2 * linux/mm/page_alloc.c
4 * Manages the free list, the system allocates free pages here.
5 * Note that kmalloc() lives in slab.c
7 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
8 * Swap reorganised 29.12.95, Stephen Tweedie
9 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
10 * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999
11 * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
12 * Zone balancing, Kanoj Sarcar, SGI, Jan 2000
13 * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002
14 * (lots of bits borrowed from Ingo Molnar & Andrew Morton)
17 #include <linux/stddef.h>
19 #include <linux/swap.h>
20 #include <linux/interrupt.h>
21 #include <linux/pagemap.h>
22 #include <linux/jiffies.h>
23 #include <linux/bootmem.h>
24 #include <linux/memblock.h>
25 #include <linux/compiler.h>
26 #include <linux/kernel.h>
27 #include <linux/kmemcheck.h>
28 #include <linux/module.h>
29 #include <linux/suspend.h>
30 #include <linux/pagevec.h>
31 #include <linux/blkdev.h>
32 #include <linux/slab.h>
33 #include <linux/oom.h>
34 #include <linux/notifier.h>
35 #include <linux/topology.h>
36 #include <linux/sysctl.h>
37 #include <linux/cpu.h>
38 #include <linux/cpuset.h>
39 #include <linux/memory_hotplug.h>
40 #include <linux/nodemask.h>
41 #include <linux/vmalloc.h>
42 #include <linux/mempolicy.h>
43 #include <linux/stop_machine.h>
44 #include <linux/sort.h>
45 #include <linux/pfn.h>
46 #include <linux/backing-dev.h>
47 #include <linux/fault-inject.h>
48 #include <linux/page-isolation.h>
49 #include <linux/page_cgroup.h>
50 #include <linux/debugobjects.h>
51 #include <linux/kmemleak.h>
52 #include <linux/memory.h>
53 #include <linux/compaction.h>
54 #include <trace/events/kmem.h>
55 #include <linux/ftrace_event.h>
57 #include <asm/tlbflush.h>
58 #include <asm/div64.h>
61 #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
62 DEFINE_PER_CPU(int, numa_node);
63 EXPORT_PER_CPU_SYMBOL(numa_node);
66 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
68 * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
69 * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
70 * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
71 * defined in <linux/topology.h>.
73 DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */
74 EXPORT_PER_CPU_SYMBOL(_numa_mem_);
78 * Array of node states.
80 nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
81 [N_POSSIBLE] = NODE_MASK_ALL,
82 [N_ONLINE] = { { [0] = 1UL } },
84 [N_NORMAL_MEMORY] = { { [0] = 1UL } },
86 [N_HIGH_MEMORY] = { { [0] = 1UL } },
88 [N_CPU] = { { [0] = 1UL } },
91 EXPORT_SYMBOL(node_states);
93 unsigned long totalram_pages __read_mostly;
94 unsigned long totalreserve_pages __read_mostly;
95 int percpu_pagelist_fraction;
96 gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK;
98 #ifdef CONFIG_PM_SLEEP
100 * The following functions are used by the suspend/hibernate code to temporarily
101 * change gfp_allowed_mask in order to avoid using I/O during memory allocations
102 * while devices are suspended. To avoid races with the suspend/hibernate code,
103 * they should always be called with pm_mutex held (gfp_allowed_mask also should
104 * only be modified with pm_mutex held, unless the suspend/hibernate code is
105 * guaranteed not to run in parallel with that modification).
108 static gfp_t saved_gfp_mask;
110 void pm_restore_gfp_mask(void)
112 WARN_ON(!mutex_is_locked(&pm_mutex));
113 if (saved_gfp_mask) {
114 gfp_allowed_mask = saved_gfp_mask;
119 void pm_restrict_gfp_mask(void)
121 WARN_ON(!mutex_is_locked(&pm_mutex));
122 WARN_ON(saved_gfp_mask);
123 saved_gfp_mask = gfp_allowed_mask;
124 gfp_allowed_mask &= ~GFP_IOFS;
126 #endif /* CONFIG_PM_SLEEP */
128 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
129 int pageblock_order __read_mostly;
132 static void __free_pages_ok(struct page *page, unsigned int order);
135 * results with 256, 32 in the lowmem_reserve sysctl:
136 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
137 * 1G machine -> (16M dma, 784M normal, 224M high)
138 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
139 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
140 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
142 * TBD: should special case ZONE_DMA32 machines here - in those we normally
143 * don't need any ZONE_NORMAL reservation
145 int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
146 #ifdef CONFIG_ZONE_DMA
149 #ifdef CONFIG_ZONE_DMA32
152 #ifdef CONFIG_HIGHMEM
158 EXPORT_SYMBOL(totalram_pages);
160 static char * const zone_names[MAX_NR_ZONES] = {
161 #ifdef CONFIG_ZONE_DMA
164 #ifdef CONFIG_ZONE_DMA32
168 #ifdef CONFIG_HIGHMEM
174 int min_free_kbytes = 1024;
176 static unsigned long __meminitdata nr_kernel_pages;
177 static unsigned long __meminitdata nr_all_pages;
178 static unsigned long __meminitdata dma_reserve;
180 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
182 * MAX_ACTIVE_REGIONS determines the maximum number of distinct
183 * ranges of memory (RAM) that may be registered with add_active_range().
184 * Ranges passed to add_active_range() will be merged if possible
185 * so the number of times add_active_range() can be called is
186 * related to the number of nodes and the number of holes
188 #ifdef CONFIG_MAX_ACTIVE_REGIONS
189 /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
190 #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
192 #if MAX_NUMNODES >= 32
193 /* If there can be many nodes, allow up to 50 holes per node */
194 #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
196 /* By default, allow up to 256 distinct regions */
197 #define MAX_ACTIVE_REGIONS 256
201 static struct node_active_region __meminitdata early_node_map[MAX_ACTIVE_REGIONS];
202 static int __meminitdata nr_nodemap_entries;
203 static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
204 static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
205 static unsigned long __initdata required_kernelcore;
206 static unsigned long __initdata required_movablecore;
207 static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
209 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
211 EXPORT_SYMBOL(movable_zone);
212 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
215 int nr_node_ids __read_mostly = MAX_NUMNODES;
216 int nr_online_nodes __read_mostly = 1;
217 EXPORT_SYMBOL(nr_node_ids);
218 EXPORT_SYMBOL(nr_online_nodes);
221 int page_group_by_mobility_disabled __read_mostly;
223 static void set_pageblock_migratetype(struct page *page, int migratetype)
226 if (unlikely(page_group_by_mobility_disabled))
227 migratetype = MIGRATE_UNMOVABLE;
229 set_pageblock_flags_group(page, (unsigned long)migratetype,
230 PB_migrate, PB_migrate_end);
233 bool oom_killer_disabled __read_mostly;
235 #ifdef CONFIG_DEBUG_VM
236 static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
240 unsigned long pfn = page_to_pfn(page);
243 seq = zone_span_seqbegin(zone);
244 if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
246 else if (pfn < zone->zone_start_pfn)
248 } while (zone_span_seqretry(zone, seq));
253 static int page_is_consistent(struct zone *zone, struct page *page)
255 if (!pfn_valid_within(page_to_pfn(page)))
257 if (zone != page_zone(page))
263 * Temporary debugging check for pages not lying within a given zone.
265 static int bad_range(struct zone *zone, struct page *page)
267 if (page_outside_zone_boundaries(zone, page))
269 if (!page_is_consistent(zone, page))
275 static inline int bad_range(struct zone *zone, struct page *page)
281 static void bad_page(struct page *page)
283 static unsigned long resume;
284 static unsigned long nr_shown;
285 static unsigned long nr_unshown;
287 /* Don't complain about poisoned pages */
288 if (PageHWPoison(page)) {
289 __ClearPageBuddy(page);
294 * Allow a burst of 60 reports, then keep quiet for that minute;
295 * or allow a steady drip of one report per second.
297 if (nr_shown == 60) {
298 if (time_before(jiffies, resume)) {
304 "BUG: Bad page state: %lu messages suppressed\n",
311 resume = jiffies + 60 * HZ;
313 printk(KERN_ALERT "BUG: Bad page state in process %s pfn:%05lx\n",
314 current->comm, page_to_pfn(page));
319 /* Leave bad fields for debug, except PageBuddy could make trouble */
320 __ClearPageBuddy(page);
321 add_taint(TAINT_BAD_PAGE);
325 * Higher-order pages are called "compound pages". They are structured thusly:
327 * The first PAGE_SIZE page is called the "head page".
329 * The remaining PAGE_SIZE pages are called "tail pages".
331 * All pages have PG_compound set. All pages have their ->private pointing at
332 * the head page (even the head page has this).
334 * The first tail page's ->lru.next holds the address of the compound page's
335 * put_page() function. Its ->lru.prev holds the order of allocation.
336 * This usage means that zero-order pages may not be compound.
339 static void free_compound_page(struct page *page)
341 __free_pages_ok(page, compound_order(page));
344 void prep_compound_page(struct page *page, unsigned long order)
347 int nr_pages = 1 << order;
349 set_compound_page_dtor(page, free_compound_page);
350 set_compound_order(page, order);
352 for (i = 1; i < nr_pages; i++) {
353 struct page *p = page + i;
356 p->first_page = page;
360 static int destroy_compound_page(struct page *page, unsigned long order)
363 int nr_pages = 1 << order;
366 if (unlikely(compound_order(page) != order) ||
367 unlikely(!PageHead(page))) {
372 __ClearPageHead(page);
374 for (i = 1; i < nr_pages; i++) {
375 struct page *p = page + i;
377 if (unlikely(!PageTail(p) || (p->first_page != page))) {
387 static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
392 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
393 * and __GFP_HIGHMEM from hard or soft interrupt context.
395 VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
396 for (i = 0; i < (1 << order); i++)
397 clear_highpage(page + i);
400 static inline void set_page_order(struct page *page, int order)
402 set_page_private(page, order);
403 __SetPageBuddy(page);
406 static inline void rmv_page_order(struct page *page)
408 __ClearPageBuddy(page);
409 set_page_private(page, 0);
413 * Locate the struct page for both the matching buddy in our
414 * pair (buddy1) and the combined O(n+1) page they form (page).
416 * 1) Any buddy B1 will have an order O twin B2 which satisfies
417 * the following equation:
419 * For example, if the starting buddy (buddy2) is #8 its order
421 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
423 * 2) Any buddy B will have an order O+1 parent P which
424 * satisfies the following equation:
427 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
429 static inline struct page *
430 __page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order)
432 unsigned long buddy_idx = page_idx ^ (1 << order);
434 return page + (buddy_idx - page_idx);
437 static inline unsigned long
438 __find_combined_index(unsigned long page_idx, unsigned int order)
440 return (page_idx & ~(1 << order));
444 * This function checks whether a page is free && is the buddy
445 * we can do coalesce a page and its buddy if
446 * (a) the buddy is not in a hole &&
447 * (b) the buddy is in the buddy system &&
448 * (c) a page and its buddy have the same order &&
449 * (d) a page and its buddy are in the same zone.
451 * For recording whether a page is in the buddy system, we use PG_buddy.
452 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
454 * For recording page's order, we use page_private(page).
456 static inline int page_is_buddy(struct page *page, struct page *buddy,
459 if (!pfn_valid_within(page_to_pfn(buddy)))
462 if (page_zone_id(page) != page_zone_id(buddy))
465 if (PageBuddy(buddy) && page_order(buddy) == order) {
466 VM_BUG_ON(page_count(buddy) != 0);
473 * Freeing function for a buddy system allocator.
475 * The concept of a buddy system is to maintain direct-mapped table
476 * (containing bit values) for memory blocks of various "orders".
477 * The bottom level table contains the map for the smallest allocatable
478 * units of memory (here, pages), and each level above it describes
479 * pairs of units from the levels below, hence, "buddies".
480 * At a high level, all that happens here is marking the table entry
481 * at the bottom level available, and propagating the changes upward
482 * as necessary, plus some accounting needed to play nicely with other
483 * parts of the VM system.
484 * At each level, we keep a list of pages, which are heads of continuous
485 * free pages of length of (1 << order) and marked with PG_buddy. Page's
486 * order is recorded in page_private(page) field.
487 * So when we are allocating or freeing one, we can derive the state of the
488 * other. That is, if we allocate a small block, and both were
489 * free, the remainder of the region must be split into blocks.
490 * If a block is freed, and its buddy is also free, then this
491 * triggers coalescing into a block of larger size.
496 static inline void __free_one_page(struct page *page,
497 struct zone *zone, unsigned int order,
500 unsigned long page_idx;
501 unsigned long combined_idx;
504 if (unlikely(PageCompound(page)))
505 if (unlikely(destroy_compound_page(page, order)))
508 VM_BUG_ON(migratetype == -1);
510 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
512 VM_BUG_ON(page_idx & ((1 << order) - 1));
513 VM_BUG_ON(bad_range(zone, page));
515 while (order < MAX_ORDER-1) {
516 buddy = __page_find_buddy(page, page_idx, order);
517 if (!page_is_buddy(page, buddy, order))
520 /* Our buddy is free, merge with it and move up one order. */
521 list_del(&buddy->lru);
522 zone->free_area[order].nr_free--;
523 rmv_page_order(buddy);
524 combined_idx = __find_combined_index(page_idx, order);
525 page = page + (combined_idx - page_idx);
526 page_idx = combined_idx;
529 set_page_order(page, order);
532 * If this is not the largest possible page, check if the buddy
533 * of the next-highest order is free. If it is, it's possible
534 * that pages are being freed that will coalesce soon. In case,
535 * that is happening, add the free page to the tail of the list
536 * so it's less likely to be used soon and more likely to be merged
537 * as a higher order page
539 if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) {
540 struct page *higher_page, *higher_buddy;
541 combined_idx = __find_combined_index(page_idx, order);
542 higher_page = page + combined_idx - page_idx;
543 higher_buddy = __page_find_buddy(higher_page, combined_idx, order + 1);
544 if (page_is_buddy(higher_page, higher_buddy, order + 1)) {
545 list_add_tail(&page->lru,
546 &zone->free_area[order].free_list[migratetype]);
551 list_add(&page->lru, &zone->free_area[order].free_list[migratetype]);
553 zone->free_area[order].nr_free++;
557 * free_page_mlock() -- clean up attempts to free and mlocked() page.
558 * Page should not be on lru, so no need to fix that up.
559 * free_pages_check() will verify...
561 static inline void free_page_mlock(struct page *page)
563 __dec_zone_page_state(page, NR_MLOCK);
564 __count_vm_event(UNEVICTABLE_MLOCKFREED);
567 static inline int free_pages_check(struct page *page)
569 if (unlikely(page_mapcount(page) |
570 (page->mapping != NULL) |
571 (atomic_read(&page->_count) != 0) |
572 (page->flags & PAGE_FLAGS_CHECK_AT_FREE))) {
576 if (page->flags & PAGE_FLAGS_CHECK_AT_PREP)
577 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
582 * Frees a number of pages from the PCP lists
583 * Assumes all pages on list are in same zone, and of same order.
584 * count is the number of pages to free.
586 * If the zone was previously in an "all pages pinned" state then look to
587 * see if this freeing clears that state.
589 * And clear the zone's pages_scanned counter, to hold off the "all pages are
590 * pinned" detection logic.
592 static void free_pcppages_bulk(struct zone *zone, int count,
593 struct per_cpu_pages *pcp)
599 spin_lock(&zone->lock);
600 zone->all_unreclaimable = 0;
601 zone->pages_scanned = 0;
605 struct list_head *list;
608 * Remove pages from lists in a round-robin fashion. A
609 * batch_free count is maintained that is incremented when an
610 * empty list is encountered. This is so more pages are freed
611 * off fuller lists instead of spinning excessively around empty
616 if (++migratetype == MIGRATE_PCPTYPES)
618 list = &pcp->lists[migratetype];
619 } while (list_empty(list));
622 page = list_entry(list->prev, struct page, lru);
623 /* must delete as __free_one_page list manipulates */
624 list_del(&page->lru);
625 /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */
626 __free_one_page(page, zone, 0, page_private(page));
627 trace_mm_page_pcpu_drain(page, 0, page_private(page));
628 } while (--to_free && --batch_free && !list_empty(list));
630 __mod_zone_page_state(zone, NR_FREE_PAGES, count);
631 spin_unlock(&zone->lock);
634 static void free_one_page(struct zone *zone, struct page *page, int order,
637 spin_lock(&zone->lock);
638 zone->all_unreclaimable = 0;
639 zone->pages_scanned = 0;
641 __free_one_page(page, zone, order, migratetype);
642 __mod_zone_page_state(zone, NR_FREE_PAGES, 1 << order);
643 spin_unlock(&zone->lock);
646 static bool free_pages_prepare(struct page *page, unsigned int order)
651 trace_mm_page_free_direct(page, order);
652 kmemcheck_free_shadow(page, order);
654 for (i = 0; i < (1 << order); i++) {
655 struct page *pg = page + i;
659 bad += free_pages_check(pg);
664 if (!PageHighMem(page)) {
665 debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order);
666 debug_check_no_obj_freed(page_address(page),
669 arch_free_page(page, order);
670 kernel_map_pages(page, 1 << order, 0);
675 static void __free_pages_ok(struct page *page, unsigned int order)
678 int wasMlocked = __TestClearPageMlocked(page);
680 if (!free_pages_prepare(page, order))
683 local_irq_save(flags);
684 if (unlikely(wasMlocked))
685 free_page_mlock(page);
686 __count_vm_events(PGFREE, 1 << order);
687 free_one_page(page_zone(page), page, order,
688 get_pageblock_migratetype(page));
689 local_irq_restore(flags);
693 * permit the bootmem allocator to evade page validation on high-order frees
695 void __meminit __free_pages_bootmem(struct page *page, unsigned int order)
698 __ClearPageReserved(page);
699 set_page_count(page, 0);
700 set_page_refcounted(page);
706 for (loop = 0; loop < BITS_PER_LONG; loop++) {
707 struct page *p = &page[loop];
709 if (loop + 1 < BITS_PER_LONG)
711 __ClearPageReserved(p);
712 set_page_count(p, 0);
715 set_page_refcounted(page);
716 __free_pages(page, order);
722 * The order of subdivision here is critical for the IO subsystem.
723 * Please do not alter this order without good reasons and regression
724 * testing. Specifically, as large blocks of memory are subdivided,
725 * the order in which smaller blocks are delivered depends on the order
726 * they're subdivided in this function. This is the primary factor
727 * influencing the order in which pages are delivered to the IO
728 * subsystem according to empirical testing, and this is also justified
729 * by considering the behavior of a buddy system containing a single
730 * large block of memory acted on by a series of small allocations.
