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>
44 #include <asm/tlbflush.h>
45 #include <asm/div64.h>
49 * MCD - HACK: Find somewhere to initialize this EARLY, or make this
52 nodemask_t node_online_map __read_mostly = { { [0] = 1UL } };
53 EXPORT_SYMBOL(node_online_map);
54 nodemask_t node_possible_map __read_mostly = NODE_MASK_ALL;
55 EXPORT_SYMBOL(node_possible_map);
56 unsigned long totalram_pages __read_mostly;
57 unsigned long totalreserve_pages __read_mostly;
59 int percpu_pagelist_fraction;
61 static void __free_pages_ok(struct page *page, unsigned int order);
64 * results with 256, 32 in the lowmem_reserve sysctl:
65 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
66 * 1G machine -> (16M dma, 784M normal, 224M high)
67 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
68 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
69 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
71 * TBD: should special case ZONE_DMA32 machines here - in those we normally
72 * don't need any ZONE_NORMAL reservation
74 int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
76 #ifdef CONFIG_ZONE_DMA32
84 EXPORT_SYMBOL(totalram_pages);
86 static char *zone_names[MAX_NR_ZONES] = {
88 #ifdef CONFIG_ZONE_DMA32
97 int min_free_kbytes = 1024;
99 unsigned long __meminitdata nr_kernel_pages;
100 unsigned long __meminitdata nr_all_pages;
101 static unsigned long __initdata dma_reserve;
103 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
105 * MAX_ACTIVE_REGIONS determines the maxmimum number of distinct
106 * ranges of memory (RAM) that may be registered with add_active_range().
107 * Ranges passed to add_active_range() will be merged if possible
108 * so the number of times add_active_range() can be called is
109 * related to the number of nodes and the number of holes
111 #ifdef CONFIG_MAX_ACTIVE_REGIONS
112 /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
113 #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
115 #if MAX_NUMNODES >= 32
116 /* If there can be many nodes, allow up to 50 holes per node */
117 #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
119 /* By default, allow up to 256 distinct regions */
120 #define MAX_ACTIVE_REGIONS 256
124 struct node_active_region __initdata early_node_map[MAX_ACTIVE_REGIONS];
125 int __initdata nr_nodemap_entries;
126 unsigned long __initdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
127 unsigned long __initdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
128 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
129 unsigned long __initdata node_boundary_start_pfn[MAX_NUMNODES];
130 unsigned long __initdata node_boundary_end_pfn[MAX_NUMNODES];
131 #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
132 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
134 #ifdef CONFIG_DEBUG_VM
135 static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
139 unsigned long pfn = page_to_pfn(page);
142 seq = zone_span_seqbegin(zone);
143 if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
145 else if (pfn < zone->zone_start_pfn)
147 } while (zone_span_seqretry(zone, seq));
152 static int page_is_consistent(struct zone *zone, struct page *page)
154 #ifdef CONFIG_HOLES_IN_ZONE
155 if (!pfn_valid(page_to_pfn(page)))
158 if (zone != page_zone(page))
164 * Temporary debugging check for pages not lying within a given zone.
166 static int bad_range(struct zone *zone, struct page *page)
168 if (page_outside_zone_boundaries(zone, page))
170 if (!page_is_consistent(zone, page))
176 static inline int bad_range(struct zone *zone, struct page *page)
182 static void bad_page(struct page *page)
184 printk(KERN_EMERG "Bad page state in process '%s'\n"
185 KERN_EMERG "page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n"
186 KERN_EMERG "Trying to fix it up, but a reboot is needed\n"
187 KERN_EMERG "Backtrace:\n",
188 current->comm, page, (int)(2*sizeof(unsigned long)),
189 (unsigned long)page->flags, page->mapping,
190 page_mapcount(page), page_count(page));
192 page->flags &= ~(1 << PG_lru |
202 set_page_count(page, 0);
203 reset_page_mapcount(page);
204 page->mapping = NULL;
205 add_taint(TAINT_BAD_PAGE);
209 * Higher-order pages are called "compound pages". They are structured thusly:
211 * The first PAGE_SIZE page is called the "head page".
213 * The remaining PAGE_SIZE pages are called "tail pages".
215 * All pages have PG_compound set. All pages have their ->private pointing at
216 * the head page (even the head page has this).
218 * The first tail page's ->lru.next holds the address of the compound page's
219 * put_page() function. Its ->lru.prev holds the order of allocation.
220 * This usage means that zero-order pages may not be compound.
223 static void free_compound_page(struct page *page)
225 __free_pages_ok(page, (unsigned long)page[1].lru.prev);
228 static void prep_compound_page(struct page *page, unsigned long order)
231 int nr_pages = 1 << order;
233 page[1].lru.next = (void *)free_compound_page; /* set dtor */
234 page[1].lru.prev = (void *)order;
235 for (i = 0; i < nr_pages; i++) {
236 struct page *p = page + i;
238 __SetPageCompound(p);
239 set_page_private(p, (unsigned long)page);
243 static void destroy_compound_page(struct page *page, unsigned long order)
246 int nr_pages = 1 << order;
248 if (unlikely((unsigned long)page[1].lru.prev != order))
251 for (i = 0; i < nr_pages; i++) {
252 struct page *p = page + i;
254 if (unlikely(!PageCompound(p) |
255 (page_private(p) != (unsigned long)page)))
257 __ClearPageCompound(p);
261 static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
265 VM_BUG_ON((gfp_flags & (__GFP_WAIT | __GFP_HIGHMEM)) == __GFP_HIGHMEM);
267 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
268 * and __GFP_HIGHMEM from hard or soft interrupt context.
270 VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
271 for (i = 0; i < (1 << order); i++)
272 clear_highpage(page + i);
276 * function for dealing with page's order in buddy system.
277 * zone->lock is already acquired when we use these.
278 * So, we don't need atomic page->flags operations here.
280 static inline unsigned long page_order(struct page *page)
282 return page_private(page);
285 static inline void set_page_order(struct page *page, int order)
287 set_page_private(page, order);
288 __SetPageBuddy(page);
291 static inline void rmv_page_order(struct page *page)
293 __ClearPageBuddy(page);
294 set_page_private(page, 0);
298 * Locate the struct page for both the matching buddy in our
299 * pair (buddy1) and the combined O(n+1) page they form (page).
301 * 1) Any buddy B1 will have an order O twin B2 which satisfies
302 * the following equation:
304 * For example, if the starting buddy (buddy2) is #8 its order
306 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
308 * 2) Any buddy B will have an order O+1 parent P which
309 * satisfies the following equation:
312 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
314 static inline struct page *
315 __page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order)
317 unsigned long buddy_idx = page_idx ^ (1 << order);
319 return page + (buddy_idx - page_idx);
322 static inline unsigned long
323 __find_combined_index(unsigned long page_idx, unsigned int order)
325 return (page_idx & ~(1 << order));
329 * This function checks whether a page is free && is the buddy
330 * we can do coalesce a page and its buddy if
331 * (a) the buddy is not in a hole &&
332 * (b) the buddy is in the buddy system &&
333 * (c) a page and its buddy have the same order &&
334 * (d) a page and its buddy are in the same zone.
336 * For recording whether a page is in the buddy system, we use PG_buddy.
337 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
339 * For recording page's order, we use page_private(page).
341 static inline int page_is_buddy(struct page *page, struct page *buddy,
344 #ifdef CONFIG_HOLES_IN_ZONE
345 if (!pfn_valid(page_to_pfn(buddy)))
349 if (page_zone_id(page) != page_zone_id(buddy))
352 if (PageBuddy(buddy) && page_order(buddy) == order) {
353 BUG_ON(page_count(buddy) != 0);
360 * Freeing function for a buddy system allocator.
362 * The concept of a buddy system is to maintain direct-mapped table
363 * (containing bit values) for memory blocks of various "orders".
364 * The bottom level table contains the map for the smallest allocatable
365 * units of memory (here, pages), and each level above it describes
366 * pairs of units from the levels below, hence, "buddies".
367 * At a high level, all that happens here is marking the table entry
368 * at the bottom level available, and propagating the changes upward
369 * as necessary, plus some accounting needed to play nicely with other
370 * parts of the VM system.
371 * At each level, we keep a list of pages, which are heads of continuous
372 * free pages of length of (1 << order) and marked with PG_buddy. Page's
373 * order is recorded in page_private(page) field.
374 * So when we are allocating or freeing one, we can derive the state of the
375 * other. That is, if we allocate a small block, and both were
376 * free, the remainder of the region must be split into blocks.
377 * If a block is freed, and its buddy is also free, then this
378 * triggers coalescing into a block of larger size.
383 static inline void __free_one_page(struct page *page,
384 struct zone *zone, unsigned int order)
386 unsigned long page_idx;
387 int order_size = 1 << order;
389 if (unlikely(PageCompound(page)))
390 destroy_compound_page(page, order);
392 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
394 VM_BUG_ON(page_idx & (order_size - 1));
395 VM_BUG_ON(bad_range(zone, page));
397 zone->free_pages += order_size;
398 while (order < MAX_ORDER-1) {
399 unsigned long combined_idx;
400 struct free_area *area;
403 buddy = __page_find_buddy(page, page_idx, order);
404 if (!page_is_buddy(page, buddy, order))
405 break; /* Move the buddy up one level. */
407 list_del(&buddy->lru);
408 area = zone->free_area + order;
410 rmv_page_order(buddy);
411 combined_idx = __find_combined_index(page_idx, order);
412 page = page + (combined_idx - page_idx);
413 page_idx = combined_idx;
416 set_page_order(page, order);
417 list_add(&page->lru, &zone->free_area[order].free_list);
418 zone->free_area[order].nr_free++;
421 static inline int free_pages_check(struct page *page)
423 if (unlikely(page_mapcount(page) |
424 (page->mapping != NULL) |
425 (page_count(page) != 0) |
439 __ClearPageDirty(page);
441 * For now, we report if PG_reserved was found set, but do not
442 * clear it, and do not free the page. But we shall soon need
443 * to do more, for when the ZERO_PAGE count wraps negative.
445 return PageReserved(page);
449 * Frees a list of pages.
450 * Assumes all pages on list are in same zone, and of same order.
451 * count is the number of pages to free.
453 * If the zone was previously in an "all pages pinned" state then look to
454 * see if this freeing clears that state.
456 * And clear the zone's pages_scanned counter, to hold off the "all pages are
457 * pinned" detection logic.
