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>
60 #include <linux/page-debug-flags.h>
62 #include <asm/tlbflush.h>
63 #include <asm/div64.h>
66 #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
67 DEFINE_PER_CPU(int, numa_node);
68 EXPORT_PER_CPU_SYMBOL(numa_node);
71 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
73 * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
74 * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
75 * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
76 * defined in <linux/topology.h>.
78 DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */
79 EXPORT_PER_CPU_SYMBOL(_numa_mem_);
83 * Array of node states.
85 nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
86 [N_POSSIBLE] = NODE_MASK_ALL,
87 [N_ONLINE] = { { [0] = 1UL } },
89 [N_NORMAL_MEMORY] = { { [0] = 1UL } },
91 [N_HIGH_MEMORY] = { { [0] = 1UL } },
93 [N_CPU] = { { [0] = 1UL } },
96 EXPORT_SYMBOL(node_states);
98 unsigned long totalram_pages __read_mostly;
99 unsigned long totalreserve_pages __read_mostly;
101 * When calculating the number of globally allowed dirty pages, there
102 * is a certain number of per-zone reserves that should not be
103 * considered dirtyable memory. This is the sum of those reserves
104 * over all existing zones that contribute dirtyable memory.
106 unsigned long dirty_balance_reserve __read_mostly;
108 int percpu_pagelist_fraction;
109 gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK;
111 #ifdef CONFIG_PM_SLEEP
113 * The following functions are used by the suspend/hibernate code to temporarily
114 * change gfp_allowed_mask in order to avoid using I/O during memory allocations
115 * while devices are suspended. To avoid races with the suspend/hibernate code,
116 * they should always be called with pm_mutex held (gfp_allowed_mask also should
117 * only be modified with pm_mutex held, unless the suspend/hibernate code is
118 * guaranteed not to run in parallel with that modification).
121 static gfp_t saved_gfp_mask;
123 void pm_restore_gfp_mask(void)
125 WARN_ON(!mutex_is_locked(&pm_mutex));
126 if (saved_gfp_mask) {
127 gfp_allowed_mask = saved_gfp_mask;
132 void pm_restrict_gfp_mask(void)
134 WARN_ON(!mutex_is_locked(&pm_mutex));
135 WARN_ON(saved_gfp_mask);
136 saved_gfp_mask = gfp_allowed_mask;
137 gfp_allowed_mask &= ~GFP_IOFS;
140 bool pm_suspended_storage(void)
142 if ((gfp_allowed_mask & GFP_IOFS) == GFP_IOFS)
146 #endif /* CONFIG_PM_SLEEP */
148 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
149 int pageblock_order __read_mostly;
152 static void __free_pages_ok(struct page *page, unsigned int order);
155 * results with 256, 32 in the lowmem_reserve sysctl:
156 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
157 * 1G machine -> (16M dma, 784M normal, 224M high)
158 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
159 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
160 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
162 * TBD: should special case ZONE_DMA32 machines here - in those we normally
163 * don't need any ZONE_NORMAL reservation
165 int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
166 #ifdef CONFIG_ZONE_DMA
169 #ifdef CONFIG_ZONE_DMA32
172 #ifdef CONFIG_HIGHMEM
178 EXPORT_SYMBOL(totalram_pages);
180 static char * const zone_names[MAX_NR_ZONES] = {
181 #ifdef CONFIG_ZONE_DMA
184 #ifdef CONFIG_ZONE_DMA32
188 #ifdef CONFIG_HIGHMEM
195 * Try to keep at least this much lowmem free. Do not allow normal
196 * allocations below this point, only high priority ones. Automatically
197 * tuned according to the amount of memory in the system.
199 int min_free_kbytes = 1024;
202 * Extra memory for the system to try freeing between the min and
203 * low watermarks. Useful for workloads that require low latency
204 * memory allocations in bursts larger than the normal gap between
207 int extra_free_kbytes;
209 static unsigned long __meminitdata nr_kernel_pages;
210 static unsigned long __meminitdata nr_all_pages;
211 static unsigned long __meminitdata dma_reserve;
213 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
214 static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
215 static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
216 static unsigned long __initdata required_kernelcore;
217 static unsigned long __initdata required_movablecore;
218 static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
220 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
222 EXPORT_SYMBOL(movable_zone);
223 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
226 int nr_node_ids __read_mostly = MAX_NUMNODES;
227 int nr_online_nodes __read_mostly = 1;
228 EXPORT_SYMBOL(nr_node_ids);
229 EXPORT_SYMBOL(nr_online_nodes);
232 int page_group_by_mobility_disabled __read_mostly;
234 static void set_pageblock_migratetype(struct page *page, int migratetype)
237 if (unlikely(page_group_by_mobility_disabled))
238 migratetype = MIGRATE_UNMOVABLE;
240 set_pageblock_flags_group(page, (unsigned long)migratetype,
241 PB_migrate, PB_migrate_end);
244 bool oom_killer_disabled __read_mostly;
246 #ifdef CONFIG_DEBUG_VM
247 static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
251 unsigned long pfn = page_to_pfn(page);
254 seq = zone_span_seqbegin(zone);
255 if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
257 else if (pfn < zone->zone_start_pfn)
259 } while (zone_span_seqretry(zone, seq));
264 static int page_is_consistent(struct zone *zone, struct page *page)
266 if (!pfn_valid_within(page_to_pfn(page)))
268 if (zone != page_zone(page))
274 * Temporary debugging check for pages not lying within a given zone.
276 static int bad_range(struct zone *zone, struct page *page)
278 if (page_outside_zone_boundaries(zone, page))
280 if (!page_is_consistent(zone, page))
286 static inline int bad_range(struct zone *zone, struct page *page)
292 static void bad_page(struct page *page)
294 static unsigned long resume;
295 static unsigned long nr_shown;
296 static unsigned long nr_unshown;
298 /* Don't complain about poisoned pages */
299 if (PageHWPoison(page)) {
300 reset_page_mapcount(page); /* remove PageBuddy */
305 * Allow a burst of 60 reports, then keep quiet for that minute;
306 * or allow a steady drip of one report per second.
308 if (nr_shown == 60) {
309 if (time_before(jiffies, resume)) {
315 "BUG: Bad page state: %lu messages suppressed\n",
322 resume = jiffies + 60 * HZ;
324 printk(KERN_ALERT "BUG: Bad page state in process %s pfn:%05lx\n",
325 current->comm, page_to_pfn(page));
331 /* Leave bad fields for debug, except PageBuddy could make trouble */
332 reset_page_mapcount(page); /* remove PageBuddy */
333 add_taint(TAINT_BAD_PAGE);
337 * Higher-order pages are called "compound pages". They are structured thusly:
339 * The first PAGE_SIZE page is called the "head page".
341 * The remaining PAGE_SIZE pages are called "tail pages".
343 * All pages have PG_compound set. All tail pages have their ->first_page
344 * pointing at the head page.
346 * The first tail page's ->lru.next holds the address of the compound page's
347 * put_page() function. Its ->lru.prev holds the order of allocation.
348 * This usage means that zero-order pages may not be compound.
351 static void free_compound_page(struct page *page)
353 __free_pages_ok(page, compound_order(page));
356 void prep_compound_page(struct page *page, unsigned long order)
359 int nr_pages = 1 << order;
361 set_compound_page_dtor(page, free_compound_page);
362 set_compound_order(page, order);
364 for (i = 1; i < nr_pages; i++) {
365 struct page *p = page + i;
367 set_page_count(p, 0);
368 p->first_page = page;
372 /* update __split_huge_page_refcount if you change this function */
373 static int destroy_compound_page(struct page *page, unsigned long order)
376 int nr_pages = 1 << order;
379 if (unlikely(compound_order(page) != order) ||
380 unlikely(!PageHead(page))) {
385 __ClearPageHead(page);
387 for (i = 1; i < nr_pages; i++) {
388 struct page *p = page + i;
390 if (unlikely(!PageTail(p) || (p->first_page != page))) {
400 static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
405 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
406 * and __GFP_HIGHMEM from hard or soft interrupt context.
408 VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
409 for (i = 0; i < (1 << order); i++)
410 clear_highpage(page + i);
413 #ifdef CONFIG_DEBUG_PAGEALLOC
414 unsigned int _debug_guardpage_minorder;
416 static int __init debug_guardpage_minorder_setup(char *buf)
420 if (kstrtoul(buf, 10, &res) < 0 || res > MAX_ORDER / 2) {
421 printk(KERN_ERR "Bad debug_guardpage_minorder value\n");
424 _debug_guardpage_minorder = res;
425 printk(KERN_INFO "Setting debug_guardpage_minorder to %lu\n", res);
428 __setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup);
430 static inline void set_page_guard_flag(struct page *page)
432 __set_bit(PAGE_DEBUG_FLAG_GUARD, &page->debug_flags);
435 static inline void clear_page_guard_flag(struct page *page)
437 __clear_bit(PAGE_DEBUG_FLAG_GUARD, &page->debug_flags);
440 static inline void set_page_guard_flag(struct page *page) { }
441 static inline void clear_page_guard_flag(struct page *page) { }
444 static inline void set_page_order(struct page *page, int order)
446 set_page_private(page, order);
447 __SetPageBuddy(page);
450 static inline void rmv_page_order(struct page *page)
452 __ClearPageBuddy(page);
453 set_page_private(page, 0);
457 * Locate the struct page for both the matching buddy in our
458 * pair (buddy1) and the combined O(n+1) page they form (page).
460 * 1) Any buddy B1 will have an order O twin B2 which satisfies
461 * the following equation:
463 * For example, if the starting buddy (buddy2) is #8 its order
465 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
467 * 2) Any buddy B will have an order O+1 parent P which
468 * satisfies the following equation:
471 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
473 static inline unsigned long
474 __find_buddy_index(unsigned long page_idx, unsigned int order)
476 return page_idx ^ (1 << order);
480 * This function checks whether a page is free && is the buddy
481 * we can do coalesce a page and its buddy if
482 * (a) the buddy is not in a hole &&
483 * (b) the buddy is in the buddy system &&
484 * (c) a page and its buddy have the same order &&
485 * (d) a page and its buddy are in the same zone.
487 * For recording whether a page is in the buddy system, we set ->_mapcount -2.
488 * Setting, clearing, and testing _mapcount -2 is serialized by zone->lock.
490 * For recording page's order, we use page_private(page).
492 static inline int page_is_buddy(struct page *page, struct page *buddy,
495 if (!pfn_valid_within(page_to_pfn(buddy)))
498 if (page_zone_id(page) != page_zone_id(buddy))
501 if (page_is_guard(buddy) && page_order(buddy) == order) {
502 VM_BUG_ON(page_count(buddy) != 0);
506 if (PageBuddy(buddy) && page_order(buddy) == order) {
507 VM_BUG_ON(page_count(buddy) != 0);
514 * Freeing function for a buddy system allocator.
516 * The concept of a buddy system is to maintain direct-mapped table
517 * (containing bit values) for memory blocks of various "orders".
518 * The bottom level table contains the map for the smallest allocatable
519 * units of memory (here, pages), and each level above it describes
520 * pairs of units from the levels below, hence, "buddies".
521 * At a high level, all that happens here is marking the table entry
522 * at the bottom level available, and propagating the changes upward
523 * as necessary, plus some accounting needed to play nicely with other
524 * parts of the VM system.
525 * At each level, we keep a list of pages, which are heads of continuous
526 * free pages of length of (1 << order) and marked with _mapcount -2. Page's
527 * order is recorded in page_private(page) field.
528 * So when we are allocating or freeing one, we can derive the state of the
529 * other. That is, if we allocate a small block, and both were
530 * free, the remainder of the region must be split into blocks.
531 * If a block is freed, and its buddy is also free, then this
532 * triggers coalescing into a block of larger size.
537 static inline void __free_one_page(struct page *page,
538 struct zone *zone, unsigned int order,
541 unsigned long page_idx;
542 unsigned long combined_idx;
543 unsigned long uninitialized_var(buddy_idx);
546 if (unlikely(PageCompound(page)))
547 if (unlikely(destroy_compound_page(page, order)))
550 VM_BUG_ON(migratetype == -1);
552 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
554 VM_BUG_ON(page_idx & ((1 << order) - 1));
555 VM_BUG_ON(bad_range(zone, page));
557 while (order < MAX_ORDER-1) {
558 buddy_idx = __find_buddy_index(page_idx, order);
559 buddy = page + (buddy_idx - page_idx);
560 if (!page_is_buddy(page, buddy, order))
563 * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
564 * merge with it and move up one order.
566 if (page_is_guard(buddy)) {
567 clear_page_guard_flag(buddy);
568 set_page_private(page, 0);
569 __mod_zone_page_state(zone, NR_FREE_PAGES, 1 << order);
571 list_del(&buddy->lru);
572 zone->free_area[order].nr_free--;
573 rmv_page_order(buddy);
575 combined_idx = buddy_idx & page_idx;
576 page = page + (combined_idx - page_idx);
577 page_idx = combined_idx;
580 set_page_order(page, order);
583 * If this is not the largest possible page, check if the buddy
584 * of the next-highest order is free. If it is, it's possible
585 * that pages are being freed that will coalesce soon. In case,
586 * that is happening, add the free page to the tail of the list
587 * so it's less likely to be used soon and more likely to be merged
588 * as a higher order page
590 if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) {
591 struct page *higher_page, *higher_buddy;
592 combined_idx = buddy_idx & page_idx;
593 higher_page = page + (combined_idx - page_idx);
594 buddy_idx = __find_buddy_index(combined_idx, order + 1);
595 higher_buddy = page + (buddy_idx - combined_idx);
596 if (page_is_buddy(higher_page, higher_buddy, order + 1)) {
597 list_add_tail(&page->lru,
598 &zone->free_area[order].free_list[migratetype]);
603 list_add(&page->lru, &zone->free_area[order].free_list[migratetype]);
605 zone->free_area[order].nr_free++;
609 * free_page_mlock() -- clean up attempts to free and mlocked() page.
610 * Page should not be on lru, so no need to fix that up.
611 * free_pages_check() will verify...
613 static inline void free_page_mlock(struct page *page)
615 __dec_zone_page_state(page, NR_MLOCK);
616 __count_vm_event(UNEVICTABLE_MLOCKFREED);
619 static inline int free_pages_check(struct page *page)
621 if (unlikely(page_mapcount(page) |
622 (page->mapping != NULL) |
623 (atomic_read(&page->_count) != 0) |
624 (page->flags & PAGE_FLAGS_CHECK_AT_FREE) |
625 (mem_cgroup_bad_page_check(page)))) {
629 if (page->flags & PAGE_FLAGS_CHECK_AT_PREP)
630 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
635 * Frees a number of pages from the PCP lists
636 * Assumes all pages on list are in same zone, and of same order.
637 * count is the number of pages to free.
639 * If the zone was previously in an "all pages pinned" state then look to
640 * see if this freeing clears that state.
642 * And clear the zone's pages_scanned counter, to hold off the "all pages are
643 * pinned" detection logic.
645 static void free_pcppages_bulk(struct zone *zone, int count,
646 struct per_cpu_pages *pcp)
652 spin_lock(&zone->lock);
653 zone->all_unreclaimable = 0;
654 zone->pages_scanned = 0;
658 struct list_head *list;
661 * Remove pages from lists in a round-robin fashion. A
662 * batch_free count is maintained that is incremented when an
663 * empty list is encountered. This is so more pages are freed
664 * off fuller lists instead of spinning excessively around empty
669 if (++migratetype == MIGRATE_PCPTYPES)
671 list = &pcp->lists[migratetype];
672 } while (list_empty(list));
674 /* This is the only non-empty list. Free them all. */
675 if (batch_free == MIGRATE_PCPTYPES)
676 batch_free = to_free;
679 page = list_entry(list->prev, struct page, lru);
680 /* must delete as __free_one_page list manipulates */
681 list_del(&page->lru);
682 /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */
683 __free_one_page(page, zone, 0, page_private(page));
684 trace_mm_page_pcpu_drain(page, 0, page_private(page));
685 } while (--to_free && --batch_free && !list_empty(list));
687 __mod_zone_page_state(zone, NR_FREE_PAGES, count);
688 spin_unlock(&zone->lock);
691 static void free_one_page(struct zone *zone, struct page *page, int order,
694 spin_lock(&zone->lock);
695 zone->all_unreclaimable = 0;
696 zone->pages_scanned = 0;
698 __free_one_page(page, zone, order, migratetype);
699 __mod_zone_page_state(zone, NR_FREE_PAGES, 1 << order);
700 spin_unlock(&zone->lock);
703 static bool free_pages_prepare(struct page *page, unsigned int order)
708 trace_mm_page_free(page, order);
709 kmemcheck_free_shadow(page, order);
712 page->mapping = NULL;
713 for (i = 0; i < (1 << order); i++)
714 bad += free_pages_check(page + i);
718 if (!PageHighMem(page)) {
719 debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order);
720 debug_check_no_obj_freed(page_address(page),
723 arch_free_page(page, order);
724 kernel_map_pages(page, 1 << order, 0);
729 static void __free_pages_ok(struct page *page, unsigned int order)
732 int wasMlocked = __TestClearPageMlocked(page);
734 if (!free_pages_prepare(page, order))
737 local_irq_save(flags);
738 if (unlikely(wasMlocked))
739 free_page_mlock(page);
740 __count_vm_events(PGFREE, 1 << order);
741 free_one_page(page_zone(page), page, order,
742 get_pageblock_migratetype(page));
743 local_irq_restore(flags);
747 * permit the bootmem allocator to evade page validation on high-order frees
749 void __meminit __free_pages_bootmem(struct page *page, unsigned int order)
752 __ClearPageReserved(page);
753 set_page_count(page, 0);
754 set_page_refcounted(page);
760 for (loop = 0; loop < (1 << order); loop++) {
761 struct page *p = &page[loop];
763 if (loop + 1 < (1 << order))
765 __ClearPageReserved(p);
766 set_page_count(p, 0);
769 set_page_refcounted(page);
770 __free_pages(page, order);
776 * The order of subdivision here is critical for the IO subsystem.
