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/config.h>
18 #include <linux/stddef.h>
20 #include <linux/swap.h>
21 #include <linux/interrupt.h>
22 #include <linux/pagemap.h>
23 #include <linux/bootmem.h>
24 #include <linux/compiler.h>
25 #include <linux/kernel.h>
26 #include <linux/module.h>
27 #include <linux/suspend.h>
28 #include <linux/pagevec.h>
29 #include <linux/blkdev.h>
30 #include <linux/slab.h>
31 #include <linux/notifier.h>
32 #include <linux/topology.h>
33 #include <linux/sysctl.h>
34 #include <linux/cpu.h>
35 #include <linux/cpuset.h>
36 #include <linux/memory_hotplug.h>
37 #include <linux/nodemask.h>
38 #include <linux/vmalloc.h>
39 #include <linux/mempolicy.h>
41 #include <asm/tlbflush.h>
45 * MCD - HACK: Find somewhere to initialize this EARLY, or make this
48 nodemask_t node_online_map __read_mostly = { { [0] = 1UL } };
49 EXPORT_SYMBOL(node_online_map);
50 nodemask_t node_possible_map __read_mostly = NODE_MASK_ALL;
51 EXPORT_SYMBOL(node_possible_map);
52 unsigned long totalram_pages __read_mostly;
53 unsigned long totalhigh_pages __read_mostly;
54 unsigned long totalreserve_pages __read_mostly;
56 int percpu_pagelist_fraction;
58 static void __free_pages_ok(struct page *page, unsigned int order);
61 * results with 256, 32 in the lowmem_reserve sysctl:
62 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
63 * 1G machine -> (16M dma, 784M normal, 224M high)
64 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
65 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
66 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
68 * TBD: should special case ZONE_DMA32 machines here - in those we normally
69 * don't need any ZONE_NORMAL reservation
71 int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = { 256, 256, 32 };
73 EXPORT_SYMBOL(totalram_pages);
76 * Used by page_zone() to look up the address of the struct zone whose
77 * id is encoded in the upper bits of page->flags
79 struct zone *zone_table[1 << ZONETABLE_SHIFT] __read_mostly;
80 EXPORT_SYMBOL(zone_table);
82 static char *zone_names[MAX_NR_ZONES] = { "DMA", "DMA32", "Normal", "HighMem" };
83 int min_free_kbytes = 1024;
85 unsigned long __initdata nr_kernel_pages;
86 unsigned long __initdata nr_all_pages;
88 #ifdef CONFIG_DEBUG_VM
89 static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
93 unsigned long pfn = page_to_pfn(page);
96 seq = zone_span_seqbegin(zone);
97 if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
99 else if (pfn < zone->zone_start_pfn)
101 } while (zone_span_seqretry(zone, seq));
106 static int page_is_consistent(struct zone *zone, struct page *page)
108 #ifdef CONFIG_HOLES_IN_ZONE
109 if (!pfn_valid(page_to_pfn(page)))
112 if (zone != page_zone(page))
118 * Temporary debugging check for pages not lying within a given zone.
120 static int bad_range(struct zone *zone, struct page *page)
122 if (page_outside_zone_boundaries(zone, page))
124 if (!page_is_consistent(zone, page))
131 static inline int bad_range(struct zone *zone, struct page *page)
137 static void bad_page(struct page *page)
139 printk(KERN_EMERG "Bad page state in process '%s'\n"
140 KERN_EMERG "page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n"
141 KERN_EMERG "Trying to fix it up, but a reboot is needed\n"
142 KERN_EMERG "Backtrace:\n",
143 current->comm, page, (int)(2*sizeof(unsigned long)),
144 (unsigned long)page->flags, page->mapping,
145 page_mapcount(page), page_count(page));
147 page->flags &= ~(1 << PG_lru |
157 set_page_count(page, 0);
158 reset_page_mapcount(page);
159 page->mapping = NULL;
160 add_taint(TAINT_BAD_PAGE);
164 * Higher-order pages are called "compound pages". They are structured thusly:
166 * The first PAGE_SIZE page is called the "head page".
168 * The remaining PAGE_SIZE pages are called "tail pages".
170 * All pages have PG_compound set. All pages have their ->private pointing at
171 * the head page (even the head page has this).
173 * The first tail page's ->lru.next holds the address of the compound page's
174 * put_page() function. Its ->lru.prev holds the order of allocation.
175 * This usage means that zero-order pages may not be compound.
178 static void free_compound_page(struct page *page)
180 __free_pages_ok(page, (unsigned long)page[1].lru.prev);
183 static void prep_compound_page(struct page *page, unsigned long order)
186 int nr_pages = 1 << order;
188 page[1].lru.next = (void *)free_compound_page; /* set dtor */
189 page[1].lru.prev = (void *)order;
190 for (i = 0; i < nr_pages; i++) {
191 struct page *p = page + i;
193 __SetPageCompound(p);
194 set_page_private(p, (unsigned long)page);
198 static void destroy_compound_page(struct page *page, unsigned long order)
201 int nr_pages = 1 << order;
203 if (unlikely((unsigned long)page[1].lru.prev != order))
206 for (i = 0; i < nr_pages; i++) {
207 struct page *p = page + i;
209 if (unlikely(!PageCompound(p) |
210 (page_private(p) != (unsigned long)page)))
212 __ClearPageCompound(p);
216 static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
220 BUG_ON((gfp_flags & (__GFP_WAIT | __GFP_HIGHMEM)) == __GFP_HIGHMEM);
222 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
223 * and __GFP_HIGHMEM from hard or soft interrupt context.
225 BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
226 for (i = 0; i < (1 << order); i++)
227 clear_highpage(page + i);
231 * function for dealing with page's order in buddy system.
232 * zone->lock is already acquired when we use these.
233 * So, we don't need atomic page->flags operations here.
235 static inline unsigned long page_order(struct page *page) {
236 return page_private(page);
239 static inline void set_page_order(struct page *page, int order) {
240 set_page_private(page, order);
241 __SetPageBuddy(page);
244 static inline void rmv_page_order(struct page *page)
246 __ClearPageBuddy(page);
247 set_page_private(page, 0);
251 * Locate the struct page for both the matching buddy in our
252 * pair (buddy1) and the combined O(n+1) page they form (page).
254 * 1) Any buddy B1 will have an order O twin B2 which satisfies
255 * the following equation:
257 * For example, if the starting buddy (buddy2) is #8 its order
259 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
261 * 2) Any buddy B will have an order O+1 parent P which
262 * satisfies the following equation:
265 * Assumption: *_mem_map is contigious at least up to MAX_ORDER
267 static inline struct page *
268 __page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order)
270 unsigned long buddy_idx = page_idx ^ (1 << order);
272 return page + (buddy_idx - page_idx);
275 static inline unsigned long
276 __find_combined_index(unsigned long page_idx, unsigned int order)
278 return (page_idx & ~(1 << order));
282 * This function checks whether a page is free && is the buddy
283 * we can do coalesce a page and its buddy if
284 * (a) the buddy is not in a hole &&
285 * (b) the buddy is in the buddy system &&
286 * (c) a page and its buddy have the same order.
288 * For recording whether a page is in the buddy system, we use PG_buddy.
289 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
291 * For recording page's order, we use page_private(page).
293 static inline int page_is_buddy(struct page *page, int order)
295 #ifdef CONFIG_HOLES_IN_ZONE
296 if (!pfn_valid(page_to_pfn(page)))
300 if (PageBuddy(page) && page_order(page) == order) {
301 BUG_ON(page_count(page) != 0);
308 * Freeing function for a buddy system allocator.
310 * The concept of a buddy system is to maintain direct-mapped table
311 * (containing bit values) for memory blocks of various "orders".
312 * The bottom level table contains the map for the smallest allocatable
313 * units of memory (here, pages), and each level above it describes
314 * pairs of units from the levels below, hence, "buddies".
315 * At a high level, all that happens here is marking the table entry
316 * at the bottom level available, and propagating the changes upward
317 * as necessary, plus some accounting needed to play nicely with other
318 * parts of the VM system.
319 * At each level, we keep a list of pages, which are heads of continuous
320 * free pages of length of (1 << order) and marked with PG_buddy. Page's
321 * order is recorded in page_private(page) field.
322 * So when we are allocating or freeing one, we can derive the state of the
323 * other. That is, if we allocate a small block, and both were
324 * free, the remainder of the region must be split into blocks.
325 * If a block is freed, and its buddy is also free, then this
326 * triggers coalescing into a block of larger size.