459 static void free_pages_bulk(struct zone *zone, int count,
460 struct list_head *list, int order)
462 spin_lock(&zone->lock);
463 zone->all_unreclaimable = 0;
464 zone->pages_scanned = 0;
468 VM_BUG_ON(list_empty(list));
469 page = list_entry(list->prev, struct page, lru);
470 /* have to delete it as __free_one_page list manipulates */
471 list_del(&page->lru);
472 __free_one_page(page, zone, order);
474 spin_unlock(&zone->lock);
477 static void free_one_page(struct zone *zone, struct page *page, int order)
479 spin_lock(&zone->lock);
480 zone->all_unreclaimable = 0;
481 zone->pages_scanned = 0;
482 __free_one_page(page, zone, order);
483 spin_unlock(&zone->lock);
486 static void __free_pages_ok(struct page *page, unsigned int order)
492 for (i = 0 ; i < (1 << order) ; ++i)
493 reserved += free_pages_check(page + i);
497 if (!PageHighMem(page))
498 debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order);
499 arch_free_page(page, order);
500 kernel_map_pages(page, 1 << order, 0);
502 local_irq_save(flags);
503 __count_vm_events(PGFREE, 1 << order);
504 free_one_page(page_zone(page), page, order);
505 local_irq_restore(flags);
509 * permit the bootmem allocator to evade page validation on high-order frees
511 void fastcall __init __free_pages_bootmem(struct page *page, unsigned int order)
514 __ClearPageReserved(page);
515 set_page_count(page, 0);
516 set_page_refcounted(page);
522 for (loop = 0; loop < BITS_PER_LONG; loop++) {
523 struct page *p = &page[loop];
525 if (loop + 1 < BITS_PER_LONG)
527 __ClearPageReserved(p);
528 set_page_count(p, 0);
531 set_page_refcounted(page);
532 __free_pages(page, order);
538 * The order of subdivision here is critical for the IO subsystem.
539 * Please do not alter this order without good reasons and regression
540 * testing. Specifically, as large blocks of memory are subdivided,
541 * the order in which smaller blocks are delivered depends on the order
542 * they're subdivided in this function. This is the primary factor
543 * influencing the order in which pages are delivered to the IO
544 * subsystem according to empirical testing, and this is also justified
545 * by considering the behavior of a buddy system containing a single
546 * large block of memory acted on by a series of small allocations.
547 * This behavior is a critical factor in sglist merging's success.
551 static inline void expand(struct zone *zone, struct page *page,
552 int low, int high, struct free_area *area)
554 unsigned long size = 1 << high;
560 VM_BUG_ON(bad_range(zone, &page[size]));
561 list_add(&page[size].lru, &area->free_list);
563 set_page_order(&page[size], high);
568 * This page is about to be returned from the page allocator
570 static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
572 if (unlikely(page_mapcount(page) |
573 (page->mapping != NULL) |
574 (page_count(page) != 0) |
590 * For now, we report if PG_reserved was found set, but do not
591 * clear it, and do not allocate the page: as a safety net.
593 if (PageReserved(page))
596 page->flags &= ~(1 << PG_uptodate | 1 << PG_error |
597 1 << PG_referenced | 1 << PG_arch_1 |
598 1 << PG_checked | 1 << PG_mappedtodisk);
599 set_page_private(page, 0);
600 set_page_refcounted(page);
601 kernel_map_pages(page, 1 << order, 1);
603 if (gfp_flags & __GFP_ZERO)
604 prep_zero_page(page, order, gfp_flags);
606 if (order && (gfp_flags & __GFP_COMP))
607 prep_compound_page(page, order);
613 * Do the hard work of removing an element from the buddy allocator.
614 * Call me with the zone->lock already held.
616 static struct page *__rmqueue(struct zone *zone, unsigned int order)
618 struct free_area * area;
619 unsigned int current_order;
622 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
623 area = zone->free_area + current_order;
624 if (list_empty(&area->free_list))
627 page = list_entry(area->free_list.next, struct page, lru);
628 list_del(&page->lru);
629 rmv_page_order(page);
631 zone->free_pages -= 1UL << order;
632 expand(zone, page, order, current_order, area);
640 * Obtain a specified number of elements from the buddy allocator, all under
641 * a single hold of the lock, for efficiency. Add them to the supplied list.
642 * Returns the number of new pages which were placed at *list.
644 static int rmqueue_bulk(struct zone *zone, unsigned int order,
645 unsigned long count, struct list_head *list)
649 spin_lock(&zone->lock);
650 for (i = 0; i < count; ++i) {
651 struct page *page = __rmqueue(zone, order);
652 if (unlikely(page == NULL))
654 list_add_tail(&page->lru, list);
656 spin_unlock(&zone->lock);
662 * Called from the slab reaper to drain pagesets on a particular node that
663 * belongs to the currently executing processor.
664 * Note that this function must be called with the thread pinned to
665 * a single processor.
667 void drain_node_pages(int nodeid)
673 for (z = 0; z < MAX_NR_ZONES; z++) {
674 struct zone *zone = NODE_DATA(nodeid)->node_zones + z;
675 struct per_cpu_pageset *pset;
677 if (!populated_zone(zone))
680 pset = zone_pcp(zone, smp_processor_id());
681 for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) {
682 struct per_cpu_pages *pcp;
686 local_irq_save(flags);
687 free_pages_bulk(zone, pcp->count, &pcp->list, 0);
689 local_irq_restore(flags);
696 #if defined(CONFIG_PM) || defined(CONFIG_HOTPLUG_CPU)
697 static void __drain_pages(unsigned int cpu)
703 for_each_zone(zone) {
704 struct per_cpu_pageset *pset;
706 pset = zone_pcp(zone, cpu);
707 for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) {
708 struct per_cpu_pages *pcp;
711 local_irq_save(flags);
712 free_pages_bulk(zone, pcp->count, &pcp->list, 0);
714 local_irq_restore(flags);
718 #endif /* CONFIG_PM || CONFIG_HOTPLUG_CPU */
722 void mark_free_pages(struct zone *zone)
724 unsigned long pfn, max_zone_pfn;
727 struct list_head *curr;
729 if (!zone->spanned_pages)
732 spin_lock_irqsave(&zone->lock, flags);
734 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
735 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
736 if (pfn_valid(pfn)) {
737 struct page *page = pfn_to_page(pfn);
739 if (!PageNosave(page))
740 ClearPageNosaveFree(page);
743 for (order = MAX_ORDER - 1; order >= 0; --order)
744 list_for_each(curr, &zone->free_area[order].free_list) {
747 pfn = page_to_pfn(list_entry(curr, struct page, lru));
748 for (i = 0; i < (1UL << order); i++)
749 SetPageNosaveFree(pfn_to_page(pfn + i));
752 spin_unlock_irqrestore(&zone->lock, flags);
756 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
758 void drain_local_pages(void)
762 local_irq_save(flags);
763 __drain_pages(smp_processor_id());
764 local_irq_restore(flags);
766 #endif /* CONFIG_PM */
769 * Free a 0-order page
771 static void fastcall free_hot_cold_page(struct page *page, int cold)
773 struct zone *zone = page_zone(page);
774 struct per_cpu_pages *pcp;
778 page->mapping = NULL;
779 if (free_pages_check(page))
782 if (!PageHighMem(page))
783 debug_check_no_locks_freed(page_address(page), PAGE_SIZE);
784 arch_free_page(page, 0);
785 kernel_map_pages(page, 1, 0);
787 pcp = &zone_pcp(zone, get_cpu())->pcp[cold];
788 local_irq_save(flags);
789 __count_vm_event(PGFREE);
790 list_add(&page->lru, &pcp->list);
792 if (pcp->count >= pcp->high) {
793 free_pages_bulk(zone, pcp->batch, &pcp->list, 0);
794 pcp->count -= pcp->batch;
796 local_irq_restore(flags);
800 void fastcall free_hot_page(struct page *page)
802 free_hot_cold_page(page, 0);
805 void fastcall free_cold_page(struct page *page)
807 free_hot_cold_page(page, 1);
811 * split_page takes a non-compound higher-order page, and splits it into
812 * n (1<<order) sub-pages: page[0..n]
813 * Each sub-page must be freed individually.
815 * Note: this is probably too low level an operation for use in drivers.
816 * Please consult with lkml before using this in your driver.
818 void split_page(struct page *page, unsigned int order)
822 VM_BUG_ON(PageCompound(page));
823 VM_BUG_ON(!page_count(page));
824 for (i = 1; i < (1 << order); i++)
825 set_page_refcounted(page + i);
829 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
830 * we cheat by calling it from here, in the order > 0 path. Saves a branch
833 static struct page *buffered_rmqueue(struct zonelist *zonelist,
834 struct zone *zone, int order, gfp_t gfp_flags)
838 int cold = !!(gfp_flags & __GFP_COLD);
843 if (likely(order == 0)) {
844 struct per_cpu_pages *pcp;
846 pcp = &zone_pcp(zone, cpu)->pcp[cold];
847 local_irq_save(flags);
849 pcp->count = rmqueue_bulk(zone, 0,
850 pcp->batch, &pcp->list);
851 if (unlikely(!pcp->count))
854 page = list_entry(pcp->list.next, struct page, lru);
855 list_del(&page->lru);
858 spin_lock_irqsave(&zone->lock, flags);
859 page = __rmqueue(zone, order);
860 spin_unlock(&zone->lock);
865 __count_zone_vm_events(PGALLOC, zone, 1 << order);
866 zone_statistics(zonelist, zone);
867 local_irq_restore(flags);
870 VM_BUG_ON(bad_range(zone, page));
871 if (prep_new_page(page, order, gfp_flags))
876 local_irq_restore(flags);
881 #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
882 #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
883 #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
884 #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
885 #define ALLOC_HARDER 0x10 /* try to alloc harder */
886 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
887 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
890 * Return 1 if free pages are above 'mark'. This takes into account the order
893 int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
894 int classzone_idx, int alloc_flags)
896 /* free_pages my go negative - that's OK */
897 unsigned long min = mark;
898 long 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)
923 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
924 * skip over zones that are not allowed by the cpuset, or that have
925 * been recently (in last second) found to be nearly full. See further
926 * comments in mmzone.h. Reduces cache footprint of zonelist scans
927 * that have to skip over alot of full or unallowed zones.
929 * If the zonelist cache is present in the passed in zonelist, then
930 * returns a pointer to the allowed node mask (either the current
931 * tasks mems_allowed, or node_online_map.)
933 * If the zonelist cache is not available for this zonelist, does
934 * nothing and returns NULL.
936 * If the fullzones BITMAP in the zonelist cache is stale (more than
937 * a second since last zap'd) then we zap it out (clear its bits.)