777 * Please do not alter this order without good reasons and regression
778 * testing. Specifically, as large blocks of memory are subdivided,
779 * the order in which smaller blocks are delivered depends on the order
780 * they're subdivided in this function. This is the primary factor
781 * influencing the order in which pages are delivered to the IO
782 * subsystem according to empirical testing, and this is also justified
783 * by considering the behavior of a buddy system containing a single
784 * large block of memory acted on by a series of small allocations.
785 * This behavior is a critical factor in sglist merging's success.
789 static inline void expand(struct zone *zone, struct page *page,
790 int low, int high, struct free_area *area,
793 unsigned long size = 1 << high;
799 VM_BUG_ON(bad_range(zone, &page[size]));
801 #ifdef CONFIG_DEBUG_PAGEALLOC
802 if (high < debug_guardpage_minorder()) {
804 * Mark as guard pages (or page), that will allow to
805 * merge back to allocator when buddy will be freed.
806 * Corresponding page table entries will not be touched,
807 * pages will stay not present in virtual address space
809 INIT_LIST_HEAD(&page[size].lru);
810 set_page_guard_flag(&page[size]);
811 set_page_private(&page[size], high);
812 /* Guard pages are not available for any usage */
813 __mod_zone_page_state(zone, NR_FREE_PAGES, -(1 << high));
817 list_add(&page[size].lru, &area->free_list[migratetype]);
819 set_page_order(&page[size], high);
824 * This page is about to be returned from the page allocator
826 static inline int check_new_page(struct page *page)
828 if (unlikely(page_mapcount(page) |
829 (page->mapping != NULL) |
830 (atomic_read(&page->_count) != 0) |
831 (page->flags & PAGE_FLAGS_CHECK_AT_PREP) |
832 (mem_cgroup_bad_page_check(page)))) {
839 static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
843 for (i = 0; i < (1 << order); i++) {
844 struct page *p = page + i;
845 if (unlikely(check_new_page(p)))
849 set_page_private(page, 0);
850 set_page_refcounted(page);
852 arch_alloc_page(page, order);
853 kernel_map_pages(page, 1 << order, 1);
855 if (gfp_flags & __GFP_ZERO)
856 prep_zero_page(page, order, gfp_flags);
858 if (order && (gfp_flags & __GFP_COMP))
859 prep_compound_page(page, order);
865 * Go through the free lists for the given migratetype and remove
866 * the smallest available page from the freelists
869 struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
872 unsigned int current_order;
873 struct free_area * area;
876 /* Find a page of the appropriate size in the preferred list */
877 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
878 area = &(zone->free_area[current_order]);
879 if (list_empty(&area->free_list[migratetype]))
882 page = list_entry(area->free_list[migratetype].next,
884 list_del(&page->lru);
885 rmv_page_order(page);
887 expand(zone, page, order, current_order, area, migratetype);
896 * This array describes the order lists are fallen back to when
897 * the free lists for the desirable migrate type are depleted
899 static int fallbacks[MIGRATE_TYPES][MIGRATE_TYPES-1] = {
900 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
901 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
902 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
903 [MIGRATE_RESERVE] = { MIGRATE_RESERVE, MIGRATE_RESERVE, MIGRATE_RESERVE }, /* Never used */
907 * Move the free pages in a range to the free lists of the requested type.
908 * Note that start_page and end_pages are not aligned on a pageblock
909 * boundary. If alignment is required, use move_freepages_block()
911 static int move_freepages(struct zone *zone,
912 struct page *start_page, struct page *end_page,
919 #ifndef CONFIG_HOLES_IN_ZONE
921 * page_zone is not safe to call in this context when
922 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
923 * anyway as we check zone boundaries in move_freepages_block().
924 * Remove at a later date when no bug reports exist related to
925 * grouping pages by mobility
927 BUG_ON(page_zone(start_page) != page_zone(end_page));
930 for (page = start_page; page <= end_page;) {
931 /* Make sure we are not inadvertently changing nodes */
932 VM_BUG_ON(page_to_nid(page) != zone_to_nid(zone));
934 if (!pfn_valid_within(page_to_pfn(page))) {
939 if (!PageBuddy(page)) {
944 order = page_order(page);
945 list_move(&page->lru,
946 &zone->free_area[order].free_list[migratetype]);
948 pages_moved += 1 << order;
954 static int move_freepages_block(struct zone *zone, struct page *page,
957 unsigned long start_pfn, end_pfn;
958 struct page *start_page, *end_page;
960 start_pfn = page_to_pfn(page);
961 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
962 start_page = pfn_to_page(start_pfn);
963 end_page = start_page + pageblock_nr_pages - 1;
964 end_pfn = start_pfn + pageblock_nr_pages - 1;
966 /* Do not cross zone boundaries */
967 if (start_pfn < zone->zone_start_pfn)
969 if (end_pfn >= zone->zone_start_pfn + zone->spanned_pages)
972 return move_freepages(zone, start_page, end_page, migratetype);
975 static void change_pageblock_range(struct page *pageblock_page,
976 int start_order, int migratetype)
978 int nr_pageblocks = 1 << (start_order - pageblock_order);
980 while (nr_pageblocks--) {
981 set_pageblock_migratetype(pageblock_page, migratetype);
982 pageblock_page += pageblock_nr_pages;
986 /* Remove an element from the buddy allocator from the fallback list */
987 static inline struct page *
988 __rmqueue_fallback(struct zone *zone, int order, int start_migratetype)
990 struct free_area * area;
995 /* Find the largest possible block of pages in the other list */
996 for (current_order = MAX_ORDER-1; current_order >= order;
998 for (i = 0; i < MIGRATE_TYPES - 1; i++) {
999 migratetype = fallbacks[start_migratetype][i];
1001 /* MIGRATE_RESERVE handled later if necessary */
1002 if (migratetype == MIGRATE_RESERVE)
1005 area = &(zone->free_area[current_order]);
1006 if (list_empty(&area->free_list[migratetype]))
1009 page = list_entry(area->free_list[migratetype].next,
1014 * If breaking a large block of pages, move all free
1015 * pages to the preferred allocation list. If falling
1016 * back for a reclaimable kernel allocation, be more
1017 * aggressive about taking ownership of free pages
1019 if (unlikely(current_order >= (pageblock_order >> 1)) ||
1020 start_migratetype == MIGRATE_RECLAIMABLE ||
1021 page_group_by_mobility_disabled) {
1022 unsigned long pages;
1023 pages = move_freepages_block(zone, page,
1026 /* Claim the whole block if over half of it is free */
1027 if (pages >= (1 << (pageblock_order-1)) ||
1028 page_group_by_mobility_disabled)
1029 set_pageblock_migratetype(page,
1032 migratetype = start_migratetype;
1035 /* Remove the page from the freelists */
1036 list_del(&page->lru);
1037 rmv_page_order(page);
1039 /* Take ownership for orders >= pageblock_order */
1040 if (current_order >= pageblock_order)
1041 change_pageblock_range(page, current_order,
1044 expand(zone, page, order, current_order, area, migratetype);
1046 trace_mm_page_alloc_extfrag(page, order, current_order,
1047 start_migratetype, migratetype);
1057 * Do the hard work of removing an element from the buddy allocator.
1058 * Call me with the zone->lock already held.
1060 static struct page *__rmqueue(struct zone *zone, unsigned int order,
1066 page = __rmqueue_smallest(zone, order, migratetype);
1068 if (unlikely(!page) && migratetype != MIGRATE_RESERVE) {
1069 page = __rmqueue_fallback(zone, order, migratetype);
1072 * Use MIGRATE_RESERVE rather than fail an allocation. goto
1073 * is used because __rmqueue_smallest is an inline function
1074 * and we want just one call site
1077 migratetype = MIGRATE_RESERVE;
1082 trace_mm_page_alloc_zone_locked(page, order, migratetype);
1087 * Obtain a specified number of elements from the buddy allocator, all under
1088 * a single hold of the lock, for efficiency. Add them to the supplied list.
1089 * Returns the number of new pages which were placed at *list.
1091 static int rmqueue_bulk(struct zone *zone, unsigned int order,
1092 unsigned long count, struct list_head *list,
1093 int migratetype, int cold)
1097 spin_lock(&zone->lock);
1098 for (i = 0; i < count; ++i) {
1099 struct page *page = __rmqueue(zone, order, migratetype);
1100 if (unlikely(page == NULL))
1104 * Split buddy pages returned by expand() are received here
1105 * in physical page order. The page is added to the callers and
1106 * list and the list head then moves forward. From the callers
1107 * perspective, the linked list is ordered by page number in
1108 * some conditions. This is useful for IO devices that can
1109 * merge IO requests if the physical pages are ordered
1112 if (likely(cold == 0))
1113 list_add(&page->lru, list);
1115 list_add_tail(&page->lru, list);
1116 set_page_private(page, migratetype);
1119 __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
1120 spin_unlock(&zone->lock);
1126 * Called from the vmstat counter updater to drain pagesets of this
1127 * currently executing processor on remote nodes after they have
1130 * Note that this function must be called with the thread pinned to
1131 * a single processor.
1133 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
1135 unsigned long flags;
1138 local_irq_save(flags);
1139 if (pcp->count >= pcp->batch)
1140 to_drain = pcp->batch;
1142 to_drain = pcp->count;
1143 free_pcppages_bulk(zone, to_drain, pcp);
1144 pcp->count -= to_drain;
1145 local_irq_restore(flags);
1150 * Drain pages of the indicated processor.
1152 * The processor must either be the current processor and the
1153 * thread pinned to the current processor or a processor that
1156 static void drain_pages(unsigned int cpu)
1158 unsigned long flags;
1161 for_each_populated_zone(zone) {
1162 struct per_cpu_pageset *pset;
1163 struct per_cpu_pages *pcp;
1165 local_irq_save(flags);
1166 pset = per_cpu_ptr(zone->pageset, cpu);
1170 free_pcppages_bulk(zone, pcp->count, pcp);
1173 local_irq_restore(flags);
1178 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
1180 void drain_local_pages(void *arg)
1182 drain_pages(smp_processor_id());
1186 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
1188 void drain_all_pages(void)
1190 on_each_cpu(drain_local_pages, NULL, 1);
1193 #ifdef CONFIG_HIBERNATION
1195 void mark_free_pages(struct zone *zone)
1197 unsigned long pfn, max_zone_pfn;
1198 unsigned long flags;
1200 struct list_head *curr;
1202 if (!zone->spanned_pages)
1205 spin_lock_irqsave(&zone->lock, flags);
1207 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
1208 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1209 if (pfn_valid(pfn)) {
1210 struct page *page = pfn_to_page(pfn);
1212 if (!swsusp_page_is_forbidden(page))
1213 swsusp_unset_page_free(page);
1216 for_each_migratetype_order(order, t) {
1217 list_for_each(curr, &zone->free_area[order].free_list[t]) {
1220 pfn = page_to_pfn(list_entry(curr, struct page, lru));
1221 for (i = 0; i < (1UL << order); i++)
1222 swsusp_set_page_free(pfn_to_page(pfn + i));
1225 spin_unlock_irqrestore(&zone->lock, flags);
1227 #endif /* CONFIG_PM */
1230 * Free a 0-order page
1231 * cold == 1 ? free a cold page : free a hot page
1233 void free_hot_cold_page(struct page *page, int cold)
1235 struct zone *zone = page_zone(page);
1236 struct per_cpu_pages *pcp;
1237 unsigned long flags;
1239 int wasMlocked = __TestClearPageMlocked(page);
1241 if (!free_pages_prepare(page, 0))
1244 migratetype = get_pageblock_migratetype(page);
1245 set_page_private(page, migratetype);
1246 local_irq_save(flags);
1247 if (unlikely(wasMlocked))
1248 free_page_mlock(page);
1249 __count_vm_event(PGFREE);
1252 * We only track unmovable, reclaimable and movable on pcp lists.
1253 * Free ISOLATE pages back to the allocator because they are being
1254 * offlined but treat RESERVE as movable pages so we can get those
1255 * areas back if necessary. Otherwise, we may have to free
1256 * excessively into the page allocator
1258 if (migratetype >= MIGRATE_PCPTYPES) {
1259 if (unlikely(migratetype == MIGRATE_ISOLATE)) {
1260 free_one_page(zone, page, 0, migratetype);
1263 migratetype = MIGRATE_MOVABLE;
1266 pcp = &this_cpu_ptr(zone->pageset)->pcp;
1268 list_add_tail(&page->lru, &pcp->lists[migratetype]);
1270 list_add(&page->lru, &pcp->lists[migratetype]);
1272 if (pcp->count >= pcp->high) {
1273 free_pcppages_bulk(zone, pcp->batch, pcp);
1274 pcp->count -= pcp->batch;
1278 local_irq_restore(flags);
1282 * Free a list of 0-order pages
1284 void free_hot_cold_page_list(struct list_head *list, int cold)
1286 struct page *page, *next;
1288 list_for_each_entry_safe(page, next, list, lru) {
1289 trace_mm_page_free_batched(page, cold);
1290 free_hot_cold_page(page, cold);
1295 * split_page takes a non-compound higher-order page, and splits it into
1296 * n (1<<order) sub-pages: page[0..n]
1297 * Each sub-page must be freed individually.
1299 * Note: this is probably too low level an operation for use in drivers.
1300 * Please consult with lkml before using this in your driver.
1302 void split_page(struct page *page, unsigned int order)
1306 VM_BUG_ON(PageCompound(page));
1307 VM_BUG_ON(!page_count(page));
1309 #ifdef CONFIG_KMEMCHECK
1311 * Split shadow pages too, because free(page[0]) would
1312 * otherwise free the whole shadow.
1314 if (kmemcheck_page_is_tracked(page))
1315 split_page(virt_to_page(page[0].shadow), order);
1318 for (i = 1; i < (1 << order); i++)
1319 set_page_refcounted(page + i);
1323 * Similar to split_page except the page is already free. As this is only
1324 * being used for migration, the migratetype of the block also changes.
1325 * As this is called with interrupts disabled, the caller is responsible
1326 * for calling arch_alloc_page() and kernel_map_page() after interrupts
1329 * Note: this is probably too low level an operation for use in drivers.
1330 * Please consult with lkml before using this in your driver.
1332 int split_free_page(struct page *page)
1335 unsigned long watermark;
1338 BUG_ON(!PageBuddy(page));
1340 zone = page_zone(page);
1341 order = page_order(page);
1343 /* Obey watermarks as if the page was being allocated */
1344 watermark = low_wmark_pages(zone) + (1 << order);
1345 if (!zone_watermark_ok(zone, 0, watermark, 0, 0))
1348 /* Remove page from free list */
1349 list_del(&page->lru);
1350 zone->free_area[order].nr_free--;
1351 rmv_page_order(page);
1352 __mod_zone_page_state(zone, NR_FREE_PAGES, -(1UL << order));
1354 /* Split into individual pages */
1355 set_page_refcounted(page);
1356 split_page(page, order);
1358 if (order >= pageblock_order - 1) {
1359 struct page *endpage = page + (1 << order) - 1;
1360 for (; page < endpage; page += pageblock_nr_pages)
1361 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
1368 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1369 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1373 struct page *buffered_rmqueue(struct zone *preferred_zone,
1374 struct zone *zone, int order, gfp_t gfp_flags,
1377 unsigned long flags;
1379 int cold = !!(gfp_flags & __GFP_COLD);
1382 if (likely(order == 0)) {
1383 struct per_cpu_pages *pcp;
1384 struct list_head *list;
1386 local_irq_save(flags);
1387 pcp = &this_cpu_ptr(zone->pageset)->pcp;
1388 list = &pcp->lists[migratetype];
1389 if (list_empty(list)) {
1390 pcp->count += rmqueue_bulk(zone, 0,
1393 if (unlikely(list_empty(list)))
1398 page = list_entry(list->prev, struct page, lru);
1400 page = list_entry(list->next, struct page, lru);
1402 list_del(&page->lru);
1405 if (unlikely(gfp_flags & __GFP_NOFAIL)) {
1407 * __GFP_NOFAIL is not to be used in new code.
1409 * All __GFP_NOFAIL callers should be fixed so that they
1410 * properly detect and handle allocation failures.
1412 * We most definitely don't want callers attempting to
1413 * allocate greater than order-1 page units with
1416 WARN_ON_ONCE(order > 1);
1418 spin_lock_irqsave(&zone->lock, flags);
1419 page = __rmqueue(zone, order, migratetype);
1420 spin_unlock(&zone->lock);
1423 __mod_zone_page_state(zone, NR_FREE_PAGES, -(1 << order));
1426 __count_zone_vm_events(PGALLOC, zone, 1 << order);
1427 zone_statistics(preferred_zone, zone, gfp_flags);
1428 local_irq_restore(flags);
1430 VM_BUG_ON(bad_range(zone, page));
1431 if (prep_new_page(page, order, gfp_flags))
1436 local_irq_restore(flags);
1440 /* The ALLOC_WMARK bits are used as an index to zone->watermark */
1441 #define ALLOC_WMARK_MIN WMARK_MIN
1442 #define ALLOC_WMARK_LOW WMARK_LOW
1443 #define ALLOC_WMARK_HIGH WMARK_HIGH
1444 #define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */
1446 /* Mask to get the watermark bits */
1447 #define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1)
1449 #define ALLOC_HARDER 0x10 /* try to alloc harder */
1450 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1451 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
1453 #ifdef CONFIG_FAIL_PAGE_ALLOC
1456 struct fault_attr attr;
1458 u32 ignore_gfp_highmem;
1459 u32 ignore_gfp_wait;
1461 } fail_page_alloc = {
1462 .attr = FAULT_ATTR_INITIALIZER,
1463 .ignore_gfp_wait = 1,
1464 .ignore_gfp_highmem = 1,
1468 static int __init setup_fail_page_alloc(char *str)
1470 return setup_fault_attr(&fail_page_alloc.attr, str);
1472 __setup("fail_page_alloc=", setup_fail_page_alloc);
1474 static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1476 if (order < fail_page_alloc.min_order)
1478 if (gfp_mask & __GFP_NOFAIL)
1480 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
1482 if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
1485 return should_fail(&fail_page_alloc.attr, 1 << order);
1488 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1490 static int __init fail_page_alloc_debugfs(void)
1492 mode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
1495 dir = fault_create_debugfs_attr("fail_page_alloc", NULL,
1496 &fail_page_alloc.attr);
1498 return PTR_ERR(dir);
1500 if (!debugfs_create_bool("ignore-gfp-wait", mode, dir,
1501 &fail_page_alloc.ignore_gfp_wait))
1503 if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir,
1504 &fail_page_alloc.ignore_gfp_highmem))
1506 if (!debugfs_create_u32("min-order", mode, dir,
1507 &fail_page_alloc.min_order))
1512 debugfs_remove_recursive(dir);
1517 late_initcall(fail_page_alloc_debugfs);
1519 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1521 #else /* CONFIG_FAIL_PAGE_ALLOC */
1523 static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1528 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1531 * Return true if free pages are above 'mark'. This takes into account the order
1532 * of the allocation.
