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/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/nodemask.h>
36 #include <linux/vmalloc.h>
38 #include <asm/tlbflush.h>
42 * MCD - HACK: Find somewhere to initialize this EARLY, or make this
45 nodemask_t node_online_map = { { [0] = 1UL } };
46 EXPORT_SYMBOL(node_online_map);
47 nodemask_t node_possible_map = NODE_MASK_ALL;
48 EXPORT_SYMBOL(node_possible_map);
49 struct pglist_data *pgdat_list;
50 unsigned long totalram_pages;
51 unsigned long totalhigh_pages;
55 * results with 256, 32 in the lowmem_reserve sysctl:
56 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
57 * 1G machine -> (16M dma, 784M normal, 224M high)
58 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
59 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
60 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
62 int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = { 256, 32 };
64 EXPORT_SYMBOL(totalram_pages);
65 EXPORT_SYMBOL(nr_swap_pages);
68 * Used by page_zone() to look up the address of the struct zone whose
69 * id is encoded in the upper bits of page->flags
71 struct zone *zone_table[1 << (ZONES_SHIFT + NODES_SHIFT)];
72 EXPORT_SYMBOL(zone_table);
75 static struct per_cpu_pageset
76 pageset_table[MAX_NR_ZONES*MAX_NUMNODES*NR_CPUS] __initdata;
79 static char *zone_names[MAX_NR_ZONES] = { "DMA", "Normal", "HighMem" };
80 int min_free_kbytes = 1024;
82 unsigned long __initdata nr_kernel_pages;
83 unsigned long __initdata nr_all_pages;
86 * Temporary debugging check for pages not lying within a given zone.
88 static int bad_range(struct zone *zone, struct page *page)
90 if (page_to_pfn(page) >= zone->zone_start_pfn + zone->spanned_pages)
92 if (page_to_pfn(page) < zone->zone_start_pfn)
94 #ifdef CONFIG_HOLES_IN_ZONE
95 if (!pfn_valid(page_to_pfn(page)))
98 if (zone != page_zone(page))
103 static void bad_page(const char *function, struct page *page)
105 printk(KERN_EMERG "Bad page state at %s (in process '%s', page %p)\n",
106 function, current->comm, page);
107 printk(KERN_EMERG "flags:0x%0*lx mapping:%p mapcount:%d count:%d\n",
108 (int)(2*sizeof(page_flags_t)), (unsigned long)page->flags,
109 page->mapping, page_mapcount(page), page_count(page));
110 printk(KERN_EMERG "Backtrace:\n");
112 printk(KERN_EMERG "Trying to fix it up, but a reboot is needed\n");
113 page->flags &= ~(1 << PG_private |
120 set_page_count(page, 0);
121 reset_page_mapcount(page);
122 page->mapping = NULL;
123 tainted |= TAINT_BAD_PAGE;
126 #ifndef CONFIG_HUGETLB_PAGE
127 #define prep_compound_page(page, order) do { } while (0)
128 #define destroy_compound_page(page, order) do { } while (0)
131 * Higher-order pages are called "compound pages". They are structured thusly:
133 * The first PAGE_SIZE page is called the "head page".
135 * The remaining PAGE_SIZE pages are called "tail pages".
137 * All pages have PG_compound set. All pages have their ->private pointing at
138 * the head page (even the head page has this).
140 * The first tail page's ->mapping, if non-zero, holds the address of the
141 * compound page's put_page() function.
143 * The order of the allocation is stored in the first tail page's ->index
144 * This is only for debug at present. This usage means that zero-order pages
145 * may not be compound.
147 static void prep_compound_page(struct page *page, unsigned long order)
150 int nr_pages = 1 << order;
152 page[1].mapping = NULL;
153 page[1].index = order;
154 for (i = 0; i < nr_pages; i++) {
155 struct page *p = page + i;
158 p->private = (unsigned long)page;
162 static void destroy_compound_page(struct page *page, unsigned long order)
165 int nr_pages = 1 << order;
167 if (!PageCompound(page))
170 if (page[1].index != order)
171 bad_page(__FUNCTION__, page);
173 for (i = 0; i < nr_pages; i++) {
174 struct page *p = page + i;
176 if (!PageCompound(p))
177 bad_page(__FUNCTION__, page);
178 if (p->private != (unsigned long)page)
179 bad_page(__FUNCTION__, page);
180 ClearPageCompound(p);
183 #endif /* CONFIG_HUGETLB_PAGE */
186 * function for dealing with page's order in buddy system.
187 * zone->lock is already acquired when we use these.
188 * So, we don't need atomic page->flags operations here.
190 static inline unsigned long page_order(struct page *page) {
191 return page->private;
194 static inline void set_page_order(struct page *page, int order) {
195 page->private = order;
196 __SetPagePrivate(page);
199 static inline void rmv_page_order(struct page *page)
201 __ClearPagePrivate(page);
206 * Locate the struct page for both the matching buddy in our
207 * pair (buddy1) and the combined O(n+1) page they form (page).
209 * 1) Any buddy B1 will have an order O twin B2 which satisfies
210 * the following equation:
212 * For example, if the starting buddy (buddy2) is #8 its order
214 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
216 * 2) Any buddy B will have an order O+1 parent P which
217 * satisfies the following equation:
220 * Assumption: *_mem_map is contigious at least up to MAX_ORDER
222 static inline struct page *
223 __page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order)
225 unsigned long buddy_idx = page_idx ^ (1 << order);
227 return page + (buddy_idx - page_idx);
230 static inline unsigned long
231 __find_combined_index(unsigned long page_idx, unsigned int order)
233 return (page_idx & ~(1 << order));
237 * This function checks whether a page is free && is the buddy
238 * we can do coalesce a page and its buddy if
239 * (a) the buddy is free &&
240 * (b) the buddy is on the buddy system &&
241 * (c) a page and its buddy have the same order.
242 * for recording page's order, we use page->private and PG_private.
245 static inline int page_is_buddy(struct page *page, int order)
247 if (PagePrivate(page) &&
248 (page_order(page) == order) &&
249 !PageReserved(page) &&
250 page_count(page) == 0)
256 * Freeing function for a buddy system allocator.
258 * The concept of a buddy system is to maintain direct-mapped table
259 * (containing bit values) for memory blocks of various "orders".
260 * The bottom level table contains the map for the smallest allocatable
261 * units of memory (here, pages), and each level above it describes
262 * pairs of units from the levels below, hence, "buddies".
263 * At a high level, all that happens here is marking the table entry
264 * at the bottom level available, and propagating the changes upward
265 * as necessary, plus some accounting needed to play nicely with other
266 * parts of the VM system.
267 * At each level, we keep a list of pages, which are heads of continuous
268 * free pages of length of (1 << order) and marked with PG_Private.Page's
269 * order is recorded in page->private field.
270 * So when we are allocating or freeing one, we can derive the state of the
271 * other. That is, if we allocate a small block, and both were
272 * free, the remainder of the region must be split into blocks.
273 * If a block is freed, and its buddy is also free, then this
274 * triggers coalescing into a block of larger size.
279 static inline void __free_pages_bulk (struct page *page,
280 struct zone *zone, unsigned int order)
282 unsigned long page_idx;
283 int order_size = 1 << order;
286 destroy_compound_page(page, order);
288 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
290 BUG_ON(page_idx & (order_size - 1));
291 BUG_ON(bad_range(zone, page));
293 zone->free_pages += order_size;
294 while (order < MAX_ORDER-1) {
295 unsigned long combined_idx;
296 struct free_area *area;
299 combined_idx = __find_combined_index(page_idx, order);
300 buddy = __page_find_buddy(page, page_idx, order);
302 if (bad_range(zone, buddy))
304 if (!page_is_buddy(buddy, order))
305 break; /* Move the buddy up one level. */
306 list_del(&buddy->lru);
307 area = zone->free_area + order;
309 rmv_page_order(buddy);
310 page = page + (combined_idx - page_idx);
311 page_idx = combined_idx;
314 set_page_order(page, order);
315 list_add(&page->lru, &zone->free_area[order].free_list);
316 zone->free_area[order].nr_free++;
319 static inline void free_pages_check(const char *function, struct page *page)
321 if ( page_mapcount(page) ||
322 page->mapping != NULL ||
323 page_count(page) != 0 ||
332 1 << PG_writeback )))
333 bad_page(function, page);
335 ClearPageDirty(page);
339 * Frees a list of pages.