939 * We hold off even calling zlc_setup, until after we've checked the
940 * first zone in the zonelist, on the theory that most allocations will
941 * be satisfied from that first zone, so best to examine that zone as
944 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
946 struct zonelist_cache *zlc; /* cached zonelist speedup info */
947 nodemask_t *allowednodes; /* zonelist_cache approximation */
949 zlc = zonelist->zlcache_ptr;
953 if (jiffies - zlc->last_full_zap > 1 * HZ) {
954 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
955 zlc->last_full_zap = jiffies;
958 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
959 &cpuset_current_mems_allowed :
965 * Given 'z' scanning a zonelist, run a couple of quick checks to see
966 * if it is worth looking at further for free memory:
967 * 1) Check that the zone isn't thought to be full (doesn't have its
968 * bit set in the zonelist_cache fullzones BITMAP).
969 * 2) Check that the zones node (obtained from the zonelist_cache
970 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
971 * Return true (non-zero) if zone is worth looking at further, or
972 * else return false (zero) if it is not.
974 * This check -ignores- the distinction between various watermarks,
975 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
976 * found to be full for any variation of these watermarks, it will
977 * be considered full for up to one second by all requests, unless
978 * we are so low on memory on all allowed nodes that we are forced
979 * into the second scan of the zonelist.
981 * In the second scan we ignore this zonelist cache and exactly
982 * apply the watermarks to all zones, even it is slower to do so.
983 * We are low on memory in the second scan, and should leave no stone
984 * unturned looking for a free page.
986 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zone **z,
987 nodemask_t *allowednodes)
989 struct zonelist_cache *zlc; /* cached zonelist speedup info */
990 int i; /* index of *z in zonelist zones */
991 int n; /* node that zone *z is on */
993 zlc = zonelist->zlcache_ptr;
997 i = z - zonelist->zones;
1000 /* This zone is worth trying if it is allowed but not full */
1001 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1005 * Given 'z' scanning a zonelist, set the corresponding bit in
1006 * zlc->fullzones, so that subsequent attempts to allocate a page
1007 * from that zone don't waste time re-examining it.
1009 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zone **z)
1011 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1012 int i; /* index of *z in zonelist zones */
1014 zlc = zonelist->zlcache_ptr;
1018 i = z - zonelist->zones;
1020 set_bit(i, zlc->fullzones);
1023 #else /* CONFIG_NUMA */
1025 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1030 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zone **z,
1031 nodemask_t *allowednodes)
1036 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zone **z)
1039 #endif /* CONFIG_NUMA */
1042 * get_page_from_freelist goes through the zonelist trying to allocate
1045 static struct page *
1046 get_page_from_freelist(gfp_t gfp_mask, unsigned int order,
1047 struct zonelist *zonelist, int alloc_flags)
1050 struct page *page = NULL;
1051 int classzone_idx = zone_idx(zonelist->zones[0]);
1053 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1054 int zlc_active = 0; /* set if using zonelist_cache */
1055 int did_zlc_setup = 0; /* just call zlc_setup() one time */
1059 * Scan zonelist, looking for a zone with enough free.
1060 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1062 z = zonelist->zones;
1065 if (NUMA_BUILD && zlc_active &&
1066 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1069 if (unlikely(NUMA_BUILD && (gfp_mask & __GFP_THISNODE) &&
1070 zone->zone_pgdat != zonelist->zones[0]->zone_pgdat))
1072 if ((alloc_flags & ALLOC_CPUSET) &&
1073 !cpuset_zone_allowed(zone, gfp_mask))
1076 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
1078 if (alloc_flags & ALLOC_WMARK_MIN)
1079 mark = zone->pages_min;
1080 else if (alloc_flags & ALLOC_WMARK_LOW)
1081 mark = zone->pages_low;
1083 mark = zone->pages_high;
1084 if (!zone_watermark_ok(zone, order, mark,
1085 classzone_idx, alloc_flags)) {
1086 if (!zone_reclaim_mode ||
1087 !zone_reclaim(zone, gfp_mask, order))
1088 goto this_zone_full;
1092 page = buffered_rmqueue(zonelist, zone, order, gfp_mask);
1097 zlc_mark_zone_full(zonelist, z);
1099 if (NUMA_BUILD && !did_zlc_setup) {
1100 /* we do zlc_setup after the first zone is tried */
1101 allowednodes = zlc_setup(zonelist, alloc_flags);
1105 } while (*(++z) != NULL);
1107 if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) {
1108 /* Disable zlc cache for second zonelist scan */
1116 * This is the 'heart' of the zoned buddy allocator.
1118 struct page * fastcall
1119 __alloc_pages(gfp_t gfp_mask, unsigned int order,
1120 struct zonelist *zonelist)
1122 const gfp_t wait = gfp_mask & __GFP_WAIT;
1125 struct reclaim_state reclaim_state;
1126 struct task_struct *p = current;
1129 int did_some_progress;
1131 might_sleep_if(wait);
1134 z = zonelist->zones; /* the list of zones suitable for gfp_mask */
1136 if (unlikely(*z == NULL)) {
1137 /* Should this ever happen?? */
1141 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
1142 zonelist, ALLOC_WMARK_LOW|ALLOC_CPUSET);
1146 for (z = zonelist->zones; *z; z++)
1147 wakeup_kswapd(*z, order);
1150 * OK, we're below the kswapd watermark and have kicked background
1151 * reclaim. Now things get more complex, so set up alloc_flags according
1152 * to how we want to proceed.
1154 * The caller may dip into page reserves a bit more if the caller
1155 * cannot run direct reclaim, or if the caller has realtime scheduling
1156 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1157 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1159 alloc_flags = ALLOC_WMARK_MIN;
1160 if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait)
1161 alloc_flags |= ALLOC_HARDER;
1162 if (gfp_mask & __GFP_HIGH)
1163 alloc_flags |= ALLOC_HIGH;
1165 alloc_flags |= ALLOC_CPUSET;
1168 * Go through the zonelist again. Let __GFP_HIGH and allocations
1169 * coming from realtime tasks go deeper into reserves.
1171 * This is the last chance, in general, before the goto nopage.
1172 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1173 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1175 page = get_page_from_freelist(gfp_mask, order, zonelist, alloc_flags);
1179 /* This allocation should allow future memory freeing. */
1181 if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE)))
1182 && !in_interrupt()) {
1183 if (!(gfp_mask & __GFP_NOMEMALLOC)) {
1185 /* go through the zonelist yet again, ignoring mins */
1186 page = get_page_from_freelist(gfp_mask, order,
1187 zonelist, ALLOC_NO_WATERMARKS);
1190 if (gfp_mask & __GFP_NOFAIL) {
1191 congestion_wait(WRITE, HZ/50);
1198 /* Atomic allocations - we can't balance anything */
1205 /* We now go into synchronous reclaim */
1206 cpuset_memory_pressure_bump();
1207 p->flags |= PF_MEMALLOC;
1208 reclaim_state.reclaimed_slab = 0;
1209 p->reclaim_state = &reclaim_state;
1211 did_some_progress = try_to_free_pages(zonelist->zones, gfp_mask);
1213 p->reclaim_state = NULL;
1214 p->flags &= ~PF_MEMALLOC;
1218 if (likely(did_some_progress)) {
1219 page = get_page_from_freelist(gfp_mask, order,
1220 zonelist, alloc_flags);
1223 } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
1225 * Go through the zonelist yet one more time, keep
1226 * very high watermark here, this is only to catch
1227 * a parallel oom killing, we must fail if we're still
1228 * under heavy pressure.
1230 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
1231 zonelist, ALLOC_WMARK_HIGH|ALLOC_CPUSET);
1235 out_of_memory(zonelist, gfp_mask, order);
1240 * Don't let big-order allocations loop unless the caller explicitly
1241 * requests that. Wait for some write requests to complete then retry.
1243 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
1244 * <= 3, but that may not be true in other implementations.
1247 if (!(gfp_mask & __GFP_NORETRY)) {
1248 if ((order <= 3) || (gfp_mask & __GFP_REPEAT))
1250 if (gfp_mask & __GFP_NOFAIL)
1254 congestion_wait(WRITE, HZ/50);
1259 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
1260 printk(KERN_WARNING "%s: page allocation failure."
1261 " order:%d, mode:0x%x\n",
1262 p->comm, order, gfp_mask);
1270 EXPORT_SYMBOL(__alloc_pages);
1273 * Common helper functions.
1275 fastcall unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
1278 page = alloc_pages(gfp_mask, order);
1281 return (unsigned long) page_address(page);
1284 EXPORT_SYMBOL(__get_free_pages);
1286 fastcall unsigned long get_zeroed_page(gfp_t gfp_mask)
1291 * get_zeroed_page() returns a 32-bit address, which cannot represent
1294 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
1296 page = alloc_pages(gfp_mask | __GFP_ZERO, 0);
1298 return (unsigned long) page_address(page);
1302 EXPORT_SYMBOL(get_zeroed_page);
1304 void __pagevec_free(struct pagevec *pvec)
1306 int i = pagevec_count(pvec);
1309 free_hot_cold_page(pvec->pages[i], pvec->cold);
1312 fastcall void __free_pages(struct page *page, unsigned int order)
1314 if (put_page_testzero(page)) {
1316 free_hot_page(page);
1318 __free_pages_ok(page, order);
1322 EXPORT_SYMBOL(__free_pages);
1324 fastcall void free_pages(unsigned long addr, unsigned int order)
1327 VM_BUG_ON(!virt_addr_valid((void *)addr));
1328 __free_pages(virt_to_page((void *)addr), order);
1332 EXPORT_SYMBOL(free_pages);
1335 * Total amount of free (allocatable) RAM:
1337 unsigned int nr_free_pages(void)
1339 unsigned int sum = 0;
1343 sum += zone->free_pages;
1348 EXPORT_SYMBOL(nr_free_pages);
1351 unsigned int nr_free_pages_pgdat(pg_data_t *pgdat)
1353 unsigned int sum = 0;
1356 for (i = 0; i < MAX_NR_ZONES; i++)
1357 sum += pgdat->node_zones[i].free_pages;
1363 static unsigned int nr_free_zone_pages(int offset)
1365 /* Just pick one node, since fallback list is circular */
1366 pg_data_t *pgdat = NODE_DATA(numa_node_id());
1367 unsigned int sum = 0;
1369 struct zonelist *zonelist = pgdat->node_zonelists + offset;
1370 struct zone **zonep = zonelist->zones;
1373 for (zone = *zonep++; zone; zone = *zonep++) {
1374 unsigned long size = zone->present_pages;
1375 unsigned long high = zone->pages_high;
1384 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1386 unsigned int nr_free_buffer_pages(void)
1388 return nr_free_zone_pages(gfp_zone(GFP_USER));
1392 * Amount of free RAM allocatable within all zones
1394 unsigned int nr_free_pagecache_pages(void)
1396 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER));
1399 static inline void show_node(struct zone *zone)
1402 printk("Node %ld ", zone_to_nid(zone));
1405 void si_meminfo(struct sysinfo *val)
1407 val->totalram = totalram_pages;
1409 val->freeram = nr_free_pages();
1410 val->bufferram = nr_blockdev_pages();
1411 val->totalhigh = totalhigh_pages;
1412 val->freehigh = nr_free_highpages();
1413 val->mem_unit = PAGE_SIZE;
1416 EXPORT_SYMBOL(si_meminfo);
1419 void si_meminfo_node(struct sysinfo *val, int nid)
1421 pg_data_t *pgdat = NODE_DATA(nid);
1423 val->totalram = pgdat->node_present_pages;
1424 val->freeram = nr_free_pages_pgdat(pgdat);
1425 #ifdef CONFIG_HIGHMEM
1426 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1427 val->freehigh = pgdat->node_zones[ZONE_HIGHMEM].free_pages;
1432 val->mem_unit = PAGE_SIZE;
1436 #define K(x) ((x) << (PAGE_SHIFT-10))
1439 * Show free area list (used inside shift_scroll-lock stuff)
1440 * We also calculate the percentage fragmentation. We do this by counting the
1441 * memory on each free list with the exception of the first item on the list.
