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/ratelimit.h>
34 #include <linux/oom.h>
35 #include <linux/notifier.h>
36 #include <linux/topology.h>
37 #include <linux/sysctl.h>
38 #include <linux/cpu.h>
39 #include <linux/cpuset.h>
40 #include <linux/memory_hotplug.h>
41 #include <linux/nodemask.h>
42 #include <linux/vmalloc.h>
43 #include <linux/vmstat.h>
44 #include <linux/mempolicy.h>
45 #include <linux/stop_machine.h>
46 #include <linux/sort.h>
47 #include <linux/pfn.h>
48 #include <linux/backing-dev.h>
49 #include <linux/fault-inject.h>
50 #include <linux/page-isolation.h>
51 #include <linux/page_cgroup.h>
52 #include <linux/debugobjects.h>
53 #include <linux/kmemleak.h>
54 #include <linux/memory.h>
55 #include <linux/compaction.h>
56 #include <trace/events/kmem.h>
57 #include <linux/ftrace_event.h>
58 #include <linux/memcontrol.h>
59 #include <linux/prefetch.h>
61 #include <asm/tlbflush.h>
62 #include <asm/div64.h>
65 #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
66 DEFINE_PER_CPU(int, numa_node);
67 EXPORT_PER_CPU_SYMBOL(numa_node);
70 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
72 * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
73 * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
74 * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
75 * defined in <linux/topology.h>.
77 DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */
78 EXPORT_PER_CPU_SYMBOL(_numa_mem_);
82 * Array of node states.
84 nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
85 [N_POSSIBLE] = NODE_MASK_ALL,
86 [N_ONLINE] = { { [0] = 1UL } },
88 [N_NORMAL_MEMORY] = { { [0] = 1UL } },
90 [N_HIGH_MEMORY] = { { [0] = 1UL } },
92 [N_CPU] = { { [0] = 1UL } },
95 EXPORT_SYMBOL(node_states);
97 unsigned long totalram_pages __read_mostly;
98 unsigned long totalreserve_pages __read_mostly;
99 int percpu_pagelist_fraction;
100 gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK;
102 #ifdef CONFIG_PM_SLEEP
104 * The following functions are used by the suspend/hibernate code to temporarily
105 * change gfp_allowed_mask in order to avoid using I/O during memory allocations
106 * while devices are suspended. To avoid races with the suspend/hibernate code,
107 * they should always be called with pm_mutex held (gfp_allowed_mask also should
108 * only be modified with pm_mutex held, unless the suspend/hibernate code is
109 * guaranteed not to run in parallel with that modification).
112 static gfp_t saved_gfp_mask;
114 void pm_restore_gfp_mask(void)
116 WARN_ON(!mutex_is_locked(&pm_mutex));
117 if (saved_gfp_mask) {
118 gfp_allowed_mask = saved_gfp_mask;
123 void pm_restrict_gfp_mask(void)
125 WARN_ON(!mutex_is_locked(&pm_mutex));
126 WARN_ON(saved_gfp_mask);
127 saved_gfp_mask = gfp_allowed_mask;
128 gfp_allowed_mask &= ~GFP_IOFS;
131 static bool pm_suspending(void)
133 if ((gfp_allowed_mask & GFP_IOFS) == GFP_IOFS)
140 static bool pm_suspending(void)
144 #endif /* CONFIG_PM_SLEEP */
146 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
147 int pageblock_order __read_mostly;
150 static void __free_pages_ok(struct page *page, unsigned int order);
153 * results with 256, 32 in the lowmem_reserve sysctl:
154 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
155 * 1G machine -> (16M dma, 784M normal, 224M high)
156 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
157 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
158 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
160 * TBD: should special case ZONE_DMA32 machines here - in those we normally
161 * don't need any ZONE_NORMAL reservation
163 int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
164 #ifdef CONFIG_ZONE_DMA
167 #ifdef CONFIG_ZONE_DMA32
170 #ifdef CONFIG_HIGHMEM
176 EXPORT_SYMBOL(totalram_pages);
178 static char * const zone_names[MAX_NR_ZONES] = {
179 #ifdef CONFIG_ZONE_DMA
182 #ifdef CONFIG_ZONE_DMA32
186 #ifdef CONFIG_HIGHMEM
192 int min_free_kbytes = 1024;
193 int min_free_order_shift = 1;
195 static unsigned long __meminitdata nr_kernel_pages;
196 static unsigned long __meminitdata nr_all_pages;
197 static unsigned long __meminitdata dma_reserve;
199 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
201 * MAX_ACTIVE_REGIONS determines the maximum number of distinct
202 * ranges of memory (RAM) that may be registered with add_active_range().
203 * Ranges passed to add_active_range() will be merged if possible
204 * so the number of times add_active_range() can be called is
205 * related to the number of nodes and the number of holes
207 #ifdef CONFIG_MAX_ACTIVE_REGIONS
208 /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
209 #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
211 #if MAX_NUMNODES >= 32
212 /* If there can be many nodes, allow up to 50 holes per node */
213 #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
215 /* By default, allow up to 256 distinct regions */
216 #define MAX_ACTIVE_REGIONS 256
220 static struct node_active_region __meminitdata early_node_map[MAX_ACTIVE_REGIONS];
221 static int __meminitdata nr_nodemap_entries;
222 static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
223 static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
224 static unsigned long __initdata required_kernelcore;
225 static unsigned long __initdata required_movablecore;
226 static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
228 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
230 EXPORT_SYMBOL(movable_zone);
231 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
234 int nr_node_ids __read_mostly = MAX_NUMNODES;
235 int nr_online_nodes __read_mostly = 1;
236 EXPORT_SYMBOL(nr_node_ids);
237 EXPORT_SYMBOL(nr_online_nodes);
240 int page_group_by_mobility_disabled __read_mostly;
242 static void set_pageblock_migratetype(struct page *page, int migratetype)
245 if (unlikely(page_group_by_mobility_disabled))
246 migratetype = MIGRATE_UNMOVABLE;
248 set_pageblock_flags_group(page, (unsigned long)migratetype,
249 PB_migrate, PB_migrate_end);
252 bool oom_killer_disabled __read_mostly;
254 #ifdef CONFIG_DEBUG_VM
255 static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
259 unsigned long pfn = page_to_pfn(page);
262 seq = zone_span_seqbegin(zone);
263 if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
265 else if (pfn < zone->zone_start_pfn)
267 } while (zone_span_seqretry(zone, seq));
272 static int page_is_consistent(struct zone *zone, struct page *page)
274 if (!pfn_valid_within(page_to_pfn(page)))
276 if (zone != page_zone(page))
282 * Temporary debugging check for pages not lying within a given zone.
284 static int bad_range(struct zone *zone, struct page *page)
286 if (page_outside_zone_boundaries(zone, page))
288 if (!page_is_consistent(zone, page))
294 static inline int bad_range(struct zone *zone, struct page *page)
300 static void bad_page(struct page *page)
302 static unsigned long resume;
303 static unsigned long nr_shown;
304 static unsigned long nr_unshown;
306 /* Don't complain about poisoned pages */
307 if (PageHWPoison(page)) {
308 reset_page_mapcount(page); /* remove PageBuddy */
313 * Allow a burst of 60 reports, then keep quiet for that minute;
314 * or allow a steady drip of one report per second.
316 if (nr_shown == 60) {
317 if (time_before(jiffies, resume)) {
323 "BUG: Bad page state: %lu messages suppressed\n",
330 resume = jiffies + 60 * HZ;
332 printk(KERN_ALERT "BUG: Bad page state in process %s pfn:%05lx\n",
333 current->comm, page_to_pfn(page));
338 /* Leave bad fields for debug, except PageBuddy could make trouble */
339 reset_page_mapcount(page); /* remove PageBuddy */
340 add_taint(TAINT_BAD_PAGE);
344 * Higher-order pages are called "compound pages". They are structured thusly:
346 * The first PAGE_SIZE page is called the "head page".
348 * The remaining PAGE_SIZE pages are called "tail pages".
350 * All pages have PG_compound set. All pages have their ->private pointing at
351 * the head page (even the head page has this).
353 * The first tail page's ->lru.next holds the address of the compound page's
354 * put_page() function. Its ->lru.prev holds the order of allocation.
355 * This usage means that zero-order pages may not be compound.
358 static void free_compound_page(struct page *page)
360 __free_pages_ok(page, compound_order(page));
363 void prep_compound_page(struct page *page, unsigned long order)
366 int nr_pages = 1 << order;
368 set_compound_page_dtor(page, free_compound_page);
369 set_compound_order(page, order);
371 for (i = 1; i < nr_pages; i++) {
372 struct page *p = page + i;
375 p->first_page = page;
379 /* update __split_huge_page_refcount if you change this function */
380 static int destroy_compound_page(struct page *page, unsigned long order)
383 int nr_pages = 1 << order;
386 if (unlikely(compound_order(page) != order) ||
387 unlikely(!PageHead(page))) {
392 __ClearPageHead(page);
394 for (i = 1; i < nr_pages; i++) {
395 struct page *p = page + i;
397 if (unlikely(!PageTail(p) || (p->first_page != page))) {
407 static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
412 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
413 * and __GFP_HIGHMEM from hard or soft interrupt context.
415 VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
416 for (i = 0; i < (1 << order); i++)
417 clear_highpage(page + i);
420 static inline void set_page_order(struct page *page, int order)
422 set_page_private(page, order);
423 __SetPageBuddy(page);
426 static inline void rmv_page_order(struct page *page)
428 __ClearPageBuddy(page);
429 set_page_private(page, 0);
433 * Locate the struct page for both the matching buddy in our
434 * pair (buddy1) and the combined O(n+1) page they form (page).
436 * 1) Any buddy B1 will have an order O twin B2 which satisfies
437 * the following equation:
439 * For example, if the starting buddy (buddy2) is #8 its order
441 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
443 * 2) Any buddy B will have an order O+1 parent P which
444 * satisfies the following equation:
447 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
449 static inline unsigned long
450 __find_buddy_index(unsigned long page_idx, unsigned int order)
452 return page_idx ^ (1 << order);
456 * This function checks whether a page is free && is the buddy
457 * we can do coalesce a page and its buddy if
458 * (a) the buddy is not in a hole &&
459 * (b) the buddy is in the buddy system &&
460 * (c) a page and its buddy have the same order &&
461 * (d) a page and its buddy are in the same zone.
463 * For recording whether a page is in the buddy system, we set ->_mapcount -2.
464 * Setting, clearing, and testing _mapcount -2 is serialized by zone->lock.
466 * For recording page's order, we use page_private(page).
468 static inline int page_is_buddy(struct page *page, struct page *buddy,
471 if (!pfn_valid_within(page_to_pfn(buddy)))
474 if (page_zone_id(page) != page_zone_id(buddy))
477 if (PageBuddy(buddy) && page_order(buddy) == order) {
478 VM_BUG_ON(page_count(buddy) != 0);
485 * Freeing function for a buddy system allocator.
487 * The concept of a buddy system is to maintain direct-mapped table
488 * (containing bit values) for memory blocks of various "orders".
489 * The bottom level table contains the map for the smallest allocatable
490 * units of memory (here, pages), and each level above it describes
491 * pairs of units from the levels below, hence, "buddies".
492 * At a high level, all that happens here is marking the table entry
493 * at the bottom level available, and propagating the changes upward
494 * as necessary, plus some accounting needed to play nicely with other
495 * parts of the VM system.
496 * At each level, we keep a list of pages, which are heads of continuous
497 * free pages of length of (1 << order) and marked with _mapcount -2. Page's
498 * order is recorded in page_private(page) field.
499 * So when we are allocating or freeing one, we can derive the state of the
500 * other. That is, if we allocate a small block, and both were
501 * free, the remainder of the region must be split into blocks.
502 * If a block is freed, and its buddy is also free, then this
503 * triggers coalescing into a block of larger size.
508 static inline void __free_one_page(struct page *page,
509 struct zone *zone, unsigned int order,
512 unsigned long page_idx;
513 unsigned long combined_idx;
514 unsigned long uninitialized_var(buddy_idx);
517 if (unlikely(PageCompound(page)))
518 if (unlikely(destroy_compound_page(page, order)))
521 VM_BUG_ON(migratetype == -1);
523 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
525 VM_BUG_ON(page_idx & ((1 << order) - 1));
526 VM_BUG_ON(bad_range(zone, page));
528 while (order < MAX_ORDER-1) {
529 buddy_idx = __find_buddy_index(page_idx, order);
530 buddy = page + (buddy_idx - page_idx);
531 if (!page_is_buddy(page, buddy, order))
534 /* Our buddy is free, merge with it and move up one order. */
535 list_del(&buddy->lru);
536 zone->free_area[order].nr_free--;
537 rmv_page_order(buddy);
538 combined_idx = buddy_idx & page_idx;
539 page = page + (combined_idx - page_idx);
540 page_idx = combined_idx;
543 set_page_order(page, order);
546 * If this is not the largest possible page, check if the buddy
547 * of the next-highest order is free. If it is, it's possible
548 * that pages are being freed that will coalesce soon. In case,
549 * that is happening, add the free page to the tail of the list
550 * so it's less likely to be used soon and more likely to be merged
551 * as a higher order page
553 if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) {
554 struct page *higher_page, *higher_buddy;
555 combined_idx = buddy_idx & page_idx;
556 higher_page = page + (combined_idx - page_idx);
557 buddy_idx = __find_buddy_index(combined_idx, order + 1);
558 higher_buddy = page + (buddy_idx - combined_idx);
559 if (page_is_buddy(higher_page, higher_buddy, order + 1)) {
560 list_add_tail(&page->lru,
561 &zone->free_area[order].free_list[migratetype]);
566 list_add(&page->lru, &zone->free_area[order].free_list[migratetype]);
568 zone->free_area[order].nr_free++;
572 * free_page_mlock() -- clean up attempts to free and mlocked() page.
573 * Page should not be on lru, so no need to fix that up.
574 * free_pages_check() will verify...
576 static inline void free_page_mlock(struct page *page)
578 __dec_zone_page_state(page, NR_MLOCK);
579 __count_vm_event(UNEVICTABLE_MLOCKFREED);
582 static inline int free_pages_check(struct page *page)
584 if (unlikely(page_mapcount(page) |
585 (page->mapping != NULL) |
586 (atomic_read(&page->_count) != 0) |
587 (page->flags & PAGE_FLAGS_CHECK_AT_FREE) |
588 (mem_cgroup_bad_page_check(page)))) {
592 if (page->flags & PAGE_FLAGS_CHECK_AT_PREP)
593 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
598 * Frees a number of pages from the PCP lists
599 * Assumes all pages on list are in same zone, and of same order.
600 * count is the number of pages to free.
602 * If the zone was previously in an "all pages pinned" state then look to
603 * see if this freeing clears that state.
605 * And clear the zone's pages_scanned counter, to hold off the "all pages are
606 * pinned" detection logic.
608 static void free_pcppages_bulk(struct zone *zone, int count,
609 struct per_cpu_pages *pcp)
615 spin_lock(&zone->lock);
616 zone->all_unreclaimable = 0;
617 zone->pages_scanned = 0;
621 struct list_head *list;
624 * Remove pages from lists in a round-robin fashion. A
625 * batch_free count is maintained that is incremented when an
626 * empty list is encountered. This is so more pages are freed
627 * off fuller lists instead of spinning excessively around empty
632 if (++migratetype == MIGRATE_PCPTYPES)
634 list = &pcp->lists[migratetype];
635 } while (list_empty(list));
637 /* This is the only non-empty list. Free them all. */
638 if (batch_free == MIGRATE_PCPTYPES)
639 batch_free = to_free;
642 page = list_entry(list->prev, struct page, lru);
643 /* must delete as __free_one_page list manipulates */
644 list_del(&page->lru);
645 /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */
646 __free_one_page(page, zone, 0, page_private(page));
647 trace_mm_page_pcpu_drain(page, 0, page_private(page));
648 } while (--to_free && --batch_free && !list_empty(list));
650 __mod_zone_page_state(zone, NR_FREE_PAGES, count);
651 spin_unlock(&zone->lock);
654 static void free_one_page(struct zone *zone, struct page *page, int order,
657 spin_lock(&zone->lock);
658 zone->all_unreclaimable = 0;
659 zone->pages_scanned = 0;
661 __free_one_page(page, zone, order, migratetype);
662 __mod_zone_page_state(zone, NR_FREE_PAGES, 1 << order);
663 spin_unlock(&zone->lock);
666 static bool free_pages_prepare(struct page *page, unsigned int order)
671 trace_mm_page_free_direct(page, order);
672 kmemcheck_free_shadow(page, order);
675 page->mapping = NULL;
676 for (i = 0; i < (1 << order); i++)
677 bad += free_pages_check(page + i);
681 if (!PageHighMem(page)) {
682 debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order);
683 debug_check_no_obj_freed(page_address(page),
686 arch_free_page(page, order);
687 kernel_map_pages(page, 1 << order, 0);
692 static void __free_pages_ok(struct page *page, unsigned int order)
695 int wasMlocked = __TestClearPageMlocked(page);
697 if (!free_pages_prepare(page, order))
700 local_irq_save(flags);
701 if (unlikely(wasMlocked))
702 free_page_mlock(page);
703 __count_vm_events(PGFREE, 1 << order);
704 free_one_page(page_zone(page), page, order,
705 get_pageblock_migratetype(page));
706 local_irq_restore(flags);
710 * permit the bootmem allocator to evade page validation on high-order frees
712 void __meminit __free_pages_bootmem(struct page *page, unsigned int order)
715 __ClearPageReserved(page);
716 set_page_count(page, 0);
717 set_page_refcounted(page);
723 for (loop = 0; loop < BITS_PER_LONG; loop++) {
724 struct page *p = &page[loop];
726 if (loop + 1 < BITS_PER_LONG)
728 __ClearPageReserved(p);
729 set_page_count(p, 0);
732 set_page_refcounted(page);
733 __free_pages(page, order);
739 * The order of subdivision here is critical for the IO subsystem.
740 * Please do not alter this order without good reasons and regression
741 * testing. Specifically, as large blocks of memory are subdivided,
742 * the order in which smaller blocks are delivered depends on the order
743 * they're subdivided in this function. This is the primary factor
744 * influencing the order in which pages are delivered to the IO
745 * subsystem according to empirical testing, and this is also justified
746 * by considering the behavior of a buddy system containing a single
747 * large block of memory acted on by a series of small allocations.
748 * This behavior is a critical factor in sglist merging's success.