340 * Assumes all pages on list are in same zone, and of same order.
341 * count is the number of pages to free, or 0 for all on the list.
343 * If the zone was previously in an "all pages pinned" state then look to
344 * see if this freeing clears that state.
346 * And clear the zone's pages_scanned counter, to hold off the "all pages are
347 * pinned" detection logic.
350 free_pages_bulk(struct zone *zone, int count,
351 struct list_head *list, unsigned int order)
354 struct page *page = NULL;
357 spin_lock_irqsave(&zone->lock, flags);
358 zone->all_unreclaimable = 0;
359 zone->pages_scanned = 0;
360 while (!list_empty(list) && count--) {
361 page = list_entry(list->prev, struct page, lru);
362 /* have to delete it as __free_pages_bulk list manipulates */
363 list_del(&page->lru);
364 __free_pages_bulk(page, zone, order);
367 spin_unlock_irqrestore(&zone->lock, flags);
371 void __free_pages_ok(struct page *page, unsigned int order)
376 arch_free_page(page, order);
378 mod_page_state(pgfree, 1 << order);
382 for (i = 1 ; i < (1 << order) ; ++i)
383 __put_page(page + i);
386 for (i = 0 ; i < (1 << order) ; ++i)
387 free_pages_check(__FUNCTION__, page + i);
388 list_add(&page->lru, &list);
389 kernel_map_pages(page, 1<<order, 0);
390 free_pages_bulk(page_zone(page), 1, &list, order);
395 * The order of subdivision here is critical for the IO subsystem.
396 * Please do not alter this order without good reasons and regression
397 * testing. Specifically, as large blocks of memory are subdivided,
398 * the order in which smaller blocks are delivered depends on the order
399 * they're subdivided in this function. This is the primary factor
400 * influencing the order in which pages are delivered to the IO
401 * subsystem according to empirical testing, and this is also justified
402 * by considering the behavior of a buddy system containing a single
403 * large block of memory acted on by a series of small allocations.
404 * This behavior is a critical factor in sglist merging's success.
408 static inline struct page *
409 expand(struct zone *zone, struct page *page,
410 int low, int high, struct free_area *area)
412 unsigned long size = 1 << high;
418 BUG_ON(bad_range(zone, &page[size]));
419 list_add(&page[size].lru, &area->free_list);
421 set_page_order(&page[size], high);
426 void set_page_refs(struct page *page, int order)
429 set_page_count(page, 1);
434 * We need to reference all the pages for this order, otherwise if
435 * anyone accesses one of the pages with (get/put) it will be freed.
436 * - eg: access_process_vm()
438 for (i = 0; i < (1 << order); i++)
439 set_page_count(page + i, 1);
440 #endif /* CONFIG_MMU */
444 * This page is about to be returned from the page allocator
446 static void prep_new_page(struct page *page, int order)
448 if (page->mapping || page_mapcount(page) ||
457 1 << PG_writeback )))
458 bad_page(__FUNCTION__, page);
460 page->flags &= ~(1 << PG_uptodate | 1 << PG_error |
461 1 << PG_referenced | 1 << PG_arch_1 |
462 1 << PG_checked | 1 << PG_mappedtodisk);
464 set_page_refs(page, order);
465 kernel_map_pages(page, 1 << order, 1);
469 * Do the hard work of removing an element from the buddy allocator.
470 * Call me with the zone->lock already held.
472 static struct page *__rmqueue(struct zone *zone, unsigned int order)
474 struct free_area * area;
475 unsigned int current_order;
478 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
479 area = zone->free_area + current_order;
480 if (list_empty(&area->free_list))
483 page = list_entry(area->free_list.next, struct page, lru);
484 list_del(&page->lru);
485 rmv_page_order(page);
487 zone->free_pages -= 1UL << order;
488 return expand(zone, page, order, current_order, area);
495 * Obtain a specified number of elements from the buddy allocator, all under
496 * a single hold of the lock, for efficiency. Add them to the supplied list.
497 * Returns the number of new pages which were placed at *list.
499 static int rmqueue_bulk(struct zone *zone, unsigned int order,
500 unsigned long count, struct list_head *list)
507 spin_lock_irqsave(&zone->lock, flags);
508 for (i = 0; i < count; ++i) {
509 page = __rmqueue(zone, order);
513 list_add_tail(&page->lru, list);
515 spin_unlock_irqrestore(&zone->lock, flags);
519 #if defined(CONFIG_PM) || defined(CONFIG_HOTPLUG_CPU)
520 static void __drain_pages(unsigned int cpu)
525 for_each_zone(zone) {
526 struct per_cpu_pageset *pset;
528 pset = zone_pcp(zone, cpu);
529 for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) {
530 struct per_cpu_pages *pcp;
533 pcp->count -= free_pages_bulk(zone, pcp->count,
538 #endif /* CONFIG_PM || CONFIG_HOTPLUG_CPU */
542 void mark_free_pages(struct zone *zone)
544 unsigned long zone_pfn, flags;
546 struct list_head *curr;
548 if (!zone->spanned_pages)
551 spin_lock_irqsave(&zone->lock, flags);
552 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn)
553 ClearPageNosaveFree(pfn_to_page(zone_pfn + zone->zone_start_pfn));
555 for (order = MAX_ORDER - 1; order >= 0; --order)
556 list_for_each(curr, &zone->free_area[order].free_list) {
557 unsigned long start_pfn, i;
559 start_pfn = page_to_pfn(list_entry(curr, struct page, lru));
561 for (i=0; i < (1<<order); i++)
562 SetPageNosaveFree(pfn_to_page(start_pfn+i));
564 spin_unlock_irqrestore(&zone->lock, flags);
568 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
570 void drain_local_pages(void)
574 local_irq_save(flags);
575 __drain_pages(smp_processor_id());
576 local_irq_restore(flags);
578 #endif /* CONFIG_PM */
580 static void zone_statistics(struct zonelist *zonelist, struct zone *z)
585 pg_data_t *pg = z->zone_pgdat;
586 pg_data_t *orig = zonelist->zones[0]->zone_pgdat;
587 struct per_cpu_pageset *p;
589 local_irq_save(flags);
590 cpu = smp_processor_id();
596 zone_pcp(zonelist->zones[0], cpu)->numa_foreign++;
598 if (pg == NODE_DATA(numa_node_id()))
602 local_irq_restore(flags);
607 * Free a 0-order page
609 static void FASTCALL(free_hot_cold_page(struct page *page, int cold));
610 static void fastcall free_hot_cold_page(struct page *page, int cold)
612 struct zone *zone = page_zone(page);
613 struct per_cpu_pages *pcp;
616 arch_free_page(page, 0);
618 kernel_map_pages(page, 1, 0);
619 inc_page_state(pgfree);
621 page->mapping = NULL;
622 free_pages_check(__FUNCTION__, page);
623 pcp = &zone_pcp(zone, get_cpu())->pcp[cold];
624 local_irq_save(flags);
625 if (pcp->count >= pcp->high)
626 pcp->count -= free_pages_bulk(zone, pcp->batch, &pcp->list, 0);
627 list_add(&page->lru, &pcp->list);
629 local_irq_restore(flags);
