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/mm_inline.h>
60 #include <linux/migrate.h>
61 #include <linux/page-debug-flags.h>
62 #include <linux/hugetlb.h>
63 #include <linux/sched/rt.h>
65 #include <asm/tlbflush.h>
66 #include <asm/div64.h>
69 #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
70 DEFINE_PER_CPU(int, numa_node);
71 EXPORT_PER_CPU_SYMBOL(numa_node);
74 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
76 * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
77 * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
78 * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
79 * defined in <linux/topology.h>.
81 DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */
82 EXPORT_PER_CPU_SYMBOL(_numa_mem_);
86 * Array of node states.
88 nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
89 [N_POSSIBLE] = NODE_MASK_ALL,
90 [N_ONLINE] = { { [0] = 1UL } },
92 [N_NORMAL_MEMORY] = { { [0] = 1UL } },
94 [N_HIGH_MEMORY] = { { [0] = 1UL } },
96 #ifdef CONFIG_MOVABLE_NODE
97 [N_MEMORY] = { { [0] = 1UL } },
99 [N_CPU] = { { [0] = 1UL } },
102 EXPORT_SYMBOL(node_states);
104 unsigned long totalram_pages __read_mostly;
105 unsigned long totalreserve_pages __read_mostly;
107 * When calculating the number of globally allowed dirty pages, there
108 * is a certain number of per-zone reserves that should not be
109 * considered dirtyable memory. This is the sum of those reserves
110 * over all existing zones that contribute dirtyable memory.
112 unsigned long dirty_balance_reserve __read_mostly;
114 int percpu_pagelist_fraction;
115 gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK;
117 #ifdef CONFIG_PM_SLEEP
119 * The following functions are used by the suspend/hibernate code to temporarily
120 * change gfp_allowed_mask in order to avoid using I/O during memory allocations
121 * while devices are suspended. To avoid races with the suspend/hibernate code,
122 * they should always be called with pm_mutex held (gfp_allowed_mask also should
123 * only be modified with pm_mutex held, unless the suspend/hibernate code is
124 * guaranteed not to run in parallel with that modification).
127 static gfp_t saved_gfp_mask;
129 void pm_restore_gfp_mask(void)
131 WARN_ON(!mutex_is_locked(&pm_mutex));
132 if (saved_gfp_mask) {
133 gfp_allowed_mask = saved_gfp_mask;
138 void pm_restrict_gfp_mask(void)
140 WARN_ON(!mutex_is_locked(&pm_mutex));
141 WARN_ON(saved_gfp_mask);
142 saved_gfp_mask = gfp_allowed_mask;
143 gfp_allowed_mask &= ~GFP_IOFS;
146 bool pm_suspended_storage(void)
148 if ((gfp_allowed_mask & GFP_IOFS) == GFP_IOFS)
152 #endif /* CONFIG_PM_SLEEP */
154 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
155 int pageblock_order __read_mostly;
158 static void __free_pages_ok(struct page *page, unsigned int order);
161 * results with 256, 32 in the lowmem_reserve sysctl:
162 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
163 * 1G machine -> (16M dma, 784M normal, 224M high)
164 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
165 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
166 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
168 * TBD: should special case ZONE_DMA32 machines here - in those we normally
169 * don't need any ZONE_NORMAL reservation
171 int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
172 #ifdef CONFIG_ZONE_DMA
175 #ifdef CONFIG_ZONE_DMA32
178 #ifdef CONFIG_HIGHMEM
184 EXPORT_SYMBOL(totalram_pages);
186 static char * const zone_names[MAX_NR_ZONES] = {
187 #ifdef CONFIG_ZONE_DMA
190 #ifdef CONFIG_ZONE_DMA32
194 #ifdef CONFIG_HIGHMEM
201 * Try to keep at least this much lowmem free. Do not allow normal
202 * allocations below this point, only high priority ones. Automatically
203 * tuned according to the amount of memory in the system.
205 int min_free_kbytes = 1024;
206 int min_free_order_shift = 1;
209 * Extra memory for the system to try freeing. Used to temporarily
210 * free memory, to make space for new workloads. Anyone can allocate
211 * down to the min watermarks controlled by min_free_kbytes above.
213 int extra_free_kbytes = 0;
215 static unsigned long __meminitdata nr_kernel_pages;
216 static unsigned long __meminitdata nr_all_pages;
217 static unsigned long __meminitdata dma_reserve;
219 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
220 static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
221 static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
222 static unsigned long __initdata required_kernelcore;
223 static unsigned long __initdata required_movablecore;
224 static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
226 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
228 EXPORT_SYMBOL(movable_zone);
229 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
232 int nr_node_ids __read_mostly = MAX_NUMNODES;
233 int nr_online_nodes __read_mostly = 1;
234 EXPORT_SYMBOL(nr_node_ids);
235 EXPORT_SYMBOL(nr_online_nodes);
238 int page_group_by_mobility_disabled __read_mostly;
240 void set_pageblock_migratetype(struct page *page, int migratetype)
243 if (unlikely(page_group_by_mobility_disabled))
244 migratetype = MIGRATE_UNMOVABLE;
246 set_pageblock_flags_group(page, (unsigned long)migratetype,
247 PB_migrate, PB_migrate_end);
250 bool oom_killer_disabled __read_mostly;
252 #ifdef CONFIG_DEBUG_VM
253 static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
257 unsigned long pfn = page_to_pfn(page);
258 unsigned long sp, start_pfn;
261 seq = zone_span_seqbegin(zone);
262 start_pfn = zone->zone_start_pfn;
263 sp = zone->spanned_pages;
264 if (!zone_spans_pfn(zone, pfn))
266 } while (zone_span_seqretry(zone, seq));
269 pr_err("page %lu outside zone [ %lu - %lu ]\n",
270 pfn, start_pfn, start_pfn + sp);
275 static int page_is_consistent(struct zone *zone, struct page *page)
277 if (!pfn_valid_within(page_to_pfn(page)))
279 if (zone != page_zone(page))
285 * Temporary debugging check for pages not lying within a given zone.
287 static int bad_range(struct zone *zone, struct page *page)
289 if (page_outside_zone_boundaries(zone, page))
291 if (!page_is_consistent(zone, page))
297 static inline int bad_range(struct zone *zone, struct page *page)
303 static void bad_page(struct page *page)
305 static unsigned long resume;
306 static unsigned long nr_shown;
307 static unsigned long nr_unshown;
309 /* Don't complain about poisoned pages */
310 if (PageHWPoison(page)) {
311 page_mapcount_reset(page); /* remove PageBuddy */
316 * Allow a burst of 60 reports, then keep quiet for that minute;
317 * or allow a steady drip of one report per second.
319 if (nr_shown == 60) {
320 if (time_before(jiffies, resume)) {
326 "BUG: Bad page state: %lu messages suppressed\n",
333 resume = jiffies + 60 * HZ;
335 printk(KERN_ALERT "BUG: Bad page state in process %s pfn:%05lx\n",
336 current->comm, page_to_pfn(page));
342 /* Leave bad fields for debug, except PageBuddy could make trouble */
343 page_mapcount_reset(page); /* remove PageBuddy */
344 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
348 * Higher-order pages are called "compound pages". They are structured thusly:
350 * The first PAGE_SIZE page is called the "head page".
352 * The remaining PAGE_SIZE pages are called "tail pages".
354 * All pages have PG_compound set. All tail pages have their ->first_page
355 * pointing at the head page.
357 * The first tail page's ->lru.next holds the address of the compound page's
358 * put_page() function. Its ->lru.prev holds the order of allocation.
359 * This usage means that zero-order pages may not be compound.
362 static void free_compound_page(struct page *page)
364 __free_pages_ok(page, compound_order(page));
367 void prep_compound_page(struct page *page, unsigned long order)
370 int nr_pages = 1 << order;
372 set_compound_page_dtor(page, free_compound_page);
373 set_compound_order(page, order);
375 for (i = 1; i < nr_pages; i++) {
376 struct page *p = page + i;
377 set_page_count(p, 0);
378 p->first_page = page;
379 /* Make sure p->first_page is always valid for PageTail() */
385 /* update __split_huge_page_refcount if you change this function */
386 static int destroy_compound_page(struct page *page, unsigned long order)
389 int nr_pages = 1 << order;
392 if (unlikely(compound_order(page) != order)) {
397 __ClearPageHead(page);
399 for (i = 1; i < nr_pages; i++) {
400 struct page *p = page + i;
402 if (unlikely(!PageTail(p) || (p->first_page != page))) {
412 static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
417 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
418 * and __GFP_HIGHMEM from hard or soft interrupt context.
420 VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
421 for (i = 0; i < (1 << order); i++)
422 clear_highpage(page + i);
425 #ifdef CONFIG_DEBUG_PAGEALLOC
426 unsigned int _debug_guardpage_minorder;
428 static int __init debug_guardpage_minorder_setup(char *buf)
432 if (kstrtoul(buf, 10, &res) < 0 || res > MAX_ORDER / 2) {
433 printk(KERN_ERR "Bad debug_guardpage_minorder value\n");
436 _debug_guardpage_minorder = res;
437 printk(KERN_INFO "Setting debug_guardpage_minorder to %lu\n", res);
440 __setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup);
442 static inline void set_page_guard_flag(struct page *page)
444 __set_bit(PAGE_DEBUG_FLAG_GUARD, &page->debug_flags);
447 static inline void clear_page_guard_flag(struct page *page)
449 __clear_bit(PAGE_DEBUG_FLAG_GUARD, &page->debug_flags);
452 static inline void set_page_guard_flag(struct page *page) { }
453 static inline void clear_page_guard_flag(struct page *page) { }
456 static inline void set_page_order(struct page *page, int order)
458 set_page_private(page, order);
459 __SetPageBuddy(page);
462 static inline void rmv_page_order(struct page *page)
464 __ClearPageBuddy(page);
465 set_page_private(page, 0);
469 * Locate the struct page for both the matching buddy in our
470 * pair (buddy1) and the combined O(n+1) page they form (page).
472 * 1) Any buddy B1 will have an order O twin B2 which satisfies
473 * the following equation:
475 * For example, if the starting buddy (buddy2) is #8 its order
477 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
479 * 2) Any buddy B will have an order O+1 parent P which
480 * satisfies the following equation:
483 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
485 static inline unsigned long
486 __find_buddy_index(unsigned long page_idx, unsigned int order)
488 return page_idx ^ (1 << order);
492 * This function checks whether a page is free && is the buddy
493 * we can do coalesce a page and its buddy if
494 * (a) the buddy is not in a hole &&
495 * (b) the buddy is in the buddy system &&
496 * (c) a page and its buddy have the same order &&
497 * (d) a page and its buddy are in the same zone.
499 * For recording whether a page is in the buddy system, we set ->_mapcount -2.
500 * Setting, clearing, and testing _mapcount -2 is serialized by zone->lock.
502 * For recording page's order, we use page_private(page).
504 static inline int page_is_buddy(struct page *page, struct page *buddy,
507 if (!pfn_valid_within(page_to_pfn(buddy)))
510 if (page_zone_id(page) != page_zone_id(buddy))
513 if (page_is_guard(buddy) && page_order(buddy) == order) {
514 VM_BUG_ON(page_count(buddy) != 0);
518 if (PageBuddy(buddy) && page_order(buddy) == order) {
519 VM_BUG_ON(page_count(buddy) != 0);
526 * Freeing function for a buddy system allocator.
528 * The concept of a buddy system is to maintain direct-mapped table
529 * (containing bit values) for memory blocks of various "orders".
530 * The bottom level table contains the map for the smallest allocatable
531 * units of memory (here, pages), and each level above it describes
532 * pairs of units from the levels below, hence, "buddies".
533 * At a high level, all that happens here is marking the table entry
534 * at the bottom level available, and propagating the changes upward
535 * as necessary, plus some accounting needed to play nicely with other
536 * parts of the VM system.
537 * At each level, we keep a list of pages, which are heads of continuous
538 * free pages of length of (1 << order) and marked with _mapcount -2. Page's
539 * order is recorded in page_private(page) field.
540 * So when we are allocating or freeing one, we can derive the state of the
541 * other. That is, if we allocate a small block, and both were
542 * free, the remainder of the region must be split into blocks.
543 * If a block is freed, and its buddy is also free, then this
544 * triggers coalescing into a block of larger size.
549 static inline void __free_one_page(struct page *page,
550 struct zone *zone, unsigned int order,
553 unsigned long page_idx;
554 unsigned long combined_idx;
555 unsigned long uninitialized_var(buddy_idx);
558 VM_BUG_ON(!zone_is_initialized(zone));
560 if (unlikely(PageCompound(page)))
561 if (unlikely(destroy_compound_page(page, order)))
564 VM_BUG_ON(migratetype == -1);
566 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
568 VM_BUG_ON(page_idx & ((1 << order) - 1));
569 VM_BUG_ON(bad_range(zone, page));
571 while (order < MAX_ORDER-1) {
572 buddy_idx = __find_buddy_index(page_idx, order);
573 buddy = page + (buddy_idx - page_idx);
574 if (!page_is_buddy(page, buddy, order))
577 * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
578 * merge with it and move up one order.
580 if (page_is_guard(buddy)) {
581 clear_page_guard_flag(buddy);
582 set_page_private(page, 0);
583 __mod_zone_freepage_state(zone, 1 << order,
586 list_del(&buddy->lru);
587 zone->free_area[order].nr_free--;
588 rmv_page_order(buddy);
590 combined_idx = buddy_idx & page_idx;
591 page = page + (combined_idx - page_idx);
592 page_idx = combined_idx;
595 set_page_order(page, order);
598 * If this is not the largest possible page, check if the buddy
599 * of the next-highest order is free. If it is, it's possible
600 * that pages are being freed that will coalesce soon. In case,
601 * that is happening, add the free page to the tail of the list
602 * so it's less likely to be used soon and more likely to be merged
603 * as a higher order page
605 if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) {
606 struct page *higher_page, *higher_buddy;
607 combined_idx = buddy_idx & page_idx;
608 higher_page = page + (combined_idx - page_idx);
609 buddy_idx = __find_buddy_index(combined_idx, order + 1);
610 higher_buddy = higher_page + (buddy_idx - combined_idx);
611 if (page_is_buddy(higher_page, higher_buddy, order + 1)) {
612 list_add_tail(&page->lru,
613 &zone->free_area[order].free_list[migratetype]);
618 list_add(&page->lru, &zone->free_area[order].free_list[migratetype]);
620 zone->free_area[order].nr_free++;
623 static inline int free_pages_check(struct page *page)
625 if (unlikely(page_mapcount(page) |
626 (page->mapping != NULL) |
627 (atomic_read(&page->_count) != 0) |
628 (page->flags & PAGE_FLAGS_CHECK_AT_FREE) |
629 (mem_cgroup_bad_page_check(page)))) {
633 page_nid_reset_last(page);
634 if (page->flags & PAGE_FLAGS_CHECK_AT_PREP)
635 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
640 * Frees a number of pages from the PCP lists
641 * Assumes all pages on list are in same zone, and of same order.
642 * count is the number of pages to free.
644 * If the zone was previously in an "all pages pinned" state then look to
645 * see if this freeing clears that state.
647 * And clear the zone's pages_scanned counter, to hold off the "all pages are
648 * pinned" detection logic.
650 static void free_pcppages_bulk(struct zone *zone, int count,
651 struct per_cpu_pages *pcp)
657 spin_lock(&zone->lock);
658 zone->pages_scanned = 0;
662 struct list_head *list;
665 * Remove pages from lists in a round-robin fashion. A
666 * batch_free count is maintained that is incremented when an
667 * empty list is encountered. This is so more pages are freed
668 * off fuller lists instead of spinning excessively around empty
673 if (++migratetype == MIGRATE_PCPTYPES)
675 list = &pcp->lists[migratetype];
676 } while (list_empty(list));
678 /* This is the only non-empty list. Free them all. */
679 if (batch_free == MIGRATE_PCPTYPES)
680 batch_free = to_free;
683 int mt; /* migratetype of the to-be-freed page */
685 page = list_entry(list->prev, struct page, lru);
686 /* must delete as __free_one_page list manipulates */
687 list_del(&page->lru);
688 mt = get_freepage_migratetype(page);
689 /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */
690 __free_one_page(page, zone, 0, mt);
691 trace_mm_page_pcpu_drain(page, 0, mt);
692 if (likely(!is_migrate_isolate_page(page))) {
693 __mod_zone_page_state(zone, NR_FREE_PAGES, 1);
694 if (is_migrate_cma(mt))
695 __mod_zone_page_state(zone, NR_FREE_CMA_PAGES, 1);
697 } while (--to_free && --batch_free && !list_empty(list));
699 spin_unlock(&zone->lock);
702 static void free_one_page(struct zone *zone, struct page *page, int order,
705 spin_lock(&zone->lock);
706 zone->pages_scanned = 0;
708 __free_one_page(page, zone, order, migratetype);
709 if (unlikely(!is_migrate_isolate(migratetype)))
710 __mod_zone_freepage_state(zone, 1 << order, migratetype);
711 spin_unlock(&zone->lock);
714 static bool free_pages_prepare(struct page *page, unsigned int order)
719 trace_mm_page_free(page, order);
720 kmemcheck_free_shadow(page, order);
723 page->mapping = NULL;
724 for (i = 0; i < (1 << order); i++)
725 bad += free_pages_check(page + i);
729 if (!PageHighMem(page)) {
730 debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order);
731 debug_check_no_obj_freed(page_address(page),
734 arch_free_page(page, order);
735 kernel_map_pages(page, 1 << order, 0);
740 static void __free_pages_ok(struct page *page, unsigned int order)
745 if (!free_pages_prepare(page, order))
748 local_irq_save(flags);
749 __count_vm_events(PGFREE, 1 << order);
750 migratetype = get_pageblock_migratetype(page);
751 set_freepage_migratetype(page, migratetype);
752 free_one_page(page_zone(page), page, order, migratetype);
753 local_irq_restore(flags);
757 * Read access to zone->managed_pages is safe because it's unsigned long,
758 * but we still need to serialize writers. Currently all callers of
759 * __free_pages_bootmem() except put_page_bootmem() should only be used
760 * at boot time. So for shorter boot time, we shift the burden to
761 * put_page_bootmem() to serialize writers.
763 void __meminit __free_pages_bootmem(struct page *page, unsigned int order)
765 unsigned int nr_pages = 1 << order;
769 for (loop = 0; loop < nr_pages; loop++) {
770 struct page *p = &page[loop];
772 if (loop + 1 < nr_pages)
774 __ClearPageReserved(p);
775 set_page_count(p, 0);
778 page_zone(page)->managed_pages += 1 << order;
779 set_page_refcounted(page);
780 #ifndef CONFIG_SPRD_PAGERECORDER
781 __free_pages(page, order);
783 __free_pages_nopagedebug(page, order);
788 bool is_cma_pageblock(struct page *page)
790 return get_pageblock_migratetype(page) == MIGRATE_CMA;
793 /* Free whole pageblock and set it's migration type to MIGRATE_CMA. */
794 void __init init_cma_reserved_pageblock(struct page *page)
796 unsigned i = pageblock_nr_pages;
797 struct page *p = page;
800 __ClearPageReserved(p);
801 set_page_count(p, 0);
804 set_page_refcounted(page);
805 set_pageblock_migratetype(page, MIGRATE_CMA);
806 __free_pages(page, pageblock_order);
807 totalram_pages += pageblock_nr_pages;
808 #ifdef CONFIG_HIGHMEM
809 if (PageHighMem(page))
810 totalhigh_pages += pageblock_nr_pages;
816 * The order of subdivision here is critical for the IO subsystem.
817 * Please do not alter this order without good reasons and regression
818 * testing. Specifically, as large blocks of memory are subdivided,
819 * the order in which smaller blocks are delivered depends on the order
820 * they're subdivided in this function. This is the primary factor
821 * influencing the order in which pages are delivered to the IO
822 * subsystem according to empirical testing, and this is also justified
823 * by considering the behavior of a buddy system containing a single
824 * large block of memory acted on by a series of small allocations.
825 * This behavior is a critical factor in sglist merging's success.
829 static inline void expand(struct zone *zone, struct page *page,
830 int low, int high, struct free_area *area,
833 unsigned long size = 1 << high;
839 VM_BUG_ON(bad_range(zone, &page[size]));
841 #ifdef CONFIG_DEBUG_PAGEALLOC
842 if (high < debug_guardpage_minorder()) {
844 * Mark as guard pages (or page), that will allow to
845 * merge back to allocator when buddy will be freed.
846 * Corresponding page table entries will not be touched,
847 * pages will stay not present in virtual address space
849 INIT_LIST_HEAD(&page[size].lru);
850 set_page_guard_flag(&page[size]);
851 set_page_private(&page[size], high);
852 /* Guard pages are not available for any usage */
853 __mod_zone_freepage_state(zone, -(1 << high),
858 list_add(&page[size].lru, &area->free_list[migratetype]);
860 set_page_order(&page[size], high);
865 * This page is about to be returned from the page allocator
867 static inline int check_new_page(struct page *page)
869 if (unlikely(page_mapcount(page) |
870 (page->mapping != NULL) |
871 (atomic_read(&page->_count) != 0) |
872 (page->flags & PAGE_FLAGS_CHECK_AT_PREP) |
873 (mem_cgroup_bad_page_check(page)))) {
880 static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
884 for (i = 0; i < (1 << order); i++) {
885 struct page *p = page + i;
886 if (unlikely(check_new_page(p)))
890 set_page_private(page, 0);
891 set_page_refcounted(page);
893 arch_alloc_page(page, order);
894 kernel_map_pages(page, 1 << order, 1);
896 if (gfp_flags & __GFP_ZERO)
897 prep_zero_page(page, order, gfp_flags);
899 if (order && (gfp_flags & __GFP_COMP))
900 prep_compound_page(page, order);
906 * Go through the free lists for the given migratetype and remove
907 * the smallest available page from the freelists
910 struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
913 unsigned int current_order;
914 struct free_area * area;
917 /* Find a page of the appropriate size in the preferred list */
918 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
919 area = &(zone->free_area[current_order]);
920 if (list_empty(&area->free_list[migratetype]))
923 page = list_entry(area->free_list[migratetype].next,
925 list_del(&page->lru);
926 rmv_page_order(page);
928 expand(zone, page, order, current_order, area, migratetype);
937 * This array describes the order lists are fallen back to when
938 * the free lists for the desirable migrate type are depleted
940 static int fallbacks[MIGRATE_TYPES][4] = {
941 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
942 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
944 [MIGRATE_MOVABLE] = { MIGRATE_CMA, MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
945 [MIGRATE_CMA] = { MIGRATE_RESERVE }, /* Never used */
947 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
949 [MIGRATE_RESERVE] = { MIGRATE_RESERVE }, /* Never used */
950 #ifdef CONFIG_MEMORY_ISOLATION
951 [MIGRATE_ISOLATE] = { MIGRATE_RESERVE }, /* Never used */
955 int *get_migratetype_fallbacks(int mtype)
957 return fallbacks[mtype];
961 * Move the free pages in a range to the free lists of the requested type.
962 * Note that start_page and end_pages are not aligned on a pageblock
963 * boundary. If alignment is required, use move_freepages_block()
965 int move_freepages(struct zone *zone,
966 struct page *start_page, struct page *end_page,
973 #ifndef CONFIG_HOLES_IN_ZONE
975 * page_zone is not safe to call in this context when
976 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
977 * anyway as we check zone boundaries in move_freepages_block().
978 * Remove at a later date when no bug reports exist related to
979 * grouping pages by mobility
981 BUG_ON(page_zone(start_page) != page_zone(end_page));
984 for (page = start_page; page <= end_page;) {
985 /* Make sure we are not inadvertently changing nodes */
986 VM_BUG_ON(page_to_nid(page) != zone_to_nid(zone));
988 if (!pfn_valid_within(page_to_pfn(page))) {
993 if (!PageBuddy(page)) {
998 order = page_order(page);
999 list_move(&page->lru,
1000 &zone->free_area[order].free_list[migratetype]);
1001 set_freepage_migratetype(page, migratetype);
1003 pages_moved += 1 << order;
1009 int move_freepages_block(struct zone *zone, struct page *page,
1012 unsigned long start_pfn, end_pfn;
1013 struct page *start_page, *end_page;
1015 start_pfn = page_to_pfn(page);
1016 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
1017 start_page = pfn_to_page(start_pfn);
1018 end_page = start_page + pageblock_nr_pages - 1;
1019 end_pfn = start_pfn + pageblock_nr_pages - 1;
1021 /* Do not cross zone boundaries */
1022 if (!zone_spans_pfn(zone, start_pfn))
1024 if (!zone_spans_pfn(zone, end_pfn))
1027 return move_freepages(zone, start_page, end_page, migratetype);
1030 static void change_pageblock_range(struct page *pageblock_page,
1031 int start_order, int migratetype)
1033 int nr_pageblocks = 1 << (start_order - pageblock_order);
1035 while (nr_pageblocks--) {
1036 set_pageblock_migratetype(pageblock_page, migratetype);
1037 pageblock_page += pageblock_nr_pages;
1041 /* Remove an element from the buddy allocator from the fallback list */
1042 static inline struct page *
1043 __rmqueue_fallback(struct zone *zone, int order, int start_migratetype)
1045 struct free_area * area;
1050 /* Find the largest possible block of pages in the other list */
1051 for (current_order = MAX_ORDER-1; current_order >= order;
1054 migratetype = fallbacks[start_migratetype][i];
1056 /* MIGRATE_RESERVE handled later if necessary */
1057 if (migratetype == MIGRATE_RESERVE)
1060 area = &(zone->free_area[current_order]);
1061 if (list_empty(&area->free_list[migratetype]))
1064 page = list_entry(area->free_list[migratetype].next,
1069 * If breaking a large block of pages, move all free
1070 * pages to the preferred allocation list. If falling
1071 * back for a reclaimable kernel allocation, be more
1072 * aggressive about taking ownership of free pages
1074 * On the other hand, never change migration
1075 * type of MIGRATE_CMA pageblocks nor move CMA
1076 * pages on different free lists. We don't
1077 * want unmovable pages to be allocated from
1078 * MIGRATE_CMA areas.
1080 if (!is_migrate_cma(migratetype) &&
1081 (unlikely(current_order >= pageblock_order / 2) ||
1082 start_migratetype == MIGRATE_RECLAIMABLE ||
1083 page_group_by_mobility_disabled)) {
1085 pages = move_freepages_block(zone, page,
1088 /* Claim the whole block if over half of it is free */
1089 if (pages >= (1 << (pageblock_order-1)) ||
1090 page_group_by_mobility_disabled)
1091 set_pageblock_migratetype(page,
1094 migratetype = start_migratetype;
1097 /* Remove the page from the freelists */
1098 list_del(&page->lru);
1099 rmv_page_order(page);
1101 /* Take ownership for orders >= pageblock_order */
1102 if (current_order >= pageblock_order &&
1103 !is_migrate_cma(migratetype))
1104 change_pageblock_range(page, current_order,
1107 expand(zone, page, order, current_order, area,
1108 is_migrate_cma(migratetype)
1109 ? migratetype : start_migratetype);
1111 trace_mm_page_alloc_extfrag(page, order, current_order,
1112 start_migratetype, migratetype);
1122 * Do the hard work of removing an element from the buddy allocator.
1123 * Call me with the zone->lock already held.
1125 static struct page *__rmqueue(struct zone *zone, unsigned int order,
1131 page = __rmqueue_smallest(zone, order, migratetype);
1133 if (unlikely(!page) && migratetype != MIGRATE_RESERVE) {
1134 page = __rmqueue_fallback(zone, order, migratetype);
1137 * Use MIGRATE_RESERVE rather than fail an allocation. goto
1138 * is used because __rmqueue_smallest is an inline function
1139 * and we want just one call site
1142 migratetype = MIGRATE_RESERVE;
1147 trace_mm_page_alloc_zone_locked(page, order, migratetype);
1151 #ifdef CONFIG_CMA_RMQUEUE
1152 static struct page *__rmqueue_cma(struct zone *zone, unsigned int order,
1155 struct page *page = 0;
1157 if (migratetype == MIGRATE_MOVABLE && !zone->cma_alloc)
1158 page = __rmqueue_smallest(zone, order, MIGRATE_CMA);
1162 page = __rmqueue_smallest(zone, order, migratetype);
1165 if (unlikely(!page) && migratetype != MIGRATE_RESERVE) {
1166 page = __rmqueue_fallback(zone, order, migratetype);
1169 * Use MIGRATE_RESERVE rather than fail an allocation. goto
1170 * is used because __rmqueue_smallest is an inline function
1171 * and we want just one call site
1174 migratetype = MIGRATE_RESERVE;
1179 trace_mm_page_alloc_zone_locked(page, order, migratetype);
1185 * Obtain a specified number of elements from the buddy allocator, all under
1186 * a single hold of the lock, for efficiency. Add them to the supplied list.
1187 * Returns the number of new pages which were placed at *list.
1189 #ifndef CONFIG_CMA_RMQUEUE
1190 static int rmqueue_bulk(struct zone *zone, unsigned int order,
1191 unsigned long count, struct list_head *list,
1192 int migratetype, int cold)
1194 static int rmqueue_bulk(struct zone *zone, unsigned int order,
1195 unsigned long count, struct list_head *list,
1196 int migratetype, int cold, int cma)
1199 int mt = migratetype, i;
1201 spin_lock(&zone->lock);
1202 for (i = 0; i < count; ++i) {
1204 #ifdef CONFIG_CMA_RMQUEUE
1206 page = __rmqueue_cma(zone, order, migratetype);
1209 page = __rmqueue(zone, order, migratetype);
1210 if (unlikely(page == NULL))
1214 * Split buddy pages returned by expand() are received here
1215 * in physical page order. The page is added to the callers and
1216 * list and the list head then moves forward. From the callers
1217 * perspective, the linked list is ordered by page number in
1218 * some conditions. This is useful for IO devices that can
1219 * merge IO requests if the physical pages are ordered
1222 if (likely(cold == 0))
1223 list_add(&page->lru, list);
1225 list_add_tail(&page->lru, list);
1226 if (IS_ENABLED(CONFIG_CMA)) {
1227 mt = get_pageblock_migratetype(page);
1228 if (!is_migrate_cma(mt) && !is_migrate_isolate(mt))
1231 set_freepage_migratetype(page, mt);
1233 if (is_migrate_cma(mt))
1234 __mod_zone_page_state(zone, NR_FREE_CMA_PAGES,
1237 __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
1238 spin_unlock(&zone->lock);
1244 * Called from the vmstat counter updater to drain pagesets of this
1245 * currently executing processor on remote nodes after they have
1248 * Note that this function must be called with the thread pinned to
1249 * a single processor.
1251 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
1253 unsigned long flags;
1256 local_irq_save(flags);
1257 if (pcp->count >= pcp->batch)
1258 to_drain = pcp->batch;
1260 to_drain = pcp->count;
1262 free_pcppages_bulk(zone, to_drain, pcp);
1263 pcp->count -= to_drain;
1265 local_irq_restore(flags);
1270 * Drain pages of the indicated processor.
1272 * The processor must either be the current processor and the
1273 * thread pinned to the current processor or a processor that
1276 static void drain_pages(unsigned int cpu)
1278 unsigned long flags;
1281 for_each_populated_zone(zone) {
1282 struct per_cpu_pageset *pset;
1283 struct per_cpu_pages *pcp;
1285 local_irq_save(flags);
1286 pset = per_cpu_ptr(zone->pageset, cpu);
1290 free_pcppages_bulk(zone, pcp->count, pcp);
1293 local_irq_restore(flags);
1298 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
1300 void drain_local_pages(void *arg)
1302 drain_pages(smp_processor_id());
1306 * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
1308 * Note that this code is protected against sending an IPI to an offline
1309 * CPU but does not guarantee sending an IPI to newly hotplugged CPUs:
1310 * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but
1311 * nothing keeps CPUs from showing up after we populated the cpumask and
1312 * before the call to on_each_cpu_mask().
1314 void drain_all_pages(void)
1317 struct per_cpu_pageset *pcp;
1321 * Allocate in the BSS so we wont require allocation in
1322 * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
1324 static cpumask_t cpus_with_pcps;
1327 * We don't care about racing with CPU hotplug event
1328 * as offline notification will cause the notified
1329 * cpu to drain that CPU pcps and on_each_cpu_mask
1330 * disables preemption as part of its processing
1332 for_each_online_cpu(cpu) {
1333 bool has_pcps = false;
1334 for_each_populated_zone(zone) {
1335 pcp = per_cpu_ptr(zone->pageset, cpu);
1336 if (pcp->pcp.count) {
1342 cpumask_set_cpu(cpu, &cpus_with_pcps);
1344 cpumask_clear_cpu(cpu, &cpus_with_pcps);
1346 on_each_cpu_mask(&cpus_with_pcps, drain_local_pages, NULL, 1);
1349 #ifdef CONFIG_HIBERNATION
1351 void mark_free_pages(struct zone *zone)
1353 unsigned long pfn, max_zone_pfn;
1354 unsigned long flags;
1356 struct list_head *curr;
1358 if (!zone->spanned_pages)
1361 spin_lock_irqsave(&zone->lock, flags);
1363 max_zone_pfn = zone_end_pfn(zone);
1364 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1365 if (pfn_valid(pfn)) {
1366 struct page *page = pfn_to_page(pfn);
1368 if (!swsusp_page_is_forbidden(page))
1369 swsusp_unset_page_free(page);
1372 for_each_migratetype_order(order, t) {
1373 list_for_each(curr, &zone->free_area[order].free_list[t]) {
1376 pfn = page_to_pfn(list_entry(curr, struct page, lru));
1377 for (i = 0; i < (1UL << order); i++)
1378 swsusp_set_page_free(pfn_to_page(pfn + i));
1381 spin_unlock_irqrestore(&zone->lock, flags);
1383 #endif /* CONFIG_PM */
1386 * Free a 0-order page
1387 * cold == 1 ? free a cold page : free a hot page
1389 void free_hot_cold_page(struct page *page, int cold)
1391 struct zone *zone = page_zone(page);
1392 struct per_cpu_pages *pcp;
1393 unsigned long flags;
1396 if (!free_pages_prepare(page, 0))
1399 migratetype = get_pageblock_migratetype(page);
1400 set_freepage_migratetype(page, migratetype);
1401 local_irq_save(flags);
1402 __count_vm_event(PGFREE);
1405 * We only track unmovable, reclaimable and movable on pcp lists.
1406 * Free ISOLATE pages back to the allocator because they are being
1407 * offlined but treat RESERVE as movable pages so we can get those
1408 * areas back if necessary. Otherwise, we may have to free
1409 * excessively into the page allocator
1411 if (migratetype >= MIGRATE_PCPTYPES) {
1412 if (unlikely(is_migrate_isolate(migratetype)) ||
1413 is_migrate_cma(migratetype)) {
1414 free_one_page(zone, page, 0, migratetype);
1417 migratetype = MIGRATE_MOVABLE;
1420 pcp = &this_cpu_ptr(zone->pageset)->pcp;
1422 list_add_tail(&page->lru, &pcp->lists[migratetype]);
1424 list_add(&page->lru, &pcp->lists[migratetype]);
1426 if (pcp->count >= pcp->high) {
1427 free_pcppages_bulk(zone, pcp->batch, pcp);
1428 pcp->count -= pcp->batch;
1432 local_irq_restore(flags);
1436 * Free a list of 0-order pages
1438 void free_hot_cold_page_list(struct list_head *list, int cold)
1440 struct page *page, *next;
1442 list_for_each_entry_safe(page, next, list, lru) {
1443 trace_mm_page_free_batched(page, cold);
1444 free_hot_cold_page(page, cold);
1449 * split_page takes a non-compound higher-order page, and splits it into
1450 * n (1<<order) sub-pages: page[0..n]
1451 * Each sub-page must be freed individually.
1453 * Note: this is probably too low level an operation for use in drivers.
1454 * Please consult with lkml before using this in your driver.
1456 void split_page(struct page *page, unsigned int order)
1460 VM_BUG_ON(PageCompound(page));
1461 VM_BUG_ON(!page_count(page));
1463 #ifdef CONFIG_KMEMCHECK
1465 * Split shadow pages too, because free(page[0]) would
1466 * otherwise free the whole shadow.
1468 if (kmemcheck_page_is_tracked(page))
1469 split_page(virt_to_page(page[0].shadow), order);
1472 for (i = 1; i < (1 << order); i++)
1473 set_page_refcounted(page + i);
1475 EXPORT_SYMBOL_GPL(split_page);
1477 static int __isolate_free_page(struct page *page, unsigned int order)
1479 unsigned long watermark;
1483 BUG_ON(!PageBuddy(page));
1485 zone = page_zone(page);
1486 mt = get_pageblock_migratetype(page);
1488 if (!is_migrate_isolate(mt)) {
1489 /* Obey watermarks as if the page was being allocated */
1490 watermark = low_wmark_pages(zone) + (1 << order);
1491 if (!is_migrate_cma(mt) &&
1492 !zone_watermark_ok(zone, 0, watermark, 0, 0))
1495 __mod_zone_freepage_state(zone, -(1UL << order), mt);
1498 /* Remove page from free list */
1499 list_del(&page->lru);
1500 zone->free_area[order].nr_free--;
1501 rmv_page_order(page);
1503 /* Set the pageblock if the isolated page is at least a pageblock */
1504 if (order >= pageblock_order - 1) {
1505 struct page *endpage = page + (1 << order) - 1;
1506 for (; page < endpage; page += pageblock_nr_pages) {
1507 int mt = get_pageblock_migratetype(page);
1508 if (!is_migrate_isolate(mt) && !is_migrate_cma(mt))
1509 set_pageblock_migratetype(page,
1514 return 1UL << order;
1518 * Similar to split_page except the page is already free. As this is only
1519 * being used for migration, the migratetype of the block also changes.
1520 * As this is called with interrupts disabled, the caller is responsible
1521 * for calling arch_alloc_page() and kernel_map_page() after interrupts
1524 * Note: this is probably too low level an operation for use in drivers.
1525 * Please consult with lkml before using this in your driver.
1527 int split_free_page(struct page *page)
1532 order = page_order(page);
1534 nr_pages = __isolate_free_page(page, order);
1538 /* Split into individual pages */
1539 set_page_refcounted(page);
1540 split_page(page, order);
1545 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1546 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1550 struct page *buffered_rmqueue(struct zone *preferred_zone,
1551 struct zone *zone, int order, gfp_t gfp_flags,
1554 unsigned long flags;
1556 int cold = !!(gfp_flags & __GFP_COLD);
1559 if (likely(order == 0)) {
1560 struct per_cpu_pages *pcp;
1561 struct list_head *list;
1563 local_irq_save(flags);
1564 pcp = &this_cpu_ptr(zone->pageset)->pcp;
1565 list = &pcp->lists[migratetype];
1566 if (list_empty(list)) {
1567 #ifndef CONFIG_CMA_RMQUEUE
1568 pcp->count += rmqueue_bulk(zone, 0,
1572 pcp->count += rmqueue_bulk(zone, 0,
1575 gfp_flags & __GFP_CMA);
1577 if (unlikely(list_empty(list)))
1582 page = list_entry(list->prev, struct page, lru);
1584 page = list_entry(list->next, struct page, lru);
1586 list_del(&page->lru);
1589 if (unlikely(gfp_flags & __GFP_NOFAIL)) {
1591 * __GFP_NOFAIL is not to be used in new code.
1593 * All __GFP_NOFAIL callers should be fixed so that they
1594 * properly detect and handle allocation failures.
1596 * We most definitely don't want callers attempting to
1597 * allocate greater than order-1 page units with
1600 WARN_ON_ONCE(order > 1);
1602 spin_lock_irqsave(&zone->lock, flags);
1603 #ifdef CONFIG_CMA_RMQUEUE
1604 if (gfp_flags & __GFP_CMA)
1605 page = __rmqueue_cma(zone, order, migratetype);
1608 page = __rmqueue(zone, order, migratetype);
1609 spin_unlock(&zone->lock);
1612 __mod_zone_freepage_state(zone, -(1 << order),
1613 get_pageblock_migratetype(page));
1616 __count_zone_vm_events(PGALLOC, zone, 1 << order);
1617 zone_statistics(preferred_zone, zone, gfp_flags);
1618 local_irq_restore(flags);
1620 VM_BUG_ON(bad_range(zone, page));
1621 if (prep_new_page(page, order, gfp_flags))
1626 local_irq_restore(flags);
1630 #ifdef CONFIG_FAIL_PAGE_ALLOC
1633 struct fault_attr attr;
1635 u32 ignore_gfp_highmem;
1636 u32 ignore_gfp_wait;
1638 } fail_page_alloc = {
1639 .attr = FAULT_ATTR_INITIALIZER,
1640 .ignore_gfp_wait = 1,
1641 .ignore_gfp_highmem = 1,
1645 static int __init setup_fail_page_alloc(char *str)
1647 return setup_fault_attr(&fail_page_alloc.attr, str);
1649 __setup("fail_page_alloc=", setup_fail_page_alloc);
1651 static bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1653 if (order < fail_page_alloc.min_order)
1655 if (gfp_mask & __GFP_NOFAIL)
1657 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
1659 if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
1662 return should_fail(&fail_page_alloc.attr, 1 << order);
1665 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1667 static int __init fail_page_alloc_debugfs(void)
1669 umode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
1672 dir = fault_create_debugfs_attr("fail_page_alloc", NULL,
1673 &fail_page_alloc.attr);
1675 return PTR_ERR(dir);
1677 if (!debugfs_create_bool("ignore-gfp-wait", mode, dir,
1678 &fail_page_alloc.ignore_gfp_wait))
1680 if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir,
1681 &fail_page_alloc.ignore_gfp_highmem))
1683 if (!debugfs_create_u32("min-order", mode, dir,
1684 &fail_page_alloc.min_order))
1689 debugfs_remove_recursive(dir);
1694 late_initcall(fail_page_alloc_debugfs);
1696 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1698 #else /* CONFIG_FAIL_PAGE_ALLOC */
1700 static inline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1705 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1708 * Return true if free pages are above 'mark'. This takes into account the order
1709 * of the allocation.
1711 static bool __zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1712 int classzone_idx, int alloc_flags, long free_pages)
1714 /* free_pages my go negative - that's OK */
1716 long lowmem_reserve = z->lowmem_reserve[classzone_idx];
1720 free_pages -= (1 << order) - 1;
1721 if (alloc_flags & ALLOC_HIGH)
1723 if (alloc_flags & ALLOC_HARDER)
1726 /* If allocation can't use CMA areas don't use free CMA pages */
1727 if (!(alloc_flags & ALLOC_CMA))
1728 free_cma = zone_page_state(z, NR_FREE_CMA_PAGES);
1731 if (free_pages - free_cma <= min + lowmem_reserve)
1733 for (o = 0; o < order; o++) {
1734 /* At the next order, this order's pages become unavailable */
1735 free_pages -= z->free_area[o].nr_free << o;
1737 /* Require fewer higher order pages to be free */
1738 min >>= min_free_order_shift;
1740 if (free_pages <= min)
1746 bool zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1747 int classzone_idx, int alloc_flags)
1749 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
1750 zone_page_state(z, NR_FREE_PAGES));
1753 bool zone_watermark_ok_safe(struct zone *z, int order, unsigned long mark,
1754 int classzone_idx, int alloc_flags)
1756 long free_pages = zone_page_state(z, NR_FREE_PAGES);
1758 if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark)
1759 free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES);
1761 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
1767 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1768 * skip over zones that are not allowed by the cpuset, or that have
1769 * been recently (in last second) found to be nearly full. See further
1770 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1771 * that have to skip over a lot of full or unallowed zones.
1773 * If the zonelist cache is present in the passed in zonelist, then
1774 * returns a pointer to the allowed node mask (either the current
1775 * tasks mems_allowed, or node_states[N_MEMORY].)
1777 * If the zonelist cache is not available for this zonelist, does
1778 * nothing and returns NULL.
1780 * If the fullzones BITMAP in the zonelist cache is stale (more than
1781 * a second since last zap'd) then we zap it out (clear its bits.)
1783 * We hold off even calling zlc_setup, until after we've checked the
1784 * first zone in the zonelist, on the theory that most allocations will
1785 * be satisfied from that first zone, so best to examine that zone as
1786 * quickly as we can.
1788 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1790 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1791 nodemask_t *allowednodes; /* zonelist_cache approximation */
1793 zlc = zonelist->zlcache_ptr;
1797 if (time_after(jiffies, zlc->last_full_zap + HZ)) {
1798 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1799 zlc->last_full_zap = jiffies;
1802 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1803 &cpuset_current_mems_allowed :
1804 &node_states[N_MEMORY];
1805 return allowednodes;
1809 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1810 * if it is worth looking at further for free memory:
1811 * 1) Check that the zone isn't thought to be full (doesn't have its
1812 * bit set in the zonelist_cache fullzones BITMAP).
1813 * 2) Check that the zones node (obtained from the zonelist_cache
1814 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1815 * Return true (non-zero) if zone is worth looking at further, or
1816 * else return false (zero) if it is not.
1818 * This check -ignores- the distinction between various watermarks,
1819 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1820 * found to be full for any variation of these watermarks, it will
1821 * be considered full for up to one second by all requests, unless
1822 * we are so low on memory on all allowed nodes that we are forced
1823 * into the second scan of the zonelist.
1825 * In the second scan we ignore this zonelist cache and exactly
1826 * apply the watermarks to all zones, even it is slower to do so.
1827 * We are low on memory in the second scan, and should leave no stone
1828 * unturned looking for a free page.
1830 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
1831 nodemask_t *allowednodes)
1833 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1834 int i; /* index of *z in zonelist zones */
1835 int n; /* node that zone *z is on */
1837 zlc = zonelist->zlcache_ptr;
1841 i = z - zonelist->_zonerefs;
1844 /* This zone is worth trying if it is allowed but not full */
1845 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1849 * Given 'z' scanning a zonelist, set the corresponding bit in
1850 * zlc->fullzones, so that subsequent attempts to allocate a page
1851 * from that zone don't waste time re-examining it.
1853 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
1855 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1856 int i; /* index of *z in zonelist zones */
1858 zlc = zonelist->zlcache_ptr;
1862 i = z - zonelist->_zonerefs;
1864 set_bit(i, zlc->fullzones);
1868 * clear all zones full, called after direct reclaim makes progress so that
1869 * a zone that was recently full is not skipped over for up to a second
1871 static void zlc_clear_zones_full(struct zonelist *zonelist)
1873 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1875 zlc = zonelist->zlcache_ptr;
1879 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1882 static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
1884 return node_isset(local_zone->node, zone->zone_pgdat->reclaim_nodes);
1887 static void __paginginit init_zone_allows_reclaim(int nid)
1891 for_each_online_node(i)
1892 if (node_distance(nid, i) <= RECLAIM_DISTANCE)
1893 node_set(i, NODE_DATA(nid)->reclaim_nodes);
1895 zone_reclaim_mode = 1;
1898 #else /* CONFIG_NUMA */
1900 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1905 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
1906 nodemask_t *allowednodes)
1911 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
1915 static void zlc_clear_zones_full(struct zonelist *zonelist)
1919 static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
1924 static inline void init_zone_allows_reclaim(int nid)
1927 #endif /* CONFIG_NUMA */
1930 * get_page_from_freelist goes through the zonelist trying to allocate
1933 static struct page *
1934 get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order,
1935 struct zonelist *zonelist, int high_zoneidx, int alloc_flags,
1936 struct zone *preferred_zone, int migratetype)
1939 struct page *page = NULL;
1942 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1943 int zlc_active = 0; /* set if using zonelist_cache */
1944 int did_zlc_setup = 0; /* just call zlc_setup() one time */
1946 classzone_idx = zone_idx(preferred_zone);
1949 * Scan zonelist, looking for a zone with enough free.
1950 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1952 for_each_zone_zonelist_nodemask(zone, z, zonelist,
1953 high_zoneidx, nodemask) {
1954 if (IS_ENABLED(CONFIG_NUMA) && zlc_active &&
1955 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1957 if ((alloc_flags & ALLOC_CPUSET) &&
1958 !cpuset_zone_allowed_softwall(zone, gfp_mask))
1961 * When allocating a page cache page for writing, we
1962 * want to get it from a zone that is within its dirty
1963 * limit, such that no single zone holds more than its
1964 * proportional share of globally allowed dirty pages.
1965 * The dirty limits take into account the zone's
1966 * lowmem reserves and high watermark so that kswapd
1967 * should be able to balance it without having to
1968 * write pages from its LRU list.
1970 * This may look like it could increase pressure on
1971 * lower zones by failing allocations in higher zones
1972 * before they are full. But the pages that do spill
1973 * over are limited as the lower zones are protected
1974 * by this very same mechanism. It should not become
1975 * a practical burden to them.
1977 * XXX: For now, allow allocations to potentially
1978 * exceed the per-zone dirty limit in the slowpath
1979 * (ALLOC_WMARK_LOW unset) before going into reclaim,
1980 * which is important when on a NUMA setup the allowed
1981 * zones are together not big enough to reach the
1982 * global limit. The proper fix for these situations
1983 * will require awareness of zones in the
1984 * dirty-throttling and the flusher threads.
1986 if ((alloc_flags & ALLOC_WMARK_LOW) &&
1987 (gfp_mask & __GFP_WRITE) && !zone_dirty_ok(zone))
1988 goto this_zone_full;
1990 BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
1991 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
1995 mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
1996 if (zone_watermark_ok(zone, order, mark,
1997 classzone_idx, alloc_flags))
2000 if (IS_ENABLED(CONFIG_NUMA) &&
2001 !did_zlc_setup && nr_online_nodes > 1) {
2003 * we do zlc_setup if there are multiple nodes
2004 * and before considering the first zone allowed
2007 allowednodes = zlc_setup(zonelist, alloc_flags);
2012 if (zone_reclaim_mode == 0 ||
2013 !zone_allows_reclaim(preferred_zone, zone))
2014 goto this_zone_full;
2017 * As we may have just activated ZLC, check if the first
2018 * eligible zone has failed zone_reclaim recently.
2020 if (IS_ENABLED(CONFIG_NUMA) && zlc_active &&
2021 !zlc_zone_worth_trying(zonelist, z, allowednodes))
2024 ret = zone_reclaim(zone, gfp_mask, order);
2026 case ZONE_RECLAIM_NOSCAN:
2029 case ZONE_RECLAIM_FULL:
2030 /* scanned but unreclaimable */
2033 /* did we reclaim enough */
2034 if (zone_watermark_ok(zone, order, mark,
2035 classzone_idx, alloc_flags))
2039 * Failed to reclaim enough to meet watermark.
2040 * Only mark the zone full if checking the min
2041 * watermark or if we failed to reclaim just
2042 * 1<<order pages or else the page allocator
2043 * fastpath will prematurely mark zones full
2044 * when the watermark is between the low and
2047 if (((alloc_flags & ALLOC_WMARK_MASK) == ALLOC_WMARK_MIN) ||
2048 ret == ZONE_RECLAIM_SOME)
2049 goto this_zone_full;
2056 page = buffered_rmqueue(preferred_zone, zone, order,
2057 gfp_mask, migratetype);
2061 if (IS_ENABLED(CONFIG_NUMA))
2062 zlc_mark_zone_full(zonelist, z);
2065 if (unlikely(IS_ENABLED(CONFIG_NUMA) && page == NULL && zlc_active)) {
2066 /* Disable zlc cache for second zonelist scan */
2073 * page->pfmemalloc is set when ALLOC_NO_WATERMARKS was
2074 * necessary to allocate the page. The expectation is
2075 * that the caller is taking steps that will free more
2076 * memory. The caller should avoid the page being used
2077 * for !PFMEMALLOC purposes.
2079 page->pfmemalloc = !!(alloc_flags & ALLOC_NO_WATERMARKS);
2085 * Large machines with many possible nodes should not always dump per-node
2086 * meminfo in irq context.
2088 static inline bool should_suppress_show_mem(void)
2093 ret = in_interrupt();
2098 static DEFINE_RATELIMIT_STATE(nopage_rs,
2099 DEFAULT_RATELIMIT_INTERVAL,
2100 DEFAULT_RATELIMIT_BURST);
2102 void warn_alloc_failed(gfp_t gfp_mask, int order, const char *fmt, ...)
2104 unsigned int filter = SHOW_MEM_FILTER_NODES;
2106 if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs) ||
2107 debug_guardpage_minorder() > 0)
2111 * Walking all memory to count page types is very expensive and should
2112 * be inhibited in non-blockable contexts.
2114 if (!(gfp_mask & __GFP_WAIT))
2115 filter |= SHOW_MEM_FILTER_PAGE_COUNT;
2118 * This documents exceptions given to allocations in certain
2119 * contexts that are allowed to allocate outside current's set
2122 if (!(gfp_mask & __GFP_NOMEMALLOC))
2123 if (test_thread_flag(TIF_MEMDIE) ||
2124 (current->flags & (PF_MEMALLOC | PF_EXITING)))
2125 filter &= ~SHOW_MEM_FILTER_NODES;
2126 if (in_interrupt() || !(gfp_mask & __GFP_WAIT))
2127 filter &= ~SHOW_MEM_FILTER_NODES;
2130 struct va_format vaf;
2133 va_start(args, fmt);
2138 pr_warn("%pV", &vaf);
2143 pr_warn("%s: page allocation failure: order:%d, mode:0x%x\n",
2144 current->comm, order, gfp_mask);
2147 if (!should_suppress_show_mem())
2152 should_alloc_retry(gfp_t gfp_mask, unsigned int order,
2153 unsigned long did_some_progress,
2154 unsigned long pages_reclaimed)
2156 /* Do not loop if specifically requested */
2157 if (gfp_mask & __GFP_NORETRY)
2160 /* Always retry if specifically requested */
2161 if (gfp_mask & __GFP_NOFAIL)
2165 * Suspend converts GFP_KERNEL to __GFP_WAIT which can prevent reclaim
2166 * making forward progress without invoking OOM. Suspend also disables
2167 * storage devices so kswapd will not help. Bail if we are suspending.
2169 if (!did_some_progress && pm_suspended_storage())
2173 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
2174 * means __GFP_NOFAIL, but that may not be true in other
2177 if (order <= PAGE_ALLOC_COSTLY_ORDER)
2181 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
2182 * specified, then we retry until we no longer reclaim any pages
2183 * (above), or we've reclaimed an order of pages at least as
2184 * large as the allocation's order. In both cases, if the
2185 * allocation still fails, we stop retrying.
2187 if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order))
2193 static inline struct page *
2194 __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
2195 struct zonelist *zonelist, enum zone_type high_zoneidx,
2196 nodemask_t *nodemask, struct zone *preferred_zone,
2201 /* Acquire the OOM killer lock for the zones in zonelist */
2202 if (!try_set_zonelist_oom(zonelist, gfp_mask)) {
2203 schedule_timeout_uninterruptible(1);
2208 * PM-freezer should be notified that there might be an OOM killer on
2209 * its way to kill and wake somebody up. This is too early and we might
2210 * end up not killing anything but false positives are acceptable.
2211 * See freeze_processes.
2216 * Go through the zonelist yet one more time, keep very high watermark
2217 * here, this is only to catch a parallel oom killing, we must fail if
2218 * we're still under heavy pressure.
2220 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask,
2221 order, zonelist, high_zoneidx,
2222 ALLOC_WMARK_HIGH|ALLOC_CPUSET,
2223 preferred_zone, migratetype);
2227 if (!(gfp_mask & __GFP_NOFAIL)) {
2228 /* The OOM killer will not help higher order allocs */
2229 if (order > PAGE_ALLOC_COSTLY_ORDER)
2231 /* The OOM killer does not needlessly kill tasks for lowmem */
2232 if (high_zoneidx < ZONE_NORMAL)
2235 * GFP_THISNODE contains __GFP_NORETRY and we never hit this.
2236 * Sanity check for bare calls of __GFP_THISNODE, not real OOM.
2237 * The caller should handle page allocation failure by itself if
2238 * it specifies __GFP_THISNODE.
2239 * Note: Hugepage uses it but will hit PAGE_ALLOC_COSTLY_ORDER.
2241 if (gfp_mask & __GFP_THISNODE)
2244 /* Exhausted what can be done so it's blamo time */
2245 out_of_memory(zonelist, gfp_mask, order, nodemask, false);
2248 clear_zonelist_oom(zonelist, gfp_mask);
2252 #ifdef CONFIG_COMPACTION
2253 /* Try memory compaction for high-order allocations before reclaim */
2254 static struct page *
2255 __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
2256 struct zonelist *zonelist, enum zone_type high_zoneidx,
2257 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
2258 int migratetype, bool sync_migration,
2259 bool *contended_compaction, bool *deferred_compaction,
2260 unsigned long *did_some_progress)
2265 if (compaction_deferred(preferred_zone, order)) {
2266 *deferred_compaction = true;
2270 current->flags |= PF_MEMALLOC;
2271 *did_some_progress = try_to_compact_pages(zonelist, order, gfp_mask,
2272 nodemask, sync_migration,
2273 contended_compaction);
2274 current->flags &= ~PF_MEMALLOC;
2276 if (*did_some_progress != COMPACT_SKIPPED) {
2279 /* Page migration frees to the PCP lists but we want merging */
2280 drain_pages(get_cpu());
2283 page = get_page_from_freelist(gfp_mask, nodemask,
2284 order, zonelist, high_zoneidx,
2285 alloc_flags & ~ALLOC_NO_WATERMARKS,
2286 preferred_zone, migratetype);
2288 preferred_zone->compact_blockskip_flush = false;
2289 preferred_zone->compact_considered = 0;
2290 preferred_zone->compact_defer_shift = 0;
2291 if (order >= preferred_zone->compact_order_failed)
2292 preferred_zone->compact_order_failed = order + 1;
2293 count_vm_event(COMPACTSUCCESS);
2298 * It's bad if compaction run occurs and fails.
2299 * The most likely reason is that pages exist,
2300 * but not enough to satisfy watermarks.
2302 count_vm_event(COMPACTFAIL);
2305 * As async compaction considers a subset of pageblocks, only
2306 * defer if the failure was a sync compaction failure.
2309 defer_compaction(preferred_zone, order);
2317 static inline struct page *
2318 __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
2319 struct zonelist *zonelist, enum zone_type high_zoneidx,
2320 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
2321 int migratetype, bool sync_migration,
2322 bool *contended_compaction, bool *deferred_compaction,
2323 unsigned long *did_some_progress)
2325 /* Mark deferred_compaction as true to avoid direct reclaim
2326 * for high-order allocation */
2327 *deferred_compaction = true;
2330 #endif /* CONFIG_COMPACTION */
2332 /* Perform direct synchronous page reclaim */
2334 __perform_reclaim(gfp_t gfp_mask, unsigned int order, struct zonelist *zonelist,
2335 nodemask_t *nodemask)
2337 struct reclaim_state reclaim_state;
2342 /* We now go into synchronous reclaim */
2343 cpuset_memory_pressure_bump();
2344 current->flags |= PF_MEMALLOC;
2345 lockdep_set_current_reclaim_state(gfp_mask);
2346 reclaim_state.reclaimed_slab = 0;
2347 current->reclaim_state = &reclaim_state;
2349 progress = try_to_free_pages(zonelist, order, gfp_mask, nodemask);
2351 current->reclaim_state = NULL;
2352 lockdep_clear_current_reclaim_state();
2353 current->flags &= ~PF_MEMALLOC;
2360 /* The really slow allocator path where we enter direct reclaim */
2361 static inline struct page *
2362 __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
2363 struct zonelist *zonelist, enum zone_type high_zoneidx,
2364 nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
2365 int migratetype, unsigned long *did_some_progress)
2367 struct page *page = NULL;
2368 bool drained = false;
2370 *did_some_progress = __perform_reclaim(gfp_mask, order, zonelist,
2372 if (unlikely(!(*did_some_progress)))
2375 /* After successful reclaim, reconsider all zones for allocation */
2376 if (IS_ENABLED(CONFIG_NUMA))
2377 zlc_clear_zones_full(zonelist);
2380 page = get_page_from_freelist(gfp_mask, nodemask, order,
2381 zonelist, high_zoneidx,
2382 alloc_flags & ~ALLOC_NO_WATERMARKS,
2383 preferred_zone, migratetype);
2386 * If an allocation failed after direct reclaim, it could be because
2387 * pages are pinned on the per-cpu lists. Drain them and try again
2389 if (!page && !drained) {
2399 * This is called in the allocator slow-path if the allocation request is of
2400 * sufficient urgency to ignore watermarks and take other desperate measures
2402 static inline struct page *
2403 __alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order,
2404 struct zonelist *zonelist, enum zone_type high_zoneidx,
2405 nodemask_t *nodemask, struct zone *preferred_zone,
2411 page = get_page_from_freelist(gfp_mask, nodemask, order,
2412 zonelist, high_zoneidx, ALLOC_NO_WATERMARKS,
2413 preferred_zone, migratetype);
2415 if (!page && gfp_mask & __GFP_NOFAIL)
2416 wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50);
2417 } while (!page && (gfp_mask & __GFP_NOFAIL));
2423 void wake_all_kswapd(unsigned int order, struct zonelist *zonelist,
2424 enum zone_type high_zoneidx,
2425 enum zone_type classzone_idx)
2430 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx)
2431 wakeup_kswapd(zone, order, classzone_idx);
2435 gfp_to_alloc_flags(gfp_t gfp_mask)
2437 int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
2438 const bool atomic = !(gfp_mask & (__GFP_WAIT | __GFP_NO_KSWAPD));
2440 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
2441 BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
2444 * The caller may dip into page reserves a bit more if the caller
2445 * cannot run direct reclaim, or if the caller has realtime scheduling
2446 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
2447 * set both ALLOC_HARDER (atomic == true) and ALLOC_HIGH (__GFP_HIGH).
2449 alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH);
2453 * Not worth trying to allocate harder for __GFP_NOMEMALLOC even
2454 * if it can't schedule.
2456 if (!(gfp_mask & __GFP_NOMEMALLOC))
2457 alloc_flags |= ALLOC_HARDER;
2459 * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the
2460 * comment for __cpuset_node_allowed_softwall().
2462 alloc_flags &= ~ALLOC_CPUSET;
2463 } else if (unlikely(rt_task(current)) && !in_interrupt())
2464 alloc_flags |= ALLOC_HARDER;
2466 if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) {
2467 if (gfp_mask & __GFP_MEMALLOC)
2468 alloc_flags |= ALLOC_NO_WATERMARKS;
2469 else if (in_serving_softirq() && (current->flags & PF_MEMALLOC))
2470 alloc_flags |= ALLOC_NO_WATERMARKS;
2471 else if (!in_interrupt() &&
2472 ((current->flags & PF_MEMALLOC) ||
2473 unlikely(test_thread_flag(TIF_MEMDIE))))
2474 alloc_flags |= ALLOC_NO_WATERMARKS;
2477 if (allocflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE)
2478 alloc_flags |= ALLOC_CMA;
2483 bool gfp_pfmemalloc_allowed(gfp_t gfp_mask)
2485 return !!(gfp_to_alloc_flags(gfp_mask) & ALLOC_NO_WATERMARKS);
2488 static uint debug_high_order_alloc = 0;
2490 module_param_named(debug_high_order_alloc, debug_high_order_alloc, uint, S_IRUGO | S_IWUSR);
2493 static inline struct page *
2494 __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
2495 struct zonelist *zonelist, enum zone_type high_zoneidx,
2496 nodemask_t *nodemask, struct zone *preferred_zone,
2499 const gfp_t wait = gfp_mask & __GFP_WAIT;
2500 struct page *page = NULL;
2502 unsigned long pages_reclaimed = 0;
2503 unsigned long did_some_progress;
2504 bool sync_migration = false;
2505 bool deferred_compaction = false;
2506 bool contended_compaction = false;
2508 #ifdef CONFIG_SPRD_MEM_POOL
2510 if(-1 == sprd_page_mask_check(current->pid))
2514 * In the slowpath, we sanity check order to avoid ever trying to
2515 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
2516 * be using allocators in order of preference for an area that is
2519 if (order >= MAX_ORDER) {
2520 WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
2525 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
2526 * __GFP_NOWARN set) should not cause reclaim since the subsystem
2527 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
2528 * using a larger set of nodes after it has established that the
2529 * allowed per node queues are empty and that nodes are
2532 if (IS_ENABLED(CONFIG_NUMA) &&
2533 (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
2536 count_vm_event(SLOWPATH_ENTERED);
2539 if (!(gfp_mask & __GFP_NO_KSWAPD))
2540 wake_all_kswapd(order, zonelist, high_zoneidx,
2541 zone_idx(preferred_zone));
2544 * OK, we're below the kswapd watermark and have kicked background
2545 * reclaim. Now things get more complex, so set up alloc_flags according
2546 * to how we want to proceed.
2548 alloc_flags = gfp_to_alloc_flags(gfp_mask);
2551 * Find the true preferred zone if the allocation is unconstrained by
2554 if (!(alloc_flags & ALLOC_CPUSET) && !nodemask)
2555 first_zones_zonelist(zonelist, high_zoneidx, NULL,
2559 /* This is the last chance, in general, before the goto nopage. */
2560 page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist,
2561 high_zoneidx, alloc_flags & ~ALLOC_NO_WATERMARKS,
2562 preferred_zone, migratetype);
2566 /* Allocate without watermarks if the context allows */
2567 if (alloc_flags & ALLOC_NO_WATERMARKS) {
2569 * Ignore mempolicies if ALLOC_NO_WATERMARKS on the grounds
2570 * the allocation is high priority and these type of
2571 * allocations are system rather than user orientated
2573 zonelist = node_zonelist(numa_node_id(), gfp_mask);
2575 page = __alloc_pages_high_priority(gfp_mask, order,
2576 zonelist, high_zoneidx, nodemask,
2577 preferred_zone, migratetype);
2583 /* Atomic allocations - we can't balance anything */
2587 /* Avoid recursion of direct reclaim */
2588 if (current->flags & PF_MEMALLOC)
2591 /* Avoid allocations with no watermarks from looping endlessly */
2592 if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL))
2596 * Try direct compaction. The first pass is asynchronous. Subsequent
2597 * attempts after direct reclaim are synchronous
2599 page = __alloc_pages_direct_compact(gfp_mask, order,
2600 zonelist, high_zoneidx,
2602 alloc_flags, preferred_zone,
2603 migratetype, sync_migration,
2604 &contended_compaction,
2605 &deferred_compaction,
2606 &did_some_progress);
2609 sync_migration = true;
2612 * If compaction is deferred for high-order allocations, it is because
2613 * sync compaction recently failed. In this is the case and the caller
2614 * requested a movable allocation that does not heavily disrupt the
2615 * system then fail the allocation instead of entering direct reclaim.
2617 if ((deferred_compaction || contended_compaction) &&
2618 (gfp_mask & __GFP_NO_KSWAPD))
2622 if(debug_high_order_alloc && (order > 1))
2624 printk("%s: pid:%d, name:%s, mask:0x%X, order:%d \r\n", __func__, current->pid, current->comm, gfp_mask, order);
2628 /* Try direct reclaim and then allocating */
2629 page = __alloc_pages_direct_reclaim(gfp_mask, order,
2630 zonelist, high_zoneidx,
2632 alloc_flags, preferred_zone,
2633 migratetype, &did_some_progress);
2637 #ifdef CONFIG_SPRD_MEM_POOL
2638 /*for sprd page alloc*/
2639 page = sprd_page_alloc(gfp_mask, order, high_zoneidx);
2645 * If we failed to make any progress reclaiming, then we are
2646 * running out of options and have to consider going OOM
2648 if (!did_some_progress) {
2649 if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
2650 if (oom_killer_disabled)
2652 /* Coredumps can quickly deplete all memory reserves */
2653 if ((current->flags & PF_DUMPCORE) &&
2654 !(gfp_mask & __GFP_NOFAIL))
2656 page = __alloc_pages_may_oom(gfp_mask, order,
2657 zonelist, high_zoneidx,
2658 nodemask, preferred_zone,
2663 if (!(gfp_mask & __GFP_NOFAIL)) {
2665 * The oom killer is not called for high-order
2666 * allocations that may fail, so if no progress
2667 * is being made, there are no other options and
2668 * retrying is unlikely to help.
2670 if (order > PAGE_ALLOC_COSTLY_ORDER)
2673 * The oom killer is not called for lowmem
2674 * allocations to prevent needlessly killing
2677 if (high_zoneidx < ZONE_NORMAL)
2685 /* Check if we should retry the allocation */
2686 pages_reclaimed += did_some_progress;
2687 if (should_alloc_retry(gfp_mask, order, did_some_progress,
2689 /* Wait for some write requests to complete then retry */
2690 wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50);
2694 * High-order allocations do not necessarily loop after
2695 * direct reclaim and reclaim/compaction depends on compaction
2696 * being called after reclaim so call directly if necessary
2698 page = __alloc_pages_direct_compact(gfp_mask, order,
2699 zonelist, high_zoneidx,
2701 alloc_flags, preferred_zone,
2702 migratetype, sync_migration,
2703 &contended_compaction,
2704 &deferred_compaction,
2705 &did_some_progress);
2711 warn_alloc_failed(gfp_mask, order, NULL);
2714 if (kmemcheck_enabled)
2715 kmemcheck_pagealloc_alloc(page, order, gfp_mask);
2721 * This is the 'heart' of the zoned buddy allocator.
2724 __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
2725 struct zonelist *zonelist, nodemask_t *nodemask)
2727 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
2728 struct zone *preferred_zone;
2729 struct page *page = NULL;
2730 int migratetype = allocflags_to_migratetype(gfp_mask);
2731 unsigned int cpuset_mems_cookie;
2732 int alloc_flags = ALLOC_WMARK_LOW|ALLOC_CPUSET;
2733 struct mem_cgroup *memcg = NULL;
2735 gfp_mask &= gfp_allowed_mask;
2737 lockdep_trace_alloc(gfp_mask);
2739 might_sleep_if(gfp_mask & __GFP_WAIT);
2741 if (should_fail_alloc_page(gfp_mask, order))
2745 * Check the zones suitable for the gfp_mask contain at least one
2746 * valid zone. It's possible to have an empty zonelist as a result
2747 * of GFP_THISNODE and a memoryless node
2749 if (unlikely(!zonelist->_zonerefs->zone))
2753 * Will only have any effect when __GFP_KMEMCG is set. This is
2754 * verified in the (always inline) callee
2756 if (!memcg_kmem_newpage_charge(gfp_mask, &memcg, order))
2760 cpuset_mems_cookie = get_mems_allowed();
2762 /* The preferred zone is used for statistics later */
2763 first_zones_zonelist(zonelist, high_zoneidx,
2764 nodemask ? : &cpuset_current_mems_allowed,
2766 if (!preferred_zone)
2770 if (allocflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE)
2771 alloc_flags |= ALLOC_CMA;
2773 /* First allocation attempt */
2774 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order,
2775 zonelist, high_zoneidx, alloc_flags,
2776 preferred_zone, migratetype);
2777 if (unlikely(!page)) {
2779 * Runtime PM, block IO and its error handling path
2780 * can deadlock because I/O on the device might not
2783 gfp_mask = memalloc_noio_flags(gfp_mask);
2784 page = __alloc_pages_slowpath(gfp_mask, order,
2785 zonelist, high_zoneidx, nodemask,
2786 preferred_zone, migratetype);
2789 trace_mm_page_alloc(page, order, gfp_mask, migratetype);
2793 * When updating a task's mems_allowed, it is possible to race with
2794 * parallel threads in such a way that an allocation can fail while
2795 * the mask is being updated. If a page allocation is about to fail,
2796 * check if the cpuset changed during allocation and if so, retry.
2798 if (unlikely(!put_mems_allowed(cpuset_mems_cookie) && !page))
2801 memcg_kmem_commit_charge(page, memcg, order);
2805 EXPORT_SYMBOL(__alloc_pages_nodemask);
2808 * Common helper functions.
2810 unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
2815 * __get_free_pages() returns a 32-bit address, which cannot represent
2818 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
2820 page = alloc_pages(gfp_mask, order);
2823 return (unsigned long) page_address(page);
2825 EXPORT_SYMBOL(__get_free_pages);
2827 #ifdef CONFIG_SPRD_PAGERECORDER
2829 * Common helper functions.
2831 unsigned long __get_free_pages_nopagedebug(gfp_t gfp_mask, unsigned int order)
2836 * __get_free_pages() returns a 32-bit address, which cannot represent
2839 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
2841 page = alloc_pages_nopagedebug(gfp_mask, order);
2844 return (unsigned long) page_address(page);
2846 EXPORT_SYMBOL(__get_free_pages_nopagedebug);
2848 unsigned long get_zeroed_page_nopagedebug(gfp_t gfp_mask)
2850 return __get_free_pages_nopagedebug(gfp_mask | __GFP_ZERO, 0);
2852 EXPORT_SYMBOL(get_zeroed_page_nopagedebug);
2857 unsigned long get_zeroed_page(gfp_t gfp_mask)
2859 return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
2861 EXPORT_SYMBOL(get_zeroed_page);
2863 void __free_pages(struct page *page, unsigned int order)
2865 #ifdef CONFIG_SPRD_PAGERECORDER
2868 remove_page_record((void *)page,order);
2871 if (put_page_testzero(page)) {
2873 free_hot_cold_page(page, 0);
2875 __free_pages_ok(page, order);
2879 EXPORT_SYMBOL(__free_pages);
2881 void free_pages(unsigned long addr, unsigned int order)
2884 VM_BUG_ON(!virt_addr_valid((void *)addr));
2885 __free_pages(virt_to_page((void *)addr), order);
2889 EXPORT_SYMBOL(free_pages);
2891 #ifdef CONFIG_SPRD_PAGERECORDER
2892 void __free_pages_nopagedebug(struct page *page, unsigned int order)
2894 if (put_page_testzero(page)) {
2896 free_hot_cold_page(page, 0);
2898 __free_pages_ok(page, order);
2902 EXPORT_SYMBOL(__free_pages_nopagedebug);
2904 void free_pages_nopagedebug(unsigned long addr, unsigned int order)
2907 VM_BUG_ON(!virt_addr_valid((void *)addr));
2908 __free_pages_nopagedebug(virt_to_page((void *)addr), order);
2912 EXPORT_SYMBOL(free_pages_nopagedebug);
2916 * __free_memcg_kmem_pages and free_memcg_kmem_pages will free
2917 * pages allocated with __GFP_KMEMCG.
2919 * Those pages are accounted to a particular memcg, embedded in the
2920 * corresponding page_cgroup. To avoid adding a hit in the allocator to search
2921 * for that information only to find out that it is NULL for users who have no
2922 * interest in that whatsoever, we provide these functions.
2924 * The caller knows better which flags it relies on.
2926 void __free_memcg_kmem_pages(struct page *page, unsigned int order)
2928 memcg_kmem_uncharge_pages(page, order);
2929 __free_pages(page, order);
2932 #ifdef CONFIG_SPRD_PAGERECORDER
2933 void __free_memcg_kmem_pages_nopagedebug(struct page *page, unsigned int order)
2935 memcg_kmem_uncharge_pages(page, order);
2936 __free_pages_nopagedebug(page, order);
2940 void free_memcg_kmem_pages(unsigned long addr, unsigned int order)
2943 VM_BUG_ON(!virt_addr_valid((void *)addr));
2944 __free_memcg_kmem_pages(virt_to_page((void *)addr), order);
2948 static void *make_alloc_exact(unsigned long addr, unsigned order, size_t size)
2951 unsigned long alloc_end = addr + (PAGE_SIZE << order);
2952 unsigned long used = addr + PAGE_ALIGN(size);
2954 split_page(virt_to_page((void *)addr), order);
2955 while (used < alloc_end) {
2960 return (void *)addr;
2964 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
2965 * @size: the number of bytes to allocate
2966 * @gfp_mask: GFP flags for the allocation
2968 * This function is similar to alloc_pages(), except that it allocates the
2969 * minimum number of pages to satisfy the request. alloc_pages() can only
2970 * allocate memory in power-of-two pages.
2972 * This function is also limited by MAX_ORDER.
2974 * Memory allocated by this function must be released by free_pages_exact().
2976 void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
2978 unsigned int order = get_order(size);
2981 addr = __get_free_pages(gfp_mask, order);
2982 return make_alloc_exact(addr, order, size);
2984 EXPORT_SYMBOL(alloc_pages_exact);
2987 * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
2989 * @nid: the preferred node ID where memory should be allocated
2990 * @size: the number of bytes to allocate
2991 * @gfp_mask: GFP flags for the allocation
2993 * Like alloc_pages_exact(), but try to allocate on node nid first before falling
2995 * Note this is not alloc_pages_exact_node() which allocates on a specific node,
2998 void *alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask)
3000 unsigned order = get_order(size);
3001 struct page *p = alloc_pages_node(nid, gfp_mask, order);
3004 return make_alloc_exact((unsigned long)page_address(p), order, size);
3006 EXPORT_SYMBOL(alloc_pages_exact_nid);
3009 * free_pages_exact - release memory allocated via alloc_pages_exact()
3010 * @virt: the value returned by alloc_pages_exact.
3011 * @size: size of allocation, same value as passed to alloc_pages_exact().
3013 * Release the memory allocated by a previous call to alloc_pages_exact.
3015 void free_pages_exact(void *virt, size_t size)
3017 unsigned long addr = (unsigned long)virt;
3018 unsigned long end = addr + PAGE_ALIGN(size);
3020 while (addr < end) {
3025 EXPORT_SYMBOL(free_pages_exact);
3028 * nr_free_zone_pages - count number of pages beyond high watermark
3029 * @offset: The zone index of the highest zone
3031 * nr_free_zone_pages() counts the number of counts pages which are beyond the
3032 * high watermark within all zones at or below a given zone index. For each
3033 * zone, the number of pages is calculated as:
3034 * present_pages - high_pages
3036 static unsigned long nr_free_zone_pages(int offset)
3041 /* Just pick one node, since fallback list is circular */
3042 unsigned long sum = 0;
3044 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
3046 for_each_zone_zonelist(zone, z, zonelist, offset) {
3047 unsigned long size = zone->managed_pages;
3048 unsigned long high = high_wmark_pages(zone);
3057 * nr_free_buffer_pages - count number of pages beyond high watermark
3059 * nr_free_buffer_pages() counts the number of pages which are beyond the high
3060 * watermark within ZONE_DMA and ZONE_NORMAL.
3062 unsigned long nr_free_buffer_pages(void)
3064 return nr_free_zone_pages(gfp_zone(GFP_USER));
3066 EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
3069 * nr_free_pagecache_pages - count number of pages beyond high watermark
3071 * nr_free_pagecache_pages() counts the number of pages which are beyond the
3072 * high watermark within all zones.
3074 unsigned long nr_free_pagecache_pages(void)
3076 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
3079 static inline void show_node(struct zone *zone)
3081 if (IS_ENABLED(CONFIG_NUMA))
3082 printk("Node %d ", zone_to_nid(zone));
3085 void si_meminfo(struct sysinfo *val)
3087 val->totalram = totalram_pages;
3089 val->freeram = global_page_state(NR_FREE_PAGES);
3090 val->bufferram = nr_blockdev_pages();
3091 val->totalhigh = totalhigh_pages;
3092 val->freehigh = nr_free_highpages();
3093 val->mem_unit = PAGE_SIZE;
3096 EXPORT_SYMBOL(si_meminfo);
3099 void si_meminfo_node(struct sysinfo *val, int nid)
3101 pg_data_t *pgdat = NODE_DATA(nid);
3103 val->totalram = pgdat->node_present_pages;
3104 val->freeram = node_page_state(nid, NR_FREE_PAGES);
3105 #ifdef CONFIG_HIGHMEM
3106 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].managed_pages;
3107 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
3113 val->mem_unit = PAGE_SIZE;
3118 * Determine whether the node should be displayed or not, depending on whether
3119 * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
3121 bool skip_free_areas_node(unsigned int flags, int nid)
3124 unsigned int cpuset_mems_cookie;
3126 if (!(flags & SHOW_MEM_FILTER_NODES))
3130 cpuset_mems_cookie = get_mems_allowed();
3131 ret = !node_isset(nid, cpuset_current_mems_allowed);
3132 } while (!put_mems_allowed(cpuset_mems_cookie));
3137 #define K(x) ((x) << (PAGE_SHIFT-10))
3139 static void show_migration_types(unsigned char type, unsigned long *nr_migrate)
3141 static const char types[MIGRATE_TYPES] = {
3142 [MIGRATE_UNMOVABLE] = 'U',
3143 [MIGRATE_RECLAIMABLE] = 'E',
3144 [MIGRATE_MOVABLE] = 'M',
3145 [MIGRATE_RESERVE] = 'R',
3147 [MIGRATE_CMA] = 'C',
3149 #ifdef CONFIG_MEMORY_ISOLATION
3150 [MIGRATE_ISOLATE] = 'I',
3157 for (i = 0; i < MIGRATE_TYPES; i++) {
3158 if (type & (1 << i))
3159 p += sprintf(p, "%c%d", types[i], nr_migrate[i]);
3163 printk("(%s) ", tmp);
3167 * Show free area list (used inside shift_scroll-lock stuff)
3168 * We also calculate the percentage fragmentation. We do this by counting the
3169 * memory on each free list with the exception of the first item on the list.
3170 * Suppresses nodes that are not allowed by current's cpuset if
3171 * SHOW_MEM_FILTER_NODES is passed.
3173 void show_free_areas(unsigned int filter)
3178 for_each_populated_zone(zone) {
3179 if (skip_free_areas_node(filter, zone_to_nid(zone)))
3182 printk("%s per-cpu:\n", zone->name);
3184 for_each_online_cpu(cpu) {
3185 struct per_cpu_pageset *pageset;
3187 pageset = per_cpu_ptr(zone->pageset, cpu);
3189 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
3190 cpu, pageset->pcp.high,
3191 pageset->pcp.batch, pageset->pcp.count);
3195 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
3196 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
3198 " dirty:%lu writeback:%lu unstable:%lu\n"
3199 " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n"
3200 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
3202 global_page_state(NR_ACTIVE_ANON),
3203 global_page_state(NR_INACTIVE_ANON),
3204 global_page_state(NR_ISOLATED_ANON),
3205 global_page_state(NR_ACTIVE_FILE),
3206 global_page_state(NR_INACTIVE_FILE),
3207 global_page_state(NR_ISOLATED_FILE),
3208 global_page_state(NR_UNEVICTABLE),
3209 global_page_state(NR_FILE_DIRTY),
3210 global_page_state(NR_WRITEBACK),
3211 global_page_state(NR_UNSTABLE_NFS),
3212 global_page_state(NR_FREE_PAGES),
3213 global_page_state(NR_SLAB_RECLAIMABLE),
3214 global_page_state(NR_SLAB_UNRECLAIMABLE),
3215 global_page_state(NR_FILE_MAPPED),
3216 global_page_state(NR_SHMEM),
3217 global_page_state(NR_PAGETABLE),
3218 global_page_state(NR_BOUNCE),
3219 global_page_state(NR_FREE_CMA_PAGES));
3221 for_each_populated_zone(zone) {
3224 if (skip_free_areas_node(filter, zone_to_nid(zone)))
3232 " active_anon:%lukB"
3233 " inactive_anon:%lukB"
3234 " active_file:%lukB"
3235 " inactive_file:%lukB"
3236 " unevictable:%lukB"
3237 " isolated(anon):%lukB"
3238 " isolated(file):%lukB"
3246 " slab_reclaimable:%lukB"
3247 " slab_unreclaimable:%lukB"
3248 " kernel_stack:%lukB"
3253 " writeback_tmp:%lukB"
3254 " pages_scanned:%lu"
3255 " all_unreclaimable? %s"
3258 K(zone_page_state(zone, NR_FREE_PAGES)),
3259 K(min_wmark_pages(zone)),
3260 K(low_wmark_pages(zone)),
3261 K(high_wmark_pages(zone)),
3262 K(zone_page_state(zone, NR_ACTIVE_ANON)),
3263 K(zone_page_state(zone, NR_INACTIVE_ANON)),
3264 K(zone_page_state(zone, NR_ACTIVE_FILE)),
3265 K(zone_page_state(zone, NR_INACTIVE_FILE)),
3266 K(zone_page_state(zone, NR_UNEVICTABLE)),
3267 K(zone_page_state(zone, NR_ISOLATED_ANON)),
3268 K(zone_page_state(zone, NR_ISOLATED_FILE)),
3269 K(zone->present_pages),
3270 K(zone->managed_pages),
3271 K(zone_page_state(zone, NR_MLOCK)),
3272 K(zone_page_state(zone, NR_FILE_DIRTY)),
3273 K(zone_page_state(zone, NR_WRITEBACK)),
3274 K(zone_page_state(zone, NR_FILE_MAPPED)),
3275 K(zone_page_state(zone, NR_SHMEM)),
3276 K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)),
3277 K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)),
3278 zone_page_state(zone, NR_KERNEL_STACK) *
3280 K(zone_page_state(zone, NR_PAGETABLE)),
3281 K(zone_page_state(zone, NR_UNSTABLE_NFS)),
3282 K(zone_page_state(zone, NR_BOUNCE)),
3283 K(zone_page_state(zone, NR_FREE_CMA_PAGES)),
3284 K(zone_page_state(zone, NR_WRITEBACK_TEMP)),
3285 zone->pages_scanned,
3286 (!zone_reclaimable(zone) ? "yes" : "no")
3288 printk("lowmem_reserve[]:");
3289 for (i = 0; i < MAX_NR_ZONES; i++)
3290 printk(" %lu", zone->lowmem_reserve[i]);
3294 for_each_populated_zone(zone) {
3295 unsigned long nr[MAX_ORDER], flags, order, total = 0;
3296 unsigned char types[MAX_ORDER];
3297 unsigned long nr_migrate[MAX_ORDER][MIGRATE_TYPES] = {0};
3299 if (skip_free_areas_node(filter, zone_to_nid(zone)))
3302 printk("%s: ", zone->name);
3304 spin_lock_irqsave(&zone->lock, flags);
3305 for (order = 0; order < MAX_ORDER; order++) {
3306 struct free_area *area = &zone->free_area[order];
3308 struct list_head *temp;
3310 nr[order] = area->nr_free;
3311 total += nr[order] << order;
3314 for (type = 0; type < MIGRATE_TYPES; type++) {
3315 if (!list_empty(&area->free_list[type]))
3316 types[order] |= 1 << type;
3317 list_for_each(temp, &area->free_list[type]) {
3318 nr_migrate[order][type]++;
3322 spin_unlock_irqrestore(&zone->lock, flags);
3323 for (order = 0; order < MAX_ORDER; order++) {
3324 printk("%lu*%lukB ", nr[order], K(1UL) << order);
3326 show_migration_types(types[order], nr_migrate[order]);
3328 printk("= %lukB\n", K(total));
3331 hugetlb_show_meminfo();
3333 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
3335 show_swap_cache_info();
3338 static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
3340 zoneref->zone = zone;
3341 zoneref->zone_idx = zone_idx(zone);
3345 * Builds allocation fallback zone lists.
3347 * Add all populated zones of a node to the zonelist.
3349 static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
3350 int nr_zones, enum zone_type zone_type)
3354 BUG_ON(zone_type >= MAX_NR_ZONES);
3359 zone = pgdat->node_zones + zone_type;
3360 if (populated_zone(zone)) {
3361 zoneref_set_zone(zone,
3362 &zonelist->_zonerefs[nr_zones++]);
3363 check_highest_zone(zone_type);
3366 } while (zone_type);
3373 * 0 = automatic detection of better ordering.
3374 * 1 = order by ([node] distance, -zonetype)
3375 * 2 = order by (-zonetype, [node] distance)
3377 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
3378 * the same zonelist. So only NUMA can configure this param.
3380 #define ZONELIST_ORDER_DEFAULT 0
3381 #define ZONELIST_ORDER_NODE 1
3382 #define ZONELIST_ORDER_ZONE 2
3384 /* zonelist order in the kernel.
3385 * set_zonelist_order() will set this to NODE or ZONE.
3387 static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
3388 static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
3392 /* The value user specified ....changed by config */
3393 static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
3394 /* string for sysctl */
3395 #define NUMA_ZONELIST_ORDER_LEN 16
3396 char numa_zonelist_order[16] = "default";
3399 * interface for configure zonelist ordering.
3400 * command line option "numa_zonelist_order"
3401 * = "[dD]efault - default, automatic configuration.
3402 * = "[nN]ode - order by node locality, then by zone within node
3403 * = "[zZ]one - order by zone, then by locality within zone
3406 static int __parse_numa_zonelist_order(char *s)
3408 if (*s == 'd' || *s == 'D') {
3409 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
3410 } else if (*s == 'n' || *s == 'N') {
3411 user_zonelist_order = ZONELIST_ORDER_NODE;
3412 } else if (*s == 'z' || *s == 'Z') {
3413 user_zonelist_order = ZONELIST_ORDER_ZONE;
3416 "Ignoring invalid numa_zonelist_order value: "
3423 static __init int setup_numa_zonelist_order(char *s)
3430 ret = __parse_numa_zonelist_order(s);
3432 strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN);
3436 early_param("numa_zonelist_order", setup_numa_zonelist_order);
3439 * sysctl handler for numa_zonelist_order
3441 int numa_zonelist_order_handler(ctl_table *table, int write,
3442 void __user *buffer, size_t *length,
3445 char saved_string[NUMA_ZONELIST_ORDER_LEN];
3447 static DEFINE_MUTEX(zl_order_mutex);
3449 mutex_lock(&zl_order_mutex);
3451 strcpy(saved_string, (char*)table->data);
3452 ret = proc_dostring(table, write, buffer, length, ppos);
3456 int oldval = user_zonelist_order;
3457 if (__parse_numa_zonelist_order((char*)table->data)) {
3459 * bogus value. restore saved string
3461 strncpy((char*)table->data, saved_string,
3462 NUMA_ZONELIST_ORDER_LEN);
3463 user_zonelist_order = oldval;
3464 } else if (oldval != user_zonelist_order) {
3465 mutex_lock(&zonelists_mutex);
3466 build_all_zonelists(NULL, NULL);
3467 mutex_unlock(&zonelists_mutex);
3471 mutex_unlock(&zl_order_mutex);
3476 #define MAX_NODE_LOAD (nr_online_nodes)
3477 static int node_load[MAX_NUMNODES];
3480 * find_next_best_node - find the next node that should appear in a given node's fallback list
3481 * @node: node whose fallback list we're appending
3482 * @used_node_mask: nodemask_t of already used nodes
3484 * We use a number of factors to determine which is the next node that should
3485 * appear on a given node's fallback list. The node should not have appeared
3486 * already in @node's fallback list, and it should be the next closest node
3487 * according to the distance array (which contains arbitrary distance values
3488 * from each node to each node in the system), and should also prefer nodes
3489 * with no CPUs, since presumably they'll have very little allocation pressure
3490 * on them otherwise.
3491 * It returns -1 if no node is found.
3493 static int find_next_best_node(int node, nodemask_t *used_node_mask)
3496 int min_val = INT_MAX;
3497 int best_node = NUMA_NO_NODE;
3498 const struct cpumask *tmp = cpumask_of_node(0);
3500 /* Use the local node if we haven't already */
3501 if (!node_isset(node, *used_node_mask)) {
3502 node_set(node, *used_node_mask);
3506 for_each_node_state(n, N_MEMORY) {
3508 /* Don't want a node to appear more than once */
3509 if (node_isset(n, *used_node_mask))
3512 /* Use the distance array to find the distance */
3513 val = node_distance(node, n);
3515 /* Penalize nodes under us ("prefer the next node") */
3518 /* Give preference to headless and unused nodes */
3519 tmp = cpumask_of_node(n);
3520 if (!cpumask_empty(tmp))
3521 val += PENALTY_FOR_NODE_WITH_CPUS;
3523 /* Slight preference for less loaded node */
3524 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
3525 val += node_load[n];
3527 if (val < min_val) {
3534 node_set(best_node, *used_node_mask);
3541 * Build zonelists ordered by node and zones within node.
3542 * This results in maximum locality--normal zone overflows into local
3543 * DMA zone, if any--but risks exhausting DMA zone.
3545 static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
3548 struct zonelist *zonelist;
3550 zonelist = &pgdat->node_zonelists[0];
3551 for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
3553 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
3555 zonelist->_zonerefs[j].zone = NULL;
3556 zonelist->_zonerefs[j].zone_idx = 0;
3560 * Build gfp_thisnode zonelists
3562 static void build_thisnode_zonelists(pg_data_t *pgdat)
3565 struct zonelist *zonelist;
3567 zonelist = &pgdat->node_zonelists[1];
3568 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
3569 zonelist->_zonerefs[j].zone = NULL;
3570 zonelist->_zonerefs[j].zone_idx = 0;
3574 * Build zonelists ordered by zone and nodes within zones.
3575 * This results in conserving DMA zone[s] until all Normal memory is
3576 * exhausted, but results in overflowing to remote node while memory
3577 * may still exist in local DMA zone.
3579 static int node_order[MAX_NUMNODES];
3581 static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
3584 int zone_type; /* needs to be signed */
3586 struct zonelist *zonelist;
3588 zonelist = &pgdat->node_zonelists[0];
3590 for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
3591 for (j = 0; j < nr_nodes; j++) {
3592 node = node_order[j];
3593 z = &NODE_DATA(node)->node_zones[zone_type];
3594 if (populated_zone(z)) {
3596 &zonelist->_zonerefs[pos++]);
3597 check_highest_zone(zone_type);
3601 zonelist->_zonerefs[pos].zone = NULL;
3602 zonelist->_zonerefs[pos].zone_idx = 0;
3605 static int default_zonelist_order(void)
3608 unsigned long low_kmem_size,total_size;
3612 * ZONE_DMA and ZONE_DMA32 can be very small area in the system.
3613 * If they are really small and used heavily, the system can fall
3614 * into OOM very easily.
3615 * This function detect ZONE_DMA/DMA32 size and configures zone order.
3617 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
3620 for_each_online_node(nid) {
3621 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
3622 z = &NODE_DATA(nid)->node_zones[zone_type];
3623 if (populated_zone(z)) {
3624 if (zone_type < ZONE_NORMAL)
3625 low_kmem_size += z->present_pages;
3626 total_size += z->present_pages;
3627 } else if (zone_type == ZONE_NORMAL) {
3629 * If any node has only lowmem, then node order
3630 * is preferred to allow kernel allocations
3631 * locally; otherwise, they can easily infringe
3632 * on other nodes when there is an abundance of
3633 * lowmem available to allocate from.
3635 return ZONELIST_ORDER_NODE;
3639 if (!low_kmem_size || /* there are no DMA area. */
3640 low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
3641 return ZONELIST_ORDER_NODE;
3643 * look into each node's config.
3644 * If there is a node whose DMA/DMA32 memory is very big area on
3645 * local memory, NODE_ORDER may be suitable.
3647 average_size = total_size /
3648 (nodes_weight(node_states[N_MEMORY]) + 1);
3649 for_each_online_node(nid) {
3652 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
3653 z = &NODE_DATA(nid)->node_zones[zone_type];
3654 if (populated_zone(z)) {
3655 if (zone_type < ZONE_NORMAL)
3656 low_kmem_size += z->present_pages;
3657 total_size += z->present_pages;
3660 if (low_kmem_size &&
3661 total_size > average_size && /* ignore small node */
3662 low_kmem_size > total_size * 70/100)
3663 return ZONELIST_ORDER_NODE;
3665 return ZONELIST_ORDER_ZONE;
3668 static void set_zonelist_order(void)
3670 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
3671 current_zonelist_order = default_zonelist_order();
3673 current_zonelist_order = user_zonelist_order;
3676 static void build_zonelists(pg_data_t *pgdat)
3680 nodemask_t used_mask;
3681 int local_node, prev_node;
3682 struct zonelist *zonelist;
3683 int order = current_zonelist_order;
3685 /* initialize zonelists */
3686 for (i = 0; i < MAX_ZONELISTS; i++) {
3687 zonelist = pgdat->node_zonelists + i;
3688 zonelist->_zonerefs[0].zone = NULL;
3689 zonelist->_zonerefs[0].zone_idx = 0;
3692 /* NUMA-aware ordering of nodes */
3693 local_node = pgdat->node_id;
3694 load = nr_online_nodes;
3695 prev_node = local_node;
3696 nodes_clear(used_mask);
3698 memset(node_order, 0, sizeof(node_order));
3701 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
3703 * We don't want to pressure a particular node.
3704 * So adding penalty to the first node in same
3705 * distance group to make it round-robin.
3707 if (node_distance(local_node, node) !=
3708 node_distance(local_node, prev_node))
3709 node_load[node] = load;
3713 if (order == ZONELIST_ORDER_NODE)
3714 build_zonelists_in_node_order(pgdat, node);
3716 node_order[j++] = node; /* remember order */
3719 if (order == ZONELIST_ORDER_ZONE) {
3720 /* calculate node order -- i.e., DMA last! */
3721 build_zonelists_in_zone_order(pgdat, j);
3724 build_thisnode_zonelists(pgdat);
3727 /* Construct the zonelist performance cache - see further mmzone.h */
3728 static void build_zonelist_cache(pg_data_t *pgdat)
3730 struct zonelist *zonelist;
3731 struct zonelist_cache *zlc;
3734 zonelist = &pgdat->node_zonelists[0];
3735 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
3736 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
3737 for (z = zonelist->_zonerefs; z->zone; z++)
3738 zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z);
3741 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
3743 * Return node id of node used for "local" allocations.
3744 * I.e., first node id of first zone in arg node's generic zonelist.
3745 * Used for initializing percpu 'numa_mem', which is used primarily
3746 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
3748 int local_memory_node(int node)
3752 (void)first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
3753 gfp_zone(GFP_KERNEL),
3760 #else /* CONFIG_NUMA */
3762 static void set_zonelist_order(void)
3764 current_zonelist_order = ZONELIST_ORDER_ZONE;
3767 static void build_zonelists(pg_data_t *pgdat)
3769 int node, local_node;
3771 struct zonelist *zonelist;
3773 local_node = pgdat->node_id;
3775 zonelist = &pgdat->node_zonelists[0];
3776 j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
3779 * Now we build the zonelist so that it contains the zones
3780 * of all the other nodes.
3781 * We don't want to pressure a particular node, so when
3782 * building the zones for node N, we make sure that the
3783 * zones coming right after the local ones are those from
3784 * node N+1 (modulo N)
3786 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
3787 if (!node_online(node))
3789 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
3792 for (node = 0; node < local_node; node++) {
3793 if (!node_online(node))
3795 j = build_zonelists_node(NODE_DATA(node), zonelist, j,
3799 zonelist->_zonerefs[j].zone = NULL;
3800 zonelist->_zonerefs[j].zone_idx = 0;
3803 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
3804 static void build_zonelist_cache(pg_data_t *pgdat)
3806 pgdat->node_zonelists[0].zlcache_ptr = NULL;
3809 #endif /* CONFIG_NUMA */
3812 * Boot pageset table. One per cpu which is going to be used for all
3813 * zones and all nodes. The parameters will be set in such a way
3814 * that an item put on a list will immediately be handed over to
3815 * the buddy list. This is safe since pageset manipulation is done
3816 * with interrupts disabled.
3818 * The boot_pagesets must be kept even after bootup is complete for
3819 * unused processors and/or zones. They do play a role for bootstrapping
3820 * hotplugged processors.
3822 * zoneinfo_show() and maybe other functions do
3823 * not check if the processor is online before following the pageset pointer.
3824 * Other parts of the kernel may not check if the zone is available.
3826 static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch);
3827 static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset);
3828 static void setup_zone_pageset(struct zone *zone);
3831 * Global mutex to protect against size modification of zonelists
3832 * as well as to serialize pageset setup for the new populated zone.
3834 DEFINE_MUTEX(zonelists_mutex);
3836 /* return values int ....just for stop_machine() */
3837 static int __build_all_zonelists(void *data)
3841 pg_data_t *self = data;
3844 memset(node_load, 0, sizeof(node_load));
3847 if (self && !node_online(self->node_id)) {
3848 build_zonelists(self);
3849 build_zonelist_cache(self);
3852 for_each_online_node(nid) {
3853 pg_data_t *pgdat = NODE_DATA(nid);
3855 build_zonelists(pgdat);
3856 build_zonelist_cache(pgdat);
3860 * Initialize the boot_pagesets that are going to be used
3861 * for bootstrapping processors. The real pagesets for
3862 * each zone will be allocated later when the per cpu
3863 * allocator is available.
3865 * boot_pagesets are used also for bootstrapping offline
3866 * cpus if the system is already booted because the pagesets
3867 * are needed to initialize allocators on a specific cpu too.
3868 * F.e. the percpu allocator needs the page allocator which
3869 * needs the percpu allocator in order to allocate its pagesets
3870 * (a chicken-egg dilemma).
3872 for_each_possible_cpu(cpu) {
3873 setup_pageset(&per_cpu(boot_pageset, cpu), 0);
3875 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
3877 * We now know the "local memory node" for each node--
3878 * i.e., the node of the first zone in the generic zonelist.
3879 * Set up numa_mem percpu variable for on-line cpus. During
3880 * boot, only the boot cpu should be on-line; we'll init the
3881 * secondary cpus' numa_mem as they come on-line. During
3882 * node/memory hotplug, we'll fixup all on-line cpus.
3884 if (cpu_online(cpu))
3885 set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
3893 * Called with zonelists_mutex held always
3894 * unless system_state == SYSTEM_BOOTING.
3896 void __ref build_all_zonelists(pg_data_t *pgdat, struct zone *zone)
3898 set_zonelist_order();
3900 if (system_state == SYSTEM_BOOTING) {
3901 __build_all_zonelists(NULL);
3902 mminit_verify_zonelist();
3903 cpuset_init_current_mems_allowed();
3905 /* we have to stop all cpus to guarantee there is no user
3907 #ifdef CONFIG_MEMORY_HOTPLUG
3909 setup_zone_pageset(zone);
3911 stop_machine(__build_all_zonelists, pgdat, NULL);
3912 /* cpuset refresh routine should be here */
3914 vm_total_pages = nr_free_pagecache_pages();
3916 * Disable grouping by mobility if the number of pages in the
3917 * system is too low to allow the mechanism to work. It would be
3918 * more accurate, but expensive to check per-zone. This check is
3919 * made on memory-hotadd so a system can start with mobility
3920 * disabled and enable it later
3922 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
3923 page_group_by_mobility_disabled = 1;
3925 page_group_by_mobility_disabled = 0;
3927 printk("Built %i zonelists in %s order, mobility grouping %s. "
3928 "Total pages: %ld\n",
3930 zonelist_order_name[current_zonelist_order],
3931 page_group_by_mobility_disabled ? "off" : "on",
3934 printk("Policy zone: %s\n", zone_names[policy_zone]);
3939 * Helper functions to size the waitqueue hash table.
3940 * Essentially these want to choose hash table sizes sufficiently
3941 * large so that collisions trying to wait on pages are rare.
3942 * But in fact, the number of active page waitqueues on typical
3943 * systems is ridiculously low, less than 200. So this is even
3944 * conservative, even though it seems large.
3946 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
3947 * waitqueues, i.e. the size of the waitq table given the number of pages.
3949 #define PAGES_PER_WAITQUEUE 256
3951 #ifndef CONFIG_MEMORY_HOTPLUG
3952 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
3954 unsigned long size = 1;
3956 pages /= PAGES_PER_WAITQUEUE;
3958 while (size < pages)
3962 * Once we have dozens or even hundreds of threads sleeping
3963 * on IO we've got bigger problems than wait queue collision.
3964 * Limit the size of the wait table to a reasonable size.
3966 size = min(size, 4096UL);
3968 return max(size, 4UL);
3972 * A zone's size might be changed by hot-add, so it is not possible to determine
3973 * a suitable size for its wait_table. So we use the maximum size now.
3975 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
3977 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
3978 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
3979 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
3981 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
3982 * or more by the traditional way. (See above). It equals:
3984 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
3985 * ia64(16K page size) : = ( 8G + 4M)byte.
3986 * powerpc (64K page size) : = (32G +16M)byte.
3988 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
3995 * This is an integer logarithm so that shifts can be used later
3996 * to extract the more random high bits from the multiplicative
3997 * hash function before the remainder is taken.
3999 static inline unsigned long wait_table_bits(unsigned long size)
4004 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
4007 * Check if a pageblock contains reserved pages
4009 static int pageblock_is_reserved(unsigned long start_pfn, unsigned long end_pfn)
4013 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
4014 if (!pfn_valid_within(pfn) || PageReserved(pfn_to_page(pfn)))
4021 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
4022 * of blocks reserved is based on min_wmark_pages(zone). The memory within
4023 * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
4024 * higher will lead to a bigger reserve which will get freed as contiguous
4025 * blocks as reclaim kicks in
4027 static void setup_zone_migrate_reserve(struct zone *zone)
4029 unsigned long start_pfn, pfn, end_pfn, block_end_pfn;
4031 unsigned long block_migratetype;
4035 * Get the start pfn, end pfn and the number of blocks to reserve
4036 * We have to be careful to be aligned to pageblock_nr_pages to
4037 * make sure that we always check pfn_valid for the first page in
4040 start_pfn = zone->zone_start_pfn;
4041 end_pfn = zone_end_pfn(zone);
4042 start_pfn = roundup(start_pfn, pageblock_nr_pages);
4043 reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >>
4047 * Reserve blocks are generally in place to help high-order atomic
4048 * allocations that are short-lived. A min_free_kbytes value that
4049 * would result in more than 2 reserve blocks for atomic allocations
4050 * is assumed to be in place to help anti-fragmentation for the
4051 * future allocation of hugepages at runtime.
4053 reserve = min(2, reserve);
4055 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
4056 if (!pfn_valid(pfn))
4058 page = pfn_to_page(pfn);
4060 /* Watch out for overlapping nodes */
4061 if (page_to_nid(page) != zone_to_nid(zone))
4064 block_migratetype = get_pageblock_migratetype(page);
4066 /* Only test what is necessary when the reserves are not met */
4069 * Blocks with reserved pages will never free, skip
4072 block_end_pfn = min(pfn + pageblock_nr_pages, end_pfn);
4073 if (pageblock_is_reserved(pfn, block_end_pfn))
4076 /* If this block is reserved, account for it */
4077 if (block_migratetype == MIGRATE_RESERVE) {
4082 /* Suitable for reserving if this block is movable */
4083 if (block_migratetype == MIGRATE_MOVABLE) {
4084 set_pageblock_migratetype(page,
4086 move_freepages_block(zone, page,
4094 * If the reserve is met and this is a previous reserved block,
4097 if (block_migratetype == MIGRATE_RESERVE) {
4098 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
4099 move_freepages_block(zone, page, MIGRATE_MOVABLE);
4105 * Initially all pages are reserved - free ones are freed
4106 * up by free_all_bootmem() once the early boot process is
4107 * done. Non-atomic initialization, single-pass.
4109 void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
4110 unsigned long start_pfn, enum memmap_context context)
4113 unsigned long end_pfn = start_pfn + size;
4117 if (highest_memmap_pfn < end_pfn - 1)
4118 highest_memmap_pfn = end_pfn - 1;
4120 z = &NODE_DATA(nid)->node_zones[zone];
4121 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
4123 * There can be holes in boot-time mem_map[]s
4124 * handed to this function. They do not
4125 * exist on hotplugged memory.
4127 if (context == MEMMAP_EARLY) {
4128 if (!early_pfn_valid(pfn))
4130 if (!early_pfn_in_nid(pfn, nid))
4133 page = pfn_to_page(pfn);
4134 set_page_links(page, zone, nid, pfn);
4135 mminit_verify_page_links(page, zone, nid, pfn);
4136 init_page_count(page);
4137 page_mapcount_reset(page);
4138 page_nid_reset_last(page);
4139 SetPageReserved(page);
4141 * Mark the block movable so that blocks are reserved for
4142 * movable at startup. This will force kernel allocations
4143 * to reserve their blocks rather than leaking throughout
4144 * the address space during boot when many long-lived
4145 * kernel allocations are made. Later some blocks near
4146 * the start are marked MIGRATE_RESERVE by
4147 * setup_zone_migrate_reserve()
4149 * bitmap is created for zone's valid pfn range. but memmap
4150 * can be created for invalid pages (for alignment)
4151 * check here not to call set_pageblock_migratetype() against
4154 if ((z->zone_start_pfn <= pfn)
4155 && (pfn < zone_end_pfn(z))
4156 && !(pfn & (pageblock_nr_pages - 1)))
4157 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
4159 INIT_LIST_HEAD(&page->lru);
4160 #ifdef WANT_PAGE_VIRTUAL
4161 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
4162 if (!is_highmem_idx(zone))
4163 set_page_address(page, __va(pfn << PAGE_SHIFT));
4168 static void __meminit zone_init_free_lists(struct zone *zone)
4171 for_each_migratetype_order(order, t) {
4172 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
4173 zone->free_area[order].nr_free = 0;
4177 #ifndef __HAVE_ARCH_MEMMAP_INIT
4178 #define memmap_init(size, nid, zone, start_pfn) \
4179 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
4182 static int __meminit zone_batchsize(struct zone *zone)
4188 * The per-cpu-pages pools are set to around 1000th of the
4189 * size of the zone. But no more than 1/2 of a meg.
4191 * OK, so we don't know how big the cache is. So guess.
4193 batch = zone->managed_pages / 1024;
4194 if (batch * PAGE_SIZE > 512 * 1024)
4195 batch = (512 * 1024) / PAGE_SIZE;
4196 batch /= 4; /* We effectively *= 4 below */
4201 * Clamp the batch to a 2^n - 1 value. Having a power
4202 * of 2 value was found to be more likely to have
4203 * suboptimal cache aliasing properties in some cases.
4205 * For example if 2 tasks are alternately allocating
4206 * batches of pages, one task can end up with a lot
4207 * of pages of one half of the possible page colors
4208 * and the other with pages of the other colors.
4210 batch = rounddown_pow_of_two(batch + batch/2) - 1;
4215 /* The deferral and batching of frees should be suppressed under NOMMU
4218 * The problem is that NOMMU needs to be able to allocate large chunks
4219 * of contiguous memory as there's no hardware page translation to
4220 * assemble apparent contiguous memory from discontiguous pages.
4222 * Queueing large contiguous runs of pages for batching, however,
4223 * causes the pages to actually be freed in smaller chunks. As there
4224 * can be a significant delay between the individual batches being
4225 * recycled, this leads to the once large chunks of space being
4226 * fragmented and becoming unavailable for high-order allocations.
4232 static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
4234 struct per_cpu_pages *pcp;
4237 memset(p, 0, sizeof(*p));
4241 pcp->high = 6 * batch;
4242 pcp->batch = max(1UL, 1 * batch);
4243 for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
4244 INIT_LIST_HEAD(&pcp->lists[migratetype]);
4248 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
4249 * to the value high for the pageset p.
4252 static void setup_pagelist_highmark(struct per_cpu_pageset *p,
4255 struct per_cpu_pages *pcp;
4259 pcp->batch = max(1UL, high/4);
4260 if ((high/4) > (PAGE_SHIFT * 8))
4261 pcp->batch = PAGE_SHIFT * 8;
4264 static void __meminit setup_zone_pageset(struct zone *zone)
4268 zone->pageset = alloc_percpu(struct per_cpu_pageset);
4270 for_each_possible_cpu(cpu) {
4271 struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu);
4273 setup_pageset(pcp, zone_batchsize(zone));
4275 if (percpu_pagelist_fraction)
4276 setup_pagelist_highmark(pcp,
4277 (zone->managed_pages /
4278 percpu_pagelist_fraction));
4283 * Allocate per cpu pagesets and initialize them.
4284 * Before this call only boot pagesets were available.
4286 void __init setup_per_cpu_pageset(void)
4290 for_each_populated_zone(zone)
4291 setup_zone_pageset(zone);
4294 static noinline __init_refok
4295 int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
4298 struct pglist_data *pgdat = zone->zone_pgdat;
4302 * The per-page waitqueue mechanism uses hashed waitqueues
4305 zone->wait_table_hash_nr_entries =
4306 wait_table_hash_nr_entries(zone_size_pages);
4307 zone->wait_table_bits =
4308 wait_table_bits(zone->wait_table_hash_nr_entries);
4309 alloc_size = zone->wait_table_hash_nr_entries
4310 * sizeof(wait_queue_head_t);
4312 if (!slab_is_available()) {
4313 zone->wait_table = (wait_queue_head_t *)
4314 alloc_bootmem_node_nopanic(pgdat, alloc_size);
4317 * This case means that a zone whose size was 0 gets new memory
4318 * via memory hot-add.
4319 * But it may be the case that a new node was hot-added. In
4320 * this case vmalloc() will not be able to use this new node's
4321 * memory - this wait_table must be initialized to use this new
4322 * node itself as well.
4323 * To use this new node's memory, further consideration will be
4326 zone->wait_table = vmalloc(alloc_size);
4328 if (!zone->wait_table)
4331 for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
4332 init_waitqueue_head(zone->wait_table + i);
4337 static __meminit void zone_pcp_init(struct zone *zone)
4340 * per cpu subsystem is not up at this point. The following code
4341 * relies on the ability of the linker to provide the
4342 * offset of a (static) per cpu variable into the per cpu area.
4344 zone->pageset = &boot_pageset;
4346 if (zone->present_pages)
4347 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n",
4348 zone->name, zone->present_pages,
4349 zone_batchsize(zone));
4352 int __meminit init_currently_empty_zone(struct zone *zone,
4353 unsigned long zone_start_pfn,
4355 enum memmap_context context)
4357 struct pglist_data *pgdat = zone->zone_pgdat;
4359 ret = zone_wait_table_init(zone, size);
4362 pgdat->nr_zones = zone_idx(zone) + 1;
4364 zone->zone_start_pfn = zone_start_pfn;
4366 mminit_dprintk(MMINIT_TRACE, "memmap_init",
4367 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
4369 (unsigned long)zone_idx(zone),
4370 zone_start_pfn, (zone_start_pfn + size));
4372 zone_init_free_lists(zone);
4377 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4378 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
4380 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
4381 * Architectures may implement their own version but if add_active_range()
4382 * was used and there are no special requirements, this is a convenient
4385 int __meminit __early_pfn_to_nid(unsigned long pfn)
4387 unsigned long start_pfn, end_pfn;
4390 * NOTE: The following SMP-unsafe globals are only used early in boot
4391 * when the kernel is running single-threaded.
4393 static unsigned long __meminitdata last_start_pfn, last_end_pfn;
4394 static int __meminitdata last_nid;
4396 if (last_start_pfn <= pfn && pfn < last_end_pfn)
4399 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid)
4400 if (start_pfn <= pfn && pfn < end_pfn) {
4401 last_start_pfn = start_pfn;
4402 last_end_pfn = end_pfn;
4406 /* This is a memory hole */
4409 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
4411 int __meminit early_pfn_to_nid(unsigned long pfn)
4415 nid = __early_pfn_to_nid(pfn);
4418 /* just returns 0 */
4422 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
4423 bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
4427 nid = __early_pfn_to_nid(pfn);
4428 if (nid >= 0 && nid != node)
4435 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
4436 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
4437 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
4439 * If an architecture guarantees that all ranges registered with
4440 * add_active_ranges() contain no holes and may be freed, this
4441 * this function may be used instead of calling free_bootmem() manually.
4443 void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn)
4445 unsigned long start_pfn, end_pfn;
4448 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) {
4449 start_pfn = min(start_pfn, max_low_pfn);
4450 end_pfn = min(end_pfn, max_low_pfn);
4452 if (start_pfn < end_pfn)
4453 free_bootmem_node(NODE_DATA(this_nid),
4454 PFN_PHYS(start_pfn),
4455 (end_pfn - start_pfn) << PAGE_SHIFT);
4460 * sparse_memory_present_with_active_regions - Call memory_present for each active range
4461 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
4463 * If an architecture guarantees that all ranges registered with
4464 * add_active_ranges() contain no holes and may be freed, this
4465 * function may be used instead of calling memory_present() manually.
4467 void __init sparse_memory_present_with_active_regions(int nid)
4469 unsigned long start_pfn, end_pfn;
4472 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid)
4473 memory_present(this_nid, start_pfn, end_pfn);
4477 * get_pfn_range_for_nid - Return the start and end page frames for a node
4478 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
4479 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
4480 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
4482 * It returns the start and end page frame of a node based on information
4483 * provided by an arch calling add_active_range(). If called for a node
4484 * with no available memory, a warning is printed and the start and end
4487 void __meminit get_pfn_range_for_nid(unsigned int nid,
4488 unsigned long *start_pfn, unsigned long *end_pfn)
4490 unsigned long this_start_pfn, this_end_pfn;
4496 for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) {
4497 *start_pfn = min(*start_pfn, this_start_pfn);
4498 *end_pfn = max(*end_pfn, this_end_pfn);
4501 if (*start_pfn == -1UL)
4506 * This finds a zone that can be used for ZONE_MOVABLE pages. The
4507 * assumption is made that zones within a node are ordered in monotonic
4508 * increasing memory addresses so that the "highest" populated zone is used
4510 static void __init find_usable_zone_for_movable(void)
4513 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
4514 if (zone_index == ZONE_MOVABLE)
4517 if (arch_zone_highest_possible_pfn[zone_index] >
4518 arch_zone_lowest_possible_pfn[zone_index])
4522 VM_BUG_ON(zone_index == -1);
4523 movable_zone = zone_index;
4527 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
4528 * because it is sized independent of architecture. Unlike the other zones,
4529 * the starting point for ZONE_MOVABLE is not fixed. It may be different
4530 * in each node depending on the size of each node and how evenly kernelcore
4531 * is distributed. This helper function adjusts the zone ranges
4532 * provided by the architecture for a given node by using the end of the
4533 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
4534 * zones within a node are in order of monotonic increases memory addresses
4536 static void __meminit adjust_zone_range_for_zone_movable(int nid,
4537 unsigned long zone_type,
4538 unsigned long node_start_pfn,
4539 unsigned long node_end_pfn,
4540 unsigned long *zone_start_pfn,
4541 unsigned long *zone_end_pfn)
4543 /* Only adjust if ZONE_MOVABLE is on this node */
4544 if (zone_movable_pfn[nid]) {
4545 /* Size ZONE_MOVABLE */
4546 if (zone_type == ZONE_MOVABLE) {
4547 *zone_start_pfn = zone_movable_pfn[nid];
4548 *zone_end_pfn = min(node_end_pfn,
4549 arch_zone_highest_possible_pfn[movable_zone]);
4551 /* Adjust for ZONE_MOVABLE starting within this range */
4552 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
4553 *zone_end_pfn > zone_movable_pfn[nid]) {
4554 *zone_end_pfn = zone_movable_pfn[nid];
4556 /* Check if this whole range is within ZONE_MOVABLE */
4557 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
4558 *zone_start_pfn = *zone_end_pfn;
4563 * Return the number of pages a zone spans in a node, including holes
4564 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
4566 static unsigned long __meminit zone_spanned_pages_in_node(int nid,
4567 unsigned long zone_type,
4568 unsigned long *ignored)
4570 unsigned long node_start_pfn, node_end_pfn;
4571 unsigned long zone_start_pfn, zone_end_pfn;
4573 /* Get the start and end of the node and zone */
4574 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
4575 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
4576 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
4577 adjust_zone_range_for_zone_movable(nid, zone_type,
4578 node_start_pfn, node_end_pfn,
4579 &zone_start_pfn, &zone_end_pfn);
4581 /* Check that this node has pages within the zone's required range */
4582 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
4585 /* Move the zone boundaries inside the node if necessary */
4586 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
4587 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
4589 /* Return the spanned pages */
4590 return zone_end_pfn - zone_start_pfn;
4594 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
4595 * then all holes in the requested range will be accounted for.
4597 unsigned long __meminit __absent_pages_in_range(int nid,
4598 unsigned long range_start_pfn,
4599 unsigned long range_end_pfn)
4601 unsigned long nr_absent = range_end_pfn - range_start_pfn;
4602 unsigned long start_pfn, end_pfn;
4605 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
4606 start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn);
4607 end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn);
4608 nr_absent -= end_pfn - start_pfn;
4614 * absent_pages_in_range - Return number of page frames in holes within a range
4615 * @start_pfn: The start PFN to start searching for holes
4616 * @end_pfn: The end PFN to stop searching for holes
4618 * It returns the number of pages frames in memory holes within a range.
4620 unsigned long __init absent_pages_in_range(unsigned long start_pfn,
4621 unsigned long end_pfn)
4623 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
4626 /* Return the number of page frames in holes in a zone on a node */
4627 static unsigned long __meminit zone_absent_pages_in_node(int nid,
4628 unsigned long zone_type,
4629 unsigned long *ignored)
4631 unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type];
4632 unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type];
4633 unsigned long node_start_pfn, node_end_pfn;
4634 unsigned long zone_start_pfn, zone_end_pfn;
4636 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
4637 zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high);
4638 zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high);
4640 adjust_zone_range_for_zone_movable(nid, zone_type,
4641 node_start_pfn, node_end_pfn,
4642 &zone_start_pfn, &zone_end_pfn);
4643 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
4646 #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
4647 static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
4648 unsigned long zone_type,
4649 unsigned long *zones_size)
4651 return zones_size[zone_type];
4654 static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
4655 unsigned long zone_type,
4656 unsigned long *zholes_size)
4661 return zholes_size[zone_type];
4664 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
4666 static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
4667 unsigned long *zones_size, unsigned long *zholes_size)
4669 unsigned long realtotalpages, totalpages = 0;
4672 for (i = 0; i < MAX_NR_ZONES; i++)
4673 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
4675 pgdat->node_spanned_pages = totalpages;
4677 realtotalpages = totalpages;
4678 for (i = 0; i < MAX_NR_ZONES; i++)
4680 zone_absent_pages_in_node(pgdat->node_id, i,
4682 pgdat->node_present_pages = realtotalpages;
4683 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
4687 #ifndef CONFIG_SPARSEMEM
4689 * Calculate the size of the zone->blockflags rounded to an unsigned long
4690 * Start by making sure zonesize is a multiple of pageblock_order by rounding
4691 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
4692 * round what is now in bits to nearest long in bits, then return it in
4695 static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize)
4697 unsigned long usemapsize;
4699 zonesize += zone_start_pfn & (pageblock_nr_pages-1);
4700 usemapsize = roundup(zonesize, pageblock_nr_pages);
4701 usemapsize = usemapsize >> pageblock_order;
4702 usemapsize *= NR_PAGEBLOCK_BITS;
4703 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
4705 return usemapsize / 8;
4708 static void __init setup_usemap(struct pglist_data *pgdat,
4710 unsigned long zone_start_pfn,
4711 unsigned long zonesize)
4713 unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize);
4714 zone->pageblock_flags = NULL;
4716 zone->pageblock_flags = alloc_bootmem_node_nopanic(pgdat,
4720 static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone,
4721 unsigned long zone_start_pfn, unsigned long zonesize) {}
4722 #endif /* CONFIG_SPARSEMEM */
4724 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
4726 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
4727 void __init set_pageblock_order(void)
4731 /* Check that pageblock_nr_pages has not already been setup */
4732 if (pageblock_order)
4735 if (HPAGE_SHIFT > PAGE_SHIFT)
4736 order = HUGETLB_PAGE_ORDER;
4738 order = MAX_ORDER - 1;
4741 * Assume the largest contiguous order of interest is a huge page.
4742 * This value may be variable depending on boot parameters on IA64 and
4745 pageblock_order = order;
4747 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4750 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
4751 * is unused as pageblock_order is set at compile-time. See
4752 * include/linux/pageblock-flags.h for the values of pageblock_order based on
4755 void __init set_pageblock_order(void)
4759 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4761 static unsigned long __paginginit calc_memmap_size(unsigned long spanned_pages,
4762 unsigned long present_pages)
4764 unsigned long pages = spanned_pages;
4767 * Provide a more accurate estimation if there are holes within
4768 * the zone and SPARSEMEM is in use. If there are holes within the
4769 * zone, each populated memory region may cost us one or two extra
4770 * memmap pages due to alignment because memmap pages for each
4771 * populated regions may not naturally algined on page boundary.
4772 * So the (present_pages >> 4) heuristic is a tradeoff for that.
4774 if (spanned_pages > present_pages + (present_pages >> 4) &&
4775 IS_ENABLED(CONFIG_SPARSEMEM))
4776 pages = present_pages;
4778 return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT;
4782 * Set up the zone data structures:
4783 * - mark all pages reserved
4784 * - mark all memory queues empty
4785 * - clear the memory bitmaps
4787 * NOTE: pgdat should get zeroed by caller.
4789 static void __paginginit free_area_init_core(struct pglist_data *pgdat,
4790 unsigned long *zones_size, unsigned long *zholes_size)
4793 int nid = pgdat->node_id;
4794 unsigned long zone_start_pfn = pgdat->node_start_pfn;
4797 pgdat_resize_init(pgdat);
4798 #ifdef CONFIG_NUMA_BALANCING
4799 spin_lock_init(&pgdat->numabalancing_migrate_lock);
4800 pgdat->numabalancing_migrate_nr_pages = 0;
4801 pgdat->numabalancing_migrate_next_window = jiffies;
4803 init_waitqueue_head(&pgdat->kswapd_wait);
4804 init_waitqueue_head(&pgdat->pfmemalloc_wait);
4805 pgdat_page_cgroup_init(pgdat);
4807 for (j = 0; j < MAX_NR_ZONES; j++) {
4808 struct zone *zone = pgdat->node_zones + j;
4809 unsigned long size, realsize, freesize, memmap_pages;
4811 size = zone_spanned_pages_in_node(nid, j, zones_size);
4812 realsize = freesize = size - zone_absent_pages_in_node(nid, j,
4816 * Adjust freesize so that it accounts for how much memory
4817 * is used by this zone for memmap. This affects the watermark
4818 * and per-cpu initialisations
4820 memmap_pages = calc_memmap_size(size, realsize);
4821 if (freesize >= memmap_pages) {
4822 freesize -= memmap_pages;
4825 " %s zone: %lu pages used for memmap\n",
4826 zone_names[j], memmap_pages);
4829 " %s zone: %lu pages exceeds freesize %lu\n",
4830 zone_names[j], memmap_pages, freesize);
4832 /* Account for reserved pages */
4833 if (j == 0 && freesize > dma_reserve) {
4834 freesize -= dma_reserve;
4835 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
4836 zone_names[0], dma_reserve);
4839 if (!is_highmem_idx(j))
4840 nr_kernel_pages += freesize;
4841 /* Charge for highmem memmap if there are enough kernel pages */
4842 else if (nr_kernel_pages > memmap_pages * 2)
4843 nr_kernel_pages -= memmap_pages;
4844 nr_all_pages += freesize;
4846 zone->spanned_pages = size;
4847 zone->present_pages = realsize;
4849 * Set an approximate value for lowmem here, it will be adjusted
4850 * when the bootmem allocator frees pages into the buddy system.
4851 * And all highmem pages will be managed by the buddy system.
4853 zone->managed_pages = is_highmem_idx(j) ? realsize : freesize;
4856 zone->min_unmapped_pages = (freesize*sysctl_min_unmapped_ratio)
4858 zone->min_slab_pages = (freesize * sysctl_min_slab_ratio) / 100;
4860 zone->name = zone_names[j];
4861 spin_lock_init(&zone->lock);
4862 spin_lock_init(&zone->lru_lock);
4863 zone_seqlock_init(zone);
4864 zone->zone_pgdat = pgdat;
4866 zone_pcp_init(zone);
4867 lruvec_init(&zone->lruvec);
4871 set_pageblock_order();
4872 setup_usemap(pgdat, zone, zone_start_pfn, size);
4873 ret = init_currently_empty_zone(zone, zone_start_pfn,
4874 size, MEMMAP_EARLY);
4876 memmap_init(size, nid, j, zone_start_pfn);
4877 zone_start_pfn += size;
4881 static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
4883 /* Skip empty nodes */
4884 if (!pgdat->node_spanned_pages)
4887 #ifdef CONFIG_FLAT_NODE_MEM_MAP
4888 /* ia64 gets its own node_mem_map, before this, without bootmem */
4889 if (!pgdat->node_mem_map) {
4890 unsigned long size, start, end;
4894 * The zone's endpoints aren't required to be MAX_ORDER
4895 * aligned but the node_mem_map endpoints must be in order
4896 * for the buddy allocator to function correctly.
4898 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
4899 end = pgdat_end_pfn(pgdat);
4900 end = ALIGN(end, MAX_ORDER_NR_PAGES);
4901 size = (end - start) * sizeof(struct page);
4902 map = alloc_remap(pgdat->node_id, size);
4904 map = alloc_bootmem_node_nopanic(pgdat, size);
4905 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
4907 #ifndef CONFIG_NEED_MULTIPLE_NODES
4909 * With no DISCONTIG, the global mem_map is just set as node 0's
4911 if (pgdat == NODE_DATA(0)) {
4912 mem_map = NODE_DATA(0)->node_mem_map;
4913 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4914 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
4915 mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
4916 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
4919 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
4922 void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
4923 unsigned long node_start_pfn, unsigned long *zholes_size)
4925 pg_data_t *pgdat = NODE_DATA(nid);
4927 /* pg_data_t should be reset to zero when it's allocated */
4928 WARN_ON(pgdat->nr_zones || pgdat->classzone_idx);
4930 pgdat->node_id = nid;
4931 pgdat->node_start_pfn = node_start_pfn;
4932 init_zone_allows_reclaim(nid);
4933 calculate_node_totalpages(pgdat, zones_size, zholes_size);
4935 alloc_node_mem_map(pgdat);
4936 #ifdef CONFIG_FLAT_NODE_MEM_MAP
4937 printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
4938 nid, (unsigned long)pgdat,
4939 (unsigned long)pgdat->node_mem_map);
4942 free_area_init_core(pgdat, zones_size, zholes_size);
4945 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4947 #if MAX_NUMNODES > 1
4949 * Figure out the number of possible node ids.
4951 void __init setup_nr_node_ids(void)
4954 unsigned int highest = 0;
4956 for_each_node_mask(node, node_possible_map)
4958 nr_node_ids = highest + 1;
4963 * node_map_pfn_alignment - determine the maximum internode alignment
4965 * This function should be called after node map is populated and sorted.
4966 * It calculates the maximum power of two alignment which can distinguish
4969 * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
4970 * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the
4971 * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is
4972 * shifted, 1GiB is enough and this function will indicate so.
4974 * This is used to test whether pfn -> nid mapping of the chosen memory
4975 * model has fine enough granularity to avoid incorrect mapping for the
4976 * populated node map.
4978 * Returns the determined alignment in pfn's. 0 if there is no alignment
4979 * requirement (single node).
4981 unsigned long __init node_map_pfn_alignment(void)
4983 unsigned long accl_mask = 0, last_end = 0;
4984 unsigned long start, end, mask;
4988 for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) {
4989 if (!start || last_nid < 0 || last_nid == nid) {
4996 * Start with a mask granular enough to pin-point to the
4997 * start pfn and tick off bits one-by-one until it becomes
4998 * too coarse to separate the current node from the last.
5000 mask = ~((1 << __ffs(start)) - 1);
5001 while (mask && last_end <= (start & (mask << 1)))
5004 /* accumulate all internode masks */
5008 /* convert mask to number of pages */
5009 return ~accl_mask + 1;
5012 /* Find the lowest pfn for a node */
5013 static unsigned long __init find_min_pfn_for_node(int nid)
5015 unsigned long min_pfn = ULONG_MAX;
5016 unsigned long start_pfn;
5019 for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL)
5020 min_pfn = min(min_pfn, start_pfn);
5022 if (min_pfn == ULONG_MAX) {
5024 "Could not find start_pfn for node %d\n", nid);
5032 * find_min_pfn_with_active_regions - Find the minimum PFN registered
5034 * It returns the minimum PFN based on information provided via
5035 * add_active_range().
5037 unsigned long __init find_min_pfn_with_active_regions(void)
5039 return find_min_pfn_for_node(MAX_NUMNODES);
5043 * early_calculate_totalpages()
5044 * Sum pages in active regions for movable zone.
5045 * Populate N_MEMORY for calculating usable_nodes.
5047 static unsigned long __init early_calculate_totalpages(void)
5049 unsigned long totalpages = 0;
5050 unsigned long start_pfn, end_pfn;
5053 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
5054 unsigned long pages = end_pfn - start_pfn;
5056 totalpages += pages;
5058 node_set_state(nid, N_MEMORY);
5064 * Find the PFN the Movable zone begins in each node. Kernel memory
5065 * is spread evenly between nodes as long as the nodes have enough
5066 * memory. When they don't, some nodes will have more kernelcore than
5069 static void __init find_zone_movable_pfns_for_nodes(void)
5072 unsigned long usable_startpfn;
5073 unsigned long kernelcore_node, kernelcore_remaining;
5074 /* save the state before borrow the nodemask */
5075 nodemask_t saved_node_state = node_states[N_MEMORY];
5076 unsigned long totalpages = early_calculate_totalpages();
5077 int usable_nodes = nodes_weight(node_states[N_MEMORY]);
5080 * If movablecore was specified, calculate what size of
5081 * kernelcore that corresponds so that memory usable for
5082 * any allocation type is evenly spread. If both kernelcore
5083 * and movablecore are specified, then the value of kernelcore
5084 * will be used for required_kernelcore if it's greater than
5085 * what movablecore would have allowed.
5087 if (required_movablecore) {
5088 unsigned long corepages;
5091 * Round-up so that ZONE_MOVABLE is at least as large as what
5092 * was requested by the user
5094 required_movablecore =
5095 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
5096 corepages = totalpages - required_movablecore;
5098 required_kernelcore = max(required_kernelcore, corepages);
5101 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
5102 if (!required_kernelcore)
5105 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
5106 find_usable_zone_for_movable();
5107 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
5110 /* Spread kernelcore memory as evenly as possible throughout nodes */
5111 kernelcore_node = required_kernelcore / usable_nodes;
5112 for_each_node_state(nid, N_MEMORY) {
5113 unsigned long start_pfn, end_pfn;
5116 * Recalculate kernelcore_node if the division per node
5117 * now exceeds what is necessary to satisfy the requested
5118 * amount of memory for the kernel
5120 if (required_kernelcore < kernelcore_node)
5121 kernelcore_node = required_kernelcore / usable_nodes;
5124 * As the map is walked, we track how much memory is usable
5125 * by the kernel using kernelcore_remaining. When it is
5126 * 0, the rest of the node is usable by ZONE_MOVABLE
5128 kernelcore_remaining = kernelcore_node;
5130 /* Go through each range of PFNs within this node */
5131 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
5132 unsigned long size_pages;
5134 start_pfn = max(start_pfn, zone_movable_pfn[nid]);
5135 if (start_pfn >= end_pfn)
5138 /* Account for what is only usable for kernelcore */
5139 if (start_pfn < usable_startpfn) {
5140 unsigned long kernel_pages;
5141 kernel_pages = min(end_pfn, usable_startpfn)
5144 kernelcore_remaining -= min(kernel_pages,
5145 kernelcore_remaining);
5146 required_kernelcore -= min(kernel_pages,
5147 required_kernelcore);
5149 /* Continue if range is now fully accounted */
5150 if (end_pfn <= usable_startpfn) {
5153 * Push zone_movable_pfn to the end so
5154 * that if we have to rebalance
5155 * kernelcore across nodes, we will
5156 * not double account here
5158 zone_movable_pfn[nid] = end_pfn;
5161 start_pfn = usable_startpfn;
5165 * The usable PFN range for ZONE_MOVABLE is from
5166 * start_pfn->end_pfn. Calculate size_pages as the
5167 * number of pages used as kernelcore
5169 size_pages = end_pfn - start_pfn;
5170 if (size_pages > kernelcore_remaining)
5171 size_pages = kernelcore_remaining;
5172 zone_movable_pfn[nid] = start_pfn + size_pages;
5175 * Some kernelcore has been met, update counts and
5176 * break if the kernelcore for this node has been
5179 required_kernelcore -= min(required_kernelcore,
5181 kernelcore_remaining -= size_pages;
5182 if (!kernelcore_remaining)
5188 * If there is still required_kernelcore, we do another pass with one
5189 * less node in the count. This will push zone_movable_pfn[nid] further
5190 * along on the nodes that still have memory until kernelcore is
5194 if (usable_nodes && required_kernelcore > usable_nodes)
5197 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
5198 for (nid = 0; nid < MAX_NUMNODES; nid++)
5199 zone_movable_pfn[nid] =
5200 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
5203 /* restore the node_state */
5204 node_states[N_MEMORY] = saved_node_state;
5207 /* Any regular or high memory on that node ? */
5208 static void check_for_memory(pg_data_t *pgdat, int nid)
5210 enum zone_type zone_type;
5212 if (N_MEMORY == N_NORMAL_MEMORY)
5215 for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) {
5216 struct zone *zone = &pgdat->node_zones[zone_type];
5217 if (zone->present_pages) {
5218 node_set_state(nid, N_HIGH_MEMORY);
5219 if (N_NORMAL_MEMORY != N_HIGH_MEMORY &&
5220 zone_type <= ZONE_NORMAL)
5221 node_set_state(nid, N_NORMAL_MEMORY);
5228 * free_area_init_nodes - Initialise all pg_data_t and zone data
5229 * @max_zone_pfn: an array of max PFNs for each zone
5231 * This will call free_area_init_node() for each active node in the system.
5232 * Using the page ranges provided by add_active_range(), the size of each
5233 * zone in each node and their holes is calculated. If the maximum PFN
5234 * between two adjacent zones match, it is assumed that the zone is empty.
5235 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
5236 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
5237 * starts where the previous one ended. For example, ZONE_DMA32 starts
5238 * at arch_max_dma_pfn.
5240 void __init free_area_init_nodes(unsigned long *max_zone_pfn)
5242 unsigned long start_pfn, end_pfn;
5245 /* Record where the zone boundaries are */
5246 memset(arch_zone_lowest_possible_pfn, 0,
5247 sizeof(arch_zone_lowest_possible_pfn));
5248 memset(arch_zone_highest_possible_pfn, 0,
5249 sizeof(arch_zone_highest_possible_pfn));
5250 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
5251 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
5252 for (i = 1; i < MAX_NR_ZONES; i++) {
5253 if (i == ZONE_MOVABLE)
5255 arch_zone_lowest_possible_pfn[i] =
5256 arch_zone_highest_possible_pfn[i-1];
5257 arch_zone_highest_possible_pfn[i] =
5258 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
5260 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
5261 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
5263 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
5264 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
5265 find_zone_movable_pfns_for_nodes();
5267 /* Print out the zone ranges */
5268 printk("Zone ranges:\n");
5269 for (i = 0; i < MAX_NR_ZONES; i++) {
5270 if (i == ZONE_MOVABLE)
5272 printk(KERN_CONT " %-8s ", zone_names[i]);
5273 if (arch_zone_lowest_possible_pfn[i] ==
5274 arch_zone_highest_possible_pfn[i])
5275 printk(KERN_CONT "empty\n");
5277 printk(KERN_CONT "[mem %0#10lx-%0#10lx]\n",
5278 arch_zone_lowest_possible_pfn[i] << PAGE_SHIFT,
5279 (arch_zone_highest_possible_pfn[i]
5280 << PAGE_SHIFT) - 1);
5283 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
5284 printk("Movable zone start for each node\n");
5285 for (i = 0; i < MAX_NUMNODES; i++) {
5286 if (zone_movable_pfn[i])
5287 printk(" Node %d: %#010lx\n", i,
5288 zone_movable_pfn[i] << PAGE_SHIFT);
5291 /* Print out the early node map */
5292 printk("Early memory node ranges\n");
5293 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid)
5294 printk(" node %3d: [mem %#010lx-%#010lx]\n", nid,
5295 start_pfn << PAGE_SHIFT, (end_pfn << PAGE_SHIFT) - 1);
5297 /* Initialise every node */
5298 mminit_verify_pageflags_layout();
5299 setup_nr_node_ids();
5300 for_each_online_node(nid) {
5301 pg_data_t *pgdat = NODE_DATA(nid);
5302 free_area_init_node(nid, NULL,
5303 find_min_pfn_for_node(nid), NULL);
5305 /* Any memory on that node */
5306 if (pgdat->node_present_pages)
5307 node_set_state(nid, N_MEMORY);
5308 check_for_memory(pgdat, nid);
5312 static int __init cmdline_parse_core(char *p, unsigned long *core)
5314 unsigned long long coremem;
5318 coremem = memparse(p, &p);
5319 *core = coremem >> PAGE_SHIFT;
5321 /* Paranoid check that UL is enough for the coremem value */
5322 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
5328 * kernelcore=size sets the amount of memory for use for allocations that
5329 * cannot be reclaimed or migrated.
5331 static int __init cmdline_parse_kernelcore(char *p)
5333 return cmdline_parse_core(p, &required_kernelcore);
5337 * movablecore=size sets the amount of memory for use for allocations that
5338 * can be reclaimed or migrated.
5340 static int __init cmdline_parse_movablecore(char *p)
5342 return cmdline_parse_core(p, &required_movablecore);
5345 early_param("kernelcore", cmdline_parse_kernelcore);
5346 early_param("movablecore", cmdline_parse_movablecore);
5348 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5350 unsigned long free_reserved_area(unsigned long start, unsigned long end,
5351 int poison, char *s)
5353 unsigned long pages, pos;
5355 pos = start = PAGE_ALIGN(start);
5357 for (pages = 0; pos < end; pos += PAGE_SIZE, pages++) {
5359 memset((void *)pos, poison, PAGE_SIZE);
5360 free_reserved_page(virt_to_page((void *)pos));
5364 pr_info("Freeing %s memory: %ldK (%lx - %lx)\n",
5365 s, pages << (PAGE_SHIFT - 10), start, end);
5370 #ifdef CONFIG_HIGHMEM
5371 void free_highmem_page(struct page *page)
5373 __free_reserved_page(page);
5380 * set_dma_reserve - set the specified number of pages reserved in the first zone
5381 * @new_dma_reserve: The number of pages to mark reserved
5383 * The per-cpu batchsize and zone watermarks are determined by present_pages.
5384 * In the DMA zone, a significant percentage may be consumed by kernel image
5385 * and other unfreeable allocations which can skew the watermarks badly. This
5386 * function may optionally be used to account for unfreeable pages in the
5387 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
5388 * smaller per-cpu batchsize.
5390 void __init set_dma_reserve(unsigned long new_dma_reserve)
5392 dma_reserve = new_dma_reserve;
5395 void __init free_area_init(unsigned long *zones_size)
5397 free_area_init_node(0, zones_size,
5398 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
5401 static int page_alloc_cpu_notify(struct notifier_block *self,
5402 unsigned long action, void *hcpu)
5404 int cpu = (unsigned long)hcpu;
5406 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
5407 lru_add_drain_cpu(cpu);
5411 * Spill the event counters of the dead processor
5412 * into the current processors event counters.
5413 * This artificially elevates the count of the current
5416 vm_events_fold_cpu(cpu);
5419 * Zero the differential counters of the dead processor
5420 * so that the vm statistics are consistent.
5422 * This is only okay since the processor is dead and cannot
5423 * race with what we are doing.
5425 refresh_cpu_vm_stats(cpu);
5430 void __init page_alloc_init(void)
5432 hotcpu_notifier(page_alloc_cpu_notify, 0);
5436 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
5437 * or min_free_kbytes changes.
5439 static void calculate_totalreserve_pages(void)
5441 struct pglist_data *pgdat;
5442 unsigned long reserve_pages = 0;
5443 enum zone_type i, j;
5445 for_each_online_pgdat(pgdat) {
5446 for (i = 0; i < MAX_NR_ZONES; i++) {
5447 struct zone *zone = pgdat->node_zones + i;
5448 unsigned long max = 0;
5450 /* Find valid and maximum lowmem_reserve in the zone */
5451 for (j = i; j < MAX_NR_ZONES; j++) {
5452 if (zone->lowmem_reserve[j] > max)
5453 max = zone->lowmem_reserve[j];
5456 /* we treat the high watermark as reserved pages. */
5457 max += high_wmark_pages(zone);
5459 if (max > zone->managed_pages)
5460 max = zone->managed_pages;
5461 reserve_pages += max;
5463 * Lowmem reserves are not available to
5464 * GFP_HIGHUSER page cache allocations and
5465 * kswapd tries to balance zones to their high
5466 * watermark. As a result, neither should be
5467 * regarded as dirtyable memory, to prevent a
5468 * situation where reclaim has to clean pages
5469 * in order to balance the zones.
5471 zone->dirty_balance_reserve = max;
5474 dirty_balance_reserve = reserve_pages;
5475 totalreserve_pages = reserve_pages;
5479 * setup_per_zone_lowmem_reserve - called whenever
5480 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
5481 * has a correct pages reserved value, so an adequate number of
5482 * pages are left in the zone after a successful __alloc_pages().
5484 static void setup_per_zone_lowmem_reserve(void)
5486 struct pglist_data *pgdat;
5487 enum zone_type j, idx;
5489 for_each_online_pgdat(pgdat) {
5490 for (j = 0; j < MAX_NR_ZONES; j++) {
5491 struct zone *zone = pgdat->node_zones + j;
5492 unsigned long managed_pages = zone->managed_pages;
5494 zone->lowmem_reserve[j] = 0;
5498 struct zone *lower_zone;
5502 if (sysctl_lowmem_reserve_ratio[idx] < 1)
5503 sysctl_lowmem_reserve_ratio[idx] = 1;
5505 lower_zone = pgdat->node_zones + idx;
5506 lower_zone->lowmem_reserve[j] = managed_pages /
5507 sysctl_lowmem_reserve_ratio[idx];
5508 managed_pages += lower_zone->managed_pages;
5513 /* update totalreserve_pages */
5514 calculate_totalreserve_pages();
5517 static void __setup_per_zone_wmarks(void)
5519 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
5520 unsigned long pages_low = extra_free_kbytes >> (PAGE_SHIFT - 10);
5521 unsigned long lowmem_pages = 0;
5523 unsigned long flags;
5525 /* Calculate total number of !ZONE_HIGHMEM pages */
5526 for_each_zone(zone) {
5527 if (!is_highmem(zone))
5528 lowmem_pages += zone->managed_pages;
5531 for_each_zone(zone) {
5534 spin_lock_irqsave(&zone->lock, flags);
5535 min = (u64)pages_min * zone->managed_pages;
5536 do_div(min, lowmem_pages);
5537 low = (u64)pages_low * zone->managed_pages;
5538 do_div(low, vm_total_pages);
5540 if (is_highmem(zone)) {
5542 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
5543 * need highmem pages, so cap pages_min to a small
5546 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
5547 * deltas controls asynch page reclaim, and so should
5548 * not be capped for highmem.
5550 unsigned long min_pages;
5552 min_pages = zone->managed_pages / 1024;
5553 min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL);
5554 zone->watermark[WMARK_MIN] = min_pages;
5557 * If it's a lowmem zone, reserve a number of pages
5558 * proportionate to the zone's size.
5560 zone->watermark[WMARK_MIN] = min;
5563 zone->watermark[WMARK_LOW] = min_wmark_pages(zone) +
5565 zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) +
5568 setup_zone_migrate_reserve(zone);
5569 spin_unlock_irqrestore(&zone->lock, flags);
5572 /* update totalreserve_pages */
5573 calculate_totalreserve_pages();
5577 * setup_per_zone_wmarks - called when min_free_kbytes changes
5578 * or when memory is hot-{added|removed}
5580 * Ensures that the watermark[min,low,high] values for each zone are set
5581 * correctly with respect to min_free_kbytes.
5583 void setup_per_zone_wmarks(void)
5585 mutex_lock(&zonelists_mutex);
5586 __setup_per_zone_wmarks();
5587 mutex_unlock(&zonelists_mutex);
5591 * The inactive anon list should be small enough that the VM never has to
5592 * do too much work, but large enough that each inactive page has a chance
5593 * to be referenced again before it is swapped out.
5595 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
5596 * INACTIVE_ANON pages on this zone's LRU, maintained by the
5597 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
5598 * the anonymous pages are kept on the inactive list.
5601 * memory ratio inactive anon
5602 * -------------------------------------
5611 static void __meminit calculate_zone_inactive_ratio(struct zone *zone)
5613 unsigned int gb, ratio;
5615 /* Zone size in gigabytes */
5616 gb = zone->managed_pages >> (30 - PAGE_SHIFT);
5618 ratio = int_sqrt(10 * gb);
5622 zone->inactive_ratio = ratio;
5625 static void __meminit setup_per_zone_inactive_ratio(void)
5630 calculate_zone_inactive_ratio(zone);
5634 * Initialise min_free_kbytes.
5636 * For small machines we want it small (128k min). For large machines
5637 * we want it large (64MB max). But it is not linear, because network
5638 * bandwidth does not increase linearly with machine size. We use
5640 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
5641 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
5657 int __meminit init_per_zone_wmark_min(void)
5659 unsigned long lowmem_kbytes;
5661 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
5663 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
5664 if (min_free_kbytes < 128)
5665 min_free_kbytes = 128;
5666 if (min_free_kbytes > 65536)
5667 min_free_kbytes = 65536;
5668 setup_per_zone_wmarks();
5669 refresh_zone_stat_thresholds();
5670 setup_per_zone_lowmem_reserve();
5671 setup_per_zone_inactive_ratio();
5674 module_init(init_per_zone_wmark_min)
5677 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
5678 * that we can call two helper functions whenever min_free_kbytes
5679 * or extra_free_kbytes changes.
5681 int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
5682 void __user *buffer, size_t *length, loff_t *ppos)
5684 proc_dointvec(table, write, buffer, length, ppos);
5686 setup_per_zone_wmarks();
5691 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
5692 void __user *buffer, size_t *length, loff_t *ppos)
5697 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
5702 zone->min_unmapped_pages = (zone->managed_pages *
5703 sysctl_min_unmapped_ratio) / 100;
5707 int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
5708 void __user *buffer, size_t *length, loff_t *ppos)
5713 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
5718 zone->min_slab_pages = (zone->managed_pages *
5719 sysctl_min_slab_ratio) / 100;
5725 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
5726 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
5727 * whenever sysctl_lowmem_reserve_ratio changes.
5729 * The reserve ratio obviously has absolutely no relation with the
5730 * minimum watermarks. The lowmem reserve ratio can only make sense
5731 * if in function of the boot time zone sizes.
5733 int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
5734 void __user *buffer, size_t *length, loff_t *ppos)
5736 proc_dointvec_minmax(table, write, buffer, length, ppos);
5737 setup_per_zone_lowmem_reserve();
5742 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
5743 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
5744 * can have before it gets flushed back to buddy allocator.
5747 int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
5748 void __user *buffer, size_t *length, loff_t *ppos)
5754 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
5755 if (!write || (ret < 0))
5757 for_each_populated_zone(zone) {
5758 for_each_possible_cpu(cpu) {
5760 high = zone->managed_pages / percpu_pagelist_fraction;
5761 setup_pagelist_highmark(
5762 per_cpu_ptr(zone->pageset, cpu), high);
5768 int hashdist = HASHDIST_DEFAULT;
5771 static int __init set_hashdist(char *str)
5775 hashdist = simple_strtoul(str, &str, 0);
5778 __setup("hashdist=", set_hashdist);
5782 * allocate a large system hash table from bootmem
5783 * - it is assumed that the hash table must contain an exact power-of-2
5784 * quantity of entries
5785 * - limit is the number of hash buckets, not the total allocation size
5787 void *__init alloc_large_system_hash(const char *tablename,
5788 unsigned long bucketsize,
5789 unsigned long numentries,
5792 unsigned int *_hash_shift,
5793 unsigned int *_hash_mask,
5794 unsigned long low_limit,
5795 unsigned long high_limit)
5797 unsigned long long max = high_limit;
5798 unsigned long log2qty, size;
5801 /* allow the kernel cmdline to have a say */
5803 /* round applicable memory size up to nearest megabyte */
5804 numentries = nr_kernel_pages;
5805 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
5806 numentries >>= 20 - PAGE_SHIFT;
5807 numentries <<= 20 - PAGE_SHIFT;
5809 /* limit to 1 bucket per 2^scale bytes of low memory */
5810 if (scale > PAGE_SHIFT)
5811 numentries >>= (scale - PAGE_SHIFT);
5813 numentries <<= (PAGE_SHIFT - scale);
5815 /* Make sure we've got at least a 0-order allocation.. */
5816 if (unlikely(flags & HASH_SMALL)) {
5817 /* Makes no sense without HASH_EARLY */
5818 WARN_ON(!(flags & HASH_EARLY));
5819 if (!(numentries >> *_hash_shift)) {
5820 numentries = 1UL << *_hash_shift;
5821 BUG_ON(!numentries);
5823 } else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
5824 numentries = PAGE_SIZE / bucketsize;
5826 numentries = roundup_pow_of_two(numentries);
5828 /* limit allocation size to 1/16 total memory by default */
5830 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
5831 do_div(max, bucketsize);
5833 max = min(max, 0x80000000ULL);
5835 if (numentries < low_limit)
5836 numentries = low_limit;
5837 if (numentries > max)
5840 log2qty = ilog2(numentries);
5843 size = bucketsize << log2qty;
5844 if (flags & HASH_EARLY)
5845 table = alloc_bootmem_nopanic(size);
5847 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
5850 * If bucketsize is not a power-of-two, we may free
5851 * some pages at the end of hash table which
5852 * alloc_pages_exact() automatically does
5854 if (get_order(size) < MAX_ORDER) {
5855 table = alloc_pages_exact(size, GFP_ATOMIC);
5856 kmemleak_alloc(table, size, 1, GFP_ATOMIC);
5859 } while (!table && size > PAGE_SIZE && --log2qty);
5862 panic("Failed to allocate %s hash table\n", tablename);
5864 printk(KERN_INFO "%s hash table entries: %ld (order: %d, %lu bytes)\n",
5867 ilog2(size) - PAGE_SHIFT,
5871 *_hash_shift = log2qty;
5873 *_hash_mask = (1 << log2qty) - 1;
5878 /* Return a pointer to the bitmap storing bits affecting a block of pages */
5879 static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
5882 #ifdef CONFIG_SPARSEMEM
5883 return __pfn_to_section(pfn)->pageblock_flags;
5885 return zone->pageblock_flags;
5886 #endif /* CONFIG_SPARSEMEM */
5889 static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
5891 #ifdef CONFIG_SPARSEMEM
5892 pfn &= (PAGES_PER_SECTION-1);
5893 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
5895 pfn = pfn - round_down(zone->zone_start_pfn, pageblock_nr_pages);
5896 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
5897 #endif /* CONFIG_SPARSEMEM */
5901 * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
5902 * @page: The page within the block of interest
5903 * @start_bitidx: The first bit of interest to retrieve
5904 * @end_bitidx: The last bit of interest
5905 * returns pageblock_bits flags
5907 unsigned long get_pageblock_flags_group(struct page *page,
5908 int start_bitidx, int end_bitidx)
5911 unsigned long *bitmap;
5912 unsigned long pfn, bitidx;
5913 unsigned long flags = 0;
5914 unsigned long value = 1;
5916 zone = page_zone(page);
5917 pfn = page_to_pfn(page);
5918 bitmap = get_pageblock_bitmap(zone, pfn);
5919 bitidx = pfn_to_bitidx(zone, pfn);
5921 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
5922 if (test_bit(bitidx + start_bitidx, bitmap))
5929 * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
5930 * @page: The page within the block of interest
5931 * @start_bitidx: The first bit of interest
5932 * @end_bitidx: The last bit of interest
5933 * @flags: The flags to set
5935 void set_pageblock_flags_group(struct page *page, unsigned long flags,
5936 int start_bitidx, int end_bitidx)
5939 unsigned long *bitmap;
5940 unsigned long pfn, bitidx;
5941 unsigned long value = 1;
5943 zone = page_zone(page);
5944 pfn = page_to_pfn(page);
5945 bitmap = get_pageblock_bitmap(zone, pfn);
5946 bitidx = pfn_to_bitidx(zone, pfn);
5947 VM_BUG_ON(!zone_spans_pfn(zone, pfn));
5949 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
5951 __set_bit(bitidx + start_bitidx, bitmap);
5953 __clear_bit(bitidx + start_bitidx, bitmap);
5957 * This function checks whether pageblock includes unmovable pages or not.
5958 * If @count is not zero, it is okay to include less @count unmovable pages
5960 * PageLRU check wihtout isolation or lru_lock could race so that
5961 * MIGRATE_MOVABLE block might include unmovable pages. It means you can't
5962 * expect this function should be exact.
5964 bool has_unmovable_pages(struct zone *zone, struct page *page, int count,
5965 bool skip_hwpoisoned_pages)
5967 unsigned long pfn, iter, found;
5971 * For avoiding noise data, lru_add_drain_all() should be called
5972 * If ZONE_MOVABLE, the zone never contains unmovable pages
5974 if (zone_idx(zone) == ZONE_MOVABLE)
5976 mt = get_pageblock_migratetype(page);
5977 if (mt == MIGRATE_MOVABLE || is_migrate_cma(mt))
5980 pfn = page_to_pfn(page);
5981 for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) {
5982 unsigned long check = pfn + iter;
5984 if (!pfn_valid_within(check))
5987 page = pfn_to_page(check);
5989 * We can't use page_count without pin a page
5990 * because another CPU can free compound page.
5991 * This check already skips compound tails of THP
5992 * because their page->_count is zero at all time.
5994 if (!atomic_read(&page->_count)) {
5995 if (PageBuddy(page))
5996 iter += (1 << page_order(page)) - 1;
6001 * The HWPoisoned page may be not in buddy system, and
6002 * page_count() is not 0.
6004 if (skip_hwpoisoned_pages && PageHWPoison(page))
6010 * If there are RECLAIMABLE pages, we need to check it.
6011 * But now, memory offline itself doesn't call shrink_slab()
6012 * and it still to be fixed.
6015 * If the page is not RAM, page_count()should be 0.
6016 * we don't need more check. This is an _used_ not-movable page.
6018 * The problematic thing here is PG_reserved pages. PG_reserved
6019 * is set to both of a memory hole page and a _used_ kernel
6028 bool is_pageblock_removable_nolock(struct page *page)
6034 * We have to be careful here because we are iterating over memory
6035 * sections which are not zone aware so we might end up outside of
6036 * the zone but still within the section.
6037 * We have to take care about the node as well. If the node is offline
6038 * its NODE_DATA will be NULL - see page_zone.
6040 if (!node_online(page_to_nid(page)))
6043 zone = page_zone(page);
6044 pfn = page_to_pfn(page);
6045 if (!zone_spans_pfn(zone, pfn))
6048 return !has_unmovable_pages(zone, page, 0, true);
6053 static unsigned long pfn_max_align_down(unsigned long pfn)
6055 return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES,
6056 pageblock_nr_pages) - 1);
6059 static unsigned long pfn_max_align_up(unsigned long pfn)
6061 return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES,
6062 pageblock_nr_pages));
6065 /* [start, end) must belong to a single zone. */
6066 static int __alloc_contig_migrate_range(struct compact_control *cc,
6067 unsigned long start, unsigned long end)
6069 /* This function is based on compact_zone() from compaction.c. */
6070 unsigned long nr_reclaimed;
6071 unsigned long pfn = start;
6072 unsigned int tries = 0;
6077 while (pfn < end || !list_empty(&cc->migratepages)) {
6078 if (fatal_signal_pending(current)) {
6083 if (list_empty(&cc->migratepages)) {
6084 cc->nr_migratepages = 0;
6085 pfn = isolate_migratepages_range(cc->zone, cc,
6092 } else if (++tries == 5) {
6093 ret = ret < 0 ? ret : -EBUSY;
6097 nr_reclaimed = reclaim_clean_pages_from_list(cc->zone,
6099 cc->nr_migratepages -= nr_reclaimed;
6101 ret = migrate_pages(&cc->migratepages, alloc_migrate_target,
6102 0, MIGRATE_SYNC, MR_CMA);
6105 putback_movable_pages(&cc->migratepages);
6112 * alloc_contig_range() -- tries to allocate given range of pages
6113 * @start: start PFN to allocate
6114 * @end: one-past-the-last PFN to allocate
6115 * @migratetype: migratetype of the underlaying pageblocks (either
6116 * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks
6117 * in range must have the same migratetype and it must
6118 * be either of the two.
6120 * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
6121 * aligned, however it's the caller's responsibility to guarantee that
6122 * we are the only thread that changes migrate type of pageblocks the
6125 * The PFN range must belong to a single zone.
6127 * Returns zero on success or negative error code. On success all
6128 * pages which PFN is in [start, end) are allocated for the caller and
6129 * need to be freed with free_contig_range().
6131 int alloc_contig_range(unsigned long start, unsigned long end,
6132 unsigned migratetype)
6134 unsigned long outer_start, outer_end;
6137 struct compact_control cc = {
6138 .nr_migratepages = 0,
6140 .zone = page_zone(pfn_to_page(start)),
6142 .ignore_skip_hint = true,
6144 INIT_LIST_HEAD(&cc.migratepages);
6147 * What we do here is we mark all pageblocks in range as
6148 * MIGRATE_ISOLATE. Because pageblock and max order pages may
6149 * have different sizes, and due to the way page allocator
6150 * work, we align the range to biggest of the two pages so
6151 * that page allocator won't try to merge buddies from
6152 * different pageblocks and change MIGRATE_ISOLATE to some
6153 * other migration type.
6155 * Once the pageblocks are marked as MIGRATE_ISOLATE, we
6156 * migrate the pages from an unaligned range (ie. pages that
6157 * we are interested in). This will put all the pages in
6158 * range back to page allocator as MIGRATE_ISOLATE.
6160 * When this is done, we take the pages in range from page
6161 * allocator removing them from the buddy system. This way
6162 * page allocator will never consider using them.
6164 * This lets us mark the pageblocks back as
6165 * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
6166 * aligned range but not in the unaligned, original range are
6167 * put back to page allocator so that buddy can use them.
6170 ret = start_isolate_page_range(pfn_max_align_down(start),
6171 pfn_max_align_up(end), migratetype,
6175 #ifdef CONFIG_CMA_RMQUEUE
6176 cc.zone->cma_alloc = 1;
6178 ret = __alloc_contig_migrate_range(&cc, start, end);
6183 * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
6184 * aligned blocks that are marked as MIGRATE_ISOLATE. What's
6185 * more, all pages in [start, end) are free in page allocator.
6186 * What we are going to do is to allocate all pages from
6187 * [start, end) (that is remove them from page allocator).
6189 * The only problem is that pages at the beginning and at the
6190 * end of interesting range may be not aligned with pages that
6191 * page allocator holds, ie. they can be part of higher order
6192 * pages. Because of this, we reserve the bigger range and
6193 * once this is done free the pages we are not interested in.
6195 * We don't have to hold zone->lock here because the pages are
6196 * isolated thus they won't get removed from buddy.
6199 lru_add_drain_all();
6203 outer_start = start;
6204 while (!PageBuddy(pfn_to_page(outer_start))) {
6205 if (++order >= MAX_ORDER) {
6209 outer_start &= ~0UL << order;
6212 /* Make sure the range is really isolated. */
6213 if (test_pages_isolated(outer_start, end, false)) {
6214 pr_warn("alloc_contig_range test_pages_isolated(%lx, %lx) failed\n",
6221 /* Grab isolated pages from freelists. */
6222 outer_end = isolate_freepages_range(&cc, outer_start, end);
6228 /* Free head and tail (if any) */
6229 if (start != outer_start)
6230 free_contig_range(outer_start, start - outer_start);
6231 if (end != outer_end)
6232 free_contig_range(end, outer_end - end);
6235 undo_isolate_page_range(pfn_max_align_down(start),
6236 pfn_max_align_up(end), migratetype);
6237 #ifdef CONFIG_CMA_RMQUEUE
6238 cc.zone->cma_alloc = 0;
6243 void free_contig_range(unsigned long pfn, unsigned nr_pages)
6245 unsigned int count = 0;
6247 for (; nr_pages--; pfn++) {
6248 struct page *page = pfn_to_page(pfn);
6250 count += page_count(page) != 1;
6253 WARN(count != 0, "%d pages are still in use!\n", count);
6257 #ifdef CONFIG_MEMORY_HOTPLUG
6258 static int __meminit __zone_pcp_update(void *data)
6260 struct zone *zone = data;
6262 unsigned long batch = zone_batchsize(zone), flags;
6264 for_each_possible_cpu(cpu) {
6265 struct per_cpu_pageset *pset;
6266 struct per_cpu_pages *pcp;
6268 pset = per_cpu_ptr(zone->pageset, cpu);
6271 local_irq_save(flags);
6273 free_pcppages_bulk(zone, pcp->count, pcp);
6274 drain_zonestat(zone, pset);
6275 setup_pageset(pset, batch);
6276 local_irq_restore(flags);
6281 void __meminit zone_pcp_update(struct zone *zone)
6283 stop_machine(__zone_pcp_update, zone, NULL);
6287 void zone_pcp_reset(struct zone *zone)
6289 unsigned long flags;
6291 struct per_cpu_pageset *pset;
6293 /* avoid races with drain_pages() */
6294 local_irq_save(flags);
6295 if (zone->pageset != &boot_pageset) {
6296 for_each_online_cpu(cpu) {
6297 pset = per_cpu_ptr(zone->pageset, cpu);
6298 drain_zonestat(zone, pset);
6300 free_percpu(zone->pageset);
6301 zone->pageset = &boot_pageset;
6303 local_irq_restore(flags);
6306 #ifdef CONFIG_MEMORY_HOTREMOVE
6308 * All pages in the range must be isolated before calling this.
6311 __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
6317 unsigned long flags;
6318 /* find the first valid pfn */
6319 for (pfn = start_pfn; pfn < end_pfn; pfn++)
6324 zone = page_zone(pfn_to_page(pfn));
6325 spin_lock_irqsave(&zone->lock, flags);
6327 while (pfn < end_pfn) {
6328 if (!pfn_valid(pfn)) {
6332 page = pfn_to_page(pfn);
6334 * The HWPoisoned page may be not in buddy system, and
6335 * page_count() is not 0.
6337 if (unlikely(!PageBuddy(page) && PageHWPoison(page))) {
6339 SetPageReserved(page);
6343 BUG_ON(page_count(page));
6344 BUG_ON(!PageBuddy(page));
6345 order = page_order(page);
6346 #ifdef CONFIG_DEBUG_VM
6347 printk(KERN_INFO "remove from free list %lx %d %lx\n",
6348 pfn, 1 << order, end_pfn);
6350 list_del(&page->lru);
6351 rmv_page_order(page);
6352 zone->free_area[order].nr_free--;
6353 #ifdef CONFIG_HIGHMEM
6354 if (PageHighMem(page))
6355 totalhigh_pages -= 1 << order;
6357 for (i = 0; i < (1 << order); i++)
6358 SetPageReserved((page+i));
6359 pfn += (1 << order);
6361 spin_unlock_irqrestore(&zone->lock, flags);
6365 #ifdef CONFIG_MEMORY_FAILURE
6366 bool is_free_buddy_page(struct page *page)
6368 struct zone *zone = page_zone(page);
6369 unsigned long pfn = page_to_pfn(page);
6370 unsigned long flags;
6373 spin_lock_irqsave(&zone->lock, flags);
6374 for (order = 0; order < MAX_ORDER; order++) {
6375 struct page *page_head = page - (pfn & ((1 << order) - 1));
6377 if (PageBuddy(page_head) && page_order(page_head) >= order)
6380 spin_unlock_irqrestore(&zone->lock, flags);
6382 return order < MAX_ORDER;
6386 static const struct trace_print_flags pageflag_names[] = {
6387 {1UL << PG_locked, "locked" },
6388 {1UL << PG_error, "error" },
6389 {1UL << PG_referenced, "referenced" },
6390 {1UL << PG_uptodate, "uptodate" },
6391 {1UL << PG_dirty, "dirty" },
6392 {1UL << PG_lru, "lru" },
6393 {1UL << PG_active, "active" },
6394 {1UL << PG_slab, "slab" },
6395 {1UL << PG_owner_priv_1, "owner_priv_1" },
6396 {1UL << PG_arch_1, "arch_1" },
6397 {1UL << PG_reserved, "reserved" },
6398 {1UL << PG_private, "private" },
6399 {1UL << PG_private_2, "private_2" },
6400 {1UL << PG_writeback, "writeback" },
6401 #ifdef CONFIG_PAGEFLAGS_EXTENDED
6402 {1UL << PG_head, "head" },
6403 {1UL << PG_tail, "tail" },
6405 {1UL << PG_compound, "compound" },
6407 {1UL << PG_swapcache, "swapcache" },
6408 {1UL << PG_mappedtodisk, "mappedtodisk" },
6409 {1UL << PG_reclaim, "reclaim" },
6410 {1UL << PG_swapbacked, "swapbacked" },
6411 {1UL << PG_unevictable, "unevictable" },
6413 {1UL << PG_mlocked, "mlocked" },
6415 #ifdef CONFIG_ARCH_USES_PG_UNCACHED
6416 {1UL << PG_uncached, "uncached" },
6418 #ifdef CONFIG_MEMORY_FAILURE
6419 {1UL << PG_hwpoison, "hwpoison" },
6421 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
6422 {1UL << PG_compound_lock, "compound_lock" },
6426 static void dump_page_flags(unsigned long flags)
6428 const char *delim = "";
6432 BUILD_BUG_ON(ARRAY_SIZE(pageflag_names) != __NR_PAGEFLAGS);
6434 printk(KERN_ALERT "page flags: %#lx(", flags);
6436 /* remove zone id */
6437 flags &= (1UL << NR_PAGEFLAGS) - 1;
6439 for (i = 0; i < ARRAY_SIZE(pageflag_names) && flags; i++) {
6441 mask = pageflag_names[i].mask;
6442 if ((flags & mask) != mask)
6446 printk("%s%s", delim, pageflag_names[i].name);
6450 /* check for left over flags */
6452 printk("%s%#lx", delim, flags);
6457 void dump_page(struct page *page)
6460 "page:%p count:%d mapcount:%d mapping:%p index:%#lx\n",
6461 page, atomic_read(&page->_count), page_mapcount(page),
6462 page->mapping, page->index);
6463 dump_page_flags(page->flags);
6464 mem_cgroup_print_bad_page(page);