752 static inline void expand(struct zone *zone, struct page *page,
753 int low, int high, struct free_area *area,
756 unsigned long size = 1 << high;
762 VM_BUG_ON(bad_range(zone, &page[size]));
763 list_add(&page[size].lru, &area->free_list[migratetype]);
765 set_page_order(&page[size], high);
770 * This page is about to be returned from the page allocator
772 static inline int check_new_page(struct page *page)
774 if (unlikely(page_mapcount(page) |
775 (page->mapping != NULL) |
776 (atomic_read(&page->_count) != 0) |
777 (page->flags & PAGE_FLAGS_CHECK_AT_PREP) |
778 (mem_cgroup_bad_page_check(page)))) {
785 static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
789 for (i = 0; i < (1 << order); i++) {
790 struct page *p = page + i;
791 if (unlikely(check_new_page(p)))
795 set_page_private(page, 0);
796 set_page_refcounted(page);
798 arch_alloc_page(page, order);
799 kernel_map_pages(page, 1 << order, 1);
801 if (gfp_flags & __GFP_ZERO)
802 prep_zero_page(page, order, gfp_flags);
804 if (order && (gfp_flags & __GFP_COMP))
805 prep_compound_page(page, order);
811 * Go through the free lists for the given migratetype and remove
812 * the smallest available page from the freelists
815 struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
818 unsigned int current_order;
819 struct free_area * area;
822 /* Find a page of the appropriate size in the preferred list */
823 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
824 area = &(zone->free_area[current_order]);
825 if (list_empty(&area->free_list[migratetype]))
828 page = list_entry(area->free_list[migratetype].next,
830 list_del(&page->lru);
831 rmv_page_order(page);
833 expand(zone, page, order, current_order, area, migratetype);
842 * This array describes the order lists are fallen back to when
843 * the free lists for the desirable migrate type are depleted
845 static int fallbacks[MIGRATE_TYPES][MIGRATE_TYPES-1] = {
846 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
847 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
848 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
849 [MIGRATE_RESERVE] = { MIGRATE_RESERVE, MIGRATE_RESERVE, MIGRATE_RESERVE }, /* Never used */
853 * Move the free pages in a range to the free lists of the requested type.
854 * Note that start_page and end_pages are not aligned on a pageblock
855 * boundary. If alignment is required, use move_freepages_block()
857 static int move_freepages(struct zone *zone,
858 struct page *start_page, struct page *end_page,
865 #ifndef CONFIG_HOLES_IN_ZONE
867 * page_zone is not safe to call in this context when
868 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
869 * anyway as we check zone boundaries in move_freepages_block().
870 * Remove at a later date when no bug reports exist related to
871 * grouping pages by mobility
873 BUG_ON(page_zone(start_page) != page_zone(end_page));
876 for (page = start_page; page <= end_page;) {
877 /* Make sure we are not inadvertently changing nodes */
878 VM_BUG_ON(page_to_nid(page) != zone_to_nid(zone));
880 if (!pfn_valid_within(page_to_pfn(page))) {
885 if (!PageBuddy(page)) {
890 order = page_order(page);
891 list_move(&page->lru,
892 &zone->free_area[order].free_list[migratetype]);
894 pages_moved += 1 << order;
900 static int move_freepages_block(struct zone *zone, struct page *page,
903 unsigned long start_pfn, end_pfn;
904 struct page *start_page, *end_page;
906 start_pfn = page_to_pfn(page);
907 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
908 start_page = pfn_to_page(start_pfn);
909 end_page = start_page + pageblock_nr_pages - 1;
910 end_pfn = start_pfn + pageblock_nr_pages - 1;
912 /* Do not cross zone boundaries */
913 if (start_pfn < zone->zone_start_pfn)
915 if (end_pfn >= zone->zone_start_pfn + zone->spanned_pages)
918 return move_freepages(zone, start_page, end_page, migratetype);
921 static void change_pageblock_range(struct page *pageblock_page,
922 int start_order, int migratetype)
924 int nr_pageblocks = 1 << (start_order - pageblock_order);
926 while (nr_pageblocks--) {
927 set_pageblock_migratetype(pageblock_page, migratetype);
928 pageblock_page += pageblock_nr_pages;
932 /* Remove an element from the buddy allocator from the fallback list */
933 static inline struct page *
934 __rmqueue_fallback(struct zone *zone, int order, int start_migratetype)
936 struct free_area * area;
941 /* Find the largest possible block of pages in the other list */
942 for (current_order = MAX_ORDER-1; current_order >= order;
944 for (i = 0; i < MIGRATE_TYPES - 1; i++) {
945 migratetype = fallbacks[start_migratetype][i];
947 /* MIGRATE_RESERVE handled later if necessary */
948 if (migratetype == MIGRATE_RESERVE)
951 area = &(zone->free_area[current_order]);
952 if (list_empty(&area->free_list[migratetype]))
955 page = list_entry(area->free_list[migratetype].next,
960 * If breaking a large block of pages, move all free
961 * pages to the preferred allocation list. If falling
962 * back for a reclaimable kernel allocation, be more
963 * aggressive about taking ownership of free pages
965 if (unlikely(current_order >= (pageblock_order >> 1)) ||
966 start_migratetype == MIGRATE_RECLAIMABLE ||
967 page_group_by_mobility_disabled) {
969 pages = move_freepages_block(zone, page,
972 /* Claim the whole block if over half of it is free */
973 if (pages >= (1 << (pageblock_order-1)) ||
974 page_group_by_mobility_disabled)
975 set_pageblock_migratetype(page,
978 migratetype = start_migratetype;
981 /* Remove the page from the freelists */
982 list_del(&page->lru);
983 rmv_page_order(page);
985 /* Take ownership for orders >= pageblock_order */
986 if (current_order >= pageblock_order)
987 change_pageblock_range(page, current_order,
990 expand(zone, page, order, current_order, area, migratetype);
992 trace_mm_page_alloc_extfrag(page, order, current_order,
993 start_migratetype, migratetype);
1003 * Do the hard work of removing an element from the buddy allocator.
1004 * Call me with the zone->lock already held.
1006 static struct page *__rmqueue(struct zone *zone, unsigned int order,
1012 page = __rmqueue_smallest(zone, order, migratetype);
1014 if (unlikely(!page) && migratetype != MIGRATE_RESERVE) {
1015 page = __rmqueue_fallback(zone, order, migratetype);
1018 * Use MIGRATE_RESERVE rather than fail an allocation. goto
1019 * is used because __rmqueue_smallest is an inline function
1020 * and we want just one call site
1023 migratetype = MIGRATE_RESERVE;
1028 trace_mm_page_alloc_zone_locked(page, order, migratetype);
1033 * Obtain a specified number of elements from the buddy allocator, all under
1034 * a single hold of the lock, for efficiency. Add them to the supplied list.
1035 * Returns the number of new pages which were placed at *list.
1037 static int rmqueue_bulk(struct zone *zone, unsigned int order,
1038 unsigned long count, struct list_head *list,
1039 int migratetype, int cold)
1043 spin_lock(&zone->lock);
1044 for (i = 0; i < count; ++i) {
1045 struct page *page = __rmqueue(zone, order, migratetype);
1046 if (unlikely(page == NULL))
1050 * Split buddy pages returned by expand() are received here
1051 * in physical page order. The page is added to the callers and
1052 * list and the list head then moves forward. From the callers
1053 * perspective, the linked list is ordered by page number in
1054 * some conditions. This is useful for IO devices that can
1055 * merge IO requests if the physical pages are ordered
1058 if (likely(cold == 0))
1059 list_add(&page->lru, list);
1061 list_add_tail(&page->lru, list);
1062 set_page_private(page, migratetype);
1065 __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
1066 spin_unlock(&zone->lock);
1072 * Called from the vmstat counter updater to drain pagesets of this
1073 * currently executing processor on remote nodes after they have
1076 * Note that this function must be called with the thread pinned to
1077 * a single processor.
1079 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
1081 unsigned long flags;
1084 local_irq_save(flags);
1085 if (pcp->count >= pcp->batch)
1086 to_drain = pcp->batch;
1088 to_drain = pcp->count;
1089 free_pcppages_bulk(zone, to_drain, pcp);
1090 pcp->count -= to_drain;
1091 local_irq_restore(flags);
1096 * Drain pages of the indicated processor.
1098 * The processor must either be the current processor and the
1099 * thread pinned to the current processor or a processor that
1102 static void drain_pages(unsigned int cpu)
1104 unsigned long flags;
1107 for_each_populated_zone(zone) {
1108 struct per_cpu_pageset *pset;
1109 struct per_cpu_pages *pcp;
1111 local_irq_save(flags);
1112 pset = per_cpu_ptr(zone->pageset, cpu);
1116 free_pcppages_bulk(zone, pcp->count, pcp);
1119 local_irq_restore(flags);
1124 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
1126 void drain_local_pages(void *arg)
1128 drain_pages(smp_processor_id());
1132 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
1134 void drain_all_pages(void)
1136 on_each_cpu(drain_local_pages, NULL, 1);
1139 #ifdef CONFIG_HIBERNATION
1141 void mark_free_pages(struct zone *zone)
1143 unsigned long pfn, max_zone_pfn;
1144 unsigned long flags;
1146 struct list_head *curr;
1148 if (!zone->spanned_pages)
1151 spin_lock_irqsave(&zone->lock, flags);
1153 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
1154 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1155 if (pfn_valid(pfn)) {
1156 struct page *page = pfn_to_page(pfn);
1158 if (!swsusp_page_is_forbidden(page))
1159 swsusp_unset_page_free(page);
1162 for_each_migratetype_order(order, t) {
1163 list_for_each(curr, &zone->free_area[order].free_list[t]) {
1166 pfn = page_to_pfn(list_entry(curr, struct page, lru));
1167 for (i = 0; i < (1UL << order); i++)
1168 swsusp_set_page_free(pfn_to_page(pfn + i));
1171 spin_unlock_irqrestore(&zone->lock, flags);
1173 #endif /* CONFIG_PM */
1176 * Free a 0-order page
1177 * cold == 1 ? free a cold page : free a hot page
1179 void free_hot_cold_page(struct page *page, int cold)
1181 struct zone *zone = page_zone(page);
1182 struct per_cpu_pages *pcp;
1183 unsigned long flags;
1185 int wasMlocked = __TestClearPageMlocked(page);
1187 if (!free_pages_prepare(page, 0))
1190 migratetype = get_pageblock_migratetype(page);
1191 set_page_private(page, migratetype);
1192 local_irq_save(flags);
1193 if (unlikely(wasMlocked))
1194 free_page_mlock(page);
1195 __count_vm_event(PGFREE);
1198 * We only track unmovable, reclaimable and movable on pcp lists.
1199 * Free ISOLATE pages back to the allocator because they are being
1200 * offlined but treat RESERVE as movable pages so we can get those
1201 * areas back if necessary. Otherwise, we may have to free
1202 * excessively into the page allocator
1204 if (migratetype >= MIGRATE_PCPTYPES) {
1205 if (unlikely(migratetype == MIGRATE_ISOLATE)) {
1206 free_one_page(zone, page, 0, migratetype);
1209 migratetype = MIGRATE_MOVABLE;
1212 pcp = &this_cpu_ptr(zone->pageset)->pcp;
1214 list_add_tail(&page->lru, &pcp->lists[migratetype]);
1216 list_add(&page->lru, &pcp->lists[migratetype]);
1218 if (pcp->count >= pcp->high) {
1219 free_pcppages_bulk(zone, pcp->batch, pcp);
1220 pcp->count -= pcp->batch;
1224 local_irq_restore(flags);
1228 * split_page takes a non-compound higher-order page, and splits it into
1229 * n (1<<order) sub-pages: page[0..n]
1230 * Each sub-page must be freed individually.
1232 * Note: this is probably too low level an operation for use in drivers.
1233 * Please consult with lkml before using this in your driver.
1235 void split_page(struct page *page, unsigned int order)
1239 VM_BUG_ON(PageCompound(page));
1240 VM_BUG_ON(!page_count(page));
1242 #ifdef CONFIG_KMEMCHECK
1244 * Split shadow pages too, because free(page[0]) would
1245 * otherwise free the whole shadow.
1247 if (kmemcheck_page_is_tracked(page))
1248 split_page(virt_to_page(page[0].shadow), order);
1251 for (i = 1; i < (1 << order); i++)
1252 set_page_refcounted(page + i);
1256 * Similar to split_page except the page is already free. As this is only
1257 * being used for migration, the migratetype of the block also changes.
1258 * As this is called with interrupts disabled, the caller is responsible
1259 * for calling arch_alloc_page() and kernel_map_page() after interrupts
1262 * Note: this is probably too low level an operation for use in drivers.
1263 * Please consult with lkml before using this in your driver.
1265 int split_free_page(struct page *page)
1268 unsigned long watermark;
1271 BUG_ON(!PageBuddy(page));
1273 zone = page_zone(page);
1274 order = page_order(page);
1276 /* Obey watermarks as if the page was being allocated */
1277 watermark = low_wmark_pages(zone) + (1 << order);
1278 if (!zone_watermark_ok(zone, 0, watermark, 0, 0))
1281 /* Remove page from free list */
1282 list_del(&page->lru);
1283 zone->free_area[order].nr_free--;
1284 rmv_page_order(page);
1285 __mod_zone_page_state(zone, NR_FREE_PAGES, -(1UL << order));
1287 /* Split into individual pages */
1288 set_page_refcounted(page);
1289 split_page(page, order);
1291 if (order >= pageblock_order - 1) {
1292 struct page *endpage = page + (1 << order) - 1;
1293 for (; page < endpage; page += pageblock_nr_pages)
1294 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
1301 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1302 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1306 struct page *buffered_rmqueue(struct zone *preferred_zone,
1307 struct zone *zone, int order, gfp_t gfp_flags,
1310 unsigned long flags;
1312 int cold = !!(gfp_flags & __GFP_COLD);
1315 if (likely(order == 0)) {
1316 struct per_cpu_pages *pcp;
1317 struct list_head *list;
1319 local_irq_save(flags);
1320 pcp = &this_cpu_ptr(zone->pageset)->pcp;
1321 list = &pcp->lists[migratetype];
1322 if (list_empty(list)) {
1323 pcp->count += rmqueue_bulk(zone, 0,
1326 if (unlikely(list_empty(list)))
1331 page = list_entry(list->prev, struct page, lru);
1333 page = list_entry(list->next, struct page, lru);
1335 list_del(&page->lru);
1338 if (unlikely(gfp_flags & __GFP_NOFAIL)) {
1340 * __GFP_NOFAIL is not to be used in new code.
1342 * All __GFP_NOFAIL callers should be fixed so that they
1343 * properly detect and handle allocation failures.
1345 * We most definitely don't want callers attempting to
1346 * allocate greater than order-1 page units with
1349 WARN_ON_ONCE(order > 1);
1351 spin_lock_irqsave(&zone->lock, flags);
1352 page = __rmqueue(zone, order, migratetype);
1353 spin_unlock(&zone->lock);
1356 __mod_zone_page_state(zone, NR_FREE_PAGES, -(1 << order));
1359 __count_zone_vm_events(PGALLOC, zone, 1 << order);
1360 zone_statistics(preferred_zone, zone, gfp_flags);
1361 local_irq_restore(flags);
1363 VM_BUG_ON(bad_range(zone, page));
1364 if (prep_new_page(page, order, gfp_flags))
1369 local_irq_restore(flags);
1373 /* The ALLOC_WMARK bits are used as an index to zone->watermark */
1374 #define ALLOC_WMARK_MIN WMARK_MIN
1375 #define ALLOC_WMARK_LOW WMARK_LOW
1376 #define ALLOC_WMARK_HIGH WMARK_HIGH
1377 #define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */
1379 /* Mask to get the watermark bits */
1380 #define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1)
1382 #define ALLOC_HARDER 0x10 /* try to alloc harder */
1383 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1384 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
1386 #ifdef CONFIG_FAIL_PAGE_ALLOC
1388 static struct fail_page_alloc_attr {
1389 struct fault_attr attr;
1391 u32 ignore_gfp_highmem;
1392 u32 ignore_gfp_wait;
1395 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1397 struct dentry *ignore_gfp_highmem_file;
1398 struct dentry *ignore_gfp_wait_file;
1399 struct dentry *min_order_file;
1401 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1403 } fail_page_alloc = {
1404 .attr = FAULT_ATTR_INITIALIZER,
1405 .ignore_gfp_wait = 1,
1406 .ignore_gfp_highmem = 1,
1410 static int __init setup_fail_page_alloc(char *str)
1412 return setup_fault_attr(&fail_page_alloc.attr, str);
1414 __setup("fail_page_alloc=", setup_fail_page_alloc);
1416 static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1418 if (order < fail_page_alloc.min_order)
1420 if (gfp_mask & __GFP_NOFAIL)
1422 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
1424 if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
1427 return should_fail(&fail_page_alloc.attr, 1 << order);
1430 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1432 static int __init fail_page_alloc_debugfs(void)
1434 mode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
1438 err = init_fault_attr_dentries(&fail_page_alloc.attr,
1442 dir = fail_page_alloc.attr.dentries.dir;
1444 fail_page_alloc.ignore_gfp_wait_file =
1445 debugfs_create_bool("ignore-gfp-wait", mode, dir,
1446 &fail_page_alloc.ignore_gfp_wait);
1448 fail_page_alloc.ignore_gfp_highmem_file =
1449 debugfs_create_bool("ignore-gfp-highmem", mode, dir,
1450 &fail_page_alloc.ignore_gfp_highmem);
1451 fail_page_alloc.min_order_file =
1452 debugfs_create_u32("min-order", mode, dir,
1453 &fail_page_alloc.min_order);
1455 if (!fail_page_alloc.ignore_gfp_wait_file ||
1456 !fail_page_alloc.ignore_gfp_highmem_file ||
1457 !fail_page_alloc.min_order_file) {
1459 debugfs_remove(fail_page_alloc.ignore_gfp_wait_file);
1460 debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file);
1461 debugfs_remove(fail_page_alloc.min_order_file);
1462 cleanup_fault_attr_dentries(&fail_page_alloc.attr);
1468 late_initcall(fail_page_alloc_debugfs);
1470 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1472 #else /* CONFIG_FAIL_PAGE_ALLOC */
1474 static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1479 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1482 * Return true if free pages are above 'mark'. This takes into account the order
1483 * of the allocation.
1485 static bool __zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1486 int classzone_idx, int alloc_flags, long free_pages)
1488 /* free_pages my go negative - that's OK */
1492 free_pages -= (1 << order) + 1;
1493 if (alloc_flags & ALLOC_HIGH)
1495 if (alloc_flags & ALLOC_HARDER)
1498 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
1500 for (o = 0; o < order; o++) {
1501 /* At the next order, this order's pages become unavailable */
1502 free_pages -= z->free_area[o].nr_free << o;
1504 /* Require fewer higher order pages to be free */
1505 min >>= min_free_order_shift;
1507 if (free_pages <= min)
1513 bool zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1514 int classzone_idx, int alloc_flags)
1516 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
1517 zone_page_state(z, NR_FREE_PAGES));
1520 bool zone_watermark_ok_safe(struct zone *z, int order, unsigned long mark,
1521 int classzone_idx, int alloc_flags)
1523 long free_pages = zone_page_state(z, NR_FREE_PAGES);
1525 if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark)
1526 free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES);
1528 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
1534 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1535 * skip over zones that are not allowed by the cpuset, or that have
1536 * been recently (in last second) found to be nearly full. See further
1537 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1538 * that have to skip over a lot of full or unallowed zones.
1540 * If the zonelist cache is present in the passed in zonelist, then
1541 * returns a pointer to the allowed node mask (either the current
1542 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
1544 * If the zonelist cache is not available for this zonelist, does
1545 * nothing and returns NULL.