331 static inline void __free_one_page(struct page *page,
332 struct zone *zone, unsigned int order)
334 unsigned long page_idx;
335 int order_size = 1 << order;
337 if (unlikely(PageCompound(page)))
338 destroy_compound_page(page, order);
340 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
342 BUG_ON(page_idx & (order_size - 1));
343 BUG_ON(bad_range(zone, page));
345 zone->free_pages += order_size;
346 while (order < MAX_ORDER-1) {
347 unsigned long combined_idx;
348 struct free_area *area;
351 buddy = __page_find_buddy(page, page_idx, order);
352 if (!page_is_buddy(buddy, order))
353 break; /* Move the buddy up one level. */
355 list_del(&buddy->lru);
356 area = zone->free_area + order;
358 rmv_page_order(buddy);
359 combined_idx = __find_combined_index(page_idx, order);
360 page = page + (combined_idx - page_idx);
361 page_idx = combined_idx;
364 set_page_order(page, order);
365 list_add(&page->lru, &zone->free_area[order].free_list);
366 zone->free_area[order].nr_free++;
369 static inline int free_pages_check(struct page *page)
371 if (unlikely(page_mapcount(page) |
372 (page->mapping != NULL) |
373 (page_count(page) != 0) |
387 __ClearPageDirty(page);
389 * For now, we report if PG_reserved was found set, but do not
390 * clear it, and do not free the page. But we shall soon need
391 * to do more, for when the ZERO_PAGE count wraps negative.
393 return PageReserved(page);
397 * Frees a list of pages.
398 * Assumes all pages on list are in same zone, and of same order.
399 * count is the number of pages to free.
401 * If the zone was previously in an "all pages pinned" state then look to
402 * see if this freeing clears that state.
404 * And clear the zone's pages_scanned counter, to hold off the "all pages are
405 * pinned" detection logic.
407 static void free_pages_bulk(struct zone *zone, int count,
408 struct list_head *list, int order)
410 spin_lock(&zone->lock);
411 zone->all_unreclaimable = 0;
412 zone->pages_scanned = 0;
416 BUG_ON(list_empty(list));
417 page = list_entry(list->prev, struct page, lru);
418 /* have to delete it as __free_one_page list manipulates */
419 list_del(&page->lru);
420 __free_one_page(page, zone, order);
422 spin_unlock(&zone->lock);
425 static void free_one_page(struct zone *zone, struct page *page, int order)
428 list_add(&page->lru, &list);
429 free_pages_bulk(zone, 1, &list, order);
432 static void __free_pages_ok(struct page *page, unsigned int order)
438 arch_free_page(page, order);
439 if (!PageHighMem(page))
440 mutex_debug_check_no_locks_freed(page_address(page),
443 for (i = 0 ; i < (1 << order) ; ++i)
444 reserved += free_pages_check(page + i);
448 kernel_map_pages(page, 1 << order, 0);
449 local_irq_save(flags);
450 __mod_page_state(pgfree, 1 << order);
451 free_one_page(page_zone(page), page, order);
452 local_irq_restore(flags);
456 * permit the bootmem allocator to evade page validation on high-order frees
458 void fastcall __init __free_pages_bootmem(struct page *page, unsigned int order)
461 __ClearPageReserved(page);
462 set_page_count(page, 0);
463 set_page_refcounted(page);
469 for (loop = 0; loop < BITS_PER_LONG; loop++) {
470 struct page *p = &page[loop];
472 if (loop + 1 < BITS_PER_LONG)
474 __ClearPageReserved(p);
475 set_page_count(p, 0);
478 set_page_refcounted(page);
479 __free_pages(page, order);
485 * The order of subdivision here is critical for the IO subsystem.
486 * Please do not alter this order without good reasons and regression
487 * testing. Specifically, as large blocks of memory are subdivided,
488 * the order in which smaller blocks are delivered depends on the order
489 * they're subdivided in this function. This is the primary factor
490 * influencing the order in which pages are delivered to the IO
491 * subsystem according to empirical testing, and this is also justified
492 * by considering the behavior of a buddy system containing a single
493 * large block of memory acted on by a series of small allocations.
494 * This behavior is a critical factor in sglist merging's success.
498 static inline void expand(struct zone *zone, struct page *page,
499 int low, int high, struct free_area *area)
501 unsigned long size = 1 << high;
507 BUG_ON(bad_range(zone, &page[size]));
508 list_add(&page[size].lru, &area->free_list);
510 set_page_order(&page[size], high);
515 * This page is about to be returned from the page allocator
517 static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
519 if (unlikely(page_mapcount(page) |
520 (page->mapping != NULL) |
521 (page_count(page) != 0) |
537 * For now, we report if PG_reserved was found set, but do not
538 * clear it, and do not allocate the page: as a safety net.
540 if (PageReserved(page))
543 page->flags &= ~(1 << PG_uptodate | 1 << PG_error |
544 1 << PG_referenced | 1 << PG_arch_1 |
545 1 << PG_checked | 1 << PG_mappedtodisk);
546 set_page_private(page, 0);
547 set_page_refcounted(page);
548 kernel_map_pages(page, 1 << order, 1);
550 if (gfp_flags & __GFP_ZERO)
551 prep_zero_page(page, order, gfp_flags);
553 if (order && (gfp_flags & __GFP_COMP))
554 prep_compound_page(page, order);
560 * Do the hard work of removing an element from the buddy allocator.
561 * Call me with the zone->lock already held.
563 static struct page *__rmqueue(struct zone *zone, unsigned int order)
565 struct free_area * area;
566 unsigned int current_order;
569 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
570 area = zone->free_area + current_order;
571 if (list_empty(&area->free_list))
574 page = list_entry(area->free_list.next, struct page, lru);
575 list_del(&page->lru);
576 rmv_page_order(page);
578 zone->free_pages -= 1UL << order;
579 expand(zone, page, order, current_order, area);
587 * Obtain a specified number of elements from the buddy allocator, all under
588 * a single hold of the lock, for efficiency. Add them to the supplied list.
589 * Returns the number of new pages which were placed at *list.
591 static int rmqueue_bulk(struct zone *zone, unsigned int order,
592 unsigned long count, struct list_head *list)
596 spin_lock(&zone->lock);
597 for (i = 0; i < count; ++i) {
598 struct page *page = __rmqueue(zone, order);
599 if (unlikely(page == NULL))
601 list_add_tail(&page->lru, list);
603 spin_unlock(&zone->lock);
609 * Called from the slab reaper to drain pagesets on a particular node that
610 * belong to the currently executing processor.
611 * Note that this function must be called with the thread pinned to
612 * a single processor.
614 void drain_node_pages(int nodeid)
619 for (z = 0; z < MAX_NR_ZONES; z++) {
620 struct zone *zone = NODE_DATA(nodeid)->node_zones + z;
621 struct per_cpu_pageset *pset;
623 pset = zone_pcp(zone, smp_processor_id());
624 for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) {
625 struct per_cpu_pages *pcp;
629 local_irq_save(flags);
630 free_pages_bulk(zone, pcp->count, &pcp->list, 0);
632 local_irq_restore(flags);
639 #if defined(CONFIG_PM) || defined(CONFIG_HOTPLUG_CPU)
640 static void __drain_pages(unsigned int cpu)
646 for_each_zone(zone) {
647 struct per_cpu_pageset *pset;
649 pset = zone_pcp(zone, cpu);
650 for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) {
651 struct per_cpu_pages *pcp;
654 local_irq_save(flags);
655 free_pages_bulk(zone, pcp->count, &pcp->list, 0);
657 local_irq_restore(flags);
661 #endif /* CONFIG_PM || CONFIG_HOTPLUG_CPU */
665 void mark_free_pages(struct zone *zone)
667 unsigned long zone_pfn, flags;
669 struct list_head *curr;
671 if (!zone->spanned_pages)
674 spin_lock_irqsave(&zone->lock, flags);
675 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn)
676 ClearPageNosaveFree(pfn_to_page(zone_pfn + zone->zone_start_pfn));
678 for (order = MAX_ORDER - 1; order >= 0; --order)
679 list_for_each(curr, &zone->free_area[order].free_list) {
680 unsigned long start_pfn, i;
682 start_pfn = page_to_pfn(list_entry(curr, struct page, lru));
684 for (i=0; i < (1<<order); i++)
685 SetPageNosaveFree(pfn_to_page(start_pfn+i));
687 spin_unlock_irqrestore(&zone->lock, flags);
691 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
693 void drain_local_pages(void)
697 local_irq_save(flags);
698 __drain_pages(smp_processor_id());
699 local_irq_restore(flags);
701 #endif /* CONFIG_PM */
703 static void zone_statistics(struct zonelist *zonelist, struct zone *z, int cpu)
706 pg_data_t *pg = z->zone_pgdat;
707 pg_data_t *orig = zonelist->zones[0]->zone_pgdat;
708 struct per_cpu_pageset *p;
710 p = zone_pcp(z, cpu);
715 zone_pcp(zonelist->zones[0], cpu)->numa_foreign++;
717 if (pg == NODE_DATA(numa_node_id()))
725 * Free a 0-order page
727 static void fastcall free_hot_cold_page(struct page *page, int cold)
729 struct zone *zone = page_zone(page);
730 struct per_cpu_pages *pcp;
733 arch_free_page(page, 0);
736 page->mapping = NULL;
737 if (free_pages_check(page))
740 kernel_map_pages(page, 1, 0);
742 pcp = &zone_pcp(zone, get_cpu())->pcp[cold];
743 local_irq_save(flags);
744 __inc_page_state(pgfree);
745 list_add(&page->lru, &pcp->list);
747 if (pcp->count >= pcp->high) {
748 free_pages_bulk(zone, pcp->batch, &pcp->list, 0);
749 pcp->count -= pcp->batch;
751 local_irq_restore(flags);
755 void fastcall free_hot_page(struct page *page)
757 free_hot_cold_page(page, 0);
760 void fastcall free_cold_page(struct page *page)
762 free_hot_cold_page(page, 1);
766 * split_page takes a non-compound higher-order page, and splits it into
767 * n (1<<order) sub-pages: page[0..n]
768 * Each sub-page must be freed individually.
