2 * linux/mm/page_alloc.c
4 * Manages the free list, the system allocates free pages here.
5 * Note that kmalloc() lives in slab.c
7 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
8 * Swap reorganised 29.12.95, Stephen Tweedie
9 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
10 * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999
11 * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
12 * Zone balancing, Kanoj Sarcar, SGI, Jan 2000
13 * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002
14 * (lots of bits borrowed from Ingo Molnar & Andrew Morton)
17 #include <linux/stddef.h>
19 #include <linux/swap.h>
20 #include <linux/interrupt.h>
21 #include <linux/pagemap.h>
22 #include <linux/jiffies.h>
23 #include <linux/bootmem.h>
24 #include <linux/memblock.h>
25 #include <linux/compiler.h>
26 #include <linux/kernel.h>
27 #include <linux/kmemcheck.h>
28 #include <linux/module.h>
29 #include <linux/suspend.h>
30 #include <linux/pagevec.h>
31 #include <linux/blkdev.h>
32 #include <linux/slab.h>
33 #include <linux/oom.h>
34 #include <linux/notifier.h>
35 #include <linux/topology.h>
36 #include <linux/sysctl.h>
37 #include <linux/cpu.h>
38 #include <linux/cpuset.h>
39 #include <linux/memory_hotplug.h>
40 #include <linux/nodemask.h>
41 #include <linux/vmalloc.h>
42 #include <linux/mempolicy.h>
43 #include <linux/stop_machine.h>
44 #include <linux/sort.h>
45 #include <linux/pfn.h>
46 #include <linux/backing-dev.h>
47 #include <linux/fault-inject.h>
48 #include <linux/page-isolation.h>
49 #include <linux/page_cgroup.h>
50 #include <linux/debugobjects.h>
51 #include <linux/kmemleak.h>
52 #include <linux/memory.h>
53 #include <linux/compaction.h>
54 #include <trace/events/kmem.h>
55 #include <linux/ftrace_event.h>
57 #include <asm/tlbflush.h>
58 #include <asm/div64.h>
61 #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
62 DEFINE_PER_CPU(int, numa_node);
63 EXPORT_PER_CPU_SYMBOL(numa_node);
66 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
68 * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
69 * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
70 * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
71 * defined in <linux/topology.h>.
73 DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */
74 EXPORT_PER_CPU_SYMBOL(_numa_mem_);
78 * Array of node states.
80 nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
81 [N_POSSIBLE] = NODE_MASK_ALL,
82 [N_ONLINE] = { { [0] = 1UL } },
84 [N_NORMAL_MEMORY] = { { [0] = 1UL } },
86 [N_HIGH_MEMORY] = { { [0] = 1UL } },
88 [N_CPU] = { { [0] = 1UL } },
91 EXPORT_SYMBOL(node_states);
93 unsigned long totalram_pages __read_mostly;
94 unsigned long totalreserve_pages __read_mostly;
95 int percpu_pagelist_fraction;
96 gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK;
98 #ifdef CONFIG_PM_SLEEP
100 * The following functions are used by the suspend/hibernate code to temporarily
101 * change gfp_allowed_mask in order to avoid using I/O during memory allocations
102 * while devices are suspended. To avoid races with the suspend/hibernate code,
103 * they should always be called with pm_mutex held (gfp_allowed_mask also should
104 * only be modified with pm_mutex held, unless the suspend/hibernate code is
105 * guaranteed not to run in parallel with that modification).
108 static gfp_t saved_gfp_mask;
110 void pm_restore_gfp_mask(void)
112 WARN_ON(!mutex_is_locked(&pm_mutex));
113 if (saved_gfp_mask) {
114 gfp_allowed_mask = saved_gfp_mask;
119 void pm_restrict_gfp_mask(void)
121 WARN_ON(!mutex_is_locked(&pm_mutex));
122 WARN_ON(saved_gfp_mask);
123 saved_gfp_mask = gfp_allowed_mask;
124 gfp_allowed_mask &= ~GFP_IOFS;
126 #endif /* CONFIG_PM_SLEEP */
128 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
129 int pageblock_order __read_mostly;
132 static void __free_pages_ok(struct page *page, unsigned int order);
135 * results with 256, 32 in the lowmem_reserve sysctl:
136 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
137 * 1G machine -> (16M dma, 784M normal, 224M high)
138 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
139 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
140 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
142 * TBD: should special case ZONE_DMA32 machines here - in those we normally
143 * don't need any ZONE_NORMAL reservation
145 int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
146 #ifdef CONFIG_ZONE_DMA
149 #ifdef CONFIG_ZONE_DMA32
152 #ifdef CONFIG_HIGHMEM
158 EXPORT_SYMBOL(totalram_pages);
160 static char * const zone_names[MAX_NR_ZONES] = {
161 #ifdef CONFIG_ZONE_DMA
164 #ifdef CONFIG_ZONE_DMA32
168 #ifdef CONFIG_HIGHMEM
174 int min_free_kbytes = 1024;
176 static unsigned long __meminitdata nr_kernel_pages;
177 static unsigned long __meminitdata nr_all_pages;
178 static unsigned long __meminitdata dma_reserve;
180 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
182 * MAX_ACTIVE_REGIONS determines the maximum number of distinct
183 * ranges of memory (RAM) that may be registered with add_active_range().
184 * Ranges passed to add_active_range() will be merged if possible
185 * so the number of times add_active_range() can be called is
186 * related to the number of nodes and the number of holes
188 #ifdef CONFIG_MAX_ACTIVE_REGIONS
189 /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
190 #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
192 #if MAX_NUMNODES >= 32
193 /* If there can be many nodes, allow up to 50 holes per node */
194 #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
196 /* By default, allow up to 256 distinct regions */
197 #define MAX_ACTIVE_REGIONS 256
201 static struct node_active_region __meminitdata early_node_map[MAX_ACTIVE_REGIONS];
202 static int __meminitdata nr_nodemap_entries;
203 static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
204 static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
205 static unsigned long __initdata required_kernelcore;
206 static unsigned long __initdata required_movablecore;
207 static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
209 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
211 EXPORT_SYMBOL(movable_zone);
212 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
215 int nr_node_ids __read_mostly = MAX_NUMNODES;
216 int nr_online_nodes __read_mostly = 1;
217 EXPORT_SYMBOL(nr_node_ids);
218 EXPORT_SYMBOL(nr_online_nodes);
221 int page_group_by_mobility_disabled __read_mostly;
223 static void set_pageblock_migratetype(struct page *page, int migratetype)
226 if (unlikely(page_group_by_mobility_disabled))
227 migratetype = MIGRATE_UNMOVABLE;
229 set_pageblock_flags_group(page, (unsigned long)migratetype,
230 PB_migrate, PB_migrate_end);
233 bool oom_killer_disabled __read_mostly;
235 #ifdef CONFIG_DEBUG_VM
236 static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
240 unsigned long pfn = page_to_pfn(page);
243 seq = zone_span_seqbegin(zone);
244 if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
246 else if (pfn < zone->zone_start_pfn)
248 } while (zone_span_seqretry(zone, seq));
253 static int page_is_consistent(struct zone *zone, struct page *page)
255 if (!pfn_valid_within(page_to_pfn(page)))
257 if (zone != page_zone(page))
263 * Temporary debugging check for pages not lying within a given zone.
265 static int bad_range(struct zone *zone, struct page *page)
267 if (page_outside_zone_boundaries(zone, page))
269 if (!page_is_consistent(zone, page))
275 static inline int bad_range(struct zone *zone, struct page *page)
281 static void bad_page(struct page *page)
283 static unsigned long resume;
284 static unsigned long nr_shown;
285 static unsigned long nr_unshown;
287 /* Don't complain about poisoned pages */
288 if (PageHWPoison(page)) {
289 reset_page_mapcount(page); /* remove PageBuddy */
294 * Allow a burst of 60 reports, then keep quiet for that minute;
295 * or allow a steady drip of one report per second.
297 if (nr_shown == 60) {
298 if (time_before(jiffies, resume)) {
304 "BUG: Bad page state: %lu messages suppressed\n",
311 resume = jiffies + 60 * HZ;
313 printk(KERN_ALERT "BUG: Bad page state in process %s pfn:%05lx\n",
314 current->comm, page_to_pfn(page));
319 /* Leave bad fields for debug, except PageBuddy could make trouble */
320 reset_page_mapcount(page); /* remove PageBuddy */
321 add_taint(TAINT_BAD_PAGE);
325 * Higher-order pages are called "compound pages". They are structured thusly:
327 * The first PAGE_SIZE page is called the "head page".
329 * The remaining PAGE_SIZE pages are called "tail pages".
331 * All pages have PG_compound set. All pages have their ->private pointing at
332 * the head page (even the head page has this).
334 * The first tail page's ->lru.next holds the address of the compound page's
335 * put_page() function. Its ->lru.prev holds the order of allocation.
336 * This usage means that zero-order pages may not be compound.
339 static void free_compound_page(struct page *page)
341 __free_pages_ok(page, compound_order(page));
344 void prep_compound_page(struct page *page, unsigned long order)
347 int nr_pages = 1 << order;
349 set_compound_page_dtor(page, free_compound_page);
350 set_compound_order(page, order);
352 for (i = 1; i < nr_pages; i++) {
353 struct page *p = page + i;
356 p->first_page = page;
360 /* update __split_huge_page_refcount if you change this function */
361 static int destroy_compound_page(struct page *page, unsigned long order)
364 int nr_pages = 1 << order;
367 if (unlikely(compound_order(page) != order) ||
368 unlikely(!PageHead(page))) {
373 __ClearPageHead(page);
375 for (i = 1; i < nr_pages; i++) {
376 struct page *p = page + i;
378 if (unlikely(!PageTail(p) || (p->first_page != page))) {
388 static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
393 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
394 * and __GFP_HIGHMEM from hard or soft interrupt context.
396 VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
397 for (i = 0; i < (1 << order); i++)
398 clear_highpage(page + i);
401 static inline void set_page_order(struct page *page, int order)
403 set_page_private(page, order);
404 __SetPageBuddy(page);
407 static inline void rmv_page_order(struct page *page)
409 __ClearPageBuddy(page);
410 set_page_private(page, 0);
414 * Locate the struct page for both the matching buddy in our
415 * pair (buddy1) and the combined O(n+1) page they form (page).
417 * 1) Any buddy B1 will have an order O twin B2 which satisfies
418 * the following equation:
420 * For example, if the starting buddy (buddy2) is #8 its order
422 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
424 * 2) Any buddy B will have an order O+1 parent P which
425 * satisfies the following equation:
428 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
430 static inline unsigned long
431 __find_buddy_index(unsigned long page_idx, unsigned int order)
433 return page_idx ^ (1 << order);
437 * This function checks whether a page is free && is the buddy
438 * we can do coalesce a page and its buddy if
439 * (a) the buddy is not in a hole &&
440 * (b) the buddy is in the buddy system &&
441 * (c) a page and its buddy have the same order &&
442 * (d) a page and its buddy are in the same zone.
444 * For recording whether a page is in the buddy system, we set ->_mapcount -2.
445 * Setting, clearing, and testing _mapcount -2 is serialized by zone->lock.
447 * For recording page's order, we use page_private(page).
449 static inline int page_is_buddy(struct page *page, struct page *buddy,
452 if (!pfn_valid_within(page_to_pfn(buddy)))
455 if (page_zone_id(page) != page_zone_id(buddy))
458 if (PageBuddy(buddy) && page_order(buddy) == order) {
459 VM_BUG_ON(page_count(buddy) != 0);
466 * Freeing function for a buddy system allocator.
468 * The concept of a buddy system is to maintain direct-mapped table
469 * (containing bit values) for memory blocks of various "orders".
470 * The bottom level table contains the map for the smallest allocatable
471 * units of memory (here, pages), and each level above it describes
472 * pairs of units from the levels below, hence, "buddies".
473 * At a high level, all that happens here is marking the table entry
474 * at the bottom level available, and propagating the changes upward
475 * as necessary, plus some accounting needed to play nicely with other
476 * parts of the VM system.
477 * At each level, we keep a list of pages, which are heads of continuous
478 * free pages of length of (1 << order) and marked with _mapcount -2. Page's
479 * order is recorded in page_private(page) field.
480 * So when we are allocating or freeing one, we can derive the state of the
481 * other. That is, if we allocate a small block, and both were
482 * free, the remainder of the region must be split into blocks.
483 * If a block is freed, and its buddy is also free, then this
484 * triggers coalescing into a block of larger size.
489 static inline void __free_one_page(struct page *page,
490 struct zone *zone, unsigned int order,
493 unsigned long page_idx;
494 unsigned long combined_idx;
495 unsigned long uninitialized_var(buddy_idx);
498 if (unlikely(PageCompound(page)))
499 if (unlikely(destroy_compound_page(page, order)))
502 VM_BUG_ON(migratetype == -1);
504 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
506 VM_BUG_ON(page_idx & ((1 << order) - 1));
507 VM_BUG_ON(bad_range(zone, page));
509 while (order < MAX_ORDER-1) {
510 buddy_idx = __find_buddy_index(page_idx, order);
511 buddy = page + (buddy_idx - page_idx);
512 if (!page_is_buddy(page, buddy, order))
515 /* Our buddy is free, merge with it and move up one order. */
516 list_del(&buddy->lru);
517 zone->free_area[order].nr_free--;
518 rmv_page_order(buddy);
519 combined_idx = buddy_idx & page_idx;
520 page = page + (combined_idx - page_idx);
521 page_idx = combined_idx;
524 set_page_order(page, order);
527 * If this is not the largest possible page, check if the buddy
528 * of the next-highest order is free. If it is, it's possible
529 * that pages are being freed that will coalesce soon. In case,
530 * that is happening, add the free page to the tail of the list
531 * so it's less likely to be used soon and more likely to be merged
532 * as a higher order page
534 if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) {
535 struct page *higher_page, *higher_buddy;
536 combined_idx = buddy_idx & page_idx;
537 higher_page = page + (combined_idx - page_idx);
538 buddy_idx = __find_buddy_index(combined_idx, order + 1);
539 higher_buddy = page + (buddy_idx - combined_idx);
540 if (page_is_buddy(higher_page, higher_buddy, order + 1)) {
541 list_add_tail(&page->lru,
542 &zone->free_area[order].free_list[migratetype]);
547 list_add(&page->lru, &zone->free_area[order].free_list[migratetype]);
549 zone->free_area[order].nr_free++;
553 * free_page_mlock() -- clean up attempts to free and mlocked() page.
554 * Page should not be on lru, so no need to fix that up.
555 * free_pages_check() will verify...
557 static inline void free_page_mlock(struct page *page)
559 __dec_zone_page_state(page, NR_MLOCK);
560 __count_vm_event(UNEVICTABLE_MLOCKFREED);
563 static inline int free_pages_check(struct page *page)
565 if (unlikely(page_mapcount(page) |
566 (page->mapping != NULL) |
567 (atomic_read(&page->_count) != 0) |
568 (page->flags & PAGE_FLAGS_CHECK_AT_FREE))) {
572 if (page->flags & PAGE_FLAGS_CHECK_AT_PREP)
573 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
578 * Frees a number of pages from the PCP lists
579 * Assumes all pages on list are in same zone, and of same order.
580 * count is the number of pages to free.
582 * If the zone was previously in an "all pages pinned" state then look to
583 * see if this freeing clears that state.
585 * And clear the zone's pages_scanned counter, to hold off the "all pages are
586 * pinned" detection logic.
588 static void free_pcppages_bulk(struct zone *zone, int count,
589 struct per_cpu_pages *pcp)
595 spin_lock(&zone->lock);
596 zone->all_unreclaimable = 0;
597 zone->pages_scanned = 0;
601 struct list_head *list;
604 * Remove pages from lists in a round-robin fashion. A
605 * batch_free count is maintained that is incremented when an
606 * empty list is encountered. This is so more pages are freed
607 * off fuller lists instead of spinning excessively around empty
612 if (++migratetype == MIGRATE_PCPTYPES)
614 list = &pcp->lists[migratetype];
615 } while (list_empty(list));
618 page = list_entry(list->prev, struct page, lru);
619 /* must delete as __free_one_page list manipulates */
620 list_del(&page->lru);
621 /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */
622 __free_one_page(page, zone, 0, page_private(page));
623 trace_mm_page_pcpu_drain(page, 0, page_private(page));
624 } while (--to_free && --batch_free && !list_empty(list));
626 __mod_zone_page_state(zone, NR_FREE_PAGES, count);
627 spin_unlock(&zone->lock);
630 static void free_one_page(struct zone *zone, struct page *page, int order,
633 spin_lock(&zone->lock);
634 zone->all_unreclaimable = 0;
635 zone->pages_scanned = 0;
637 __free_one_page(page, zone, order, migratetype);
638 __mod_zone_page_state(zone, NR_FREE_PAGES, 1 << order);
639 spin_unlock(&zone->lock);
642 static bool free_pages_prepare(struct page *page, unsigned int order)
647 trace_mm_page_free_direct(page, order);
648 kmemcheck_free_shadow(page, order);
651 page->mapping = NULL;
652 for (i = 0; i < (1 << order); i++)
653 bad += free_pages_check(page + i);
657 if (!PageHighMem(page)) {
658 debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order);
659 debug_check_no_obj_freed(page_address(page),
662 arch_free_page(page, order);
663 kernel_map_pages(page, 1 << order, 0);
668 static void __free_pages_ok(struct page *page, unsigned int order)
671 int wasMlocked = __TestClearPageMlocked(page);
673 if (!free_pages_prepare(page, order))
676 local_irq_save(flags);
677 if (unlikely(wasMlocked))
678 free_page_mlock(page);
679 __count_vm_events(PGFREE, 1 << order);
680 free_one_page(page_zone(page), page, order,
681 get_pageblock_migratetype(page));
682 local_irq_restore(flags);
686 * permit the bootmem allocator to evade page validation on high-order frees
688 void __meminit __free_pages_bootmem(struct page *page, unsigned int order)
691 __ClearPageReserved(page);
692 set_page_count(page, 0);
693 set_page_refcounted(page);
699 for (loop = 0; loop < BITS_PER_LONG; loop++) {
700 struct page *p = &page[loop];
702 if (loop + 1 < BITS_PER_LONG)
704 __ClearPageReserved(p);
705 set_page_count(p, 0);
708 set_page_refcounted(page);
709 __free_pages(page, order);
715 * The order of subdivision here is critical for the IO subsystem.
716 * Please do not alter this order without good reasons and regression
717 * testing. Specifically, as large blocks of memory are subdivided,
718 * the order in which smaller blocks are delivered depends on the order
719 * they're subdivided in this function. This is the primary factor
720 * influencing the order in which pages are delivered to the IO
721 * subsystem according to empirical testing, and this is also justified
722 * by considering the behavior of a buddy system containing a single
723 * large block of memory acted on by a series of small allocations.