1534 static bool __zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1535 int classzone_idx, int alloc_flags, long free_pages)
1537 /* free_pages my go negative - that's OK */
1541 free_pages -= (1 << order) - 1;
1542 if (alloc_flags & ALLOC_HIGH)
1544 if (alloc_flags & ALLOC_HARDER)
1547 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
1549 for (o = 0; o < order; o++) {
1550 /* At the next order, this order's pages become unavailable */
1551 free_pages -= z->free_area[o].nr_free << o;
1553 /* Require fewer higher order pages to be free */
1556 if (free_pages <= min)
1562 bool zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1563 int classzone_idx, int alloc_flags)
1565 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
1566 zone_page_state(z, NR_FREE_PAGES));
1569 bool zone_watermark_ok_safe(struct zone *z, int order, unsigned long mark,
1570 int classzone_idx, int alloc_flags)
1572 long free_pages = zone_page_state(z, NR_FREE_PAGES);
1574 if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark)
1575 free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES);
1577 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
1583 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1584 * skip over zones that are not allowed by the cpuset, or that have
1585 * been recently (in last second) found to be nearly full. See further
1586 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1587 * that have to skip over a lot of full or unallowed zones.
1589 * If the zonelist cache is present in the passed in zonelist, then
1590 * returns a pointer to the allowed node mask (either the current
1591 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
1593 * If the zonelist cache is not available for this zonelist, does
1594 * nothing and returns NULL.
1596 * If the fullzones BITMAP in the zonelist cache is stale (more than
1597 * a second since last zap'd) then we zap it out (clear its bits.)
1599 * We hold off even calling zlc_setup, until after we've checked the
1600 * first zone in the zonelist, on the theory that most allocations will
1601 * be satisfied from that first zone, so best to examine that zone as
1602 * quickly as we can.
1604 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1606 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1607 nodemask_t *allowednodes; /* zonelist_cache approximation */
1609 zlc = zonelist->zlcache_ptr;
1613 if (time_after(jiffies, zlc->last_full_zap + HZ)) {
1614 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1615 zlc->last_full_zap = jiffies;
1618 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1619 &cpuset_current_mems_allowed :
1620 &node_states[N_HIGH_MEMORY];
1621 return allowednodes;
1625 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1626 * if it is worth looking at further for free memory:
1627 * 1) Check that the zone isn't thought to be full (doesn't have its
1628 * bit set in the zonelist_cache fullzones BITMAP).
1629 * 2) Check that the zones node (obtained from the zonelist_cache
1630 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1631 * Return true (non-zero) if zone is worth looking at further, or
1632 * else return false (zero) if it is not.
1634 * This check -ignores- the distinction between various watermarks,
1635 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1636 * found to be full for any variation of these watermarks, it will
1637 * be considered full for up to one second by all requests, unless
1638 * we are so low on memory on all allowed nodes that we are forced
1639 * into the second scan of the zonelist.
1641 * In the second scan we ignore this zonelist cache and exactly
1642 * apply the watermarks to all zones, even it is slower to do so.
1643 * We are low on memory in the second scan, and should leave no stone
1644 * unturned looking for a free page.
1646 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
1647 nodemask_t *allowednodes)
1649 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1650 int i; /* index of *z in zonelist zones */
1651 int n; /* node that zone *z is on */
1653 zlc = zonelist->zlcache_ptr;
1657 i = z - zonelist->_zonerefs;
1660 /* This zone is worth trying if it is allowed but not full */
1661 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1665 * Given 'z' scanning a zonelist, set the corresponding bit in
1666 * zlc->fullzones, so that subsequent attempts to allocate a page
1667 * from that zone don't waste time re-examining it.
1669 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
1671 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1672 int i; /* index of *z in zonelist zones */
1674 zlc = zonelist->zlcache_ptr;
1678 i = z - zonelist->_zonerefs;
1680 set_bit(i, zlc->fullzones);
1684 * clear all zones full, called after direct reclaim makes progress so that
1685 * a zone that was recently full is not skipped over for up to a second
1687 static void zlc_clear_zones_full(struct zonelist *zonelist)
1689 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1691 zlc = zonelist->zlcache_ptr;
1695 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1698 #else /* CONFIG_NUMA */
1700 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1705 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
1706 nodemask_t *allowednodes)
1711 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
1715 static void zlc_clear_zones_full(struct zonelist *zonelist)
1718 #endif /* CONFIG_NUMA */
1721 * get_page_from_freelist goes through the zonelist trying to allocate
1724 static struct page *
1725 get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order,
1726 struct zonelist *zonelist, int high_zoneidx, int alloc_flags,
1727 struct zone *preferred_zone, int migratetype)
1730 struct page *page = NULL;
1733 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1734 int zlc_active = 0; /* set if using zonelist_cache */
1735 int did_zlc_setup = 0; /* just call zlc_setup() one time */
1737 classzone_idx = zone_idx(preferred_zone);
1740 * Scan zonelist, looking for a zone with enough free.
1741 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1743 for_each_zone_zonelist_nodemask(zone, z, zonelist,
1744 high_zoneidx, nodemask) {
1745 if (NUMA_BUILD && zlc_active &&
1746 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1748 if ((alloc_flags & ALLOC_CPUSET) &&
1749 !cpuset_zone_allowed_softwall(zone, gfp_mask))
1752 * When allocating a page cache page for writing, we
1753 * want to get it from a zone that is within its dirty
1754 * limit, such that no single zone holds more than its
1755 * proportional share of globally allowed dirty pages.
1756 * The dirty limits take into account the zone's
1757 * lowmem reserves and high watermark so that kswapd
1758 * should be able to balance it without having to
1759 * write pages from its LRU list.
1761 * This may look like it could increase pressure on
1762 * lower zones by failing allocations in higher zones
1763 * before they are full. But the pages that do spill
1764 * over are limited as the lower zones are protected
1765 * by this very same mechanism. It should not become
1766 * a practical burden to them.
1768 * XXX: For now, allow allocations to potentially
1769 * exceed the per-zone dirty limit in the slowpath
1770 * (ALLOC_WMARK_LOW unset) before going into reclaim,
1771 * which is important when on a NUMA setup the allowed
1772 * zones are together not big enough to reach the
1773 * global limit. The proper fix for these situations
1774 * will require awareness of zones in the
1775 * dirty-throttling and the flusher threads.
1777 if ((alloc_flags & ALLOC_WMARK_LOW) &&
1778 (gfp_mask & __GFP_WRITE) && !zone_dirty_ok(zone))
1779 goto this_zone_full;
1781 BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
1782 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
1786 mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
1787 if (zone_watermark_ok(zone, order, mark,
1788 classzone_idx, alloc_flags))
1791 if (NUMA_BUILD && !did_zlc_setup && nr_online_nodes > 1) {
1793 * we do zlc_setup if there are multiple nodes
1794 * and before considering the first zone allowed
1797 allowednodes = zlc_setup(zonelist, alloc_flags);
1802 if (zone_reclaim_mode == 0)
1803 goto this_zone_full;
1806 * As we may have just activated ZLC, check if the first
1807 * eligible zone has failed zone_reclaim recently.
1809 if (NUMA_BUILD && zlc_active &&
1810 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1813 ret = zone_reclaim(zone, gfp_mask, order);
1815 case ZONE_RECLAIM_NOSCAN:
1818 case ZONE_RECLAIM_FULL:
1819 /* scanned but unreclaimable */
1822 /* did we reclaim enough */
1823 if (!zone_watermark_ok(zone, order, mark,
1824 classzone_idx, alloc_flags))
1825 goto this_zone_full;
1830 page = buffered_rmqueue(preferred_zone, zone, order,
1831 gfp_mask, migratetype);
1836 zlc_mark_zone_full(zonelist, z);
1839 if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) {
1840 /* Disable zlc cache for second zonelist scan */
1848 * Large machines with many possible nodes should not always dump per-node
1849 * meminfo in irq context.
1851 static inline bool should_suppress_show_mem(void)
1856 ret = in_interrupt();
1861 static DEFINE_RATELIMIT_STATE(nopage_rs,
1862 DEFAULT_RATELIMIT_INTERVAL,
1863 DEFAULT_RATELIMIT_BURST);
1865 void warn_alloc_failed(gfp_t gfp_mask, int order, const char *fmt, ...)
1867 unsigned int filter = SHOW_MEM_FILTER_NODES;
1869 if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs) ||
1870 debug_guardpage_minorder() > 0)
1874 * This documents exceptions given to allocations in certain
1875 * contexts that are allowed to allocate outside current's set
1878 if (!(gfp_mask & __GFP_NOMEMALLOC))
1879 if (test_thread_flag(TIF_MEMDIE) ||
1880 (current->flags & (PF_MEMALLOC | PF_EXITING)))
1881 filter &= ~SHOW_MEM_FILTER_NODES;
1882 if (in_interrupt() || !(gfp_mask & __GFP_WAIT))
1883 filter &= ~SHOW_MEM_FILTER_NODES;
1886 struct va_format vaf;
1889 va_start(args, fmt);
1894 pr_warn("%pV", &vaf);
1899 pr_warn("%s: page allocation failure: order:%d, mode:0x%x\n",
1900 current->comm, order, gfp_mask);
1903 if (!should_suppress_show_mem())
1908 should_alloc_retry(gfp_t gfp_mask, unsigned int order,
1909 unsigned long did_some_progress,
1910 unsigned long pages_reclaimed)
1912 /* Do not loop if specifically requested */
1913 if (gfp_mask & __GFP_NORETRY)
1916 /* Always retry if specifically requested */
1917 if (gfp_mask & __GFP_NOFAIL)
1921 * Suspend converts GFP_KERNEL to __GFP_WAIT which can prevent reclaim
1922 * making forward progress without invoking OOM. Suspend also disables
1923 * storage devices so kswapd will not help. Bail if we are suspending.
1925 if (!did_some_progress && pm_suspended_storage())
1929 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
1930 * means __GFP_NOFAIL, but that may not be true in other
1933 if (order <= PAGE_ALLOC_COSTLY_ORDER)
1937 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
1938 * specified, then we retry until we no longer reclaim any pages
1939 * (above), or we've reclaimed an order of pages at least as
1940 * large as the allocation's order. In both cases, if the
1941 * allocation still fails, we stop retrying.
1943 if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order))
1949 static inline struct page *
1950 __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
1951 struct zonelist *zonelist, enum zone_type high_zoneidx,
1952 nodemask_t *nodemask, struct zone *preferred_zone,
1957 /* Acquire the OOM killer lock for the zones in zonelist */
1958 if (!try_set_zonelist_oom(zonelist, gfp_mask)) {
1959 schedule_timeout_uninterruptible(1);
1964 * Go through the zonelist yet one more time, keep very high watermark
1965 * here, this is only to catch a parallel oom killing, we must fail if
1966 * we're still under heavy pressure.
1968 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask,
1969 order, zonelist, high_zoneidx,
1970 ALLOC_WMARK_HIGH|ALLOC_CPUSET,
1971 preferred_zone, migratetype);
1975 if (!(gfp_mask & __GFP_NOFAIL)) {
1976 /* The OOM killer will not help higher order allocs */
1977 if (order > PAGE_ALLOC_COSTLY_ORDER)
1979 /* The OOM killer does not needlessly kill tasks for lowmem */
1980 if (high_zoneidx < ZONE_NORMAL)
1983 * GFP_THISNODE contains __GFP_NORETRY and we never hit this.
1984 * Sanity check for bare calls of __GFP_THISNODE, not real OOM.
1985 * The caller should handle page allocation failure by itself if
1986 * it specifies __GFP_THISNODE.
1987 * Note: Hugepage uses it but will hit PAGE_ALLOC_COSTLY_ORDER.
1989 if (gfp_mask & __GFP_THISNODE)
1992 /* Exhausted what can be done so it's blamo time */
1993 out_of_memory(zonelist, gfp_mask, order, nodemask);
1996 clear_zonelist_oom(zonelist, gfp_mask);
2000 #ifdef CONFIG_COMPACTION
2001 /* Try memory compaction for high-order allocations before reclaim */
2002 static struct page *
2003 __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
2004 struct zonelist *zonelist, enum zone_type high_zoneidx,
2005 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
2006 int migratetype, unsigned long *did_some_progress,
2007 bool sync_migration)
2011 if (!order || compaction_deferred(preferred_zone))
2014 current->flags |= PF_MEMALLOC;
2015 *did_some_progress = try_to_compact_pages(zonelist, order, gfp_mask,
2016 nodemask, sync_migration);
2017 current->flags &= ~PF_MEMALLOC;
2018 if (*did_some_progress != COMPACT_SKIPPED) {
2020 /* Page migration frees to the PCP lists but we want merging */
2021 drain_pages(get_cpu());
2024 page = get_page_from_freelist(gfp_mask, nodemask,
2025 order, zonelist, high_zoneidx,
2026 alloc_flags, preferred_zone,
2029 preferred_zone->compact_considered = 0;
2030 preferred_zone->compact_defer_shift = 0;
2031 count_vm_event(COMPACTSUCCESS);
2036 * It's bad if compaction run occurs and fails.
2037 * The most likely reason is that pages exist,
2038 * but not enough to satisfy watermarks.
2040 count_vm_event(COMPACTFAIL);
2041 defer_compaction(preferred_zone);
2049 static inline struct page *
2050 __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
2051 struct zonelist *zonelist, enum zone_type high_zoneidx,
2052 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
2053 int migratetype, unsigned long *did_some_progress,
2054 bool sync_migration)
2058 #endif /* CONFIG_COMPACTION */
2060 /* The really slow allocator path where we enter direct reclaim */
2061 static inline struct page *
2062 __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
2063 struct zonelist *zonelist, enum zone_type high_zoneidx,
2064 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
2065 int migratetype, unsigned long *did_some_progress)
2067 struct page *page = NULL;
2068 struct reclaim_state reclaim_state;
2069 bool drained = false;
2073 /* We now go into synchronous reclaim */
2074 cpuset_memory_pressure_bump();
2075 current->flags |= PF_MEMALLOC;
2076 lockdep_set_current_reclaim_state(gfp_mask);
2077 reclaim_state.reclaimed_slab = 0;
2078 current->reclaim_state = &reclaim_state;
2080 *did_some_progress = try_to_free_pages(zonelist, order, gfp_mask, nodemask);
2082 current->reclaim_state = NULL;
2083 lockdep_clear_current_reclaim_state();
2084 current->flags &= ~PF_MEMALLOC;
2088 if (unlikely(!(*did_some_progress)))
2091 /* After successful reclaim, reconsider all zones for allocation */
2093 zlc_clear_zones_full(zonelist);
2096 page = get_page_from_freelist(gfp_mask, nodemask, order,
2097 zonelist, high_zoneidx,
2098 alloc_flags, preferred_zone,
2102 * If an allocation failed after direct reclaim, it could be because
2103 * pages are pinned on the per-cpu lists. Drain them and try again
2105 if (!page && !drained) {
2115 * This is called in the allocator slow-path if the allocation request is of
2116 * sufficient urgency to ignore watermarks and take other desperate measures
2118 static inline struct page *
2119 __alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order,
2120 struct zonelist *zonelist, enum zone_type high_zoneidx,
2121 nodemask_t *nodemask, struct zone *preferred_zone,
2127 page = get_page_from_freelist(gfp_mask, nodemask, order,
2128 zonelist, high_zoneidx, ALLOC_NO_WATERMARKS,
2129 preferred_zone, migratetype);
2131 if (!page && gfp_mask & __GFP_NOFAIL)
2132 wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50);
2133 } while (!page && (gfp_mask & __GFP_NOFAIL));
2139 void wake_all_kswapd(unsigned int order, struct zonelist *zonelist,
2140 enum zone_type high_zoneidx,
2141 enum zone_type classzone_idx)
2146 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx)
2147 wakeup_kswapd(zone, order, classzone_idx);
2151 gfp_to_alloc_flags(gfp_t gfp_mask)
2153 int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
2154 const gfp_t wait = gfp_mask & __GFP_WAIT;
2156 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
2157 BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
2160 * The caller may dip into page reserves a bit more if the caller
2161 * cannot run direct reclaim, or if the caller has realtime scheduling
2162 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
2163 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
2165 alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH);
2169 * Not worth trying to allocate harder for
2170 * __GFP_NOMEMALLOC even if it can't schedule.
2172 if (!(gfp_mask & __GFP_NOMEMALLOC))
2173 alloc_flags |= ALLOC_HARDER;
2175 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
2176 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
2178 alloc_flags &= ~ALLOC_CPUSET;
2179 } else if (unlikely(rt_task(current)) && !in_interrupt())
2180 alloc_flags |= ALLOC_HARDER;
2182 if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) {
2183 if (!in_interrupt() &&
2184 ((current->flags & PF_MEMALLOC) ||
2185 unlikely(test_thread_flag(TIF_MEMDIE))))
2186 alloc_flags |= ALLOC_NO_WATERMARKS;
2192 static inline struct page *
2193 __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
2194 struct zonelist *zonelist, enum zone_type high_zoneidx,
2195 nodemask_t *nodemask, struct zone *preferred_zone,
2198 const gfp_t wait = gfp_mask & __GFP_WAIT;
2199 struct page *page = NULL;
2201 unsigned long pages_reclaimed = 0;
2202 unsigned long did_some_progress;
2203 bool sync_migration = false;
2206 * In the slowpath, we sanity check order to avoid ever trying to
2207 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
2208 * be using allocators in order of preference for an area that is
2211 if (order >= MAX_ORDER) {
2212 WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
2217 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
2218 * __GFP_NOWARN set) should not cause reclaim since the subsystem
2219 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
2220 * using a larger set of nodes after it has established that the
2221 * allowed per node queues are empty and that nodes are
2224 if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
2228 if (!(gfp_mask & __GFP_NO_KSWAPD))
2229 wake_all_kswapd(order, zonelist, high_zoneidx,
2230 zone_idx(preferred_zone));
2233 * OK, we're below the kswapd watermark and have kicked background
2234 * reclaim. Now things get more complex, so set up alloc_flags according
2235 * to how we want to proceed.
