy(page);
295 set_page_private(page, 0);
299 * Locate the struct page for both the matching buddy in our
300 * pair (buddy1) and the combined O(n+1) page they form (page).
302 * 1) Any buddy B1 will have an order O twin B2 which satisfies
303 * the following equation:
305 * For example, if the starting buddy (buddy2) is #8 its order
307 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
309 * 2) Any buddy B will have an order O+1 parent P which
310 * satisfies the following equation:
313 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
315 static inline struct page *
316 __page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order)
318 unsigned long buddy_idx = page_idx ^ (1 << order);
320 return page + (buddy_idx - page_idx);
323 static inline unsigned long
324 __find_combined_index(unsigned long page_idx, unsigned int order)
326 return (page_idx & ~(1 << order));
330 * This function checks whether a page is free && is the buddy
331 * we can do coalesce a page and its buddy if
332 * (a) the buddy is not in a hole &&
333 * (b) the buddy is in the buddy system &&
334 * (c) a page and its buddy have the same order &&
335 * (d) a page and its buddy are in the same zone.
337 * For recording whether a page is in the buddy system, we use PG_buddy.
338 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
340 * For recording page's order, we use page_private(page).
342 static inline int page_is_buddy(struct page *page, struct page *buddy,
345 #ifdef CONFIG_HOLES_IN_ZONE
346 if (!pfn_valid(page_to_pfn(buddy)))
350 if (page_zone_id(page) != page_zone_id(buddy))
353 if (PageBuddy(buddy) && page_order(buddy) == order) {
354 BUG_ON(page_count(buddy) != 0);
361 * Freeing function for a buddy system allocator.
363 * The concept of a buddy system is to maintain direct-mapped table
364 * (containing bit values) for memory blocks of various "orders".
365 * The bottom level table contains the map for the smallest allocatable
366 * units of memory (here, pages), and each level above it describes
367 * pairs of units from the levels below, hence, "buddies".
368 * At a high level, all that happens here is marking the table entry
369 * at the bottom level available, and propagating the changes upward
370 * as necessary, plus some accounting needed to play nicely with other
371 * parts of the VM system.
372 * At each level, we keep a list of pages, which are heads of continuous
373 * free pages of length of (1 << order) and marked with PG_buddy. Page's
374 * order is recorded in page_private(page) field.
375 * So when we are allocating or freeing one, we can derive the state of the
376 * other. That is, if we allocate a small block, and both were
377 * free, the remainder of the region must be split into blocks.
378 * If a block is freed, and its buddy is also free, then this
379 * triggers coalescing into a block of larger size.
384 static inline void __free_one_page(struct page *page,
385 struct zone *zone, unsigned int order)
387 unsigned long page_idx;
388 int order_size = 1 << order;
390 if (unlikely(PageCompound(page)))
391 destroy_compound_page(page, order);
393 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
395 VM_BUG_ON(page_idx & (order_size - 1));
396 VM_BUG_ON(bad_range(zone, page));
398 zone->free_pages += order_size;
399 while (order < MAX_ORDER-1) {
400 unsigned long combined_idx;
401 struct free_area *area;
404 buddy = __page_find_buddy(page, page_idx, order);
405 if (!page_is_buddy(page, buddy, order))
406 break; /* Move the buddy up one level. */
408 list_del(&buddy->lru);
409 area = zone->free_area + order;
411 rmv_page_order(buddy);
412 combined_idx = __find_combined_index(page_idx, order);
413 page = page + (combined_idx - page_idx);
414 page_idx = combined_idx;
417 set_page_order(page, order);
418 list_add(&page->lru, &zone->free_area[order].free_list);
419 zone->free_area[order].nr_free++;
422 static inline int free_pages_check(struct page *page)
424 if (unlikely(page_mapcount(page) |
425 (page->mapping != NULL) |
426 (page_count(page) != 0) |
440 __ClearPageDirty(page);
442 * For now, we report if PG_reserved was found set, but do not
443 * clear it, and do not free the page. But we shall soon need
444 * to do more, for when the ZERO_PAGE count wraps negative.
446 return PageReserved(page);
450 * Frees a list of pages.
451 * Assumes all pages on list are in same zone, and of same order.
452 * count is the number of pages to free.
454 * If the zone was previously in an "all pages pinned" state then look to
455 * see if this freeing clears that state.
457 * And clear the zone's pages_scanned counter, to hold off the "all pages are
458 * pinned" detection logic.
460 static void free_pages_bulk(struct zone *zone, int count,
461 struct list_head *list, int order)
463 spin_lock(&zone->lock);
464 zone->all_unreclaimable = 0;
465 zone->pages_scanned = 0;
469 VM_BUG_ON(list_empty(list));
470 page = list_entry(list->prev, struct page, lru);
471 /* have to delete it as __free_one_page list manipulates */
472 list_del(&page->lru);
473 __free_one_page(page, zone, order);
475 spin_unlock(&zone->lock);
478 static void free_one_page(struct zone *zone, struct page *page, int order)
480 spin_lock(&zone->lock);
481 zone->all_unreclaimable = 0;
482 zone->pages_scanned = 0;
483 __free_one_page(page, zone ,order);
484 spin_unlock(&zone->lock);
487 static void __free_pages_ok(struct page *page, unsigned int order)
493 arch_free_page(page, order);
494 if (!PageHighMem(page))
495 debug_check_no_locks_freed(page_address(page),
498 for (i = 0 ; i < (1 << order) ; ++i)
499 reserved += free_pages_check(page + i);
503 kernel_map_pages(page, 1 << order, 0);
504 local_irq_save(flags);
505 __count_vm_events(PGFREE, 1 << order);
506 free_one_page(page_zone(page), page, order);
507 local_irq_restore(flags);
511 * permit the bootmem allocator to evade page validation on high-order frees
513 void fastcall __init __free_pages_bootmem(struct page *page, unsigned int order)
516 __ClearPageReserved(page);
517 set_page_count(page, 0);
518 set_page_refcounted(page);
524 for (loop = 0; loop < BITS_PER_LONG; loop++) {
525 struct page *p = &page[loop];
527 if (loop + 1 < BITS_PER_LONG)
529 __ClearPageReserved(p);
530 set_page_count(p, 0);
533 set_page_refcounted(page);
534 __free_pages(page, order);
540 * The order of subdivision here is critical for the IO subsystem.
541 * Please do not alter this order without good reasons and regression
542 * testing. Specifically, as large blocks of memory are subdivided,
543 * the order in which smaller blocks are delivered depends on the order
544 * they're subdivided in this function. This is the primary factor
545 * influencing the order in which pages are delivered to the IO
546 * subsystem according to empirical testing, and this is also justified
547 * by considering the behavior of a buddy system containing a single
548 * large block of memory acted on by a series of small allocations.
549 * This behavior is a critical factor in sglist merging's success.
553 static inline void expand(struct zone *zone, struct page *page,
554 int low, int high, struct free_area *area)
556 unsigned long size = 1 << high;
562 VM_BUG_ON(bad_range(zone, &page[size]));
563 list_add(&page[size].lru, &area->free_list);
565 set_page_order(&page[size], high);
570 * This page is about to be returned from the page allocator
572 static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
574 if (unlikely(page_mapcount(page) |
575 (page->mapping != NULL) |
576 (page_count(page) != 0) |
592 * For now, we report if PG_reserved was found set, but do not
593 * clear it, and do not allocate the page: as a safety net.
595 if (PageReserved(page))
598 page->flags &= ~(1 << PG_uptodate | 1 << PG_error |
599 1 << PG_referenced | 1 << PG_arch_1 |
600 1 << PG_checked | 1 << PG_mappedtodisk);
601 set_page_private(page, 0);
602 set_page_refcounted(page);
603 kernel_map_pages(page, 1 << order, 1);
605 if (gfp_flags & __GFP_ZERO)
606 prep_zero_page(page, order, gfp_flags);
608 if (order && (gfp_flags & __GFP_COMP))
609 prep_compound_page(page, order);
615 * Do the hard work of removing an element from the buddy allocator.
616 * Call me with the zone->lock already held.
618 static struct page *__rmqueue(struct zone *zone, unsigned int order)
620 struct free_area * area;
621 unsigned int current_order;
624 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
625 area = zone->free_area + current_order;
626 if (list_empty(&area->free_list))
629 page = list_entry(area->free_list.next, struct page, lru);
630 list_del(&page->lru);
631 rmv_page_order(page);
633 zone->free_pages -= 1UL << order;
634 expand(zone, page, order, current_order, area);
642 * Obtain a specified number of elements from the buddy allocator, all under
643 * a single hold of the lock, for efficiency. Add them to the supplied list.
644 * Returns the number of new pages which were placed at *list.