633 void fastcall free_hot_page(struct page *page)
635 free_hot_cold_page(page, 0);
638 void fastcall free_cold_page(struct page *page)
640 free_hot_cold_page(page, 1);
643 static inline void prep_zero_page(struct page *page, int order, unsigned int __nocast gfp_flags)
647 BUG_ON((gfp_flags & (__GFP_WAIT | __GFP_HIGHMEM)) == __GFP_HIGHMEM);
648 for(i = 0; i < (1 << order); i++)
649 clear_highpage(page + i);
653 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
654 * we cheat by calling it from here, in the order > 0 path. Saves a branch
658 buffered_rmqueue(struct zone *zone, int order, unsigned int __nocast gfp_flags)
661 struct page *page = NULL;
662 int cold = !!(gfp_flags & __GFP_COLD);
665 struct per_cpu_pages *pcp;
667 pcp = &zone_pcp(zone, get_cpu())->pcp[cold];
668 local_irq_save(flags);
669 if (pcp->count <= pcp->low)
670 pcp->count += rmqueue_bulk(zone, 0,
671 pcp->batch, &pcp->list);
673 page = list_entry(pcp->list.next, struct page, lru);
674 list_del(&page->lru);
677 local_irq_restore(flags);
682 spin_lock_irqsave(&zone->lock, flags);
683 page = __rmqueue(zone, order);
684 spin_unlock_irqrestore(&zone->lock, flags);
688 BUG_ON(bad_range(zone, page));
689 mod_page_state_zone(zone, pgalloc, 1 << order);
690 prep_new_page(page, order);
692 if (gfp_flags & __GFP_ZERO)
693 prep_zero_page(page, order, gfp_flags);
695 if (order && (gfp_flags & __GFP_COMP))
696 prep_compound_page(page, order);
702 * Return 1 if free pages are above 'mark'. This takes into account the order
705 int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
706 int classzone_idx, int can_try_harder, int gfp_high)
708 /* free_pages my go negative - that's OK */
709 long min = mark, free_pages = z->free_pages - (1 << order) + 1;
717 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
719 for (o = 0; o < order; o++) {
720 /* At the next order, this order's pages become unavailable */
721 free_pages -= z->free_area[o].nr_free << o;
723 /* Require fewer higher order pages to be free */
726 if (free_pages <= min)
733 should_reclaim_zone(struct zone *z, unsigned int gfp_mask)
735 if (!z->reclaim_pages)
737 if (gfp_mask & __GFP_NORECLAIM)
743 * This is the 'heart' of the zoned buddy allocator.
745 struct page * fastcall
746 __alloc_pages(unsigned int __nocast gfp_mask, unsigned int order,
747 struct zonelist *zonelist)
749 const int wait = gfp_mask & __GFP_WAIT;
750 struct zone **zones, *z;
752 struct reclaim_state reclaim_state;
753 struct task_struct *p = current;
758 int did_some_progress;
760 might_sleep_if(wait);
763 * The caller may dip into page reserves a bit more if the caller
764 * cannot run direct reclaim, or is the caller has realtime scheduling
767 can_try_harder = (unlikely(rt_task(p)) && !in_interrupt()) || !wait;
769 zones = zonelist->zones; /* the list of zones suitable for gfp_mask */
771 if (unlikely(zones[0] == NULL)) {
772 /* Should this ever happen?? */
776 classzone_idx = zone_idx(zones[0]);
779 /* Go through the zonelist once, looking for a zone with enough free */
780 for (i = 0; (z = zones[i]) != NULL; i++) {
781 int do_reclaim = should_reclaim_zone(z, gfp_mask);
783 if (!cpuset_zone_allowed(z))
787 * If the zone is to attempt early page reclaim then this loop
788 * will try to reclaim pages and check the watermark a second
789 * time before giving up and falling back to the next zone.
792 if (!zone_watermark_ok(z, order, z->pages_low,
793 classzone_idx, 0, 0)) {
797 zone_reclaim(z, gfp_mask, order);
798 /* Only try reclaim once */
800 goto zone_reclaim_retry;
804 page = buffered_rmqueue(z, order, gfp_mask);
809 for (i = 0; (z = zones[i]) != NULL; i++)
810 wakeup_kswapd(z, order);
813 * Go through the zonelist again. Let __GFP_HIGH and allocations
814 * coming from realtime tasks to go deeper into reserves
816 * This is the last chance, in general, before the goto nopage.
817 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
819 for (i = 0; (z = zones[i]) != NULL; i++) {
820 if (!zone_watermark_ok(z, order, z->pages_min,
821 classzone_idx, can_try_harder,
822 gfp_mask & __GFP_HIGH))
825 if (wait && !cpuset_zone_allowed(z))
828 page = buffered_rmqueue(z, order, gfp_mask);
833 /* This allocation should allow future memory freeing. */
835 if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE)))
836 && !in_interrupt()) {
837 if (!(gfp_mask & __GFP_NOMEMALLOC)) {
838 /* go through the zonelist yet again, ignoring mins */
839 for (i = 0; (z = zones[i]) != NULL; i++) {
840 if (!cpuset_zone_allowed(z))
842 page = buffered_rmqueue(z, order, gfp_mask);
850 /* Atomic allocations - we can't balance anything */
857 /* We now go into synchronous reclaim */
858 p->flags |= PF_MEMALLOC;
859 reclaim_state.reclaimed_slab = 0;
860 p->reclaim_state = &reclaim_state;
862 did_some_progress = try_to_free_pages(zones, gfp_mask);
864 p->reclaim_state = NULL;
865 p->flags &= ~PF_MEMALLOC;
869 if (likely(did_some_progress)) {
871 * Go through the zonelist yet one more time, keep
872 * very high watermark here, this is only to catch
873 * a parallel oom killing, we must fail if we're still
874 * under heavy pressure.
876 for (i = 0; (z = zones[i]) != NULL; i++) {
877 if (!zone_watermark_ok(z, order, z->pages_min,
878 classzone_idx, can_try_harder,
879 gfp_mask & __GFP_HIGH))
882 if (!cpuset_zone_allowed(z))
885 page = buffered_rmqueue(z, order, gfp_mask);
889 } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
891 * Go through the zonelist yet one more time, keep
892 * very high watermark here, this is only to catch
893 * a parallel oom killing, we must fail if we're still
894 * under heavy pressure.
896 for (i = 0; (z = zones[i]) != NULL; i++) {
897 if (!zone_watermark_ok(z, order, z->pages_high,
898 classzone_idx, 0, 0))
901 if (!cpuset_zone_allowed(z))
904 page = buffered_rmqueue(z, order, gfp_mask);
909 out_of_memory(gfp_mask);
914 * Don't let big-order allocations loop unless the caller explicitly
915 * requests that. Wait for some write requests to complete then retry.
917 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
918 * <= 3, but that may not be true in other implementations.
921 if (!(gfp_mask & __GFP_NORETRY)) {
922 if ((order <= 3) || (gfp_mask & __GFP_REPEAT))
924 if (gfp_mask & __GFP_NOFAIL)
928 blk_congestion_wait(WRITE, HZ/50);
933 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
934 printk(KERN_WARNING "%s: page allocation failure."
935 " order:%d, mode:0x%x\n",
936 p->comm, order, gfp_mask);
942 zone_statistics(zonelist, z);
946 EXPORT_SYMBOL(__alloc_pages);
949 * Common helper functions.