2237 alloc_flags = gfp_to_alloc_flags(gfp_mask);
2240 * Find the true preferred zone if the allocation is unconstrained by
2243 if (!(alloc_flags & ALLOC_CPUSET) && !nodemask)
2244 first_zones_zonelist(zonelist, high_zoneidx, NULL,
2248 /* This is the last chance, in general, before the goto nopage. */
2249 page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist,
2250 high_zoneidx, alloc_flags & ~ALLOC_NO_WATERMARKS,
2251 preferred_zone, migratetype);
2255 /* Allocate without watermarks if the context allows */
2256 if (alloc_flags & ALLOC_NO_WATERMARKS) {
2257 page = __alloc_pages_high_priority(gfp_mask, order,
2258 zonelist, high_zoneidx, nodemask,
2259 preferred_zone, migratetype);
2264 /* Atomic allocations - we can't balance anything */
2268 /* Avoid recursion of direct reclaim */
2269 if (current->flags & PF_MEMALLOC)
2272 /* Avoid allocations with no watermarks from looping endlessly */
2273 if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL))
2277 * Try direct compaction. The first pass is asynchronous. Subsequent
2278 * attempts after direct reclaim are synchronous
2280 page = __alloc_pages_direct_compact(gfp_mask, order,
2281 zonelist, high_zoneidx,
2283 alloc_flags, preferred_zone,
2284 migratetype, &did_some_progress,
2290 * Do not use sync migration if __GFP_NO_KSWAPD is used to indicate
2291 * the system should not be heavily disrupted. In practice, this is
2292 * to avoid THP callers being stalled in writeback during migration
2293 * as it's preferable for the the allocations to fail than to stall
2295 sync_migration = !(gfp_mask & __GFP_NO_KSWAPD);
2297 /* Try direct reclaim and then allocating */
2298 page = __alloc_pages_direct_reclaim(gfp_mask, order,
2299 zonelist, high_zoneidx,
2301 alloc_flags, preferred_zone,
2302 migratetype, &did_some_progress);
2307 * If we failed to make any progress reclaiming, then we are
2308 * running out of options and have to consider going OOM
2310 if (!did_some_progress) {
2311 if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
2312 if (oom_killer_disabled)
2314 page = __alloc_pages_may_oom(gfp_mask, order,
2315 zonelist, high_zoneidx,
2316 nodemask, preferred_zone,
2321 if (!(gfp_mask & __GFP_NOFAIL)) {
2323 * The oom killer is not called for high-order
2324 * allocations that may fail, so if no progress
2325 * is being made, there are no other options and
2326 * retrying is unlikely to help.
2328 if (order > PAGE_ALLOC_COSTLY_ORDER)
2331 * The oom killer is not called for lowmem
2332 * allocations to prevent needlessly killing
2335 if (high_zoneidx < ZONE_NORMAL)
2343 /* Check if we should retry the allocation */
2344 pages_reclaimed += did_some_progress;
2345 if (should_alloc_retry(gfp_mask, order, did_some_progress,
2347 /* Wait for some write requests to complete then retry */
2348 wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50);
2352 * High-order allocations do not necessarily loop after
2353 * direct reclaim and reclaim/compaction depends on compaction
2354 * being called after reclaim so call directly if necessary
2356 page = __alloc_pages_direct_compact(gfp_mask, order,
2357 zonelist, high_zoneidx,
2359 alloc_flags, preferred_zone,
2360 migratetype, &did_some_progress,
2367 warn_alloc_failed(gfp_mask, order, NULL);
2370 if (kmemcheck_enabled)
2371 kmemcheck_pagealloc_alloc(page, order, gfp_mask);
2377 * This is the 'heart' of the zoned buddy allocator.
2380 __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
2381 struct zonelist *zonelist, nodemask_t *nodemask)
2383 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
2384 struct zone *preferred_zone;
2386 int migratetype = allocflags_to_migratetype(gfp_mask);
2388 gfp_mask &= gfp_allowed_mask;
2390 lockdep_trace_alloc(gfp_mask);
2392 might_sleep_if(gfp_mask & __GFP_WAIT);
2394 if (should_fail_alloc_page(gfp_mask, order))
2398 * Check the zones suitable for the gfp_mask contain at least one
2399 * valid zone. It's possible to have an empty zonelist as a result
2400 * of GFP_THISNODE and a memoryless node
2402 if (unlikely(!zonelist->_zonerefs->zone))
2406 /* The preferred zone is used for statistics later */
2407 first_zones_zonelist(zonelist, high_zoneidx,
2408 nodemask ? : &cpuset_current_mems_allowed,
2410 if (!preferred_zone) {
2415 /* First allocation attempt */
2416 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order,
2417 zonelist, high_zoneidx, ALLOC_WMARK_LOW|ALLOC_CPUSET,
2418 preferred_zone, migratetype);
2419 if (unlikely(!page))
2420 page = __alloc_pages_slowpath(gfp_mask, order,
2421 zonelist, high_zoneidx, nodemask,
2422 preferred_zone, migratetype);
2425 trace_mm_page_alloc(page, order, gfp_mask, migratetype);
2428 EXPORT_SYMBOL(__alloc_pages_nodemask);
2431 * Common helper functions.
2433 unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
2438 * __get_free_pages() returns a 32-bit address, which cannot represent
2441 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
2443 page = alloc_pages(gfp_mask, order);
2446 return (unsigned long) page_address(page);
2448 EXPORT_SYMBOL(__get_free_pages);
2450 unsigned long get_zeroed_page(gfp_t gfp_mask)
2452 return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
2454 EXPORT_SYMBOL(get_zeroed_page);
2456 void __free_pages(struct page *page, unsigned int order)
2458 if (put_page_testzero(page)) {
2460 free_hot_cold_page(page, 0);
2462 __free_pages_ok(page, order);
2466 EXPORT_SYMBOL(__free_pages);
2468 void free_pages(unsigned long addr, unsigned int order)
2471 VM_BUG_ON(!virt_addr_valid((void *)addr));
2472 __free_pages(virt_to_page((void *)addr), order);
2476 EXPORT_SYMBOL(free_pages);
2478 static void *make_alloc_exact(unsigned long addr, unsigned order, size_t size)
2481 unsigned long alloc_end = addr + (PAGE_SIZE << order);
2482 unsigned long used = addr + PAGE_ALIGN(size);
2484 split_page(virt_to_page((void *)addr), order);
2485 while (used < alloc_end) {
2490 return (void *)addr;
2494 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
2495 * @size: the number of bytes to allocate
2496 * @gfp_mask: GFP flags for the allocation
2498 * This function is similar to alloc_pages(), except that it allocates the
2499 * minimum number of pages to satisfy the request. alloc_pages() can only
2500 * allocate memory in power-of-two pages.
2502 * This function is also limited by MAX_ORDER.
2504 * Memory allocated by this function must be released by free_pages_exact().
2506 void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
2508 unsigned int order = get_order(size);
2511 addr = __get_free_pages(gfp_mask, order);
2512 return make_alloc_exact(addr, order, size);
2514 EXPORT_SYMBOL(alloc_pages_exact);
2517 * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
2519 * @nid: the preferred node ID where memory should be allocated
2520 * @size: the number of bytes to allocate
2521 * @gfp_mask: GFP flags for the allocation
2523 * Like alloc_pages_exact(), but try to allocate on node nid first before falling
2525 * Note this is not alloc_pages_exact_node() which allocates on a specific node,
2528 void *alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask)
2530 unsigned order = get_order(size);
2531 struct page *p = alloc_pages_node(nid, gfp_mask, order);
2534 return make_alloc_exact((unsigned long)page_address(p), order, size);
2536 EXPORT_SYMBOL(alloc_pages_exact_nid);
2539 * free_pages_exact - release memory allocated via alloc_pages_exact()
2540 * @virt: the value returned by alloc_pages_exact.
2541 * @size: size of allocation, same value as passed to alloc_pages_exact().
2543 * Release the memory allocated by a previous call to alloc_pages_exact.
2545 void free_pages_exact(void *virt, size_t size)
2547 unsigned long addr = (unsigned long)virt;
2548 unsigned long end = addr + PAGE_ALIGN(size);
2550 while (addr < end) {
2555 EXPORT_SYMBOL(free_pages_exact);
2557 static unsigned int nr_free_zone_pages(int offset)
2562 /* Just pick one node, since fallback list is circular */
2563 unsigned int sum = 0;
2565 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
2567 for_each_zone_zonelist(zone, z, zonelist, offset) {
2568 unsigned long size = zone->present_pages;
2569 unsigned long high = high_wmark_pages(zone);
2578 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
2580 unsigned int nr_free_buffer_pages(void)
2582 return nr_free_zone_pages(gfp_zone(GFP_USER));
2584 EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
2587 * Amount of free RAM allocatable within all zones
2589 unsigned int nr_free_pagecache_pages(void)
2591 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
2594 static inline void show_node(struct zone *zone)
2597 printk("Node %d ", zone_to_nid(zone));
2600 void si_meminfo(struct sysinfo *val)
2602 val->totalram = totalram_pages;
2604 val->freeram = global_page_state(NR_FREE_PAGES);
2605 val->bufferram = nr_blockdev_pages();
2606 val->totalhigh = totalhigh_pages;
2607 val->freehigh = nr_free_highpages();
2608 val->mem_unit = PAGE_SIZE;
2611 EXPORT_SYMBOL(si_meminfo);
2614 void si_meminfo_node(struct sysinfo *val, int nid)
2616 pg_data_t *pgdat = NODE_DATA(nid);
2618 val->totalram = pgdat->node_present_pages;
2619 val->freeram = node_page_state(nid, NR_FREE_PAGES);
2620 #ifdef CONFIG_HIGHMEM
2621 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
2622 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
2628 val->mem_unit = PAGE_SIZE;
2633 * Determine whether the node should be displayed or not, depending on whether
2634 * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
2636 bool skip_free_areas_node(unsigned int flags, int nid)
2640 if (!(flags & SHOW_MEM_FILTER_NODES))
2644 ret = !node_isset(nid, cpuset_current_mems_allowed);
2650 #define K(x) ((x) << (PAGE_SHIFT-10))
2653 * Show free area list (used inside shift_scroll-lock stuff)
2654 * We also calculate the percentage fragmentation. We do this by counting the
2655 * memory on each free list with the exception of the first item on the list.
2656 * Suppresses nodes that are not allowed by current's cpuset if
2657 * SHOW_MEM_FILTER_NODES is passed.
2659 void show_free_areas(unsigned int filter)
2664 for_each_populated_zone(zone) {
2665 if (skip_free_areas_node(filter, zone_to_nid(zone)))
2668 printk("%s per-cpu:\n", zone->name);
2670 for_each_online_cpu(cpu) {
2671 struct per_cpu_pageset *pageset;
2673 pageset = per_cpu_ptr(zone->pageset, cpu);
2675 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
2676 cpu, pageset->pcp.high,
2677 pageset->pcp.batch, pageset->pcp.count);
2681 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
2682 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
2684 " dirty:%lu writeback:%lu unstable:%lu\n"
2685 " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n"
2686 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n",
2687 global_page_state(NR_ACTIVE_ANON),
2688 global_page_state(NR_INACTIVE_ANON),
2689 global_page_state(NR_ISOLATED_ANON),
2690 global_page_state(NR_ACTIVE_FILE),
2691 global_page_state(NR_INACTIVE_FILE),
2692 global_page_state(NR_ISOLATED_FILE),
2693 global_page_state(NR_UNEVICTABLE),
2694 global_page_state(NR_FILE_DIRTY),
2695 global_page_state(NR_WRITEBACK),
2696 global_page_state(NR_UNSTABLE_NFS),
2697 global_page_state(NR_FREE_PAGES),
2698 global_page_state(NR_SLAB_RECLAIMABLE),
2699 global_page_state(NR_SLAB_UNRECLAIMABLE),
2700 global_page_state(NR_FILE_MAPPED),
2701 global_page_state(NR_SHMEM),
2702 global_page_state(NR_PAGETABLE),
2703 global_page_state(NR_BOUNCE));
2705 for_each_populated_zone(zone) {
2708 if (skip_free_areas_node(filter, zone_to_nid(zone)))
2716 " active_anon:%lukB"
2717 " inactive_anon:%lukB"
2718 " active_file:%lukB"
2719 " inactive_file:%lukB"
2720 " unevictable:%lukB"
2721 " isolated(anon):%lukB"
2722 " isolated(file):%lukB"
2729 " slab_reclaimable:%lukB"
2730 " slab_unreclaimable:%lukB"
2731 " kernel_stack:%lukB"
2735 " writeback_tmp:%lukB"
2736 " pages_scanned:%lu"
2737 " all_unreclaimable? %s"
2740 K(zone_page_state(zone, NR_FREE_PAGES)),
2741 K(min_wmark_pages(zone)),
2742 K(low_wmark_pages(zone)),
2743 K(high_wmark_pages(zone)),
2744 K(zone_page_state(zone, NR_ACTIVE_ANON)),
2745 K(zone_page_state(zone, NR_INACTIVE_ANON)),
2746 K(zone_page_state(zone, NR_ACTIVE_FILE)),
2747 K(zone_page_state(zone, NR_INACTIVE_FILE)),
2748 K(zone_page_state(zone, NR_UNEVICTABLE)),
2749 K(zone_page_state(zone, NR_ISOLATED_ANON)),
2750 K(zone_page_state(zone, NR_ISOLATED_FILE)),
2751 K(zone->present_pages),
2752 K(zone_page_state(zone, NR_MLOCK)),
2753 K(zone_page_state(zone, NR_FILE_DIRTY)),
2754 K(zone_page_state(zone, NR_WRITEBACK)),
2755 K(zone_page_state(zone, NR_FILE_MAPPED)),
2756 K(zone_page_state(zone, NR_SHMEM)),
2757 K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)),
2758 K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)),
2759 zone_page_state(zone, NR_KERNEL_STACK) *
2761 K(zone_page_state(zone, NR_PAGETABLE)),
2762 K(zone_page_state(zone, NR_UNSTABLE_NFS)),
2763 K(zone_page_state(zone, NR_BOUNCE)),
2764 K(zone_page_state(zone, NR_WRITEBACK_TEMP)),
2765 zone->pages_scanned,
2766 (zone->all_unreclaimable ? "yes" : "no")
2768 printk("lowmem_reserve[]:");
2769 for (i = 0; i < MAX_NR_ZONES; i++)
2770 printk(" %lu", zone->lowmem_reserve[i]);
2774 for_each_populated_zone(zone) {
2775 unsigned long nr[MAX_ORDER], flags, order, total = 0;
2777 if (skip_free_areas_node(filter, zone_to_nid(zone)))
2780 printk("%s: ", zone->name);
2782 spin_lock_irqsave(&zone->lock, flags);
2783 for (order = 0; order < MAX_ORDER; order++) {
2784 nr[order] = zone->free_area[order].nr_free;
2785 total += nr[order] << order;
2787 spin_unlock_irqrestore(&zone->lock, flags);
2788 for (order = 0; order < MAX_ORDER; order++)
2789 printk("%lu*%lukB ", nr[order], K(1UL) << order);
2790 printk("= %lukB\n", K(total));
2793 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
2795 show_swap_cache_info();
2798 static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
2800 zoneref->zone = zone;
2801 zoneref->zone_idx = zone_idx(zone);
2805 * Builds allocation fallback zone lists.
2807 * Add all populated zones of a node to the zonelist.
2809 static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
2810 int nr_zones, enum zone_type zone_type)
2814 BUG_ON(zone_type >= MAX_NR_ZONES);
2819 zone = pgdat->node_zones + zone_type;
2820 if (populated_zone(zone)) {
2821 zoneref_set_zone(zone,
2822 &zonelist->_zonerefs[nr_zones++]);
2823 check_highest_zone(zone_type);
2826 } while (zone_type);
2833 * 0 = automatic detection of better ordering.
2834 * 1 = order by ([node] distance, -zonetype)
2835 * 2 = order by (-zonetype, [node] distance)
2837 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
2838 * the same zonelist. So only NUMA can configure this param.
2840 #define ZONELIST_ORDER_DEFAULT 0
2841 #define ZONELIST_ORDER_NODE 1
2842 #define ZONELIST_ORDER_ZONE 2
2844 /* zonelist order in the kernel.
2845 * set_zonelist_order() will set this to NODE or ZONE.
2847 static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
2848 static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
2852 /* The value user specified ....changed by config */
2853 static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2854 /* string for sysctl */
2855 #define NUMA_ZONELIST_ORDER_LEN 16
2856 char numa_zonelist_order[16] = "default";
2859 * interface for configure zonelist ordering.
2860 * command line option "numa_zonelist_order"
2861 * = "[dD]efault - default, automatic configuration.