724 * This behavior is a critical factor in sglist merging's success.
728 static inline void expand(struct zone *zone, struct page *page,
729 int low, int high, struct free_area *area,
732 unsigned long size = 1 << high;
738 VM_BUG_ON(bad_range(zone, &page[size]));
739 list_add(&page[size].lru, &area->free_list[migratetype]);
741 set_page_order(&page[size], high);
746 * This page is about to be returned from the page allocator
748 static inline int check_new_page(struct page *page)
750 if (unlikely(page_mapcount(page) |
751 (page->mapping != NULL) |
752 (atomic_read(&page->_count) != 0) |
753 (page->flags & PAGE_FLAGS_CHECK_AT_PREP))) {
760 static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
764 for (i = 0; i < (1 << order); i++) {
765 struct page *p = page + i;
766 if (unlikely(check_new_page(p)))
770 set_page_private(page, 0);
771 set_page_refcounted(page);
773 arch_alloc_page(page, order);
774 kernel_map_pages(page, 1 << order, 1);
776 if (gfp_flags & __GFP_ZERO)
777 prep_zero_page(page, order, gfp_flags);
779 if (order && (gfp_flags & __GFP_COMP))
780 prep_compound_page(page, order);
786 * Go through the free lists for the given migratetype and remove
787 * the smallest available page from the freelists
790 struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
793 unsigned int current_order;
794 struct free_area * area;
797 /* Find a page of the appropriate size in the preferred list */
798 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
799 area = &(zone->free_area[current_order]);
800 if (list_empty(&area->free_list[migratetype]))
803 page = list_entry(area->free_list[migratetype].next,
805 list_del(&page->lru);
806 rmv_page_order(page);
808 expand(zone, page, order, current_order, area, migratetype);
817 * This array describes the order lists are fallen back to when
818 * the free lists for the desirable migrate type are depleted
820 static int fallbacks[MIGRATE_TYPES][MIGRATE_TYPES-1] = {
821 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
822 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
823 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
824 [MIGRATE_RESERVE] = { MIGRATE_RESERVE, MIGRATE_RESERVE, MIGRATE_RESERVE }, /* Never used */
828 * Move the free pages in a range to the free lists of the requested type.
829 * Note that start_page and end_pages are not aligned on a pageblock
830 * boundary. If alignment is required, use move_freepages_block()
832 static int move_freepages(struct zone *zone,
833 struct page *start_page, struct page *end_page,
840 #ifndef CONFIG_HOLES_IN_ZONE
842 * page_zone is not safe to call in this context when
843 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
844 * anyway as we check zone boundaries in move_freepages_block().
845 * Remove at a later date when no bug reports exist related to
846 * grouping pages by mobility
848 BUG_ON(page_zone(start_page) != page_zone(end_page));
851 for (page = start_page; page <= end_page;) {
852 /* Make sure we are not inadvertently changing nodes */
853 VM_BUG_ON(page_to_nid(page) != zone_to_nid(zone));
855 if (!pfn_valid_within(page_to_pfn(page))) {
860 if (!PageBuddy(page)) {
865 order = page_order(page);
866 list_del(&page->lru);
868 &zone->free_area[order].free_list[migratetype]);
870 pages_moved += 1 << order;
876 static int move_freepages_block(struct zone *zone, struct page *page,
879 unsigned long start_pfn, end_pfn;
880 struct page *start_page, *end_page;
882 start_pfn = page_to_pfn(page);
883 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
884 start_page = pfn_to_page(start_pfn);
885 end_page = start_page + pageblock_nr_pages - 1;
886 end_pfn = start_pfn + pageblock_nr_pages - 1;
888 /* Do not cross zone boundaries */
889 if (start_pfn < zone->zone_start_pfn)
891 if (end_pfn >= zone->zone_start_pfn + zone->spanned_pages)
894 return move_freepages(zone, start_page, end_page, migratetype);
897 static void change_pageblock_range(struct page *pageblock_page,
898 int start_order, int migratetype)
900 int nr_pageblocks = 1 << (start_order - pageblock_order);
902 while (nr_pageblocks--) {
903 set_pageblock_migratetype(pageblock_page, migratetype);
904 pageblock_page += pageblock_nr_pages;
908 /* Remove an element from the buddy allocator from the fallback list */
909 static inline struct page *
910 __rmqueue_fallback(struct zone *zone, int order, int start_migratetype)
912 struct free_area * area;
917 /* Find the largest possible block of pages in the other list */
918 for (current_order = MAX_ORDER-1; current_order >= order;
920 for (i = 0; i < MIGRATE_TYPES - 1; i++) {
921 migratetype = fallbacks[start_migratetype][i];
923 /* MIGRATE_RESERVE handled later if necessary */
924 if (migratetype == MIGRATE_RESERVE)
927 area = &(zone->free_area[current_order]);
928 if (list_empty(&area->free_list[migratetype]))
931 page = list_entry(area->free_list[migratetype].next,
936 * If breaking a large block of pages, move all free
937 * pages to the preferred allocation list. If falling
938 * back for a reclaimable kernel allocation, be more
939 * agressive about taking ownership of free pages
941 if (unlikely(current_order >= (pageblock_order >> 1)) ||
942 start_migratetype == MIGRATE_RECLAIMABLE ||
943 page_group_by_mobility_disabled) {
945 pages = move_freepages_block(zone, page,
948 /* Claim the whole block if over half of it is free */
949 if (pages >= (1 << (pageblock_order-1)) ||
950 page_group_by_mobility_disabled)
951 set_pageblock_migratetype(page,
954 migratetype = start_migratetype;
957 /* Remove the page from the freelists */
958 list_del(&page->lru);
959 rmv_page_order(page);
961 /* Take ownership for orders >= pageblock_order */
962 if (current_order >= pageblock_order)
963 change_pageblock_range(page, current_order,
966 expand(zone, page, order, current_order, area, migratetype);
968 trace_mm_page_alloc_extfrag(page, order, current_order,
969 start_migratetype, migratetype);
979 * Do the hard work of removing an element from the buddy allocator.
980 * Call me with the zone->lock already held.
982 static struct page *__rmqueue(struct zone *zone, unsigned int order,
988 page = __rmqueue_smallest(zone, order, migratetype);
990 if (unlikely(!page) && migratetype != MIGRATE_RESERVE) {
991 page = __rmqueue_fallback(zone, order, migratetype);
994 * Use MIGRATE_RESERVE rather than fail an allocation. goto
995 * is used because __rmqueue_smallest is an inline function
996 * and we want just one call site
999 migratetype = MIGRATE_RESERVE;
1004 trace_mm_page_alloc_zone_locked(page, order, migratetype);
1009 * Obtain a specified number of elements from the buddy allocator, all under
1010 * a single hold of the lock, for efficiency. Add them to the supplied list.
1011 * Returns the number of new pages which were placed at *list.
1013 static int rmqueue_bulk(struct zone *zone, unsigned int order,
1014 unsigned long count, struct list_head *list,
1015 int migratetype, int cold)
1019 spin_lock(&zone->lock);
1020 for (i = 0; i < count; ++i) {
1021 struct page *page = __rmqueue(zone, order, migratetype);
1022 if (unlikely(page == NULL))
1026 * Split buddy pages returned by expand() are received here
1027 * in physical page order. The page is added to the callers and
1028 * list and the list head then moves forward. From the callers
1029 * perspective, the linked list is ordered by page number in
1030 * some conditions. This is useful for IO devices that can
1031 * merge IO requests if the physical pages are ordered
1034 if (likely(cold == 0))
1035 list_add(&page->lru, list);
1037 list_add_tail(&page->lru, list);
1038 set_page_private(page, migratetype);
1041 __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
1042 spin_unlock(&zone->lock);
1048 * Called from the vmstat counter updater to drain pagesets of this
1049 * currently executing processor on remote nodes after they have
1052 * Note that this function must be called with the thread pinned to
1053 * a single processor.
1055 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
1057 unsigned long flags;
1060 local_irq_save(flags);
1061 if (pcp->count >= pcp->batch)
1062 to_drain = pcp->batch;
1064 to_drain = pcp->count;
1065 free_pcppages_bulk(zone, to_drain, pcp);
1066 pcp->count -= to_drain;
1067 local_irq_restore(flags);
1072 * Drain pages of the indicated processor.
1074 * The processor must either be the current processor and the
1075 * thread pinned to the current processor or a processor that
1078 static void drain_pages(unsigned int cpu)
1080 unsigned long flags;
1083 for_each_populated_zone(zone) {
1084 struct per_cpu_pageset *pset;
1085 struct per_cpu_pages *pcp;
1087 local_irq_save(flags);
1088 pset = per_cpu_ptr(zone->pageset, cpu);
1092 free_pcppages_bulk(zone, pcp->count, pcp);
1095 local_irq_restore(flags);
1100 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
1102 void drain_local_pages(void *arg)
1104 drain_pages(smp_processor_id());
1108 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
1110 void drain_all_pages(void)
1112 on_each_cpu(drain_local_pages, NULL, 1);
1115 #ifdef CONFIG_HIBERNATION
1117 void mark_free_pages(struct zone *zone)
1119 unsigned long pfn, max_zone_pfn;
1120 unsigned long flags;
1122 struct list_head *curr;
1124 if (!zone->spanned_pages)
1127 spin_lock_irqsave(&zone->lock, flags);
1129 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
1130 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1131 if (pfn_valid(pfn)) {
1132 struct page *page = pfn_to_page(pfn);
1134 if (!swsusp_page_is_forbidden(page))
1135 swsusp_unset_page_free(page);
1138 for_each_migratetype_order(order, t) {
1139 list_for_each(curr, &zone->free_area[order].free_list[t]) {
1142 pfn = page_to_pfn(list_entry(curr, struct page, lru));
1143 for (i = 0; i < (1UL << order); i++)
1144 swsusp_set_page_free(pfn_to_page(pfn + i));
1147 spin_unlock_irqrestore(&zone->lock, flags);
1149 #endif /* CONFIG_PM */
1152 * Free a 0-order page
1153 * cold == 1 ? free a cold page : free a hot page
1155 void free_hot_cold_page(struct page *page, int cold)
1157 struct zone *zone = page_zone(page);
1158 struct per_cpu_pages *pcp;
1159 unsigned long flags;
1161 int wasMlocked = __TestClearPageMlocked(page);
1163 if (!free_pages_prepare(page, 0))
1166 migratetype = get_pageblock_migratetype(page);
1167 set_page_private(page, migratetype);
1168 local_irq_save(flags);
1169 if (unlikely(wasMlocked))
1170 free_page_mlock(page);
1171 __count_vm_event(PGFREE);
1174 * We only track unmovable, reclaimable and movable on pcp lists.
1175 * Free ISOLATE pages back to the allocator because they are being
1176 * offlined but treat RESERVE as movable pages so we can get those
1177 * areas back if necessary. Otherwise, we may have to free
1178 * excessively into the page allocator
1180 if (migratetype >= MIGRATE_PCPTYPES) {
1181 if (unlikely(migratetype == MIGRATE_ISOLATE)) {
1182 free_one_page(zone, page, 0, migratetype);
1185 migratetype = MIGRATE_MOVABLE;
1188 pcp = &this_cpu_ptr(zone->pageset)->pcp;
1190 list_add_tail(&page->lru, &pcp->lists[migratetype]);
1192 list_add(&page->lru, &pcp->lists[migratetype]);
1194 if (pcp->count >= pcp->high) {
1195 free_pcppages_bulk(zone, pcp->batch, pcp);
1196 pcp->count -= pcp->batch;
1200 local_irq_restore(flags);
1204 * split_page takes a non-compound higher-order page, and splits it into
1205 * n (1<<order) sub-pages: page[0..n]
1206 * Each sub-page must be freed individually.
1208 * Note: this is probably too low level an operation for use in drivers.
1209 * Please consult with lkml before using this in your driver.
1211 void split_page(struct page *page, unsigned int order)
1215 VM_BUG_ON(PageCompound(page));
1216 VM_BUG_ON(!page_count(page));
1218 #ifdef CONFIG_KMEMCHECK
1220 * Split shadow pages too, because free(page[0]) would
1221 * otherwise free the whole shadow.
1223 if (kmemcheck_page_is_tracked(page))
1224 split_page(virt_to_page(page[0].shadow), order);
1227 for (i = 1; i < (1 << order); i++)
1228 set_page_refcounted(page + i);
1232 * Similar to split_page except the page is already free. As this is only
1233 * being used for migration, the migratetype of the block also changes.
1234 * As this is called with interrupts disabled, the caller is responsible
1235 * for calling arch_alloc_page() and kernel_map_page() after interrupts
1238 * Note: this is probably too low level an operation for use in drivers.
1239 * Please consult with lkml before using this in your driver.
1241 int split_free_page(struct page *page)
1244 unsigned long watermark;
1247 BUG_ON(!PageBuddy(page));
1249 zone = page_zone(page);
1250 order = page_order(page);
1252 /* Obey watermarks as if the page was being allocated */
1253 watermark = low_wmark_pages(zone) + (1 << order);
1254 if (!zone_watermark_ok(zone, 0, watermark, 0, 0))
1257 /* Remove page from free list */
1258 list_del(&page->lru);
1259 zone->free_area[order].nr_free--;
1260 rmv_page_order(page);
1261 __mod_zone_page_state(zone, NR_FREE_PAGES, -(1UL << order));
1263 /* Split into individual pages */
1264 set_page_refcounted(page);
1265 split_page(page, order);
1267 if (order >= pageblock_order - 1) {
1268 struct page *endpage = page + (1 << order) - 1;
1269 for (; page < endpage; page += pageblock_nr_pages)
1270 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
1277 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1278 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1282 struct page *buffered_rmqueue(struct zone *preferred_zone,
1283 struct zone *zone, int order, gfp_t gfp_flags,
1286 unsigned long flags;
1288 int cold = !!(gfp_flags & __GFP_COLD);
1291 if (likely(order == 0)) {
1292 struct per_cpu_pages *pcp;
1293 struct list_head *list;
1295 local_irq_save(flags);
1296 pcp = &this_cpu_ptr(zone->pageset)->pcp;
1297 list = &pcp->lists[migratetype];
1298 if (list_empty(list)) {
1299 pcp->count += rmqueue_bulk(zone, 0,
1302 if (unlikely(list_empty(list)))
1307 page = list_entry(list->prev, struct page, lru);
1309 page = list_entry(list->next, struct page, lru);
1311 list_del(&page->lru);
1314 if (unlikely(gfp_flags & __GFP_NOFAIL)) {
1316 * __GFP_NOFAIL is not to be used in new code.
1318 * All __GFP_NOFAIL callers should be fixed so that they
1319 * properly detect and handle allocation failures.
1321 * We most definitely don't want callers attempting to
1322 * allocate greater than order-1 page units with
1325 WARN_ON_ONCE(order > 1);
1327 spin_lock_irqsave(&zone->lock, flags);
1328 page = __rmqueue(zone, order, migratetype);
1329 spin_unlock(&zone->lock);
1332 __mod_zone_page_state(zone, NR_FREE_PAGES, -(1 << order));
1335 __count_zone_vm_events(PGALLOC, zone, 1 << order);
1336 zone_statistics(preferred_zone, zone);
1337 local_irq_restore(flags);
1339 VM_BUG_ON(bad_range(zone, page));
1340 if (prep_new_page(page, order, gfp_flags))
1345 local_irq_restore(flags);
1349 /* The ALLOC_WMARK bits are used as an index to zone->watermark */
1350 #define ALLOC_WMARK_MIN WMARK_MIN
1351 #define ALLOC_WMARK_LOW WMARK_LOW
1352 #define ALLOC_WMARK_HIGH WMARK_HIGH
1353 #define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */
1355 /* Mask to get the watermark bits */
1356 #define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1)
1358 #define ALLOC_HARDER 0x10 /* try to alloc harder */
1359 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1360 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
1362 #ifdef CONFIG_FAIL_PAGE_ALLOC
1364 static struct fail_page_alloc_attr {
1365 struct fault_attr attr;
1367 u32 ignore_gfp_highmem;
1368 u32 ignore_gfp_wait;
1371 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1373 struct dentry *ignore_gfp_highmem_file;
1374 struct dentry *ignore_gfp_wait_file;
1375 struct dentry *min_order_file;
1377 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1379 } fail_page_alloc = {
1380 .attr = FAULT_ATTR_INITIALIZER,
1381 .ignore_gfp_wait = 1,
1382 .ignore_gfp_highmem = 1,
1386 static int __init setup_fail_page_alloc(char *str)
1388 return setup_fault_attr(&fail_page_alloc.attr, str);
1390 __setup("fail_page_alloc=", setup_fail_page_alloc);
1392 static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1394 if (order < fail_page_alloc.min_order)
1396 if (gfp_mask & __GFP_NOFAIL)
1398 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
1400 if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
1403 return should_fail(&fail_page_alloc.attr, 1 << order);
1406 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1408 static int __init fail_page_alloc_debugfs(void)
1410 mode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
1414 err = init_fault_attr_dentries(&fail_page_alloc.attr,
1418 dir = fail_page_alloc.attr.dentries.dir;
1420 fail_page_alloc.ignore_gfp_wait_file =
1421 debugfs_create_bool("ignore-gfp-wait", mode, dir,
1422 &fail_page_alloc.ignore_gfp_wait);
1424 fail_page_alloc.ignore_gfp_highmem_file =
1425 debugfs_create_bool("ignore-gfp-highmem", mode, dir,
1426 &fail_page_alloc.ignore_gfp_highmem);
1427 fail_page_alloc.min_order_file =
1428 debugfs_create_u32("min-order", mode, dir,
1429 &fail_page_alloc.min_order);
1431 if (!fail_page_alloc.ignore_gfp_wait_file ||
1432 !fail_page_alloc.ignore_gfp_highmem_file ||
1433 !fail_page_alloc.min_order_file) {
1435 debugfs_remove(fail_page_alloc.ignore_gfp_wait_file);
1436 debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file);
1437 debugfs_remove(fail_page_alloc.min_order_file);
1438 cleanup_fault_attr_dentries(&fail_page_alloc.attr);
1444 late_initcall(fail_page_alloc_debugfs);
1446 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1448 #else /* CONFIG_FAIL_PAGE_ALLOC */
1450 static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1455 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1458 * Return true if free pages are above 'mark'. This takes into account the order
1459 * of the allocation.