731 * This behavior is a critical factor in sglist merging's success.
735 static inline void expand(struct zone *zone, struct page *page,
736 int low, int high, struct free_area *area,
739 unsigned long size = 1 << high;
745 VM_BUG_ON(bad_range(zone, &page[size]));
746 list_add(&page[size].lru, &area->free_list[migratetype]);
748 set_page_order(&page[size], high);
753 * This page is about to be returned from the page allocator
755 static inline int check_new_page(struct page *page)
757 if (unlikely(page_mapcount(page) |
758 (page->mapping != NULL) |
759 (atomic_read(&page->_count) != 0) |
760 (page->flags & PAGE_FLAGS_CHECK_AT_PREP))) {
767 static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
771 for (i = 0; i < (1 << order); i++) {
772 struct page *p = page + i;
773 if (unlikely(check_new_page(p)))
777 set_page_private(page, 0);
778 set_page_refcounted(page);
780 arch_alloc_page(page, order);
781 kernel_map_pages(page, 1 << order, 1);
783 if (gfp_flags & __GFP_ZERO)
784 prep_zero_page(page, order, gfp_flags);
786 if (order && (gfp_flags & __GFP_COMP))
787 prep_compound_page(page, order);
793 * Go through the free lists for the given migratetype and remove
794 * the smallest available page from the freelists
797 struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
800 unsigned int current_order;
801 struct free_area * area;
804 /* Find a page of the appropriate size in the preferred list */
805 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
806 area = &(zone->free_area[current_order]);
807 if (list_empty(&area->free_list[migratetype]))
810 page = list_entry(area->free_list[migratetype].next,
812 list_del(&page->lru);
813 rmv_page_order(page);
815 expand(zone, page, order, current_order, area, migratetype);
824 * This array describes the order lists are fallen back to when
825 * the free lists for the desirable migrate type are depleted
827 static int fallbacks[MIGRATE_TYPES][MIGRATE_TYPES-1] = {
828 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
829 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
830 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
831 [MIGRATE_RESERVE] = { MIGRATE_RESERVE, MIGRATE_RESERVE, MIGRATE_RESERVE }, /* Never used */
835 * Move the free pages in a range to the free lists of the requested type.
836 * Note that start_page and end_pages are not aligned on a pageblock
837 * boundary. If alignment is required, use move_freepages_block()
839 static int move_freepages(struct zone *zone,
840 struct page *start_page, struct page *end_page,
847 #ifndef CONFIG_HOLES_IN_ZONE
849 * page_zone is not safe to call in this context when
850 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
851 * anyway as we check zone boundaries in move_freepages_block().
852 * Remove at a later date when no bug reports exist related to
853 * grouping pages by mobility
855 BUG_ON(page_zone(start_page) != page_zone(end_page));
858 for (page = start_page; page <= end_page;) {
859 /* Make sure we are not inadvertently changing nodes */
860 VM_BUG_ON(page_to_nid(page) != zone_to_nid(zone));
862 if (!pfn_valid_within(page_to_pfn(page))) {
867 if (!PageBuddy(page)) {
872 order = page_order(page);
873 list_del(&page->lru);
875 &zone->free_area[order].free_list[migratetype]);
877 pages_moved += 1 << order;
883 static int move_freepages_block(struct zone *zone, struct page *page,
886 unsigned long start_pfn, end_pfn;
887 struct page *start_page, *end_page;
889 start_pfn = page_to_pfn(page);
890 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
891 start_page = pfn_to_page(start_pfn);
892 end_page = start_page + pageblock_nr_pages - 1;
893 end_pfn = start_pfn + pageblock_nr_pages - 1;
895 /* Do not cross zone boundaries */
896 if (start_pfn < zone->zone_start_pfn)
898 if (end_pfn >= zone->zone_start_pfn + zone->spanned_pages)
901 return move_freepages(zone, start_page, end_page, migratetype);
904 static void change_pageblock_range(struct page *pageblock_page,
905 int start_order, int migratetype)
907 int nr_pageblocks = 1 << (start_order - pageblock_order);
909 while (nr_pageblocks--) {
910 set_pageblock_migratetype(pageblock_page, migratetype);
911 pageblock_page += pageblock_nr_pages;
915 /* Remove an element from the buddy allocator from the fallback list */
916 static inline struct page *
917 __rmqueue_fallback(struct zone *zone, int order, int start_migratetype)
919 struct free_area * area;
924 /* Find the largest possible block of pages in the other list */
925 for (current_order = MAX_ORDER-1; current_order >= order;
927 for (i = 0; i < MIGRATE_TYPES - 1; i++) {
928 migratetype = fallbacks[start_migratetype][i];
930 /* MIGRATE_RESERVE handled later if necessary */
931 if (migratetype == MIGRATE_RESERVE)
934 area = &(zone->free_area[current_order]);
935 if (list_empty(&area->free_list[migratetype]))
938 page = list_entry(area->free_list[migratetype].next,
943 * If breaking a large block of pages, move all free
944 * pages to the preferred allocation list. If falling
945 * back for a reclaimable kernel allocation, be more
946 * agressive about taking ownership of free pages
948 if (unlikely(current_order >= (pageblock_order >> 1)) ||
949 start_migratetype == MIGRATE_RECLAIMABLE ||
950 page_group_by_mobility_disabled) {
952 pages = move_freepages_block(zone, page,
955 /* Claim the whole block if over half of it is free */
956 if (pages >= (1 << (pageblock_order-1)) ||
957 page_group_by_mobility_disabled)
958 set_pageblock_migratetype(page,
961 migratetype = start_migratetype;
964 /* Remove the page from the freelists */
965 list_del(&page->lru);
966 rmv_page_order(page);
968 /* Take ownership for orders >= pageblock_order */
969 if (current_order >= pageblock_order)
970 change_pageblock_range(page, current_order,
973 expand(zone, page, order, current_order, area, migratetype);
975 trace_mm_page_alloc_extfrag(page, order, current_order,
976 start_migratetype, migratetype);
986 * Do the hard work of removing an element from the buddy allocator.
987 * Call me with the zone->lock already held.
989 static struct page *__rmqueue(struct zone *zone, unsigned int order,
995 page = __rmqueue_smallest(zone, order, migratetype);
997 if (unlikely(!page) && migratetype != MIGRATE_RESERVE) {
998 page = __rmqueue_fallback(zone, order, migratetype);
1001 * Use MIGRATE_RESERVE rather than fail an allocation. goto
1002 * is used because __rmqueue_smallest is an inline function
1003 * and we want just one call site
1006 migratetype = MIGRATE_RESERVE;
1011 trace_mm_page_alloc_zone_locked(page, order, migratetype);
1016 * Obtain a specified number of elements from the buddy allocator, all under
1017 * a single hold of the lock, for efficiency. Add them to the supplied list.
1018 * Returns the number of new pages which were placed at *list.
1020 static int rmqueue_bulk(struct zone *zone, unsigned int order,
1021 unsigned long count, struct list_head *list,
1022 int migratetype, int cold)
1026 spin_lock(&zone->lock);
1027 for (i = 0; i < count; ++i) {
1028 struct page *page = __rmqueue(zone, order, migratetype);
1029 if (unlikely(page == NULL))
1033 * Split buddy pages returned by expand() are received here
1034 * in physical page order. The page is added to the callers and
1035 * list and the list head then moves forward. From the callers
1036 * perspective, the linked list is ordered by page number in
1037 * some conditions. This is useful for IO devices that can
1038 * merge IO requests if the physical pages are ordered
1041 if (likely(cold == 0))
1042 list_add(&page->lru, list);
1044 list_add_tail(&page->lru, list);
1045 set_page_private(page, migratetype);
1048 __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
1049 spin_unlock(&zone->lock);
1055 * Called from the vmstat counter updater to drain pagesets of this
1056 * currently executing processor on remote nodes after they have
1059 * Note that this function must be called with the thread pinned to
1060 * a single processor.
1062 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
1064 unsigned long flags;
1067 local_irq_save(flags);
1068 if (pcp->count >= pcp->batch)
1069 to_drain = pcp->batch;
1071 to_drain = pcp->count;
1072 free_pcppages_bulk(zone, to_drain, pcp);
1073 pcp->count -= to_drain;
1074 local_irq_restore(flags);
1079 * Drain pages of the indicated processor.
1081 * The processor must either be the current processor and the
1082 * thread pinned to the current processor or a processor that
1085 static void drain_pages(unsigned int cpu)
1087 unsigned long flags;
1090 for_each_populated_zone(zone) {
1091 struct per_cpu_pageset *pset;
1092 struct per_cpu_pages *pcp;
1094 local_irq_save(flags);
1095 pset = per_cpu_ptr(zone->pageset, cpu);
1098 free_pcppages_bulk(zone, pcp->count, pcp);
1100 local_irq_restore(flags);
1105 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
1107 void drain_local_pages(void *arg)
1109 drain_pages(smp_processor_id());
1113 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
1115 void drain_all_pages(void)
1117 on_each_cpu(drain_local_pages, NULL, 1);
1120 #ifdef CONFIG_HIBERNATION
1122 void mark_free_pages(struct zone *zone)
1124 unsigned long pfn, max_zone_pfn;
1125 unsigned long flags;
1127 struct list_head *curr;
1129 if (!zone->spanned_pages)
1132 spin_lock_irqsave(&zone->lock, flags);
1134 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
1135 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1136 if (pfn_valid(pfn)) {
1137 struct page *page = pfn_to_page(pfn);
1139 if (!swsusp_page_is_forbidden(page))
1140 swsusp_unset_page_free(page);
1143 for_each_migratetype_order(order, t) {
1144 list_for_each(curr, &zone->free_area[order].free_list[t]) {
1147 pfn = page_to_pfn(list_entry(curr, struct page, lru));
1148 for (i = 0; i < (1UL << order); i++)
1149 swsusp_set_page_free(pfn_to_page(pfn + i));
1152 spin_unlock_irqrestore(&zone->lock, flags);
1154 #endif /* CONFIG_PM */
1157 * Free a 0-order page
1158 * cold == 1 ? free a cold page : free a hot page
1160 void free_hot_cold_page(struct page *page, int cold)
1162 struct zone *zone = page_zone(page);
1163 struct per_cpu_pages *pcp;
1164 unsigned long flags;
1166 int wasMlocked = __TestClearPageMlocked(page);
1168 if (!free_pages_prepare(page, 0))
1171 migratetype = get_pageblock_migratetype(page);
1172 set_page_private(page, migratetype);
1173 local_irq_save(flags);
1174 if (unlikely(wasMlocked))
1175 free_page_mlock(page);
1176 __count_vm_event(PGFREE);
1179 * We only track unmovable, reclaimable and movable on pcp lists.
1180 * Free ISOLATE pages back to the allocator because they are being
1181 * offlined but treat RESERVE as movable pages so we can get those
1182 * areas back if necessary. Otherwise, we may have to free
1183 * excessively into the page allocator
1185 if (migratetype >= MIGRATE_PCPTYPES) {
1186 if (unlikely(migratetype == MIGRATE_ISOLATE)) {
1187 free_one_page(zone, page, 0, migratetype);
1190 migratetype = MIGRATE_MOVABLE;
1193 pcp = &this_cpu_ptr(zone->pageset)->pcp;
1195 list_add_tail(&page->lru, &pcp->lists[migratetype]);
1197 list_add(&page->lru, &pcp->lists[migratetype]);
1199 if (pcp->count >= pcp->high) {
1200 free_pcppages_bulk(zone, pcp->batch, pcp);
1201 pcp->count -= pcp->batch;
1205 local_irq_restore(flags);
1209 * split_page takes a non-compound higher-order page, and splits it into
1210 * n (1<<order) sub-pages: page[0..n]
1211 * Each sub-page must be freed individually.
1213 * Note: this is probably too low level an operation for use in drivers.
1214 * Please consult with lkml before using this in your driver.
1216 void split_page(struct page *page, unsigned int order)
1220 VM_BUG_ON(PageCompound(page));
1221 VM_BUG_ON(!page_count(page));
1223 #ifdef CONFIG_KMEMCHECK
1225 * Split shadow pages too, because free(page[0]) would
1226 * otherwise free the whole shadow.
1228 if (kmemcheck_page_is_tracked(page))
1229 split_page(virt_to_page(page[0].shadow), order);
1232 for (i = 1; i < (1 << order); i++)
1233 set_page_refcounted(page + i);
1237 * Similar to split_page except the page is already free. As this is only
1238 * being used for migration, the migratetype of the block also changes.
1239 * As this is called with interrupts disabled, the caller is responsible
1240 * for calling arch_alloc_page() and kernel_map_page() after interrupts
1243 * Note: this is probably too low level an operation for use in drivers.
1244 * Please consult with lkml before using this in your driver.
1246 int split_free_page(struct page *page)
1249 unsigned long watermark;
1252 BUG_ON(!PageBuddy(page));
1254 zone = page_zone(page);
1255 order = page_order(page);
1257 /* Obey watermarks as if the page was being allocated */
1258 watermark = low_wmark_pages(zone) + (1 << order);
1259 if (!zone_watermark_ok(zone, 0, watermark, 0, 0))
1262 /* Remove page from free list */
1263 list_del(&page->lru);
1264 zone->free_area[order].nr_free--;
1265 rmv_page_order(page);
1266 __mod_zone_page_state(zone, NR_FREE_PAGES, -(1UL << order));
1268 /* Split into individual pages */
1269 set_page_refcounted(page);
1270 split_page(page, order);
1272 if (order >= pageblock_order - 1) {
1273 struct page *endpage = page + (1 << order) - 1;
1274 for (; page < endpage; page += pageblock_nr_pages)
1275 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
1282 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1283 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1287 struct page *buffered_rmqueue(struct zone *preferred_zone,
1288 struct zone *zone, int order, gfp_t gfp_flags,
1291 unsigned long flags;
1293 int cold = !!(gfp_flags & __GFP_COLD);
1296 if (likely(order == 0)) {
1297 struct per_cpu_pages *pcp;
1298 struct list_head *list;
1300 local_irq_save(flags);
1301 pcp = &this_cpu_ptr(zone->pageset)->pcp;
1302 list = &pcp->lists[migratetype];
1303 if (list_empty(list)) {
1304 pcp->count += rmqueue_bulk(zone, 0,
1307 if (unlikely(list_empty(list)))
1312 page = list_entry(list->prev, struct page, lru);
1314 page = list_entry(list->next, struct page, lru);
1316 list_del(&page->lru);
1319 if (unlikely(gfp_flags & __GFP_NOFAIL)) {
1321 * __GFP_NOFAIL is not to be used in new code.
1323 * All __GFP_NOFAIL callers should be fixed so that they
1324 * properly detect and handle allocation failures.
1326 * We most definitely don't want callers attempting to
1327 * allocate greater than order-1 page units with
1330 WARN_ON_ONCE(order > 1);
1332 spin_lock_irqsave(&zone->lock, flags);
1333 page = __rmqueue(zone, order, migratetype);
1334 spin_unlock(&zone->lock);
1337 __mod_zone_page_state(zone, NR_FREE_PAGES, -(1 << order));
1340 __count_zone_vm_events(PGALLOC, zone, 1 << order);
1341 zone_statistics(preferred_zone, zone);
1342 local_irq_restore(flags);
1344 VM_BUG_ON(bad_range(zone, page));
1345 if (prep_new_page(page, order, gfp_flags))
1350 local_irq_restore(flags);
1354 /* The ALLOC_WMARK bits are used as an index to zone->watermark */
1355 #define ALLOC_WMARK_MIN WMARK_MIN
1356 #define ALLOC_WMARK_LOW WMARK_LOW
1357 #define ALLOC_WMARK_HIGH WMARK_HIGH
1358 #define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */
1360 /* Mask to get the watermark bits */
1361 #define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1)
1363 #define ALLOC_HARDER 0x10 /* try to alloc harder */
1364 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1365 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
1367 #ifdef CONFIG_FAIL_PAGE_ALLOC
1369 static struct fail_page_alloc_attr {
1370 struct fault_attr attr;
1372 u32 ignore_gfp_highmem;
1373 u32 ignore_gfp_wait;
1376 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1378 struct dentry *ignore_gfp_highmem_file;
1379 struct dentry *ignore_gfp_wait_file;
1380 struct dentry *min_order_file;
1382 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1384 } fail_page_alloc = {
1385 .attr = FAULT_ATTR_INITIALIZER,
1386 .ignore_gfp_wait = 1,
1387 .ignore_gfp_highmem = 1,
1391 static int __init setup_fail_page_alloc(char *str)
1393 return setup_fault_attr(&fail_page_alloc.attr, str);
1395 __setup("fail_page_alloc=", setup_fail_page_alloc);
1397 static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1399 if (order < fail_page_alloc.min_order)
1401 if (gfp_mask & __GFP_NOFAIL)
1403 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
1405 if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
1408 return should_fail(&fail_page_alloc.attr, 1 << order);
1411 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1413 static int __init fail_page_alloc_debugfs(void)
1415 mode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
1419 err = init_fault_attr_dentries(&fail_page_alloc.attr,
1423 dir = fail_page_alloc.attr.dentries.dir;
1425 fail_page_alloc.ignore_gfp_wait_file =
1426 debugfs_create_bool("ignore-gfp-wait", mode, dir,
1427 &fail_page_alloc.ignore_gfp_wait);
1429 fail_page_alloc.ignore_gfp_highmem_file =
1430 debugfs_create_bool("ignore-gfp-highmem", mode, dir,
1431 &fail_page_alloc.ignore_gfp_highmem);
1432 fail_page_alloc.min_order_file =
1433 debugfs_create_u32("min-order", mode, dir,
1434 &fail_page_alloc.min_order);
1436 if (!fail_page_alloc.ignore_gfp_wait_file ||
1437 !fail_page_alloc.ignore_gfp_highmem_file ||
1438 !fail_page_alloc.min_order_file) {
1440 debugfs_remove(fail_page_alloc.ignore_gfp_wait_file);
1441 debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file);
1442 debugfs_remove(fail_page_alloc.min_order_file);
1443 cleanup_fault_attr_dentries(&fail_page_alloc.attr);
1449 late_initcall(fail_page_alloc_debugfs);
1451 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1453 #else /* CONFIG_FAIL_PAGE_ALLOC */
1455 static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1460 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1463 * Return true if free pages are above 'mark'. This takes into account the order
1464 * of the allocation.