951 fastcall unsigned long __get_free_pages(unsigned int __nocast gfp_mask, unsigned int order)
954 page = alloc_pages(gfp_mask, order);
957 return (unsigned long) page_address(page);
960 EXPORT_SYMBOL(__get_free_pages);
962 fastcall unsigned long get_zeroed_page(unsigned int __nocast gfp_mask)
967 * get_zeroed_page() returns a 32-bit address, which cannot represent
970 BUG_ON(gfp_mask & __GFP_HIGHMEM);
972 page = alloc_pages(gfp_mask | __GFP_ZERO, 0);
974 return (unsigned long) page_address(page);
978 EXPORT_SYMBOL(get_zeroed_page);
980 void __pagevec_free(struct pagevec *pvec)
982 int i = pagevec_count(pvec);
985 free_hot_cold_page(pvec->pages[i], pvec->cold);
988 fastcall void __free_pages(struct page *page, unsigned int order)
990 if (!PageReserved(page) && put_page_testzero(page)) {
994 __free_pages_ok(page, order);
998 EXPORT_SYMBOL(__free_pages);
1000 fastcall void free_pages(unsigned long addr, unsigned int order)
1003 BUG_ON(!virt_addr_valid((void *)addr));
1004 __free_pages(virt_to_page((void *)addr), order);
1008 EXPORT_SYMBOL(free_pages);
1011 * Total amount of free (allocatable) RAM:
1013 unsigned int nr_free_pages(void)
1015 unsigned int sum = 0;
1019 sum += zone->free_pages;
1024 EXPORT_SYMBOL(nr_free_pages);
1027 unsigned int nr_free_pages_pgdat(pg_data_t *pgdat)
1029 unsigned int i, sum = 0;
1031 for (i = 0; i < MAX_NR_ZONES; i++)
1032 sum += pgdat->node_zones[i].free_pages;
1038 static unsigned int nr_free_zone_pages(int offset)
1041 unsigned int sum = 0;
1043 for_each_pgdat(pgdat) {
1044 struct zonelist *zonelist = pgdat->node_zonelists + offset;
1045 struct zone **zonep = zonelist->zones;
1048 for (zone = *zonep++; zone; zone = *zonep++) {
1049 unsigned long size = zone->present_pages;
1050 unsigned long high = zone->pages_high;
1060 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1062 unsigned int nr_free_buffer_pages(void)
1064 return nr_free_zone_pages(GFP_USER & GFP_ZONEMASK);
1068 * Amount of free RAM allocatable within all zones
1070 unsigned int nr_free_pagecache_pages(void)
1072 return nr_free_zone_pages(GFP_HIGHUSER & GFP_ZONEMASK);
1075 #ifdef CONFIG_HIGHMEM
1076 unsigned int nr_free_highpages (void)
1079 unsigned int pages = 0;
1081 for_each_pgdat(pgdat)
1082 pages += pgdat->node_zones[ZONE_HIGHMEM].free_pages;
1089 static void show_node(struct zone *zone)
1091 printk("Node %d ", zone->zone_pgdat->node_id);
1094 #define show_node(zone) do { } while (0)
1098 * Accumulate the page_state information across all CPUs.
1099 * The result is unavoidably approximate - it can change
1100 * during and after execution of this function.
1102 static DEFINE_PER_CPU(struct page_state, page_states) = {0};
1104 atomic_t nr_pagecache = ATOMIC_INIT(0);
1105 EXPORT_SYMBOL(nr_pagecache);
1107 DEFINE_PER_CPU(long, nr_pagecache_local) = 0;
1110 void __get_page_state(struct page_state *ret, int nr)
1114 memset(ret, 0, sizeof(*ret));
1116 cpu = first_cpu(cpu_online_map);
1117 while (cpu < NR_CPUS) {
1118 unsigned long *in, *out, off;
1120 in = (unsigned long *)&per_cpu(page_states, cpu);
1122 cpu = next_cpu(cpu, cpu_online_map);
1125 prefetch(&per_cpu(page_states, cpu));
1127 out = (unsigned long *)ret;
1128 for (off = 0; off < nr; off++)
1133 void get_page_state(struct page_state *ret)
1137 nr = offsetof(struct page_state, GET_PAGE_STATE_LAST);
1138 nr /= sizeof(unsigned long);
1140 __get_page_state(ret, nr + 1);
1143 void get_full_page_state(struct page_state *ret)
1145 __get_page_state(ret, sizeof(*ret) / sizeof(unsigned long));
1148 unsigned long __read_page_state(unsigned long offset)
1150 unsigned long ret = 0;
1153 for_each_online_cpu(cpu) {
1156 in = (unsigned long)&per_cpu(page_states, cpu) + offset;
1157 ret += *((unsigned long *)in);
1162 void __mod_page_state(unsigned long offset, unsigned long delta)
1164 unsigned long flags;
1167 local_irq_save(flags);
1168 ptr = &__get_cpu_var(page_states);
1169 *(unsigned long*)(ptr + offset) += delta;
1170 local_irq_restore(flags);
1173 EXPORT_SYMBOL(__mod_page_state);
1175 void __get_zone_counts(unsigned long *active, unsigned long *inactive,
1176 unsigned long *free, struct pglist_data *pgdat)
1178 struct zone *zones = pgdat->node_zones;
1184 for (i = 0; i < MAX_NR_ZONES; i++) {
1185 *active += zones[i].nr_active;
1186 *inactive += zones[i].nr_inactive;
1187 *free += zones[i].free_pages;
1191 void get_zone_counts(unsigned long *active,
1192 unsigned long *inactive, unsigned long *free)
1194 struct pglist_data *pgdat;
1199 for_each_pgdat(pgdat) {
1200 unsigned long l, m, n;
1201 __get_zone_counts(&l, &m, &n, pgdat);
1208 void si_meminfo(struct sysinfo *val)
1210 val->totalram = totalram_pages;
1212 val->freeram = nr_free_pages();
1213 val->bufferram = nr_blockdev_pages();
1214 #ifdef CONFIG_HIGHMEM
1215 val->totalhigh = totalhigh_pages;
1216 val->freehigh = nr_free_highpages();
1221 val->mem_unit = PAGE_SIZE;
1224 EXPORT_SYMBOL(si_meminfo);
1227 void si_meminfo_node(struct sysinfo *val, int nid)
1229 pg_data_t *pgdat = NODE_DATA(nid);
1231 val->totalram = pgdat->node_present_pages;
1232 val->freeram = nr_free_pages_pgdat(pgdat);
1233 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1234 val->freehigh = pgdat->node_zones[ZONE_HIGHMEM].free_pages;
1235 val->mem_unit = PAGE_SIZE;
1239 #define K(x) ((x) << (PAGE_SHIFT-10))
1242 * Show free area list (used inside shift_scroll-lock stuff)
1243 * We also calculate the percentage fragmentation. We do this by counting the
1244 * memory on each free list with the exception of the first item on the list.
