2 * linux/mm/page_alloc.c
4 * Manages the free list, the system allocates free pages here.
5 * Note that kmalloc() lives in slab.c
7 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
8 * Swap reorganised 29.12.95, Stephen Tweedie
9 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
10 * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999
11 * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
12 * Zone balancing, Kanoj Sarcar, SGI, Jan 2000
13 * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002
14 * (lots of bits borrowed from Ingo Molnar & Andrew Morton)
17 #include <linux/stddef.h>
19 #include <linux/swap.h>
20 #include <linux/interrupt.h>
21 #include <linux/pagemap.h>
22 #include <linux/bootmem.h>
23 #include <linux/compiler.h>
24 #include <linux/kernel.h>
25 #include <linux/module.h>
26 #include <linux/suspend.h>
27 #include <linux/pagevec.h>
28 #include <linux/blkdev.h>
29 #include <linux/slab.h>
30 #include <linux/notifier.h>
31 #include <linux/topology.h>
32 #include <linux/sysctl.h>
33 #include <linux/cpu.h>
34 #include <linux/cpuset.h>
35 #include <linux/memory_hotplug.h>
36 #include <linux/nodemask.h>
37 #include <linux/vmalloc.h>
38 #include <linux/mempolicy.h>
39 #include <linux/stop_machine.h>
40 #include <linux/sort.h>
41 #include <linux/pfn.h>
42 #include <linux/backing-dev.h>
43 #include <linux/fault-inject.h>
44 #include <linux/vs_base.h>
45 #include <linux/vs_limit.h>
47 #include <asm/tlbflush.h>
48 #include <asm/div64.h>
52 * MCD - HACK: Find somewhere to initialize this EARLY, or make this
55 nodemask_t node_online_map __read_mostly = { { [0] = 1UL } };
56 EXPORT_SYMBOL(node_online_map);
57 nodemask_t node_possible_map __read_mostly = NODE_MASK_ALL;
58 EXPORT_SYMBOL(node_possible_map);
59 unsigned long totalram_pages __read_mostly;
60 unsigned long totalreserve_pages __read_mostly;
62 int percpu_pagelist_fraction;
64 static void __free_pages_ok(struct page *page, unsigned int order);
67 * results with 256, 32 in the lowmem_reserve sysctl:
68 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
69 * 1G machine -> (16M dma, 784M normal, 224M high)
70 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
71 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
72 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
74 * TBD: should special case ZONE_DMA32 machines here - in those we normally
75 * don't need any ZONE_NORMAL reservation
77 int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
79 #ifdef CONFIG_ZONE_DMA32
87 EXPORT_SYMBOL(totalram_pages);
89 static char * const zone_names[MAX_NR_ZONES] = {
91 #ifdef CONFIG_ZONE_DMA32
100 int min_free_kbytes = 1024;
102 unsigned long __meminitdata nr_kernel_pages;
103 unsigned long __meminitdata nr_all_pages;
104 static unsigned long __initdata dma_reserve;
106 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
108 * MAX_ACTIVE_REGIONS determines the maxmimum number of distinct
109 * ranges of memory (RAM) that may be registered with add_active_range().
110 * Ranges passed to add_active_range() will be merged if possible
111 * so the number of times add_active_range() can be called is
112 * related to the number of nodes and the number of holes
114 #ifdef CONFIG_MAX_ACTIVE_REGIONS
115 /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
116 #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
118 #if MAX_NUMNODES >= 32
119 /* If there can be many nodes, allow up to 50 holes per node */
120 #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
122 /* By default, allow up to 256 distinct regions */
123 #define MAX_ACTIVE_REGIONS 256
127 struct node_active_region __initdata early_node_map[MAX_ACTIVE_REGIONS];
128 int __initdata nr_nodemap_entries;
129 unsigned long __initdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
130 unsigned long __initdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
131 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
132 unsigned long __initdata node_boundary_start_pfn[MAX_NUMNODES];
133 unsigned long __initdata node_boundary_end_pfn[MAX_NUMNODES];
134 #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
135 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
137 #ifdef CONFIG_DEBUG_VM
138 static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
142 unsigned long pfn = page_to_pfn(page);
145 seq = zone_span_seqbegin(zone);
146 if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
148 else if (pfn < zone->zone_start_pfn)
150 } while (zone_span_seqretry(zone, seq));
155 static int page_is_consistent(struct zone *zone, struct page *page)
157 #ifdef CONFIG_HOLES_IN_ZONE
158 if (!pfn_valid(page_to_pfn(page)))
161 if (zone != page_zone(page))
167 * Temporary debugging check for pages not lying within a given zone.
169 static int bad_range(struct zone *zone, struct page *page)
171 if (page_outside_zone_boundaries(zone, page))
173 if (!page_is_consistent(zone, page))
179 static inline int bad_range(struct zone *zone, struct page *page)
185 static void bad_page(struct page *page)
187 printk(KERN_EMERG "Bad page state in process '%s'\n"
188 KERN_EMERG "page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d (%s)\n"
189 KERN_EMERG "Trying to fix it up, but a reboot is needed\n"
190 KERN_EMERG "Backtrace:\n",
191 current->comm, page, (int)(2*sizeof(unsigned long)),
192 (unsigned long)page->flags, page->mapping,
193 page_mapcount(page), page_count(page), print_tainted());
195 page->flags &= ~(1 << PG_lru |
205 set_page_count(page, 0);
206 reset_page_mapcount(page);
207 page->mapping = NULL;
208 add_taint(TAINT_BAD_PAGE);
212 * Higher-order pages are called "compound pages". They are structured thusly:
214 * The first PAGE_SIZE page is called the "head page".
216 * The remaining PAGE_SIZE pages are called "tail pages".
218 * All pages have PG_compound set. All pages have their ->private pointing at
219 * the head page (even the head page has this).
221 * The first tail page's ->lru.next holds the address of the compound page's
222 * put_page() function. Its ->lru.prev holds the order of allocation.
223 * This usage means that zero-order pages may not be compound.
226 static void free_compound_page(struct page *page)
228 __free_pages_ok(page, (unsigned long)page[1].lru.prev);
231 static void prep_compound_page(struct page *page, unsigned long order)
234 int nr_pages = 1 << order;
236 set_compound_page_dtor(page, free_compound_page);
237 page[1].lru.prev = (void *)order;
238 for (i = 0; i < nr_pages; i++) {
239 struct page *p = page + i;
241 __SetPageCompound(p);
242 set_page_private(p, (unsigned long)page);
246 static void destroy_compound_page(struct page *page, unsigned long order)
249 int nr_pages = 1 << order;
251 if (unlikely((unsigned long)page[1].lru.prev != order))
254 for (i = 0; i < nr_pages; i++) {
255 struct page *p = page + i;
257 if (unlikely(!PageCompound(p) |
258 (page_private(p) != (unsigned long)page)))
260 __ClearPageCompound(p);
264 static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
268 VM_BUG_ON((gfp_flags & (__GFP_WAIT | __GFP_HIGHMEM)) == __GFP_HIGHMEM);
270 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
271 * and __GFP_HIGHMEM from hard or soft interrupt context.
273 VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
274 for (i = 0; i < (1 << order); i++)
275 clear_highpage(page + i);
279 * function for dealing with page's order in buddy system.
280 * zone->lock is already acquired when we use these.
281 * So, we don't need atomic page->flags operations here.
283 static inline unsigned long page_order(struct page *page)
285 return page_private(page);
288 static inline void set_page_order(struct page *page, int order)
290 set_page_private(page, order);
291 __SetPageBuddy(page);
294 static inline void rmv_page_order(struct page *page)
296 __ClearPageBuddy(page);
297 set_page_private(page, 0);
301 * Locate the struct page for both the matching buddy in our
302 * pair (buddy1) and the combined O(n+1) page they form (page).
304 * 1) Any buddy B1 will have an order O twin B2 which satisfies
305 * the following equation:
307 * For example, if the starting buddy (buddy2) is #8 its order
309 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
311 * 2) Any buddy B will have an order O+1 parent P which
312 * satisfies the following equation:
315 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
317 static inline struct page *
318 __page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order)
320 unsigned long buddy_idx = page_idx ^ (1 << order);
322 return page + (buddy_idx - page_idx);
325 static inline unsigned long
326 __find_combined_index(unsigned long page_idx, unsigned int order)
328 return (page_idx & ~(1 << order));
332 * This function checks whether a page is free && is the buddy
333 * we can do coalesce a page and its buddy if
334 * (a) the buddy is not in a hole &&
335 * (b) the buddy is in the buddy system &&
336 * (c) a page and its buddy have the same order &&
337 * (d) a page and its buddy are in the same zone.
339 * For recording whether a page is in the buddy system, we use PG_buddy.
340 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
342 * For recording page's order, we use page_private(page).
344 static inline int page_is_buddy(struct page *page, struct page *buddy,
347 #ifdef CONFIG_HOLES_IN_ZONE
348 if (!pfn_valid(page_to_pfn(buddy)))
352 if (page_zone_id(page) != page_zone_id(buddy))
355 if (PageBuddy(buddy) && page_order(buddy) == order) {
356 BUG_ON(page_count(buddy) != 0);
363 * Freeing function for a buddy system allocator.
365 * The concept of a buddy system is to maintain direct-mapped table
366 * (containing bit values) for memory blocks of various "orders".
367 * The bottom level table contains the map for the smallest allocatable
368 * units of memory (here, pages), and each level above it describes
369 * pairs of units from the levels below, hence, "buddies".
370 * At a high level, all that happens here is marking the table entry
371 * at the bottom level available, and propagating the changes upward
372 * as necessary, plus some accounting needed to play nicely with other
373 * parts of the VM system.
374 * At each level, we keep a list of pages, which are heads of continuous
375 * free pages of length of (1 << order) and marked with PG_buddy. Page's
376 * order is recorded in page_private(page) field.
377 * So when we are allocating or freeing one, we can derive the state of the
378 * other. That is, if we allocate a small block, and both were
379 * free, the remainder of the region must be split into blocks.
380 * If a block is freed, and its buddy is also free, then this
381 * triggers coalescing into a block of larger size.
386 static inline void __free_one_page(struct page *page,
387 struct zone *zone, unsigned int order)
389 unsigned long page_idx;
390 int order_size = 1 << order;
392 if (unlikely(PageCompound(page)))
393 destroy_compound_page(page, order);
395 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
397 VM_BUG_ON(page_idx & (order_size - 1));
398 VM_BUG_ON(bad_range(zone, page));
400 zone->free_pages += order_size;
401 while (order < MAX_ORDER-1) {
402 unsigned long combined_idx;
403 struct free_area *area;
406 buddy = __page_find_buddy(page, page_idx, order);
407 if (!page_is_buddy(page, buddy, order))
408 break; /* Move the buddy up one level. */
410 list_del(&buddy->lru);
411 area = zone->free_area + order;
413 rmv_page_order(buddy);
414 combined_idx = __find_combined_index(page_idx, order);
415 page = page + (combined_idx - page_idx);
416 page_idx = combined_idx;
419 set_page_order(page, order);
420 list_add(&page->lru, &zone->free_area[order].free_list);
421 zone->free_area[order].nr_free++;
424 static inline int free_pages_check(struct page *page)
426 if (unlikely(page_mapcount(page) |
427 (page->mapping != NULL) |
428 (page_count(page) != 0) |
442 __ClearPageDirty(page);
444 * For now, we report if PG_reserved was found set, but do not
445 * clear it, and do not free the page. But we shall soon need
446 * to do more, for when the ZERO_PAGE count wraps negative.
448 return PageReserved(page);
452 * Frees a list of pages.
453 * Assumes all pages on list are in same zone, and of same order.
454 * count is the number of pages to free.
456 * If the zone was previously in an "all pages pinned" state then look to
457 * see if this freeing clears that state.
459 * And clear the zone's pages_scanned counter, to hold off the "all pages are
460 * pinned" detection logic.
462 static void free_pages_bulk(struct zone *zone, int count,
463 struct list_head *list, int order)
465 spin_lock(&zone->lock);
466 zone->all_unreclaimable = 0;
467 zone->pages_scanned = 0;
471 VM_BUG_ON(list_empty(list));
472 page = list_entry(list->prev, struct page, lru);
473 /* have to delete it as __free_one_page list manipulates */
474 list_del(&page->lru);
475 __free_one_page(page, zone, order);
477 spin_unlock(&zone->lock);
480 static void free_one_page(struct zone *zone, struct page *page, int order)
482 spin_lock(&zone->lock);
483 zone->all_unreclaimable = 0;
484 zone->pages_scanned = 0;
485 __free_one_page(page, zone, order);
486 spin_unlock(&zone->lock);
489 static void __free_pages_ok(struct page *page, unsigned int order)
495 if (arch_free_page(page, order))
497 for (i = 0 ; i < (1 << order) ; ++i)
498 reserved += free_pages_check(page + i);
502 if (!PageHighMem(page))
503 debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order);
504 kernel_map_pages(page, 1 << order, 0);
506 local_irq_save(flags);
507 __count_vm_events(PGFREE, 1 << order);
508 free_one_page(page_zone(page), page, order);
509 local_irq_restore(flags);
513 * permit the bootmem allocator to evade page validation on high-order frees
515 void fastcall __init __free_pages_bootmem(struct page *page, unsigned int order)
518 __ClearPageReserved(page);
519 set_page_count(page, 0);
520 set_page_refcounted(page);
526 for (loop = 0; loop < BITS_PER_LONG; loop++) {
527 struct page *p = &page[loop];
529 if (loop + 1 < BITS_PER_LONG)
531 __ClearPageReserved(p);
532 set_page_count(p, 0);
535 set_page_refcounted(page);
536 __free_pages(page, order);
542 * The order of subdivision here is critical for the IO subsystem.
543 * Please do not alter this order without good reasons and regression
544 * testing. Specifically, as large blocks of memory are subdivided,
545 * the order in which smaller blocks are delivered depends on the order
546 * they're subdivided in this function. This is the primary factor
547 * influencing the order in which pages are delivered to the IO
548 * subsystem according to empirical testing, and this is also justified
549 * by considering the behavior of a buddy system containing a single
550 * large block of memory acted on by a series of small allocations.
551 * This behavior is a critical factor in sglist merging's success.
555 static inline void expand(struct zone *zone, struct page *page,
556 int low, int high, struct free_area *area)
558 unsigned long size = 1 << high;
564 VM_BUG_ON(bad_range(zone, &page[size]));
565 list_add(&page[size].lru, &area->free_list);
567 set_page_order(&page[size], high);
572 * This page is about to be returned from the page allocator
574 static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
576 if (unlikely(page_mapcount(page) |
577 (page->mapping != NULL) |
578 (page_count(page) != 0) |
594 * For now, we report if PG_reserved was found set, but do not
595 * clear it, and do not allocate the page: as a safety net.
597 if (PageReserved(page))
600 page->flags &= ~(1 << PG_uptodate | 1 << PG_error |
601 1 << PG_referenced | 1 << PG_arch_1 |
602 1 << PG_checked | 1 << PG_mappedtodisk);
603 set_page_private(page, 0);
604 set_page_refcounted(page);
606 arch_alloc_page(page, order);
607 kernel_map_pages(page, 1 << order, 1);
609 if (gfp_flags & __GFP_ZERO)
610 prep_zero_page(page, order, gfp_flags);
612 if (order && (gfp_flags & __GFP_COMP))
613 prep_compound_page(page, order);
619 * Do the hard work of removing an element from the buddy allocator.
620 * Call me with the zone->lock already held.
622 static struct page *__rmqueue(struct zone *zone, unsigned int order)
624 struct free_area * area;
625 unsigned int current_order;
628 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
629 area = zone->free_area + current_order;
630 if (list_empty(&area->free_list))
633 page = list_entry(area->free_list.next, struct page, lru);
634 list_del(&page->lru);
635 rmv_page_order(page);
637 zone->free_pages -= 1UL << order;
638 expand(zone, page, order, current_order, area);
646 * Obtain a specified number of elements from the buddy allocator, all under
647 * a single hold of the lock, for efficiency. Add them to the supplied list.
648 * Returns the number of new pages which were placed at *list.
650 static int rmqueue_bulk(struct zone *zone, unsigned int order,
651 unsigned long count, struct list_head *list)
655 spin_lock(&zone->lock);
656 for (i = 0; i < count; ++i) {
657 struct page *page = __rmqueue(zone, order);
658 if (unlikely(page == NULL))
660 list_add_tail(&page->lru, list);
662 spin_unlock(&zone->lock);
668 * Called from the slab reaper to drain pagesets on a particular node that
669 * belongs to the currently executing processor.
670 * Note that this function must be called with the thread pinned to
671 * a single processor.
673 void drain_node_pages(int nodeid)
679 for (z = 0; z < MAX_NR_ZONES; z++) {
680 struct zone *zone = NODE_DATA(nodeid)->node_zones + z;
681 struct per_cpu_pageset *pset;
683 if (!populated_zone(zone))
686 pset = zone_pcp(zone, smp_processor_id());
687 for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) {
688 struct per_cpu_pages *pcp;
694 local_irq_save(flags);
695 if (pcp->count >= pcp->batch)
696 to_drain = pcp->batch;
698 to_drain = pcp->count;
699 free_pages_bulk(zone, to_drain, &pcp->list, 0);
700 pcp->count -= to_drain;
701 local_irq_restore(flags);
708 static void __drain_pages(unsigned int cpu)
714 for_each_zone(zone) {
715 struct per_cpu_pageset *pset;
717 if (!populated_zone(zone))
720 pset = zone_pcp(zone, cpu);
721 for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) {
722 struct per_cpu_pages *pcp;
725 local_irq_save(flags);
726 free_pages_bulk(zone, pcp->count, &pcp->list, 0);
728 local_irq_restore(flags);
735 void mark_free_pages(struct zone *zone)
737 unsigned long pfn, max_zone_pfn;
740 struct list_head *curr;
742 if (!zone->spanned_pages)
745 spin_lock_irqsave(&zone->lock, flags);
747 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
748 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
749 if (pfn_valid(pfn)) {
750 struct page *page = pfn_to_page(pfn);
752 if (!PageNosave(page))
753 ClearPageNosaveFree(page);
756 for (order = MAX_ORDER - 1; order >= 0; --order)
757 list_for_each(curr, &zone->free_area[order].free_list) {
760 pfn = page_to_pfn(list_entry(curr, struct page, lru));
761 for (i = 0; i < (1UL << order); i++)
762 SetPageNosaveFree(pfn_to_page(pfn + i));
765 spin_unlock_irqrestore(&zone->lock, flags);
769 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
771 void drain_local_pages(void)
775 local_irq_save(flags);
776 __drain_pages(smp_processor_id());
777 local_irq_restore(flags);
779 #endif /* CONFIG_PM */
782 * Free a 0-order page
784 static void fastcall free_hot_cold_page(struct page *page, int cold)
786 struct zone *zone = page_zone(page);
787 struct per_cpu_pages *pcp;
790 if (arch_free_page(page, 0))
794 page->mapping = NULL;
795 if (free_pages_check(page))
798 if (!PageHighMem(page))
799 debug_check_no_locks_freed(page_address(page), PAGE_SIZE);
800 arch_free_page(page, 0);
801 kernel_map_pages(page, 1, 0);
803 pcp = &zone_pcp(zone, get_cpu())->pcp[cold];
804 local_irq_save(flags);
805 __count_vm_event(PGFREE);
806 list_add(&page->lru, &pcp->list);
808 if (pcp->count >= pcp->high) {
809 free_pages_bulk(zone, pcp->batch, &pcp->list, 0);
810 pcp->count -= pcp->batch;
812 local_irq_restore(flags);
816 void fastcall free_hot_page(struct page *page)
818 free_hot_cold_page(page, 0);
821 void fastcall free_cold_page(struct page *page)
823 free_hot_cold_page(page, 1);
827 * split_page takes a non-compound higher-order page, and splits it into
828 * n (1<<order) sub-pages: page[0..n]
829 * Each sub-page must be freed individually.
831 * Note: this is probably too low level an operation for use in drivers.
832 * Please consult with lkml before using this in your driver.
834 void split_page(struct page *page, unsigned int order)
838 VM_BUG_ON(PageCompound(page));
839 VM_BUG_ON(!page_count(page));
840 for (i = 1; i < (1 << order); i++)
841 set_page_refcounted(page + i);
845 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
846 * we cheat by calling it from here, in the order > 0 path. Saves a branch
849 static struct page *buffered_rmqueue(struct zonelist *zonelist,
850 struct zone *zone, int order, gfp_t gfp_flags)
854 int cold = !!(gfp_flags & __GFP_COLD);
859 if (likely(order == 0)) {
860 struct per_cpu_pages *pcp;
862 pcp = &zone_pcp(zone, cpu)->pcp[cold];
863 local_irq_save(flags);
865 pcp->count = rmqueue_bulk(zone, 0,
866 pcp->batch, &pcp->list);
867 if (unlikely(!pcp->count))
870 page = list_entry(pcp->list.next, struct page, lru);
871 list_del(&page->lru);
874 spin_lock_irqsave(&zone->lock, flags);
875 page = __rmqueue(zone, order);
876 spin_unlock(&zone->lock);
881 __count_zone_vm_events(PGALLOC, zone, 1 << order);
882 zone_statistics(zonelist, zone);
883 local_irq_restore(flags);
886 VM_BUG_ON(bad_range(zone, page));
887 if (prep_new_page(page, order, gfp_flags))
892 local_irq_restore(flags);
897 #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
898 #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
899 #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
900 #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
901 #define ALLOC_HARDER 0x10 /* try to alloc harder */
902 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
903 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
905 #ifdef CONFIG_FAIL_PAGE_ALLOC
907 static struct fail_page_alloc_attr {
908 struct fault_attr attr;
910 u32 ignore_gfp_highmem;
913 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
915 struct dentry *ignore_gfp_highmem_file;
916 struct dentry *ignore_gfp_wait_file;
918 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
920 } fail_page_alloc = {
921 .attr = FAULT_ATTR_INITIALIZER,
922 .ignore_gfp_wait = 1,
923 .ignore_gfp_highmem = 1,
926 static int __init setup_fail_page_alloc(char *str)
928 return setup_fault_attr(&fail_page_alloc.attr, str);
930 __setup("fail_page_alloc=", setup_fail_page_alloc);
932 static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
934 if (gfp_mask & __GFP_NOFAIL)
936 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
938 if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
941 return should_fail(&fail_page_alloc.attr, 1 << order);
944 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
946 static int __init fail_page_alloc_debugfs(void)
948 mode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
952 err = init_fault_attr_dentries(&fail_page_alloc.attr,
956 dir = fail_page_alloc.attr.dentries.dir;
958 fail_page_alloc.ignore_gfp_wait_file =
959 debugfs_create_bool("ignore-gfp-wait", mode, dir,
960 &fail_page_alloc.ignore_gfp_wait);
962 fail_page_alloc.ignore_gfp_highmem_file =
963 debugfs_create_bool("ignore-gfp-highmem", mode, dir,
964 &fail_page_alloc.ignore_gfp_highmem);
966 if (!fail_page_alloc.ignore_gfp_wait_file ||
967 !fail_page_alloc.ignore_gfp_highmem_file) {
969 debugfs_remove(fail_page_alloc.ignore_gfp_wait_file);
970 debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file);
971 cleanup_fault_attr_dentries(&fail_page_alloc.attr);
977 late_initcall(fail_page_alloc_debugfs);
979 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
981 #else /* CONFIG_FAIL_PAGE_ALLOC */
983 static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
988 #endif /* CONFIG_FAIL_PAGE_ALLOC */
991 * Return 1 if free pages are above 'mark'. This takes into account the order
994 int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
995 int classzone_idx, int alloc_flags)
997 /* free_pages my go negative - that's OK */
998 long min = mark, free_pages = z->free_pages - (1 << order) + 1;
1001 if (alloc_flags & ALLOC_HIGH)
1003 if (alloc_flags & ALLOC_HARDER)
1006 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
1008 for (o = 0; o < order; o++) {
1009 /* At the next order, this order's pages become unavailable */
1010 free_pages -= z->free_area[o].nr_free << o;
1012 /* Require fewer higher order pages to be free */
1015 if (free_pages <= min)
1023 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1024 * skip over zones that are not allowed by the cpuset, or that have
1025 * been recently (in last second) found to be nearly full. See further
1026 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1027 * that have to skip over alot of full or unallowed zones.
1029 * If the zonelist cache is present in the passed in zonelist, then
1030 * returns a pointer to the allowed node mask (either the current
1031 * tasks mems_allowed, or node_online_map.)
1033 * If the zonelist cache is not available for this zonelist, does
1034 * nothing and returns NULL.
1036 * If the fullzones BITMAP in the zonelist cache is stale (more than
1037 * a second since last zap'd) then we zap it out (clear its bits.)
1039 * We hold off even calling zlc_setup, until after we've checked the
1040 * first zone in the zonelist, on the theory that most allocations will
1041 * be satisfied from that first zone, so best to examine that zone as
1042 * quickly as we can.
1044 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1046 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1047 nodemask_t *allowednodes; /* zonelist_cache approximation */
1049 zlc = zonelist->zlcache_ptr;
1053 if (jiffies - zlc->last_full_zap > 1 * HZ) {
1054 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1055 zlc->last_full_zap = jiffies;
1058 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1059 &cpuset_current_mems_allowed :
1061 return allowednodes;
1065 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1066 * if it is worth looking at further for free memory:
1067 * 1) Check that the zone isn't thought to be full (doesn't have its
1068 * bit set in the zonelist_cache fullzones BITMAP).
1069 * 2) Check that the zones node (obtained from the zonelist_cache
1070 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1071 * Return true (non-zero) if zone is worth looking at further, or
1072 * else return false (zero) if it is not.
1074 * This check -ignores- the distinction between various watermarks,
1075 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1076 * found to be full for any variation of these watermarks, it will
1077 * be considered full for up to one second by all requests, unless
1078 * we are so low on memory on all allowed nodes that we are forced
1079 * into the second scan of the zonelist.
1081 * In the second scan we ignore this zonelist cache and exactly
1082 * apply the watermarks to all zones, even it is slower to do so.
1083 * We are low on memory in the second scan, and should leave no stone
1084 * unturned looking for a free page.
1086 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zone **z,
1087 nodemask_t *allowednodes)
1089 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1090 int i; /* index of *z in zonelist zones */
1091 int n; /* node that zone *z is on */
1093 zlc = zonelist->zlcache_ptr;
1097 i = z - zonelist->zones;
1100 /* This zone is worth trying if it is allowed but not full */
1101 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1105 * Given 'z' scanning a zonelist, set the corresponding bit in
1106 * zlc->fullzones, so that subsequent attempts to allocate a page
1107 * from that zone don't waste time re-examining it.
1109 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zone **z)
1111 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1112 int i; /* index of *z in zonelist zones */
1114 zlc = zonelist->zlcache_ptr;
1118 i = z - zonelist->zones;
1120 set_bit(i, zlc->fullzones);
1123 #else /* CONFIG_NUMA */
1125 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1130 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zone **z,
1131 nodemask_t *allowednodes)
1136 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zone **z)
1139 #endif /* CONFIG_NUMA */
1142 * get_page_from_freelist goes through the zonelist trying to allocate
1145 static struct page *
1146 get_page_from_freelist(gfp_t gfp_mask, unsigned int order,
1147 struct zonelist *zonelist, int alloc_flags)
1150 struct page *page = NULL;
1151 int classzone_idx = zone_idx(zonelist->zones[0]);
1153 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1154 int zlc_active = 0; /* set if using zonelist_cache */
1155 int did_zlc_setup = 0; /* just call zlc_setup() one time */
1159 * Scan zonelist, looking for a zone with enough free.
1160 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1162 z = zonelist->zones;
1165 if (NUMA_BUILD && zlc_active &&
1166 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1169 if (unlikely(NUMA_BUILD && (gfp_mask & __GFP_THISNODE) &&
1170 zone->zone_pgdat != zonelist->zones[0]->zone_pgdat))
1172 if ((alloc_flags & ALLOC_CPUSET) &&
1173 !cpuset_zone_allowed_softwall(zone, gfp_mask))
1176 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
1178 if (alloc_flags & ALLOC_WMARK_MIN)
1179 mark = zone->pages_min;
1180 else if (alloc_flags & ALLOC_WMARK_LOW)
1181 mark = zone->pages_low;
1183 mark = zone->pages_high;
1184 if (!zone_watermark_ok(zone, order, mark,
1185 classzone_idx, alloc_flags)) {
1186 if (!zone_reclaim_mode ||
1187 !zone_reclaim(zone, gfp_mask, order))
1188 goto this_zone_full;
1192 page = buffered_rmqueue(zonelist, zone, order, gfp_mask);
1197 zlc_mark_zone_full(zonelist, z);
1199 if (NUMA_BUILD && !did_zlc_setup) {
1200 /* we do zlc_setup after the first zone is tried */
1201 allowednodes = zlc_setup(zonelist, alloc_flags);
1205 } while (*(++z) != NULL);
1207 if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) {
1208 /* Disable zlc cache for second zonelist scan */
1216 * This is the 'heart' of the zoned buddy allocator.
1218 struct page * fastcall
1219 __alloc_pages(gfp_t gfp_mask, unsigned int order,
1220 struct zonelist *zonelist)
1222 const gfp_t wait = gfp_mask & __GFP_WAIT;
1225 struct reclaim_state reclaim_state;
1226 struct task_struct *p = current;
1229 int did_some_progress;
1231 might_sleep_if(wait);
1233 if (should_fail_alloc_page(gfp_mask, order))
1237 z = zonelist->zones; /* the list of zones suitable for gfp_mask */
1239 if (unlikely(*z == NULL)) {
1240 /* Should this ever happen?? */
1244 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
1245 zonelist, ALLOC_WMARK_LOW|ALLOC_CPUSET);
1250 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1251 * __GFP_NOWARN set) should not cause reclaim since the subsystem
1252 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1253 * using a larger set of nodes after it has established that the
1254 * allowed per node queues are empty and that nodes are
1257 if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
1260 for (z = zonelist->zones; *z; z++)
1261 wakeup_kswapd(*z, order);
1264 * OK, we're below the kswapd watermark and have kicked background
1265 * reclaim. Now things get more complex, so set up alloc_flags according
1266 * to how we want to proceed.
1268 * The caller may dip into page reserves a bit more if the caller
1269 * cannot run direct reclaim, or if the caller has realtime scheduling
1270 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1271 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1273 alloc_flags = ALLOC_WMARK_MIN;
1274 if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait)
1275 alloc_flags |= ALLOC_HARDER;
1276 if (gfp_mask & __GFP_HIGH)
1277 alloc_flags |= ALLOC_HIGH;
1279 alloc_flags |= ALLOC_CPUSET;
1282 * Go through the zonelist again. Let __GFP_HIGH and allocations
1283 * coming from realtime tasks go deeper into reserves.
1285 * This is the last chance, in general, before the goto nopage.
1286 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1287 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1289 page = get_page_from_freelist(gfp_mask, order, zonelist, alloc_flags);
1293 /* This allocation should allow future memory freeing. */
1296 if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE)))
1297 && !in_interrupt()) {
1298 if (!(gfp_mask & __GFP_NOMEMALLOC)) {
1300 /* go through the zonelist yet again, ignoring mins */
1301 page = get_page_from_freelist(gfp_mask, order,
1302 zonelist, ALLOC_NO_WATERMARKS);
1305 if (gfp_mask & __GFP_NOFAIL) {
1306 congestion_wait(WRITE, HZ/50);
1313 /* Atomic allocations - we can't balance anything */
1319 /* We now go into synchronous reclaim */
1320 cpuset_memory_pressure_bump();
1321 p->flags |= PF_MEMALLOC;
1322 reclaim_state.reclaimed_slab = 0;
1323 p->reclaim_state = &reclaim_state;
1325 did_some_progress = try_to_free_pages(zonelist->zones, gfp_mask);
1327 p->reclaim_state = NULL;
1328 p->flags &= ~PF_MEMALLOC;
1332 if (likely(did_some_progress)) {
1333 page = get_page_from_freelist(gfp_mask, order,
1334 zonelist, alloc_flags);
1337 } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
1339 * Go through the zonelist yet one more time, keep
1340 * very high watermark here, this is only to catch
1341 * a parallel oom killing, we must fail if we're still
1342 * under heavy pressure.
1344 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
1345 zonelist, ALLOC_WMARK_HIGH|ALLOC_CPUSET);
1349 out_of_memory(zonelist, gfp_mask, order);
1354 * Don't let big-order allocations loop unless the caller explicitly
1355 * requests that. Wait for some write requests to complete then retry.
1357 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
1358 * <= 3, but that may not be true in other implementations.
1361 if (!(gfp_mask & __GFP_NORETRY)) {
1362 if ((order <= 3) || (gfp_mask & __GFP_REPEAT))
1364 if (gfp_mask & __GFP_NOFAIL)
1368 congestion_wait(WRITE, HZ/50);
1373 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
1374 printk(KERN_WARNING "%s: page allocation failure."
1375 " order:%d, mode:0x%x\n",
1376 p->comm, order, gfp_mask);
1384 EXPORT_SYMBOL(__alloc_pages);
1387 * Common helper functions.
1389 fastcall unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
1392 page = alloc_pages(gfp_mask, order);
1395 return (unsigned long) page_address(page);
1398 EXPORT_SYMBOL(__get_free_pages);
1400 fastcall unsigned long get_zeroed_page(gfp_t gfp_mask)
1405 * get_zeroed_page() returns a 32-bit address, which cannot represent
1408 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
1410 page = alloc_pages(gfp_mask | __GFP_ZERO, 0);
1412 return (unsigned long) page_address(page);
1416 EXPORT_SYMBOL(get_zeroed_page);
1418 void __pagevec_free(struct pagevec *pvec)
1420 int i = pagevec_count(pvec);
1423 free_hot_cold_page(pvec->pages[i], pvec->cold);
1426 fastcall void __free_pages(struct page *page, unsigned int order)
1428 if (put_page_testzero(page)) {
1430 free_hot_page(page);
1432 __free_pages_ok(page, order);
1436 EXPORT_SYMBOL(__free_pages);
1438 fastcall void free_pages(unsigned long addr, unsigned int order)
1441 VM_BUG_ON(!virt_addr_valid((void *)addr));
1442 __free_pages(virt_to_page((void *)addr), order);
1446 EXPORT_SYMBOL(free_pages);
1449 * Total amount of free (allocatable) RAM:
1451 unsigned int nr_free_pages(void)
1453 unsigned int sum = 0;
1457 sum += zone->free_pages;
1462 EXPORT_SYMBOL(nr_free_pages);
1465 unsigned int nr_free_pages_pgdat(pg_data_t *pgdat)
1467 unsigned int sum = 0;
1470 for (i = 0; i < MAX_NR_ZONES; i++)
1471 sum += pgdat->node_zones[i].free_pages;
1477 static unsigned int nr_free_zone_pages(int offset)
1479 /* Just pick one node, since fallback list is circular */
1480 pg_data_t *pgdat = NODE_DATA(numa_node_id());
1481 unsigned int sum = 0;
1483 struct zonelist *zonelist = pgdat->node_zonelists + offset;
1484 struct zone **zonep = zonelist->zones;
1487 for (zone = *zonep++; zone; zone = *zonep++) {
1488 unsigned long size = zone->present_pages;
1489 unsigned long high = zone->pages_high;
1498 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1500 unsigned int nr_free_buffer_pages(void)
1502 return nr_free_zone_pages(gfp_zone(GFP_USER));
1506 * Amount of free RAM allocatable within all zones
1508 unsigned int nr_free_pagecache_pages(void)
1510 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER));
1513 static inline void show_node(struct zone *zone)
1516 printk("Node %d ", zone_to_nid(zone));
1519 void si_meminfo(struct sysinfo *val)
1521 val->totalram = totalram_pages;
1523 val->freeram = nr_free_pages();
1524 val->bufferram = nr_blockdev_pages();
1525 val->totalhigh = totalhigh_pages;
1526 val->freehigh = nr_free_highpages();
1527 val->mem_unit = PAGE_SIZE;
1529 if (vx_flags(VXF_VIRT_MEM, 0))
1530 vx_vsi_meminfo(val);
1533 EXPORT_SYMBOL(si_meminfo);
1536 void si_meminfo_node(struct sysinfo *val, int nid)
1538 pg_data_t *pgdat = NODE_DATA(nid);
1540 val->totalram = pgdat->node_present_pages;
1541 val->freeram = nr_free_pages_pgdat(pgdat);
1542 #ifdef CONFIG_HIGHMEM
1543 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1544 val->freehigh = pgdat->node_zones[ZONE_HIGHMEM].free_pages;
1549 val->mem_unit = PAGE_SIZE;
1551 if (vx_flags(VXF_VIRT_MEM, 0))
1552 vx_vsi_meminfo(val);
1556 #define K(x) ((x) << (PAGE_SHIFT-10))
1559 * Show free area list (used inside shift_scroll-lock stuff)
1560 * We also calculate the percentage fragmentation. We do this by counting the
1561 * memory on each free list with the exception of the first item on the list.
1563 void show_free_areas(void)
1566 unsigned long active;
1567 unsigned long inactive;
1571 for_each_zone(zone) {
1572 if (!populated_zone(zone))
1576 printk("%s per-cpu:\n", zone->name);
1578 for_each_online_cpu(cpu) {
1579 struct per_cpu_pageset *pageset;
1581 pageset = zone_pcp(zone, cpu);
1583 printk("CPU %4d: Hot: hi:%5d, btch:%4d usd:%4d "
1584 "Cold: hi:%5d, btch:%4d usd:%4d\n",
1585 cpu, pageset->pcp[0].high,
1586 pageset->pcp[0].batch, pageset->pcp[0].count,
1587 pageset->pcp[1].high, pageset->pcp[1].batch,
1588 pageset->pcp[1].count);
1592 get_zone_counts(&active, &inactive, &free);
1594 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu "
1595 "unstable:%lu free:%u slab:%lu mapped:%lu pagetables:%lu\n",
1598 global_page_state(NR_FILE_DIRTY),
1599 global_page_state(NR_WRITEBACK),
1600 global_page_state(NR_UNSTABLE_NFS),
1602 global_page_state(NR_SLAB_RECLAIMABLE) +
1603 global_page_state(NR_SLAB_UNRECLAIMABLE),
1604 global_page_state(NR_FILE_MAPPED),
1605 global_page_state(NR_PAGETABLE));
1607 for_each_zone(zone) {
1610 if (!populated_zone(zone))
1622 " pages_scanned:%lu"
1623 " all_unreclaimable? %s"
1626 K(zone->free_pages),
1629 K(zone->pages_high),
1631 K(zone->nr_inactive),
1632 K(zone->present_pages),
1633 zone->pages_scanned,
1634 (zone->all_unreclaimable ? "yes" : "no")
1636 printk("lowmem_reserve[]:");
1637 for (i = 0; i < MAX_NR_ZONES; i++)
1638 printk(" %lu", zone->lowmem_reserve[i]);
1642 for_each_zone(zone) {
1643 unsigned long nr[MAX_ORDER], flags, order, total = 0;
1645 if (!populated_zone(zone))
1649 printk("%s: ", zone->name);
1651 spin_lock_irqsave(&zone->lock, flags);
1652 for (order = 0; order < MAX_ORDER; order++) {
1653 nr[order] = zone->free_area[order].nr_free;
1654 total += nr[order] << order;
1656 spin_unlock_irqrestore(&zone->lock, flags);
1657 for (order = 0; order < MAX_ORDER; order++)
1658 printk("%lu*%lukB ", nr[order], K(1UL) << order);
1659 printk("= %lukB\n", K(total));
1662 show_swap_cache_info();
1666 * Builds allocation fallback zone lists.
1668 * Add all populated zones of a node to the zonelist.
1670 static int __meminit build_zonelists_node(pg_data_t *pgdat,
1671 struct zonelist *zonelist, int nr_zones, enum zone_type zone_type)
1675 BUG_ON(zone_type >= MAX_NR_ZONES);
1680 zone = pgdat->node_zones + zone_type;
1681 if (populated_zone(zone)) {
1682 zonelist->zones[nr_zones++] = zone;
1683 check_highest_zone(zone_type);
1686 } while (zone_type);
1691 #define MAX_NODE_LOAD (num_online_nodes())
1692 static int __meminitdata node_load[MAX_NUMNODES];
1694 * find_next_best_node - find the next node that should appear in a given node's fallback list
1695 * @node: node whose fallback list we're appending
1696 * @used_node_mask: nodemask_t of already used nodes
1698 * We use a number of factors to determine which is the next node that should
1699 * appear on a given node's fallback list. The node should not have appeared
1700 * already in @node's fallback list, and it should be the next closest node
1701 * according to the distance array (which contains arbitrary distance values
1702 * from each node to each node in the system), and should also prefer nodes
1703 * with no CPUs, since presumably they'll have very little allocation pressure
1704 * on them otherwise.
1705 * It returns -1 if no node is found.
1707 static int __meminit find_next_best_node(int node, nodemask_t *used_node_mask)
1710 int min_val = INT_MAX;
1713 /* Use the local node if we haven't already */
1714 if (!node_isset(node, *used_node_mask)) {
1715 node_set(node, *used_node_mask);
1719 for_each_online_node(n) {
1722 /* Don't want a node to appear more than once */
1723 if (node_isset(n, *used_node_mask))
1726 /* Use the distance array to find the distance */
1727 val = node_distance(node, n);
1729 /* Penalize nodes under us ("prefer the next node") */
1732 /* Give preference to headless and unused nodes */
1733 tmp = node_to_cpumask(n);
1734 if (!cpus_empty(tmp))
1735 val += PENALTY_FOR_NODE_WITH_CPUS;
1737 /* Slight preference for less loaded node */
1738 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
1739 val += node_load[n];
1741 if (val < min_val) {
1748 node_set(best_node, *used_node_mask);
1753 static void __meminit build_zonelists(pg_data_t *pgdat)
1755 int j, node, local_node;
1757 int prev_node, load;
1758 struct zonelist *zonelist;
1759 nodemask_t used_mask;
1761 /* initialize zonelists */
1762 for (i = 0; i < MAX_NR_ZONES; i++) {
1763 zonelist = pgdat->node_zonelists + i;
1764 zonelist->zones[0] = NULL;
1767 /* NUMA-aware ordering of nodes */
1768 local_node = pgdat->node_id;
1769 load = num_online_nodes();
1770 prev_node = local_node;
1771 nodes_clear(used_mask);
1772 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
1773 int distance = node_distance(local_node, node);
1776 * If another node is sufficiently far away then it is better
1777 * to reclaim pages in a zone before going off node.
1779 if (distance > RECLAIM_DISTANCE)
1780 zone_reclaim_mode = 1;
1783 * We don't want to pressure a particular node.
1784 * So adding penalty to the first node in same
1785 * distance group to make it round-robin.
1788 if (distance != node_distance(local_node, prev_node))
1789 node_load[node] += load;
1792 for (i = 0; i < MAX_NR_ZONES; i++) {
1793 zonelist = pgdat->node_zonelists + i;
1794 for (j = 0; zonelist->zones[j] != NULL; j++);
1796 j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
1797 zonelist->zones[j] = NULL;
1802 /* Construct the zonelist performance cache - see further mmzone.h */
1803 static void __meminit build_zonelist_cache(pg_data_t *pgdat)
1807 for (i = 0; i < MAX_NR_ZONES; i++) {
1808 struct zonelist *zonelist;
1809 struct zonelist_cache *zlc;
1812 zonelist = pgdat->node_zonelists + i;
1813 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
1814 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1815 for (z = zonelist->zones; *z; z++)
1816 zlc->z_to_n[z - zonelist->zones] = zone_to_nid(*z);
1820 #else /* CONFIG_NUMA */
1822 static void __meminit build_zonelists(pg_data_t *pgdat)
1824 int node, local_node;
1827 local_node = pgdat->node_id;
1828 for (i = 0; i < MAX_NR_ZONES; i++) {
1829 struct zonelist *zonelist;
1831 zonelist = pgdat->node_zonelists + i;
1833 j = build_zonelists_node(pgdat, zonelist, 0, i);
1835 * Now we build the zonelist so that it contains the zones
1836 * of all the other nodes.
1837 * We don't want to pressure a particular node, so when
1838 * building the zones for node N, we make sure that the
1839 * zones coming right after the local ones are those from
1840 * node N+1 (modulo N)
1842 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
1843 if (!node_online(node))
1845 j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
1847 for (node = 0; node < local_node; node++) {
1848 if (!node_online(node))
1850 j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
1853 zonelist->zones[j] = NULL;
1857 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
1858 static void __meminit build_zonelist_cache(pg_data_t *pgdat)
1862 for (i = 0; i < MAX_NR_ZONES; i++)
1863 pgdat->node_zonelists[i].zlcache_ptr = NULL;
1866 #endif /* CONFIG_NUMA */
1868 /* return values int ....just for stop_machine_run() */
1869 static int __meminit __build_all_zonelists(void *dummy)
1873 for_each_online_node(nid) {
1874 build_zonelists(NODE_DATA(nid));
1875 build_zonelist_cache(NODE_DATA(nid));
1880 void __meminit build_all_zonelists(void)
1882 if (system_state == SYSTEM_BOOTING) {
1883 __build_all_zonelists(NULL);
1884 cpuset_init_current_mems_allowed();
1886 /* we have to stop all cpus to guaranntee there is no user
1888 stop_machine_run(__build_all_zonelists, NULL, NR_CPUS);
1889 /* cpuset refresh routine should be here */
1891 vm_total_pages = nr_free_pagecache_pages();
1892 printk("Built %i zonelists. Total pages: %ld\n",
1893 num_online_nodes(), vm_total_pages);
1897 * Helper functions to size the waitqueue hash table.
1898 * Essentially these want to choose hash table sizes sufficiently
1899 * large so that collisions trying to wait on pages are rare.
1900 * But in fact, the number of active page waitqueues on typical
1901 * systems is ridiculously low, less than 200. So this is even
1902 * conservative, even though it seems large.
1904 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
1905 * waitqueues, i.e. the size of the waitq table given the number of pages.
1907 #define PAGES_PER_WAITQUEUE 256
1909 #ifndef CONFIG_MEMORY_HOTPLUG
1910 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
1912 unsigned long size = 1;
1914 pages /= PAGES_PER_WAITQUEUE;
1916 while (size < pages)
1920 * Once we have dozens or even hundreds of threads sleeping
1921 * on IO we've got bigger problems than wait queue collision.
1922 * Limit the size of the wait table to a reasonable size.
1924 size = min(size, 4096UL);
1926 return max(size, 4UL);
1930 * A zone's size might be changed by hot-add, so it is not possible to determine
1931 * a suitable size for its wait_table. So we use the maximum size now.
1933 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
1935 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
1936 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
1937 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
1939 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
1940 * or more by the traditional way. (See above). It equals:
1942 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
1943 * ia64(16K page size) : = ( 8G + 4M)byte.
1944 * powerpc (64K page size) : = (32G +16M)byte.
1946 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
1953 * This is an integer logarithm so that shifts can be used later
1954 * to extract the more random high bits from the multiplicative
1955 * hash function before the remainder is taken.
1957 static inline unsigned long wait_table_bits(unsigned long size)
1962 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
1965 * Initially all pages are reserved - free ones are freed
1966 * up by free_all_bootmem() once the early boot process is
1967 * done. Non-atomic initialization, single-pass.
1969 void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
1970 unsigned long start_pfn, enum memmap_context context)
1973 unsigned long end_pfn = start_pfn + size;
1976 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1978 * There can be holes in boot-time mem_map[]s
1979 * handed to this function. They do not
1980 * exist on hotplugged memory.
1982 if (context == MEMMAP_EARLY) {
1983 if (!early_pfn_valid(pfn))
1985 if (!early_pfn_in_nid(pfn, nid))
1988 page = pfn_to_page(pfn);
1989 set_page_links(page, zone, nid, pfn);
1990 init_page_count(page);
1991 reset_page_mapcount(page);
1992 SetPageReserved(page);
1993 INIT_LIST_HEAD(&page->lru);
1994 #ifdef WANT_PAGE_VIRTUAL
1995 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
1996 if (!is_highmem_idx(zone))
1997 set_page_address(page, __va(pfn << PAGE_SHIFT));
2002 void zone_init_free_lists(struct pglist_data *pgdat, struct zone *zone,
2006 for (order = 0; order < MAX_ORDER ; order++) {
2007 INIT_LIST_HEAD(&zone->free_area[order].free_list);
2008 zone->free_area[order].nr_free = 0;
2012 #ifndef __HAVE_ARCH_MEMMAP_INIT
2013 #define memmap_init(size, nid, zone, start_pfn) \
2014 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
2017 static int __cpuinit zone_batchsize(struct zone *zone)
2022 * The per-cpu-pages pools are set to around 1000th of the
2023 * size of the zone. But no more than 1/2 of a meg.
2025 * OK, so we don't know how big the cache is. So guess.
2027 batch = zone->present_pages / 1024;
2028 if (batch * PAGE_SIZE > 512 * 1024)
2029 batch = (512 * 1024) / PAGE_SIZE;
2030 batch /= 4; /* We effectively *= 4 below */
2035 * Clamp the batch to a 2^n - 1 value. Having a power
2036 * of 2 value was found to be more likely to have
2037 * suboptimal cache aliasing properties in some cases.
2039 * For example if 2 tasks are alternately allocating
2040 * batches of pages, one task can end up with a lot
2041 * of pages of one half of the possible page colors
2042 * and the other with pages of the other colors.
2044 batch = (1 << (fls(batch + batch/2)-1)) - 1;
2049 inline void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
2051 struct per_cpu_pages *pcp;
2053 memset(p, 0, sizeof(*p));
2055 pcp = &p->pcp[0]; /* hot */
2057 pcp->high = 6 * batch;
2058 pcp->batch = max(1UL, 1 * batch);
2059 INIT_LIST_HEAD(&pcp->list);
2061 pcp = &p->pcp[1]; /* cold*/
2063 pcp->high = 2 * batch;
2064 pcp->batch = max(1UL, batch/2);
2065 INIT_LIST_HEAD(&pcp->list);
2069 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
2070 * to the value high for the pageset p.
2073 static void setup_pagelist_highmark(struct per_cpu_pageset *p,
2076 struct per_cpu_pages *pcp;
2078 pcp = &p->pcp[0]; /* hot list */
2080 pcp->batch = max(1UL, high/4);
2081 if ((high/4) > (PAGE_SHIFT * 8))
2082 pcp->batch = PAGE_SHIFT * 8;
2088 * Boot pageset table. One per cpu which is going to be used for all
2089 * zones and all nodes. The parameters will be set in such a way
2090 * that an item put on a list will immediately be handed over to
2091 * the buddy list. This is safe since pageset manipulation is done
2092 * with interrupts disabled.
2094 * Some NUMA counter updates may also be caught by the boot pagesets.
2096 * The boot_pagesets must be kept even after bootup is complete for
2097 * unused processors and/or zones. They do play a role for bootstrapping
2098 * hotplugged processors.
2100 * zoneinfo_show() and maybe other functions do
2101 * not check if the processor is online before following the pageset pointer.
2102 * Other parts of the kernel may not check if the zone is available.
2104 static struct per_cpu_pageset boot_pageset[NR_CPUS];
2107 * Dynamically allocate memory for the
2108 * per cpu pageset array in struct zone.
2110 static int __cpuinit process_zones(int cpu)
2112 struct zone *zone, *dzone;
2114 for_each_zone(zone) {
2116 if (!populated_zone(zone))
2119 zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset),
2120 GFP_KERNEL, cpu_to_node(cpu));
2121 if (!zone_pcp(zone, cpu))
2124 setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone));
2126 if (percpu_pagelist_fraction)
2127 setup_pagelist_highmark(zone_pcp(zone, cpu),
2128 (zone->present_pages / percpu_pagelist_fraction));
2133 for_each_zone(dzone) {
2136 kfree(zone_pcp(dzone, cpu));
2137 zone_pcp(dzone, cpu) = NULL;
2142 static inline void free_zone_pagesets(int cpu)
2146 for_each_zone(zone) {
2147 struct per_cpu_pageset *pset = zone_pcp(zone, cpu);
2149 /* Free per_cpu_pageset if it is slab allocated */
2150 if (pset != &boot_pageset[cpu])
2152 zone_pcp(zone, cpu) = NULL;
2156 static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb,
2157 unsigned long action,
2160 int cpu = (long)hcpu;
2161 int ret = NOTIFY_OK;
2164 case CPU_UP_PREPARE:
2165 if (process_zones(cpu))
2168 case CPU_UP_CANCELED:
2170 free_zone_pagesets(cpu);
2178 static struct notifier_block __cpuinitdata pageset_notifier =
2179 { &pageset_cpuup_callback, NULL, 0 };
2181 void __init setup_per_cpu_pageset(void)
2185 /* Initialize per_cpu_pageset for cpu 0.
2186 * A cpuup callback will do this for every cpu
2187 * as it comes online
2189 err = process_zones(smp_processor_id());
2191 register_cpu_notifier(&pageset_notifier);
2197 int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
2200 struct pglist_data *pgdat = zone->zone_pgdat;
2204 * The per-page waitqueue mechanism uses hashed waitqueues
2207 zone->wait_table_hash_nr_entries =
2208 wait_table_hash_nr_entries(zone_size_pages);
2209 zone->wait_table_bits =
2210 wait_table_bits(zone->wait_table_hash_nr_entries);
2211 alloc_size = zone->wait_table_hash_nr_entries
2212 * sizeof(wait_queue_head_t);
2214 if (system_state == SYSTEM_BOOTING) {
2215 zone->wait_table = (wait_queue_head_t *)
2216 alloc_bootmem_node(pgdat, alloc_size);
2219 * This case means that a zone whose size was 0 gets new memory
2220 * via memory hot-add.
2221 * But it may be the case that a new node was hot-added. In
2222 * this case vmalloc() will not be able to use this new node's
2223 * memory - this wait_table must be initialized to use this new
2224 * node itself as well.
2225 * To use this new node's memory, further consideration will be
2228 zone->wait_table = (wait_queue_head_t *)vmalloc(alloc_size);
2230 if (!zone->wait_table)
2233 for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
2234 init_waitqueue_head(zone->wait_table + i);
2239 static __meminit void zone_pcp_init(struct zone *zone)
2242 unsigned long batch = zone_batchsize(zone);
2244 for (cpu = 0; cpu < NR_CPUS; cpu++) {
2246 /* Early boot. Slab allocator not functional yet */
2247 zone_pcp(zone, cpu) = &boot_pageset[cpu];
2248 setup_pageset(&boot_pageset[cpu],0);
2250 setup_pageset(zone_pcp(zone,cpu), batch);
2253 if (zone->present_pages)
2254 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n",
2255 zone->name, zone->present_pages, batch);
2258 __meminit int init_currently_empty_zone(struct zone *zone,
2259 unsigned long zone_start_pfn,
2261 enum memmap_context context)
2263 struct pglist_data *pgdat = zone->zone_pgdat;
2265 ret = zone_wait_table_init(zone, size);
2268 pgdat->nr_zones = zone_idx(zone) + 1;
2270 zone->zone_start_pfn = zone_start_pfn;
2272 memmap_init(size, pgdat->node_id, zone_idx(zone), zone_start_pfn);
2274 zone_init_free_lists(pgdat, zone, zone->spanned_pages);
2279 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2281 * Basic iterator support. Return the first range of PFNs for a node
2282 * Note: nid == MAX_NUMNODES returns first region regardless of node
2284 static int __init first_active_region_index_in_nid(int nid)
2288 for (i = 0; i < nr_nodemap_entries; i++)
2289 if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
2296 * Basic iterator support. Return the next active range of PFNs for a node
2297 * Note: nid == MAX_NUMNODES returns next region regardles of node
2299 static int __init next_active_region_index_in_nid(int index, int nid)
2301 for (index = index + 1; index < nr_nodemap_entries; index++)
2302 if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
2308 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
2310 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
2311 * Architectures may implement their own version but if add_active_range()
2312 * was used and there are no special requirements, this is a convenient
2315 int __init early_pfn_to_nid(unsigned long pfn)
2319 for (i = 0; i < nr_nodemap_entries; i++) {
2320 unsigned long start_pfn = early_node_map[i].start_pfn;
2321 unsigned long end_pfn = early_node_map[i].end_pfn;
2323 if (start_pfn <= pfn && pfn < end_pfn)
2324 return early_node_map[i].nid;
2329 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
2331 /* Basic iterator support to walk early_node_map[] */
2332 #define for_each_active_range_index_in_nid(i, nid) \
2333 for (i = first_active_region_index_in_nid(nid); i != -1; \
2334 i = next_active_region_index_in_nid(i, nid))
2337 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
2338 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
2339 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
2341 * If an architecture guarantees that all ranges registered with
2342 * add_active_ranges() contain no holes and may be freed, this
2343 * this function may be used instead of calling free_bootmem() manually.
2345 void __init free_bootmem_with_active_regions(int nid,
2346 unsigned long max_low_pfn)
2350 for_each_active_range_index_in_nid(i, nid) {
2351 unsigned long size_pages = 0;
2352 unsigned long end_pfn = early_node_map[i].end_pfn;
2354 if (early_node_map[i].start_pfn >= max_low_pfn)
2357 if (end_pfn > max_low_pfn)
2358 end_pfn = max_low_pfn;
2360 size_pages = end_pfn - early_node_map[i].start_pfn;
2361 free_bootmem_node(NODE_DATA(early_node_map[i].nid),
2362 PFN_PHYS(early_node_map[i].start_pfn),
2363 size_pages << PAGE_SHIFT);
2368 * sparse_memory_present_with_active_regions - Call memory_present for each active range
2369 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
2371 * If an architecture guarantees that all ranges registered with
2372 * add_active_ranges() contain no holes and may be freed, this
2373 * function may be used instead of calling memory_present() manually.
2375 void __init sparse_memory_present_with_active_regions(int nid)
2379 for_each_active_range_index_in_nid(i, nid)
2380 memory_present(early_node_map[i].nid,
2381 early_node_map[i].start_pfn,
2382 early_node_map[i].end_pfn);
2386 * push_node_boundaries - Push node boundaries to at least the requested boundary
2387 * @nid: The nid of the node to push the boundary for
2388 * @start_pfn: The start pfn of the node
2389 * @end_pfn: The end pfn of the node
2391 * In reserve-based hot-add, mem_map is allocated that is unused until hotadd
2392 * time. Specifically, on x86_64, SRAT will report ranges that can potentially
2393 * be hotplugged even though no physical memory exists. This function allows
2394 * an arch to push out the node boundaries so mem_map is allocated that can
2397 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
2398 void __init push_node_boundaries(unsigned int nid,
2399 unsigned long start_pfn, unsigned long end_pfn)
2401 printk(KERN_DEBUG "Entering push_node_boundaries(%u, %lu, %lu)\n",
2402 nid, start_pfn, end_pfn);
2404 /* Initialise the boundary for this node if necessary */
2405 if (node_boundary_end_pfn[nid] == 0)
2406 node_boundary_start_pfn[nid] = -1UL;
2408 /* Update the boundaries */
2409 if (node_boundary_start_pfn[nid] > start_pfn)
2410 node_boundary_start_pfn[nid] = start_pfn;
2411 if (node_boundary_end_pfn[nid] < end_pfn)
2412 node_boundary_end_pfn[nid] = end_pfn;
2415 /* If necessary, push the node boundary out for reserve hotadd */
2416 static void __init account_node_boundary(unsigned int nid,
2417 unsigned long *start_pfn, unsigned long *end_pfn)
2419 printk(KERN_DEBUG "Entering account_node_boundary(%u, %lu, %lu)\n",
2420 nid, *start_pfn, *end_pfn);
2422 /* Return if boundary information has not been provided */
2423 if (node_boundary_end_pfn[nid] == 0)
2426 /* Check the boundaries and update if necessary */
2427 if (node_boundary_start_pfn[nid] < *start_pfn)
2428 *start_pfn = node_boundary_start_pfn[nid];
2429 if (node_boundary_end_pfn[nid] > *end_pfn)
2430 *end_pfn = node_boundary_end_pfn[nid];
2433 void __init push_node_boundaries(unsigned int nid,
2434 unsigned long start_pfn, unsigned long end_pfn) {}
2436 static void __init account_node_boundary(unsigned int nid,
2437 unsigned long *start_pfn, unsigned long *end_pfn) {}
2442 * get_pfn_range_for_nid - Return the start and end page frames for a node
2443 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
2444 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
2445 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
2447 * It returns the start and end page frame of a node based on information
2448 * provided by an arch calling add_active_range(). If called for a node
2449 * with no available memory, a warning is printed and the start and end
2452 void __init get_pfn_range_for_nid(unsigned int nid,
2453 unsigned long *start_pfn, unsigned long *end_pfn)
2459 for_each_active_range_index_in_nid(i, nid) {
2460 *start_pfn = min(*start_pfn, early_node_map[i].start_pfn);
2461 *end_pfn = max(*end_pfn, early_node_map[i].end_pfn);
2464 if (*start_pfn == -1UL) {
2465 printk(KERN_WARNING "Node %u active with no memory\n", nid);
2469 /* Push the node boundaries out if requested */
2470 account_node_boundary(nid, start_pfn, end_pfn);
2474 * Return the number of pages a zone spans in a node, including holes
2475 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
2477 unsigned long __init zone_spanned_pages_in_node(int nid,
2478 unsigned long zone_type,
2479 unsigned long *ignored)
2481 unsigned long node_start_pfn, node_end_pfn;
2482 unsigned long zone_start_pfn, zone_end_pfn;
2484 /* Get the start and end of the node and zone */
2485 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
2486 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
2487 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
2489 /* Check that this node has pages within the zone's required range */
2490 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
2493 /* Move the zone boundaries inside the node if necessary */
2494 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
2495 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
2497 /* Return the spanned pages */
2498 return zone_end_pfn - zone_start_pfn;
2502 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
2503 * then all holes in the requested range will be accounted for.
2505 unsigned long __init __absent_pages_in_range(int nid,
2506 unsigned long range_start_pfn,
2507 unsigned long range_end_pfn)
2510 unsigned long prev_end_pfn = 0, hole_pages = 0;
2511 unsigned long start_pfn;
2513 /* Find the end_pfn of the first active range of pfns in the node */
2514 i = first_active_region_index_in_nid(nid);
2518 /* Account for ranges before physical memory on this node */
2519 if (early_node_map[i].start_pfn > range_start_pfn)
2520 hole_pages = early_node_map[i].start_pfn - range_start_pfn;
2522 prev_end_pfn = early_node_map[i].start_pfn;
2524 /* Find all holes for the zone within the node */
2525 for (; i != -1; i = next_active_region_index_in_nid(i, nid)) {
2527 /* No need to continue if prev_end_pfn is outside the zone */
2528 if (prev_end_pfn >= range_end_pfn)
2531 /* Make sure the end of the zone is not within the hole */
2532 start_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
2533 prev_end_pfn = max(prev_end_pfn, range_start_pfn);
2535 /* Update the hole size cound and move on */
2536 if (start_pfn > range_start_pfn) {
2537 BUG_ON(prev_end_pfn > start_pfn);
2538 hole_pages += start_pfn - prev_end_pfn;
2540 prev_end_pfn = early_node_map[i].end_pfn;
2543 /* Account for ranges past physical memory on this node */
2544 if (range_end_pfn > prev_end_pfn)
2545 hole_pages += range_end_pfn -
2546 max(range_start_pfn, prev_end_pfn);
2552 * absent_pages_in_range - Return number of page frames in holes within a range
2553 * @start_pfn: The start PFN to start searching for holes
2554 * @end_pfn: The end PFN to stop searching for holes
2556 * It returns the number of pages frames in memory holes within a range.
2558 unsigned long __init absent_pages_in_range(unsigned long start_pfn,
2559 unsigned long end_pfn)
2561 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
2564 /* Return the number of page frames in holes in a zone on a node */
2565 unsigned long __init zone_absent_pages_in_node(int nid,
2566 unsigned long zone_type,
2567 unsigned long *ignored)
2569 unsigned long node_start_pfn, node_end_pfn;
2570 unsigned long zone_start_pfn, zone_end_pfn;
2572 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
2573 zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type],
2575 zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type],
2578 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
2582 static inline unsigned long zone_spanned_pages_in_node(int nid,
2583 unsigned long zone_type,
2584 unsigned long *zones_size)
2586 return zones_size[zone_type];
2589 static inline unsigned long zone_absent_pages_in_node(int nid,
2590 unsigned long zone_type,
2591 unsigned long *zholes_size)
2596 return zholes_size[zone_type];
2601 static void __init calculate_node_totalpages(struct pglist_data *pgdat,
2602 unsigned long *zones_size, unsigned long *zholes_size)
2604 unsigned long realtotalpages, totalpages = 0;
2607 for (i = 0; i < MAX_NR_ZONES; i++)
2608 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
2610 pgdat->node_spanned_pages = totalpages;
2612 realtotalpages = totalpages;
2613 for (i = 0; i < MAX_NR_ZONES; i++)
2615 zone_absent_pages_in_node(pgdat->node_id, i,
2617 pgdat->node_present_pages = realtotalpages;
2618 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
2623 * Set up the zone data structures:
2624 * - mark all pages reserved
2625 * - mark all memory queues empty
2626 * - clear the memory bitmaps
2628 static void __meminit free_area_init_core(struct pglist_data *pgdat,
2629 unsigned long *zones_size, unsigned long *zholes_size)
2632 int nid = pgdat->node_id;
2633 unsigned long zone_start_pfn = pgdat->node_start_pfn;
2636 pgdat_resize_init(pgdat);
2637 pgdat->nr_zones = 0;
2638 init_waitqueue_head(&pgdat->kswapd_wait);
2639 pgdat->kswapd_max_order = 0;
2641 for (j = 0; j < MAX_NR_ZONES; j++) {
2642 struct zone *zone = pgdat->node_zones + j;
2643 unsigned long size, realsize, memmap_pages;
2645 size = zone_spanned_pages_in_node(nid, j, zones_size);
2646 realsize = size - zone_absent_pages_in_node(nid, j,
2650 * Adjust realsize so that it accounts for how much memory
2651 * is used by this zone for memmap. This affects the watermark
2652 * and per-cpu initialisations
2654 memmap_pages = (size * sizeof(struct page)) >> PAGE_SHIFT;
2655 if (realsize >= memmap_pages) {
2656 realsize -= memmap_pages;
2658 " %s zone: %lu pages used for memmap\n",
2659 zone_names[j], memmap_pages);
2662 " %s zone: %lu pages exceeds realsize %lu\n",
2663 zone_names[j], memmap_pages, realsize);
2665 /* Account for reserved DMA pages */
2666 if (j == ZONE_DMA && realsize > dma_reserve) {
2667 realsize -= dma_reserve;
2668 printk(KERN_DEBUG " DMA zone: %lu pages reserved\n",
2672 if (!is_highmem_idx(j))
2673 nr_kernel_pages += realsize;
2674 nr_all_pages += realsize;
2676 zone->spanned_pages = size;
2677 zone->present_pages = realsize;
2680 zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
2682 zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100;
2684 zone->name = zone_names[j];
2685 spin_lock_init(&zone->lock);
2686 spin_lock_init(&zone->lru_lock);
2687 zone_seqlock_init(zone);
2688 zone->zone_pgdat = pgdat;
2689 zone->free_pages = 0;
2691 zone->prev_priority = DEF_PRIORITY;
2693 zone_pcp_init(zone);
2694 INIT_LIST_HEAD(&zone->active_list);
2695 INIT_LIST_HEAD(&zone->inactive_list);
2696 zone->nr_scan_active = 0;
2697 zone->nr_scan_inactive = 0;
2698 zone->nr_active = 0;
2699 zone->nr_inactive = 0;
2700 zap_zone_vm_stats(zone);
2701 atomic_set(&zone->reclaim_in_progress, 0);
2705 ret = init_currently_empty_zone(zone, zone_start_pfn,
2706 size, MEMMAP_EARLY);
2708 zone_start_pfn += size;
2712 static void __init alloc_node_mem_map(struct pglist_data *pgdat)
2714 /* Skip empty nodes */
2715 if (!pgdat->node_spanned_pages)
2718 #ifdef CONFIG_FLAT_NODE_MEM_MAP
2719 /* ia64 gets its own node_mem_map, before this, without bootmem */
2720 if (!pgdat->node_mem_map) {
2721 unsigned long size, start, end;
2725 * The zone's endpoints aren't required to be MAX_ORDER
2726 * aligned but the node_mem_map endpoints must be in order
2727 * for the buddy allocator to function correctly.
2729 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
2730 end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
2731 end = ALIGN(end, MAX_ORDER_NR_PAGES);
2732 size = (end - start) * sizeof(struct page);
2733 map = alloc_remap(pgdat->node_id, size);
2735 map = alloc_bootmem_node(pgdat, size);
2736 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
2738 #ifdef CONFIG_FLATMEM
2740 * With no DISCONTIG, the global mem_map is just set as node 0's
2742 if (pgdat == NODE_DATA(0)) {
2743 mem_map = NODE_DATA(0)->node_mem_map;
2744 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2745 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
2746 mem_map -= pgdat->node_start_pfn;
2747 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
2750 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
2753 void __meminit free_area_init_node(int nid, struct pglist_data *pgdat,
2754 unsigned long *zones_size, unsigned long node_start_pfn,
2755 unsigned long *zholes_size)
2757 pgdat->node_id = nid;
2758 pgdat->node_start_pfn = node_start_pfn;
2759 calculate_node_totalpages(pgdat, zones_size, zholes_size);
2761 alloc_node_mem_map(pgdat);
2763 free_area_init_core(pgdat, zones_size, zholes_size);
2766 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2768 * add_active_range - Register a range of PFNs backed by physical memory
2769 * @nid: The node ID the range resides on
2770 * @start_pfn: The start PFN of the available physical memory
2771 * @end_pfn: The end PFN of the available physical memory
2773 * These ranges are stored in an early_node_map[] and later used by
2774 * free_area_init_nodes() to calculate zone sizes and holes. If the
2775 * range spans a memory hole, it is up to the architecture to ensure
2776 * the memory is not freed by the bootmem allocator. If possible
2777 * the range being registered will be merged with existing ranges.
2779 void __init add_active_range(unsigned int nid, unsigned long start_pfn,
2780 unsigned long end_pfn)
2784 printk(KERN_DEBUG "Entering add_active_range(%d, %lu, %lu) "
2785 "%d entries of %d used\n",
2786 nid, start_pfn, end_pfn,
2787 nr_nodemap_entries, MAX_ACTIVE_REGIONS);
2789 /* Merge with existing active regions if possible */
2790 for (i = 0; i < nr_nodemap_entries; i++) {
2791 if (early_node_map[i].nid != nid)
2794 /* Skip if an existing region covers this new one */
2795 if (start_pfn >= early_node_map[i].start_pfn &&
2796 end_pfn <= early_node_map[i].end_pfn)
2799 /* Merge forward if suitable */
2800 if (start_pfn <= early_node_map[i].end_pfn &&
2801 end_pfn > early_node_map[i].end_pfn) {
2802 early_node_map[i].end_pfn = end_pfn;
2806 /* Merge backward if suitable */
2807 if (start_pfn < early_node_map[i].end_pfn &&
2808 end_pfn >= early_node_map[i].start_pfn) {
2809 early_node_map[i].start_pfn = start_pfn;
2814 /* Check that early_node_map is large enough */
2815 if (i >= MAX_ACTIVE_REGIONS) {
2816 printk(KERN_CRIT "More than %d memory regions, truncating\n",
2817 MAX_ACTIVE_REGIONS);
2821 early_node_map[i].nid = nid;
2822 early_node_map[i].start_pfn = start_pfn;
2823 early_node_map[i].end_pfn = end_pfn;
2824 nr_nodemap_entries = i + 1;
2828 * shrink_active_range - Shrink an existing registered range of PFNs
2829 * @nid: The node id the range is on that should be shrunk
2830 * @old_end_pfn: The old end PFN of the range
2831 * @new_end_pfn: The new PFN of the range
2833 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
2834 * The map is kept at the end physical page range that has already been
2835 * registered with add_active_range(). This function allows an arch to shrink
2836 * an existing registered range.
2838 void __init shrink_active_range(unsigned int nid, unsigned long old_end_pfn,
2839 unsigned long new_end_pfn)
2843 /* Find the old active region end and shrink */
2844 for_each_active_range_index_in_nid(i, nid)
2845 if (early_node_map[i].end_pfn == old_end_pfn) {
2846 early_node_map[i].end_pfn = new_end_pfn;
2852 * remove_all_active_ranges - Remove all currently registered regions
2854 * During discovery, it may be found that a table like SRAT is invalid
2855 * and an alternative discovery method must be used. This function removes
2856 * all currently registered regions.
2858 void __init remove_all_active_ranges(void)
2860 memset(early_node_map, 0, sizeof(early_node_map));
2861 nr_nodemap_entries = 0;
2862 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
2863 memset(node_boundary_start_pfn, 0, sizeof(node_boundary_start_pfn));
2864 memset(node_boundary_end_pfn, 0, sizeof(node_boundary_end_pfn));
2865 #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
2868 /* Compare two active node_active_regions */
2869 static int __init cmp_node_active_region(const void *a, const void *b)
2871 struct node_active_region *arange = (struct node_active_region *)a;
2872 struct node_active_region *brange = (struct node_active_region *)b;
2874 /* Done this way to avoid overflows */
2875 if (arange->start_pfn > brange->start_pfn)
2877 if (arange->start_pfn < brange->start_pfn)
2883 /* sort the node_map by start_pfn */
2884 static void __init sort_node_map(void)
2886 sort(early_node_map, (size_t)nr_nodemap_entries,
2887 sizeof(struct node_active_region),
2888 cmp_node_active_region, NULL);
2891 /* Find the lowest pfn for a node. This depends on a sorted early_node_map */
2892 unsigned long __init find_min_pfn_for_node(unsigned long nid)
2896 /* Regions in the early_node_map can be in any order */
2899 /* Assuming a sorted map, the first range found has the starting pfn */
2900 for_each_active_range_index_in_nid(i, nid)
2901 return early_node_map[i].start_pfn;
2903 printk(KERN_WARNING "Could not find start_pfn for node %lu\n", nid);
2908 * find_min_pfn_with_active_regions - Find the minimum PFN registered
2910 * It returns the minimum PFN based on information provided via
2911 * add_active_range().
2913 unsigned long __init find_min_pfn_with_active_regions(void)
2915 return find_min_pfn_for_node(MAX_NUMNODES);
2919 * find_max_pfn_with_active_regions - Find the maximum PFN registered
2921 * It returns the maximum PFN based on information provided via
2922 * add_active_range().
2924 unsigned long __init find_max_pfn_with_active_regions(void)
2927 unsigned long max_pfn = 0;
2929 for (i = 0; i < nr_nodemap_entries; i++)
2930 max_pfn = max(max_pfn, early_node_map[i].end_pfn);
2936 * free_area_init_nodes - Initialise all pg_data_t and zone data
2937 * @max_zone_pfn: an array of max PFNs for each zone
2939 * This will call free_area_init_node() for each active node in the system.
2940 * Using the page ranges provided by add_active_range(), the size of each
2941 * zone in each node and their holes is calculated. If the maximum PFN
2942 * between two adjacent zones match, it is assumed that the zone is empty.
2943 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
2944 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
2945 * starts where the previous one ended. For example, ZONE_DMA32 starts
2946 * at arch_max_dma_pfn.
2948 void __init free_area_init_nodes(unsigned long *max_zone_pfn)
2953 /* Record where the zone boundaries are */
2954 memset(arch_zone_lowest_possible_pfn, 0,
2955 sizeof(arch_zone_lowest_possible_pfn));
2956 memset(arch_zone_highest_possible_pfn, 0,
2957 sizeof(arch_zone_highest_possible_pfn));
2958 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
2959 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
2960 for (i = 1; i < MAX_NR_ZONES; i++) {
2961 arch_zone_lowest_possible_pfn[i] =
2962 arch_zone_highest_possible_pfn[i-1];
2963 arch_zone_highest_possible_pfn[i] =
2964 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
2967 /* Print out the zone ranges */
2968 printk("Zone PFN ranges:\n");
2969 for (i = 0; i < MAX_NR_ZONES; i++)
2970 printk(" %-8s %8lu -> %8lu\n",
2972 arch_zone_lowest_possible_pfn[i],
2973 arch_zone_highest_possible_pfn[i]);
2975 /* Print out the early_node_map[] */
2976 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries);
2977 for (i = 0; i < nr_nodemap_entries; i++)
2978 printk(" %3d: %8lu -> %8lu\n", early_node_map[i].nid,
2979 early_node_map[i].start_pfn,
2980 early_node_map[i].end_pfn);
2982 /* Initialise every node */
2983 for_each_online_node(nid) {
2984 pg_data_t *pgdat = NODE_DATA(nid);
2985 free_area_init_node(nid, pgdat, NULL,
2986 find_min_pfn_for_node(nid), NULL);
2989 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
2992 * set_dma_reserve - set the specified number of pages reserved in the first zone
2993 * @new_dma_reserve: The number of pages to mark reserved
2995 * The per-cpu batchsize and zone watermarks are determined by present_pages.
2996 * In the DMA zone, a significant percentage may be consumed by kernel image
2997 * and other unfreeable allocations which can skew the watermarks badly. This
2998 * function may optionally be used to account for unfreeable pages in the
2999 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
3000 * smaller per-cpu batchsize.
3002 void __init set_dma_reserve(unsigned long new_dma_reserve)
3004 dma_reserve = new_dma_reserve;
3007 #ifndef CONFIG_NEED_MULTIPLE_NODES
3008 static bootmem_data_t contig_bootmem_data;
3009 struct pglist_data contig_page_data = { .bdata = &contig_bootmem_data };
3011 EXPORT_SYMBOL(contig_page_data);
3014 void __init free_area_init(unsigned long *zones_size)
3016 free_area_init_node(0, NODE_DATA(0), zones_size,
3017 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
3020 static int page_alloc_cpu_notify(struct notifier_block *self,
3021 unsigned long action, void *hcpu)
3023 int cpu = (unsigned long)hcpu;
3025 if (action == CPU_DEAD) {
3026 local_irq_disable();
3028 vm_events_fold_cpu(cpu);
3030 refresh_cpu_vm_stats(cpu);
3035 void __init page_alloc_init(void)
3037 hotcpu_notifier(page_alloc_cpu_notify, 0);
3041 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
3042 * or min_free_kbytes changes.
3044 static void calculate_totalreserve_pages(void)
3046 struct pglist_data *pgdat;
3047 unsigned long reserve_pages = 0;
3048 enum zone_type i, j;
3050 for_each_online_pgdat(pgdat) {
3051 for (i = 0; i < MAX_NR_ZONES; i++) {
3052 struct zone *zone = pgdat->node_zones + i;
3053 unsigned long max = 0;
3055 /* Find valid and maximum lowmem_reserve in the zone */
3056 for (j = i; j < MAX_NR_ZONES; j++) {
3057 if (zone->lowmem_reserve[j] > max)
3058 max = zone->lowmem_reserve[j];
3061 /* we treat pages_high as reserved pages. */
3062 max += zone->pages_high;
3064 if (max > zone->present_pages)
3065 max = zone->present_pages;
3066 reserve_pages += max;
3069 totalreserve_pages = reserve_pages;
3073 * setup_per_zone_lowmem_reserve - called whenever
3074 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
3075 * has a correct pages reserved value, so an adequate number of
3076 * pages are left in the zone after a successful __alloc_pages().
3078 static void setup_per_zone_lowmem_reserve(void)
3080 struct pglist_data *pgdat;
3081 enum zone_type j, idx;
3083 for_each_online_pgdat(pgdat) {
3084 for (j = 0; j < MAX_NR_ZONES; j++) {
3085 struct zone *zone = pgdat->node_zones + j;
3086 unsigned long present_pages = zone->present_pages;
3088 zone->lowmem_reserve[j] = 0;
3092 struct zone *lower_zone;
3096 if (sysctl_lowmem_reserve_ratio[idx] < 1)
3097 sysctl_lowmem_reserve_ratio[idx] = 1;
3099 lower_zone = pgdat->node_zones + idx;
3100 lower_zone->lowmem_reserve[j] = present_pages /
3101 sysctl_lowmem_reserve_ratio[idx];
3102 present_pages += lower_zone->present_pages;
3107 /* update totalreserve_pages */
3108 calculate_totalreserve_pages();
3112 * setup_per_zone_pages_min - called when min_free_kbytes changes.
3114 * Ensures that the pages_{min,low,high} values for each zone are set correctly
3115 * with respect to min_free_kbytes.
3117 void setup_per_zone_pages_min(void)
3119 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
3120 unsigned long lowmem_pages = 0;
3122 unsigned long flags;
3124 /* Calculate total number of !ZONE_HIGHMEM pages */
3125 for_each_zone(zone) {
3126 if (!is_highmem(zone))
3127 lowmem_pages += zone->present_pages;
3130 for_each_zone(zone) {
3133 spin_lock_irqsave(&zone->lru_lock, flags);
3134 tmp = (u64)pages_min * zone->present_pages;
3135 do_div(tmp, lowmem_pages);
3136 if (is_highmem(zone)) {
3138 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
3139 * need highmem pages, so cap pages_min to a small
3142 * The (pages_high-pages_low) and (pages_low-pages_min)
3143 * deltas controls asynch page reclaim, and so should
3144 * not be capped for highmem.
3148 min_pages = zone->present_pages / 1024;
3149 if (min_pages < SWAP_CLUSTER_MAX)
3150 min_pages = SWAP_CLUSTER_MAX;
3151 if (min_pages > 128)
3153 zone->pages_min = min_pages;
3156 * If it's a lowmem zone, reserve a number of pages
3157 * proportionate to the zone's size.
3159 zone->pages_min = tmp;
3162 zone->pages_low = zone->pages_min + (tmp >> 2);
3163 zone->pages_high = zone->pages_min + (tmp >> 1);
3164 spin_unlock_irqrestore(&zone->lru_lock, flags);
3167 /* update totalreserve_pages */
3168 calculate_totalreserve_pages();
3172 * Initialise min_free_kbytes.
3174 * For small machines we want it small (128k min). For large machines
3175 * we want it large (64MB max). But it is not linear, because network
3176 * bandwidth does not increase linearly with machine size. We use
3178 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
3179 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
3195 static int __init init_per_zone_pages_min(void)
3197 unsigned long lowmem_kbytes;
3199 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
3201 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
3202 if (min_free_kbytes < 128)
3203 min_free_kbytes = 128;
3204 if (min_free_kbytes > 65536)
3205 min_free_kbytes = 65536;
3206 setup_per_zone_pages_min();
3207 setup_per_zone_lowmem_reserve();
3210 module_init(init_per_zone_pages_min)
3213 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
3214 * that we can call two helper functions whenever min_free_kbytes
3217 int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
3218 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
3220 proc_dointvec(table, write, file, buffer, length, ppos);
3221 setup_per_zone_pages_min();
3226 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
3227 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
3232 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
3237 zone->min_unmapped_pages = (zone->present_pages *
3238 sysctl_min_unmapped_ratio) / 100;
3242 int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
3243 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
3248 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
3253 zone->min_slab_pages = (zone->present_pages *
3254 sysctl_min_slab_ratio) / 100;
3260 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
3261 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
3262 * whenever sysctl_lowmem_reserve_ratio changes.
3264 * The reserve ratio obviously has absolutely no relation with the
3265 * pages_min watermarks. The lowmem reserve ratio can only make sense
3266 * if in function of the boot time zone sizes.
3268 int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
3269 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
3271 proc_dointvec_minmax(table, write, file, buffer, length, ppos);
3272 setup_per_zone_lowmem_reserve();
3277 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
3278 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
3279 * can have before it gets flushed back to buddy allocator.
3282 int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
3283 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
3289 ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
3290 if (!write || (ret == -EINVAL))
3292 for_each_zone(zone) {
3293 for_each_online_cpu(cpu) {
3295 high = zone->present_pages / percpu_pagelist_fraction;
3296 setup_pagelist_highmark(zone_pcp(zone, cpu), high);
3302 int hashdist = HASHDIST_DEFAULT;
3305 static int __init set_hashdist(char *str)
3309 hashdist = simple_strtoul(str, &str, 0);
3312 __setup("hashdist=", set_hashdist);
3316 * allocate a large system hash table from bootmem
3317 * - it is assumed that the hash table must contain an exact power-of-2
3318 * quantity of entries
3319 * - limit is the number of hash buckets, not the total allocation size
3321 void *__init alloc_large_system_hash(const char *tablename,
3322 unsigned long bucketsize,
3323 unsigned long numentries,
3326 unsigned int *_hash_shift,
3327 unsigned int *_hash_mask,
3328 unsigned long limit)
3330 unsigned long long max = limit;
3331 unsigned long log2qty, size;
3334 /* allow the kernel cmdline to have a say */
3336 /* round applicable memory size up to nearest megabyte */
3337 numentries = nr_kernel_pages;
3338 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
3339 numentries >>= 20 - PAGE_SHIFT;
3340 numentries <<= 20 - PAGE_SHIFT;
3342 /* limit to 1 bucket per 2^scale bytes of low memory */
3343 if (scale > PAGE_SHIFT)
3344 numentries >>= (scale - PAGE_SHIFT);
3346 numentries <<= (PAGE_SHIFT - scale);
3348 /* Make sure we've got at least a 0-order allocation.. */
3349 if (unlikely((numentries * bucketsize) < PAGE_SIZE))
3350 numentries = PAGE_SIZE / bucketsize;
3352 numentries = roundup_pow_of_two(numentries);
3354 /* limit allocation size to 1/16 total memory by default */
3356 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
3357 do_div(max, bucketsize);
3360 if (numentries > max)
3363 log2qty = ilog2(numentries);
3366 size = bucketsize << log2qty;
3367 if (flags & HASH_EARLY)
3368 table = alloc_bootmem(size);
3370 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
3372 unsigned long order;
3373 for (order = 0; ((1UL << order) << PAGE_SHIFT) < size; order++)
3375 table = (void*) __get_free_pages(GFP_ATOMIC, order);
3377 } while (!table && size > PAGE_SIZE && --log2qty);
3380 panic("Failed to allocate %s hash table\n", tablename);
3382 printk("%s hash table entries: %d (order: %d, %lu bytes)\n",
3385 ilog2(size) - PAGE_SHIFT,
3389 *_hash_shift = log2qty;
3391 *_hash_mask = (1 << log2qty) - 1;
3396 #ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE
3397 struct page *pfn_to_page(unsigned long pfn)
3399 return __pfn_to_page(pfn);
3401 unsigned long page_to_pfn(struct page *page)
3403 return __page_to_pfn(page);
3405 EXPORT_SYMBOL(pfn_to_page);
3406 EXPORT_SYMBOL(page_to_pfn);
3407 #endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */
3409 #if MAX_NUMNODES > 1
3411 * Find the highest possible node id.
3413 int highest_possible_node_id(void)
3416 unsigned int highest = 0;
3418 for_each_node_mask(node, node_possible_map)
3422 EXPORT_SYMBOL(highest_possible_node_id);