770 * Note: this is probably too low level an operation for use in drivers.
771 * Please consult with lkml before using this in your driver.
773 void split_page(struct page *page, unsigned int order)
777 BUG_ON(PageCompound(page));
778 BUG_ON(!page_count(page));
779 for (i = 1; i < (1 << order); i++)
780 set_page_refcounted(page + i);
784 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
785 * we cheat by calling it from here, in the order > 0 path. Saves a branch
788 static struct page *buffered_rmqueue(struct zonelist *zonelist,
789 struct zone *zone, int order, gfp_t gfp_flags)
793 int cold = !!(gfp_flags & __GFP_COLD);
798 if (likely(order == 0)) {
799 struct per_cpu_pages *pcp;
801 pcp = &zone_pcp(zone, cpu)->pcp[cold];
802 local_irq_save(flags);
804 pcp->count += rmqueue_bulk(zone, 0,
805 pcp->batch, &pcp->list);
806 if (unlikely(!pcp->count))
809 page = list_entry(pcp->list.next, struct page, lru);
810 list_del(&page->lru);
813 spin_lock_irqsave(&zone->lock, flags);
814 page = __rmqueue(zone, order);
815 spin_unlock(&zone->lock);
820 __mod_page_state_zone(zone, pgalloc, 1 << order);
821 zone_statistics(zonelist, zone, cpu);
822 local_irq_restore(flags);
825 BUG_ON(bad_range(zone, page));
826 if (prep_new_page(page, order, gfp_flags))
831 local_irq_restore(flags);
836 #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
837 #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
838 #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
839 #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
840 #define ALLOC_HARDER 0x10 /* try to alloc harder */
841 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
842 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
845 * Return 1 if free pages are above 'mark'. This takes into account the order
848 int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
849 int classzone_idx, int alloc_flags)
851 /* free_pages my go negative - that's OK */
852 long min = mark, free_pages = z->free_pages - (1 << order) + 1;
855 if (alloc_flags & ALLOC_HIGH)
857 if (alloc_flags & ALLOC_HARDER)
860 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
862 for (o = 0; o < order; o++) {
863 /* At the next order, this order's pages become unavailable */
864 free_pages -= z->free_area[o].nr_free << o;
866 /* Require fewer higher order pages to be free */
869 if (free_pages <= min)
876 * get_page_from_freeliest goes through the zonelist trying to allocate
880 get_page_from_freelist(gfp_t gfp_mask, unsigned int order,
881 struct zonelist *zonelist, int alloc_flags)
883 struct zone **z = zonelist->zones;
884 struct page *page = NULL;
885 int classzone_idx = zone_idx(*z);
888 * Go through the zonelist once, looking for a zone with enough free.
889 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
892 if ((alloc_flags & ALLOC_CPUSET) &&
893 !cpuset_zone_allowed(*z, gfp_mask))
896 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
898 if (alloc_flags & ALLOC_WMARK_MIN)
899 mark = (*z)->pages_min;
900 else if (alloc_flags & ALLOC_WMARK_LOW)
901 mark = (*z)->pages_low;
903 mark = (*z)->pages_high;
904 if (!zone_watermark_ok(*z, order, mark,
905 classzone_idx, alloc_flags))
906 if (!zone_reclaim_mode ||
907 !zone_reclaim(*z, gfp_mask, order))
911 page = buffered_rmqueue(zonelist, *z, order, gfp_mask);
915 } while (*(++z) != NULL);
920 * This is the 'heart' of the zoned buddy allocator.
922 struct page * fastcall
923 __alloc_pages(gfp_t gfp_mask, unsigned int order,
924 struct zonelist *zonelist)
926 const gfp_t wait = gfp_mask & __GFP_WAIT;
929 struct reclaim_state reclaim_state;
930 struct task_struct *p = current;
933 int did_some_progress;
935 might_sleep_if(wait);
938 z = zonelist->zones; /* the list of zones suitable for gfp_mask */
940 if (unlikely(*z == NULL)) {
941 /* Should this ever happen?? */
945 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
946 zonelist, ALLOC_WMARK_LOW|ALLOC_CPUSET);
951 if (cpuset_zone_allowed(*z, gfp_mask))
952 wakeup_kswapd(*z, order);
956 * OK, we're below the kswapd watermark and have kicked background
957 * reclaim. Now things get more complex, so set up alloc_flags according
958 * to how we want to proceed.
960 * The caller may dip into page reserves a bit more if the caller
961 * cannot run direct reclaim, or if the caller has realtime scheduling
962 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
963 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
965 alloc_flags = ALLOC_WMARK_MIN;
966 if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait)
967 alloc_flags |= ALLOC_HARDER;
968 if (gfp_mask & __GFP_HIGH)
969 alloc_flags |= ALLOC_HIGH;
970 alloc_flags |= ALLOC_CPUSET;
973 * Go through the zonelist again. Let __GFP_HIGH and allocations
974 * coming from realtime tasks go deeper into reserves.
976 * This is the last chance, in general, before the goto nopage.
977 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
978 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
980 page = get_page_from_freelist(gfp_mask, order, zonelist, alloc_flags);
984 /* This allocation should allow future memory freeing. */
986 if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE)))
987 && !in_interrupt()) {
988 if (!(gfp_mask & __GFP_NOMEMALLOC)) {
990 /* go through the zonelist yet again, ignoring mins */
991 page = get_page_from_freelist(gfp_mask, order,
992 zonelist, ALLOC_NO_WATERMARKS);
995 if (gfp_mask & __GFP_NOFAIL) {
996 blk_congestion_wait(WRITE, HZ/50);
1003 /* Atomic allocations - we can't balance anything */
1010 /* We now go into synchronous reclaim */
1011 cpuset_memory_pressure_bump();
1012 p->flags |= PF_MEMALLOC;
1013 reclaim_state.reclaimed_slab = 0;
1014 p->reclaim_state = &reclaim_state;
1016 did_some_progress = try_to_free_pages(zonelist->zones, gfp_mask);
1018 p->reclaim_state = NULL;
1019 p->flags &= ~PF_MEMALLOC;
1023 if (likely(did_some_progress)) {
1024 page = get_page_from_freelist(gfp_mask, order,
1025 zonelist, alloc_flags);
1028 } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
1030 * Go through the zonelist yet one more time, keep
1031 * very high watermark here, this is only to catch
1032 * a parallel oom killing, we must fail if we're still
1033 * under heavy pressure.
1035 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
1036 zonelist, ALLOC_WMARK_HIGH|ALLOC_CPUSET);
1040 out_of_memory(zonelist, gfp_mask, order);
1045 * Don't let big-order allocations loop unless the caller explicitly
1046 * requests that. Wait for some write requests to complete then retry.
1048 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
1049 * <= 3, but that may not be true in other implementations.
1052 if (!(gfp_mask & __GFP_NORETRY)) {
1053 if ((order <= 3) || (gfp_mask & __GFP_REPEAT))
1055 if (gfp_mask & __GFP_NOFAIL)
1059 blk_congestion_wait(WRITE, HZ/50);
1064 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
1065 printk(KERN_WARNING "%s: page allocation failure."
1066 " order:%d, mode:0x%x\n",
1067 p->comm, order, gfp_mask);
1075 EXPORT_SYMBOL(__alloc_pages);
1078 * Common helper functions.