2862 * = "[nN]ode - order by node locality, then by zone within node
2863 * = "[zZ]one - order by zone, then by locality within zone
2866 static int __parse_numa_zonelist_order(char *s)
2868 if (*s == 'd' || *s == 'D') {
2869 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2870 } else if (*s == 'n' || *s == 'N') {
2871 user_zonelist_order = ZONELIST_ORDER_NODE;
2872 } else if (*s == 'z' || *s == 'Z') {
2873 user_zonelist_order = ZONELIST_ORDER_ZONE;
2876 "Ignoring invalid numa_zonelist_order value: "
2883 static __init int setup_numa_zonelist_order(char *s)
2890 ret = __parse_numa_zonelist_order(s);
2892 strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN);
2896 early_param("numa_zonelist_order", setup_numa_zonelist_order);
2899 * sysctl handler for numa_zonelist_order
2901 int numa_zonelist_order_handler(ctl_table *table, int write,
2902 void __user *buffer, size_t *length,
2905 char saved_string[NUMA_ZONELIST_ORDER_LEN];
2907 static DEFINE_MUTEX(zl_order_mutex);
2909 mutex_lock(&zl_order_mutex);
2911 strcpy(saved_string, (char*)table->data);
2912 ret = proc_dostring(table, write, buffer, length, ppos);
2916 int oldval = user_zonelist_order;
2917 if (__parse_numa_zonelist_order((char*)table->data)) {
2919 * bogus value. restore saved string
2921 strncpy((char*)table->data, saved_string,
2922 NUMA_ZONELIST_ORDER_LEN);
2923 user_zonelist_order = oldval;
2924 } else if (oldval != user_zonelist_order) {
2925 mutex_lock(&zonelists_mutex);
2926 build_all_zonelists(NULL);
2927 mutex_unlock(&zonelists_mutex);
2931 mutex_unlock(&zl_order_mutex);
2936 #define MAX_NODE_LOAD (nr_online_nodes)
2937 static int node_load[MAX_NUMNODES];
2940 * find_next_best_node - find the next node that should appear in a given node's fallback list
2941 * @node: node whose fallback list we're appending
2942 * @used_node_mask: nodemask_t of already used nodes
2944 * We use a number of factors to determine which is the next node that should
2945 * appear on a given node's fallback list. The node should not have appeared
2946 * already in @node's fallback list, and it should be the next closest node
2947 * according to the distance array (which contains arbitrary distance values
2948 * from each node to each node in the system), and should also prefer nodes
2949 * with no CPUs, since presumably they'll have very little allocation pressure
2950 * on them otherwise.
2951 * It returns -1 if no node is found.
2953 static int find_next_best_node(int node, nodemask_t *used_node_mask)
2956 int min_val = INT_MAX;
2958 const struct cpumask *tmp = cpumask_of_node(0);
2960 /* Use the local node if we haven't already */
2961 if (!node_isset(node, *used_node_mask)) {
2962 node_set(node, *used_node_mask);
2966 for_each_node_state(n, N_HIGH_MEMORY) {
2968 /* Don't want a node to appear more than once */
2969 if (node_isset(n, *used_node_mask))
2972 /* Use the distance array to find the distance */
2973 val = node_distance(node, n);
2975 /* Penalize nodes under us ("prefer the next node") */
2978 /* Give preference to headless and unused nodes */
2979 tmp = cpumask_of_node(n);
2980 if (!cpumask_empty(tmp))
2981 val += PENALTY_FOR_NODE_WITH_CPUS;
2983 /* Slight preference for less loaded node */
2984 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
2985 val += node_load[n];
2987 if (val < min_val) {
2994 node_set(best_node, *used_node_mask);
3001 * Build zonelists ordered by node and zones within node.
3002 * This results in maximum locality--normal zone overflows into local
3003 * DMA zone, if any--but risks exhausting DMA zone.
3005 static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
3008 struct zonelist *zonelist;
3010 zonelist = &pgdat->node_zonelists[0];
3011 for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
3013 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
3015 zonelist->_zonerefs[j].zone = NULL;
3016 zonelist->_zonerefs[j].zone_idx = 0;
3020 * Build gfp_thisnode zonelists
3022 static void build_thisnode_zonelists(pg_data_t *pgdat)
3025 struct zonelist *zonelist;
3027 zonelist = &pgdat->node_zonelists[1];
3028 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
3029 zonelist->_zonerefs[j].zone = NULL;
3030 zonelist->_zonerefs[j].zone_idx = 0;
3034 * Build zonelists ordered by zone and nodes within zones.
3035 * This results in conserving DMA zone[s] until all Normal memory is
3036 * exhausted, but results in overflowing to remote node while memory
3037 * may still exist in local DMA zone.
3039 static int node_order[MAX_NUMNODES];
3041 static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
3044 int zone_type; /* needs to be signed */
3046 struct zonelist *zonelist;
3048 zonelist = &pgdat->node_zonelists[0];
3050 for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
3051 for (j = 0; j < nr_nodes; j++) {
3052 node = node_order[j];
3053 z = &NODE_DATA(node)->node_zones[zone_type];
3054 if (populated_zone(z)) {
3056 &zonelist->_zonerefs[pos++]);
3057 check_highest_zone(zone_type);
3061 zonelist->_zonerefs[pos].zone = NULL;
3062 zonelist->_zonerefs[pos].zone_idx = 0;
3065 static int default_zonelist_order(void)
3068 unsigned long low_kmem_size,total_size;
3072 * ZONE_DMA and ZONE_DMA32 can be very small area in the system.
3073 * If they are really small and used heavily, the system can fall
3074 * into OOM very easily.
3075 * This function detect ZONE_DMA/DMA32 size and configures zone order.
3077 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
3080 for_each_online_node(nid) {
3081 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
3082 z = &NODE_DATA(nid)->node_zones[zone_type];
3083 if (populated_zone(z)) {
3084 if (zone_type < ZONE_NORMAL)
3085 low_kmem_size += z->present_pages;
3086 total_size += z->present_pages;
3087 } else if (zone_type == ZONE_NORMAL) {
3089 * If any node has only lowmem, then node order
3090 * is preferred to allow kernel allocations
3091 * locally; otherwise, they can easily infringe
3092 * on other nodes when there is an abundance of
3093 * lowmem available to allocate from.
3095 return ZONELIST_ORDER_NODE;
3099 if (!low_kmem_size || /* there are no DMA area. */
3100 low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
3101 return ZONELIST_ORDER_NODE;
3103 * look into each node's config.
3104 * If there is a node whose DMA/DMA32 memory is very big area on
3105 * local memory, NODE_ORDER may be suitable.
3107 average_size = total_size /
3108 (nodes_weight(node_states[N_HIGH_MEMORY]) + 1);
3109 for_each_online_node(nid) {
3112 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
3113 z = &NODE_DATA(nid)->node_zones[zone_type];
3114 if (populated_zone(z)) {
3115 if (zone_type < ZONE_NORMAL)
3116 low_kmem_size += z->present_pages;
3117 total_size += z->present_pages;
3120 if (low_kmem_size &&
3121 total_size > average_size && /* ignore small node */
3122 low_kmem_size > total_size * 70/100)
3123 return ZONELIST_ORDER_NODE;
3125 return ZONELIST_ORDER_ZONE;
3128 static void set_zonelist_order(void)
3130 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
3131 current_zonelist_order = default_zonelist_order();
3133 current_zonelist_order = user_zonelist_order;
3136 static void build_zonelists(pg_data_t *pgdat)
3140 nodemask_t used_mask;
3141 int local_node, prev_node;
3142 struct zonelist *zonelist;
3143 int order = current_zonelist_order;
3145 /* initialize zonelists */
3146 for (i = 0; i < MAX_ZONELISTS; i++) {
3147 zonelist = pgdat->node_zonelists + i;
3148 zonelist->_zonerefs[0].zone = NULL;
3149 zonelist->_zonerefs[0].zone_idx = 0;
3152 /* NUMA-aware ordering of nodes */
3153 local_node = pgdat->node_id;
3154 load = nr_online_nodes;
3155 prev_node = local_node;
3156 nodes_clear(used_mask);
3158 memset(node_order, 0, sizeof(node_order));
3161 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
3162 int distance = node_distance(local_node, node);
3165 * If another node is sufficiently far away then it is better
3166 * to reclaim pages in a zone before going off node.
3168 if (distance > RECLAIM_DISTANCE)
3169 zone_reclaim_mode = 1;
3172 * We don't want to pressure a particular node.
3173 * So adding penalty to the first node in same
3174 * distance group to make it round-robin.
3176 if (distance != node_distance(local_node, prev_node))
3177 node_load[node] = load;
3181 if (order == ZONELIST_ORDER_NODE)
3182 build_zonelists_in_node_order(pgdat, node);
3184 node_order[j++] = node; /* remember order */
3187 if (order == ZONELIST_ORDER_ZONE) {
3188 /* calculate node order -- i.e., DMA last! */
3189 build_zonelists_in_zone_order(pgdat, j);
3192 build_thisnode_zonelists(pgdat);
3195 /* Construct the zonelist performance cache - see further mmzone.h */
3196 static void build_zonelist_cache(pg_data_t *pgdat)
3198 struct zonelist *zonelist;
3199 struct zonelist_cache *zlc;
3202 zonelist = &pgdat->node_zonelists[0];
3203 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
3204 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
3205 for (z = zonelist->_zonerefs; z->zone; z++)
3206 zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z);
3209 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
3211 * Return node id of node used for "local" allocations.
3212 * I.e., first node id of first zone in arg node's generic zonelist.
3213 * Used for initializing percpu 'numa_mem', which is used primarily
3214 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
3216 int local_memory_node(int node)
3220 (void)first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
3221 gfp_zone(GFP_KERNEL),
3228 #else /* CONFIG_NUMA */
3230 static void set_zonelist_order(void)
3232 current_zonelist_order = ZONELIST_ORDER_ZONE;
3235 static void build_zonelists(pg_data_t *pgdat)
3237 int node, local_node;
3239 struct zonelist *zonelist;
3241 local_node = pgdat->node_id;
3243 zonelist = &pgdat->node_zonelists[0];
3244 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
3247 * Now we build the zonelist so that it contains the zones
3248 * of all the other nodes.
3249 * We don't want to pressure a particular node, so when
3250 * building the zones for node N, we make sure that the
3251 * zones coming right after the local ones are those from
3252 * node N+1 (modulo N)
3254 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
3255 if (!node_online(node))
3257 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
3260 for (node = 0; node < local_node; node++) {
3261 if (!node_online(node))
3263 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
3267 zonelist->_zonerefs[j].zone = NULL;
3268 zonelist->_zonerefs[j].zone_idx = 0;
3271 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
3272 static void build_zonelist_cache(pg_data_t *pgdat)
3274 pgdat->node_zonelists[0].zlcache_ptr = NULL;
3277 #endif /* CONFIG_NUMA */
3280 * Boot pageset table. One per cpu which is going to be used for all
3281 * zones and all nodes. The parameters will be set in such a way
3282 * that an item put on a list will immediately be handed over to
3283 * the buddy list. This is safe since pageset manipulation is done
3284 * with interrupts disabled.
3286 * The boot_pagesets must be kept even after bootup is complete for
3287 * unused processors and/or zones. They do play a role for bootstrapping
3288 * hotplugged processors.
3290 * zoneinfo_show() and maybe other functions do
3291 * not check if the processor is online before following the pageset pointer.
3292 * Other parts of the kernel may not check if the zone is available.
3294 static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch);
3295 static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset);
3296 static void setup_zone_pageset(struct zone *zone);
3299 * Global mutex to protect against size modification of zonelists
3300 * as well as to serialize pageset setup for the new populated zone.
3302 DEFINE_MUTEX(zonelists_mutex);
3304 /* return values int ....just for stop_machine() */
3305 static __init_refok int __build_all_zonelists(void *data)
3311 memset(node_load, 0, sizeof(node_load));
3313 for_each_online_node(nid) {
3314 pg_data_t *pgdat = NODE_DATA(nid);
3316 build_zonelists(pgdat);
3317 build_zonelist_cache(pgdat);
3321 * Initialize the boot_pagesets that are going to be used
3322 * for bootstrapping processors. The real pagesets for
3323 * each zone will be allocated later when the per cpu
3324 * allocator is available.
3326 * boot_pagesets are used also for bootstrapping offline
3327 * cpus if the system is already booted because the pagesets
3328 * are needed to initialize allocators on a specific cpu too.
3329 * F.e. the percpu allocator needs the page allocator which
3330 * needs the percpu allocator in order to allocate its pagesets
3331 * (a chicken-egg dilemma).
3333 for_each_possible_cpu(cpu) {
3334 setup_pageset(&per_cpu(boot_pageset, cpu), 0);
3336 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
3338 * We now know the "local memory node" for each node--
3339 * i.e., the node of the first zone in the generic zonelist.
3340 * Set up numa_mem percpu variable for on-line cpus. During
3341 * boot, only the boot cpu should be on-line; we'll init the
3342 * secondary cpus' numa_mem as they come on-line. During
3343 * node/memory hotplug, we'll fixup all on-line cpus.
3345 if (cpu_online(cpu))
3346 set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
3354 * Called with zonelists_mutex held always
3355 * unless system_state == SYSTEM_BOOTING.
3357 void __ref build_all_zonelists(void *data)
3359 set_zonelist_order();
3361 if (system_state == SYSTEM_BOOTING) {
3362 __build_all_zonelists(NULL);
3363 mminit_verify_zonelist();
3364 cpuset_init_current_mems_allowed();
3366 /* we have to stop all cpus to guarantee there is no user
3368 #ifdef CONFIG_MEMORY_HOTPLUG
3370 setup_zone_pageset((struct zone *)data);
3372 stop_machine(__build_all_zonelists, NULL, NULL);
3373 /* cpuset refresh routine should be here */
3375 vm_total_pages = nr_free_pagecache_pages();
3377 * Disable grouping by mobility if the number of pages in the
3378 * system is too low to allow the mechanism to work. It would be
3379 * more accurate, but expensive to check per-zone. This check is
3380 * made on memory-hotadd so a system can start with mobility
3381 * disabled and enable it later
3383 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
3384 page_group_by_mobility_disabled = 1;
3386 page_group_by_mobility_disabled = 0;
3388 printk("Built %i zonelists in %s order, mobility grouping %s. "
3389 "Total pages: %ld\n",
3391 zonelist_order_name[current_zonelist_order],
3392 page_group_by_mobility_disabled ? "off" : "on",
3395 printk("Policy zone: %s\n", zone_names[policy_zone]);
3400 * Helper functions to size the waitqueue hash table.
3401 * Essentially these want to choose hash table sizes sufficiently
3402 * large so that collisions trying to wait on pages are rare.
3403 * But in fact, the number of active page waitqueues on typical
3404 * systems is ridiculously low, less than 200. So this is even
3405 * conservative, even though it seems large.
3407 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
3408 * waitqueues, i.e. the size of the waitq table given the number of pages.
3410 #define PAGES_PER_WAITQUEUE 256
3412 #ifndef CONFIG_MEMORY_HOTPLUG
3413 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
3415 unsigned long size = 1;
3417 pages /= PAGES_PER_WAITQUEUE;
3419 while (size < pages)
3423 * Once we have dozens or even hundreds of threads sleeping
3424 * on IO we've got bigger problems than wait queue collision.
3425 * Limit the size of the wait table to a reasonable size.
3427 size = min(size, 4096UL);
3429 return max(size, 4UL);
3433 * A zone's size might be changed by hot-add, so it is not possible to determine
3434 * a suitable size for its wait_table. So we use the maximum size now.
3436 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
3438 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
3439 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
3440 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
3442 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
3443 * or more by the traditional way. (See above). It equals:
3445 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
3446 * ia64(16K page size) : = ( 8G + 4M)byte.
3447 * powerpc (64K page size) : = (32G +16M)byte.
3449 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
3456 * This is an integer logarithm so that shifts can be used later
3457 * to extract the more random high bits from the multiplicative
3458 * hash function before the remainder is taken.
3460 static inline unsigned long wait_table_bits(unsigned long size)
3465 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
3468 * Check if a pageblock contains reserved pages
3470 static int pageblock_is_reserved(unsigned long start_pfn, unsigned long end_pfn)
3474 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
3475 if (!pfn_valid_within(pfn) || PageReserved(pfn_to_page(pfn)))
3482 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
3483 * of blocks reserved is based on min_wmark_pages(zone). The memory within
3484 * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
3485 * higher will lead to a bigger reserve which will get freed as contiguous
3486 * blocks as reclaim kicks in
3488 static void setup_zone_migrate_reserve(struct zone *zone)
3490 unsigned long start_pfn, pfn, end_pfn, block_end_pfn;
3492 unsigned long block_migratetype;
3496 * Get the start pfn, end pfn and the number of blocks to reserve
3497 * We have to be careful to be aligned to pageblock_nr_pages to
3498 * make sure that we always check pfn_valid for the first page in
3501 start_pfn = zone->zone_start_pfn;
3502 end_pfn = start_pfn + zone->spanned_pages;
3503 start_pfn = roundup(start_pfn, pageblock_nr_pages);
3504 reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >>
3508 * Reserve blocks are generally in place to help high-order atomic
3509 * allocations that are short-lived. A min_free_kbytes value that
3510 * would result in more than 2 reserve blocks for atomic allocations
3511 * is assumed to be in place to help anti-fragmentation for the
3512 * future allocation of hugepages at runtime.
3514 reserve = min(2, reserve);
3516 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
3517 if (!pfn_valid(pfn))
3519 page = pfn_to_page(pfn);
3521 /* Watch out for overlapping nodes */
3522 if (page_to_nid(page) != zone_to_nid(zone))
3525 block_migratetype = get_pageblock_migratetype(page);
3527 /* Only test what is necessary when the reserves are not met */
3530 * Blocks with reserved pages will never free, skip
3533 block_end_pfn = min(pfn + pageblock_nr_pages, end_pfn);
3534 if (pageblock_is_reserved(pfn, block_end_pfn))
3537 /* If this block is reserved, account for it */
3538 if (block_migratetype == MIGRATE_RESERVE) {
3543 /* Suitable for reserving if this block is movable */
3544 if (block_migratetype == MIGRATE_MOVABLE) {
3545 set_pageblock_migratetype(page,
3547 move_freepages_block(zone, page,
3555 * If the reserve is met and this is a previous reserved block,
3558 if (block_migratetype == MIGRATE_RESERVE) {
3559 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
3560 move_freepages_block(zone, page, MIGRATE_MOVABLE);
3566 * Initially all pages are reserved - free ones are freed
3567 * up by free_all_bootmem() once the early boot process is
3568 * done. Non-atomic initialization, single-pass.
3570 void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
3571 unsigned long start_pfn, enum memmap_context context)
3574 unsigned long end_pfn = start_pfn + size;
3578 if (highest_memmap_pfn < end_pfn - 1)
3579 highest_memmap_pfn = end_pfn - 1;
3581 z = &NODE_DATA(nid)->node_zones[zone];
3582 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
3584 * There can be holes in boot-time mem_map[]s
3585 * handed to this function. They do not
3586 * exist on hotplugged memory.
