1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17#include <linux/stddef.h>
18#include <linux/mm.h>
19#include <linux/highmem.h>
20#include <linux/swap.h>
21#include <linux/interrupt.h>
22#include <linux/pagemap.h>
23#include <linux/jiffies.h>
24#include <linux/memblock.h>
25#include <linux/compiler.h>
26#include <linux/kernel.h>
27#include <linux/kasan.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/topology.h>
36#include <linux/sysctl.h>
37#include <linux/cpu.h>
38#include <linux/cpuset.h>
39#include <linux/memory_hotplug.h>
40#include <linux/nodemask.h>
41#include <linux/vmalloc.h>
42#include <linux/vmstat.h>
43#include <linux/mempolicy.h>
44#include <linux/memremap.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_ext.h>
52#include <linux/debugobjects.h>
53#include <linux/kmemleak.h>
54#include <linux/compaction.h>
55#include <trace/events/kmem.h>
56#include <trace/events/oom.h>
57#include <linux/prefetch.h>
58#include <linux/mm_inline.h>
59#include <linux/migrate.h>
60#include <linux/hugetlb.h>
61#include <linux/sched/rt.h>
62#include <linux/sched/mm.h>
63#include <linux/page_owner.h>
64#include <linux/kthread.h>
65#include <linux/memcontrol.h>
66#include <linux/ftrace.h>
67#include <linux/lockdep.h>
68#include <linux/nmi.h>
69#include <linux/psi.h>
70
71#include <asm/sections.h>
72#include <asm/tlbflush.h>
73#include <asm/div64.h>
74#include "internal.h"
75
76
77static DEFINE_MUTEX(pcp_batch_high_lock);
78#define MIN_PERCPU_PAGELIST_FRACTION (8)
79
80#ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
81DEFINE_PER_CPU(int, numa_node);
82EXPORT_PER_CPU_SYMBOL(numa_node);
83#endif
84
85DEFINE_STATIC_KEY_TRUE(vm_numa_stat_key);
86
87#ifdef CONFIG_HAVE_MEMORYLESS_NODES
88
89
90
91
92
93
94DEFINE_PER_CPU(int, _numa_mem_);
95EXPORT_PER_CPU_SYMBOL(_numa_mem_);
96int _node_numa_mem_[MAX_NUMNODES];
97#endif
98
99
100struct pcpu_drain {
101 struct zone *zone;
102 struct work_struct work;
103};
104DEFINE_MUTEX(pcpu_drain_mutex);
105DEFINE_PER_CPU(struct pcpu_drain, pcpu_drain);
106
107#ifdef CONFIG_GCC_PLUGIN_LATENT_ENTROPY
108volatile unsigned long latent_entropy __latent_entropy;
109EXPORT_SYMBOL(latent_entropy);
110#endif
111
112
113
114
115nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
116 [N_POSSIBLE] = NODE_MASK_ALL,
117 [N_ONLINE] = { { [0] = 1UL } },
118#ifndef CONFIG_NUMA
119 [N_NORMAL_MEMORY] = { { [0] = 1UL } },
120#ifdef CONFIG_HIGHMEM
121 [N_HIGH_MEMORY] = { { [0] = 1UL } },
122#endif
123 [N_MEMORY] = { { [0] = 1UL } },
124 [N_CPU] = { { [0] = 1UL } },
125#endif
126};
127EXPORT_SYMBOL(node_states);
128
129atomic_long_t _totalram_pages __read_mostly;
130EXPORT_SYMBOL(_totalram_pages);
131unsigned long totalreserve_pages __read_mostly;
132unsigned long totalcma_pages __read_mostly;
133
134int percpu_pagelist_fraction;
135gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK;
136
137
138
139
140
141
142
143
144
145static inline int get_pcppage_migratetype(struct page *page)
146{
147 return page->index;
148}
149
150static inline void set_pcppage_migratetype(struct page *page, int migratetype)
151{
152 page->index = migratetype;
153}
154
155#ifdef CONFIG_PM_SLEEP
156
157
158
159
160
161
162
163
164
165
166static gfp_t saved_gfp_mask;
167
168void pm_restore_gfp_mask(void)
169{
170 WARN_ON(!mutex_is_locked(&system_transition_mutex));
171 if (saved_gfp_mask) {
172 gfp_allowed_mask = saved_gfp_mask;
173 saved_gfp_mask = 0;
174 }
175}
176
177void pm_restrict_gfp_mask(void)
178{
179 WARN_ON(!mutex_is_locked(&system_transition_mutex));
180 WARN_ON(saved_gfp_mask);
181 saved_gfp_mask = gfp_allowed_mask;
182 gfp_allowed_mask &= ~(__GFP_IO | __GFP_FS);
183}
184
185bool pm_suspended_storage(void)
186{
187 if ((gfp_allowed_mask & (__GFP_IO | __GFP_FS)) == (__GFP_IO | __GFP_FS))
188 return false;
189 return true;
190}
191#endif
192
193#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
194unsigned int pageblock_order __read_mostly;
195#endif
196
197static void __free_pages_ok(struct page *page, unsigned int order);
198
199
200
201
202
203
204
205
206
207
208
209
210int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES] = {
211#ifdef CONFIG_ZONE_DMA
212 [ZONE_DMA] = 256,
213#endif
214#ifdef CONFIG_ZONE_DMA32
215 [ZONE_DMA32] = 256,
216#endif
217 [ZONE_NORMAL] = 32,
218#ifdef CONFIG_HIGHMEM
219 [ZONE_HIGHMEM] = 0,
220#endif
221 [ZONE_MOVABLE] = 0,
222};
223
224EXPORT_SYMBOL(totalram_pages);
225
226static char * const zone_names[MAX_NR_ZONES] = {
227#ifdef CONFIG_ZONE_DMA
228 "DMA",
229#endif
230#ifdef CONFIG_ZONE_DMA32
231 "DMA32",
232#endif
233 "Normal",
234#ifdef CONFIG_HIGHMEM
235 "HighMem",
236#endif
237 "Movable",
238#ifdef CONFIG_ZONE_DEVICE
239 "Device",
240#endif
241};
242
243const char * const migratetype_names[MIGRATE_TYPES] = {
244 "Unmovable",
245 "Movable",
246 "Reclaimable",
247 "HighAtomic",
248#ifdef CONFIG_CMA
249 "CMA",
250#endif
251#ifdef CONFIG_MEMORY_ISOLATION
252 "Isolate",
253#endif
254};
255
256compound_page_dtor * const compound_page_dtors[] = {
257 NULL,
258 free_compound_page,
259#ifdef CONFIG_HUGETLB_PAGE
260 free_huge_page,
261#endif
262#ifdef CONFIG_TRANSPARENT_HUGEPAGE
263 free_transhuge_page,
264#endif
265};
266
267int min_free_kbytes = 1024;
268int user_min_free_kbytes = -1;
269int watermark_boost_factor __read_mostly = 15000;
270int watermark_scale_factor = 10;
271
272static unsigned long nr_kernel_pages __initdata;
273static unsigned long nr_all_pages __initdata;
274static unsigned long dma_reserve __initdata;
275
276#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
277static unsigned long arch_zone_lowest_possible_pfn[MAX_NR_ZONES] __initdata;
278static unsigned long arch_zone_highest_possible_pfn[MAX_NR_ZONES] __initdata;
279static unsigned long required_kernelcore __initdata;
280static unsigned long required_kernelcore_percent __initdata;
281static unsigned long required_movablecore __initdata;
282static unsigned long required_movablecore_percent __initdata;
283static unsigned long zone_movable_pfn[MAX_NUMNODES] __initdata;
284static bool mirrored_kernelcore __meminitdata;
285
286
287int movable_zone;
288EXPORT_SYMBOL(movable_zone);
289#endif
290
291#if MAX_NUMNODES > 1
292int nr_node_ids __read_mostly = MAX_NUMNODES;
293int nr_online_nodes __read_mostly = 1;
294EXPORT_SYMBOL(nr_node_ids);
295EXPORT_SYMBOL(nr_online_nodes);
296#endif
297
298int page_group_by_mobility_disabled __read_mostly;
299
300#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
301
302
303
304
305
306static DEFINE_STATIC_KEY_TRUE(deferred_pages);
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321static inline void kasan_free_nondeferred_pages(struct page *page, int order)
322{
323 if (!static_branch_unlikely(&deferred_pages))
324 kasan_free_pages(page, order);
325}
326
327
328static inline bool __meminit early_page_uninitialised(unsigned long pfn)
329{
330 int nid = early_pfn_to_nid(pfn);
331
332 if (node_online(nid) && pfn >= NODE_DATA(nid)->first_deferred_pfn)
333 return true;
334
335 return false;
336}
337
338
339
340
341
342static bool __meminit
343defer_init(int nid, unsigned long pfn, unsigned long end_pfn)
344{
345 static unsigned long prev_end_pfn, nr_initialised;
346
347
348
349
350
351 if (prev_end_pfn != end_pfn) {
352 prev_end_pfn = end_pfn;
353 nr_initialised = 0;
354 }
355
356
357 if (end_pfn < pgdat_end_pfn(NODE_DATA(nid)))
358 return false;
359
360
361
362
363
364 nr_initialised++;
365 if ((nr_initialised > PAGES_PER_SECTION) &&
366 (pfn & (PAGES_PER_SECTION - 1)) == 0) {
367 NODE_DATA(nid)->first_deferred_pfn = pfn;
368 return true;
369 }
370 return false;
371}
372#else
373#define kasan_free_nondeferred_pages(p, o) kasan_free_pages(p, o)
374
375static inline bool early_page_uninitialised(unsigned long pfn)
376{
377 return false;
378}
379