1080 fastcall unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
1083 page = alloc_pages(gfp_mask, order);
1086 return (unsigned long) page_address(page);
1089 EXPORT_SYMBOL(__get_free_pages);
1091 fastcall unsigned long get_zeroed_page(gfp_t gfp_mask)
1096 * get_zeroed_page() returns a 32-bit address, which cannot represent
1099 BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
1101 page = alloc_pages(gfp_mask | __GFP_ZERO, 0);
1103 return (unsigned long) page_address(page);
1107 EXPORT_SYMBOL(get_zeroed_page);
1109 void __pagevec_free(struct pagevec *pvec)
1111 int i = pagevec_count(pvec);
1114 free_hot_cold_page(pvec->pages[i], pvec->cold);
1117 fastcall void __free_pages(struct page *page, unsigned int order)
1119 if (put_page_testzero(page)) {
1121 free_hot_page(page);
1123 __free_pages_ok(page, order);
1127 EXPORT_SYMBOL(__free_pages);
1129 fastcall void free_pages(unsigned long addr, unsigned int order)
1132 BUG_ON(!virt_addr_valid((void *)addr));
1133 __free_pages(virt_to_page((void *)addr), order);
1137 EXPORT_SYMBOL(free_pages);
1140 * Total amount of free (allocatable) RAM:
1142 unsigned int nr_free_pages(void)
1144 unsigned int sum = 0;
1148 sum += zone->free_pages;
1153 EXPORT_SYMBOL(nr_free_pages);
1156 unsigned int nr_free_pages_pgdat(pg_data_t *pgdat)
1158 unsigned int i, sum = 0;
1160 for (i = 0; i < MAX_NR_ZONES; i++)
1161 sum += pgdat->node_zones[i].free_pages;
1167 static unsigned int nr_free_zone_pages(int offset)
1169 /* Just pick one node, since fallback list is circular */
1170 pg_data_t *pgdat = NODE_DATA(numa_node_id());
1171 unsigned int sum = 0;
1173 struct zonelist *zonelist = pgdat->node_zonelists + offset;
1174 struct zone **zonep = zonelist->zones;
1177 for (zone = *zonep++; zone; zone = *zonep++) {
1178 unsigned long size = zone->present_pages;
1179 unsigned long high = zone->pages_high;
1188 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1190 unsigned int nr_free_buffer_pages(void)
1192 return nr_free_zone_pages(gfp_zone(GFP_USER));
1196 * Amount of free RAM allocatable within all zones
1198 unsigned int nr_free_pagecache_pages(void)
1200 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER));
1203 #ifdef CONFIG_HIGHMEM
1204 unsigned int nr_free_highpages (void)
1207 unsigned int pages = 0;
1209 for_each_online_pgdat(pgdat)
1210 pages += pgdat->node_zones[ZONE_HIGHMEM].free_pages;
1217 static void show_node(struct zone *zone)
1219 printk("Node %d ", zone->zone_pgdat->node_id);
1222 #define show_node(zone) do { } while (0)
1226 * Accumulate the page_state information across all CPUs.
1227 * The result is unavoidably approximate - it can change
1228 * during and after execution of this function.
1230 static DEFINE_PER_CPU(struct page_state, page_states) = {0};
1232 atomic_t nr_pagecache = ATOMIC_INIT(0);
1233 EXPORT_SYMBOL(nr_pagecache);
1235 DEFINE_PER_CPU(long, nr_pagecache_local) = 0;
1238 static void __get_page_state(struct page_state *ret, int nr, cpumask_t *cpumask)
1242 memset(ret, 0, nr * sizeof(unsigned long));
1243 cpus_and(*cpumask, *cpumask, cpu_online_map);
1245 for_each_cpu_mask(cpu, *cpumask) {
1251 in = (unsigned long *)&per_cpu(page_states, cpu);
1253 next_cpu = next_cpu(cpu, *cpumask);
1254 if (likely(next_cpu < NR_CPUS))
1255 prefetch(&per_cpu(page_states, next_cpu));
1257 out = (unsigned long *)ret;
1258 for (off = 0; off < nr; off++)
1263 void get_page_state_node(struct page_state *ret, int node)
1266 cpumask_t mask = node_to_cpumask(node);
1268 nr = offsetof(struct page_state, GET_PAGE_STATE_LAST);
1269 nr /= sizeof(unsigned long);
1271 __get_page_state(ret, nr+1, &mask);
1274 void get_page_state(struct page_state *ret)
1277 cpumask_t mask = CPU_MASK_ALL;
1279 nr = offsetof(struct page_state, GET_PAGE_STATE_LAST);
1280 nr /= sizeof(unsigned long);
1282 __get_page_state(ret, nr + 1, &mask);
1285 void get_full_page_state(struct page_state *ret)
1287 cpumask_t mask = CPU_MASK_ALL;
1289 __get_page_state(ret, sizeof(*ret) / sizeof(unsigned long), &mask);
1292 unsigned long read_page_state_offset(unsigned long offset)
1294 unsigned long ret = 0;
1297 for_each_online_cpu(cpu) {
1300 in = (unsigned long)&per_cpu(page_states, cpu) + offset;
1301 ret += *((unsigned long *)in);
1306 void __mod_page_state_offset(unsigned long offset, unsigned long delta)
1310 ptr = &__get_cpu_var(page_states);
1311 *(unsigned long *)(ptr + offset) += delta;
1313 EXPORT_SYMBOL(__mod_page_state_offset);
1315 void mod_page_state_offset(unsigned long offset, unsigned long delta)
1317 unsigned long flags;
1320 local_irq_save(flags);
1321 ptr = &__get_cpu_var(page_states);
1322 *(unsigned long *)(ptr + offset) += delta;
1323 local_irq_restore(flags);
1325 EXPORT_SYMBOL(mod_page_state_offset);
1327 void __get_zone_counts(unsigned long *active, unsigned long *inactive,
1328 unsigned long *free, struct pglist_data *pgdat)
1330 struct zone *zones = pgdat->node_zones;
1336 for (i = 0; i < MAX_NR_ZONES; i++) {
1337 *active += zones[i].nr_active;
1338 *inactive += zones[i].nr_inactive;
1339 *free += zones[i].free_pages;
1343 void get_zone_counts(unsigned long *active,
1344 unsigned long *inactive, unsigned long *free)
1346 struct pglist_data *pgdat;
1351 for_each_online_pgdat(pgdat) {
1352 unsigned long l, m, n;
1353 __get_zone_counts(&l, &m, &n, pgdat);
1360 void si_meminfo(struct sysinfo *val)
1362 val->totalram = totalram_pages;
1364 val->freeram = nr_free_pages();
1365 val->bufferram = nr_blockdev_pages();
1366 #ifdef CONFIG_HIGHMEM
1367 val->totalhigh = totalhigh_pages;
1368 val->freehigh = nr_free_highpages();
1373 val->mem_unit = PAGE_SIZE;
1376 EXPORT_SYMBOL(si_meminfo);
1379 void si_meminfo_node(struct sysinfo *val, int nid)
1381 pg_data_t *pgdat = NODE_DATA(nid);
1383 val->totalram = pgdat->node_present_pages;
1384 val->freeram = nr_free_pages_pgdat(pgdat);
1385 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1386 val->freehigh = pgdat->node_zones[ZONE_HIGHMEM].free_pages;
1387 val->mem_unit = PAGE_SIZE;
1391 #define K(x) ((x) << (PAGE_SHIFT-10))
1394 * Show free area list (used inside shift_scroll-lock stuff)
1395 * We also calculate the percentage fragmentation. We do this by counting the
1396 * memory on each free list with the exception of the first item on the list.
1398 void show_free_areas(void)
1400 struct page_state ps;
1401 int cpu, temperature;
1402 unsigned long active;
1403 unsigned long inactive;
1407 for_each_zone(zone) {
1409 printk("%s per-cpu:", zone->name);
1411 if (!populated_zone(zone)) {
1417 for_each_online_cpu(cpu) {
1418 struct per_cpu_pageset *pageset;
1420 pageset = zone_pcp(zone, cpu);
1422 for (temperature = 0; temperature < 2; temperature++)
1423 printk("cpu %d %s: high %d, batch %d used:%d\n",
1425 temperature ? "cold" : "hot",
1426 pageset->pcp[temperature].high,
1427 pageset->pcp[temperature].batch,
1428 pageset->pcp[temperature].count);
1432 get_page_state(&ps);
1433 get_zone_counts(&active, &inactive, &free);
1435 printk("Free pages: %11ukB (%ukB HighMem)\n",
1437 K(nr_free_highpages()));
1439 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu "
1440 "unstable:%lu free:%u slab:%lu mapped:%lu pagetables:%lu\n",
1449 ps.nr_page_table_pages);
1451 for_each_zone(zone) {
1463 " pages_scanned:%lu"
1464 " all_unreclaimable? %s"
1467 K(zone->free_pages),
1470 K(zone->pages_high),
1472 K(zone->nr_inactive),
1473 K(zone->present_pages),
1474 zone->pages_scanned,
1475 (zone->all_unreclaimable ? "yes" : "no")
1477 printk("lowmem_reserve[]:");
1478 for (i = 0; i < MAX_NR_ZONES; i++)
1479 printk(" %lu", zone->lowmem_reserve[i]);
1483 for_each_zone(zone) {
1484 unsigned long nr, flags, order, total = 0;
1487 printk("%s: ", zone->name);
1488 if (!populated_zone(zone)) {
1493 spin_lock_irqsave(&zone->lock, flags);
1494 for (order = 0; order < MAX_ORDER; order++) {
1495 nr = zone->free_area[order].nr_free;
1496 total += nr << order;
1497 printk("%lu*%lukB ", nr, K(1UL) << order);
1499 spin_unlock_irqrestore(&zone->lock, flags);
1500 printk("= %lukB\n", K(total));
1503 show_swap_cache_info();
1507 * Builds allocation fallback zone lists.