646 static int rmqueue_bulk(struct zone *zone, unsigned int order,
647 unsigned long count, struct list_head *list)
651 spin_lock(&zone->lock);
652 for (i = 0; i < count; ++i) {
653 struct page *page = __rmqueue(zone, order);
654 if (unlikely(page == NULL))
656 list_add_tail(&page->lru, list);
658 spin_unlock(&zone->lock);
664 * Called from the slab reaper to drain pagesets on a particular node that
665 * belongs to the currently executing processor.
666 * Note that this function must be called with the thread pinned to
667 * a single processor.
669 void drain_node_pages(int nodeid)
675 for (z = 0; z < MAX_NR_ZONES; z++) {
676 struct zone *zone = NODE_DATA(nodeid)->node_zones + z;
677 struct per_cpu_pageset *pset;
679 if (!populated_zone(zone))
682 pset = zone_pcp(zone, smp_processor_id());
683 for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) {
684 struct per_cpu_pages *pcp;
688 local_irq_save(flags);
689 free_pages_bulk(zone, pcp->count, &pcp->list, 0);
691 local_irq_restore(flags);
698 #if defined(CONFIG_PM) || defined(CONFIG_HOTPLUG_CPU)
699 static void __drain_pages(unsigned int cpu)
705 for_each_zone(zone) {
706 struct per_cpu_pageset *pset;
708 pset = zone_pcp(zone, cpu);
709 for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) {
710 struct per_cpu_pages *pcp;
713 local_irq_save(flags);
714 free_pages_bulk(zone, pcp->count, &pcp->list, 0);
716 local_irq_restore(flags);
720 #endif /* CONFIG_PM || CONFIG_HOTPLUG_CPU */
724 void mark_free_pages(struct zone *zone)
726 unsigned long pfn, max_zone_pfn;
729 struct list_head *curr;
731 if (!zone->spanned_pages)
734 spin_lock_irqsave(&zone->lock, flags);
736 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
737 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
738 if (pfn_valid(pfn)) {
739 struct page *page = pfn_to_page(pfn);
741 if (!PageNosave(page))
742 ClearPageNosaveFree(page);
745 for (order = MAX_ORDER - 1; order >= 0; --order)
746 list_for_each(curr, &zone->free_area[order].free_list) {
749 pfn = page_to_pfn(list_entry(curr, struct page, lru));
750 for (i = 0; i < (1UL << order); i++)
751 SetPageNosaveFree(pfn_to_page(pfn + i));
754 spin_unlock_irqrestore(&zone->lock, flags);
758 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
760 void drain_local_pages(void)
764 local_irq_save(flags);
765 __drain_pages(smp_processor_id());
766 local_irq_restore(flags);
768 #endif /* CONFIG_PM */
771 * Free a 0-order page
773 static void fastcall free_hot_cold_page(struct page *page, int cold)
775 struct zone *zone = page_zone(page);
776 struct per_cpu_pages *pcp;
779 arch_free_page(page, 0);
782 page->mapping = NULL;
783 if (free_pages_check(page))
786 kernel_map_pages(page, 1, 0);
788 pcp = &zone_pcp(zone, get_cpu())->pcp[cold];
789 local_irq_save(flags);
790 __count_vm_event(PGFREE);
791 list_add(&page->lru, &pcp->list);
793 if (pcp->count >= pcp->high) {
794 free_pages_bulk(zone, pcp->batch, &pcp->list, 0);
795 pcp->count -= pcp->batch;
797 local_irq_restore(flags);
801 void fastcall free_hot_page(struct page *page)
803 free_hot_cold_page(page, 0);
806 void fastcall free_cold_page(struct page *page)
808 free_hot_cold_page(page, 1);
812 * split_page takes a non-compound higher-order page, and splits it into
813 * n (1<<order) sub-pages: page[0..n]
814 * Each sub-page must be freed individually.
816 * Note: this is probably too low level an operation for use in drivers.
817 * Please consult with lkml before using this in your driver.
819 void split_page(struct page *page, unsigned int order)
823 VM_BUG_ON(PageCompound(page));
824 VM_BUG_ON(!page_count(page));
825 for (i = 1; i < (1 << order); i++)
826 set_page_refcounted(page + i);
830 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
831 * we cheat by calling it from here, in the order > 0 path. Saves a branch
834 static struct page *buffered_rmqueue(struct zonelist *zonelist,
835 struct zone *zone, int order, gfp_t gfp_flags)
839 int cold = !!(gfp_flags & __GFP_COLD);
844 if (likely(order == 0)) {
845 struct per_cpu_pages *pcp;
847 pcp = &zone_pcp(zone, cpu)->pcp[cold];
848 local_irq_save(flags);
850 pcp->count += rmqueue_bulk(zone, 0,
851 pcp->batch, &pcp->list);
852 if (unlikely(!pcp->count))
855 page = list_entry(pcp->list.next, struct page, lru);
856 list_del(&page->lru);
859 spin_lock_irqsave(&zone->lock, flags);
860 page = __rmqueue(zone, order);
861 spin_unlock(&zone->lock);
866 __count_zone_vm_events(PGALLOC, zone, 1 << order);
867 zone_statistics(zonelist, zone);
868 local_irq_restore(flags);
871 VM_BUG_ON(bad_range(zone, page));
872 if (prep_new_page(page, order, gfp_flags))
877 local_irq_restore(flags);
882 #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
883 #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
884 #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
885 #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
886 #define ALLOC_HARDER 0x10 /* try to alloc harder */
887 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
888 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
891 * Return 1 if free pages are above 'mark'. This takes into account the order
894 int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
895 int classzone_idx, int alloc_flags)
897 /* free_pages my go negative - that's OK */
898 long min = mark, free_pages = z->free_pages - (1 << order) + 1;
901 if (alloc_flags & ALLOC_HIGH)
903 if (alloc_flags & ALLOC_HARDER)
906 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
908 for (o = 0; o < order; o++) {
909 /* At the next order, this order's pages become unavailable */
910 free_pages -= z->free_area[o].nr_free << o;
912 /* Require fewer higher order pages to be free */
915 if (free_pages <= min)
922 * get_page_from_freeliest goes through the zonelist trying to allocate
926 get_page_from_freelist(gfp_t gfp_mask, unsigned int order,
927 struct zonelist *zonelist, int alloc_flags)
929 struct zone **z = zonelist->zones;
930 struct page *page = NULL;
931 int classzone_idx = zone_idx(*z);
935 * Go through the zonelist once, looking for a zone with enough free.
936 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
940 if (unlikely((gfp_mask & __GFP_THISNODE) &&
941 zone->zone_pgdat != zonelist->zones[0]->zone_pgdat))
943 if ((alloc_flags & ALLOC_CPUSET) &&
944 !cpuset_zone_allowed(zone, gfp_mask))
947 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
949 if (alloc_flags & ALLOC_WMARK_MIN)
950 mark = zone->pages_min;
951 else if (alloc_flags & ALLOC_WMARK_LOW)
952 mark = zone->pages_low;
954 mark = zone->pages_high;
955 if (!zone_watermark_ok(zone , order, mark,
956 classzone_idx, alloc_flags))
957 if (!zone_reclaim_mode ||
958 !zone_reclaim(zone, gfp_mask, order))
962 page = buffered_rmqueue(zonelist, zone, order, gfp_mask);
966 } while (*(++z) != NULL);
971 * This is the 'heart' of the zoned buddy allocator.
973 struct page * fastcall
974 __alloc_pages(gfp_t gfp_mask, unsigned int order,
975 struct zonelist *zonelist)
977 const gfp_t wait = gfp_mask & __GFP_WAIT;
980 struct reclaim_state reclaim_state;
981 struct task_struct *p = current;
984 int did_some_progress;
986 might_sleep_if(wait);
989 z = zonelist->zones; /* the list of zones suitable for gfp_mask */
991 if (unlikely(*z == NULL)) {
992 /* Should this ever happen?? */
996 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
997 zonelist, ALLOC_WMARK_LOW|ALLOC_CPUSET);
1002 wakeup_kswapd(*z, order);
1006 * OK, we're below the kswapd watermark and have kicked background
1007 * reclaim. Now things get more complex, so set up alloc_flags according
1008 * to how we want to proceed.
1010 * The caller may dip into page reserves a bit more if the caller
1011 * cannot run direct reclaim, or if the caller has realtime scheduling
1012 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1013 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1015 alloc_flags = ALLOC_WMARK_MIN;
1016 if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait)
1017 alloc_flags |= ALLOC_HARDER;
1018 if (gfp_mask & __GFP_HIGH)
1019 alloc_flags |= ALLOC_HIGH;
1021 alloc_flags |= ALLOC_CPUSET;
1024 * Go through the zonelist again. Let __GFP_HIGH and allocations
1025 * coming from realtime tasks go deeper into reserves.
1027 * This is the last chance, in general, before the goto nopage.