1466 static bool __zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1467 int classzone_idx, int alloc_flags, long free_pages)
1469 /* free_pages my go negative - that's OK */
1473 free_pages -= (1 << order) + 1;
1474 if (alloc_flags & ALLOC_HIGH)
1476 if (alloc_flags & ALLOC_HARDER)
1479 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
1481 for (o = 0; o < order; o++) {
1482 /* At the next order, this order's pages become unavailable */
1483 free_pages -= z->free_area[o].nr_free << o;
1485 /* Require fewer higher order pages to be free */
1488 if (free_pages <= min)
1494 bool zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1495 int classzone_idx, int alloc_flags)
1497 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
1498 zone_page_state(z, NR_FREE_PAGES));
1501 bool zone_watermark_ok_safe(struct zone *z, int order, unsigned long mark,
1502 int classzone_idx, int alloc_flags)
1504 long free_pages = zone_page_state(z, NR_FREE_PAGES);
1506 if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark)
1507 free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES);
1509 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
1515 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1516 * skip over zones that are not allowed by the cpuset, or that have
1517 * been recently (in last second) found to be nearly full. See further
1518 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1519 * that have to skip over a lot of full or unallowed zones.
1521 * If the zonelist cache is present in the passed in zonelist, then
1522 * returns a pointer to the allowed node mask (either the current
1523 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
1525 * If the zonelist cache is not available for this zonelist, does
1526 * nothing and returns NULL.
1528 * If the fullzones BITMAP in the zonelist cache is stale (more than
1529 * a second since last zap'd) then we zap it out (clear its bits.)
1531 * We hold off even calling zlc_setup, until after we've checked the
1532 * first zone in the zonelist, on the theory that most allocations will
1533 * be satisfied from that first zone, so best to examine that zone as
1534 * quickly as we can.
1536 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1538 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1539 nodemask_t *allowednodes; /* zonelist_cache approximation */
1541 zlc = zonelist->zlcache_ptr;
1545 if (time_after(jiffies, zlc->last_full_zap + HZ)) {
1546 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1547 zlc->last_full_zap = jiffies;
1550 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1551 &cpuset_current_mems_allowed :
1552 &node_states[N_HIGH_MEMORY];
1553 return allowednodes;
1557 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1558 * if it is worth looking at further for free memory:
1559 * 1) Check that the zone isn't thought to be full (doesn't have its
1560 * bit set in the zonelist_cache fullzones BITMAP).
1561 * 2) Check that the zones node (obtained from the zonelist_cache
1562 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1563 * Return true (non-zero) if zone is worth looking at further, or
1564 * else return false (zero) if it is not.
1566 * This check -ignores- the distinction between various watermarks,
1567 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1568 * found to be full for any variation of these watermarks, it will
1569 * be considered full for up to one second by all requests, unless
1570 * we are so low on memory on all allowed nodes that we are forced
1571 * into the second scan of the zonelist.
1573 * In the second scan we ignore this zonelist cache and exactly
1574 * apply the watermarks to all zones, even it is slower to do so.
1575 * We are low on memory in the second scan, and should leave no stone
1576 * unturned looking for a free page.
1578 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
1579 nodemask_t *allowednodes)
1581 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1582 int i; /* index of *z in zonelist zones */
1583 int n; /* node that zone *z is on */
1585 zlc = zonelist->zlcache_ptr;
1589 i = z - zonelist->_zonerefs;
1592 /* This zone is worth trying if it is allowed but not full */
1593 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1597 * Given 'z' scanning a zonelist, set the corresponding bit in
1598 * zlc->fullzones, so that subsequent attempts to allocate a page
1599 * from that zone don't waste time re-examining it.
1601 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
1603 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1604 int i; /* index of *z in zonelist zones */
1606 zlc = zonelist->zlcache_ptr;
1610 i = z - zonelist->_zonerefs;
1612 set_bit(i, zlc->fullzones);
1615 #else /* CONFIG_NUMA */
1617 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1622 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
1623 nodemask_t *allowednodes)
1628 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
1631 #endif /* CONFIG_NUMA */
1634 * get_page_from_freelist goes through the zonelist trying to allocate
1637 static struct page *
1638 get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order,
1639 struct zonelist *zonelist, int high_zoneidx, int alloc_flags,
1640 struct zone *preferred_zone, int migratetype)
1643 struct page *page = NULL;
1646 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1647 int zlc_active = 0; /* set if using zonelist_cache */
1648 int did_zlc_setup = 0; /* just call zlc_setup() one time */
1650 classzone_idx = zone_idx(preferred_zone);
1653 * Scan zonelist, looking for a zone with enough free.
1654 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1656 for_each_zone_zonelist_nodemask(zone, z, zonelist,
1657 high_zoneidx, nodemask) {
1658 if (NUMA_BUILD && zlc_active &&
1659 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1661 if ((alloc_flags & ALLOC_CPUSET) &&
1662 !cpuset_zone_allowed_softwall(zone, gfp_mask))
1665 BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
1666 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
1670 mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
1671 if (zone_watermark_ok(zone, order, mark,
1672 classzone_idx, alloc_flags))
1675 if (zone_reclaim_mode == 0)
1676 goto this_zone_full;
1678 ret = zone_reclaim(zone, gfp_mask, order);
1680 case ZONE_RECLAIM_NOSCAN:
1683 case ZONE_RECLAIM_FULL:
1684 /* scanned but unreclaimable */
1685 goto this_zone_full;
1687 /* did we reclaim enough */
1688 if (!zone_watermark_ok(zone, order, mark,
1689 classzone_idx, alloc_flags))
1690 goto this_zone_full;
1695 page = buffered_rmqueue(preferred_zone, zone, order,
1696 gfp_mask, migratetype);
1701 zlc_mark_zone_full(zonelist, z);
1703 if (NUMA_BUILD && !did_zlc_setup && nr_online_nodes > 1) {
1705 * we do zlc_setup after the first zone is tried but only
1706 * if there are multiple nodes make it worthwhile
1708 allowednodes = zlc_setup(zonelist, alloc_flags);
1714 if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) {
1715 /* Disable zlc cache for second zonelist scan */
1723 should_alloc_retry(gfp_t gfp_mask, unsigned int order,
1724 unsigned long pages_reclaimed)
1726 /* Do not loop if specifically requested */
1727 if (gfp_mask & __GFP_NORETRY)
1731 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
1732 * means __GFP_NOFAIL, but that may not be true in other
1735 if (order <= PAGE_ALLOC_COSTLY_ORDER)
1739 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
1740 * specified, then we retry until we no longer reclaim any pages
1741 * (above), or we've reclaimed an order of pages at least as
1742 * large as the allocation's order. In both cases, if the
1743 * allocation still fails, we stop retrying.
1745 if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order))
1749 * Don't let big-order allocations loop unless the caller
1750 * explicitly requests that.
1752 if (gfp_mask & __GFP_NOFAIL)
1758 static inline struct page *
1759 __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
1760 struct zonelist *zonelist, enum zone_type high_zoneidx,
1761 nodemask_t *nodemask, struct zone *preferred_zone,
1766 /* Acquire the OOM killer lock for the zones in zonelist */
1767 if (!try_set_zonelist_oom(zonelist, gfp_mask)) {
1768 schedule_timeout_uninterruptible(1);
1773 * Go through the zonelist yet one more time, keep very high watermark
1774 * here, this is only to catch a parallel oom killing, we must fail if
1775 * we're still under heavy pressure.
1777 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask,
1778 order, zonelist, high_zoneidx,
1779 ALLOC_WMARK_HIGH|ALLOC_CPUSET,
1780 preferred_zone, migratetype);
1784 if (!(gfp_mask & __GFP_NOFAIL)) {
1785 /* The OOM killer will not help higher order allocs */
1786 if (order > PAGE_ALLOC_COSTLY_ORDER)
1788 /* The OOM killer does not needlessly kill tasks for lowmem */
1789 if (high_zoneidx < ZONE_NORMAL)
1792 * GFP_THISNODE contains __GFP_NORETRY and we never hit this.
1793 * Sanity check for bare calls of __GFP_THISNODE, not real OOM.
1794 * The caller should handle page allocation failure by itself if
1795 * it specifies __GFP_THISNODE.
1796 * Note: Hugepage uses it but will hit PAGE_ALLOC_COSTLY_ORDER.
1798 if (gfp_mask & __GFP_THISNODE)
1801 /* Exhausted what can be done so it's blamo time */
1802 out_of_memory(zonelist, gfp_mask, order, nodemask);
1805 clear_zonelist_oom(zonelist, gfp_mask);
1809 #ifdef CONFIG_COMPACTION
1810 /* Try memory compaction for high-order allocations before reclaim */
1811 static struct page *
1812 __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
1813 struct zonelist *zonelist, enum zone_type high_zoneidx,
1814 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
1815 int migratetype, unsigned long *did_some_progress)
1819 if (!order || compaction_deferred(preferred_zone))
1822 *did_some_progress = try_to_compact_pages(zonelist, order, gfp_mask,
1824 if (*did_some_progress != COMPACT_SKIPPED) {
1826 /* Page migration frees to the PCP lists but we want merging */
1827 drain_pages(get_cpu());
1830 page = get_page_from_freelist(gfp_mask, nodemask,
1831 order, zonelist, high_zoneidx,
1832 alloc_flags, preferred_zone,
1835 preferred_zone->compact_considered = 0;
1836 preferred_zone->compact_defer_shift = 0;
1837 count_vm_event(COMPACTSUCCESS);
1842 * It's bad if compaction run occurs and fails.
1843 * The most likely reason is that pages exist,
1844 * but not enough to satisfy watermarks.
1846 count_vm_event(COMPACTFAIL);
1847 defer_compaction(preferred_zone);
1855 static inline struct page *
1856 __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
1857 struct zonelist *zonelist, enum zone_type high_zoneidx,
1858 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
1859 int migratetype, unsigned long *did_some_progress)
1863 #endif /* CONFIG_COMPACTION */
1865 /* The really slow allocator path where we enter direct reclaim */
1866 static inline struct page *
1867 __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
1868 struct zonelist *zonelist, enum zone_type high_zoneidx,
1869 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
1870 int migratetype, unsigned long *did_some_progress)
1872 struct page *page = NULL;
1873 struct reclaim_state reclaim_state;
1874 struct task_struct *p = current;
1875 bool drained = false;
1879 /* We now go into synchronous reclaim */
1880 cpuset_memory_pressure_bump();
1881 p->flags |= PF_MEMALLOC;
1882 lockdep_set_current_reclaim_state(gfp_mask);
1883 reclaim_state.reclaimed_slab = 0;
1884 p->reclaim_state = &reclaim_state;
1886 *did_some_progress = try_to_free_pages(zonelist, order, gfp_mask, nodemask);
1888 p->reclaim_state = NULL;
1889 lockdep_clear_current_reclaim_state();
1890 p->flags &= ~PF_MEMALLOC;
1894 if (unlikely(!(*did_some_progress)))
1898 page = get_page_from_freelist(gfp_mask, nodemask, order,
1899 zonelist, high_zoneidx,
1900 alloc_flags, preferred_zone,
1904 * If an allocation failed after direct reclaim, it could be because
1905 * pages are pinned on the per-cpu lists. Drain them and try again
1907 if (!page && !drained) {
1917 * This is called in the allocator slow-path if the allocation request is of
1918 * sufficient urgency to ignore watermarks and take other desperate measures
1920 static inline struct page *
1921 __alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order,
1922 struct zonelist *zonelist, enum zone_type high_zoneidx,
1923 nodemask_t *nodemask, struct zone *preferred_zone,
1929 page = get_page_from_freelist(gfp_mask, nodemask, order,
1930 zonelist, high_zoneidx, ALLOC_NO_WATERMARKS,
1931 preferred_zone, migratetype);
1933 if (!page && gfp_mask & __GFP_NOFAIL)
1934 wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50);
1935 } while (!page && (gfp_mask & __GFP_NOFAIL));
1941 void wake_all_kswapd(unsigned int order, struct zonelist *zonelist,
1942 enum zone_type high_zoneidx)
1947 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx)
1948 wakeup_kswapd(zone, order);
1952 gfp_to_alloc_flags(gfp_t gfp_mask)
1954 struct task_struct *p = current;
1955 int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
1956 const gfp_t wait = gfp_mask & __GFP_WAIT;
1958 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
1959 BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
1962 * The caller may dip into page reserves a bit more if the caller
1963 * cannot run direct reclaim, or if the caller has realtime scheduling
1964 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1965 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1967 alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH);
1970 alloc_flags |= ALLOC_HARDER;
1972 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1973 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1975 alloc_flags &= ~ALLOC_CPUSET;
1976 } else if (unlikely(rt_task(p)) && !in_interrupt())
1977 alloc_flags |= ALLOC_HARDER;
1979 if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) {
1980 if (!in_interrupt() &&
1981 ((p->flags & PF_MEMALLOC) ||
1982 unlikely(test_thread_flag(TIF_MEMDIE))))
1983 alloc_flags |= ALLOC_NO_WATERMARKS;
1989 static inline struct page *
1990 __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
1991 struct zonelist *zonelist, enum zone_type high_zoneidx,
1992 nodemask_t *nodemask, struct zone *preferred_zone,
1995 const gfp_t wait = gfp_mask & __GFP_WAIT;
1996 struct page *page = NULL;
1998 unsigned long pages_reclaimed = 0;
1999 unsigned long did_some_progress;
2000 struct task_struct *p = current;
2003 * In the slowpath, we sanity check order to avoid ever trying to
2004 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
2005 * be using allocators in order of preference for an area that is
2008 if (order >= MAX_ORDER) {
2009 WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
2014 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
2015 * __GFP_NOWARN set) should not cause reclaim since the subsystem
2016 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
2017 * using a larger set of nodes after it has established that the
2018 * allowed per node queues are empty and that nodes are
2021 if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
2025 wake_all_kswapd(order, zonelist, high_zoneidx);
2028 * OK, we're below the kswapd watermark and have kicked background
2029 * reclaim. Now things get more complex, so set up alloc_flags according
2030 * to how we want to proceed.
2032 alloc_flags = gfp_to_alloc_flags(gfp_mask);
2034 /* This is the last chance, in general, before the goto nopage. */
2035 page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist,
2036 high_zoneidx, alloc_flags & ~ALLOC_NO_WATERMARKS,
2037 preferred_zone, migratetype);
2042 /* Allocate without watermarks if the context allows */
2043 if (alloc_flags & ALLOC_NO_WATERMARKS) {
2044 page = __alloc_pages_high_priority(gfp_mask, order,
2045 zonelist, high_zoneidx, nodemask,
2046 preferred_zone, migratetype);
2051 /* Atomic allocations - we can't balance anything */
2055 /* Avoid recursion of direct reclaim */
2056 if (p->flags & PF_MEMALLOC)
2059 /* Avoid allocations with no watermarks from looping endlessly */
2060 if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL))
2063 /* Try direct compaction */
2064 page = __alloc_pages_direct_compact(gfp_mask, order,
2065 zonelist, high_zoneidx,
2067 alloc_flags, preferred_zone,
2068 migratetype, &did_some_progress);
2072 /* Try direct reclaim and then allocating */
2073 page = __alloc_pages_direct_reclaim(gfp_mask, order,
2074 zonelist, high_zoneidx,
2076 alloc_flags, preferred_zone,
2077 migratetype, &did_some_progress);
2082 * If we failed to make any progress reclaiming, then we are
2083 * running out of options and have to consider going OOM
2085 if (!did_some_progress) {
2086 if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
2087 if (oom_killer_disabled)
2089 page = __alloc_pages_may_oom(gfp_mask, order,
2090 zonelist, high_zoneidx,
2091 nodemask, preferred_zone,
2096 if (!(gfp_mask & __GFP_NOFAIL)) {
2098 * The oom killer is not called for high-order
2099 * allocations that may fail, so if no progress
2100 * is being made, there are no other options and
2101 * retrying is unlikely to help.
2103 if (order > PAGE_ALLOC_COSTLY_ORDER)
2106 * The oom killer is not called for lowmem
2107 * allocations to prevent needlessly killing
2110 if (high_zoneidx < ZONE_NORMAL)
2118 /* Check if we should retry the allocation */
2119 pages_reclaimed += did_some_progress;
2120 if (should_alloc_retry(gfp_mask, order, pages_reclaimed)) {
2121 /* Wait for some write requests to complete then retry */
2122 wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50);
2127 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
2128 printk(KERN_WARNING "%s: page allocation failure."
2129 " order:%d, mode:0x%x\n",
2130 p->comm, order, gfp_mask);
2136 if (kmemcheck_enabled)
2137 kmemcheck_pagealloc_alloc(page, order, gfp_mask);
2143 * This is the 'heart' of the zoned buddy allocator.
2146 __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
2147 struct zonelist *zonelist, nodemask_t *nodemask)
2149 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
2150 struct zone *preferred_zone;
2152 int migratetype = allocflags_to_migratetype(gfp_mask);
2154 gfp_mask &= gfp_allowed_mask;
2156 lockdep_trace_alloc(gfp_mask);
2158 might_sleep_if(gfp_mask & __GFP_WAIT);
2160 if (should_fail_alloc_page(gfp_mask, order))
2164 * Check the zones suitable for the gfp_mask contain at least one
2165 * valid zone. It's possible to have an empty zonelist as a result
2166 * of GFP_THISNODE and a memoryless node
2168 if (unlikely(!zonelist->_zonerefs->zone))
2172 /* The preferred zone is used for statistics later */
2173 first_zones_zonelist(zonelist, high_zoneidx, nodemask, &preferred_zone);
2174 if (!preferred_zone) {
2179 /* First allocation attempt */
2180 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order,
2181 zonelist, high_zoneidx, ALLOC_WMARK_LOW|ALLOC_CPUSET,
2182 preferred_zone, migratetype);
2183 if (unlikely(!page))
2184 page = __alloc_pages_slowpath(gfp_mask, order,
2185 zonelist, high_zoneidx, nodemask,
2186 preferred_zone, migratetype);
2189 trace_mm_page_alloc(page, order, gfp_mask, migratetype);
2192 EXPORT_SYMBOL(__alloc_pages_nodemask);
2195 * Common helper functions.
2197 unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
2202 * __get_free_pages() returns a 32-bit address, which cannot represent
2205 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
2207 page = alloc_pages(gfp_mask, order);
2210 return (unsigned long) page_address(page);
2212 EXPORT_SYMBOL(__get_free_pages);
2214 unsigned long get_zeroed_page(gfp_t gfp_mask)
2216 return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
2218 EXPORT_SYMBOL(get_zeroed_page);
2220 void __pagevec_free(struct pagevec *pvec)
2222 int i = pagevec_count(pvec);
2225 trace_mm_pagevec_free(pvec->pages[i], pvec->cold);
2226 free_hot_cold_page(pvec->pages[i], pvec->cold);
2230 void __free_pages(struct page *page, unsigned int order)
2232 if (put_page_testzero(page)) {
2234 free_hot_cold_page(page, 0);
2236 __free_pages_ok(page, order);
2240 EXPORT_SYMBOL(__free_pages);
2242 void free_pages(unsigned long addr, unsigned int order)
2245 VM_BUG_ON(!virt_addr_valid((void *)addr));
2246 __free_pages(virt_to_page((void *)addr), order);
2250 EXPORT_SYMBOL(free_pages);
2253 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
2254 * @size: the number of bytes to allocate
2255 * @gfp_mask: GFP flags for the allocation
2257 * This function is similar to alloc_pages(), except that it allocates the
2258 * minimum number of pages to satisfy the request. alloc_pages() can only
2259 * allocate memory in power-of-two pages.
2261 * This function is also limited by MAX_ORDER.
2263 * Memory allocated by this function must be released by free_pages_exact().
2265 void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
2267 unsigned int order = get_order(size);
2270 addr = __get_free_pages(gfp_mask, order);
2272 unsigned long alloc_end = addr + (PAGE_SIZE << order);
2273 unsigned long used = addr + PAGE_ALIGN(size);
2275 split_page(virt_to_page((void *)addr), order);
2276 while (used < alloc_end) {
2282 return (void *)addr;
2284 EXPORT_SYMBOL(alloc_pages_exact);
2287 * free_pages_exact - release memory allocated via alloc_pages_exact()
2288 * @virt: the value returned by alloc_pages_exact.
2289 * @size: size of allocation, same value as passed to alloc_pages_exact().
2291 * Release the memory allocated by a previous call to alloc_pages_exact.
2293 void free_pages_exact(void *virt, size_t size)
2295 unsigned long addr = (unsigned long)virt;
2296 unsigned long end = addr + PAGE_ALIGN(size);
2298 while (addr < end) {
2303 EXPORT_SYMBOL(free_pages_exact);
2305 static unsigned int nr_free_zone_pages(int offset)
2310 /* Just pick one node, since fallback list is circular */
2311 unsigned int sum = 0;
2313 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
2315 for_each_zone_zonelist(zone, z, zonelist, offset) {
2316 unsigned long size = zone->present_pages;
2317 unsigned long high = high_wmark_pages(zone);
2326 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
2328 unsigned int nr_free_buffer_pages(void)
2330 return nr_free_zone_pages(gfp_zone(GFP_USER));
2332 EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
2335 * Amount of free RAM allocatable within all zones
2337 unsigned int nr_free_pagecache_pages(void)
2339 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
2342 static inline void show_node(struct zone *zone)
2345 printk("Node %d ", zone_to_nid(zone));
2348 void si_meminfo(struct sysinfo *val)
2350 val->totalram = totalram_pages;
2352 val->freeram = global_page_state(NR_FREE_PAGES);
2353 val->bufferram = nr_blockdev_pages();
2354 val->totalhigh = totalhigh_pages;
2355 val->freehigh = nr_free_highpages();
2356 val->mem_unit = PAGE_SIZE;
2359 EXPORT_SYMBOL(si_meminfo);
2362 void si_meminfo_node(struct sysinfo *val, int nid)
2364 pg_data_t *pgdat = NODE_DATA(nid);
2366 val->totalram = pgdat->node_present_pages;
2367 val->freeram = node_page_state(nid, NR_FREE_PAGES);
2368 #ifdef CONFIG_HIGHMEM
2369 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
2370 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
2376 val->mem_unit = PAGE_SIZE;
2380 #define K(x) ((x) << (PAGE_SHIFT-10))
2383 * Show free area list (used inside shift_scroll-lock stuff)
2384 * We also calculate the percentage fragmentation. We do this by counting the
2385 * memory on each free list with the exception of the first item on the list.
2387 void show_free_areas(void)
2392 for_each_populated_zone(zone) {
2394 printk("%s per-cpu:\n", zone->name);
2396 for_each_online_cpu(cpu) {
2397 struct per_cpu_pageset *pageset;
2399 pageset = per_cpu_ptr(zone->pageset, cpu);
2401 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
2402 cpu, pageset->pcp.high,
2403 pageset->pcp.batch, pageset->pcp.count);
2407 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
2408 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
2410 " dirty:%lu writeback:%lu unstable:%lu\n"
2411 " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n"
2412 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n",
2413 global_page_state(NR_ACTIVE_ANON),
2414 global_page_state(NR_INACTIVE_ANON),
2415 global_page_state(NR_ISOLATED_ANON),
2416 global_page_state(NR_ACTIVE_FILE),
2417 global_page_state(NR_INACTIVE_FILE),
2418 global_page_state(NR_ISOLATED_FILE),
2419 global_page_state(NR_UNEVICTABLE),
2420 global_page_state(NR_FILE_DIRTY),
2421 global_page_state(NR_WRITEBACK),
2422 global_page_state(NR_UNSTABLE_NFS),
2423 global_page_state(NR_FREE_PAGES),
2424 global_page_state(NR_SLAB_RECLAIMABLE),
2425 global_page_state(NR_SLAB_UNRECLAIMABLE),
2426 global_page_state(NR_FILE_MAPPED),
2427 global_page_state(NR_SHMEM),
2428 global_page_state(NR_PAGETABLE),
2429 global_page_state(NR_BOUNCE));
2431 for_each_populated_zone(zone) {
2440 " active_anon:%lukB"
2441 " inactive_anon:%lukB"
2442 " active_file:%lukB"
2443 " inactive_file:%lukB"
2444 " unevictable:%lukB"
2445 " isolated(anon):%lukB"
2446 " isolated(file):%lukB"
2453 " slab_reclaimable:%lukB"
2454 " slab_unreclaimable:%lukB"
2455 " kernel_stack:%lukB"
2459 " writeback_tmp:%lukB"
2460 " pages_scanned:%lu"
2461 " all_unreclaimable? %s"
2464 K(zone_page_state(zone, NR_FREE_PAGES)),
2465 K(min_wmark_pages(zone)),
2466 K(low_wmark_pages(zone)),
2467 K(high_wmark_pages(zone)),
2468 K(zone_page_state(zone, NR_ACTIVE_ANON)),
2469 K(zone_page_state(zone, NR_INACTIVE_ANON)),
2470 K(zone_page_state(zone, NR_ACTIVE_FILE)),
2471 K(zone_page_state(zone, NR_INACTIVE_FILE)),
2472 K(zone_page_state(zone, NR_UNEVICTABLE)),
2473 K(zone_page_state(zone, NR_ISOLATED_ANON)),
2474 K(zone_page_state(zone, NR_ISOLATED_FILE)),
2475 K(zone->present_pages),
2476 K(zone_page_state(zone, NR_MLOCK)),
2477 K(zone_page_state(zone, NR_FILE_DIRTY)),
2478 K(zone_page_state(zone, NR_WRITEBACK)),
2479 K(zone_page_state(zone, NR_FILE_MAPPED)),
2480 K(zone_page_state(zone, NR_SHMEM)),
2481 K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)),
2482 K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)),
2483 zone_page_state(zone, NR_KERNEL_STACK) *
2485 K(zone_page_state(zone, NR_PAGETABLE)),
2486 K(zone_page_state(zone, NR_UNSTABLE_NFS)),
2487 K(zone_page_state(zone, NR_BOUNCE)),
2488 K(zone_page_state(zone, NR_WRITEBACK_TEMP)),
2489 zone->pages_scanned,
2490 (zone->all_unreclaimable ? "yes" : "no")
2492 printk("lowmem_reserve[]:");
2493 for (i = 0; i < MAX_NR_ZONES; i++)
2494 printk(" %lu", zone->lowmem_reserve[i]);
2498 for_each_populated_zone(zone) {
2499 unsigned long nr[MAX_ORDER], flags, order, total = 0;
2502 printk("%s: ", zone->name);
2504 spin_lock_irqsave(&zone->lock, flags);
2505 for (order = 0; order < MAX_ORDER; order++) {
2506 nr[order] = zone->free_area[order].nr_free;
2507 total += nr[order] << order;
2509 spin_unlock_irqrestore(&zone->lock, flags);
2510 for (order = 0; order < MAX_ORDER; order++)
2511 printk("%lu*%lukB ", nr[order], K(1UL) << order);
2512 printk("= %lukB\n", K(total));
2515 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
2517 show_swap_cache_info();
2520 static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
2522 zoneref->zone = zone;
2523 zoneref->zone_idx = zone_idx(zone);
2527 * Builds allocation fallback zone lists.
2529 * Add all populated zones of a node to the zonelist.
2531 static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
2532 int nr_zones, enum zone_type zone_type)
2536 BUG_ON(zone_type >= MAX_NR_ZONES);
2541 zone = pgdat->node_zones + zone_type;
2542 if (populated_zone(zone)) {
2543 zoneref_set_zone(zone,
2544 &zonelist->_zonerefs[nr_zones++]);
2545 check_highest_zone(zone_type);
2548 } while (zone_type);
2555 * 0 = automatic detection of better ordering.
2556 * 1 = order by ([node] distance, -zonetype)
2557 * 2 = order by (-zonetype, [node] distance)
2559 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
2560 * the same zonelist. So only NUMA can configure this param.
2562 #define ZONELIST_ORDER_DEFAULT 0
2563 #define ZONELIST_ORDER_NODE 1
2564 #define ZONELIST_ORDER_ZONE 2
2566 /* zonelist order in the kernel.
2567 * set_zonelist_order() will set this to NODE or ZONE.
2569 static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
2570 static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
2574 /* The value user specified ....changed by config */
2575 static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2576 /* string for sysctl */
2577 #define NUMA_ZONELIST_ORDER_LEN 16
2578 char numa_zonelist_order[16] = "default";
2581 * interface for configure zonelist ordering.
2582 * command line option "numa_zonelist_order"
2583 * = "[dD]efault - default, automatic configuration.
2584 * = "[nN]ode - order by node locality, then by zone within node
2585 * = "[zZ]one - order by zone, then by locality within zone
2588 static int __parse_numa_zonelist_order(char *s)
2590 if (*s == 'd' || *s == 'D') {
2591 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2592 } else if (*s == 'n' || *s == 'N') {
2593 user_zonelist_order = ZONELIST_ORDER_NODE;
2594 } else if (*s == 'z' || *s == 'Z') {
2595 user_zonelist_order = ZONELIST_ORDER_ZONE;
2598 "Ignoring invalid numa_zonelist_order value: "
2605 static __init int setup_numa_zonelist_order(char *s)
2608 return __parse_numa_zonelist_order(s);
2611 early_param("numa_zonelist_order", setup_numa_zonelist_order);
2614 * sysctl handler for numa_zonelist_order
2616 int numa_zonelist_order_handler(ctl_table *table, int write,
2617 void __user *buffer, size_t *length,
2620 char saved_string[NUMA_ZONELIST_ORDER_LEN];
2622 static DEFINE_MUTEX(zl_order_mutex);
2624 mutex_lock(&zl_order_mutex);
2626 strcpy(saved_string, (char*)table->data);
2627 ret = proc_dostring(table, write, buffer, length, ppos);
2631 int oldval = user_zonelist_order;
2632 if (__parse_numa_zonelist_order((char*)table->data)) {
2634 * bogus value. restore saved string
2636 strncpy((char*)table->data, saved_string,
2637 NUMA_ZONELIST_ORDER_LEN);
2638 user_zonelist_order = oldval;
2639 } else if (oldval != user_zonelist_order) {
2640 mutex_lock(&zonelists_mutex);
2641 build_all_zonelists(NULL);
2642 mutex_unlock(&zonelists_mutex);
2646 mutex_unlock(&zl_order_mutex);
2651 #define MAX_NODE_LOAD (nr_online_nodes)
2652 static int node_load[MAX_NUMNODES];
2655 * find_next_best_node - find the next node that should appear in a given node's fallback list
2656 * @node: node whose fallback list we're appending
2657 * @used_node_mask: nodemask_t of already used nodes
2659 * We use a number of factors to determine which is the next node that should
2660 * appear on a given node's fallback list. The node should not have appeared
2661 * already in @node's fallback list, and it should be the next closest node
2662 * according to the distance array (which contains arbitrary distance values
2663 * from each node to each node in the system), and should also prefer nodes
2664 * with no CPUs, since presumably they'll have very little allocation pressure
2665 * on them otherwise.
2666 * It returns -1 if no node is found.
2668 static int find_next_best_node(int node, nodemask_t *used_node_mask)
2671 int min_val = INT_MAX;
2673 const struct cpumask *tmp = cpumask_of_node(0);
2675 /* Use the local node if we haven't already */
2676 if (!node_isset(node, *used_node_mask)) {
2677 node_set(node, *used_node_mask);
2681 for_each_node_state(n, N_HIGH_MEMORY) {
2683 /* Don't want a node to appear more than once */
2684 if (node_isset(n, *used_node_mask))
2687 /* Use the distance array to find the distance */
2688 val = node_distance(node, n);
2690 /* Penalize nodes under us ("prefer the next node") */
2693 /* Give preference to headless and unused nodes */
2694 tmp = cpumask_of_node(n);
2695 if (!cpumask_empty(tmp))
2696 val += PENALTY_FOR_NODE_WITH_CPUS;
2698 /* Slight preference for less loaded node */
2699 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
2700 val += node_load[n];
2702 if (val < min_val) {
2709 node_set(best_node, *used_node_mask);
2716 * Build zonelists ordered by node and zones within node.
2717 * This results in maximum locality--normal zone overflows into local
2718 * DMA zone, if any--but risks exhausting DMA zone.
2720 static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
2723 struct zonelist *zonelist;
2725 zonelist = &pgdat->node_zonelists[0];
2726 for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
2728 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2730 zonelist->_zonerefs[j].zone = NULL;
2731 zonelist->_zonerefs[j].zone_idx = 0;
2735 * Build gfp_thisnode zonelists
2737 static void build_thisnode_zonelists(pg_data_t *pgdat)
2740 struct zonelist *zonelist;
2742 zonelist = &pgdat->node_zonelists[1];
2743 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
2744 zonelist->_zonerefs[j].zone = NULL;
2745 zonelist->_zonerefs[j].zone_idx = 0;
2749 * Build zonelists ordered by zone and nodes within zones.
2750 * This results in conserving DMA zone[s] until all Normal memory is
2751 * exhausted, but results in overflowing to remote node while memory
2752 * may still exist in local DMA zone.
2754 static int node_order[MAX_NUMNODES];
2756 static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
2759 int zone_type; /* needs to be signed */
2761 struct zonelist *zonelist;
2763 zonelist = &pgdat->node_zonelists[0];
2765 for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
2766 for (j = 0; j < nr_nodes; j++) {
2767 node = node_order[j];
2768 z = &NODE_DATA(node)->node_zones[zone_type];
2769 if (populated_zone(z)) {
2771 &zonelist->_zonerefs[pos++]);
2772 check_highest_zone(zone_type);
2776 zonelist->_zonerefs[pos].zone = NULL;
2777 zonelist->_zonerefs[pos].zone_idx = 0;
2780 static int default_zonelist_order(void)
2783 unsigned long low_kmem_size,total_size;
2787 * ZONE_DMA and ZONE_DMA32 can be very small area in the system.
2788 * If they are really small and used heavily, the system can fall
2789 * into OOM very easily.
2790 * This function detect ZONE_DMA/DMA32 size and configures zone order.
2792 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2795 for_each_online_node(nid) {
2796 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2797 z = &NODE_DATA(nid)->node_zones[zone_type];
2798 if (populated_zone(z)) {
2799 if (zone_type < ZONE_NORMAL)
2800 low_kmem_size += z->present_pages;
2801 total_size += z->present_pages;
2802 } else if (zone_type == ZONE_NORMAL) {
2804 * If any node has only lowmem, then node order
2805 * is preferred to allow kernel allocations
2806 * locally; otherwise, they can easily infringe
2807 * on other nodes when there is an abundance of
2808 * lowmem available to allocate from.
2810 return ZONELIST_ORDER_NODE;
2814 if (!low_kmem_size || /* there are no DMA area. */
2815 low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
2816 return ZONELIST_ORDER_NODE;
2818 * look into each node's config.
2819 * If there is a node whose DMA/DMA32 memory is very big area on
2820 * local memory, NODE_ORDER may be suitable.
2822 average_size = total_size /
2823 (nodes_weight(node_states[N_HIGH_MEMORY]) + 1);
2824 for_each_online_node(nid) {
2827 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2828 z = &NODE_DATA(nid)->node_zones[zone_type];
2829 if (populated_zone(z)) {
2830 if (zone_type < ZONE_NORMAL)
2831 low_kmem_size += z->present_pages;
2832 total_size += z->present_pages;
2835 if (low_kmem_size &&
2836 total_size > average_size && /* ignore small node */
2837 low_kmem_size > total_size * 70/100)
2838 return ZONELIST_ORDER_NODE;
2840 return ZONELIST_ORDER_ZONE;
2843 static void set_zonelist_order(void)
2845 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
2846 current_zonelist_order = default_zonelist_order();
2848 current_zonelist_order = user_zonelist_order;
2851 static void build_zonelists(pg_data_t *pgdat)
2855 nodemask_t used_mask;
2856 int local_node, prev_node;
2857 struct zonelist *zonelist;
2858 int order = current_zonelist_order;
2860 /* initialize zonelists */
2861 for (i = 0; i < MAX_ZONELISTS; i++) {
2862 zonelist = pgdat->node_zonelists + i;
2863 zonelist->_zonerefs[0].zone = NULL;
2864 zonelist->_zonerefs[0].zone_idx = 0;
2867 /* NUMA-aware ordering of nodes */
2868 local_node = pgdat->node_id;
2869 load = nr_online_nodes;
2870 prev_node = local_node;
2871 nodes_clear(used_mask);
2873 memset(node_order, 0, sizeof(node_order));
2876 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
2877 int distance = node_distance(local_node, node);
2880 * If another node is sufficiently far away then it is better
2881 * to reclaim pages in a zone before going off node.
2883 if (distance > RECLAIM_DISTANCE)
2884 zone_reclaim_mode = 1;
2887 * We don't want to pressure a particular node.
2888 * So adding penalty to the first node in same
2889 * distance group to make it round-robin.
2891 if (distance != node_distance(local_node, prev_node))
2892 node_load[node] = load;
2896 if (order == ZONELIST_ORDER_NODE)
2897 build_zonelists_in_node_order(pgdat, node);
2899 node_order[j++] = node; /* remember order */
2902 if (order == ZONELIST_ORDER_ZONE) {
2903 /* calculate node order -- i.e., DMA last! */
2904 build_zonelists_in_zone_order(pgdat, j);
2907 build_thisnode_zonelists(pgdat);
2910 /* Construct the zonelist performance cache - see further mmzone.h */
2911 static void build_zonelist_cache(pg_data_t *pgdat)
2913 struct zonelist *zonelist;
2914 struct zonelist_cache *zlc;
2917 zonelist = &pgdat->node_zonelists[0];
2918 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
2919 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
2920 for (z = zonelist->_zonerefs; z->zone; z++)
2921 zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z);
2924 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
2926 * Return node id of node used for "local" allocations.
2927 * I.e., first node id of first zone in arg node's generic zonelist.
2928 * Used for initializing percpu 'numa_mem', which is used primarily
2929 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
2931 int local_memory_node(int node)
2935 (void)first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
2936 gfp_zone(GFP_KERNEL),
2943 #else /* CONFIG_NUMA */
2945 static void set_zonelist_order(void)
2947 current_zonelist_order = ZONELIST_ORDER_ZONE;
2950 static void build_zonelists(pg_data_t *pgdat)
2952 int node, local_node;
2954 struct zonelist *zonelist;
2956 local_node = pgdat->node_id;
2958 zonelist = &pgdat->node_zonelists[0];
2959 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
2962 * Now we build the zonelist so that it contains the zones
2963 * of all the other nodes.
2964 * We don't want to pressure a particular node, so when
2965 * building the zones for node N, we make sure that the
2966 * zones coming right after the local ones are those from
2967 * node N+1 (modulo N)
2969 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
2970 if (!node_online(node))
2972 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2975 for (node = 0; node < local_node; node++) {
2976 if (!node_online(node))
2978 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2982 zonelist->_zonerefs[j].zone = NULL;
2983 zonelist->_zonerefs[j].zone_idx = 0;
2986 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
2987 static void build_zonelist_cache(pg_data_t *pgdat)
2989 pgdat->node_zonelists[0].zlcache_ptr = NULL;
2992 #endif /* CONFIG_NUMA */
2995 * Boot pageset table. One per cpu which is going to be used for all
2996 * zones and all nodes. The parameters will be set in such a way
2997 * that an item put on a list will immediately be handed over to
2998 * the buddy list. This is safe since pageset manipulation is done
2999 * with interrupts disabled.
3001 * The boot_pagesets must be kept even after bootup is complete for
3002 * unused processors and/or zones. They do play a role for bootstrapping
3003 * hotplugged processors.
3005 * zoneinfo_show() and maybe other functions do
3006 * not check if the processor is online before following the pageset pointer.
3007 * Other parts of the kernel may not check if the zone is available.
3009 static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch);
3010 static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset);
3011 static void setup_zone_pageset(struct zone *zone);
3014 * Global mutex to protect against size modification of zonelists
3015 * as well as to serialize pageset setup for the new populated zone.
3017 DEFINE_MUTEX(zonelists_mutex);
3019 /* return values int ....just for stop_machine() */
3020 static __init_refok int __build_all_zonelists(void *data)
3026 memset(node_load, 0, sizeof(node_load));
3028 for_each_online_node(nid) {
3029 pg_data_t *pgdat = NODE_DATA(nid);
3031 build_zonelists(pgdat);
3032 build_zonelist_cache(pgdat);
3036 * Initialize the boot_pagesets that are going to be used
3037 * for bootstrapping processors. The real pagesets for
3038 * each zone will be allocated later when the per cpu
3039 * allocator is available.
3041 * boot_pagesets are used also for bootstrapping offline
3042 * cpus if the system is already booted because the pagesets
3043 * are needed to initialize allocators on a specific cpu too.
3044 * F.e. the percpu allocator needs the page allocator which
3045 * needs the percpu allocator in order to allocate its pagesets
3046 * (a chicken-egg dilemma).
3048 for_each_possible_cpu(cpu) {
3049 setup_pageset(&per_cpu(boot_pageset, cpu), 0);
3051 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
3053 * We now know the "local memory node" for each node--
3054 * i.e., the node of the first zone in the generic zonelist.
3055 * Set up numa_mem percpu variable for on-line cpus. During
3056 * boot, only the boot cpu should be on-line; we'll init the
3057 * secondary cpus' numa_mem as they come on-line. During
3058 * node/memory hotplug, we'll fixup all on-line cpus.
3060 if (cpu_online(cpu))
3061 set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
3069 * Called with zonelists_mutex held always
3070 * unless system_state == SYSTEM_BOOTING.
3072 void build_all_zonelists(void *data)
3074 set_zonelist_order();
3076 if (system_state == SYSTEM_BOOTING) {
3077 __build_all_zonelists(NULL);
3078 mminit_verify_zonelist();
3079 cpuset_init_current_mems_allowed();
3081 /* we have to stop all cpus to guarantee there is no user
3083 #ifdef CONFIG_MEMORY_HOTPLUG
3085 setup_zone_pageset((struct zone *)data);
3087 stop_machine(__build_all_zonelists, NULL, NULL);
3088 /* cpuset refresh routine should be here */
3090 vm_total_pages = nr_free_pagecache_pages();
3092 * Disable grouping by mobility if the number of pages in the
3093 * system is too low to allow the mechanism to work. It would be
3094 * more accurate, but expensive to check per-zone. This check is
3095 * made on memory-hotadd so a system can start with mobility
3096 * disabled and enable it later
3098 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
3099 page_group_by_mobility_disabled = 1;
3101 page_group_by_mobility_disabled = 0;
3103 printk("Built %i zonelists in %s order, mobility grouping %s. "
3104 "Total pages: %ld\n",
3106 zonelist_order_name[current_zonelist_order],
3107 page_group_by_mobility_disabled ? "off" : "on",
3110 printk("Policy zone: %s\n", zone_names[policy_zone]);
3115 * Helper functions to size the waitqueue hash table.
3116 * Essentially these want to choose hash table sizes sufficiently
3117 * large so that collisions trying to wait on pages are rare.
3118 * But in fact, the number of active page waitqueues on typical
3119 * systems is ridiculously low, less than 200. So this is even
3120 * conservative, even though it seems large.
3122 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
3123 * waitqueues, i.e. the size of the waitq table given the number of pages.
3125 #define PAGES_PER_WAITQUEUE 256
3127 #ifndef CONFIG_MEMORY_HOTPLUG
3128 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
3130 unsigned long size = 1;
3132 pages /= PAGES_PER_WAITQUEUE;
3134 while (size < pages)
3138 * Once we have dozens or even hundreds of threads sleeping
3139 * on IO we've got bigger problems than wait queue collision.
3140 * Limit the size of the wait table to a reasonable size.
3142 size = min(size, 4096UL);
3144 return max(size, 4UL);
3148 * A zone's size might be changed by hot-add, so it is not possible to determine
3149 * a suitable size for its wait_table. So we use the maximum size now.
3151 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
3153 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
3154 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
3155 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
3157 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
3158 * or more by the traditional way. (See above). It equals:
3160 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
3161 * ia64(16K page size) : = ( 8G + 4M)byte.
3162 * powerpc (64K page size) : = (32G +16M)byte.
3164 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
3171 * This is an integer logarithm so that shifts can be used later
3172 * to extract the more random high bits from the multiplicative
3173 * hash function before the remainder is taken.
3175 static inline unsigned long wait_table_bits(unsigned long size)
3180 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
3183 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
3184 * of blocks reserved is based on min_wmark_pages(zone). The memory within
3185 * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
3186 * higher will lead to a bigger reserve which will get freed as contiguous
3187 * blocks as reclaim kicks in
3189 static void setup_zone_migrate_reserve(struct zone *zone)
3191 unsigned long start_pfn, pfn, end_pfn;
3193 unsigned long block_migratetype;
3196 /* Get the start pfn, end pfn and the number of blocks to reserve */
3197 start_pfn = zone->zone_start_pfn;
3198 end_pfn = start_pfn + zone->spanned_pages;
3199 reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >>
3203 * Reserve blocks are generally in place to help high-order atomic
3204 * allocations that are short-lived. A min_free_kbytes value that
3205 * would result in more than 2 reserve blocks for atomic allocations
3206 * is assumed to be in place to help anti-fragmentation for the
3207 * future allocation of hugepages at runtime.
3209 reserve = min(2, reserve);
3211 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
3212 if (!pfn_valid(pfn))
3214 page = pfn_to_page(pfn);
3216 /* Watch out for overlapping nodes */
3217 if (page_to_nid(page) != zone_to_nid(zone))
3220 /* Blocks with reserved pages will never free, skip them. */
3221 if (PageReserved(page))
3224 block_migratetype = get_pageblock_migratetype(page);
3226 /* If this block is reserved, account for it */
3227 if (reserve > 0 && block_migratetype == MIGRATE_RESERVE) {
3232 /* Suitable for reserving if this block is movable */
3233 if (reserve > 0 && block_migratetype == MIGRATE_MOVABLE) {
3234 set_pageblock_migratetype(page, MIGRATE_RESERVE);
3235 move_freepages_block(zone, page, MIGRATE_RESERVE);
3241 * If the reserve is met and this is a previous reserved block,
3244 if (block_migratetype == MIGRATE_RESERVE) {
3245 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
3246 move_freepages_block(zone, page, MIGRATE_MOVABLE);
3252 * Initially all pages are reserved - free ones are freed
3253 * up by free_all_bootmem() once the early boot process is
3254 * done. Non-atomic initialization, single-pass.
3256 void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
3257 unsigned long start_pfn, enum memmap_context context)
3260 unsigned long end_pfn = start_pfn + size;
3264 if (highest_memmap_pfn < end_pfn - 1)
3265 highest_memmap_pfn = end_pfn - 1;
3267 z = &NODE_DATA(nid)->node_zones[zone];
3268 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
3270 * There can be holes in boot-time mem_map[]s
3271 * handed to this function. They do not
3272 * exist on hotplugged memory.
3274 if (context == MEMMAP_EARLY) {
3275 if (!early_pfn_valid(pfn))
3277 if (!early_pfn_in_nid(pfn, nid))
3280 page = pfn_to_page(pfn);
3281 set_page_links(page, zone, nid, pfn);
3282 mminit_verify_page_links(page, zone, nid, pfn);
3283 init_page_count(page);
3284 reset_page_mapcount(page);
3285 SetPageReserved(page);
3287 * Mark the block movable so that blocks are reserved for
3288 * movable at startup. This will force kernel allocations
3289 * to reserve their blocks rather than leaking throughout
3290 * the address space during boot when many long-lived
3291 * kernel allocations are made. Later some blocks near
3292 * the start are marked MIGRATE_RESERVE by
3293 * setup_zone_migrate_reserve()
3295 * bitmap is created for zone's valid pfn range. but memmap
3296 * can be created for invalid pages (for alignment)
3297 * check here not to call set_pageblock_migratetype() against
3300 if ((z->zone_start_pfn <= pfn)
3301 && (pfn < z->zone_start_pfn + z->spanned_pages)
3302 && !(pfn & (pageblock_nr_pages - 1)))
3303 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
3305 INIT_LIST_HEAD(&page->lru);
3306 #ifdef WANT_PAGE_VIRTUAL
3307 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
3308 if (!is_highmem_idx(zone))
3309 set_page_address(page, __va(pfn << PAGE_SHIFT));
3314 static void __meminit zone_init_free_lists(struct zone *zone)
3317 for_each_migratetype_order(order, t) {
3318 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
3319 zone->free_area[order].nr_free = 0;
3323 #ifndef __HAVE_ARCH_MEMMAP_INIT
3324 #define memmap_init(size, nid, zone, start_pfn) \
3325 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
3328 static int zone_batchsize(struct zone *zone)
3334 * The per-cpu-pages pools are set to around 1000th of the
3335 * size of the zone. But no more than 1/2 of a meg.
3337 * OK, so we don't know how big the cache is. So guess.
3339 batch = zone->present_pages / 1024;
3340 if (batch * PAGE_SIZE > 512 * 1024)
3341 batch = (512 * 1024) / PAGE_SIZE;
3342 batch /= 4; /* We effectively *= 4 below */
3347 * Clamp the batch to a 2^n - 1 value. Having a power
3348 * of 2 value was found to be more likely to have
3349 * suboptimal cache aliasing properties in some cases.
3351 * For example if 2 tasks are alternately allocating
3352 * batches of pages, one task can end up with a lot
3353 * of pages of one half of the possible page colors
3354 * and the other with pages of the other colors.
3356 batch = rounddown_pow_of_two(batch + batch/2) - 1;
3361 /* The deferral and batching of frees should be suppressed under NOMMU
3364 * The problem is that NOMMU needs to be able to allocate large chunks
3365 * of contiguous memory as there's no hardware page translation to
3366 * assemble apparent contiguous memory from discontiguous pages.
3368 * Queueing large contiguous runs of pages for batching, however,
3369 * causes the pages to actually be freed in smaller chunks. As there
3370 * can be a significant delay between the individual batches being
3371 * recycled, this leads to the once large chunks of space being
3372 * fragmented and becoming unavailable for high-order allocations.
3378 static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
3380 struct per_cpu_pages *pcp;
3383 memset(p, 0, sizeof(*p));
3387 pcp->high = 6 * batch;
3388 pcp->batch = max(1UL, 1 * batch);
3389 for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
3390 INIT_LIST_HEAD(&pcp->lists[migratetype]);
3394 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
3395 * to the value high for the pageset p.
3398 static void setup_pagelist_highmark(struct per_cpu_pageset *p,
3401 struct per_cpu_pages *pcp;
3405 pcp->batch = max(1UL, high/4);
3406 if ((high/4) > (PAGE_SHIFT * 8))
3407 pcp->batch = PAGE_SHIFT * 8;
3410 static __meminit void setup_zone_pageset(struct zone *zone)
3414 zone->pageset = alloc_percpu(struct per_cpu_pageset);
3416 for_each_possible_cpu(cpu) {
3417 struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu);
3419 setup_pageset(pcp, zone_batchsize(zone));
3421 if (percpu_pagelist_fraction)
3422 setup_pagelist_highmark(pcp,
3423 (zone->present_pages /
3424 percpu_pagelist_fraction));
3429 * Allocate per cpu pagesets and initialize them.
3430 * Before this call only boot pagesets were available.
3432 void __init setup_per_cpu_pageset(void)
3436 for_each_populated_zone(zone)
3437 setup_zone_pageset(zone);
3440 static noinline __init_refok
3441 int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
3444 struct pglist_data *pgdat = zone->zone_pgdat;
3448 * The per-page waitqueue mechanism uses hashed waitqueues
3451 zone->wait_table_hash_nr_entries =
3452 wait_table_hash_nr_entries(zone_size_pages);
3453 zone->wait_table_bits =
3454 wait_table_bits(zone->wait_table_hash_nr_entries);
3455 alloc_size = zone->wait_table_hash_nr_entries
3456 * sizeof(wait_queue_head_t);
3458 if (!slab_is_available()) {
3459 zone->wait_table = (wait_queue_head_t *)
3460 alloc_bootmem_node(pgdat, alloc_size);
3463 * This case means that a zone whose size was 0 gets new memory
3464 * via memory hot-add.
3465 * But it may be the case that a new node was hot-added. In
3466 * this case vmalloc() will not be able to use this new node's
3467 * memory - this wait_table must be initialized to use this new
3468 * node itself as well.
3469 * To use this new node's memory, further consideration will be
3472 zone->wait_table = vmalloc(alloc_size);
3474 if (!zone->wait_table)
3477 for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
3478 init_waitqueue_head(zone->wait_table + i);
3483 static int __zone_pcp_update(void *data)
3485 struct zone *zone = data;
3487 unsigned long batch = zone_batchsize(zone), flags;
3489 for_each_possible_cpu(cpu) {
3490 struct per_cpu_pageset *pset;
3491 struct per_cpu_pages *pcp;
3493 pset = per_cpu_ptr(zone->pageset, cpu);
3496 local_irq_save(flags);
3497 free_pcppages_bulk(zone, pcp->count, pcp);
3498 setup_pageset(pset, batch);
3499 local_irq_restore(flags);
3504 void zone_pcp_update(struct zone *zone)
3506 stop_machine(__zone_pcp_update, zone, NULL);
3509 static __meminit void zone_pcp_init(struct zone *zone)
3512 * per cpu subsystem is not up at this point. The following code
3513 * relies on the ability of the linker to provide the
3514 * offset of a (static) per cpu variable into the per cpu area.
3516 zone->pageset = &boot_pageset;
3518 if (zone->present_pages)
3519 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n",
3520 zone->name, zone->present_pages,
3521 zone_batchsize(zone));
3524 __meminit int init_currently_empty_zone(struct zone *zone,
3525 unsigned long zone_start_pfn,
3527 enum memmap_context context)
3529 struct pglist_data *pgdat = zone->zone_pgdat;
3531 ret = zone_wait_table_init(zone, size);
3534 pgdat->nr_zones = zone_idx(zone) + 1;
3536 zone->zone_start_pfn = zone_start_pfn;
3538 mminit_dprintk(MMINIT_TRACE, "memmap_init",
3539 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
3541 (unsigned long)zone_idx(zone),
3542 zone_start_pfn, (zone_start_pfn + size));
3544 zone_init_free_lists(zone);
3549 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3551 * Basic iterator support. Return the first range of PFNs for a node
3552 * Note: nid == MAX_NUMNODES returns first region regardless of node
3554 static int __meminit first_active_region_index_in_nid(int nid)
3558 for (i = 0; i < nr_nodemap_entries; i++)
3559 if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
3566 * Basic iterator support. Return the next active range of PFNs for a node
3567 * Note: nid == MAX_NUMNODES returns next region regardless of node
3569 static int __meminit next_active_region_index_in_nid(int index, int nid)
3571 for (index = index + 1; index < nr_nodemap_entries; index++)
3572 if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
3578 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
3580 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
3581 * Architectures may implement their own version but if add_active_range()
3582 * was used and there are no special requirements, this is a convenient
3585 int __meminit __early_pfn_to_nid(unsigned long pfn)
3589 for (i = 0; i < nr_nodemap_entries; i++) {
3590 unsigned long start_pfn = early_node_map[i].start_pfn;
3591 unsigned long end_pfn = early_node_map[i].end_pfn;
3593 if (start_pfn <= pfn && pfn < end_pfn)
3594 return early_node_map[i].nid;
3596 /* This is a memory hole */
3599 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
3601 int __meminit early_pfn_to_nid(unsigned long pfn)
3605 nid = __early_pfn_to_nid(pfn);
3608 /* just returns 0 */
3612 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
3613 bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
3617 nid = __early_pfn_to_nid(pfn);
3618 if (nid >= 0 && nid != node)
3624 /* Basic iterator support to walk early_node_map[] */
3625 #define for_each_active_range_index_in_nid(i, nid) \
3626 for (i = first_active_region_index_in_nid(nid); i != -1; \
3627 i = next_active_region_index_in_nid(i, nid))
3630 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
3631 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
3632 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
3634 * If an architecture guarantees that all ranges registered with
3635 * add_active_ranges() contain no holes and may be freed, this
3636 * this function may be used instead of calling free_bootmem() manually.
3638 void __init free_bootmem_with_active_regions(int nid,
3639 unsigned long max_low_pfn)
3643 for_each_active_range_index_in_nid(i, nid) {
3644 unsigned long size_pages = 0;
3645 unsigned long end_pfn = early_node_map[i].end_pfn;
3647 if (early_node_map[i].start_pfn >= max_low_pfn)
3650 if (end_pfn > max_low_pfn)
3651 end_pfn = max_low_pfn;
3653 size_pages = end_pfn - early_node_map[i].start_pfn;
3654 free_bootmem_node(NODE_DATA(early_node_map[i].nid),
3655 PFN_PHYS(early_node_map[i].start_pfn),
3656 size_pages << PAGE_SHIFT);
3660 #ifdef CONFIG_HAVE_MEMBLOCK
3661 u64 __init find_memory_core_early(int nid, u64 size, u64 align,
3662 u64 goal, u64 limit)
3666 /* Need to go over early_node_map to find out good range for node */
3667 for_each_active_range_index_in_nid(i, nid) {
3669 u64 ei_start, ei_last;
3670 u64 final_start, final_end;
3672 ei_last = early_node_map[i].end_pfn;
3673 ei_last <<= PAGE_SHIFT;
3674 ei_start = early_node_map[i].start_pfn;
3675 ei_start <<= PAGE_SHIFT;
3677 final_start = max(ei_start, goal);
3678 final_end = min(ei_last, limit);
3680 if (final_start >= final_end)
3683 addr = memblock_find_in_range(final_start, final_end, size, align);
3685 if (addr == MEMBLOCK_ERROR)
3691 return MEMBLOCK_ERROR;
3695 int __init add_from_early_node_map(struct range *range, int az,
3696 int nr_range, int nid)
3701 /* need to go over early_node_map to find out good range for node */
3702 for_each_active_range_index_in_nid(i, nid) {
3703 start = early_node_map[i].start_pfn;
3704 end = early_node_map[i].end_pfn;
3705 nr_range = add_range(range, az, nr_range, start, end);
3710 #ifdef CONFIG_NO_BOOTMEM
3711 void * __init __alloc_memory_core_early(int nid, u64 size, u64 align,
3712 u64 goal, u64 limit)
3717 if (limit > memblock.current_limit)
3718 limit = memblock.current_limit;
3720 addr = find_memory_core_early(nid, size, align, goal, limit);
3722 if (addr == MEMBLOCK_ERROR)
3725 ptr = phys_to_virt(addr);
3726 memset(ptr, 0, size);
3727 memblock_x86_reserve_range(addr, addr + size, "BOOTMEM");
3729 * The min_count is set to 0 so that bootmem allocated blocks
3730 * are never reported as leaks.
3732 kmemleak_alloc(ptr, size, 0, 0);
3738 void __init work_with_active_regions(int nid, work_fn_t work_fn, void *data)
3743 for_each_active_range_index_in_nid(i, nid) {
3744 ret = work_fn(early_node_map[i].start_pfn,
3745 early_node_map[i].end_pfn, data);
3751 * sparse_memory_present_with_active_regions - Call memory_present for each active range
3752 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
3754 * If an architecture guarantees that all ranges registered with
3755 * add_active_ranges() contain no holes and may be freed, this
3756 * function may be used instead of calling memory_present() manually.
3758 void __init sparse_memory_present_with_active_regions(int nid)
3762 for_each_active_range_index_in_nid(i, nid)
3763 memory_present(early_node_map[i].nid,
3764 early_node_map[i].start_pfn,
3765 early_node_map[i].end_pfn);
3769 * get_pfn_range_for_nid - Return the start and end page frames for a node
3770 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
3771 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
3772 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
3774 * It returns the start and end page frame of a node based on information
3775 * provided by an arch calling add_active_range(). If called for a node
3776 * with no available memory, a warning is printed and the start and end
3779 void __meminit get_pfn_range_for_nid(unsigned int nid,
3780 unsigned long *start_pfn, unsigned long *end_pfn)
3786 for_each_active_range_index_in_nid(i, nid) {
3787 *start_pfn = min(*start_pfn, early_node_map[i].start_pfn);
3788 *end_pfn = max(*end_pfn, early_node_map[i].end_pfn);
3791 if (*start_pfn == -1UL)
3796 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3797 * assumption is made that zones within a node are ordered in monotonic
3798 * increasing memory addresses so that the "highest" populated zone is used
3800 static void __init find_usable_zone_for_movable(void)
3803 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
3804 if (zone_index == ZONE_MOVABLE)
3807 if (arch_zone_highest_possible_pfn[zone_index] >
3808 arch_zone_lowest_possible_pfn[zone_index])
3812 VM_BUG_ON(zone_index == -1);
3813 movable_zone = zone_index;
3817 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
3818 * because it is sized independant of architecture. Unlike the other zones,
3819 * the starting point for ZONE_MOVABLE is not fixed. It may be different
3820 * in each node depending on the size of each node and how evenly kernelcore
3821 * is distributed. This helper function adjusts the zone ranges
3822 * provided by the architecture for a given node by using the end of the
3823 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3824 * zones within a node are in order of monotonic increases memory addresses
3826 static void __meminit adjust_zone_range_for_zone_movable(int nid,
3827 unsigned long zone_type,
3828 unsigned long node_start_pfn,
3829 unsigned long node_end_pfn,
3830 unsigned long *zone_start_pfn,
3831 unsigned long *zone_end_pfn)
3833 /* Only adjust if ZONE_MOVABLE is on this node */
3834 if (zone_movable_pfn[nid]) {
3835 /* Size ZONE_MOVABLE */
3836 if (zone_type == ZONE_MOVABLE) {
3837 *zone_start_pfn = zone_movable_pfn[nid];
3838 *zone_end_pfn = min(node_end_pfn,
3839 arch_zone_highest_possible_pfn[movable_zone]);
3841 /* Adjust for ZONE_MOVABLE starting within this range */
3842 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
3843 *zone_end_pfn > zone_movable_pfn[nid]) {
3844 *zone_end_pfn = zone_movable_pfn[nid];
3846 /* Check if this whole range is within ZONE_MOVABLE */
3847 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
3848 *zone_start_pfn = *zone_end_pfn;
3853 * Return the number of pages a zone spans in a node, including holes
3854 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3856 static unsigned long __meminit zone_spanned_pages_in_node(int nid,
3857 unsigned long zone_type,
3858 unsigned long *ignored)
3860 unsigned long node_start_pfn, node_end_pfn;
3861 unsigned long zone_start_pfn, zone_end_pfn;
3863 /* Get the start and end of the node and zone */
3864 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3865 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
3866 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
3867 adjust_zone_range_for_zone_movable(nid, zone_type,
3868 node_start_pfn, node_end_pfn,
3869 &zone_start_pfn, &zone_end_pfn);
3871 /* Check that this node has pages within the zone's required range */
3872 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
3875 /* Move the zone boundaries inside the node if necessary */
3876 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
3877 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
3879 /* Return the spanned pages */
3880 return zone_end_pfn - zone_start_pfn;
3884 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
3885 * then all holes in the requested range will be accounted for.
3887 unsigned long __meminit __absent_pages_in_range(int nid,
3888 unsigned long range_start_pfn,
3889 unsigned long range_end_pfn)
3892 unsigned long prev_end_pfn = 0, hole_pages = 0;
3893 unsigned long start_pfn;
3895 /* Find the end_pfn of the first active range of pfns in the node */
3896 i = first_active_region_index_in_nid(nid);
3900 prev_end_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3902 /* Account for ranges before physical memory on this node */
3903 if (early_node_map[i].start_pfn > range_start_pfn)
3904 hole_pages = prev_end_pfn - range_start_pfn;
3906 /* Find all holes for the zone within the node */
3907 for (; i != -1; i = next_active_region_index_in_nid(i, nid)) {
3909 /* No need to continue if prev_end_pfn is outside the zone */
3910 if (prev_end_pfn >= range_end_pfn)
3913 /* Make sure the end of the zone is not within the hole */
3914 start_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3915 prev_end_pfn = max(prev_end_pfn, range_start_pfn);
3917 /* Update the hole size cound and move on */
3918 if (start_pfn > range_start_pfn) {
3919 BUG_ON(prev_end_pfn > start_pfn);
3920 hole_pages += start_pfn - prev_end_pfn;
3922 prev_end_pfn = early_node_map[i].end_pfn;
3925 /* Account for ranges past physical memory on this node */
3926 if (range_end_pfn > prev_end_pfn)
3927 hole_pages += range_end_pfn -
3928 max(range_start_pfn, prev_end_pfn);
3934 * absent_pages_in_range - Return number of page frames in holes within a range
3935 * @start_pfn: The start PFN to start searching for holes
3936 * @end_pfn: The end PFN to stop searching for holes
3938 * It returns the number of pages frames in memory holes within a range.
3940 unsigned long __init absent_pages_in_range(unsigned long start_pfn,
3941 unsigned long end_pfn)
3943 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
3946 /* Return the number of page frames in holes in a zone on a node */
3947 static unsigned long __meminit zone_absent_pages_in_node(int nid,
3948 unsigned long zone_type,
3949 unsigned long *ignored)
3951 unsigned long node_start_pfn, node_end_pfn;
3952 unsigned long zone_start_pfn, zone_end_pfn;
3954 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3955 zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type],
3957 zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type],
3960 adjust_zone_range_for_zone_movable(nid, zone_type,
3961 node_start_pfn, node_end_pfn,
3962 &zone_start_pfn, &zone_end_pfn);
3963 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
3967 static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
3968 unsigned long zone_type,
3969 unsigned long *zones_size)
3971 return zones_size[zone_type];
3974 static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
3975 unsigned long zone_type,
3976 unsigned long *zholes_size)
3981 return zholes_size[zone_type];
3986 static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
3987 unsigned long *zones_size, unsigned long *zholes_size)
3989 unsigned long realtotalpages, totalpages = 0;
3992 for (i = 0; i < MAX_NR_ZONES; i++)
3993 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
3995 pgdat->node_spanned_pages = totalpages;
3997 realtotalpages = totalpages;
3998 for (i = 0; i < MAX_NR_ZONES; i++)
4000 zone_absent_pages_in_node(pgdat->node_id, i,
4002 pgdat->node_present_pages = realtotalpages;
4003 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
4007 #ifndef CONFIG_SPARSEMEM
4009 * Calculate the size of the zone->blockflags rounded to an unsigned long
4010 * Start by making sure zonesize is a multiple of pageblock_order by rounding
4011 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
4012 * round what is now in bits to nearest long in bits, then return it in
4015 static unsigned long __init usemap_size(unsigned long zonesize)
4017 unsigned long usemapsize;
4019 usemapsize = roundup(zonesize, pageblock_nr_pages);
4020 usemapsize = usemapsize >> pageblock_order;
4021 usemapsize *= NR_PAGEBLOCK_BITS;
4022 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
4024 return usemapsize / 8;
4027 static void __init setup_usemap(struct pglist_data *pgdat,
4028 struct zone *zone, unsigned long zonesize)
4030 unsigned long usemapsize = usemap_size(zonesize);
4031 zone->pageblock_flags = NULL;
4033 zone->pageblock_flags = alloc_bootmem_node(pgdat, usemapsize);
4036 static inline void setup_usemap(struct pglist_data *pgdat,
4037 struct zone *zone, unsigned long zonesize) {}
4038 #endif /* CONFIG_SPARSEMEM */
4040 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
4042 /* Return a sensible default order for the pageblock size. */
4043 static inline int pageblock_default_order(void)
4045 if (HPAGE_SHIFT > PAGE_SHIFT)
4046 return HUGETLB_PAGE_ORDER;
4051 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
4052 static inline void __init set_pageblock_order(unsigned int order)
4054 /* Check that pageblock_nr_pages has not already been setup */
4055 if (pageblock_order)
4059 * Assume the largest contiguous order of interest is a huge page.
4060 * This value may be variable depending on boot parameters on IA64
4062 pageblock_order = order;
4064 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4067 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
4068 * and pageblock_default_order() are unused as pageblock_order is set
4069 * at compile-time. See include/linux/pageblock-flags.h for the values of
4070 * pageblock_order based on the kernel config
4072 static inline int pageblock_default_order(unsigned int order)
4076 #define set_pageblock_order(x) do {} while (0)
4078 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4081 * Set up the zone data structures:
4082 * - mark all pages reserved
4083 * - mark all memory queues empty
4084 * - clear the memory bitmaps
4086 static void __paginginit free_area_init_core(struct pglist_data *pgdat,
4087 unsigned long *zones_size, unsigned long *zholes_size)
4090 int nid = pgdat->node_id;
4091 unsigned long zone_start_pfn = pgdat->node_start_pfn;
4094 pgdat_resize_init(pgdat);
4095 pgdat->nr_zones = 0;
4096 init_waitqueue_head(&pgdat->kswapd_wait);
4097 pgdat->kswapd_max_order = 0;
4098 pgdat_page_cgroup_init(pgdat);
4100 for (j = 0; j < MAX_NR_ZONES; j++) {
4101 struct zone *zone = pgdat->node_zones + j;
4102 unsigned long size, realsize, memmap_pages;
4105 size = zone_spanned_pages_in_node(nid, j, zones_size);
4106 realsize = size - zone_absent_pages_in_node(nid, j,
4110 * Adjust realsize so that it accounts for how much memory
4111 * is used by this zone for memmap. This affects the watermark
4112 * and per-cpu initialisations
4115 PAGE_ALIGN(size * sizeof(struct page)) >> PAGE_SHIFT;
4116 if (realsize >= memmap_pages) {
4117 realsize -= memmap_pages;
4120 " %s zone: %lu pages used for memmap\n",
4121 zone_names[j], memmap_pages);
4124 " %s zone: %lu pages exceeds realsize %lu\n",
4125 zone_names[j], memmap_pages, realsize);
4127 /* Account for reserved pages */
4128 if (j == 0 && realsize > dma_reserve) {
4129 realsize -= dma_reserve;
4130 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
4131 zone_names[0], dma_reserve);
4134 if (!is_highmem_idx(j))
4135 nr_kernel_pages += realsize;
4136 nr_all_pages += realsize;
4138 zone->spanned_pages = size;
4139 zone->present_pages = realsize;
4142 zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
4144 zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100;
4146 zone->name = zone_names[j];
4147 spin_lock_init(&zone->lock);
4148 spin_lock_init(&zone->lru_lock);
4149 zone_seqlock_init(zone);
4150 zone->zone_pgdat = pgdat;
4152 zone_pcp_init(zone);
4154 INIT_LIST_HEAD(&zone->lru[l].list);
4155 zone->reclaim_stat.nr_saved_scan[l] = 0;
4157 zone->reclaim_stat.recent_rotated[0] = 0;
4158 zone->reclaim_stat.recent_rotated[1] = 0;
4159 zone->reclaim_stat.recent_scanned[0] = 0;
4160 zone->reclaim_stat.recent_scanned[1] = 0;
4161 zap_zone_vm_stats(zone);
4166 set_pageblock_order(pageblock_default_order());
4167 setup_usemap(pgdat, zone, size);
4168 ret = init_currently_empty_zone(zone, zone_start_pfn,
4169 size, MEMMAP_EARLY);
4171 memmap_init(size, nid, j, zone_start_pfn);
4172 zone_start_pfn += size;
4176 static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
4178 /* Skip empty nodes */
4179 if (!pgdat->node_spanned_pages)
4182 #ifdef CONFIG_FLAT_NODE_MEM_MAP
4183 /* ia64 gets its own node_mem_map, before this, without bootmem */
4184 if (!pgdat->node_mem_map) {
4185 unsigned long size, start, end;
4189 * The zone's endpoints aren't required to be MAX_ORDER
4190 * aligned but the node_mem_map endpoints must be in order
4191 * for the buddy allocator to function correctly.
4193 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
4194 end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
4195 end = ALIGN(end, MAX_ORDER_NR_PAGES);
4196 size = (end - start) * sizeof(struct page);
4197 map = alloc_remap(pgdat->node_id, size);
4199 map = alloc_bootmem_node(pgdat, size);
4200 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
4202 #ifndef CONFIG_NEED_MULTIPLE_NODES
4204 * With no DISCONTIG, the global mem_map is just set as node 0's
4206 if (pgdat == NODE_DATA(0)) {
4207 mem_map = NODE_DATA(0)->node_mem_map;
4208 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
4209 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
4210 mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
4211 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4214 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
4217 void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
4218 unsigned long node_start_pfn, unsigned long *zholes_size)
4220 pg_data_t *pgdat = NODE_DATA(nid);
4222 pgdat->node_id = nid;
4223 pgdat->node_start_pfn = node_start_pfn;
4224 calculate_node_totalpages(pgdat, zones_size, zholes_size);
4226 alloc_node_mem_map(pgdat);
4227 #ifdef CONFIG_FLAT_NODE_MEM_MAP
4228 printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
4229 nid, (unsigned long)pgdat,
4230 (unsigned long)pgdat->node_mem_map);
4233 free_area_init_core(pgdat, zones_size, zholes_size);
4236 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
4238 #if MAX_NUMNODES > 1
4240 * Figure out the number of possible node ids.
4242 static void __init setup_nr_node_ids(void)
4245 unsigned int highest = 0;
4247 for_each_node_mask(node, node_possible_map)
4249 nr_node_ids = highest + 1;
4252 static inline void setup_nr_node_ids(void)
4258 * add_active_range - Register a range of PFNs backed by physical memory
4259 * @nid: The node ID the range resides on
4260 * @start_pfn: The start PFN of the available physical memory
4261 * @end_pfn: The end PFN of the available physical memory
4263 * These ranges are stored in an early_node_map[] and later used by
4264 * free_area_init_nodes() to calculate zone sizes and holes. If the
4265 * range spans a memory hole, it is up to the architecture to ensure
4266 * the memory is not freed by the bootmem allocator. If possible
4267 * the range being registered will be merged with existing ranges.
4269 void __init add_active_range(unsigned int nid, unsigned long start_pfn,
4270 unsigned long end_pfn)
4274 mminit_dprintk(MMINIT_TRACE, "memory_register",
4275 "Entering add_active_range(%d, %#lx, %#lx) "
4276 "%d entries of %d used\n",
4277 nid, start_pfn, end_pfn,
4278 nr_nodemap_entries, MAX_ACTIVE_REGIONS);
4280 mminit_validate_memmodel_limits(&start_pfn, &end_pfn);
4282 /* Merge with existing active regions if possible */
4283 for (i = 0; i < nr_nodemap_entries; i++) {
4284 if (early_node_map[i].nid != nid)
4287 /* Skip if an existing region covers this new one */
4288 if (start_pfn >= early_node_map[i].start_pfn &&
4289 end_pfn <= early_node_map[i].end_pfn)
4292 /* Merge forward if suitable */
4293 if (start_pfn <= early_node_map[i].end_pfn &&
4294 end_pfn > early_node_map[i].end_pfn) {
4295 early_node_map[i].end_pfn = end_pfn;
4299 /* Merge backward if suitable */
4300 if (start_pfn < early_node_map[i].start_pfn &&
4301 end_pfn >= early_node_map[i].start_pfn) {
4302 early_node_map[i].start_pfn = start_pfn;
4307 /* Check that early_node_map is large enough */
4308 if (i >= MAX_ACTIVE_REGIONS) {
4309 printk(KERN_CRIT "More than %d memory regions, truncating\n",
4310 MAX_ACTIVE_REGIONS);
4314 early_node_map[i].nid = nid;
4315 early_node_map[i].start_pfn = start_pfn;
4316 early_node_map[i].end_pfn = end_pfn;
4317 nr_nodemap_entries = i + 1;
4321 * remove_active_range - Shrink an existing registered range of PFNs
4322 * @nid: The node id the range is on that should be shrunk
4323 * @start_pfn: The new PFN of the range
4324 * @end_pfn: The new PFN of the range
4326 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
4327 * The map is kept near the end physical page range that has already been
4328 * registered. This function allows an arch to shrink an existing registered
4331 void __init remove_active_range(unsigned int nid, unsigned long start_pfn,
4332 unsigned long end_pfn)
4337 printk(KERN_DEBUG "remove_active_range (%d, %lu, %lu)\n",
4338 nid, start_pfn, end_pfn);
4340 /* Find the old active region end and shrink */
4341 for_each_active_range_index_in_nid(i, nid) {
4342 if (early_node_map[i].start_pfn >= start_pfn &&
4343 early_node_map[i].end_pfn <= end_pfn) {
4345 early_node_map[i].start_pfn = 0;
4346 early_node_map[i].end_pfn = 0;
4350 if (early_node_map[i].start_pfn < start_pfn &&
4351 early_node_map[i].end_pfn > start_pfn) {
4352 unsigned long temp_end_pfn = early_node_map[i].end_pfn;
4353 early_node_map[i].end_pfn = start_pfn;
4354 if (temp_end_pfn > end_pfn)
4355 add_active_range(nid, end_pfn, temp_end_pfn);
4358 if (early_node_map[i].start_pfn >= start_pfn &&
4359 early_node_map[i].end_pfn > end_pfn &&
4360 early_node_map[i].start_pfn < end_pfn) {
4361 early_node_map[i].start_pfn = end_pfn;
4369 /* remove the blank ones */
4370 for (i = nr_nodemap_entries - 1; i > 0; i--) {
4371 if (early_node_map[i].nid != nid)
4373 if (early_node_map[i].end_pfn)
4375 /* we found it, get rid of it */
4376 for (j = i; j < nr_nodemap_entries - 1; j++)
4377 memcpy(&early_node_map[j], &early_node_map[j+1],
4378 sizeof(early_node_map[j]));
4379 j = nr_nodemap_entries - 1;
4380 memset(&early_node_map[j], 0, sizeof(early_node_map[j]));
4381 nr_nodemap_entries--;
4386 * remove_all_active_ranges - Remove all currently registered regions
4388 * During discovery, it may be found that a table like SRAT is invalid
4389 * and an alternative discovery method must be used. This function removes
4390 * all currently registered regions.
4392 void __init remove_all_active_ranges(void)
4394 memset(early_node_map, 0, sizeof(early_node_map));
4395 nr_nodemap_entries = 0;
4398 /* Compare two active node_active_regions */
4399 static int __init cmp_node_active_region(const void *a, const void *b)
4401 struct node_active_region *arange = (struct node_active_region *)a;
4402 struct node_active_region *brange = (struct node_active_region *)b;
4404 /* Done this way to avoid overflows */
4405 if (arange->start_pfn > brange->start_pfn)
4407 if (arange->start_pfn < brange->start_pfn)
4413 /* sort the node_map by start_pfn */
4414 void __init sort_node_map(void)
4416 sort(early_node_map, (size_t)nr_nodemap_entries,
4417 sizeof(struct node_active_region),
4418 cmp_node_active_region, NULL);
4421 /* Find the lowest pfn for a node */
4422 static unsigned long __init find_min_pfn_for_node(int nid)
4425 unsigned long min_pfn = ULONG_MAX;
4427 /* Assuming a sorted map, the first range found has the starting pfn */
4428 for_each_active_range_index_in_nid(i, nid)
4429 min_pfn = min(min_pfn, early_node_map[i].start_pfn);
4431 if (min_pfn == ULONG_MAX) {
4433 "Could not find start_pfn for node %d\n", nid);
4441 * find_min_pfn_with_active_regions - Find the minimum PFN registered
4443 * It returns the minimum PFN based on information provided via
4444 * add_active_range().
4446 unsigned long __init find_min_pfn_with_active_regions(void)
4448 return find_min_pfn_for_node(MAX_NUMNODES);
4452 * early_calculate_totalpages()
4453 * Sum pages in active regions for movable zone.
4454 * Populate N_HIGH_MEMORY for calculating usable_nodes.
4456 static unsigned long __init early_calculate_totalpages(void)
4459 unsigned long totalpages = 0;
4461 for (i = 0; i < nr_nodemap_entries; i++) {
4462 unsigned long pages = early_node_map[i].end_pfn -
4463 early_node_map[i].start_pfn;
4464 totalpages += pages;
4466 node_set_state(early_node_map[i].nid, N_HIGH_MEMORY);
4472 * Find the PFN the Movable zone begins in each node. Kernel memory
4473 * is spread evenly between nodes as long as the nodes have enough
4474 * memory. When they don't, some nodes will have more kernelcore than
4477 static void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn)
4480 unsigned long usable_startpfn;
4481 unsigned long kernelcore_node, kernelcore_remaining;
4482 /* save the state before borrow the nodemask */
4483 nodemask_t saved_node_state = node_states[N_HIGH_MEMORY];
4484 unsigned long totalpages = early_calculate_totalpages();
4485 int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]);
4488 * If movablecore was specified, calculate what size of
4489 * kernelcore that corresponds so that memory usable for
4490 * any allocation type is evenly spread. If both kernelcore
4491 * and movablecore are specified, then the value of kernelcore
4492 * will be used for required_kernelcore if it's greater than
4493 * what movablecore would have allowed.
4495 if (required_movablecore) {
4496 unsigned long corepages;
4499 * Round-up so that ZONE_MOVABLE is at least as large as what
4500 * was requested by the user
4502 required_movablecore =
4503 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
4504 corepages = totalpages - required_movablecore;
4506 required_kernelcore = max(required_kernelcore, corepages);
4509 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
4510 if (!required_kernelcore)
4513 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
4514 find_usable_zone_for_movable();
4515 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
4518 /* Spread kernelcore memory as evenly as possible throughout nodes */
4519 kernelcore_node = required_kernelcore / usable_nodes;
4520 for_each_node_state(nid, N_HIGH_MEMORY) {
4522 * Recalculate kernelcore_node if the division per node
4523 * now exceeds what is necessary to satisfy the requested
4524 * amount of memory for the kernel
4526 if (required_kernelcore < kernelcore_node)
4527 kernelcore_node = required_kernelcore / usable_nodes;
4530 * As the map is walked, we track how much memory is usable
4531 * by the kernel using kernelcore_remaining. When it is
4532 * 0, the rest of the node is usable by ZONE_MOVABLE
4534 kernelcore_remaining = kernelcore_node;
4536 /* Go through each range of PFNs within this node */
4537 for_each_active_range_index_in_nid(i, nid) {
4538 unsigned long start_pfn, end_pfn;
4539 unsigned long size_pages;
4541 start_pfn = max(early_node_map[i].start_pfn,
4542 zone_movable_pfn[nid]);
4543 end_pfn = early_node_map[i].end_pfn;
4544 if (start_pfn >= end_pfn)
4547 /* Account for what is only usable for kernelcore */
4548 if (start_pfn < usable_startpfn) {
4549 unsigned long kernel_pages;
4550 kernel_pages = min(end_pfn, usable_startpfn)
4553 kernelcore_remaining -= min(kernel_pages,
4554 kernelcore_remaining);
4555 required_kernelcore -= min(kernel_pages,
4556 required_kernelcore);
4558 /* Continue if range is now fully accounted */
4559 if (end_pfn <= usable_startpfn) {
4562 * Push zone_movable_pfn to the end so
4563 * that if we have to rebalance
4564 * kernelcore across nodes, we will
4565 * not double account here
4567 zone_movable_pfn[nid] = end_pfn;
4570 start_pfn = usable_startpfn;
4574 * The usable PFN range for ZONE_MOVABLE is from
4575 * start_pfn->end_pfn. Calculate size_pages as the
4576 * number of pages used as kernelcore
4578 size_pages = end_pfn - start_pfn;
4579 if (size_pages > kernelcore_remaining)
4580 size_pages = kernelcore_remaining;
4581 zone_movable_pfn[nid] = start_pfn + size_pages;
4584 * Some kernelcore has been met, update counts and
4585 * break if the kernelcore for this node has been
4588 required_kernelcore -= min(required_kernelcore,
4590 kernelcore_remaining -= size_pages;
4591 if (!kernelcore_remaining)
4597 * If there is still required_kernelcore, we do another pass with one
4598 * less node in the count. This will push zone_movable_pfn[nid] further
4599 * along on the nodes that still have memory until kernelcore is
4603 if (usable_nodes && required_kernelcore > usable_nodes)
4606 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
4607 for (nid = 0; nid < MAX_NUMNODES; nid++)
4608 zone_movable_pfn[nid] =
4609 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
4612 /* restore the node_state */
4613 node_states[N_HIGH_MEMORY] = saved_node_state;
4616 /* Any regular memory on that node ? */
4617 static void check_for_regular_memory(pg_data_t *pgdat)
4619 #ifdef CONFIG_HIGHMEM
4620 enum zone_type zone_type;
4622 for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) {
4623 struct zone *zone = &pgdat->node_zones[zone_type];
4624 if (zone->present_pages)
4625 node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY);
4631 * free_area_init_nodes - Initialise all pg_data_t and zone data
4632 * @max_zone_pfn: an array of max PFNs for each zone
4634 * This will call free_area_init_node() for each active node in the system.
4635 * Using the page ranges provided by add_active_range(), the size of each
4636 * zone in each node and their holes is calculated. If the maximum PFN
4637 * between two adjacent zones match, it is assumed that the zone is empty.
4638 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
4639 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
4640 * starts where the previous one ended. For example, ZONE_DMA32 starts
4641 * at arch_max_dma_pfn.
4643 void __init free_area_init_nodes(unsigned long *max_zone_pfn)
4648 /* Sort early_node_map as initialisation assumes it is sorted */
4651 /* Record where the zone boundaries are */
4652 memset(arch_zone_lowest_possible_pfn, 0,
4653 sizeof(arch_zone_lowest_possible_pfn));
4654 memset(arch_zone_highest_possible_pfn, 0,
4655 sizeof(arch_zone_highest_possible_pfn));
4656 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
4657 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
4658 for (i = 1; i < MAX_NR_ZONES; i++) {
4659 if (i == ZONE_MOVABLE)
4661 arch_zone_lowest_possible_pfn[i] =
4662 arch_zone_highest_possible_pfn[i-1];
4663 arch_zone_highest_possible_pfn[i] =
4664 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
4666 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
4667 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
4669 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
4670 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
4671 find_zone_movable_pfns_for_nodes(zone_movable_pfn);
4673 /* Print out the zone ranges */
4674 printk("Zone PFN ranges:\n");
4675 for (i = 0; i < MAX_NR_ZONES; i++) {
4676 if (i == ZONE_MOVABLE)
4678 printk(" %-8s ", zone_names[i]);
4679 if (arch_zone_lowest_possible_pfn[i] ==
4680 arch_zone_highest_possible_pfn[i])
4683 printk("%0#10lx -> %0#10lx\n",
4684 arch_zone_lowest_possible_pfn[i],
4685 arch_zone_highest_possible_pfn[i]);
4688 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
4689 printk("Movable zone start PFN for each node\n");
4690 for (i = 0; i < MAX_NUMNODES; i++) {
4691 if (zone_movable_pfn[i])
4692 printk(" Node %d: %lu\n", i, zone_movable_pfn[i]);
4695 /* Print out the early_node_map[] */
4696 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries);
4697 for (i = 0; i < nr_nodemap_entries; i++)
4698 printk(" %3d: %0#10lx -> %0#10lx\n", early_node_map[i].nid,
4699 early_node_map[i].start_pfn,
4700 early_node_map[i].end_pfn);
4702 /* Initialise every node */
4703 mminit_verify_pageflags_layout();
4704 setup_nr_node_ids();
4705 for_each_online_node(nid) {
4706 pg_data_t *pgdat = NODE_DATA(nid);
4707 free_area_init_node(nid, NULL,
4708 find_min_pfn_for_node(nid), NULL);
4710 /* Any memory on that node */
4711 if (pgdat->node_present_pages)
4712 node_set_state(nid, N_HIGH_MEMORY);
4713 check_for_regular_memory(pgdat);
4717 static int __init cmdline_parse_core(char *p, unsigned long *core)
4719 unsigned long long coremem;
4723 coremem = memparse(p, &p);
4724 *core = coremem >> PAGE_SHIFT;
4726 /* Paranoid check that UL is enough for the coremem value */
4727 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
4733 * kernelcore=size sets the amount of memory for use for allocations that
4734 * cannot be reclaimed or migrated.
4736 static int __init cmdline_parse_kernelcore(char *p)
4738 return cmdline_parse_core(p, &required_kernelcore);
4742 * movablecore=size sets the amount of memory for use for allocations that
4743 * can be reclaimed or migrated.
4745 static int __init cmdline_parse_movablecore(char *p)
4747 return cmdline_parse_core(p, &required_movablecore);
4750 early_param("kernelcore", cmdline_parse_kernelcore);
4751 early_param("movablecore", cmdline_parse_movablecore);
4753 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4756 * set_dma_reserve - set the specified number of pages reserved in the first zone
4757 * @new_dma_reserve: The number of pages to mark reserved
4759 * The per-cpu batchsize and zone watermarks are determined by present_pages.
4760 * In the DMA zone, a significant percentage may be consumed by kernel image
4761 * and other unfreeable allocations which can skew the watermarks badly. This
4762 * function may optionally be used to account for unfreeable pages in the
4763 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
4764 * smaller per-cpu batchsize.
4766 void __init set_dma_reserve(unsigned long new_dma_reserve)
4768 dma_reserve = new_dma_reserve;
4771 #ifndef CONFIG_NEED_MULTIPLE_NODES
4772 struct pglist_data __refdata contig_page_data = {
4773 #ifndef CONFIG_NO_BOOTMEM
4774 .bdata = &bootmem_node_data[0]
4777 EXPORT_SYMBOL(contig_page_data);
4780 void __init free_area_init(unsigned long *zones_size)
4782 free_area_init_node(0, zones_size,
4783 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
4786 static int page_alloc_cpu_notify(struct notifier_block *self,
4787 unsigned long action, void *hcpu)
4789 int cpu = (unsigned long)hcpu;
4791 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
4795 * Spill the event counters of the dead processor
4796 * into the current processors event counters.
4797 * This artificially elevates the count of the current
4800 vm_events_fold_cpu(cpu);
4803 * Zero the differential counters of the dead processor
4804 * so that the vm statistics are consistent.
4806 * This is only okay since the processor is dead and cannot
4807 * race with what we are doing.
4809 refresh_cpu_vm_stats(cpu);
4814 void __init page_alloc_init(void)
4816 hotcpu_notifier(page_alloc_cpu_notify, 0);
4820 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
4821 * or min_free_kbytes changes.
4823 static void calculate_totalreserve_pages(void)
4825 struct pglist_data *pgdat;
4826 unsigned long reserve_pages = 0;
4827 enum zone_type i, j;
4829 for_each_online_pgdat(pgdat) {
4830 for (i = 0; i < MAX_NR_ZONES; i++) {
4831 struct zone *zone = pgdat->node_zones + i;
4832 unsigned long max = 0;
4834 /* Find valid and maximum lowmem_reserve in the zone */
4835 for (j = i; j < MAX_NR_ZONES; j++) {
4836 if (zone->lowmem_reserve[j] > max)
4837 max = zone->lowmem_reserve[j];
4840 /* we treat the high watermark as reserved pages. */
4841 max += high_wmark_pages(zone);
4843 if (max > zone->present_pages)
4844 max = zone->present_pages;
4845 reserve_pages += max;
4848 totalreserve_pages = reserve_pages;
4852 * setup_per_zone_lowmem_reserve - called whenever
4853 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
4854 * has a correct pages reserved value, so an adequate number of
4855 * pages are left in the zone after a successful __alloc_pages().
4857 static void setup_per_zone_lowmem_reserve(void)
4859 struct pglist_data *pgdat;
4860 enum zone_type j, idx;
4862 for_each_online_pgdat(pgdat) {
4863 for (j = 0; j < MAX_NR_ZONES; j++) {
4864 struct zone *zone = pgdat->node_zones + j;
4865 unsigned long present_pages = zone->present_pages;
4867 zone->lowmem_reserve[j] = 0;
4871 struct zone *lower_zone;
4875 if (sysctl_lowmem_reserve_ratio[idx] < 1)
4876 sysctl_lowmem_reserve_ratio[idx] = 1;
4878 lower_zone = pgdat->node_zones + idx;
4879 lower_zone->lowmem_reserve[j] = present_pages /
4880 sysctl_lowmem_reserve_ratio[idx];
4881 present_pages += lower_zone->present_pages;
4886 /* update totalreserve_pages */
4887 calculate_totalreserve_pages();
4891 * setup_per_zone_wmarks - called when min_free_kbytes changes
4892 * or when memory is hot-{added|removed}
4894 * Ensures that the watermark[min,low,high] values for each zone are set
4895 * correctly with respect to min_free_kbytes.
4897 void setup_per_zone_wmarks(void)
4899 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
4900 unsigned long lowmem_pages = 0;
4902 unsigned long flags;
4904 /* Calculate total number of !ZONE_HIGHMEM pages */
4905 for_each_zone(zone) {
4906 if (!is_highmem(zone))
4907 lowmem_pages += zone->present_pages;
4910 for_each_zone(zone) {
4913 spin_lock_irqsave(&zone->lock, flags);
4914 tmp = (u64)pages_min * zone->present_pages;
4915 do_div(tmp, lowmem_pages);
4916 if (is_highmem(zone)) {
4918 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
4919 * need highmem pages, so cap pages_min to a small
4922 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
4923 * deltas controls asynch page reclaim, and so should
4924 * not be capped for highmem.
4928 min_pages = zone->present_pages / 1024;
4929 if (min_pages < SWAP_CLUSTER_MAX)
4930 min_pages = SWAP_CLUSTER_MAX;
4931 if (min_pages > 128)
4933 zone->watermark[WMARK_MIN] = min_pages;
4936 * If it's a lowmem zone, reserve a number of pages
4937 * proportionate to the zone's size.
4939 zone->watermark[WMARK_MIN] = tmp;
4942 zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + (tmp >> 2);
4943 zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1);
4944 setup_zone_migrate_reserve(zone);
4945 spin_unlock_irqrestore(&zone->lock, flags);
4948 /* update totalreserve_pages */
4949 calculate_totalreserve_pages();
4953 * The inactive anon list should be small enough that the VM never has to
4954 * do too much work, but large enough that each inactive page has a chance
4955 * to be referenced again before it is swapped out.
4957 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
4958 * INACTIVE_ANON pages on this zone's LRU, maintained by the
4959 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
4960 * the anonymous pages are kept on the inactive list.
4963 * memory ratio inactive anon
4964 * -------------------------------------
4973 void calculate_zone_inactive_ratio(struct zone *zone)
4975 unsigned int gb, ratio;
4977 /* Zone size in gigabytes */
4978 gb = zone->present_pages >> (30 - PAGE_SHIFT);
4980 ratio = int_sqrt(10 * gb);
4984 zone->inactive_ratio = ratio;
4987 static void __init setup_per_zone_inactive_ratio(void)
4992 calculate_zone_inactive_ratio(zone);
4996 * Initialise min_free_kbytes.
4998 * For small machines we want it small (128k min). For large machines
4999 * we want it large (64MB max). But it is not linear, because network
5000 * bandwidth does not increase linearly with machine size. We use
5002 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
5003 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
5019 static int __init init_per_zone_wmark_min(void)
5021 unsigned long lowmem_kbytes;
5023 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
5025 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
5026 if (min_free_kbytes < 128)
5027 min_free_kbytes = 128;
5028 if (min_free_kbytes > 65536)
5029 min_free_kbytes = 65536;
5030 setup_per_zone_wmarks();
5031 setup_per_zone_lowmem_reserve();
5032 setup_per_zone_inactive_ratio();
5035 module_init(init_per_zone_wmark_min)
5038 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
5039 * that we can call two helper functions whenever min_free_kbytes
5042 int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
5043 void __user *buffer, size_t *length, loff_t *ppos)
5045 proc_dointvec(table, write, buffer, length, ppos);
5047 setup_per_zone_wmarks();
5052 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
5053 void __user *buffer, size_t *length, loff_t *ppos)
5058 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
5063 zone->min_unmapped_pages = (zone->present_pages *
5064 sysctl_min_unmapped_ratio) / 100;
5068 int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
5069 void __user *buffer, size_t *length, loff_t *ppos)
5074 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
5079 zone->min_slab_pages = (zone->present_pages *
5080 sysctl_min_slab_ratio) / 100;
5086 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
5087 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
5088 * whenever sysctl_lowmem_reserve_ratio changes.
5090 * The reserve ratio obviously has absolutely no relation with the
5091 * minimum watermarks. The lowmem reserve ratio can only make sense
5092 * if in function of the boot time zone sizes.
5094 int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
5095 void __user *buffer, size_t *length, loff_t *ppos)
5097 proc_dointvec_minmax(table, write, buffer, length, ppos);
5098 setup_per_zone_lowmem_reserve();
5103 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
5104 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
5105 * can have before it gets flushed back to buddy allocator.
5108 int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
5109 void __user *buffer, size_t *length, loff_t *ppos)
5115 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
5116 if (!write || (ret == -EINVAL))
5118 for_each_populated_zone(zone) {
5119 for_each_possible_cpu(cpu) {
5121 high = zone->present_pages / percpu_pagelist_fraction;
5122 setup_pagelist_highmark(
5123 per_cpu_ptr(zone->pageset, cpu), high);
5129 int hashdist = HASHDIST_DEFAULT;
5132 static int __init set_hashdist(char *str)
5136 hashdist = simple_strtoul(str, &str, 0);
5139 __setup("hashdist=", set_hashdist);
5143 * allocate a large system hash table from bootmem
5144 * - it is assumed that the hash table must contain an exact power-of-2
5145 * quantity of entries
5146 * - limit is the number of hash buckets, not the total allocation size
5148 void *__init alloc_large_system_hash(const char *tablename,
5149 unsigned long bucketsize,
5150 unsigned long numentries,
5153 unsigned int *_hash_shift,
5154 unsigned int *_hash_mask,
5155 unsigned long limit)
5157 unsigned long long max = limit;
5158 unsigned long log2qty, size;
5161 /* allow the kernel cmdline to have a say */
5163 /* round applicable memory size up to nearest megabyte */
5164 numentries = nr_kernel_pages;
5165 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
5166 numentries >>= 20 - PAGE_SHIFT;
5167 numentries <<= 20 - PAGE_SHIFT;
5169 /* limit to 1 bucket per 2^scale bytes of low memory */
5170 if (scale > PAGE_SHIFT)
5171 numentries >>= (scale - PAGE_SHIFT);
5173 numentries <<= (PAGE_SHIFT - scale);
5175 /* Make sure we've got at least a 0-order allocation.. */
5176 if (unlikely(flags & HASH_SMALL)) {
5177 /* Makes no sense without HASH_EARLY */
5178 WARN_ON(!(flags & HASH_EARLY));
5179 if (!(numentries >> *_hash_shift)) {
5180 numentries = 1UL << *_hash_shift;
5181 BUG_ON(!numentries);
5183 } else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
5184 numentries = PAGE_SIZE / bucketsize;
5186 numentries = roundup_pow_of_two(numentries);
5188 /* limit allocation size to 1/16 total memory by default */
5190 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
5191 do_div(max, bucketsize);
5194 if (numentries > max)
5197 log2qty = ilog2(numentries);
5200 size = bucketsize << log2qty;
5201 if (flags & HASH_EARLY)
5202 table = alloc_bootmem_nopanic(size);
5204 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
5207 * If bucketsize is not a power-of-two, we may free
5208 * some pages at the end of hash table which
5209 * alloc_pages_exact() automatically does
5211 if (get_order(size) < MAX_ORDER) {
5212 table = alloc_pages_exact(size, GFP_ATOMIC);
5213 kmemleak_alloc(table, size, 1, GFP_ATOMIC);
5216 } while (!table && size > PAGE_SIZE && --log2qty);
5219 panic("Failed to allocate %s hash table\n", tablename);
5221 printk(KERN_INFO "%s hash table entries: %ld (order: %d, %lu bytes)\n",
5224 ilog2(size) - PAGE_SHIFT,
5228 *_hash_shift = log2qty;
5230 *_hash_mask = (1 << log2qty) - 1;
5235 /* Return a pointer to the bitmap storing bits affecting a block of pages */
5236 static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
5239 #ifdef CONFIG_SPARSEMEM
5240 return __pfn_to_section(pfn)->pageblock_flags;
5242 return zone->pageblock_flags;
5243 #endif /* CONFIG_SPARSEMEM */
5246 static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
5248 #ifdef CONFIG_SPARSEMEM
5249 pfn &= (PAGES_PER_SECTION-1);
5250 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
5252 pfn = pfn - zone->zone_start_pfn;
5253 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
5254 #endif /* CONFIG_SPARSEMEM */
5258 * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
5259 * @page: The page within the block of interest
5260 * @start_bitidx: The first bit of interest to retrieve
5261 * @end_bitidx: The last bit of interest
5262 * returns pageblock_bits flags
5264 unsigned long get_pageblock_flags_group(struct page *page,
5265 int start_bitidx, int end_bitidx)
5268 unsigned long *bitmap;
5269 unsigned long pfn, bitidx;
5270 unsigned long flags = 0;
5271 unsigned long value = 1;
5273 zone = page_zone(page);
5274 pfn = page_to_pfn(page);
5275 bitmap = get_pageblock_bitmap(zone, pfn);
5276 bitidx = pfn_to_bitidx(zone, pfn);
5278 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
5279 if (test_bit(bitidx + start_bitidx, bitmap))
5286 * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
5287 * @page: The page within the block of interest
5288 * @start_bitidx: The first bit of interest
5289 * @end_bitidx: The last bit of interest
5290 * @flags: The flags to set
5292 void set_pageblock_flags_group(struct page *page, unsigned long flags,
5293 int start_bitidx, int end_bitidx)
5296 unsigned long *bitmap;
5297 unsigned long pfn, bitidx;
5298 unsigned long value = 1;
5300 zone = page_zone(page);
5301 pfn = page_to_pfn(page);
5302 bitmap = get_pageblock_bitmap(zone, pfn);
5303 bitidx = pfn_to_bitidx(zone, pfn);
5304 VM_BUG_ON(pfn < zone->zone_start_pfn);
5305 VM_BUG_ON(pfn >= zone->zone_start_pfn + zone->spanned_pages);
5307 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
5309 __set_bit(bitidx + start_bitidx, bitmap);
5311 __clear_bit(bitidx + start_bitidx, bitmap);
5315 * This is designed as sub function...plz see page_isolation.c also.
5316 * set/clear page block's type to be ISOLATE.
5317 * page allocater never alloc memory from ISOLATE block.
5321 __count_immobile_pages(struct zone *zone, struct page *page, int count)
5323 unsigned long pfn, iter, found;
5325 * For avoiding noise data, lru_add_drain_all() should be called
5326 * If ZONE_MOVABLE, the zone never contains immobile pages
5328 if (zone_idx(zone) == ZONE_MOVABLE)
5331 if (get_pageblock_migratetype(page) == MIGRATE_MOVABLE)
5334 pfn = page_to_pfn(page);
5335 for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) {
5336 unsigned long check = pfn + iter;
5338 if (!pfn_valid_within(check)) {
5342 page = pfn_to_page(check);
5343 if (!page_count(page)) {
5344 if (PageBuddy(page))
5345 iter += (1 << page_order(page)) - 1;
5351 * If there are RECLAIMABLE pages, we need to check it.
5352 * But now, memory offline itself doesn't call shrink_slab()
5353 * and it still to be fixed.
5356 * If the page is not RAM, page_count()should be 0.
5357 * we don't need more check. This is an _used_ not-movable page.
5359 * The problematic thing here is PG_reserved pages. PG_reserved
5360 * is set to both of a memory hole page and a _used_ kernel
5369 bool is_pageblock_removable_nolock(struct page *page)
5371 struct zone *zone = page_zone(page);
5372 return __count_immobile_pages(zone, page, 0);
5375 int set_migratetype_isolate(struct page *page)
5378 unsigned long flags, pfn;
5379 struct memory_isolate_notify arg;
5384 zone = page_zone(page);
5385 zone_idx = zone_idx(zone);
5387 spin_lock_irqsave(&zone->lock, flags);
5389 pfn = page_to_pfn(page);
5390 arg.start_pfn = pfn;
5391 arg.nr_pages = pageblock_nr_pages;
5392 arg.pages_found = 0;
5395 * It may be possible to isolate a pageblock even if the
5396 * migratetype is not MIGRATE_MOVABLE. The memory isolation
5397 * notifier chain is used by balloon drivers to return the
5398 * number of pages in a range that are held by the balloon
5399 * driver to shrink memory. If all the pages are accounted for
5400 * by balloons, are free, or on the LRU, isolation can continue.
5401 * Later, for example, when memory hotplug notifier runs, these
5402 * pages reported as "can be isolated" should be isolated(freed)
5403 * by the balloon driver through the memory notifier chain.
5405 notifier_ret = memory_isolate_notify(MEM_ISOLATE_COUNT, &arg);
5406 notifier_ret = notifier_to_errno(notifier_ret);
5410 * FIXME: Now, memory hotplug doesn't call shrink_slab() by itself.
5411 * We just check MOVABLE pages.
5413 if (__count_immobile_pages(zone, page, arg.pages_found))
5417 * immobile means "not-on-lru" paes. If immobile is larger than
5418 * removable-by-driver pages reported by notifier, we'll fail.
5423 set_pageblock_migratetype(page, MIGRATE_ISOLATE);
5424 move_freepages_block(zone, page, MIGRATE_ISOLATE);
5427 spin_unlock_irqrestore(&zone->lock, flags);
5433 void unset_migratetype_isolate(struct page *page)
5436 unsigned long flags;
5437 zone = page_zone(page);
5438 spin_lock_irqsave(&zone->lock, flags);
5439 if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE)
5441 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
5442 move_freepages_block(zone, page, MIGRATE_MOVABLE);
5444 spin_unlock_irqrestore(&zone->lock, flags);
5447 #ifdef CONFIG_MEMORY_HOTREMOVE
5449 * All pages in the range must be isolated before calling this.
5452 __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
5458 unsigned long flags;
5459 /* find the first valid pfn */
5460 for (pfn = start_pfn; pfn < end_pfn; pfn++)
5465 zone = page_zone(pfn_to_page(pfn));
5466 spin_lock_irqsave(&zone->lock, flags);
5468 while (pfn < end_pfn) {
5469 if (!pfn_valid(pfn)) {
5473 page = pfn_to_page(pfn);
5474 BUG_ON(page_count(page));
5475 BUG_ON(!PageBuddy(page));
5476 order = page_order(page);
5477 #ifdef CONFIG_DEBUG_VM
5478 printk(KERN_INFO "remove from free list %lx %d %lx\n",
5479 pfn, 1 << order, end_pfn);
5481 list_del(&page->lru);
5482 rmv_page_order(page);
5483 zone->free_area[order].nr_free--;
5484 __mod_zone_page_state(zone, NR_FREE_PAGES,
5486 for (i = 0; i < (1 << order); i++)
5487 SetPageReserved((page+i));
5488 pfn += (1 << order);
5490 spin_unlock_irqrestore(&zone->lock, flags);
5494 #ifdef CONFIG_MEMORY_FAILURE
5495 bool is_free_buddy_page(struct page *page)
5497 struct zone *zone = page_zone(page);
5498 unsigned long pfn = page_to_pfn(page);
5499 unsigned long flags;
5502 spin_lock_irqsave(&zone->lock, flags);
5503 for (order = 0; order < MAX_ORDER; order++) {
5504 struct page *page_head = page - (pfn & ((1 << order) - 1));
5506 if (PageBuddy(page_head) && page_order(page_head) >= order)
5509 spin_unlock_irqrestore(&zone->lock, flags);
5511 return order < MAX_ORDER;
5515 static struct trace_print_flags pageflag_names[] = {
5516 {1UL << PG_locked, "locked" },
5517 {1UL << PG_error, "error" },
5518 {1UL << PG_referenced, "referenced" },
5519 {1UL << PG_uptodate, "uptodate" },
5520 {1UL << PG_dirty, "dirty" },
5521 {1UL << PG_lru, "lru" },
5522 {1UL << PG_active, "active" },
5523 {1UL << PG_slab, "slab" },
5524 {1UL << PG_owner_priv_1, "owner_priv_1" },
5525 {1UL << PG_arch_1, "arch_1" },
5526 {1UL << PG_reserved, "reserved" },
5527 {1UL << PG_private, "private" },
5528 {1UL << PG_private_2, "private_2" },
5529 {1UL << PG_writeback, "writeback" },
5530 #ifdef CONFIG_PAGEFLAGS_EXTENDED
5531 {1UL << PG_head, "head" },
5532 {1UL << PG_tail, "tail" },
5534 {1UL << PG_compound, "compound" },
5536 {1UL << PG_swapcache, "swapcache" },
5537 {1UL << PG_mappedtodisk, "mappedtodisk" },
5538 {1UL << PG_reclaim, "reclaim" },
5539 {1UL << PG_buddy, "buddy" },
5540 {1UL << PG_swapbacked, "swapbacked" },
5541 {1UL << PG_unevictable, "unevictable" },
5543 {1UL << PG_mlocked, "mlocked" },
5545 #ifdef CONFIG_ARCH_USES_PG_UNCACHED
5546 {1UL << PG_uncached, "uncached" },
5548 #ifdef CONFIG_MEMORY_FAILURE
5549 {1UL << PG_hwpoison, "hwpoison" },
5554 static void dump_page_flags(unsigned long flags)
5556 const char *delim = "";
5560 printk(KERN_ALERT "page flags: %#lx(", flags);
5562 /* remove zone id */
5563 flags &= (1UL << NR_PAGEFLAGS) - 1;
5565 for (i = 0; pageflag_names[i].name && flags; i++) {
5567 mask = pageflag_names[i].mask;
5568 if ((flags & mask) != mask)
5572 printk("%s%s", delim, pageflag_names[i].name);
5576 /* check for left over flags */
5578 printk("%s%#lx", delim, flags);
5583 void dump_page(struct page *page)
5586 "page:%p count:%d mapcount:%d mapping:%p index:%#lx\n",
5587 page, page_count(page), page_mapcount(page),
5588 page->mapping, page->index);
5589 dump_page_flags(page->flags);