1509 * Add all populated zones of a node to the zonelist.
1511 static int __init build_zonelists_node(pg_data_t *pgdat,
1512 struct zonelist *zonelist, int nr_zones, int zone_type)
1516 BUG_ON(zone_type > ZONE_HIGHMEM);
1519 zone = pgdat->node_zones + zone_type;
1520 if (populated_zone(zone)) {
1521 #ifndef CONFIG_HIGHMEM
1522 BUG_ON(zone_type > ZONE_NORMAL);
1524 zonelist->zones[nr_zones++] = zone;
1525 check_highest_zone(zone_type);
1529 } while (zone_type >= 0);
1533 static inline int highest_zone(int zone_bits)
1535 int res = ZONE_NORMAL;
1536 if (zone_bits & (__force int)__GFP_HIGHMEM)
1538 if (zone_bits & (__force int)__GFP_DMA32)
1540 if (zone_bits & (__force int)__GFP_DMA)
1546 #define MAX_NODE_LOAD (num_online_nodes())
1547 static int __initdata node_load[MAX_NUMNODES];
1549 * find_next_best_node - find the next node that should appear in a given node's fallback list
1550 * @node: node whose fallback list we're appending
1551 * @used_node_mask: nodemask_t of already used nodes
1553 * We use a number of factors to determine which is the next node that should
1554 * appear on a given node's fallback list. The node should not have appeared
1555 * already in @node's fallback list, and it should be the next closest node
1556 * according to the distance array (which contains arbitrary distance values
1557 * from each node to each node in the system), and should also prefer nodes
1558 * with no CPUs, since presumably they'll have very little allocation pressure
1559 * on them otherwise.
1560 * It returns -1 if no node is found.
1562 static int __init find_next_best_node(int node, nodemask_t *used_node_mask)
1565 int min_val = INT_MAX;
1568 /* Use the local node if we haven't already */
1569 if (!node_isset(node, *used_node_mask)) {
1570 node_set(node, *used_node_mask);
1574 for_each_online_node(n) {
1577 /* Don't want a node to appear more than once */
1578 if (node_isset(n, *used_node_mask))
1581 /* Use the distance array to find the distance */
1582 val = node_distance(node, n);
1584 /* Penalize nodes under us ("prefer the next node") */
1587 /* Give preference to headless and unused nodes */
1588 tmp = node_to_cpumask(n);
1589 if (!cpus_empty(tmp))
1590 val += PENALTY_FOR_NODE_WITH_CPUS;
1592 /* Slight preference for less loaded node */
1593 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
1594 val += node_load[n];
1596 if (val < min_val) {
1603 node_set(best_node, *used_node_mask);
1608 static void __init build_zonelists(pg_data_t *pgdat)
1610 int i, j, k, node, local_node;
1611 int prev_node, load;
1612 struct zonelist *zonelist;
1613 nodemask_t used_mask;
1615 /* initialize zonelists */
1616 for (i = 0; i < GFP_ZONETYPES; i++) {
1617 zonelist = pgdat->node_zonelists + i;
1618 zonelist->zones[0] = NULL;
1621 /* NUMA-aware ordering of nodes */
1622 local_node = pgdat->node_id;
1623 load = num_online_nodes();
1624 prev_node = local_node;
1625 nodes_clear(used_mask);
1626 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
1627 int distance = node_distance(local_node, node);
1630 * If another node is sufficiently far away then it is better
1631 * to reclaim pages in a zone before going off node.
1633 if (distance > RECLAIM_DISTANCE)
1634 zone_reclaim_mode = 1;
1637 * We don't want to pressure a particular node.
1638 * So adding penalty to the first node in same
1639 * distance group to make it round-robin.
1642 if (distance != node_distance(local_node, prev_node))
1643 node_load[node] += load;
1646 for (i = 0; i < GFP_ZONETYPES; i++) {
1647 zonelist = pgdat->node_zonelists + i;
1648 for (j = 0; zonelist->zones[j] != NULL; j++);
1650 k = highest_zone(i);
1652 j = build_zonelists_node(NODE_DATA(node), zonelist, j, k);
1653 zonelist->zones[j] = NULL;
1658 #else /* CONFIG_NUMA */
1660 static void __init build_zonelists(pg_data_t *pgdat)
1662 int i, j, k, node, local_node;
1664 local_node = pgdat->node_id;
1665 for (i = 0; i < GFP_ZONETYPES; i++) {
1666 struct zonelist *zonelist;
1668 zonelist = pgdat->node_zonelists + i;
1671 k = highest_zone(i);
1672 j = build_zonelists_node(pgdat, zonelist, j, k);
1674 * Now we build the zonelist so that it contains the zones
1675 * of all the other nodes.
1676 * We don't want to pressure a particular node, so when
1677 * building the zones for node N, we make sure that the
1678 * zones coming right after the local ones are those from
1679 * node N+1 (modulo N)
1681 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
1682 if (!node_online(node))
1684 j = build_zonelists_node(NODE_DATA(node), zonelist, j, k);
1686 for (node = 0; node < local_node; node++) {
1687 if (!node_online(node))
1689 j = build_zonelists_node(NODE_DATA(node), zonelist, j, k);
1692 zonelist->zones[j] = NULL;
1696 #endif /* CONFIG_NUMA */
1698 void __init build_all_zonelists(void)
1702 for_each_online_node(i)
1703 build_zonelists(NODE_DATA(i));
1704 printk("Built %i zonelists\n", num_online_nodes());
1705 cpuset_init_current_mems_allowed();
1709 * Helper functions to size the waitqueue hash table.
1710 * Essentially these want to choose hash table sizes sufficiently
1711 * large so that collisions trying to wait on pages are rare.
1712 * But in fact, the number of active page waitqueues on typical
1713 * systems is ridiculously low, less than 200. So this is even
1714 * conservative, even though it seems large.
1716 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
1717 * waitqueues, i.e. the size of the waitq table given the number of pages.
1719 #define PAGES_PER_WAITQUEUE 256
1721 static inline unsigned long wait_table_size(unsigned long pages)
1723 unsigned long size = 1;
1725 pages /= PAGES_PER_WAITQUEUE;
1727 while (size < pages)
1731 * Once we have dozens or even hundreds of threads sleeping
1732 * on IO we've got bigger problems than wait queue collision.
1733 * Limit the size of the wait table to a reasonable size.
1735 size = min(size, 4096UL);
1737 return max(size, 4UL);
1741 * This is an integer logarithm so that shifts can be used later
1742 * to extract the more random high bits from the multiplicative
1743 * hash function before the remainder is taken.
1745 static inline unsigned long wait_table_bits(unsigned long size)
1750 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
1752 static void __init calculate_zone_totalpages(struct pglist_data *pgdat,
1753 unsigned long *zones_size, unsigned long *zholes_size)
1755 unsigned long realtotalpages, totalpages = 0;
1758 for (i = 0; i < MAX_NR_ZONES; i++)
1759 totalpages += zones_size[i];
1760 pgdat->node_spanned_pages = totalpages;
1762 realtotalpages = totalpages;
1764 for (i = 0; i < MAX_NR_ZONES; i++)
1765 realtotalpages -= zholes_size[i];
1766 pgdat->node_present_pages = realtotalpages;
1767 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, realtotalpages);
1772 * Initially all pages are reserved - free ones are freed
1773 * up by free_all_bootmem() once the early boot process is
1774 * done. Non-atomic initialization, single-pass.