1028 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1029 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1031 page = get_page_from_freelist(gfp_mask, order, zonelist, alloc_flags);
1035 /* This allocation should allow future memory freeing. */
1037 if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE)))
1038 && !in_interrupt()) {
1039 if (!(gfp_mask & __GFP_NOMEMALLOC)) {
1041 /* go through the zonelist yet again, ignoring mins */
1042 page = get_page_from_freelist(gfp_mask, order,
1043 zonelist, ALLOC_NO_WATERMARKS);
1046 if (gfp_mask & __GFP_NOFAIL) {
1047 blk_congestion_wait(WRITE, HZ/50);
1054 /* Atomic allocations - we can't balance anything */
1061 /* We now go into synchronous reclaim */
1062 cpuset_memory_pressure_bump();
1063 p->flags |= PF_MEMALLOC;
1064 reclaim_state.reclaimed_slab = 0;
1065 p->reclaim_state = &reclaim_state;
1067 did_some_progress = try_to_free_pages(zonelist->zones, gfp_mask);
1069 p->reclaim_state = NULL;
1070 p->flags &= ~PF_MEMALLOC;
1074 if (likely(did_some_progress)) {
1075 page = get_page_from_freelist(gfp_mask, order,
1076 zonelist, alloc_flags);
1079 } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
1081 * Go through the zonelist yet one more time, keep
1082 * very high watermark here, this is only to catch
1083 * a parallel oom killing, we must fail if we're still
1084 * under heavy pressure.
1086 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
1087 zonelist, ALLOC_WMARK_HIGH|ALLOC_CPUSET);
1091 out_of_memory(zonelist, gfp_mask, order);
1096 * Don't let big-order allocations loop unless the caller explicitly
1097 * requests that. Wait for some write requests to complete then retry.
1099 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
1100 * <= 3, but that may not be true in other implementations.
1103 if (!(gfp_mask & __GFP_NORETRY)) {
1104 if ((order <= 3) || (gfp_mask & __GFP_REPEAT))
1106 if (gfp_mask & __GFP_NOFAIL)
1110 blk_congestion_wait(WRITE, HZ/50);
1115 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
1116 printk(KERN_WARNING "%s: page allocation failure."
1117 " order:%d, mode:0x%x\n",
1118 p->comm, order, gfp_mask);
1126 EXPORT_SYMBOL(__alloc_pages);
1129 * Common helper functions.
1131 fastcall unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
1134 page = alloc_pages(gfp_mask, order);
1137 return (unsigned long) page_address(page);
1140 EXPORT_SYMBOL(__get_free_pages);
1142 fastcall unsigned long get_zeroed_page(gfp_t gfp_mask)
1147 * get_zeroed_page() returns a 32-bit address, which cannot represent
1150 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
1152 page = alloc_pages(gfp_mask | __GFP_ZERO, 0);
1154 return (unsigned long) page_address(page);
1158 EXPORT_SYMBOL(get_zeroed_page);
1160 void __pagevec_free(struct pagevec *pvec)
1162 int i = pagevec_count(pvec);
1165 free_hot_cold_page(pvec->pages[i], pvec->cold);
1168 fastcall void __free_pages(struct page *page, unsigned int order)
1170 if (put_page_testzero(page)) {
1172 free_hot_page(page);
1174 __free_pages_ok(page, order);
1178 EXPORT_SYMBOL(__free_pages);
1180 fastcall void free_pages(unsigned long addr, unsigned int order)
1183 VM_BUG_ON(!virt_addr_valid((void *)addr));
1184 __free_pages(virt_to_page((void *)addr), order);
1188 EXPORT_SYMBOL(free_pages);
1191 * Total amount of free (allocatable) RAM:
1193 unsigned int nr_free_pages(void)
1195 unsigned int sum = 0;
1199 sum += zone->free_pages;
1204 EXPORT_SYMBOL(nr_free_pages);
1207 unsigned int nr_free_pages_pgdat(pg_data_t *pgdat)
1209 unsigned int sum = 0;
1212 for (i = 0; i < MAX_NR_ZONES; i++)
1213 sum += pgdat->node_zones[i].free_pages;
1219 static unsigned int nr_free_zone_pages(int offset)
1221 /* Just pick one node, since fallback list is circular */
1222 pg_data_t *pgdat = NODE_DATA(numa_node_id());
1223 unsigned int sum = 0;
1225 struct zonelist *zonelist = pgdat->node_zonelists + offset;
1226 struct zone **zonep = zonelist->zones;
1229 for (zone = *zonep++; zone; zone = *zonep++) {
1230 unsigned long size = zone->present_pages;
1231 unsigned long high = zone->pages_high;
1240 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1242 unsigned int nr_free_buffer_pages(void)
1244 return nr_free_zone_pages(gfp_zone(GFP_USER));
1248 * Amount of free RAM allocatable within all zones
1250 unsigned int nr_free_pagecache_pages(void)
1252 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER));
1255 static void show_node(struct zone *zone)
1257 printk("Node %ld ", zone_to_nid(zone));
1260 #define show_node(zone) do { } while (0)
1263 void si_meminfo(struct sysinfo *val)
1265 val->totalram = totalram_pages;
1267 val->freeram = nr_free_pages();
1268 val->bufferram = nr_blockdev_pages();
1269 val->totalhigh = totalhigh_pages;
1270 val->freehigh = nr_free_highpages();
1271 val->mem_unit = PAGE_SIZE;
1274 EXPORT_SYMBOL(si_meminfo);
1277 void si_meminfo_node(struct sysinfo *val, int nid)
1279 pg_data_t *pgdat = NODE_DATA(nid);
1281 val->totalram = pgdat->node_present_pages;
1282 val->freeram = nr_free_pages_pgdat(pgdat);
1283 #ifdef CONFIG_HIGHMEM
1284 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1285 val->freehigh = pgdat->node_zones[ZONE_HIGHMEM].free_pages;
1290 val->mem_unit = PAGE_SIZE;
1294 #define K(x) ((x) << (PAGE_SHIFT-10))
1297 * Show free area list (used inside shift_scroll-lock stuff)
1298 * We also calculate the percentage fragmentation. We do this by counting the
1299 * memory on each free list with the exception of the first item on the list.
1301 void show_free_areas(void)
1303 int cpu, temperature;
1304 unsigned long active;
1305 unsigned long inactive;
1309 for_each_zone(zone) {
1311 printk("%s per-cpu:", zone->name);
1313 if (!populated_zone(zone)) {
1319 for_each_online_cpu(cpu) {
1320 struct per_cpu_pageset *pageset;
1322 pageset = zone_pcp(zone, cpu);
1324 for (temperature = 0; temperature < 2; temperature++)
1325 printk("cpu %d %s: high %d, batch %d used:%d\n",
1327 temperature ? "cold" : "hot",
1328 pageset->pcp[temperature].high,
1329 pageset->pcp[temperature].batch,
1330 pageset->pcp[temperature].count);
1334 get_zone_counts(&active, &inactive, &free);
1336 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu "
1337 "unstable:%lu free:%u slab:%lu mapped:%lu pagetables:%lu\n",
1340 global_page_state(NR_FILE_DIRTY),
1341 global_page_state(NR_WRITEBACK),
1342 global_page_state(NR_UNSTABLE_NFS),
1344 global_page_state(NR_SLAB_RECLAIMABLE) +
1345 global_page_state(NR_SLAB_UNRECLAIMABLE),
1346 global_page_state(NR_FILE_MAPPED),
1347 global_page_state(NR_PAGETABLE));
1349 for_each_zone(zone) {
1361 " pages_scanned:%lu"
1362 " all_unreclaimable? %s"
1365 K(zone->free_pages),
1368 K(zone->pages_high),
1370 K(zone->nr_inactive),
1371 K(zone->present_pages),
1372 zone->pages_scanned,
1373 (zone->all_unreclaimable ? "yes" : "no")
1375 printk("lowmem_reserve[]:");
1376 for (i = 0; i < MAX_NR_ZONES; i++)
1377 printk(" %lu", zone->lowmem_reserve[i]);
1381 for_each_zone(zone) {
1382 unsigned long nr[MAX_ORDER], flags, order, total = 0;
1385 printk("%s: ", zone->name);
1386 if (!populated_zone(zone)) {
1391 spin_lock_irqsave(&zone->lock, flags);
1392 for (order = 0; order < MAX_ORDER; order++) {
1393 nr[order] = zone->free_area[order].nr_free;
1394 total += nr[order] << order;
1396 spin_unlock_irqrestore(&zone->lock, flags);
1397 for (order = 0; order < MAX_ORDER; order++)
1398 printk("%lu*%lukB ", nr[order], K(1UL) << order);
1399 printk("= %lukB\n", K(total));
1402 show_swap_cache_info();
1406 * Builds allocation fallback zone lists.
1408 * Add all populated zones of a node to the zonelist.
1410 static int __meminit build_zonelists_node(pg_data_t *pgdat,
1411 struct zonelist *zonelist, int nr_zones, enum zone_type zone_type)
1415 BUG_ON(zone_type >= MAX_NR_ZONES);
1420 zone = pgdat->node_zones + zone_type;
1421 if (populated_zone(zone)) {
1422 zonelist->zones[nr_zones++] = zone;
1423 check_highest_zone(zone_type);
1426 } while (zone_type);
1431 #define MAX_NODE_LOAD (num_online_nodes())
1432 static int __meminitdata node_load[MAX_NUMNODES];
1434 * find_next_best_node - find the next node that should appear in a given node's fallback list
1435 * @node: node whose fallback list we're appending
1436 * @used_node_mask: nodemask_t of already used nodes
1438 * We use a number of factors to determine which is the next node that should
1439 * appear on a given node's fallback list. The node should not have appeared
1440 * already in @node's fallback list, and it should be the next closest node
1441 * according to the distance array (which contains arbitrary distance values
1442 * from each node to each node in the system), and should also prefer nodes
1443 * with no CPUs, since presumably they'll have very little allocation pressure
1444 * on them otherwise.
1445 * It returns -1 if no node is found.
1447 static int __meminit find_next_best_node(int node, nodemask_t *used_node_mask)
1450 int min_val = INT_MAX;
1453 /* Use the local node if we haven't already */
1454 if (!node_isset(node, *used_node_mask)) {
1455 node_set(node, *used_node_mask);
1459 for_each_online_node(n) {
1462 /* Don't want a node to appear more than once */
1463 if (node_isset(n, *used_node_mask))
1466 /* Use the distance array to find the distance */
1467 val = node_distance(node, n);
1469 /* Penalize nodes under us ("prefer the next node") */
1472 /* Give preference to headless and unused nodes */
1473 tmp = node_to_cpumask(n);
1474 if (!cpus_empty(tmp))
1475 val += PENALTY_FOR_NODE_WITH_CPUS;
1477 /* Slight preference for less loaded node */
1478 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
1479 val += node_load[n];
1481 if (val < min_val) {
1488 node_set(best_node, *used_node_mask);
1493 static void __meminit build_zonelists(pg_data_t *pgdat)
1495 int j, node, local_node;
1497 int prev_node, load;
1498 struct zonelist *zonelist;
1499 nodemask_t used_mask;
1501 /* initialize zonelists */
1502 for (i = 0; i < MAX_NR_ZONES; i++) {
1503 zonelist = pgdat->node_zonelists + i;
1504 zonelist->zones[0] = NULL;
1507 /* NUMA-aware ordering of nodes */
1508 local_node = pgdat->node_id;
1509 load = num_online_nodes();
1510 prev_node = local_node;
1511 nodes_clear(used_mask);
1512 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
1513 int distance = node_distance(local_node, node);
1516 * If another node is sufficiently far away then it is better
1517 * to reclaim pages in a zone before going off node.
1519 if (distance > RECLAIM_DISTANCE)
1520 zone_reclaim_mode = 1;
1523 * We don't want to pressure a particular node.
1524 * So adding penalty to the first node in same
1525 * distance group to make it round-robin.
1528 if (distance != node_distance(local_node, prev_node))
1529 node_load[node] += load;
1532 for (i = 0; i < MAX_NR_ZONES; i++) {
1533 zonelist = pgdat->node_zonelists + i;
1534 for (j = 0; zonelist->zones[j] != NULL; j++);
1536 j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
1537 zonelist->zones[j] = NULL;
1542 #else /* CONFIG_NUMA */
1544 static void __meminit build_zonelists(pg_data_t *pgdat)
1546 int node, local_node;
1549 local_node = pgdat->node_id;
1550 for (i = 0; i < MAX_NR_ZONES; i++) {
1551 struct zonelist *zonelist;
1553 zonelist = pgdat->node_zonelists + i;
1555 j = build_zonelists_node(pgdat, zonelist, 0, i);
1557 * Now we build the zonelist so that it contains the zones
1558 * of all the other nodes.
1559 * We don't want to pressure a particular node, so when
1560 * building the zones for node N, we make sure that the
1561 * zones coming right after the local ones are those from
1562 * node N+1 (modulo N)
1564 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
1565 if (!node_online(node))
1567 j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
1569 for (node = 0; node < local_node; node++) {
1570 if (!node_online(node))
1572 j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
1575 zonelist->zones[j] = NULL;
1579 #endif /* CONFIG_NUMA */
1581 /* return values int ....just for stop_machine_run() */
1582 static int __meminit __build_all_zonelists(void *dummy)
1585 for_each_online_node(nid)
1586 build_zonelists(NODE_DATA(nid));
1590 void __meminit build_all_zonelists(void)
1592 if (system_state == SYSTEM_BOOTING) {
1593 __build_all_zonelists(0);
1594 cpuset_init_current_mems_allowed();
1596 /* we have to stop all cpus to guaranntee there is no user
1598 stop_machine_run(__build_all_zonelists, NULL, NR_CPUS);
1599 /* cpuset refresh routine should be here */
1601 vm_total_pages = nr_free_pagecache_pages();
1602 printk("Built %i zonelists. Total pages: %ld\n",
1603 num_online_nodes(), vm_total_pages);
1607 * Helper functions to size the waitqueue hash table.
1608 * Essentially these want to choose hash table sizes sufficiently
1609 * large so that collisions trying to wait on pages are rare.
1610 * But in fact, the number of active page waitqueues on typical
1611 * systems is ridiculously low, less than 200. So this is even
1612 * conservative, even though it seems large.
1614 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
1615 * waitqueues, i.e. the size of the waitq table given the number of pages.
1617 #define PAGES_PER_WAITQUEUE 256
1619 #ifndef CONFIG_MEMORY_HOTPLUG
1620 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
1622 unsigned long size = 1;
1624 pages /= PAGES_PER_WAITQUEUE;
1626 while (size < pages)
1630 * Once we have dozens or even hundreds of threads sleeping
1631 * on IO we've got bigger problems than wait queue collision.
1632 * Limit the size of the wait table to a reasonable size.
1634 size = min(size, 4096UL);
1636 return max(size, 4UL);
1640 * A zone's size might be changed by hot-add, so it is not possible to determine
1641 * a suitable size for its wait_table. So we use the maximum size now.
1643 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
1645 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
1646 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
1647 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
1649 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
1650 * or more by the traditional way. (See above). It equals:
1652 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
1653 * ia64(16K page size) : = ( 8G + 4M)byte.
1654 * powerpc (64K page size) : = (32G +16M)byte.
1656 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
1663 * This is an integer logarithm so that shifts can be used later
1664 * to extract the more random high bits from the multiplicative
1665 * hash function before the remainder is taken.
1667 static inline unsigned long wait_table_bits(unsigned long size)
1672 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
1675 * Initially all pages are reserved - free ones are freed
1676 * up by free_all_bootmem() once the early boot process is
1677 * done. Non-atomic initialization, single-pass.
1679 void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
1680 unsigned long start_pfn)
1683 unsigned long end_pfn = start_pfn + size;
1686 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1687 if (!early_pfn_valid(pfn))
1689 page = pfn_to_page(pfn);
1690 set_page_links(page, zone, nid, pfn);
1691 init_page_count(page);
1692 reset_page_mapcount(page);
1693 SetPageReserved(page);
1694 INIT_LIST_HEAD(&page->lru);
1695 #ifdef WANT_PAGE_VIRTUAL
1696 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
1697 if (!is_highmem_idx(zone))
1698 set_page_address(page, __va(pfn << PAGE_SHIFT));
1703 void zone_init_free_lists(struct pglist_data *pgdat, struct zone *zone,
1707 for (order = 0; order < MAX_ORDER ; order++) {
1708 INIT_LIST_HEAD(&zone->free_area[order].free_list);
1709 zone->free_area[order].nr_free = 0;
1713 #define ZONETABLE_INDEX(x, zone_nr) ((x << ZONES_SHIFT) | zone_nr)
1714 void zonetable_add(struct zone *zone, int nid, enum zone_type zid,
1715 unsigned long pfn, unsigned long size)
1717 unsigned long snum = pfn_to_section_nr(pfn);
1718 unsigned long end = pfn_to_section_nr(pfn + size);
1721 zone_table[ZONETABLE_INDEX(nid, zid)] = zone;
1723 for (; snum <= end; snum++)
1724 zone_table[ZONETABLE_INDEX(snum, zid)] = zone;
1727 #ifndef __HAVE_ARCH_MEMMAP_INIT
1728 #define memmap_init(size, nid, zone, start_pfn) \
1729 memmap_init_zone((size), (nid), (zone), (start_pfn))
1732 static int __cpuinit zone_batchsize(struct zone *zone)
1737 * The per-cpu-pages pools are set to around 1000th of the
1738 * size of the zone. But no more than 1/2 of a meg.
1740 * OK, so we don't know how big the cache is. So guess.
1742 batch = zone->present_pages / 1024;
1743 if (batch * PAGE_SIZE > 512 * 1024)
1744 batch = (512 * 1024) / PAGE_SIZE;
1745 batch /= 4; /* We effectively *= 4 below */
1750 * Clamp the batch to a 2^n - 1 value. Having a power
1751 * of 2 value was found to be more likely to have
1752 * suboptimal cache aliasing properties in some cases.
1754 * For example if 2 tasks are alternately allocating
1755 * batches of pages, one task can end up with a lot
1756 * of pages of one half of the possible page colors
1757 * and the other with pages of the other colors.
1759 batch = (1 << (fls(batch + batch/2)-1)) - 1;
1764 inline void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
1766 struct per_cpu_pages *pcp;
1768 memset(p, 0, sizeof(*p));
1770 pcp = &p->pcp[0]; /* hot */
1772 pcp->high = 6 * batch;
1773 pcp->batch = max(1UL, 1 * batch);
1774 INIT_LIST_HEAD(&pcp->list);
1776 pcp = &p->pcp[1]; /* cold*/
1778 pcp->high = 2 * batch;
1779 pcp->batch = max(1UL, batch/2);
1780 INIT_LIST_HEAD(&pcp->list);
1784 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
1785 * to the value high for the pageset p.
1788 static void setup_pagelist_highmark(struct per_cpu_pageset *p,
1791 struct per_cpu_pages *pcp;
1793 pcp = &p->pcp[0]; /* hot list */
1795 pcp->batch = max(1UL, high/4);
1796 if ((high/4) > (PAGE_SHIFT * 8))
1797 pcp->batch = PAGE_SHIFT * 8;
1803 * Boot pageset table. One per cpu which is going to be used for all
1804 * zones and all nodes. The parameters will be set in such a way
1805 * that an item put on a list will immediately be handed over to
1806 * the buddy list. This is safe since pageset manipulation is done
1807 * with interrupts disabled.
1809 * Some NUMA counter updates may also be caught by the boot pagesets.
1811 * The boot_pagesets must be kept even after bootup is complete for
1812 * unused processors and/or zones. They do play a role for bootstrapping
1813 * hotplugged processors.
1815 * zoneinfo_show() and maybe other functions do
1816 * not check if the processor is online before following the pageset pointer.
1817 * Other parts of the kernel may not check if the zone is available.
1819 static struct per_cpu_pageset boot_pageset[NR_CPUS];
1822 * Dynamically allocate memory for the
1823 * per cpu pageset array in struct zone.
1825 static int __cpuinit process_zones(int cpu)
1827 struct zone *zone, *dzone;
1829 for_each_zone(zone) {
1831 zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset),
1832 GFP_KERNEL, cpu_to_node(cpu));
1833 if (!zone_pcp(zone, cpu))
1836 setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone));
1838 if (percpu_pagelist_fraction)
1839 setup_pagelist_highmark(zone_pcp(zone, cpu),
1840 (zone->present_pages / percpu_pagelist_fraction));
1845 for_each_zone(dzone) {
1848 kfree(zone_pcp(dzone, cpu));
1849 zone_pcp(dzone, cpu) = NULL;
1854 static inline void free_zone_pagesets(int cpu)
1858 for_each_zone(zone) {
1859 struct per_cpu_pageset *pset = zone_pcp(zone, cpu);
1861 /* Free per_cpu_pageset if it is slab allocated */
1862 if (pset != &boot_pageset[cpu])
1864 zone_pcp(zone, cpu) = NULL;
1868 static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb,
1869 unsigned long action,
1872 int cpu = (long)hcpu;
1873 int ret = NOTIFY_OK;
1876 case CPU_UP_PREPARE:
1877 if (process_zones(cpu))
1880 case CPU_UP_CANCELED:
1882 free_zone_pagesets(cpu);
1890 static struct notifier_block __cpuinitdata pageset_notifier =
1891 { &pageset_cpuup_callback, NULL, 0 };
1893 void __init setup_per_cpu_pageset(void)
1897 /* Initialize per_cpu_pageset for cpu 0.
1898 * A cpuup callback will do this for every cpu
1899 * as it comes online
1901 err = process_zones(smp_processor_id());
1903 register_cpu_notifier(&pageset_notifier);
1909 int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
1912 struct pglist_data *pgdat = zone->zone_pgdat;
1916 * The per-page waitqueue mechanism uses hashed waitqueues
1919 zone->wait_table_hash_nr_entries =
1920 wait_table_hash_nr_entries(zone_size_pages);
1921 zone->wait_table_bits =
1922 wait_table_bits(zone->wait_table_hash_nr_entries);
1923 alloc_size = zone->wait_table_hash_nr_entries
1924 * sizeof(wait_queue_head_t);
1926 if (system_state == SYSTEM_BOOTING) {
1927 zone->wait_table = (wait_queue_head_t *)
1928 alloc_bootmem_node(pgdat, alloc_size);
1931 * This case means that a zone whose size was 0 gets new memory
1932 * via memory hot-add.
1933 * But it may be the case that a new node was hot-added. In
1934 * this case vmalloc() will not be able to use this new node's
1935 * memory - this wait_table must be initialized to use this new
1936 * node itself as well.
1937 * To use this new node's memory, further consideration will be
1940 zone->wait_table = (wait_queue_head_t *)vmalloc(alloc_size);
1942 if (!zone->wait_table)
1945 for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
1946 init_waitqueue_head(zone->wait_table + i);
1951 static __meminit void zone_pcp_init(struct zone *zone)
1954 unsigned long batch = zone_batchsize(zone);
1956 for (cpu = 0; cpu < NR_CPUS; cpu++) {
1958 /* Early boot. Slab allocator not functional yet */
1959 zone_pcp(zone, cpu) = &boot_pageset[cpu];
1960 setup_pageset(&boot_pageset[cpu],0);
1962 setup_pageset(zone_pcp(zone,cpu), batch);
1965 if (zone->present_pages)
1966 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n",
1967 zone->name, zone->present_pages, batch);
1970 __meminit int init_currently_empty_zone(struct zone *zone,
1971 unsigned long zone_start_pfn,
1974 struct pglist_data *pgdat = zone->zone_pgdat;
1976 ret = zone_wait_table_init(zone, size);
1979 pgdat->nr_zones = zone_idx(zone) + 1;
1981 zone->zone_start_pfn = zone_start_pfn;
1983 memmap_init(size, pgdat->node_id, zone_idx(zone), zone_start_pfn);
1985 zone_init_free_lists(pgdat, zone, zone->spanned_pages);
1990 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
1992 * Basic iterator support. Return the first range of PFNs for a node
1993 * Note: nid == MAX_NUMNODES returns first region regardless of node
1995 static int __init first_active_region_index_in_nid(int nid)
1999 for (i = 0; i < nr_nodemap_entries; i++)
2000 if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
2007 * Basic iterator support. Return the next active range of PFNs for a node
2008 * Note: nid == MAX_NUMNODES returns next region regardles of node
2010 static int __init next_active_region_index_in_nid(int index, int nid)
2012 for (index = index + 1; index < nr_nodemap_entries; index++)
2013 if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
2019 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
2021 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
2022 * Architectures may implement their own version but if add_active_range()
2023 * was used and there are no special requirements, this is a convenient
2026 int __init early_pfn_to_nid(unsigned long pfn)
2030 for (i = 0; i < nr_nodemap_entries; i++) {
2031 unsigned long start_pfn = early_node_map[i].start_pfn;
2032 unsigned long end_pfn = early_node_map[i].end_pfn;
2034 if (start_pfn <= pfn && pfn < end_pfn)
2035 return early_node_map[i].nid;
2040 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
2042 /* Basic iterator support to walk early_node_map[] */
2043 #define for_each_active_range_index_in_nid(i, nid) \
2044 for (i = first_active_region_index_in_nid(nid); i != -1; \
2045 i = next_active_region_index_in_nid(i, nid))
2048 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
2049 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed
2050 * @max_low_pfn: The highest PFN that till be passed to free_bootmem_node
2052 * If an architecture guarantees that all ranges registered with
2053 * add_active_ranges() contain no holes and may be freed, this
2054 * this function may be used instead of calling free_bootmem() manually.
2056 void __init free_bootmem_with_active_regions(int nid,
2057 unsigned long max_low_pfn)
2061 for_each_active_range_index_in_nid(i, nid) {
2062 unsigned long size_pages = 0;
2063 unsigned long end_pfn = early_node_map[i].end_pfn;
2065 if (early_node_map[i].start_pfn >= max_low_pfn)
2068 if (end_pfn > max_low_pfn)
2069 end_pfn = max_low_pfn;
2071 size_pages = end_pfn - early_node_map[i].start_pfn;
2072 free_bootmem_node(NODE_DATA(early_node_map[i].nid),
2073 PFN_PHYS(early_node_map[i].start_pfn),
2074 size_pages << PAGE_SHIFT);
2079 * sparse_memory_present_with_active_regions - Call memory_present for each active range
2080 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used
2082 * If an architecture guarantees that all ranges registered with
2083 * add_active_ranges() contain no holes and may be freed, this
2084 * this function may be used instead of calling memory_present() manually.
2086 void __init sparse_memory_present_with_active_regions(int nid)
2090 for_each_active_range_index_in_nid(i, nid)
2091 memory_present(early_node_map[i].nid,
2092 early_node_map[i].start_pfn,
2093 early_node_map[i].end_pfn);
2097 * get_pfn_range_for_nid - Return the start and end page frames for a node
2098 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned
2099 * @start_pfn: Passed by reference. On return, it will have the node start_pfn
2100 * @end_pfn: Passed by reference. On return, it will have the node end_pfn
2102 * It returns the start and end page frame of a node based on information
2103 * provided by an arch calling add_active_range(). If called for a node
2104 * with no available memory, a warning is printed and the start and end
2107 void __init get_pfn_range_for_nid(unsigned int nid,
2108 unsigned long *start_pfn, unsigned long *end_pfn)
2114 for_each_active_range_index_in_nid(i, nid) {
2115 *start_pfn = min(*start_pfn, early_node_map[i].start_pfn);
2116 *end_pfn = max(*end_pfn, early_node_map[i].end_pfn);
2119 if (*start_pfn == -1UL) {
2120 printk(KERN_WARNING "Node %u active with no memory\n", nid);
2126 * Return the number of pages a zone spans in a node, including holes
2127 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
2129 unsigned long __init zone_spanned_pages_in_node(int nid,
2130 unsigned long zone_type,
2131 unsigned long *ignored)
2133 unsigned long node_start_pfn, node_end_pfn;
2134 unsigned long zone_start_pfn, zone_end_pfn;
2136 /* Get the start and end of the node and zone */
2137 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
2138 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
2139 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
2141 /* Check that this node has pages within the zone's required range */
2142 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
2145 /* Move the zone boundaries inside the node if necessary */
2146 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
2147 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
2149 /* Return the spanned pages */
2150 return zone_end_pfn - zone_start_pfn;
2154 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
2155 * then all holes in the requested range will be accounted for
2157 unsigned long __init __absent_pages_in_range(int nid,
2158 unsigned long range_start_pfn,
2159 unsigned long range_end_pfn)
2162 unsigned long prev_end_pfn = 0, hole_pages = 0;
2163 unsigned long start_pfn;
2165 /* Find the end_pfn of the first active range of pfns in the node */
2166 i = first_active_region_index_in_nid(nid);
2170 prev_end_pfn = early_node_map[i].start_pfn;
2172 /* Find all holes for the zone within the node */
2173 for (; i != -1; i = next_active_region_index_in_nid(i, nid)) {
2175 /* No need to continue if prev_end_pfn is outside the zone */
2176 if (prev_end_pfn >= range_end_pfn)
2179 /* Make sure the end of the zone is not within the hole */
2180 start_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
2181 prev_end_pfn = max(prev_end_pfn, range_start_pfn);
2183 /* Update the hole size cound and move on */
2184 if (start_pfn > range_start_pfn) {
2185 BUG_ON(prev_end_pfn > start_pfn);
2186 hole_pages += start_pfn - prev_end_pfn;
2188 prev_end_pfn = early_node_map[i].end_pfn;
2195 * absent_pages_in_range - Return number of page frames in holes within a range
2196 * @start_pfn: The start PFN to start searching for holes
2197 * @end_pfn: The end PFN to stop searching for holes
2199 * It returns the number of pages frames in memory holes within a range
2201 unsigned long __init absent_pages_in_range(unsigned long start_pfn,
2202 unsigned long end_pfn)
2204 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
2207 /* Return the number of page frames in holes in a zone on a node */
2208 unsigned long __init zone_absent_pages_in_node(int nid,
2209 unsigned long zone_type,
2210 unsigned long *ignored)
2212 return __absent_pages_in_range(nid,
2213 arch_zone_lowest_possible_pfn[zone_type],
2214 arch_zone_highest_possible_pfn[zone_type]);
2217 static inline unsigned long zone_spanned_pages_in_node(int nid,
2218 unsigned long zone_type,
2219 unsigned long *zones_size)
2221 return zones_size[zone_type];
2224 static inline unsigned long zone_absent_pages_in_node(int nid,
2225 unsigned long zone_type,
2226 unsigned long *zholes_size)
2231 return zholes_size[zone_type];
2235 static void __init calculate_node_totalpages(struct pglist_data *pgdat,
2236 unsigned long *zones_size, unsigned long *zholes_size)
2238 unsigned long realtotalpages, totalpages = 0;
2241 for (i = 0; i < MAX_NR_ZONES; i++)
2242 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
2244 pgdat->node_spanned_pages = totalpages;
2246 realtotalpages = totalpages;
2247 for (i = 0; i < MAX_NR_ZONES; i++)
2249 zone_absent_pages_in_node(pgdat->node_id, i,
2251 pgdat->node_present_pages = realtotalpages;
2252 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
2257 * Set up the zone data structures:
2258 * - mark all pages reserved
2259 * - mark all memory queues empty
2260 * - clear the memory bitmaps
2262 static void __meminit free_area_init_core(struct pglist_data *pgdat,
2263 unsigned long *zones_size, unsigned long *zholes_size)
2266 int nid = pgdat->node_id;
2267 unsigned long zone_start_pfn = pgdat->node_start_pfn;
2270 pgdat_resize_init(pgdat);
2271 pgdat->nr_zones = 0;
2272 init_waitqueue_head(&pgdat->kswapd_wait);
2273 pgdat->kswapd_max_order = 0;
2275 for (j = 0; j < MAX_NR_ZONES; j++) {
2276 struct zone *zone = pgdat->node_zones + j;
2277 unsigned long size, realsize;
2279 size = zone_spanned_pages_in_node(nid, j, zones_size);
2280 realsize = size - zone_absent_pages_in_node(nid, j,
2283 if (!is_highmem_idx(j))
2284 nr_kernel_pages += realsize;
2285 nr_all_pages += realsize;
2287 zone->spanned_pages = size;
2288 zone->present_pages = realsize;
2290 zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
2292 zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100;
2294 zone->name = zone_names[j];
2295 spin_lock_init(&zone->lock);
2296 spin_lock_init(&zone->lru_lock);
2297 zone_seqlock_init(zone);
2298 zone->zone_pgdat = pgdat;
2299 zone->free_pages = 0;
2301 zone->temp_priority = zone->prev_priority = DEF_PRIORITY;
2303 zone_pcp_init(zone);
2304 INIT_LIST_HEAD(&zone->active_list);
2305 INIT_LIST_HEAD(&zone->inactive_list);
2306 zone->nr_scan_active = 0;
2307 zone->nr_scan_inactive = 0;
2308 zone->nr_active = 0;
2309 zone->nr_inactive = 0;
2310 zap_zone_vm_stats(zone);
2311 atomic_set(&zone->reclaim_in_progress, 0);
2315 zonetable_add(zone, nid, j, zone_start_pfn, size);
2316 ret = init_currently_empty_zone(zone, zone_start_pfn, size);
2318 zone_start_pfn += size;
2322 static void __init alloc_node_mem_map(struct pglist_data *pgdat)
2324 /* Skip empty nodes */
2325 if (!pgdat->node_spanned_pages)
2328 #ifdef CONFIG_FLAT_NODE_MEM_MAP
2329 /* ia64 gets its own node_mem_map, before this, without bootmem */
2330 if (!pgdat->node_mem_map) {
2331 unsigned long size, start, end;
2335 * The zone's endpoints aren't required to be MAX_ORDER
2336 * aligned but the node_mem_map endpoints must be in order
2337 * for the buddy allocator to function correctly.