1461 static bool __zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1462 int classzone_idx, int alloc_flags, long free_pages)
1464 /* free_pages my go negative - that's OK */
1468 free_pages -= (1 << order) + 1;
1469 if (alloc_flags & ALLOC_HIGH)
1471 if (alloc_flags & ALLOC_HARDER)
1474 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
1476 for (o = 0; o < order; o++) {
1477 /* At the next order, this order's pages become unavailable */
1478 free_pages -= z->free_area[o].nr_free << o;
1480 /* Require fewer higher order pages to be free */
1483 if (free_pages <= min)
1489 bool zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1490 int classzone_idx, int alloc_flags)
1492 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
1493 zone_page_state(z, NR_FREE_PAGES));
1496 bool zone_watermark_ok_safe(struct zone *z, int order, unsigned long mark,
1497 int classzone_idx, int alloc_flags)
1499 long free_pages = zone_page_state(z, NR_FREE_PAGES);
1501 if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark)
1502 free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES);
1504 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
1510 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1511 * skip over zones that are not allowed by the cpuset, or that have
1512 * been recently (in last second) found to be nearly full. See further
1513 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1514 * that have to skip over a lot of full or unallowed zones.
1516 * If the zonelist cache is present in the passed in zonelist, then
1517 * returns a pointer to the allowed node mask (either the current
1518 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
1520 * If the zonelist cache is not available for this zonelist, does
1521 * nothing and returns NULL.
1523 * If the fullzones BITMAP in the zonelist cache is stale (more than
1524 * a second since last zap'd) then we zap it out (clear its bits.)
1526 * We hold off even calling zlc_setup, until after we've checked the
1527 * first zone in the zonelist, on the theory that most allocations will
1528 * be satisfied from that first zone, so best to examine that zone as
1529 * quickly as we can.
1531 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1533 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1534 nodemask_t *allowednodes; /* zonelist_cache approximation */
1536 zlc = zonelist->zlcache_ptr;
1540 if (time_after(jiffies, zlc->last_full_zap + HZ)) {
1541 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1542 zlc->last_full_zap = jiffies;
1545 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1546 &cpuset_current_mems_allowed :
1547 &node_states[N_HIGH_MEMORY];
1548 return allowednodes;
1552 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1553 * if it is worth looking at further for free memory:
1554 * 1) Check that the zone isn't thought to be full (doesn't have its
1555 * bit set in the zonelist_cache fullzones BITMAP).
1556 * 2) Check that the zones node (obtained from the zonelist_cache
1557 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1558 * Return true (non-zero) if zone is worth looking at further, or
1559 * else return false (zero) if it is not.
1561 * This check -ignores- the distinction between various watermarks,
1562 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1563 * found to be full for any variation of these watermarks, it will
1564 * be considered full for up to one second by all requests, unless
1565 * we are so low on memory on all allowed nodes that we are forced
1566 * into the second scan of the zonelist.
1568 * In the second scan we ignore this zonelist cache and exactly
1569 * apply the watermarks to all zones, even it is slower to do so.
1570 * We are low on memory in the second scan, and should leave no stone
1571 * unturned looking for a free page.
1573 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
1574 nodemask_t *allowednodes)
1576 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1577 int i; /* index of *z in zonelist zones */
1578 int n; /* node that zone *z is on */
1580 zlc = zonelist->zlcache_ptr;
1584 i = z - zonelist->_zonerefs;
1587 /* This zone is worth trying if it is allowed but not full */
1588 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1592 * Given 'z' scanning a zonelist, set the corresponding bit in
1593 * zlc->fullzones, so that subsequent attempts to allocate a page
1594 * from that zone don't waste time re-examining it.
1596 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
1598 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1599 int i; /* index of *z in zonelist zones */
1601 zlc = zonelist->zlcache_ptr;
1605 i = z - zonelist->_zonerefs;
1607 set_bit(i, zlc->fullzones);
1610 #else /* CONFIG_NUMA */
1612 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1617 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
1618 nodemask_t *allowednodes)
1623 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
1626 #endif /* CONFIG_NUMA */
1629 * get_page_from_freelist goes through the zonelist trying to allocate
1632 static struct page *
1633 get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order,
1634 struct zonelist *zonelist, int high_zoneidx, int alloc_flags,
1635 struct zone *preferred_zone, int migratetype)
1638 struct page *page = NULL;
1641 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1642 int zlc_active = 0; /* set if using zonelist_cache */
1643 int did_zlc_setup = 0; /* just call zlc_setup() one time */
1645 classzone_idx = zone_idx(preferred_zone);
1648 * Scan zonelist, looking for a zone with enough free.
1649 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1651 for_each_zone_zonelist_nodemask(zone, z, zonelist,
1652 high_zoneidx, nodemask) {
1653 if (NUMA_BUILD && zlc_active &&
1654 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1656 if ((alloc_flags & ALLOC_CPUSET) &&
1657 !cpuset_zone_allowed_softwall(zone, gfp_mask))
1660 BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
1661 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
1665 mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
1666 if (zone_watermark_ok(zone, order, mark,
1667 classzone_idx, alloc_flags))
1670 if (zone_reclaim_mode == 0)
1671 goto this_zone_full;
1673 ret = zone_reclaim(zone, gfp_mask, order);
1675 case ZONE_RECLAIM_NOSCAN:
1678 case ZONE_RECLAIM_FULL:
1679 /* scanned but unreclaimable */
1680 goto this_zone_full;
1682 /* did we reclaim enough */
1683 if (!zone_watermark_ok(zone, order, mark,
1684 classzone_idx, alloc_flags))
1685 goto this_zone_full;
1690 page = buffered_rmqueue(preferred_zone, zone, order,
1691 gfp_mask, migratetype);
1696 zlc_mark_zone_full(zonelist, z);
1698 if (NUMA_BUILD && !did_zlc_setup && nr_online_nodes > 1) {
1700 * we do zlc_setup after the first zone is tried but only
1701 * if there are multiple nodes make it worthwhile
1703 allowednodes = zlc_setup(zonelist, alloc_flags);
1709 if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) {
1710 /* Disable zlc cache for second zonelist scan */
1718 * Large machines with many possible nodes should not always dump per-node
1719 * meminfo in irq context.
1721 static inline bool should_suppress_show_mem(void)
1726 ret = in_interrupt();
1732 should_alloc_retry(gfp_t gfp_mask, unsigned int order,
1733 unsigned long pages_reclaimed)
1735 /* Do not loop if specifically requested */
1736 if (gfp_mask & __GFP_NORETRY)
1740 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
1741 * means __GFP_NOFAIL, but that may not be true in other
1744 if (order <= PAGE_ALLOC_COSTLY_ORDER)
1748 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
1749 * specified, then we retry until we no longer reclaim any pages
1750 * (above), or we've reclaimed an order of pages at least as
1751 * large as the allocation's order. In both cases, if the
1752 * allocation still fails, we stop retrying.
1754 if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order))
1758 * Don't let big-order allocations loop unless the caller
1759 * explicitly requests that.
1761 if (gfp_mask & __GFP_NOFAIL)
1767 static inline struct page *
1768 __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
1769 struct zonelist *zonelist, enum zone_type high_zoneidx,
1770 nodemask_t *nodemask, struct zone *preferred_zone,
1775 /* Acquire the OOM killer lock for the zones in zonelist */
1776 if (!try_set_zonelist_oom(zonelist, gfp_mask)) {
1777 schedule_timeout_uninterruptible(1);
1782 * Go through the zonelist yet one more time, keep very high watermark
1783 * here, this is only to catch a parallel oom killing, we must fail if
1784 * we're still under heavy pressure.
1786 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask,
1787 order, zonelist, high_zoneidx,
1788 ALLOC_WMARK_HIGH|ALLOC_CPUSET,
1789 preferred_zone, migratetype);
1793 if (!(gfp_mask & __GFP_NOFAIL)) {
1794 /* The OOM killer will not help higher order allocs */
1795 if (order > PAGE_ALLOC_COSTLY_ORDER)
1797 /* The OOM killer does not needlessly kill tasks for lowmem */
1798 if (high_zoneidx < ZONE_NORMAL)
1801 * GFP_THISNODE contains __GFP_NORETRY and we never hit this.
1802 * Sanity check for bare calls of __GFP_THISNODE, not real OOM.
1803 * The caller should handle page allocation failure by itself if
1804 * it specifies __GFP_THISNODE.
1805 * Note: Hugepage uses it but will hit PAGE_ALLOC_COSTLY_ORDER.
1807 if (gfp_mask & __GFP_THISNODE)
1810 /* Exhausted what can be done so it's blamo time */
1811 out_of_memory(zonelist, gfp_mask, order, nodemask);
1814 clear_zonelist_oom(zonelist, gfp_mask);
1818 #ifdef CONFIG_COMPACTION
1819 /* Try memory compaction for high-order allocations before reclaim */
1820 static struct page *
1821 __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
1822 struct zonelist *zonelist, enum zone_type high_zoneidx,
1823 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
1824 int migratetype, unsigned long *did_some_progress,
1825 bool sync_migration)
1829 if (!order || compaction_deferred(preferred_zone))
1832 current->flags |= PF_MEMALLOC;
1833 *did_some_progress = try_to_compact_pages(zonelist, order, gfp_mask,
1834 nodemask, sync_migration);
1835 current->flags &= ~PF_MEMALLOC;
1836 if (*did_some_progress != COMPACT_SKIPPED) {
1838 /* Page migration frees to the PCP lists but we want merging */
1839 drain_pages(get_cpu());
1842 page = get_page_from_freelist(gfp_mask, nodemask,
1843 order, zonelist, high_zoneidx,
1844 alloc_flags, preferred_zone,
1847 preferred_zone->compact_considered = 0;
1848 preferred_zone->compact_defer_shift = 0;
1849 count_vm_event(COMPACTSUCCESS);
1854 * It's bad if compaction run occurs and fails.
1855 * The most likely reason is that pages exist,
1856 * but not enough to satisfy watermarks.
1858 count_vm_event(COMPACTFAIL);
1859 defer_compaction(preferred_zone);
1867 static inline struct page *
1868 __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
1869 struct zonelist *zonelist, enum zone_type high_zoneidx,
1870 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
1871 int migratetype, unsigned long *did_some_progress,
1872 bool sync_migration)
1876 #endif /* CONFIG_COMPACTION */
1878 /* The really slow allocator path where we enter direct reclaim */
1879 static inline struct page *
1880 __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
1881 struct zonelist *zonelist, enum zone_type high_zoneidx,
1882 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
1883 int migratetype, unsigned long *did_some_progress)
1885 struct page *page = NULL;
1886 struct reclaim_state reclaim_state;
1887 bool drained = false;
1891 /* We now go into synchronous reclaim */
1892 cpuset_memory_pressure_bump();
1893 current->flags |= PF_MEMALLOC;
1894 lockdep_set_current_reclaim_state(gfp_mask);
1895 reclaim_state.reclaimed_slab = 0;
1896 current->reclaim_state = &reclaim_state;
1898 *did_some_progress = try_to_free_pages(zonelist, order, gfp_mask, nodemask);
1900 current->reclaim_state = NULL;
1901 lockdep_clear_current_reclaim_state();
1902 current->flags &= ~PF_MEMALLOC;
1906 if (unlikely(!(*did_some_progress)))
1910 page = get_page_from_freelist(gfp_mask, nodemask, order,
1911 zonelist, high_zoneidx,
1912 alloc_flags, preferred_zone,
1916 * If an allocation failed after direct reclaim, it could be because
1917 * pages are pinned on the per-cpu lists. Drain them and try again
1919 if (!page && !drained) {
1929 * This is called in the allocator slow-path if the allocation request is of
1930 * sufficient urgency to ignore watermarks and take other desperate measures
1932 static inline struct page *
1933 __alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order,
1934 struct zonelist *zonelist, enum zone_type high_zoneidx,
1935 nodemask_t *nodemask, struct zone *preferred_zone,
1941 page = get_page_from_freelist(gfp_mask, nodemask, order,
1942 zonelist, high_zoneidx, ALLOC_NO_WATERMARKS,
1943 preferred_zone, migratetype);
1945 if (!page && gfp_mask & __GFP_NOFAIL)
1946 wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50);
1947 } while (!page && (gfp_mask & __GFP_NOFAIL));
1953 void wake_all_kswapd(unsigned int order, struct zonelist *zonelist,
1954 enum zone_type high_zoneidx,
1955 enum zone_type classzone_idx)
1960 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx)
1961 wakeup_kswapd(zone, order, classzone_idx);
1965 gfp_to_alloc_flags(gfp_t gfp_mask)
1967 int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
1968 const gfp_t wait = gfp_mask & __GFP_WAIT;
1970 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
1971 BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
1974 * The caller may dip into page reserves a bit more if the caller
1975 * cannot run direct reclaim, or if the caller has realtime scheduling
1976 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1977 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1979 alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH);
1983 * Not worth trying to allocate harder for
1984 * __GFP_NOMEMALLOC even if it can't schedule.
1986 if (!(gfp_mask & __GFP_NOMEMALLOC))
1987 alloc_flags |= ALLOC_HARDER;
1989 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1990 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1992 alloc_flags &= ~ALLOC_CPUSET;
1993 } else if (unlikely(rt_task(current)) && !in_interrupt())
1994 alloc_flags |= ALLOC_HARDER;
1996 if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) {
1997 if (!in_interrupt() &&
1998 ((current->flags & PF_MEMALLOC) ||
1999 unlikely(test_thread_flag(TIF_MEMDIE))))
2000 alloc_flags |= ALLOC_NO_WATERMARKS;
2006 static inline struct page *
2007 __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
2008 struct zonelist *zonelist, enum zone_type high_zoneidx,
2009 nodemask_t *nodemask, struct zone *preferred_zone,
2012 const gfp_t wait = gfp_mask & __GFP_WAIT;
2013 struct page *page = NULL;
2015 unsigned long pages_reclaimed = 0;
2016 unsigned long did_some_progress;
2017 bool sync_migration = false;
2020 * In the slowpath, we sanity check order to avoid ever trying to
2021 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
2022 * be using allocators in order of preference for an area that is
2025 if (order >= MAX_ORDER) {
2026 WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
2031 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
2032 * __GFP_NOWARN set) should not cause reclaim since the subsystem
2033 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
2034 * using a larger set of nodes after it has established that the
2035 * allowed per node queues are empty and that nodes are
2038 if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
2042 if (!(gfp_mask & __GFP_NO_KSWAPD))
2043 wake_all_kswapd(order, zonelist, high_zoneidx,
2044 zone_idx(preferred_zone));
2047 * OK, we're below the kswapd watermark and have kicked background
2048 * reclaim. Now things get more complex, so set up alloc_flags according
2049 * to how we want to proceed.
2051 alloc_flags = gfp_to_alloc_flags(gfp_mask);
2054 * Find the true preferred zone if the allocation is unconstrained by
2057 if (!(alloc_flags & ALLOC_CPUSET) && !nodemask)
2058 first_zones_zonelist(zonelist, high_zoneidx, NULL,
2061 /* This is the last chance, in general, before the goto nopage. */
2062 page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist,
2063 high_zoneidx, alloc_flags & ~ALLOC_NO_WATERMARKS,
2064 preferred_zone, migratetype);
2069 /* Allocate without watermarks if the context allows */
2070 if (alloc_flags & ALLOC_NO_WATERMARKS) {
2071 page = __alloc_pages_high_priority(gfp_mask, order,
2072 zonelist, high_zoneidx, nodemask,
2073 preferred_zone, migratetype);
2078 /* Atomic allocations - we can't balance anything */
2082 /* Avoid recursion of direct reclaim */
2083 if (current->flags & PF_MEMALLOC)
2086 /* Avoid allocations with no watermarks from looping endlessly */
2087 if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL))
2091 * Try direct compaction. The first pass is asynchronous. Subsequent
2092 * attempts after direct reclaim are synchronous
2094 page = __alloc_pages_direct_compact(gfp_mask, order,
2095 zonelist, high_zoneidx,
2097 alloc_flags, preferred_zone,
2098 migratetype, &did_some_progress,
2102 sync_migration = true;
2104 /* Try direct reclaim and then allocating */
2105 page = __alloc_pages_direct_reclaim(gfp_mask, order,
2106 zonelist, high_zoneidx,
2108 alloc_flags, preferred_zone,
2109 migratetype, &did_some_progress);
2114 * If we failed to make any progress reclaiming, then we are
2115 * running out of options and have to consider going OOM
2117 if (!did_some_progress) {
2118 if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
2119 if (oom_killer_disabled)
2121 page = __alloc_pages_may_oom(gfp_mask, order,
2122 zonelist, high_zoneidx,
2123 nodemask, preferred_zone,
2128 if (!(gfp_mask & __GFP_NOFAIL)) {
2130 * The oom killer is not called for high-order
2131 * allocations that may fail, so if no progress
2132 * is being made, there are no other options and
2133 * retrying is unlikely to help.
2135 if (order > PAGE_ALLOC_COSTLY_ORDER)
2138 * The oom killer is not called for lowmem
2139 * allocations to prevent needlessly killing
2142 if (high_zoneidx < ZONE_NORMAL)
2150 /* Check if we should retry the allocation */
2151 pages_reclaimed += did_some_progress;
2152 if (should_alloc_retry(gfp_mask, order, pages_reclaimed)) {
2153 /* Wait for some write requests to complete then retry */
2154 wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50);
2158 * High-order allocations do not necessarily loop after
2159 * direct reclaim and reclaim/compaction depends on compaction
2160 * being called after reclaim so call directly if necessary
2162 page = __alloc_pages_direct_compact(gfp_mask, order,
2163 zonelist, high_zoneidx,
2165 alloc_flags, preferred_zone,
2166 migratetype, &did_some_progress,
2173 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
2174 unsigned int filter = SHOW_MEM_FILTER_NODES;
2177 * This documents exceptions given to allocations in certain
2178 * contexts that are allowed to allocate outside current's set
2181 if (!(gfp_mask & __GFP_NOMEMALLOC))
2182 if (test_thread_flag(TIF_MEMDIE) ||
2183 (current->flags & (PF_MEMALLOC | PF_EXITING)))
2184 filter &= ~SHOW_MEM_FILTER_NODES;
2185 if (in_interrupt() || !wait)
2186 filter &= ~SHOW_MEM_FILTER_NODES;
2188 pr_warning("%s: page allocation failure. order:%d, mode:0x%x\n",
2189 current->comm, order, gfp_mask);
2191 if (!should_suppress_show_mem())
2196 if (kmemcheck_enabled)
2197 kmemcheck_pagealloc_alloc(page, order, gfp_mask);
2203 * This is the 'heart' of the zoned buddy allocator.
2206 __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
2207 struct zonelist *zonelist, nodemask_t *nodemask)
2209 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
2210 struct zone *preferred_zone;
2212 int migratetype = allocflags_to_migratetype(gfp_mask);
2214 gfp_mask &= gfp_allowed_mask;
2216 lockdep_trace_alloc(gfp_mask);
2218 might_sleep_if(gfp_mask & __GFP_WAIT);
2220 if (should_fail_alloc_page(gfp_mask, order))
2224 * Check the zones suitable for the gfp_mask contain at least one
2225 * valid zone. It's possible to have an empty zonelist as a result
2226 * of GFP_THISNODE and a memoryless node
2228 if (unlikely(!zonelist->_zonerefs->zone))
2232 /* The preferred zone is used for statistics later */
2233 first_zones_zonelist(zonelist, high_zoneidx,
2234 nodemask ? : &cpuset_current_mems_allowed,
2236 if (!preferred_zone) {
2241 /* First allocation attempt */
2242 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order,
2243 zonelist, high_zoneidx, ALLOC_WMARK_LOW|ALLOC_CPUSET,
2244 preferred_zone, migratetype);
2245 if (unlikely(!page))
2246 page = __alloc_pages_slowpath(gfp_mask, order,
2247 zonelist, high_zoneidx, nodemask,
2248 preferred_zone, migratetype);
2251 trace_mm_page_alloc(page, order, gfp_mask, migratetype);
2254 EXPORT_SYMBOL(__alloc_pages_nodemask);
2257 * Common helper functions.
2259 unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
2264 * __get_free_pages() returns a 32-bit address, which cannot represent
2267 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
2269 page = alloc_pages(gfp_mask, order);
2272 return (unsigned long) page_address(page);
2274 EXPORT_SYMBOL(__get_free_pages);
2276 unsigned long get_zeroed_page(gfp_t gfp_mask)
2278 return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
2280 EXPORT_SYMBOL(get_zeroed_page);
2282 void __pagevec_free(struct pagevec *pvec)
2284 int i = pagevec_count(pvec);
2287 trace_mm_pagevec_free(pvec->pages[i], pvec->cold);
2288 free_hot_cold_page(pvec->pages[i], pvec->cold);
2292 void __free_pages(struct page *page, unsigned int order)
2294 if (put_page_testzero(page)) {
2296 free_hot_cold_page(page, 0);
2298 __free_pages_ok(page, order);
2302 EXPORT_SYMBOL(__free_pages);
2304 void free_pages(unsigned long addr, unsigned int order)
2307 VM_BUG_ON(!virt_addr_valid((void *)addr));
2308 __free_pages(virt_to_page((void *)addr), order);
2312 EXPORT_SYMBOL(free_pages);
2315 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
2316 * @size: the number of bytes to allocate
2317 * @gfp_mask: GFP flags for the allocation
2319 * This function is similar to alloc_pages(), except that it allocates the
2320 * minimum number of pages to satisfy the request. alloc_pages() can only
2321 * allocate memory in power-of-two pages.
2323 * This function is also limited by MAX_ORDER.
2325 * Memory allocated by this function must be released by free_pages_exact().
2327 void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
2329 unsigned int order = get_order(size);
2332 addr = __get_free_pages(gfp_mask, order);
2334 unsigned long alloc_end = addr + (PAGE_SIZE << order);
2335 unsigned long used = addr + PAGE_ALIGN(size);
2337 split_page(virt_to_page((void *)addr), order);
2338 while (used < alloc_end) {
2344 return (void *)addr;
2346 EXPORT_SYMBOL(alloc_pages_exact);
2349 * free_pages_exact - release memory allocated via alloc_pages_exact()
2350 * @virt: the value returned by alloc_pages_exact.
2351 * @size: size of allocation, same value as passed to alloc_pages_exact().
2353 * Release the memory allocated by a previous call to alloc_pages_exact.
2355 void free_pages_exact(void *virt, size_t size)
2357 unsigned long addr = (unsigned long)virt;
2358 unsigned long end = addr + PAGE_ALIGN(size);
2360 while (addr < end) {
2365 EXPORT_SYMBOL(free_pages_exact);
2367 static unsigned int nr_free_zone_pages(int offset)
2372 /* Just pick one node, since fallback list is circular */
2373 unsigned int sum = 0;
2375 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
2377 for_each_zone_zonelist(zone, z, zonelist, offset) {
2378 unsigned long size = zone->present_pages;
2379 unsigned long high = high_wmark_pages(zone);
2388 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
2390 unsigned int nr_free_buffer_pages(void)
2392 return nr_free_zone_pages(gfp_zone(GFP_USER));
2394 EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
2397 * Amount of free RAM allocatable within all zones
2399 unsigned int nr_free_pagecache_pages(void)
2401 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
2404 static inline void show_node(struct zone *zone)
2407 printk("Node %d ", zone_to_nid(zone));
2410 void si_meminfo(struct sysinfo *val)
2412 val->totalram = totalram_pages;
2414 val->freeram = global_page_state(NR_FREE_PAGES);
2415 val->bufferram = nr_blockdev_pages();
2416 val->totalhigh = totalhigh_pages;
2417 val->freehigh = nr_free_highpages();
2418 val->mem_unit = PAGE_SIZE;
2421 EXPORT_SYMBOL(si_meminfo);
2424 void si_meminfo_node(struct sysinfo *val, int nid)
2426 pg_data_t *pgdat = NODE_DATA(nid);
2428 val->totalram = pgdat->node_present_pages;
2429 val->freeram = node_page_state(nid, NR_FREE_PAGES);
2430 #ifdef CONFIG_HIGHMEM
2431 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
2432 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
2438 val->mem_unit = PAGE_SIZE;
2443 * Determine whether the zone's node should be displayed or not, depending on
2444 * whether SHOW_MEM_FILTER_NODES was passed to __show_free_areas().
2446 static bool skip_free_areas_zone(unsigned int flags, const struct zone *zone)
2450 if (!(flags & SHOW_MEM_FILTER_NODES))
2454 ret = !node_isset(zone->zone_pgdat->node_id,
2455 cpuset_current_mems_allowed);
2461 #define K(x) ((x) << (PAGE_SHIFT-10))
2464 * Show free area list (used inside shift_scroll-lock stuff)
2465 * We also calculate the percentage fragmentation. We do this by counting the
2466 * memory on each free list with the exception of the first item on the list.
2467 * Suppresses nodes that are not allowed by current's cpuset if
2468 * SHOW_MEM_FILTER_NODES is passed.
2470 void __show_free_areas(unsigned int filter)
2475 for_each_populated_zone(zone) {
2476 if (skip_free_areas_zone(filter, zone))
2479 printk("%s per-cpu:\n", zone->name);
2481 for_each_online_cpu(cpu) {
2482 struct per_cpu_pageset *pageset;
2484 pageset = per_cpu_ptr(zone->pageset, cpu);
2486 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
2487 cpu, pageset->pcp.high,
2488 pageset->pcp.batch, pageset->pcp.count);
2492 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
2493 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
2495 " dirty:%lu writeback:%lu unstable:%lu\n"
2496 " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n"
2497 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n",
2498 global_page_state(NR_ACTIVE_ANON),
2499 global_page_state(NR_INACTIVE_ANON),
2500 global_page_state(NR_ISOLATED_ANON),
2501 global_page_state(NR_ACTIVE_FILE),
2502 global_page_state(NR_INACTIVE_FILE),
2503 global_page_state(NR_ISOLATED_FILE),
2504 global_page_state(NR_UNEVICTABLE),
2505 global_page_state(NR_FILE_DIRTY),
2506 global_page_state(NR_WRITEBACK),
2507 global_page_state(NR_UNSTABLE_NFS),
2508 global_page_state(NR_FREE_PAGES),
2509 global_page_state(NR_SLAB_RECLAIMABLE),
2510 global_page_state(NR_SLAB_UNRECLAIMABLE),
2511 global_page_state(NR_FILE_MAPPED),
2512 global_page_state(NR_SHMEM),
2513 global_page_state(NR_PAGETABLE),
2514 global_page_state(NR_BOUNCE));
2516 for_each_populated_zone(zone) {
2519 if (skip_free_areas_zone(filter, zone))
2527 " active_anon:%lukB"
2528 " inactive_anon:%lukB"
2529 " active_file:%lukB"
2530 " inactive_file:%lukB"
2531 " unevictable:%lukB"
2532 " isolated(anon):%lukB"
2533 " isolated(file):%lukB"
2540 " slab_reclaimable:%lukB"
2541 " slab_unreclaimable:%lukB"
2542 " kernel_stack:%lukB"
2546 " writeback_tmp:%lukB"
2547 " pages_scanned:%lu"
2548 " all_unreclaimable? %s"
2551 K(zone_page_state(zone, NR_FREE_PAGES)),
2552 K(min_wmark_pages(zone)),
2553 K(low_wmark_pages(zone)),
2554 K(high_wmark_pages(zone)),
2555 K(zone_page_state(zone, NR_ACTIVE_ANON)),
2556 K(zone_page_state(zone, NR_INACTIVE_ANON)),
2557 K(zone_page_state(zone, NR_ACTIVE_FILE)),
2558 K(zone_page_state(zone, NR_INACTIVE_FILE)),
2559 K(zone_page_state(zone, NR_UNEVICTABLE)),
2560 K(zone_page_state(zone, NR_ISOLATED_ANON)),
2561 K(zone_page_state(zone, NR_ISOLATED_FILE)),
2562 K(zone->present_pages),
2563 K(zone_page_state(zone, NR_MLOCK)),
2564 K(zone_page_state(zone, NR_FILE_DIRTY)),
2565 K(zone_page_state(zone, NR_WRITEBACK)),
2566 K(zone_page_state(zone, NR_FILE_MAPPED)),
2567 K(zone_page_state(zone, NR_SHMEM)),
2568 K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)),
2569 K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)),
2570 zone_page_state(zone, NR_KERNEL_STACK) *
2572 K(zone_page_state(zone, NR_PAGETABLE)),
2573 K(zone_page_state(zone, NR_UNSTABLE_NFS)),
2574 K(zone_page_state(zone, NR_BOUNCE)),
2575 K(zone_page_state(zone, NR_WRITEBACK_TEMP)),
2576 zone->pages_scanned,
2577 (zone->all_unreclaimable ? "yes" : "no")
2579 printk("lowmem_reserve[]:");
2580 for (i = 0; i < MAX_NR_ZONES; i++)
2581 printk(" %lu", zone->lowmem_reserve[i]);
2585 for_each_populated_zone(zone) {
2586 unsigned long nr[MAX_ORDER], flags, order, total = 0;
2588 if (skip_free_areas_zone(filter, zone))
2591 printk("%s: ", zone->name);
2593 spin_lock_irqsave(&zone->lock, flags);
2594 for (order = 0; order < MAX_ORDER; order++) {
2595 nr[order] = zone->free_area[order].nr_free;
2596 total += nr[order] << order;
2598 spin_unlock_irqrestore(&zone->lock, flags);
2599 for (order = 0; order < MAX_ORDER; order++)
2600 printk("%lu*%lukB ", nr[order], K(1UL) << order);
2601 printk("= %lukB\n", K(total));
2604 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
2606 show_swap_cache_info();
2609 void show_free_areas(void)
2611 __show_free_areas(0);
2614 static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
2616 zoneref->zone = zone;
2617 zoneref->zone_idx = zone_idx(zone);
2621 * Builds allocation fallback zone lists.
2623 * Add all populated zones of a node to the zonelist.
2625 static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
2626 int nr_zones, enum zone_type zone_type)
2630 BUG_ON(zone_type >= MAX_NR_ZONES);
2635 zone = pgdat->node_zones + zone_type;
2636 if (populated_zone(zone)) {
2637 zoneref_set_zone(zone,
2638 &zonelist->_zonerefs[nr_zones++]);
2639 check_highest_zone(zone_type);
2642 } while (zone_type);
2649 * 0 = automatic detection of better ordering.
2650 * 1 = order by ([node] distance, -zonetype)
2651 * 2 = order by (-zonetype, [node] distance)
2653 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
2654 * the same zonelist. So only NUMA can configure this param.
2656 #define ZONELIST_ORDER_DEFAULT 0
2657 #define ZONELIST_ORDER_NODE 1
2658 #define ZONELIST_ORDER_ZONE 2
2660 /* zonelist order in the kernel.
2661 * set_zonelist_order() will set this to NODE or ZONE.
2663 static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
2664 static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
2668 /* The value user specified ....changed by config */
2669 static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2670 /* string for sysctl */
2671 #define NUMA_ZONELIST_ORDER_LEN 16
2672 char numa_zonelist_order[16] = "default";
2675 * interface for configure zonelist ordering.
2676 * command line option "numa_zonelist_order"
2677 * = "[dD]efault - default, automatic configuration.
2678 * = "[nN]ode - order by node locality, then by zone within node
2679 * = "[zZ]one - order by zone, then by locality within zone
2682 static int __parse_numa_zonelist_order(char *s)
2684 if (*s == 'd' || *s == 'D') {
2685 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2686 } else if (*s == 'n' || *s == 'N') {
2687 user_zonelist_order = ZONELIST_ORDER_NODE;
2688 } else if (*s == 'z' || *s == 'Z') {
2689 user_zonelist_order = ZONELIST_ORDER_ZONE;
2692 "Ignoring invalid numa_zonelist_order value: "
2699 static __init int setup_numa_zonelist_order(char *s)
2706 ret = __parse_numa_zonelist_order(s);
2708 strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN);
2712 early_param("numa_zonelist_order", setup_numa_zonelist_order);
2715 * sysctl handler for numa_zonelist_order
2717 int numa_zonelist_order_handler(ctl_table *table, int write,
2718 void __user *buffer, size_t *length,
2721 char saved_string[NUMA_ZONELIST_ORDER_LEN];
2723 static DEFINE_MUTEX(zl_order_mutex);
2725 mutex_lock(&zl_order_mutex);
2727 strcpy(saved_string, (char*)table->data);
2728 ret = proc_dostring(table, write, buffer, length, ppos);
2732 int oldval = user_zonelist_order;
2733 if (__parse_numa_zonelist_order((char*)table->data)) {
2735 * bogus value. restore saved string
2737 strncpy((char*)table->data, saved_string,
2738 NUMA_ZONELIST_ORDER_LEN);
2739 user_zonelist_order = oldval;
2740 } else if (oldval != user_zonelist_order) {
2741 mutex_lock(&zonelists_mutex);
2742 build_all_zonelists(NULL);
2743 mutex_unlock(&zonelists_mutex);
2747 mutex_unlock(&zl_order_mutex);
2752 #define MAX_NODE_LOAD (nr_online_nodes)
2753 static int node_load[MAX_NUMNODES];
2756 * find_next_best_node - find the next node that should appear in a given node's fallback list
2757 * @node: node whose fallback list we're appending
2758 * @used_node_mask: nodemask_t of already used nodes
2760 * We use a number of factors to determine which is the next node that should
2761 * appear on a given node's fallback list. The node should not have appeared
2762 * already in @node's fallback list, and it should be the next closest node
2763 * according to the distance array (which contains arbitrary distance values
2764 * from each node to each node in the system), and should also prefer nodes
2765 * with no CPUs, since presumably they'll have very little allocation pressure
2766 * on them otherwise.
2767 * It returns -1 if no node is found.
2769 static int find_next_best_node(int node, nodemask_t *used_node_mask)
2772 int min_val = INT_MAX;
2774 const struct cpumask *tmp = cpumask_of_node(0);
2776 /* Use the local node if we haven't already */
2777 if (!node_isset(node, *used_node_mask)) {
2778 node_set(node, *used_node_mask);
2782 for_each_node_state(n, N_HIGH_MEMORY) {
2784 /* Don't want a node to appear more than once */
2785 if (node_isset(n, *used_node_mask))
2788 /* Use the distance array to find the distance */
2789 val = node_distance(node, n);
2791 /* Penalize nodes under us ("prefer the next node") */
2794 /* Give preference to headless and unused nodes */
2795 tmp = cpumask_of_node(n);
2796 if (!cpumask_empty(tmp))
2797 val += PENALTY_FOR_NODE_WITH_CPUS;
2799 /* Slight preference for less loaded node */
2800 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
2801 val += node_load[n];
2803 if (val < min_val) {
2810 node_set(best_node, *used_node_mask);
2817 * Build zonelists ordered by node and zones within node.
2818 * This results in maximum locality--normal zone overflows into local
2819 * DMA zone, if any--but risks exhausting DMA zone.
2821 static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
2824 struct zonelist *zonelist;
2826 zonelist = &pgdat->node_zonelists[0];
2827 for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
2829 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2831 zonelist->_zonerefs[j].zone = NULL;
2832 zonelist->_zonerefs[j].zone_idx = 0;
2836 * Build gfp_thisnode zonelists
2838 static void build_thisnode_zonelists(pg_data_t *pgdat)
2841 struct zonelist *zonelist;
2843 zonelist = &pgdat->node_zonelists[1];
2844 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
2845 zonelist->_zonerefs[j].zone = NULL;
2846 zonelist->_zonerefs[j].zone_idx = 0;
2850 * Build zonelists ordered by zone and nodes within zones.
2851 * This results in conserving DMA zone[s] until all Normal memory is
2852 * exhausted, but results in overflowing to remote node while memory
2853 * may still exist in local DMA zone.
2855 static int node_order[MAX_NUMNODES];
2857 static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
2860 int zone_type; /* needs to be signed */
2862 struct zonelist *zonelist;
2864 zonelist = &pgdat->node_zonelists[0];
2866 for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
2867 for (j = 0; j < nr_nodes; j++) {
2868 node = node_order[j];
2869 z = &NODE_DATA(node)->node_zones[zone_type];
2870 if (populated_zone(z)) {
2872 &zonelist->_zonerefs[pos++]);
2873 check_highest_zone(zone_type);
2877 zonelist->_zonerefs[pos].zone = NULL;
2878 zonelist->_zonerefs[pos].zone_idx = 0;
2881 static int default_zonelist_order(void)
2884 unsigned long low_kmem_size,total_size;
2888 * ZONE_DMA and ZONE_DMA32 can be very small area in the system.
2889 * If they are really small and used heavily, the system can fall
2890 * into OOM very easily.
2891 * This function detect ZONE_DMA/DMA32 size and configures zone order.
2893 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2896 for_each_online_node(nid) {
2897 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2898 z = &NODE_DATA(nid)->node_zones[zone_type];
2899 if (populated_zone(z)) {
2900 if (zone_type < ZONE_NORMAL)
2901 low_kmem_size += z->present_pages;
2902 total_size += z->present_pages;
2903 } else if (zone_type == ZONE_NORMAL) {
2905 * If any node has only lowmem, then node order
2906 * is preferred to allow kernel allocations
2907 * locally; otherwise, they can easily infringe
2908 * on other nodes when there is an abundance of
2909 * lowmem available to allocate from.
2911 return ZONELIST_ORDER_NODE;
2915 if (!low_kmem_size || /* there are no DMA area. */
2916 low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
2917 return ZONELIST_ORDER_NODE;
2919 * look into each node's config.
2920 * If there is a node whose DMA/DMA32 memory is very big area on
2921 * local memory, NODE_ORDER may be suitable.
2923 average_size = total_size /
2924 (nodes_weight(node_states[N_HIGH_MEMORY]) + 1);
2925 for_each_online_node(nid) {
2928 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2929 z = &NODE_DATA(nid)->node_zones[zone_type];
2930 if (populated_zone(z)) {
2931 if (zone_type < ZONE_NORMAL)
2932 low_kmem_size += z->present_pages;
2933 total_size += z->present_pages;
2936 if (low_kmem_size &&
2937 total_size > average_size && /* ignore small node */
2938 low_kmem_size > total_size * 70/100)
2939 return ZONELIST_ORDER_NODE;
2941 return ZONELIST_ORDER_ZONE;
2944 static void set_zonelist_order(void)
2946 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
2947 current_zonelist_order = default_zonelist_order();
2949 current_zonelist_order = user_zonelist_order;
2952 static void build_zonelists(pg_data_t *pgdat)
2956 nodemask_t used_mask;
2957 int local_node, prev_node;
2958 struct zonelist *zonelist;
2959 int order = current_zonelist_order;
2961 /* initialize zonelists */
2962 for (i = 0; i < MAX_ZONELISTS; i++) {
2963 zonelist = pgdat->node_zonelists + i;
2964 zonelist->_zonerefs[0].zone = NULL;
2965 zonelist->_zonerefs[0].zone_idx = 0;
2968 /* NUMA-aware ordering of nodes */
2969 local_node = pgdat->node_id;
2970 load = nr_online_nodes;
2971 prev_node = local_node;
2972 nodes_clear(used_mask);
2974 memset(node_order, 0, sizeof(node_order));
2977 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
2978 int distance = node_distance(local_node, node);
2981 * If another node is sufficiently far away then it is better
2982 * to reclaim pages in a zone before going off node.
2984 if (distance > RECLAIM_DISTANCE)
2985 zone_reclaim_mode = 1;
2988 * We don't want to pressure a particular node.
2989 * So adding penalty to the first node in same
2990 * distance group to make it round-robin.
2992 if (distance != node_distance(local_node, prev_node))
2993 node_load[node] = load;
2997 if (order == ZONELIST_ORDER_NODE)
2998 build_zonelists_in_node_order(pgdat, node);
3000 node_order[j++] = node; /* remember order */
3003 if (order == ZONELIST_ORDER_ZONE) {
3004 /* calculate node order -- i.e., DMA last! */
3005 build_zonelists_in_zone_order(pgdat, j);
3008 build_thisnode_zonelists(pgdat);
3011 /* Construct the zonelist performance cache - see further mmzone.h */
3012 static void build_zonelist_cache(pg_data_t *pgdat)
3014 struct zonelist *zonelist;
3015 struct zonelist_cache *zlc;
3018 zonelist = &pgdat->node_zonelists[0];
3019 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
3020 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
3021 for (z = zonelist->_zonerefs; z->zone; z++)
3022 zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z);
3025 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
3027 * Return node id of node used for "local" allocations.
3028 * I.e., first node id of first zone in arg node's generic zonelist.
3029 * Used for initializing percpu 'numa_mem', which is used primarily
3030 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
3032 int local_memory_node(int node)
3036 (void)first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
3037 gfp_zone(GFP_KERNEL),
3044 #else /* CONFIG_NUMA */
3046 static void set_zonelist_order(void)
3048 current_zonelist_order = ZONELIST_ORDER_ZONE;
3051 static void build_zonelists(pg_data_t *pgdat)
3053 int node, local_node;
3055 struct zonelist *zonelist;
3057 local_node = pgdat->node_id;
3059 zonelist = &pgdat->node_zonelists[0];
3060 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
3063 * Now we build the zonelist so that it contains the zones
3064 * of all the other nodes.
3065 * We don't want to pressure a particular node, so when
3066 * building the zones for node N, we make sure that the
3067 * zones coming right after the local ones are those from
3068 * node N+1 (modulo N)
3070 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
3071 if (!node_online(node))
3073 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
3076 for (node = 0; node < local_node; node++) {
3077 if (!node_online(node))
3079 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
3083 zonelist->_zonerefs[j].zone = NULL;
3084 zonelist->_zonerefs[j].zone_idx = 0;
3087 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
3088 static void build_zonelist_cache(pg_data_t *pgdat)
3090 pgdat->node_zonelists[0].zlcache_ptr = NULL;
3093 #endif /* CONFIG_NUMA */
3096 * Boot pageset table. One per cpu which is going to be used for all
3097 * zones and all nodes. The parameters will be set in such a way
3098 * that an item put on a list will immediately be handed over to
3099 * the buddy list. This is safe since pageset manipulation is done
3100 * with interrupts disabled.
3102 * The boot_pagesets must be kept even after bootup is complete for
3103 * unused processors and/or zones. They do play a role for bootstrapping
3104 * hotplugged processors.
3106 * zoneinfo_show() and maybe other functions do
3107 * not check if the processor is online before following the pageset pointer.
3108 * Other parts of the kernel may not check if the zone is available.
3110 static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch);
3111 static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset);
3112 static void setup_zone_pageset(struct zone *zone);
3115 * Global mutex to protect against size modification of zonelists
3116 * as well as to serialize pageset setup for the new populated zone.
3118 DEFINE_MUTEX(zonelists_mutex);
3120 /* return values int ....just for stop_machine() */
3121 static __init_refok int __build_all_zonelists(void *data)
3127 memset(node_load, 0, sizeof(node_load));
3129 for_each_online_node(nid) {
3130 pg_data_t *pgdat = NODE_DATA(nid);
3132 build_zonelists(pgdat);
3133 build_zonelist_cache(pgdat);
3137 * Initialize the boot_pagesets that are going to be used
3138 * for bootstrapping processors. The real pagesets for
3139 * each zone will be allocated later when the per cpu
3140 * allocator is available.
3142 * boot_pagesets are used also for bootstrapping offline
3143 * cpus if the system is already booted because the pagesets
3144 * are needed to initialize allocators on a specific cpu too.
3145 * F.e. the percpu allocator needs the page allocator which
3146 * needs the percpu allocator in order to allocate its pagesets
3147 * (a chicken-egg dilemma).
3149 for_each_possible_cpu(cpu) {
3150 setup_pageset(&per_cpu(boot_pageset, cpu), 0);
3152 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
3154 * We now know the "local memory node" for each node--
3155 * i.e., the node of the first zone in the generic zonelist.
3156 * Set up numa_mem percpu variable for on-line cpus. During
3157 * boot, only the boot cpu should be on-line; we'll init the
3158 * secondary cpus' numa_mem as they come on-line. During
3159 * node/memory hotplug, we'll fixup all on-line cpus.
3161 if (cpu_online(cpu))
3162 set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
3170 * Called with zonelists_mutex held always
3171 * unless system_state == SYSTEM_BOOTING.
3173 void build_all_zonelists(void *data)
3175 set_zonelist_order();
3177 if (system_state == SYSTEM_BOOTING) {
3178 __build_all_zonelists(NULL);
3179 mminit_verify_zonelist();
3180 cpuset_init_current_mems_allowed();
3182 /* we have to stop all cpus to guarantee there is no user
3184 #ifdef CONFIG_MEMORY_HOTPLUG
3186 setup_zone_pageset((struct zone *)data);
3188 stop_machine(__build_all_zonelists, NULL, NULL);
3189 /* cpuset refresh routine should be here */
3191 vm_total_pages = nr_free_pagecache_pages();
3193 * Disable grouping by mobility if the number of pages in the
3194 * system is too low to allow the mechanism to work. It would be
3195 * more accurate, but expensive to check per-zone. This check is
3196 * made on memory-hotadd so a system can start with mobility
3197 * disabled and enable it later
3199 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
3200 page_group_by_mobility_disabled = 1;
3202 page_group_by_mobility_disabled = 0;
3204 printk("Built %i zonelists in %s order, mobility grouping %s. "
3205 "Total pages: %ld\n",
3207 zonelist_order_name[current_zonelist_order],
3208 page_group_by_mobility_disabled ? "off" : "on",
3211 printk("Policy zone: %s\n", zone_names[policy_zone]);
3216 * Helper functions to size the waitqueue hash table.
3217 * Essentially these want to choose hash table sizes sufficiently
3218 * large so that collisions trying to wait on pages are rare.
3219 * But in fact, the number of active page waitqueues on typical
3220 * systems is ridiculously low, less than 200. So this is even
3221 * conservative, even though it seems large.
3223 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
3224 * waitqueues, i.e. the size of the waitq table given the number of pages.
3226 #define PAGES_PER_WAITQUEUE 256
3228 #ifndef CONFIG_MEMORY_HOTPLUG
3229 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
3231 unsigned long size = 1;
3233 pages /= PAGES_PER_WAITQUEUE;
3235 while (size < pages)
3239 * Once we have dozens or even hundreds of threads sleeping
3240 * on IO we've got bigger problems than wait queue collision.
3241 * Limit the size of the wait table to a reasonable size.
3243 size = min(size, 4096UL);
3245 return max(size, 4UL);
3249 * A zone's size might be changed by hot-add, so it is not possible to determine
3250 * a suitable size for its wait_table. So we use the maximum size now.
3252 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
3254 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
3255 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
3256 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
3258 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
3259 * or more by the traditional way. (See above). It equals:
3261 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
3262 * ia64(16K page size) : = ( 8G + 4M)byte.
3263 * powerpc (64K page size) : = (32G +16M)byte.
3265 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
3272 * This is an integer logarithm so that shifts can be used later
3273 * to extract the more random high bits from the multiplicative
3274 * hash function before the remainder is taken.
3276 static inline unsigned long wait_table_bits(unsigned long size)
3281 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
3284 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
3285 * of blocks reserved is based on min_wmark_pages(zone). The memory within
3286 * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
3287 * higher will lead to a bigger reserve which will get freed as contiguous
3288 * blocks as reclaim kicks in
3290 static void setup_zone_migrate_reserve(struct zone *zone)
3292 unsigned long start_pfn, pfn, end_pfn;
3294 unsigned long block_migratetype;
3297 /* Get the start pfn, end pfn and the number of blocks to reserve */
3298 start_pfn = zone->zone_start_pfn;
3299 end_pfn = start_pfn + zone->spanned_pages;
3300 reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >>
3304 * Reserve blocks are generally in place to help high-order atomic
3305 * allocations that are short-lived. A min_free_kbytes value that
3306 * would result in more than 2 reserve blocks for atomic allocations
3307 * is assumed to be in place to help anti-fragmentation for the
3308 * future allocation of hugepages at runtime.
3310 reserve = min(2, reserve);
3312 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
3313 if (!pfn_valid(pfn))
3315 page = pfn_to_page(pfn);
3317 /* Watch out for overlapping nodes */
3318 if (page_to_nid(page) != zone_to_nid(zone))
3321 /* Blocks with reserved pages will never free, skip them. */
3322 if (PageReserved(page))
3325 block_migratetype = get_pageblock_migratetype(page);
3327 /* If this block is reserved, account for it */
3328 if (reserve > 0 && block_migratetype == MIGRATE_RESERVE) {
3333 /* Suitable for reserving if this block is movable */
3334 if (reserve > 0 && block_migratetype == MIGRATE_MOVABLE) {
3335 set_pageblock_migratetype(page, MIGRATE_RESERVE);
3336 move_freepages_block(zone, page, MIGRATE_RESERVE);
3342 * If the reserve is met and this is a previous reserved block,
3345 if (block_migratetype == MIGRATE_RESERVE) {
3346 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
3347 move_freepages_block(zone, page, MIGRATE_MOVABLE);
3353 * Initially all pages are reserved - free ones are freed
3354 * up by free_all_bootmem() once the early boot process is
3355 * done. Non-atomic initialization, single-pass.
3357 void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
3358 unsigned long start_pfn, enum memmap_context context)
3361 unsigned long end_pfn = start_pfn + size;
3365 if (highest_memmap_pfn < end_pfn - 1)
3366 highest_memmap_pfn = end_pfn - 1;
3368 z = &NODE_DATA(nid)->node_zones[zone];
3369 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
3371 * There can be holes in boot-time mem_map[]s
3372 * handed to this function. They do not
3373 * exist on hotplugged memory.
3375 if (context == MEMMAP_EARLY) {
3376 if (!early_pfn_valid(pfn))
3378 if (!early_pfn_in_nid(pfn, nid))
3381 page = pfn_to_page(pfn);
3382 set_page_links(page, zone, nid, pfn);
3383 mminit_verify_page_links(page, zone, nid, pfn);
3384 init_page_count(page);
3385 reset_page_mapcount(page);
3386 SetPageReserved(page);
3388 * Mark the block movable so that blocks are reserved for
3389 * movable at startup. This will force kernel allocations
3390 * to reserve their blocks rather than leaking throughout
3391 * the address space during boot when many long-lived
3392 * kernel allocations are made. Later some blocks near
3393 * the start are marked MIGRATE_RESERVE by
3394 * setup_zone_migrate_reserve()
3396 * bitmap is created for zone's valid pfn range. but memmap
3397 * can be created for invalid pages (for alignment)
3398 * check here not to call set_pageblock_migratetype() against
3401 if ((z->zone_start_pfn <= pfn)
3402 && (pfn < z->zone_start_pfn + z->spanned_pages)
3403 && !(pfn & (pageblock_nr_pages - 1)))
3404 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
3406 INIT_LIST_HEAD(&page->lru);
3407 #ifdef WANT_PAGE_VIRTUAL
3408 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
3409 if (!is_highmem_idx(zone))
3410 set_page_address(page, __va(pfn << PAGE_SHIFT));
3415 static void __meminit zone_init_free_lists(struct zone *zone)
3418 for_each_migratetype_order(order, t) {
3419 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
3420 zone->free_area[order].nr_free = 0;
3424 #ifndef __HAVE_ARCH_MEMMAP_INIT
3425 #define memmap_init(size, nid, zone, start_pfn) \
3426 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
3429 static int zone_batchsize(struct zone *zone)
3435 * The per-cpu-pages pools are set to around 1000th of the
3436 * size of the zone. But no more than 1/2 of a meg.
3438 * OK, so we don't know how big the cache is. So guess.
3440 batch = zone->present_pages / 1024;
3441 if (batch * PAGE_SIZE > 512 * 1024)
3442 batch = (512 * 1024) / PAGE_SIZE;
3443 batch /= 4; /* We effectively *= 4 below */
3448 * Clamp the batch to a 2^n - 1 value. Having a power
3449 * of 2 value was found to be more likely to have
3450 * suboptimal cache aliasing properties in some cases.
3452 * For example if 2 tasks are alternately allocating
3453 * batches of pages, one task can end up with a lot
3454 * of pages of one half of the possible page colors
3455 * and the other with pages of the other colors.
3457 batch = rounddown_pow_of_two(batch + batch/2) - 1;
3462 /* The deferral and batching of frees should be suppressed under NOMMU
3465 * The problem is that NOMMU needs to be able to allocate large chunks
3466 * of contiguous memory as there's no hardware page translation to
3467 * assemble apparent contiguous memory from discontiguous pages.
3469 * Queueing large contiguous runs of pages for batching, however,
3470 * causes the pages to actually be freed in smaller chunks. As there
3471 * can be a significant delay between the individual batches being
3472 * recycled, this leads to the once large chunks of space being
3473 * fragmented and becoming unavailable for high-order allocations.
3479 static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
3481 struct per_cpu_pages *pcp;
3484 memset(p, 0, sizeof(*p));
3488 pcp->high = 6 * batch;
3489 pcp->batch = max(1UL, 1 * batch);
3490 for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
3491 INIT_LIST_HEAD(&pcp->lists[migratetype]);
3495 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
3496 * to the value high for the pageset p.
3499 static void setup_pagelist_highmark(struct per_cpu_pageset *p,
3502 struct per_cpu_pages *pcp;
3506 pcp->batch = max(1UL, high/4);
3507 if ((high/4) > (PAGE_SHIFT * 8))
3508 pcp->batch = PAGE_SHIFT * 8;
3511 static __meminit void setup_zone_pageset(struct zone *zone)
3515 zone->pageset = alloc_percpu(struct per_cpu_pageset);
3517 for_each_possible_cpu(cpu) {
3518 struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu);
3520 setup_pageset(pcp, zone_batchsize(zone));
3522 if (percpu_pagelist_fraction)
3523 setup_pagelist_highmark(pcp,
3524 (zone->present_pages /
3525 percpu_pagelist_fraction));
3530 * Allocate per cpu pagesets and initialize them.
3531 * Before this call only boot pagesets were available.
3533 void __init setup_per_cpu_pageset(void)
3537 for_each_populated_zone(zone)
3538 setup_zone_pageset(zone);
3541 static noinline __init_refok
3542 int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
3545 struct pglist_data *pgdat = zone->zone_pgdat;
3549 * The per-page waitqueue mechanism uses hashed waitqueues
3552 zone->wait_table_hash_nr_entries =
3553 wait_table_hash_nr_entries(zone_size_pages);
3554 zone->wait_table_bits =
3555 wait_table_bits(zone->wait_table_hash_nr_entries);
3556 alloc_size = zone->wait_table_hash_nr_entries
3557 * sizeof(wait_queue_head_t);
3559 if (!slab_is_available()) {
3560 zone->wait_table = (wait_queue_head_t *)
3561 alloc_bootmem_node(pgdat, alloc_size);
3564 * This case means that a zone whose size was 0 gets new memory
3565 * via memory hot-add.
3566 * But it may be the case that a new node was hot-added. In
3567 * this case vmalloc() will not be able to use this new node's
3568 * memory - this wait_table must be initialized to use this new
3569 * node itself as well.
3570 * To use this new node's memory, further consideration will be
3573 zone->wait_table = vmalloc(alloc_size);
3575 if (!zone->wait_table)
3578 for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
3579 init_waitqueue_head(zone->wait_table + i);
3584 static int __zone_pcp_update(void *data)
3586 struct zone *zone = data;
3588 unsigned long batch = zone_batchsize(zone), flags;
3590 for_each_possible_cpu(cpu) {
3591 struct per_cpu_pageset *pset;
3592 struct per_cpu_pages *pcp;
3594 pset = per_cpu_ptr(zone->pageset, cpu);
3597 local_irq_save(flags);
3598 free_pcppages_bulk(zone, pcp->count, pcp);
3599 setup_pageset(pset, batch);
3600 local_irq_restore(flags);
3605 void zone_pcp_update(struct zone *zone)
3607 stop_machine(__zone_pcp_update, zone, NULL);
3610 static __meminit void zone_pcp_init(struct zone *zone)
3613 * per cpu subsystem is not up at this point. The following code
3614 * relies on the ability of the linker to provide the
3615 * offset of a (static) per cpu variable into the per cpu area.
3617 zone->pageset = &boot_pageset;
3619 if (zone->present_pages)
3620 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n",
3621 zone->name, zone->present_pages,
3622 zone_batchsize(zone));
3625 __meminit int init_currently_empty_zone(struct zone *zone,
3626 unsigned long zone_start_pfn,
3628 enum memmap_context context)
3630 struct pglist_data *pgdat = zone->zone_pgdat;
3632 ret = zone_wait_table_init(zone, size);
3635 pgdat->nr_zones = zone_idx(zone) + 1;
3637 zone->zone_start_pfn = zone_start_pfn;
3639 mminit_dprintk(MMINIT_TRACE, "memmap_init",
3640 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
3642 (unsigned long)zone_idx(zone),
3643 zone_start_pfn, (zone_start_pfn + size));
3645 zone_init_free_lists(zone);
3650 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3652 * Basic iterator support. Return the first range of PFNs for a node
3653 * Note: nid == MAX_NUMNODES returns first region regardless of node
3655 static int __meminit first_active_region_index_in_nid(int nid)
3659 for (i = 0; i < nr_nodemap_entries; i++)
3660 if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
3667 * Basic iterator support. Return the next active range of PFNs for a node
3668 * Note: nid == MAX_NUMNODES returns next region regardless of node
3670 static int __meminit next_active_region_index_in_nid(int index, int nid)
3672 for (index = index + 1; index < nr_nodemap_entries; index++)
3673 if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
3679 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
3681 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
3682 * Architectures may implement their own version but if add_active_range()
3683 * was used and there are no special requirements, this is a convenient
3686 int __meminit __early_pfn_to_nid(unsigned long pfn)
3690 for (i = 0; i < nr_nodemap_entries; i++) {
3691 unsigned long start_pfn = early_node_map[i].start_pfn;
3692 unsigned long end_pfn = early_node_map[i].end_pfn;
3694 if (start_pfn <= pfn && pfn < end_pfn)
3695 return early_node_map[i].nid;
3697 /* This is a memory hole */
3700 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
3702 int __meminit early_pfn_to_nid(unsigned long pfn)
3706 nid = __early_pfn_to_nid(pfn);
3709 /* just returns 0 */
3713 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
3714 bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
3718 nid = __early_pfn_to_nid(pfn);
3719 if (nid >= 0 && nid != node)
3725 /* Basic iterator support to walk early_node_map[] */
3726 #define for_each_active_range_index_in_nid(i, nid) \
3727 for (i = first_active_region_index_in_nid(nid); i != -1; \
3728 i = next_active_region_index_in_nid(i, nid))
3731 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
3732 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
3733 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
3735 * If an architecture guarantees that all ranges registered with
3736 * add_active_ranges() contain no holes and may be freed, this
3737 * this function may be used instead of calling free_bootmem() manually.
3739 void __init free_bootmem_with_active_regions(int nid,
3740 unsigned long max_low_pfn)
3744 for_each_active_range_index_in_nid(i, nid) {
3745 unsigned long size_pages = 0;
3746 unsigned long end_pfn = early_node_map[i].end_pfn;
3748 if (early_node_map[i].start_pfn >= max_low_pfn)
3751 if (end_pfn > max_low_pfn)
3752 end_pfn = max_low_pfn;
3754 size_pages = end_pfn - early_node_map[i].start_pfn;
3755 free_bootmem_node(NODE_DATA(early_node_map[i].nid),
3756 PFN_PHYS(early_node_map[i].start_pfn),
3757 size_pages << PAGE_SHIFT);
3761 #ifdef CONFIG_HAVE_MEMBLOCK
3763 * Basic iterator support. Return the last range of PFNs for a node
3764 * Note: nid == MAX_NUMNODES returns last region regardless of node
3766 static int __meminit last_active_region_index_in_nid(int nid)
3770 for (i = nr_nodemap_entries - 1; i >= 0; i--)
3771 if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
3778 * Basic iterator support. Return the previous active range of PFNs for a node
3779 * Note: nid == MAX_NUMNODES returns next region regardless of node
3781 static int __meminit previous_active_region_index_in_nid(int index, int nid)
3783 for (index = index - 1; index >= 0; index--)
3784 if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
3790 #define for_each_active_range_index_in_nid_reverse(i, nid) \
3791 for (i = last_active_region_index_in_nid(nid); i != -1; \
3792 i = previous_active_region_index_in_nid(i, nid))
3794 u64 __init find_memory_core_early(int nid, u64 size, u64 align,
3795 u64 goal, u64 limit)
3799 /* Need to go over early_node_map to find out good range for node */
3800 for_each_active_range_index_in_nid_reverse(i, nid) {
3802 u64 ei_start, ei_last;
3803 u64 final_start, final_end;
3805 ei_last = early_node_map[i].end_pfn;
3806 ei_last <<= PAGE_SHIFT;
3807 ei_start = early_node_map[i].start_pfn;
3808 ei_start <<= PAGE_SHIFT;
3810 final_start = max(ei_start, goal);
3811 final_end = min(ei_last, limit);
3813 if (final_start >= final_end)
3816 addr = memblock_find_in_range(final_start, final_end, size, align);
3818 if (addr == MEMBLOCK_ERROR)
3824 return MEMBLOCK_ERROR;
3828 int __init add_from_early_node_map(struct range *range, int az,
3829 int nr_range, int nid)
3834 /* need to go over early_node_map to find out good range for node */
3835 for_each_active_range_index_in_nid(i, nid) {
3836 start = early_node_map[i].start_pfn;
3837 end = early_node_map[i].end_pfn;
3838 nr_range = add_range(range, az, nr_range, start, end);
3843 void __init work_with_active_regions(int nid, work_fn_t work_fn, void *data)
3848 for_each_active_range_index_in_nid(i, nid) {
3849 ret = work_fn(early_node_map[i].start_pfn,
3850 early_node_map[i].end_pfn, data);
3856 * sparse_memory_present_with_active_regions - Call memory_present for each active range
3857 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
3859 * If an architecture guarantees that all ranges registered with
3860 * add_active_ranges() contain no holes and may be freed, this
3861 * function may be used instead of calling memory_present() manually.
3863 void __init sparse_memory_present_with_active_regions(int nid)
3867 for_each_active_range_index_in_nid(i, nid)
3868 memory_present(early_node_map[i].nid,
3869 early_node_map[i].start_pfn,
3870 early_node_map[i].end_pfn);
3874 * get_pfn_range_for_nid - Return the start and end page frames for a node
3875 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
3876 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
3877 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
3879 * It returns the start and end page frame of a node based on information
3880 * provided by an arch calling add_active_range(). If called for a node
3881 * with no available memory, a warning is printed and the start and end
3884 void __meminit get_pfn_range_for_nid(unsigned int nid,
3885 unsigned long *start_pfn, unsigned long *end_pfn)
3891 for_each_active_range_index_in_nid(i, nid) {
3892 *start_pfn = min(*start_pfn, early_node_map[i].start_pfn);
3893 *end_pfn = max(*end_pfn, early_node_map[i].end_pfn);
3896 if (*start_pfn == -1UL)
3901 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3902 * assumption is made that zones within a node are ordered in monotonic
3903 * increasing memory addresses so that the "highest" populated zone is used
3905 static void __init find_usable_zone_for_movable(void)
3908 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
3909 if (zone_index == ZONE_MOVABLE)
3912 if (arch_zone_highest_possible_pfn[zone_index] >
3913 arch_zone_lowest_possible_pfn[zone_index])
3917 VM_BUG_ON(zone_index == -1);
3918 movable_zone = zone_index;
3922 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
3923 * because it is sized independant of architecture. Unlike the other zones,
3924 * the starting point for ZONE_MOVABLE is not fixed. It may be different
3925 * in each node depending on the size of each node and how evenly kernelcore
3926 * is distributed. This helper function adjusts the zone ranges
3927 * provided by the architecture for a given node by using the end of the
3928 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3929 * zones within a node are in order of monotonic increases memory addresses
3931 static void __meminit adjust_zone_range_for_zone_movable(int nid,
3932 unsigned long zone_type,
3933 unsigned long node_start_pfn,
3934 unsigned long node_end_pfn,
3935 unsigned long *zone_start_pfn,
3936 unsigned long *zone_end_pfn)
3938 /* Only adjust if ZONE_MOVABLE is on this node */
3939 if (zone_movable_pfn[nid]) {
3940 /* Size ZONE_MOVABLE */
3941 if (zone_type == ZONE_MOVABLE) {
3942 *zone_start_pfn = zone_movable_pfn[nid];
3943 *zone_end_pfn = min(node_end_pfn,
3944 arch_zone_highest_possible_pfn[movable_zone]);
3946 /* Adjust for ZONE_MOVABLE starting within this range */
3947 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
3948 *zone_end_pfn > zone_movable_pfn[nid]) {
3949 *zone_end_pfn = zone_movable_pfn[nid];
3951 /* Check if this whole range is within ZONE_MOVABLE */
3952 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
3953 *zone_start_pfn = *zone_end_pfn;
3958 * Return the number of pages a zone spans in a node, including holes
3959 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3961 static unsigned long __meminit zone_spanned_pages_in_node(int nid,
3962 unsigned long zone_type,
3963 unsigned long *ignored)
3965 unsigned long node_start_pfn, node_end_pfn;
3966 unsigned long zone_start_pfn, zone_end_pfn;
3968 /* Get the start and end of the node and zone */
3969 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3970 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
3971 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
3972 adjust_zone_range_for_zone_movable(nid, zone_type,
3973 node_start_pfn, node_end_pfn,
3974 &zone_start_pfn, &zone_end_pfn);
3976 /* Check that this node has pages within the zone's required range */
3977 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
3980 /* Move the zone boundaries inside the node if necessary */
3981 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
3982 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
3984 /* Return the spanned pages */
3985 return zone_end_pfn - zone_start_pfn;
3989 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
3990 * then all holes in the requested range will be accounted for.
3992 unsigned long __meminit __absent_pages_in_range(int nid,
3993 unsigned long range_start_pfn,
3994 unsigned long range_end_pfn)
3997 unsigned long prev_end_pfn = 0, hole_pages = 0;
3998 unsigned long start_pfn;
4000 /* Find the end_pfn of the first active range of pfns in the node */
4001 i = first_active_region_index_in_nid(nid);
4005 prev_end_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
4007 /* Account for ranges before physical memory on this node */
4008 if (early_node_map[i].start_pfn > range_start_pfn)
4009 hole_pages = prev_end_pfn - range_start_pfn;
4011 /* Find all holes for the zone within the node */
4012 for (; i != -1; i = next_active_region_index_in_nid(i, nid)) {
4014 /* No need to continue if prev_end_pfn is outside the zone */
4015 if (prev_end_pfn >= range_end_pfn)
4018 /* Make sure the end of the zone is not within the hole */
4019 start_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
4020 prev_end_pfn = max(prev_end_pfn, range_start_pfn);
4022 /* Update the hole size cound and move on */
4023 if (start_pfn > range_start_pfn) {
4024 BUG_ON(prev_end_pfn > start_pfn);
4025 hole_pages += start_pfn - prev_end_pfn;
4027 prev_end_pfn = early_node_map[i].end_pfn;
4030 /* Account for ranges past physical memory on this node */
4031 if (range_end_pfn > prev_end_pfn)
4032 hole_pages += range_end_pfn -
4033 max(range_start_pfn, prev_end_pfn);
4039 * absent_pages_in_range - Return number of page frames in holes within a range
4040 * @start_pfn: The start PFN to start searching for holes
4041 * @end_pfn: The end PFN to stop searching for holes
4043 * It returns the number of pages frames in memory holes within a range.
4045 unsigned long __init absent_pages_in_range(unsigned long start_pfn,
4046 unsigned long end_pfn)
4048 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
4051 /* Return the number of page frames in holes in a zone on a node */
4052 static unsigned long __meminit zone_absent_pages_in_node(int nid,
4053 unsigned long zone_type,
4054 unsigned long *ignored)
4056 unsigned long node_start_pfn, node_end_pfn;
4057 unsigned long zone_start_pfn, zone_end_pfn;
4059 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
4060 zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type],
4062 zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type],
4065 adjust_zone_range_for_zone_movable(nid, zone_type,
4066 node_start_pfn, node_end_pfn,
4067 &zone_start_pfn, &zone_end_pfn);
4068 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
4072 static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
4073 unsigned long zone_type,
4074 unsigned long *zones_size)
4076 return zones_size[zone_type];
4079 static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
4080 unsigned long zone_type,
4081 unsigned long *zholes_size)
4086 return zholes_size[zone_type];
4091 static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
4092 unsigned long *zones_size, unsigned long *zholes_size)
4094 unsigned long realtotalpages, totalpages = 0;
4097 for (i = 0; i < MAX_NR_ZONES; i++)
4098 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
4100 pgdat->node_spanned_pages = totalpages;
4102 realtotalpages = totalpages;
4103 for (i = 0; i < MAX_NR_ZONES; i++)
4105 zone_absent_pages_in_node(pgdat->node_id, i,
4107 pgdat->node_present_pages = realtotalpages;
4108 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
4112 #ifndef CONFIG_SPARSEMEM
4114 * Calculate the size of the zone->blockflags rounded to an unsigned long
4115 * Start by making sure zonesize is a multiple of pageblock_order by rounding
4116 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
4117 * round what is now in bits to nearest long in bits, then return it in
4120 static unsigned long __init usemap_size(unsigned long zonesize)
4122 unsigned long usemapsize;
4124 usemapsize = roundup(zonesize, pageblock_nr_pages);
4125 usemapsize = usemapsize >> pageblock_order;
4126 usemapsize *= NR_PAGEBLOCK_BITS;
4127 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
4129 return usemapsize / 8;
4132 static void __init setup_usemap(struct pglist_data *pgdat,
4133 struct zone *zone, unsigned long zonesize)
4135 unsigned long usemapsize = usemap_size(zonesize);
4136 zone->pageblock_flags = NULL;
4138 zone->pageblock_flags = alloc_bootmem_node(pgdat, usemapsize);
4141 static inline void setup_usemap(struct pglist_data *pgdat,
4142 struct zone *zone, unsigned long zonesize) {}
4143 #endif /* CONFIG_SPARSEMEM */
4145 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
4147 /* Return a sensible default order for the pageblock size. */
4148 static inline int pageblock_default_order(void)
4150 if (HPAGE_SHIFT > PAGE_SHIFT)
4151 return HUGETLB_PAGE_ORDER;
4156 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
4157 static inline void __init set_pageblock_order(unsigned int order)
4159 /* Check that pageblock_nr_pages has not already been setup */
4160 if (pageblock_order)
4164 * Assume the largest contiguous order of interest is a huge page.
4165 * This value may be variable depending on boot parameters on IA64
4167 pageblock_order = order;
4169 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4172 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
4173 * and pageblock_default_order() are unused as pageblock_order is set
4174 * at compile-time. See include/linux/pageblock-flags.h for the values of
4175 * pageblock_order based on the kernel config
4177 static inline int pageblock_default_order(unsigned int order)
4181 #define set_pageblock_order(x) do {} while (0)
4183 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4186 * Set up the zone data structures:
4187 * - mark all pages reserved
4188 * - mark all memory queues empty
4189 * - clear the memory bitmaps
4191 static void __paginginit free_area_init_core(struct pglist_data *pgdat,
4192 unsigned long *zones_size, unsigned long *zholes_size)
4195 int nid = pgdat->node_id;
4196 unsigned long zone_start_pfn = pgdat->node_start_pfn;
4199 pgdat_resize_init(pgdat);
4200 pgdat->nr_zones = 0;
4201 init_waitqueue_head(&pgdat->kswapd_wait);
4202 pgdat->kswapd_max_order = 0;
4203 pgdat_page_cgroup_init(pgdat);
4205 for (j = 0; j < MAX_NR_ZONES; j++) {
4206 struct zone *zone = pgdat->node_zones + j;
4207 unsigned long size, realsize, memmap_pages;
4210 size = zone_spanned_pages_in_node(nid, j, zones_size);
4211 realsize = size - zone_absent_pages_in_node(nid, j,
4215 * Adjust realsize so that it accounts for how much memory
4216 * is used by this zone for memmap. This affects the watermark
4217 * and per-cpu initialisations
4220 PAGE_ALIGN(size * sizeof(struct page)) >> PAGE_SHIFT;
4221 if (realsize >= memmap_pages) {
4222 realsize -= memmap_pages;
4225 " %s zone: %lu pages used for memmap\n",
4226 zone_names[j], memmap_pages);
4229 " %s zone: %lu pages exceeds realsize %lu\n",
4230 zone_names[j], memmap_pages, realsize);
4232 /* Account for reserved pages */
4233 if (j == 0 && realsize > dma_reserve) {
4234 realsize -= dma_reserve;
4235 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
4236 zone_names[0], dma_reserve);
4239 if (!is_highmem_idx(j))
4240 nr_kernel_pages += realsize;
4241 nr_all_pages += realsize;
4243 zone->spanned_pages = size;
4244 zone->present_pages = realsize;
4247 zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
4249 zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100;
4251 zone->name = zone_names[j];
4252 spin_lock_init(&zone->lock);
4253 spin_lock_init(&zone->lru_lock);
4254 zone_seqlock_init(zone);
4255 zone->zone_pgdat = pgdat;
4257 zone_pcp_init(zone);
4259 INIT_LIST_HEAD(&zone->lru[l].list);
4260 zone->reclaim_stat.nr_saved_scan[l] = 0;
4262 zone->reclaim_stat.recent_rotated[0] = 0;
4263 zone->reclaim_stat.recent_rotated[1] = 0;
4264 zone->reclaim_stat.recent_scanned[0] = 0;
4265 zone->reclaim_stat.recent_scanned[1] = 0;
4266 zap_zone_vm_stats(zone);
4271 set_pageblock_order(pageblock_default_order());
4272 setup_usemap(pgdat, zone, size);
4273 ret = init_currently_empty_zone(zone, zone_start_pfn,
4274 size, MEMMAP_EARLY);
4276 memmap_init(size, nid, j, zone_start_pfn);
4277 zone_start_pfn += size;
4281 static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
4283 /* Skip empty nodes */
4284 if (!pgdat->node_spanned_pages)
4287 #ifdef CONFIG_FLAT_NODE_MEM_MAP
4288 /* ia64 gets its own node_mem_map, before this, without bootmem */
4289 if (!pgdat->node_mem_map) {
4290 unsigned long size, start, end;
4294 * The zone's endpoints aren't required to be MAX_ORDER
4295 * aligned but the node_mem_map endpoints must be in order
4296 * for the buddy allocator to function correctly.
4298 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
4299 end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
4300 end = ALIGN(end, MAX_ORDER_NR_PAGES);
4301 size = (end - start) * sizeof(struct page);
4302 map = alloc_remap(pgdat->node_id, size);
4304 map = alloc_bootmem_node(pgdat, size);
4305 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
4307 #ifndef CONFIG_NEED_MULTIPLE_NODES
4309 * With no DISCONTIG, the global mem_map is just set as node 0's
4311 if (pgdat == NODE_DATA(0)) {
4312 mem_map = NODE_DATA(0)->node_mem_map;
4313 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
4314 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
4315 mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
4316 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4319 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
4322 void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
4323 unsigned long node_start_pfn, unsigned long *zholes_size)
4325 pg_data_t *pgdat = NODE_DATA(nid);
4327 pgdat->node_id = nid;
4328 pgdat->node_start_pfn = node_start_pfn;
4329 calculate_node_totalpages(pgdat, zones_size, zholes_size);
4331 alloc_node_mem_map(pgdat);
4332 #ifdef CONFIG_FLAT_NODE_MEM_MAP
4333 printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
4334 nid, (unsigned long)pgdat,
4335 (unsigned long)pgdat->node_mem_map);
4338 free_area_init_core(pgdat, zones_size, zholes_size);
4341 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
4343 #if MAX_NUMNODES > 1
4345 * Figure out the number of possible node ids.
4347 static void __init setup_nr_node_ids(void)
4350 unsigned int highest = 0;
4352 for_each_node_mask(node, node_possible_map)
4354 nr_node_ids = highest + 1;
4357 static inline void setup_nr_node_ids(void)
4363 * add_active_range - Register a range of PFNs backed by physical memory
4364 * @nid: The node ID the range resides on
4365 * @start_pfn: The start PFN of the available physical memory
4366 * @end_pfn: The end PFN of the available physical memory
4368 * These ranges are stored in an early_node_map[] and later used by
4369 * free_area_init_nodes() to calculate zone sizes and holes. If the
4370 * range spans a memory hole, it is up to the architecture to ensure
4371 * the memory is not freed by the bootmem allocator. If possible
4372 * the range being registered will be merged with existing ranges.
4374 void __init add_active_range(unsigned int nid, unsigned long start_pfn,
4375 unsigned long end_pfn)
4379 mminit_dprintk(MMINIT_TRACE, "memory_register",
4380 "Entering add_active_range(%d, %#lx, %#lx) "
4381 "%d entries of %d used\n",
4382 nid, start_pfn, end_pfn,
4383 nr_nodemap_entries, MAX_ACTIVE_REGIONS);
4385 mminit_validate_memmodel_limits(&start_pfn, &end_pfn);
4387 /* Merge with existing active regions if possible */
4388 for (i = 0; i < nr_nodemap_entries; i++) {
4389 if (early_node_map[i].nid != nid)
4392 /* Skip if an existing region covers this new one */
4393 if (start_pfn >= early_node_map[i].start_pfn &&
4394 end_pfn <= early_node_map[i].end_pfn)
4397 /* Merge forward if suitable */
4398 if (start_pfn <= early_node_map[i].end_pfn &&
4399 end_pfn > early_node_map[i].end_pfn) {
4400 early_node_map[i].end_pfn = end_pfn;
4404 /* Merge backward if suitable */
4405 if (start_pfn < early_node_map[i].start_pfn &&
4406 end_pfn >= early_node_map[i].start_pfn) {
4407 early_node_map[i].start_pfn = start_pfn;
4412 /* Check that early_node_map is large enough */
4413 if (i >= MAX_ACTIVE_REGIONS) {
4414 printk(KERN_CRIT "More than %d memory regions, truncating\n",
4415 MAX_ACTIVE_REGIONS);
4419 early_node_map[i].nid = nid;
4420 early_node_map[i].start_pfn = start_pfn;
4421 early_node_map[i].end_pfn = end_pfn;
4422 nr_nodemap_entries = i + 1;
4426 * remove_active_range - Shrink an existing registered range of PFNs
4427 * @nid: The node id the range is on that should be shrunk
4428 * @start_pfn: The new PFN of the range
4429 * @end_pfn: The new PFN of the range
4431 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
4432 * The map is kept near the end physical page range that has already been
4433 * registered. This function allows an arch to shrink an existing registered
4436 void __init remove_active_range(unsigned int nid, unsigned long start_pfn,
4437 unsigned long end_pfn)
4442 printk(KERN_DEBUG "remove_active_range (%d, %lu, %lu)\n",
4443 nid, start_pfn, end_pfn);
4445 /* Find the old active region end and shrink */
4446 for_each_active_range_index_in_nid(i, nid) {
4447 if (early_node_map[i].start_pfn >= start_pfn &&
4448 early_node_map[i].end_pfn <= end_pfn) {
4450 early_node_map[i].start_pfn = 0;
4451 early_node_map[i].end_pfn = 0;
4455 if (early_node_map[i].start_pfn < start_pfn &&
4456 early_node_map[i].end_pfn > start_pfn) {
4457 unsigned long temp_end_pfn = early_node_map[i].end_pfn;
4458 early_node_map[i].end_pfn = start_pfn;
4459 if (temp_end_pfn > end_pfn)
4460 add_active_range(nid, end_pfn, temp_end_pfn);
4463 if (early_node_map[i].start_pfn >= start_pfn &&
4464 early_node_map[i].end_pfn > end_pfn &&
4465 early_node_map[i].start_pfn < end_pfn) {
4466 early_node_map[i].start_pfn = end_pfn;
4474 /* remove the blank ones */
4475 for (i = nr_nodemap_entries - 1; i > 0; i--) {
4476 if (early_node_map[i].nid != nid)
4478 if (early_node_map[i].end_pfn)
4480 /* we found it, get rid of it */
4481 for (j = i; j < nr_nodemap_entries - 1; j++)
4482 memcpy(&early_node_map[j], &early_node_map[j+1],
4483 sizeof(early_node_map[j]));
4484 j = nr_nodemap_entries - 1;
4485 memset(&early_node_map[j], 0, sizeof(early_node_map[j]));
4486 nr_nodemap_entries--;
4491 * remove_all_active_ranges - Remove all currently registered regions
4493 * During discovery, it may be found that a table like SRAT is invalid
4494 * and an alternative discovery method must be used. This function removes
4495 * all currently registered regions.
4497 void __init remove_all_active_ranges(void)
4499 memset(early_node_map, 0, sizeof(early_node_map));
4500 nr_nodemap_entries = 0;
4503 /* Compare two active node_active_regions */
4504 static int __init cmp_node_active_region(const void *a, const void *b)
4506 struct node_active_region *arange = (struct node_active_region *)a;
4507 struct node_active_region *brange = (struct node_active_region *)b;
4509 /* Done this way to avoid overflows */
4510 if (arange->start_pfn > brange->start_pfn)
4512 if (arange->start_pfn < brange->start_pfn)
4518 /* sort the node_map by start_pfn */
4519 void __init sort_node_map(void)
4521 sort(early_node_map, (size_t)nr_nodemap_entries,
4522 sizeof(struct node_active_region),
4523 cmp_node_active_region, NULL);
4526 /* Find the lowest pfn for a node */
4527 static unsigned long __init find_min_pfn_for_node(int nid)
4530 unsigned long min_pfn = ULONG_MAX;
4532 /* Assuming a sorted map, the first range found has the starting pfn */
4533 for_each_active_range_index_in_nid(i, nid)
4534 min_pfn = min(min_pfn, early_node_map[i].start_pfn);
4536 if (min_pfn == ULONG_MAX) {
4538 "Could not find start_pfn for node %d\n", nid);
4546 * find_min_pfn_with_active_regions - Find the minimum PFN registered
4548 * It returns the minimum PFN based on information provided via
4549 * add_active_range().
4551 unsigned long __init find_min_pfn_with_active_regions(void)
4553 return find_min_pfn_for_node(MAX_NUMNODES);
4557 * early_calculate_totalpages()
4558 * Sum pages in active regions for movable zone.
4559 * Populate N_HIGH_MEMORY for calculating usable_nodes.
4561 static unsigned long __init early_calculate_totalpages(void)
4564 unsigned long totalpages = 0;
4566 for (i = 0; i < nr_nodemap_entries; i++) {
4567 unsigned long pages = early_node_map[i].end_pfn -
4568 early_node_map[i].start_pfn;
4569 totalpages += pages;
4571 node_set_state(early_node_map[i].nid, N_HIGH_MEMORY);
4577 * Find the PFN the Movable zone begins in each node. Kernel memory
4578 * is spread evenly between nodes as long as the nodes have enough
4579 * memory. When they don't, some nodes will have more kernelcore than
4582 static void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn)
4585 unsigned long usable_startpfn;
4586 unsigned long kernelcore_node, kernelcore_remaining;
4587 /* save the state before borrow the nodemask */
4588 nodemask_t saved_node_state = node_states[N_HIGH_MEMORY];
4589 unsigned long totalpages = early_calculate_totalpages();
4590 int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]);
4593 * If movablecore was specified, calculate what size of
4594 * kernelcore that corresponds so that memory usable for
4595 * any allocation type is evenly spread. If both kernelcore
4596 * and movablecore are specified, then the value of kernelcore
4597 * will be used for required_kernelcore if it's greater than
4598 * what movablecore would have allowed.
4600 if (required_movablecore) {
4601 unsigned long corepages;
4604 * Round-up so that ZONE_MOVABLE is at least as large as what
4605 * was requested by the user
4607 required_movablecore =
4608 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
4609 corepages = totalpages - required_movablecore;
4611 required_kernelcore = max(required_kernelcore, corepages);
4614 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
4615 if (!required_kernelcore)
4618 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
4619 find_usable_zone_for_movable();
4620 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
4623 /* Spread kernelcore memory as evenly as possible throughout nodes */
4624 kernelcore_node = required_kernelcore / usable_nodes;
4625 for_each_node_state(nid, N_HIGH_MEMORY) {
4627 * Recalculate kernelcore_node if the division per node
4628 * now exceeds what is necessary to satisfy the requested
4629 * amount of memory for the kernel
4631 if (required_kernelcore < kernelcore_node)
4632 kernelcore_node = required_kernelcore / usable_nodes;
4635 * As the map is walked, we track how much memory is usable
4636 * by the kernel using kernelcore_remaining. When it is
4637 * 0, the rest of the node is usable by ZONE_MOVABLE
4639 kernelcore_remaining = kernelcore_node;
4641 /* Go through each range of PFNs within this node */
4642 for_each_active_range_index_in_nid(i, nid) {
4643 unsigned long start_pfn, end_pfn;
4644 unsigned long size_pages;
4646 start_pfn = max(early_node_map[i].start_pfn,
4647 zone_movable_pfn[nid]);
4648 end_pfn = early_node_map[i].end_pfn;
4649 if (start_pfn >= end_pfn)
4652 /* Account for what is only usable for kernelcore */
4653 if (start_pfn < usable_startpfn) {
4654 unsigned long kernel_pages;
4655 kernel_pages = min(end_pfn, usable_startpfn)
4658 kernelcore_remaining -= min(kernel_pages,
4659 kernelcore_remaining);
4660 required_kernelcore -= min(kernel_pages,
4661 required_kernelcore);
4663 /* Continue if range is now fully accounted */
4664 if (end_pfn <= usable_startpfn) {
4667 * Push zone_movable_pfn to the end so
4668 * that if we have to rebalance
4669 * kernelcore across nodes, we will
4670 * not double account here
4672 zone_movable_pfn[nid] = end_pfn;
4675 start_pfn = usable_startpfn;
4679 * The usable PFN range for ZONE_MOVABLE is from
4680 * start_pfn->end_pfn. Calculate size_pages as the
4681 * number of pages used as kernelcore
4683 size_pages = end_pfn - start_pfn;
4684 if (size_pages > kernelcore_remaining)
4685 size_pages = kernelcore_remaining;
4686 zone_movable_pfn[nid] = start_pfn + size_pages;
4689 * Some kernelcore has been met, update counts and
4690 * break if the kernelcore for this node has been
4693 required_kernelcore -= min(required_kernelcore,
4695 kernelcore_remaining -= size_pages;
4696 if (!kernelcore_remaining)
4702 * If there is still required_kernelcore, we do another pass with one
4703 * less node in the count. This will push zone_movable_pfn[nid] further
4704 * along on the nodes that still have memory until kernelcore is
4708 if (usable_nodes && required_kernelcore > usable_nodes)
4711 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
4712 for (nid = 0; nid < MAX_NUMNODES; nid++)
4713 zone_movable_pfn[nid] =
4714 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
4717 /* restore the node_state */
4718 node_states[N_HIGH_MEMORY] = saved_node_state;
4721 /* Any regular memory on that node ? */
4722 static void check_for_regular_memory(pg_data_t *pgdat)
4724 #ifdef CONFIG_HIGHMEM
4725 enum zone_type zone_type;
4727 for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) {
4728 struct zone *zone = &pgdat->node_zones[zone_type];
4729 if (zone->present_pages)
4730 node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY);
4736 * free_area_init_nodes - Initialise all pg_data_t and zone data
4737 * @max_zone_pfn: an array of max PFNs for each zone
4739 * This will call free_area_init_node() for each active node in the system.
4740 * Using the page ranges provided by add_active_range(), the size of each
4741 * zone in each node and their holes is calculated. If the maximum PFN
4742 * between two adjacent zones match, it is assumed that the zone is empty.
4743 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
4744 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
4745 * starts where the previous one ended. For example, ZONE_DMA32 starts
4746 * at arch_max_dma_pfn.
4748 void __init free_area_init_nodes(unsigned long *max_zone_pfn)
4753 /* Sort early_node_map as initialisation assumes it is sorted */
4756 /* Record where the zone boundaries are */
4757 memset(arch_zone_lowest_possible_pfn, 0,
4758 sizeof(arch_zone_lowest_possible_pfn));
4759 memset(arch_zone_highest_possible_pfn, 0,
4760 sizeof(arch_zone_highest_possible_pfn));
4761 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
4762 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
4763 for (i = 1; i < MAX_NR_ZONES; i++) {
4764 if (i == ZONE_MOVABLE)
4766 arch_zone_lowest_possible_pfn[i] =
4767 arch_zone_highest_possible_pfn[i-1];
4768 arch_zone_highest_possible_pfn[i] =
4769 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
4771 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
4772 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
4774 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
4775 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
4776 find_zone_movable_pfns_for_nodes(zone_movable_pfn);
4778 /* Print out the zone ranges */
4779 printk("Zone PFN ranges:\n");
4780 for (i = 0; i < MAX_NR_ZONES; i++) {
4781 if (i == ZONE_MOVABLE)
4783 printk(" %-8s ", zone_names[i]);
4784 if (arch_zone_lowest_possible_pfn[i] ==
4785 arch_zone_highest_possible_pfn[i])
4788 printk("%0#10lx -> %0#10lx\n",
4789 arch_zone_lowest_possible_pfn[i],
4790 arch_zone_highest_possible_pfn[i]);
4793 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
4794 printk("Movable zone start PFN for each node\n");
4795 for (i = 0; i < MAX_NUMNODES; i++) {
4796 if (zone_movable_pfn[i])
4797 printk(" Node %d: %lu\n", i, zone_movable_pfn[i]);
4800 /* Print out the early_node_map[] */
4801 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries);
4802 for (i = 0; i < nr_nodemap_entries; i++)
4803 printk(" %3d: %0#10lx -> %0#10lx\n", early_node_map[i].nid,
4804 early_node_map[i].start_pfn,
4805 early_node_map[i].end_pfn);
4807 /* Initialise every node */
4808 mminit_verify_pageflags_layout();
4809 setup_nr_node_ids();
4810 for_each_online_node(nid) {
4811 pg_data_t *pgdat = NODE_DATA(nid);
4812 free_area_init_node(nid, NULL,
4813 find_min_pfn_for_node(nid), NULL);
4815 /* Any memory on that node */
4816 if (pgdat->node_present_pages)
4817 node_set_state(nid, N_HIGH_MEMORY);
4818 check_for_regular_memory(pgdat);
4822 static int __init cmdline_parse_core(char *p, unsigned long *core)
4824 unsigned long long coremem;
4828 coremem = memparse(p, &p);
4829 *core = coremem >> PAGE_SHIFT;
4831 /* Paranoid check that UL is enough for the coremem value */
4832 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
4838 * kernelcore=size sets the amount of memory for use for allocations that
4839 * cannot be reclaimed or migrated.
4841 static int __init cmdline_parse_kernelcore(char *p)
4843 return cmdline_parse_core(p, &required_kernelcore);
4847 * movablecore=size sets the amount of memory for use for allocations that
4848 * can be reclaimed or migrated.
4850 static int __init cmdline_parse_movablecore(char *p)
4852 return cmdline_parse_core(p, &required_movablecore);
4855 early_param("kernelcore", cmdline_parse_kernelcore);
4856 early_param("movablecore", cmdline_parse_movablecore);
4858 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4861 * set_dma_reserve - set the specified number of pages reserved in the first zone
4862 * @new_dma_reserve: The number of pages to mark reserved
4864 * The per-cpu batchsize and zone watermarks are determined by present_pages.
4865 * In the DMA zone, a significant percentage may be consumed by kernel image
4866 * and other unfreeable allocations which can skew the watermarks badly. This
4867 * function may optionally be used to account for unfreeable pages in the
4868 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
4869 * smaller per-cpu batchsize.
4871 void __init set_dma_reserve(unsigned long new_dma_reserve)
4873 dma_reserve = new_dma_reserve;
4876 void __init free_area_init(unsigned long *zones_size)
4878 free_area_init_node(0, zones_size,
4879 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
4882 static int page_alloc_cpu_notify(struct notifier_block *self,
4883 unsigned long action, void *hcpu)
4885 int cpu = (unsigned long)hcpu;
4887 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
4891 * Spill the event counters of the dead processor
4892 * into the current processors event counters.
4893 * This artificially elevates the count of the current
4896 vm_events_fold_cpu(cpu);
4899 * Zero the differential counters of the dead processor
4900 * so that the vm statistics are consistent.
4902 * This is only okay since the processor is dead and cannot
4903 * race with what we are doing.
4905 refresh_cpu_vm_stats(cpu);
4910 void __init page_alloc_init(void)
4912 hotcpu_notifier(page_alloc_cpu_notify, 0);
4916 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
4917 * or min_free_kbytes changes.
4919 static void calculate_totalreserve_pages(void)
4921 struct pglist_data *pgdat;
4922 unsigned long reserve_pages = 0;
4923 enum zone_type i, j;
4925 for_each_online_pgdat(pgdat) {
4926 for (i = 0; i < MAX_NR_ZONES; i++) {
4927 struct zone *zone = pgdat->node_zones + i;
4928 unsigned long max = 0;
4930 /* Find valid and maximum lowmem_reserve in the zone */
4931 for (j = i; j < MAX_NR_ZONES; j++) {
4932 if (zone->lowmem_reserve[j] > max)
4933 max = zone->lowmem_reserve[j];
4936 /* we treat the high watermark as reserved pages. */
4937 max += high_wmark_pages(zone);
4939 if (max > zone->present_pages)
4940 max = zone->present_pages;
4941 reserve_pages += max;
4944 totalreserve_pages = reserve_pages;
4948 * setup_per_zone_lowmem_reserve - called whenever
4949 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
4950 * has a correct pages reserved value, so an adequate number of
4951 * pages are left in the zone after a successful __alloc_pages().
4953 static void setup_per_zone_lowmem_reserve(void)
4955 struct pglist_data *pgdat;
4956 enum zone_type j, idx;
4958 for_each_online_pgdat(pgdat) {
4959 for (j = 0; j < MAX_NR_ZONES; j++) {
4960 struct zone *zone = pgdat->node_zones + j;
4961 unsigned long present_pages = zone->present_pages;
4963 zone->lowmem_reserve[j] = 0;
4967 struct zone *lower_zone;
4971 if (sysctl_lowmem_reserve_ratio[idx] < 1)
4972 sysctl_lowmem_reserve_ratio[idx] = 1;
4974 lower_zone = pgdat->node_zones + idx;
4975 lower_zone->lowmem_reserve[j] = present_pages /
4976 sysctl_lowmem_reserve_ratio[idx];
4977 present_pages += lower_zone->present_pages;
4982 /* update totalreserve_pages */
4983 calculate_totalreserve_pages();
4987 * setup_per_zone_wmarks - called when min_free_kbytes changes
4988 * or when memory is hot-{added|removed}
4990 * Ensures that the watermark[min,low,high] values for each zone are set
4991 * correctly with respect to min_free_kbytes.
4993 void setup_per_zone_wmarks(void)
4995 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
4996 unsigned long lowmem_pages = 0;
4998 unsigned long flags;
5000 /* Calculate total number of !ZONE_HIGHMEM pages */
5001 for_each_zone(zone) {
5002 if (!is_highmem(zone))
5003 lowmem_pages += zone->present_pages;
5006 for_each_zone(zone) {
5009 spin_lock_irqsave(&zone->lock, flags);
5010 tmp = (u64)pages_min * zone->present_pages;
5011 do_div(tmp, lowmem_pages);
5012 if (is_highmem(zone)) {
5014 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
5015 * need highmem pages, so cap pages_min to a small
5018 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
5019 * deltas controls asynch page reclaim, and so should
5020 * not be capped for highmem.
5024 min_pages = zone->present_pages / 1024;
5025 if (min_pages < SWAP_CLUSTER_MAX)
5026 min_pages = SWAP_CLUSTER_MAX;
5027 if (min_pages > 128)
5029 zone->watermark[WMARK_MIN] = min_pages;
5032 * If it's a lowmem zone, reserve a number of pages
5033 * proportionate to the zone's size.
5035 zone->watermark[WMARK_MIN] = tmp;
5038 zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + (tmp >> 2);
5039 zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1);
5040 setup_zone_migrate_reserve(zone);
5041 spin_unlock_irqrestore(&zone->lock, flags);
5044 /* update totalreserve_pages */
5045 calculate_totalreserve_pages();
5049 * The inactive anon list should be small enough that the VM never has to
5050 * do too much work, but large enough that each inactive page has a chance
5051 * to be referenced again before it is swapped out.
5053 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
5054 * INACTIVE_ANON pages on this zone's LRU, maintained by the
5055 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
5056 * the anonymous pages are kept on the inactive list.
5059 * memory ratio inactive anon
5060 * -------------------------------------
5069 void calculate_zone_inactive_ratio(struct zone *zone)
5071 unsigned int gb, ratio;
5073 /* Zone size in gigabytes */
5074 gb = zone->present_pages >> (30 - PAGE_SHIFT);
5076 ratio = int_sqrt(10 * gb);
5080 zone->inactive_ratio = ratio;
5083 static void __init setup_per_zone_inactive_ratio(void)
5088 calculate_zone_inactive_ratio(zone);
5092 * Initialise min_free_kbytes.
5094 * For small machines we want it small (128k min). For large machines
5095 * we want it large (64MB max). But it is not linear, because network
5096 * bandwidth does not increase linearly with machine size. We use
5098 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
5099 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
5115 static int __init init_per_zone_wmark_min(void)
5117 unsigned long lowmem_kbytes;
5119 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
5121 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
5122 if (min_free_kbytes < 128)
5123 min_free_kbytes = 128;
5124 if (min_free_kbytes > 65536)
5125 min_free_kbytes = 65536;
5126 setup_per_zone_wmarks();
5127 setup_per_zone_lowmem_reserve();
5128 setup_per_zone_inactive_ratio();
5131 module_init(init_per_zone_wmark_min)
5134 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
5135 * that we can call two helper functions whenever min_free_kbytes
5138 int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
5139 void __user *buffer, size_t *length, loff_t *ppos)
5141 proc_dointvec(table, write, buffer, length, ppos);
5143 setup_per_zone_wmarks();
5148 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
5149 void __user *buffer, size_t *length, loff_t *ppos)
5154 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
5159 zone->min_unmapped_pages = (zone->present_pages *
5160 sysctl_min_unmapped_ratio) / 100;
5164 int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
5165 void __user *buffer, size_t *length, loff_t *ppos)
5170 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
5175 zone->min_slab_pages = (zone->present_pages *
5176 sysctl_min_slab_ratio) / 100;
5182 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
5183 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
5184 * whenever sysctl_lowmem_reserve_ratio changes.
5186 * The reserve ratio obviously has absolutely no relation with the
5187 * minimum watermarks. The lowmem reserve ratio can only make sense
5188 * if in function of the boot time zone sizes.
5190 int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
5191 void __user *buffer, size_t *length, loff_t *ppos)
5193 proc_dointvec_minmax(table, write, buffer, length, ppos);
5194 setup_per_zone_lowmem_reserve();
5199 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
5200 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
5201 * can have before it gets flushed back to buddy allocator.
5204 int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
5205 void __user *buffer, size_t *length, loff_t *ppos)
5211 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
5212 if (!write || (ret == -EINVAL))
5214 for_each_populated_zone(zone) {
5215 for_each_possible_cpu(cpu) {
5217 high = zone->present_pages / percpu_pagelist_fraction;
5218 setup_pagelist_highmark(
5219 per_cpu_ptr(zone->pageset, cpu), high);
5225 int hashdist = HASHDIST_DEFAULT;
5228 static int __init set_hashdist(char *str)
5232 hashdist = simple_strtoul(str, &str, 0);
5235 __setup("hashdist=", set_hashdist);
5239 * allocate a large system hash table from bootmem
5240 * - it is assumed that the hash table must contain an exact power-of-2
5241 * quantity of entries
5242 * - limit is the number of hash buckets, not the total allocation size
5244 void *__init alloc_large_system_hash(const char *tablename,
5245 unsigned long bucketsize,
5246 unsigned long numentries,
5249 unsigned int *_hash_shift,
5250 unsigned int *_hash_mask,
5251 unsigned long limit)
5253 unsigned long long max = limit;
5254 unsigned long log2qty, size;
5257 /* allow the kernel cmdline to have a say */
5259 /* round applicable memory size up to nearest megabyte */
5260 numentries = nr_kernel_pages;
5261 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
5262 numentries >>= 20 - PAGE_SHIFT;
5263 numentries <<= 20 - PAGE_SHIFT;
5265 /* limit to 1 bucket per 2^scale bytes of low memory */
5266 if (scale > PAGE_SHIFT)
5267 numentries >>= (scale - PAGE_SHIFT);
5269 numentries <<= (PAGE_SHIFT - scale);
5271 /* Make sure we've got at least a 0-order allocation.. */
5272 if (unlikely(flags & HASH_SMALL)) {
5273 /* Makes no sense without HASH_EARLY */
5274 WARN_ON(!(flags & HASH_EARLY));
5275 if (!(numentries >> *_hash_shift)) {
5276 numentries = 1UL << *_hash_shift;
5277 BUG_ON(!numentries);
5279 } else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
5280 numentries = PAGE_SIZE / bucketsize;
5282 numentries = roundup_pow_of_two(numentries);
5284 /* limit allocation size to 1/16 total memory by default */
5286 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
5287 do_div(max, bucketsize);
5290 if (numentries > max)
5293 log2qty = ilog2(numentries);
5296 size = bucketsize << log2qty;
5297 if (flags & HASH_EARLY)
5298 table = alloc_bootmem_nopanic(size);
5300 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
5303 * If bucketsize is not a power-of-two, we may free
5304 * some pages at the end of hash table which
5305 * alloc_pages_exact() automatically does
5307 if (get_order(size) < MAX_ORDER) {
5308 table = alloc_pages_exact(size, GFP_ATOMIC);
5309 kmemleak_alloc(table, size, 1, GFP_ATOMIC);
5312 } while (!table && size > PAGE_SIZE && --log2qty);
5315 panic("Failed to allocate %s hash table\n", tablename);
5317 printk(KERN_INFO "%s hash table entries: %ld (order: %d, %lu bytes)\n",
5320 ilog2(size) - PAGE_SHIFT,
5324 *_hash_shift = log2qty;
5326 *_hash_mask = (1 << log2qty) - 1;
5331 /* Return a pointer to the bitmap storing bits affecting a block of pages */
5332 static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
5335 #ifdef CONFIG_SPARSEMEM
5336 return __pfn_to_section(pfn)->pageblock_flags;
5338 return zone->pageblock_flags;
5339 #endif /* CONFIG_SPARSEMEM */
5342 static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
5344 #ifdef CONFIG_SPARSEMEM
5345 pfn &= (PAGES_PER_SECTION-1);
5346 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
5348 pfn = pfn - zone->zone_start_pfn;
5349 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
5350 #endif /* CONFIG_SPARSEMEM */
5354 * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
5355 * @page: The page within the block of interest
5356 * @start_bitidx: The first bit of interest to retrieve
5357 * @end_bitidx: The last bit of interest
5358 * returns pageblock_bits flags
5360 unsigned long get_pageblock_flags_group(struct page *page,
5361 int start_bitidx, int end_bitidx)
5364 unsigned long *bitmap;
5365 unsigned long pfn, bitidx;
5366 unsigned long flags = 0;
5367 unsigned long value = 1;
5369 zone = page_zone(page);
5370 pfn = page_to_pfn(page);
5371 bitmap = get_pageblock_bitmap(zone, pfn);
5372 bitidx = pfn_to_bitidx(zone, pfn);
5374 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
5375 if (test_bit(bitidx + start_bitidx, bitmap))
5382 * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
5383 * @page: The page within the block of interest
5384 * @start_bitidx: The first bit of interest
5385 * @end_bitidx: The last bit of interest
5386 * @flags: The flags to set
5388 void set_pageblock_flags_group(struct page *page, unsigned long flags,
5389 int start_bitidx, int end_bitidx)
5392 unsigned long *bitmap;
5393 unsigned long pfn, bitidx;
5394 unsigned long value = 1;
5396 zone = page_zone(page);
5397 pfn = page_to_pfn(page);
5398 bitmap = get_pageblock_bitmap(zone, pfn);
5399 bitidx = pfn_to_bitidx(zone, pfn);
5400 VM_BUG_ON(pfn < zone->zone_start_pfn);
5401 VM_BUG_ON(pfn >= zone->zone_start_pfn + zone->spanned_pages);
5403 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
5405 __set_bit(bitidx + start_bitidx, bitmap);
5407 __clear_bit(bitidx + start_bitidx, bitmap);
5411 * This is designed as sub function...plz see page_isolation.c also.
5412 * set/clear page block's type to be ISOLATE.
5413 * page allocater never alloc memory from ISOLATE block.
5417 __count_immobile_pages(struct zone *zone, struct page *page, int count)
5419 unsigned long pfn, iter, found;
5421 * For avoiding noise data, lru_add_drain_all() should be called
5422 * If ZONE_MOVABLE, the zone never contains immobile pages
5424 if (zone_idx(zone) == ZONE_MOVABLE)
5427 if (get_pageblock_migratetype(page) == MIGRATE_MOVABLE)
5430 pfn = page_to_pfn(page);
5431 for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) {
5432 unsigned long check = pfn + iter;
5434 if (!pfn_valid_within(check))
5437 page = pfn_to_page(check);
5438 if (!page_count(page)) {
5439 if (PageBuddy(page))
5440 iter += (1 << page_order(page)) - 1;
5446 * If there are RECLAIMABLE pages, we need to check it.
5447 * But now, memory offline itself doesn't call shrink_slab()
5448 * and it still to be fixed.
5451 * If the page is not RAM, page_count()should be 0.
5452 * we don't need more check. This is an _used_ not-movable page.
5454 * The problematic thing here is PG_reserved pages. PG_reserved
5455 * is set to both of a memory hole page and a _used_ kernel
5464 bool is_pageblock_removable_nolock(struct page *page)
5466 struct zone *zone = page_zone(page);
5467 return __count_immobile_pages(zone, page, 0);
5470 int set_migratetype_isolate(struct page *page)
5473 unsigned long flags, pfn;
5474 struct memory_isolate_notify arg;
5479 zone = page_zone(page);
5480 zone_idx = zone_idx(zone);
5482 spin_lock_irqsave(&zone->lock, flags);
5484 pfn = page_to_pfn(page);
5485 arg.start_pfn = pfn;
5486 arg.nr_pages = pageblock_nr_pages;
5487 arg.pages_found = 0;
5490 * It may be possible to isolate a pageblock even if the
5491 * migratetype is not MIGRATE_MOVABLE. The memory isolation
5492 * notifier chain is used by balloon drivers to return the
5493 * number of pages in a range that are held by the balloon
5494 * driver to shrink memory. If all the pages are accounted for
5495 * by balloons, are free, or on the LRU, isolation can continue.
5496 * Later, for example, when memory hotplug notifier runs, these
5497 * pages reported as "can be isolated" should be isolated(freed)
5498 * by the balloon driver through the memory notifier chain.
5500 notifier_ret = memory_isolate_notify(MEM_ISOLATE_COUNT, &arg);
5501 notifier_ret = notifier_to_errno(notifier_ret);
5505 * FIXME: Now, memory hotplug doesn't call shrink_slab() by itself.
5506 * We just check MOVABLE pages.
5508 if (__count_immobile_pages(zone, page, arg.pages_found))
5512 * immobile means "not-on-lru" paes. If immobile is larger than
5513 * removable-by-driver pages reported by notifier, we'll fail.
5518 set_pageblock_migratetype(page, MIGRATE_ISOLATE);
5519 move_freepages_block(zone, page, MIGRATE_ISOLATE);
5522 spin_unlock_irqrestore(&zone->lock, flags);
5528 void unset_migratetype_isolate(struct page *page)
5531 unsigned long flags;
5532 zone = page_zone(page);
5533 spin_lock_irqsave(&zone->lock, flags);
5534 if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE)
5536 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
5537 move_freepages_block(zone, page, MIGRATE_MOVABLE);
5539 spin_unlock_irqrestore(&zone->lock, flags);
5542 #ifdef CONFIG_MEMORY_HOTREMOVE
5544 * All pages in the range must be isolated before calling this.
5547 __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
5553 unsigned long flags;
5554 /* find the first valid pfn */
5555 for (pfn = start_pfn; pfn < end_pfn; pfn++)
5560 zone = page_zone(pfn_to_page(pfn));
5561 spin_lock_irqsave(&zone->lock, flags);
5563 while (pfn < end_pfn) {
5564 if (!pfn_valid(pfn)) {
5568 page = pfn_to_page(pfn);
5569 BUG_ON(page_count(page));
5570 BUG_ON(!PageBuddy(page));
5571 order = page_order(page);
5572 #ifdef CONFIG_DEBUG_VM
5573 printk(KERN_INFO "remove from free list %lx %d %lx\n",
5574 pfn, 1 << order, end_pfn);
5576 list_del(&page->lru);
5577 rmv_page_order(page);
5578 zone->free_area[order].nr_free--;
5579 __mod_zone_page_state(zone, NR_FREE_PAGES,
5581 for (i = 0; i < (1 << order); i++)
5582 SetPageReserved((page+i));
5583 pfn += (1 << order);
5585 spin_unlock_irqrestore(&zone->lock, flags);
5589 #ifdef CONFIG_MEMORY_FAILURE
5590 bool is_free_buddy_page(struct page *page)
5592 struct zone *zone = page_zone(page);
5593 unsigned long pfn = page_to_pfn(page);
5594 unsigned long flags;
5597 spin_lock_irqsave(&zone->lock, flags);
5598 for (order = 0; order < MAX_ORDER; order++) {
5599 struct page *page_head = page - (pfn & ((1 << order) - 1));
5601 if (PageBuddy(page_head) && page_order(page_head) >= order)
5604 spin_unlock_irqrestore(&zone->lock, flags);
5606 return order < MAX_ORDER;
5610 static struct trace_print_flags pageflag_names[] = {
5611 {1UL << PG_locked, "locked" },
5612 {1UL << PG_error, "error" },
5613 {1UL << PG_referenced, "referenced" },
5614 {1UL << PG_uptodate, "uptodate" },
5615 {1UL << PG_dirty, "dirty" },
5616 {1UL << PG_lru, "lru" },
5617 {1UL << PG_active, "active" },
5618 {1UL << PG_slab, "slab" },
5619 {1UL << PG_owner_priv_1, "owner_priv_1" },
5620 {1UL << PG_arch_1, "arch_1" },
5621 {1UL << PG_reserved, "reserved" },
5622 {1UL << PG_private, "private" },
5623 {1UL << PG_private_2, "private_2" },
5624 {1UL << PG_writeback, "writeback" },
5625 #ifdef CONFIG_PAGEFLAGS_EXTENDED
5626 {1UL << PG_head, "head" },
5627 {1UL << PG_tail, "tail" },
5629 {1UL << PG_compound, "compound" },
5631 {1UL << PG_swapcache, "swapcache" },
5632 {1UL << PG_mappedtodisk, "mappedtodisk" },
5633 {1UL << PG_reclaim, "reclaim" },
5634 {1UL << PG_swapbacked, "swapbacked" },
5635 {1UL << PG_unevictable, "unevictable" },
5637 {1UL << PG_mlocked, "mlocked" },
5639 #ifdef CONFIG_ARCH_USES_PG_UNCACHED
5640 {1UL << PG_uncached, "uncached" },
5642 #ifdef CONFIG_MEMORY_FAILURE
5643 {1UL << PG_hwpoison, "hwpoison" },
5648 static void dump_page_flags(unsigned long flags)
5650 const char *delim = "";
5654 printk(KERN_ALERT "page flags: %#lx(", flags);
5656 /* remove zone id */
5657 flags &= (1UL << NR_PAGEFLAGS) - 1;
5659 for (i = 0; pageflag_names[i].name && flags; i++) {
5661 mask = pageflag_names[i].mask;
5662 if ((flags & mask) != mask)
5666 printk("%s%s", delim, pageflag_names[i].name);
5670 /* check for left over flags */
5672 printk("%s%#lx", delim, flags);
5677 void dump_page(struct page *page)
5680 "page:%p count:%d mapcount:%d mapping:%p index:%#lx\n",
5681 page, atomic_read(&page->_count), page_mapcount(page),
5682 page->mapping, page->index);
5683 dump_page_flags(page->flags);