1547 * If the fullzones BITMAP in the zonelist cache is stale (more than
1548 * a second since last zap'd) then we zap it out (clear its bits.)
1550 * We hold off even calling zlc_setup, until after we've checked the
1551 * first zone in the zonelist, on the theory that most allocations will
1552 * be satisfied from that first zone, so best to examine that zone as
1553 * quickly as we can.
1555 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1557 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1558 nodemask_t *allowednodes; /* zonelist_cache approximation */
1560 zlc = zonelist->zlcache_ptr;
1564 if (time_after(jiffies, zlc->last_full_zap + HZ)) {
1565 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1566 zlc->last_full_zap = jiffies;
1569 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1570 &cpuset_current_mems_allowed :
1571 &node_states[N_HIGH_MEMORY];
1572 return allowednodes;
1576 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1577 * if it is worth looking at further for free memory:
1578 * 1) Check that the zone isn't thought to be full (doesn't have its
1579 * bit set in the zonelist_cache fullzones BITMAP).
1580 * 2) Check that the zones node (obtained from the zonelist_cache
1581 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1582 * Return true (non-zero) if zone is worth looking at further, or
1583 * else return false (zero) if it is not.
1585 * This check -ignores- the distinction between various watermarks,
1586 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1587 * found to be full for any variation of these watermarks, it will
1588 * be considered full for up to one second by all requests, unless
1589 * we are so low on memory on all allowed nodes that we are forced
1590 * into the second scan of the zonelist.
1592 * In the second scan we ignore this zonelist cache and exactly
1593 * apply the watermarks to all zones, even it is slower to do so.
1594 * We are low on memory in the second scan, and should leave no stone
1595 * unturned looking for a free page.
1597 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
1598 nodemask_t *allowednodes)
1600 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1601 int i; /* index of *z in zonelist zones */
1602 int n; /* node that zone *z is on */
1604 zlc = zonelist->zlcache_ptr;
1608 i = z - zonelist->_zonerefs;
1611 /* This zone is worth trying if it is allowed but not full */
1612 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1616 * Given 'z' scanning a zonelist, set the corresponding bit in
1617 * zlc->fullzones, so that subsequent attempts to allocate a page
1618 * from that zone don't waste time re-examining it.
1620 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
1622 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1623 int i; /* index of *z in zonelist zones */
1625 zlc = zonelist->zlcache_ptr;
1629 i = z - zonelist->_zonerefs;
1631 set_bit(i, zlc->fullzones);
1635 * clear all zones full, called after direct reclaim makes progress so that
1636 * a zone that was recently full is not skipped over for up to a second
1638 static void zlc_clear_zones_full(struct zonelist *zonelist)
1640 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1642 zlc = zonelist->zlcache_ptr;
1646 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1649 #else /* CONFIG_NUMA */
1651 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1656 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
1657 nodemask_t *allowednodes)
1662 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
1666 static void zlc_clear_zones_full(struct zonelist *zonelist)
1669 #endif /* CONFIG_NUMA */
1672 * get_page_from_freelist goes through the zonelist trying to allocate
1675 static struct page *
1676 get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order,
1677 struct zonelist *zonelist, int high_zoneidx, int alloc_flags,
1678 struct zone *preferred_zone, int migratetype)
1681 struct page *page = NULL;
1684 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1685 int zlc_active = 0; /* set if using zonelist_cache */
1686 int did_zlc_setup = 0; /* just call zlc_setup() one time */
1688 classzone_idx = zone_idx(preferred_zone);
1691 * Scan zonelist, looking for a zone with enough free.
1692 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1694 for_each_zone_zonelist_nodemask(zone, z, zonelist,
1695 high_zoneidx, nodemask) {
1696 if (NUMA_BUILD && zlc_active &&
1697 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1699 if ((alloc_flags & ALLOC_CPUSET) &&
1700 !cpuset_zone_allowed_softwall(zone, gfp_mask))
1703 BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
1704 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
1708 mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
1709 if (zone_watermark_ok(zone, order, mark,
1710 classzone_idx, alloc_flags))
1713 if (NUMA_BUILD && !did_zlc_setup && nr_online_nodes > 1) {
1715 * we do zlc_setup if there are multiple nodes
1716 * and before considering the first zone allowed
1719 allowednodes = zlc_setup(zonelist, alloc_flags);
1724 if (zone_reclaim_mode == 0)
1725 goto this_zone_full;
1728 * As we may have just activated ZLC, check if the first
1729 * eligible zone has failed zone_reclaim recently.
1731 if (NUMA_BUILD && zlc_active &&
1732 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1735 ret = zone_reclaim(zone, gfp_mask, order);
1737 case ZONE_RECLAIM_NOSCAN:
1740 case ZONE_RECLAIM_FULL:
1741 /* scanned but unreclaimable */
1744 /* did we reclaim enough */
1745 if (!zone_watermark_ok(zone, order, mark,
1746 classzone_idx, alloc_flags))
1747 goto this_zone_full;
1752 page = buffered_rmqueue(preferred_zone, zone, order,
1753 gfp_mask, migratetype);
1758 zlc_mark_zone_full(zonelist, z);
1761 if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) {
1762 /* Disable zlc cache for second zonelist scan */
1770 * Large machines with many possible nodes should not always dump per-node
1771 * meminfo in irq context.
1773 static inline bool should_suppress_show_mem(void)
1778 ret = in_interrupt();
1783 static DEFINE_RATELIMIT_STATE(nopage_rs,
1784 DEFAULT_RATELIMIT_INTERVAL,
1785 DEFAULT_RATELIMIT_BURST);
1787 void warn_alloc_failed(gfp_t gfp_mask, int order, const char *fmt, ...)
1790 unsigned int filter = SHOW_MEM_FILTER_NODES;
1792 if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs))
1796 * This documents exceptions given to allocations in certain
1797 * contexts that are allowed to allocate outside current's set
1800 if (!(gfp_mask & __GFP_NOMEMALLOC))
1801 if (test_thread_flag(TIF_MEMDIE) ||
1802 (current->flags & (PF_MEMALLOC | PF_EXITING)))
1803 filter &= ~SHOW_MEM_FILTER_NODES;
1804 if (in_interrupt() || !(gfp_mask & __GFP_WAIT))
1805 filter &= ~SHOW_MEM_FILTER_NODES;
1808 printk(KERN_WARNING);
1809 va_start(args, fmt);
1814 pr_warning("%s: page allocation failure: order:%d, mode:0x%x\n",
1815 current->comm, order, gfp_mask);
1818 if (!should_suppress_show_mem())
1823 should_alloc_retry(gfp_t gfp_mask, unsigned int order,
1824 unsigned long pages_reclaimed)
1826 /* Do not loop if specifically requested */
1827 if (gfp_mask & __GFP_NORETRY)
1831 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
1832 * means __GFP_NOFAIL, but that may not be true in other
1835 if (order <= PAGE_ALLOC_COSTLY_ORDER)
1839 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
1840 * specified, then we retry until we no longer reclaim any pages
1841 * (above), or we've reclaimed an order of pages at least as
1842 * large as the allocation's order. In both cases, if the
1843 * allocation still fails, we stop retrying.
1845 if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order))
1849 * Don't let big-order allocations loop unless the caller
1850 * explicitly requests that.
1852 if (gfp_mask & __GFP_NOFAIL)
1858 static inline struct page *
1859 __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
1860 struct zonelist *zonelist, enum zone_type high_zoneidx,
1861 nodemask_t *nodemask, struct zone *preferred_zone,
1866 /* Acquire the OOM killer lock for the zones in zonelist */
1867 if (!try_set_zonelist_oom(zonelist, gfp_mask)) {
1868 schedule_timeout_uninterruptible(1);
1873 * Go through the zonelist yet one more time, keep very high watermark
1874 * here, this is only to catch a parallel oom killing, we must fail if
1875 * we're still under heavy pressure.
1877 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask,
1878 order, zonelist, high_zoneidx,
1879 ALLOC_WMARK_HIGH|ALLOC_CPUSET,
1880 preferred_zone, migratetype);
1884 if (!(gfp_mask & __GFP_NOFAIL)) {
1885 /* The OOM killer will not help higher order allocs */
1886 if (order > PAGE_ALLOC_COSTLY_ORDER)
1888 /* The OOM killer does not needlessly kill tasks for lowmem */
1889 if (high_zoneidx < ZONE_NORMAL)
1892 * GFP_THISNODE contains __GFP_NORETRY and we never hit this.
1893 * Sanity check for bare calls of __GFP_THISNODE, not real OOM.
1894 * The caller should handle page allocation failure by itself if
1895 * it specifies __GFP_THISNODE.
1896 * Note: Hugepage uses it but will hit PAGE_ALLOC_COSTLY_ORDER.
1898 if (gfp_mask & __GFP_THISNODE)
1901 /* Exhausted what can be done so it's blamo time */
1902 out_of_memory(zonelist, gfp_mask, order, nodemask);
1905 clear_zonelist_oom(zonelist, gfp_mask);
1909 #ifdef CONFIG_COMPACTION
1910 /* Try memory compaction for high-order allocations before reclaim */
1911 static struct page *
1912 __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
1913 struct zonelist *zonelist, enum zone_type high_zoneidx,
1914 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
1915 int migratetype, unsigned long *did_some_progress,
1916 bool sync_migration)
1920 if (!order || compaction_deferred(preferred_zone))
1923 current->flags |= PF_MEMALLOC;
1924 *did_some_progress = try_to_compact_pages(zonelist, order, gfp_mask,
1925 nodemask, sync_migration);
1926 current->flags &= ~PF_MEMALLOC;
1927 if (*did_some_progress != COMPACT_SKIPPED) {
1929 /* Page migration frees to the PCP lists but we want merging */
1930 drain_pages(get_cpu());
1933 page = get_page_from_freelist(gfp_mask, nodemask,
1934 order, zonelist, high_zoneidx,
1935 alloc_flags, preferred_zone,
1938 preferred_zone->compact_considered = 0;
1939 preferred_zone->compact_defer_shift = 0;
1940 count_vm_event(COMPACTSUCCESS);
1945 * It's bad if compaction run occurs and fails.
1946 * The most likely reason is that pages exist,
1947 * but not enough to satisfy watermarks.
1949 count_vm_event(COMPACTFAIL);
1950 defer_compaction(preferred_zone);
1958 static inline struct page *
1959 __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
1960 struct zonelist *zonelist, enum zone_type high_zoneidx,
1961 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
1962 int migratetype, unsigned long *did_some_progress,
1963 bool sync_migration)
1967 #endif /* CONFIG_COMPACTION */
1969 /* The really slow allocator path where we enter direct reclaim */
1970 static inline struct page *
1971 __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
1972 struct zonelist *zonelist, enum zone_type high_zoneidx,
1973 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
1974 int migratetype, unsigned long *did_some_progress)
1976 struct page *page = NULL;
1977 struct reclaim_state reclaim_state;
1978 bool drained = false;
1982 /* We now go into synchronous reclaim */
1983 cpuset_memory_pressure_bump();
1984 current->flags |= PF_MEMALLOC;
1985 lockdep_set_current_reclaim_state(gfp_mask);
1986 reclaim_state.reclaimed_slab = 0;
1987 current->reclaim_state = &reclaim_state;
1989 *did_some_progress = try_to_free_pages(zonelist, order, gfp_mask, nodemask);
1991 current->reclaim_state = NULL;
1992 lockdep_clear_current_reclaim_state();
1993 current->flags &= ~PF_MEMALLOC;
1997 if (unlikely(!(*did_some_progress)))
2000 /* After successful reclaim, reconsider all zones for allocation */
2002 zlc_clear_zones_full(zonelist);
2005 page = get_page_from_freelist(gfp_mask, nodemask, order,
2006 zonelist, high_zoneidx,
2007 alloc_flags, preferred_zone,
2011 * If an allocation failed after direct reclaim, it could be because
2012 * pages are pinned on the per-cpu lists. Drain them and try again
2014 if (!page && !drained) {
2024 * This is called in the allocator slow-path if the allocation request is of
2025 * sufficient urgency to ignore watermarks and take other desperate measures
2027 static inline struct page *
2028 __alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order,
2029 struct zonelist *zonelist, enum zone_type high_zoneidx,
2030 nodemask_t *nodemask, struct zone *preferred_zone,
2036 page = get_page_from_freelist(gfp_mask, nodemask, order,
2037 zonelist, high_zoneidx, ALLOC_NO_WATERMARKS,
2038 preferred_zone, migratetype);
2040 if (!page && gfp_mask & __GFP_NOFAIL)
2041 wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50);
2042 } while (!page && (gfp_mask & __GFP_NOFAIL));
2048 void wake_all_kswapd(unsigned int order, struct zonelist *zonelist,
2049 enum zone_type high_zoneidx,
2050 enum zone_type classzone_idx)
2055 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx)
2056 wakeup_kswapd(zone, order, classzone_idx);
2060 gfp_to_alloc_flags(gfp_t gfp_mask)
2062 int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
2063 const gfp_t wait = gfp_mask & __GFP_WAIT;
2065 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
2066 BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
2069 * The caller may dip into page reserves a bit more if the caller
2070 * cannot run direct reclaim, or if the caller has realtime scheduling
2071 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
2072 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
2074 alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH);
2078 * Not worth trying to allocate harder for
2079 * __GFP_NOMEMALLOC even if it can't schedule.
2081 if (!(gfp_mask & __GFP_NOMEMALLOC))
2082 alloc_flags |= ALLOC_HARDER;
2084 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
2085 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
2087 alloc_flags &= ~ALLOC_CPUSET;
2088 } else if (unlikely(rt_task(current)) && !in_interrupt())
2089 alloc_flags |= ALLOC_HARDER;
2091 if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) {
2092 if (!in_interrupt() &&
2093 ((current->flags & PF_MEMALLOC) ||
2094 unlikely(test_thread_flag(TIF_MEMDIE))))
2095 alloc_flags |= ALLOC_NO_WATERMARKS;
2101 static inline struct page *
2102 __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
2103 struct zonelist *zonelist, enum zone_type high_zoneidx,
2104 nodemask_t *nodemask, struct zone *preferred_zone,
2107 const gfp_t wait = gfp_mask & __GFP_WAIT;
2108 struct page *page = NULL;
2110 unsigned long pages_reclaimed = 0;
2111 unsigned long did_some_progress;
2112 bool sync_migration = false;
2115 * In the slowpath, we sanity check order to avoid ever trying to
2116 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
2117 * be using allocators in order of preference for an area that is
2120 if (order >= MAX_ORDER) {
2121 WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
2126 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
2127 * __GFP_NOWARN set) should not cause reclaim since the subsystem
2128 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
2129 * using a larger set of nodes after it has established that the
2130 * allowed per node queues are empty and that nodes are
2133 if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
2137 if (!(gfp_mask & __GFP_NO_KSWAPD))
2138 wake_all_kswapd(order, zonelist, high_zoneidx,
2139 zone_idx(preferred_zone));
2142 * OK, we're below the kswapd watermark and have kicked background
2143 * reclaim. Now things get more complex, so set up alloc_flags according
2144 * to how we want to proceed.
2146 alloc_flags = gfp_to_alloc_flags(gfp_mask);
2149 * Find the true preferred zone if the allocation is unconstrained by
2152 if (!(alloc_flags & ALLOC_CPUSET) && !nodemask)
2153 first_zones_zonelist(zonelist, high_zoneidx, NULL,
2157 /* This is the last chance, in general, before the goto nopage. */
2158 page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist,
2159 high_zoneidx, alloc_flags & ~ALLOC_NO_WATERMARKS,
2160 preferred_zone, migratetype);
2164 /* Allocate without watermarks if the context allows */
2165 if (alloc_flags & ALLOC_NO_WATERMARKS) {
2166 page = __alloc_pages_high_priority(gfp_mask, order,
2167 zonelist, high_zoneidx, nodemask,
2168 preferred_zone, migratetype);
2173 /* Atomic allocations - we can't balance anything */
2177 /* Avoid recursion of direct reclaim */
2178 if (current->flags & PF_MEMALLOC)
2181 /* Avoid allocations with no watermarks from looping endlessly */
2182 if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL))
2186 * Try direct compaction. The first pass is asynchronous. Subsequent
2187 * attempts after direct reclaim are synchronous
2189 page = __alloc_pages_direct_compact(gfp_mask, order,
2190 zonelist, high_zoneidx,
2192 alloc_flags, preferred_zone,
2193 migratetype, &did_some_progress,
2197 sync_migration = true;
2199 /* Try direct reclaim and then allocating */
2200 page = __alloc_pages_direct_reclaim(gfp_mask, order,
2201 zonelist, high_zoneidx,
2203 alloc_flags, preferred_zone,
2204 migratetype, &did_some_progress);
2209 * If we failed to make any progress reclaiming, then we are
2210 * running out of options and have to consider going OOM
2212 if (!did_some_progress) {
2213 if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
2214 if (oom_killer_disabled)
2216 page = __alloc_pages_may_oom(gfp_mask, order,
2217 zonelist, high_zoneidx,
2218 nodemask, preferred_zone,
2223 if (!(gfp_mask & __GFP_NOFAIL)) {
2225 * The oom killer is not called for high-order
2226 * allocations that may fail, so if no progress
2227 * is being made, there are no other options and
2228 * retrying is unlikely to help.
2230 if (order > PAGE_ALLOC_COSTLY_ORDER)
2233 * The oom killer is not called for lowmem
2234 * allocations to prevent needlessly killing
2237 if (high_zoneidx < ZONE_NORMAL)
2245 * Suspend converts GFP_KERNEL to __GFP_WAIT which can
2246 * prevent reclaim making forward progress without
2247 * invoking OOM. Bail if we are suspending
2249 if (pm_suspending())
2253 /* Check if we should retry the allocation */
2254 pages_reclaimed += did_some_progress;
2255 if (should_alloc_retry(gfp_mask, order, pages_reclaimed)) {
2256 /* Wait for some write requests to complete then retry */
2257 wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50);
2261 * High-order allocations do not necessarily loop after
2262 * direct reclaim and reclaim/compaction depends on compaction
2263 * being called after reclaim so call directly if necessary
2265 page = __alloc_pages_direct_compact(gfp_mask, order,
2266 zonelist, high_zoneidx,
2268 alloc_flags, preferred_zone,
2269 migratetype, &did_some_progress,
2276 warn_alloc_failed(gfp_mask, order, NULL);
2279 if (kmemcheck_enabled)
2280 kmemcheck_pagealloc_alloc(page, order, gfp_mask);
2286 * This is the 'heart' of the zoned buddy allocator.