1776 void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
1777 unsigned long start_pfn)
1780 unsigned long end_pfn = start_pfn + size;
1783 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1784 if (!early_pfn_valid(pfn))
1786 page = pfn_to_page(pfn);
1787 set_page_links(page, zone, nid, pfn);
1788 init_page_count(page);
1789 reset_page_mapcount(page);
1790 SetPageReserved(page);
1791 INIT_LIST_HEAD(&page->lru);
1792 #ifdef WANT_PAGE_VIRTUAL
1793 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
1794 if (!is_highmem_idx(zone))
1795 set_page_address(page, __va(pfn << PAGE_SHIFT));
1800 void zone_init_free_lists(struct pglist_data *pgdat, struct zone *zone,
1804 for (order = 0; order < MAX_ORDER ; order++) {
1805 INIT_LIST_HEAD(&zone->free_area[order].free_list);
1806 zone->free_area[order].nr_free = 0;
1810 #define ZONETABLE_INDEX(x, zone_nr) ((x << ZONES_SHIFT) | zone_nr)
1811 void zonetable_add(struct zone *zone, int nid, int zid, unsigned long pfn,
1814 unsigned long snum = pfn_to_section_nr(pfn);
1815 unsigned long end = pfn_to_section_nr(pfn + size);
1818 zone_table[ZONETABLE_INDEX(nid, zid)] = zone;
1820 for (; snum <= end; snum++)
1821 zone_table[ZONETABLE_INDEX(snum, zid)] = zone;
1824 #ifndef __HAVE_ARCH_MEMMAP_INIT
1825 #define memmap_init(size, nid, zone, start_pfn) \
1826 memmap_init_zone((size), (nid), (zone), (start_pfn))
1829 static int __cpuinit zone_batchsize(struct zone *zone)
1834 * The per-cpu-pages pools are set to around 1000th of the
1835 * size of the zone. But no more than 1/2 of a meg.
1837 * OK, so we don't know how big the cache is. So guess.
1839 batch = zone->present_pages / 1024;
1840 if (batch * PAGE_SIZE > 512 * 1024)
1841 batch = (512 * 1024) / PAGE_SIZE;
1842 batch /= 4; /* We effectively *= 4 below */
1847 * Clamp the batch to a 2^n - 1 value. Having a power
1848 * of 2 value was found to be more likely to have
1849 * suboptimal cache aliasing properties in some cases.
1851 * For example if 2 tasks are alternately allocating
1852 * batches of pages, one task can end up with a lot
1853 * of pages of one half of the possible page colors
1854 * and the other with pages of the other colors.
1856 batch = (1 << (fls(batch + batch/2)-1)) - 1;
1861 inline void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
1863 struct per_cpu_pages *pcp;
1865 memset(p, 0, sizeof(*p));
1867 pcp = &p->pcp[0]; /* hot */
1869 pcp->high = 6 * batch;
1870 pcp->batch = max(1UL, 1 * batch);
1871 INIT_LIST_HEAD(&pcp->list);
1873 pcp = &p->pcp[1]; /* cold*/
1875 pcp->high = 2 * batch;
1876 pcp->batch = max(1UL, batch/2);
1877 INIT_LIST_HEAD(&pcp->list);
1881 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
1882 * to the value high for the pageset p.
1885 static void setup_pagelist_highmark(struct per_cpu_pageset *p,
1888 struct per_cpu_pages *pcp;
1890 pcp = &p->pcp[0]; /* hot list */
1892 pcp->batch = max(1UL, high/4);
1893 if ((high/4) > (PAGE_SHIFT * 8))
1894 pcp->batch = PAGE_SHIFT * 8;
1900 * Boot pageset table. One per cpu which is going to be used for all
1901 * zones and all nodes. The parameters will be set in such a way
1902 * that an item put on a list will immediately be handed over to
1903 * the buddy list. This is safe since pageset manipulation is done
1904 * with interrupts disabled.
1906 * Some NUMA counter updates may also be caught by the boot pagesets.
1908 * The boot_pagesets must be kept even after bootup is complete for
1909 * unused processors and/or zones. They do play a role for bootstrapping
1910 * hotplugged processors.
1912 * zoneinfo_show() and maybe other functions do
1913 * not check if the processor is online before following the pageset pointer.
1914 * Other parts of the kernel may not check if the zone is available.
1916 static struct per_cpu_pageset boot_pageset[NR_CPUS];
1919 * Dynamically allocate memory for the
1920 * per cpu pageset array in struct zone.
1922 static int __cpuinit process_zones(int cpu)
1924 struct zone *zone, *dzone;
1926 for_each_zone(zone) {
1928 zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset),
1929 GFP_KERNEL, cpu_to_node(cpu));
1930 if (!zone_pcp(zone, cpu))
1933 setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone));
1935 if (percpu_pagelist_fraction)
1936 setup_pagelist_highmark(zone_pcp(zone, cpu),
1937 (zone->present_pages / percpu_pagelist_fraction));
1942 for_each_zone(dzone) {
1945 kfree(zone_pcp(dzone, cpu));
1946 zone_pcp(dzone, cpu) = NULL;
1951 static inline void free_zone_pagesets(int cpu)
1955 for_each_zone(zone) {
1956 struct per_cpu_pageset *pset = zone_pcp(zone, cpu);
1958 zone_pcp(zone, cpu) = NULL;
1963 static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb,
1964 unsigned long action,
1967 int cpu = (long)hcpu;
1968 int ret = NOTIFY_OK;
1971 case CPU_UP_PREPARE:
1972 if (process_zones(cpu))
1975 case CPU_UP_CANCELED:
1977 free_zone_pagesets(cpu);
1985 static struct notifier_block pageset_notifier =
1986 { &pageset_cpuup_callback, NULL, 0 };
1988 void __init setup_per_cpu_pageset(void)
1992 /* Initialize per_cpu_pageset for cpu 0.
1993 * A cpuup callback will do this for every cpu
1994 * as it comes online
1996 err = process_zones(smp_processor_id());
1998 register_cpu_notifier(&pageset_notifier);
2004 void zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
2007 struct pglist_data *pgdat = zone->zone_pgdat;
2010 * The per-page waitqueue mechanism uses hashed waitqueues
2013 zone->wait_table_size = wait_table_size(zone_size_pages);
2014 zone->wait_table_bits = wait_table_bits(zone->wait_table_size);
2015 zone->wait_table = (wait_queue_head_t *)
2016 alloc_bootmem_node(pgdat, zone->wait_table_size
2017 * sizeof(wait_queue_head_t));
2019 for(i = 0; i < zone->wait_table_size; ++i)
2020 init_waitqueue_head(zone->wait_table + i);
2023 static __meminit void zone_pcp_init(struct zone *zone)
2026 unsigned long batch = zone_batchsize(zone);
2028 for (cpu = 0; cpu < NR_CPUS; cpu++) {
2030 /* Early boot. Slab allocator not functional yet */
2031 zone_pcp(zone, cpu) = &boot_pageset[cpu];
2032 setup_pageset(&boot_pageset[cpu],0);
2034 setup_pageset(zone_pcp(zone,cpu), batch);
2037 if (zone->present_pages)
2038 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n",
2039 zone->name, zone->present_pages, batch);
2042 static __meminit void init_currently_empty_zone(struct zone *zone,
2043 unsigned long zone_start_pfn, unsigned long size)
2045 struct pglist_data *pgdat = zone->zone_pgdat;
2047 zone_wait_table_init(zone, size);
2048 pgdat->nr_zones = zone_idx(zone) + 1;
2050 zone->zone_start_pfn = zone_start_pfn;
2052 memmap_init(size, pgdat->node_id, zone_idx(zone), zone_start_pfn);
2054 zone_init_free_lists(pgdat, zone, zone->spanned_pages);
2058 * Set up the zone data structures:
2059 * - mark all pages reserved
2060 * - mark all memory queues empty
2061 * - clear the memory bitmaps
2063 static void __init free_area_init_core(struct pglist_data *pgdat,
2064 unsigned long *zones_size, unsigned long *zholes_size)
2067 int nid = pgdat->node_id;
2068 unsigned long zone_start_pfn = pgdat->node_start_pfn;
2070 pgdat_resize_init(pgdat);
2071 pgdat->nr_zones = 0;
2072 init_waitqueue_head(&pgdat->kswapd_wait);
2073 pgdat->kswapd_max_order = 0;
2075 for (j = 0; j < MAX_NR_ZONES; j++) {
2076 struct zone *zone = pgdat->node_zones + j;
2077 unsigned long size, realsize;
2079 realsize = size = zones_size[j];
2081 realsize -= zholes_size[j];
2083 if (j < ZONE_HIGHMEM)
2084 nr_kernel_pages += realsize;
2085 nr_all_pages += realsize;
2087 zone->spanned_pages = size;
2088 zone->present_pages = realsize;
2089 zone->name = zone_names[j];
2090 spin_lock_init(&zone->lock);
2091 spin_lock_init(&zone->lru_lock);
2092 zone_seqlock_init(zone);
2093 zone->zone_pgdat = pgdat;
2094 zone->free_pages = 0;
2096 zone->temp_priority = zone->prev_priority = DEF_PRIORITY;