1443 void show_free_areas(void)
1446 unsigned long active;
1447 unsigned long inactive;
1451 for_each_zone(zone) {
1452 if (!populated_zone(zone))
1456 printk("%s per-cpu:\n", zone->name);
1458 for_each_online_cpu(cpu) {
1459 struct per_cpu_pageset *pageset;
1461 pageset = zone_pcp(zone, cpu);
1463 printk("CPU %4d: Hot: hi:%5d, btch:%4d usd:%4d "
1464 "Cold: hi:%5d, btch:%4d usd:%4d\n",
1465 cpu, pageset->pcp[0].high,
1466 pageset->pcp[0].batch, pageset->pcp[0].count,
1467 pageset->pcp[1].high, pageset->pcp[1].batch,
1468 pageset->pcp[1].count);
1472 get_zone_counts(&active, &inactive, &free);
1474 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu "
1475 "unstable:%lu free:%u slab:%lu mapped:%lu pagetables:%lu\n",
1478 global_page_state(NR_FILE_DIRTY),
1479 global_page_state(NR_WRITEBACK),
1480 global_page_state(NR_UNSTABLE_NFS),
1482 global_page_state(NR_SLAB_RECLAIMABLE) +
1483 global_page_state(NR_SLAB_UNRECLAIMABLE),
1484 global_page_state(NR_FILE_MAPPED),
1485 global_page_state(NR_PAGETABLE));
1487 for_each_zone(zone) {
1490 if (!populated_zone(zone))
1502 " pages_scanned:%lu"
1503 " all_unreclaimable? %s"
1506 K(zone->free_pages),
1509 K(zone->pages_high),
1511 K(zone->nr_inactive),
1512 K(zone->present_pages),
1513 zone->pages_scanned,
1514 (zone->all_unreclaimable ? "yes" : "no")
1516 printk("lowmem_reserve[]:");
1517 for (i = 0; i < MAX_NR_ZONES; i++)
1518 printk(" %lu", zone->lowmem_reserve[i]);
1522 for_each_zone(zone) {
1523 unsigned long nr[MAX_ORDER], flags, order, total = 0;
1525 if (!populated_zone(zone))
1529 printk("%s: ", zone->name);
1531 spin_lock_irqsave(&zone->lock, flags);
1532 for (order = 0; order < MAX_ORDER; order++) {
1533 nr[order] = zone->free_area[order].nr_free;
1534 total += nr[order] << order;
1536 spin_unlock_irqrestore(&zone->lock, flags);
1537 for (order = 0; order < MAX_ORDER; order++)
1538 printk("%lu*%lukB ", nr[order], K(1UL) << order);
1539 printk("= %lukB\n", K(total));
1542 show_swap_cache_info();
1546 * Builds allocation fallback zone lists.
1548 * Add all populated zones of a node to the zonelist.
1550 static int __meminit build_zonelists_node(pg_data_t *pgdat,
1551 struct zonelist *zonelist, int nr_zones, enum zone_type zone_type)
1555 BUG_ON(zone_type >= MAX_NR_ZONES);
1560 zone = pgdat->node_zones + zone_type;
1561 if (populated_zone(zone)) {
1562 zonelist->zones[nr_zones++] = zone;
1563 check_highest_zone(zone_type);
1566 } while (zone_type);
1571 #define MAX_NODE_LOAD (num_online_nodes())
1572 static int __meminitdata node_load[MAX_NUMNODES];
1574 * find_next_best_node - find the next node that should appear in a given node's fallback list
1575 * @node: node whose fallback list we're appending
1576 * @used_node_mask: nodemask_t of already used nodes
1578 * We use a number of factors to determine which is the next node that should
1579 * appear on a given node's fallback list. The node should not have appeared
1580 * already in @node's fallback list, and it should be the next closest node
1581 * according to the distance array (which contains arbitrary distance values
1582 * from each node to each node in the system), and should also prefer nodes
1583 * with no CPUs, since presumably they'll have very little allocation pressure
1584 * on them otherwise.
1585 * It returns -1 if no node is found.
1587 static int __meminit find_next_best_node(int node, nodemask_t *used_node_mask)
1590 int min_val = INT_MAX;
1593 /* Use the local node if we haven't already */
1594 if (!node_isset(node, *used_node_mask)) {
1595 node_set(node, *used_node_mask);
1599 for_each_online_node(n) {
1602 /* Don't want a node to appear more than once */
1603 if (node_isset(n, *used_node_mask))
1606 /* Use the distance array to find the distance */
1607 val = node_distance(node, n);
1609 /* Penalize nodes under us ("prefer the next node") */
1612 /* Give preference to headless and unused nodes */
1613 tmp = node_to_cpumask(n);
1614 if (!cpus_empty(tmp))
1615 val += PENALTY_FOR_NODE_WITH_CPUS;
1617 /* Slight preference for less loaded node */
1618 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
1619 val += node_load[n];
1621 if (val < min_val) {
1628 node_set(best_node, *used_node_mask);
1633 static void __meminit build_zonelists(pg_data_t *pgdat)
1635 int j, node, local_node;
1637 int prev_node, load;
1638 struct zonelist *zonelist;
1639 nodemask_t used_mask;
1641 /* initialize zonelists */
1642 for (i = 0; i < MAX_NR_ZONES; i++) {
1643 zonelist = pgdat->node_zonelists + i;
1644 zonelist->zones[0] = NULL;
1647 /* NUMA-aware ordering of nodes */
1648 local_node = pgdat->node_id;
1649 load = num_online_nodes();
1650 prev_node = local_node;
1651 nodes_clear(used_mask);
1652 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
1653 int distance = node_distance(local_node, node);
1656 * If another node is sufficiently far away then it is better
1657 * to reclaim pages in a zone before going off node.
1659 if (distance > RECLAIM_DISTANCE)
1660 zone_reclaim_mode = 1;
1663 * We don't want to pressure a particular node.
1664 * So adding penalty to the first node in same
1665 * distance group to make it round-robin.
1668 if (distance != node_distance(local_node, prev_node))
1669 node_load[node] += load;
1672 for (i = 0; i < MAX_NR_ZONES; i++) {
1673 zonelist = pgdat->node_zonelists + i;
1674 for (j = 0; zonelist->zones[j] != NULL; j++);
1676 j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
1677 zonelist->zones[j] = NULL;
1682 /* Construct the zonelist performance cache - see further mmzone.h */
1683 static void __meminit build_zonelist_cache(pg_data_t *pgdat)
1687 for (i = 0; i < MAX_NR_ZONES; i++) {
1688 struct zonelist *zonelist;
1689 struct zonelist_cache *zlc;
1692 zonelist = pgdat->node_zonelists + i;
1693 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
1694 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1695 for (z = zonelist->zones; *z; z++)
1696 zlc->z_to_n[z - zonelist->zones] = zone_to_nid(*z);
1700 #else /* CONFIG_NUMA */
1702 static void __meminit build_zonelists(pg_data_t *pgdat)
1704 int node, local_node;
1707 local_node = pgdat->node_id;
1708 for (i = 0; i < MAX_NR_ZONES; i++) {
1709 struct zonelist *zonelist;
1711 zonelist = pgdat->node_zonelists + i;
1713 j = build_zonelists_node(pgdat, zonelist, 0, i);
1715 * Now we build the zonelist so that it contains the zones
1716 * of all the other nodes.
1717 * We don't want to pressure a particular node, so when
1718 * building the zones for node N, we make sure that the
1719 * zones coming right after the local ones are those from
1720 * node N+1 (modulo N)
1722 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
1723 if (!node_online(node))
1725 j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
1727 for (node = 0; node < local_node; node++) {
1728 if (!node_online(node))
1730 j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
1733 zonelist->zones[j] = NULL;
1737 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
1738 static void __meminit build_zonelist_cache(pg_data_t *pgdat)
1742 for (i = 0; i < MAX_NR_ZONES; i++)
1743 pgdat->node_zonelists[i].zlcache_ptr = NULL;
1746 #endif /* CONFIG_NUMA */
1748 /* return values int ....just for stop_machine_run() */
1749 static int __meminit __build_all_zonelists(void *dummy)
1753 for_each_online_node(nid) {
1754 build_zonelists(NODE_DATA(nid));
1755 build_zonelist_cache(NODE_DATA(nid));
1760 void __meminit build_all_zonelists(void)
1762 if (system_state == SYSTEM_BOOTING) {
1763 __build_all_zonelists(NULL);
1764 cpuset_init_current_mems_allowed();
1766 /* we have to stop all cpus to guaranntee there is no user
1768 stop_machine_run(__build_all_zonelists, NULL, NR_CPUS);
1769 /* cpuset refresh routine should be here */
1771 vm_total_pages = nr_free_pagecache_pages();
1772 printk("Built %i zonelists. Total pages: %ld\n",
1773 num_online_nodes(), vm_total_pages);
1777 * Helper functions to size the waitqueue hash table.
1778 * Essentially these want to choose hash table sizes sufficiently
1779 * large so that collisions trying to wait on pages are rare.
1780 * But in fact, the number of active page waitqueues on typical
1781 * systems is ridiculously low, less than 200. So this is even
1782 * conservative, even though it seems large.
1784 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
1785 * waitqueues, i.e. the size of the waitq table given the number of pages.
1787 #define PAGES_PER_WAITQUEUE 256
1789 #ifndef CONFIG_MEMORY_HOTPLUG
1790 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
1792 unsigned long size = 1;
1794 pages /= PAGES_PER_WAITQUEUE;
1796 while (size < pages)
1800 * Once we have dozens or even hundreds of threads sleeping
1801 * on IO we've got bigger problems than wait queue collision.
1802 * Limit the size of the wait table to a reasonable size.
1804 size = min(size, 4096UL);
1806 return max(size, 4UL);
1810 * A zone's size might be changed by hot-add, so it is not possible to determine
1811 * a suitable size for its wait_table. So we use the maximum size now.
1813 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
1815 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
1816 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
1817 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
1819 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
1820 * or more by the traditional way. (See above). It equals:
1822 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
1823 * ia64(16K page size) : = ( 8G + 4M)byte.
1824 * powerpc (64K page size) : = (32G +16M)byte.
1826 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
1833 * This is an integer logarithm so that shifts can be used later
1834 * to extract the more random high bits from the multiplicative
1835 * hash function before the remainder is taken.
1837 static inline unsigned long wait_table_bits(unsigned long size)
1842 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
1845 * Initially all pages are reserved - free ones are freed
1846 * up by free_all_bootmem() once the early boot process is
1847 * done. Non-atomic initialization, single-pass.
1849 void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
1850 unsigned long start_pfn)
1853 unsigned long end_pfn = start_pfn + size;
1856 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1857 if (!early_pfn_valid(pfn))
1859 if (!early_pfn_in_nid(pfn, nid))
1861 page = pfn_to_page(pfn);
1862 set_page_links(page, zone, nid, pfn);
1863 init_page_count(page);
1864 reset_page_mapcount(page);
1865 SetPageReserved(page);
1866 INIT_LIST_HEAD(&page->lru);
1867 #ifdef WANT_PAGE_VIRTUAL
1868 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
1869 if (!is_highmem_idx(zone))
1870 set_page_address(page, __va(pfn << PAGE_SHIFT));
1875 void zone_init_free_lists(struct pglist_data *pgdat, struct zone *zone,
1879 for (order = 0; order < MAX_ORDER ; order++) {
1880 INIT_LIST_HEAD(&zone->free_area[order].free_list);
1881 zone->free_area[order].nr_free = 0;
1885 #ifndef __HAVE_ARCH_MEMMAP_INIT
1886 #define memmap_init(size, nid, zone, start_pfn) \
1887 memmap_init_zone((size), (nid), (zone), (start_pfn))
1890 static int __cpuinit zone_batchsize(struct zone *zone)
1895 * The per-cpu-pages pools are set to around 1000th of the
1896 * size of the zone. But no more than 1/2 of a meg.
1898 * OK, so we don't know how big the cache is. So guess.
1900 batch = zone->present_pages / 1024;
1901 if (batch * PAGE_SIZE > 512 * 1024)
1902 batch = (512 * 1024) / PAGE_SIZE;
1903 batch /= 4; /* We effectively *= 4 below */
1908 * Clamp the batch to a 2^n - 1 value. Having a power
1909 * of 2 value was found to be more likely to have
1910 * suboptimal cache aliasing properties in some cases.
1912 * For example if 2 tasks are alternately allocating
1913 * batches of pages, one task can end up with a lot
1914 * of pages of one half of the possible page colors
1915 * and the other with pages of the other colors.
1917 batch = (1 << (fls(batch + batch/2)-1)) - 1;
1922 inline void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
1924 struct per_cpu_pages *pcp;
1926 memset(p, 0, sizeof(*p));
1928 pcp = &p->pcp[0]; /* hot */
1930 pcp->high = 6 * batch;
1931 pcp->batch = max(1UL, 1 * batch);
1932 INIT_LIST_HEAD(&pcp->list);
1934 pcp = &p->pcp[1]; /* cold*/
1936 pcp->high = 2 * batch;
1937 pcp->batch = max(1UL, batch/2);
1938 INIT_LIST_HEAD(&pcp->list);
1942 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
1943 * to the value high for the pageset p.
1946 static void setup_pagelist_highmark(struct per_cpu_pageset *p,
1949 struct per_cpu_pages *pcp;
1951 pcp = &p->pcp[0]; /* hot list */
1953 pcp->batch = max(1UL, high/4);
1954 if ((high/4) > (PAGE_SHIFT * 8))
1955 pcp->batch = PAGE_SHIFT * 8;
1961 * Boot pageset table. One per cpu which is going to be used for all
1962 * zones and all nodes. The parameters will be set in such a way
1963 * that an item put on a list will immediately be handed over to
1964 * the buddy list. This is safe since pageset manipulation is done
1965 * with interrupts disabled.
1967 * Some NUMA counter updates may also be caught by the boot pagesets.
1969 * The boot_pagesets must be kept even after bootup is complete for
1970 * unused processors and/or zones. They do play a role for bootstrapping
1971 * hotplugged processors.
1973 * zoneinfo_show() and maybe other functions do
1974 * not check if the processor is online before following the pageset pointer.
1975 * Other parts of the kernel may not check if the zone is available.
1977 static struct per_cpu_pageset boot_pageset[NR_CPUS];
1980 * Dynamically allocate memory for the
1981 * per cpu pageset array in struct zone.
1983 static int __cpuinit process_zones(int cpu)
1985 struct zone *zone, *dzone;
1987 for_each_zone(zone) {
1989 if (!populated_zone(zone))
1992 zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset),
1993 GFP_KERNEL, cpu_to_node(cpu));
1994 if (!zone_pcp(zone, cpu))
1997 setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone));
1999 if (percpu_pagelist_fraction)
2000 setup_pagelist_highmark(zone_pcp(zone, cpu),
2001 (zone->present_pages / percpu_pagelist_fraction));
2006 for_each_zone(dzone) {
2009 kfree(zone_pcp(dzone, cpu));
2010 zone_pcp(dzone, cpu) = NULL;
2015 static inline void free_zone_pagesets(int cpu)
2019 for_each_zone(zone) {
2020 struct per_cpu_pageset *pset = zone_pcp(zone, cpu);
2022 /* Free per_cpu_pageset if it is slab allocated */
2023 if (pset != &boot_pageset[cpu])
2025 zone_pcp(zone, cpu) = NULL;
2029 static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb,
2030 unsigned long action,
2033 int cpu = (long)hcpu;
2034 int ret = NOTIFY_OK;
2037 case CPU_UP_PREPARE:
2038 if (process_zones(cpu))
2041 case CPU_UP_CANCELED:
2043 free_zone_pagesets(cpu);
2051 static struct notifier_block __cpuinitdata pageset_notifier =
2052 { &pageset_cpuup_callback, NULL, 0 };
2054 void __init setup_per_cpu_pageset(void)
2058 /* Initialize per_cpu_pageset for cpu 0.
2059 * A cpuup callback will do this for every cpu
2060 * as it comes online
2062 err = process_zones(smp_processor_id());
2064 register_cpu_notifier(&pageset_notifier);
2070 int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
2073 struct pglist_data *pgdat = zone->zone_pgdat;
2077 * The per-page waitqueue mechanism uses hashed waitqueues
2080 zone->wait_table_hash_nr_entries =
2081 wait_table_hash_nr_entries(zone_size_pages);
2082 zone->wait_table_bits =
2083 wait_table_bits(zone->wait_table_hash_nr_entries);
2084 alloc_size = zone->wait_table_hash_nr_entries
2085 * sizeof(wait_queue_head_t);
2087 if (system_state == SYSTEM_BOOTING) {
2088 zone->wait_table = (wait_queue_head_t *)
2089 alloc_bootmem_node(pgdat, alloc_size);
2092 * This case means that a zone whose size was 0 gets new memory
2093 * via memory hot-add.
2094 * But it may be the case that a new node was hot-added. In
2095 * this case vmalloc() will not be able to use this new node's
2096 * memory - this wait_table must be initialized to use this new
2097 * node itself as well.
2098 * To use this new node's memory, further consideration will be
2101 zone->wait_table = (wait_queue_head_t *)vmalloc(alloc_size);
2103 if (!zone->wait_table)
2106 for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
2107 init_waitqueue_head(zone->wait_table + i);
2112 static __meminit void zone_pcp_init(struct zone *zone)
2115 unsigned long batch = zone_batchsize(zone);
2117 for (cpu = 0; cpu < NR_CPUS; cpu++) {
2119 /* Early boot. Slab allocator not functional yet */
2120 zone_pcp(zone, cpu) = &boot_pageset[cpu];
2121 setup_pageset(&boot_pageset[cpu],0);
2123 setup_pageset(zone_pcp(zone,cpu), batch);
2126 if (zone->present_pages)
2127 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n",
2128 zone->name, zone->present_pages, batch);
2131 __meminit int init_currently_empty_zone(struct zone *zone,
2132 unsigned long zone_start_pfn,
2135 struct pglist_data *pgdat = zone->zone_pgdat;
2137 ret = zone_wait_table_init(zone, size);
2140 pgdat->nr_zones = zone_idx(zone) + 1;
2142 zone->zone_start_pfn = zone_start_pfn;
2144 memmap_init(size, pgdat->node_id, zone_idx(zone), zone_start_pfn);
2146 zone_init_free_lists(pgdat, zone, zone->spanned_pages);
2151 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2153 * Basic iterator support. Return the first range of PFNs for a node
2154 * Note: nid == MAX_NUMNODES returns first region regardless of node
2156 static int __init first_active_region_index_in_nid(int nid)
2160 for (i = 0; i < nr_nodemap_entries; i++)
2161 if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
2168 * Basic iterator support. Return the next active range of PFNs for a node
2169 * Note: nid == MAX_NUMNODES returns next region regardles of node
2171 static int __init next_active_region_index_in_nid(int index, int nid)
2173 for (index = index + 1; index < nr_nodemap_entries; index++)
2174 if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
2180 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
2182 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
2183 * Architectures may implement their own version but if add_active_range()
2184 * was used and there are no special requirements, this is a convenient
2187 int __init early_pfn_to_nid(unsigned long pfn)
2191 for (i = 0; i < nr_nodemap_entries; i++) {
2192 unsigned long start_pfn = early_node_map[i].start_pfn;
2193 unsigned long end_pfn = early_node_map[i].end_pfn;
2195 if (start_pfn <= pfn && pfn < end_pfn)
2196 return early_node_map[i].nid;
2201 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
2203 /* Basic iterator support to walk early_node_map[] */
2204 #define for_each_active_range_index_in_nid(i, nid) \
2205 for (i = first_active_region_index_in_nid(nid); i != -1; \
2206 i = next_active_region_index_in_nid(i, nid))
2209 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
2210 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
2211 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
2213 * If an architecture guarantees that all ranges registered with
2214 * add_active_ranges() contain no holes and may be freed, this
2215 * this function may be used instead of calling free_bootmem() manually.
2217 void __init free_bootmem_with_active_regions(int nid,
2218 unsigned long max_low_pfn)
2222 for_each_active_range_index_in_nid(i, nid) {
2223 unsigned long size_pages = 0;
2224 unsigned long end_pfn = early_node_map[i].end_pfn;
2226 if (early_node_map[i].start_pfn >= max_low_pfn)
2229 if (end_pfn > max_low_pfn)
2230 end_pfn = max_low_pfn;
2232 size_pages = end_pfn - early_node_map[i].start_pfn;
2233 free_bootmem_node(NODE_DATA(early_node_map[i].nid),
2234 PFN_PHYS(early_node_map[i].start_pfn),
2235 size_pages << PAGE_SHIFT);
2240 * sparse_memory_present_with_active_regions - Call memory_present for each active range
2241 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
2243 * If an architecture guarantees that all ranges registered with
2244 * add_active_ranges() contain no holes and may be freed, this
2245 * function may be used instead of calling memory_present() manually.
2247 void __init sparse_memory_present_with_active_regions(int nid)
2251 for_each_active_range_index_in_nid(i, nid)
2252 memory_present(early_node_map[i].nid,
2253 early_node_map[i].start_pfn,
2254 early_node_map[i].end_pfn);
2258 * push_node_boundaries - Push node boundaries to at least the requested boundary
2259 * @nid: The nid of the node to push the boundary for
2260 * @start_pfn: The start pfn of the node
2261 * @end_pfn: The end pfn of the node
2263 * In reserve-based hot-add, mem_map is allocated that is unused until hotadd
2264 * time. Specifically, on x86_64, SRAT will report ranges that can potentially
2265 * be hotplugged even though no physical memory exists. This function allows
2266 * an arch to push out the node boundaries so mem_map is allocated that can
2269 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
2270 void __init push_node_boundaries(unsigned int nid,
2271 unsigned long start_pfn, unsigned long end_pfn)
2273 printk(KERN_DEBUG "Entering push_node_boundaries(%u, %lu, %lu)\n",
2274 nid, start_pfn, end_pfn);
2276 /* Initialise the boundary for this node if necessary */
2277 if (node_boundary_end_pfn[nid] == 0)
2278 node_boundary_start_pfn[nid] = -1UL;
2280 /* Update the boundaries */
2281 if (node_boundary_start_pfn[nid] > start_pfn)
2282 node_boundary_start_pfn[nid] = start_pfn;
2283 if (node_boundary_end_pfn[nid] < end_pfn)
2284 node_boundary_end_pfn[nid] = end_pfn;
2287 /* If necessary, push the node boundary out for reserve hotadd */
2288 static void __init account_node_boundary(unsigned int nid,
2289 unsigned long *start_pfn, unsigned long *end_pfn)
2291 printk(KERN_DEBUG "Entering account_node_boundary(%u, %lu, %lu)\n",
2292 nid, *start_pfn, *end_pfn);
2294 /* Return if boundary information has not been provided */
2295 if (node_boundary_end_pfn[nid] == 0)
2298 /* Check the boundaries and update if necessary */
2299 if (node_boundary_start_pfn[nid] < *start_pfn)
2300 *start_pfn = node_boundary_start_pfn[nid];
2301 if (node_boundary_end_pfn[nid] > *end_pfn)
2302 *end_pfn = node_boundary_end_pfn[nid];
2305 void __init push_node_boundaries(unsigned int nid,
2306 unsigned long start_pfn, unsigned long end_pfn) {}
2308 static void __init account_node_boundary(unsigned int nid,
2309 unsigned long *start_pfn, unsigned long *end_pfn) {}
2314 * get_pfn_range_for_nid - Return the start and end page frames for a node
2315 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
2316 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
2317 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
2319 * It returns the start and end page frame of a node based on information
2320 * provided by an arch calling add_active_range(). If called for a node
2321 * with no available memory, a warning is printed and the start and end
2324 void __init get_pfn_range_for_nid(unsigned int nid,
2325 unsigned long *start_pfn, unsigned long *end_pfn)
2331 for_each_active_range_index_in_nid(i, nid) {
2332 *start_pfn = min(*start_pfn, early_node_map[i].start_pfn);
2333 *end_pfn = max(*end_pfn, early_node_map[i].end_pfn);
2336 if (*start_pfn == -1UL) {
2337 printk(KERN_WARNING "Node %u active with no memory\n", nid);
2341 /* Push the node boundaries out if requested */
2342 account_node_boundary(nid, start_pfn, end_pfn);
2346 * Return the number of pages a zone spans in a node, including holes
2347 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
2349 unsigned long __init zone_spanned_pages_in_node(int nid,
2350 unsigned long zone_type,
2351 unsigned long *ignored)
2353 unsigned long node_start_pfn, node_end_pfn;
2354 unsigned long zone_start_pfn, zone_end_pfn;
2356 /* Get the start and end of the node and zone */
2357 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
2358 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
2359 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
2361 /* Check that this node has pages within the zone's required range */
2362 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
2365 /* Move the zone boundaries inside the node if necessary */
2366 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
2367 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
2369 /* Return the spanned pages */
2370 return zone_end_pfn - zone_start_pfn;
2374 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
2375 * then all holes in the requested range will be accounted for.
2377 unsigned long __init __absent_pages_in_range(int nid,
2378 unsigned long range_start_pfn,
2379 unsigned long range_end_pfn)
2382 unsigned long prev_end_pfn = 0, hole_pages = 0;
2383 unsigned long start_pfn;
2385 /* Find the end_pfn of the first active range of pfns in the node */
2386 i = first_active_region_index_in_nid(nid);
2390 /* Account for ranges before physical memory on this node */
2391 if (early_node_map[i].start_pfn > range_start_pfn)
2392 hole_pages = early_node_map[i].start_pfn - range_start_pfn;
2394 prev_end_pfn = early_node_map[i].start_pfn;
2396 /* Find all holes for the zone within the node */
2397 for (; i != -1; i = next_active_region_index_in_nid(i, nid)) {
2399 /* No need to continue if prev_end_pfn is outside the zone */
2400 if (prev_end_pfn >= range_end_pfn)
2403 /* Make sure the end of the zone is not within the hole */
2404 start_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
2405 prev_end_pfn = max(prev_end_pfn, range_start_pfn);
2407 /* Update the hole size cound and move on */
2408 if (start_pfn > range_start_pfn) {
2409 BUG_ON(prev_end_pfn > start_pfn);
2410 hole_pages += start_pfn - prev_end_pfn;
2412 prev_end_pfn = early_node_map[i].end_pfn;
2415 /* Account for ranges past physical memory on this node */
2416 if (range_end_pfn > prev_end_pfn)
2417 hole_pages += range_end_pfn -
2418 max(range_start_pfn, prev_end_pfn);
2424 * absent_pages_in_range - Return number of page frames in holes within a range
2425 * @start_pfn: The start PFN to start searching for holes
2426 * @end_pfn: The end PFN to stop searching for holes
2428 * It returns the number of pages frames in memory holes within a range.
2430 unsigned long __init absent_pages_in_range(unsigned long start_pfn,
2431 unsigned long end_pfn)
2433 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
2436 /* Return the number of page frames in holes in a zone on a node */
2437 unsigned long __init zone_absent_pages_in_node(int nid,
2438 unsigned long zone_type,
2439 unsigned long *ignored)
2441 unsigned long node_start_pfn, node_end_pfn;
2442 unsigned long zone_start_pfn, zone_end_pfn;
2444 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
2445 zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type],
2447 zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type],
2450 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
2454 static inline unsigned long zone_spanned_pages_in_node(int nid,
2455 unsigned long zone_type,
2456 unsigned long *zones_size)
2458 return zones_size[zone_type];
2461 static inline unsigned long zone_absent_pages_in_node(int nid,
2462 unsigned long zone_type,
2463 unsigned long *zholes_size)
2468 return zholes_size[zone_type];
2473 static void __init calculate_node_totalpages(struct pglist_data *pgdat,
2474 unsigned long *zones_size, unsigned long *zholes_size)
2476 unsigned long realtotalpages, totalpages = 0;
2479 for (i = 0; i < MAX_NR_ZONES; i++)
2480 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
2482 pgdat->node_spanned_pages = totalpages;
2484 realtotalpages = totalpages;
2485 for (i = 0; i < MAX_NR_ZONES; i++)
2487 zone_absent_pages_in_node(pgdat->node_id, i,
2489 pgdat->node_present_pages = realtotalpages;
2490 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
2495 * Set up the zone data structures:
2496 * - mark all pages reserved
2497 * - mark all memory queues empty
2498 * - clear the memory bitmaps
2500 static void __meminit free_area_init_core(struct pglist_data *pgdat,
2501 unsigned long *zones_size, unsigned long *zholes_size)
2504 int nid = pgdat->node_id;
2505 unsigned long zone_start_pfn = pgdat->node_start_pfn;
2508 pgdat_resize_init(pgdat);
2509 pgdat->nr_zones = 0;
2510 init_waitqueue_head(&pgdat->kswapd_wait);
2511 pgdat->kswapd_max_order = 0;
2513 for (j = 0; j < MAX_NR_ZONES; j++) {
2514 struct zone *zone = pgdat->node_zones + j;
2515 unsigned long size, realsize, memmap_pages;
2517 size = zone_spanned_pages_in_node(nid, j, zones_size);
2518 realsize = size - zone_absent_pages_in_node(nid, j,
2522 * Adjust realsize so that it accounts for how much memory
2523 * is used by this zone for memmap. This affects the watermark
2524 * and per-cpu initialisations
2526 memmap_pages = (size * sizeof(struct page)) >> PAGE_SHIFT;
2527 if (realsize >= memmap_pages) {
2528 realsize -= memmap_pages;
2530 " %s zone: %lu pages used for memmap\n",
2531 zone_names[j], memmap_pages);
2534 " %s zone: %lu pages exceeds realsize %lu\n",
2535 zone_names[j], memmap_pages, realsize);
2537 /* Account for reserved DMA pages */
2538 if (j == ZONE_DMA && realsize > dma_reserve) {
2539 realsize -= dma_reserve;
2540 printk(KERN_DEBUG " DMA zone: %lu pages reserved\n",
2544 if (!is_highmem_idx(j))
2545 nr_kernel_pages += realsize;
2546 nr_all_pages += realsize;
2548 zone->spanned_pages = size;
2549 zone->present_pages = realsize;
2552 zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
2554 zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100;
2556 zone->name = zone_names[j];
2557 spin_lock_init(&zone->lock);
2558 spin_lock_init(&zone->lru_lock);
2559 zone_seqlock_init(zone);
2560 zone->zone_pgdat = pgdat;
2561 zone->free_pages = 0;
2563 zone->prev_priority = DEF_PRIORITY;
2565 zone_pcp_init(zone);
2566 INIT_LIST_HEAD(&zone->active_list);
2567 INIT_LIST_HEAD(&zone->inactive_list);
2568 zone->nr_scan_active = 0;
2569 zone->nr_scan_inactive = 0;
2570 zone->nr_active = 0;
2571 zone->nr_inactive = 0;
2572 zap_zone_vm_stats(zone);
2573 atomic_set(&zone->reclaim_in_progress, 0);
2577 ret = init_currently_empty_zone(zone, zone_start_pfn, size);
2579 zone_start_pfn += size;
2583 static void __init alloc_node_mem_map(struct pglist_data *pgdat)
2585 /* Skip empty nodes */
2586 if (!pgdat->node_spanned_pages)
2589 #ifdef CONFIG_FLAT_NODE_MEM_MAP
2590 /* ia64 gets its own node_mem_map, before this, without bootmem */
2591 if (!pgdat->node_mem_map) {
2592 unsigned long size, start, end;
2596 * The zone's endpoints aren't required to be MAX_ORDER
2597 * aligned but the node_mem_map endpoints must be in order
2598 * for the buddy allocator to function correctly.
2600 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
2601 end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
2602 end = ALIGN(end, MAX_ORDER_NR_PAGES);
2603 size = (end - start) * sizeof(struct page);
2604 map = alloc_remap(pgdat->node_id, size);
2606 map = alloc_bootmem_node(pgdat, size);
2607 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
2609 #ifdef CONFIG_FLATMEM
2611 * With no DISCONTIG, the global mem_map is just set as node 0's
2613 if (pgdat == NODE_DATA(0)) {
2614 mem_map = NODE_DATA(0)->node_mem_map;
2615 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2616 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
2617 mem_map -= pgdat->node_start_pfn;
2618 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
2621 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
2624 void __meminit free_area_init_node(int nid, struct pglist_data *pgdat,
2625 unsigned long *zones_size, unsigned long node_start_pfn,
2626 unsigned long *zholes_size)
2628 pgdat->node_id = nid;
2629 pgdat->node_start_pfn = node_start_pfn;
2630 calculate_node_totalpages(pgdat, zones_size, zholes_size);
2632 alloc_node_mem_map(pgdat);
2634 free_area_init_core(pgdat, zones_size, zholes_size);
2637 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2639 * add_active_range - Register a range of PFNs backed by physical memory
2640 * @nid: The node ID the range resides on
2641 * @start_pfn: The start PFN of the available physical memory
2642 * @end_pfn: The end PFN of the available physical memory
2644 * These ranges are stored in an early_node_map[] and later used by
2645 * free_area_init_nodes() to calculate zone sizes and holes. If the
2646 * range spans a memory hole, it is up to the architecture to ensure
2647 * the memory is not freed by the bootmem allocator. If possible
2648 * the range being registered will be merged with existing ranges.
2650 void __init add_active_range(unsigned int nid, unsigned long start_pfn,
2651 unsigned long end_pfn)
2655 printk(KERN_DEBUG "Entering add_active_range(%d, %lu, %lu) "
2656 "%d entries of %d used\n",
2657 nid, start_pfn, end_pfn,
2658 nr_nodemap_entries, MAX_ACTIVE_REGIONS);
2660 /* Merge with existing active regions if possible */
2661 for (i = 0; i < nr_nodemap_entries; i++) {
2662 if (early_node_map[i].nid != nid)
2665 /* Skip if an existing region covers this new one */
2666 if (start_pfn >= early_node_map[i].start_pfn &&
2667 end_pfn <= early_node_map[i].end_pfn)
2670 /* Merge forward if suitable */
2671 if (start_pfn <= early_node_map[i].end_pfn &&
2672 end_pfn > early_node_map[i].end_pfn) {
2673 early_node_map[i].end_pfn = end_pfn;
2677 /* Merge backward if suitable */
2678 if (start_pfn < early_node_map[i].end_pfn &&
2679 end_pfn >= early_node_map[i].start_pfn) {
2680 early_node_map[i].start_pfn = start_pfn;
2685 /* Check that early_node_map is large enough */
2686 if (i >= MAX_ACTIVE_REGIONS) {
2687 printk(KERN_CRIT "More than %d memory regions, truncating\n",
2688 MAX_ACTIVE_REGIONS);
2692 early_node_map[i].nid = nid;
2693 early_node_map[i].start_pfn = start_pfn;
2694 early_node_map[i].end_pfn = end_pfn;
2695 nr_nodemap_entries = i + 1;
2699 * shrink_active_range - Shrink an existing registered range of PFNs
2700 * @nid: The node id the range is on that should be shrunk
2701 * @old_end_pfn: The old end PFN of the range
2702 * @new_end_pfn: The new PFN of the range
2704 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
2705 * The map is kept at the end physical page range that has already been
2706 * registered with add_active_range(). This function allows an arch to shrink
2707 * an existing registered range.
2709 void __init shrink_active_range(unsigned int nid, unsigned long old_end_pfn,
2710 unsigned long new_end_pfn)
2714 /* Find the old active region end and shrink */
2715 for_each_active_range_index_in_nid(i, nid)
2716 if (early_node_map[i].end_pfn == old_end_pfn) {
2717 early_node_map[i].end_pfn = new_end_pfn;
2723 * remove_all_active_ranges - Remove all currently registered regions
2725 * During discovery, it may be found that a table like SRAT is invalid
2726 * and an alternative discovery method must be used. This function removes
2727 * all currently registered regions.
2729 void __init remove_all_active_ranges(void)
2731 memset(early_node_map, 0, sizeof(early_node_map));
2732 nr_nodemap_entries = 0;
2733 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
2734 memset(node_boundary_start_pfn, 0, sizeof(node_boundary_start_pfn));
2735 memset(node_boundary_end_pfn, 0, sizeof(node_boundary_end_pfn));
2736 #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
2739 /* Compare two active node_active_regions */
2740 static int __init cmp_node_active_region(const void *a, const void *b)
2742 struct node_active_region *arange = (struct node_active_region *)a;
2743 struct node_active_region *brange = (struct node_active_region *)b;
2745 /* Done this way to avoid overflows */
2746 if (arange->start_pfn > brange->start_pfn)
2748 if (arange->start_pfn < brange->start_pfn)
2754 /* sort the node_map by start_pfn */
2755 static void __init sort_node_map(void)
2757 sort(early_node_map, (size_t)nr_nodemap_entries,
2758 sizeof(struct node_active_region),
2759 cmp_node_active_region, NULL);
2762 /* Find the lowest pfn for a node. This depends on a sorted early_node_map */
2763 unsigned long __init find_min_pfn_for_node(unsigned long nid)
2767 /* Regions in the early_node_map can be in any order */
2770 /* Assuming a sorted map, the first range found has the starting pfn */
2771 for_each_active_range_index_in_nid(i, nid)
2772 return early_node_map[i].start_pfn;
2774 printk(KERN_WARNING "Could not find start_pfn for node %lu\n", nid);
2779 * find_min_pfn_with_active_regions - Find the minimum PFN registered
2781 * It returns the minimum PFN based on information provided via
2782 * add_active_range().
2784 unsigned long __init find_min_pfn_with_active_regions(void)
2786 return find_min_pfn_for_node(MAX_NUMNODES);
2790 * find_max_pfn_with_active_regions - Find the maximum PFN registered
2792 * It returns the maximum PFN based on information provided via
2793 * add_active_range().
2795 unsigned long __init find_max_pfn_with_active_regions(void)
2798 unsigned long max_pfn = 0;
2800 for (i = 0; i < nr_nodemap_entries; i++)
2801 max_pfn = max(max_pfn, early_node_map[i].end_pfn);
2807 * free_area_init_nodes - Initialise all pg_data_t and zone data
2808 * @max_zone_pfn: an array of max PFNs for each zone
2810 * This will call free_area_init_node() for each active node in the system.
2811 * Using the page ranges provided by add_active_range(), the size of each
2812 * zone in each node and their holes is calculated. If the maximum PFN
2813 * between two adjacent zones match, it is assumed that the zone is empty.
2814 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
2815 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
2816 * starts where the previous one ended. For example, ZONE_DMA32 starts
2817 * at arch_max_dma_pfn.
2819 void __init free_area_init_nodes(unsigned long *max_zone_pfn)
2824 /* Record where the zone boundaries are */
2825 memset(arch_zone_lowest_possible_pfn, 0,
2826 sizeof(arch_zone_lowest_possible_pfn));
2827 memset(arch_zone_highest_possible_pfn, 0,
2828 sizeof(arch_zone_highest_possible_pfn));
2829 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
2830 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
2831 for (i = 1; i < MAX_NR_ZONES; i++) {
2832 arch_zone_lowest_possible_pfn[i] =
2833 arch_zone_highest_possible_pfn[i-1];
2834 arch_zone_highest_possible_pfn[i] =
2835 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
2838 /* Print out the zone ranges */
2839 printk("Zone PFN ranges:\n");
2840 for (i = 0; i < MAX_NR_ZONES; i++)
2841 printk(" %-8s %8lu -> %8lu\n",
2843 arch_zone_lowest_possible_pfn[i],
2844 arch_zone_highest_possible_pfn[i]);
2846 /* Print out the early_node_map[] */
2847 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries);
2848 for (i = 0; i < nr_nodemap_entries; i++)
2849 printk(" %3d: %8lu -> %8lu\n", early_node_map[i].nid,
2850 early_node_map[i].start_pfn,
2851 early_node_map[i].end_pfn);
2853 /* Initialise every node */
2854 for_each_online_node(nid) {
2855 pg_data_t *pgdat = NODE_DATA(nid);
2856 free_area_init_node(nid, pgdat, NULL,
2857 find_min_pfn_for_node(nid), NULL);
2860 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
2863 * set_dma_reserve - set the specified number of pages reserved in the first zone
2864 * @new_dma_reserve: The number of pages to mark reserved
2866 * The per-cpu batchsize and zone watermarks are determined by present_pages.