2289 __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
2290 struct zonelist *zonelist, nodemask_t *nodemask)
2292 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
2293 struct zone *preferred_zone;
2295 int migratetype = allocflags_to_migratetype(gfp_mask);
2297 gfp_mask &= gfp_allowed_mask;
2299 lockdep_trace_alloc(gfp_mask);
2301 might_sleep_if(gfp_mask & __GFP_WAIT);
2303 if (should_fail_alloc_page(gfp_mask, order))
2307 * Check the zones suitable for the gfp_mask contain at least one
2308 * valid zone. It's possible to have an empty zonelist as a result
2309 * of GFP_THISNODE and a memoryless node
2311 if (unlikely(!zonelist->_zonerefs->zone))
2315 /* The preferred zone is used for statistics later */
2316 first_zones_zonelist(zonelist, high_zoneidx,
2317 nodemask ? : &cpuset_current_mems_allowed,
2319 if (!preferred_zone) {
2324 /* First allocation attempt */
2325 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order,
2326 zonelist, high_zoneidx, ALLOC_WMARK_LOW|ALLOC_CPUSET,
2327 preferred_zone, migratetype);
2328 if (unlikely(!page))
2329 page = __alloc_pages_slowpath(gfp_mask, order,
2330 zonelist, high_zoneidx, nodemask,
2331 preferred_zone, migratetype);
2334 trace_mm_page_alloc(page, order, gfp_mask, migratetype);
2337 EXPORT_SYMBOL(__alloc_pages_nodemask);
2340 * Common helper functions.
2342 unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
2347 * __get_free_pages() returns a 32-bit address, which cannot represent
2350 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
2352 page = alloc_pages(gfp_mask, order);
2355 return (unsigned long) page_address(page);
2357 EXPORT_SYMBOL(__get_free_pages);
2359 unsigned long get_zeroed_page(gfp_t gfp_mask)
2361 return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
2363 EXPORT_SYMBOL(get_zeroed_page);
2365 void __pagevec_free(struct pagevec *pvec)
2367 int i = pagevec_count(pvec);
2370 trace_mm_pagevec_free(pvec->pages[i], pvec->cold);
2371 free_hot_cold_page(pvec->pages[i], pvec->cold);
2375 void __free_pages(struct page *page, unsigned int order)
2377 if (put_page_testzero(page)) {
2379 free_hot_cold_page(page, 0);
2381 __free_pages_ok(page, order);
2385 EXPORT_SYMBOL(__free_pages);
2387 void free_pages(unsigned long addr, unsigned int order)
2390 VM_BUG_ON(!virt_addr_valid((void *)addr));
2391 __free_pages(virt_to_page((void *)addr), order);
2395 EXPORT_SYMBOL(free_pages);
2397 static void *make_alloc_exact(unsigned long addr, unsigned order, size_t size)
2400 unsigned long alloc_end = addr + (PAGE_SIZE << order);
2401 unsigned long used = addr + PAGE_ALIGN(size);
2403 split_page(virt_to_page((void *)addr), order);
2404 while (used < alloc_end) {
2409 return (void *)addr;
2413 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
2414 * @size: the number of bytes to allocate
2415 * @gfp_mask: GFP flags for the allocation
2417 * This function is similar to alloc_pages(), except that it allocates the
2418 * minimum number of pages to satisfy the request. alloc_pages() can only
2419 * allocate memory in power-of-two pages.
2421 * This function is also limited by MAX_ORDER.
2423 * Memory allocated by this function must be released by free_pages_exact().
2425 void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
2427 unsigned int order = get_order(size);
2430 addr = __get_free_pages(gfp_mask, order);
2431 return make_alloc_exact(addr, order, size);
2433 EXPORT_SYMBOL(alloc_pages_exact);
2436 * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
2438 * @nid: the preferred node ID where memory should be allocated
2439 * @size: the number of bytes to allocate
2440 * @gfp_mask: GFP flags for the allocation
2442 * Like alloc_pages_exact(), but try to allocate on node nid first before falling
2444 * Note this is not alloc_pages_exact_node() which allocates on a specific node,
2447 void *alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask)
2449 unsigned order = get_order(size);
2450 struct page *p = alloc_pages_node(nid, gfp_mask, order);
2453 return make_alloc_exact((unsigned long)page_address(p), order, size);
2455 EXPORT_SYMBOL(alloc_pages_exact_nid);
2458 * free_pages_exact - release memory allocated via alloc_pages_exact()
2459 * @virt: the value returned by alloc_pages_exact.
2460 * @size: size of allocation, same value as passed to alloc_pages_exact().
2462 * Release the memory allocated by a previous call to alloc_pages_exact.
2464 void free_pages_exact(void *virt, size_t size)
2466 unsigned long addr = (unsigned long)virt;
2467 unsigned long end = addr + PAGE_ALIGN(size);
2469 while (addr < end) {
2474 EXPORT_SYMBOL(free_pages_exact);
2476 static unsigned int nr_free_zone_pages(int offset)
2481 /* Just pick one node, since fallback list is circular */
2482 unsigned int sum = 0;
2484 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
2486 for_each_zone_zonelist(zone, z, zonelist, offset) {
2487 unsigned long size = zone->present_pages;
2488 unsigned long high = high_wmark_pages(zone);
2497 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
2499 unsigned int nr_free_buffer_pages(void)
2501 return nr_free_zone_pages(gfp_zone(GFP_USER));
2503 EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
2506 * Amount of free RAM allocatable within all zones
2508 unsigned int nr_free_pagecache_pages(void)
2510 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
2513 static inline void show_node(struct zone *zone)
2516 printk("Node %d ", zone_to_nid(zone));
2519 void si_meminfo(struct sysinfo *val)
2521 val->totalram = totalram_pages;
2523 val->freeram = global_page_state(NR_FREE_PAGES);
2524 val->bufferram = nr_blockdev_pages();
2525 val->totalhigh = totalhigh_pages;
2526 val->freehigh = nr_free_highpages();
2527 val->mem_unit = PAGE_SIZE;
2530 EXPORT_SYMBOL(si_meminfo);
2533 void si_meminfo_node(struct sysinfo *val, int nid)
2535 pg_data_t *pgdat = NODE_DATA(nid);
2537 val->totalram = pgdat->node_present_pages;
2538 val->freeram = node_page_state(nid, NR_FREE_PAGES);
2539 #ifdef CONFIG_HIGHMEM
2540 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
2541 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
2547 val->mem_unit = PAGE_SIZE;
2552 * Determine whether the node should be displayed or not, depending on whether
2553 * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
2555 bool skip_free_areas_node(unsigned int flags, int nid)
2559 if (!(flags & SHOW_MEM_FILTER_NODES))
2563 ret = !node_isset(nid, cpuset_current_mems_allowed);
2569 #define K(x) ((x) << (PAGE_SHIFT-10))
2572 * Show free area list (used inside shift_scroll-lock stuff)
2573 * We also calculate the percentage fragmentation. We do this by counting the
2574 * memory on each free list with the exception of the first item on the list.
2575 * Suppresses nodes that are not allowed by current's cpuset if
2576 * SHOW_MEM_FILTER_NODES is passed.
2578 void show_free_areas(unsigned int filter)
2583 for_each_populated_zone(zone) {
2584 if (skip_free_areas_node(filter, zone_to_nid(zone)))
2587 printk("%s per-cpu:\n", zone->name);
2589 for_each_online_cpu(cpu) {
2590 struct per_cpu_pageset *pageset;
2592 pageset = per_cpu_ptr(zone->pageset, cpu);
2594 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
2595 cpu, pageset->pcp.high,
2596 pageset->pcp.batch, pageset->pcp.count);
2600 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
2601 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
2603 " dirty:%lu writeback:%lu unstable:%lu\n"
2604 " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n"
2605 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n",
2606 global_page_state(NR_ACTIVE_ANON),
2607 global_page_state(NR_INACTIVE_ANON),
2608 global_page_state(NR_ISOLATED_ANON),
2609 global_page_state(NR_ACTIVE_FILE),
2610 global_page_state(NR_INACTIVE_FILE),
2611 global_page_state(NR_ISOLATED_FILE),
2612 global_page_state(NR_UNEVICTABLE),
2613 global_page_state(NR_FILE_DIRTY),
2614 global_page_state(NR_WRITEBACK),
2615 global_page_state(NR_UNSTABLE_NFS),
2616 global_page_state(NR_FREE_PAGES),
2617 global_page_state(NR_SLAB_RECLAIMABLE),
2618 global_page_state(NR_SLAB_UNRECLAIMABLE),
2619 global_page_state(NR_FILE_MAPPED),
2620 global_page_state(NR_SHMEM),
2621 global_page_state(NR_PAGETABLE),
2622 global_page_state(NR_BOUNCE));
2624 for_each_populated_zone(zone) {
2627 if (skip_free_areas_node(filter, zone_to_nid(zone)))
2635 " active_anon:%lukB"
2636 " inactive_anon:%lukB"
2637 " active_file:%lukB"
2638 " inactive_file:%lukB"
2639 " unevictable:%lukB"
2640 " isolated(anon):%lukB"
2641 " isolated(file):%lukB"
2648 " slab_reclaimable:%lukB"
2649 " slab_unreclaimable:%lukB"
2650 " kernel_stack:%lukB"
2654 " writeback_tmp:%lukB"
2655 " pages_scanned:%lu"
2656 " all_unreclaimable? %s"
2659 K(zone_page_state(zone, NR_FREE_PAGES)),
2660 K(min_wmark_pages(zone)),
2661 K(low_wmark_pages(zone)),
2662 K(high_wmark_pages(zone)),
2663 K(zone_page_state(zone, NR_ACTIVE_ANON)),
2664 K(zone_page_state(zone, NR_INACTIVE_ANON)),
2665 K(zone_page_state(zone, NR_ACTIVE_FILE)),
2666 K(zone_page_state(zone, NR_INACTIVE_FILE)),
2667 K(zone_page_state(zone, NR_UNEVICTABLE)),
2668 K(zone_page_state(zone, NR_ISOLATED_ANON)),
2669 K(zone_page_state(zone, NR_ISOLATED_FILE)),
2670 K(zone->present_pages),
2671 K(zone_page_state(zone, NR_MLOCK)),
2672 K(zone_page_state(zone, NR_FILE_DIRTY)),
2673 K(zone_page_state(zone, NR_WRITEBACK)),
2674 K(zone_page_state(zone, NR_FILE_MAPPED)),
2675 K(zone_page_state(zone, NR_SHMEM)),
2676 K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)),
2677 K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)),
2678 zone_page_state(zone, NR_KERNEL_STACK) *
2680 K(zone_page_state(zone, NR_PAGETABLE)),
2681 K(zone_page_state(zone, NR_UNSTABLE_NFS)),
2682 K(zone_page_state(zone, NR_BOUNCE)),
2683 K(zone_page_state(zone, NR_WRITEBACK_TEMP)),
2684 zone->pages_scanned,
2685 (zone->all_unreclaimable ? "yes" : "no")
2687 printk("lowmem_reserve[]:");
2688 for (i = 0; i < MAX_NR_ZONES; i++)
2689 printk(" %lu", zone->lowmem_reserve[i]);
2693 for_each_populated_zone(zone) {
2694 unsigned long nr[MAX_ORDER], flags, order, total = 0;
2696 if (skip_free_areas_node(filter, zone_to_nid(zone)))
2699 printk("%s: ", zone->name);
2701 spin_lock_irqsave(&zone->lock, flags);
2702 for (order = 0; order < MAX_ORDER; order++) {
2703 nr[order] = zone->free_area[order].nr_free;
2704 total += nr[order] << order;
2706 spin_unlock_irqrestore(&zone->lock, flags);
2707 for (order = 0; order < MAX_ORDER; order++)
2708 printk("%lu*%lukB ", nr[order], K(1UL) << order);
2709 printk("= %lukB\n", K(total));
2712 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
2714 show_swap_cache_info();
2717 static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
2719 zoneref->zone = zone;
2720 zoneref->zone_idx = zone_idx(zone);
2724 * Builds allocation fallback zone lists.
2726 * Add all populated zones of a node to the zonelist.
2728 static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
2729 int nr_zones, enum zone_type zone_type)
2733 BUG_ON(zone_type >= MAX_NR_ZONES);
2738 zone = pgdat->node_zones + zone_type;
2739 if (populated_zone(zone)) {
2740 zoneref_set_zone(zone,
2741 &zonelist->_zonerefs[nr_zones++]);
2742 check_highest_zone(zone_type);
2745 } while (zone_type);
2752 * 0 = automatic detection of better ordering.
2753 * 1 = order by ([node] distance, -zonetype)
2754 * 2 = order by (-zonetype, [node] distance)
2756 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
2757 * the same zonelist. So only NUMA can configure this param.
2759 #define ZONELIST_ORDER_DEFAULT 0
2760 #define ZONELIST_ORDER_NODE 1
2761 #define ZONELIST_ORDER_ZONE 2
2763 /* zonelist order in the kernel.
2764 * set_zonelist_order() will set this to NODE or ZONE.
2766 static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
2767 static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
2771 /* The value user specified ....changed by config */
2772 static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2773 /* string for sysctl */
2774 #define NUMA_ZONELIST_ORDER_LEN 16
2775 char numa_zonelist_order[16] = "default";
2778 * interface for configure zonelist ordering.
2779 * command line option "numa_zonelist_order"
2780 * = "[dD]efault - default, automatic configuration.
2781 * = "[nN]ode - order by node locality, then by zone within node
2782 * = "[zZ]one - order by zone, then by locality within zone
2785 static int __parse_numa_zonelist_order(char *s)
2787 if (*s == 'd' || *s == 'D') {
2788 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2789 } else if (*s == 'n' || *s == 'N') {
2790 user_zonelist_order = ZONELIST_ORDER_NODE;
2791 } else if (*s == 'z' || *s == 'Z') {
2792 user_zonelist_order = ZONELIST_ORDER_ZONE;
2795 "Ignoring invalid numa_zonelist_order value: "
2802 static __init int setup_numa_zonelist_order(char *s)
2809 ret = __parse_numa_zonelist_order(s);
2811 strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN);
2815 early_param("numa_zonelist_order", setup_numa_zonelist_order);
2818 * sysctl handler for numa_zonelist_order
2820 int numa_zonelist_order_handler(ctl_table *table, int write,
2821 void __user *buffer, size_t *length,
2824 char saved_string[NUMA_ZONELIST_ORDER_LEN];
2826 static DEFINE_MUTEX(zl_order_mutex);
2828 mutex_lock(&zl_order_mutex);
2830 strcpy(saved_string, (char*)table->data);
2831 ret = proc_dostring(table, write, buffer, length, ppos);
2835 int oldval = user_zonelist_order;
2836 if (__parse_numa_zonelist_order((char*)table->data)) {
2838 * bogus value. restore saved string
2840 strncpy((char*)table->data, saved_string,
2841 NUMA_ZONELIST_ORDER_LEN);
2842 user_zonelist_order = oldval;
2843 } else if (oldval != user_zonelist_order) {
2844 mutex_lock(&zonelists_mutex);
2845 build_all_zonelists(NULL);
2846 mutex_unlock(&zonelists_mutex);
2850 mutex_unlock(&zl_order_mutex);
2855 #define MAX_NODE_LOAD (nr_online_nodes)
2856 static int node_load[MAX_NUMNODES];
2859 * find_next_best_node - find the next node that should appear in a given node's fallback list
2860 * @node: node whose fallback list we're appending
2861 * @used_node_mask: nodemask_t of already used nodes
2863 * We use a number of factors to determine which is the next node that should
2864 * appear on a given node's fallback list. The node should not have appeared
2865 * already in @node's fallback list, and it should be the next closest node
2866 * according to the distance array (which contains arbitrary distance values
2867 * from each node to each node in the system), and should also prefer nodes
2868 * with no CPUs, since presumably they'll have very little allocation pressure
2869 * on them otherwise.
2870 * It returns -1 if no node is found.
2872 static int find_next_best_node(int node, nodemask_t *used_node_mask)
2875 int min_val = INT_MAX;
2877 const struct cpumask *tmp = cpumask_of_node(0);
2879 /* Use the local node if we haven't already */
2880 if (!node_isset(node, *used_node_mask)) {
2881 node_set(node, *used_node_mask);
2885 for_each_node_state(n, N_HIGH_MEMORY) {
2887 /* Don't want a node to appear more than once */
2888 if (node_isset(n, *used_node_mask))
2891 /* Use the distance array to find the distance */
2892 val = node_distance(node, n);
2894 /* Penalize nodes under us ("prefer the next node") */
2897 /* Give preference to headless and unused nodes */
2898 tmp = cpumask_of_node(n);
2899 if (!cpumask_empty(tmp))
2900 val += PENALTY_FOR_NODE_WITH_CPUS;
2902 /* Slight preference for less loaded node */
2903 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
2904 val += node_load[n];
2906 if (val < min_val) {
2913 node_set(best_node, *used_node_mask);
2920 * Build zonelists ordered by node and zones within node.
2921 * This results in maximum locality--normal zone overflows into local
2922 * DMA zone, if any--but risks exhausting DMA zone.
2924 static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
2927 struct zonelist *zonelist;
2929 zonelist = &pgdat->node_zonelists[0];
2930 for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
2932 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2934 zonelist->_zonerefs[j].zone = NULL;
2935 zonelist->_zonerefs[j].zone_idx = 0;
2939 * Build gfp_thisnode zonelists
2941 static void build_thisnode_zonelists(pg_data_t *pgdat)
2944 struct zonelist *zonelist;
2946 zonelist = &pgdat->node_zonelists[1];
2947 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
2948 zonelist->_zonerefs[j].zone = NULL;
2949 zonelist->_zonerefs[j].zone_idx = 0;
2953 * Build zonelists ordered by zone and nodes within zones.
2954 * This results in conserving DMA zone[s] until all Normal memory is
2955 * exhausted, but results in overflowing to remote node while memory
2956 * may still exist in local DMA zone.
2958 static int node_order[MAX_NUMNODES];
2960 static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
2963 int zone_type; /* needs to be signed */
2965 struct zonelist *zonelist;
2967 zonelist = &pgdat->node_zonelists[0];
2969 for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
2970 for (j = 0; j < nr_nodes; j++) {
2971 node = node_order[j];
2972 z = &NODE_DATA(node)->node_zones[zone_type];
2973 if (populated_zone(z)) {
2975 &zonelist->_zonerefs[pos++]);
2976 check_highest_zone(zone_type);
2980 zonelist->_zonerefs[pos].zone = NULL;
2981 zonelist->_zonerefs[pos].zone_idx = 0;
2984 static int default_zonelist_order(void)
2987 unsigned long low_kmem_size,total_size;
2991 * ZONE_DMA and ZONE_DMA32 can be very small area in the system.
2992 * If they are really small and used heavily, the system can fall
2993 * into OOM very easily.
2994 * This function detect ZONE_DMA/DMA32 size and configures zone order.
2996 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2999 for_each_online_node(nid) {
3000 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
3001 z = &NODE_DATA(nid)->node_zones[zone_type];
3002 if (populated_zone(z)) {
3003 if (zone_type < ZONE_NORMAL)
3004 low_kmem_size += z->present_pages;
3005 total_size += z->present_pages;
3006 } else if (zone_type == ZONE_NORMAL) {
3008 * If any node has only lowmem, then node order
3009 * is preferred to allow kernel allocations
3010 * locally; otherwise, they can easily infringe
3011 * on other nodes when there is an abundance of
3012 * lowmem available to allocate from.
3014 return ZONELIST_ORDER_NODE;
3018 if (!low_kmem_size || /* there are no DMA area. */
3019 low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
3020 return ZONELIST_ORDER_NODE;
3022 * look into each node's config.
3023 * If there is a node whose DMA/DMA32 memory is very big area on
3024 * local memory, NODE_ORDER may be suitable.
3026 average_size = total_size /
3027 (nodes_weight(node_states[N_HIGH_MEMORY]) + 1);
3028 for_each_online_node(nid) {
3031 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
3032 z = &NODE_DATA(nid)->node_zones[zone_type];
3033 if (populated_zone(z)) {
3034 if (zone_type < ZONE_NORMAL)
3035 low_kmem_size += z->present_pages;
3036 total_size += z->present_pages;
3039 if (low_kmem_size &&
3040 total_size > average_size && /* ignore small node */
3041 low_kmem_size > total_size * 70/100)
3042 return ZONELIST_ORDER_NODE;
3044 return ZONELIST_ORDER_ZONE;
3047 static void set_zonelist_order(void)
3049 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
3050 current_zonelist_order = default_zonelist_order();
3052 current_zonelist_order = user_zonelist_order;
3055 static void build_zonelists(pg_data_t *pgdat)
3059 nodemask_t used_mask;
3060 int local_node, prev_node;
3061 struct zonelist *zonelist;
3062 int order = current_zonelist_order;
3064 /* initialize zonelists */
3065 for (i = 0; i < MAX_ZONELISTS; i++) {
3066 zonelist = pgdat->node_zonelists + i;
3067 zonelist->_zonerefs[0].zone = NULL;
3068 zonelist->_zonerefs[0].zone_idx = 0;
3071 /* NUMA-aware ordering of nodes */
3072 local_node = pgdat->node_id;
3073 load = nr_online_nodes;
3074 prev_node = local_node;
3075 nodes_clear(used_mask);
3077 memset(node_order, 0, sizeof(node_order));
3080 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
3081 int distance = node_distance(local_node, node);
3084 * If another node is sufficiently far away then it is better
3085 * to reclaim pages in a zone before going off node.
3087 if (distance > RECLAIM_DISTANCE)
3088 zone_reclaim_mode = 1;
3091 * We don't want to pressure a particular node.
3092 * So adding penalty to the first node in same
3093 * distance group to make it round-robin.
3095 if (distance != node_distance(local_node, prev_node))
3096 node_load[node] = load;
3100 if (order == ZONELIST_ORDER_NODE)
3101 build_zonelists_in_node_order(pgdat, node);
3103 node_order[j++] = node; /* remember order */
3106 if (order == ZONELIST_ORDER_ZONE) {
3107 /* calculate node order -- i.e., DMA last! */
3108 build_zonelists_in_zone_order(pgdat, j);
3111 build_thisnode_zonelists(pgdat);
3114 /* Construct the zonelist performance cache - see further mmzone.h */
3115 static void build_zonelist_cache(pg_data_t *pgdat)
3117 struct zonelist *zonelist;
3118 struct zonelist_cache *zlc;
3121 zonelist = &pgdat->node_zonelists[0];
3122 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
3123 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
3124 for (z = zonelist->_zonerefs; z->zone; z++)
3125 zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z);
3128 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
3130 * Return node id of node used for "local" allocations.
3131 * I.e., first node id of first zone in arg node's generic zonelist.
3132 * Used for initializing percpu 'numa_mem', which is used primarily
3133 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
3135 int local_memory_node(int node)
3139 (void)first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
3140 gfp_zone(GFP_KERNEL),
3147 #else /* CONFIG_NUMA */
3149 static void set_zonelist_order(void)
3151 current_zonelist_order = ZONELIST_ORDER_ZONE;
3154 static void build_zonelists(pg_data_t *pgdat)
3156 int node, local_node;
3158 struct zonelist *zonelist;
3160 local_node = pgdat->node_id;
3162 zonelist = &pgdat->node_zonelists[0];
3163 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
3166 * Now we build the zonelist so that it contains the zones
3167 * of all the other nodes.
3168 * We don't want to pressure a particular node, so when
3169 * building the zones for node N, we make sure that the
3170 * zones coming right after the local ones are those from
3171 * node N+1 (modulo N)
3173 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
3174 if (!node_online(node))
3176 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
3179 for (node = 0; node < local_node; node++) {
3180 if (!node_online(node))
3182 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
3186 zonelist->_zonerefs[j].zone = NULL;
3187 zonelist->_zonerefs[j].zone_idx = 0;
3190 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
3191 static void build_zonelist_cache(pg_data_t *pgdat)
3193 pgdat->node_zonelists[0].zlcache_ptr = NULL;
3196 #endif /* CONFIG_NUMA */
3199 * Boot pageset table. One per cpu which is going to be used for all
3200 * zones and all nodes. The parameters will be set in such a way
3201 * that an item put on a list will immediately be handed over to
3202 * the buddy list. This is safe since pageset manipulation is done
3203 * with interrupts disabled.
3205 * The boot_pagesets must be kept even after bootup is complete for
3206 * unused processors and/or zones. They do play a role for bootstrapping
3207 * hotplugged processors.
3209 * zoneinfo_show() and maybe other functions do
3210 * not check if the processor is online before following the pageset pointer.
3211 * Other parts of the kernel may not check if the zone is available.
3213 static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch);
3214 static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset);
3215 static void setup_zone_pageset(struct zone *zone);
3218 * Global mutex to protect against size modification of zonelists
3219 * as well as to serialize pageset setup for the new populated zone.
3221 DEFINE_MUTEX(zonelists_mutex);
3223 /* return values int ....just for stop_machine() */
3224 static __init_refok int __build_all_zonelists(void *data)
3230 memset(node_load, 0, sizeof(node_load));
3232 for_each_online_node(nid) {
3233 pg_data_t *pgdat = NODE_DATA(nid);
3235 build_zonelists(pgdat);
3236 build_zonelist_cache(pgdat);
3240 * Initialize the boot_pagesets that are going to be used
3241 * for bootstrapping processors. The real pagesets for
3242 * each zone will be allocated later when the per cpu
3243 * allocator is available.
3245 * boot_pagesets are used also for bootstrapping offline
3246 * cpus if the system is already booted because the pagesets
3247 * are needed to initialize allocators on a specific cpu too.
3248 * F.e. the percpu allocator needs the page allocator which
3249 * needs the percpu allocator in order to allocate its pagesets
3250 * (a chicken-egg dilemma).
3252 for_each_possible_cpu(cpu) {
3253 setup_pageset(&per_cpu(boot_pageset, cpu), 0);
3255 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
3257 * We now know the "local memory node" for each node--
3258 * i.e., the node of the first zone in the generic zonelist.
3259 * Set up numa_mem percpu variable for on-line cpus. During
3260 * boot, only the boot cpu should be on-line; we'll init the
3261 * secondary cpus' numa_mem as they come on-line. During
3262 * node/memory hotplug, we'll fixup all on-line cpus.
3264 if (cpu_online(cpu))
3265 set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
3273 * Called with zonelists_mutex held always
3274 * unless system_state == SYSTEM_BOOTING.
3276 void __ref build_all_zonelists(void *data)
3278 set_zonelist_order();
3280 if (system_state == SYSTEM_BOOTING) {
3281 __build_all_zonelists(NULL);
3282 mminit_verify_zonelist();
3283 cpuset_init_current_mems_allowed();
3285 /* we have to stop all cpus to guarantee there is no user
3287 #ifdef CONFIG_MEMORY_HOTPLUG
3289 setup_zone_pageset((struct zone *)data);
3291 stop_machine(__build_all_zonelists, NULL, NULL);
3292 /* cpuset refresh routine should be here */
3294 vm_total_pages = nr_free_pagecache_pages();
3296 * Disable grouping by mobility if the number of pages in the
3297 * system is too low to allow the mechanism to work. It would be
3298 * more accurate, but expensive to check per-zone. This check is
3299 * made on memory-hotadd so a system can start with mobility
3300 * disabled and enable it later
3302 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
3303 page_group_by_mobility_disabled = 1;
3305 page_group_by_mobility_disabled = 0;
3307 printk("Built %i zonelists in %s order, mobility grouping %s. "
3308 "Total pages: %ld\n",
3310 zonelist_order_name[current_zonelist_order],
3311 page_group_by_mobility_disabled ? "off" : "on",
3314 printk("Policy zone: %s\n", zone_names[policy_zone]);
3319 * Helper functions to size the waitqueue hash table.
3320 * Essentially these want to choose hash table sizes sufficiently
3321 * large so that collisions trying to wait on pages are rare.
3322 * But in fact, the number of active page waitqueues on typical
3323 * systems is ridiculously low, less than 200. So this is even
3324 * conservative, even though it seems large.
3326 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
3327 * waitqueues, i.e. the size of the waitq table given the number of pages.
3329 #define PAGES_PER_WAITQUEUE 256
3331 #ifndef CONFIG_MEMORY_HOTPLUG
3332 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
3334 unsigned long size = 1;
3336 pages /= PAGES_PER_WAITQUEUE;
3338 while (size < pages)
3342 * Once we have dozens or even hundreds of threads sleeping
3343 * on IO we've got bigger problems than wait queue collision.
3344 * Limit the size of the wait table to a reasonable size.
3346 size = min(size, 4096UL);
3348 return max(size, 4UL);
3352 * A zone's size might be changed by hot-add, so it is not possible to determine
3353 * a suitable size for its wait_table. So we use the maximum size now.
3355 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
3357 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
3358 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
3359 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
3361 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
3362 * or more by the traditional way. (See above). It equals:
3364 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
3365 * ia64(16K page size) : = ( 8G + 4M)byte.
3366 * powerpc (64K page size) : = (32G +16M)byte.
3368 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
3375 * This is an integer logarithm so that shifts can be used later
3376 * to extract the more random high bits from the multiplicative
3377 * hash function before the remainder is taken.
3379 static inline unsigned long wait_table_bits(unsigned long size)
3384 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
3387 * Check if a pageblock contains reserved pages
3389 static int pageblock_is_reserved(unsigned long start_pfn, unsigned long end_pfn)
3393 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
3394 if (!pfn_valid_within(pfn) || PageReserved(pfn_to_page(pfn)))
3401 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
3402 * of blocks reserved is based on min_wmark_pages(zone). The memory within
3403 * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
3404 * higher will lead to a bigger reserve which will get freed as contiguous
3405 * blocks as reclaim kicks in
3407 static void setup_zone_migrate_reserve(struct zone *zone)
3409 unsigned long start_pfn, pfn, end_pfn, block_end_pfn;
3411 unsigned long block_migratetype;
3414 /* Get the start pfn, end pfn and the number of blocks to reserve */
3415 start_pfn = zone->zone_start_pfn;
3416 end_pfn = start_pfn + zone->spanned_pages;
3417 reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >>
3421 * Reserve blocks are generally in place to help high-order atomic
3422 * allocations that are short-lived. A min_free_kbytes value that
3423 * would result in more than 2 reserve blocks for atomic allocations
3424 * is assumed to be in place to help anti-fragmentation for the
3425 * future allocation of hugepages at runtime.
3427 reserve = min(2, reserve);
3429 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
3430 if (!pfn_valid(pfn))
3432 page = pfn_to_page(pfn);
3434 /* Watch out for overlapping nodes */
3435 if (page_to_nid(page) != zone_to_nid(zone))
3438 /* Blocks with reserved pages will never free, skip them. */
3439 block_end_pfn = min(pfn + pageblock_nr_pages, end_pfn);
3440 if (pageblock_is_reserved(pfn, block_end_pfn))
3443 block_migratetype = get_pageblock_migratetype(page);
3445 /* If this block is reserved, account for it */
3446 if (reserve > 0 && block_migratetype == MIGRATE_RESERVE) {
3451 /* Suitable for reserving if this block is movable */
3452 if (reserve > 0 && block_migratetype == MIGRATE_MOVABLE) {
3453 set_pageblock_migratetype(page, MIGRATE_RESERVE);
3454 move_freepages_block(zone, page, MIGRATE_RESERVE);
3460 * If the reserve is met and this is a previous reserved block,
3463 if (block_migratetype == MIGRATE_RESERVE) {
3464 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
3465 move_freepages_block(zone, page, MIGRATE_MOVABLE);
3471 * Initially all pages are reserved - free ones are freed
3472 * up by free_all_bootmem() once the early boot process is
3473 * done. Non-atomic initialization, single-pass.
3475 void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
3476 unsigned long start_pfn, enum memmap_context context)
3479 unsigned long end_pfn = start_pfn + size;
3483 if (highest_memmap_pfn < end_pfn - 1)
3484 highest_memmap_pfn = end_pfn - 1;
3486 z = &NODE_DATA(nid)->node_zones[zone];
3487 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
3489 * There can be holes in boot-time mem_map[]s
3490 * handed to this function. They do not
3491 * exist on hotplugged memory.
3493 if (context == MEMMAP_EARLY) {
3494 if (!early_pfn_valid(pfn))
3496 if (!early_pfn_in_nid(pfn, nid))
3499 page = pfn_to_page(pfn);
3500 set_page_links(page, zone, nid, pfn);
3501 mminit_verify_page_links(page, zone, nid, pfn);
3502 init_page_count(page);
3503 reset_page_mapcount(page);
3504 SetPageReserved(page);
3506 * Mark the block movable so that blocks are reserved for
3507 * movable at startup. This will force kernel allocations
3508 * to reserve their blocks rather than leaking throughout
3509 * the address space during boot when many long-lived
3510 * kernel allocations are made. Later some blocks near
3511 * the start are marked MIGRATE_RESERVE by
3512 * setup_zone_migrate_reserve()
3514 * bitmap is created for zone's valid pfn range. but memmap
3515 * can be created for invalid pages (for alignment)
3516 * check here not to call set_pageblock_migratetype() against
3519 if ((z->zone_start_pfn <= pfn)
3520 && (pfn < z->zone_start_pfn + z->spanned_pages)
3521 && !(pfn & (pageblock_nr_pages - 1)))
3522 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
3524 INIT_LIST_HEAD(&page->lru);
3525 #ifdef WANT_PAGE_VIRTUAL
3526 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
3527 if (!is_highmem_idx(zone))
3528 set_page_address(page, __va(pfn << PAGE_SHIFT));
3533 static void __meminit zone_init_free_lists(struct zone *zone)
3536 for_each_migratetype_order(order, t) {
3537 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
3538 zone->free_area[order].nr_free = 0;
3542 #ifndef __HAVE_ARCH_MEMMAP_INIT
3543 #define memmap_init(size, nid, zone, start_pfn) \
3544 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
3547 static int zone_batchsize(struct zone *zone)
3553 * The per-cpu-pages pools are set to around 1000th of the
3554 * size of the zone. But no more than 1/2 of a meg.
3556 * OK, so we don't know how big the cache is. So guess.
3558 batch = zone->present_pages / 1024;
3559 if (batch * PAGE_SIZE > 512 * 1024)
3560 batch = (512 * 1024) / PAGE_SIZE;
3561 batch /= 4; /* We effectively *= 4 below */
3566 * Clamp the batch to a 2^n - 1 value. Having a power
3567 * of 2 value was found to be more likely to have
3568 * suboptimal cache aliasing properties in some cases.
3570 * For example if 2 tasks are alternately allocating
3571 * batches of pages, one task can end up with a lot
3572 * of pages of one half of the possible page colors
3573 * and the other with pages of the other colors.
3575 batch = rounddown_pow_of_two(batch + batch/2) - 1;
3580 /* The deferral and batching of frees should be suppressed under NOMMU
3583 * The problem is that NOMMU needs to be able to allocate large chunks
3584 * of contiguous memory as there's no hardware page translation to
3585 * assemble apparent contiguous memory from discontiguous pages.
3587 * Queueing large contiguous runs of pages for batching, however,
3588 * causes the pages to actually be freed in smaller chunks. As there
3589 * can be a significant delay between the individual batches being
3590 * recycled, this leads to the once large chunks of space being
3591 * fragmented and becoming unavailable for high-order allocations.
3597 static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
3599 struct per_cpu_pages *pcp;
3602 memset(p, 0, sizeof(*p));
3606 pcp->high = 6 * batch;
3607 pcp->batch = max(1UL, 1 * batch);
3608 for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
3609 INIT_LIST_HEAD(&pcp->lists[migratetype]);
3613 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
3614 * to the value high for the pageset p.
3617 static void setup_pagelist_highmark(struct per_cpu_pageset *p,
3620 struct per_cpu_pages *pcp;
3624 pcp->batch = max(1UL, high/4);
3625 if ((high/4) > (PAGE_SHIFT * 8))
3626 pcp->batch = PAGE_SHIFT * 8;
3629 static void setup_zone_pageset(struct zone *zone)
3633 zone->pageset = alloc_percpu(struct per_cpu_pageset);
3635 for_each_possible_cpu(cpu) {
3636 struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu);
3638 setup_pageset(pcp, zone_batchsize(zone));
3640 if (percpu_pagelist_fraction)
3641 setup_pagelist_highmark(pcp,
3642 (zone->present_pages /
3643 percpu_pagelist_fraction));
3648 * Allocate per cpu pagesets and initialize them.
3649 * Before this call only boot pagesets were available.
3651 void __init setup_per_cpu_pageset(void)
3655 for_each_populated_zone(zone)
3656 setup_zone_pageset(zone);
3659 static noinline __init_refok
3660 int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
3663 struct pglist_data *pgdat = zone->zone_pgdat;
3667 * The per-page waitqueue mechanism uses hashed waitqueues
3670 zone->wait_table_hash_nr_entries =
3671 wait_table_hash_nr_entries(zone_size_pages);
3672 zone->wait_table_bits =
3673 wait_table_bits(zone->wait_table_hash_nr_entries);
3674 alloc_size = zone->wait_table_hash_nr_entries
3675 * sizeof(wait_queue_head_t);
3677 if (!slab_is_available()) {
3678 zone->wait_table = (wait_queue_head_t *)
3679 alloc_bootmem_node_nopanic(pgdat, alloc_size);
3682 * This case means that a zone whose size was 0 gets new memory
3683 * via memory hot-add.
3684 * But it may be the case that a new node was hot-added. In
3685 * this case vmalloc() will not be able to use this new node's
3686 * memory - this wait_table must be initialized to use this new
3687 * node itself as well.
3688 * To use this new node's memory, further consideration will be
3691 zone->wait_table = vmalloc(alloc_size);
3693 if (!zone->wait_table)
3696 for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
3697 init_waitqueue_head(zone->wait_table + i);
3702 static int __zone_pcp_update(void *data)
3704 struct zone *zone = data;
3706 unsigned long batch = zone_batchsize(zone), flags;
3708 for_each_possible_cpu(cpu) {
3709 struct per_cpu_pageset *pset;
3710 struct per_cpu_pages *pcp;
3712 pset = per_cpu_ptr(zone->pageset, cpu);
3715 local_irq_save(flags);
3716 free_pcppages_bulk(zone, pcp->count, pcp);
3717 setup_pageset(pset, batch);
3718 local_irq_restore(flags);
3723 void zone_pcp_update(struct zone *zone)
3725 stop_machine(__zone_pcp_update, zone, NULL);
3728 static __meminit void zone_pcp_init(struct zone *zone)
3731 * per cpu subsystem is not up at this point. The following code
3732 * relies on the ability of the linker to provide the
3733 * offset of a (static) per cpu variable into the per cpu area.
3735 zone->pageset = &boot_pageset;
3737 if (zone->present_pages)
3738 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n",
3739 zone->name, zone->present_pages,
3740 zone_batchsize(zone));
3743 __meminit int init_currently_empty_zone(struct zone *zone,
3744 unsigned long zone_start_pfn,
3746 enum memmap_context context)
3748 struct pglist_data *pgdat = zone->zone_pgdat;
3750 ret = zone_wait_table_init(zone, size);
3753 pgdat->nr_zones = zone_idx(zone) + 1;
3755 zone->zone_start_pfn = zone_start_pfn;
3757 mminit_dprintk(MMINIT_TRACE, "memmap_init",
3758 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
3760 (unsigned long)zone_idx(zone),
3761 zone_start_pfn, (zone_start_pfn + size));
3763 zone_init_free_lists(zone);
3768 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3770 * Basic iterator support. Return the first range of PFNs for a node
3771 * Note: nid == MAX_NUMNODES returns first region regardless of node
3773 static int __meminit first_active_region_index_in_nid(int nid)
3777 for (i = 0; i < nr_nodemap_entries; i++)
3778 if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
3785 * Basic iterator support. Return the next active range of PFNs for a node
3786 * Note: nid == MAX_NUMNODES returns next region regardless of node
3788 static int __meminit next_active_region_index_in_nid(int index, int nid)
3790 for (index = index + 1; index < nr_nodemap_entries; index++)
3791 if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
3797 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
3799 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
3800 * Architectures may implement their own version but if add_active_range()
3801 * was used and there are no special requirements, this is a convenient
3804 int __meminit __early_pfn_to_nid(unsigned long pfn)
3808 for (i = 0; i < nr_nodemap_entries; i++) {
3809 unsigned long start_pfn = early_node_map[i].start_pfn;
3810 unsigned long end_pfn = early_node_map[i].end_pfn;
3812 if (start_pfn <= pfn && pfn < end_pfn)
3813 return early_node_map[i].nid;
3815 /* This is a memory hole */
3818 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
3820 int __meminit early_pfn_to_nid(unsigned long pfn)
3824 nid = __early_pfn_to_nid(pfn);
3827 /* just returns 0 */
3831 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
3832 bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
3836 nid = __early_pfn_to_nid(pfn);
3837 if (nid >= 0 && nid != node)
3843 /* Basic iterator support to walk early_node_map[] */
3844 #define for_each_active_range_index_in_nid(i, nid) \
3845 for (i = first_active_region_index_in_nid(nid); i != -1; \
3846 i = next_active_region_index_in_nid(i, nid))
3849 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
3850 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
3851 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
3853 * If an architecture guarantees that all ranges registered with
3854 * add_active_ranges() contain no holes and may be freed, this
3855 * this function may be used instead of calling free_bootmem() manually.
3857 void __init free_bootmem_with_active_regions(int nid,
3858 unsigned long max_low_pfn)
3862 for_each_active_range_index_in_nid(i, nid) {
3863 unsigned long size_pages = 0;
3864 unsigned long end_pfn = early_node_map[i].end_pfn;
3866 if (early_node_map[i].start_pfn >= max_low_pfn)
3869 if (end_pfn > max_low_pfn)
3870 end_pfn = max_low_pfn;
3872 size_pages = end_pfn - early_node_map[i].start_pfn;
3873 free_bootmem_node(NODE_DATA(early_node_map[i].nid),
3874 PFN_PHYS(early_node_map[i].start_pfn),
3875 size_pages << PAGE_SHIFT);
3879 #ifdef CONFIG_HAVE_MEMBLOCK
3881 * Basic iterator support. Return the last range of PFNs for a node
3882 * Note: nid == MAX_NUMNODES returns last region regardless of node
3884 static int __meminit last_active_region_index_in_nid(int nid)
3888 for (i = nr_nodemap_entries - 1; i >= 0; i--)
3889 if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
3896 * Basic iterator support. Return the previous active range of PFNs for a node
3897 * Note: nid == MAX_NUMNODES returns next region regardless of node
3899 static int __meminit previous_active_region_index_in_nid(int index, int nid)
3901 for (index = index - 1; index >= 0; index--)
3902 if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
3908 #define for_each_active_range_index_in_nid_reverse(i, nid) \
3909 for (i = last_active_region_index_in_nid(nid); i != -1; \
3910 i = previous_active_region_index_in_nid(i, nid))
3912 u64 __init find_memory_core_early(int nid, u64 size, u64 align,
3913 u64 goal, u64 limit)
3917 /* Need to go over early_node_map to find out good range for node */
3918 for_each_active_range_index_in_nid_reverse(i, nid) {
3920 u64 ei_start, ei_last;
3921 u64 final_start, final_end;
3923 ei_last = early_node_map[i].end_pfn;
3924 ei_last <<= PAGE_SHIFT;
3925 ei_start = early_node_map[i].start_pfn;
3926 ei_start <<= PAGE_SHIFT;
3928 final_start = max(ei_start, goal);
3929 final_end = min(ei_last, limit);
3931 if (final_start >= final_end)
3934 addr = memblock_find_in_range(final_start, final_end, size, align);
3936 if (addr == MEMBLOCK_ERROR)
3942 return MEMBLOCK_ERROR;
3946 int __init add_from_early_node_map(struct range *range, int az,
3947 int nr_range, int nid)
3952 /* need to go over early_node_map to find out good range for node */
3953 for_each_active_range_index_in_nid(i, nid) {
3954 start = early_node_map[i].start_pfn;
3955 end = early_node_map[i].end_pfn;
3956 nr_range = add_range(range, az, nr_range, start, end);
3961 void __init work_with_active_regions(int nid, work_fn_t work_fn, void *data)
3966 for_each_active_range_index_in_nid(i, nid) {
3967 ret = work_fn(early_node_map[i].start_pfn,
3968 early_node_map[i].end_pfn, data);
3974 * sparse_memory_present_with_active_regions - Call memory_present for each active range
3975 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
3977 * If an architecture guarantees that all ranges registered with
3978 * add_active_ranges() contain no holes and may be freed, this
3979 * function may be used instead of calling memory_present() manually.
3981 void __init sparse_memory_present_with_active_regions(int nid)
3985 for_each_active_range_index_in_nid(i, nid)
3986 memory_present(early_node_map[i].nid,
3987 early_node_map[i].start_pfn,
3988 early_node_map[i].end_pfn);
3992 * get_pfn_range_for_nid - Return the start and end page frames for a node
3993 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
3994 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
3995 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
3997 * It returns the start and end page frame of a node based on information
3998 * provided by an arch calling add_active_range(). If called for a node
3999 * with no available memory, a warning is printed and the start and end
4002 void __meminit get_pfn_range_for_nid(unsigned int nid,
4003 unsigned long *start_pfn, unsigned long *end_pfn)
4009 for_each_active_range_index_in_nid(i, nid) {
4010 *start_pfn = min(*start_pfn, early_node_map[i].start_pfn);
4011 *end_pfn = max(*end_pfn, early_node_map[i].end_pfn);
4014 if (*start_pfn == -1UL)
4019 * This finds a zone that can be used for ZONE_MOVABLE pages. The
4020 * assumption is made that zones within a node are ordered in monotonic
4021 * increasing memory addresses so that the "highest" populated zone is used
4023 static void __init find_usable_zone_for_movable(void)
4026 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
4027 if (zone_index == ZONE_MOVABLE)
4030 if (arch_zone_highest_possible_pfn[zone_index] >
4031 arch_zone_lowest_possible_pfn[zone_index])
4035 VM_BUG_ON(zone_index == -1);
4036 movable_zone = zone_index;
4040 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
4041 * because it is sized independent of architecture. Unlike the other zones,
4042 * the starting point for ZONE_MOVABLE is not fixed. It may be different
4043 * in each node depending on the size of each node and how evenly kernelcore
4044 * is distributed. This helper function adjusts the zone ranges
4045 * provided by the architecture for a given node by using the end of the
4046 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
4047 * zones within a node are in order of monotonic increases memory addresses
4049 static void __meminit adjust_zone_range_for_zone_movable(int nid,
4050 unsigned long zone_type,
4051 unsigned long node_start_pfn,
4052 unsigned long node_end_pfn,
4053 unsigned long *zone_start_pfn,
4054 unsigned long *zone_end_pfn)
4056 /* Only adjust if ZONE_MOVABLE is on this node */
4057 if (zone_movable_pfn[nid]) {
4058 /* Size ZONE_MOVABLE */
4059 if (zone_type == ZONE_MOVABLE) {
4060 *zone_start_pfn = zone_movable_pfn[nid];
4061 *zone_end_pfn = min(node_end_pfn,
4062 arch_zone_highest_possible_pfn[movable_zone]);
4064 /* Adjust for ZONE_MOVABLE starting within this range */
4065 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
4066 *zone_end_pfn > zone_movable_pfn[nid]) {
4067 *zone_end_pfn = zone_movable_pfn[nid];
4069 /* Check if this whole range is within ZONE_MOVABLE */
4070 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
4071 *zone_start_pfn = *zone_end_pfn;
4076 * Return the number of pages a zone spans in a node, including holes
4077 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
4079 static unsigned long __meminit zone_spanned_pages_in_node(int nid,
4080 unsigned long zone_type,
4081 unsigned long *ignored)
4083 unsigned long node_start_pfn, node_end_pfn;
4084 unsigned long zone_start_pfn, zone_end_pfn;
4086 /* Get the start and end of the node and zone */
4087 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
4088 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
4089 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
4090 adjust_zone_range_for_zone_movable(nid, zone_type,
4091 node_start_pfn, node_end_pfn,
4092 &zone_start_pfn, &zone_end_pfn);
4094 /* Check that this node has pages within the zone's required range */
4095 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
4098 /* Move the zone boundaries inside the node if necessary */
4099 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
4100 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
4102 /* Return the spanned pages */
4103 return zone_end_pfn - zone_start_pfn;
4107 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
4108 * then all holes in the requested range will be accounted for.
4110 unsigned long __meminit __absent_pages_in_range(int nid,
4111 unsigned long range_start_pfn,
4112 unsigned long range_end_pfn)
4115 unsigned long prev_end_pfn = 0, hole_pages = 0;
4116 unsigned long start_pfn;
4118 /* Find the end_pfn of the first active range of pfns in the node */
4119 i = first_active_region_index_in_nid(nid);
4123 prev_end_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
4125 /* Account for ranges before physical memory on this node */
4126 if (early_node_map[i].start_pfn > range_start_pfn)
4127 hole_pages = prev_end_pfn - range_start_pfn;
4129 /* Find all holes for the zone within the node */
4130 for (; i != -1; i = next_active_region_index_in_nid(i, nid)) {
4132 /* No need to continue if prev_end_pfn is outside the zone */
4133 if (prev_end_pfn >= range_end_pfn)
4136 /* Make sure the end of the zone is not within the hole */
4137 start_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
4138 prev_end_pfn = max(prev_end_pfn, range_start_pfn);
4140 /* Update the hole size cound and move on */
4141 if (start_pfn > range_start_pfn) {
4142 BUG_ON(prev_end_pfn > start_pfn);
4143 hole_pages += start_pfn - prev_end_pfn;
4145 prev_end_pfn = early_node_map[i].end_pfn;
4148 /* Account for ranges past physical memory on this node */
4149 if (range_end_pfn > prev_end_pfn)
4150 hole_pages += range_end_pfn -
4151 max(range_start_pfn, prev_end_pfn);
4157 * absent_pages_in_range - Return number of page frames in holes within a range
4158 * @start_pfn: The start PFN to start searching for holes
4159 * @end_pfn: The end PFN to stop searching for holes
4161 * It returns the number of pages frames in memory holes within a range.
4163 unsigned long __init absent_pages_in_range(unsigned long start_pfn,
4164 unsigned long end_pfn)
4166 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
4169 /* Return the number of page frames in holes in a zone on a node */
4170 static unsigned long __meminit zone_absent_pages_in_node(int nid,
4171 unsigned long zone_type,
4172 unsigned long *ignored)
4174 unsigned long node_start_pfn, node_end_pfn;
4175 unsigned long zone_start_pfn, zone_end_pfn;
4177 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
4178 zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type],
4180 zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type],
4183 adjust_zone_range_for_zone_movable(nid, zone_type,
4184 node_start_pfn, node_end_pfn,
4185 &zone_start_pfn, &zone_end_pfn);
4186 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
4190 static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
4191 unsigned long zone_type,
4192 unsigned long *zones_size)
4194 return zones_size[zone_type];
4197 static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
4198 unsigned long zone_type,
4199 unsigned long *zholes_size)
4204 return zholes_size[zone_type];
4209 static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
4210 unsigned long *zones_size, unsigned long *zholes_size)
4212 unsigned long realtotalpages, totalpages = 0;
4215 for (i = 0; i < MAX_NR_ZONES; i++)
4216 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
4218 pgdat->node_spanned_pages = totalpages;
4220 realtotalpages = totalpages;
4221 for (i = 0; i < MAX_NR_ZONES; i++)
4223 zone_absent_pages_in_node(pgdat->node_id, i,
4225 pgdat->node_present_pages = realtotalpages;
4226 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
4230 #ifndef CONFIG_SPARSEMEM
4232 * Calculate the size of the zone->blockflags rounded to an unsigned long
4233 * Start by making sure zonesize is a multiple of pageblock_order by rounding
4234 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
4235 * round what is now in bits to nearest long in bits, then return it in
4238 static unsigned long __init usemap_size(unsigned long zonesize)
4240 unsigned long usemapsize;
4242 usemapsize = roundup(zonesize, pageblock_nr_pages);
4243 usemapsize = usemapsize >> pageblock_order;
4244 usemapsize *= NR_PAGEBLOCK_BITS;
4245 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
4247 return usemapsize / 8;
4250 static void __init setup_usemap(struct pglist_data *pgdat,
4251 struct zone *zone, unsigned long zonesize)
4253 unsigned long usemapsize = usemap_size(zonesize);
4254 zone->pageblock_flags = NULL;
4256 zone->pageblock_flags = alloc_bootmem_node_nopanic(pgdat,
4260 static inline void setup_usemap(struct pglist_data *pgdat,
4261 struct zone *zone, unsigned long zonesize) {}
4262 #endif /* CONFIG_SPARSEMEM */
4264 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
4266 /* Return a sensible default order for the pageblock size. */
4267 static inline int pageblock_default_order(void)
4269 if (HPAGE_SHIFT > PAGE_SHIFT)
4270 return HUGETLB_PAGE_ORDER;
4275 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
4276 static inline void __init set_pageblock_order(unsigned int order)
4278 /* Check that pageblock_nr_pages has not already been setup */
4279 if (pageblock_order)
4283 * Assume the largest contiguous order of interest is a huge page.
4284 * This value may be variable depending on boot parameters on IA64
4286 pageblock_order = order;
4288 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4291 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
4292 * and pageblock_default_order() are unused as pageblock_order is set
4293 * at compile-time. See include/linux/pageblock-flags.h for the values of
4294 * pageblock_order based on the kernel config
4296 static inline int pageblock_default_order(unsigned int order)
4300 #define set_pageblock_order(x) do {} while (0)
4302 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4305 * Set up the zone data structures:
4306 * - mark all pages reserved
4307 * - mark all memory queues empty
4308 * - clear the memory bitmaps
4310 static void __paginginit free_area_init_core(struct pglist_data *pgdat,
4311 unsigned long *zones_size, unsigned long *zholes_size)
4314 int nid = pgdat->node_id;
4315 unsigned long zone_start_pfn = pgdat->node_start_pfn;
4318 pgdat_resize_init(pgdat);
4319 pgdat->nr_zones = 0;
4320 init_waitqueue_head(&pgdat->kswapd_wait);
4321 pgdat->kswapd_max_order = 0;
4322 pgdat_page_cgroup_init(pgdat);
4324 for (j = 0; j < MAX_NR_ZONES; j++) {
4325 struct zone *zone = pgdat->node_zones + j;
4326 unsigned long size, realsize, memmap_pages;
4329 size = zone_spanned_pages_in_node(nid, j, zones_size);
4330 realsize = size - zone_absent_pages_in_node(nid, j,
4334 * Adjust realsize so that it accounts for how much memory
4335 * is used by this zone for memmap. This affects the watermark
4336 * and per-cpu initialisations
4339 PAGE_ALIGN(size * sizeof(struct page)) >> PAGE_SHIFT;
4340 if (realsize >= memmap_pages) {
4341 realsize -= memmap_pages;
4344 " %s zone: %lu pages used for memmap\n",
4345 zone_names[j], memmap_pages);
4348 " %s zone: %lu pages exceeds realsize %lu\n",
4349 zone_names[j], memmap_pages, realsize);
4351 /* Account for reserved pages */
4352 if (j == 0 && realsize > dma_reserve) {
4353 realsize -= dma_reserve;
4354 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
4355 zone_names[0], dma_reserve);
4358 if (!is_highmem_idx(j))
4359 nr_kernel_pages += realsize;
4360 nr_all_pages += realsize;
4362 zone->spanned_pages = size;
4363 zone->present_pages = realsize;
4366 zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
4368 zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100;
4370 zone->name = zone_names[j];
4371 spin_lock_init(&zone->lock);
4372 spin_lock_init(&zone->lru_lock);
4373 zone_seqlock_init(zone);
4374 zone->zone_pgdat = pgdat;
4376 zone_pcp_init(zone);
4378 INIT_LIST_HEAD(&zone->lru[l].list);
4379 zone->reclaim_stat.recent_rotated[0] = 0;
4380 zone->reclaim_stat.recent_rotated[1] = 0;
4381 zone->reclaim_stat.recent_scanned[0] = 0;
4382 zone->reclaim_stat.recent_scanned[1] = 0;
4383 zap_zone_vm_stats(zone);
4388 set_pageblock_order(pageblock_default_order());
4389 setup_usemap(pgdat, zone, size);
4390 ret = init_currently_empty_zone(zone, zone_start_pfn,
4391 size, MEMMAP_EARLY);
4393 memmap_init(size, nid, j, zone_start_pfn);
4394 zone_start_pfn += size;
4398 static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
4400 /* Skip empty nodes */
4401 if (!pgdat->node_spanned_pages)
4404 #ifdef CONFIG_FLAT_NODE_MEM_MAP
4405 /* ia64 gets its own node_mem_map, before this, without bootmem */
4406 if (!pgdat->node_mem_map) {
4407 unsigned long size, start, end;
4411 * The zone's endpoints aren't required to be MAX_ORDER
4412 * aligned but the node_mem_map endpoints must be in order
4413 * for the buddy allocator to function correctly.
4415 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
4416 end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
4417 end = ALIGN(end, MAX_ORDER_NR_PAGES);
4418 size = (end - start) * sizeof(struct page);
4419 map = alloc_remap(pgdat->node_id, size);
4421 map = alloc_bootmem_node_nopanic(pgdat, size);
4422 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
4424 #ifndef CONFIG_NEED_MULTIPLE_NODES
4426 * With no DISCONTIG, the global mem_map is just set as node 0's
4428 if (pgdat == NODE_DATA(0)) {
4429 mem_map = NODE_DATA(0)->node_mem_map;
4430 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
4431 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
4432 mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
4433 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4436 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
4439 void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
4440 unsigned long node_start_pfn, unsigned long *zholes_size)
4442 pg_data_t *pgdat = NODE_DATA(nid);
4444 pgdat->node_id = nid;
4445 pgdat->node_start_pfn = node_start_pfn;
4446 calculate_node_totalpages(pgdat, zones_size, zholes_size);
4448 alloc_node_mem_map(pgdat);
4449 #ifdef CONFIG_FLAT_NODE_MEM_MAP
4450 printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
4451 nid, (unsigned long)pgdat,
4452 (unsigned long)pgdat->node_mem_map);
4455 free_area_init_core(pgdat, zones_size, zholes_size);
4458 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
4460 #if MAX_NUMNODES > 1
4462 * Figure out the number of possible node ids.
4464 static void __init setup_nr_node_ids(void)
4467 unsigned int highest = 0;
4469 for_each_node_mask(node, node_possible_map)
4471 nr_node_ids = highest + 1;
4474 static inline void setup_nr_node_ids(void)
4480 * add_active_range - Register a range of PFNs backed by physical memory
4481 * @nid: The node ID the range resides on
4482 * @start_pfn: The start PFN of the available physical memory
4483 * @end_pfn: The end PFN of the available physical memory
4485 * These ranges are stored in an early_node_map[] and later used by
4486 * free_area_init_nodes() to calculate zone sizes and holes. If the
4487 * range spans a memory hole, it is up to the architecture to ensure
4488 * the memory is not freed by the bootmem allocator. If possible
4489 * the range being registered will be merged with existing ranges.
4491 void __init add_active_range(unsigned int nid, unsigned long start_pfn,
4492 unsigned long end_pfn)
4496 mminit_dprintk(MMINIT_TRACE, "memory_register",
4497 "Entering add_active_range(%d, %#lx, %#lx) "
4498 "%d entries of %d used\n",
4499 nid, start_pfn, end_pfn,
4500 nr_nodemap_entries, MAX_ACTIVE_REGIONS);
4502 mminit_validate_memmodel_limits(&start_pfn, &end_pfn);
4504 /* Merge with existing active regions if possible */
4505 for (i = 0; i < nr_nodemap_entries; i++) {
4506 if (early_node_map[i].nid != nid)
4509 /* Skip if an existing region covers this new one */
4510 if (start_pfn >= early_node_map[i].start_pfn &&
4511 end_pfn <= early_node_map[i].end_pfn)
4514 /* Merge forward if suitable */
4515 if (start_pfn <= early_node_map[i].end_pfn &&
4516 end_pfn > early_node_map[i].end_pfn) {
4517 early_node_map[i].end_pfn = end_pfn;
4521 /* Merge backward if suitable */
4522 if (start_pfn < early_node_map[i].start_pfn &&
4523 end_pfn >= early_node_map[i].start_pfn) {
4524 early_node_map[i].start_pfn = start_pfn;
4529 /* Check that early_node_map is large enough */
4530 if (i >= MAX_ACTIVE_REGIONS) {
4531 printk(KERN_CRIT "More than %d memory regions, truncating\n",
4532 MAX_ACTIVE_REGIONS);
4536 early_node_map[i].nid = nid;
4537 early_node_map[i].start_pfn = start_pfn;
4538 early_node_map[i].end_pfn = end_pfn;
4539 nr_nodemap_entries = i + 1;
4543 * remove_active_range - Shrink an existing registered range of PFNs
4544 * @nid: The node id the range is on that should be shrunk
4545 * @start_pfn: The new PFN of the range
4546 * @end_pfn: The new PFN of the range
4548 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
4549 * The map is kept near the end physical page range that has already been
4550 * registered. This function allows an arch to shrink an existing registered
4553 void __init remove_active_range(unsigned int nid, unsigned long start_pfn,
4554 unsigned long end_pfn)
4559 printk(KERN_DEBUG "remove_active_range (%d, %lu, %lu)\n",
4560 nid, start_pfn, end_pfn);
4562 /* Find the old active region end and shrink */
4563 for_each_active_range_index_in_nid(i, nid) {
4564 if (early_node_map[i].start_pfn >= start_pfn &&
4565 early_node_map[i].end_pfn <= end_pfn) {
4567 early_node_map[i].start_pfn = 0;
4568 early_node_map[i].end_pfn = 0;
4572 if (early_node_map[i].start_pfn < start_pfn &&
4573 early_node_map[i].end_pfn > start_pfn) {
4574 unsigned long temp_end_pfn = early_node_map[i].end_pfn;
4575 early_node_map[i].end_pfn = start_pfn;
4576 if (temp_end_pfn > end_pfn)
4577 add_active_range(nid, end_pfn, temp_end_pfn);
4580 if (early_node_map[i].start_pfn >= start_pfn &&
4581 early_node_map[i].end_pfn > end_pfn &&
4582 early_node_map[i].start_pfn < end_pfn) {
4583 early_node_map[i].start_pfn = end_pfn;
4591 /* remove the blank ones */
4592 for (i = nr_nodemap_entries - 1; i > 0; i--) {
4593 if (early_node_map[i].nid != nid)
4595 if (early_node_map[i].end_pfn)
4597 /* we found it, get rid of it */
4598 for (j = i; j < nr_nodemap_entries - 1; j++)
4599 memcpy(&early_node_map[j], &early_node_map[j+1],
4600 sizeof(early_node_map[j]));
4601 j = nr_nodemap_entries - 1;
4602 memset(&early_node_map[j], 0, sizeof(early_node_map[j]));
4603 nr_nodemap_entries--;
4608 * remove_all_active_ranges - Remove all currently registered regions
4610 * During discovery, it may be found that a table like SRAT is invalid
4611 * and an alternative discovery method must be used. This function removes
4612 * all currently registered regions.
4614 void __init remove_all_active_ranges(void)
4616 memset(early_node_map, 0, sizeof(early_node_map));
4617 nr_nodemap_entries = 0;
4620 /* Compare two active node_active_regions */
4621 static int __init cmp_node_active_region(const void *a, const void *b)
4623 struct node_active_region *arange = (struct node_active_region *)a;
4624 struct node_active_region *brange = (struct node_active_region *)b;
4626 /* Done this way to avoid overflows */
4627 if (arange->start_pfn > brange->start_pfn)
4629 if (arange->start_pfn < brange->start_pfn)
4635 /* sort the node_map by start_pfn */
4636 void __init sort_node_map(void)
4638 sort(early_node_map, (size_t)nr_nodemap_entries,
4639 sizeof(struct node_active_region),
4640 cmp_node_active_region, NULL);
4643 /* Find the lowest pfn for a node */
4644 static unsigned long __init find_min_pfn_for_node(int nid)
4647 unsigned long min_pfn = ULONG_MAX;
4649 /* Assuming a sorted map, the first range found has the starting pfn */
4650 for_each_active_range_index_in_nid(i, nid)
4651 min_pfn = min(min_pfn, early_node_map[i].start_pfn);
4653 if (min_pfn == ULONG_MAX) {
4655 "Could not find start_pfn for node %d\n", nid);
4663 * find_min_pfn_with_active_regions - Find the minimum PFN registered
4665 * It returns the minimum PFN based on information provided via
4666 * add_active_range().
4668 unsigned long __init find_min_pfn_with_active_regions(void)
4670 return find_min_pfn_for_node(MAX_NUMNODES);
4674 * early_calculate_totalpages()
4675 * Sum pages in active regions for movable zone.
4676 * Populate N_HIGH_MEMORY for calculating usable_nodes.
4678 static unsigned long __init early_calculate_totalpages(void)
4681 unsigned long totalpages = 0;
4683 for (i = 0; i < nr_nodemap_entries; i++) {
4684 unsigned long pages = early_node_map[i].end_pfn -
4685 early_node_map[i].start_pfn;
4686 totalpages += pages;
4688 node_set_state(early_node_map[i].nid, N_HIGH_MEMORY);
4694 * Find the PFN the Movable zone begins in each node. Kernel memory
4695 * is spread evenly between nodes as long as the nodes have enough
4696 * memory. When they don't, some nodes will have more kernelcore than
4699 static void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn)
4702 unsigned long usable_startpfn;
4703 unsigned long kernelcore_node, kernelcore_remaining;
4704 /* save the state before borrow the nodemask */
4705 nodemask_t saved_node_state = node_states[N_HIGH_MEMORY];
4706 unsigned long totalpages = early_calculate_totalpages();
4707 int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]);
4710 * If movablecore was specified, calculate what size of
4711 * kernelcore that corresponds so that memory usable for
4712 * any allocation type is evenly spread. If both kernelcore
4713 * and movablecore are specified, then the value of kernelcore
4714 * will be used for required_kernelcore if it's greater than
4715 * what movablecore would have allowed.
4717 if (required_movablecore) {
4718 unsigned long corepages;
4721 * Round-up so that ZONE_MOVABLE is at least as large as what
4722 * was requested by the user
4724 required_movablecore =
4725 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
4726 corepages = totalpages - required_movablecore;
4728 required_kernelcore = max(required_kernelcore, corepages);
4731 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
4732 if (!required_kernelcore)
4735 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
4736 find_usable_zone_for_movable();
4737 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
4740 /* Spread kernelcore memory as evenly as possible throughout nodes */
4741 kernelcore_node = required_kernelcore / usable_nodes;
4742 for_each_node_state(nid, N_HIGH_MEMORY) {
4744 * Recalculate kernelcore_node if the division per node
4745 * now exceeds what is necessary to satisfy the requested
4746 * amount of memory for the kernel
4748 if (required_kernelcore < kernelcore_node)
4749 kernelcore_node = required_kernelcore / usable_nodes;
4752 * As the map is walked, we track how much memory is usable
4753 * by the kernel using kernelcore_remaining. When it is
4754 * 0, the rest of the node is usable by ZONE_MOVABLE
4756 kernelcore_remaining = kernelcore_node;
4758 /* Go through each range of PFNs within this node */
4759 for_each_active_range_index_in_nid(i, nid) {
4760 unsigned long start_pfn, end_pfn;
4761 unsigned long size_pages;
4763 start_pfn = max(early_node_map[i].start_pfn,
4764 zone_movable_pfn[nid]);
4765 end_pfn = early_node_map[i].end_pfn;
4766 if (start_pfn >= end_pfn)
4769 /* Account for what is only usable for kernelcore */
4770 if (start_pfn < usable_startpfn) {
4771 unsigned long kernel_pages;
4772 kernel_pages = min(end_pfn, usable_startpfn)
4775 kernelcore_remaining -= min(kernel_pages,
4776 kernelcore_remaining);
4777 required_kernelcore -= min(kernel_pages,
4778 required_kernelcore);
4780 /* Continue if range is now fully accounted */
4781 if (end_pfn <= usable_startpfn) {
4784 * Push zone_movable_pfn to the end so
4785 * that if we have to rebalance
4786 * kernelcore across nodes, we will
4787 * not double account here
4789 zone_movable_pfn[nid] = end_pfn;
4792 start_pfn = usable_startpfn;
4796 * The usable PFN range for ZONE_MOVABLE is from
4797 * start_pfn->end_pfn. Calculate size_pages as the
4798 * number of pages used as kernelcore
4800 size_pages = end_pfn - start_pfn;
4801 if (size_pages > kernelcore_remaining)
4802 size_pages = kernelcore_remaining;
4803 zone_movable_pfn[nid] = start_pfn + size_pages;
4806 * Some kernelcore has been met, update counts and
4807 * break if the kernelcore for this node has been
4810 required_kernelcore -= min(required_kernelcore,
4812 kernelcore_remaining -= size_pages;
4813 if (!kernelcore_remaining)
4819 * If there is still required_kernelcore, we do another pass with one
4820 * less node in the count. This will push zone_movable_pfn[nid] further
4821 * along on the nodes that still have memory until kernelcore is
4825 if (usable_nodes && required_kernelcore > usable_nodes)
4828 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
4829 for (nid = 0; nid < MAX_NUMNODES; nid++)
4830 zone_movable_pfn[nid] =
4831 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
4834 /* restore the node_state */
4835 node_states[N_HIGH_MEMORY] = saved_node_state;
4838 /* Any regular memory on that node ? */
4839 static void check_for_regular_memory(pg_data_t *pgdat)
4841 #ifdef CONFIG_HIGHMEM
4842 enum zone_type zone_type;
4844 for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) {
4845 struct zone *zone = &pgdat->node_zones[zone_type];
4846 if (zone->present_pages)
4847 node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY);
4853 * free_area_init_nodes - Initialise all pg_data_t and zone data
4854 * @max_zone_pfn: an array of max PFNs for each zone
4856 * This will call free_area_init_node() for each active node in the system.
4857 * Using the page ranges provided by add_active_range(), the size of each
4858 * zone in each node and their holes is calculated. If the maximum PFN
4859 * between two adjacent zones match, it is assumed that the zone is empty.
4860 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
4861 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
4862 * starts where the previous one ended. For example, ZONE_DMA32 starts
4863 * at arch_max_dma_pfn.
4865 void __init free_area_init_nodes(unsigned long *max_zone_pfn)
4870 /* Sort early_node_map as initialisation assumes it is sorted */
4873 /* Record where the zone boundaries are */
4874 memset(arch_zone_lowest_possible_pfn, 0,
4875 sizeof(arch_zone_lowest_possible_pfn));
4876 memset(arch_zone_highest_possible_pfn, 0,
4877 sizeof(arch_zone_highest_possible_pfn));
4878 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
4879 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
4880 for (i = 1; i < MAX_NR_ZONES; i++) {
4881 if (i == ZONE_MOVABLE)
4883 arch_zone_lowest_possible_pfn[i] =
4884 arch_zone_highest_possible_pfn[i-1];
4885 arch_zone_highest_possible_pfn[i] =
4886 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
4888 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
4889 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
4891 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
4892 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
4893 find_zone_movable_pfns_for_nodes(zone_movable_pfn);
4895 /* Print out the zone ranges */
4896 printk("Zone PFN ranges:\n");
4897 for (i = 0; i < MAX_NR_ZONES; i++) {
4898 if (i == ZONE_MOVABLE)
4900 printk(" %-8s ", zone_names[i]);
4901 if (arch_zone_lowest_possible_pfn[i] ==
4902 arch_zone_highest_possible_pfn[i])
4905 printk("%0#10lx -> %0#10lx\n",
4906 arch_zone_lowest_possible_pfn[i],
4907 arch_zone_highest_possible_pfn[i]);
4910 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
4911 printk("Movable zone start PFN for each node\n");
4912 for (i = 0; i < MAX_NUMNODES; i++) {
4913 if (zone_movable_pfn[i])
4914 printk(" Node %d: %lu\n", i, zone_movable_pfn[i]);
4917 /* Print out the early_node_map[] */
4918 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries);
4919 for (i = 0; i < nr_nodemap_entries; i++)
4920 printk(" %3d: %0#10lx -> %0#10lx\n", early_node_map[i].nid,
4921 early_node_map[i].start_pfn,
4922 early_node_map[i].end_pfn);
4924 /* Initialise every node */
4925 mminit_verify_pageflags_layout();
4926 setup_nr_node_ids();
4927 for_each_online_node(nid) {
4928 pg_data_t *pgdat = NODE_DATA(nid);
4929 free_area_init_node(nid, NULL,
4930 find_min_pfn_for_node(nid), NULL);
4932 /* Any memory on that node */
4933 if (pgdat->node_present_pages)
4934 node_set_state(nid, N_HIGH_MEMORY);
4935 check_for_regular_memory(pgdat);
4939 static int __init cmdline_parse_core(char *p, unsigned long *core)
4941 unsigned long long coremem;
4945 coremem = memparse(p, &p);
4946 *core = coremem >> PAGE_SHIFT;
4948 /* Paranoid check that UL is enough for the coremem value */
4949 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
4955 * kernelcore=size sets the amount of memory for use for allocations that
4956 * cannot be reclaimed or migrated.
4958 static int __init cmdline_parse_kernelcore(char *p)
4960 return cmdline_parse_core(p, &required_kernelcore);
4964 * movablecore=size sets the amount of memory for use for allocations that
4965 * can be reclaimed or migrated.
4967 static int __init cmdline_parse_movablecore(char *p)
4969 return cmdline_parse_core(p, &required_movablecore);
4972 early_param("kernelcore", cmdline_parse_kernelcore);
4973 early_param("movablecore", cmdline_parse_movablecore);
4975 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4978 * set_dma_reserve - set the specified number of pages reserved in the first zone
4979 * @new_dma_reserve: The number of pages to mark reserved
4981 * The per-cpu batchsize and zone watermarks are determined by present_pages.
4982 * In the DMA zone, a significant percentage may be consumed by kernel image
4983 * and other unfreeable allocations which can skew the watermarks badly. This
4984 * function may optionally be used to account for unfreeable pages in the
4985 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
4986 * smaller per-cpu batchsize.
4988 void __init set_dma_reserve(unsigned long new_dma_reserve)
4990 dma_reserve = new_dma_reserve;
4993 void __init free_area_init(unsigned long *zones_size)
4995 free_area_init_node(0, zones_size,
4996 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
4999 static int page_alloc_cpu_notify(struct notifier_block *self,
5000 unsigned long action, void *hcpu)
5002 int cpu = (unsigned long)hcpu;
5004 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
5008 * Spill the event counters of the dead processor
5009 * into the current processors event counters.
5010 * This artificially elevates the count of the current
5013 vm_events_fold_cpu(cpu);
5016 * Zero the differential counters of the dead processor
5017 * so that the vm statistics are consistent.
5019 * This is only okay since the processor is dead and cannot
5020 * race with what we are doing.
5022 refresh_cpu_vm_stats(cpu);
5027 void __init page_alloc_init(void)
5029 hotcpu_notifier(page_alloc_cpu_notify, 0);
5033 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
5034 * or min_free_kbytes changes.
5036 static void calculate_totalreserve_pages(void)
5038 struct pglist_data *pgdat;
5039 unsigned long reserve_pages = 0;
5040 enum zone_type i, j;
5042 for_each_online_pgdat(pgdat) {
5043 for (i = 0; i < MAX_NR_ZONES; i++) {
5044 struct zone *zone = pgdat->node_zones + i;
5045 unsigned long max = 0;
5047 /* Find valid and maximum lowmem_reserve in the zone */
5048 for (j = i; j < MAX_NR_ZONES; j++) {
5049 if (zone->lowmem_reserve[j] > max)
5050 max = zone->lowmem_reserve[j];
5053 /* we treat the high watermark as reserved pages. */
5054 max += high_wmark_pages(zone);
5056 if (max > zone->present_pages)
5057 max = zone->present_pages;
5058 reserve_pages += max;
5061 totalreserve_pages = reserve_pages;
5065 * setup_per_zone_lowmem_reserve - called whenever
5066 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
5067 * has a correct pages reserved value, so an adequate number of
5068 * pages are left in the zone after a successful __alloc_pages().
5070 static void setup_per_zone_lowmem_reserve(void)
5072 struct pglist_data *pgdat;
5073 enum zone_type j, idx;
5075 for_each_online_pgdat(pgdat) {
5076 for (j = 0; j < MAX_NR_ZONES; j++) {
5077 struct zone *zone = pgdat->node_zones + j;
5078 unsigned long present_pages = zone->present_pages;
5080 zone->lowmem_reserve[j] = 0;
5084 struct zone *lower_zone;
5088 if (sysctl_lowmem_reserve_ratio[idx] < 1)
5089 sysctl_lowmem_reserve_ratio[idx] = 1;
5091 lower_zone = pgdat->node_zones + idx;
5092 lower_zone->lowmem_reserve[j] = present_pages /
5093 sysctl_lowmem_reserve_ratio[idx];
5094 present_pages += lower_zone->present_pages;
5099 /* update totalreserve_pages */
5100 calculate_totalreserve_pages();
5104 * setup_per_zone_wmarks - called when min_free_kbytes changes
5105 * or when memory is hot-{added|removed}
5107 * Ensures that the watermark[min,low,high] values for each zone are set
5108 * correctly with respect to min_free_kbytes.
5110 void setup_per_zone_wmarks(void)
5112 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
5113 unsigned long lowmem_pages = 0;
5115 unsigned long flags;
5117 /* Calculate total number of !ZONE_HIGHMEM pages */
5118 for_each_zone(zone) {
5119 if (!is_highmem(zone))
5120 lowmem_pages += zone->present_pages;
5123 for_each_zone(zone) {
5126 spin_lock_irqsave(&zone->lock, flags);
5127 tmp = (u64)pages_min * zone->present_pages;
5128 do_div(tmp, lowmem_pages);
5129 if (is_highmem(zone)) {
5131 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
5132 * need highmem pages, so cap pages_min to a small
5135 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
5136 * deltas controls asynch page reclaim, and so should
5137 * not be capped for highmem.
5141 min_pages = zone->present_pages / 1024;
5142 if (min_pages < SWAP_CLUSTER_MAX)
5143 min_pages = SWAP_CLUSTER_MAX;
5144 if (min_pages > 128)
5146 zone->watermark[WMARK_MIN] = min_pages;
5149 * If it's a lowmem zone, reserve a number of pages
5150 * proportionate to the zone's size.
5152 zone->watermark[WMARK_MIN] = tmp;
5155 zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + (tmp >> 2);
5156 zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1);
5157 setup_zone_migrate_reserve(zone);
5158 spin_unlock_irqrestore(&zone->lock, flags);
5161 /* update totalreserve_pages */
5162 calculate_totalreserve_pages();
5166 * The inactive anon list should be small enough that the VM never has to
5167 * do too much work, but large enough that each inactive page has a chance
5168 * to be referenced again before it is swapped out.
5170 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
5171 * INACTIVE_ANON pages on this zone's LRU, maintained by the
5172 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
5173 * the anonymous pages are kept on the inactive list.
5176 * memory ratio inactive anon
5177 * -------------------------------------
5186 static void __meminit calculate_zone_inactive_ratio(struct zone *zone)
5188 unsigned int gb, ratio;
5190 /* Zone size in gigabytes */
5191 gb = zone->present_pages >> (30 - PAGE_SHIFT);
5193 ratio = int_sqrt(10 * gb);
5197 zone->inactive_ratio = ratio;
5200 static void __meminit setup_per_zone_inactive_ratio(void)
5205 calculate_zone_inactive_ratio(zone);
5209 * Initialise min_free_kbytes.
5211 * For small machines we want it small (128k min). For large machines
5212 * we want it large (64MB max). But it is not linear, because network
5213 * bandwidth does not increase linearly with machine size. We use
5215 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
5216 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
5232 int __meminit init_per_zone_wmark_min(void)
5234 unsigned long lowmem_kbytes;
5236 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
5238 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
5239 if (min_free_kbytes < 128)
5240 min_free_kbytes = 128;
5241 if (min_free_kbytes > 65536)
5242 min_free_kbytes = 65536;
5243 setup_per_zone_wmarks();
5244 refresh_zone_stat_thresholds();
5245 setup_per_zone_lowmem_reserve();
5246 setup_per_zone_inactive_ratio();
5249 module_init(init_per_zone_wmark_min)
5252 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
5253 * that we can call two helper functions whenever min_free_kbytes
5256 int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
5257 void __user *buffer, size_t *length, loff_t *ppos)
5259 proc_dointvec(table, write, buffer, length, ppos);
5261 setup_per_zone_wmarks();
5266 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
5267 void __user *buffer, size_t *length, loff_t *ppos)
5272 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
5277 zone->min_unmapped_pages = (zone->present_pages *
5278 sysctl_min_unmapped_ratio) / 100;
5282 int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
5283 void __user *buffer, size_t *length, loff_t *ppos)
5288 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
5293 zone->min_slab_pages = (zone->present_pages *
5294 sysctl_min_slab_ratio) / 100;
5300 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
5301 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
5302 * whenever sysctl_lowmem_reserve_ratio changes.
5304 * The reserve ratio obviously has absolutely no relation with the
5305 * minimum watermarks. The lowmem reserve ratio can only make sense
5306 * if in function of the boot time zone sizes.
5308 int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
5309 void __user *buffer, size_t *length, loff_t *ppos)
5311 proc_dointvec_minmax(table, write, buffer, length, ppos);
5312 setup_per_zone_lowmem_reserve();
5317 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
5318 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
5319 * can have before it gets flushed back to buddy allocator.
5322 int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
5323 void __user *buffer, size_t *length, loff_t *ppos)
5329 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
5330 if (!write || (ret == -EINVAL))
5332 for_each_populated_zone(zone) {
5333 for_each_possible_cpu(cpu) {
5335 high = zone->present_pages / percpu_pagelist_fraction;
5336 setup_pagelist_highmark(
5337 per_cpu_ptr(zone->pageset, cpu), high);
5343 int hashdist = HASHDIST_DEFAULT;
5346 static int __init set_hashdist(char *str)
5350 hashdist = simple_strtoul(str, &str, 0);
5353 __setup("hashdist=", set_hashdist);
5357 * allocate a large system hash table from bootmem
5358 * - it is assumed that the hash table must contain an exact power-of-2
5359 * quantity of entries
5360 * - limit is the number of hash buckets, not the total allocation size
5362 void *__init alloc_large_system_hash(const char *tablename,
5363 unsigned long bucketsize,
5364 unsigned long numentries,
5367 unsigned int *_hash_shift,
5368 unsigned int *_hash_mask,
5369 unsigned long limit)
5371 unsigned long long max = limit;
5372 unsigned long log2qty, size;
5375 /* allow the kernel cmdline to have a say */
5377 /* round applicable memory size up to nearest megabyte */
5378 numentries = nr_kernel_pages;
5379 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
5380 numentries >>= 20 - PAGE_SHIFT;
5381 numentries <<= 20 - PAGE_SHIFT;
5383 /* limit to 1 bucket per 2^scale bytes of low memory */
5384 if (scale > PAGE_SHIFT)
5385 numentries >>= (scale - PAGE_SHIFT);
5387 numentries <<= (PAGE_SHIFT - scale);
5389 /* Make sure we've got at least a 0-order allocation.. */
5390 if (unlikely(flags & HASH_SMALL)) {
5391 /* Makes no sense without HASH_EARLY */
5392 WARN_ON(!(flags & HASH_EARLY));
5393 if (!(numentries >> *_hash_shift)) {
5394 numentries = 1UL << *_hash_shift;
5395 BUG_ON(!numentries);
5397 } else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
5398 numentries = PAGE_SIZE / bucketsize;
5400 numentries = roundup_pow_of_two(numentries);
5402 /* limit allocation size to 1/16 total memory by default */
5404 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
5405 do_div(max, bucketsize);
5408 if (numentries > max)
5411 log2qty = ilog2(numentries);
5414 size = bucketsize << log2qty;
5415 if (flags & HASH_EARLY)
5416 table = alloc_bootmem_nopanic(size);
5418 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
5421 * If bucketsize is not a power-of-two, we may free
5422 * some pages at the end of hash table which
5423 * alloc_pages_exact() automatically does
5425 if (get_order(size) < MAX_ORDER) {
5426 table = alloc_pages_exact(size, GFP_ATOMIC);
5427 kmemleak_alloc(table, size, 1, GFP_ATOMIC);
5430 } while (!table && size > PAGE_SIZE && --log2qty);
5433 panic("Failed to allocate %s hash table\n", tablename);
5435 printk(KERN_INFO "%s hash table entries: %ld (order: %d, %lu bytes)\n",
5438 ilog2(size) - PAGE_SHIFT,
5442 *_hash_shift = log2qty;
5444 *_hash_mask = (1 << log2qty) - 1;
5449 /* Return a pointer to the bitmap storing bits affecting a block of pages */
5450 static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
5453 #ifdef CONFIG_SPARSEMEM
5454 return __pfn_to_section(pfn)->pageblock_flags;
5456 return zone->pageblock_flags;
5457 #endif /* CONFIG_SPARSEMEM */
5460 static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
5462 #ifdef CONFIG_SPARSEMEM
5463 pfn &= (PAGES_PER_SECTION-1);
5464 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
5466 pfn = pfn - zone->zone_start_pfn;
5467 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
5468 #endif /* CONFIG_SPARSEMEM */
5472 * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
5473 * @page: The page within the block of interest
5474 * @start_bitidx: The first bit of interest to retrieve
5475 * @end_bitidx: The last bit of interest
5476 * returns pageblock_bits flags
5478 unsigned long get_pageblock_flags_group(struct page *page,
5479 int start_bitidx, int end_bitidx)
5482 unsigned long *bitmap;
5483 unsigned long pfn, bitidx;
5484 unsigned long flags = 0;
5485 unsigned long value = 1;
5487 zone = page_zone(page);
5488 pfn = page_to_pfn(page);
5489 bitmap = get_pageblock_bitmap(zone, pfn);
5490 bitidx = pfn_to_bitidx(zone, pfn);
5492 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
5493 if (test_bit(bitidx + start_bitidx, bitmap))
5500 * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
5501 * @page: The page within the block of interest
5502 * @start_bitidx: The first bit of interest
5503 * @end_bitidx: The last bit of interest
5504 * @flags: The flags to set
5506 void set_pageblock_flags_group(struct page *page, unsigned long flags,
5507 int start_bitidx, int end_bitidx)
5510 unsigned long *bitmap;
5511 unsigned long pfn, bitidx;
5512 unsigned long value = 1;
5514 zone = page_zone(page);
5515 pfn = page_to_pfn(page);
5516 bitmap = get_pageblock_bitmap(zone, pfn);
5517 bitidx = pfn_to_bitidx(zone, pfn);
5518 VM_BUG_ON(pfn < zone->zone_start_pfn);
5519 VM_BUG_ON(pfn >= zone->zone_start_pfn + zone->spanned_pages);
5521 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
5523 __set_bit(bitidx + start_bitidx, bitmap);
5525 __clear_bit(bitidx + start_bitidx, bitmap);
5529 * This is designed as sub function...plz see page_isolation.c also.
5530 * set/clear page block's type to be ISOLATE.
5531 * page allocater never alloc memory from ISOLATE block.
5535 __count_immobile_pages(struct zone *zone, struct page *page, int count)
5537 unsigned long pfn, iter, found;
5539 * For avoiding noise data, lru_add_drain_all() should be called
5540 * If ZONE_MOVABLE, the zone never contains immobile pages
5542 if (zone_idx(zone) == ZONE_MOVABLE)
5545 if (get_pageblock_migratetype(page) == MIGRATE_MOVABLE)
5548 pfn = page_to_pfn(page);
5549 for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) {
5550 unsigned long check = pfn + iter;
5552 if (!pfn_valid_within(check))
5555 page = pfn_to_page(check);
5556 if (!page_count(page)) {
5557 if (PageBuddy(page))
5558 iter += (1 << page_order(page)) - 1;
5564 * If there are RECLAIMABLE pages, we need to check it.
5565 * But now, memory offline itself doesn't call shrink_slab()
5566 * and it still to be fixed.
5569 * If the page is not RAM, page_count()should be 0.
5570 * we don't need more check. This is an _used_ not-movable page.
5572 * The problematic thing here is PG_reserved pages. PG_reserved
5573 * is set to both of a memory hole page and a _used_ kernel
5582 bool is_pageblock_removable_nolock(struct page *page)
5584 struct zone *zone = page_zone(page);
5585 return __count_immobile_pages(zone, page, 0);
5588 int set_migratetype_isolate(struct page *page)
5591 unsigned long flags, pfn;
5592 struct memory_isolate_notify arg;
5596 zone = page_zone(page);
5598 spin_lock_irqsave(&zone->lock, flags);
5600 pfn = page_to_pfn(page);
5601 arg.start_pfn = pfn;
5602 arg.nr_pages = pageblock_nr_pages;
5603 arg.pages_found = 0;
5606 * It may be possible to isolate a pageblock even if the
5607 * migratetype is not MIGRATE_MOVABLE. The memory isolation
5608 * notifier chain is used by balloon drivers to return the
5609 * number of pages in a range that are held by the balloon
5610 * driver to shrink memory. If all the pages are accounted for
5611 * by balloons, are free, or on the LRU, isolation can continue.
5612 * Later, for example, when memory hotplug notifier runs, these
5613 * pages reported as "can be isolated" should be isolated(freed)
5614 * by the balloon driver through the memory notifier chain.
5616 notifier_ret = memory_isolate_notify(MEM_ISOLATE_COUNT, &arg);
5617 notifier_ret = notifier_to_errno(notifier_ret);
5621 * FIXME: Now, memory hotplug doesn't call shrink_slab() by itself.
5622 * We just check MOVABLE pages.
5624 if (__count_immobile_pages(zone, page, arg.pages_found))
5628 * immobile means "not-on-lru" paes. If immobile is larger than
5629 * removable-by-driver pages reported by notifier, we'll fail.
5634 set_pageblock_migratetype(page, MIGRATE_ISOLATE);
5635 move_freepages_block(zone, page, MIGRATE_ISOLATE);
5638 spin_unlock_irqrestore(&zone->lock, flags);
5644 void unset_migratetype_isolate(struct page *page)
5647 unsigned long flags;
5648 zone = page_zone(page);
5649 spin_lock_irqsave(&zone->lock, flags);
5650 if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE)
5652 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
5653 move_freepages_block(zone, page, MIGRATE_MOVABLE);
5655 spin_unlock_irqrestore(&zone->lock, flags);
5658 #ifdef CONFIG_MEMORY_HOTREMOVE
5660 * All pages in the range must be isolated before calling this.
5663 __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
5669 unsigned long flags;
5670 /* find the first valid pfn */
5671 for (pfn = start_pfn; pfn < end_pfn; pfn++)
5676 zone = page_zone(pfn_to_page(pfn));
5677 spin_lock_irqsave(&zone->lock, flags);
5679 while (pfn < end_pfn) {
5680 if (!pfn_valid(pfn)) {
5684 page = pfn_to_page(pfn);
5685 BUG_ON(page_count(page));
5686 BUG_ON(!PageBuddy(page));
5687 order = page_order(page);
5688 #ifdef CONFIG_DEBUG_VM
5689 printk(KERN_INFO "remove from free list %lx %d %lx\n",
5690 pfn, 1 << order, end_pfn);
5692 list_del(&page->lru);
5693 rmv_page_order(page);
5694 zone->free_area[order].nr_free--;
5695 __mod_zone_page_state(zone, NR_FREE_PAGES,
5697 for (i = 0; i < (1 << order); i++)
5698 SetPageReserved((page+i));
5699 pfn += (1 << order);
5701 spin_unlock_irqrestore(&zone->lock, flags);
5705 #ifdef CONFIG_MEMORY_FAILURE
5706 bool is_free_buddy_page(struct page *page)
5708 struct zone *zone = page_zone(page);
5709 unsigned long pfn = page_to_pfn(page);
5710 unsigned long flags;
5713 spin_lock_irqsave(&zone->lock, flags);
5714 for (order = 0; order < MAX_ORDER; order++) {
5715 struct page *page_head = page - (pfn & ((1 << order) - 1));
5717 if (PageBuddy(page_head) && page_order(page_head) >= order)
5720 spin_unlock_irqrestore(&zone->lock, flags);
5722 return order < MAX_ORDER;
5726 static struct trace_print_flags pageflag_names[] = {
5727 {1UL << PG_locked, "locked" },
5728 {1UL << PG_error, "error" },
5729 {1UL << PG_referenced, "referenced" },
5730 {1UL << PG_uptodate, "uptodate" },
5731 {1UL << PG_dirty, "dirty" },
5732 {1UL << PG_lru, "lru" },
5733 {1UL << PG_active, "active" },
5734 {1UL << PG_slab, "slab" },
5735 {1UL << PG_owner_priv_1, "owner_priv_1" },
5736 {1UL << PG_arch_1, "arch_1" },
5737 {1UL << PG_reserved, "reserved" },
5738 {1UL << PG_private, "private" },
5739 {1UL << PG_private_2, "private_2" },
5740 {1UL << PG_writeback, "writeback" },
5741 #ifdef CONFIG_PAGEFLAGS_EXTENDED
5742 {1UL << PG_head, "head" },
5743 {1UL << PG_tail, "tail" },
5745 {1UL << PG_compound, "compound" },
5747 {1UL << PG_swapcache, "swapcache" },
5748 {1UL << PG_mappedtodisk, "mappedtodisk" },
5749 {1UL << PG_reclaim, "reclaim" },
5750 {1UL << PG_swapbacked, "swapbacked" },
5751 {1UL << PG_unevictable, "unevictable" },
5753 {1UL << PG_mlocked, "mlocked" },
5755 #ifdef CONFIG_ARCH_USES_PG_UNCACHED
5756 {1UL << PG_uncached, "uncached" },
5758 #ifdef CONFIG_MEMORY_FAILURE
5759 {1UL << PG_hwpoison, "hwpoison" },
5764 static void dump_page_flags(unsigned long flags)
5766 const char *delim = "";
5770 printk(KERN_ALERT "page flags: %#lx(", flags);
5772 /* remove zone id */
5773 flags &= (1UL << NR_PAGEFLAGS) - 1;
5775 for (i = 0; pageflag_names[i].name && flags; i++) {
5777 mask = pageflag_names[i].mask;
5778 if ((flags & mask) != mask)
5782 printk("%s%s", delim, pageflag_names[i].name);
5786 /* check for left over flags */
5788 printk("%s%#lx", delim, flags);
5793 void dump_page(struct page *page)
5796 "page:%p count:%d mapcount:%d mapping:%p index:%#lx\n",
5797 page, atomic_read(&page->_count), page_mapcount(page),
5798 page->mapping, page->index);
5799 dump_page_flags(page->flags);
5800 mem_cgroup_print_bad_page(page);