2098 zone_pcp_init(zone);
2099 INIT_LIST_HEAD(&zone->active_list);
2100 INIT_LIST_HEAD(&zone->inactive_list);
2101 zone->nr_scan_active = 0;
2102 zone->nr_scan_inactive = 0;
2103 zone->nr_active = 0;
2104 zone->nr_inactive = 0;
2105 atomic_set(&zone->reclaim_in_progress, 0);
2109 zonetable_add(zone, nid, j, zone_start_pfn, size);
2110 init_currently_empty_zone(zone, zone_start_pfn, size);
2111 zone_start_pfn += size;
2115 static void __init alloc_node_mem_map(struct pglist_data *pgdat)
2117 /* Skip empty nodes */
2118 if (!pgdat->node_spanned_pages)
2121 #ifdef CONFIG_FLAT_NODE_MEM_MAP
2122 /* ia64 gets its own node_mem_map, before this, without bootmem */
2123 if (!pgdat->node_mem_map) {
2127 size = (pgdat->node_spanned_pages + 1) * sizeof(struct page);
2128 map = alloc_remap(pgdat->node_id, size);
2130 map = alloc_bootmem_node(pgdat, size);
2131 pgdat->node_mem_map = map;
2133 #ifdef CONFIG_FLATMEM
2135 * With no DISCONTIG, the global mem_map is just set as node 0's
2137 if (pgdat == NODE_DATA(0))
2138 mem_map = NODE_DATA(0)->node_mem_map;
2140 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
2143 void __init free_area_init_node(int nid, struct pglist_data *pgdat,
2144 unsigned long *zones_size, unsigned long node_start_pfn,
2145 unsigned long *zholes_size)
2147 pgdat->node_id = nid;
2148 pgdat->node_start_pfn = node_start_pfn;
2149 calculate_zone_totalpages(pgdat, zones_size, zholes_size);
2151 alloc_node_mem_map(pgdat);
2153 free_area_init_core(pgdat, zones_size, zholes_size);
2156 #ifndef CONFIG_NEED_MULTIPLE_NODES
2157 static bootmem_data_t contig_bootmem_data;
2158 struct pglist_data contig_page_data = { .bdata = &contig_bootmem_data };
2160 EXPORT_SYMBOL(contig_page_data);
2163 void __init free_area_init(unsigned long *zones_size)
2165 free_area_init_node(0, NODE_DATA(0), zones_size,
2166 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
2169 #ifdef CONFIG_PROC_FS
2171 #include <linux/seq_file.h>
2173 static void *frag_start(struct seq_file *m, loff_t *pos)
2177 for (pgdat = first_online_pgdat();
2179 pgdat = next_online_pgdat(pgdat))
2185 static void *frag_next(struct seq_file *m, void *arg, loff_t *pos)
2187 pg_data_t *pgdat = (pg_data_t *)arg;
2190 return next_online_pgdat(pgdat);
2193 static void frag_stop(struct seq_file *m, void *arg)
2198 * This walks the free areas for each zone.
2200 static int frag_show(struct seq_file *m, void *arg)
2202 pg_data_t *pgdat = (pg_data_t *)arg;
2204 struct zone *node_zones = pgdat->node_zones;
2205 unsigned long flags;
2208 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
2209 if (!populated_zone(zone))
2212 spin_lock_irqsave(&zone->lock, flags);
2213 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
2214 for (order = 0; order < MAX_ORDER; ++order)
2215 seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
2216 spin_unlock_irqrestore(&zone->lock, flags);
2222 struct seq_operations fragmentation_op = {
2223 .start = frag_start,
2230 * Output information about zones in @pgdat.
2232 static int zoneinfo_show(struct seq_file *m, void *arg)
2234 pg_data_t *pgdat = arg;
2236 struct zone *node_zones = pgdat->node_zones;
2237 unsigned long flags;
2239 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; zone++) {
2242 if (!populated_zone(zone))
2245 spin_lock_irqsave(&zone->lock, flags);
2246 seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name);
2254 "\n scanned %lu (a: %lu i: %lu)"
2263 zone->pages_scanned,
2264 zone->nr_scan_active, zone->nr_scan_inactive,
2265 zone->spanned_pages,
2266 zone->present_pages);
2268 "\n protection: (%lu",
2269 zone->lowmem_reserve[0]);
2270 for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
2271 seq_printf(m, ", %lu", zone->lowmem_reserve[i]);
2275 for_each_online_cpu(i) {
2276 struct per_cpu_pageset *pageset;
2279 pageset = zone_pcp(zone, i);
2280 for (j = 0; j < ARRAY_SIZE(pageset->pcp); j++) {
2281 if (pageset->pcp[j].count)
2284 if (j == ARRAY_SIZE(pageset->pcp))
2286 for (j = 0; j < ARRAY_SIZE(pageset->pcp); j++) {
2288 "\n cpu: %i pcp: %i"
2293 pageset->pcp[j].count,
2294 pageset->pcp[j].high,
2295 pageset->pcp[j].batch);
2301 "\n numa_foreign: %lu"
2302 "\n interleave_hit: %lu"
2303 "\n local_node: %lu"
2304 "\n other_node: %lu",
2307 pageset->numa_foreign,
2308 pageset->interleave_hit,
2309 pageset->local_node,
2310 pageset->other_node);
2314 "\n all_unreclaimable: %u"
2315 "\n prev_priority: %i"
2316 "\n temp_priority: %i"
2317 "\n start_pfn: %lu",
2318 zone->all_unreclaimable,
2319 zone->prev_priority,
2320 zone->temp_priority,
2321 zone->zone_start_pfn);
2322 spin_unlock_irqrestore(&zone->lock, flags);
2328 struct seq_operations zoneinfo_op = {
2329 .start = frag_start, /* iterate over all zones. The same as in
2333 .show = zoneinfo_show,
2336 static char *vmstat_text[] = {
2340 "nr_page_table_pages",
2371 "pgscan_kswapd_high",
2372 "pgscan_kswapd_normal",
2373 "pgscan_kswapd_dma32",
2374 "pgscan_kswapd_dma",
2376 "pgscan_direct_high",
2377 "pgscan_direct_normal",
2378 "pgscan_direct_dma32",
2379 "pgscan_direct_dma",
2384 "kswapd_inodesteal",
2392 static void *vmstat_start(struct seq_file *m, loff_t *pos)
2394 struct page_state *ps;
2396 if (*pos >= ARRAY_SIZE(vmstat_text))
2399 ps = kmalloc(sizeof(*ps), GFP_KERNEL);
2402 return ERR_PTR(-ENOMEM);
2403 get_full_page_state(ps);
2404 ps->pgpgin /= 2; /* sectors -> kbytes */
2406 return (unsigned long *)ps + *pos;
2409 static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
2412 if (*pos >= ARRAY_SIZE(vmstat_text))
2414 return (unsigned long *)m->private + *pos;
2417 static int vmstat_show(struct seq_file *m, void *arg)
2419 unsigned long *l = arg;
2420 unsigned long off = l - (unsigned long *)m->private;
2422 seq_printf(m, "%s %lu\n", vmstat_text[off], *l);
2426 static void vmstat_stop(struct seq_file *m, void *arg)
2432 struct seq_operations vmstat_op = {
2433 .start = vmstat_start,
2434 .next = vmstat_next,
2435 .stop = vmstat_stop,
2436 .show = vmstat_show,
2439 #endif /* CONFIG_PROC_FS */
2441 #ifdef CONFIG_HOTPLUG_CPU
2442 static int page_alloc_cpu_notify(struct notifier_block *self,
2443 unsigned long action, void *hcpu)
2445 int cpu = (unsigned long)hcpu;
2447 unsigned long *src, *dest;
2449 if (action == CPU_DEAD) {
2452 /* Drain local pagecache count. */
2453 count = &per_cpu(nr_pagecache_local, cpu);
2454 atomic_add(*count, &nr_pagecache);
2456 local_irq_disable();
2459 /* Add dead cpu's page_states to our own. */
2460 dest = (unsigned long *)&__get_cpu_var(page_states);
2461 src = (unsigned long *)&per_cpu(page_states, cpu);
2463 for (i = 0; i < sizeof(struct page_state)/sizeof(unsigned long);
2473 #endif /* CONFIG_HOTPLUG_CPU */
2475 void __init page_alloc_init(void)
2477 hotcpu_notifier(page_alloc_cpu_notify, 0);
2481 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
2482 * or min_free_kbytes changes.
2484 static void calculate_totalreserve_pages(void)
2486 struct pglist_data *pgdat;
2487 unsigned long reserve_pages = 0;
2490 for_each_online_pgdat(pgdat) {
2491 for (i = 0; i < MAX_NR_ZONES; i++) {
2492 struct zone *zone = pgdat->node_zones + i;
2493 unsigned long max = 0;
2495 /* Find valid and maximum lowmem_reserve in the zone */
2496 for (j = i; j < MAX_NR_ZONES; j++) {
2497 if (zone->lowmem_reserve[j] > max)
2498 max = zone->lowmem_reserve[j];
2501 /* we treat pages_high as reserved pages. */
2502 max += zone->pages_high;
2504 if (max > zone->present_pages)
2505 max = zone->present_pages;
2506 reserve_pages += max;
2509 totalreserve_pages = reserve_pages;
2513 * setup_per_zone_lowmem_reserve - called whenever
2514 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
2515 * has a correct pages reserved value, so an adequate number of
2516 * pages are left in the zone after a successful __alloc_pages().
2518 static void setup_per_zone_lowmem_reserve(void)
2520 struct pglist_data *pgdat;
2523 for_each_online_pgdat(pgdat) {
2524 for (j = 0; j < MAX_NR_ZONES; j++) {
2525 struct zone *zone = pgdat->node_zones + j;
2526 unsigned long present_pages = zone->present_pages;
2528 zone->lowmem_reserve[j] = 0;
2530 for (idx = j-1; idx >= 0; idx--) {
2531 struct zone *lower_zone;
2533 if (sysctl_lowmem_reserve_ratio[idx] < 1)
2534 sysctl_lowmem_reserve_ratio[idx] = 1;
2536 lower_zone = pgdat->node_zones + idx;
2537 lower_zone->lowmem_reserve[j] = present_pages /
2538 sysctl_lowmem_reserve_ratio[idx];
2539 present_pages += lower_zone->present_pages;
2544 /* update totalreserve_pages */
2545 calculate_totalreserve_pages();
2549 * setup_per_zone_pages_min - called when min_free_kbytes changes. Ensures
2550 * that the pages_{min,low,high} values for each zone are set correctly
2551 * with respect to min_free_kbytes.
2553 void setup_per_zone_pages_min(void)
2555 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
2556 unsigned long lowmem_pages = 0;
2558 unsigned long flags;
2560 /* Calculate total number of !ZONE_HIGHMEM pages */
2561 for_each_zone(zone) {
2562 if (!is_highmem(zone))
2563 lowmem_pages += zone->present_pages;
2566 for_each_zone(zone) {
2568 spin_lock_irqsave(&zone->lru_lock, flags);
2569 tmp = (pages_min * zone->present_pages) / lowmem_pages;
2570 if (is_highmem(zone)) {
2572 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
2573 * need highmem pages, so cap pages_min to a small
2576 * The (pages_high-pages_low) and (pages_low-pages_min)
2577 * deltas controls asynch page reclaim, and so should
2578 * not be capped for highmem.
2582 min_pages = zone->present_pages / 1024;
2583 if (min_pages < SWAP_CLUSTER_MAX)
2584 min_pages = SWAP_CLUSTER_MAX;
2585 if (min_pages > 128)
2587 zone->pages_min = min_pages;
2590 * If it's a lowmem zone, reserve a number of pages
2591 * proportionate to the zone's size.
2593 zone->pages_min = tmp;
2596 zone->pages_low = zone->pages_min + tmp / 4;
2597 zone->pages_high = zone->pages_min + tmp / 2;
2598 spin_unlock_irqrestore(&zone->lru_lock, flags);
2601 /* update totalreserve_pages */
2602 calculate_totalreserve_pages();
2606 * Initialise min_free_kbytes.
2608 * For small machines we want it small (128k min). For large machines
2609 * we want it large (64MB max). But it is not linear, because network
2610 * bandwidth does not increase linearly with machine size. We use
2612 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
2613 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
2629 static int __init init_per_zone_pages_min(void)
2631 unsigned long lowmem_kbytes;
2633 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
2635 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
2636 if (min_free_kbytes < 128)
2637 min_free_kbytes = 128;
2638 if (min_free_kbytes > 65536)
2639 min_free_kbytes = 65536;
2640 setup_per_zone_pages_min();
2641 setup_per_zone_lowmem_reserve();
2644 module_init(init_per_zone_pages_min)
2647 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
2648 * that we can call two helper functions whenever min_free_kbytes
2651 int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
2652 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2654 proc_dointvec(table, write, file, buffer, length, ppos);
2655 setup_per_zone_pages_min();
2660 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
2661 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
2662 * whenever sysctl_lowmem_reserve_ratio changes.
2664 * The reserve ratio obviously has absolutely no relation with the
2665 * pages_min watermarks. The lowmem reserve ratio can only make sense
2666 * if in function of the boot time zone sizes.
2668 int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
2669 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2671 proc_dointvec_minmax(table, write, file, buffer, length, ppos);
2672 setup_per_zone_lowmem_reserve();
2677 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
2678 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
2679 * can have before it gets flushed back to buddy allocator.
2682 int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
2683 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2689 ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
2690 if (!write || (ret == -EINVAL))
2692 for_each_zone(zone) {
2693 for_each_online_cpu(cpu) {
2695 high = zone->present_pages / percpu_pagelist_fraction;
2696 setup_pagelist_highmark(zone_pcp(zone, cpu), high);
2702 __initdata int hashdist = HASHDIST_DEFAULT;
2705 static int __init set_hashdist(char *str)
2709 hashdist = simple_strtoul(str, &str, 0);
2712 __setup("hashdist=", set_hashdist);
2716 * allocate a large system hash table from bootmem
2717 * - it is assumed that the hash table must contain an exact power-of-2
2718 * quantity of entries
2719 * - limit is the number of hash buckets, not the total allocation size
2721 void *__init alloc_large_system_hash(const char *tablename,
2722 unsigned long bucketsize,
2723 unsigned long numentries,
2726 unsigned int *_hash_shift,
2727 unsigned int *_hash_mask,
2728 unsigned long limit)
2730 unsigned long long max = limit;
2731 unsigned long log2qty, size;
2734 /* allow the kernel cmdline to have a say */
2736 /* round applicable memory size up to nearest megabyte */
2737 numentries = (flags & HASH_HIGHMEM) ? nr_all_pages : nr_kernel_pages;
2738 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
2739 numentries >>= 20 - PAGE_SHIFT;
2740 numentries <<= 20 - PAGE_SHIFT;
2742 /* limit to 1 bucket per 2^scale bytes of low memory */
2743 if (scale > PAGE_SHIFT)
2744 numentries >>= (scale - PAGE_SHIFT);
2746 numentries <<= (PAGE_SHIFT - scale);
2748 numentries = roundup_pow_of_two(numentries);
2750 /* limit allocation size to 1/16 total memory by default */
2752 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
2753 do_div(max, bucketsize);
2756 if (numentries > max)
2759 log2qty = long_log2(numentries);
2762 size = bucketsize << log2qty;
2763 if (flags & HASH_EARLY)
2764 table = alloc_bootmem(size);
2766 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
2768 unsigned long order;
2769 for (order = 0; ((1UL << order) << PAGE_SHIFT) < size; order++)
2771 table = (void*) __get_free_pages(GFP_ATOMIC, order);
2773 } while (!table && size > PAGE_SIZE && --log2qty);
2776 panic("Failed to allocate %s hash table\n", tablename);
2778 printk("%s hash table entries: %d (order: %d, %lu bytes)\n",
2781 long_log2(size) - PAGE_SHIFT,
2785 *_hash_shift = log2qty;
2787 *_hash_mask = (1 << log2qty) - 1;
2792 #ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE
2794 * pfn <-> page translation. out-of-line version.
2795 * (see asm-generic/memory_model.h)
2797 #if defined(CONFIG_FLATMEM)
2798 struct page *pfn_to_page(unsigned long pfn)
2800 return mem_map + (pfn - ARCH_PFN_OFFSET);
2802 unsigned long page_to_pfn(struct page *page)
2804 return (page - mem_map) + ARCH_PFN_OFFSET;
2806 #elif defined(CONFIG_DISCONTIGMEM)
2807 struct page *pfn_to_page(unsigned long pfn)
2809 int nid = arch_pfn_to_nid(pfn);
2810 return NODE_DATA(nid)->node_mem_map + arch_local_page_offset(pfn,nid);
2812 unsigned long page_to_pfn(struct page *page)
2814 struct pglist_data *pgdat = NODE_DATA(page_to_nid(page));
2815 return (page - pgdat->node_mem_map) + pgdat->node_start_pfn;
2817 #elif defined(CONFIG_SPARSEMEM)
2818 struct page *pfn_to_page(unsigned long pfn)
2820 return __section_mem_map_addr(__pfn_to_section(pfn)) + pfn;
2823 unsigned long page_to_pfn(struct page *page)
2825 long section_id = page_to_section(page);
2826 return page - __section_mem_map_addr(__nr_to_section(section_id));
2828 #endif /* CONFIG_FLATMEM/DISCONTIGMME/SPARSEMEM */
2829 EXPORT_SYMBOL(pfn_to_page);
2830 EXPORT_SYMBOL(page_to_pfn);
2831 #endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */