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/compaction.h>
55 #include <trace/events/kmem.h>
56 #include <linux/ftrace_event.h>
57 #include <linux/memcontrol.h>
58 #include <linux/prefetch.h>
59 #include <linux/migrate.h>
60 #include <linux/page-debug-flags.h>
61 #include <linux/sched/rt.h>
63 #include <asm/tlbflush.h>
64 #include <asm/div64.h>
67 #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
68 DEFINE_PER_CPU(int, numa_node);
69 EXPORT_PER_CPU_SYMBOL(numa_node);
72 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
74 * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
75 * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
76 * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
77 * defined in <linux/topology.h>.
79 DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */
80 EXPORT_PER_CPU_SYMBOL(_numa_mem_);
84 * Array of node states.
86 nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
87 [N_POSSIBLE] = NODE_MASK_ALL,
88 [N_ONLINE] = { { [0] = 1UL } },
90 [N_NORMAL_MEMORY] = { { [0] = 1UL } },
92 [N_HIGH_MEMORY] = { { [0] = 1UL } },
94 #ifdef CONFIG_MOVABLE_NODE
95 [N_MEMORY] = { { [0] = 1UL } },
97 [N_CPU] = { { [0] = 1UL } },
100 EXPORT_SYMBOL(node_states);
102 unsigned long totalram_pages __read_mostly;
103 unsigned long totalreserve_pages __read_mostly;
105 * When calculating the number of globally allowed dirty pages, there
106 * is a certain number of per-zone reserves that should not be
107 * considered dirtyable memory. This is the sum of those reserves
108 * over all existing zones that contribute dirtyable memory.
110 unsigned long dirty_balance_reserve __read_mostly;
112 int percpu_pagelist_fraction;
113 gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK;
115 #ifdef CONFIG_PM_SLEEP
117 * The following functions are used by the suspend/hibernate code to temporarily
118 * change gfp_allowed_mask in order to avoid using I/O during memory allocations
119 * while devices are suspended. To avoid races with the suspend/hibernate code,
120 * they should always be called with pm_mutex held (gfp_allowed_mask also should
121 * only be modified with pm_mutex held, unless the suspend/hibernate code is
122 * guaranteed not to run in parallel with that modification).
125 static gfp_t saved_gfp_mask;
127 void pm_restore_gfp_mask(void)
129 WARN_ON(!mutex_is_locked(&pm_mutex));
130 if (saved_gfp_mask) {
131 gfp_allowed_mask = saved_gfp_mask;
136 void pm_restrict_gfp_mask(void)
138 WARN_ON(!mutex_is_locked(&pm_mutex));
139 WARN_ON(saved_gfp_mask);
140 saved_gfp_mask = gfp_allowed_mask;
141 gfp_allowed_mask &= ~GFP_IOFS;
144 bool pm_suspended_storage(void)
146 if ((gfp_allowed_mask & GFP_IOFS) == GFP_IOFS)
150 #endif /* CONFIG_PM_SLEEP */
152 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
153 int pageblock_order __read_mostly;
156 static void __free_pages_ok(struct page *page, unsigned int order);
159 * results with 256, 32 in the lowmem_reserve sysctl:
160 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
161 * 1G machine -> (16M dma, 784M normal, 224M high)
162 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
163 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
164 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
166 * TBD: should special case ZONE_DMA32 machines here - in those we normally
167 * don't need any ZONE_NORMAL reservation
169 int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
170 #ifdef CONFIG_ZONE_DMA
173 #ifdef CONFIG_ZONE_DMA32
176 #ifdef CONFIG_HIGHMEM
182 EXPORT_SYMBOL(totalram_pages);
184 static char * const zone_names[MAX_NR_ZONES] = {
185 #ifdef CONFIG_ZONE_DMA
188 #ifdef CONFIG_ZONE_DMA32
192 #ifdef CONFIG_HIGHMEM
198 int min_free_kbytes = 1024;
200 static unsigned long __meminitdata nr_kernel_pages;
201 static unsigned long __meminitdata nr_all_pages;
202 static unsigned long __meminitdata dma_reserve;
204 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
205 /* Movable memory ranges, will also be used by memblock subsystem. */
206 struct movablemem_map movablemem_map = {
211 static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
212 static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
213 static unsigned long __initdata required_kernelcore;
214 static unsigned long __initdata required_movablecore;
215 static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
216 static unsigned long __meminitdata zone_movable_limit[MAX_NUMNODES];
218 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
220 EXPORT_SYMBOL(movable_zone);
221 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
224 int nr_node_ids __read_mostly = MAX_NUMNODES;
225 int nr_online_nodes __read_mostly = 1;
226 EXPORT_SYMBOL(nr_node_ids);
227 EXPORT_SYMBOL(nr_online_nodes);
230 int page_group_by_mobility_disabled __read_mostly;
232 void set_pageblock_migratetype(struct page *page, int migratetype)
235 if (unlikely(page_group_by_mobility_disabled))
236 migratetype = MIGRATE_UNMOVABLE;
238 set_pageblock_flags_group(page, (unsigned long)migratetype,
239 PB_migrate, PB_migrate_end);
242 bool oom_killer_disabled __read_mostly;
244 #ifdef CONFIG_DEBUG_VM
245 static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
249 unsigned long pfn = page_to_pfn(page);
252 seq = zone_span_seqbegin(zone);
253 if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
255 else if (pfn < zone->zone_start_pfn)
257 } while (zone_span_seqretry(zone, seq));
262 static int page_is_consistent(struct zone *zone, struct page *page)
264 if (!pfn_valid_within(page_to_pfn(page)))
266 if (zone != page_zone(page))
272 * Temporary debugging check for pages not lying within a given zone.
274 static int bad_range(struct zone *zone, struct page *page)
276 if (page_outside_zone_boundaries(zone, page))
278 if (!page_is_consistent(zone, page))
284 static inline int bad_range(struct zone *zone, struct page *page)
290 static void bad_page(struct page *page)
292 static unsigned long resume;
293 static unsigned long nr_shown;
294 static unsigned long nr_unshown;
296 /* Don't complain about poisoned pages */
297 if (PageHWPoison(page)) {
298 reset_page_mapcount(page); /* remove PageBuddy */
303 * Allow a burst of 60 reports, then keep quiet for that minute;
304 * or allow a steady drip of one report per second.
306 if (nr_shown == 60) {
307 if (time_before(jiffies, resume)) {
313 "BUG: Bad page state: %lu messages suppressed\n",
320 resume = jiffies + 60 * HZ;
322 printk(KERN_ALERT "BUG: Bad page state in process %s pfn:%05lx\n",
323 current->comm, page_to_pfn(page));
329 /* Leave bad fields for debug, except PageBuddy could make trouble */
330 reset_page_mapcount(page); /* remove PageBuddy */
331 add_taint(TAINT_BAD_PAGE);
335 * Higher-order pages are called "compound pages". They are structured thusly:
337 * The first PAGE_SIZE page is called the "head page".
339 * The remaining PAGE_SIZE pages are called "tail pages".
341 * All pages have PG_compound set. All tail pages have their ->first_page
342 * pointing at the head page.
344 * The first tail page's ->lru.next holds the address of the compound page's
345 * put_page() function. Its ->lru.prev holds the order of allocation.
346 * This usage means that zero-order pages may not be compound.
349 static void free_compound_page(struct page *page)
351 __free_pages_ok(page, compound_order(page));
354 void prep_compound_page(struct page *page, unsigned long order)
357 int nr_pages = 1 << order;
359 set_compound_page_dtor(page, free_compound_page);
360 set_compound_order(page, order);
362 for (i = 1; i < nr_pages; i++) {
363 struct page *p = page + i;
365 set_page_count(p, 0);
366 p->first_page = page;
370 /* update __split_huge_page_refcount if you change this function */
371 static int destroy_compound_page(struct page *page, unsigned long order)
374 int nr_pages = 1 << order;
377 if (unlikely(compound_order(page) != order)) {
382 __ClearPageHead(page);
384 for (i = 1; i < nr_pages; i++) {
385 struct page *p = page + i;
387 if (unlikely(!PageTail(p) || (p->first_page != page))) {
397 static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
402 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
403 * and __GFP_HIGHMEM from hard or soft interrupt context.
405 VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
406 for (i = 0; i < (1 << order); i++)
407 clear_highpage(page + i);
410 #ifdef CONFIG_DEBUG_PAGEALLOC
411 unsigned int _debug_guardpage_minorder;
413 static int __init debug_guardpage_minorder_setup(char *buf)
417 if (kstrtoul(buf, 10, &res) < 0 || res > MAX_ORDER / 2) {
418 printk(KERN_ERR "Bad debug_guardpage_minorder value\n");
421 _debug_guardpage_minorder = res;
422 printk(KERN_INFO "Setting debug_guardpage_minorder to %lu\n", res);
425 __setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup);
427 static inline void set_page_guard_flag(struct page *page)
429 __set_bit(PAGE_DEBUG_FLAG_GUARD, &page->debug_flags);
432 static inline void clear_page_guard_flag(struct page *page)
434 __clear_bit(PAGE_DEBUG_FLAG_GUARD, &page->debug_flags);
437 static inline void set_page_guard_flag(struct page *page) { }
438 static inline void clear_page_guard_flag(struct page *page) { }
441 static inline void set_page_order(struct page *page, int order)
443 set_page_private(page, order);
444 __SetPageBuddy(page);
447 static inline void rmv_page_order(struct page *page)
449 __ClearPageBuddy(page);
450 set_page_private(page, 0);
454 * Locate the struct page for both the matching buddy in our
455 * pair (buddy1) and the combined O(n+1) page they form (page).
457 * 1) Any buddy B1 will have an order O twin B2 which satisfies
458 * the following equation:
460 * For example, if the starting buddy (buddy2) is #8 its order
462 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
464 * 2) Any buddy B will have an order O+1 parent P which
465 * satisfies the following equation:
468 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
470 static inline unsigned long
471 __find_buddy_index(unsigned long page_idx, unsigned int order)
473 return page_idx ^ (1 << order);
477 * This function checks whether a page is free && is the buddy
478 * we can do coalesce a page and its buddy if
479 * (a) the buddy is not in a hole &&
480 * (b) the buddy is in the buddy system &&
481 * (c) a page and its buddy have the same order &&
482 * (d) a page and its buddy are in the same zone.
484 * For recording whether a page is in the buddy system, we set ->_mapcount -2.
485 * Setting, clearing, and testing _mapcount -2 is serialized by zone->lock.
487 * For recording page's order, we use page_private(page).
489 static inline int page_is_buddy(struct page *page, struct page *buddy,
492 if (!pfn_valid_within(page_to_pfn(buddy)))
495 if (page_zone_id(page) != page_zone_id(buddy))
498 if (page_is_guard(buddy) && page_order(buddy) == order) {
499 VM_BUG_ON(page_count(buddy) != 0);
503 if (PageBuddy(buddy) && page_order(buddy) == order) {
504 VM_BUG_ON(page_count(buddy) != 0);
511 * Freeing function for a buddy system allocator.
513 * The concept of a buddy system is to maintain direct-mapped table
514 * (containing bit values) for memory blocks of various "orders".
515 * The bottom level table contains the map for the smallest allocatable
516 * units of memory (here, pages), and each level above it describes
517 * pairs of units from the levels below, hence, "buddies".
518 * At a high level, all that happens here is marking the table entry
519 * at the bottom level available, and propagating the changes upward
520 * as necessary, plus some accounting needed to play nicely with other
521 * parts of the VM system.
522 * At each level, we keep a list of pages, which are heads of continuous
523 * free pages of length of (1 << order) and marked with _mapcount -2. Page's
524 * order is recorded in page_private(page) field.
525 * So when we are allocating or freeing one, we can derive the state of the
526 * other. That is, if we allocate a small block, and both were
527 * free, the remainder of the region must be split into blocks.
528 * If a block is freed, and its buddy is also free, then this
529 * triggers coalescing into a block of larger size.
534 static inline void __free_one_page(struct page *page,
535 struct zone *zone, unsigned int order,
538 unsigned long page_idx;
539 unsigned long combined_idx;
540 unsigned long uninitialized_var(buddy_idx);
543 if (unlikely(PageCompound(page)))
544 if (unlikely(destroy_compound_page(page, order)))
547 VM_BUG_ON(migratetype == -1);
549 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
551 VM_BUG_ON(page_idx & ((1 << order) - 1));
552 VM_BUG_ON(bad_range(zone, page));
554 while (order < MAX_ORDER-1) {
555 buddy_idx = __find_buddy_index(page_idx, order);
556 buddy = page + (buddy_idx - page_idx);
557 if (!page_is_buddy(page, buddy, order))
560 * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
561 * merge with it and move up one order.
563 if (page_is_guard(buddy)) {
564 clear_page_guard_flag(buddy);
565 set_page_private(page, 0);
566 __mod_zone_freepage_state(zone, 1 << order,
569 list_del(&buddy->lru);
570 zone->free_area[order].nr_free--;
571 rmv_page_order(buddy);
573 combined_idx = buddy_idx & page_idx;
574 page = page + (combined_idx - page_idx);
575 page_idx = combined_idx;
578 set_page_order(page, order);
581 * If this is not the largest possible page, check if the buddy
582 * of the next-highest order is free. If it is, it's possible
583 * that pages are being freed that will coalesce soon. In case,
584 * that is happening, add the free page to the tail of the list
585 * so it's less likely to be used soon and more likely to be merged
586 * as a higher order page
588 if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) {
589 struct page *higher_page, *higher_buddy;
590 combined_idx = buddy_idx & page_idx;
591 higher_page = page + (combined_idx - page_idx);
592 buddy_idx = __find_buddy_index(combined_idx, order + 1);
593 higher_buddy = higher_page + (buddy_idx - combined_idx);
594 if (page_is_buddy(higher_page, higher_buddy, order + 1)) {
595 list_add_tail(&page->lru,
596 &zone->free_area[order].free_list[migratetype]);
601 list_add(&page->lru, &zone->free_area[order].free_list[migratetype]);
603 zone->free_area[order].nr_free++;
606 static inline int free_pages_check(struct page *page)
608 if (unlikely(page_mapcount(page) |
609 (page->mapping != NULL) |
610 (atomic_read(&page->_count) != 0) |
611 (page->flags & PAGE_FLAGS_CHECK_AT_FREE) |
612 (mem_cgroup_bad_page_check(page)))) {
616 reset_page_last_nid(page);
617 if (page->flags & PAGE_FLAGS_CHECK_AT_PREP)
618 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
623 * Frees a number of pages from the PCP lists
624 * Assumes all pages on list are in same zone, and of same order.
625 * count is the number of pages to free.
627 * If the zone was previously in an "all pages pinned" state then look to
628 * see if this freeing clears that state.
630 * And clear the zone's pages_scanned counter, to hold off the "all pages are
631 * pinned" detection logic.
633 static void free_pcppages_bulk(struct zone *zone, int count,
634 struct per_cpu_pages *pcp)
640 spin_lock(&zone->lock);
641 zone->all_unreclaimable = 0;
642 zone->pages_scanned = 0;
646 struct list_head *list;
649 * Remove pages from lists in a round-robin fashion. A
650 * batch_free count is maintained that is incremented when an
651 * empty list is encountered. This is so more pages are freed
652 * off fuller lists instead of spinning excessively around empty
657 if (++migratetype == MIGRATE_PCPTYPES)
659 list = &pcp->lists[migratetype];
660 } while (list_empty(list));
662 /* This is the only non-empty list. Free them all. */
663 if (batch_free == MIGRATE_PCPTYPES)
664 batch_free = to_free;
667 int mt; /* migratetype of the to-be-freed page */
669 page = list_entry(list->prev, struct page, lru);
670 /* must delete as __free_one_page list manipulates */
671 list_del(&page->lru);
672 mt = get_freepage_migratetype(page);
673 /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */
674 __free_one_page(page, zone, 0, mt);
675 trace_mm_page_pcpu_drain(page, 0, mt);
676 if (likely(!is_migrate_isolate_page(page))) {
677 __mod_zone_page_state(zone, NR_FREE_PAGES, 1);
678 if (is_migrate_cma(mt))
679 __mod_zone_page_state(zone, NR_FREE_CMA_PAGES, 1);
681 } while (--to_free && --batch_free && !list_empty(list));
683 spin_unlock(&zone->lock);
686 static void free_one_page(struct zone *zone, struct page *page, int order,
689 spin_lock(&zone->lock);
690 zone->all_unreclaimable = 0;
691 zone->pages_scanned = 0;
693 __free_one_page(page, zone, order, migratetype);
694 if (unlikely(!is_migrate_isolate(migratetype)))
695 __mod_zone_freepage_state(zone, 1 << order, migratetype);
696 spin_unlock(&zone->lock);
699 static bool free_pages_prepare(struct page *page, unsigned int order)
704 trace_mm_page_free(page, order);
705 kmemcheck_free_shadow(page, order);
708 page->mapping = NULL;
709 for (i = 0; i < (1 << order); i++)
710 bad += free_pages_check(page + i);
714 if (!PageHighMem(page)) {
715 debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order);
716 debug_check_no_obj_freed(page_address(page),
719 arch_free_page(page, order);
720 kernel_map_pages(page, 1 << order, 0);
725 static void __free_pages_ok(struct page *page, unsigned int order)
730 if (!free_pages_prepare(page, order))
733 local_irq_save(flags);
734 __count_vm_events(PGFREE, 1 << order);
735 migratetype = get_pageblock_migratetype(page);
736 set_freepage_migratetype(page, migratetype);
737 free_one_page(page_zone(page), page, order, migratetype);
738 local_irq_restore(flags);
742 * Read access to zone->managed_pages is safe because it's unsigned long,
743 * but we still need to serialize writers. Currently all callers of
744 * __free_pages_bootmem() except put_page_bootmem() should only be used
745 * at boot time. So for shorter boot time, we shift the burden to
746 * put_page_bootmem() to serialize writers.
748 void __meminit __free_pages_bootmem(struct page *page, unsigned int order)
750 unsigned int nr_pages = 1 << order;
754 for (loop = 0; loop < nr_pages; loop++) {
755 struct page *p = &page[loop];
757 if (loop + 1 < nr_pages)
759 __ClearPageReserved(p);
760 set_page_count(p, 0);
763 page_zone(page)->managed_pages += 1 << order;
764 set_page_refcounted(page);
765 __free_pages(page, order);
769 /* Free whole pageblock and set it's migration type to MIGRATE_CMA. */
770 void __init init_cma_reserved_pageblock(struct page *page)
772 unsigned i = pageblock_nr_pages;
773 struct page *p = page;
776 __ClearPageReserved(p);
777 set_page_count(p, 0);
780 set_page_refcounted(page);
781 set_pageblock_migratetype(page, MIGRATE_CMA);
782 __free_pages(page, pageblock_order);
783 totalram_pages += pageblock_nr_pages;
784 #ifdef CONFIG_HIGHMEM
785 if (PageHighMem(page))
786 totalhigh_pages += pageblock_nr_pages;
792 * The order of subdivision here is critical for the IO subsystem.
793 * Please do not alter this order without good reasons and regression
794 * testing. Specifically, as large blocks of memory are subdivided,
795 * the order in which smaller blocks are delivered depends on the order
796 * they're subdivided in this function. This is the primary factor
797 * influencing the order in which pages are delivered to the IO
798 * subsystem according to empirical testing, and this is also justified
799 * by considering the behavior of a buddy system containing a single
800 * large block of memory acted on by a series of small allocations.
801 * This behavior is a critical factor in sglist merging's success.
805 static inline void expand(struct zone *zone, struct page *page,
806 int low, int high, struct free_area *area,
809 unsigned long size = 1 << high;
815 VM_BUG_ON(bad_range(zone, &page[size]));
817 #ifdef CONFIG_DEBUG_PAGEALLOC
818 if (high < debug_guardpage_minorder()) {
820 * Mark as guard pages (or page), that will allow to
821 * merge back to allocator when buddy will be freed.
822 * Corresponding page table entries will not be touched,
823 * pages will stay not present in virtual address space
825 INIT_LIST_HEAD(&page[size].lru);
826 set_page_guard_flag(&page[size]);
827 set_page_private(&page[size], high);
828 /* Guard pages are not available for any usage */
829 __mod_zone_freepage_state(zone, -(1 << high),
834 list_add(&page[size].lru, &area->free_list[migratetype]);
836 set_page_order(&page[size], high);
841 * This page is about to be returned from the page allocator
843 static inline int check_new_page(struct page *page)
845 if (unlikely(page_mapcount(page) |
846 (page->mapping != NULL) |
847 (atomic_read(&page->_count) != 0) |
848 (page->flags & PAGE_FLAGS_CHECK_AT_PREP) |
849 (mem_cgroup_bad_page_check(page)))) {
856 static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
860 for (i = 0; i < (1 << order); i++) {
861 struct page *p = page + i;
862 if (unlikely(check_new_page(p)))
866 set_page_private(page, 0);
867 set_page_refcounted(page);
869 arch_alloc_page(page, order);
870 kernel_map_pages(page, 1 << order, 1);
872 if (gfp_flags & __GFP_ZERO)
873 prep_zero_page(page, order, gfp_flags);
875 if (order && (gfp_flags & __GFP_COMP))
876 prep_compound_page(page, order);
882 * Go through the free lists for the given migratetype and remove
883 * the smallest available page from the freelists
886 struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
889 unsigned int current_order;
890 struct free_area * area;
893 /* Find a page of the appropriate size in the preferred list */
894 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
895 area = &(zone->free_area[current_order]);
896 if (list_empty(&area->free_list[migratetype]))
899 page = list_entry(area->free_list[migratetype].next,
901 list_del(&page->lru);
902 rmv_page_order(page);
904 expand(zone, page, order, current_order, area, migratetype);
913 * This array describes the order lists are fallen back to when
914 * the free lists for the desirable migrate type are depleted
916 static int fallbacks[MIGRATE_TYPES][4] = {
917 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
918 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
920 [MIGRATE_MOVABLE] = { MIGRATE_CMA, MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
921 [MIGRATE_CMA] = { MIGRATE_RESERVE }, /* Never used */
923 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
925 [MIGRATE_RESERVE] = { MIGRATE_RESERVE }, /* Never used */
926 #ifdef CONFIG_MEMORY_ISOLATION
927 [MIGRATE_ISOLATE] = { MIGRATE_RESERVE }, /* Never used */
932 * Move the free pages in a range to the free lists of the requested type.
933 * Note that start_page and end_pages are not aligned on a pageblock
934 * boundary. If alignment is required, use move_freepages_block()
936 int move_freepages(struct zone *zone,
937 struct page *start_page, struct page *end_page,
944 #ifndef CONFIG_HOLES_IN_ZONE
946 * page_zone is not safe to call in this context when
947 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
948 * anyway as we check zone boundaries in move_freepages_block().
949 * Remove at a later date when no bug reports exist related to
950 * grouping pages by mobility
952 BUG_ON(page_zone(start_page) != page_zone(end_page));
955 for (page = start_page; page <= end_page;) {
956 /* Make sure we are not inadvertently changing nodes */
957 VM_BUG_ON(page_to_nid(page) != zone_to_nid(zone));
959 if (!pfn_valid_within(page_to_pfn(page))) {
964 if (!PageBuddy(page)) {
969 order = page_order(page);
970 list_move(&page->lru,
971 &zone->free_area[order].free_list[migratetype]);
972 set_freepage_migratetype(page, migratetype);
974 pages_moved += 1 << order;
980 int move_freepages_block(struct zone *zone, struct page *page,
983 unsigned long start_pfn, end_pfn;
984 struct page *start_page, *end_page;
986 start_pfn = page_to_pfn(page);
987 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
988 start_page = pfn_to_page(start_pfn);
989 end_page = start_page + pageblock_nr_pages - 1;
990 end_pfn = start_pfn + pageblock_nr_pages - 1;
992 /* Do not cross zone boundaries */
993 if (start_pfn < zone->zone_start_pfn)
995 if (end_pfn >= zone->zone_start_pfn + zone->spanned_pages)
998 return move_freepages(zone, start_page, end_page, migratetype);
1001 static void change_pageblock_range(struct page *pageblock_page,
1002 int start_order, int migratetype)
1004 int nr_pageblocks = 1 << (start_order - pageblock_order);
1006 while (nr_pageblocks--) {
1007 set_pageblock_migratetype(pageblock_page, migratetype);
1008 pageblock_page += pageblock_nr_pages;
1012 /* Remove an element from the buddy allocator from the fallback list */
1013 static inline struct page *
1014 __rmqueue_fallback(struct zone *zone, int order, int start_migratetype)
1016 struct free_area * area;
1021 /* Find the largest possible block of pages in the other list */
1022 for (current_order = MAX_ORDER-1; current_order >= order;
1025 migratetype = fallbacks[start_migratetype][i];
1027 /* MIGRATE_RESERVE handled later if necessary */
1028 if (migratetype == MIGRATE_RESERVE)
1031 area = &(zone->free_area[current_order]);
1032 if (list_empty(&area->free_list[migratetype]))
1035 page = list_entry(area->free_list[migratetype].next,
1040 * If breaking a large block of pages, move all free
1041 * pages to the preferred allocation list. If falling
1042 * back for a reclaimable kernel allocation, be more
1043 * aggressive about taking ownership of free pages
1045 * On the other hand, never change migration
1046 * type of MIGRATE_CMA pageblocks nor move CMA
1047 * pages on different free lists. We don't
1048 * want unmovable pages to be allocated from
1049 * MIGRATE_CMA areas.
1051 if (!is_migrate_cma(migratetype) &&
1052 (unlikely(current_order >= pageblock_order / 2) ||
1053 start_migratetype == MIGRATE_RECLAIMABLE ||
1054 page_group_by_mobility_disabled)) {
1056 pages = move_freepages_block(zone, page,
1059 /* Claim the whole block if over half of it is free */
1060 if (pages >= (1 << (pageblock_order-1)) ||
1061 page_group_by_mobility_disabled)
1062 set_pageblock_migratetype(page,
1065 migratetype = start_migratetype;
1068 /* Remove the page from the freelists */
1069 list_del(&page->lru);
1070 rmv_page_order(page);
1072 /* Take ownership for orders >= pageblock_order */
1073 if (current_order >= pageblock_order &&
1074 !is_migrate_cma(migratetype))
1075 change_pageblock_range(page, current_order,
1078 expand(zone, page, order, current_order, area,
1079 is_migrate_cma(migratetype)
1080 ? migratetype : start_migratetype);
1082 trace_mm_page_alloc_extfrag(page, order, current_order,
1083 start_migratetype, migratetype);
1093 * Do the hard work of removing an element from the buddy allocator.
1094 * Call me with the zone->lock already held.
1096 static struct page *__rmqueue(struct zone *zone, unsigned int order,
1102 page = __rmqueue_smallest(zone, order, migratetype);
1104 if (unlikely(!page) && migratetype != MIGRATE_RESERVE) {
1105 page = __rmqueue_fallback(zone, order, migratetype);
1108 * Use MIGRATE_RESERVE rather than fail an allocation. goto
1109 * is used because __rmqueue_smallest is an inline function
1110 * and we want just one call site
1113 migratetype = MIGRATE_RESERVE;
1118 trace_mm_page_alloc_zone_locked(page, order, migratetype);
1123 * Obtain a specified number of elements from the buddy allocator, all under
1124 * a single hold of the lock, for efficiency. Add them to the supplied list.
1125 * Returns the number of new pages which were placed at *list.
1127 static int rmqueue_bulk(struct zone *zone, unsigned int order,
1128 unsigned long count, struct list_head *list,
1129 int migratetype, int cold)
1131 int mt = migratetype, i;
1133 spin_lock(&zone->lock);
1134 for (i = 0; i < count; ++i) {
1135 struct page *page = __rmqueue(zone, order, migratetype);
1136 if (unlikely(page == NULL))
1140 * Split buddy pages returned by expand() are received here
1141 * in physical page order. The page is added to the callers and
1142 * list and the list head then moves forward. From the callers
1143 * perspective, the linked list is ordered by page number in
1144 * some conditions. This is useful for IO devices that can
1145 * merge IO requests if the physical pages are ordered
1148 if (likely(cold == 0))
1149 list_add(&page->lru, list);
1151 list_add_tail(&page->lru, list);
1152 if (IS_ENABLED(CONFIG_CMA)) {
1153 mt = get_pageblock_migratetype(page);
1154 if (!is_migrate_cma(mt) && !is_migrate_isolate(mt))
1157 set_freepage_migratetype(page, mt);
1159 if (is_migrate_cma(mt))
1160 __mod_zone_page_state(zone, NR_FREE_CMA_PAGES,
1163 __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
1164 spin_unlock(&zone->lock);
1170 * Called from the vmstat counter updater to drain pagesets of this
1171 * currently executing processor on remote nodes after they have
1174 * Note that this function must be called with the thread pinned to
1175 * a single processor.
1177 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
1179 unsigned long flags;
1182 local_irq_save(flags);
1183 if (pcp->count >= pcp->batch)
1184 to_drain = pcp->batch;
1186 to_drain = pcp->count;
1188 free_pcppages_bulk(zone, to_drain, pcp);
1189 pcp->count -= to_drain;
1191 local_irq_restore(flags);
1196 * Drain pages of the indicated processor.
1198 * The processor must either be the current processor and the
1199 * thread pinned to the current processor or a processor that
1202 static void drain_pages(unsigned int cpu)
1204 unsigned long flags;
1207 for_each_populated_zone(zone) {
1208 struct per_cpu_pageset *pset;
1209 struct per_cpu_pages *pcp;
1211 local_irq_save(flags);
1212 pset = per_cpu_ptr(zone->pageset, cpu);
1216 free_pcppages_bulk(zone, pcp->count, pcp);
1219 local_irq_restore(flags);
1224 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
1226 void drain_local_pages(void *arg)
1228 drain_pages(smp_processor_id());
1232 * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
1234 * Note that this code is protected against sending an IPI to an offline
1235 * CPU but does not guarantee sending an IPI to newly hotplugged CPUs:
1236 * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but
1237 * nothing keeps CPUs from showing up after we populated the cpumask and
1238 * before the call to on_each_cpu_mask().
1240 void drain_all_pages(void)
1243 struct per_cpu_pageset *pcp;
1247 * Allocate in the BSS so we wont require allocation in
1248 * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
1250 static cpumask_t cpus_with_pcps;
1253 * We don't care about racing with CPU hotplug event
1254 * as offline notification will cause the notified
1255 * cpu to drain that CPU pcps and on_each_cpu_mask
1256 * disables preemption as part of its processing
1258 for_each_online_cpu(cpu) {
1259 bool has_pcps = false;
1260 for_each_populated_zone(zone) {
1261 pcp = per_cpu_ptr(zone->pageset, cpu);
1262 if (pcp->pcp.count) {
1268 cpumask_set_cpu(cpu, &cpus_with_pcps);
1270 cpumask_clear_cpu(cpu, &cpus_with_pcps);
1272 on_each_cpu_mask(&cpus_with_pcps, drain_local_pages, NULL, 1);
1275 #ifdef CONFIG_HIBERNATION
1277 void mark_free_pages(struct zone *zone)
1279 unsigned long pfn, max_zone_pfn;
1280 unsigned long flags;
1282 struct list_head *curr;
1284 if (!zone->spanned_pages)
1287 spin_lock_irqsave(&zone->lock, flags);
1289 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
1290 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1291 if (pfn_valid(pfn)) {
1292 struct page *page = pfn_to_page(pfn);
1294 if (!swsusp_page_is_forbidden(page))
1295 swsusp_unset_page_free(page);
1298 for_each_migratetype_order(order, t) {
1299 list_for_each(curr, &zone->free_area[order].free_list[t]) {
1302 pfn = page_to_pfn(list_entry(curr, struct page, lru));
1303 for (i = 0; i < (1UL << order); i++)
1304 swsusp_set_page_free(pfn_to_page(pfn + i));
1307 spin_unlock_irqrestore(&zone->lock, flags);
1309 #endif /* CONFIG_PM */
1312 * Free a 0-order page
1313 * cold == 1 ? free a cold page : free a hot page
1315 void free_hot_cold_page(struct page *page, int cold)
1317 struct zone *zone = page_zone(page);
1318 struct per_cpu_pages *pcp;
1319 unsigned long flags;
1322 if (!free_pages_prepare(page, 0))
1325 migratetype = get_pageblock_migratetype(page);
1326 set_freepage_migratetype(page, migratetype);
1327 local_irq_save(flags);
1328 __count_vm_event(PGFREE);
1331 * We only track unmovable, reclaimable and movable on pcp lists.
1332 * Free ISOLATE pages back to the allocator because they are being
1333 * offlined but treat RESERVE as movable pages so we can get those
1334 * areas back if necessary. Otherwise, we may have to free
1335 * excessively into the page allocator
1337 if (migratetype >= MIGRATE_PCPTYPES) {
1338 if (unlikely(is_migrate_isolate(migratetype))) {
1339 free_one_page(zone, page, 0, migratetype);
1342 migratetype = MIGRATE_MOVABLE;
1345 pcp = &this_cpu_ptr(zone->pageset)->pcp;
1347 list_add_tail(&page->lru, &pcp->lists[migratetype]);
1349 list_add(&page->lru, &pcp->lists[migratetype]);
1351 if (pcp->count >= pcp->high) {
1352 free_pcppages_bulk(zone, pcp->batch, pcp);
1353 pcp->count -= pcp->batch;
1357 local_irq_restore(flags);
1361 * Free a list of 0-order pages
1363 void free_hot_cold_page_list(struct list_head *list, int cold)
1365 struct page *page, *next;
1367 list_for_each_entry_safe(page, next, list, lru) {
1368 trace_mm_page_free_batched(page, cold);
1369 free_hot_cold_page(page, cold);
1374 * split_page takes a non-compound higher-order page, and splits it into
1375 * n (1<<order) sub-pages: page[0..n]
1376 * Each sub-page must be freed individually.
1378 * Note: this is probably too low level an operation for use in drivers.
1379 * Please consult with lkml before using this in your driver.
1381 void split_page(struct page *page, unsigned int order)
1385 VM_BUG_ON(PageCompound(page));
1386 VM_BUG_ON(!page_count(page));
1388 #ifdef CONFIG_KMEMCHECK
1390 * Split shadow pages too, because free(page[0]) would
1391 * otherwise free the whole shadow.
1393 if (kmemcheck_page_is_tracked(page))
1394 split_page(virt_to_page(page[0].shadow), order);
1397 for (i = 1; i < (1 << order); i++)
1398 set_page_refcounted(page + i);
1401 static int __isolate_free_page(struct page *page, unsigned int order)
1403 unsigned long watermark;
1407 BUG_ON(!PageBuddy(page));
1409 zone = page_zone(page);
1410 mt = get_pageblock_migratetype(page);
1412 if (!is_migrate_isolate(mt)) {
1413 /* Obey watermarks as if the page was being allocated */
1414 watermark = low_wmark_pages(zone) + (1 << order);
1415 if (!zone_watermark_ok(zone, 0, watermark, 0, 0))
1418 __mod_zone_freepage_state(zone, -(1UL << order), mt);
1421 /* Remove page from free list */
1422 list_del(&page->lru);
1423 zone->free_area[order].nr_free--;
1424 rmv_page_order(page);
1426 /* Set the pageblock if the isolated page is at least a pageblock */
1427 if (order >= pageblock_order - 1) {
1428 struct page *endpage = page + (1 << order) - 1;
1429 for (; page < endpage; page += pageblock_nr_pages) {
1430 int mt = get_pageblock_migratetype(page);
1431 if (!is_migrate_isolate(mt) && !is_migrate_cma(mt))
1432 set_pageblock_migratetype(page,
1437 return 1UL << order;
1441 * Similar to split_page except the page is already free. As this is only
1442 * being used for migration, the migratetype of the block also changes.
1443 * As this is called with interrupts disabled, the caller is responsible
1444 * for calling arch_alloc_page() and kernel_map_page() after interrupts
1447 * Note: this is probably too low level an operation for use in drivers.
1448 * Please consult with lkml before using this in your driver.
1450 int split_free_page(struct page *page)
1455 order = page_order(page);
1457 nr_pages = __isolate_free_page(page, order);
1461 /* Split into individual pages */
1462 set_page_refcounted(page);
1463 split_page(page, order);
1468 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1469 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1473 struct page *buffered_rmqueue(struct zone *preferred_zone,
1474 struct zone *zone, int order, gfp_t gfp_flags,
1477 unsigned long flags;
1479 int cold = !!(gfp_flags & __GFP_COLD);
1482 if (likely(order == 0)) {
1483 struct per_cpu_pages *pcp;
1484 struct list_head *list;
1486 local_irq_save(flags);
1487 pcp = &this_cpu_ptr(zone->pageset)->pcp;
1488 list = &pcp->lists[migratetype];
1489 if (list_empty(list)) {
1490 pcp->count += rmqueue_bulk(zone, 0,
1493 if (unlikely(list_empty(list)))
1498 page = list_entry(list->prev, struct page, lru);
1500 page = list_entry(list->next, struct page, lru);
1502 list_del(&page->lru);
1505 if (unlikely(gfp_flags & __GFP_NOFAIL)) {
1507 * __GFP_NOFAIL is not to be used in new code.
1509 * All __GFP_NOFAIL callers should be fixed so that they
1510 * properly detect and handle allocation failures.
1512 * We most definitely don't want callers attempting to
1513 * allocate greater than order-1 page units with
1516 WARN_ON_ONCE(order > 1);
1518 spin_lock_irqsave(&zone->lock, flags);
1519 page = __rmqueue(zone, order, migratetype);
1520 spin_unlock(&zone->lock);
1523 __mod_zone_freepage_state(zone, -(1 << order),
1524 get_pageblock_migratetype(page));
1527 __count_zone_vm_events(PGALLOC, zone, 1 << order);
1528 zone_statistics(preferred_zone, zone, gfp_flags);
1529 local_irq_restore(flags);
1531 VM_BUG_ON(bad_range(zone, page));
1532 if (prep_new_page(page, order, gfp_flags))
1537 local_irq_restore(flags);
1541 #ifdef CONFIG_FAIL_PAGE_ALLOC
1544 struct fault_attr attr;
1546 u32 ignore_gfp_highmem;
1547 u32 ignore_gfp_wait;
1549 } fail_page_alloc = {
1550 .attr = FAULT_ATTR_INITIALIZER,
1551 .ignore_gfp_wait = 1,
1552 .ignore_gfp_highmem = 1,
1556 static int __init setup_fail_page_alloc(char *str)
1558 return setup_fault_attr(&fail_page_alloc.attr, str);
1560 __setup("fail_page_alloc=", setup_fail_page_alloc);
1562 static bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1564 if (order < fail_page_alloc.min_order)
1566 if (gfp_mask & __GFP_NOFAIL)
1568 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
1570 if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
1573 return should_fail(&fail_page_alloc.attr, 1 << order);
1576 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1578 static int __init fail_page_alloc_debugfs(void)
1580 umode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
1583 dir = fault_create_debugfs_attr("fail_page_alloc", NULL,
1584 &fail_page_alloc.attr);
1586 return PTR_ERR(dir);
1588 if (!debugfs_create_bool("ignore-gfp-wait", mode, dir,
1589 &fail_page_alloc.ignore_gfp_wait))
1591 if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir,
1592 &fail_page_alloc.ignore_gfp_highmem))
1594 if (!debugfs_create_u32("min-order", mode, dir,
1595 &fail_page_alloc.min_order))
1600 debugfs_remove_recursive(dir);
1605 late_initcall(fail_page_alloc_debugfs);
1607 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1609 #else /* CONFIG_FAIL_PAGE_ALLOC */
1611 static inline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1616 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1619 * Return true if free pages are above 'mark'. This takes into account the order
1620 * of the allocation.
1622 static bool __zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1623 int classzone_idx, int alloc_flags, long free_pages)
1625 /* free_pages my go negative - that's OK */
1627 long lowmem_reserve = z->lowmem_reserve[classzone_idx];
1630 free_pages -= (1 << order) - 1;
1631 if (alloc_flags & ALLOC_HIGH)
1633 if (alloc_flags & ALLOC_HARDER)
1636 /* If allocation can't use CMA areas don't use free CMA pages */
1637 if (!(alloc_flags & ALLOC_CMA))
1638 free_pages -= zone_page_state(z, NR_FREE_CMA_PAGES);
1640 if (free_pages <= min + lowmem_reserve)
1642 for (o = 0; o < order; o++) {
1643 /* At the next order, this order's pages become unavailable */
1644 free_pages -= z->free_area[o].nr_free << o;
1646 /* Require fewer higher order pages to be free */
1649 if (free_pages <= min)
1655 bool zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1656 int classzone_idx, int alloc_flags)
1658 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
1659 zone_page_state(z, NR_FREE_PAGES));
1662 bool zone_watermark_ok_safe(struct zone *z, int order, unsigned long mark,
1663 int classzone_idx, int alloc_flags)
1665 long free_pages = zone_page_state(z, NR_FREE_PAGES);
1667 if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark)
1668 free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES);
1670 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
1676 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1677 * skip over zones that are not allowed by the cpuset, or that have
1678 * been recently (in last second) found to be nearly full. See further
1679 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1680 * that have to skip over a lot of full or unallowed zones.
1682 * If the zonelist cache is present in the passed in zonelist, then
1683 * returns a pointer to the allowed node mask (either the current
1684 * tasks mems_allowed, or node_states[N_MEMORY].)
1686 * If the zonelist cache is not available for this zonelist, does
1687 * nothing and returns NULL.
1689 * If the fullzones BITMAP in the zonelist cache is stale (more than
1690 * a second since last zap'd) then we zap it out (clear its bits.)
1692 * We hold off even calling zlc_setup, until after we've checked the
1693 * first zone in the zonelist, on the theory that most allocations will
1694 * be satisfied from that first zone, so best to examine that zone as
1695 * quickly as we can.
1697 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1699 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1700 nodemask_t *allowednodes; /* zonelist_cache approximation */
1702 zlc = zonelist->zlcache_ptr;
1706 if (time_after(jiffies, zlc->last_full_zap + HZ)) {
1707 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1708 zlc->last_full_zap = jiffies;
1711 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1712 &cpuset_current_mems_allowed :
1713 &node_states[N_MEMORY];
1714 return allowednodes;
1718 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1719 * if it is worth looking at further for free memory:
1720 * 1) Check that the zone isn't thought to be full (doesn't have its
1721 * bit set in the zonelist_cache fullzones BITMAP).
1722 * 2) Check that the zones node (obtained from the zonelist_cache
1723 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1724 * Return true (non-zero) if zone is worth looking at further, or
1725 * else return false (zero) if it is not.
1727 * This check -ignores- the distinction between various watermarks,
1728 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1729 * found to be full for any variation of these watermarks, it will
1730 * be considered full for up to one second by all requests, unless
1731 * we are so low on memory on all allowed nodes that we are forced
1732 * into the second scan of the zonelist.
1734 * In the second scan we ignore this zonelist cache and exactly
1735 * apply the watermarks to all zones, even it is slower to do so.
1736 * We are low on memory in the second scan, and should leave no stone
1737 * unturned looking for a free page.
1739 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
1740 nodemask_t *allowednodes)
1742 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1743 int i; /* index of *z in zonelist zones */
1744 int n; /* node that zone *z is on */
1746 zlc = zonelist->zlcache_ptr;
1750 i = z - zonelist->_zonerefs;
1753 /* This zone is worth trying if it is allowed but not full */
1754 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1758 * Given 'z' scanning a zonelist, set the corresponding bit in
1759 * zlc->fullzones, so that subsequent attempts to allocate a page
1760 * from that zone don't waste time re-examining it.
1762 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
1764 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1765 int i; /* index of *z in zonelist zones */
1767 zlc = zonelist->zlcache_ptr;
1771 i = z - zonelist->_zonerefs;
1773 set_bit(i, zlc->fullzones);
1777 * clear all zones full, called after direct reclaim makes progress so that
1778 * a zone that was recently full is not skipped over for up to a second
1780 static void zlc_clear_zones_full(struct zonelist *zonelist)
1782 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1784 zlc = zonelist->zlcache_ptr;
1788 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1791 static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
1793 return node_isset(local_zone->node, zone->zone_pgdat->reclaim_nodes);
1796 static void __paginginit init_zone_allows_reclaim(int nid)
1800 for_each_online_node(i)
1801 if (node_distance(nid, i) <= RECLAIM_DISTANCE)
1802 node_set(i, NODE_DATA(nid)->reclaim_nodes);
1804 zone_reclaim_mode = 1;
1807 #else /* CONFIG_NUMA */
1809 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1814 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
1815 nodemask_t *allowednodes)
1820 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
1824 static void zlc_clear_zones_full(struct zonelist *zonelist)
1828 static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
1833 static inline void init_zone_allows_reclaim(int nid)
1836 #endif /* CONFIG_NUMA */
1839 * get_page_from_freelist goes through the zonelist trying to allocate
1842 static struct page *
1843 get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order,
1844 struct zonelist *zonelist, int high_zoneidx, int alloc_flags,
1845 struct zone *preferred_zone, int migratetype)
1848 struct page *page = NULL;
1851 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1852 int zlc_active = 0; /* set if using zonelist_cache */
1853 int did_zlc_setup = 0; /* just call zlc_setup() one time */
1855 classzone_idx = zone_idx(preferred_zone);
1858 * Scan zonelist, looking for a zone with enough free.
1859 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1861 for_each_zone_zonelist_nodemask(zone, z, zonelist,
1862 high_zoneidx, nodemask) {
1863 if (IS_ENABLED(CONFIG_NUMA) && zlc_active &&
1864 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1866 if ((alloc_flags & ALLOC_CPUSET) &&
1867 !cpuset_zone_allowed_softwall(zone, gfp_mask))
1870 * When allocating a page cache page for writing, we
1871 * want to get it from a zone that is within its dirty
1872 * limit, such that no single zone holds more than its
1873 * proportional share of globally allowed dirty pages.
1874 * The dirty limits take into account the zone's
1875 * lowmem reserves and high watermark so that kswapd
1876 * should be able to balance it without having to
1877 * write pages from its LRU list.
1879 * This may look like it could increase pressure on
1880 * lower zones by failing allocations in higher zones
1881 * before they are full. But the pages that do spill
1882 * over are limited as the lower zones are protected
1883 * by this very same mechanism. It should not become
1884 * a practical burden to them.
1886 * XXX: For now, allow allocations to potentially
1887 * exceed the per-zone dirty limit in the slowpath
1888 * (ALLOC_WMARK_LOW unset) before going into reclaim,
1889 * which is important when on a NUMA setup the allowed
1890 * zones are together not big enough to reach the
1891 * global limit. The proper fix for these situations
1892 * will require awareness of zones in the
1893 * dirty-throttling and the flusher threads.
1895 if ((alloc_flags & ALLOC_WMARK_LOW) &&
1896 (gfp_mask & __GFP_WRITE) && !zone_dirty_ok(zone))
1897 goto this_zone_full;
1899 BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
1900 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
1904 mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
1905 if (zone_watermark_ok(zone, order, mark,
1906 classzone_idx, alloc_flags))
1909 if (IS_ENABLED(CONFIG_NUMA) &&
1910 !did_zlc_setup && nr_online_nodes > 1) {
1912 * we do zlc_setup if there are multiple nodes
1913 * and before considering the first zone allowed
1916 allowednodes = zlc_setup(zonelist, alloc_flags);
1921 if (zone_reclaim_mode == 0 ||
1922 !zone_allows_reclaim(preferred_zone, zone))
1923 goto this_zone_full;
1926 * As we may have just activated ZLC, check if the first
1927 * eligible zone has failed zone_reclaim recently.
1929 if (IS_ENABLED(CONFIG_NUMA) && zlc_active &&
1930 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1933 ret = zone_reclaim(zone, gfp_mask, order);
1935 case ZONE_RECLAIM_NOSCAN:
1938 case ZONE_RECLAIM_FULL:
1939 /* scanned but unreclaimable */
1942 /* did we reclaim enough */
1943 if (!zone_watermark_ok(zone, order, mark,
1944 classzone_idx, alloc_flags))
1945 goto this_zone_full;
1950 page = buffered_rmqueue(preferred_zone, zone, order,
1951 gfp_mask, migratetype);
1955 if (IS_ENABLED(CONFIG_NUMA))
1956 zlc_mark_zone_full(zonelist, z);
1959 if (unlikely(IS_ENABLED(CONFIG_NUMA) && page == NULL && zlc_active)) {
1960 /* Disable zlc cache for second zonelist scan */
1967 * page->pfmemalloc is set when ALLOC_NO_WATERMARKS was
1968 * necessary to allocate the page. The expectation is
1969 * that the caller is taking steps that will free more
1970 * memory. The caller should avoid the page being used
1971 * for !PFMEMALLOC purposes.
1973 page->pfmemalloc = !!(alloc_flags & ALLOC_NO_WATERMARKS);
1979 * Large machines with many possible nodes should not always dump per-node
1980 * meminfo in irq context.
1982 static inline bool should_suppress_show_mem(void)
1987 ret = in_interrupt();
1992 static DEFINE_RATELIMIT_STATE(nopage_rs,
1993 DEFAULT_RATELIMIT_INTERVAL,
1994 DEFAULT_RATELIMIT_BURST);
1996 void warn_alloc_failed(gfp_t gfp_mask, int order, const char *fmt, ...)
1998 unsigned int filter = SHOW_MEM_FILTER_NODES;
2000 if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs) ||
2001 debug_guardpage_minorder() > 0)
2005 * This documents exceptions given to allocations in certain
2006 * contexts that are allowed to allocate outside current's set
2009 if (!(gfp_mask & __GFP_NOMEMALLOC))
2010 if (test_thread_flag(TIF_MEMDIE) ||
2011 (current->flags & (PF_MEMALLOC | PF_EXITING)))
2012 filter &= ~SHOW_MEM_FILTER_NODES;
2013 if (in_interrupt() || !(gfp_mask & __GFP_WAIT))
2014 filter &= ~SHOW_MEM_FILTER_NODES;
2017 struct va_format vaf;
2020 va_start(args, fmt);
2025 pr_warn("%pV", &vaf);
2030 pr_warn("%s: page allocation failure: order:%d, mode:0x%x\n",
2031 current->comm, order, gfp_mask);
2034 if (!should_suppress_show_mem())
2039 should_alloc_retry(gfp_t gfp_mask, unsigned int order,
2040 unsigned long did_some_progress,
2041 unsigned long pages_reclaimed)
2043 /* Do not loop if specifically requested */
2044 if (gfp_mask & __GFP_NORETRY)
2047 /* Always retry if specifically requested */
2048 if (gfp_mask & __GFP_NOFAIL)
2052 * Suspend converts GFP_KERNEL to __GFP_WAIT which can prevent reclaim
2053 * making forward progress without invoking OOM. Suspend also disables
2054 * storage devices so kswapd will not help. Bail if we are suspending.
2056 if (!did_some_progress && pm_suspended_storage())
2060 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
2061 * means __GFP_NOFAIL, but that may not be true in other
2064 if (order <= PAGE_ALLOC_COSTLY_ORDER)
2068 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
2069 * specified, then we retry until we no longer reclaim any pages
2070 * (above), or we've reclaimed an order of pages at least as
2071 * large as the allocation's order. In both cases, if the
2072 * allocation still fails, we stop retrying.
2074 if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order))
2080 static inline struct page *
2081 __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
2082 struct zonelist *zonelist, enum zone_type high_zoneidx,
2083 nodemask_t *nodemask, struct zone *preferred_zone,
2088 /* Acquire the OOM killer lock for the zones in zonelist */
2089 if (!try_set_zonelist_oom(zonelist, gfp_mask)) {
2090 schedule_timeout_uninterruptible(1);
2095 * Go through the zonelist yet one more time, keep very high watermark
2096 * here, this is only to catch a parallel oom killing, we must fail if
2097 * we're still under heavy pressure.
2099 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask,
2100 order, zonelist, high_zoneidx,
2101 ALLOC_WMARK_HIGH|ALLOC_CPUSET,
2102 preferred_zone, migratetype);
2106 if (!(gfp_mask & __GFP_NOFAIL)) {
2107 /* The OOM killer will not help higher order allocs */
2108 if (order > PAGE_ALLOC_COSTLY_ORDER)
2110 /* The OOM killer does not needlessly kill tasks for lowmem */
2111 if (high_zoneidx < ZONE_NORMAL)
2114 * GFP_THISNODE contains __GFP_NORETRY and we never hit this.
2115 * Sanity check for bare calls of __GFP_THISNODE, not real OOM.
2116 * The caller should handle page allocation failure by itself if
2117 * it specifies __GFP_THISNODE.
2118 * Note: Hugepage uses it but will hit PAGE_ALLOC_COSTLY_ORDER.
2120 if (gfp_mask & __GFP_THISNODE)
2123 /* Exhausted what can be done so it's blamo time */
2124 out_of_memory(zonelist, gfp_mask, order, nodemask, false);
2127 clear_zonelist_oom(zonelist, gfp_mask);
2131 #ifdef CONFIG_COMPACTION
2132 /* Try memory compaction for high-order allocations before reclaim */
2133 static struct page *
2134 __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
2135 struct zonelist *zonelist, enum zone_type high_zoneidx,
2136 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
2137 int migratetype, bool sync_migration,
2138 bool *contended_compaction, bool *deferred_compaction,
2139 unsigned long *did_some_progress)
2144 if (compaction_deferred(preferred_zone, order)) {
2145 *deferred_compaction = true;
2149 current->flags |= PF_MEMALLOC;
2150 *did_some_progress = try_to_compact_pages(zonelist, order, gfp_mask,
2151 nodemask, sync_migration,
2152 contended_compaction);
2153 current->flags &= ~PF_MEMALLOC;
2155 if (*did_some_progress != COMPACT_SKIPPED) {
2158 /* Page migration frees to the PCP lists but we want merging */
2159 drain_pages(get_cpu());
2162 page = get_page_from_freelist(gfp_mask, nodemask,
2163 order, zonelist, high_zoneidx,
2164 alloc_flags & ~ALLOC_NO_WATERMARKS,
2165 preferred_zone, migratetype);
2167 preferred_zone->compact_blockskip_flush = false;
2168 preferred_zone->compact_considered = 0;
2169 preferred_zone->compact_defer_shift = 0;
2170 if (order >= preferred_zone->compact_order_failed)
2171 preferred_zone->compact_order_failed = order + 1;
2172 count_vm_event(COMPACTSUCCESS);
2177 * It's bad if compaction run occurs and fails.
2178 * The most likely reason is that pages exist,
2179 * but not enough to satisfy watermarks.
2181 count_vm_event(COMPACTFAIL);
2184 * As async compaction considers a subset of pageblocks, only
2185 * defer if the failure was a sync compaction failure.
2188 defer_compaction(preferred_zone, order);
2196 static inline struct page *
2197 __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
2198 struct zonelist *zonelist, enum zone_type high_zoneidx,
2199 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
2200 int migratetype, bool sync_migration,
2201 bool *contended_compaction, bool *deferred_compaction,
2202 unsigned long *did_some_progress)
2206 #endif /* CONFIG_COMPACTION */
2208 /* Perform direct synchronous page reclaim */
2210 __perform_reclaim(gfp_t gfp_mask, unsigned int order, struct zonelist *zonelist,
2211 nodemask_t *nodemask)
2213 struct reclaim_state reclaim_state;
2218 /* We now go into synchronous reclaim */
2219 cpuset_memory_pressure_bump();
2220 current->flags |= PF_MEMALLOC;
2221 lockdep_set_current_reclaim_state(gfp_mask);
2222 reclaim_state.reclaimed_slab = 0;
2223 current->reclaim_state = &reclaim_state;
2225 progress = try_to_free_pages(zonelist, order, gfp_mask, nodemask);
2227 current->reclaim_state = NULL;
2228 lockdep_clear_current_reclaim_state();
2229 current->flags &= ~PF_MEMALLOC;
2236 /* The really slow allocator path where we enter direct reclaim */
2237 static inline struct page *
2238 __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
2239 struct zonelist *zonelist, enum zone_type high_zoneidx,
2240 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
2241 int migratetype, unsigned long *did_some_progress)
2243 struct page *page = NULL;
2244 bool drained = false;
2246 *did_some_progress = __perform_reclaim(gfp_mask, order, zonelist,
2248 if (unlikely(!(*did_some_progress)))
2251 /* After successful reclaim, reconsider all zones for allocation */
2252 if (IS_ENABLED(CONFIG_NUMA))
2253 zlc_clear_zones_full(zonelist);
2256 page = get_page_from_freelist(gfp_mask, nodemask, order,
2257 zonelist, high_zoneidx,
2258 alloc_flags & ~ALLOC_NO_WATERMARKS,
2259 preferred_zone, migratetype);
2262 * If an allocation failed after direct reclaim, it could be because
2263 * pages are pinned on the per-cpu lists. Drain them and try again
2265 if (!page && !drained) {
2275 * This is called in the allocator slow-path if the allocation request is of
2276 * sufficient urgency to ignore watermarks and take other desperate measures
2278 static inline struct page *
2279 __alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order,
2280 struct zonelist *zonelist, enum zone_type high_zoneidx,
2281 nodemask_t *nodemask, struct zone *preferred_zone,
2287 page = get_page_from_freelist(gfp_mask, nodemask, order,
2288 zonelist, high_zoneidx, ALLOC_NO_WATERMARKS,
2289 preferred_zone, migratetype);
2291 if (!page && gfp_mask & __GFP_NOFAIL)
2292 wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50);
2293 } while (!page && (gfp_mask & __GFP_NOFAIL));
2299 void wake_all_kswapd(unsigned int order, struct zonelist *zonelist,
2300 enum zone_type high_zoneidx,
2301 enum zone_type classzone_idx)
2306 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx)
2307 wakeup_kswapd(zone, order, classzone_idx);
2311 gfp_to_alloc_flags(gfp_t gfp_mask)
2313 int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
2314 const gfp_t wait = gfp_mask & __GFP_WAIT;
2316 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
2317 BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
2320 * The caller may dip into page reserves a bit more if the caller
2321 * cannot run direct reclaim, or if the caller has realtime scheduling
2322 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
2323 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
2325 alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH);
2329 * Not worth trying to allocate harder for
2330 * __GFP_NOMEMALLOC even if it can't schedule.
2332 if (!(gfp_mask & __GFP_NOMEMALLOC))
2333 alloc_flags |= ALLOC_HARDER;
2335 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
2336 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
2338 alloc_flags &= ~ALLOC_CPUSET;
2339 } else if (unlikely(rt_task(current)) && !in_interrupt())
2340 alloc_flags |= ALLOC_HARDER;
2342 if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) {
2343 if (gfp_mask & __GFP_MEMALLOC)
2344 alloc_flags |= ALLOC_NO_WATERMARKS;
2345 else if (in_serving_softirq() && (current->flags & PF_MEMALLOC))
2346 alloc_flags |= ALLOC_NO_WATERMARKS;
2347 else if (!in_interrupt() &&
2348 ((current->flags & PF_MEMALLOC) ||
2349 unlikely(test_thread_flag(TIF_MEMDIE))))
2350 alloc_flags |= ALLOC_NO_WATERMARKS;
2353 if (allocflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE)
2354 alloc_flags |= ALLOC_CMA;
2359 bool gfp_pfmemalloc_allowed(gfp_t gfp_mask)
2361 return !!(gfp_to_alloc_flags(gfp_mask) & ALLOC_NO_WATERMARKS);
2364 static inline struct page *
2365 __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
2366 struct zonelist *zonelist, enum zone_type high_zoneidx,
2367 nodemask_t *nodemask, struct zone *preferred_zone,
2370 const gfp_t wait = gfp_mask & __GFP_WAIT;
2371 struct page *page = NULL;
2373 unsigned long pages_reclaimed = 0;
2374 unsigned long did_some_progress;
2375 bool sync_migration = false;
2376 bool deferred_compaction = false;
2377 bool contended_compaction = false;
2380 * In the slowpath, we sanity check order to avoid ever trying to
2381 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
2382 * be using allocators in order of preference for an area that is
2385 if (order >= MAX_ORDER) {
2386 WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
2391 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
2392 * __GFP_NOWARN set) should not cause reclaim since the subsystem
2393 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
2394 * using a larger set of nodes after it has established that the
2395 * allowed per node queues are empty and that nodes are
2398 if (IS_ENABLED(CONFIG_NUMA) &&
2399 (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
2403 if (!(gfp_mask & __GFP_NO_KSWAPD))
2404 wake_all_kswapd(order, zonelist, high_zoneidx,
2405 zone_idx(preferred_zone));
2408 * OK, we're below the kswapd watermark and have kicked background
2409 * reclaim. Now things get more complex, so set up alloc_flags according
2410 * to how we want to proceed.
2412 alloc_flags = gfp_to_alloc_flags(gfp_mask);
2415 * Find the true preferred zone if the allocation is unconstrained by
2418 if (!(alloc_flags & ALLOC_CPUSET) && !nodemask)
2419 first_zones_zonelist(zonelist, high_zoneidx, NULL,
2423 /* This is the last chance, in general, before the goto nopage. */
2424 page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist,
2425 high_zoneidx, alloc_flags & ~ALLOC_NO_WATERMARKS,
2426 preferred_zone, migratetype);
2430 /* Allocate without watermarks if the context allows */
2431 if (alloc_flags & ALLOC_NO_WATERMARKS) {
2433 * Ignore mempolicies if ALLOC_NO_WATERMARKS on the grounds
2434 * the allocation is high priority and these type of
2435 * allocations are system rather than user orientated
2437 zonelist = node_zonelist(numa_node_id(), gfp_mask);
2439 page = __alloc_pages_high_priority(gfp_mask, order,
2440 zonelist, high_zoneidx, nodemask,
2441 preferred_zone, migratetype);
2447 /* Atomic allocations - we can't balance anything */
2451 /* Avoid recursion of direct reclaim */
2452 if (current->flags & PF_MEMALLOC)
2455 /* Avoid allocations with no watermarks from looping endlessly */
2456 if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL))
2460 * Try direct compaction. The first pass is asynchronous. Subsequent
2461 * attempts after direct reclaim are synchronous
2463 page = __alloc_pages_direct_compact(gfp_mask, order,
2464 zonelist, high_zoneidx,
2466 alloc_flags, preferred_zone,
2467 migratetype, sync_migration,
2468 &contended_compaction,
2469 &deferred_compaction,
2470 &did_some_progress);
2473 sync_migration = true;
2476 * If compaction is deferred for high-order allocations, it is because
2477 * sync compaction recently failed. In this is the case and the caller
2478 * requested a movable allocation that does not heavily disrupt the
2479 * system then fail the allocation instead of entering direct reclaim.
2481 if ((deferred_compaction || contended_compaction) &&
2482 (gfp_mask & __GFP_NO_KSWAPD))
2485 /* Try direct reclaim and then allocating */
2486 page = __alloc_pages_direct_reclaim(gfp_mask, order,
2487 zonelist, high_zoneidx,
2489 alloc_flags, preferred_zone,
2490 migratetype, &did_some_progress);
2495 * If we failed to make any progress reclaiming, then we are
2496 * running out of options and have to consider going OOM
2498 if (!did_some_progress) {
2499 if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
2500 if (oom_killer_disabled)
2502 /* Coredumps can quickly deplete all memory reserves */
2503 if ((current->flags & PF_DUMPCORE) &&
2504 !(gfp_mask & __GFP_NOFAIL))
2506 page = __alloc_pages_may_oom(gfp_mask, order,
2507 zonelist, high_zoneidx,
2508 nodemask, preferred_zone,
2513 if (!(gfp_mask & __GFP_NOFAIL)) {
2515 * The oom killer is not called for high-order
2516 * allocations that may fail, so if no progress
2517 * is being made, there are no other options and
2518 * retrying is unlikely to help.
2520 if (order > PAGE_ALLOC_COSTLY_ORDER)
2523 * The oom killer is not called for lowmem
2524 * allocations to prevent needlessly killing
2527 if (high_zoneidx < ZONE_NORMAL)
2535 /* Check if we should retry the allocation */
2536 pages_reclaimed += did_some_progress;
2537 if (should_alloc_retry(gfp_mask, order, did_some_progress,
2539 /* Wait for some write requests to complete then retry */
2540 wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50);
2544 * High-order allocations do not necessarily loop after
2545 * direct reclaim and reclaim/compaction depends on compaction
2546 * being called after reclaim so call directly if necessary
2548 page = __alloc_pages_direct_compact(gfp_mask, order,
2549 zonelist, high_zoneidx,
2551 alloc_flags, preferred_zone,
2552 migratetype, sync_migration,
2553 &contended_compaction,
2554 &deferred_compaction,
2555 &did_some_progress);
2561 warn_alloc_failed(gfp_mask, order, NULL);
2564 if (kmemcheck_enabled)
2565 kmemcheck_pagealloc_alloc(page, order, gfp_mask);
2571 * This is the 'heart' of the zoned buddy allocator.
2574 __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
2575 struct zonelist *zonelist, nodemask_t *nodemask)
2577 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
2578 struct zone *preferred_zone;
2579 struct page *page = NULL;
2580 int migratetype = allocflags_to_migratetype(gfp_mask);
2581 unsigned int cpuset_mems_cookie;
2582 int alloc_flags = ALLOC_WMARK_LOW|ALLOC_CPUSET;
2583 struct mem_cgroup *memcg = NULL;
2585 gfp_mask &= gfp_allowed_mask;
2587 lockdep_trace_alloc(gfp_mask);
2589 might_sleep_if(gfp_mask & __GFP_WAIT);
2591 if (should_fail_alloc_page(gfp_mask, order))
2595 * Check the zones suitable for the gfp_mask contain at least one
2596 * valid zone. It's possible to have an empty zonelist as a result
2597 * of GFP_THISNODE and a memoryless node
2599 if (unlikely(!zonelist->_zonerefs->zone))
2603 * Will only have any effect when __GFP_KMEMCG is set. This is
2604 * verified in the (always inline) callee
2606 if (!memcg_kmem_newpage_charge(gfp_mask, &memcg, order))
2610 cpuset_mems_cookie = get_mems_allowed();
2612 /* The preferred zone is used for statistics later */
2613 first_zones_zonelist(zonelist, high_zoneidx,
2614 nodemask ? : &cpuset_current_mems_allowed,
2616 if (!preferred_zone)
2620 if (allocflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE)
2621 alloc_flags |= ALLOC_CMA;
2623 /* First allocation attempt */
2624 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order,
2625 zonelist, high_zoneidx, alloc_flags,
2626 preferred_zone, migratetype);
2627 if (unlikely(!page))
2628 page = __alloc_pages_slowpath(gfp_mask, order,
2629 zonelist, high_zoneidx, nodemask,
2630 preferred_zone, migratetype);
2632 trace_mm_page_alloc(page, order, gfp_mask, migratetype);
2636 * When updating a task's mems_allowed, it is possible to race with
2637 * parallel threads in such a way that an allocation can fail while
2638 * the mask is being updated. If a page allocation is about to fail,
2639 * check if the cpuset changed during allocation and if so, retry.
2641 if (unlikely(!put_mems_allowed(cpuset_mems_cookie) && !page))
2644 memcg_kmem_commit_charge(page, memcg, order);
2648 EXPORT_SYMBOL(__alloc_pages_nodemask);
2651 * Common helper functions.
2653 unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
2658 * __get_free_pages() returns a 32-bit address, which cannot represent
2661 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
2663 page = alloc_pages(gfp_mask, order);
2666 return (unsigned long) page_address(page);
2668 EXPORT_SYMBOL(__get_free_pages);
2670 unsigned long get_zeroed_page(gfp_t gfp_mask)
2672 return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
2674 EXPORT_SYMBOL(get_zeroed_page);
2676 void __free_pages(struct page *page, unsigned int order)
2678 if (put_page_testzero(page)) {
2680 free_hot_cold_page(page, 0);
2682 __free_pages_ok(page, order);
2686 EXPORT_SYMBOL(__free_pages);
2688 void free_pages(unsigned long addr, unsigned int order)
2691 VM_BUG_ON(!virt_addr_valid((void *)addr));
2692 __free_pages(virt_to_page((void *)addr), order);
2696 EXPORT_SYMBOL(free_pages);
2699 * __free_memcg_kmem_pages and free_memcg_kmem_pages will free
2700 * pages allocated with __GFP_KMEMCG.
2702 * Those pages are accounted to a particular memcg, embedded in the
2703 * corresponding page_cgroup. To avoid adding a hit in the allocator to search
2704 * for that information only to find out that it is NULL for users who have no
2705 * interest in that whatsoever, we provide these functions.
2707 * The caller knows better which flags it relies on.
2709 void __free_memcg_kmem_pages(struct page *page, unsigned int order)
2711 memcg_kmem_uncharge_pages(page, order);
2712 __free_pages(page, order);
2715 void free_memcg_kmem_pages(unsigned long addr, unsigned int order)
2718 VM_BUG_ON(!virt_addr_valid((void *)addr));
2719 __free_memcg_kmem_pages(virt_to_page((void *)addr), order);
2723 static void *make_alloc_exact(unsigned long addr, unsigned order, size_t size)
2726 unsigned long alloc_end = addr + (PAGE_SIZE << order);
2727 unsigned long used = addr + PAGE_ALIGN(size);
2729 split_page(virt_to_page((void *)addr), order);
2730 while (used < alloc_end) {
2735 return (void *)addr;
2739 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
2740 * @size: the number of bytes to allocate
2741 * @gfp_mask: GFP flags for the allocation
2743 * This function is similar to alloc_pages(), except that it allocates the
2744 * minimum number of pages to satisfy the request. alloc_pages() can only
2745 * allocate memory in power-of-two pages.
2747 * This function is also limited by MAX_ORDER.
2749 * Memory allocated by this function must be released by free_pages_exact().
2751 void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
2753 unsigned int order = get_order(size);
2756 addr = __get_free_pages(gfp_mask, order);
2757 return make_alloc_exact(addr, order, size);
2759 EXPORT_SYMBOL(alloc_pages_exact);
2762 * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
2764 * @nid: the preferred node ID where memory should be allocated
2765 * @size: the number of bytes to allocate
2766 * @gfp_mask: GFP flags for the allocation
2768 * Like alloc_pages_exact(), but try to allocate on node nid first before falling
2770 * Note this is not alloc_pages_exact_node() which allocates on a specific node,
2773 void *alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask)
2775 unsigned order = get_order(size);
2776 struct page *p = alloc_pages_node(nid, gfp_mask, order);
2779 return make_alloc_exact((unsigned long)page_address(p), order, size);
2781 EXPORT_SYMBOL(alloc_pages_exact_nid);
2784 * free_pages_exact - release memory allocated via alloc_pages_exact()
2785 * @virt: the value returned by alloc_pages_exact.
2786 * @size: size of allocation, same value as passed to alloc_pages_exact().
2788 * Release the memory allocated by a previous call to alloc_pages_exact.
2790 void free_pages_exact(void *virt, size_t size)
2792 unsigned long addr = (unsigned long)virt;
2793 unsigned long end = addr + PAGE_ALIGN(size);
2795 while (addr < end) {
2800 EXPORT_SYMBOL(free_pages_exact);
2802 static unsigned int nr_free_zone_pages(int offset)
2807 /* Just pick one node, since fallback list is circular */
2808 unsigned int sum = 0;
2810 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
2812 for_each_zone_zonelist(zone, z, zonelist, offset) {
2813 unsigned long size = zone->managed_pages;
2814 unsigned long high = high_wmark_pages(zone);
2823 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
2825 unsigned int nr_free_buffer_pages(void)
2827 return nr_free_zone_pages(gfp_zone(GFP_USER));
2829 EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
2832 * Amount of free RAM allocatable within all zones
2834 unsigned int nr_free_pagecache_pages(void)
2836 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
2839 static inline void show_node(struct zone *zone)
2841 if (IS_ENABLED(CONFIG_NUMA))
2842 printk("Node %d ", zone_to_nid(zone));
2845 void si_meminfo(struct sysinfo *val)
2847 val->totalram = totalram_pages;
2849 val->freeram = global_page_state(NR_FREE_PAGES);
2850 val->bufferram = nr_blockdev_pages();
2851 val->totalhigh = totalhigh_pages;
2852 val->freehigh = nr_free_highpages();
2853 val->mem_unit = PAGE_SIZE;
2856 EXPORT_SYMBOL(si_meminfo);
2859 void si_meminfo_node(struct sysinfo *val, int nid)
2861 pg_data_t *pgdat = NODE_DATA(nid);
2863 val->totalram = pgdat->node_present_pages;
2864 val->freeram = node_page_state(nid, NR_FREE_PAGES);
2865 #ifdef CONFIG_HIGHMEM
2866 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].managed_pages;
2867 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
2873 val->mem_unit = PAGE_SIZE;
2878 * Determine whether the node should be displayed or not, depending on whether
2879 * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
2881 bool skip_free_areas_node(unsigned int flags, int nid)
2884 unsigned int cpuset_mems_cookie;
2886 if (!(flags & SHOW_MEM_FILTER_NODES))
2890 cpuset_mems_cookie = get_mems_allowed();
2891 ret = !node_isset(nid, cpuset_current_mems_allowed);
2892 } while (!put_mems_allowed(cpuset_mems_cookie));
2897 #define K(x) ((x) << (PAGE_SHIFT-10))
2899 static void show_migration_types(unsigned char type)
2901 static const char types[MIGRATE_TYPES] = {
2902 [MIGRATE_UNMOVABLE] = 'U',
2903 [MIGRATE_RECLAIMABLE] = 'E',
2904 [MIGRATE_MOVABLE] = 'M',
2905 [MIGRATE_RESERVE] = 'R',
2907 [MIGRATE_CMA] = 'C',
2909 #ifdef CONFIG_MEMORY_ISOLATION
2910 [MIGRATE_ISOLATE] = 'I',
2913 char tmp[MIGRATE_TYPES + 1];
2917 for (i = 0; i < MIGRATE_TYPES; i++) {
2918 if (type & (1 << i))
2923 printk("(%s) ", tmp);
2927 * Show free area list (used inside shift_scroll-lock stuff)
2928 * We also calculate the percentage fragmentation. We do this by counting the
2929 * memory on each free list with the exception of the first item on the list.
2930 * Suppresses nodes that are not allowed by current's cpuset if
2931 * SHOW_MEM_FILTER_NODES is passed.
2933 void show_free_areas(unsigned int filter)
2938 for_each_populated_zone(zone) {
2939 if (skip_free_areas_node(filter, zone_to_nid(zone)))
2942 printk("%s per-cpu:\n", zone->name);
2944 for_each_online_cpu(cpu) {
2945 struct per_cpu_pageset *pageset;
2947 pageset = per_cpu_ptr(zone->pageset, cpu);
2949 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
2950 cpu, pageset->pcp.high,
2951 pageset->pcp.batch, pageset->pcp.count);
2955 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
2956 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
2958 " dirty:%lu writeback:%lu unstable:%lu\n"
2959 " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n"
2960 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
2962 global_page_state(NR_ACTIVE_ANON),
2963 global_page_state(NR_INACTIVE_ANON),
2964 global_page_state(NR_ISOLATED_ANON),
2965 global_page_state(NR_ACTIVE_FILE),
2966 global_page_state(NR_INACTIVE_FILE),
2967 global_page_state(NR_ISOLATED_FILE),
2968 global_page_state(NR_UNEVICTABLE),
2969 global_page_state(NR_FILE_DIRTY),
2970 global_page_state(NR_WRITEBACK),
2971 global_page_state(NR_UNSTABLE_NFS),
2972 global_page_state(NR_FREE_PAGES),
2973 global_page_state(NR_SLAB_RECLAIMABLE),
2974 global_page_state(NR_SLAB_UNRECLAIMABLE),
2975 global_page_state(NR_FILE_MAPPED),
2976 global_page_state(NR_SHMEM),
2977 global_page_state(NR_PAGETABLE),
2978 global_page_state(NR_BOUNCE),
2979 global_page_state(NR_FREE_CMA_PAGES));
2981 for_each_populated_zone(zone) {
2984 if (skip_free_areas_node(filter, zone_to_nid(zone)))
2992 " active_anon:%lukB"
2993 " inactive_anon:%lukB"
2994 " active_file:%lukB"
2995 " inactive_file:%lukB"
2996 " unevictable:%lukB"
2997 " isolated(anon):%lukB"
2998 " isolated(file):%lukB"
3006 " slab_reclaimable:%lukB"
3007 " slab_unreclaimable:%lukB"
3008 " kernel_stack:%lukB"
3013 " writeback_tmp:%lukB"
3014 " pages_scanned:%lu"
3015 " all_unreclaimable? %s"
3018 K(zone_page_state(zone, NR_FREE_PAGES)),
3019 K(min_wmark_pages(zone)),
3020 K(low_wmark_pages(zone)),
3021 K(high_wmark_pages(zone)),
3022 K(zone_page_state(zone, NR_ACTIVE_ANON)),
3023 K(zone_page_state(zone, NR_INACTIVE_ANON)),
3024 K(zone_page_state(zone, NR_ACTIVE_FILE)),
3025 K(zone_page_state(zone, NR_INACTIVE_FILE)),
3026 K(zone_page_state(zone, NR_UNEVICTABLE)),
3027 K(zone_page_state(zone, NR_ISOLATED_ANON)),
3028 K(zone_page_state(zone, NR_ISOLATED_FILE)),
3029 K(zone->present_pages),
3030 K(zone->managed_pages),
3031 K(zone_page_state(zone, NR_MLOCK)),
3032 K(zone_page_state(zone, NR_FILE_DIRTY)),
3033 K(zone_page_state(zone, NR_WRITEBACK)),
3034 K(zone_page_state(zone, NR_FILE_MAPPED)),
3035 K(zone_page_state(zone, NR_SHMEM)),
3036 K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)),
3037 K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)),
3038 zone_page_state(zone, NR_KERNEL_STACK) *
3040 K(zone_page_state(zone, NR_PAGETABLE)),
3041 K(zone_page_state(zone, NR_UNSTABLE_NFS)),
3042 K(zone_page_state(zone, NR_BOUNCE)),
3043 K(zone_page_state(zone, NR_FREE_CMA_PAGES)),
3044 K(zone_page_state(zone, NR_WRITEBACK_TEMP)),
3045 zone->pages_scanned,
3046 (zone->all_unreclaimable ? "yes" : "no")
3048 printk("lowmem_reserve[]:");
3049 for (i = 0; i < MAX_NR_ZONES; i++)
3050 printk(" %lu", zone->lowmem_reserve[i]);
3054 for_each_populated_zone(zone) {
3055 unsigned long nr[MAX_ORDER], flags, order, total = 0;
3056 unsigned char types[MAX_ORDER];
3058 if (skip_free_areas_node(filter, zone_to_nid(zone)))
3061 printk("%s: ", zone->name);
3063 spin_lock_irqsave(&zone->lock, flags);
3064 for (order = 0; order < MAX_ORDER; order++) {
3065 struct free_area *area = &zone->free_area[order];
3068 nr[order] = area->nr_free;
3069 total += nr[order] << order;
3072 for (type = 0; type < MIGRATE_TYPES; type++) {
3073 if (!list_empty(&area->free_list[type]))
3074 types[order] |= 1 << type;
3077 spin_unlock_irqrestore(&zone->lock, flags);
3078 for (order = 0; order < MAX_ORDER; order++) {
3079 printk("%lu*%lukB ", nr[order], K(1UL) << order);
3081 show_migration_types(types[order]);
3083 printk("= %lukB\n", K(total));
3086 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
3088 show_swap_cache_info();
3091 static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
3093 zoneref->zone = zone;
3094 zoneref->zone_idx = zone_idx(zone);
3098 * Builds allocation fallback zone lists.
3100 * Add all populated zones of a node to the zonelist.
3102 static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
3103 int nr_zones, enum zone_type zone_type)
3107 BUG_ON(zone_type >= MAX_NR_ZONES);
3112 zone = pgdat->node_zones + zone_type;
3113 if (populated_zone(zone)) {
3114 zoneref_set_zone(zone,
3115 &zonelist->_zonerefs[nr_zones++]);
3116 check_highest_zone(zone_type);
3119 } while (zone_type);
3126 * 0 = automatic detection of better ordering.
3127 * 1 = order by ([node] distance, -zonetype)
3128 * 2 = order by (-zonetype, [node] distance)
3130 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
3131 * the same zonelist. So only NUMA can configure this param.
3133 #define ZONELIST_ORDER_DEFAULT 0
3134 #define ZONELIST_ORDER_NODE 1
3135 #define ZONELIST_ORDER_ZONE 2
3137 /* zonelist order in the kernel.
3138 * set_zonelist_order() will set this to NODE or ZONE.
3140 static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
3141 static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
3145 /* The value user specified ....changed by config */
3146 static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
3147 /* string for sysctl */
3148 #define NUMA_ZONELIST_ORDER_LEN 16
3149 char numa_zonelist_order[16] = "default";
3152 * interface for configure zonelist ordering.
3153 * command line option "numa_zonelist_order"
3154 * = "[dD]efault - default, automatic configuration.
3155 * = "[nN]ode - order by node locality, then by zone within node
3156 * = "[zZ]one - order by zone, then by locality within zone
3159 static int __parse_numa_zonelist_order(char *s)
3161 if (*s == 'd' || *s == 'D') {
3162 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
3163 } else if (*s == 'n' || *s == 'N') {
3164 user_zonelist_order = ZONELIST_ORDER_NODE;
3165 } else if (*s == 'z' || *s == 'Z') {
3166 user_zonelist_order = ZONELIST_ORDER_ZONE;
3169 "Ignoring invalid numa_zonelist_order value: "
3176 static __init int setup_numa_zonelist_order(char *s)
3183 ret = __parse_numa_zonelist_order(s);
3185 strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN);
3189 early_param("numa_zonelist_order", setup_numa_zonelist_order);
3192 * sysctl handler for numa_zonelist_order
3194 int numa_zonelist_order_handler(ctl_table *table, int write,
3195 void __user *buffer, size_t *length,
3198 char saved_string[NUMA_ZONELIST_ORDER_LEN];
3200 static DEFINE_MUTEX(zl_order_mutex);
3202 mutex_lock(&zl_order_mutex);
3204 strcpy(saved_string, (char*)table->data);
3205 ret = proc_dostring(table, write, buffer, length, ppos);
3209 int oldval = user_zonelist_order;
3210 if (__parse_numa_zonelist_order((char*)table->data)) {
3212 * bogus value. restore saved string
3214 strncpy((char*)table->data, saved_string,
3215 NUMA_ZONELIST_ORDER_LEN);
3216 user_zonelist_order = oldval;
3217 } else if (oldval != user_zonelist_order) {
3218 mutex_lock(&zonelists_mutex);
3219 build_all_zonelists(NULL, NULL);
3220 mutex_unlock(&zonelists_mutex);
3224 mutex_unlock(&zl_order_mutex);
3229 #define MAX_NODE_LOAD (nr_online_nodes)
3230 static int node_load[MAX_NUMNODES];
3233 * find_next_best_node - find the next node that should appear in a given node's fallback list
3234 * @node: node whose fallback list we're appending
3235 * @used_node_mask: nodemask_t of already used nodes
3237 * We use a number of factors to determine which is the next node that should
3238 * appear on a given node's fallback list. The node should not have appeared
3239 * already in @node's fallback list, and it should be the next closest node
3240 * according to the distance array (which contains arbitrary distance values
3241 * from each node to each node in the system), and should also prefer nodes
3242 * with no CPUs, since presumably they'll have very little allocation pressure
3243 * on them otherwise.
3244 * It returns -1 if no node is found.
3246 static int find_next_best_node(int node, nodemask_t *used_node_mask)
3249 int min_val = INT_MAX;
3251 const struct cpumask *tmp = cpumask_of_node(0);
3253 /* Use the local node if we haven't already */
3254 if (!node_isset(node, *used_node_mask)) {
3255 node_set(node, *used_node_mask);
3259 for_each_node_state(n, N_MEMORY) {
3261 /* Don't want a node to appear more than once */
3262 if (node_isset(n, *used_node_mask))
3265 /* Use the distance array to find the distance */
3266 val = node_distance(node, n);
3268 /* Penalize nodes under us ("prefer the next node") */
3271 /* Give preference to headless and unused nodes */
3272 tmp = cpumask_of_node(n);
3273 if (!cpumask_empty(tmp))
3274 val += PENALTY_FOR_NODE_WITH_CPUS;
3276 /* Slight preference for less loaded node */
3277 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
3278 val += node_load[n];
3280 if (val < min_val) {
3287 node_set(best_node, *used_node_mask);
3294 * Build zonelists ordered by node and zones within node.
3295 * This results in maximum locality--normal zone overflows into local
3296 * DMA zone, if any--but risks exhausting DMA zone.
3298 static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
3301 struct zonelist *zonelist;
3303 zonelist = &pgdat->node_zonelists[0];
3304 for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
3306 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
3308 zonelist->_zonerefs[j].zone = NULL;
3309 zonelist->_zonerefs[j].zone_idx = 0;
3313 * Build gfp_thisnode zonelists
3315 static void build_thisnode_zonelists(pg_data_t *pgdat)
3318 struct zonelist *zonelist;
3320 zonelist = &pgdat->node_zonelists[1];
3321 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
3322 zonelist->_zonerefs[j].zone = NULL;
3323 zonelist->_zonerefs[j].zone_idx = 0;
3327 * Build zonelists ordered by zone and nodes within zones.
3328 * This results in conserving DMA zone[s] until all Normal memory is
3329 * exhausted, but results in overflowing to remote node while memory
3330 * may still exist in local DMA zone.
3332 static int node_order[MAX_NUMNODES];
3334 static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
3337 int zone_type; /* needs to be signed */
3339 struct zonelist *zonelist;
3341 zonelist = &pgdat->node_zonelists[0];
3343 for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
3344 for (j = 0; j < nr_nodes; j++) {
3345 node = node_order[j];
3346 z = &NODE_DATA(node)->node_zones[zone_type];
3347 if (populated_zone(z)) {
3349 &zonelist->_zonerefs[pos++]);
3350 check_highest_zone(zone_type);
3354 zonelist->_zonerefs[pos].zone = NULL;
3355 zonelist->_zonerefs[pos].zone_idx = 0;
3358 static int default_zonelist_order(void)
3361 unsigned long low_kmem_size,total_size;
3365 * ZONE_DMA and ZONE_DMA32 can be very small area in the system.
3366 * If they are really small and used heavily, the system can fall
3367 * into OOM very easily.
3368 * This function detect ZONE_DMA/DMA32 size and configures zone order.
3370 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
3373 for_each_online_node(nid) {
3374 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
3375 z = &NODE_DATA(nid)->node_zones[zone_type];
3376 if (populated_zone(z)) {
3377 if (zone_type < ZONE_NORMAL)
3378 low_kmem_size += z->present_pages;
3379 total_size += z->present_pages;
3380 } else if (zone_type == ZONE_NORMAL) {
3382 * If any node has only lowmem, then node order
3383 * is preferred to allow kernel allocations
3384 * locally; otherwise, they can easily infringe
3385 * on other nodes when there is an abundance of
3386 * lowmem available to allocate from.
3388 return ZONELIST_ORDER_NODE;
3392 if (!low_kmem_size || /* there are no DMA area. */
3393 low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
3394 return ZONELIST_ORDER_NODE;
3396 * look into each node's config.
3397 * If there is a node whose DMA/DMA32 memory is very big area on
3398 * local memory, NODE_ORDER may be suitable.
3400 average_size = total_size /
3401 (nodes_weight(node_states[N_MEMORY]) + 1);
3402 for_each_online_node(nid) {
3405 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
3406 z = &NODE_DATA(nid)->node_zones[zone_type];
3407 if (populated_zone(z)) {
3408 if (zone_type < ZONE_NORMAL)
3409 low_kmem_size += z->present_pages;
3410 total_size += z->present_pages;
3413 if (low_kmem_size &&
3414 total_size > average_size && /* ignore small node */
3415 low_kmem_size > total_size * 70/100)
3416 return ZONELIST_ORDER_NODE;
3418 return ZONELIST_ORDER_ZONE;
3421 static void set_zonelist_order(void)
3423 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
3424 current_zonelist_order = default_zonelist_order();
3426 current_zonelist_order = user_zonelist_order;
3429 static void build_zonelists(pg_data_t *pgdat)
3433 nodemask_t used_mask;
3434 int local_node, prev_node;
3435 struct zonelist *zonelist;
3436 int order = current_zonelist_order;
3438 /* initialize zonelists */
3439 for (i = 0; i < MAX_ZONELISTS; i++) {
3440 zonelist = pgdat->node_zonelists + i;
3441 zonelist->_zonerefs[0].zone = NULL;
3442 zonelist->_zonerefs[0].zone_idx = 0;
3445 /* NUMA-aware ordering of nodes */
3446 local_node = pgdat->node_id;
3447 load = nr_online_nodes;
3448 prev_node = local_node;
3449 nodes_clear(used_mask);
3451 memset(node_order, 0, sizeof(node_order));
3454 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
3456 * We don't want to pressure a particular node.
3457 * So adding penalty to the first node in same
3458 * distance group to make it round-robin.
3460 if (node_distance(local_node, node) !=
3461 node_distance(local_node, prev_node))
3462 node_load[node] = load;
3466 if (order == ZONELIST_ORDER_NODE)
3467 build_zonelists_in_node_order(pgdat, node);
3469 node_order[j++] = node; /* remember order */
3472 if (order == ZONELIST_ORDER_ZONE) {
3473 /* calculate node order -- i.e., DMA last! */
3474 build_zonelists_in_zone_order(pgdat, j);
3477 build_thisnode_zonelists(pgdat);
3480 /* Construct the zonelist performance cache - see further mmzone.h */
3481 static void build_zonelist_cache(pg_data_t *pgdat)
3483 struct zonelist *zonelist;
3484 struct zonelist_cache *zlc;
3487 zonelist = &pgdat->node_zonelists[0];
3488 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
3489 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
3490 for (z = zonelist->_zonerefs; z->zone; z++)
3491 zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z);
3494 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
3496 * Return node id of node used for "local" allocations.
3497 * I.e., first node id of first zone in arg node's generic zonelist.
3498 * Used for initializing percpu 'numa_mem', which is used primarily
3499 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
3501 int local_memory_node(int node)
3505 (void)first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
3506 gfp_zone(GFP_KERNEL),
3513 #else /* CONFIG_NUMA */
3515 static void set_zonelist_order(void)
3517 current_zonelist_order = ZONELIST_ORDER_ZONE;
3520 static void build_zonelists(pg_data_t *pgdat)
3522 int node, local_node;
3524 struct zonelist *zonelist;
3526 local_node = pgdat->node_id;
3528 zonelist = &pgdat->node_zonelists[0];
3529 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
3532 * Now we build the zonelist so that it contains the zones
3533 * of all the other nodes.
3534 * We don't want to pressure a particular node, so when
3535 * building the zones for node N, we make sure that the
3536 * zones coming right after the local ones are those from
3537 * node N+1 (modulo N)
3539 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
3540 if (!node_online(node))
3542 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
3545 for (node = 0; node < local_node; node++) {
3546 if (!node_online(node))
3548 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
3552 zonelist->_zonerefs[j].zone = NULL;
3553 zonelist->_zonerefs[j].zone_idx = 0;
3556 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
3557 static void build_zonelist_cache(pg_data_t *pgdat)
3559 pgdat->node_zonelists[0].zlcache_ptr = NULL;
3562 #endif /* CONFIG_NUMA */
3565 * Boot pageset table. One per cpu which is going to be used for all
3566 * zones and all nodes. The parameters will be set in such a way
3567 * that an item put on a list will immediately be handed over to
3568 * the buddy list. This is safe since pageset manipulation is done
3569 * with interrupts disabled.
3571 * The boot_pagesets must be kept even after bootup is complete for
3572 * unused processors and/or zones. They do play a role for bootstrapping
3573 * hotplugged processors.
3575 * zoneinfo_show() and maybe other functions do
3576 * not check if the processor is online before following the pageset pointer.
3577 * Other parts of the kernel may not check if the zone is available.
3579 static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch);
3580 static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset);
3581 static void setup_zone_pageset(struct zone *zone);
3584 * Global mutex to protect against size modification of zonelists
3585 * as well as to serialize pageset setup for the new populated zone.
3587 DEFINE_MUTEX(zonelists_mutex);
3589 /* return values int ....just for stop_machine() */
3590 static int __build_all_zonelists(void *data)
3594 pg_data_t *self = data;
3597 memset(node_load, 0, sizeof(node_load));
3600 if (self && !node_online(self->node_id)) {
3601 build_zonelists(self);
3602 build_zonelist_cache(self);
3605 for_each_online_node(nid) {
3606 pg_data_t *pgdat = NODE_DATA(nid);
3608 build_zonelists(pgdat);
3609 build_zonelist_cache(pgdat);
3613 * Initialize the boot_pagesets that are going to be used
3614 * for bootstrapping processors. The real pagesets for
3615 * each zone will be allocated later when the per cpu
3616 * allocator is available.
3618 * boot_pagesets are used also for bootstrapping offline
3619 * cpus if the system is already booted because the pagesets
3620 * are needed to initialize allocators on a specific cpu too.
3621 * F.e. the percpu allocator needs the page allocator which
3622 * needs the percpu allocator in order to allocate its pagesets
3623 * (a chicken-egg dilemma).
3625 for_each_possible_cpu(cpu) {
3626 setup_pageset(&per_cpu(boot_pageset, cpu), 0);
3628 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
3630 * We now know the "local memory node" for each node--
3631 * i.e., the node of the first zone in the generic zonelist.
3632 * Set up numa_mem percpu variable for on-line cpus. During
3633 * boot, only the boot cpu should be on-line; we'll init the
3634 * secondary cpus' numa_mem as they come on-line. During
3635 * node/memory hotplug, we'll fixup all on-line cpus.
3637 if (cpu_online(cpu))
3638 set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
3646 * Called with zonelists_mutex held always
3647 * unless system_state == SYSTEM_BOOTING.
3649 void __ref build_all_zonelists(pg_data_t *pgdat, struct zone *zone)
3651 set_zonelist_order();
3653 if (system_state == SYSTEM_BOOTING) {
3654 __build_all_zonelists(NULL);
3655 mminit_verify_zonelist();
3656 cpuset_init_current_mems_allowed();
3658 /* we have to stop all cpus to guarantee there is no user
3660 #ifdef CONFIG_MEMORY_HOTPLUG
3662 setup_zone_pageset(zone);
3664 stop_machine(__build_all_zonelists, pgdat, NULL);
3665 /* cpuset refresh routine should be here */
3667 vm_total_pages = nr_free_pagecache_pages();
3669 * Disable grouping by mobility if the number of pages in the
3670 * system is too low to allow the mechanism to work. It would be
3671 * more accurate, but expensive to check per-zone. This check is
3672 * made on memory-hotadd so a system can start with mobility
3673 * disabled and enable it later
3675 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
3676 page_group_by_mobility_disabled = 1;
3678 page_group_by_mobility_disabled = 0;
3680 printk("Built %i zonelists in %s order, mobility grouping %s. "
3681 "Total pages: %ld\n",
3683 zonelist_order_name[current_zonelist_order],
3684 page_group_by_mobility_disabled ? "off" : "on",
3687 printk("Policy zone: %s\n", zone_names[policy_zone]);
3692 * Helper functions to size the waitqueue hash table.
3693 * Essentially these want to choose hash table sizes sufficiently
3694 * large so that collisions trying to wait on pages are rare.
3695 * But in fact, the number of active page waitqueues on typical
3696 * systems is ridiculously low, less than 200. So this is even
3697 * conservative, even though it seems large.
3699 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
3700 * waitqueues, i.e. the size of the waitq table given the number of pages.
3702 #define PAGES_PER_WAITQUEUE 256
3704 #ifndef CONFIG_MEMORY_HOTPLUG
3705 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
3707 unsigned long size = 1;
3709 pages /= PAGES_PER_WAITQUEUE;
3711 while (size < pages)
3715 * Once we have dozens or even hundreds of threads sleeping
3716 * on IO we've got bigger problems than wait queue collision.
3717 * Limit the size of the wait table to a reasonable size.
3719 size = min(size, 4096UL);
3721 return max(size, 4UL);
3725 * A zone's size might be changed by hot-add, so it is not possible to determine
3726 * a suitable size for its wait_table. So we use the maximum size now.
3728 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
3730 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
3731 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
3732 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
3734 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
3735 * or more by the traditional way. (See above). It equals:
3737 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
3738 * ia64(16K page size) : = ( 8G + 4M)byte.
3739 * powerpc (64K page size) : = (32G +16M)byte.
3741 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
3748 * This is an integer logarithm so that shifts can be used later
3749 * to extract the more random high bits from the multiplicative
3750 * hash function before the remainder is taken.
3752 static inline unsigned long wait_table_bits(unsigned long size)
3757 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
3760 * Check if a pageblock contains reserved pages
3762 static int pageblock_is_reserved(unsigned long start_pfn, unsigned long end_pfn)
3766 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
3767 if (!pfn_valid_within(pfn) || PageReserved(pfn_to_page(pfn)))
3774 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
3775 * of blocks reserved is based on min_wmark_pages(zone). The memory within
3776 * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
3777 * higher will lead to a bigger reserve which will get freed as contiguous
3778 * blocks as reclaim kicks in
3780 static void setup_zone_migrate_reserve(struct zone *zone)
3782 unsigned long start_pfn, pfn, end_pfn, block_end_pfn;
3784 unsigned long block_migratetype;
3788 * Get the start pfn, end pfn and the number of blocks to reserve
3789 * We have to be careful to be aligned to pageblock_nr_pages to
3790 * make sure that we always check pfn_valid for the first page in
3793 start_pfn = zone->zone_start_pfn;
3794 end_pfn = start_pfn + zone->spanned_pages;
3795 start_pfn = roundup(start_pfn, pageblock_nr_pages);
3796 reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >>
3800 * Reserve blocks are generally in place to help high-order atomic
3801 * allocations that are short-lived. A min_free_kbytes value that
3802 * would result in more than 2 reserve blocks for atomic allocations
3803 * is assumed to be in place to help anti-fragmentation for the
3804 * future allocation of hugepages at runtime.
3806 reserve = min(2, reserve);
3808 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
3809 if (!pfn_valid(pfn))
3811 page = pfn_to_page(pfn);
3813 /* Watch out for overlapping nodes */
3814 if (page_to_nid(page) != zone_to_nid(zone))
3817 block_migratetype = get_pageblock_migratetype(page);
3819 /* Only test what is necessary when the reserves are not met */
3822 * Blocks with reserved pages will never free, skip
3825 block_end_pfn = min(pfn + pageblock_nr_pages, end_pfn);
3826 if (pageblock_is_reserved(pfn, block_end_pfn))
3829 /* If this block is reserved, account for it */
3830 if (block_migratetype == MIGRATE_RESERVE) {
3835 /* Suitable for reserving if this block is movable */
3836 if (block_migratetype == MIGRATE_MOVABLE) {
3837 set_pageblock_migratetype(page,
3839 move_freepages_block(zone, page,
3847 * If the reserve is met and this is a previous reserved block,
3850 if (block_migratetype == MIGRATE_RESERVE) {
3851 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
3852 move_freepages_block(zone, page, MIGRATE_MOVABLE);
3858 * Initially all pages are reserved - free ones are freed
3859 * up by free_all_bootmem() once the early boot process is
3860 * done. Non-atomic initialization, single-pass.
3862 void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
3863 unsigned long start_pfn, enum memmap_context context)
3866 unsigned long end_pfn = start_pfn + size;
3870 if (highest_memmap_pfn < end_pfn - 1)
3871 highest_memmap_pfn = end_pfn - 1;
3873 z = &NODE_DATA(nid)->node_zones[zone];
3874 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
3876 * There can be holes in boot-time mem_map[]s
3877 * handed to this function. They do not
3878 * exist on hotplugged memory.
3880 if (context == MEMMAP_EARLY) {
3881 if (!early_pfn_valid(pfn))
3883 if (!early_pfn_in_nid(pfn, nid))
3886 page = pfn_to_page(pfn);
3887 set_page_links(page, zone, nid, pfn);
3888 mminit_verify_page_links(page, zone, nid, pfn);
3889 init_page_count(page);
3890 reset_page_mapcount(page);
3891 reset_page_last_nid(page);
3892 SetPageReserved(page);
3894 * Mark the block movable so that blocks are reserved for
3895 * movable at startup. This will force kernel allocations
3896 * to reserve their blocks rather than leaking throughout
3897 * the address space during boot when many long-lived
3898 * kernel allocations are made. Later some blocks near
3899 * the start are marked MIGRATE_RESERVE by
3900 * setup_zone_migrate_reserve()
3902 * bitmap is created for zone's valid pfn range. but memmap
3903 * can be created for invalid pages (for alignment)
3904 * check here not to call set_pageblock_migratetype() against
3907 if ((z->zone_start_pfn <= pfn)
3908 && (pfn < z->zone_start_pfn + z->spanned_pages)
3909 && !(pfn & (pageblock_nr_pages - 1)))
3910 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
3912 INIT_LIST_HEAD(&page->lru);
3913 #ifdef WANT_PAGE_VIRTUAL
3914 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
3915 if (!is_highmem_idx(zone))
3916 set_page_address(page, __va(pfn << PAGE_SHIFT));
3921 static void __meminit zone_init_free_lists(struct zone *zone)
3924 for_each_migratetype_order(order, t) {
3925 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
3926 zone->free_area[order].nr_free = 0;
3930 #ifndef __HAVE_ARCH_MEMMAP_INIT
3931 #define memmap_init(size, nid, zone, start_pfn) \
3932 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
3935 static int __meminit zone_batchsize(struct zone *zone)
3941 * The per-cpu-pages pools are set to around 1000th of the
3942 * size of the zone. But no more than 1/2 of a meg.
3944 * OK, so we don't know how big the cache is. So guess.
3946 batch = zone->managed_pages / 1024;
3947 if (batch * PAGE_SIZE > 512 * 1024)
3948 batch = (512 * 1024) / PAGE_SIZE;
3949 batch /= 4; /* We effectively *= 4 below */
3954 * Clamp the batch to a 2^n - 1 value. Having a power
3955 * of 2 value was found to be more likely to have
3956 * suboptimal cache aliasing properties in some cases.
3958 * For example if 2 tasks are alternately allocating
3959 * batches of pages, one task can end up with a lot
3960 * of pages of one half of the possible page colors
3961 * and the other with pages of the other colors.
3963 batch = rounddown_pow_of_two(batch + batch/2) - 1;
3968 /* The deferral and batching of frees should be suppressed under NOMMU
3971 * The problem is that NOMMU needs to be able to allocate large chunks
3972 * of contiguous memory as there's no hardware page translation to
3973 * assemble apparent contiguous memory from discontiguous pages.
3975 * Queueing large contiguous runs of pages for batching, however,
3976 * causes the pages to actually be freed in smaller chunks. As there
3977 * can be a significant delay between the individual batches being
3978 * recycled, this leads to the once large chunks of space being
3979 * fragmented and becoming unavailable for high-order allocations.
3985 static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
3987 struct per_cpu_pages *pcp;
3990 memset(p, 0, sizeof(*p));
3994 pcp->high = 6 * batch;
3995 pcp->batch = max(1UL, 1 * batch);
3996 for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
3997 INIT_LIST_HEAD(&pcp->lists[migratetype]);
4001 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
4002 * to the value high for the pageset p.
4005 static void setup_pagelist_highmark(struct per_cpu_pageset *p,
4008 struct per_cpu_pages *pcp;
4012 pcp->batch = max(1UL, high/4);
4013 if ((high/4) > (PAGE_SHIFT * 8))
4014 pcp->batch = PAGE_SHIFT * 8;
4017 static void __meminit setup_zone_pageset(struct zone *zone)
4021 zone->pageset = alloc_percpu(struct per_cpu_pageset);
4023 for_each_possible_cpu(cpu) {
4024 struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu);
4026 setup_pageset(pcp, zone_batchsize(zone));
4028 if (percpu_pagelist_fraction)
4029 setup_pagelist_highmark(pcp,
4030 (zone->managed_pages /
4031 percpu_pagelist_fraction));
4036 * Allocate per cpu pagesets and initialize them.
4037 * Before this call only boot pagesets were available.
4039 void __init setup_per_cpu_pageset(void)
4043 for_each_populated_zone(zone)
4044 setup_zone_pageset(zone);
4047 static noinline __init_refok
4048 int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
4051 struct pglist_data *pgdat = zone->zone_pgdat;
4055 * The per-page waitqueue mechanism uses hashed waitqueues
4058 zone->wait_table_hash_nr_entries =
4059 wait_table_hash_nr_entries(zone_size_pages);
4060 zone->wait_table_bits =
4061 wait_table_bits(zone->wait_table_hash_nr_entries);
4062 alloc_size = zone->wait_table_hash_nr_entries
4063 * sizeof(wait_queue_head_t);
4065 if (!slab_is_available()) {
4066 zone->wait_table = (wait_queue_head_t *)
4067 alloc_bootmem_node_nopanic(pgdat, alloc_size);
4070 * This case means that a zone whose size was 0 gets new memory
4071 * via memory hot-add.
4072 * But it may be the case that a new node was hot-added. In
4073 * this case vmalloc() will not be able to use this new node's
4074 * memory - this wait_table must be initialized to use this new
4075 * node itself as well.
4076 * To use this new node's memory, further consideration will be
4079 zone->wait_table = vmalloc(alloc_size);
4081 if (!zone->wait_table)
4084 for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
4085 init_waitqueue_head(zone->wait_table + i);
4090 static __meminit void zone_pcp_init(struct zone *zone)
4093 * per cpu subsystem is not up at this point. The following code
4094 * relies on the ability of the linker to provide the
4095 * offset of a (static) per cpu variable into the per cpu area.
4097 zone->pageset = &boot_pageset;
4099 if (zone->present_pages)
4100 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n",
4101 zone->name, zone->present_pages,
4102 zone_batchsize(zone));
4105 int __meminit init_currently_empty_zone(struct zone *zone,
4106 unsigned long zone_start_pfn,
4108 enum memmap_context context)
4110 struct pglist_data *pgdat = zone->zone_pgdat;
4112 ret = zone_wait_table_init(zone, size);
4115 pgdat->nr_zones = zone_idx(zone) + 1;
4117 zone->zone_start_pfn = zone_start_pfn;
4119 mminit_dprintk(MMINIT_TRACE, "memmap_init",
4120 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
4122 (unsigned long)zone_idx(zone),
4123 zone_start_pfn, (zone_start_pfn + size));
4125 zone_init_free_lists(zone);
4130 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4131 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
4133 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
4134 * Architectures may implement their own version but if add_active_range()
4135 * was used and there are no special requirements, this is a convenient
4138 int __meminit __early_pfn_to_nid(unsigned long pfn)
4140 unsigned long start_pfn, end_pfn;
4143 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid)
4144 if (start_pfn <= pfn && pfn < end_pfn)
4146 /* This is a memory hole */
4149 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
4151 int __meminit early_pfn_to_nid(unsigned long pfn)
4155 nid = __early_pfn_to_nid(pfn);
4158 /* just returns 0 */
4162 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
4163 bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
4167 nid = __early_pfn_to_nid(pfn);
4168 if (nid >= 0 && nid != node)
4175 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
4176 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
4177 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
4179 * If an architecture guarantees that all ranges registered with
4180 * add_active_ranges() contain no holes and may be freed, this
4181 * this function may be used instead of calling free_bootmem() manually.
4183 void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn)
4185 unsigned long start_pfn, end_pfn;
4188 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) {
4189 start_pfn = min(start_pfn, max_low_pfn);
4190 end_pfn = min(end_pfn, max_low_pfn);
4192 if (start_pfn < end_pfn)
4193 free_bootmem_node(NODE_DATA(this_nid),
4194 PFN_PHYS(start_pfn),
4195 (end_pfn - start_pfn) << PAGE_SHIFT);
4200 * sparse_memory_present_with_active_regions - Call memory_present for each active range
4201 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
4203 * If an architecture guarantees that all ranges registered with
4204 * add_active_ranges() contain no holes and may be freed, this
4205 * function may be used instead of calling memory_present() manually.
4207 void __init sparse_memory_present_with_active_regions(int nid)
4209 unsigned long start_pfn, end_pfn;
4212 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid)
4213 memory_present(this_nid, start_pfn, end_pfn);
4217 * get_pfn_range_for_nid - Return the start and end page frames for a node
4218 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
4219 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
4220 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
4222 * It returns the start and end page frame of a node based on information
4223 * provided by an arch calling add_active_range(). If called for a node
4224 * with no available memory, a warning is printed and the start and end
4227 void __meminit get_pfn_range_for_nid(unsigned int nid,
4228 unsigned long *start_pfn, unsigned long *end_pfn)
4230 unsigned long this_start_pfn, this_end_pfn;
4236 for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) {
4237 *start_pfn = min(*start_pfn, this_start_pfn);
4238 *end_pfn = max(*end_pfn, this_end_pfn);
4241 if (*start_pfn == -1UL)
4246 * This finds a zone that can be used for ZONE_MOVABLE pages. The
4247 * assumption is made that zones within a node are ordered in monotonic
4248 * increasing memory addresses so that the "highest" populated zone is used
4250 static void __init find_usable_zone_for_movable(void)
4253 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
4254 if (zone_index == ZONE_MOVABLE)
4257 if (arch_zone_highest_possible_pfn[zone_index] >
4258 arch_zone_lowest_possible_pfn[zone_index])
4262 VM_BUG_ON(zone_index == -1);
4263 movable_zone = zone_index;
4267 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
4268 * because it is sized independent of architecture. Unlike the other zones,
4269 * the starting point for ZONE_MOVABLE is not fixed. It may be different
4270 * in each node depending on the size of each node and how evenly kernelcore
4271 * is distributed. This helper function adjusts the zone ranges
4272 * provided by the architecture for a given node by using the end of the
4273 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
4274 * zones within a node are in order of monotonic increases memory addresses
4276 static void __meminit adjust_zone_range_for_zone_movable(int nid,
4277 unsigned long zone_type,
4278 unsigned long node_start_pfn,
4279 unsigned long node_end_pfn,
4280 unsigned long *zone_start_pfn,
4281 unsigned long *zone_end_pfn)
4283 /* Only adjust if ZONE_MOVABLE is on this node */
4284 if (zone_movable_pfn[nid]) {
4285 /* Size ZONE_MOVABLE */
4286 if (zone_type == ZONE_MOVABLE) {
4287 *zone_start_pfn = zone_movable_pfn[nid];
4288 *zone_end_pfn = min(node_end_pfn,
4289 arch_zone_highest_possible_pfn[movable_zone]);
4291 /* Adjust for ZONE_MOVABLE starting within this range */
4292 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
4293 *zone_end_pfn > zone_movable_pfn[nid]) {
4294 *zone_end_pfn = zone_movable_pfn[nid];
4296 /* Check if this whole range is within ZONE_MOVABLE */
4297 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
4298 *zone_start_pfn = *zone_end_pfn;
4303 * Return the number of pages a zone spans in a node, including holes
4304 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
4306 static unsigned long __meminit zone_spanned_pages_in_node(int nid,
4307 unsigned long zone_type,
4308 unsigned long *ignored)
4310 unsigned long node_start_pfn, node_end_pfn;
4311 unsigned long zone_start_pfn, zone_end_pfn;
4313 /* Get the start and end of the node and zone */
4314 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
4315 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
4316 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
4317 adjust_zone_range_for_zone_movable(nid, zone_type,
4318 node_start_pfn, node_end_pfn,
4319 &zone_start_pfn, &zone_end_pfn);
4321 /* Check that this node has pages within the zone's required range */
4322 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
4325 /* Move the zone boundaries inside the node if necessary */
4326 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
4327 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
4329 /* Return the spanned pages */
4330 return zone_end_pfn - zone_start_pfn;
4334 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
4335 * then all holes in the requested range will be accounted for.
4337 unsigned long __meminit __absent_pages_in_range(int nid,
4338 unsigned long range_start_pfn,
4339 unsigned long range_end_pfn)
4341 unsigned long nr_absent = range_end_pfn - range_start_pfn;
4342 unsigned long start_pfn, end_pfn;
4345 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
4346 start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn);
4347 end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn);
4348 nr_absent -= end_pfn - start_pfn;
4354 * absent_pages_in_range - Return number of page frames in holes within a range
4355 * @start_pfn: The start PFN to start searching for holes
4356 * @end_pfn: The end PFN to stop searching for holes
4358 * It returns the number of pages frames in memory holes within a range.
4360 unsigned long __init absent_pages_in_range(unsigned long start_pfn,
4361 unsigned long end_pfn)
4363 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
4366 /* Return the number of page frames in holes in a zone on a node */
4367 static unsigned long __meminit zone_absent_pages_in_node(int nid,
4368 unsigned long zone_type,
4369 unsigned long *ignored)
4371 unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type];
4372 unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type];
4373 unsigned long node_start_pfn, node_end_pfn;
4374 unsigned long zone_start_pfn, zone_end_pfn;
4376 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
4377 zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high);
4378 zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high);
4380 adjust_zone_range_for_zone_movable(nid, zone_type,
4381 node_start_pfn, node_end_pfn,
4382 &zone_start_pfn, &zone_end_pfn);
4383 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
4387 * sanitize_zone_movable_limit - Sanitize the zone_movable_limit array.
4389 * zone_movable_limit is initialized as 0. This function will try to get
4390 * the first ZONE_MOVABLE pfn of each node from movablemem_map, and
4391 * assigne them to zone_movable_limit.
4392 * zone_movable_limit[nid] == 0 means no limit for the node.
4394 * Note: Each range is represented as [start_pfn, end_pfn)
4396 static void __meminit sanitize_zone_movable_limit(void)
4398 int map_pos = 0, i, nid;
4399 unsigned long start_pfn, end_pfn;
4401 if (!movablemem_map.nr_map)
4404 /* Iterate all ranges from minimum to maximum */
4405 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
4407 * If we have found lowest pfn of ZONE_MOVABLE of the node
4408 * specified by user, just go on to check next range.
4410 if (zone_movable_limit[nid])
4413 #ifdef CONFIG_ZONE_DMA
4414 /* Skip DMA memory. */
4415 if (start_pfn < arch_zone_highest_possible_pfn[ZONE_DMA])
4416 start_pfn = arch_zone_highest_possible_pfn[ZONE_DMA];
4419 #ifdef CONFIG_ZONE_DMA32
4420 /* Skip DMA32 memory. */
4421 if (start_pfn < arch_zone_highest_possible_pfn[ZONE_DMA32])
4422 start_pfn = arch_zone_highest_possible_pfn[ZONE_DMA32];
4425 #ifdef CONFIG_HIGHMEM
4426 /* Skip lowmem if ZONE_MOVABLE is highmem. */
4427 if (zone_movable_is_highmem() &&
4428 start_pfn < arch_zone_lowest_possible_pfn[ZONE_HIGHMEM])
4429 start_pfn = arch_zone_lowest_possible_pfn[ZONE_HIGHMEM];
4432 if (start_pfn >= end_pfn)
4435 while (map_pos < movablemem_map.nr_map) {
4436 if (end_pfn <= movablemem_map.map[map_pos].start_pfn)
4439 if (start_pfn >= movablemem_map.map[map_pos].end_pfn) {
4445 * The start_pfn of ZONE_MOVABLE is either the minimum
4446 * pfn specified by movablemem_map, or 0, which means
4447 * the node has no ZONE_MOVABLE.
4449 zone_movable_limit[nid] = max(start_pfn,
4450 movablemem_map.map[map_pos].start_pfn);
4457 #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
4458 static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
4459 unsigned long zone_type,
4460 unsigned long *zones_size)
4462 return zones_size[zone_type];
4465 static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
4466 unsigned long zone_type,
4467 unsigned long *zholes_size)
4472 return zholes_size[zone_type];
4474 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
4476 static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
4477 unsigned long *zones_size, unsigned long *zholes_size)
4479 unsigned long realtotalpages, totalpages = 0;
4482 for (i = 0; i < MAX_NR_ZONES; i++)
4483 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
4485 pgdat->node_spanned_pages = totalpages;
4487 realtotalpages = totalpages;
4488 for (i = 0; i < MAX_NR_ZONES; i++)
4490 zone_absent_pages_in_node(pgdat->node_id, i,
4492 pgdat->node_present_pages = realtotalpages;
4493 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
4497 #ifndef CONFIG_SPARSEMEM
4499 * Calculate the size of the zone->blockflags rounded to an unsigned long
4500 * Start by making sure zonesize is a multiple of pageblock_order by rounding
4501 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
4502 * round what is now in bits to nearest long in bits, then return it in
4505 static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize)
4507 unsigned long usemapsize;
4509 zonesize += zone_start_pfn & (pageblock_nr_pages-1);
4510 usemapsize = roundup(zonesize, pageblock_nr_pages);
4511 usemapsize = usemapsize >> pageblock_order;
4512 usemapsize *= NR_PAGEBLOCK_BITS;
4513 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
4515 return usemapsize / 8;
4518 static void __init setup_usemap(struct pglist_data *pgdat,
4520 unsigned long zone_start_pfn,
4521 unsigned long zonesize)
4523 unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize);
4524 zone->pageblock_flags = NULL;
4526 zone->pageblock_flags = alloc_bootmem_node_nopanic(pgdat,
4530 static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone,
4531 unsigned long zone_start_pfn, unsigned long zonesize) {}
4532 #endif /* CONFIG_SPARSEMEM */
4534 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
4536 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
4537 void __init set_pageblock_order(void)
4541 /* Check that pageblock_nr_pages has not already been setup */
4542 if (pageblock_order)
4545 if (HPAGE_SHIFT > PAGE_SHIFT)
4546 order = HUGETLB_PAGE_ORDER;
4548 order = MAX_ORDER - 1;
4551 * Assume the largest contiguous order of interest is a huge page.
4552 * This value may be variable depending on boot parameters on IA64 and
4555 pageblock_order = order;
4557 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4560 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
4561 * is unused as pageblock_order is set at compile-time. See
4562 * include/linux/pageblock-flags.h for the values of pageblock_order based on
4565 void __init set_pageblock_order(void)
4569 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4571 static unsigned long __paginginit calc_memmap_size(unsigned long spanned_pages,
4572 unsigned long present_pages)
4574 unsigned long pages = spanned_pages;
4577 * Provide a more accurate estimation if there are holes within
4578 * the zone and SPARSEMEM is in use. If there are holes within the
4579 * zone, each populated memory region may cost us one or two extra
4580 * memmap pages due to alignment because memmap pages for each
4581 * populated regions may not naturally algined on page boundary.
4582 * So the (present_pages >> 4) heuristic is a tradeoff for that.
4584 if (spanned_pages > present_pages + (present_pages >> 4) &&
4585 IS_ENABLED(CONFIG_SPARSEMEM))
4586 pages = present_pages;
4588 return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT;
4592 * Set up the zone data structures:
4593 * - mark all pages reserved
4594 * - mark all memory queues empty
4595 * - clear the memory bitmaps
4597 * NOTE: pgdat should get zeroed by caller.
4599 static void __paginginit free_area_init_core(struct pglist_data *pgdat,
4600 unsigned long *zones_size, unsigned long *zholes_size)
4603 int nid = pgdat->node_id;
4604 unsigned long zone_start_pfn = pgdat->node_start_pfn;
4607 pgdat_resize_init(pgdat);
4608 #ifdef CONFIG_NUMA_BALANCING
4609 spin_lock_init(&pgdat->numabalancing_migrate_lock);
4610 pgdat->numabalancing_migrate_nr_pages = 0;
4611 pgdat->numabalancing_migrate_next_window = jiffies;
4613 init_waitqueue_head(&pgdat->kswapd_wait);
4614 init_waitqueue_head(&pgdat->pfmemalloc_wait);
4615 pgdat_page_cgroup_init(pgdat);
4617 for (j = 0; j < MAX_NR_ZONES; j++) {
4618 struct zone *zone = pgdat->node_zones + j;
4619 unsigned long size, realsize, freesize, memmap_pages;
4621 size = zone_spanned_pages_in_node(nid, j, zones_size);
4622 realsize = freesize = size - zone_absent_pages_in_node(nid, j,
4626 * Adjust freesize so that it accounts for how much memory
4627 * is used by this zone for memmap. This affects the watermark
4628 * and per-cpu initialisations
4630 memmap_pages = calc_memmap_size(size, realsize);
4631 if (freesize >= memmap_pages) {
4632 freesize -= memmap_pages;
4635 " %s zone: %lu pages used for memmap\n",
4636 zone_names[j], memmap_pages);
4639 " %s zone: %lu pages exceeds freesize %lu\n",
4640 zone_names[j], memmap_pages, freesize);
4642 /* Account for reserved pages */
4643 if (j == 0 && freesize > dma_reserve) {
4644 freesize -= dma_reserve;
4645 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
4646 zone_names[0], dma_reserve);
4649 if (!is_highmem_idx(j))
4650 nr_kernel_pages += freesize;
4651 /* Charge for highmem memmap if there are enough kernel pages */
4652 else if (nr_kernel_pages > memmap_pages * 2)
4653 nr_kernel_pages -= memmap_pages;
4654 nr_all_pages += freesize;
4656 zone->spanned_pages = size;
4657 zone->present_pages = realsize;
4659 * Set an approximate value for lowmem here, it will be adjusted
4660 * when the bootmem allocator frees pages into the buddy system.
4661 * And all highmem pages will be managed by the buddy system.
4663 zone->managed_pages = is_highmem_idx(j) ? realsize : freesize;
4666 zone->min_unmapped_pages = (freesize*sysctl_min_unmapped_ratio)
4668 zone->min_slab_pages = (freesize * sysctl_min_slab_ratio) / 100;
4670 zone->name = zone_names[j];
4671 spin_lock_init(&zone->lock);
4672 spin_lock_init(&zone->lru_lock);
4673 zone_seqlock_init(zone);
4674 zone->zone_pgdat = pgdat;
4676 zone_pcp_init(zone);
4677 lruvec_init(&zone->lruvec);
4681 set_pageblock_order();
4682 setup_usemap(pgdat, zone, zone_start_pfn, size);
4683 ret = init_currently_empty_zone(zone, zone_start_pfn,
4684 size, MEMMAP_EARLY);
4686 memmap_init(size, nid, j, zone_start_pfn);
4687 zone_start_pfn += size;
4691 static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
4693 /* Skip empty nodes */
4694 if (!pgdat->node_spanned_pages)
4697 #ifdef CONFIG_FLAT_NODE_MEM_MAP
4698 /* ia64 gets its own node_mem_map, before this, without bootmem */
4699 if (!pgdat->node_mem_map) {
4700 unsigned long size, start, end;
4704 * The zone's endpoints aren't required to be MAX_ORDER
4705 * aligned but the node_mem_map endpoints must be in order
4706 * for the buddy allocator to function correctly.
4708 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
4709 end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
4710 end = ALIGN(end, MAX_ORDER_NR_PAGES);
4711 size = (end - start) * sizeof(struct page);
4712 map = alloc_remap(pgdat->node_id, size);
4714 map = alloc_bootmem_node_nopanic(pgdat, size);
4715 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
4717 #ifndef CONFIG_NEED_MULTIPLE_NODES
4719 * With no DISCONTIG, the global mem_map is just set as node 0's
4721 if (pgdat == NODE_DATA(0)) {
4722 mem_map = NODE_DATA(0)->node_mem_map;
4723 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4724 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
4725 mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
4726 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
4729 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
4732 void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
4733 unsigned long node_start_pfn, unsigned long *zholes_size)
4735 pg_data_t *pgdat = NODE_DATA(nid);
4737 /* pg_data_t should be reset to zero when it's allocated */
4738 WARN_ON(pgdat->nr_zones || pgdat->classzone_idx);
4740 pgdat->node_id = nid;
4741 pgdat->node_start_pfn = node_start_pfn;
4742 init_zone_allows_reclaim(nid);
4743 calculate_node_totalpages(pgdat, zones_size, zholes_size);
4745 alloc_node_mem_map(pgdat);
4746 #ifdef CONFIG_FLAT_NODE_MEM_MAP
4747 printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
4748 nid, (unsigned long)pgdat,
4749 (unsigned long)pgdat->node_mem_map);
4752 free_area_init_core(pgdat, zones_size, zholes_size);
4755 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4757 #if MAX_NUMNODES > 1
4759 * Figure out the number of possible node ids.
4761 static void __init setup_nr_node_ids(void)
4764 unsigned int highest = 0;
4766 for_each_node_mask(node, node_possible_map)
4768 nr_node_ids = highest + 1;
4771 static inline void setup_nr_node_ids(void)
4777 * node_map_pfn_alignment - determine the maximum internode alignment
4779 * This function should be called after node map is populated and sorted.
4780 * It calculates the maximum power of two alignment which can distinguish
4783 * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
4784 * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the
4785 * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is
4786 * shifted, 1GiB is enough and this function will indicate so.
4788 * This is used to test whether pfn -> nid mapping of the chosen memory
4789 * model has fine enough granularity to avoid incorrect mapping for the
4790 * populated node map.
4792 * Returns the determined alignment in pfn's. 0 if there is no alignment
4793 * requirement (single node).
4795 unsigned long __init node_map_pfn_alignment(void)
4797 unsigned long accl_mask = 0, last_end = 0;
4798 unsigned long start, end, mask;
4802 for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) {
4803 if (!start || last_nid < 0 || last_nid == nid) {
4810 * Start with a mask granular enough to pin-point to the
4811 * start pfn and tick off bits one-by-one until it becomes
4812 * too coarse to separate the current node from the last.
4814 mask = ~((1 << __ffs(start)) - 1);
4815 while (mask && last_end <= (start & (mask << 1)))
4818 /* accumulate all internode masks */
4822 /* convert mask to number of pages */
4823 return ~accl_mask + 1;
4826 /* Find the lowest pfn for a node */
4827 static unsigned long __init find_min_pfn_for_node(int nid)
4829 unsigned long min_pfn = ULONG_MAX;
4830 unsigned long start_pfn;
4833 for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL)
4834 min_pfn = min(min_pfn, start_pfn);
4836 if (min_pfn == ULONG_MAX) {
4838 "Could not find start_pfn for node %d\n", nid);
4846 * find_min_pfn_with_active_regions - Find the minimum PFN registered
4848 * It returns the minimum PFN based on information provided via
4849 * add_active_range().
4851 unsigned long __init find_min_pfn_with_active_regions(void)
4853 return find_min_pfn_for_node(MAX_NUMNODES);
4857 * early_calculate_totalpages()
4858 * Sum pages in active regions for movable zone.
4859 * Populate N_MEMORY for calculating usable_nodes.
4861 static unsigned long __init early_calculate_totalpages(void)
4863 unsigned long totalpages = 0;
4864 unsigned long start_pfn, end_pfn;
4867 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
4868 unsigned long pages = end_pfn - start_pfn;
4870 totalpages += pages;
4872 node_set_state(nid, N_MEMORY);
4878 * Find the PFN the Movable zone begins in each node. Kernel memory
4879 * is spread evenly between nodes as long as the nodes have enough
4880 * memory. When they don't, some nodes will have more kernelcore than
4883 static void __init find_zone_movable_pfns_for_nodes(void)
4886 unsigned long usable_startpfn;
4887 unsigned long kernelcore_node, kernelcore_remaining;
4888 /* save the state before borrow the nodemask */
4889 nodemask_t saved_node_state = node_states[N_MEMORY];
4890 unsigned long totalpages = early_calculate_totalpages();
4891 int usable_nodes = nodes_weight(node_states[N_MEMORY]);
4894 * If movablecore was specified, calculate what size of
4895 * kernelcore that corresponds so that memory usable for
4896 * any allocation type is evenly spread. If both kernelcore
4897 * and movablecore are specified, then the value of kernelcore
4898 * will be used for required_kernelcore if it's greater than
4899 * what movablecore would have allowed.
4901 if (required_movablecore) {
4902 unsigned long corepages;
4905 * Round-up so that ZONE_MOVABLE is at least as large as what
4906 * was requested by the user
4908 required_movablecore =
4909 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
4910 corepages = totalpages - required_movablecore;
4912 required_kernelcore = max(required_kernelcore, corepages);
4916 * If neither kernelcore/movablecore nor movablemem_map is specified,
4917 * there is no ZONE_MOVABLE. But if movablemem_map is specified, the
4918 * start pfn of ZONE_MOVABLE has been stored in zone_movable_limit[].
4920 if (!required_kernelcore) {
4921 if (movablemem_map.nr_map)
4922 memcpy(zone_movable_pfn, zone_movable_limit,
4923 sizeof(zone_movable_pfn));
4927 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
4928 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
4931 /* Spread kernelcore memory as evenly as possible throughout nodes */
4932 kernelcore_node = required_kernelcore / usable_nodes;
4933 for_each_node_state(nid, N_MEMORY) {
4934 unsigned long start_pfn, end_pfn;
4937 * Recalculate kernelcore_node if the division per node
4938 * now exceeds what is necessary to satisfy the requested
4939 * amount of memory for the kernel
4941 if (required_kernelcore < kernelcore_node)
4942 kernelcore_node = required_kernelcore / usable_nodes;
4945 * As the map is walked, we track how much memory is usable
4946 * by the kernel using kernelcore_remaining. When it is
4947 * 0, the rest of the node is usable by ZONE_MOVABLE
4949 kernelcore_remaining = kernelcore_node;
4951 /* Go through each range of PFNs within this node */
4952 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
4953 unsigned long size_pages;
4956 * Find more memory for kernelcore in
4957 * [zone_movable_pfn[nid], zone_movable_limit[nid]).
4959 start_pfn = max(start_pfn, zone_movable_pfn[nid]);
4960 if (start_pfn >= end_pfn)
4963 if (zone_movable_limit[nid]) {
4964 end_pfn = min(end_pfn, zone_movable_limit[nid]);
4965 /* No range left for kernelcore in this node */
4966 if (start_pfn >= end_pfn) {
4967 zone_movable_pfn[nid] =
4968 zone_movable_limit[nid];
4973 /* Account for what is only usable for kernelcore */
4974 if (start_pfn < usable_startpfn) {
4975 unsigned long kernel_pages;
4976 kernel_pages = min(end_pfn, usable_startpfn)
4979 kernelcore_remaining -= min(kernel_pages,
4980 kernelcore_remaining);
4981 required_kernelcore -= min(kernel_pages,
4982 required_kernelcore);
4984 /* Continue if range is now fully accounted */
4985 if (end_pfn <= usable_startpfn) {
4988 * Push zone_movable_pfn to the end so
4989 * that if we have to rebalance
4990 * kernelcore across nodes, we will
4991 * not double account here
4993 zone_movable_pfn[nid] = end_pfn;
4996 start_pfn = usable_startpfn;
5000 * The usable PFN range for ZONE_MOVABLE is from
5001 * start_pfn->end_pfn. Calculate size_pages as the
5002 * number of pages used as kernelcore
5004 size_pages = end_pfn - start_pfn;
5005 if (size_pages > kernelcore_remaining)
5006 size_pages = kernelcore_remaining;
5007 zone_movable_pfn[nid] = start_pfn + size_pages;
5010 * Some kernelcore has been met, update counts and
5011 * break if the kernelcore for this node has been
5014 required_kernelcore -= min(required_kernelcore,
5016 kernelcore_remaining -= size_pages;
5017 if (!kernelcore_remaining)
5023 * If there is still required_kernelcore, we do another pass with one
5024 * less node in the count. This will push zone_movable_pfn[nid] further
5025 * along on the nodes that still have memory until kernelcore is
5029 if (usable_nodes && required_kernelcore > usable_nodes)
5033 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
5034 for (nid = 0; nid < MAX_NUMNODES; nid++)
5035 zone_movable_pfn[nid] =
5036 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
5038 /* restore the node_state */
5039 node_states[N_MEMORY] = saved_node_state;
5042 /* Any regular or high memory on that node ? */
5043 static void check_for_memory(pg_data_t *pgdat, int nid)
5045 enum zone_type zone_type;
5047 if (N_MEMORY == N_NORMAL_MEMORY)
5050 for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) {
5051 struct zone *zone = &pgdat->node_zones[zone_type];
5052 if (zone->present_pages) {
5053 node_set_state(nid, N_HIGH_MEMORY);
5054 if (N_NORMAL_MEMORY != N_HIGH_MEMORY &&
5055 zone_type <= ZONE_NORMAL)
5056 node_set_state(nid, N_NORMAL_MEMORY);
5063 * free_area_init_nodes - Initialise all pg_data_t and zone data
5064 * @max_zone_pfn: an array of max PFNs for each zone
5066 * This will call free_area_init_node() for each active node in the system.
5067 * Using the page ranges provided by add_active_range(), the size of each
5068 * zone in each node and their holes is calculated. If the maximum PFN
5069 * between two adjacent zones match, it is assumed that the zone is empty.
5070 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
5071 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
5072 * starts where the previous one ended. For example, ZONE_DMA32 starts
5073 * at arch_max_dma_pfn.
5075 void __init free_area_init_nodes(unsigned long *max_zone_pfn)
5077 unsigned long start_pfn, end_pfn;
5080 /* Record where the zone boundaries are */
5081 memset(arch_zone_lowest_possible_pfn, 0,
5082 sizeof(arch_zone_lowest_possible_pfn));
5083 memset(arch_zone_highest_possible_pfn, 0,
5084 sizeof(arch_zone_highest_possible_pfn));
5085 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
5086 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
5087 for (i = 1; i < MAX_NR_ZONES; i++) {
5088 if (i == ZONE_MOVABLE)
5090 arch_zone_lowest_possible_pfn[i] =
5091 arch_zone_highest_possible_pfn[i-1];
5092 arch_zone_highest_possible_pfn[i] =
5093 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
5095 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
5096 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
5098 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
5099 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
5100 find_usable_zone_for_movable();
5101 sanitize_zone_movable_limit();
5102 find_zone_movable_pfns_for_nodes();
5104 /* Print out the zone ranges */
5105 printk("Zone ranges:\n");
5106 for (i = 0; i < MAX_NR_ZONES; i++) {
5107 if (i == ZONE_MOVABLE)
5109 printk(KERN_CONT " %-8s ", zone_names[i]);
5110 if (arch_zone_lowest_possible_pfn[i] ==
5111 arch_zone_highest_possible_pfn[i])
5112 printk(KERN_CONT "empty\n");
5114 printk(KERN_CONT "[mem %0#10lx-%0#10lx]\n",
5115 arch_zone_lowest_possible_pfn[i] << PAGE_SHIFT,
5116 (arch_zone_highest_possible_pfn[i]
5117 << PAGE_SHIFT) - 1);
5120 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
5121 printk("Movable zone start for each node\n");
5122 for (i = 0; i < MAX_NUMNODES; i++) {
5123 if (zone_movable_pfn[i])
5124 printk(" Node %d: %#010lx\n", i,
5125 zone_movable_pfn[i] << PAGE_SHIFT);
5128 /* Print out the early node map */
5129 printk("Early memory node ranges\n");
5130 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid)
5131 printk(" node %3d: [mem %#010lx-%#010lx]\n", nid,
5132 start_pfn << PAGE_SHIFT, (end_pfn << PAGE_SHIFT) - 1);
5134 /* Initialise every node */
5135 mminit_verify_pageflags_layout();
5136 setup_nr_node_ids();
5137 for_each_online_node(nid) {
5138 pg_data_t *pgdat = NODE_DATA(nid);
5139 free_area_init_node(nid, NULL,
5140 find_min_pfn_for_node(nid), NULL);
5142 /* Any memory on that node */
5143 if (pgdat->node_present_pages)
5144 node_set_state(nid, N_MEMORY);
5145 check_for_memory(pgdat, nid);
5149 static int __init cmdline_parse_core(char *p, unsigned long *core)
5151 unsigned long long coremem;
5155 coremem = memparse(p, &p);
5156 *core = coremem >> PAGE_SHIFT;
5158 /* Paranoid check that UL is enough for the coremem value */
5159 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
5165 * kernelcore=size sets the amount of memory for use for allocations that
5166 * cannot be reclaimed or migrated.
5168 static int __init cmdline_parse_kernelcore(char *p)
5170 return cmdline_parse_core(p, &required_kernelcore);
5174 * movablecore=size sets the amount of memory for use for allocations that
5175 * can be reclaimed or migrated.
5177 static int __init cmdline_parse_movablecore(char *p)
5179 return cmdline_parse_core(p, &required_movablecore);
5182 early_param("kernelcore", cmdline_parse_kernelcore);
5183 early_param("movablecore", cmdline_parse_movablecore);
5186 * movablemem_map_overlap() - Check if a range overlaps movablemem_map.map[].
5187 * @start_pfn: start pfn of the range to be checked
5188 * @end_pfn: end pfn of the range to be checked (exclusive)
5190 * This function checks if a given memory range [start_pfn, end_pfn) overlaps
5191 * the movablemem_map.map[] array.
5193 * Return: index of the first overlapped element in movablemem_map.map[]
5194 * or -1 if they don't overlap each other.
5196 int __init movablemem_map_overlap(unsigned long start_pfn,
5197 unsigned long end_pfn)
5201 if (!movablemem_map.nr_map)
5204 for (overlap = 0; overlap < movablemem_map.nr_map; overlap++)
5205 if (start_pfn < movablemem_map.map[overlap].end_pfn)
5208 if (overlap == movablemem_map.nr_map ||
5209 end_pfn <= movablemem_map.map[overlap].start_pfn)
5216 * insert_movablemem_map - Insert a memory range in to movablemem_map.map.
5217 * @start_pfn: start pfn of the range
5218 * @end_pfn: end pfn of the range
5220 * This function will also merge the overlapped ranges, and sort the array
5221 * by start_pfn in monotonic increasing order.
5223 void __init insert_movablemem_map(unsigned long start_pfn,
5224 unsigned long end_pfn)
5229 * pos will be at the 1st overlapped range, or the position
5230 * where the element should be inserted.
5232 for (pos = 0; pos < movablemem_map.nr_map; pos++)
5233 if (start_pfn <= movablemem_map.map[pos].end_pfn)
5236 /* If there is no overlapped range, just insert the element. */
5237 if (pos == movablemem_map.nr_map ||
5238 end_pfn < movablemem_map.map[pos].start_pfn) {
5240 * If pos is not the end of array, we need to move all
5241 * the rest elements backward.
5243 if (pos < movablemem_map.nr_map)
5244 memmove(&movablemem_map.map[pos+1],
5245 &movablemem_map.map[pos],
5246 sizeof(struct movablemem_entry) *
5247 (movablemem_map.nr_map - pos));
5248 movablemem_map.map[pos].start_pfn = start_pfn;
5249 movablemem_map.map[pos].end_pfn = end_pfn;
5250 movablemem_map.nr_map++;
5254 /* overlap will be at the last overlapped range */
5255 for (overlap = pos + 1; overlap < movablemem_map.nr_map; overlap++)
5256 if (end_pfn < movablemem_map.map[overlap].start_pfn)
5260 * If there are more ranges overlapped, we need to merge them,
5261 * and move the rest elements forward.
5264 movablemem_map.map[pos].start_pfn = min(start_pfn,
5265 movablemem_map.map[pos].start_pfn);
5266 movablemem_map.map[pos].end_pfn = max(end_pfn,
5267 movablemem_map.map[overlap].end_pfn);
5269 if (pos != overlap && overlap + 1 != movablemem_map.nr_map)
5270 memmove(&movablemem_map.map[pos+1],
5271 &movablemem_map.map[overlap+1],
5272 sizeof(struct movablemem_entry) *
5273 (movablemem_map.nr_map - overlap - 1));
5275 movablemem_map.nr_map -= overlap - pos;
5279 * movablemem_map_add_region - Add a memory range into movablemem_map.
5280 * @start: physical start address of range
5281 * @end: physical end address of range
5283 * This function transform the physical address into pfn, and then add the
5284 * range into movablemem_map by calling insert_movablemem_map().
5286 static void __init movablemem_map_add_region(u64 start, u64 size)
5288 unsigned long start_pfn, end_pfn;
5290 /* In case size == 0 or start + size overflows */
5291 if (start + size <= start)
5294 if (movablemem_map.nr_map >= ARRAY_SIZE(movablemem_map.map)) {
5295 pr_err("movablemem_map: too many entries;"
5296 " ignoring [mem %#010llx-%#010llx]\n",
5297 (unsigned long long) start,
5298 (unsigned long long) (start + size - 1));
5302 start_pfn = PFN_DOWN(start);
5303 end_pfn = PFN_UP(start + size);
5304 insert_movablemem_map(start_pfn, end_pfn);
5308 * cmdline_parse_movablemem_map - Parse boot option movablemem_map.
5309 * @p: The boot option of the following format:
5310 * movablemem_map=nn[KMG]@ss[KMG]
5312 * This option sets the memory range [ss, ss+nn) to be used as movable memory.
5314 * Return: 0 on success or -EINVAL on failure.
5316 static int __init cmdline_parse_movablemem_map(char *p)
5319 u64 start_at, mem_size;
5324 if (!strcmp(p, "acpi"))
5325 movablemem_map.acpi = true;
5328 * If user decide to use info from BIOS, all the other user specified
5329 * ranges will be ingored.
5331 if (movablemem_map.acpi) {
5332 if (movablemem_map.nr_map) {
5333 memset(movablemem_map.map, 0,
5334 sizeof(struct movablemem_entry)
5335 * movablemem_map.nr_map);
5336 movablemem_map.nr_map = 0;
5342 mem_size = memparse(p, &p);
5348 start_at = memparse(p, &p);
5349 if (p == oldp || *p != '\0')
5352 movablemem_map_add_region(start_at, mem_size);
5358 early_param("movablemem_map", cmdline_parse_movablemem_map);
5360 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5363 * set_dma_reserve - set the specified number of pages reserved in the first zone
5364 * @new_dma_reserve: The number of pages to mark reserved
5366 * The per-cpu batchsize and zone watermarks are determined by present_pages.
5367 * In the DMA zone, a significant percentage may be consumed by kernel image
5368 * and other unfreeable allocations which can skew the watermarks badly. This
5369 * function may optionally be used to account for unfreeable pages in the
5370 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
5371 * smaller per-cpu batchsize.
5373 void __init set_dma_reserve(unsigned long new_dma_reserve)
5375 dma_reserve = new_dma_reserve;
5378 void __init free_area_init(unsigned long *zones_size)
5380 free_area_init_node(0, zones_size,
5381 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
5384 static int page_alloc_cpu_notify(struct notifier_block *self,
5385 unsigned long action, void *hcpu)
5387 int cpu = (unsigned long)hcpu;
5389 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
5390 lru_add_drain_cpu(cpu);
5394 * Spill the event counters of the dead processor
5395 * into the current processors event counters.
5396 * This artificially elevates the count of the current
5399 vm_events_fold_cpu(cpu);
5402 * Zero the differential counters of the dead processor
5403 * so that the vm statistics are consistent.
5405 * This is only okay since the processor is dead and cannot
5406 * race with what we are doing.
5408 refresh_cpu_vm_stats(cpu);
5413 void __init page_alloc_init(void)
5415 hotcpu_notifier(page_alloc_cpu_notify, 0);
5419 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
5420 * or min_free_kbytes changes.
5422 static void calculate_totalreserve_pages(void)
5424 struct pglist_data *pgdat;
5425 unsigned long reserve_pages = 0;
5426 enum zone_type i, j;
5428 for_each_online_pgdat(pgdat) {
5429 for (i = 0; i < MAX_NR_ZONES; i++) {
5430 struct zone *zone = pgdat->node_zones + i;
5431 unsigned long max = 0;
5433 /* Find valid and maximum lowmem_reserve in the zone */
5434 for (j = i; j < MAX_NR_ZONES; j++) {
5435 if (zone->lowmem_reserve[j] > max)
5436 max = zone->lowmem_reserve[j];
5439 /* we treat the high watermark as reserved pages. */
5440 max += high_wmark_pages(zone);
5442 if (max > zone->managed_pages)
5443 max = zone->managed_pages;
5444 reserve_pages += max;
5446 * Lowmem reserves are not available to
5447 * GFP_HIGHUSER page cache allocations and
5448 * kswapd tries to balance zones to their high
5449 * watermark. As a result, neither should be
5450 * regarded as dirtyable memory, to prevent a
5451 * situation where reclaim has to clean pages
5452 * in order to balance the zones.
5454 zone->dirty_balance_reserve = max;
5457 dirty_balance_reserve = reserve_pages;
5458 totalreserve_pages = reserve_pages;
5462 * setup_per_zone_lowmem_reserve - called whenever
5463 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
5464 * has a correct pages reserved value, so an adequate number of
5465 * pages are left in the zone after a successful __alloc_pages().
5467 static void setup_per_zone_lowmem_reserve(void)
5469 struct pglist_data *pgdat;
5470 enum zone_type j, idx;
5472 for_each_online_pgdat(pgdat) {
5473 for (j = 0; j < MAX_NR_ZONES; j++) {
5474 struct zone *zone = pgdat->node_zones + j;
5475 unsigned long managed_pages = zone->managed_pages;
5477 zone->lowmem_reserve[j] = 0;
5481 struct zone *lower_zone;
5485 if (sysctl_lowmem_reserve_ratio[idx] < 1)
5486 sysctl_lowmem_reserve_ratio[idx] = 1;
5488 lower_zone = pgdat->node_zones + idx;
5489 lower_zone->lowmem_reserve[j] = managed_pages /
5490 sysctl_lowmem_reserve_ratio[idx];
5491 managed_pages += lower_zone->managed_pages;
5496 /* update totalreserve_pages */
5497 calculate_totalreserve_pages();
5500 static void __setup_per_zone_wmarks(void)
5502 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
5503 unsigned long lowmem_pages = 0;
5505 unsigned long flags;
5507 /* Calculate total number of !ZONE_HIGHMEM pages */
5508 for_each_zone(zone) {
5509 if (!is_highmem(zone))
5510 lowmem_pages += zone->managed_pages;
5513 for_each_zone(zone) {
5516 spin_lock_irqsave(&zone->lock, flags);
5517 tmp = (u64)pages_min * zone->managed_pages;
5518 do_div(tmp, lowmem_pages);
5519 if (is_highmem(zone)) {
5521 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
5522 * need highmem pages, so cap pages_min to a small
5525 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
5526 * deltas controls asynch page reclaim, and so should
5527 * not be capped for highmem.
5529 unsigned long min_pages;
5531 min_pages = zone->managed_pages / 1024;
5532 min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL);
5533 zone->watermark[WMARK_MIN] = min_pages;
5536 * If it's a lowmem zone, reserve a number of pages
5537 * proportionate to the zone's size.
5539 zone->watermark[WMARK_MIN] = tmp;
5542 zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + (tmp >> 2);
5543 zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1);
5545 setup_zone_migrate_reserve(zone);
5546 spin_unlock_irqrestore(&zone->lock, flags);
5549 /* update totalreserve_pages */
5550 calculate_totalreserve_pages();
5554 * setup_per_zone_wmarks - called when min_free_kbytes changes
5555 * or when memory is hot-{added|removed}
5557 * Ensures that the watermark[min,low,high] values for each zone are set
5558 * correctly with respect to min_free_kbytes.
5560 void setup_per_zone_wmarks(void)
5562 mutex_lock(&zonelists_mutex);
5563 __setup_per_zone_wmarks();
5564 mutex_unlock(&zonelists_mutex);
5568 * The inactive anon list should be small enough that the VM never has to
5569 * do too much work, but large enough that each inactive page has a chance
5570 * to be referenced again before it is swapped out.
5572 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
5573 * INACTIVE_ANON pages on this zone's LRU, maintained by the
5574 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
5575 * the anonymous pages are kept on the inactive list.
5578 * memory ratio inactive anon
5579 * -------------------------------------
5588 static void __meminit calculate_zone_inactive_ratio(struct zone *zone)
5590 unsigned int gb, ratio;
5592 /* Zone size in gigabytes */
5593 gb = zone->managed_pages >> (30 - PAGE_SHIFT);
5595 ratio = int_sqrt(10 * gb);
5599 zone->inactive_ratio = ratio;
5602 static void __meminit setup_per_zone_inactive_ratio(void)
5607 calculate_zone_inactive_ratio(zone);
5611 * Initialise min_free_kbytes.
5613 * For small machines we want it small (128k min). For large machines
5614 * we want it large (64MB max). But it is not linear, because network
5615 * bandwidth does not increase linearly with machine size. We use
5617 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
5618 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
5634 int __meminit init_per_zone_wmark_min(void)
5636 unsigned long lowmem_kbytes;
5638 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
5640 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
5641 if (min_free_kbytes < 128)
5642 min_free_kbytes = 128;
5643 if (min_free_kbytes > 65536)
5644 min_free_kbytes = 65536;
5645 setup_per_zone_wmarks();
5646 refresh_zone_stat_thresholds();
5647 setup_per_zone_lowmem_reserve();
5648 setup_per_zone_inactive_ratio();
5651 module_init(init_per_zone_wmark_min)
5654 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
5655 * that we can call two helper functions whenever min_free_kbytes
5658 int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
5659 void __user *buffer, size_t *length, loff_t *ppos)
5661 proc_dointvec(table, write, buffer, length, ppos);
5663 setup_per_zone_wmarks();
5668 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
5669 void __user *buffer, size_t *length, loff_t *ppos)
5674 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
5679 zone->min_unmapped_pages = (zone->managed_pages *
5680 sysctl_min_unmapped_ratio) / 100;
5684 int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
5685 void __user *buffer, size_t *length, loff_t *ppos)
5690 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
5695 zone->min_slab_pages = (zone->managed_pages *
5696 sysctl_min_slab_ratio) / 100;
5702 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
5703 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
5704 * whenever sysctl_lowmem_reserve_ratio changes.
5706 * The reserve ratio obviously has absolutely no relation with the
5707 * minimum watermarks. The lowmem reserve ratio can only make sense
5708 * if in function of the boot time zone sizes.
5710 int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
5711 void __user *buffer, size_t *length, loff_t *ppos)
5713 proc_dointvec_minmax(table, write, buffer, length, ppos);
5714 setup_per_zone_lowmem_reserve();
5719 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
5720 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
5721 * can have before it gets flushed back to buddy allocator.
5724 int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
5725 void __user *buffer, size_t *length, loff_t *ppos)
5731 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
5732 if (!write || (ret < 0))
5734 for_each_populated_zone(zone) {
5735 for_each_possible_cpu(cpu) {
5737 high = zone->managed_pages / percpu_pagelist_fraction;
5738 setup_pagelist_highmark(
5739 per_cpu_ptr(zone->pageset, cpu), high);
5745 int hashdist = HASHDIST_DEFAULT;
5748 static int __init set_hashdist(char *str)
5752 hashdist = simple_strtoul(str, &str, 0);
5755 __setup("hashdist=", set_hashdist);
5759 * allocate a large system hash table from bootmem
5760 * - it is assumed that the hash table must contain an exact power-of-2
5761 * quantity of entries
5762 * - limit is the number of hash buckets, not the total allocation size
5764 void *__init alloc_large_system_hash(const char *tablename,
5765 unsigned long bucketsize,
5766 unsigned long numentries,
5769 unsigned int *_hash_shift,
5770 unsigned int *_hash_mask,
5771 unsigned long low_limit,
5772 unsigned long high_limit)
5774 unsigned long long max = high_limit;
5775 unsigned long log2qty, size;
5778 /* allow the kernel cmdline to have a say */
5780 /* round applicable memory size up to nearest megabyte */
5781 numentries = nr_kernel_pages;
5782 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
5783 numentries >>= 20 - PAGE_SHIFT;
5784 numentries <<= 20 - PAGE_SHIFT;
5786 /* limit to 1 bucket per 2^scale bytes of low memory */
5787 if (scale > PAGE_SHIFT)
5788 numentries >>= (scale - PAGE_SHIFT);
5790 numentries <<= (PAGE_SHIFT - scale);
5792 /* Make sure we've got at least a 0-order allocation.. */
5793 if (unlikely(flags & HASH_SMALL)) {
5794 /* Makes no sense without HASH_EARLY */
5795 WARN_ON(!(flags & HASH_EARLY));
5796 if (!(numentries >> *_hash_shift)) {
5797 numentries = 1UL << *_hash_shift;
5798 BUG_ON(!numentries);
5800 } else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
5801 numentries = PAGE_SIZE / bucketsize;
5803 numentries = roundup_pow_of_two(numentries);
5805 /* limit allocation size to 1/16 total memory by default */
5807 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
5808 do_div(max, bucketsize);
5810 max = min(max, 0x80000000ULL);
5812 if (numentries < low_limit)
5813 numentries = low_limit;
5814 if (numentries > max)
5817 log2qty = ilog2(numentries);
5820 size = bucketsize << log2qty;
5821 if (flags & HASH_EARLY)
5822 table = alloc_bootmem_nopanic(size);
5824 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
5827 * If bucketsize is not a power-of-two, we may free
5828 * some pages at the end of hash table which
5829 * alloc_pages_exact() automatically does
5831 if (get_order(size) < MAX_ORDER) {
5832 table = alloc_pages_exact(size, GFP_ATOMIC);
5833 kmemleak_alloc(table, size, 1, GFP_ATOMIC);
5836 } while (!table && size > PAGE_SIZE && --log2qty);
5839 panic("Failed to allocate %s hash table\n", tablename);
5841 printk(KERN_INFO "%s hash table entries: %ld (order: %d, %lu bytes)\n",
5844 ilog2(size) - PAGE_SHIFT,
5848 *_hash_shift = log2qty;
5850 *_hash_mask = (1 << log2qty) - 1;
5855 /* Return a pointer to the bitmap storing bits affecting a block of pages */
5856 static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
5859 #ifdef CONFIG_SPARSEMEM
5860 return __pfn_to_section(pfn)->pageblock_flags;
5862 return zone->pageblock_flags;
5863 #endif /* CONFIG_SPARSEMEM */
5866 static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
5868 #ifdef CONFIG_SPARSEMEM
5869 pfn &= (PAGES_PER_SECTION-1);
5870 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
5872 pfn = pfn - round_down(zone->zone_start_pfn, pageblock_nr_pages);
5873 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
5874 #endif /* CONFIG_SPARSEMEM */
5878 * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
5879 * @page: The page within the block of interest
5880 * @start_bitidx: The first bit of interest to retrieve
5881 * @end_bitidx: The last bit of interest
5882 * returns pageblock_bits flags
5884 unsigned long get_pageblock_flags_group(struct page *page,
5885 int start_bitidx, int end_bitidx)
5888 unsigned long *bitmap;
5889 unsigned long pfn, bitidx;
5890 unsigned long flags = 0;
5891 unsigned long value = 1;
5893 zone = page_zone(page);
5894 pfn = page_to_pfn(page);
5895 bitmap = get_pageblock_bitmap(zone, pfn);
5896 bitidx = pfn_to_bitidx(zone, pfn);
5898 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
5899 if (test_bit(bitidx + start_bitidx, bitmap))
5906 * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
5907 * @page: The page within the block of interest
5908 * @start_bitidx: The first bit of interest
5909 * @end_bitidx: The last bit of interest
5910 * @flags: The flags to set
5912 void set_pageblock_flags_group(struct page *page, unsigned long flags,
5913 int start_bitidx, int end_bitidx)
5916 unsigned long *bitmap;
5917 unsigned long pfn, bitidx;
5918 unsigned long value = 1;
5920 zone = page_zone(page);
5921 pfn = page_to_pfn(page);
5922 bitmap = get_pageblock_bitmap(zone, pfn);
5923 bitidx = pfn_to_bitidx(zone, pfn);
5924 VM_BUG_ON(pfn < zone->zone_start_pfn);
5925 VM_BUG_ON(pfn >= zone->zone_start_pfn + zone->spanned_pages);
5927 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
5929 __set_bit(bitidx + start_bitidx, bitmap);
5931 __clear_bit(bitidx + start_bitidx, bitmap);
5935 * This function checks whether pageblock includes unmovable pages or not.
5936 * If @count is not zero, it is okay to include less @count unmovable pages
5938 * PageLRU check wihtout isolation or lru_lock could race so that
5939 * MIGRATE_MOVABLE block might include unmovable pages. It means you can't
5940 * expect this function should be exact.
5942 bool has_unmovable_pages(struct zone *zone, struct page *page, int count,
5943 bool skip_hwpoisoned_pages)
5945 unsigned long pfn, iter, found;
5949 * For avoiding noise data, lru_add_drain_all() should be called
5950 * If ZONE_MOVABLE, the zone never contains unmovable pages
5952 if (zone_idx(zone) == ZONE_MOVABLE)
5954 mt = get_pageblock_migratetype(page);
5955 if (mt == MIGRATE_MOVABLE || is_migrate_cma(mt))
5958 pfn = page_to_pfn(page);
5959 for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) {
5960 unsigned long check = pfn + iter;
5962 if (!pfn_valid_within(check))
5965 page = pfn_to_page(check);
5967 * We can't use page_count without pin a page
5968 * because another CPU can free compound page.
5969 * This check already skips compound tails of THP
5970 * because their page->_count is zero at all time.
5972 if (!atomic_read(&page->_count)) {
5973 if (PageBuddy(page))
5974 iter += (1 << page_order(page)) - 1;
5979 * The HWPoisoned page may be not in buddy system, and
5980 * page_count() is not 0.
5982 if (skip_hwpoisoned_pages && PageHWPoison(page))
5988 * If there are RECLAIMABLE pages, we need to check it.
5989 * But now, memory offline itself doesn't call shrink_slab()
5990 * and it still to be fixed.
5993 * If the page is not RAM, page_count()should be 0.
5994 * we don't need more check. This is an _used_ not-movable page.
5996 * The problematic thing here is PG_reserved pages. PG_reserved
5997 * is set to both of a memory hole page and a _used_ kernel
6006 bool is_pageblock_removable_nolock(struct page *page)
6012 * We have to be careful here because we are iterating over memory
6013 * sections which are not zone aware so we might end up outside of
6014 * the zone but still within the section.
6015 * We have to take care about the node as well. If the node is offline
6016 * its NODE_DATA will be NULL - see page_zone.
6018 if (!node_online(page_to_nid(page)))
6021 zone = page_zone(page);
6022 pfn = page_to_pfn(page);
6023 if (zone->zone_start_pfn > pfn ||
6024 zone->zone_start_pfn + zone->spanned_pages <= pfn)
6027 return !has_unmovable_pages(zone, page, 0, true);
6032 static unsigned long pfn_max_align_down(unsigned long pfn)
6034 return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES,
6035 pageblock_nr_pages) - 1);
6038 static unsigned long pfn_max_align_up(unsigned long pfn)
6040 return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES,
6041 pageblock_nr_pages));
6044 /* [start, end) must belong to a single zone. */
6045 static int __alloc_contig_migrate_range(struct compact_control *cc,
6046 unsigned long start, unsigned long end)
6048 /* This function is based on compact_zone() from compaction.c. */
6049 unsigned long nr_reclaimed;
6050 unsigned long pfn = start;
6051 unsigned int tries = 0;
6056 while (pfn < end || !list_empty(&cc->migratepages)) {
6057 if (fatal_signal_pending(current)) {
6062 if (list_empty(&cc->migratepages)) {
6063 cc->nr_migratepages = 0;
6064 pfn = isolate_migratepages_range(cc->zone, cc,
6071 } else if (++tries == 5) {
6072 ret = ret < 0 ? ret : -EBUSY;
6076 nr_reclaimed = reclaim_clean_pages_from_list(cc->zone,
6078 cc->nr_migratepages -= nr_reclaimed;
6080 ret = migrate_pages(&cc->migratepages,
6081 alloc_migrate_target,
6082 0, false, MIGRATE_SYNC,
6086 putback_movable_pages(&cc->migratepages);
6093 * alloc_contig_range() -- tries to allocate given range of pages
6094 * @start: start PFN to allocate
6095 * @end: one-past-the-last PFN to allocate
6096 * @migratetype: migratetype of the underlaying pageblocks (either
6097 * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks
6098 * in range must have the same migratetype and it must
6099 * be either of the two.
6101 * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
6102 * aligned, however it's the caller's responsibility to guarantee that
6103 * we are the only thread that changes migrate type of pageblocks the
6106 * The PFN range must belong to a single zone.
6108 * Returns zero on success or negative error code. On success all
6109 * pages which PFN is in [start, end) are allocated for the caller and
6110 * need to be freed with free_contig_range().
6112 int alloc_contig_range(unsigned long start, unsigned long end,
6113 unsigned migratetype)
6115 unsigned long outer_start, outer_end;
6118 struct compact_control cc = {
6119 .nr_migratepages = 0,
6121 .zone = page_zone(pfn_to_page(start)),
6123 .ignore_skip_hint = true,
6125 INIT_LIST_HEAD(&cc.migratepages);
6128 * What we do here is we mark all pageblocks in range as
6129 * MIGRATE_ISOLATE. Because pageblock and max order pages may
6130 * have different sizes, and due to the way page allocator
6131 * work, we align the range to biggest of the two pages so
6132 * that page allocator won't try to merge buddies from
6133 * different pageblocks and change MIGRATE_ISOLATE to some
6134 * other migration type.
6136 * Once the pageblocks are marked as MIGRATE_ISOLATE, we
6137 * migrate the pages from an unaligned range (ie. pages that
6138 * we are interested in). This will put all the pages in
6139 * range back to page allocator as MIGRATE_ISOLATE.
6141 * When this is done, we take the pages in range from page
6142 * allocator removing them from the buddy system. This way
6143 * page allocator will never consider using them.
6145 * This lets us mark the pageblocks back as
6146 * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
6147 * aligned range but not in the unaligned, original range are
6148 * put back to page allocator so that buddy can use them.
6151 ret = start_isolate_page_range(pfn_max_align_down(start),
6152 pfn_max_align_up(end), migratetype,
6157 ret = __alloc_contig_migrate_range(&cc, start, end);
6162 * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
6163 * aligned blocks that are marked as MIGRATE_ISOLATE. What's
6164 * more, all pages in [start, end) are free in page allocator.
6165 * What we are going to do is to allocate all pages from
6166 * [start, end) (that is remove them from page allocator).
6168 * The only problem is that pages at the beginning and at the
6169 * end of interesting range may be not aligned with pages that
6170 * page allocator holds, ie. they can be part of higher order
6171 * pages. Because of this, we reserve the bigger range and
6172 * once this is done free the pages we are not interested in.
6174 * We don't have to hold zone->lock here because the pages are
6175 * isolated thus they won't get removed from buddy.
6178 lru_add_drain_all();
6182 outer_start = start;
6183 while (!PageBuddy(pfn_to_page(outer_start))) {
6184 if (++order >= MAX_ORDER) {
6188 outer_start &= ~0UL << order;
6191 /* Make sure the range is really isolated. */
6192 if (test_pages_isolated(outer_start, end, false)) {
6193 pr_warn("alloc_contig_range test_pages_isolated(%lx, %lx) failed\n",
6200 /* Grab isolated pages from freelists. */
6201 outer_end = isolate_freepages_range(&cc, outer_start, end);
6207 /* Free head and tail (if any) */
6208 if (start != outer_start)
6209 free_contig_range(outer_start, start - outer_start);
6210 if (end != outer_end)
6211 free_contig_range(end, outer_end - end);
6214 undo_isolate_page_range(pfn_max_align_down(start),
6215 pfn_max_align_up(end), migratetype);
6219 void free_contig_range(unsigned long pfn, unsigned nr_pages)
6221 unsigned int count = 0;
6223 for (; nr_pages--; pfn++) {
6224 struct page *page = pfn_to_page(pfn);
6226 count += page_count(page) != 1;
6229 WARN(count != 0, "%d pages are still in use!\n", count);
6233 #ifdef CONFIG_MEMORY_HOTPLUG
6234 static int __meminit __zone_pcp_update(void *data)
6236 struct zone *zone = data;
6238 unsigned long batch = zone_batchsize(zone), flags;
6240 for_each_possible_cpu(cpu) {
6241 struct per_cpu_pageset *pset;
6242 struct per_cpu_pages *pcp;
6244 pset = per_cpu_ptr(zone->pageset, cpu);
6247 local_irq_save(flags);
6249 free_pcppages_bulk(zone, pcp->count, pcp);
6250 drain_zonestat(zone, pset);
6251 setup_pageset(pset, batch);
6252 local_irq_restore(flags);
6257 void __meminit zone_pcp_update(struct zone *zone)
6259 stop_machine(__zone_pcp_update, zone, NULL);
6263 void zone_pcp_reset(struct zone *zone)
6265 unsigned long flags;
6267 struct per_cpu_pageset *pset;
6269 /* avoid races with drain_pages() */
6270 local_irq_save(flags);
6271 if (zone->pageset != &boot_pageset) {
6272 for_each_online_cpu(cpu) {
6273 pset = per_cpu_ptr(zone->pageset, cpu);
6274 drain_zonestat(zone, pset);
6276 free_percpu(zone->pageset);
6277 zone->pageset = &boot_pageset;
6279 local_irq_restore(flags);
6282 #ifdef CONFIG_MEMORY_HOTREMOVE
6284 * All pages in the range must be isolated before calling this.
6287 __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
6293 unsigned long flags;
6294 /* find the first valid pfn */
6295 for (pfn = start_pfn; pfn < end_pfn; pfn++)
6300 zone = page_zone(pfn_to_page(pfn));
6301 spin_lock_irqsave(&zone->lock, flags);
6303 while (pfn < end_pfn) {
6304 if (!pfn_valid(pfn)) {
6308 page = pfn_to_page(pfn);
6310 * The HWPoisoned page may be not in buddy system, and
6311 * page_count() is not 0.
6313 if (unlikely(!PageBuddy(page) && PageHWPoison(page))) {
6315 SetPageReserved(page);
6319 BUG_ON(page_count(page));
6320 BUG_ON(!PageBuddy(page));
6321 order = page_order(page);
6322 #ifdef CONFIG_DEBUG_VM
6323 printk(KERN_INFO "remove from free list %lx %d %lx\n",
6324 pfn, 1 << order, end_pfn);
6326 list_del(&page->lru);
6327 rmv_page_order(page);
6328 zone->free_area[order].nr_free--;
6329 for (i = 0; i < (1 << order); i++)
6330 SetPageReserved((page+i));
6331 pfn += (1 << order);
6333 spin_unlock_irqrestore(&zone->lock, flags);
6337 #ifdef CONFIG_MEMORY_FAILURE
6338 bool is_free_buddy_page(struct page *page)
6340 struct zone *zone = page_zone(page);
6341 unsigned long pfn = page_to_pfn(page);
6342 unsigned long flags;
6345 spin_lock_irqsave(&zone->lock, flags);
6346 for (order = 0; order < MAX_ORDER; order++) {
6347 struct page *page_head = page - (pfn & ((1 << order) - 1));
6349 if (PageBuddy(page_head) && page_order(page_head) >= order)
6352 spin_unlock_irqrestore(&zone->lock, flags);
6354 return order < MAX_ORDER;
6358 static const struct trace_print_flags pageflag_names[] = {
6359 {1UL << PG_locked, "locked" },
6360 {1UL << PG_error, "error" },
6361 {1UL << PG_referenced, "referenced" },
6362 {1UL << PG_uptodate, "uptodate" },
6363 {1UL << PG_dirty, "dirty" },
6364 {1UL << PG_lru, "lru" },
6365 {1UL << PG_active, "active" },
6366 {1UL << PG_slab, "slab" },
6367 {1UL << PG_owner_priv_1, "owner_priv_1" },
6368 {1UL << PG_arch_1, "arch_1" },
6369 {1UL << PG_reserved, "reserved" },
6370 {1UL << PG_private, "private" },
6371 {1UL << PG_private_2, "private_2" },
6372 {1UL << PG_writeback, "writeback" },
6373 #ifdef CONFIG_PAGEFLAGS_EXTENDED
6374 {1UL << PG_head, "head" },
6375 {1UL << PG_tail, "tail" },
6377 {1UL << PG_compound, "compound" },
6379 {1UL << PG_swapcache, "swapcache" },
6380 {1UL << PG_mappedtodisk, "mappedtodisk" },
6381 {1UL << PG_reclaim, "reclaim" },
6382 {1UL << PG_swapbacked, "swapbacked" },
6383 {1UL << PG_unevictable, "unevictable" },
6385 {1UL << PG_mlocked, "mlocked" },
6387 #ifdef CONFIG_ARCH_USES_PG_UNCACHED
6388 {1UL << PG_uncached, "uncached" },
6390 #ifdef CONFIG_MEMORY_FAILURE
6391 {1UL << PG_hwpoison, "hwpoison" },
6393 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
6394 {1UL << PG_compound_lock, "compound_lock" },
6398 static void dump_page_flags(unsigned long flags)
6400 const char *delim = "";
6404 BUILD_BUG_ON(ARRAY_SIZE(pageflag_names) != __NR_PAGEFLAGS);
6406 printk(KERN_ALERT "page flags: %#lx(", flags);
6408 /* remove zone id */
6409 flags &= (1UL << NR_PAGEFLAGS) - 1;
6411 for (i = 0; i < ARRAY_SIZE(pageflag_names) && flags; i++) {
6413 mask = pageflag_names[i].mask;
6414 if ((flags & mask) != mask)
6418 printk("%s%s", delim, pageflag_names[i].name);
6422 /* check for left over flags */
6424 printk("%s%#lx", delim, flags);
6429 void dump_page(struct page *page)
6432 "page:%p count:%d mapcount:%d mapping:%p index:%#lx\n",
6433 page, atomic_read(&page->_count), page_mapcount(page),
6434 page->mapping, page->index);
6435 dump_page_flags(page->flags);
6436 mem_cgroup_print_bad_page(page);