1246 void show_free_areas(void)
1248 struct page_state ps;
1249 int cpu, temperature;
1250 unsigned long active;
1251 unsigned long inactive;
1255 for_each_zone(zone) {
1257 printk("%s per-cpu:", zone->name);
1259 if (!zone->present_pages) {
1265 for (cpu = 0; cpu < NR_CPUS; ++cpu) {
1266 struct per_cpu_pageset *pageset;
1268 if (!cpu_possible(cpu))
1271 pageset = zone_pcp(zone, cpu);
1273 for (temperature = 0; temperature < 2; temperature++)
1274 printk("cpu %d %s: low %d, high %d, batch %d\n",
1276 temperature ? "cold" : "hot",
1277 pageset->pcp[temperature].low,
1278 pageset->pcp[temperature].high,
1279 pageset->pcp[temperature].batch);
1283 get_page_state(&ps);
1284 get_zone_counts(&active, &inactive, &free);
1286 printk("\nFree pages: %11ukB (%ukB HighMem)\n",
1288 K(nr_free_highpages()));
1290 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu "
1291 "unstable:%lu free:%u slab:%lu mapped:%lu pagetables:%lu\n",
1300 ps.nr_page_table_pages);
1302 for_each_zone(zone) {
1314 " pages_scanned:%lu"
1315 " all_unreclaimable? %s"
1318 K(zone->free_pages),
1321 K(zone->pages_high),
1323 K(zone->nr_inactive),
1324 K(zone->present_pages),
1325 zone->pages_scanned,
1326 (zone->all_unreclaimable ? "yes" : "no")
1328 printk("lowmem_reserve[]:");
1329 for (i = 0; i < MAX_NR_ZONES; i++)
1330 printk(" %lu", zone->lowmem_reserve[i]);
1334 for_each_zone(zone) {
1335 unsigned long nr, flags, order, total = 0;
1338 printk("%s: ", zone->name);
1339 if (!zone->present_pages) {
1344 spin_lock_irqsave(&zone->lock, flags);
1345 for (order = 0; order < MAX_ORDER; order++) {
1346 nr = zone->free_area[order].nr_free;
1347 total += nr << order;
1348 printk("%lu*%lukB ", nr, K(1UL) << order);
1350 spin_unlock_irqrestore(&zone->lock, flags);
1351 printk("= %lukB\n", K(total));
1354 show_swap_cache_info();
1358 * Builds allocation fallback zone lists.
1360 static int __init build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist, int j, int k)
1367 zone = pgdat->node_zones + ZONE_HIGHMEM;
1368 if (zone->present_pages) {
1369 #ifndef CONFIG_HIGHMEM
1372 zonelist->zones[j++] = zone;
1375 zone = pgdat->node_zones + ZONE_NORMAL;
1376 if (zone->present_pages)
1377 zonelist->zones[j++] = zone;
1379 zone = pgdat->node_zones + ZONE_DMA;
1380 if (zone->present_pages)
1381 zonelist->zones[j++] = zone;
1388 #define MAX_NODE_LOAD (num_online_nodes())
1389 static int __initdata node_load[MAX_NUMNODES];
1391 * find_next_best_node - find the next node that should appear in a given node's fallback list
1392 * @node: node whose fallback list we're appending
1393 * @used_node_mask: nodemask_t of already used nodes
1395 * We use a number of factors to determine which is the next node that should
1396 * appear on a given node's fallback list. The node should not have appeared
1397 * already in @node's fallback list, and it should be the next closest node
1398 * according to the distance array (which contains arbitrary distance values
1399 * from each node to each node in the system), and should also prefer nodes
1400 * with no CPUs, since presumably they'll have very little allocation pressure
1401 * on them otherwise.
1402 * It returns -1 if no node is found.
1404 static int __init find_next_best_node(int node, nodemask_t *used_node_mask)
1407 int min_val = INT_MAX;
1410 for_each_online_node(i) {
1413 /* Start from local node */
1414 n = (node+i) % num_online_nodes();
1416 /* Don't want a node to appear more than once */
1417 if (node_isset(n, *used_node_mask))
1420 /* Use the local node if we haven't already */
1421 if (!node_isset(node, *used_node_mask)) {
1426 /* Use the distance array to find the distance */
1427 val = node_distance(node, n);
1429 /* Give preference to headless and unused nodes */
1430 tmp = node_to_cpumask(n);
1431 if (!cpus_empty(tmp))
1432 val += PENALTY_FOR_NODE_WITH_CPUS;
1434 /* Slight preference for less loaded node */
1435 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
1436 val += node_load[n];
1438 if (val < min_val) {
1445 node_set(best_node, *used_node_mask);
1450 static void __init build_zonelists(pg_data_t *pgdat)
1452 int i, j, k, node, local_node;
1453 int prev_node, load;
1454 struct zonelist *zonelist;
1455 nodemask_t used_mask;
1457 /* initialize zonelists */
1458 for (i = 0; i < GFP_ZONETYPES; i++) {
1459 zonelist = pgdat->node_zonelists + i;
1460 zonelist->zones[0] = NULL;
1463 /* NUMA-aware ordering of nodes */
1464 local_node = pgdat->node_id;
1465 load = num_online_nodes();
1466 prev_node = local_node;
1467 nodes_clear(used_mask);
1468 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
1470 * We don't want to pressure a particular node.
1471 * So adding penalty to the first node in same
1472 * distance group to make it round-robin.
1474 if (node_distance(local_node, node) !=
1475 node_distance(local_node, prev_node))
1476 node_load[node] += load;
1479 for (i = 0; i < GFP_ZONETYPES; i++) {
1480 zonelist = pgdat->node_zonelists + i;
1481 for (j = 0; zonelist->zones[j] != NULL; j++);
1484 if (i & __GFP_HIGHMEM)
1489 j = build_zonelists_node(NODE_DATA(node), zonelist, j, k);
1490 zonelist->zones[j] = NULL;
1495 #else /* CONFIG_NUMA */
1497 static void __init build_zonelists(pg_data_t *pgdat)
1499 int i, j, k, node, local_node;
1501 local_node = pgdat->node_id;
1502 for (i = 0; i < GFP_ZONETYPES; i++) {
1503 struct zonelist *zonelist;
1505 zonelist = pgdat->node_zonelists + i;
1509 if (i & __GFP_HIGHMEM)
1514 j = build_zonelists_node(pgdat, zonelist, j, k);
1516 * Now we build the zonelist so that it contains the zones
1517 * of all the other nodes.
1518 * We don't want to pressure a particular node, so when
1519 * building the zones for node N, we make sure that the
1520 * zones coming right after the local ones are those from
1521 * node N+1 (modulo N)
1523 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
1524 if (!node_online(node))
1526 j = build_zonelists_node(NODE_DATA(node), zonelist, j, k);
1528 for (node = 0; node < local_node; node++) {
1529 if (!node_online(node))
1531 j = build_zonelists_node(NODE_DATA(node), zonelist, j, k);
1534 zonelist->zones[j] = NULL;
1538 #endif /* CONFIG_NUMA */
1540 void __init build_all_zonelists(void)
1544 for_each_online_node(i)
1545 build_zonelists(NODE_DATA(i));
1546 printk("Built %i zonelists\n", num_online_nodes());
1547 cpuset_init_current_mems_allowed();
1551 * Helper functions to size the waitqueue hash table.
1552 * Essentially these want to choose hash table sizes sufficiently
1553 * large so that collisions trying to wait on pages are rare.
1554 * But in fact, the number of active page waitqueues on typical
1555 * systems is ridiculously low, less than 200. So this is even
1556 * conservative, even though it seems large.
1558 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
1559 * waitqueues, i.e. the size of the waitq table given the number of pages.
1561 #define PAGES_PER_WAITQUEUE 256
1563 static inline unsigned long wait_table_size(unsigned long pages)
1565 unsigned long size = 1;
1567 pages /= PAGES_PER_WAITQUEUE;
1569 while (size < pages)
1573 * Once we have dozens or even hundreds of threads sleeping
1574 * on IO we've got bigger problems than wait queue collision.
1575 * Limit the size of the wait table to a reasonable size.
1577 size = min(size, 4096UL);
1579 return max(size, 4UL);
1583 * This is an integer logarithm so that shifts can be used later
1584 * to extract the more random high bits from the multiplicative
1585 * hash function before the remainder is taken.
1587 static inline unsigned long wait_table_bits(unsigned long size)
1592 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
1594 static void __init calculate_zone_totalpages(struct pglist_data *pgdat,
1595 unsigned long *zones_size, unsigned long *zholes_size)
1597 unsigned long realtotalpages, totalpages = 0;
1600 for (i = 0; i < MAX_NR_ZONES; i++)
1601 totalpages += zones_size[i];
1602 pgdat->node_spanned_pages = totalpages;
1604 realtotalpages = totalpages;
1606 for (i = 0; i < MAX_NR_ZONES; i++)
1607 realtotalpages -= zholes_size[i];
1608 pgdat->node_present_pages = realtotalpages;
1609 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, realtotalpages);
1614 * Initially all pages are reserved - free ones are freed
1615 * up by free_all_bootmem() once the early boot process is
1616 * done. Non-atomic initialization, single-pass.
1618 void __init memmap_init_zone(unsigned long size, int nid, unsigned long zone,
1619 unsigned long start_pfn)
1621 struct page *start = pfn_to_page(start_pfn);
1624 for (page = start; page < (start + size); page++) {
1625 set_page_zone(page, NODEZONE(nid, zone));
1626 set_page_count(page, 0);
1627 reset_page_mapcount(page);
1628 SetPageReserved(page);
1629 INIT_LIST_HEAD(&page->lru);
1630 #ifdef WANT_PAGE_VIRTUAL
1631 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
1632 if (!is_highmem_idx(zone))
1633 set_page_address(page, __va(start_pfn << PAGE_SHIFT));
1639 void zone_init_free_lists(struct pglist_data *pgdat, struct zone *zone,
1643 for (order = 0; order < MAX_ORDER ; order++) {
1644 INIT_LIST_HEAD(&zone->free_area[order].free_list);
1645 zone->free_area[order].nr_free = 0;
1649 #ifndef __HAVE_ARCH_MEMMAP_INIT
1650 #define memmap_init(size, nid, zone, start_pfn) \
1651 memmap_init_zone((size), (nid), (zone), (start_pfn))
1654 static int __devinit zone_batchsize(struct zone *zone)
1659 * The per-cpu-pages pools are set to around 1000th of the
1660 * size of the zone. But no more than 1/4 of a meg - there's
1661 * no point in going beyond the size of L2 cache.
1663 * OK, so we don't know how big the cache is. So guess.
1665 batch = zone->present_pages / 1024;
1666 if (batch * PAGE_SIZE > 256 * 1024)
1667 batch = (256 * 1024) / PAGE_SIZE;
1668 batch /= 4; /* We effectively *= 4 below */
1673 * Clamp the batch to a 2^n - 1 value. Having a power
1674 * of 2 value was found to be more likely to have
1675 * suboptimal cache aliasing properties in some cases.
1677 * For example if 2 tasks are alternately allocating
1678 * batches of pages, one task can end up with a lot
1679 * of pages of one half of the possible page colors
1680 * and the other with pages of the other colors.
1682 batch = (1 << fls(batch + batch/2)) - 1;
1688 * Dynamicaly allocate memory for the
1689 * per cpu pageset array in struct zone.
1691 static int __devinit process_zones(int cpu)
1693 struct zone *zone, *dzone;
1696 for_each_zone(zone) {
1697 struct per_cpu_pageset *npageset = NULL;
1699 npageset = kmalloc_node(sizeof(struct per_cpu_pageset),
1700 GFP_KERNEL, cpu_to_node(cpu));
1702 zone->pageset[cpu] = NULL;
1706 if (zone->pageset[cpu]) {
1707 memcpy(npageset, zone->pageset[cpu],
1708 sizeof(struct per_cpu_pageset));
1710 /* Relocate lists */
1711 for (i = 0; i < 2; i++) {
1712 INIT_LIST_HEAD(&npageset->pcp[i].list);
1713 list_splice(&zone->pageset[cpu]->pcp[i].list,
1714 &npageset->pcp[i].list);
1717 struct per_cpu_pages *pcp;
1718 unsigned long batch;
1720 batch = zone_batchsize(zone);
1722 pcp = &npageset->pcp[0]; /* hot */
1724 pcp->low = 2 * batch;
1725 pcp->high = 6 * batch;
1726 pcp->batch = 1 * batch;
1727 INIT_LIST_HEAD(&pcp->list);
1729 pcp = &npageset->pcp[1]; /* cold*/
1732 pcp->high = 2 * batch;
1733 pcp->batch = 1 * batch;
1734 INIT_LIST_HEAD(&pcp->list);
1736 zone->pageset[cpu] = npageset;
1741 for_each_zone(dzone) {
1744 kfree(dzone->pageset[cpu]);
1745 dzone->pageset[cpu] = NULL;
1750 static inline void free_zone_pagesets(int cpu)
1755 for_each_zone(zone) {
1756 struct per_cpu_pageset *pset = zone_pcp(zone, cpu);
1758 zone_pcp(zone, cpu) = NULL;
1764 static int __devinit pageset_cpuup_callback(struct notifier_block *nfb,
1765 unsigned long action,
1768 int cpu = (long)hcpu;
1769 int ret = NOTIFY_OK;
1772 case CPU_UP_PREPARE:
1773 if (process_zones(cpu))
1776 #ifdef CONFIG_HOTPLUG_CPU
1778 free_zone_pagesets(cpu);
1787 static struct notifier_block pageset_notifier =
1788 { &pageset_cpuup_callback, NULL, 0 };
1790 void __init setup_per_cpu_pageset()
1794 /* Initialize per_cpu_pageset for cpu 0.
1795 * A cpuup callback will do this for every cpu
1796 * as it comes online
1798 err = process_zones(smp_processor_id());
1800 register_cpu_notifier(&pageset_notifier);
1806 * Set up the zone data structures:
1807 * - mark all pages reserved
1808 * - mark all memory queues empty
1809 * - clear the memory bitmaps
1811 static void __init free_area_init_core(struct pglist_data *pgdat,
1812 unsigned long *zones_size, unsigned long *zholes_size)
1815 const unsigned long zone_required_alignment = 1UL << (MAX_ORDER-1);
1816 int cpu, nid = pgdat->node_id;
1817 unsigned long zone_start_pfn = pgdat->node_start_pfn;
1819 pgdat->nr_zones = 0;
1820 init_waitqueue_head(&pgdat->kswapd_wait);
1821 pgdat->kswapd_max_order = 0;
1823 for (j = 0; j < MAX_NR_ZONES; j++) {
1824 struct zone *zone = pgdat->node_zones + j;
1825 unsigned long size, realsize;
1826 unsigned long batch;
1828 zone_table[NODEZONE(nid, j)] = zone;
1829 realsize = size = zones_size[j];
1831 realsize -= zholes_size[j];
1833 if (j == ZONE_DMA || j == ZONE_NORMAL)
1834 nr_kernel_pages += realsize;
1835 nr_all_pages += realsize;
1837 zone->spanned_pages = size;
1838 zone->present_pages = realsize;
1839 zone->name = zone_names[j];
1840 spin_lock_init(&zone->lock);
1841 spin_lock_init(&zone->lru_lock);
1842 zone->zone_pgdat = pgdat;
1843 zone->free_pages = 0;
1845 zone->temp_priority = zone->prev_priority = DEF_PRIORITY;
1847 batch = zone_batchsize(zone);
1849 for (cpu = 0; cpu < NR_CPUS; cpu++) {
1850 struct per_cpu_pages *pcp;
1852 struct per_cpu_pageset *pgset;
1853 pgset = &pageset_table[nid*MAX_NR_ZONES*NR_CPUS +
1854 (j * NR_CPUS) + cpu];
1856 zone->pageset[cpu] = pgset;
1858 struct per_cpu_pageset *pgset = zone_pcp(zone, cpu);
1861 pcp = &pgset->pcp[0]; /* hot */
1863 pcp->low = 2 * batch;
1864 pcp->high = 6 * batch;
1865 pcp->batch = 1 * batch;
1866 INIT_LIST_HEAD(&pcp->list);
1868 pcp = &pgset->pcp[1]; /* cold */
1871 pcp->high = 2 * batch;
1872 pcp->batch = 1 * batch;
1873 INIT_LIST_HEAD(&pcp->list);
1875 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n",
1876 zone_names[j], realsize, batch);
1877 INIT_LIST_HEAD(&zone->active_list);
1878 INIT_LIST_HEAD(&zone->inactive_list);
1879 zone->nr_scan_active = 0;
1880 zone->nr_scan_inactive = 0;
1881 zone->nr_active = 0;
1882 zone->nr_inactive = 0;
1883 atomic_set(&zone->reclaim_in_progress, -1);
1888 * The per-page waitqueue mechanism uses hashed waitqueues
1891 zone->wait_table_size = wait_table_size(size);
1892 zone->wait_table_bits =
1893 wait_table_bits(zone->wait_table_size);
1894 zone->wait_table = (wait_queue_head_t *)
1895 alloc_bootmem_node(pgdat, zone->wait_table_size
1896 * sizeof(wait_queue_head_t));
1898 for(i = 0; i < zone->wait_table_size; ++i)
1899 init_waitqueue_head(zone->wait_table + i);
1901 pgdat->nr_zones = j+1;
1903 zone->zone_mem_map = pfn_to_page(zone_start_pfn);
1904 zone->zone_start_pfn = zone_start_pfn;
1906 if ((zone_start_pfn) & (zone_required_alignment-1))
1907 printk(KERN_CRIT "BUG: wrong zone alignment, it will crash\n");
1909 memmap_init(size, nid, j, zone_start_pfn);
1911 zone_start_pfn += size;
1913 zone_init_free_lists(pgdat, zone, zone->spanned_pages);
1917 static void __init alloc_node_mem_map(struct pglist_data *pgdat)
1921 /* Skip empty nodes */
1922 if (!pgdat->node_spanned_pages)
1925 /* ia64 gets its own node_mem_map, before this, without bootmem */
1926 if (!pgdat->node_mem_map) {
1927 size = (pgdat->node_spanned_pages + 1) * sizeof(struct page);
1928 pgdat->node_mem_map = alloc_bootmem_node(pgdat, size);
1930 #ifndef CONFIG_DISCONTIGMEM
1932 * With no DISCONTIG, the global mem_map is just set as node 0's
1934 if (pgdat == NODE_DATA(0))
1935 mem_map = NODE_DATA(0)->node_mem_map;
1939 void __init free_area_init_node(int nid, struct pglist_data *pgdat,
1940 unsigned long *zones_size, unsigned long node_start_pfn,
1941 unsigned long *zholes_size)
1943 pgdat->node_id = nid;
1944 pgdat->node_start_pfn = node_start_pfn;
1945 calculate_zone_totalpages(pgdat, zones_size, zholes_size);
1947 alloc_node_mem_map(pgdat);
1949 free_area_init_core(pgdat, zones_size, zholes_size);
1952 #ifndef CONFIG_DISCONTIGMEM
1953 static bootmem_data_t contig_bootmem_data;
1954 struct pglist_data contig_page_data = { .bdata = &contig_bootmem_data };
1956 EXPORT_SYMBOL(contig_page_data);
1958 void __init free_area_init(unsigned long *zones_size)
1960 free_area_init_node(0, &contig_page_data, zones_size,
1961 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
1965 #ifdef CONFIG_PROC_FS
1967 #include <linux/seq_file.h>
1969 static void *frag_start(struct seq_file *m, loff_t *pos)
1974 for (pgdat = pgdat_list; pgdat && node; pgdat = pgdat->pgdat_next)
1980 static void *frag_next(struct seq_file *m, void *arg, loff_t *pos)
1982 pg_data_t *pgdat = (pg_data_t *)arg;
1985 return pgdat->pgdat_next;
1988 static void frag_stop(struct seq_file *m, void *arg)
1993 * This walks the free areas for each zone.
1995 static int frag_show(struct seq_file *m, void *arg)
1997 pg_data_t *pgdat = (pg_data_t *)arg;
1999 struct zone *node_zones = pgdat->node_zones;
2000 unsigned long flags;
2003 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
2004 if (!zone->present_pages)
2007 spin_lock_irqsave(&zone->lock, flags);
2008 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
2009 for (order = 0; order < MAX_ORDER; ++order)
2010 seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
2011 spin_unlock_irqrestore(&zone->lock, flags);
2017 struct seq_operations fragmentation_op = {
2018 .start = frag_start,
2025 * Output information about zones in @pgdat.
2027 static int zoneinfo_show(struct seq_file *m, void *arg)
2029 pg_data_t *pgdat = arg;
2031 struct zone *node_zones = pgdat->node_zones;
2032 unsigned long flags;
2034 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; zone++) {
2037 if (!zone->present_pages)
2040 spin_lock_irqsave(&zone->lock, flags);
2041 seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name);
2049 "\n scanned %lu (a: %lu i: %lu)"
2058 zone->pages_scanned,
2059 zone->nr_scan_active, zone->nr_scan_inactive,
2060 zone->spanned_pages,
2061 zone->present_pages);
2063 "\n protection: (%lu",
2064 zone->lowmem_reserve[0]);
2065 for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
2066 seq_printf(m, ", %lu", zone->lowmem_reserve[i]);
2070 for (i = 0; i < ARRAY_SIZE(zone->pageset); i++) {
2071 struct per_cpu_pageset *pageset;
2074 pageset = zone_pcp(zone, i);
2075 for (j = 0; j < ARRAY_SIZE(pageset->pcp); j++) {
2076 if (pageset->pcp[j].count)
2079 if (j == ARRAY_SIZE(pageset->pcp))
2081 for (j = 0; j < ARRAY_SIZE(pageset->pcp); j++) {
2083 "\n cpu: %i pcp: %i"
2089 pageset->pcp[j].count,
2090 pageset->pcp[j].low,
2091 pageset->pcp[j].high,
2092 pageset->pcp[j].batch);
2098 "\n numa_foreign: %lu"
2099 "\n interleave_hit: %lu"
2100 "\n local_node: %lu"
2101 "\n other_node: %lu",
2104 pageset->numa_foreign,
2105 pageset->interleave_hit,
2106 pageset->local_node,
2107 pageset->other_node);
2111 "\n all_unreclaimable: %u"
2112 "\n prev_priority: %i"
2113 "\n temp_priority: %i"
2114 "\n start_pfn: %lu",
2115 zone->all_unreclaimable,
2116 zone->prev_priority,
2117 zone->temp_priority,
2118 zone->zone_start_pfn);
2119 spin_unlock_irqrestore(&zone->lock, flags);
2125 struct seq_operations zoneinfo_op = {
2126 .start = frag_start, /* iterate over all zones. The same as in
2130 .show = zoneinfo_show,
2133 static char *vmstat_text[] = {
2137 "nr_page_table_pages",
2162 "pgscan_kswapd_high",
2163 "pgscan_kswapd_normal",
2165 "pgscan_kswapd_dma",
2166 "pgscan_direct_high",
2167 "pgscan_direct_normal",
2168 "pgscan_direct_dma",
2173 "kswapd_inodesteal",
2181 static void *vmstat_start(struct seq_file *m, loff_t *pos)
2183 struct page_state *ps;
2185 if (*pos >= ARRAY_SIZE(vmstat_text))
2188 ps = kmalloc(sizeof(*ps), GFP_KERNEL);
2191 return ERR_PTR(-ENOMEM);
2192 get_full_page_state(ps);
2193 ps->pgpgin /= 2; /* sectors -> kbytes */
2195 return (unsigned long *)ps + *pos;
2198 static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
2201 if (*pos >= ARRAY_SIZE(vmstat_text))
2203 return (unsigned long *)m->private + *pos;
2206 static int vmstat_show(struct seq_file *m, void *arg)
2208 unsigned long *l = arg;
2209 unsigned long off = l - (unsigned long *)m->private;
2211 seq_printf(m, "%s %lu\n", vmstat_text[off], *l);
2215 static void vmstat_stop(struct seq_file *m, void *arg)
2221 struct seq_operations vmstat_op = {
2222 .start = vmstat_start,
2223 .next = vmstat_next,
2224 .stop = vmstat_stop,
2225 .show = vmstat_show,
2228 #endif /* CONFIG_PROC_FS */
2230 #ifdef CONFIG_HOTPLUG_CPU
2231 static int page_alloc_cpu_notify(struct notifier_block *self,
2232 unsigned long action, void *hcpu)
2234 int cpu = (unsigned long)hcpu;
2236 unsigned long *src, *dest;
2238 if (action == CPU_DEAD) {
2241 /* Drain local pagecache count. */
2242 count = &per_cpu(nr_pagecache_local, cpu);
2243 atomic_add(*count, &nr_pagecache);
2245 local_irq_disable();
2248 /* Add dead cpu's page_states to our own. */
2249 dest = (unsigned long *)&__get_cpu_var(page_states);
2250 src = (unsigned long *)&per_cpu(page_states, cpu);
2252 for (i = 0; i < sizeof(struct page_state)/sizeof(unsigned long);
2262 #endif /* CONFIG_HOTPLUG_CPU */
2264 void __init page_alloc_init(void)
2266 hotcpu_notifier(page_alloc_cpu_notify, 0);
2270 * setup_per_zone_lowmem_reserve - called whenever
2271 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
2272 * has a correct pages reserved value, so an adequate number of
2273 * pages are left in the zone after a successful __alloc_pages().
2275 static void setup_per_zone_lowmem_reserve(void)
2277 struct pglist_data *pgdat;
2280 for_each_pgdat(pgdat) {
2281 for (j = 0; j < MAX_NR_ZONES; j++) {
2282 struct zone *zone = pgdat->node_zones + j;
2283 unsigned long present_pages = zone->present_pages;
2285 zone->lowmem_reserve[j] = 0;
2287 for (idx = j-1; idx >= 0; idx--) {
2288 struct zone *lower_zone;
2290 if (sysctl_lowmem_reserve_ratio[idx] < 1)
2291 sysctl_lowmem_reserve_ratio[idx] = 1;
2293 lower_zone = pgdat->node_zones + idx;
2294 lower_zone->lowmem_reserve[j] = present_pages /
2295 sysctl_lowmem_reserve_ratio[idx];
2296 present_pages += lower_zone->present_pages;
2303 * setup_per_zone_pages_min - called when min_free_kbytes changes. Ensures
2304 * that the pages_{min,low,high} values for each zone are set correctly
2305 * with respect to min_free_kbytes.
2307 static void setup_per_zone_pages_min(void)
2309 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
2310 unsigned long lowmem_pages = 0;
2312 unsigned long flags;
2314 /* Calculate total number of !ZONE_HIGHMEM pages */
2315 for_each_zone(zone) {
2316 if (!is_highmem(zone))
2317 lowmem_pages += zone->present_pages;
2320 for_each_zone(zone) {
2321 spin_lock_irqsave(&zone->lru_lock, flags);
2322 if (is_highmem(zone)) {
2324 * Often, highmem doesn't need to reserve any pages.
2325 * But the pages_min/low/high values are also used for
2326 * batching up page reclaim activity so we need a
2327 * decent value here.
2331 min_pages = zone->present_pages / 1024;
2332 if (min_pages < SWAP_CLUSTER_MAX)
2333 min_pages = SWAP_CLUSTER_MAX;
2334 if (min_pages > 128)
2336 zone->pages_min = min_pages;
2338 /* if it's a lowmem zone, reserve a number of pages
2339 * proportionate to the zone's size.
2341 zone->pages_min = (pages_min * zone->present_pages) /
2346 * When interpreting these watermarks, just keep in mind that:
2347 * zone->pages_min == (zone->pages_min * 4) / 4;
2349 zone->pages_low = (zone->pages_min * 5) / 4;
2350 zone->pages_high = (zone->pages_min * 6) / 4;
2351 spin_unlock_irqrestore(&zone->lru_lock, flags);
2356 * Initialise min_free_kbytes.
2358 * For small machines we want it small (128k min). For large machines
2359 * we want it large (64MB max). But it is not linear, because network
2360 * bandwidth does not increase linearly with machine size. We use
2362 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
2363 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
2379 static int __init init_per_zone_pages_min(void)
2381 unsigned long lowmem_kbytes;
2383 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
2385 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
2386 if (min_free_kbytes < 128)
2387 min_free_kbytes = 128;
2388 if (min_free_kbytes > 65536)
2389 min_free_kbytes = 65536;
2390 setup_per_zone_pages_min();
2391 setup_per_zone_lowmem_reserve();
2394 module_init(init_per_zone_pages_min)
2397 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
2398 * that we can call two helper functions whenever min_free_kbytes
2401 int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
2402 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2404 proc_dointvec(table, write, file, buffer, length, ppos);
2405 setup_per_zone_pages_min();
2410 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
2411 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
2412 * whenever sysctl_lowmem_reserve_ratio changes.
2414 * The reserve ratio obviously has absolutely no relation with the
2415 * pages_min watermarks. The lowmem reserve ratio can only make sense
2416 * if in function of the boot time zone sizes.
2418 int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
2419 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2421 proc_dointvec_minmax(table, write, file, buffer, length, ppos);
2422 setup_per_zone_lowmem_reserve();
2426 __initdata int hashdist = HASHDIST_DEFAULT;
2429 static int __init set_hashdist(char *str)
2433 hashdist = simple_strtoul(str, &str, 0);
2436 __setup("hashdist=", set_hashdist);
2440 * allocate a large system hash table from bootmem
2441 * - it is assumed that the hash table must contain an exact power-of-2
2442 * quantity of entries
2443 * - limit is the number of hash buckets, not the total allocation size
2445 void *__init alloc_large_system_hash(const char *tablename,
2446 unsigned long bucketsize,
2447 unsigned long numentries,
2450 unsigned int *_hash_shift,
2451 unsigned int *_hash_mask,
2452 unsigned long limit)
2454 unsigned long long max = limit;
2455 unsigned long log2qty, size;
2458 /* allow the kernel cmdline to have a say */
2460 /* round applicable memory size up to nearest megabyte */
2461 numentries = (flags & HASH_HIGHMEM) ? nr_all_pages : nr_kernel_pages;
2462 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
2463 numentries >>= 20 - PAGE_SHIFT;
2464 numentries <<= 20 - PAGE_SHIFT;
2466 /* limit to 1 bucket per 2^scale bytes of low memory */
2467 if (scale > PAGE_SHIFT)
2468 numentries >>= (scale - PAGE_SHIFT);
2470 numentries <<= (PAGE_SHIFT - scale);
2472 /* rounded up to nearest power of 2 in size */
2473 numentries = 1UL << (long_log2(numentries) + 1);
2475 /* limit allocation size to 1/16 total memory by default */
2477 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
2478 do_div(max, bucketsize);
2481 if (numentries > max)
2484 log2qty = long_log2(numentries);
2487 size = bucketsize << log2qty;
2488 if (flags & HASH_EARLY)
2489 table = alloc_bootmem(size);
2491 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
2493 unsigned long order;
2494 for (order = 0; ((1UL << order) << PAGE_SHIFT) < size; order++)
2496 table = (void*) __get_free_pages(GFP_ATOMIC, order);
2498 } while (!table && size > PAGE_SIZE && --log2qty);
2501 panic("Failed to allocate %s hash table\n", tablename);
2503 printk("%s hash table entries: %d (order: %d, %lu bytes)\n",
2506 long_log2(size) - PAGE_SHIFT,
2510 *_hash_shift = log2qty;
2512 *_hash_mask = (1 << log2qty) - 1;