2339 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
2340 end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
2341 end = ALIGN(end, MAX_ORDER_NR_PAGES);
2342 size = (end - start) * sizeof(struct page);
2343 map = alloc_remap(pgdat->node_id, size);
2345 map = alloc_bootmem_node(pgdat, size);
2346 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
2348 #ifdef CONFIG_FLATMEM
2350 * With no DISCONTIG, the global mem_map is just set as node 0's
2352 if (pgdat == NODE_DATA(0)) {
2353 mem_map = NODE_DATA(0)->node_mem_map;
2354 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2355 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
2356 mem_map -= pgdat->node_start_pfn;
2357 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
2360 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
2363 void __meminit free_area_init_node(int nid, struct pglist_data *pgdat,
2364 unsigned long *zones_size, unsigned long node_start_pfn,
2365 unsigned long *zholes_size)
2367 pgdat->node_id = nid;
2368 pgdat->node_start_pfn = node_start_pfn;
2369 calculate_node_totalpages(pgdat, zones_size, zholes_size);
2371 alloc_node_mem_map(pgdat);
2373 free_area_init_core(pgdat, zones_size, zholes_size);
2376 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2378 * add_active_range - Register a range of PFNs backed by physical memory
2379 * @nid: The node ID the range resides on
2380 * @start_pfn: The start PFN of the available physical memory
2381 * @end_pfn: The end PFN of the available physical memory
2383 * These ranges are stored in an early_node_map[] and later used by
2384 * free_area_init_nodes() to calculate zone sizes and holes. If the
2385 * range spans a memory hole, it is up to the architecture to ensure
2386 * the memory is not freed by the bootmem allocator. If possible
2387 * the range being registered will be merged with existing ranges.
2389 void __init add_active_range(unsigned int nid, unsigned long start_pfn,
2390 unsigned long end_pfn)
2394 printk(KERN_DEBUG "Entering add_active_range(%d, %lu, %lu) "
2395 "%d entries of %d used\n",
2396 nid, start_pfn, end_pfn,
2397 nr_nodemap_entries, MAX_ACTIVE_REGIONS);
2399 /* Merge with existing active regions if possible */
2400 for (i = 0; i < nr_nodemap_entries; i++) {
2401 if (early_node_map[i].nid != nid)
2404 /* Skip if an existing region covers this new one */
2405 if (start_pfn >= early_node_map[i].start_pfn &&
2406 end_pfn <= early_node_map[i].end_pfn)
2409 /* Merge forward if suitable */
2410 if (start_pfn <= early_node_map[i].end_pfn &&
2411 end_pfn > early_node_map[i].end_pfn) {
2412 early_node_map[i].end_pfn = end_pfn;
2416 /* Merge backward if suitable */
2417 if (start_pfn < early_node_map[i].end_pfn &&
2418 end_pfn >= early_node_map[i].start_pfn) {
2419 early_node_map[i].start_pfn = start_pfn;
2424 /* Check that early_node_map is large enough */
2425 if (i >= MAX_ACTIVE_REGIONS) {
2426 printk(KERN_CRIT "More than %d memory regions, truncating\n",
2427 MAX_ACTIVE_REGIONS);
2431 early_node_map[i].nid = nid;
2432 early_node_map[i].start_pfn = start_pfn;
2433 early_node_map[i].end_pfn = end_pfn;
2434 nr_nodemap_entries = i + 1;
2438 * shrink_active_range - Shrink an existing registered range of PFNs
2439 * @nid: The node id the range is on that should be shrunk
2440 * @old_end_pfn: The old end PFN of the range
2441 * @new_end_pfn: The new PFN of the range
2443 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
2444 * The map is kept at the end physical page range that has already been
2445 * registered with add_active_range(). This function allows an arch to shrink
2446 * an existing registered range.
2448 void __init shrink_active_range(unsigned int nid, unsigned long old_end_pfn,
2449 unsigned long new_end_pfn)
2453 /* Find the old active region end and shrink */
2454 for_each_active_range_index_in_nid(i, nid)
2455 if (early_node_map[i].end_pfn == old_end_pfn) {
2456 early_node_map[i].end_pfn = new_end_pfn;
2462 * remove_all_active_ranges - Remove all currently registered regions
2463 * During discovery, it may be found that a table like SRAT is invalid
2464 * and an alternative discovery method must be used. This function removes
2465 * all currently registered regions.
2467 void __init remove_all_active_ranges()
2469 memset(early_node_map, 0, sizeof(early_node_map));
2470 nr_nodemap_entries = 0;
2473 /* Compare two active node_active_regions */
2474 static int __init cmp_node_active_region(const void *a, const void *b)
2476 struct node_active_region *arange = (struct node_active_region *)a;
2477 struct node_active_region *brange = (struct node_active_region *)b;
2479 /* Done this way to avoid overflows */
2480 if (arange->start_pfn > brange->start_pfn)
2482 if (arange->start_pfn < brange->start_pfn)
2488 /* sort the node_map by start_pfn */
2489 static void __init sort_node_map(void)
2491 sort(early_node_map, (size_t)nr_nodemap_entries,
2492 sizeof(struct node_active_region),
2493 cmp_node_active_region, NULL);
2496 /* Find the lowest pfn for a node. This depends on a sorted early_node_map */
2497 unsigned long __init find_min_pfn_for_node(unsigned long nid)
2501 /* Assuming a sorted map, the first range found has the starting pfn */
2502 for_each_active_range_index_in_nid(i, nid)
2503 return early_node_map[i].start_pfn;
2505 printk(KERN_WARNING "Could not find start_pfn for node %lu\n", nid);
2510 * find_min_pfn_with_active_regions - Find the minimum PFN registered
2512 * It returns the minimum PFN based on information provided via
2513 * add_active_range()
2515 unsigned long __init find_min_pfn_with_active_regions(void)
2517 return find_min_pfn_for_node(MAX_NUMNODES);
2521 * find_max_pfn_with_active_regions - Find the maximum PFN registered
2523 * It returns the maximum PFN based on information provided via
2524 * add_active_range()
2526 unsigned long __init find_max_pfn_with_active_regions(void)
2529 unsigned long max_pfn = 0;
2531 for (i = 0; i < nr_nodemap_entries; i++)
2532 max_pfn = max(max_pfn, early_node_map[i].end_pfn);
2538 * free_area_init_nodes - Initialise all pg_data_t and zone data
2539 * @arch_max_dma_pfn: The maximum PFN usable for ZONE_DMA
2540 * @arch_max_dma32_pfn: The maximum PFN usable for ZONE_DMA32
2541 * @arch_max_low_pfn: The maximum PFN usable for ZONE_NORMAL
2542 * @arch_max_high_pfn: The maximum PFN usable for ZONE_HIGHMEM
2544 * This will call free_area_init_node() for each active node in the system.
2545 * Using the page ranges provided by add_active_range(), the size of each
2546 * zone in each node and their holes is calculated. If the maximum PFN
2547 * between two adjacent zones match, it is assumed that the zone is empty.
2548 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
2549 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
2550 * starts where the previous one ended. For example, ZONE_DMA32 starts
2551 * at arch_max_dma_pfn.
2553 void __init free_area_init_nodes(unsigned long *max_zone_pfn)
2558 /* Record where the zone boundaries are */
2559 memset(arch_zone_lowest_possible_pfn, 0,
2560 sizeof(arch_zone_lowest_possible_pfn));
2561 memset(arch_zone_highest_possible_pfn, 0,
2562 sizeof(arch_zone_highest_possible_pfn));
2563 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
2564 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
2565 for (i = 1; i < MAX_NR_ZONES; i++) {
2566 arch_zone_lowest_possible_pfn[i] =
2567 arch_zone_highest_possible_pfn[i-1];
2568 arch_zone_highest_possible_pfn[i] =
2569 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
2572 /* Regions in the early_node_map can be in any order */
2575 /* Print out the zone ranges */
2576 printk("Zone PFN ranges:\n");
2577 for (i = 0; i < MAX_NR_ZONES; i++)
2578 printk(" %-8s %8lu -> %8lu\n",
2580 arch_zone_lowest_possible_pfn[i],
2581 arch_zone_highest_possible_pfn[i]);
2583 /* Print out the early_node_map[] */
2584 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries);
2585 for (i = 0; i < nr_nodemap_entries; i++)
2586 printk(" %3d: %8lu -> %8lu\n", early_node_map[i].nid,
2587 early_node_map[i].start_pfn,
2588 early_node_map[i].end_pfn);
2590 /* Initialise every node */
2591 for_each_online_node(nid) {
2592 pg_data_t *pgdat = NODE_DATA(nid);
2593 free_area_init_node(nid, pgdat, NULL,
2594 find_min_pfn_for_node(nid), NULL);
2597 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
2599 #ifndef CONFIG_NEED_MULTIPLE_NODES
2600 static bootmem_data_t contig_bootmem_data;
2601 struct pglist_data contig_page_data = { .bdata = &contig_bootmem_data };
2603 EXPORT_SYMBOL(contig_page_data);
2606 void __init free_area_init(unsigned long *zones_size)
2608 free_area_init_node(0, NODE_DATA(0), zones_size,
2609 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
2612 #ifdef CONFIG_HOTPLUG_CPU
2613 static int page_alloc_cpu_notify(struct notifier_block *self,
2614 unsigned long action, void *hcpu)
2616 int cpu = (unsigned long)hcpu;
2618 if (action == CPU_DEAD) {
2619 local_irq_disable();
2621 vm_events_fold_cpu(cpu);
2623 refresh_cpu_vm_stats(cpu);
2627 #endif /* CONFIG_HOTPLUG_CPU */
2629 void __init page_alloc_init(void)
2631 hotcpu_notifier(page_alloc_cpu_notify, 0);
2635 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
2636 * or min_free_kbytes changes.
2638 static void calculate_totalreserve_pages(void)
2640 struct pglist_data *pgdat;
2641 unsigned long reserve_pages = 0;
2642 enum zone_type i, j;
2644 for_each_online_pgdat(pgdat) {
2645 for (i = 0; i < MAX_NR_ZONES; i++) {
2646 struct zone *zone = pgdat->node_zones + i;
2647 unsigned long max = 0;
2649 /* Find valid and maximum lowmem_reserve in the zone */
2650 for (j = i; j < MAX_NR_ZONES; j++) {
2651 if (zone->lowmem_reserve[j] > max)
2652 max = zone->lowmem_reserve[j];
2655 /* we treat pages_high as reserved pages. */
2656 max += zone->pages_high;
2658 if (max > zone->present_pages)
2659 max = zone->present_pages;
2660 reserve_pages += max;
2663 totalreserve_pages = reserve_pages;
2667 * setup_per_zone_lowmem_reserve - called whenever
2668 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
2669 * has a correct pages reserved value, so an adequate number of
2670 * pages are left in the zone after a successful __alloc_pages().
2672 static void setup_per_zone_lowmem_reserve(void)
2674 struct pglist_data *pgdat;
2675 enum zone_type j, idx;
2677 for_each_online_pgdat(pgdat) {
2678 for (j = 0; j < MAX_NR_ZONES; j++) {
2679 struct zone *zone = pgdat->node_zones + j;
2680 unsigned long present_pages = zone->present_pages;
2682 zone->lowmem_reserve[j] = 0;
2686 struct zone *lower_zone;
2690 if (sysctl_lowmem_reserve_ratio[idx] < 1)
2691 sysctl_lowmem_reserve_ratio[idx] = 1;
2693 lower_zone = pgdat->node_zones + idx;
2694 lower_zone->lowmem_reserve[j] = present_pages /
2695 sysctl_lowmem_reserve_ratio[idx];
2696 present_pages += lower_zone->present_pages;
2701 /* update totalreserve_pages */
2702 calculate_totalreserve_pages();
2706 * setup_per_zone_pages_min - called when min_free_kbytes changes. Ensures
2707 * that the pages_{min,low,high} values for each zone are set correctly
2708 * with respect to min_free_kbytes.
2710 void setup_per_zone_pages_min(void)
2712 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
2713 unsigned long lowmem_pages = 0;
2715 unsigned long flags;
2717 /* Calculate total number of !ZONE_HIGHMEM pages */
2718 for_each_zone(zone) {
2719 if (!is_highmem(zone))
2720 lowmem_pages += zone->present_pages;
2723 for_each_zone(zone) {
2726 spin_lock_irqsave(&zone->lru_lock, flags);
2727 tmp = (u64)pages_min * zone->present_pages;
2728 do_div(tmp, lowmem_pages);
2729 if (is_highmem(zone)) {
2731 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
2732 * need highmem pages, so cap pages_min to a small
2735 * The (pages_high-pages_low) and (pages_low-pages_min)
2736 * deltas controls asynch page reclaim, and so should
2737 * not be capped for highmem.
2741 min_pages = zone->present_pages / 1024;
2742 if (min_pages < SWAP_CLUSTER_MAX)
2743 min_pages = SWAP_CLUSTER_MAX;
2744 if (min_pages > 128)
2746 zone->pages_min = min_pages;
2749 * If it's a lowmem zone, reserve a number of pages
2750 * proportionate to the zone's size.
2752 zone->pages_min = tmp;
2755 zone->pages_low = zone->pages_min + (tmp >> 2);
2756 zone->pages_high = zone->pages_min + (tmp >> 1);
2757 spin_unlock_irqrestore(&zone->lru_lock, flags);
2760 /* update totalreserve_pages */
2761 calculate_totalreserve_pages();
2765 * Initialise min_free_kbytes.
2767 * For small machines we want it small (128k min). For large machines
2768 * we want it large (64MB max). But it is not linear, because network
2769 * bandwidth does not increase linearly with machine size. We use
2771 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
2772 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
2788 static int __init init_per_zone_pages_min(void)
2790 unsigned long lowmem_kbytes;
2792 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
2794 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
2795 if (min_free_kbytes < 128)
2796 min_free_kbytes = 128;
2797 if (min_free_kbytes > 65536)
2798 min_free_kbytes = 65536;
2799 setup_per_zone_pages_min();
2800 setup_per_zone_lowmem_reserve();
2803 module_init(init_per_zone_pages_min)
2806 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
2807 * that we can call two helper functions whenever min_free_kbytes
2810 int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
2811 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2813 proc_dointvec(table, write, file, buffer, length, ppos);
2814 setup_per_zone_pages_min();
2819 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
2820 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2825 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
2830 zone->min_unmapped_pages = (zone->present_pages *
2831 sysctl_min_unmapped_ratio) / 100;
2835 int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
2836 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2841 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
2846 zone->min_slab_pages = (zone->present_pages *
2847 sysctl_min_slab_ratio) / 100;
2853 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
2854 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
2855 * whenever sysctl_lowmem_reserve_ratio changes.
2857 * The reserve ratio obviously has absolutely no relation with the
2858 * pages_min watermarks. The lowmem reserve ratio can only make sense
2859 * if in function of the boot time zone sizes.
2861 int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
2862 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2864 proc_dointvec_minmax(table, write, file, buffer, length, ppos);
2865 setup_per_zone_lowmem_reserve();
2870 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
2871 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
2872 * can have before it gets flushed back to buddy allocator.
2875 int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
2876 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2882 ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
2883 if (!write || (ret == -EINVAL))
2885 for_each_zone(zone) {
2886 for_each_online_cpu(cpu) {
2888 high = zone->present_pages / percpu_pagelist_fraction;
2889 setup_pagelist_highmark(zone_pcp(zone, cpu), high);
2895 int hashdist = HASHDIST_DEFAULT;
2898 static int __init set_hashdist(char *str)
2902 hashdist = simple_strtoul(str, &str, 0);
2905 __setup("hashdist=", set_hashdist);
2909 * allocate a large system hash table from bootmem
2910 * - it is assumed that the hash table must contain an exact power-of-2
2911 * quantity of entries
2912 * - limit is the number of hash buckets, not the total allocation size
2914 void *__init alloc_large_system_hash(const char *tablename,
2915 unsigned long bucketsize,
2916 unsigned long numentries,
2919 unsigned int *_hash_shift,
2920 unsigned int *_hash_mask,
2921 unsigned long limit)
2923 unsigned long long max = limit;
2924 unsigned long log2qty, size;
2927 /* allow the kernel cmdline to have a say */
2929 /* round applicable memory size up to nearest megabyte */
2930 numentries = (flags & HASH_HIGHMEM) ? nr_all_pages : nr_kernel_pages;
2931 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
2932 numentries >>= 20 - PAGE_SHIFT;
2933 numentries <<= 20 - PAGE_SHIFT;
2935 /* limit to 1 bucket per 2^scale bytes of low memory */
2936 if (scale > PAGE_SHIFT)
2937 numentries >>= (scale - PAGE_SHIFT);
2939 numentries <<= (PAGE_SHIFT - scale);
2941 numentries = roundup_pow_of_two(numentries);
2943 /* limit allocation size to 1/16 total memory by default */
2945 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
2946 do_div(max, bucketsize);
2949 if (numentries > max)
2952 log2qty = long_log2(numentries);
2955 size = bucketsize << log2qty;
2956 if (flags & HASH_EARLY)
2957 table = alloc_bootmem(size);
2959 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
2961 unsigned long order;
2962 for (order = 0; ((1UL << order) << PAGE_SHIFT) < size; order++)
2964 table = (void*) __get_free_pages(GFP_ATOMIC, order);
2966 } while (!table && size > PAGE_SIZE && --log2qty);
2969 panic("Failed to allocate %s hash table\n", tablename);
2971 printk("%s hash table entries: %d (order: %d, %lu bytes)\n",
2974 long_log2(size) - PAGE_SHIFT,
2978 *_hash_shift = log2qty;
2980 *_hash_mask = (1 << log2qty) - 1;
2985 #ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE
2986 struct page *pfn_to_page(unsigned long pfn)
2988 return __pfn_to_page(pfn);
2990 unsigned long page_to_pfn(struct page *page)
2992 return __page_to_pfn(page);
2994 EXPORT_SYMBOL(pfn_to_page);
2995 EXPORT_SYMBOL(page_to_pfn);
2996 #endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */