2 * linux/mm/page_alloc.c
4 * Manages the free list, the system allocates free pages here.
5 * Note that kmalloc() lives in slab.c
7 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
8 * Swap reorganised 29.12.95, Stephen Tweedie
9 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
10 * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999
11 * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
12 * Zone balancing, Kanoj Sarcar, SGI, Jan 2000
13 * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002
14 * (lots of bits borrowed from Ingo Molnar & Andrew Morton)
17 #include <linux/stddef.h>
19 #include <linux/swap.h>
20 #include <linux/interrupt.h>
21 #include <linux/pagemap.h>
22 #include <linux/jiffies.h>
23 #include <linux/bootmem.h>
24 #include <linux/memblock.h>
25 #include <linux/compiler.h>
26 #include <linux/kernel.h>
27 #include <linux/kmemcheck.h>
28 #include <linux/module.h>
29 #include <linux/suspend.h>
30 #include <linux/pagevec.h>
31 #include <linux/blkdev.h>
32 #include <linux/slab.h>
33 #include <linux/ratelimit.h>
34 #include <linux/oom.h>
35 #include <linux/notifier.h>
36 #include <linux/topology.h>
37 #include <linux/sysctl.h>
38 #include <linux/cpu.h>
39 #include <linux/cpuset.h>
40 #include <linux/memory_hotplug.h>
41 #include <linux/nodemask.h>
42 #include <linux/vmalloc.h>
43 #include <linux/vmstat.h>
44 #include <linux/mempolicy.h>
45 #include <linux/stop_machine.h>
46 #include <linux/sort.h>
47 #include <linux/pfn.h>
48 #include <linux/backing-dev.h>
49 #include <linux/fault-inject.h>
50 #include <linux/page-isolation.h>
51 #include <linux/page_cgroup.h>
52 #include <linux/debugobjects.h>
53 #include <linux/kmemleak.h>
54 #include <linux/memory.h>
55 #include <linux/compaction.h>
56 #include <trace/events/kmem.h>
57 #include <linux/ftrace_event.h>
58 #include <linux/memcontrol.h>
59 #include <linux/prefetch.h>
61 #include <asm/tlbflush.h>
62 #include <asm/div64.h>
65 #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
66 DEFINE_PER_CPU(int, numa_node);
67 EXPORT_PER_CPU_SYMBOL(numa_node);
70 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
72 * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
73 * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
74 * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
75 * defined in <linux/topology.h>.
77 DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */
78 EXPORT_PER_CPU_SYMBOL(_numa_mem_);
82 * Array of node states.
84 nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
85 [N_POSSIBLE] = NODE_MASK_ALL,
86 [N_ONLINE] = { { [0] = 1UL } },
88 [N_NORMAL_MEMORY] = { { [0] = 1UL } },
90 [N_HIGH_MEMORY] = { { [0] = 1UL } },
92 [N_CPU] = { { [0] = 1UL } },
95 EXPORT_SYMBOL(node_states);
97 unsigned long totalram_pages __read_mostly;
98 unsigned long totalreserve_pages __read_mostly;
99 int percpu_pagelist_fraction;
100 gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK;
102 #ifdef CONFIG_PM_SLEEP
104 * The following functions are used by the suspend/hibernate code to temporarily
105 * change gfp_allowed_mask in order to avoid using I/O during memory allocations
106 * while devices are suspended. To avoid races with the suspend/hibernate code,
107 * they should always be called with pm_mutex held (gfp_allowed_mask also should
108 * only be modified with pm_mutex held, unless the suspend/hibernate code is
109 * guaranteed not to run in parallel with that modification).
112 static gfp_t saved_gfp_mask;
114 void pm_restore_gfp_mask(void)
116 WARN_ON(!mutex_is_locked(&pm_mutex));
117 if (saved_gfp_mask) {
118 gfp_allowed_mask = saved_gfp_mask;
123 void pm_restrict_gfp_mask(void)
125 WARN_ON(!mutex_is_locked(&pm_mutex));
126 WARN_ON(saved_gfp_mask);
127 saved_gfp_mask = gfp_allowed_mask;
128 gfp_allowed_mask &= ~GFP_IOFS;
130 #endif /* CONFIG_PM_SLEEP */
132 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
133 int pageblock_order __read_mostly;
136 static void __free_pages_ok(struct page *page, unsigned int order);
139 * results with 256, 32 in the lowmem_reserve sysctl:
140 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
141 * 1G machine -> (16M dma, 784M normal, 224M high)
142 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
143 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
144 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
146 * TBD: should special case ZONE_DMA32 machines here - in those we normally
147 * don't need any ZONE_NORMAL reservation
149 int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
150 #ifdef CONFIG_ZONE_DMA
153 #ifdef CONFIG_ZONE_DMA32
156 #ifdef CONFIG_HIGHMEM
162 EXPORT_SYMBOL(totalram_pages);
164 static char * const zone_names[MAX_NR_ZONES] = {
165 #ifdef CONFIG_ZONE_DMA
168 #ifdef CONFIG_ZONE_DMA32
172 #ifdef CONFIG_HIGHMEM
179 * Try to keep at least this much lowmem free. Do not allow normal
180 * allocations below this point, only high priority ones. Automatically
181 * tuned according to the amount of memory in the system.
183 int min_free_kbytes = 1024;
186 * Extra memory for the system to try freeing. Used to temporarily
187 * free memory, to make space for new workloads. Anyone can allocate
188 * down to the min watermarks controlled by min_free_kbytes above.
190 int extra_free_kbytes = 0;
192 static unsigned long __meminitdata nr_kernel_pages;
193 static unsigned long __meminitdata nr_all_pages;
194 static unsigned long __meminitdata dma_reserve;
196 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
198 * MAX_ACTIVE_REGIONS determines the maximum number of distinct
199 * ranges of memory (RAM) that may be registered with add_active_range().
200 * Ranges passed to add_active_range() will be merged if possible
201 * so the number of times add_active_range() can be called is
202 * related to the number of nodes and the number of holes
204 #ifdef CONFIG_MAX_ACTIVE_REGIONS
205 /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
206 #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
208 #if MAX_NUMNODES >= 32
209 /* If there can be many nodes, allow up to 50 holes per node */
210 #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
212 /* By default, allow up to 256 distinct regions */
213 #define MAX_ACTIVE_REGIONS 256
217 static struct node_active_region __meminitdata early_node_map[MAX_ACTIVE_REGIONS];
218 static int __meminitdata nr_nodemap_entries;
219 static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
220 static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
221 static unsigned long __initdata required_kernelcore;
222 static unsigned long __initdata required_movablecore;
223 static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
225 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
227 EXPORT_SYMBOL(movable_zone);
228 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
231 int nr_node_ids __read_mostly = MAX_NUMNODES;
232 int nr_online_nodes __read_mostly = 1;
233 EXPORT_SYMBOL(nr_node_ids);
234 EXPORT_SYMBOL(nr_online_nodes);
237 int page_group_by_mobility_disabled __read_mostly;
239 static void set_pageblock_migratetype(struct page *page, int migratetype)
242 if (unlikely(page_group_by_mobility_disabled))
243 migratetype = MIGRATE_UNMOVABLE;
245 set_pageblock_flags_group(page, (unsigned long)migratetype,
246 PB_migrate, PB_migrate_end);
249 bool oom_killer_disabled __read_mostly;
251 #ifdef CONFIG_DEBUG_VM
252 static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
256 unsigned long pfn = page_to_pfn(page);
259 seq = zone_span_seqbegin(zone);
260 if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
262 else if (pfn < zone->zone_start_pfn)
264 } while (zone_span_seqretry(zone, seq));
269 static int page_is_consistent(struct zone *zone, struct page *page)
271 if (!pfn_valid_within(page_to_pfn(page)))
273 if (zone != page_zone(page))
279 * Temporary debugging check for pages not lying within a given zone.
281 static int bad_range(struct zone *zone, struct page *page)
283 if (page_outside_zone_boundaries(zone, page))
285 if (!page_is_consistent(zone, page))
291 static inline int bad_range(struct zone *zone, struct page *page)
297 static void bad_page(struct page *page)
299 static unsigned long resume;
300 static unsigned long nr_shown;
301 static unsigned long nr_unshown;
303 /* Don't complain about poisoned pages */
304 if (PageHWPoison(page)) {
305 reset_page_mapcount(page); /* remove PageBuddy */
310 * Allow a burst of 60 reports, then keep quiet for that minute;
311 * or allow a steady drip of one report per second.
313 if (nr_shown == 60) {
314 if (time_before(jiffies, resume)) {
320 "BUG: Bad page state: %lu messages suppressed\n",
327 resume = jiffies + 60 * HZ;
329 printk(KERN_ALERT "BUG: Bad page state in process %s pfn:%05lx\n",
330 current->comm, page_to_pfn(page));
336 /* Leave bad fields for debug, except PageBuddy could make trouble */
337 reset_page_mapcount(page); /* remove PageBuddy */
338 add_taint(TAINT_BAD_PAGE);
342 * Higher-order pages are called "compound pages". They are structured thusly:
344 * The first PAGE_SIZE page is called the "head page".
346 * The remaining PAGE_SIZE pages are called "tail pages".
348 * All pages have PG_compound set. All pages have their ->private pointing at
349 * the head page (even the head page has this).
351 * The first tail page's ->lru.next holds the address of the compound page's
352 * put_page() function. Its ->lru.prev holds the order of allocation.
353 * This usage means that zero-order pages may not be compound.
356 static void free_compound_page(struct page *page)
358 __free_pages_ok(page, compound_order(page));
361 void prep_compound_page(struct page *page, unsigned long order)
364 int nr_pages = 1 << order;
366 set_compound_page_dtor(page, free_compound_page);
367 set_compound_order(page, order);
369 for (i = 1; i < nr_pages; i++) {
370 struct page *p = page + i;
373 p->first_page = page;
377 /* update __split_huge_page_refcount if you change this function */
378 static int destroy_compound_page(struct page *page, unsigned long order)
381 int nr_pages = 1 << order;
384 if (unlikely(compound_order(page) != order) ||
385 unlikely(!PageHead(page))) {
390 __ClearPageHead(page);
392 for (i = 1; i < nr_pages; i++) {
393 struct page *p = page + i;
395 if (unlikely(!PageTail(p) || (p->first_page != page))) {
405 static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
410 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
411 * and __GFP_HIGHMEM from hard or soft interrupt context.
413 VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
414 for (i = 0; i < (1 << order); i++)
415 clear_highpage(page + i);
418 static inline void set_page_order(struct page *page, int order)
420 set_page_private(page, order);
421 __SetPageBuddy(page);
424 static inline void rmv_page_order(struct page *page)
426 __ClearPageBuddy(page);
427 set_page_private(page, 0);
431 * Locate the struct page for both the matching buddy in our
432 * pair (buddy1) and the combined O(n+1) page they form (page).
434 * 1) Any buddy B1 will have an order O twin B2 which satisfies
435 * the following equation:
437 * For example, if the starting buddy (buddy2) is #8 its order
439 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
441 * 2) Any buddy B will have an order O+1 parent P which
442 * satisfies the following equation:
445 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
447 static inline unsigned long
448 __find_buddy_index(unsigned long page_idx, unsigned int order)
450 return page_idx ^ (1 << order);
454 * This function checks whether a page is free && is the buddy
455 * we can do coalesce a page and its buddy if
456 * (a) the buddy is not in a hole &&
457 * (b) the buddy is in the buddy system &&
458 * (c) a page and its buddy have the same order &&
459 * (d) a page and its buddy are in the same zone.
461 * For recording whether a page is in the buddy system, we set ->_mapcount -2.
462 * Setting, clearing, and testing _mapcount -2 is serialized by zone->lock.
464 * For recording page's order, we use page_private(page).
466 static inline int page_is_buddy(struct page *page, struct page *buddy,
469 if (!pfn_valid_within(page_to_pfn(buddy)))
472 if (page_zone_id(page) != page_zone_id(buddy))
475 if (PageBuddy(buddy) && page_order(buddy) == order) {
476 VM_BUG_ON(page_count(buddy) != 0);
483 * Freeing function for a buddy system allocator.
485 * The concept of a buddy system is to maintain direct-mapped table
486 * (containing bit values) for memory blocks of various "orders".
487 * The bottom level table contains the map for the smallest allocatable
488 * units of memory (here, pages), and each level above it describes
489 * pairs of units from the levels below, hence, "buddies".
490 * At a high level, all that happens here is marking the table entry
491 * at the bottom level available, and propagating the changes upward
492 * as necessary, plus some accounting needed to play nicely with other
493 * parts of the VM system.
494 * At each level, we keep a list of pages, which are heads of continuous
495 * free pages of length of (1 << order) and marked with _mapcount -2. Page's
496 * order is recorded in page_private(page) field.
497 * So when we are allocating or freeing one, we can derive the state of the
498 * other. That is, if we allocate a small block, and both were
499 * free, the remainder of the region must be split into blocks.
500 * If a block is freed, and its buddy is also free, then this
501 * triggers coalescing into a block of larger size.
506 static inline void __free_one_page(struct page *page,
507 struct zone *zone, unsigned int order,
510 unsigned long page_idx;
511 unsigned long combined_idx;
512 unsigned long uninitialized_var(buddy_idx);
515 if (unlikely(PageCompound(page)))
516 if (unlikely(destroy_compound_page(page, order)))
519 VM_BUG_ON(migratetype == -1);
521 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
523 VM_BUG_ON(page_idx & ((1 << order) - 1));
524 VM_BUG_ON(bad_range(zone, page));
526 while (order < MAX_ORDER-1) {
527 buddy_idx = __find_buddy_index(page_idx, order);
528 buddy = page + (buddy_idx - page_idx);
529 if (!page_is_buddy(page, buddy, order))
532 /* Our buddy is free, merge with it and move up one order. */
533 list_del(&buddy->lru);
534 zone->free_area[order].nr_free--;
535 rmv_page_order(buddy);
536 combined_idx = buddy_idx & page_idx;
537 page = page + (combined_idx - page_idx);
538 page_idx = combined_idx;
541 set_page_order(page, order);
544 * If this is not the largest possible page, check if the buddy
545 * of the next-highest order is free. If it is, it's possible
546 * that pages are being freed that will coalesce soon. In case,
547 * that is happening, add the free page to the tail of the list
548 * so it's less likely to be used soon and more likely to be merged
549 * as a higher order page
551 if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) {
552 struct page *higher_page, *higher_buddy;
553 combined_idx = buddy_idx & page_idx;
554 higher_page = page + (combined_idx - page_idx);
555 buddy_idx = __find_buddy_index(combined_idx, order + 1);
556 higher_buddy = page + (buddy_idx - combined_idx);
557 if (page_is_buddy(higher_page, higher_buddy, order + 1)) {
558 list_add_tail(&page->lru,
559 &zone->free_area[order].free_list[migratetype]);
564 list_add(&page->lru, &zone->free_area[order].free_list[migratetype]);
566 zone->free_area[order].nr_free++;
570 * free_page_mlock() -- clean up attempts to free and mlocked() page.
571 * Page should not be on lru, so no need to fix that up.
572 * free_pages_check() will verify...
574 static inline void free_page_mlock(struct page *page)
576 __dec_zone_page_state(page, NR_MLOCK);
577 __count_vm_event(UNEVICTABLE_MLOCKFREED);
580 static inline int free_pages_check(struct page *page)
582 if (unlikely(page_mapcount(page) |
583 (page->mapping != NULL) |
584 (atomic_read(&page->_count) != 0) |
585 (page->flags & PAGE_FLAGS_CHECK_AT_FREE) |
586 (mem_cgroup_bad_page_check(page)))) {
590 if (page->flags & PAGE_FLAGS_CHECK_AT_PREP)
591 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
596 * Frees a number of pages from the PCP lists
597 * Assumes all pages on list are in same zone, and of same order.
598 * count is the number of pages to free.
600 * If the zone was previously in an "all pages pinned" state then look to
601 * see if this freeing clears that state.
603 * And clear the zone's pages_scanned counter, to hold off the "all pages are
604 * pinned" detection logic.
606 static void free_pcppages_bulk(struct zone *zone, int count,
607 struct per_cpu_pages *pcp)
613 spin_lock(&zone->lock);
614 zone->all_unreclaimable = 0;
615 zone->pages_scanned = 0;
619 struct list_head *list;
622 * Remove pages from lists in a round-robin fashion. A
623 * batch_free count is maintained that is incremented when an
624 * empty list is encountered. This is so more pages are freed
625 * off fuller lists instead of spinning excessively around empty
630 if (++migratetype == MIGRATE_PCPTYPES)
632 list = &pcp->lists[migratetype];
633 } while (list_empty(list));
635 /* This is the only non-empty list. Free them all. */
636 if (batch_free == MIGRATE_PCPTYPES)
637 batch_free = to_free;
640 page = list_entry(list->prev, struct page, lru);
641 /* must delete as __free_one_page list manipulates */
642 list_del(&page->lru);
643 /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */
644 __free_one_page(page, zone, 0, page_private(page));
645 trace_mm_page_pcpu_drain(page, 0, page_private(page));
646 } while (--to_free && --batch_free && !list_empty(list));
648 __mod_zone_page_state(zone, NR_FREE_PAGES, count);
649 spin_unlock(&zone->lock);
652 static void free_one_page(struct zone *zone, struct page *page, int order,
655 spin_lock(&zone->lock);
656 zone->all_unreclaimable = 0;
657 zone->pages_scanned = 0;
659 __free_one_page(page, zone, order, migratetype);
660 __mod_zone_page_state(zone, NR_FREE_PAGES, 1 << order);
661 spin_unlock(&zone->lock);
664 static bool free_pages_prepare(struct page *page, unsigned int order)
669 trace_mm_page_free_direct(page, order);
670 kmemcheck_free_shadow(page, order);
673 page->mapping = NULL;
674 for (i = 0; i < (1 << order); i++)
675 bad += free_pages_check(page + i);
679 if (!PageHighMem(page)) {
680 debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order);
681 debug_check_no_obj_freed(page_address(page),
684 arch_free_page(page, order);
685 kernel_map_pages(page, 1 << order, 0);
690 static void __free_pages_ok(struct page *page, unsigned int order)
693 int wasMlocked = __TestClearPageMlocked(page);
695 if (!free_pages_prepare(page, order))
698 local_irq_save(flags);
699 if (unlikely(wasMlocked))
700 free_page_mlock(page);
701 __count_vm_events(PGFREE, 1 << order);
702 free_one_page(page_zone(page), page, order,
703 get_pageblock_migratetype(page));
704 local_irq_restore(flags);
708 * permit the bootmem allocator to evade page validation on high-order frees
710 void __meminit __free_pages_bootmem(struct page *page, unsigned int order)
713 __ClearPageReserved(page);
714 set_page_count(page, 0);
715 set_page_refcounted(page);
721 for (loop = 0; loop < BITS_PER_LONG; loop++) {
722 struct page *p = &page[loop];
724 if (loop + 1 < BITS_PER_LONG)
726 __ClearPageReserved(p);
727 set_page_count(p, 0);
730 set_page_refcounted(page);
731 __free_pages(page, order);
737 * The order of subdivision here is critical for the IO subsystem.
738 * Please do not alter this order without good reasons and regression
739 * testing. Specifically, as large blocks of memory are subdivided,
740 * the order in which smaller blocks are delivered depends on the order
741 * they're subdivided in this function. This is the primary factor
742 * influencing the order in which pages are delivered to the IO
743 * subsystem according to empirical testing, and this is also justified
744 * by considering the behavior of a buddy system containing a single
745 * large block of memory acted on by a series of small allocations.
746 * This behavior is a critical factor in sglist merging's success.
750 static inline void expand(struct zone *zone, struct page *page,
751 int low, int high, struct free_area *area,
754 unsigned long size = 1 << high;
760 VM_BUG_ON(bad_range(zone, &page[size]));
761 list_add(&page[size].lru, &area->free_list[migratetype]);
763 set_page_order(&page[size], high);
768 * This page is about to be returned from the page allocator
770 static inline int check_new_page(struct page *page)
772 if (unlikely(page_mapcount(page) |
773 (page->mapping != NULL) |
774 (atomic_read(&page->_count) != 0) |
775 (page->flags & PAGE_FLAGS_CHECK_AT_PREP) |
776 (mem_cgroup_bad_page_check(page)))) {
783 static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
787 for (i = 0; i < (1 << order); i++) {
788 struct page *p = page + i;
789 if (unlikely(check_new_page(p)))
793 set_page_private(page, 0);
794 set_page_refcounted(page);
796 arch_alloc_page(page, order);
797 kernel_map_pages(page, 1 << order, 1);
799 if (gfp_flags & __GFP_ZERO)
800 prep_zero_page(page, order, gfp_flags);
802 if (order && (gfp_flags & __GFP_COMP))
803 prep_compound_page(page, order);
809 * Go through the free lists for the given migratetype and remove
810 * the smallest available page from the freelists
813 struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
816 unsigned int current_order;
817 struct free_area * area;
820 /* Find a page of the appropriate size in the preferred list */
821 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
822 area = &(zone->free_area[current_order]);
823 if (list_empty(&area->free_list[migratetype]))
826 page = list_entry(area->free_list[migratetype].next,
828 list_del(&page->lru);
829 rmv_page_order(page);
831 expand(zone, page, order, current_order, area, migratetype);
840 * This array describes the order lists are fallen back to when
841 * the free lists for the desirable migrate type are depleted
843 static int fallbacks[MIGRATE_TYPES][MIGRATE_TYPES-1] = {
844 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
845 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
846 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
847 [MIGRATE_RESERVE] = { MIGRATE_RESERVE, MIGRATE_RESERVE, MIGRATE_RESERVE }, /* Never used */
851 * Move the free pages in a range to the free lists of the requested type.
852 * Note that start_page and end_pages are not aligned on a pageblock
853 * boundary. If alignment is required, use move_freepages_block()
855 static int move_freepages(struct zone *zone,
856 struct page *start_page, struct page *end_page,
863 #ifndef CONFIG_HOLES_IN_ZONE
865 * page_zone is not safe to call in this context when
866 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
867 * anyway as we check zone boundaries in move_freepages_block().
868 * Remove at a later date when no bug reports exist related to
869 * grouping pages by mobility
871 BUG_ON(page_zone(start_page) != page_zone(end_page));
874 for (page = start_page; page <= end_page;) {
875 /* Make sure we are not inadvertently changing nodes */
876 VM_BUG_ON(page_to_nid(page) != zone_to_nid(zone));
878 if (!pfn_valid_within(page_to_pfn(page))) {
883 if (!PageBuddy(page)) {
888 order = page_order(page);
889 list_move(&page->lru,
890 &zone->free_area[order].free_list[migratetype]);
892 pages_moved += 1 << order;
898 static int move_freepages_block(struct zone *zone, struct page *page,
901 unsigned long start_pfn, end_pfn;
902 struct page *start_page, *end_page;
904 start_pfn = page_to_pfn(page);
905 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
906 start_page = pfn_to_page(start_pfn);
907 end_page = start_page + pageblock_nr_pages - 1;
908 end_pfn = start_pfn + pageblock_nr_pages - 1;
910 /* Do not cross zone boundaries */
911 if (start_pfn < zone->zone_start_pfn)
913 if (end_pfn >= zone->zone_start_pfn + zone->spanned_pages)
916 return move_freepages(zone, start_page, end_page, migratetype);
919 static void change_pageblock_range(struct page *pageblock_page,
920 int start_order, int migratetype)
922 int nr_pageblocks = 1 << (start_order - pageblock_order);
924 while (nr_pageblocks--) {
925 set_pageblock_migratetype(pageblock_page, migratetype);
926 pageblock_page += pageblock_nr_pages;
930 /* Remove an element from the buddy allocator from the fallback list */
931 static inline struct page *
932 __rmqueue_fallback(struct zone *zone, int order, int start_migratetype)
934 struct free_area * area;
939 /* Find the largest possible block of pages in the other list */
940 for (current_order = MAX_ORDER-1; current_order >= order;
942 for (i = 0; i < MIGRATE_TYPES - 1; i++) {
943 migratetype = fallbacks[start_migratetype][i];
945 /* MIGRATE_RESERVE handled later if necessary */
946 if (migratetype == MIGRATE_RESERVE)
949 area = &(zone->free_area[current_order]);
950 if (list_empty(&area->free_list[migratetype]))
953 page = list_entry(area->free_list[migratetype].next,
958 * If breaking a large block of pages, move all free
959 * pages to the preferred allocation list. If falling
960 * back for a reclaimable kernel allocation, be more
961 * aggressive about taking ownership of free pages
963 if (unlikely(current_order >= (pageblock_order >> 1)) ||
964 start_migratetype == MIGRATE_RECLAIMABLE ||
965 page_group_by_mobility_disabled) {
967 pages = move_freepages_block(zone, page,
970 /* Claim the whole block if over half of it is free */
971 if (pages >= (1 << (pageblock_order-1)) ||
972 page_group_by_mobility_disabled)
973 set_pageblock_migratetype(page,
976 migratetype = start_migratetype;
979 /* Remove the page from the freelists */
980 list_del(&page->lru);
981 rmv_page_order(page);
983 /* Take ownership for orders >= pageblock_order */
984 if (current_order >= pageblock_order)
985 change_pageblock_range(page, current_order,
988 expand(zone, page, order, current_order, area, migratetype);
990 trace_mm_page_alloc_extfrag(page, order, current_order,
991 start_migratetype, migratetype);
1001 * Do the hard work of removing an element from the buddy allocator.
1002 * Call me with the zone->lock already held.
1004 static struct page *__rmqueue(struct zone *zone, unsigned int order,
1010 page = __rmqueue_smallest(zone, order, migratetype);
1012 if (unlikely(!page) && migratetype != MIGRATE_RESERVE) {
1013 page = __rmqueue_fallback(zone, order, migratetype);
1016 * Use MIGRATE_RESERVE rather than fail an allocation. goto
1017 * is used because __rmqueue_smallest is an inline function
1018 * and we want just one call site
1021 migratetype = MIGRATE_RESERVE;
1026 trace_mm_page_alloc_zone_locked(page, order, migratetype);
1031 * Obtain a specified number of elements from the buddy allocator, all under
1032 * a single hold of the lock, for efficiency. Add them to the supplied list.
1033 * Returns the number of new pages which were placed at *list.
1035 static int rmqueue_bulk(struct zone *zone, unsigned int order,
1036 unsigned long count, struct list_head *list,
1037 int migratetype, int cold)
1041 spin_lock(&zone->lock);
1042 for (i = 0; i < count; ++i) {
1043 struct page *page = __rmqueue(zone, order, migratetype);
1044 if (unlikely(page == NULL))
1048 * Split buddy pages returned by expand() are received here
1049 * in physical page order. The page is added to the callers and
1050 * list and the list head then moves forward. From the callers
1051 * perspective, the linked list is ordered by page number in
1052 * some conditions. This is useful for IO devices that can
1053 * merge IO requests if the physical pages are ordered
1056 if (likely(cold == 0))
1057 list_add(&page->lru, list);
1059 list_add_tail(&page->lru, list);
1060 set_page_private(page, migratetype);
1063 __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
1064 spin_unlock(&zone->lock);
1070 * Called from the vmstat counter updater to drain pagesets of this
1071 * currently executing processor on remote nodes after they have
1074 * Note that this function must be called with the thread pinned to
1075 * a single processor.
1077 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
1079 unsigned long flags;
1082 local_irq_save(flags);
1083 if (pcp->count >= pcp->batch)
1084 to_drain = pcp->batch;
1086 to_drain = pcp->count;
1087 free_pcppages_bulk(zone, to_drain, pcp);
1088 pcp->count -= to_drain;
1089 local_irq_restore(flags);
1094 * Drain pages of the indicated processor.
1096 * The processor must either be the current processor and the
1097 * thread pinned to the current processor or a processor that
1100 static void drain_pages(unsigned int cpu)
1102 unsigned long flags;
1105 for_each_populated_zone(zone) {
1106 struct per_cpu_pageset *pset;
1107 struct per_cpu_pages *pcp;
1109 local_irq_save(flags);
1110 pset = per_cpu_ptr(zone->pageset, cpu);
1114 free_pcppages_bulk(zone, pcp->count, pcp);
1117 local_irq_restore(flags);
1122 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
1124 void drain_local_pages(void *arg)
1126 drain_pages(smp_processor_id());
1130 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
1132 void drain_all_pages(void)
1134 on_each_cpu(drain_local_pages, NULL, 1);
1137 #ifdef CONFIG_HIBERNATION
1139 void mark_free_pages(struct zone *zone)
1141 unsigned long pfn, max_zone_pfn;
1142 unsigned long flags;
1144 struct list_head *curr;
1146 if (!zone->spanned_pages)
1149 spin_lock_irqsave(&zone->lock, flags);
1151 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
1152 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1153 if (pfn_valid(pfn)) {
1154 struct page *page = pfn_to_page(pfn);
1156 if (!swsusp_page_is_forbidden(page))
1157 swsusp_unset_page_free(page);
1160 for_each_migratetype_order(order, t) {
1161 list_for_each(curr, &zone->free_area[order].free_list[t]) {
1164 pfn = page_to_pfn(list_entry(curr, struct page, lru));
1165 for (i = 0; i < (1UL << order); i++)
1166 swsusp_set_page_free(pfn_to_page(pfn + i));
1169 spin_unlock_irqrestore(&zone->lock, flags);
1171 #endif /* CONFIG_PM */
1174 * Free a 0-order page
1175 * cold == 1 ? free a cold page : free a hot page
1177 void free_hot_cold_page(struct page *page, int cold)
1179 struct zone *zone = page_zone(page);
1180 struct per_cpu_pages *pcp;
1181 unsigned long flags;
1183 int wasMlocked = __TestClearPageMlocked(page);
1185 if (!free_pages_prepare(page, 0))
1188 migratetype = get_pageblock_migratetype(page);
1189 set_page_private(page, migratetype);
1190 local_irq_save(flags);
1191 if (unlikely(wasMlocked))
1192 free_page_mlock(page);
1193 __count_vm_event(PGFREE);
1196 * We only track unmovable, reclaimable and movable on pcp lists.
1197 * Free ISOLATE pages back to the allocator because they are being
1198 * offlined but treat RESERVE as movable pages so we can get those
1199 * areas back if necessary. Otherwise, we may have to free
1200 * excessively into the page allocator
1202 if (migratetype >= MIGRATE_PCPTYPES) {
1203 if (unlikely(migratetype == MIGRATE_ISOLATE)) {
1204 free_one_page(zone, page, 0, migratetype);
1207 migratetype = MIGRATE_MOVABLE;
1210 pcp = &this_cpu_ptr(zone->pageset)->pcp;
1212 list_add_tail(&page->lru, &pcp->lists[migratetype]);
1214 list_add(&page->lru, &pcp->lists[migratetype]);
1216 if (pcp->count >= pcp->high) {
1217 free_pcppages_bulk(zone, pcp->batch, pcp);
1218 pcp->count -= pcp->batch;
1222 local_irq_restore(flags);
1225 void free_hot_cold_page_list(struct list_head *list, int cold)
1227 struct page *page, *next;
1229 list_for_each_entry_safe(page, next, list, lru) {
1230 trace_mm_pagevec_free(page, cold);
1231 free_hot_cold_page(page, cold);
1234 INIT_LIST_HEAD(list);
1238 * split_page takes a non-compound higher-order page, and splits it into
1239 * n (1<<order) sub-pages: page[0..n]
1240 * Each sub-page must be freed individually.
1242 * Note: this is probably too low level an operation for use in drivers.
1243 * Please consult with lkml before using this in your driver.
1245 void split_page(struct page *page, unsigned int order)
1249 VM_BUG_ON(PageCompound(page));
1250 VM_BUG_ON(!page_count(page));
1252 #ifdef CONFIG_KMEMCHECK
1254 * Split shadow pages too, because free(page[0]) would
1255 * otherwise free the whole shadow.
1257 if (kmemcheck_page_is_tracked(page))
1258 split_page(virt_to_page(page[0].shadow), order);
1261 for (i = 1; i < (1 << order); i++)
1262 set_page_refcounted(page + i);
1266 * Similar to split_page except the page is already free. As this is only
1267 * being used for migration, the migratetype of the block also changes.
1268 * As this is called with interrupts disabled, the caller is responsible
1269 * for calling arch_alloc_page() and kernel_map_page() after interrupts
1272 * Note: this is probably too low level an operation for use in drivers.
1273 * Please consult with lkml before using this in your driver.
1275 int split_free_page(struct page *page)
1278 unsigned long watermark;
1281 BUG_ON(!PageBuddy(page));
1283 zone = page_zone(page);
1284 order = page_order(page);
1286 /* Obey watermarks as if the page was being allocated */
1287 watermark = low_wmark_pages(zone) + (1 << order);
1288 if (!zone_watermark_ok(zone, 0, watermark, 0, 0))
1291 /* Remove page from free list */
1292 list_del(&page->lru);
1293 zone->free_area[order].nr_free--;
1294 rmv_page_order(page);
1295 __mod_zone_page_state(zone, NR_FREE_PAGES, -(1UL << order));
1297 /* Split into individual pages */
1298 set_page_refcounted(page);
1299 split_page(page, order);
1301 if (order >= pageblock_order - 1) {
1302 struct page *endpage = page + (1 << order) - 1;
1303 for (; page < endpage; page += pageblock_nr_pages)
1304 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
1311 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1312 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1316 struct page *buffered_rmqueue(struct zone *preferred_zone,
1317 struct zone *zone, int order, gfp_t gfp_flags,
1320 unsigned long flags;
1322 int cold = !!(gfp_flags & __GFP_COLD);
1325 if (likely(order == 0)) {
1326 struct per_cpu_pages *pcp;
1327 struct list_head *list;
1329 local_irq_save(flags);
1330 pcp = &this_cpu_ptr(zone->pageset)->pcp;
1331 list = &pcp->lists[migratetype];
1332 if (list_empty(list)) {
1333 pcp->count += rmqueue_bulk(zone, 0,
1336 if (unlikely(list_empty(list)))
1341 page = list_entry(list->prev, struct page, lru);
1343 page = list_entry(list->next, struct page, lru);
1345 list_del(&page->lru);
1348 if (unlikely(gfp_flags & __GFP_NOFAIL)) {
1350 * __GFP_NOFAIL is not to be used in new code.
1352 * All __GFP_NOFAIL callers should be fixed so that they
1353 * properly detect and handle allocation failures.
1355 * We most definitely don't want callers attempting to
1356 * allocate greater than order-1 page units with
1359 WARN_ON_ONCE(order > 1);
1361 spin_lock_irqsave(&zone->lock, flags);
1362 page = __rmqueue(zone, order, migratetype);
1363 spin_unlock(&zone->lock);
1366 __mod_zone_page_state(zone, NR_FREE_PAGES, -(1 << order));
1369 __count_zone_vm_events(PGALLOC, zone, 1 << order);
1370 zone_statistics(preferred_zone, zone, gfp_flags);
1371 local_irq_restore(flags);
1373 VM_BUG_ON(bad_range(zone, page));
1374 if (prep_new_page(page, order, gfp_flags))
1379 local_irq_restore(flags);
1383 /* The ALLOC_WMARK bits are used as an index to zone->watermark */
1384 #define ALLOC_WMARK_MIN WMARK_MIN
1385 #define ALLOC_WMARK_LOW WMARK_LOW
1386 #define ALLOC_WMARK_HIGH WMARK_HIGH
1387 #define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */
1389 /* Mask to get the watermark bits */
1390 #define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1)
1392 #define ALLOC_HARDER 0x10 /* try to alloc harder */
1393 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1394 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
1396 #ifdef CONFIG_FAIL_PAGE_ALLOC
1399 struct fault_attr attr;
1401 u32 ignore_gfp_highmem;
1402 u32 ignore_gfp_wait;
1404 } fail_page_alloc = {
1405 .attr = FAULT_ATTR_INITIALIZER,
1406 .ignore_gfp_wait = 1,
1407 .ignore_gfp_highmem = 1,
1411 static int __init setup_fail_page_alloc(char *str)
1413 return setup_fault_attr(&fail_page_alloc.attr, str);
1415 __setup("fail_page_alloc=", setup_fail_page_alloc);
1417 static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1419 if (order < fail_page_alloc.min_order)
1421 if (gfp_mask & __GFP_NOFAIL)
1423 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
1425 if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
1428 return should_fail(&fail_page_alloc.attr, 1 << order);
1431 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1433 static int __init fail_page_alloc_debugfs(void)
1435 mode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
1438 dir = fault_create_debugfs_attr("fail_page_alloc", NULL,
1439 &fail_page_alloc.attr);
1441 return PTR_ERR(dir);
1443 if (!debugfs_create_bool("ignore-gfp-wait", mode, dir,
1444 &fail_page_alloc.ignore_gfp_wait))
1446 if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir,
1447 &fail_page_alloc.ignore_gfp_highmem))
1449 if (!debugfs_create_u32("min-order", mode, dir,
1450 &fail_page_alloc.min_order))
1455 debugfs_remove_recursive(dir);
1460 late_initcall(fail_page_alloc_debugfs);
1462 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1464 #else /* CONFIG_FAIL_PAGE_ALLOC */
1466 static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1471 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1474 * Return true if free pages are above 'mark'. This takes into account the order
1475 * of the allocation.
1477 static bool __zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1478 int classzone_idx, int alloc_flags, long free_pages)
1480 /* free_pages my go negative - that's OK */
1484 free_pages -= (1 << order) + 1;
1485 if (alloc_flags & ALLOC_HIGH)
1487 if (alloc_flags & ALLOC_HARDER)
1490 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
1492 for (o = 0; o < order; o++) {
1493 /* At the next order, this order's pages become unavailable */
1494 free_pages -= z->free_area[o].nr_free << o;
1496 /* Require fewer higher order pages to be free */
1499 if (free_pages <= min)
1505 bool zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1506 int classzone_idx, int alloc_flags)
1508 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
1509 zone_page_state(z, NR_FREE_PAGES));
1512 bool zone_watermark_ok_safe(struct zone *z, int order, unsigned long mark,
1513 int classzone_idx, int alloc_flags)
1515 long free_pages = zone_page_state(z, NR_FREE_PAGES);
1517 if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark)
1518 free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES);
1520 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
1526 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1527 * skip over zones that are not allowed by the cpuset, or that have
1528 * been recently (in last second) found to be nearly full. See further
1529 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1530 * that have to skip over a lot of full or unallowed zones.
1532 * If the zonelist cache is present in the passed in zonelist, then
1533 * returns a pointer to the allowed node mask (either the current
1534 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
1536 * If the zonelist cache is not available for this zonelist, does
1537 * nothing and returns NULL.
1539 * If the fullzones BITMAP in the zonelist cache is stale (more than
1540 * a second since last zap'd) then we zap it out (clear its bits.)
1542 * We hold off even calling zlc_setup, until after we've checked the
1543 * first zone in the zonelist, on the theory that most allocations will
1544 * be satisfied from that first zone, so best to examine that zone as
1545 * quickly as we can.
1547 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1549 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1550 nodemask_t *allowednodes; /* zonelist_cache approximation */
1552 zlc = zonelist->zlcache_ptr;
1556 if (time_after(jiffies, zlc->last_full_zap + HZ)) {
1557 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1558 zlc->last_full_zap = jiffies;
1561 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1562 &cpuset_current_mems_allowed :
1563 &node_states[N_HIGH_MEMORY];
1564 return allowednodes;
1568 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1569 * if it is worth looking at further for free memory:
1570 * 1) Check that the zone isn't thought to be full (doesn't have its
1571 * bit set in the zonelist_cache fullzones BITMAP).
1572 * 2) Check that the zones node (obtained from the zonelist_cache
1573 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1574 * Return true (non-zero) if zone is worth looking at further, or
1575 * else return false (zero) if it is not.
1577 * This check -ignores- the distinction between various watermarks,
1578 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1579 * found to be full for any variation of these watermarks, it will
1580 * be considered full for up to one second by all requests, unless
1581 * we are so low on memory on all allowed nodes that we are forced
1582 * into the second scan of the zonelist.
1584 * In the second scan we ignore this zonelist cache and exactly
1585 * apply the watermarks to all zones, even it is slower to do so.
1586 * We are low on memory in the second scan, and should leave no stone
1587 * unturned looking for a free page.
1589 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
1590 nodemask_t *allowednodes)
1592 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1593 int i; /* index of *z in zonelist zones */
1594 int n; /* node that zone *z is on */
1596 zlc = zonelist->zlcache_ptr;
1600 i = z - zonelist->_zonerefs;
1603 /* This zone is worth trying if it is allowed but not full */
1604 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1608 * Given 'z' scanning a zonelist, set the corresponding bit in
1609 * zlc->fullzones, so that subsequent attempts to allocate a page
1610 * from that zone don't waste time re-examining it.
1612 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
1614 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1615 int i; /* index of *z in zonelist zones */
1617 zlc = zonelist->zlcache_ptr;
1621 i = z - zonelist->_zonerefs;
1623 set_bit(i, zlc->fullzones);
1627 * clear all zones full, called after direct reclaim makes progress so that
1628 * a zone that was recently full is not skipped over for up to a second
1630 static void zlc_clear_zones_full(struct zonelist *zonelist)
1632 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1634 zlc = zonelist->zlcache_ptr;
1638 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1641 #else /* CONFIG_NUMA */
1643 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1648 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
1649 nodemask_t *allowednodes)
1654 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
1658 static void zlc_clear_zones_full(struct zonelist *zonelist)
1661 #endif /* CONFIG_NUMA */
1664 * get_page_from_freelist goes through the zonelist trying to allocate
1667 static struct page *
1668 get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order,
1669 struct zonelist *zonelist, int high_zoneidx, int alloc_flags,
1670 struct zone *preferred_zone, int migratetype)
1673 struct page *page = NULL;
1676 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1677 int zlc_active = 0; /* set if using zonelist_cache */
1678 int did_zlc_setup = 0; /* just call zlc_setup() one time */
1680 classzone_idx = zone_idx(preferred_zone);
1683 * Scan zonelist, looking for a zone with enough free.
1684 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1686 for_each_zone_zonelist_nodemask(zone, z, zonelist,
1687 high_zoneidx, nodemask) {
1688 if (NUMA_BUILD && zlc_active &&
1689 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1691 if ((alloc_flags & ALLOC_CPUSET) &&
1692 !cpuset_zone_allowed_softwall(zone, gfp_mask))
1695 BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
1696 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
1700 mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
1701 if (zone_watermark_ok(zone, order, mark,
1702 classzone_idx, alloc_flags))
1705 if (NUMA_BUILD && !did_zlc_setup && nr_online_nodes > 1) {
1707 * we do zlc_setup if there are multiple nodes
1708 * and before considering the first zone allowed
1711 allowednodes = zlc_setup(zonelist, alloc_flags);
1716 if (zone_reclaim_mode == 0)
1717 goto this_zone_full;
1720 * As we may have just activated ZLC, check if the first
1721 * eligible zone has failed zone_reclaim recently.
1723 if (NUMA_BUILD && zlc_active &&
1724 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1727 ret = zone_reclaim(zone, gfp_mask, order);
1729 case ZONE_RECLAIM_NOSCAN:
1732 case ZONE_RECLAIM_FULL:
1733 /* scanned but unreclaimable */
1736 /* did we reclaim enough */
1737 if (!zone_watermark_ok(zone, order, mark,
1738 classzone_idx, alloc_flags))
1739 goto this_zone_full;
1744 page = buffered_rmqueue(preferred_zone, zone, order,
1745 gfp_mask, migratetype);
1750 zlc_mark_zone_full(zonelist, z);
1753 if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) {
1754 /* Disable zlc cache for second zonelist scan */
1762 * Large machines with many possible nodes should not always dump per-node
1763 * meminfo in irq context.
1765 static inline bool should_suppress_show_mem(void)
1770 ret = in_interrupt();
1775 static DEFINE_RATELIMIT_STATE(nopage_rs,
1776 DEFAULT_RATELIMIT_INTERVAL,
1777 DEFAULT_RATELIMIT_BURST);
1779 void warn_alloc_failed(gfp_t gfp_mask, int order, const char *fmt, ...)
1781 unsigned int filter = SHOW_MEM_FILTER_NODES;
1783 if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs))
1787 * This documents exceptions given to allocations in certain
1788 * contexts that are allowed to allocate outside current's set
1791 if (!(gfp_mask & __GFP_NOMEMALLOC))
1792 if (test_thread_flag(TIF_MEMDIE) ||
1793 (current->flags & (PF_MEMALLOC | PF_EXITING)))
1794 filter &= ~SHOW_MEM_FILTER_NODES;
1795 if (in_interrupt() || !(gfp_mask & __GFP_WAIT))
1796 filter &= ~SHOW_MEM_FILTER_NODES;
1799 struct va_format vaf;
1802 va_start(args, fmt);
1807 pr_warn("%pV", &vaf);
1812 pr_warn("%s: page allocation failure: order:%d, mode:0x%x\n",
1813 current->comm, order, gfp_mask);
1816 if (!should_suppress_show_mem())
1821 should_alloc_retry(gfp_t gfp_mask, unsigned int order,
1822 unsigned long pages_reclaimed)
1824 /* Do not loop if specifically requested */
1825 if (gfp_mask & __GFP_NORETRY)
1829 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
1830 * means __GFP_NOFAIL, but that may not be true in other
1833 if (order <= PAGE_ALLOC_COSTLY_ORDER)
1837 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
1838 * specified, then we retry until we no longer reclaim any pages
1839 * (above), or we've reclaimed an order of pages at least as
1840 * large as the allocation's order. In both cases, if the
1841 * allocation still fails, we stop retrying.
1843 if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order))
1847 * Don't let big-order allocations loop unless the caller
1848 * explicitly requests that.
1850 if (gfp_mask & __GFP_NOFAIL)
1856 static inline struct page *
1857 __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
1858 struct zonelist *zonelist, enum zone_type high_zoneidx,
1859 nodemask_t *nodemask, struct zone *preferred_zone,
1864 /* Acquire the OOM killer lock for the zones in zonelist */
1865 if (!try_set_zonelist_oom(zonelist, gfp_mask)) {
1866 schedule_timeout_uninterruptible(1);
1871 * Go through the zonelist yet one more time, keep very high watermark
1872 * here, this is only to catch a parallel oom killing, we must fail if
1873 * we're still under heavy pressure.
1875 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask,
1876 order, zonelist, high_zoneidx,
1877 ALLOC_WMARK_HIGH|ALLOC_CPUSET,
1878 preferred_zone, migratetype);
1882 if (!(gfp_mask & __GFP_NOFAIL)) {
1883 /* The OOM killer will not help higher order allocs */
1884 if (order > PAGE_ALLOC_COSTLY_ORDER)
1886 /* The OOM killer does not needlessly kill tasks for lowmem */
1887 if (high_zoneidx < ZONE_NORMAL)
1890 * GFP_THISNODE contains __GFP_NORETRY and we never hit this.
1891 * Sanity check for bare calls of __GFP_THISNODE, not real OOM.
1892 * The caller should handle page allocation failure by itself if
1893 * it specifies __GFP_THISNODE.
1894 * Note: Hugepage uses it but will hit PAGE_ALLOC_COSTLY_ORDER.
1896 if (gfp_mask & __GFP_THISNODE)
1899 /* Exhausted what can be done so it's blamo time */
1900 out_of_memory(zonelist, gfp_mask, order, nodemask);
1903 clear_zonelist_oom(zonelist, gfp_mask);
1907 #ifdef CONFIG_COMPACTION
1908 /* Try memory compaction for high-order allocations before reclaim */
1909 static struct page *
1910 __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
1911 struct zonelist *zonelist, enum zone_type high_zoneidx,
1912 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
1913 int migratetype, unsigned long *did_some_progress,
1914 bool sync_migration)
1918 if (!order || compaction_deferred(preferred_zone))
1921 current->flags |= PF_MEMALLOC;
1922 *did_some_progress = try_to_compact_pages(zonelist, order, gfp_mask,
1923 nodemask, sync_migration);
1924 current->flags &= ~PF_MEMALLOC;
1925 if (*did_some_progress != COMPACT_SKIPPED) {
1927 /* Page migration frees to the PCP lists but we want merging */
1928 drain_pages(get_cpu());
1931 page = get_page_from_freelist(gfp_mask, nodemask,
1932 order, zonelist, high_zoneidx,
1933 alloc_flags, preferred_zone,
1936 preferred_zone->compact_considered = 0;
1937 preferred_zone->compact_defer_shift = 0;
1938 count_vm_event(COMPACTSUCCESS);
1943 * It's bad if compaction run occurs and fails.
1944 * The most likely reason is that pages exist,
1945 * but not enough to satisfy watermarks.
1947 count_vm_event(COMPACTFAIL);
1948 defer_compaction(preferred_zone);
1956 static inline struct page *
1957 __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
1958 struct zonelist *zonelist, enum zone_type high_zoneidx,
1959 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
1960 int migratetype, unsigned long *did_some_progress,
1961 bool sync_migration)
1965 #endif /* CONFIG_COMPACTION */
1967 /* The really slow allocator path where we enter direct reclaim */
1968 static inline struct page *
1969 __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
1970 struct zonelist *zonelist, enum zone_type high_zoneidx,
1971 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
1972 int migratetype, unsigned long *did_some_progress)
1974 struct page *page = NULL;
1975 struct reclaim_state reclaim_state;
1976 bool drained = false;
1980 /* We now go into synchronous reclaim */
1981 cpuset_memory_pressure_bump();
1982 current->flags |= PF_MEMALLOC;
1983 lockdep_set_current_reclaim_state(gfp_mask);
1984 reclaim_state.reclaimed_slab = 0;
1985 current->reclaim_state = &reclaim_state;
1987 *did_some_progress = try_to_free_pages(zonelist, order, gfp_mask, nodemask);
1989 current->reclaim_state = NULL;
1990 lockdep_clear_current_reclaim_state();
1991 current->flags &= ~PF_MEMALLOC;
1995 if (unlikely(!(*did_some_progress)))
1998 /* After successful reclaim, reconsider all zones for allocation */
2000 zlc_clear_zones_full(zonelist);
2003 page = get_page_from_freelist(gfp_mask, nodemask, order,
2004 zonelist, high_zoneidx,
2005 alloc_flags, preferred_zone,
2009 * If an allocation failed after direct reclaim, it could be because
2010 * pages are pinned on the per-cpu lists. Drain them and try again
2012 if (!page && !drained) {
2022 * This is called in the allocator slow-path if the allocation request is of
2023 * sufficient urgency to ignore watermarks and take other desperate measures
2025 static inline struct page *
2026 __alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order,
2027 struct zonelist *zonelist, enum zone_type high_zoneidx,
2028 nodemask_t *nodemask, struct zone *preferred_zone,
2034 page = get_page_from_freelist(gfp_mask, nodemask, order,
2035 zonelist, high_zoneidx, ALLOC_NO_WATERMARKS,
2036 preferred_zone, migratetype);
2038 if (!page && gfp_mask & __GFP_NOFAIL)
2039 wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50);
2040 } while (!page && (gfp_mask & __GFP_NOFAIL));
2046 void wake_all_kswapd(unsigned int order, struct zonelist *zonelist,
2047 enum zone_type high_zoneidx,
2048 enum zone_type classzone_idx)
2053 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx)
2054 wakeup_kswapd(zone, order, classzone_idx);
2058 gfp_to_alloc_flags(gfp_t gfp_mask)
2060 int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
2061 const gfp_t wait = gfp_mask & __GFP_WAIT;
2063 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
2064 BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
2067 * The caller may dip into page reserves a bit more if the caller
2068 * cannot run direct reclaim, or if the caller has realtime scheduling
2069 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
2070 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
2072 alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH);
2076 * Not worth trying to allocate harder for
2077 * __GFP_NOMEMALLOC even if it can't schedule.
2079 if (!(gfp_mask & __GFP_NOMEMALLOC))
2080 alloc_flags |= ALLOC_HARDER;
2082 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
2083 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
2085 alloc_flags &= ~ALLOC_CPUSET;
2086 } else if (unlikely(rt_task(current)) && !in_interrupt())
2087 alloc_flags |= ALLOC_HARDER;
2089 if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) {
2090 if (!in_interrupt() &&
2091 ((current->flags & PF_MEMALLOC) ||
2092 unlikely(test_thread_flag(TIF_MEMDIE))))
2093 alloc_flags |= ALLOC_NO_WATERMARKS;
2099 static inline struct page *
2100 __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
2101 struct zonelist *zonelist, enum zone_type high_zoneidx,
2102 nodemask_t *nodemask, struct zone *preferred_zone,
2105 const gfp_t wait = gfp_mask & __GFP_WAIT;
2106 struct page *page = NULL;
2108 unsigned long pages_reclaimed = 0;
2109 unsigned long did_some_progress;
2110 bool sync_migration = false;
2113 * In the slowpath, we sanity check order to avoid ever trying to
2114 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
2115 * be using allocators in order of preference for an area that is
2118 if (order >= MAX_ORDER) {
2119 WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
2124 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
2125 * __GFP_NOWARN set) should not cause reclaim since the subsystem
2126 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
2127 * using a larger set of nodes after it has established that the
2128 * allowed per node queues are empty and that nodes are
2131 if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
2135 if (!(gfp_mask & __GFP_NO_KSWAPD))
2136 wake_all_kswapd(order, zonelist, high_zoneidx,
2137 zone_idx(preferred_zone));
2140 * OK, we're below the kswapd watermark and have kicked background
2141 * reclaim. Now things get more complex, so set up alloc_flags according
2142 * to how we want to proceed.
2144 alloc_flags = gfp_to_alloc_flags(gfp_mask);
2147 * Find the true preferred zone if the allocation is unconstrained by
2150 if (!(alloc_flags & ALLOC_CPUSET) && !nodemask)
2151 first_zones_zonelist(zonelist, high_zoneidx, NULL,
2155 /* This is the last chance, in general, before the goto nopage. */
2156 page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist,
2157 high_zoneidx, alloc_flags & ~ALLOC_NO_WATERMARKS,
2158 preferred_zone, migratetype);
2162 /* Allocate without watermarks if the context allows */
2163 if (alloc_flags & ALLOC_NO_WATERMARKS) {
2164 page = __alloc_pages_high_priority(gfp_mask, order,
2165 zonelist, high_zoneidx, nodemask,
2166 preferred_zone, migratetype);
2171 /* Atomic allocations - we can't balance anything */
2175 /* Avoid recursion of direct reclaim */
2176 if (current->flags & PF_MEMALLOC)
2179 /* Avoid allocations with no watermarks from looping endlessly */
2180 if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL))
2184 * Try direct compaction. The first pass is asynchronous. Subsequent
2185 * attempts after direct reclaim are synchronous
2187 page = __alloc_pages_direct_compact(gfp_mask, order,
2188 zonelist, high_zoneidx,
2190 alloc_flags, preferred_zone,
2191 migratetype, &did_some_progress,
2195 sync_migration = true;
2197 /* Try direct reclaim and then allocating */
2198 page = __alloc_pages_direct_reclaim(gfp_mask, order,
2199 zonelist, high_zoneidx,
2201 alloc_flags, preferred_zone,
2202 migratetype, &did_some_progress);
2207 * If we failed to make any progress reclaiming, then we are
2208 * running out of options and have to consider going OOM
2210 if (!did_some_progress) {
2211 if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
2212 if (oom_killer_disabled)
2214 page = __alloc_pages_may_oom(gfp_mask, order,
2215 zonelist, high_zoneidx,
2216 nodemask, preferred_zone,
2221 if (!(gfp_mask & __GFP_NOFAIL)) {
2223 * The oom killer is not called for high-order
2224 * allocations that may fail, so if no progress
2225 * is being made, there are no other options and
2226 * retrying is unlikely to help.
2228 if (order > PAGE_ALLOC_COSTLY_ORDER)
2231 * The oom killer is not called for lowmem
2232 * allocations to prevent needlessly killing
2235 if (high_zoneidx < ZONE_NORMAL)
2243 /* Check if we should retry the allocation */
2244 pages_reclaimed += did_some_progress;
2245 if (should_alloc_retry(gfp_mask, order, pages_reclaimed)) {
2246 /* Wait for some write requests to complete then retry */
2247 wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50);
2251 * High-order allocations do not necessarily loop after
2252 * direct reclaim and reclaim/compaction depends on compaction
2253 * being called after reclaim so call directly if necessary
2255 page = __alloc_pages_direct_compact(gfp_mask, order,
2256 zonelist, high_zoneidx,
2258 alloc_flags, preferred_zone,
2259 migratetype, &did_some_progress,
2266 warn_alloc_failed(gfp_mask, order, NULL);
2269 if (kmemcheck_enabled)
2270 kmemcheck_pagealloc_alloc(page, order, gfp_mask);
2276 * This is the 'heart' of the zoned buddy allocator.
2279 __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
2280 struct zonelist *zonelist, nodemask_t *nodemask)
2282 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
2283 struct zone *preferred_zone;
2285 int migratetype = allocflags_to_migratetype(gfp_mask);
2287 gfp_mask &= gfp_allowed_mask;
2289 lockdep_trace_alloc(gfp_mask);
2291 might_sleep_if(gfp_mask & __GFP_WAIT);
2293 if (should_fail_alloc_page(gfp_mask, order))
2297 * Check the zones suitable for the gfp_mask contain at least one
2298 * valid zone. It's possible to have an empty zonelist as a result
2299 * of GFP_THISNODE and a memoryless node
2301 if (unlikely(!zonelist->_zonerefs->zone))
2305 /* The preferred zone is used for statistics later */
2306 first_zones_zonelist(zonelist, high_zoneidx,
2307 nodemask ? : &cpuset_current_mems_allowed,
2309 if (!preferred_zone) {
2314 /* First allocation attempt */
2315 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order,
2316 zonelist, high_zoneidx, ALLOC_WMARK_LOW|ALLOC_CPUSET,
2317 preferred_zone, migratetype);
2318 if (unlikely(!page))
2319 page = __alloc_pages_slowpath(gfp_mask, order,
2320 zonelist, high_zoneidx, nodemask,
2321 preferred_zone, migratetype);
2324 trace_mm_page_alloc(page, order, gfp_mask, migratetype);
2327 EXPORT_SYMBOL(__alloc_pages_nodemask);
2330 * Common helper functions.
2332 unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
2337 * __get_free_pages() returns a 32-bit address, which cannot represent
2340 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
2342 page = alloc_pages(gfp_mask, order);
2345 return (unsigned long) page_address(page);
2347 EXPORT_SYMBOL(__get_free_pages);
2349 unsigned long get_zeroed_page(gfp_t gfp_mask)
2351 return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
2353 EXPORT_SYMBOL(get_zeroed_page);
2355 void __pagevec_free(struct pagevec *pvec)
2357 int i = pagevec_count(pvec);
2360 trace_mm_pagevec_free(pvec->pages[i], pvec->cold);
2361 free_hot_cold_page(pvec->pages[i], pvec->cold);
2365 void __free_pages(struct page *page, unsigned int order)
2367 if (put_page_testzero(page)) {
2369 free_hot_cold_page(page, 0);
2371 __free_pages_ok(page, order);
2375 EXPORT_SYMBOL(__free_pages);
2377 void free_pages(unsigned long addr, unsigned int order)
2380 VM_BUG_ON(!virt_addr_valid((void *)addr));
2381 __free_pages(virt_to_page((void *)addr), order);
2385 EXPORT_SYMBOL(free_pages);
2387 static void *make_alloc_exact(unsigned long addr, unsigned order, size_t size)
2390 unsigned long alloc_end = addr + (PAGE_SIZE << order);
2391 unsigned long used = addr + PAGE_ALIGN(size);
2393 split_page(virt_to_page((void *)addr), order);
2394 while (used < alloc_end) {
2399 return (void *)addr;
2403 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
2404 * @size: the number of bytes to allocate
2405 * @gfp_mask: GFP flags for the allocation
2407 * This function is similar to alloc_pages(), except that it allocates the
2408 * minimum number of pages to satisfy the request. alloc_pages() can only
2409 * allocate memory in power-of-two pages.
2411 * This function is also limited by MAX_ORDER.
2413 * Memory allocated by this function must be released by free_pages_exact().
2415 void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
2417 unsigned int order = get_order(size);
2420 addr = __get_free_pages(gfp_mask, order);
2421 return make_alloc_exact(addr, order, size);
2423 EXPORT_SYMBOL(alloc_pages_exact);
2426 * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
2428 * @nid: the preferred node ID where memory should be allocated
2429 * @size: the number of bytes to allocate
2430 * @gfp_mask: GFP flags for the allocation
2432 * Like alloc_pages_exact(), but try to allocate on node nid first before falling
2434 * Note this is not alloc_pages_exact_node() which allocates on a specific node,
2437 void *alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask)
2439 unsigned order = get_order(size);
2440 struct page *p = alloc_pages_node(nid, gfp_mask, order);
2443 return make_alloc_exact((unsigned long)page_address(p), order, size);
2445 EXPORT_SYMBOL(alloc_pages_exact_nid);
2448 * free_pages_exact - release memory allocated via alloc_pages_exact()
2449 * @virt: the value returned by alloc_pages_exact.
2450 * @size: size of allocation, same value as passed to alloc_pages_exact().
2452 * Release the memory allocated by a previous call to alloc_pages_exact.
2454 void free_pages_exact(void *virt, size_t size)
2456 unsigned long addr = (unsigned long)virt;
2457 unsigned long end = addr + PAGE_ALIGN(size);
2459 while (addr < end) {
2464 EXPORT_SYMBOL(free_pages_exact);
2466 static unsigned int nr_free_zone_pages(int offset)
2471 /* Just pick one node, since fallback list is circular */
2472 unsigned int sum = 0;
2474 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
2476 for_each_zone_zonelist(zone, z, zonelist, offset) {
2477 unsigned long size = zone->present_pages;
2478 unsigned long high = high_wmark_pages(zone);
2487 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
2489 unsigned int nr_free_buffer_pages(void)
2491 return nr_free_zone_pages(gfp_zone(GFP_USER));
2493 EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
2496 * Amount of free RAM allocatable within all zones
2498 unsigned int nr_free_pagecache_pages(void)
2500 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
2503 static inline void show_node(struct zone *zone)
2506 printk("Node %d ", zone_to_nid(zone));
2509 void si_meminfo(struct sysinfo *val)
2511 val->totalram = totalram_pages;
2513 val->freeram = global_page_state(NR_FREE_PAGES);
2514 val->bufferram = nr_blockdev_pages();
2515 val->totalhigh = totalhigh_pages;
2516 val->freehigh = nr_free_highpages();
2517 val->mem_unit = PAGE_SIZE;
2520 EXPORT_SYMBOL(si_meminfo);
2523 void si_meminfo_node(struct sysinfo *val, int nid)
2525 pg_data_t *pgdat = NODE_DATA(nid);
2527 val->totalram = pgdat->node_present_pages;
2528 val->freeram = node_page_state(nid, NR_FREE_PAGES);
2529 #ifdef CONFIG_HIGHMEM
2530 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
2531 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
2537 val->mem_unit = PAGE_SIZE;
2542 * Determine whether the node should be displayed or not, depending on whether
2543 * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
2545 bool skip_free_areas_node(unsigned int flags, int nid)
2549 if (!(flags & SHOW_MEM_FILTER_NODES))
2553 ret = !node_isset(nid, cpuset_current_mems_allowed);
2559 #define K(x) ((x) << (PAGE_SHIFT-10))
2562 * Show free area list (used inside shift_scroll-lock stuff)
2563 * We also calculate the percentage fragmentation. We do this by counting the
2564 * memory on each free list with the exception of the first item on the list.
2565 * Suppresses nodes that are not allowed by current's cpuset if
2566 * SHOW_MEM_FILTER_NODES is passed.
2568 void show_free_areas(unsigned int filter)
2573 for_each_populated_zone(zone) {
2574 if (skip_free_areas_node(filter, zone_to_nid(zone)))
2577 printk("%s per-cpu:\n", zone->name);
2579 for_each_online_cpu(cpu) {
2580 struct per_cpu_pageset *pageset;
2582 pageset = per_cpu_ptr(zone->pageset, cpu);
2584 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
2585 cpu, pageset->pcp.high,
2586 pageset->pcp.batch, pageset->pcp.count);
2590 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
2591 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
2593 " dirty:%lu writeback:%lu unstable:%lu\n"
2594 " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n"
2595 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n",
2596 global_page_state(NR_ACTIVE_ANON),
2597 global_page_state(NR_INACTIVE_ANON),
2598 global_page_state(NR_ISOLATED_ANON),
2599 global_page_state(NR_ACTIVE_FILE),
2600 global_page_state(NR_INACTIVE_FILE),
2601 global_page_state(NR_ISOLATED_FILE),
2602 global_page_state(NR_UNEVICTABLE),
2603 global_page_state(NR_FILE_DIRTY),
2604 global_page_state(NR_WRITEBACK),
2605 global_page_state(NR_UNSTABLE_NFS),
2606 global_page_state(NR_FREE_PAGES),
2607 global_page_state(NR_SLAB_RECLAIMABLE),
2608 global_page_state(NR_SLAB_UNRECLAIMABLE),
2609 global_page_state(NR_FILE_MAPPED),
2610 global_page_state(NR_SHMEM),
2611 global_page_state(NR_PAGETABLE),
2612 global_page_state(NR_BOUNCE));
2614 for_each_populated_zone(zone) {
2617 if (skip_free_areas_node(filter, zone_to_nid(zone)))
2625 " active_anon:%lukB"
2626 " inactive_anon:%lukB"
2627 " active_file:%lukB"
2628 " inactive_file:%lukB"
2629 " unevictable:%lukB"
2630 " isolated(anon):%lukB"
2631 " isolated(file):%lukB"
2638 " slab_reclaimable:%lukB"
2639 " slab_unreclaimable:%lukB"
2640 " kernel_stack:%lukB"
2644 " writeback_tmp:%lukB"
2645 " pages_scanned:%lu"
2646 " all_unreclaimable? %s"
2649 K(zone_page_state(zone, NR_FREE_PAGES)),
2650 K(min_wmark_pages(zone)),
2651 K(low_wmark_pages(zone)),
2652 K(high_wmark_pages(zone)),
2653 K(zone_page_state(zone, NR_ACTIVE_ANON)),
2654 K(zone_page_state(zone, NR_INACTIVE_ANON)),
2655 K(zone_page_state(zone, NR_ACTIVE_FILE)),
2656 K(zone_page_state(zone, NR_INACTIVE_FILE)),
2657 K(zone_page_state(zone, NR_UNEVICTABLE)),
2658 K(zone_page_state(zone, NR_ISOLATED_ANON)),
2659 K(zone_page_state(zone, NR_ISOLATED_FILE)),
2660 K(zone->present_pages),
2661 K(zone_page_state(zone, NR_MLOCK)),
2662 K(zone_page_state(zone, NR_FILE_DIRTY)),
2663 K(zone_page_state(zone, NR_WRITEBACK)),
2664 K(zone_page_state(zone, NR_FILE_MAPPED)),
2665 K(zone_page_state(zone, NR_SHMEM)),
2666 K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)),
2667 K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)),
2668 zone_page_state(zone, NR_KERNEL_STACK) *
2670 K(zone_page_state(zone, NR_PAGETABLE)),
2671 K(zone_page_state(zone, NR_UNSTABLE_NFS)),
2672 K(zone_page_state(zone, NR_BOUNCE)),
2673 K(zone_page_state(zone, NR_WRITEBACK_TEMP)),
2674 zone->pages_scanned,
2675 (zone->all_unreclaimable ? "yes" : "no")
2677 printk("lowmem_reserve[]:");
2678 for (i = 0; i < MAX_NR_ZONES; i++)
2679 printk(" %lu", zone->lowmem_reserve[i]);
2683 for_each_populated_zone(zone) {
2684 unsigned long nr[MAX_ORDER], flags, order, total = 0;
2686 if (skip_free_areas_node(filter, zone_to_nid(zone)))
2689 printk("%s: ", zone->name);
2691 spin_lock_irqsave(&zone->lock, flags);
2692 for (order = 0; order < MAX_ORDER; order++) {
2693 nr[order] = zone->free_area[order].nr_free;
2694 total += nr[order] << order;
2696 spin_unlock_irqrestore(&zone->lock, flags);
2697 for (order = 0; order < MAX_ORDER; order++)
2698 printk("%lu*%lukB ", nr[order], K(1UL) << order);
2699 printk("= %lukB\n", K(total));
2702 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
2704 show_swap_cache_info();
2707 static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
2709 zoneref->zone = zone;
2710 zoneref->zone_idx = zone_idx(zone);
2714 * Builds allocation fallback zone lists.
2716 * Add all populated zones of a node to the zonelist.
2718 static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
2719 int nr_zones, enum zone_type zone_type)
2723 BUG_ON(zone_type >= MAX_NR_ZONES);
2728 zone = pgdat->node_zones + zone_type;
2729 if (populated_zone(zone)) {
2730 zoneref_set_zone(zone,
2731 &zonelist->_zonerefs[nr_zones++]);
2732 check_highest_zone(zone_type);
2735 } while (zone_type);
2742 * 0 = automatic detection of better ordering.
2743 * 1 = order by ([node] distance, -zonetype)
2744 * 2 = order by (-zonetype, [node] distance)
2746 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
2747 * the same zonelist. So only NUMA can configure this param.
2749 #define ZONELIST_ORDER_DEFAULT 0
2750 #define ZONELIST_ORDER_NODE 1
2751 #define ZONELIST_ORDER_ZONE 2
2753 /* zonelist order in the kernel.
2754 * set_zonelist_order() will set this to NODE or ZONE.
2756 static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
2757 static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
2761 /* The value user specified ....changed by config */
2762 static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2763 /* string for sysctl */
2764 #define NUMA_ZONELIST_ORDER_LEN 16
2765 char numa_zonelist_order[16] = "default";
2768 * interface for configure zonelist ordering.
2769 * command line option "numa_zonelist_order"
2770 * = "[dD]efault - default, automatic configuration.
2771 * = "[nN]ode - order by node locality, then by zone within node
2772 * = "[zZ]one - order by zone, then by locality within zone
2775 static int __parse_numa_zonelist_order(char *s)
2777 if (*s == 'd' || *s == 'D') {
2778 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2779 } else if (*s == 'n' || *s == 'N') {
2780 user_zonelist_order = ZONELIST_ORDER_NODE;
2781 } else if (*s == 'z' || *s == 'Z') {
2782 user_zonelist_order = ZONELIST_ORDER_ZONE;
2785 "Ignoring invalid numa_zonelist_order value: "
2792 static __init int setup_numa_zonelist_order(char *s)
2799 ret = __parse_numa_zonelist_order(s);
2801 strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN);
2805 early_param("numa_zonelist_order", setup_numa_zonelist_order);
2808 * sysctl handler for numa_zonelist_order
2810 int numa_zonelist_order_handler(ctl_table *table, int write,
2811 void __user *buffer, size_t *length,
2814 char saved_string[NUMA_ZONELIST_ORDER_LEN];
2816 static DEFINE_MUTEX(zl_order_mutex);
2818 mutex_lock(&zl_order_mutex);
2820 strcpy(saved_string, (char*)table->data);
2821 ret = proc_dostring(table, write, buffer, length, ppos);
2825 int oldval = user_zonelist_order;
2826 if (__parse_numa_zonelist_order((char*)table->data)) {
2828 * bogus value. restore saved string
2830 strncpy((char*)table->data, saved_string,
2831 NUMA_ZONELIST_ORDER_LEN);
2832 user_zonelist_order = oldval;
2833 } else if (oldval != user_zonelist_order) {
2834 mutex_lock(&zonelists_mutex);
2835 build_all_zonelists(NULL);
2836 mutex_unlock(&zonelists_mutex);
2840 mutex_unlock(&zl_order_mutex);
2845 #define MAX_NODE_LOAD (nr_online_nodes)
2846 static int node_load[MAX_NUMNODES];
2849 * find_next_best_node - find the next node that should appear in a given node's fallback list
2850 * @node: node whose fallback list we're appending
2851 * @used_node_mask: nodemask_t of already used nodes
2853 * We use a number of factors to determine which is the next node that should
2854 * appear on a given node's fallback list. The node should not have appeared
2855 * already in @node's fallback list, and it should be the next closest node
2856 * according to the distance array (which contains arbitrary distance values
2857 * from each node to each node in the system), and should also prefer nodes
2858 * with no CPUs, since presumably they'll have very little allocation pressure
2859 * on them otherwise.
2860 * It returns -1 if no node is found.
2862 static int find_next_best_node(int node, nodemask_t *used_node_mask)
2865 int min_val = INT_MAX;
2867 const struct cpumask *tmp = cpumask_of_node(0);
2869 /* Use the local node if we haven't already */
2870 if (!node_isset(node, *used_node_mask)) {
2871 node_set(node, *used_node_mask);
2875 for_each_node_state(n, N_HIGH_MEMORY) {
2877 /* Don't want a node to appear more than once */
2878 if (node_isset(n, *used_node_mask))
2881 /* Use the distance array to find the distance */
2882 val = node_distance(node, n);
2884 /* Penalize nodes under us ("prefer the next node") */
2887 /* Give preference to headless and unused nodes */
2888 tmp = cpumask_of_node(n);
2889 if (!cpumask_empty(tmp))
2890 val += PENALTY_FOR_NODE_WITH_CPUS;
2892 /* Slight preference for less loaded node */
2893 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
2894 val += node_load[n];
2896 if (val < min_val) {
2903 node_set(best_node, *used_node_mask);
2910 * Build zonelists ordered by node and zones within node.
2911 * This results in maximum locality--normal zone overflows into local
2912 * DMA zone, if any--but risks exhausting DMA zone.
2914 static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
2917 struct zonelist *zonelist;
2919 zonelist = &pgdat->node_zonelists[0];
2920 for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
2922 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2924 zonelist->_zonerefs[j].zone = NULL;
2925 zonelist->_zonerefs[j].zone_idx = 0;
2929 * Build gfp_thisnode zonelists
2931 static void build_thisnode_zonelists(pg_data_t *pgdat)
2934 struct zonelist *zonelist;
2936 zonelist = &pgdat->node_zonelists[1];
2937 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
2938 zonelist->_zonerefs[j].zone = NULL;
2939 zonelist->_zonerefs[j].zone_idx = 0;
2943 * Build zonelists ordered by zone and nodes within zones.
2944 * This results in conserving DMA zone[s] until all Normal memory is
2945 * exhausted, but results in overflowing to remote node while memory
2946 * may still exist in local DMA zone.
2948 static int node_order[MAX_NUMNODES];
2950 static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
2953 int zone_type; /* needs to be signed */
2955 struct zonelist *zonelist;
2957 zonelist = &pgdat->node_zonelists[0];
2959 for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
2960 for (j = 0; j < nr_nodes; j++) {
2961 node = node_order[j];
2962 z = &NODE_DATA(node)->node_zones[zone_type];
2963 if (populated_zone(z)) {
2965 &zonelist->_zonerefs[pos++]);
2966 check_highest_zone(zone_type);
2970 zonelist->_zonerefs[pos].zone = NULL;
2971 zonelist->_zonerefs[pos].zone_idx = 0;
2974 static int default_zonelist_order(void)
2977 unsigned long low_kmem_size,total_size;
2981 * ZONE_DMA and ZONE_DMA32 can be very small area in the system.
2982 * If they are really small and used heavily, the system can fall
2983 * into OOM very easily.
2984 * This function detect ZONE_DMA/DMA32 size and configures zone order.
2986 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2989 for_each_online_node(nid) {
2990 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2991 z = &NODE_DATA(nid)->node_zones[zone_type];
2992 if (populated_zone(z)) {
2993 if (zone_type < ZONE_NORMAL)
2994 low_kmem_size += z->present_pages;
2995 total_size += z->present_pages;
2996 } else if (zone_type == ZONE_NORMAL) {
2998 * If any node has only lowmem, then node order
2999 * is preferred to allow kernel allocations
3000 * locally; otherwise, they can easily infringe
3001 * on other nodes when there is an abundance of
3002 * lowmem available to allocate from.
3004 return ZONELIST_ORDER_NODE;
3008 if (!low_kmem_size || /* there are no DMA area. */
3009 low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
3010 return ZONELIST_ORDER_NODE;
3012 * look into each node's config.
3013 * If there is a node whose DMA/DMA32 memory is very big area on
3014 * local memory, NODE_ORDER may be suitable.
3016 average_size = total_size /
3017 (nodes_weight(node_states[N_HIGH_MEMORY]) + 1);
3018 for_each_online_node(nid) {
3021 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
3022 z = &NODE_DATA(nid)->node_zones[zone_type];
3023 if (populated_zone(z)) {
3024 if (zone_type < ZONE_NORMAL)
3025 low_kmem_size += z->present_pages;
3026 total_size += z->present_pages;
3029 if (low_kmem_size &&
3030 total_size > average_size && /* ignore small node */
3031 low_kmem_size > total_size * 70/100)
3032 return ZONELIST_ORDER_NODE;
3034 return ZONELIST_ORDER_ZONE;
3037 static void set_zonelist_order(void)
3039 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
3040 current_zonelist_order = default_zonelist_order();
3042 current_zonelist_order = user_zonelist_order;
3045 static void build_zonelists(pg_data_t *pgdat)
3049 nodemask_t used_mask;
3050 int local_node, prev_node;
3051 struct zonelist *zonelist;
3052 int order = current_zonelist_order;
3054 /* initialize zonelists */
3055 for (i = 0; i < MAX_ZONELISTS; i++) {
3056 zonelist = pgdat->node_zonelists + i;
3057 zonelist->_zonerefs[0].zone = NULL;
3058 zonelist->_zonerefs[0].zone_idx = 0;
3061 /* NUMA-aware ordering of nodes */
3062 local_node = pgdat->node_id;
3063 load = nr_online_nodes;
3064 prev_node = local_node;
3065 nodes_clear(used_mask);
3067 memset(node_order, 0, sizeof(node_order));
3070 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
3071 int distance = node_distance(local_node, node);
3074 * If another node is sufficiently far away then it is better
3075 * to reclaim pages in a zone before going off node.
3077 if (distance > RECLAIM_DISTANCE)
3078 zone_reclaim_mode = 1;
3081 * We don't want to pressure a particular node.
3082 * So adding penalty to the first node in same
3083 * distance group to make it round-robin.
3085 if (distance != node_distance(local_node, prev_node))
3086 node_load[node] = load;
3090 if (order == ZONELIST_ORDER_NODE)
3091 build_zonelists_in_node_order(pgdat, node);
3093 node_order[j++] = node; /* remember order */
3096 if (order == ZONELIST_ORDER_ZONE) {
3097 /* calculate node order -- i.e., DMA last! */
3098 build_zonelists_in_zone_order(pgdat, j);
3101 build_thisnode_zonelists(pgdat);
3104 /* Construct the zonelist performance cache - see further mmzone.h */
3105 static void build_zonelist_cache(pg_data_t *pgdat)
3107 struct zonelist *zonelist;
3108 struct zonelist_cache *zlc;
3111 zonelist = &pgdat->node_zonelists[0];
3112 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
3113 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
3114 for (z = zonelist->_zonerefs; z->zone; z++)
3115 zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z);
3118 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
3120 * Return node id of node used for "local" allocations.
3121 * I.e., first node id of first zone in arg node's generic zonelist.
3122 * Used for initializing percpu 'numa_mem', which is used primarily
3123 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
3125 int local_memory_node(int node)
3129 (void)first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
3130 gfp_zone(GFP_KERNEL),
3137 #else /* CONFIG_NUMA */
3139 static void set_zonelist_order(void)
3141 current_zonelist_order = ZONELIST_ORDER_ZONE;
3144 static void build_zonelists(pg_data_t *pgdat)
3146 int node, local_node;
3148 struct zonelist *zonelist;
3150 local_node = pgdat->node_id;
3152 zonelist = &pgdat->node_zonelists[0];
3153 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
3156 * Now we build the zonelist so that it contains the zones
3157 * of all the other nodes.
3158 * We don't want to pressure a particular node, so when
3159 * building the zones for node N, we make sure that the
3160 * zones coming right after the local ones are those from
3161 * node N+1 (modulo N)
3163 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
3164 if (!node_online(node))
3166 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
3169 for (node = 0; node < local_node; node++) {
3170 if (!node_online(node))
3172 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
3176 zonelist->_zonerefs[j].zone = NULL;
3177 zonelist->_zonerefs[j].zone_idx = 0;
3180 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
3181 static void build_zonelist_cache(pg_data_t *pgdat)
3183 pgdat->node_zonelists[0].zlcache_ptr = NULL;
3186 #endif /* CONFIG_NUMA */
3189 * Boot pageset table. One per cpu which is going to be used for all
3190 * zones and all nodes. The parameters will be set in such a way
3191 * that an item put on a list will immediately be handed over to
3192 * the buddy list. This is safe since pageset manipulation is done
3193 * with interrupts disabled.
3195 * The boot_pagesets must be kept even after bootup is complete for
3196 * unused processors and/or zones. They do play a role for bootstrapping
3197 * hotplugged processors.
3199 * zoneinfo_show() and maybe other functions do
3200 * not check if the processor is online before following the pageset pointer.
3201 * Other parts of the kernel may not check if the zone is available.
3203 static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch);
3204 static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset);
3205 static void setup_zone_pageset(struct zone *zone);
3208 * Global mutex to protect against size modification of zonelists
3209 * as well as to serialize pageset setup for the new populated zone.
3211 DEFINE_MUTEX(zonelists_mutex);
3213 /* return values int ....just for stop_machine() */
3214 static __init_refok int __build_all_zonelists(void *data)
3220 memset(node_load, 0, sizeof(node_load));
3222 for_each_online_node(nid) {
3223 pg_data_t *pgdat = NODE_DATA(nid);
3225 build_zonelists(pgdat);
3226 build_zonelist_cache(pgdat);
3230 * Initialize the boot_pagesets that are going to be used
3231 * for bootstrapping processors. The real pagesets for
3232 * each zone will be allocated later when the per cpu
3233 * allocator is available.
3235 * boot_pagesets are used also for bootstrapping offline
3236 * cpus if the system is already booted because the pagesets
3237 * are needed to initialize allocators on a specific cpu too.
3238 * F.e. the percpu allocator needs the page allocator which
3239 * needs the percpu allocator in order to allocate its pagesets
3240 * (a chicken-egg dilemma).
3242 for_each_possible_cpu(cpu) {
3243 setup_pageset(&per_cpu(boot_pageset, cpu), 0);
3245 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
3247 * We now know the "local memory node" for each node--
3248 * i.e., the node of the first zone in the generic zonelist.
3249 * Set up numa_mem percpu variable for on-line cpus. During
3250 * boot, only the boot cpu should be on-line; we'll init the
3251 * secondary cpus' numa_mem as they come on-line. During
3252 * node/memory hotplug, we'll fixup all on-line cpus.
3254 if (cpu_online(cpu))
3255 set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
3263 * Called with zonelists_mutex held always
3264 * unless system_state == SYSTEM_BOOTING.
3266 void __ref build_all_zonelists(void *data)
3268 set_zonelist_order();
3270 if (system_state == SYSTEM_BOOTING) {
3271 __build_all_zonelists(NULL);
3272 mminit_verify_zonelist();
3273 cpuset_init_current_mems_allowed();
3275 /* we have to stop all cpus to guarantee there is no user
3277 #ifdef CONFIG_MEMORY_HOTPLUG
3279 setup_zone_pageset((struct zone *)data);
3281 stop_machine(__build_all_zonelists, NULL, NULL);
3282 /* cpuset refresh routine should be here */
3284 vm_total_pages = nr_free_pagecache_pages();
3286 * Disable grouping by mobility if the number of pages in the
3287 * system is too low to allow the mechanism to work. It would be
3288 * more accurate, but expensive to check per-zone. This check is
3289 * made on memory-hotadd so a system can start with mobility
3290 * disabled and enable it later
3292 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
3293 page_group_by_mobility_disabled = 1;
3295 page_group_by_mobility_disabled = 0;
3297 printk("Built %i zonelists in %s order, mobility grouping %s. "
3298 "Total pages: %ld\n",
3300 zonelist_order_name[current_zonelist_order],
3301 page_group_by_mobility_disabled ? "off" : "on",
3304 printk("Policy zone: %s\n", zone_names[policy_zone]);
3309 * Helper functions to size the waitqueue hash table.
3310 * Essentially these want to choose hash table sizes sufficiently
3311 * large so that collisions trying to wait on pages are rare.
3312 * But in fact, the number of active page waitqueues on typical
3313 * systems is ridiculously low, less than 200. So this is even
3314 * conservative, even though it seems large.
3316 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
3317 * waitqueues, i.e. the size of the waitq table given the number of pages.
3319 #define PAGES_PER_WAITQUEUE 256
3321 #ifndef CONFIG_MEMORY_HOTPLUG
3322 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
3324 unsigned long size = 1;
3326 pages /= PAGES_PER_WAITQUEUE;
3328 while (size < pages)
3332 * Once we have dozens or even hundreds of threads sleeping
3333 * on IO we've got bigger problems than wait queue collision.
3334 * Limit the size of the wait table to a reasonable size.
3336 size = min(size, 4096UL);
3338 return max(size, 4UL);
3342 * A zone's size might be changed by hot-add, so it is not possible to determine
3343 * a suitable size for its wait_table. So we use the maximum size now.
3345 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
3347 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
3348 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
3349 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
3351 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
3352 * or more by the traditional way. (See above). It equals:
3354 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
3355 * ia64(16K page size) : = ( 8G + 4M)byte.
3356 * powerpc (64K page size) : = (32G +16M)byte.
3358 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
3365 * This is an integer logarithm so that shifts can be used later
3366 * to extract the more random high bits from the multiplicative
3367 * hash function before the remainder is taken.
3369 static inline unsigned long wait_table_bits(unsigned long size)
3374 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
3377 * Check if a pageblock contains reserved pages
3379 static int pageblock_is_reserved(unsigned long start_pfn, unsigned long end_pfn)
3383 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
3384 if (!pfn_valid_within(pfn) || PageReserved(pfn_to_page(pfn)))
3391 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
3392 * of blocks reserved is based on min_wmark_pages(zone). The memory within
3393 * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
3394 * higher will lead to a bigger reserve which will get freed as contiguous
3395 * blocks as reclaim kicks in
3397 static void setup_zone_migrate_reserve(struct zone *zone)
3399 unsigned long start_pfn, pfn, end_pfn, block_end_pfn;
3401 unsigned long block_migratetype;
3404 /* Get the start pfn, end pfn and the number of blocks to reserve */
3405 start_pfn = zone->zone_start_pfn;
3406 end_pfn = start_pfn + zone->spanned_pages;
3407 reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >>
3411 * Reserve blocks are generally in place to help high-order atomic
3412 * allocations that are short-lived. A min_free_kbytes value that
3413 * would result in more than 2 reserve blocks for atomic allocations
3414 * is assumed to be in place to help anti-fragmentation for the
3415 * future allocation of hugepages at runtime.
3417 reserve = min(2, reserve);
3419 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
3420 if (!pfn_valid(pfn))
3422 page = pfn_to_page(pfn);
3424 /* Watch out for overlapping nodes */
3425 if (page_to_nid(page) != zone_to_nid(zone))
3428 /* Blocks with reserved pages will never free, skip them. */
3429 block_end_pfn = min(pfn + pageblock_nr_pages, end_pfn);
3430 if (pageblock_is_reserved(pfn, block_end_pfn))
3433 block_migratetype = get_pageblock_migratetype(page);
3435 /* If this block is reserved, account for it */
3436 if (reserve > 0 && block_migratetype == MIGRATE_RESERVE) {
3441 /* Suitable for reserving if this block is movable */
3442 if (reserve > 0 && block_migratetype == MIGRATE_MOVABLE) {
3443 set_pageblock_migratetype(page, MIGRATE_RESERVE);
3444 move_freepages_block(zone, page, MIGRATE_RESERVE);
3450 * If the reserve is met and this is a previous reserved block,
3453 if (block_migratetype == MIGRATE_RESERVE) {
3454 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
3455 move_freepages_block(zone, page, MIGRATE_MOVABLE);
3461 * Initially all pages are reserved - free ones are freed
3462 * up by free_all_bootmem() once the early boot process is
3463 * done. Non-atomic initialization, single-pass.
3465 void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
3466 unsigned long start_pfn, enum memmap_context context)
3469 unsigned long end_pfn = start_pfn + size;
3473 if (highest_memmap_pfn < end_pfn - 1)
3474 highest_memmap_pfn = end_pfn - 1;
3476 z = &NODE_DATA(nid)->node_zones[zone];
3477 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
3479 * There can be holes in boot-time mem_map[]s
3480 * handed to this function. They do not
3481 * exist on hotplugged memory.
3483 if (context == MEMMAP_EARLY) {
3484 if (!early_pfn_valid(pfn))
3486 if (!early_pfn_in_nid(pfn, nid))
3489 page = pfn_to_page(pfn);
3490 set_page_links(page, zone, nid, pfn);
3491 mminit_verify_page_links(page, zone, nid, pfn);
3492 init_page_count(page);
3493 reset_page_mapcount(page);
3494 SetPageReserved(page);
3496 * Mark the block movable so that blocks are reserved for
3497 * movable at startup. This will force kernel allocations
3498 * to reserve their blocks rather than leaking throughout
3499 * the address space during boot when many long-lived
3500 * kernel allocations are made. Later some blocks near
3501 * the start are marked MIGRATE_RESERVE by
3502 * setup_zone_migrate_reserve()
3504 * bitmap is created for zone's valid pfn range. but memmap
3505 * can be created for invalid pages (for alignment)
3506 * check here not to call set_pageblock_migratetype() against
3509 if ((z->zone_start_pfn <= pfn)
3510 && (pfn < z->zone_start_pfn + z->spanned_pages)
3511 && !(pfn & (pageblock_nr_pages - 1)))
3512 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
3514 INIT_LIST_HEAD(&page->lru);
3515 #ifdef WANT_PAGE_VIRTUAL
3516 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
3517 if (!is_highmem_idx(zone))
3518 set_page_address(page, __va(pfn << PAGE_SHIFT));
3523 static void __meminit zone_init_free_lists(struct zone *zone)
3526 for_each_migratetype_order(order, t) {
3527 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
3528 zone->free_area[order].nr_free = 0;
3532 #ifndef __HAVE_ARCH_MEMMAP_INIT
3533 #define memmap_init(size, nid, zone, start_pfn) \
3534 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
3537 static int zone_batchsize(struct zone *zone)
3543 * The per-cpu-pages pools are set to around 1000th of the
3544 * size of the zone. But no more than 1/2 of a meg.
3546 * OK, so we don't know how big the cache is. So guess.
3548 batch = zone->present_pages / 1024;
3549 if (batch * PAGE_SIZE > 512 * 1024)
3550 batch = (512 * 1024) / PAGE_SIZE;
3551 batch /= 4; /* We effectively *= 4 below */
3556 * Clamp the batch to a 2^n - 1 value. Having a power
3557 * of 2 value was found to be more likely to have
3558 * suboptimal cache aliasing properties in some cases.
3560 * For example if 2 tasks are alternately allocating
3561 * batches of pages, one task can end up with a lot
3562 * of pages of one half of the possible page colors
3563 * and the other with pages of the other colors.
3565 batch = rounddown_pow_of_two(batch + batch/2) - 1;
3570 /* The deferral and batching of frees should be suppressed under NOMMU
3573 * The problem is that NOMMU needs to be able to allocate large chunks
3574 * of contiguous memory as there's no hardware page translation to
3575 * assemble apparent contiguous memory from discontiguous pages.
3577 * Queueing large contiguous runs of pages for batching, however,
3578 * causes the pages to actually be freed in smaller chunks. As there
3579 * can be a significant delay between the individual batches being
3580 * recycled, this leads to the once large chunks of space being
3581 * fragmented and becoming unavailable for high-order allocations.
3587 static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
3589 struct per_cpu_pages *pcp;
3592 memset(p, 0, sizeof(*p));
3596 pcp->high = 6 * batch;
3597 pcp->batch = max(1UL, 1 * batch);
3598 for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
3599 INIT_LIST_HEAD(&pcp->lists[migratetype]);
3603 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
3604 * to the value high for the pageset p.
3607 static void setup_pagelist_highmark(struct per_cpu_pageset *p,
3610 struct per_cpu_pages *pcp;
3614 pcp->batch = max(1UL, high/4);
3615 if ((high/4) > (PAGE_SHIFT * 8))
3616 pcp->batch = PAGE_SHIFT * 8;
3619 static void setup_zone_pageset(struct zone *zone)
3623 zone->pageset = alloc_percpu(struct per_cpu_pageset);
3625 for_each_possible_cpu(cpu) {
3626 struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu);
3628 setup_pageset(pcp, zone_batchsize(zone));
3630 if (percpu_pagelist_fraction)
3631 setup_pagelist_highmark(pcp,
3632 (zone->present_pages /
3633 percpu_pagelist_fraction));
3638 * Allocate per cpu pagesets and initialize them.
3639 * Before this call only boot pagesets were available.
3641 void __init setup_per_cpu_pageset(void)
3645 for_each_populated_zone(zone)
3646 setup_zone_pageset(zone);
3649 static noinline __init_refok
3650 int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
3653 struct pglist_data *pgdat = zone->zone_pgdat;
3657 * The per-page waitqueue mechanism uses hashed waitqueues
3660 zone->wait_table_hash_nr_entries =
3661 wait_table_hash_nr_entries(zone_size_pages);
3662 zone->wait_table_bits =
3663 wait_table_bits(zone->wait_table_hash_nr_entries);
3664 alloc_size = zone->wait_table_hash_nr_entries
3665 * sizeof(wait_queue_head_t);
3667 if (!slab_is_available()) {
3668 zone->wait_table = (wait_queue_head_t *)
3669 alloc_bootmem_node_nopanic(pgdat, alloc_size);
3672 * This case means that a zone whose size was 0 gets new memory
3673 * via memory hot-add.
3674 * But it may be the case that a new node was hot-added. In
3675 * this case vmalloc() will not be able to use this new node's
3676 * memory - this wait_table must be initialized to use this new
3677 * node itself as well.
3678 * To use this new node's memory, further consideration will be
3681 zone->wait_table = vmalloc(alloc_size);
3683 if (!zone->wait_table)
3686 for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
3687 init_waitqueue_head(zone->wait_table + i);
3692 static int __zone_pcp_update(void *data)
3694 struct zone *zone = data;
3696 unsigned long batch = zone_batchsize(zone), flags;
3698 for_each_possible_cpu(cpu) {
3699 struct per_cpu_pageset *pset;
3700 struct per_cpu_pages *pcp;
3702 pset = per_cpu_ptr(zone->pageset, cpu);
3705 local_irq_save(flags);
3706 free_pcppages_bulk(zone, pcp->count, pcp);
3707 setup_pageset(pset, batch);
3708 local_irq_restore(flags);
3713 void zone_pcp_update(struct zone *zone)
3715 stop_machine(__zone_pcp_update, zone, NULL);
3718 static __meminit void zone_pcp_init(struct zone *zone)
3721 * per cpu subsystem is not up at this point. The following code
3722 * relies on the ability of the linker to provide the
3723 * offset of a (static) per cpu variable into the per cpu area.
3725 zone->pageset = &boot_pageset;
3727 if (zone->present_pages)
3728 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n",
3729 zone->name, zone->present_pages,
3730 zone_batchsize(zone));
3733 __meminit int init_currently_empty_zone(struct zone *zone,
3734 unsigned long zone_start_pfn,
3736 enum memmap_context context)
3738 struct pglist_data *pgdat = zone->zone_pgdat;
3740 ret = zone_wait_table_init(zone, size);
3743 pgdat->nr_zones = zone_idx(zone) + 1;
3745 zone->zone_start_pfn = zone_start_pfn;
3747 mminit_dprintk(MMINIT_TRACE, "memmap_init",
3748 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
3750 (unsigned long)zone_idx(zone),
3751 zone_start_pfn, (zone_start_pfn + size));
3753 zone_init_free_lists(zone);
3758 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3760 * Basic iterator support. Return the first range of PFNs for a node
3761 * Note: nid == MAX_NUMNODES returns first region regardless of node
3763 static int __meminit first_active_region_index_in_nid(int nid)
3767 for (i = 0; i < nr_nodemap_entries; i++)
3768 if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
3775 * Basic iterator support. Return the next active range of PFNs for a node
3776 * Note: nid == MAX_NUMNODES returns next region regardless of node
3778 static int __meminit next_active_region_index_in_nid(int index, int nid)
3780 for (index = index + 1; index < nr_nodemap_entries; index++)
3781 if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
3787 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
3789 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
3790 * Architectures may implement their own version but if add_active_range()
3791 * was used and there are no special requirements, this is a convenient
3794 int __meminit __early_pfn_to_nid(unsigned long pfn)
3798 for (i = 0; i < nr_nodemap_entries; i++) {
3799 unsigned long start_pfn = early_node_map[i].start_pfn;
3800 unsigned long end_pfn = early_node_map[i].end_pfn;
3802 if (start_pfn <= pfn && pfn < end_pfn)
3803 return early_node_map[i].nid;
3805 /* This is a memory hole */
3808 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
3810 int __meminit early_pfn_to_nid(unsigned long pfn)
3814 nid = __early_pfn_to_nid(pfn);
3817 /* just returns 0 */
3821 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
3822 bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
3826 nid = __early_pfn_to_nid(pfn);
3827 if (nid >= 0 && nid != node)
3833 /* Basic iterator support to walk early_node_map[] */
3834 #define for_each_active_range_index_in_nid(i, nid) \
3835 for (i = first_active_region_index_in_nid(nid); i != -1; \
3836 i = next_active_region_index_in_nid(i, nid))
3839 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
3840 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
3841 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
3843 * If an architecture guarantees that all ranges registered with
3844 * add_active_ranges() contain no holes and may be freed, this
3845 * this function may be used instead of calling free_bootmem() manually.
3847 void __init free_bootmem_with_active_regions(int nid,
3848 unsigned long max_low_pfn)
3852 for_each_active_range_index_in_nid(i, nid) {
3853 unsigned long size_pages = 0;
3854 unsigned long end_pfn = early_node_map[i].end_pfn;
3856 if (early_node_map[i].start_pfn >= max_low_pfn)
3859 if (end_pfn > max_low_pfn)
3860 end_pfn = max_low_pfn;
3862 size_pages = end_pfn - early_node_map[i].start_pfn;
3863 free_bootmem_node(NODE_DATA(early_node_map[i].nid),
3864 PFN_PHYS(early_node_map[i].start_pfn),
3865 size_pages << PAGE_SHIFT);
3869 #ifdef CONFIG_HAVE_MEMBLOCK
3871 * Basic iterator support. Return the last range of PFNs for a node
3872 * Note: nid == MAX_NUMNODES returns last region regardless of node
3874 static int __meminit last_active_region_index_in_nid(int nid)
3878 for (i = nr_nodemap_entries - 1; i >= 0; i--)
3879 if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
3886 * Basic iterator support. Return the previous active range of PFNs for a node
3887 * Note: nid == MAX_NUMNODES returns next region regardless of node
3889 static int __meminit previous_active_region_index_in_nid(int index, int nid)
3891 for (index = index - 1; index >= 0; index--)
3892 if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
3898 #define for_each_active_range_index_in_nid_reverse(i, nid) \
3899 for (i = last_active_region_index_in_nid(nid); i != -1; \
3900 i = previous_active_region_index_in_nid(i, nid))
3902 u64 __init find_memory_core_early(int nid, u64 size, u64 align,
3903 u64 goal, u64 limit)
3907 /* Need to go over early_node_map to find out good range for node */
3908 for_each_active_range_index_in_nid_reverse(i, nid) {
3910 u64 ei_start, ei_last;
3911 u64 final_start, final_end;
3913 ei_last = early_node_map[i].end_pfn;
3914 ei_last <<= PAGE_SHIFT;
3915 ei_start = early_node_map[i].start_pfn;
3916 ei_start <<= PAGE_SHIFT;
3918 final_start = max(ei_start, goal);
3919 final_end = min(ei_last, limit);
3921 if (final_start >= final_end)
3924 addr = memblock_find_in_range(final_start, final_end, size, align);
3926 if (addr == MEMBLOCK_ERROR)
3932 return MEMBLOCK_ERROR;
3936 int __init add_from_early_node_map(struct range *range, int az,
3937 int nr_range, int nid)
3942 /* need to go over early_node_map to find out good range for node */
3943 for_each_active_range_index_in_nid(i, nid) {
3944 start = early_node_map[i].start_pfn;
3945 end = early_node_map[i].end_pfn;
3946 nr_range = add_range(range, az, nr_range, start, end);
3951 void __init work_with_active_regions(int nid, work_fn_t work_fn, void *data)
3956 for_each_active_range_index_in_nid(i, nid) {
3957 ret = work_fn(early_node_map[i].start_pfn,
3958 early_node_map[i].end_pfn, data);
3964 * sparse_memory_present_with_active_regions - Call memory_present for each active range
3965 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
3967 * If an architecture guarantees that all ranges registered with
3968 * add_active_ranges() contain no holes and may be freed, this
3969 * function may be used instead of calling memory_present() manually.
3971 void __init sparse_memory_present_with_active_regions(int nid)
3975 for_each_active_range_index_in_nid(i, nid)
3976 memory_present(early_node_map[i].nid,
3977 early_node_map[i].start_pfn,
3978 early_node_map[i].end_pfn);
3982 * get_pfn_range_for_nid - Return the start and end page frames for a node
3983 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
3984 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
3985 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
3987 * It returns the start and end page frame of a node based on information
3988 * provided by an arch calling add_active_range(). If called for a node
3989 * with no available memory, a warning is printed and the start and end
3992 void __meminit get_pfn_range_for_nid(unsigned int nid,
3993 unsigned long *start_pfn, unsigned long *end_pfn)
3999 for_each_active_range_index_in_nid(i, nid) {
4000 *start_pfn = min(*start_pfn, early_node_map[i].start_pfn);
4001 *end_pfn = max(*end_pfn, early_node_map[i].end_pfn);
4004 if (*start_pfn == -1UL)
4009 * This finds a zone that can be used for ZONE_MOVABLE pages. The
4010 * assumption is made that zones within a node are ordered in monotonic
4011 * increasing memory addresses so that the "highest" populated zone is used
4013 static void __init find_usable_zone_for_movable(void)
4016 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
4017 if (zone_index == ZONE_MOVABLE)
4020 if (arch_zone_highest_possible_pfn[zone_index] >
4021 arch_zone_lowest_possible_pfn[zone_index])
4025 VM_BUG_ON(zone_index == -1);
4026 movable_zone = zone_index;
4030 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
4031 * because it is sized independent of architecture. Unlike the other zones,
4032 * the starting point for ZONE_MOVABLE is not fixed. It may be different
4033 * in each node depending on the size of each node and how evenly kernelcore
4034 * is distributed. This helper function adjusts the zone ranges
4035 * provided by the architecture for a given node by using the end of the
4036 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
4037 * zones within a node are in order of monotonic increases memory addresses
4039 static void __meminit adjust_zone_range_for_zone_movable(int nid,
4040 unsigned long zone_type,
4041 unsigned long node_start_pfn,
4042 unsigned long node_end_pfn,
4043 unsigned long *zone_start_pfn,
4044 unsigned long *zone_end_pfn)
4046 /* Only adjust if ZONE_MOVABLE is on this node */
4047 if (zone_movable_pfn[nid]) {
4048 /* Size ZONE_MOVABLE */
4049 if (zone_type == ZONE_MOVABLE) {
4050 *zone_start_pfn = zone_movable_pfn[nid];
4051 *zone_end_pfn = min(node_end_pfn,
4052 arch_zone_highest_possible_pfn[movable_zone]);
4054 /* Adjust for ZONE_MOVABLE starting within this range */
4055 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
4056 *zone_end_pfn > zone_movable_pfn[nid]) {
4057 *zone_end_pfn = zone_movable_pfn[nid];
4059 /* Check if this whole range is within ZONE_MOVABLE */
4060 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
4061 *zone_start_pfn = *zone_end_pfn;
4066 * Return the number of pages a zone spans in a node, including holes
4067 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
4069 static unsigned long __meminit zone_spanned_pages_in_node(int nid,
4070 unsigned long zone_type,
4071 unsigned long *ignored)
4073 unsigned long node_start_pfn, node_end_pfn;
4074 unsigned long zone_start_pfn, zone_end_pfn;
4076 /* Get the start and end of the node and zone */
4077 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
4078 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
4079 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
4080 adjust_zone_range_for_zone_movable(nid, zone_type,
4081 node_start_pfn, node_end_pfn,
4082 &zone_start_pfn, &zone_end_pfn);
4084 /* Check that this node has pages within the zone's required range */
4085 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
4088 /* Move the zone boundaries inside the node if necessary */
4089 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
4090 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
4092 /* Return the spanned pages */
4093 return zone_end_pfn - zone_start_pfn;
4097 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
4098 * then all holes in the requested range will be accounted for.
4100 unsigned long __meminit __absent_pages_in_range(int nid,
4101 unsigned long range_start_pfn,
4102 unsigned long range_end_pfn)
4105 unsigned long prev_end_pfn = 0, hole_pages = 0;
4106 unsigned long start_pfn;
4108 /* Find the end_pfn of the first active range of pfns in the node */
4109 i = first_active_region_index_in_nid(nid);
4113 prev_end_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
4115 /* Account for ranges before physical memory on this node */
4116 if (early_node_map[i].start_pfn > range_start_pfn)
4117 hole_pages = prev_end_pfn - range_start_pfn;
4119 /* Find all holes for the zone within the node */
4120 for (; i != -1; i = next_active_region_index_in_nid(i, nid)) {
4122 /* No need to continue if prev_end_pfn is outside the zone */
4123 if (prev_end_pfn >= range_end_pfn)
4126 /* Make sure the end of the zone is not within the hole */
4127 start_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
4128 prev_end_pfn = max(prev_end_pfn, range_start_pfn);
4130 /* Update the hole size cound and move on */
4131 if (start_pfn > range_start_pfn) {
4132 BUG_ON(prev_end_pfn > start_pfn);
4133 hole_pages += start_pfn - prev_end_pfn;
4135 prev_end_pfn = early_node_map[i].end_pfn;
4138 /* Account for ranges past physical memory on this node */
4139 if (range_end_pfn > prev_end_pfn)
4140 hole_pages += range_end_pfn -
4141 max(range_start_pfn, prev_end_pfn);
4147 * absent_pages_in_range - Return number of page frames in holes within a range
4148 * @start_pfn: The start PFN to start searching for holes
4149 * @end_pfn: The end PFN to stop searching for holes
4151 * It returns the number of pages frames in memory holes within a range.
4153 unsigned long __init absent_pages_in_range(unsigned long start_pfn,
4154 unsigned long end_pfn)
4156 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
4159 /* Return the number of page frames in holes in a zone on a node */
4160 static unsigned long __meminit zone_absent_pages_in_node(int nid,
4161 unsigned long zone_type,
4162 unsigned long *ignored)
4164 unsigned long node_start_pfn, node_end_pfn;
4165 unsigned long zone_start_pfn, zone_end_pfn;
4167 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
4168 zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type],
4170 zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type],
4173 adjust_zone_range_for_zone_movable(nid, zone_type,
4174 node_start_pfn, node_end_pfn,
4175 &zone_start_pfn, &zone_end_pfn);
4176 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
4180 static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
4181 unsigned long zone_type,
4182 unsigned long *zones_size)
4184 return zones_size[zone_type];
4187 static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
4188 unsigned long zone_type,
4189 unsigned long *zholes_size)
4194 return zholes_size[zone_type];
4199 static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
4200 unsigned long *zones_size, unsigned long *zholes_size)
4202 unsigned long realtotalpages, totalpages = 0;
4205 for (i = 0; i < MAX_NR_ZONES; i++)
4206 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
4208 pgdat->node_spanned_pages = totalpages;
4210 realtotalpages = totalpages;
4211 for (i = 0; i < MAX_NR_ZONES; i++)
4213 zone_absent_pages_in_node(pgdat->node_id, i,
4215 pgdat->node_present_pages = realtotalpages;
4216 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
4220 #ifndef CONFIG_SPARSEMEM
4222 * Calculate the size of the zone->blockflags rounded to an unsigned long
4223 * Start by making sure zonesize is a multiple of pageblock_order by rounding
4224 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
4225 * round what is now in bits to nearest long in bits, then return it in
4228 static unsigned long __init usemap_size(unsigned long zonesize)
4230 unsigned long usemapsize;
4232 usemapsize = roundup(zonesize, pageblock_nr_pages);
4233 usemapsize = usemapsize >> pageblock_order;
4234 usemapsize *= NR_PAGEBLOCK_BITS;
4235 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
4237 return usemapsize / 8;
4240 static void __init setup_usemap(struct pglist_data *pgdat,
4241 struct zone *zone, unsigned long zonesize)
4243 unsigned long usemapsize = usemap_size(zonesize);
4244 zone->pageblock_flags = NULL;
4246 zone->pageblock_flags = alloc_bootmem_node_nopanic(pgdat,
4250 static inline void setup_usemap(struct pglist_data *pgdat,
4251 struct zone *zone, unsigned long zonesize) {}
4252 #endif /* CONFIG_SPARSEMEM */
4254 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
4256 /* Return a sensible default order for the pageblock size. */
4257 static inline int pageblock_default_order(void)
4259 if (HPAGE_SHIFT > PAGE_SHIFT)
4260 return HUGETLB_PAGE_ORDER;
4265 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
4266 static inline void __init set_pageblock_order(unsigned int order)
4268 /* Check that pageblock_nr_pages has not already been setup */
4269 if (pageblock_order)
4273 * Assume the largest contiguous order of interest is a huge page.
4274 * This value may be variable depending on boot parameters on IA64
4276 pageblock_order = order;
4278 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4281 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
4282 * and pageblock_default_order() are unused as pageblock_order is set
4283 * at compile-time. See include/linux/pageblock-flags.h for the values of
4284 * pageblock_order based on the kernel config
4286 static inline int pageblock_default_order(unsigned int order)
4290 #define set_pageblock_order(x) do {} while (0)
4292 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4295 * Set up the zone data structures:
4296 * - mark all pages reserved
4297 * - mark all memory queues empty
4298 * - clear the memory bitmaps
4300 static void __paginginit free_area_init_core(struct pglist_data *pgdat,
4301 unsigned long *zones_size, unsigned long *zholes_size)
4304 int nid = pgdat->node_id;
4305 unsigned long zone_start_pfn = pgdat->node_start_pfn;
4308 pgdat_resize_init(pgdat);
4309 pgdat->nr_zones = 0;
4310 init_waitqueue_head(&pgdat->kswapd_wait);
4311 pgdat->kswapd_max_order = 0;
4312 pgdat_page_cgroup_init(pgdat);
4314 for (j = 0; j < MAX_NR_ZONES; j++) {
4315 struct zone *zone = pgdat->node_zones + j;
4316 unsigned long size, realsize, memmap_pages;
4319 size = zone_spanned_pages_in_node(nid, j, zones_size);
4320 realsize = size - zone_absent_pages_in_node(nid, j,
4324 * Adjust realsize so that it accounts for how much memory
4325 * is used by this zone for memmap. This affects the watermark
4326 * and per-cpu initialisations
4329 PAGE_ALIGN(size * sizeof(struct page)) >> PAGE_SHIFT;
4330 if (realsize >= memmap_pages) {
4331 realsize -= memmap_pages;
4334 " %s zone: %lu pages used for memmap\n",
4335 zone_names[j], memmap_pages);
4338 " %s zone: %lu pages exceeds realsize %lu\n",
4339 zone_names[j], memmap_pages, realsize);
4341 /* Account for reserved pages */
4342 if (j == 0 && realsize > dma_reserve) {
4343 realsize -= dma_reserve;
4344 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
4345 zone_names[0], dma_reserve);
4348 if (!is_highmem_idx(j))
4349 nr_kernel_pages += realsize;
4350 nr_all_pages += realsize;
4352 zone->spanned_pages = size;
4353 zone->present_pages = realsize;
4356 zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
4358 zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100;
4360 zone->name = zone_names[j];
4361 spin_lock_init(&zone->lock);
4362 spin_lock_init(&zone->lru_lock);
4363 zone_seqlock_init(zone);
4364 zone->zone_pgdat = pgdat;
4366 zone_pcp_init(zone);
4368 INIT_LIST_HEAD(&zone->lru[l].list);
4369 zone->reclaim_stat.recent_rotated[0] = 0;
4370 zone->reclaim_stat.recent_rotated[1] = 0;
4371 zone->reclaim_stat.recent_scanned[0] = 0;
4372 zone->reclaim_stat.recent_scanned[1] = 0;
4373 zap_zone_vm_stats(zone);
4378 set_pageblock_order(pageblock_default_order());
4379 setup_usemap(pgdat, zone, size);
4380 ret = init_currently_empty_zone(zone, zone_start_pfn,
4381 size, MEMMAP_EARLY);
4383 memmap_init(size, nid, j, zone_start_pfn);
4384 zone_start_pfn += size;
4388 static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
4390 /* Skip empty nodes */
4391 if (!pgdat->node_spanned_pages)
4394 #ifdef CONFIG_FLAT_NODE_MEM_MAP
4395 /* ia64 gets its own node_mem_map, before this, without bootmem */
4396 if (!pgdat->node_mem_map) {
4397 unsigned long size, start, end;
4401 * The zone's endpoints aren't required to be MAX_ORDER
4402 * aligned but the node_mem_map endpoints must be in order
4403 * for the buddy allocator to function correctly.
4405 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
4406 end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
4407 end = ALIGN(end, MAX_ORDER_NR_PAGES);
4408 size = (end - start) * sizeof(struct page);
4409 map = alloc_remap(pgdat->node_id, size);
4411 map = alloc_bootmem_node_nopanic(pgdat, size);
4412 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
4414 #ifndef CONFIG_NEED_MULTIPLE_NODES
4416 * With no DISCONTIG, the global mem_map is just set as node 0's
4418 if (pgdat == NODE_DATA(0)) {
4419 mem_map = NODE_DATA(0)->node_mem_map;
4420 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
4421 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
4422 mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
4423 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4426 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
4429 void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
4430 unsigned long node_start_pfn, unsigned long *zholes_size)
4432 pg_data_t *pgdat = NODE_DATA(nid);
4434 pgdat->node_id = nid;
4435 pgdat->node_start_pfn = node_start_pfn;
4436 calculate_node_totalpages(pgdat, zones_size, zholes_size);
4438 alloc_node_mem_map(pgdat);
4439 #ifdef CONFIG_FLAT_NODE_MEM_MAP
4440 printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
4441 nid, (unsigned long)pgdat,
4442 (unsigned long)pgdat->node_mem_map);
4445 free_area_init_core(pgdat, zones_size, zholes_size);
4448 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
4450 #if MAX_NUMNODES > 1
4452 * Figure out the number of possible node ids.
4454 static void __init setup_nr_node_ids(void)
4457 unsigned int highest = 0;
4459 for_each_node_mask(node, node_possible_map)
4461 nr_node_ids = highest + 1;
4464 static inline void setup_nr_node_ids(void)
4470 * add_active_range - Register a range of PFNs backed by physical memory
4471 * @nid: The node ID the range resides on
4472 * @start_pfn: The start PFN of the available physical memory
4473 * @end_pfn: The end PFN of the available physical memory
4475 * These ranges are stored in an early_node_map[] and later used by
4476 * free_area_init_nodes() to calculate zone sizes and holes. If the
4477 * range spans a memory hole, it is up to the architecture to ensure
4478 * the memory is not freed by the bootmem allocator. If possible
4479 * the range being registered will be merged with existing ranges.
4481 void __init add_active_range(unsigned int nid, unsigned long start_pfn,
4482 unsigned long end_pfn)
4486 mminit_dprintk(MMINIT_TRACE, "memory_register",
4487 "Entering add_active_range(%d, %#lx, %#lx) "
4488 "%d entries of %d used\n",
4489 nid, start_pfn, end_pfn,
4490 nr_nodemap_entries, MAX_ACTIVE_REGIONS);
4492 mminit_validate_memmodel_limits(&start_pfn, &end_pfn);
4494 /* Merge with existing active regions if possible */
4495 for (i = 0; i < nr_nodemap_entries; i++) {
4496 if (early_node_map[i].nid != nid)
4499 /* Skip if an existing region covers this new one */
4500 if (start_pfn >= early_node_map[i].start_pfn &&
4501 end_pfn <= early_node_map[i].end_pfn)
4504 /* Merge forward if suitable */
4505 if (start_pfn <= early_node_map[i].end_pfn &&
4506 end_pfn > early_node_map[i].end_pfn) {
4507 early_node_map[i].end_pfn = end_pfn;
4511 /* Merge backward if suitable */
4512 if (start_pfn < early_node_map[i].start_pfn &&
4513 end_pfn >= early_node_map[i].start_pfn) {
4514 early_node_map[i].start_pfn = start_pfn;
4519 /* Check that early_node_map is large enough */
4520 if (i >= MAX_ACTIVE_REGIONS) {
4521 printk(KERN_CRIT "More than %d memory regions, truncating\n",
4522 MAX_ACTIVE_REGIONS);
4526 early_node_map[i].nid = nid;
4527 early_node_map[i].start_pfn = start_pfn;
4528 early_node_map[i].end_pfn = end_pfn;
4529 nr_nodemap_entries = i + 1;
4533 * remove_active_range - Shrink an existing registered range of PFNs
4534 * @nid: The node id the range is on that should be shrunk
4535 * @start_pfn: The new PFN of the range
4536 * @end_pfn: The new PFN of the range
4538 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
4539 * The map is kept near the end physical page range that has already been
4540 * registered. This function allows an arch to shrink an existing registered
4543 void __init remove_active_range(unsigned int nid, unsigned long start_pfn,
4544 unsigned long end_pfn)
4549 printk(KERN_DEBUG "remove_active_range (%d, %lu, %lu)\n",
4550 nid, start_pfn, end_pfn);
4552 /* Find the old active region end and shrink */
4553 for_each_active_range_index_in_nid(i, nid) {
4554 if (early_node_map[i].start_pfn >= start_pfn &&
4555 early_node_map[i].end_pfn <= end_pfn) {
4557 early_node_map[i].start_pfn = 0;
4558 early_node_map[i].end_pfn = 0;
4562 if (early_node_map[i].start_pfn < start_pfn &&
4563 early_node_map[i].end_pfn > start_pfn) {
4564 unsigned long temp_end_pfn = early_node_map[i].end_pfn;
4565 early_node_map[i].end_pfn = start_pfn;
4566 if (temp_end_pfn > end_pfn)
4567 add_active_range(nid, end_pfn, temp_end_pfn);
4570 if (early_node_map[i].start_pfn >= start_pfn &&
4571 early_node_map[i].end_pfn > end_pfn &&
4572 early_node_map[i].start_pfn < end_pfn) {
4573 early_node_map[i].start_pfn = end_pfn;
4581 /* remove the blank ones */
4582 for (i = nr_nodemap_entries - 1; i > 0; i--) {
4583 if (early_node_map[i].nid != nid)
4585 if (early_node_map[i].end_pfn)
4587 /* we found it, get rid of it */
4588 for (j = i; j < nr_nodemap_entries - 1; j++)
4589 memcpy(&early_node_map[j], &early_node_map[j+1],
4590 sizeof(early_node_map[j]));
4591 j = nr_nodemap_entries - 1;
4592 memset(&early_node_map[j], 0, sizeof(early_node_map[j]));
4593 nr_nodemap_entries--;
4598 * remove_all_active_ranges - Remove all currently registered regions
4600 * During discovery, it may be found that a table like SRAT is invalid
4601 * and an alternative discovery method must be used. This function removes
4602 * all currently registered regions.
4604 void __init remove_all_active_ranges(void)
4606 memset(early_node_map, 0, sizeof(early_node_map));
4607 nr_nodemap_entries = 0;
4610 /* Compare two active node_active_regions */
4611 static int __init cmp_node_active_region(const void *a, const void *b)
4613 struct node_active_region *arange = (struct node_active_region *)a;
4614 struct node_active_region *brange = (struct node_active_region *)b;
4616 /* Done this way to avoid overflows */
4617 if (arange->start_pfn > brange->start_pfn)
4619 if (arange->start_pfn < brange->start_pfn)
4625 /* sort the node_map by start_pfn */
4626 void __init sort_node_map(void)
4628 sort(early_node_map, (size_t)nr_nodemap_entries,
4629 sizeof(struct node_active_region),
4630 cmp_node_active_region, NULL);
4634 * node_map_pfn_alignment - determine the maximum internode alignment
4636 * This function should be called after node map is populated and sorted.
4637 * It calculates the maximum power of two alignment which can distinguish
4640 * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
4641 * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the
4642 * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is
4643 * shifted, 1GiB is enough and this function will indicate so.
4645 * This is used to test whether pfn -> nid mapping of the chosen memory
4646 * model has fine enough granularity to avoid incorrect mapping for the
4647 * populated node map.
4649 * Returns the determined alignment in pfn's. 0 if there is no alignment
4650 * requirement (single node).
4652 unsigned long __init node_map_pfn_alignment(void)
4654 unsigned long accl_mask = 0, last_end = 0;
4658 for_each_active_range_index_in_nid(i, MAX_NUMNODES) {
4659 int nid = early_node_map[i].nid;
4660 unsigned long start = early_node_map[i].start_pfn;
4661 unsigned long end = early_node_map[i].end_pfn;
4664 if (!start || last_nid < 0 || last_nid == nid) {
4671 * Start with a mask granular enough to pin-point to the
4672 * start pfn and tick off bits one-by-one until it becomes
4673 * too coarse to separate the current node from the last.
4675 mask = ~((1 << __ffs(start)) - 1);
4676 while (mask && last_end <= (start & (mask << 1)))
4679 /* accumulate all internode masks */
4683 /* convert mask to number of pages */
4684 return ~accl_mask + 1;
4687 /* Find the lowest pfn for a node */
4688 static unsigned long __init find_min_pfn_for_node(int nid)
4691 unsigned long min_pfn = ULONG_MAX;
4693 /* Assuming a sorted map, the first range found has the starting pfn */
4694 for_each_active_range_index_in_nid(i, nid)
4695 min_pfn = min(min_pfn, early_node_map[i].start_pfn);
4697 if (min_pfn == ULONG_MAX) {
4699 "Could not find start_pfn for node %d\n", nid);
4707 * find_min_pfn_with_active_regions - Find the minimum PFN registered
4709 * It returns the minimum PFN based on information provided via
4710 * add_active_range().
4712 unsigned long __init find_min_pfn_with_active_regions(void)
4714 return find_min_pfn_for_node(MAX_NUMNODES);
4718 * early_calculate_totalpages()
4719 * Sum pages in active regions for movable zone.
4720 * Populate N_HIGH_MEMORY for calculating usable_nodes.
4722 static unsigned long __init early_calculate_totalpages(void)
4725 unsigned long totalpages = 0;
4727 for (i = 0; i < nr_nodemap_entries; i++) {
4728 unsigned long pages = early_node_map[i].end_pfn -
4729 early_node_map[i].start_pfn;
4730 totalpages += pages;
4732 node_set_state(early_node_map[i].nid, N_HIGH_MEMORY);
4738 * Find the PFN the Movable zone begins in each node. Kernel memory
4739 * is spread evenly between nodes as long as the nodes have enough
4740 * memory. When they don't, some nodes will have more kernelcore than
4743 static void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn)
4746 unsigned long usable_startpfn;
4747 unsigned long kernelcore_node, kernelcore_remaining;
4748 /* save the state before borrow the nodemask */
4749 nodemask_t saved_node_state = node_states[N_HIGH_MEMORY];
4750 unsigned long totalpages = early_calculate_totalpages();
4751 int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]);
4754 * If movablecore was specified, calculate what size of
4755 * kernelcore that corresponds so that memory usable for
4756 * any allocation type is evenly spread. If both kernelcore
4757 * and movablecore are specified, then the value of kernelcore
4758 * will be used for required_kernelcore if it's greater than
4759 * what movablecore would have allowed.
4761 if (required_movablecore) {
4762 unsigned long corepages;
4765 * Round-up so that ZONE_MOVABLE is at least as large as what
4766 * was requested by the user
4768 required_movablecore =
4769 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
4770 corepages = totalpages - required_movablecore;
4772 required_kernelcore = max(required_kernelcore, corepages);
4775 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
4776 if (!required_kernelcore)
4779 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
4780 find_usable_zone_for_movable();
4781 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
4784 /* Spread kernelcore memory as evenly as possible throughout nodes */
4785 kernelcore_node = required_kernelcore / usable_nodes;
4786 for_each_node_state(nid, N_HIGH_MEMORY) {
4788 * Recalculate kernelcore_node if the division per node
4789 * now exceeds what is necessary to satisfy the requested
4790 * amount of memory for the kernel
4792 if (required_kernelcore < kernelcore_node)
4793 kernelcore_node = required_kernelcore / usable_nodes;
4796 * As the map is walked, we track how much memory is usable
4797 * by the kernel using kernelcore_remaining. When it is
4798 * 0, the rest of the node is usable by ZONE_MOVABLE
4800 kernelcore_remaining = kernelcore_node;
4802 /* Go through each range of PFNs within this node */
4803 for_each_active_range_index_in_nid(i, nid) {
4804 unsigned long start_pfn, end_pfn;
4805 unsigned long size_pages;
4807 start_pfn = max(early_node_map[i].start_pfn,
4808 zone_movable_pfn[nid]);
4809 end_pfn = early_node_map[i].end_pfn;
4810 if (start_pfn >= end_pfn)
4813 /* Account for what is only usable for kernelcore */
4814 if (start_pfn < usable_startpfn) {
4815 unsigned long kernel_pages;
4816 kernel_pages = min(end_pfn, usable_startpfn)
4819 kernelcore_remaining -= min(kernel_pages,
4820 kernelcore_remaining);
4821 required_kernelcore -= min(kernel_pages,
4822 required_kernelcore);
4824 /* Continue if range is now fully accounted */
4825 if (end_pfn <= usable_startpfn) {
4828 * Push zone_movable_pfn to the end so
4829 * that if we have to rebalance
4830 * kernelcore across nodes, we will
4831 * not double account here
4833 zone_movable_pfn[nid] = end_pfn;
4836 start_pfn = usable_startpfn;
4840 * The usable PFN range for ZONE_MOVABLE is from
4841 * start_pfn->end_pfn. Calculate size_pages as the
4842 * number of pages used as kernelcore
4844 size_pages = end_pfn - start_pfn;
4845 if (size_pages > kernelcore_remaining)
4846 size_pages = kernelcore_remaining;
4847 zone_movable_pfn[nid] = start_pfn + size_pages;
4850 * Some kernelcore has been met, update counts and
4851 * break if the kernelcore for this node has been
4854 required_kernelcore -= min(required_kernelcore,
4856 kernelcore_remaining -= size_pages;
4857 if (!kernelcore_remaining)
4863 * If there is still required_kernelcore, we do another pass with one
4864 * less node in the count. This will push zone_movable_pfn[nid] further
4865 * along on the nodes that still have memory until kernelcore is
4869 if (usable_nodes && required_kernelcore > usable_nodes)
4872 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
4873 for (nid = 0; nid < MAX_NUMNODES; nid++)
4874 zone_movable_pfn[nid] =
4875 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
4878 /* restore the node_state */
4879 node_states[N_HIGH_MEMORY] = saved_node_state;
4882 /* Any regular memory on that node ? */
4883 static void check_for_regular_memory(pg_data_t *pgdat)
4885 #ifdef CONFIG_HIGHMEM
4886 enum zone_type zone_type;
4888 for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) {
4889 struct zone *zone = &pgdat->node_zones[zone_type];
4890 if (zone->present_pages)
4891 node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY);
4897 * free_area_init_nodes - Initialise all pg_data_t and zone data
4898 * @max_zone_pfn: an array of max PFNs for each zone
4900 * This will call free_area_init_node() for each active node in the system.
4901 * Using the page ranges provided by add_active_range(), the size of each
4902 * zone in each node and their holes is calculated. If the maximum PFN
4903 * between two adjacent zones match, it is assumed that the zone is empty.
4904 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
4905 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
4906 * starts where the previous one ended. For example, ZONE_DMA32 starts
4907 * at arch_max_dma_pfn.
4909 void __init free_area_init_nodes(unsigned long *max_zone_pfn)
4914 /* Sort early_node_map as initialisation assumes it is sorted */
4917 /* Record where the zone boundaries are */
4918 memset(arch_zone_lowest_possible_pfn, 0,
4919 sizeof(arch_zone_lowest_possible_pfn));
4920 memset(arch_zone_highest_possible_pfn, 0,
4921 sizeof(arch_zone_highest_possible_pfn));
4922 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
4923 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
4924 for (i = 1; i < MAX_NR_ZONES; i++) {
4925 if (i == ZONE_MOVABLE)
4927 arch_zone_lowest_possible_pfn[i] =
4928 arch_zone_highest_possible_pfn[i-1];
4929 arch_zone_highest_possible_pfn[i] =
4930 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
4932 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
4933 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
4935 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
4936 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
4937 find_zone_movable_pfns_for_nodes(zone_movable_pfn);
4939 /* Print out the zone ranges */
4940 printk("Zone PFN ranges:\n");
4941 for (i = 0; i < MAX_NR_ZONES; i++) {
4942 if (i == ZONE_MOVABLE)
4944 printk(" %-8s ", zone_names[i]);
4945 if (arch_zone_lowest_possible_pfn[i] ==
4946 arch_zone_highest_possible_pfn[i])
4949 printk("%0#10lx -> %0#10lx\n",
4950 arch_zone_lowest_possible_pfn[i],
4951 arch_zone_highest_possible_pfn[i]);
4954 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
4955 printk("Movable zone start PFN for each node\n");
4956 for (i = 0; i < MAX_NUMNODES; i++) {
4957 if (zone_movable_pfn[i])
4958 printk(" Node %d: %lu\n", i, zone_movable_pfn[i]);
4961 /* Print out the early_node_map[] */
4962 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries);
4963 for (i = 0; i < nr_nodemap_entries; i++)
4964 printk(" %3d: %0#10lx -> %0#10lx\n", early_node_map[i].nid,
4965 early_node_map[i].start_pfn,
4966 early_node_map[i].end_pfn);
4968 /* Initialise every node */
4969 mminit_verify_pageflags_layout();
4970 setup_nr_node_ids();
4971 for_each_online_node(nid) {
4972 pg_data_t *pgdat = NODE_DATA(nid);
4973 free_area_init_node(nid, NULL,
4974 find_min_pfn_for_node(nid), NULL);
4976 /* Any memory on that node */
4977 if (pgdat->node_present_pages)
4978 node_set_state(nid, N_HIGH_MEMORY);
4979 check_for_regular_memory(pgdat);
4983 static int __init cmdline_parse_core(char *p, unsigned long *core)
4985 unsigned long long coremem;
4989 coremem = memparse(p, &p);
4990 *core = coremem >> PAGE_SHIFT;
4992 /* Paranoid check that UL is enough for the coremem value */
4993 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
4999 * kernelcore=size sets the amount of memory for use for allocations that
5000 * cannot be reclaimed or migrated.
5002 static int __init cmdline_parse_kernelcore(char *p)
5004 return cmdline_parse_core(p, &required_kernelcore);
5008 * movablecore=size sets the amount of memory for use for allocations that
5009 * can be reclaimed or migrated.
5011 static int __init cmdline_parse_movablecore(char *p)
5013 return cmdline_parse_core(p, &required_movablecore);
5016 early_param("kernelcore", cmdline_parse_kernelcore);
5017 early_param("movablecore", cmdline_parse_movablecore);
5019 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
5022 * set_dma_reserve - set the specified number of pages reserved in the first zone
5023 * @new_dma_reserve: The number of pages to mark reserved
5025 * The per-cpu batchsize and zone watermarks are determined by present_pages.
5026 * In the DMA zone, a significant percentage may be consumed by kernel image
5027 * and other unfreeable allocations which can skew the watermarks badly. This
5028 * function may optionally be used to account for unfreeable pages in the
5029 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
5030 * smaller per-cpu batchsize.
5032 void __init set_dma_reserve(unsigned long new_dma_reserve)
5034 dma_reserve = new_dma_reserve;
5037 void __init free_area_init(unsigned long *zones_size)
5039 free_area_init_node(0, zones_size,
5040 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
5043 static int page_alloc_cpu_notify(struct notifier_block *self,
5044 unsigned long action, void *hcpu)
5046 int cpu = (unsigned long)hcpu;
5048 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
5052 * Spill the event counters of the dead processor
5053 * into the current processors event counters.
5054 * This artificially elevates the count of the current
5057 vm_events_fold_cpu(cpu);
5060 * Zero the differential counters of the dead processor
5061 * so that the vm statistics are consistent.
5063 * This is only okay since the processor is dead and cannot
5064 * race with what we are doing.
5066 refresh_cpu_vm_stats(cpu);
5071 void __init page_alloc_init(void)
5073 hotcpu_notifier(page_alloc_cpu_notify, 0);
5077 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
5078 * or min_free_kbytes changes.
5080 static void calculate_totalreserve_pages(void)
5082 struct pglist_data *pgdat;
5083 unsigned long reserve_pages = 0;
5084 enum zone_type i, j;
5086 for_each_online_pgdat(pgdat) {
5087 for (i = 0; i < MAX_NR_ZONES; i++) {
5088 struct zone *zone = pgdat->node_zones + i;
5089 unsigned long max = 0;
5091 /* Find valid and maximum lowmem_reserve in the zone */
5092 for (j = i; j < MAX_NR_ZONES; j++) {
5093 if (zone->lowmem_reserve[j] > max)
5094 max = zone->lowmem_reserve[j];
5097 /* we treat the high watermark as reserved pages. */
5098 max += high_wmark_pages(zone);
5100 if (max > zone->present_pages)
5101 max = zone->present_pages;
5102 reserve_pages += max;
5105 totalreserve_pages = reserve_pages;
5109 * setup_per_zone_lowmem_reserve - called whenever
5110 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
5111 * has a correct pages reserved value, so an adequate number of
5112 * pages are left in the zone after a successful __alloc_pages().
5114 static void setup_per_zone_lowmem_reserve(void)
5116 struct pglist_data *pgdat;
5117 enum zone_type j, idx;
5119 for_each_online_pgdat(pgdat) {
5120 for (j = 0; j < MAX_NR_ZONES; j++) {
5121 struct zone *zone = pgdat->node_zones + j;
5122 unsigned long present_pages = zone->present_pages;
5124 zone->lowmem_reserve[j] = 0;
5128 struct zone *lower_zone;
5132 if (sysctl_lowmem_reserve_ratio[idx] < 1)
5133 sysctl_lowmem_reserve_ratio[idx] = 1;
5135 lower_zone = pgdat->node_zones + idx;
5136 lower_zone->lowmem_reserve[j] = present_pages /
5137 sysctl_lowmem_reserve_ratio[idx];
5138 present_pages += lower_zone->present_pages;
5143 /* update totalreserve_pages */
5144 calculate_totalreserve_pages();
5148 * setup_per_zone_wmarks - called when min_free_kbytes changes
5149 * or when memory is hot-{added|removed}
5151 * Ensures that the watermark[min,low,high] values for each zone are set
5152 * correctly with respect to min_free_kbytes.
5154 void setup_per_zone_wmarks(void)
5156 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
5157 unsigned long pages_low = extra_free_kbytes >> (PAGE_SHIFT - 10);
5158 unsigned long lowmem_pages = 0;
5160 unsigned long flags;
5162 /* Calculate total number of !ZONE_HIGHMEM pages */
5163 for_each_zone(zone) {
5164 if (!is_highmem(zone))
5165 lowmem_pages += zone->present_pages;
5168 for_each_zone(zone) {
5171 spin_lock_irqsave(&zone->lock, flags);
5172 min = (u64)pages_min * zone->present_pages;
5173 do_div(min, lowmem_pages);
5174 low = (u64)pages_low * zone->present_pages;
5175 do_div(low, vm_total_pages);
5177 if (is_highmem(zone)) {
5179 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
5180 * need highmem pages, so cap pages_min to a small
5183 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
5184 * deltas controls asynch page reclaim, and so should
5185 * not be capped for highmem.
5189 min_pages = zone->present_pages / 1024;
5190 if (min_pages < SWAP_CLUSTER_MAX)
5191 min_pages = SWAP_CLUSTER_MAX;
5192 if (min_pages > 128)
5194 zone->watermark[WMARK_MIN] = min_pages;
5197 * If it's a lowmem zone, reserve a number of pages
5198 * proportionate to the zone's size.
5200 zone->watermark[WMARK_MIN] = min;
5203 zone->watermark[WMARK_LOW] = min_wmark_pages(zone) +
5205 zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) +
5207 setup_zone_migrate_reserve(zone);
5208 spin_unlock_irqrestore(&zone->lock, flags);
5211 /* update totalreserve_pages */
5212 calculate_totalreserve_pages();
5216 * The inactive anon list should be small enough that the VM never has to
5217 * do too much work, but large enough that each inactive page has a chance
5218 * to be referenced again before it is swapped out.
5220 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
5221 * INACTIVE_ANON pages on this zone's LRU, maintained by the
5222 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
5223 * the anonymous pages are kept on the inactive list.
5226 * memory ratio inactive anon
5227 * -------------------------------------
5236 static void __meminit calculate_zone_inactive_ratio(struct zone *zone)
5238 unsigned int gb, ratio;
5240 /* Zone size in gigabytes */
5241 gb = zone->present_pages >> (30 - PAGE_SHIFT);
5243 ratio = int_sqrt(10 * gb);
5247 zone->inactive_ratio = ratio;
5250 static void __meminit setup_per_zone_inactive_ratio(void)
5255 calculate_zone_inactive_ratio(zone);
5259 * Initialise min_free_kbytes.
5261 * For small machines we want it small (128k min). For large machines
5262 * we want it large (64MB max). But it is not linear, because network
5263 * bandwidth does not increase linearly with machine size. We use
5265 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
5266 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
5282 int __meminit init_per_zone_wmark_min(void)
5284 unsigned long lowmem_kbytes;
5286 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
5288 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
5289 if (min_free_kbytes < 128)
5290 min_free_kbytes = 128;
5291 if (min_free_kbytes > 65536)
5292 min_free_kbytes = 65536;
5293 setup_per_zone_wmarks();
5294 refresh_zone_stat_thresholds();
5295 setup_per_zone_lowmem_reserve();
5296 setup_per_zone_inactive_ratio();
5299 module_init(init_per_zone_wmark_min)
5302 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
5303 * that we can call two helper functions whenever min_free_kbytes
5304 * or extra_free_kbytes changes.
5306 int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
5307 void __user *buffer, size_t *length, loff_t *ppos)
5309 proc_dointvec(table, write, buffer, length, ppos);
5311 setup_per_zone_wmarks();
5316 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
5317 void __user *buffer, size_t *length, loff_t *ppos)
5322 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
5327 zone->min_unmapped_pages = (zone->present_pages *
5328 sysctl_min_unmapped_ratio) / 100;
5332 int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
5333 void __user *buffer, size_t *length, loff_t *ppos)
5338 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
5343 zone->min_slab_pages = (zone->present_pages *
5344 sysctl_min_slab_ratio) / 100;
5350 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
5351 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
5352 * whenever sysctl_lowmem_reserve_ratio changes.
5354 * The reserve ratio obviously has absolutely no relation with the
5355 * minimum watermarks. The lowmem reserve ratio can only make sense
5356 * if in function of the boot time zone sizes.
5358 int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
5359 void __user *buffer, size_t *length, loff_t *ppos)
5361 proc_dointvec_minmax(table, write, buffer, length, ppos);
5362 setup_per_zone_lowmem_reserve();
5367 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
5368 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
5369 * can have before it gets flushed back to buddy allocator.
5372 int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
5373 void __user *buffer, size_t *length, loff_t *ppos)
5379 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
5380 if (!write || (ret == -EINVAL))
5382 for_each_populated_zone(zone) {
5383 for_each_possible_cpu(cpu) {
5385 high = zone->present_pages / percpu_pagelist_fraction;
5386 setup_pagelist_highmark(
5387 per_cpu_ptr(zone->pageset, cpu), high);
5393 int hashdist = HASHDIST_DEFAULT;
5396 static int __init set_hashdist(char *str)
5400 hashdist = simple_strtoul(str, &str, 0);
5403 __setup("hashdist=", set_hashdist);
5407 * allocate a large system hash table from bootmem
5408 * - it is assumed that the hash table must contain an exact power-of-2
5409 * quantity of entries
5410 * - limit is the number of hash buckets, not the total allocation size
5412 void *__init alloc_large_system_hash(const char *tablename,
5413 unsigned long bucketsize,
5414 unsigned long numentries,
5417 unsigned int *_hash_shift,
5418 unsigned int *_hash_mask,
5419 unsigned long limit)
5421 unsigned long long max = limit;
5422 unsigned long log2qty, size;
5425 /* allow the kernel cmdline to have a say */
5427 /* round applicable memory size up to nearest megabyte */
5428 numentries = nr_kernel_pages;
5429 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
5430 numentries >>= 20 - PAGE_SHIFT;
5431 numentries <<= 20 - PAGE_SHIFT;
5433 /* limit to 1 bucket per 2^scale bytes of low memory */
5434 if (scale > PAGE_SHIFT)
5435 numentries >>= (scale - PAGE_SHIFT);
5437 numentries <<= (PAGE_SHIFT - scale);
5439 /* Make sure we've got at least a 0-order allocation.. */
5440 if (unlikely(flags & HASH_SMALL)) {
5441 /* Makes no sense without HASH_EARLY */
5442 WARN_ON(!(flags & HASH_EARLY));
5443 if (!(numentries >> *_hash_shift)) {
5444 numentries = 1UL << *_hash_shift;
5445 BUG_ON(!numentries);
5447 } else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
5448 numentries = PAGE_SIZE / bucketsize;
5450 numentries = roundup_pow_of_two(numentries);
5452 /* limit allocation size to 1/16 total memory by default */
5454 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
5455 do_div(max, bucketsize);
5458 if (numentries > max)
5461 log2qty = ilog2(numentries);
5464 size = bucketsize << log2qty;
5465 if (flags & HASH_EARLY)
5466 table = alloc_bootmem_nopanic(size);
5468 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
5471 * If bucketsize is not a power-of-two, we may free
5472 * some pages at the end of hash table which
5473 * alloc_pages_exact() automatically does
5475 if (get_order(size) < MAX_ORDER) {
5476 table = alloc_pages_exact(size, GFP_ATOMIC);
5477 kmemleak_alloc(table, size, 1, GFP_ATOMIC);
5480 } while (!table && size > PAGE_SIZE && --log2qty);
5483 panic("Failed to allocate %s hash table\n", tablename);
5485 printk(KERN_INFO "%s hash table entries: %ld (order: %d, %lu bytes)\n",
5488 ilog2(size) - PAGE_SHIFT,
5492 *_hash_shift = log2qty;
5494 *_hash_mask = (1 << log2qty) - 1;
5499 /* Return a pointer to the bitmap storing bits affecting a block of pages */
5500 static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
5503 #ifdef CONFIG_SPARSEMEM
5504 return __pfn_to_section(pfn)->pageblock_flags;
5506 return zone->pageblock_flags;
5507 #endif /* CONFIG_SPARSEMEM */
5510 static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
5512 #ifdef CONFIG_SPARSEMEM
5513 pfn &= (PAGES_PER_SECTION-1);
5514 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
5516 pfn = pfn - zone->zone_start_pfn;
5517 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
5518 #endif /* CONFIG_SPARSEMEM */
5522 * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
5523 * @page: The page within the block of interest
5524 * @start_bitidx: The first bit of interest to retrieve
5525 * @end_bitidx: The last bit of interest
5526 * returns pageblock_bits flags
5528 unsigned long get_pageblock_flags_group(struct page *page,
5529 int start_bitidx, int end_bitidx)
5532 unsigned long *bitmap;
5533 unsigned long pfn, bitidx;
5534 unsigned long flags = 0;
5535 unsigned long value = 1;
5537 zone = page_zone(page);
5538 pfn = page_to_pfn(page);
5539 bitmap = get_pageblock_bitmap(zone, pfn);
5540 bitidx = pfn_to_bitidx(zone, pfn);
5542 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
5543 if (test_bit(bitidx + start_bitidx, bitmap))
5550 * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
5551 * @page: The page within the block of interest
5552 * @start_bitidx: The first bit of interest
5553 * @end_bitidx: The last bit of interest
5554 * @flags: The flags to set
5556 void set_pageblock_flags_group(struct page *page, unsigned long flags,
5557 int start_bitidx, int end_bitidx)
5560 unsigned long *bitmap;
5561 unsigned long pfn, bitidx;
5562 unsigned long value = 1;
5564 zone = page_zone(page);
5565 pfn = page_to_pfn(page);
5566 bitmap = get_pageblock_bitmap(zone, pfn);
5567 bitidx = pfn_to_bitidx(zone, pfn);
5568 VM_BUG_ON(pfn < zone->zone_start_pfn);
5569 VM_BUG_ON(pfn >= zone->zone_start_pfn + zone->spanned_pages);
5571 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
5573 __set_bit(bitidx + start_bitidx, bitmap);
5575 __clear_bit(bitidx + start_bitidx, bitmap);
5579 * This is designed as sub function...plz see page_isolation.c also.
5580 * set/clear page block's type to be ISOLATE.
5581 * page allocater never alloc memory from ISOLATE block.
5585 __count_immobile_pages(struct zone *zone, struct page *page, int count)
5587 unsigned long pfn, iter, found;
5589 * For avoiding noise data, lru_add_drain_all() should be called
5590 * If ZONE_MOVABLE, the zone never contains immobile pages
5592 if (zone_idx(zone) == ZONE_MOVABLE)
5595 if (get_pageblock_migratetype(page) == MIGRATE_MOVABLE)
5598 pfn = page_to_pfn(page);
5599 for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) {
5600 unsigned long check = pfn + iter;
5602 if (!pfn_valid_within(check))
5605 page = pfn_to_page(check);
5606 if (!page_count(page)) {
5607 if (PageBuddy(page))
5608 iter += (1 << page_order(page)) - 1;
5614 * If there are RECLAIMABLE pages, we need to check it.
5615 * But now, memory offline itself doesn't call shrink_slab()
5616 * and it still to be fixed.
5619 * If the page is not RAM, page_count()should be 0.
5620 * we don't need more check. This is an _used_ not-movable page.
5622 * The problematic thing here is PG_reserved pages. PG_reserved
5623 * is set to both of a memory hole page and a _used_ kernel
5632 bool is_pageblock_removable_nolock(struct page *page)
5634 struct zone *zone = page_zone(page);
5635 return __count_immobile_pages(zone, page, 0);
5638 int set_migratetype_isolate(struct page *page)
5641 unsigned long flags, pfn;
5642 struct memory_isolate_notify arg;
5646 zone = page_zone(page);
5648 spin_lock_irqsave(&zone->lock, flags);
5650 pfn = page_to_pfn(page);
5651 arg.start_pfn = pfn;
5652 arg.nr_pages = pageblock_nr_pages;
5653 arg.pages_found = 0;
5656 * It may be possible to isolate a pageblock even if the
5657 * migratetype is not MIGRATE_MOVABLE. The memory isolation
5658 * notifier chain is used by balloon drivers to return the
5659 * number of pages in a range that are held by the balloon
5660 * driver to shrink memory. If all the pages are accounted for
5661 * by balloons, are free, or on the LRU, isolation can continue.
5662 * Later, for example, when memory hotplug notifier runs, these
5663 * pages reported as "can be isolated" should be isolated(freed)
5664 * by the balloon driver through the memory notifier chain.
5666 notifier_ret = memory_isolate_notify(MEM_ISOLATE_COUNT, &arg);
5667 notifier_ret = notifier_to_errno(notifier_ret);
5671 * FIXME: Now, memory hotplug doesn't call shrink_slab() by itself.
5672 * We just check MOVABLE pages.
5674 if (__count_immobile_pages(zone, page, arg.pages_found))
5678 * immobile means "not-on-lru" paes. If immobile is larger than
5679 * removable-by-driver pages reported by notifier, we'll fail.
5684 set_pageblock_migratetype(page, MIGRATE_ISOLATE);
5685 move_freepages_block(zone, page, MIGRATE_ISOLATE);
5688 spin_unlock_irqrestore(&zone->lock, flags);
5694 void unset_migratetype_isolate(struct page *page)
5697 unsigned long flags;
5698 zone = page_zone(page);
5699 spin_lock_irqsave(&zone->lock, flags);
5700 if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE)
5702 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
5703 move_freepages_block(zone, page, MIGRATE_MOVABLE);
5705 spin_unlock_irqrestore(&zone->lock, flags);
5708 #ifdef CONFIG_MEMORY_HOTREMOVE
5710 * All pages in the range must be isolated before calling this.
5713 __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
5719 unsigned long flags;
5720 /* find the first valid pfn */
5721 for (pfn = start_pfn; pfn < end_pfn; pfn++)
5726 zone = page_zone(pfn_to_page(pfn));
5727 spin_lock_irqsave(&zone->lock, flags);
5729 while (pfn < end_pfn) {
5730 if (!pfn_valid(pfn)) {
5734 page = pfn_to_page(pfn);
5735 BUG_ON(page_count(page));
5736 BUG_ON(!PageBuddy(page));
5737 order = page_order(page);
5738 #ifdef CONFIG_DEBUG_VM
5739 printk(KERN_INFO "remove from free list %lx %d %lx\n",
5740 pfn, 1 << order, end_pfn);
5742 list_del(&page->lru);
5743 rmv_page_order(page);
5744 zone->free_area[order].nr_free--;
5745 __mod_zone_page_state(zone, NR_FREE_PAGES,
5747 for (i = 0; i < (1 << order); i++)
5748 SetPageReserved((page+i));
5749 pfn += (1 << order);
5751 spin_unlock_irqrestore(&zone->lock, flags);
5755 #ifdef CONFIG_MEMORY_FAILURE
5756 bool is_free_buddy_page(struct page *page)
5758 struct zone *zone = page_zone(page);
5759 unsigned long pfn = page_to_pfn(page);
5760 unsigned long flags;
5763 spin_lock_irqsave(&zone->lock, flags);
5764 for (order = 0; order < MAX_ORDER; order++) {
5765 struct page *page_head = page - (pfn & ((1 << order) - 1));
5767 if (PageBuddy(page_head) && page_order(page_head) >= order)
5770 spin_unlock_irqrestore(&zone->lock, flags);
5772 return order < MAX_ORDER;
5776 static struct trace_print_flags pageflag_names[] = {
5777 {1UL << PG_locked, "locked" },
5778 {1UL << PG_error, "error" },
5779 {1UL << PG_referenced, "referenced" },
5780 {1UL << PG_uptodate, "uptodate" },
5781 {1UL << PG_dirty, "dirty" },
5782 {1UL << PG_lru, "lru" },
5783 {1UL << PG_active, "active" },
5784 {1UL << PG_slab, "slab" },
5785 {1UL << PG_owner_priv_1, "owner_priv_1" },
5786 {1UL << PG_arch_1, "arch_1" },
5787 {1UL << PG_reserved, "reserved" },
5788 {1UL << PG_private, "private" },
5789 {1UL << PG_private_2, "private_2" },
5790 {1UL << PG_writeback, "writeback" },
5791 #ifdef CONFIG_PAGEFLAGS_EXTENDED
5792 {1UL << PG_head, "head" },
5793 {1UL << PG_tail, "tail" },
5795 {1UL << PG_compound, "compound" },
5797 {1UL << PG_swapcache, "swapcache" },
5798 {1UL << PG_mappedtodisk, "mappedtodisk" },
5799 {1UL << PG_reclaim, "reclaim" },
5800 {1UL << PG_swapbacked, "swapbacked" },
5801 {1UL << PG_unevictable, "unevictable" },
5803 {1UL << PG_mlocked, "mlocked" },
5805 #ifdef CONFIG_ARCH_USES_PG_UNCACHED
5806 {1UL << PG_uncached, "uncached" },
5808 #ifdef CONFIG_MEMORY_FAILURE
5809 {1UL << PG_hwpoison, "hwpoison" },
5814 static void dump_page_flags(unsigned long flags)
5816 const char *delim = "";
5820 printk(KERN_ALERT "page flags: %#lx(", flags);
5822 /* remove zone id */
5823 flags &= (1UL << NR_PAGEFLAGS) - 1;
5825 for (i = 0; pageflag_names[i].name && flags; i++) {
5827 mask = pageflag_names[i].mask;
5828 if ((flags & mask) != mask)
5832 printk("%s%s", delim, pageflag_names[i].name);
5836 /* check for left over flags */
5838 printk("%s%#lx", delim, flags);
5843 void dump_page(struct page *page)
5846 "page:%p count:%d mapcount:%d mapping:%p index:%#lx\n",
5847 page, atomic_read(&page->_count), page_mapcount(page),
5848 page->mapping, page->index);
5849 dump_page_flags(page->flags);
5850 mem_cgroup_print_bad_page(page);