2867 * In the DMA zone, a significant percentage may be consumed by kernel image
2868 * and other unfreeable allocations which can skew the watermarks badly. This
2869 * function may optionally be used to account for unfreeable pages in the
2870 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
2871 * smaller per-cpu batchsize.
2873 void __init set_dma_reserve(unsigned long new_dma_reserve)
2875 dma_reserve = new_dma_reserve;
2878 #ifndef CONFIG_NEED_MULTIPLE_NODES
2879 static bootmem_data_t contig_bootmem_data;
2880 struct pglist_data contig_page_data = { .bdata = &contig_bootmem_data };
2882 EXPORT_SYMBOL(contig_page_data);
2885 void __init free_area_init(unsigned long *zones_size)
2887 free_area_init_node(0, NODE_DATA(0), zones_size,
2888 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
2891 #ifdef CONFIG_HOTPLUG_CPU
2892 static int page_alloc_cpu_notify(struct notifier_block *self,
2893 unsigned long action, void *hcpu)
2895 int cpu = (unsigned long)hcpu;
2897 if (action == CPU_DEAD) {
2898 local_irq_disable();
2900 vm_events_fold_cpu(cpu);
2902 refresh_cpu_vm_stats(cpu);
2906 #endif /* CONFIG_HOTPLUG_CPU */
2908 void __init page_alloc_init(void)
2910 hotcpu_notifier(page_alloc_cpu_notify, 0);
2914 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
2915 * or min_free_kbytes changes.
2917 static void calculate_totalreserve_pages(void)
2919 struct pglist_data *pgdat;
2920 unsigned long reserve_pages = 0;
2921 enum zone_type i, j;
2923 for_each_online_pgdat(pgdat) {
2924 for (i = 0; i < MAX_NR_ZONES; i++) {
2925 struct zone *zone = pgdat->node_zones + i;
2926 unsigned long max = 0;
2928 /* Find valid and maximum lowmem_reserve in the zone */
2929 for (j = i; j < MAX_NR_ZONES; j++) {
2930 if (zone->lowmem_reserve[j] > max)
2931 max = zone->lowmem_reserve[j];
2934 /* we treat pages_high as reserved pages. */
2935 max += zone->pages_high;
2937 if (max > zone->present_pages)
2938 max = zone->present_pages;
2939 reserve_pages += max;
2942 totalreserve_pages = reserve_pages;
2946 * setup_per_zone_lowmem_reserve - called whenever
2947 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
2948 * has a correct pages reserved value, so an adequate number of
2949 * pages are left in the zone after a successful __alloc_pages().
2951 static void setup_per_zone_lowmem_reserve(void)
2953 struct pglist_data *pgdat;
2954 enum zone_type j, idx;
2956 for_each_online_pgdat(pgdat) {
2957 for (j = 0; j < MAX_NR_ZONES; j++) {
2958 struct zone *zone = pgdat->node_zones + j;
2959 unsigned long present_pages = zone->present_pages;
2961 zone->lowmem_reserve[j] = 0;
2965 struct zone *lower_zone;
2969 if (sysctl_lowmem_reserve_ratio[idx] < 1)
2970 sysctl_lowmem_reserve_ratio[idx] = 1;
2972 lower_zone = pgdat->node_zones + idx;
2973 lower_zone->lowmem_reserve[j] = present_pages /
2974 sysctl_lowmem_reserve_ratio[idx];
2975 present_pages += lower_zone->present_pages;
2980 /* update totalreserve_pages */
2981 calculate_totalreserve_pages();
2985 * setup_per_zone_pages_min - called when min_free_kbytes changes.
2987 * Ensures that the pages_{min,low,high} values for each zone are set correctly
2988 * with respect to min_free_kbytes.
2990 void setup_per_zone_pages_min(void)
2992 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
2993 unsigned long lowmem_pages = 0;
2995 unsigned long flags;
2997 /* Calculate total number of !ZONE_HIGHMEM pages */
2998 for_each_zone(zone) {
2999 if (!is_highmem(zone))
3000 lowmem_pages += zone->present_pages;
3003 for_each_zone(zone) {
3006 spin_lock_irqsave(&zone->lru_lock, flags);
3007 tmp = (u64)pages_min * zone->present_pages;
3008 do_div(tmp, lowmem_pages);
3009 if (is_highmem(zone)) {
3011 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
3012 * need highmem pages, so cap pages_min to a small
3015 * The (pages_high-pages_low) and (pages_low-pages_min)
3016 * deltas controls asynch page reclaim, and so should
3017 * not be capped for highmem.
3021 min_pages = zone->present_pages / 1024;
3022 if (min_pages < SWAP_CLUSTER_MAX)
3023 min_pages = SWAP_CLUSTER_MAX;
3024 if (min_pages > 128)
3026 zone->pages_min = min_pages;
3029 * If it's a lowmem zone, reserve a number of pages
3030 * proportionate to the zone's size.
3032 zone->pages_min = tmp;
3035 zone->pages_low = zone->pages_min + (tmp >> 2);
3036 zone->pages_high = zone->pages_min + (tmp >> 1);
3037 spin_unlock_irqrestore(&zone->lru_lock, flags);
3040 /* update totalreserve_pages */
3041 calculate_totalreserve_pages();
3045 * Initialise min_free_kbytes.
3047 * For small machines we want it small (128k min). For large machines
3048 * we want it large (64MB max). But it is not linear, because network
3049 * bandwidth does not increase linearly with machine size. We use
3051 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
3052 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
3068 static int __init init_per_zone_pages_min(void)
3070 unsigned long lowmem_kbytes;
3072 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
3074 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
3075 if (min_free_kbytes < 128)
3076 min_free_kbytes = 128;
3077 if (min_free_kbytes > 65536)
3078 min_free_kbytes = 65536;
3079 setup_per_zone_pages_min();
3080 setup_per_zone_lowmem_reserve();
3083 module_init(init_per_zone_pages_min)
3086 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
3087 * that we can call two helper functions whenever min_free_kbytes
3090 int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
3091 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
3093 proc_dointvec(table, write, file, buffer, length, ppos);
3094 setup_per_zone_pages_min();
3099 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
3100 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
3105 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
3110 zone->min_unmapped_pages = (zone->present_pages *
3111 sysctl_min_unmapped_ratio) / 100;
3115 int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
3116 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
3121 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
3126 zone->min_slab_pages = (zone->present_pages *
3127 sysctl_min_slab_ratio) / 100;
3133 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
3134 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
3135 * whenever sysctl_lowmem_reserve_ratio changes.
3137 * The reserve ratio obviously has absolutely no relation with the
3138 * pages_min watermarks. The lowmem reserve ratio can only make sense
3139 * if in function of the boot time zone sizes.
3141 int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
3142 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
3144 proc_dointvec_minmax(table, write, file, buffer, length, ppos);
3145 setup_per_zone_lowmem_reserve();
3150 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
3151 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
3152 * can have before it gets flushed back to buddy allocator.
3155 int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
3156 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
3162 ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
3163 if (!write || (ret == -EINVAL))
3165 for_each_zone(zone) {
3166 for_each_online_cpu(cpu) {
3168 high = zone->present_pages / percpu_pagelist_fraction;
3169 setup_pagelist_highmark(zone_pcp(zone, cpu), high);
3175 int hashdist = HASHDIST_DEFAULT;
3178 static int __init set_hashdist(char *str)
3182 hashdist = simple_strtoul(str, &str, 0);
3185 __setup("hashdist=", set_hashdist);
3189 * allocate a large system hash table from bootmem
3190 * - it is assumed that the hash table must contain an exact power-of-2
3191 * quantity of entries
3192 * - limit is the number of hash buckets, not the total allocation size
3194 void *__init alloc_large_system_hash(const char *tablename,
3195 unsigned long bucketsize,
3196 unsigned long numentries,
3199 unsigned int *_hash_shift,
3200 unsigned int *_hash_mask,
3201 unsigned long limit)
3203 unsigned long long max = limit;
3204 unsigned long log2qty, size;
3207 /* allow the kernel cmdline to have a say */
3209 /* round applicable memory size up to nearest megabyte */
3210 numentries = (flags & HASH_HIGHMEM) ? nr_all_pages : nr_kernel_pages;
3211 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
3212 numentries >>= 20 - PAGE_SHIFT;
3213 numentries <<= 20 - PAGE_SHIFT;
3215 /* limit to 1 bucket per 2^scale bytes of low memory */
3216 if (scale > PAGE_SHIFT)
3217 numentries >>= (scale - PAGE_SHIFT);
3219 numentries <<= (PAGE_SHIFT - scale);
3221 numentries = roundup_pow_of_two(numentries);
3223 /* limit allocation size to 1/16 total memory by default */
3225 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
3226 do_div(max, bucketsize);
3229 if (numentries > max)
3232 log2qty = long_log2(numentries);
3235 size = bucketsize << log2qty;
3236 if (flags & HASH_EARLY)
3237 table = alloc_bootmem(size);
3239 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
3241 unsigned long order;
3242 for (order = 0; ((1UL << order) << PAGE_SHIFT) < size; order++)
3244 table = (void*) __get_free_pages(GFP_ATOMIC, order);
3246 } while (!table && size > PAGE_SIZE && --log2qty);
3249 panic("Failed to allocate %s hash table\n", tablename);
3251 printk("%s hash table entries: %d (order: %d, %lu bytes)\n",
3254 long_log2(size) - PAGE_SHIFT,
3258 *_hash_shift = log2qty;
3260 *_hash_mask = (1 << log2qty) - 1;
3265 #ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE
3266 struct page *pfn_to_page(unsigned long pfn)
3268 return __pfn_to_page(pfn);
3270 unsigned long page_to_pfn(struct page *page)
3272 return __page_to_pfn(page);
3274 EXPORT_SYMBOL(pfn_to_page);
3275 EXPORT_SYMBOL(page_to_pfn);
3276 #endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */
3278 #if MAX_NUMNODES > 1
3280 * Find the highest possible node id.
3282 int highest_possible_node_id(void)
3285 unsigned int highest = 0;
3287 for_each_node_mask(node, node_possible_map)
3291 EXPORT_SYMBOL(highest_possible_node_id);