3588 if (context == MEMMAP_EARLY) {
3589 if (!early_pfn_valid(pfn))
3591 if (!early_pfn_in_nid(pfn, nid))
3594 page = pfn_to_page(pfn);
3595 set_page_links(page, zone, nid, pfn);
3596 mminit_verify_page_links(page, zone, nid, pfn);
3597 init_page_count(page);
3598 reset_page_mapcount(page);
3599 SetPageReserved(page);
3601 * Mark the block movable so that blocks are reserved for
3602 * movable at startup. This will force kernel allocations
3603 * to reserve their blocks rather than leaking throughout
3604 * the address space during boot when many long-lived
3605 * kernel allocations are made. Later some blocks near
3606 * the start are marked MIGRATE_RESERVE by
3607 * setup_zone_migrate_reserve()
3609 * bitmap is created for zone's valid pfn range. but memmap
3610 * can be created for invalid pages (for alignment)
3611 * check here not to call set_pageblock_migratetype() against
3614 if ((z->zone_start_pfn <= pfn)
3615 && (pfn < z->zone_start_pfn + z->spanned_pages)
3616 && !(pfn & (pageblock_nr_pages - 1)))
3617 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
3619 INIT_LIST_HEAD(&page->lru);
3620 #ifdef WANT_PAGE_VIRTUAL
3621 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
3622 if (!is_highmem_idx(zone))
3623 set_page_address(page, __va(pfn << PAGE_SHIFT));
3628 static void __meminit zone_init_free_lists(struct zone *zone)
3631 for_each_migratetype_order(order, t) {
3632 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
3633 zone->free_area[order].nr_free = 0;
3637 #ifndef __HAVE_ARCH_MEMMAP_INIT
3638 #define memmap_init(size, nid, zone, start_pfn) \
3639 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
3642 static int zone_batchsize(struct zone *zone)
3648 * The per-cpu-pages pools are set to around 1000th of the
3649 * size of the zone. But no more than 1/2 of a meg.
3651 * OK, so we don't know how big the cache is. So guess.
3653 batch = zone->present_pages / 1024;
3654 if (batch * PAGE_SIZE > 512 * 1024)
3655 batch = (512 * 1024) / PAGE_SIZE;
3656 batch /= 4; /* We effectively *= 4 below */
3661 * Clamp the batch to a 2^n - 1 value. Having a power
3662 * of 2 value was found to be more likely to have
3663 * suboptimal cache aliasing properties in some cases.
3665 * For example if 2 tasks are alternately allocating
3666 * batches of pages, one task can end up with a lot
3667 * of pages of one half of the possible page colors
3668 * and the other with pages of the other colors.
3670 batch = rounddown_pow_of_two(batch + batch/2) - 1;
3675 /* The deferral and batching of frees should be suppressed under NOMMU
3678 * The problem is that NOMMU needs to be able to allocate large chunks
3679 * of contiguous memory as there's no hardware page translation to
3680 * assemble apparent contiguous memory from discontiguous pages.
3682 * Queueing large contiguous runs of pages for batching, however,
3683 * causes the pages to actually be freed in smaller chunks. As there
3684 * can be a significant delay between the individual batches being
3685 * recycled, this leads to the once large chunks of space being
3686 * fragmented and becoming unavailable for high-order allocations.
3692 static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
3694 struct per_cpu_pages *pcp;
3697 memset(p, 0, sizeof(*p));
3701 pcp->high = 6 * batch;
3702 pcp->batch = max(1UL, 1 * batch);
3703 for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
3704 INIT_LIST_HEAD(&pcp->lists[migratetype]);
3708 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
3709 * to the value high for the pageset p.
3712 static void setup_pagelist_highmark(struct per_cpu_pageset *p,
3715 struct per_cpu_pages *pcp;
3719 pcp->batch = max(1UL, high/4);
3720 if ((high/4) > (PAGE_SHIFT * 8))
3721 pcp->batch = PAGE_SHIFT * 8;
3724 static void setup_zone_pageset(struct zone *zone)
3728 zone->pageset = alloc_percpu(struct per_cpu_pageset);
3730 for_each_possible_cpu(cpu) {
3731 struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu);
3733 setup_pageset(pcp, zone_batchsize(zone));
3735 if (percpu_pagelist_fraction)
3736 setup_pagelist_highmark(pcp,
3737 (zone->present_pages /
3738 percpu_pagelist_fraction));
3743 * Allocate per cpu pagesets and initialize them.
3744 * Before this call only boot pagesets were available.
3746 void __init setup_per_cpu_pageset(void)
3750 for_each_populated_zone(zone)
3751 setup_zone_pageset(zone);
3754 static noinline __init_refok
3755 int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
3758 struct pglist_data *pgdat = zone->zone_pgdat;
3762 * The per-page waitqueue mechanism uses hashed waitqueues
3765 zone->wait_table_hash_nr_entries =
3766 wait_table_hash_nr_entries(zone_size_pages);
3767 zone->wait_table_bits =
3768 wait_table_bits(zone->wait_table_hash_nr_entries);
3769 alloc_size = zone->wait_table_hash_nr_entries
3770 * sizeof(wait_queue_head_t);
3772 if (!slab_is_available()) {
3773 zone->wait_table = (wait_queue_head_t *)
3774 alloc_bootmem_node_nopanic(pgdat, alloc_size);
3777 * This case means that a zone whose size was 0 gets new memory
3778 * via memory hot-add.
3779 * But it may be the case that a new node was hot-added. In
3780 * this case vmalloc() will not be able to use this new node's
3781 * memory - this wait_table must be initialized to use this new
3782 * node itself as well.
3783 * To use this new node's memory, further consideration will be
3786 zone->wait_table = vmalloc(alloc_size);
3788 if (!zone->wait_table)
3791 for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
3792 init_waitqueue_head(zone->wait_table + i);
3797 static int __zone_pcp_update(void *data)
3799 struct zone *zone = data;
3801 unsigned long batch = zone_batchsize(zone), flags;
3803 for_each_possible_cpu(cpu) {
3804 struct per_cpu_pageset *pset;
3805 struct per_cpu_pages *pcp;
3807 pset = per_cpu_ptr(zone->pageset, cpu);
3810 local_irq_save(flags);
3811 free_pcppages_bulk(zone, pcp->count, pcp);
3812 setup_pageset(pset, batch);
3813 local_irq_restore(flags);
3818 void zone_pcp_update(struct zone *zone)
3820 stop_machine(__zone_pcp_update, zone, NULL);
3823 static __meminit void zone_pcp_init(struct zone *zone)
3826 * per cpu subsystem is not up at this point. The following code
3827 * relies on the ability of the linker to provide the
3828 * offset of a (static) per cpu variable into the per cpu area.
3830 zone->pageset = &boot_pageset;
3832 if (zone->present_pages)
3833 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n",
3834 zone->name, zone->present_pages,
3835 zone_batchsize(zone));
3838 __meminit int init_currently_empty_zone(struct zone *zone,
3839 unsigned long zone_start_pfn,
3841 enum memmap_context context)
3843 struct pglist_data *pgdat = zone->zone_pgdat;
3845 ret = zone_wait_table_init(zone, size);
3848 pgdat->nr_zones = zone_idx(zone) + 1;
3850 zone->zone_start_pfn = zone_start_pfn;
3852 mminit_dprintk(MMINIT_TRACE, "memmap_init",
3853 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
3855 (unsigned long)zone_idx(zone),
3856 zone_start_pfn, (zone_start_pfn + size));
3858 zone_init_free_lists(zone);
3863 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
3864 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
3866 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
3867 * Architectures may implement their own version but if add_active_range()
3868 * was used and there are no special requirements, this is a convenient
3871 int __meminit __early_pfn_to_nid(unsigned long pfn)
3873 unsigned long start_pfn, end_pfn;
3876 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid)
3877 if (start_pfn <= pfn && pfn < end_pfn)
3879 /* This is a memory hole */
3882 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
3884 int __meminit early_pfn_to_nid(unsigned long pfn)
3888 nid = __early_pfn_to_nid(pfn);
3891 /* just returns 0 */
3895 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
3896 bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
3900 nid = __early_pfn_to_nid(pfn);
3901 if (nid >= 0 && nid != node)
3908 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
3909 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
3910 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
3912 * If an architecture guarantees that all ranges registered with
3913 * add_active_ranges() contain no holes and may be freed, this
3914 * this function may be used instead of calling free_bootmem() manually.
3916 void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn)
3918 unsigned long start_pfn, end_pfn;
3921 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) {
3922 start_pfn = min(start_pfn, max_low_pfn);
3923 end_pfn = min(end_pfn, max_low_pfn);
3925 if (start_pfn < end_pfn)
3926 free_bootmem_node(NODE_DATA(this_nid),
3927 PFN_PHYS(start_pfn),
3928 (end_pfn - start_pfn) << PAGE_SHIFT);
3932 int __init add_from_early_node_map(struct range *range, int az,
3933 int nr_range, int nid)
3935 unsigned long start_pfn, end_pfn;
3938 /* need to go over early_node_map to find out good range for node */
3939 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL)
3940 nr_range = add_range(range, az, nr_range, start_pfn, end_pfn);
3945 * sparse_memory_present_with_active_regions - Call memory_present for each active range
3946 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
3948 * If an architecture guarantees that all ranges registered with
3949 * add_active_ranges() contain no holes and may be freed, this
3950 * function may be used instead of calling memory_present() manually.
3952 void __init sparse_memory_present_with_active_regions(int nid)
3954 unsigned long start_pfn, end_pfn;
3957 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid)
3958 memory_present(this_nid, start_pfn, end_pfn);
3962 * get_pfn_range_for_nid - Return the start and end page frames for a node
3963 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
3964 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
3965 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
3967 * It returns the start and end page frame of a node based on information
3968 * provided by an arch calling add_active_range(). If called for a node
3969 * with no available memory, a warning is printed and the start and end
3972 void __meminit get_pfn_range_for_nid(unsigned int nid,
3973 unsigned long *start_pfn, unsigned long *end_pfn)
3975 unsigned long this_start_pfn, this_end_pfn;
3981 for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) {
3982 *start_pfn = min(*start_pfn, this_start_pfn);
3983 *end_pfn = max(*end_pfn, this_end_pfn);
3986 if (*start_pfn == -1UL)
3991 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3992 * assumption is made that zones within a node are ordered in monotonic
3993 * increasing memory addresses so that the "highest" populated zone is used
3995 static void __init find_usable_zone_for_movable(void)
3998 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
3999 if (zone_index == ZONE_MOVABLE)
4002 if (arch_zone_highest_possible_pfn[zone_index] >
4003 arch_zone_lowest_possible_pfn[zone_index])
4007 VM_BUG_ON(zone_index == -1);
4008 movable_zone = zone_index;
4012 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
4013 * because it is sized independent of architecture. Unlike the other zones,
4014 * the starting point for ZONE_MOVABLE is not fixed. It may be different
4015 * in each node depending on the size of each node and how evenly kernelcore
4016 * is distributed. This helper function adjusts the zone ranges
4017 * provided by the architecture for a given node by using the end of the
4018 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
4019 * zones within a node are in order of monotonic increases memory addresses
4021 static void __meminit adjust_zone_range_for_zone_movable(int nid,
4022 unsigned long zone_type,
4023 unsigned long node_start_pfn,
4024 unsigned long node_end_pfn,
4025 unsigned long *zone_start_pfn,
4026 unsigned long *zone_end_pfn)
4028 /* Only adjust if ZONE_MOVABLE is on this node */
4029 if (zone_movable_pfn[nid]) {
4030 /* Size ZONE_MOVABLE */
4031 if (zone_type == ZONE_MOVABLE) {
4032 *zone_start_pfn = zone_movable_pfn[nid];
4033 *zone_end_pfn = min(node_end_pfn,
4034 arch_zone_highest_possible_pfn[movable_zone]);
4036 /* Adjust for ZONE_MOVABLE starting within this range */
4037 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
4038 *zone_end_pfn > zone_movable_pfn[nid]) {
4039 *zone_end_pfn = zone_movable_pfn[nid];
4041 /* Check if this whole range is within ZONE_MOVABLE */
4042 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
4043 *zone_start_pfn = *zone_end_pfn;
4048 * Return the number of pages a zone spans in a node, including holes
4049 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
4051 static unsigned long __meminit zone_spanned_pages_in_node(int nid,
4052 unsigned long zone_type,
4053 unsigned long *ignored)
4055 unsigned long node_start_pfn, node_end_pfn;
4056 unsigned long zone_start_pfn, zone_end_pfn;
4058 /* Get the start and end of the node and zone */
4059 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
4060 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
4061 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
4062 adjust_zone_range_for_zone_movable(nid, zone_type,
4063 node_start_pfn, node_end_pfn,
4064 &zone_start_pfn, &zone_end_pfn);
4066 /* Check that this node has pages within the zone's required range */
4067 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
4070 /* Move the zone boundaries inside the node if necessary */
4071 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
4072 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
4074 /* Return the spanned pages */
4075 return zone_end_pfn - zone_start_pfn;
4079 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
4080 * then all holes in the requested range will be accounted for.
4082 unsigned long __meminit __absent_pages_in_range(int nid,
4083 unsigned long range_start_pfn,
4084 unsigned long range_end_pfn)
4086 unsigned long nr_absent = range_end_pfn - range_start_pfn;
4087 unsigned long start_pfn, end_pfn;
4090 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
4091 start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn);
4092 end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn);
4093 nr_absent -= end_pfn - start_pfn;
4099 * absent_pages_in_range - Return number of page frames in holes within a range
4100 * @start_pfn: The start PFN to start searching for holes
4101 * @end_pfn: The end PFN to stop searching for holes
4103 * It returns the number of pages frames in memory holes within a range.
4105 unsigned long __init absent_pages_in_range(unsigned long start_pfn,
4106 unsigned long end_pfn)
4108 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
4111 /* Return the number of page frames in holes in a zone on a node */
4112 static unsigned long __meminit zone_absent_pages_in_node(int nid,
4113 unsigned long zone_type,
4114 unsigned long *ignored)
4116 unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type];
4117 unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type];
4118 unsigned long node_start_pfn, node_end_pfn;
4119 unsigned long zone_start_pfn, zone_end_pfn;
4121 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
4122 zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high);
4123 zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high);
4125 adjust_zone_range_for_zone_movable(nid, zone_type,
4126 node_start_pfn, node_end_pfn,
4127 &zone_start_pfn, &zone_end_pfn);
4128 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
4131 #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
4132 static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
4133 unsigned long zone_type,
4134 unsigned long *zones_size)
4136 return zones_size[zone_type];
4139 static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
4140 unsigned long zone_type,
4141 unsigned long *zholes_size)
4146 return zholes_size[zone_type];
4149 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
4151 static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
4152 unsigned long *zones_size, unsigned long *zholes_size)
4154 unsigned long realtotalpages, totalpages = 0;
4157 for (i = 0; i < MAX_NR_ZONES; i++)
4158 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
4160 pgdat->node_spanned_pages = totalpages;
4162 realtotalpages = totalpages;
4163 for (i = 0; i < MAX_NR_ZONES; i++)
4165 zone_absent_pages_in_node(pgdat->node_id, i,
4167 pgdat->node_present_pages = realtotalpages;
4168 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
4172 #ifndef CONFIG_SPARSEMEM
4174 * Calculate the size of the zone->blockflags rounded to an unsigned long
4175 * Start by making sure zonesize is a multiple of pageblock_order by rounding
4176 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
4177 * round what is now in bits to nearest long in bits, then return it in
4180 static unsigned long __init usemap_size(unsigned long zonesize)
4182 unsigned long usemapsize;
4184 usemapsize = roundup(zonesize, pageblock_nr_pages);
4185 usemapsize = usemapsize >> pageblock_order;
4186 usemapsize *= NR_PAGEBLOCK_BITS;
4187 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
4189 return usemapsize / 8;
4192 static void __init setup_usemap(struct pglist_data *pgdat,
4193 struct zone *zone, unsigned long zonesize)
4195 unsigned long usemapsize = usemap_size(zonesize);
4196 zone->pageblock_flags = NULL;
4198 zone->pageblock_flags = alloc_bootmem_node_nopanic(pgdat,
4202 static inline void setup_usemap(struct pglist_data *pgdat,
4203 struct zone *zone, unsigned long zonesize) {}
4204 #endif /* CONFIG_SPARSEMEM */
4206 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
4208 /* Return a sensible default order for the pageblock size. */
4209 static inline int pageblock_default_order(void)
4211 if (HPAGE_SHIFT > PAGE_SHIFT)
4212 return HUGETLB_PAGE_ORDER;
4217 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
4218 static inline void __init set_pageblock_order(unsigned int order)
4220 /* Check that pageblock_nr_pages has not already been setup */
4221 if (pageblock_order)
4225 * Assume the largest contiguous order of interest is a huge page.
4226 * This value may be variable depending on boot parameters on IA64
4228 pageblock_order = order;
4230 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4233 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
4234 * and pageblock_default_order() are unused as pageblock_order is set
4235 * at compile-time. See include/linux/pageblock-flags.h for the values of
4236 * pageblock_order based on the kernel config
4238 static inline int pageblock_default_order(unsigned int order)
4242 #define set_pageblock_order(x) do {} while (0)
4244 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4247 * Set up the zone data structures:
4248 * - mark all pages reserved
4249 * - mark all memory queues empty
4250 * - clear the memory bitmaps
4252 static void __paginginit free_area_init_core(struct pglist_data *pgdat,
4253 unsigned long *zones_size, unsigned long *zholes_size)
4256 int nid = pgdat->node_id;
4257 unsigned long zone_start_pfn = pgdat->node_start_pfn;
4260 pgdat_resize_init(pgdat);
4261 pgdat->nr_zones = 0;
4262 init_waitqueue_head(&pgdat->kswapd_wait);
4263 pgdat->kswapd_max_order = 0;
4264 pgdat_page_cgroup_init(pgdat);
4266 for (j = 0; j < MAX_NR_ZONES; j++) {
4267 struct zone *zone = pgdat->node_zones + j;
4268 unsigned long size, realsize, memmap_pages;
4271 size = zone_spanned_pages_in_node(nid, j, zones_size);
4272 realsize = size - zone_absent_pages_in_node(nid, j,
4276 * Adjust realsize so that it accounts for how much memory
4277 * is used by this zone for memmap. This affects the watermark
4278 * and per-cpu initialisations
4281 PAGE_ALIGN(size * sizeof(struct page)) >> PAGE_SHIFT;
4282 if (realsize >= memmap_pages) {
4283 realsize -= memmap_pages;
4286 " %s zone: %lu pages used for memmap\n",
4287 zone_names[j], memmap_pages);
4290 " %s zone: %lu pages exceeds realsize %lu\n",
4291 zone_names[j], memmap_pages, realsize);
4293 /* Account for reserved pages */
4294 if (j == 0 && realsize > dma_reserve) {
4295 realsize -= dma_reserve;
4296 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
4297 zone_names[0], dma_reserve);
4300 if (!is_highmem_idx(j))
4301 nr_kernel_pages += realsize;
4302 nr_all_pages += realsize;
4304 zone->spanned_pages = size;
4305 zone->present_pages = realsize;
4308 zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
4310 zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100;
4312 zone->name = zone_names[j];
4313 spin_lock_init(&zone->lock);
4314 spin_lock_init(&zone->lru_lock);
4315 zone_seqlock_init(zone);
4316 zone->zone_pgdat = pgdat;
4318 zone_pcp_init(zone);
4320 INIT_LIST_HEAD(&zone->lruvec.lists[l]);
4321 zone->reclaim_stat.recent_rotated[0] = 0;
4322 zone->reclaim_stat.recent_rotated[1] = 0;
4323 zone->reclaim_stat.recent_scanned[0] = 0;
4324 zone->reclaim_stat.recent_scanned[1] = 0;
4325 zap_zone_vm_stats(zone);
4330 set_pageblock_order(pageblock_default_order());
4331 setup_usemap(pgdat, zone, size);
4332 ret = init_currently_empty_zone(zone, zone_start_pfn,
4333 size, MEMMAP_EARLY);
4335 memmap_init(size, nid, j, zone_start_pfn);
4336 zone_start_pfn += size;
4340 static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
4342 /* Skip empty nodes */
4343 if (!pgdat->node_spanned_pages)
4346 #ifdef CONFIG_FLAT_NODE_MEM_MAP
4347 /* ia64 gets its own node_mem_map, before this, without bootmem */
4348 if (!pgdat->node_mem_map) {
4349 unsigned long size, start, end;
4353 * The zone's endpoints aren't required to be MAX_ORDER
4354 * aligned but the node_mem_map endpoints must be in order
4355 * for the buddy allocator to function correctly.
4357 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
4358 end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
4359 end = ALIGN(end, MAX_ORDER_NR_PAGES);
4360 size = (end - start) * sizeof(struct page);
4361 map = alloc_remap(pgdat->node_id, size);
4363 map = alloc_bootmem_node_nopanic(pgdat, size);
4364 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
4366 #ifndef CONFIG_NEED_MULTIPLE_NODES
4368 * With no DISCONTIG, the global mem_map is just set as node 0's
4370 if (pgdat == NODE_DATA(0)) {
4371 mem_map = NODE_DATA(0)->node_mem_map;
4372 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4373 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
4374 mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
4375 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
4378 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
4381 void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
4382 unsigned long node_start_pfn, unsigned long *zholes_size)
4384 pg_data_t *pgdat = NODE_DATA(nid);
4386 pgdat->node_id = nid;
4387 pgdat->node_start_pfn = node_start_pfn;
4388 calculate_node_totalpages(pgdat, zones_size, zholes_size);
4390 alloc_node_mem_map(pgdat);
4391 #ifdef CONFIG_FLAT_NODE_MEM_MAP
4392 printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
4393 nid, (unsigned long)pgdat,
4394 (unsigned long)pgdat->node_mem_map);
4397 free_area_init_core(pgdat, zones_size, zholes_size);
4400 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4402 #if MAX_NUMNODES > 1
4404 * Figure out the number of possible node ids.
4406 static void __init setup_nr_node_ids(void)
4409 unsigned int highest = 0;
4411 for_each_node_mask(node, node_possible_map)
4413 nr_node_ids = highest + 1;
4416 static inline void setup_nr_node_ids(void)
4422 * node_map_pfn_alignment - determine the maximum internode alignment
4424 * This function should be called after node map is populated and sorted.
4425 * It calculates the maximum power of two alignment which can distinguish
4428 * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
4429 * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the
4430 * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is
4431 * shifted, 1GiB is enough and this function will indicate so.
4433 * This is used to test whether pfn -> nid mapping of the chosen memory
4434 * model has fine enough granularity to avoid incorrect mapping for the
4435 * populated node map.
4437 * Returns the determined alignment in pfn's. 0 if there is no alignment
4438 * requirement (single node).
4440 unsigned long __init node_map_pfn_alignment(void)
4442 unsigned long accl_mask = 0, last_end = 0;
4443 unsigned long start, end, mask;
4447 for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) {
4448 if (!start || last_nid < 0 || last_nid == nid) {
4455 * Start with a mask granular enough to pin-point to the
4456 * start pfn and tick off bits one-by-one until it becomes
4457 * too coarse to separate the current node from the last.
4459 mask = ~((1 << __ffs(start)) - 1);
4460 while (mask && last_end <= (start & (mask << 1)))
4463 /* accumulate all internode masks */
4467 /* convert mask to number of pages */
4468 return ~accl_mask + 1;
4471 /* Find the lowest pfn for a node */
4472 static unsigned long __init find_min_pfn_for_node(int nid)
4474 unsigned long min_pfn = ULONG_MAX;
4475 unsigned long start_pfn;
4478 for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL)
4479 min_pfn = min(min_pfn, start_pfn);
4481 if (min_pfn == ULONG_MAX) {
4483 "Could not find start_pfn for node %d\n", nid);
4491 * find_min_pfn_with_active_regions - Find the minimum PFN registered
4493 * It returns the minimum PFN based on information provided via
4494 * add_active_range().
4496 unsigned long __init find_min_pfn_with_active_regions(void)
4498 return find_min_pfn_for_node(MAX_NUMNODES);
4502 * early_calculate_totalpages()
4503 * Sum pages in active regions for movable zone.
4504 * Populate N_HIGH_MEMORY for calculating usable_nodes.
4506 static unsigned long __init early_calculate_totalpages(void)
4508 unsigned long totalpages = 0;
4509 unsigned long start_pfn, end_pfn;
4512 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
4513 unsigned long pages = end_pfn - start_pfn;
4515 totalpages += pages;
4517 node_set_state(nid, N_HIGH_MEMORY);
4523 * Find the PFN the Movable zone begins in each node. Kernel memory
4524 * is spread evenly between nodes as long as the nodes have enough
4525 * memory. When they don't, some nodes will have more kernelcore than
4528 static void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn)
4531 unsigned long usable_startpfn;
4532 unsigned long kernelcore_node, kernelcore_remaining;
4533 /* save the state before borrow the nodemask */
4534 nodemask_t saved_node_state = node_states[N_HIGH_MEMORY];
4535 unsigned long totalpages = early_calculate_totalpages();
4536 int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]);
4539 * If movablecore was specified, calculate what size of
4540 * kernelcore that corresponds so that memory usable for
4541 * any allocation type is evenly spread. If both kernelcore
4542 * and movablecore are specified, then the value of kernelcore
4543 * will be used for required_kernelcore if it's greater than
4544 * what movablecore would have allowed.
4546 if (required_movablecore) {
4547 unsigned long corepages;
4550 * Round-up so that ZONE_MOVABLE is at least as large as what
4551 * was requested by the user
4553 required_movablecore =
4554 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
4555 corepages = totalpages - required_movablecore;
4557 required_kernelcore = max(required_kernelcore, corepages);
4560 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
4561 if (!required_kernelcore)
4564 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
4565 find_usable_zone_for_movable();
4566 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
4569 /* Spread kernelcore memory as evenly as possible throughout nodes */
4570 kernelcore_node = required_kernelcore / usable_nodes;
4571 for_each_node_state(nid, N_HIGH_MEMORY) {
4572 unsigned long start_pfn, end_pfn;
4575 * Recalculate kernelcore_node if the division per node
4576 * now exceeds what is necessary to satisfy the requested
4577 * amount of memory for the kernel
4579 if (required_kernelcore < kernelcore_node)
4580 kernelcore_node = required_kernelcore / usable_nodes;
4583 * As the map is walked, we track how much memory is usable
4584 * by the kernel using kernelcore_remaining. When it is
4585 * 0, the rest of the node is usable by ZONE_MOVABLE
4587 kernelcore_remaining = kernelcore_node;
4589 /* Go through each range of PFNs within this node */
4590 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
4591 unsigned long size_pages;
4593 start_pfn = max(start_pfn, zone_movable_pfn[nid]);
4594 if (start_pfn >= end_pfn)
4597 /* Account for what is only usable for kernelcore */
4598 if (start_pfn < usable_startpfn) {
4599 unsigned long kernel_pages;
4600 kernel_pages = min(end_pfn, usable_startpfn)
4603 kernelcore_remaining -= min(kernel_pages,
4604 kernelcore_remaining);
4605 required_kernelcore -= min(kernel_pages,
4606 required_kernelcore);
4608 /* Continue if range is now fully accounted */
4609 if (end_pfn <= usable_startpfn) {
4612 * Push zone_movable_pfn to the end so
4613 * that if we have to rebalance
4614 * kernelcore across nodes, we will
4615 * not double account here
4617 zone_movable_pfn[nid] = end_pfn;
4620 start_pfn = usable_startpfn;
4624 * The usable PFN range for ZONE_MOVABLE is from
4625 * start_pfn->end_pfn. Calculate size_pages as the
4626 * number of pages used as kernelcore
4628 size_pages = end_pfn - start_pfn;
4629 if (size_pages > kernelcore_remaining)
4630 size_pages = kernelcore_remaining;
4631 zone_movable_pfn[nid] = start_pfn + size_pages;
4634 * Some kernelcore has been met, update counts and
4635 * break if the kernelcore for this node has been
4638 required_kernelcore -= min(required_kernelcore,
4640 kernelcore_remaining -= size_pages;
4641 if (!kernelcore_remaining)
4647 * If there is still required_kernelcore, we do another pass with one
4648 * less node in the count. This will push zone_movable_pfn[nid] further
4649 * along on the nodes that still have memory until kernelcore is
4653 if (usable_nodes && required_kernelcore > usable_nodes)
4656 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
4657 for (nid = 0; nid < MAX_NUMNODES; nid++)
4658 zone_movable_pfn[nid] =
4659 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
4662 /* restore the node_state */
4663 node_states[N_HIGH_MEMORY] = saved_node_state;
4666 /* Any regular memory on that node ? */
4667 static void check_for_regular_memory(pg_data_t *pgdat)
4669 #ifdef CONFIG_HIGHMEM
4670 enum zone_type zone_type;
4672 for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) {
4673 struct zone *zone = &pgdat->node_zones[zone_type];
4674 if (zone->present_pages)
4675 node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY);
4681 * free_area_init_nodes - Initialise all pg_data_t and zone data
4682 * @max_zone_pfn: an array of max PFNs for each zone
4684 * This will call free_area_init_node() for each active node in the system.
4685 * Using the page ranges provided by add_active_range(), the size of each
4686 * zone in each node and their holes is calculated. If the maximum PFN
4687 * between two adjacent zones match, it is assumed that the zone is empty.
4688 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
4689 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
4690 * starts where the previous one ended. For example, ZONE_DMA32 starts
4691 * at arch_max_dma_pfn.
4693 void __init free_area_init_nodes(unsigned long *max_zone_pfn)
4695 unsigned long start_pfn, end_pfn;
4698 /* Record where the zone boundaries are */
4699 memset(arch_zone_lowest_possible_pfn, 0,
4700 sizeof(arch_zone_lowest_possible_pfn));
4701 memset(arch_zone_highest_possible_pfn, 0,
4702 sizeof(arch_zone_highest_possible_pfn));
4703 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
4704 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
4705 for (i = 1; i < MAX_NR_ZONES; i++) {
4706 if (i == ZONE_MOVABLE)
4708 arch_zone_lowest_possible_pfn[i] =
4709 arch_zone_highest_possible_pfn[i-1];
4710 arch_zone_highest_possible_pfn[i] =
4711 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
4713 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
4714 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
4716 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
4717 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
4718 find_zone_movable_pfns_for_nodes(zone_movable_pfn);
4720 /* Print out the zone ranges */
4721 printk("Zone PFN ranges:\n");
4722 for (i = 0; i < MAX_NR_ZONES; i++) {
4723 if (i == ZONE_MOVABLE)
4725 printk(" %-8s ", zone_names[i]);
4726 if (arch_zone_lowest_possible_pfn[i] ==
4727 arch_zone_highest_possible_pfn[i])
4730 printk("%0#10lx -> %0#10lx\n",
4731 arch_zone_lowest_possible_pfn[i],
4732 arch_zone_highest_possible_pfn[i]);
4735 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
4736 printk("Movable zone start PFN for each node\n");
4737 for (i = 0; i < MAX_NUMNODES; i++) {
4738 if (zone_movable_pfn[i])
4739 printk(" Node %d: %lu\n", i, zone_movable_pfn[i]);
4742 /* Print out the early_node_map[] */
4743 printk("Early memory PFN ranges\n");
4744 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid)
4745 printk(" %3d: %0#10lx -> %0#10lx\n", nid, start_pfn, end_pfn);
4747 /* Initialise every node */
4748 mminit_verify_pageflags_layout();
4749 setup_nr_node_ids();
4750 for_each_online_node(nid) {
4751 pg_data_t *pgdat = NODE_DATA(nid);
4752 free_area_init_node(nid, NULL,
4753 find_min_pfn_for_node(nid), NULL);
4755 /* Any memory on that node */
4756 if (pgdat->node_present_pages)
4757 node_set_state(nid, N_HIGH_MEMORY);
4758 check_for_regular_memory(pgdat);
4762 static int __init cmdline_parse_core(char *p, unsigned long *core)
4764 unsigned long long coremem;
4768 coremem = memparse(p, &p);
4769 *core = coremem >> PAGE_SHIFT;
4771 /* Paranoid check that UL is enough for the coremem value */
4772 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
4778 * kernelcore=size sets the amount of memory for use for allocations that
4779 * cannot be reclaimed or migrated.
4781 static int __init cmdline_parse_kernelcore(char *p)
4783 return cmdline_parse_core(p, &required_kernelcore);
4787 * movablecore=size sets the amount of memory for use for allocations that
4788 * can be reclaimed or migrated.
4790 static int __init cmdline_parse_movablecore(char *p)
4792 return cmdline_parse_core(p, &required_movablecore);
4795 early_param("kernelcore", cmdline_parse_kernelcore);
4796 early_param("movablecore", cmdline_parse_movablecore);
4798 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
4801 * set_dma_reserve - set the specified number of pages reserved in the first zone
4802 * @new_dma_reserve: The number of pages to mark reserved
4804 * The per-cpu batchsize and zone watermarks are determined by present_pages.
4805 * In the DMA zone, a significant percentage may be consumed by kernel image
4806 * and other unfreeable allocations which can skew the watermarks badly. This
4807 * function may optionally be used to account for unfreeable pages in the
4808 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
4809 * smaller per-cpu batchsize.
4811 void __init set_dma_reserve(unsigned long new_dma_reserve)
4813 dma_reserve = new_dma_reserve;
4816 void __init free_area_init(unsigned long *zones_size)
4818 free_area_init_node(0, zones_size,
4819 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
4822 static int page_alloc_cpu_notify(struct notifier_block *self,
4823 unsigned long action, void *hcpu)
4825 int cpu = (unsigned long)hcpu;
4827 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
4831 * Spill the event counters of the dead processor
4832 * into the current processors event counters.
4833 * This artificially elevates the count of the current
4836 vm_events_fold_cpu(cpu);
4839 * Zero the differential counters of the dead processor
4840 * so that the vm statistics are consistent.
4842 * This is only okay since the processor is dead and cannot
4843 * race with what we are doing.
4845 refresh_cpu_vm_stats(cpu);
4850 void __init page_alloc_init(void)
4852 hotcpu_notifier(page_alloc_cpu_notify, 0);
4856 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
4857 * or min_free_kbytes changes.
4859 static void calculate_totalreserve_pages(void)
4861 struct pglist_data *pgdat;
4862 unsigned long reserve_pages = 0;
4863 enum zone_type i, j;
4865 for_each_online_pgdat(pgdat) {
4866 for (i = 0; i < MAX_NR_ZONES; i++) {
4867 struct zone *zone = pgdat->node_zones + i;
4868 unsigned long max = 0;
4870 /* Find valid and maximum lowmem_reserve in the zone */
4871 for (j = i; j < MAX_NR_ZONES; j++) {
4872 if (zone->lowmem_reserve[j] > max)
4873 max = zone->lowmem_reserve[j];
4876 /* we treat the high watermark as reserved pages. */
4877 max += high_wmark_pages(zone);
4879 if (max > zone->present_pages)
4880 max = zone->present_pages;
4881 reserve_pages += max;
4883 * Lowmem reserves are not available to
4884 * GFP_HIGHUSER page cache allocations and
4885 * kswapd tries to balance zones to their high
4886 * watermark. As a result, neither should be
4887 * regarded as dirtyable memory, to prevent a
4888 * situation where reclaim has to clean pages
4889 * in order to balance the zones.
4891 zone->dirty_balance_reserve = max;
4894 dirty_balance_reserve = reserve_pages;
4895 totalreserve_pages = reserve_pages;
4899 * setup_per_zone_lowmem_reserve - called whenever
4900 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
4901 * has a correct pages reserved value, so an adequate number of
4902 * pages are left in the zone after a successful __alloc_pages().
4904 static void setup_per_zone_lowmem_reserve(void)
4906 struct pglist_data *pgdat;
4907 enum zone_type j, idx;
4909 for_each_online_pgdat(pgdat) {
4910 for (j = 0; j < MAX_NR_ZONES; j++) {
4911 struct zone *zone = pgdat->node_zones + j;
4912 unsigned long present_pages = zone->present_pages;
4914 zone->lowmem_reserve[j] = 0;
4918 struct zone *lower_zone;
4922 if (sysctl_lowmem_reserve_ratio[idx] < 1)
4923 sysctl_lowmem_reserve_ratio[idx] = 1;
4925 lower_zone = pgdat->node_zones + idx;
4926 lower_zone->lowmem_reserve[j] = present_pages /
4927 sysctl_lowmem_reserve_ratio[idx];
4928 present_pages += lower_zone->present_pages;
4933 /* update totalreserve_pages */
4934 calculate_totalreserve_pages();
4938 * setup_per_zone_wmarks - called when min_free_kbytes changes
4939 * or when memory is hot-{added|removed}
4941 * Ensures that the watermark[min,low,high] values for each zone are set
4942 * correctly with respect to min_free_kbytes.
4944 void setup_per_zone_wmarks(void)
4946 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
4947 unsigned long pages_low = extra_free_kbytes >> (PAGE_SHIFT - 10);
4948 unsigned long lowmem_pages = 0;
4950 unsigned long flags;
4952 /* Calculate total number of !ZONE_HIGHMEM pages */
4953 for_each_zone(zone) {
4954 if (!is_highmem(zone))
4955 lowmem_pages += zone->present_pages;
4958 for_each_zone(zone) {
4961 spin_lock_irqsave(&zone->lock, flags);
4962 min = (u64)pages_min * zone->present_pages;
4963 do_div(min, lowmem_pages);
4964 low = (u64)pages_low * zone->present_pages;
4965 do_div(low, vm_total_pages);
4967 if (is_highmem(zone)) {
4969 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
4970 * need highmem pages, so cap pages_min to a small
4973 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
4974 * deltas controls asynch page reclaim, and so should
4975 * not be capped for highmem.
4979 min_pages = zone->present_pages / 1024;
4980 if (min_pages < SWAP_CLUSTER_MAX)
4981 min_pages = SWAP_CLUSTER_MAX;
4982 if (min_pages > 128)
4984 zone->watermark[WMARK_MIN] = min_pages;
4987 * If it's a lowmem zone, reserve a number of pages
4988 * proportionate to the zone's size.
4990 zone->watermark[WMARK_MIN] = min;
4993 zone->watermark[WMARK_LOW] = min_wmark_pages(zone) +
4995 zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) +
4997 setup_zone_migrate_reserve(zone);
4998 spin_unlock_irqrestore(&zone->lock, flags);
5001 /* update totalreserve_pages */
5002 calculate_totalreserve_pages();
5006 * The inactive anon list should be small enough that the VM never has to
5007 * do too much work, but large enough that each inactive page has a chance
5008 * to be referenced again before it is swapped out.
5010 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
5011 * INACTIVE_ANON pages on this zone's LRU, maintained by the
5012 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
5013 * the anonymous pages are kept on the inactive list.
5016 * memory ratio inactive anon
5017 * -------------------------------------
5026 static void __meminit calculate_zone_inactive_ratio(struct zone *zone)
5028 unsigned int gb, ratio;
5030 /* Zone size in gigabytes */
5031 gb = zone->present_pages >> (30 - PAGE_SHIFT);
5033 ratio = int_sqrt(10 * gb);
5037 zone->inactive_ratio = ratio;
5040 static void __meminit setup_per_zone_inactive_ratio(void)
5045 calculate_zone_inactive_ratio(zone);
5049 * Initialise min_free_kbytes.
5051 * For small machines we want it small (128k min). For large machines
5052 * we want it large (64MB max). But it is not linear, because network
5053 * bandwidth does not increase linearly with machine size. We use
5055 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
5056 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
5072 int __meminit init_per_zone_wmark_min(void)
5074 unsigned long lowmem_kbytes;
5076 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
5078 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
5079 if (min_free_kbytes < 128)
5080 min_free_kbytes = 128;
5081 if (min_free_kbytes > 65536)
5082 min_free_kbytes = 65536;
5083 setup_per_zone_wmarks();
5084 refresh_zone_stat_thresholds();
5085 setup_per_zone_lowmem_reserve();
5086 setup_per_zone_inactive_ratio();
5089 module_init(init_per_zone_wmark_min)
5092 * free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
5093 * that we can call two helper functions whenever min_free_kbytes
5094 * or extra_free_kbytes changes.
5096 int free_kbytes_sysctl_handler(ctl_table *table, int write,
5097 void __user *buffer, size_t *length, loff_t *ppos)
5099 proc_dointvec(table, write, buffer, length, ppos);
5101 setup_per_zone_wmarks();
5106 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
5107 void __user *buffer, size_t *length, loff_t *ppos)
5112 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
5117 zone->min_unmapped_pages = (zone->present_pages *
5118 sysctl_min_unmapped_ratio) / 100;
5122 int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
5123 void __user *buffer, size_t *length, loff_t *ppos)
5128 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
5133 zone->min_slab_pages = (zone->present_pages *
5134 sysctl_min_slab_ratio) / 100;
5140 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
5141 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
5142 * whenever sysctl_lowmem_reserve_ratio changes.
5144 * The reserve ratio obviously has absolutely no relation with the
5145 * minimum watermarks. The lowmem reserve ratio can only make sense
5146 * if in function of the boot time zone sizes.
5148 int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
5149 void __user *buffer, size_t *length, loff_t *ppos)
5151 proc_dointvec_minmax(table, write, buffer, length, ppos);
5152 setup_per_zone_lowmem_reserve();
5157 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
5158 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
5159 * can have before it gets flushed back to buddy allocator.
5162 int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
5163 void __user *buffer, size_t *length, loff_t *ppos)
5169 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
5170 if (!write || (ret == -EINVAL))
5172 for_each_populated_zone(zone) {
5173 for_each_possible_cpu(cpu) {
5175 high = zone->present_pages / percpu_pagelist_fraction;
5176 setup_pagelist_highmark(
5177 per_cpu_ptr(zone->pageset, cpu), high);
5183 int hashdist = HASHDIST_DEFAULT;
5186 static int __init set_hashdist(char *str)
5190 hashdist = simple_strtoul(str, &str, 0);
5193 __setup("hashdist=", set_hashdist);
5197 * allocate a large system hash table from bootmem
5198 * - it is assumed that the hash table must contain an exact power-of-2
5199 * quantity of entries
5200 * - limit is the number of hash buckets, not the total allocation size
5202 void *__init alloc_large_system_hash(const char *tablename,
5203 unsigned long bucketsize,
5204 unsigned long numentries,
5207 unsigned int *_hash_shift,
5208 unsigned int *_hash_mask,
5209 unsigned long limit)
5211 unsigned long long max = limit;
5212 unsigned long log2qty, size;
5215 /* allow the kernel cmdline to have a say */
5217 /* round applicable memory size up to nearest megabyte */
5218 numentries = nr_kernel_pages;
5219 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
5220 numentries >>= 20 - PAGE_SHIFT;
5221 numentries <<= 20 - PAGE_SHIFT;
5223 /* limit to 1 bucket per 2^scale bytes of low memory */
5224 if (scale > PAGE_SHIFT)
5225 numentries >>= (scale - PAGE_SHIFT);
5227 numentries <<= (PAGE_SHIFT - scale);
5229 /* Make sure we've got at least a 0-order allocation.. */
5230 if (unlikely(flags & HASH_SMALL)) {
5231 /* Makes no sense without HASH_EARLY */
5232 WARN_ON(!(flags & HASH_EARLY));
5233 if (!(numentries >> *_hash_shift)) {
5234 numentries = 1UL << *_hash_shift;
5235 BUG_ON(!numentries);
5237 } else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
5238 numentries = PAGE_SIZE / bucketsize;
5240 numentries = roundup_pow_of_two(numentries);
5242 /* limit allocation size to 1/16 total memory by default */
5244 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
5245 do_div(max, bucketsize);
5248 if (numentries > max)
5251 log2qty = ilog2(numentries);
5254 size = bucketsize << log2qty;
5255 if (flags & HASH_EARLY)
5256 table = alloc_bootmem_nopanic(size);
5258 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
5261 * If bucketsize is not a power-of-two, we may free
5262 * some pages at the end of hash table which
5263 * alloc_pages_exact() automatically does
5265 if (get_order(size) < MAX_ORDER) {
5266 table = alloc_pages_exact(size, GFP_ATOMIC);
5267 kmemleak_alloc(table, size, 1, GFP_ATOMIC);
5270 } while (!table && size > PAGE_SIZE && --log2qty);
5273 panic("Failed to allocate %s hash table\n", tablename);
5275 printk(KERN_INFO "%s hash table entries: %ld (order: %d, %lu bytes)\n",
5278 ilog2(size) - PAGE_SHIFT,
5282 *_hash_shift = log2qty;
5284 *_hash_mask = (1 << log2qty) - 1;
5289 /* Return a pointer to the bitmap storing bits affecting a block of pages */
5290 static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
5293 #ifdef CONFIG_SPARSEMEM
5294 return __pfn_to_section(pfn)->pageblock_flags;
5296 return zone->pageblock_flags;
5297 #endif /* CONFIG_SPARSEMEM */
5300 static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
5302 #ifdef CONFIG_SPARSEMEM
5303 pfn &= (PAGES_PER_SECTION-1);
5304 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
5306 pfn = pfn - zone->zone_start_pfn;
5307 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
5308 #endif /* CONFIG_SPARSEMEM */
5312 * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
5313 * @page: The page within the block of interest
5314 * @start_bitidx: The first bit of interest to retrieve
5315 * @end_bitidx: The last bit of interest
5316 * returns pageblock_bits flags
5318 unsigned long get_pageblock_flags_group(struct page *page,
5319 int start_bitidx, int end_bitidx)
5322 unsigned long *bitmap;
5323 unsigned long pfn, bitidx;
5324 unsigned long flags = 0;
5325 unsigned long value = 1;
5327 zone = page_zone(page);
5328 pfn = page_to_pfn(page);
5329 bitmap = get_pageblock_bitmap(zone, pfn);
5330 bitidx = pfn_to_bitidx(zone, pfn);
5332 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
5333 if (test_bit(bitidx + start_bitidx, bitmap))
5340 * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
5341 * @page: The page within the block of interest
5342 * @start_bitidx: The first bit of interest
5343 * @end_bitidx: The last bit of interest
5344 * @flags: The flags to set
5346 void set_pageblock_flags_group(struct page *page, unsigned long flags,
5347 int start_bitidx, int end_bitidx)
5350 unsigned long *bitmap;
5351 unsigned long pfn, bitidx;
5352 unsigned long value = 1;
5354 zone = page_zone(page);
5355 pfn = page_to_pfn(page);
5356 bitmap = get_pageblock_bitmap(zone, pfn);
5357 bitidx = pfn_to_bitidx(zone, pfn);
5358 VM_BUG_ON(pfn < zone->zone_start_pfn);
5359 VM_BUG_ON(pfn >= zone->zone_start_pfn + zone->spanned_pages);
5361 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
5363 __set_bit(bitidx + start_bitidx, bitmap);
5365 __clear_bit(bitidx + start_bitidx, bitmap);
5369 * This is designed as sub function...plz see page_isolation.c also.
5370 * set/clear page block's type to be ISOLATE.
5371 * page allocater never alloc memory from ISOLATE block.
5375 __count_immobile_pages(struct zone *zone, struct page *page, int count)
5377 unsigned long pfn, iter, found;
5379 * For avoiding noise data, lru_add_drain_all() should be called
5380 * If ZONE_MOVABLE, the zone never contains immobile pages
5382 if (zone_idx(zone) == ZONE_MOVABLE)
5385 if (get_pageblock_migratetype(page) == MIGRATE_MOVABLE)
5388 pfn = page_to_pfn(page);
5389 for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) {
5390 unsigned long check = pfn + iter;
5392 if (!pfn_valid_within(check))
5395 page = pfn_to_page(check);
5396 if (!page_count(page)) {
5397 if (PageBuddy(page))
5398 iter += (1 << page_order(page)) - 1;
5404 * If there are RECLAIMABLE pages, we need to check it.
5405 * But now, memory offline itself doesn't call shrink_slab()
5406 * and it still to be fixed.
5409 * If the page is not RAM, page_count()should be 0.
5410 * we don't need more check. This is an _used_ not-movable page.
5412 * The problematic thing here is PG_reserved pages. PG_reserved
5413 * is set to both of a memory hole page and a _used_ kernel
5422 bool is_pageblock_removable_nolock(struct page *page)
5424 struct zone *zone = page_zone(page);
5425 return __count_immobile_pages(zone, page, 0);
5428 int set_migratetype_isolate(struct page *page)
5431 unsigned long flags, pfn;
5432 struct memory_isolate_notify arg;
5436 zone = page_zone(page);
5438 spin_lock_irqsave(&zone->lock, flags);
5440 pfn = page_to_pfn(page);
5441 arg.start_pfn = pfn;
5442 arg.nr_pages = pageblock_nr_pages;
5443 arg.pages_found = 0;
5446 * It may be possible to isolate a pageblock even if the
5447 * migratetype is not MIGRATE_MOVABLE. The memory isolation
5448 * notifier chain is used by balloon drivers to return the
5449 * number of pages in a range that are held by the balloon
5450 * driver to shrink memory. If all the pages are accounted for
5451 * by balloons, are free, or on the LRU, isolation can continue.
5452 * Later, for example, when memory hotplug notifier runs, these
5453 * pages reported as "can be isolated" should be isolated(freed)
5454 * by the balloon driver through the memory notifier chain.
5456 notifier_ret = memory_isolate_notify(MEM_ISOLATE_COUNT, &arg);
5457 notifier_ret = notifier_to_errno(notifier_ret);
5461 * FIXME: Now, memory hotplug doesn't call shrink_slab() by itself.
5462 * We just check MOVABLE pages.
5464 if (__count_immobile_pages(zone, page, arg.pages_found))
5468 * immobile means "not-on-lru" paes. If immobile is larger than
5469 * removable-by-driver pages reported by notifier, we'll fail.
5474 set_pageblock_migratetype(page, MIGRATE_ISOLATE);
5475 move_freepages_block(zone, page, MIGRATE_ISOLATE);
5478 spin_unlock_irqrestore(&zone->lock, flags);
5484 void unset_migratetype_isolate(struct page *page)
5487 unsigned long flags;
5488 zone = page_zone(page);
5489 spin_lock_irqsave(&zone->lock, flags);
5490 if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE)
5492 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
5493 move_freepages_block(zone, page, MIGRATE_MOVABLE);
5495 spin_unlock_irqrestore(&zone->lock, flags);
5498 #ifdef CONFIG_MEMORY_HOTREMOVE
5500 * All pages in the range must be isolated before calling this.
5503 __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
5509 unsigned long flags;
5510 /* find the first valid pfn */
5511 for (pfn = start_pfn; pfn < end_pfn; pfn++)
5516 zone = page_zone(pfn_to_page(pfn));
5517 spin_lock_irqsave(&zone->lock, flags);
5519 while (pfn < end_pfn) {
5520 if (!pfn_valid(pfn)) {
5524 page = pfn_to_page(pfn);
5525 BUG_ON(page_count(page));
5526 BUG_ON(!PageBuddy(page));
5527 order = page_order(page);
5528 #ifdef CONFIG_DEBUG_VM
5529 printk(KERN_INFO "remove from free list %lx %d %lx\n",
5530 pfn, 1 << order, end_pfn);
5532 list_del(&page->lru);
5533 rmv_page_order(page);
5534 zone->free_area[order].nr_free--;
5535 __mod_zone_page_state(zone, NR_FREE_PAGES,
5537 for (i = 0; i < (1 << order); i++)
5538 SetPageReserved((page+i));
5539 pfn += (1 << order);
5541 spin_unlock_irqrestore(&zone->lock, flags);
5545 #ifdef CONFIG_MEMORY_FAILURE
5546 bool is_free_buddy_page(struct page *page)
5548 struct zone *zone = page_zone(page);
5549 unsigned long pfn = page_to_pfn(page);
5550 unsigned long flags;
5553 spin_lock_irqsave(&zone->lock, flags);
5554 for (order = 0; order < MAX_ORDER; order++) {
5555 struct page *page_head = page - (pfn & ((1 << order) - 1));
5557 if (PageBuddy(page_head) && page_order(page_head) >= order)
5560 spin_unlock_irqrestore(&zone->lock, flags);
5562 return order < MAX_ORDER;
5566 static struct trace_print_flags pageflag_names[] = {
5567 {1UL << PG_locked, "locked" },
5568 {1UL << PG_error, "error" },
5569 {1UL << PG_referenced, "referenced" },
5570 {1UL << PG_uptodate, "uptodate" },
5571 {1UL << PG_dirty, "dirty" },
5572 {1UL << PG_lru, "lru" },
5573 {1UL << PG_active, "active" },
5574 {1UL << PG_slab, "slab" },
5575 {1UL << PG_owner_priv_1, "owner_priv_1" },
5576 {1UL << PG_arch_1, "arch_1" },
5577 {1UL << PG_reserved, "reserved" },
5578 {1UL << PG_private, "private" },
5579 {1UL << PG_private_2, "private_2" },
5580 {1UL << PG_writeback, "writeback" },
5581 #ifdef CONFIG_PAGEFLAGS_EXTENDED
5582 {1UL << PG_head, "head" },
5583 {1UL << PG_tail, "tail" },
5585 {1UL << PG_compound, "compound" },
5587 {1UL << PG_swapcache, "swapcache" },
5588 {1UL << PG_mappedtodisk, "mappedtodisk" },
5589 {1UL << PG_reclaim, "reclaim" },
5590 {1UL << PG_swapbacked, "swapbacked" },
5591 {1UL << PG_unevictable, "unevictable" },
5593 {1UL << PG_mlocked, "mlocked" },
5595 #ifdef CONFIG_ARCH_USES_PG_UNCACHED
5596 {1UL << PG_uncached, "uncached" },
5598 #ifdef CONFIG_MEMORY_FAILURE
5599 {1UL << PG_hwpoison, "hwpoison" },
5604 static void dump_page_flags(unsigned long flags)
5606 const char *delim = "";
5610 printk(KERN_ALERT "page flags: %#lx(", flags);
5612 /* remove zone id */
5613 flags &= (1UL << NR_PAGEFLAGS) - 1;
5615 for (i = 0; pageflag_names[i].name && flags; i++) {
5617 mask = pageflag_names[i].mask;
5618 if ((flags & mask) != mask)
5622 printk("%s%s", delim, pageflag_names[i].name);
5626 /* check for left over flags */
5628 printk("%s%#lx", delim, flags);
5633 void dump_page(struct page *page)
5636 "page:%p count:%d mapcount:%d mapping:%p index:%#lx\n",
5637 page, atomic_read(&page->_count), page_mapcount(page),
5638 page->mapping, page->index);
5639 dump_page_flags(page->flags);
5640 mem_cgroup_print_bad_page(page);