380static inline bool defer_init(int nid, unsigned long pfn, unsigned long end_pfn)
381{
382 return false;
383}
384#endif
385
386
387static inline unsigned long *get_pageblock_bitmap(struct page *page,
388 unsigned long pfn)
389{
390#ifdef CONFIG_SPARSEMEM
391 return __pfn_to_section(pfn)->pageblock_flags;
392#else
393 return page_zone(page)->pageblock_flags;
394#endif
395}
396
397static inline int pfn_to_bitidx(struct page *page, unsigned long pfn)
398{
399#ifdef CONFIG_SPARSEMEM
400 pfn &= (PAGES_PER_SECTION-1);
401 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
402#else
403 pfn = pfn - round_down(page_zone(page)->zone_start_pfn, pageblock_nr_pages);
404 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
405#endif
406}
407
408
409
410
411
412
413
414
415
416
417static __always_inline unsigned long __get_pfnblock_flags_mask(struct page *page,
418 unsigned long pfn,
419 unsigned long end_bitidx,
420 unsigned long mask)
421{
422 unsigned long *bitmap;
423 unsigned long bitidx, word_bitidx;
424 unsigned long word;
425
426 bitmap = get_pageblock_bitmap(page, pfn);
427 bitidx = pfn_to_bitidx(page, pfn);
428 word_bitidx = bitidx / BITS_PER_LONG;
429 bitidx &= (BITS_PER_LONG-1);
430
431 word = bitmap[word_bitidx];
432 bitidx += end_bitidx;
433 return (word >> (BITS_PER_LONG - bitidx - 1)) & mask;
434}
435
436unsigned long get_pfnblock_flags_mask(struct page *page, unsigned long pfn,
437 unsigned long end_bitidx,
438 unsigned long mask)
439{
440 return __get_pfnblock_flags_mask(page, pfn, end_bitidx, mask);
441}
442
443static __always_inline int get_pfnblock_migratetype(struct page *page, unsigned long pfn)
444{
445 return __get_pfnblock_flags_mask(page, pfn, PB_migrate_end, MIGRATETYPE_MASK);
446}
447
448
449
450
451
452
453
454
455
456void set_pfnblock_flags_mask(struct page *page, unsigned long flags,
457 unsigned long pfn,
458 unsigned long end_bitidx,
459 unsigned long mask)
460{
461 unsigned long *bitmap;
462 unsigned long bitidx, word_bitidx;
463 unsigned long old_word, word;
464
465 BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4);
466 BUILD_BUG_ON(MIGRATE_TYPES > (1 << PB_migratetype_bits));
467
468 bitmap = get_pageblock_bitmap(page, pfn);
469 bitidx = pfn_to_bitidx(page, pfn);
470 word_bitidx = bitidx / BITS_PER_LONG;
471 bitidx &= (BITS_PER_LONG-1);
472
473 VM_BUG_ON_PAGE(!zone_spans_pfn(page_zone(page), pfn), page);
474
475 bitidx += end_bitidx;
476 mask <<= (BITS_PER_LONG - bitidx - 1);
477 flags <<= (BITS_PER_LONG - bitidx - 1);
478
479 word = READ_ONCE(bitmap[word_bitidx]);
480 for (;;) {
481 old_word = cmpxchg(&bitmap[word_bitidx], word, (word & ~mask) | flags);
482 if (word == old_word)
483 break;
484 word = old_word;
485 }
486}
487
488void set_pageblock_migratetype(struct page *page, int migratetype)
489{
490 if (unlikely(page_group_by_mobility_disabled &&
491 migratetype < MIGRATE_PCPTYPES))
492 migratetype = MIGRATE_UNMOVABLE;
493
494 set_pageblock_flags_group(page, (unsigned long)migratetype,
495 PB_migrate, PB_migrate_end);
496}
497
498#ifdef CONFIG_DEBUG_VM
499static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
500{
501 int ret = 0;
502 unsigned seq;
503 unsigned long pfn = page_to_pfn(page);
504 unsigned long sp, start_pfn;
505
506 do {
507 seq = zone_span_seqbegin(zone);
508 start_pfn = zone->zone_start_pfn;
509 sp = zone->spanned_pages;
510 if (!zone_spans_pfn(zone, pfn))
511 ret = 1;
512 } while (zone_span_seqretry(zone, seq));
513
514 if (ret)
515 pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n",
516 pfn, zone_to_nid(zone), zone->name,
517 start_pfn, start_pfn + sp);
518
519 return ret;
520}
521
522static int page_is_consistent(struct zone *zone, struct page *page)
523{
524 if (!pfn_valid_within(page_to_pfn(page)))
525 return 0;
526 if (zone != page_zone(page))
527 return 0;
528
529 return 1;
530}
531
532
533
534static int __maybe_unused bad_range(struct zone *zone, struct page *page)
535{
536 if (page_outside_zone_boundaries(zone, page))
537 return 1;
538 if (!page_is_consistent(zone, page))
539 return 1;
540
541 return 0;
542}
543#else
544static inline int __maybe_unused bad_range(struct zone *zone, struct page *page)
545{
546 return 0;
547}
548#endif
549
550static void bad_page(struct page *page, const char *reason,
551 unsigned long bad_flags)
552{
553 static unsigned long resume;
554 static unsigned long nr_shown;
555 static unsigned long nr_unshown;
556
557
558
559
560
561 if (nr_shown == 60) {
562 if (time_before(jiffies, resume)) {
563 nr_unshown++;
564 goto out;
565 }
566 if (nr_unshown) {
567 pr_alert(
568 "BUG: Bad page state: %lu messages suppressed\n",
569 nr_unshown);
570 nr_unshown = 0;
571 }
572 nr_shown = 0;
573 }
574 if (nr_shown++ == 0)
575 resume = jiffies + 60 * HZ;
576
577 pr_alert("BUG: Bad page state in process %s pfn:%05lx\n",
578 current->comm, page_to_pfn(page));
579 __dump_page(page, reason);
580 bad_flags &= page->flags;
581 if (bad_flags)
582 pr_alert("bad because of flags: %#lx(%pGp)\n",
583 bad_flags, &bad_flags);
584 dump_page_owner(page);
585
586 print_modules();
587 dump_stack();
588out:
589
590 page_mapcount_reset(page);
591 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
592}
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609void free_compound_page(struct page *page)
610{
611 __free_pages_ok(page, compound_order(page));
612}
613
614void prep_compound_page(struct page *page, unsigned int order)
615{
616 int i;
617 int nr_pages = 1 << order;
618
619 set_compound_page_dtor(page, COMPOUND_PAGE_DTOR);
620 set_compound_order(page, order);
621 __SetPageHead(page);
622 for (i = 1; i < nr_pages; i++) {
623 struct page *p = page + i;
624 set_page_count(p, 0);
625 p->mapping = TAIL_MAPPING;
626 set_compound_head(p, page);
627 }
628 atomic_set(compound_mapcount_ptr(page), -1);
629}
630
631#ifdef CONFIG_DEBUG_PAGEALLOC
632unsigned int _debug_guardpage_minorder;
633bool _debug_pagealloc_enabled __read_mostly
634 = IS_ENABLED(CONFIG_DEBUG_PAGEALLOC_ENABLE_DEFAULT);
635EXPORT_SYMBOL(_debug_pagealloc_enabled);
636bool _debug_guardpage_enabled __read_mostly;
637
638static int __init early_debug_pagealloc(char *buf)
639{
640 if (!buf)
641 return -EINVAL;
642 return kstrtobool(buf, &_debug_pagealloc_enabled);
643}
644early_param("debug_pagealloc", early_debug_pagealloc);
645
646static bool need_debug_guardpage(void)
647{
648
649 if (!debug_pagealloc_enabled())
650 return false;
651
652 if (!debug_guardpage_minorder())
653 return false;
654
655 return true;
656}
657
658static void init_debug_guardpage(void)
659{
660 if (!debug_pagealloc_enabled())
661 return;
662
663 if (!debug_guardpage_minorder())
664 return;
665
666 _debug_guardpage_enabled = true;
667}
668
669struct page_ext_operations debug_guardpage_ops = {
670 .need = need_debug_guardpage,
671 .init = init_debug_guardpage,
672};
673
674static int __init debug_guardpage_minorder_setup(char *buf)
675{
676 unsigned long res;
677
678 if (kstrtoul(buf, 10, &res) < 0 || res > MAX_ORDER / 2) {
679 pr_err("Bad debug_guardpage_minorder value\n");
680 return 0;
681 }
682 _debug_guardpage_minorder = res;
683 pr_info("Setting debug_guardpage_minorder to %lu\n", res);
684 return 0;
685}
686early_param("debug_guardpage_minorder", debug_guardpage_minorder_setup);
687
688static inline bool set_page_guard(struct zone *zone, struct page *page,
689 unsigned int order, int migratetype)
690{
691 struct page_ext *page_ext;
692
693 if (!debug_guardpage_enabled())
694 return false;
695
696 if (order >= debug_guardpage_minorder())
697 return false;
698
699 page_ext = lookup_page_ext(page);
700 if (unlikely(!page_ext))
701 return false;
702
703 __set_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags);
704
705 INIT_LIST_HEAD(&page->lru);
706 set_page_private(page, order);
707
708 __mod_zone_freepage_state(zone, -(1 << order), migratetype);
709
710 return true;
711}
712
713static inline void clear_page_guard(struct zone *zone, struct page *page,
714 unsigned int order, int migratetype)
715{
716 struct page_ext *page_ext;
717
718 if (!debug_guardpage_enabled())
719 return;
720
721 page_ext = lookup_page_ext(page);
722 if (unlikely(!page_ext))
723 return;
724
725 __clear_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags);
726
727 set_page_private(page, 0);
728 if (!is_migrate_isolate(migratetype))
729 __mod_zone_freepage_state(zone, (1 << order), migratetype);
730}
731#else
732struct page_ext_operations debug_guardpage_ops;
733static inline bool set_page_guard(struct zone *zone, struct page *page,
734 unsigned int order, int migratetype) { return false; }
735static inline void clear_page_guard(struct zone *zone, struct page *page,
736 unsigned int order, int migratetype) {}
737#endif
738
739static inline void set_page_order(struct page *page, unsigned int order)
740{
741 set_page_private(page, order);
742 __SetPageBuddy(page);
743}
744
745static inline void rmv_page_order(struct page *page)
746{
747 __ClearPageBuddy(page);
748 set_page_private(page, 0);
749}
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764static inline int page_is_buddy(struct page *page, struct page *buddy,
765 unsigned int order)
766{
767 if (page_is_guard(buddy) && page_order(buddy) == order) {
768 if (page_zone_id(page) != page_zone_id(buddy))
769 return 0;
770
771 VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
772
773 return 1;
774 }
775
776 if (PageBuddy(buddy) && page_order(buddy) == order) {
777
778
779
780
781
782 if (page_zone_id(page) != page_zone_id(buddy))
783 return 0;
784
785 VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
786
787 return 1;
788 }
789 return 0;
790}
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816static inline void __free_one_page(struct page *page,
817 unsigned long pfn,
818 struct zone *zone, unsigned int order,
819 int migratetype)
820{
821 unsigned long combined_pfn;
822 unsigned long uninitialized_var(buddy_pfn);
823 struct page *buddy;
824 unsigned int max_order;
825
826 max_order = min_t(unsigned int, MAX_ORDER, pageblock_order + 1);
827
828 VM_BUG_ON(!zone_is_initialized(zone));
829 VM_BUG_ON_PAGE(page->flags & PAGE_FLAGS_CHECK_AT_PREP, page);
830
831 VM_BUG_ON(migratetype == -1);
832 if (likely(!is_migrate_isolate(migratetype)))
833 __mod_zone_freepage_state(zone, 1 << order, migratetype);
834
835 VM_BUG_ON_PAGE(pfn & ((1 << order) - 1), page);
836 VM_BUG_ON_PAGE(bad_range(zone, page), page);
837
838continue_merging:
839 while (order < max_order - 1) {
840 buddy_pfn = __find_buddy_pfn(pfn, order);
841 buddy = page + (buddy_pfn - pfn);
842
843 if (!pfn_valid_within(buddy_pfn))
844 goto done_merging;
845 if (!page_is_buddy(page, buddy, order))
846 goto done_merging;
847
848
849
850
851 if (page_is_guard(buddy)) {
852 clear_page_guard(zone, buddy, order, migratetype);
853 } else {
854 list_del(&buddy->lru);
855 zone->free_area[order].nr_free--;
856 rmv_page_order(buddy);
857 }
858 combined_pfn = buddy_pfn & pfn;
859 page = page + (combined_pfn - pfn);
860 pfn = combined_pfn;
861 order++;
862 }
863 if (max_order < MAX_ORDER) {
864
865
866
867
868
869
870
871
872 if (unlikely(has_isolate_pageblock(zone))) {
873 int buddy_mt;
874
875 buddy_pfn = __find_buddy_pfn(pfn, order);
876 buddy = page + (buddy_pfn - pfn);
877 buddy_mt = get_pageblock_migratetype(buddy);
878
879 if (migratetype != buddy_mt
880 && (is_migrate_isolate(migratetype) ||
881 is_migrate_isolate(buddy_mt)))
882 goto done_merging;
883 }
884 max_order++;
885 goto continue_merging;
886 }
887
888done_merging:
889 set_page_order(page, order);
890
891
892
893
894
895
896
897
898
899 if ((order < MAX_ORDER-2) && pfn_valid_within(buddy_pfn)) {
900 struct page *higher_page, *higher_buddy;
901 combined_pfn = buddy_pfn & pfn;
902 higher_page = page + (combined_pfn - pfn);
903 buddy_pfn = __find_buddy_pfn(combined_pfn, order + 1);
904 higher_buddy = higher_page + (buddy_pfn - combined_pfn);
905 if (pfn_valid_within(buddy_pfn) &&
906 page_is_buddy(higher_page, higher_buddy, order + 1)) {
907 list_add_tail(&page->lru,
908 &zone->free_area[order].free_list[migratetype]);
909 goto out;
910 }
911 }
912
913 list_add(&page->lru, &zone->free_area[order].free_list[migratetype]);
914out:
915 zone->free_area[order].nr_free++;
916}
917
918
919
920
921
922
923static inline bool page_expected_state(struct page *page,
924 unsigned long check_flags)
925{
926 if (unlikely(atomic_read(&page->_mapcount) != -1))
927 return false;
928
929 if (unlikely((unsigned long)page->mapping |
930 page_ref_count(page) |
931#ifdef CONFIG_MEMCG
932 (unsigned long)page->mem_cgroup |
933#endif
934 (page->flags & check_flags)))
935 return false;
936
937 return true;
938}
939
940static void free_pages_check_bad(struct page *page)
941{
942 const char *bad_reason;
943 unsigned long bad_flags;
944
945 bad_reason = NULL;
946 bad_flags = 0;
947
948 if (unlikely(atomic_read(&page->_mapcount) != -1))
949 bad_reason = "nonzero mapcount";
950 if (unlikely(page->mapping != NULL))
951 bad_reason = "non-NULL mapping";
952 if (unlikely(page_ref_count(page) != 0))
953 bad_reason = "nonzero _refcount";
954 if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_FREE)) {
955 bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set";
956 bad_flags = PAGE_FLAGS_CHECK_AT_FREE;
957 }
958#ifdef CONFIG_MEMCG
959 if (unlikely(page->mem_cgroup))
960 bad_reason = "page still charged to cgroup";
961#endif
962 bad_page(page, bad_reason, bad_flags);
963}
964
965static inline int free_pages_check(struct page *page)
966{
967 if (likely(page_expected_state(page, PAGE_FLAGS_CHECK_AT_FREE)))
968 return 0;
969
970
971 free_pages_check_bad(page);
972 return 1;
973}
974
975static int free_tail_pages_check(struct page *head_page, struct page *page)
976{
977 int ret = 1;
978
979
980
981
982
983 BUILD_BUG_ON((unsigned long)LIST_POISON1 & 1);
984
985 if (!IS_ENABLED(CONFIG_DEBUG_VM)) {
986 ret = 0;
987 goto out;
988 }
989 switch (page - head_page) {
990 case 1:
991
992 if (unlikely(compound_mapcount(page))) {
993 bad_page(page, "nonzero compound_mapcount", 0);
994 goto out;
995 }
996 break;
997 case 2:
998
999
1000
1001
1002 break;
1003 default:
1004 if (page->mapping != TAIL_MAPPING) {
1005 bad_page(page, "corrupted mapping in tail page", 0);
1006 goto out;
1007 }
1008 break;
1009 }
1010 if (unlikely(!PageTail(page))) {
1011 bad_page(page, "PageTail not set", 0);
1012 goto out;
1013 }
1014 if (unlikely(compound_head(page) != head_page)) {
1015 bad_page(page, "compound_head not consistent", 0);
1016 goto out;
1017 }
1018 ret = 0;
1019out:
1020 page->mapping = NULL;
1021 clear_compound_head(page);
1022 return ret;
1023}
1024
1025static __always_inline bool free_pages_prepare(struct page *page,
1026 unsigned int order, bool check_free)
1027{
1028 int bad = 0;
1029
1030 VM_BUG_ON_PAGE(PageTail(page), page);
1031
1032 trace_mm_page_free(page, order);
1033
1034
1035
1036
1037
1038 if (unlikely(order)) {
1039 bool compound = PageCompound(page);
1040 int i;
1041
1042 VM_BUG_ON_PAGE(compound && compound_order(page) != order, page);
1043
1044 if (compound)
1045 ClearPageDoubleMap(page);
1046 for (i = 1; i < (1 << order); i++) {
1047 if (compound)
1048 bad += free_tail_pages_check(page, page + i);
1049 if (unlikely(free_pages_check(page + i))) {
1050 bad++;
1051 continue;
1052 }
1053 (page + i)->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
1054 }
1055 }
1056 if (PageMappingFlags(page))
1057 page->mapping = NULL;
1058 if (memcg_kmem_enabled() && PageKmemcg(page))
1059 memcg_kmem_uncharge(page, order);
1060 if (check_free)
1061 bad += free_pages_check(page);
1062 if (bad)
1063 return false;
1064
1065 page_cpupid_reset_last(page);
1066 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
1067 reset_page_owner(page, order);
1068
1069 if (!PageHighMem(page)) {
1070 debug_check_no_locks_freed(page_address(page),
1071 PAGE_SIZE << order);
1072 debug_check_no_obj_freed(page_address(page),
1073 PAGE_SIZE << order);
1074 }
1075 arch_free_page(page, order);
1076 kernel_poison_pages(page, 1 << order, 0);
1077 kernel_map_pages(page, 1 << order, 0);
1078 kasan_free_nondeferred_pages(page, order);
1079
1080 return true;
1081}
1082
1083#ifdef CONFIG_DEBUG_VM
1084static inline bool free_pcp_prepare(struct page *page)
1085{
1086 return free_pages_prepare(page, 0, true);
1087}
1088
1089static inline bool bulkfree_pcp_prepare(struct page *page)
1090{
1091 return false;
1092}
1093#else
1094static bool free_pcp_prepare(struct page *page)
1095{
1096 return free_pages_prepare(page, 0, false);
1097}
1098
1099static bool bulkfree_pcp_prepare(struct page *page)
1100{
1101 return free_pages_check(page);
1102}
1103#endif
1104
1105static inline void prefetch_buddy(struct page *page)
1106{
1107 unsigned long pfn = page_to_pfn(page);
1108 unsigned long buddy_pfn = __find_buddy_pfn(pfn, 0);
1109 struct page *buddy = page + (buddy_pfn - pfn);
1110
1111 prefetch(buddy);
1112}
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125static void free_pcppages_bulk(struct zone *zone, int count,
1126 struct per_cpu_pages *pcp)
1127{
1128 int migratetype = 0;
1129 int batch_free = 0;
1130 int prefetch_nr = 0;
1131 bool isolated_pageblocks;
1132 struct page *page, *tmp;
1133 LIST_HEAD(head);
1134
1135 while (count) {
1136 struct list_head *list;
1137
1138
1139
1140
1141
1142
1143
1144
1145 do {
1146 batch_free++;
1147 if (++migratetype == MIGRATE_PCPTYPES)
1148 migratetype = 0;
1149 list = &pcp->lists[migratetype];
1150 } while (list_empty(list));
1151
1152
1153 if (batch_free == MIGRATE_PCPTYPES)
1154 batch_free = count;
1155
1156 do {
1157 page = list_last_entry(list, struct page, lru);
1158
1159 list_del(&page->lru);
1160 pcp->count--;
1161
1162 if (bulkfree_pcp_prepare(page))
1163 continue;
1164
1165 list_add_tail(&page->lru, &head);
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176 if (prefetch_nr++ < pcp->batch)
1177 prefetch_buddy(page);
1178 } while (--count && --batch_free && !list_empty(list));
1179 }
1180
1181 spin_lock(&zone->lock);
1182 isolated_pageblocks = has_isolate_pageblock(zone);
1183
1184
1185
1186
1187
1188 list_for_each_entry_safe(page, tmp, &head, lru) {
1189 int mt = get_pcppage_migratetype(page);
1190
1191 VM_BUG_ON_PAGE(is_migrate_isolate(mt), page);
1192
1193 if (unlikely(isolated_pageblocks))
1194 mt = get_pageblock_migratetype(page);
1195
1196 __free_one_page(page, page_to_pfn(page), zone, 0, mt);
1197 trace_mm_page_pcpu_drain(page, 0, mt);
1198 }
1199 spin_unlock(&zone->lock);
1200}
1201
1202static void free_one_page(struct zone *zone,
1203 struct page *page, unsigned long pfn,
1204 unsigned int order,
1205 int migratetype)
1206{
1207 spin_lock(&zone->lock);
1208 if (unlikely(has_isolate_pageblock(zone) ||
1209 is_migrate_isolate(migratetype))) {
1210 migratetype = get_pfnblock_migratetype(page, pfn);
1211 }
1212 __free_one_page(page, pfn, zone, order, migratetype);
1213 spin_unlock(&zone->lock);
1214}
1215
1216static void __meminit __init_single_page(struct page *page, unsigned long pfn,
1217 unsigned long zone, int nid)
1218{
1219 mm_zero_struct_page(page);
1220 set_page_links(page, zone, nid, pfn);
1221 init_page_count(page);
1222 page_mapcount_reset(page);
1223 page_cpupid_reset_last(page);
1224 page_kasan_tag_reset(page);
1225
1226 INIT_LIST_HEAD(&page->lru);
1227#ifdef WANT_PAGE_VIRTUAL
1228
1229 if (!is_highmem_idx(zone))
1230 set_page_address(page, __va(pfn << PAGE_SHIFT));
1231#endif
1232}
1233
1234#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1235static void __meminit init_reserved_page(unsigned long pfn)
1236{
1237 pg_data_t *pgdat;
1238 int nid, zid;
1239
1240 if (!early_page_uninitialised(pfn))
1241 return;
1242
1243 nid = early_pfn_to_nid(pfn);
1244 pgdat = NODE_DATA(nid);
1245
1246 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
1247 struct zone *zone = &pgdat->node_zones[zid];
1248
1249 if (pfn >= zone->zone_start_pfn && pfn < zone_end_pfn(zone))
1250 break;
1251 }
1252 __init_single_page(pfn_to_page(pfn), pfn, zid, nid);
1253}
1254#else
1255static inline void init_reserved_page(unsigned long pfn)
1256{
1257}
1258#endif
1259
1260
1261
1262
1263
1264
1265
1266void __meminit reserve_bootmem_region(phys_addr_t start, phys_addr_t end)
1267{
1268 unsigned long start_pfn = PFN_DOWN(start);
1269 unsigned long end_pfn = PFN_UP(end);
1270
1271 for (; start_pfn < end_pfn; start_pfn++) {
1272 if (pfn_valid(start_pfn)) {
1273 struct page *page = pfn_to_page(start_pfn);
1274
1275 init_reserved_page(start_pfn);
1276
1277
1278 INIT_LIST_HEAD(&page->lru);
1279
1280
1281
1282
1283
1284
1285 __SetPageReserved(page);
1286 }
1287 }
1288}
1289
1290static void __free_pages_ok(struct page *page, unsigned int order)
1291{
1292 unsigned long flags;
1293 int migratetype;
1294 unsigned long pfn = page_to_pfn(page);
1295
1296 if (!free_pages_prepare(page, order, true))
1297 return;
1298
1299 migratetype = get_pfnblock_migratetype(page, pfn);
1300 local_irq_save(flags);
1301 __count_vm_events(PGFREE, 1 << order);
1302 free_one_page(page_zone(page), page, pfn, order, migratetype);
1303 local_irq_restore(flags);
1304}
1305
1306static void __init __free_pages_boot_core(struct page *page, unsigned int order)
1307{
1308 unsigned int nr_pages = 1 << order;
1309 struct page *p = page;
1310 unsigned int loop;
1311
1312 prefetchw(p);
1313 for (loop = 0; loop < (nr_pages - 1); loop++, p++) {
1314 prefetchw(p + 1);
1315 __ClearPageReserved(p);
1316 set_page_count(p, 0);
1317 }
1318 __ClearPageReserved(p);
1319 set_page_count(p, 0);
1320
1321 atomic_long_add(nr_pages, &page_zone(page)->managed_pages);
1322 set_page_refcounted(page);
1323 __free_pages(page, order);
1324}
1325
1326#if defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) || \
1327 defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
1328
1329static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata;
1330
1331int __meminit early_pfn_to_nid(unsigned long pfn)
1332{
1333 static DEFINE_SPINLOCK(early_pfn_lock);
1334 int nid;
1335
1336 spin_lock(&early_pfn_lock);
1337 nid = __early_pfn_to_nid(pfn, &early_pfnnid_cache);
1338 if (nid < 0)
1339 nid = first_online_node;
1340 spin_unlock(&early_pfn_lock);
1341
1342 return nid;
1343}
1344#endif
1345
1346#ifdef CONFIG_NODES_SPAN_OTHER_NODES
1347static inline bool __meminit __maybe_unused
1348meminit_pfn_in_nid(unsigned long pfn, int node,
1349 struct mminit_pfnnid_cache *state)
1350{
1351 int nid;
1352
1353 nid = __early_pfn_to_nid(pfn, state);
1354 if (nid >= 0 && nid != node)
1355 return false;
1356 return true;
1357}
1358
1359
1360static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
1361{
1362 return meminit_pfn_in_nid(pfn, node, &early_pfnnid_cache);
1363}
1364
1365#else
1366
1367static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
1368{
1369 return true;
1370}
1371static inline bool __meminit __maybe_unused
1372meminit_pfn_in_nid(unsigned long pfn, int node,
1373 struct mminit_pfnnid_cache *state)
1374{
1375 return true;
1376}
1377#endif
1378
1379
1380void __init memblock_free_pages(struct page *page, unsigned long pfn,
1381 unsigned int order)
1382{
1383 if (early_page_uninitialised(pfn))
1384 return;
1385 return __free_pages_boot_core(page, order);
1386}
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405struct page *__pageblock_pfn_to_page(unsigned long start_pfn,
1406 unsigned long end_pfn, struct zone *zone)
1407{
1408 struct page *start_page;
1409 struct page *end_page;
1410
1411
1412 end_pfn--;
1413
1414 if (!pfn_valid(start_pfn) || !pfn_valid(end_pfn))
1415 return NULL;
1416
1417 start_page = pfn_to_online_page(start_pfn);
1418 if (!start_page)
1419 return NULL;
1420
1421 if (page_zone(start_page) != zone)
1422 return NULL;
1423
1424 end_page = pfn_to_page(end_pfn);
1425
1426
1427 if (page_zone_id(start_page) != page_zone_id(end_page))
1428 return NULL;
1429
1430 return start_page;
1431}
1432
1433void set_zone_contiguous(struct zone *zone)
1434{
1435 unsigned long block_start_pfn = zone->zone_start_pfn;
1436 unsigned long block_end_pfn;
1437
1438 block_end_pfn = ALIGN(block_start_pfn + 1, pageblock_nr_pages);
1439 for (; block_start_pfn < zone_end_pfn(zone);
1440 block_start_pfn = block_end_pfn,
1441 block_end_pfn += pageblock_nr_pages) {
1442
1443 block_end_pfn = min(block_end_pfn, zone_end_pfn(zone));
1444
1445 if (!__pageblock_pfn_to_page(block_start_pfn,
1446 block_end_pfn, zone))
1447 return;
1448 }
1449
1450
1451 zone->contiguous = true;
1452}
1453
1454void clear_zone_contiguous(struct zone *zone)
1455{
1456 zone->contiguous = false;
1457}
1458
1459#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1460static void __init deferred_free_range(unsigned long pfn,
1461 unsigned long nr_pages)
1462{
1463 struct page *page;
1464 unsigned long i;
1465
1466 if (!nr_pages)
1467 return;
1468
1469 page = pfn_to_page(pfn);
1470
1471
1472 if (nr_pages == pageblock_nr_pages &&
1473 (pfn & (pageblock_nr_pages - 1)) == 0) {
1474 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
1475 __free_pages_boot_core(page, pageblock_order);
1476 return;
1477 }
1478
1479 for (i = 0; i < nr_pages; i++, page++, pfn++) {
1480 if ((pfn & (pageblock_nr_pages - 1)) == 0)
1481 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
1482 __free_pages_boot_core(page, 0);
1483 }
1484}
1485
1486
1487static atomic_t pgdat_init_n_undone __initdata;
1488static __initdata DECLARE_COMPLETION(pgdat_init_all_done_comp);
1489
1490static inline void __init pgdat_init_report_one_done(void)
1491{
1492 if (atomic_dec_and_test(&pgdat_init_n_undone))
1493 complete(&pgdat_init_all_done_comp);
1494}
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510static inline bool __init
1511deferred_pfn_valid(int nid, unsigned long pfn,
1512 struct mminit_pfnnid_cache *nid_init_state)
1513{
1514 if (!pfn_valid_within(pfn))
1515 return false;
1516 if (!(pfn & (pageblock_nr_pages - 1)) && !pfn_valid(pfn))
1517 return false;
1518 if (!meminit_pfn_in_nid(pfn, nid, nid_init_state))
1519 return false;
1520 return true;
1521}
1522
1523
1524
1525
1526
1527static void __init deferred_free_pages(int nid, int zid, unsigned long pfn,
1528 unsigned long end_pfn)
1529{
1530 struct mminit_pfnnid_cache nid_init_state = { };
1531 unsigned long nr_pgmask = pageblock_nr_pages - 1;
1532 unsigned long nr_free = 0;
1533
1534 for (; pfn < end_pfn; pfn++) {
1535 if (!deferred_pfn_valid(nid, pfn, &nid_init_state)) {
1536 deferred_free_range(pfn - nr_free, nr_free);
1537 nr_free = 0;
1538 } else if (!(pfn & nr_pgmask)) {
1539 deferred_free_range(pfn - nr_free, nr_free);
1540 nr_free = 1;
1541 touch_nmi_watchdog();
1542 } else {
1543 nr_free++;
1544 }
1545 }
1546
1547 deferred_free_range(pfn - nr_free, nr_free);
1548}
1549
1550
1551
1552
1553
1554
1555static unsigned long __init deferred_init_pages(int nid, int zid,
1556 unsigned long pfn,
1557 unsigned long end_pfn)
1558{
1559 struct mminit_pfnnid_cache nid_init_state = { };
1560 unsigned long nr_pgmask = pageblock_nr_pages - 1;
1561 unsigned long nr_pages = 0;
1562 struct page *page = NULL;
1563
1564 for (; pfn < end_pfn; pfn++) {
1565 if (!deferred_pfn_valid(nid, pfn, &nid_init_state)) {
1566 page = NULL;
1567 continue;
1568 } else if (!page || !(pfn & nr_pgmask)) {
1569 page = pfn_to_page(pfn);
1570 touch_nmi_watchdog();
1571 } else {
1572 page++;
1573 }
1574 __init_single_page(page, pfn, zid, nid);
1575 nr_pages++;
1576 }
1577 return (nr_pages);
1578}
1579
1580
1581static int __init deferred_init_memmap(void *data)
1582{
1583 pg_data_t *pgdat = data;
1584 int nid = pgdat->node_id;
1585 unsigned long start = jiffies;
1586 unsigned long nr_pages = 0;
1587 unsigned long spfn, epfn, first_init_pfn, flags;
1588 phys_addr_t spa, epa;
1589 int zid;
1590 struct zone *zone;
1591 const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id);
1592 u64 i;
1593
1594
1595 if (!cpumask_empty(cpumask))
1596 set_cpus_allowed_ptr(current, cpumask);
1597
1598 pgdat_resize_lock(pgdat, &flags);
1599 first_init_pfn = pgdat->first_deferred_pfn;
1600 if (first_init_pfn == ULONG_MAX) {
1601 pgdat_resize_unlock(pgdat, &flags);
1602 pgdat_init_report_one_done();
1603 return 0;
1604 }
1605
1606
1607 BUG_ON(pgdat->first_deferred_pfn < pgdat->node_start_pfn);
1608 BUG_ON(pgdat->first_deferred_pfn > pgdat_end_pfn(pgdat));
1609 pgdat->first_deferred_pfn = ULONG_MAX;
1610
1611
1612 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
1613 zone = pgdat->node_zones + zid;
1614 if (first_init_pfn < zone_end_pfn(zone))
1615 break;
1616 }
1617 first_init_pfn = max(zone->zone_start_pfn, first_init_pfn);
1618
1619
1620
1621
1622
1623
1624
1625 for_each_free_mem_range(i, nid, MEMBLOCK_NONE, &spa, &epa, NULL) {
1626 spfn = max_t(unsigned long, first_init_pfn, PFN_UP(spa));
1627 epfn = min_t(unsigned long, zone_end_pfn(zone), PFN_DOWN(epa));
1628 nr_pages += deferred_init_pages(nid, zid, spfn, epfn);
1629 }
1630 for_each_free_mem_range(i, nid, MEMBLOCK_NONE, &spa, &epa, NULL) {
1631 spfn = max_t(unsigned long, first_init_pfn, PFN_UP(spa));
1632 epfn = min_t(unsigned long, zone_end_pfn(zone), PFN_DOWN(epa));
1633 deferred_free_pages(nid, zid, spfn, epfn);
1634 }
1635 pgdat_resize_unlock(pgdat, &flags);
1636
1637
1638 WARN_ON(++zid < MAX_NR_ZONES && populated_zone(++zone));
1639
1640 pr_info("node %d initialised, %lu pages in %ums\n", nid, nr_pages,
1641 jiffies_to_msecs(jiffies - start));
1642
1643 pgdat_init_report_one_done();
1644 return 0;
1645}
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662static noinline bool __init
1663deferred_grow_zone(struct zone *zone, unsigned int order)
1664{
1665 int zid = zone_idx(zone);
1666 int nid = zone_to_nid(zone);
1667 pg_data_t *pgdat = NODE_DATA(nid);
1668 unsigned long nr_pages_needed = ALIGN(1 << order, PAGES_PER_SECTION);
1669 unsigned long nr_pages = 0;
1670 unsigned long first_init_pfn, spfn, epfn, t, flags;
1671 unsigned long first_deferred_pfn = pgdat->first_deferred_pfn;
1672 phys_addr_t spa, epa;
1673 u64 i;
1674
1675
1676 if (zone_end_pfn(zone) != pgdat_end_pfn(pgdat))
1677 return false;
1678
1679 pgdat_resize_lock(pgdat, &flags);
1680
1681
1682
1683
1684
1685
1686
1687 if (!static_branch_unlikely(&deferred_pages)) {
1688 pgdat_resize_unlock(pgdat, &flags);
1689 return true;
1690 }
1691
1692
1693
1694
1695
1696 if (first_deferred_pfn != pgdat->first_deferred_pfn) {
1697 pgdat_resize_unlock(pgdat, &flags);
1698 return true;
1699 }
1700
1701 first_init_pfn = max(zone->zone_start_pfn, first_deferred_pfn);
1702
1703 if (first_init_pfn >= pgdat_end_pfn(pgdat)) {
1704 pgdat_resize_unlock(pgdat, &flags);
1705 return false;
1706 }
1707
1708 for_each_free_mem_range(i, nid, MEMBLOCK_NONE, &spa, &epa, NULL) {
1709 spfn = max_t(unsigned long, first_init_pfn, PFN_UP(spa));
1710 epfn = min_t(unsigned long, zone_end_pfn(zone), PFN_DOWN(epa));
1711
1712 while (spfn < epfn && nr_pages < nr_pages_needed) {
1713 t = ALIGN(spfn + PAGES_PER_SECTION, PAGES_PER_SECTION);
1714 first_deferred_pfn = min(t, epfn);
1715 nr_pages += deferred_init_pages(nid, zid, spfn,
1716 first_deferred_pfn);
1717 spfn = first_deferred_pfn;
1718 }
1719
1720 if (nr_pages >= nr_pages_needed)
1721 break;
1722 }
1723
1724 for_each_free_mem_range(i, nid, MEMBLOCK_NONE, &spa, &epa, NULL) {
1725 spfn = max_t(unsigned long, first_init_pfn, PFN_UP(spa));
1726 epfn = min_t(unsigned long, first_deferred_pfn, PFN_DOWN(epa));
1727 deferred_free_pages(nid, zid, spfn, epfn);
1728
1729 if (first_deferred_pfn == epfn)
1730 break;
1731 }
1732 pgdat->first_deferred_pfn = first_deferred_pfn;
1733 pgdat_resize_unlock(pgdat, &flags);
1734
1735 return nr_pages > 0;
1736}
1737
1738
1739
1740
1741
1742
1743
1744static bool __ref
1745_deferred_grow_zone(struct zone *zone, unsigned int order)
1746{
1747 return deferred_grow_zone(zone, order);
1748}
1749
1750#endif
1751
1752void __init page_alloc_init_late(void)
1753{
1754 struct zone *zone;
1755
1756#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1757 int nid;
1758
1759
1760 atomic_set(&pgdat_init_n_undone, num_node_state(N_MEMORY));
1761 for_each_node_state(nid, N_MEMORY) {
1762 kthread_run(deferred_init_memmap, NODE_DATA(nid), "pgdatinit%d", nid);
1763 }
1764
1765
1766 wait_for_completion(&pgdat_init_all_done_comp);
1767
1768
1769
1770
1771
1772 static_branch_disable(&deferred_pages);
1773
1774
1775 files_maxfiles_init();
1776#endif
1777#ifdef CONFIG_ARCH_DISCARD_MEMBLOCK
1778
1779 memblock_discard();
1780#endif
1781
1782 for_each_populated_zone(zone)
1783 set_zone_contiguous(zone);
1784}
1785
1786#ifdef CONFIG_CMA
1787
1788void __init init_cma_reserved_pageblock(struct page *page)
1789{
1790 unsigned i = pageblock_nr_pages;
1791 struct page *p = page;
1792
1793 do {
1794 __ClearPageReserved(p);
1795 set_page_count(p, 0);
1796 } while (++p, --i);
1797
1798 set_pageblock_migratetype(page, MIGRATE_CMA);
1799
1800 if (pageblock_order >= MAX_ORDER) {
1801 i = pageblock_nr_pages;
1802 p = page;
1803 do {
1804 set_page_refcounted(p);
1805 __free_pages(p, MAX_ORDER - 1);
1806 p += MAX_ORDER_NR_PAGES;
1807 } while (i -= MAX_ORDER_NR_PAGES);
1808 } else {
1809 set_page_refcounted(page);
1810 __free_pages(page, pageblock_order);
1811 }
1812
1813 adjust_managed_page_count(page, pageblock_nr_pages);
1814}
1815#endif
1816
1817
1818
1819
1820
1821
1822
1823
1824
1825
1826
1827
1828
1829
1830
1831static inline void expand(struct zone *zone, struct page *page,
1832 int low, int high, struct free_area *area,
1833 int migratetype)
1834{
1835 unsigned long size = 1 << high;
1836
1837 while (high > low) {
1838 area--;
1839 high--;
1840 size >>= 1;
1841 VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]);
1842
1843
1844
1845
1846
1847
1848
1849 if (set_page_guard(zone, &page[size], high, migratetype))
1850 continue;
1851
1852 list_add(&page[size].lru, &area->free_list[migratetype]);
1853 area->nr_free++;
1854 set_page_order(&page[size], high);
1855 }
1856}
1857
1858static void check_new_page_bad(struct page *page)
1859{
1860 const char *bad_reason = NULL;
1861 unsigned long bad_flags = 0;
1862
1863 if (unlikely(atomic_read(&page->_mapcount) != -1))
1864 bad_reason = "nonzero mapcount";
1865 if (unlikely(page->mapping != NULL))
1866 bad_reason = "non-NULL mapping";
1867 if (unlikely(page_ref_count(page) != 0))
1868 bad_reason = "nonzero _count";
1869 if (unlikely(page->flags & __PG_HWPOISON)) {
1870 bad_reason = "HWPoisoned (hardware-corrupted)";
1871 bad_flags = __PG_HWPOISON;
1872
1873 page_mapcount_reset(page);
1874 return;
1875 }
1876 if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_PREP)) {
1877 bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag set";
1878 bad_flags = PAGE_FLAGS_CHECK_AT_PREP;
1879 }
1880#ifdef CONFIG_MEMCG
1881 if (unlikely(page->mem_cgroup))
1882 bad_reason = "page still charged to cgroup";
1883#endif
1884 bad_page(page, bad_reason, bad_flags);
1885}
1886
1887
1888
1889
1890static inline int check_new_page(struct page *page)
1891{
1892 if (likely(page_expected_state(page,
1893 PAGE_FLAGS_CHECK_AT_PREP|__PG_HWPOISON)))
1894 return 0;
1895
1896 check_new_page_bad(page);
1897 return 1;
1898}
1899
1900static inline bool free_pages_prezeroed(void)
1901{
1902 return IS_ENABLED(CONFIG_PAGE_POISONING_ZERO) &&
1903 page_poisoning_enabled();
1904}
1905
1906#ifdef CONFIG_DEBUG_VM
1907static bool check_pcp_refill(struct page *page)
1908{
1909 return false;
1910}
1911
1912static bool check_new_pcp(struct page *page)
1913{
1914 return check_new_page(page);
1915}
1916#else
1917static bool check_pcp_refill(struct page *page)
1918{
1919 return check_new_page(page);
1920}
1921static bool check_new_pcp(struct page *page)
1922{
1923 return false;
1924}
1925#endif
1926
1927static bool check_new_pages(struct page *page, unsigned int order)
1928{
1929 int i;
1930 for (i = 0; i < (1 << order); i++) {
1931 struct page *p = page + i;
1932
1933 if (unlikely(check_new_page(p)))
1934 return true;
1935 }
1936
1937 return false;
1938}
1939
1940inline void post_alloc_hook(struct page *page, unsigned int order,
1941 gfp_t gfp_flags)
1942{
1943 set_page_private(page, 0);
1944 set_page_refcounted(page);
1945
1946 arch_alloc_page(page, order);
1947 kernel_map_pages(page, 1 << order, 1);
1948 kernel_poison_pages(page, 1 << order, 1);
1949 kasan_alloc_pages(page, order);
1950 set_page_owner(page, order, gfp_flags);
1951}
1952
1953static void prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags,
1954 unsigned int alloc_flags)
1955{
1956 int i;
1957
1958 post_alloc_hook(page, order, gfp_flags);
1959
1960 if (!free_pages_prezeroed() && (gfp_flags & __GFP_ZERO))
1961 for (i = 0; i < (1 << order); i++)
1962 clear_highpage(page + i);
1963
1964 if (order && (gfp_flags & __GFP_COMP))
1965 prep_compound_page(page, order);
1966
1967
1968
1969
1970
1971
1972
1973 if (alloc_flags & ALLOC_NO_WATERMARKS)
1974 set_page_pfmemalloc(page);
1975 else
1976 clear_page_pfmemalloc(page);
1977}
1978
1979
1980
1981
1982
1983static __always_inline
1984struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
1985 int migratetype)
1986{
1987 unsigned int current_order;
1988 struct free_area *area;
1989 struct page *page;
1990
1991
1992 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
1993 area = &(zone->free_area[current_order]);
1994 page = list_first_entry_or_null(&area->free_list[migratetype],
1995 struct page, lru);
1996 if (!page)
1997 continue;
1998 list_del(&page->lru);
1999 rmv_page_order(page);
2000 area->nr_free--;
2001 expand(zone, page, order, current_order, area, migratetype);
2002 set_pcppage_migratetype(page, migratetype);
2003 return page;
2004 }
2005
2006 return NULL;
2007}
2008
2009
2010
2011
2012
2013
2014static int fallbacks[MIGRATE_TYPES][4] = {
2015 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_TYPES },
2016 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_TYPES },
2017 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_TYPES },
2018#ifdef CONFIG_CMA
2019 [MIGRATE_CMA] = { MIGRATE_TYPES },
2020#endif
2021#ifdef CONFIG_MEMORY_ISOLATION
2022 [MIGRATE_ISOLATE] = { MIGRATE_TYPES },
2023#endif
2024};
2025
2026#ifdef CONFIG_CMA
2027static __always_inline struct page *__rmqueue_cma_fallback(struct zone *zone,
2028 unsigned int order)
2029{
2030 return __rmqueue_smallest(zone, order, MIGRATE_CMA);
2031}
2032#else
2033static inline struct page *__rmqueue_cma_fallback(struct zone *zone,
2034 unsigned int order) { return NULL; }
2035#endif
2036
2037
2038
2039
2040
2041
2042static int move_freepages(struct zone *zone,
2043 struct page *start_page, struct page *end_page,
2044 int migratetype, int *num_movable)
2045{
2046 struct page *page;
2047 unsigned int order;
2048 int pages_moved = 0;
2049
2050#ifndef CONFIG_HOLES_IN_ZONE
2051
2052
2053
2054
2055
2056
2057
2058 VM_BUG_ON(pfn_valid(page_to_pfn(start_page)) &&
2059 pfn_valid(page_to_pfn(end_page)) &&
2060 page_zone(start_page) != page_zone(end_page));
2061#endif
2062 for (page = start_page; page <= end_page;) {
2063 if (!pfn_valid_within(page_to_pfn(page))) {
2064 page++;
2065 continue;
2066 }
2067
2068
2069 VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page);
2070
2071 if (!PageBuddy(page)) {
2072
2073
2074
2075
2076
2077 if (num_movable &&
2078 (PageLRU(page) || __PageMovable(page)))
2079 (*num_movable)++;
2080
2081 page++;
2082 continue;
2083 }
2084
2085 order = page_order(page);
2086 list_move(&page->lru,
2087 &zone->free_area[order].free_list[migratetype]);
2088 page += 1 << order;
2089 pages_moved += 1 << order;
2090 }
2091
2092 return pages_moved;
2093}
2094
2095int move_freepages_block(struct zone *zone, struct page *page,
2096 int migratetype, int *num_movable)
2097{
2098 unsigned long start_pfn, end_pfn;
2099 struct page *start_page, *end_page;
2100
2101 if (num_movable)
2102 *num_movable = 0;
2103
2104 start_pfn = page_to_pfn(page);
2105 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
2106 start_page = pfn_to_page(start_pfn);
2107 end_page = start_page + pageblock_nr_pages - 1;
2108 end_pfn = start_pfn + pageblock_nr_pages - 1;
2109
2110
2111 if (!zone_spans_pfn(zone, start_pfn))
2112 start_page = page;
2113 if (!zone_spans_pfn(zone, end_pfn))
2114 return 0;
2115
2116 return move_freepages(zone, start_page, end_page, migratetype,
2117 num_movable);
2118}
2119
2120static void change_pageblock_range(struct page *pageblock_page,
2121 int start_order, int migratetype)
2122{
2123 int nr_pageblocks = 1 << (start_order - pageblock_order);
2124
2125 while (nr_pageblocks--) {
2126 set_pageblock_migratetype(pageblock_page, migratetype);
2127 pageblock_page += pageblock_nr_pages;
2128 }
2129}
2130
2131
2132
2133
2134
2135
2136
2137
2138
2139
2140
2141
2142
2143static bool can_steal_fallback(unsigned int order, int start_mt)
2144{
2145
2146
2147
2148
2149
2150
2151
2152 if (order >= pageblock_order)
2153 return true;
2154
2155 if (order >= pageblock_order / 2 ||
2156 start_mt == MIGRATE_RECLAIMABLE ||
2157 start_mt == MIGRATE_UNMOVABLE ||
2158 page_group_by_mobility_disabled)
2159 return true;
2160
2161 return false;
2162}
2163
2164static inline void boost_watermark(struct zone *zone)
2165{
2166 unsigned long max_boost;
2167
2168 if (!watermark_boost_factor)
2169 return;
2170
2171 max_boost = mult_frac(zone->_watermark[WMARK_HIGH],
2172 watermark_boost_factor, 10000);
2173
2174
2175
2176
2177
2178
2179
2180
2181
2182 if (!max_boost)
2183 return;
2184
2185 max_boost = max(pageblock_nr_pages, max_boost);
2186
2187 zone->watermark_boost = min(zone->watermark_boost + pageblock_nr_pages,
2188 max_boost);
2189}
2190
2191
2192
2193
2194
2195
2196
2197
2198
2199static void steal_suitable_fallback(struct zone *zone, struct page *page,
2200 unsigned int alloc_flags, int start_type, bool whole_block)
2201{
2202 unsigned int current_order = page_order(page);
2203 struct free_area *area;
2204 int free_pages, movable_pages, alike_pages;
2205 int old_block_type;
2206
2207 old_block_type = get_pageblock_migratetype(page);
2208
2209
2210
2211
2212
2213 if (is_migrate_highatomic(old_block_type))
2214 goto single_page;
2215
2216
2217 if (current_order >= pageblock_order) {
2218 change_pageblock_range(page, current_order, start_type);
2219 goto single_page;
2220 }
2221
2222
2223
2224
2225
2226
2227 boost_watermark(zone);
2228 if (alloc_flags & ALLOC_KSWAPD)
2229 set_bit(ZONE_BOOSTED_WATERMARK, &zone->flags);
2230
2231
2232 if (!whole_block)
2233 goto single_page;
2234
2235 free_pages = move_freepages_block(zone, page, start_type,
2236 &movable_pages);
2237
2238
2239
2240
2241
2242 if (start_type == MIGRATE_MOVABLE) {
2243 alike_pages = movable_pages;
2244 } else {
2245
2246
2247
2248
2249
2250
2251
2252 if (old_block_type == MIGRATE_MOVABLE)
2253 alike_pages = pageblock_nr_pages
2254 - (free_pages + movable_pages);
2255 else
2256 alike_pages = 0;
2257 }
2258
2259
2260 if (!free_pages)
2261 goto single_page;
2262
2263
2264
2265
2266
2267 if (free_pages + alike_pages >= (1 << (pageblock_order-1)) ||
2268 page_group_by_mobility_disabled)
2269 set_pageblock_migratetype(page, start_type);
2270
2271 return;
2272
2273single_page:
2274 area = &zone->free_area[current_order];
2275 list_move(&page->lru, &area->free_list[start_type]);
2276}
2277
2278
2279
2280
2281
2282
2283
2284int find_suitable_fallback(struct free_area *area, unsigned int order,
2285 int migratetype, bool only_stealable, bool *can_steal)
2286{
2287 int i;
2288 int fallback_mt;
2289
2290 if (area->nr_free == 0)
2291 return -1;
2292
2293 *can_steal = false;
2294 for (i = 0;; i++) {
2295 fallback_mt = fallbacks[migratetype][i];
2296 if (fallback_mt == MIGRATE_TYPES)
2297 break;
2298
2299 if (list_empty(&area->free_list[fallback_mt]))
2300 continue;
2301
2302 if (can_steal_fallback(order, migratetype))
2303 *can_steal = true;
2304
2305 if (!only_stealable)
2306 return fallback_mt;
2307
2308 if (*can_steal)
2309 return fallback_mt;
2310 }
2311
2312 return -1;
2313}
2314
2315
2316
2317
2318
2319static void reserve_highatomic_pageblock(struct page *page, struct zone *zone,
2320 unsigned int alloc_order)
2321{
2322 int mt;
2323 unsigned long max_managed, flags;
2324
2325
2326
2327
2328
2329 max_managed = (zone_managed_pages(zone) / 100) + pageblock_nr_pages;
2330 if (zone->nr_reserved_highatomic >= max_managed)
2331 return;
2332
2333 spin_lock_irqsave(&zone->lock, flags);
2334
2335
2336 if (zone->nr_reserved_highatomic >= max_managed)
2337 goto out_unlock;
2338
2339
2340 mt = get_pageblock_migratetype(page);
2341 if (!is_migrate_highatomic(mt) && !is_migrate_isolate(mt)
2342 && !is_migrate_cma(mt)) {
2343 zone->nr_reserved_highatomic += pageblock_nr_pages;
2344 set_pageblock_migratetype(page, MIGRATE_HIGHATOMIC);
2345 move_freepages_block(zone, page, MIGRATE_HIGHATOMIC, NULL);
2346 }
2347
2348out_unlock:
2349 spin_unlock_irqrestore(&zone->lock, flags);
2350}
2351
2352
2353
2354
2355
2356
2357
2358
2359
2360
2361static bool unreserve_highatomic_pageblock(const struct alloc_context *ac,
2362 bool force)
2363{
2364 struct zonelist *zonelist = ac->zonelist;
2365 unsigned long flags;
2366 struct zoneref *z;
2367 struct zone *zone;
2368 struct page *page;
2369 int order;
2370 bool ret;
2371
2372 for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->high_zoneidx,
2373 ac->nodemask) {
2374
2375
2376
2377
2378 if (!force && zone->nr_reserved_highatomic <=
2379 pageblock_nr_pages)
2380 continue;
2381
2382 spin_lock_irqsave(&zone->lock, flags);
2383 for (order = 0; order < MAX_ORDER; order++) {
2384 struct free_area *area = &(zone->free_area[order]);
2385
2386 page = list_first_entry_or_null(
2387 &area->free_list[MIGRATE_HIGHATOMIC],
2388 struct page, lru);
2389 if (!page)
2390 continue;
2391
2392
2393
2394
2395
2396
2397
2398
2399 if (is_migrate_highatomic_page(page)) {
2400
2401
2402
2403
2404
2405
2406
2407 zone->nr_reserved_highatomic -= min(
2408 pageblock_nr_pages,
2409 zone->nr_reserved_highatomic);
2410 }
2411
2412
2413
2414
2415
2416
2417
2418
2419
2420
2421 set_pageblock_migratetype(page, ac->migratetype);
2422 ret = move_freepages_block(zone, page, ac->migratetype,
2423 NULL);
2424 if (ret) {
2425 spin_unlock_irqrestore(&zone->lock, flags);
2426 return ret;
2427 }
2428 }
2429 spin_unlock_irqrestore(&zone->lock, flags);
2430 }
2431
2432 return false;
2433}
2434
2435
2436
2437
2438
2439
2440
2441
2442
2443
2444
2445static __always_inline bool
2446__rmqueue_fallback(struct zone *zone, int order, int start_migratetype,
2447 unsigned int alloc_flags)
2448{
2449 struct free_area *area;
2450 int current_order;
2451 int min_order = order;
2452 struct page *page;
2453 int fallback_mt;
2454 bool can_steal;
2455
2456
2457
2458
2459
2460
2461 if (alloc_flags & ALLOC_NOFRAGMENT)
2462 min_order = pageblock_order;
2463
2464
2465
2466
2467
2468
2469 for (current_order = MAX_ORDER - 1; current_order >= min_order;
2470 --current_order) {
2471 area = &(zone->free_area[current_order]);
2472 fallback_mt = find_suitable_fallback(area, current_order,
2473 start_migratetype, false, &can_steal);
2474 if (fallback_mt == -1)
2475 continue;
2476
2477
2478
2479
2480
2481
2482
2483
2484
2485 if (!can_steal && start_migratetype == MIGRATE_MOVABLE
2486 && current_order > order)
2487 goto find_smallest;
2488
2489 goto do_steal;
2490 }
2491
2492 return false;
2493
2494find_smallest:
2495 for (current_order = order; current_order < MAX_ORDER;
2496 current_order++) {
2497 area = &(zone->free_area[current_order]);
2498 fallback_mt = find_suitable_fallback(area, current_order,
2499 start_migratetype, false, &can_steal);
2500 if (fallback_mt != -1)
2501 break;
2502 }
2503
2504
2505
2506
2507
2508 VM_BUG_ON(current_order == MAX_ORDER);
2509
2510do_steal:
2511 page = list_first_entry(&area->free_list[fallback_mt],
2512 struct page, lru);
2513
2514 steal_suitable_fallback(zone, page, alloc_flags, start_migratetype,
2515 can_steal);
2516
2517 trace_mm_page_alloc_extfrag(page, order, current_order,
2518 start_migratetype, fallback_mt);
2519
2520 return true;
2521
2522}
2523
2524
2525
2526
2527
2528static __always_inline struct page *
2529__rmqueue(struct zone *zone, unsigned int order, int migratetype,
2530 unsigned int alloc_flags)
2531{
2532 struct page *page;
2533
2534retry:
2535 page = __rmqueue_smallest(zone, order, migratetype);
2536 if (unlikely(!page)) {
2537 if (migratetype == MIGRATE_MOVABLE)
2538 page = __rmqueue_cma_fallback(zone, order);
2539
2540 if (!page && __rmqueue_fallback(zone, order, migratetype,
2541 alloc_flags))
2542 goto retry;
2543 }
2544
2545 trace_mm_page_alloc_zone_locked(page, order, migratetype);
2546 return page;
2547}
2548
2549
2550
2551
2552
2553
2554static int rmqueue_bulk(struct zone *zone, unsigned int order,
2555 unsigned long count, struct list_head *list,
2556 int migratetype, unsigned int alloc_flags)
2557{
2558 int i, alloced = 0;
2559
2560 spin_lock(&zone->lock);
2561 for (i = 0; i < count; ++i) {
2562 struct page *page = __rmqueue(zone, order, migratetype,
2563 alloc_flags);
2564 if (unlikely(page == NULL))
2565 break;
2566
2567 if (unlikely(check_pcp_refill(page)))
2568 continue;
2569
2570
2571
2572
2573
2574
2575
2576
2577
2578
2579
2580 list_add_tail(&page->lru, list);
2581 alloced++;
2582 if (is_migrate_cma(get_pcppage_migratetype(page)))
2583 __mod_zone_page_state(zone, NR_FREE_CMA_PAGES,
2584 -(1 << order));
2585 }
2586
2587
2588
2589
2590
2591
2592
2593 __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
2594 spin_unlock(&zone->lock);
2595 return alloced;
2596}
2597
2598#ifdef CONFIG_NUMA
2599
2600
2601
2602
2603
2604
2605
2606
2607void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
2608{
2609 unsigned long flags;
2610 int to_drain, batch;
2611
2612 local_irq_save(flags);
2613 batch = READ_ONCE(pcp->batch);
2614 to_drain = min(pcp->count, batch);
2615 if (to_drain > 0)
2616 free_pcppages_bulk(zone, to_drain, pcp);
2617 local_irq_restore(flags);
2618}
2619#endif
2620
2621
2622
2623
2624
2625
2626
2627
2628static void drain_pages_zone(unsigned int cpu, struct zone *zone)
2629{
2630 unsigned long flags;
2631 struct per_cpu_pageset *pset;
2632 struct per_cpu_pages *pcp;
2633
2634 local_irq_save(flags);
2635 pset = per_cpu_ptr(zone->pageset, cpu);
2636
2637 pcp = &pset->pcp;
2638 if (pcp->count)
2639 free_pcppages_bulk(zone, pcp->count, pcp);
2640 local_irq_restore(flags);
2641}
2642
2643
2644
2645
2646
2647
2648
2649
2650static void drain_pages(unsigned int cpu)
2651{
2652 struct zone *zone;
2653
2654 for_each_populated_zone(zone) {
2655 drain_pages_zone(cpu, zone);
2656 }
2657}
2658
2659
2660
2661
2662
2663
2664
2665void drain_local_pages(struct zone *zone)
2666{
2667 int cpu = smp_processor_id();
2668
2669 if (zone)
2670 drain_pages_zone(cpu, zone);
2671 else
2672 drain_pages(cpu);
2673}
2674
2675static void drain_local_pages_wq(struct work_struct *work)
2676{
2677 struct pcpu_drain *drain;
2678
2679 drain = container_of(work, struct pcpu_drain, work);
2680
2681
2682
2683
2684
2685
2686
2687
2688 preempt_disable();
2689 drain_local_pages(drain->zone);
2690 preempt_enable();
2691}
2692
2693
2694
2695
2696
2697
2698
2699
2700void drain_all_pages(struct zone *zone)
2701{
2702 int cpu;
2703
2704
2705
2706
2707
2708 static cpumask_t cpus_with_pcps;
2709
2710
2711
2712
2713
2714 if (WARN_ON_ONCE(!mm_percpu_wq))
2715 return;
2716
2717
2718
2719
2720
2721
2722 if (unlikely(!mutex_trylock(&pcpu_drain_mutex))) {
2723 if (!zone)
2724 return;
2725 mutex_lock(&pcpu_drain_mutex);
2726 }
2727
2728
2729
2730
2731
2732
2733
2734 for_each_online_cpu(cpu) {
2735 struct per_cpu_pageset *pcp;
2736 struct zone *z;
2737 bool has_pcps = false;
2738
2739 if (zone) {
2740 pcp = per_cpu_ptr(zone->pageset, cpu);
2741 if (pcp->pcp.count)
2742 has_pcps = true;
2743 } else {
2744 for_each_populated_zone(z) {
2745 pcp = per_cpu_ptr(z->pageset, cpu);
2746 if (pcp->pcp.count) {
2747 has_pcps = true;
2748 break;
2749 }
2750 }
2751 }
2752
2753 if (has_pcps)
2754 cpumask_set_cpu(cpu, &cpus_with_pcps);
2755 else
2756 cpumask_clear_cpu(cpu, &cpus_with_pcps);
2757 }
2758
2759 for_each_cpu(cpu, &cpus_with_pcps) {
2760 struct pcpu_drain *drain = per_cpu_ptr(&pcpu_drain, cpu);
2761
2762 drain->zone = zone;
2763 INIT_WORK(&drain->work, drain_local_pages_wq);
2764 queue_work_on(cpu, mm_percpu_wq, &drain->work);
2765 }
2766 for_each_cpu(cpu, &cpus_with_pcps)
2767 flush_work(&per_cpu_ptr(&pcpu_drain, cpu)->work);
2768
2769 mutex_unlock(&pcpu_drain_mutex);
2770}
2771
2772#ifdef CONFIG_HIBERNATION
2773
2774
2775
2776
2777#define WD_PAGE_COUNT (128*1024)
2778
2779void mark_free_pages(struct zone *zone)
2780{
2781 unsigned long pfn, max_zone_pfn, page_count = WD_PAGE_COUNT;
2782 unsigned long flags;
2783 unsigned int order, t;
2784 struct page *page;
2785
2786 if (zone_is_empty(zone))
2787 return;
2788
2789 spin_lock_irqsave(&zone->lock, flags);
2790
2791 max_zone_pfn = zone_end_pfn(zone);
2792 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
2793 if (pfn_valid(pfn)) {
2794 page = pfn_to_page(pfn);
2795
2796 if (!--page_count) {
2797 touch_nmi_watchdog();
2798 page_count = WD_PAGE_COUNT;
2799 }
2800
2801 if (page_zone(page) != zone)
2802 continue;
2803
2804 if (!swsusp_page_is_forbidden(page))
2805 swsusp_unset_page_free(page);
2806 }
2807
2808 for_each_migratetype_order(order, t) {
2809 list_for_each_entry(page,
2810 &zone->free_area[order].free_list[t], lru) {
2811 unsigned long i;
2812
2813 pfn = page_to_pfn(page);
2814 for (i = 0; i < (1UL << order); i++) {
2815 if (!--page_count) {
2816 touch_nmi_watchdog();
2817 page_count = WD_PAGE_COUNT;
2818 }
2819 swsusp_set_page_free(pfn_to_page(pfn + i));
2820 }
2821 }
2822 }
2823 spin_unlock_irqrestore(&zone->lock, flags);
2824}
2825#endif
2826
2827static bool free_unref_page_prepare(struct page *page, unsigned long pfn)
2828{
2829 int migratetype;
2830
2831 if (!free_pcp_prepare(page))
2832 return false;
2833
2834 migratetype = get_pfnblock_migratetype(page, pfn);
2835 set_pcppage_migratetype(page, migratetype);
2836 return true;
2837}
2838
2839static void free_unref_page_commit(struct page *page, unsigned long pfn)
2840{
2841 struct zone *zone = page_zone(page);
2842 struct per_cpu_pages *pcp;
2843 int migratetype;
2844
2845 migratetype = get_pcppage_migratetype(page);
2846 __count_vm_event(PGFREE);
2847
2848
2849
2850
2851
2852
2853
2854
2855 if (migratetype >= MIGRATE_PCPTYPES) {
2856 if (unlikely(is_migrate_isolate(migratetype))) {
2857 free_one_page(zone, page, pfn, 0, migratetype);
2858 return;
2859 }
2860 migratetype = MIGRATE_MOVABLE;
2861 }
2862
2863 pcp = &this_cpu_ptr(zone->pageset)->pcp;
2864 list_add(&page->lru, &pcp->lists[migratetype]);
2865 pcp->count++;
2866 if (pcp->count >= pcp->high) {
2867 unsigned long batch = READ_ONCE(pcp->batch);
2868 free_pcppages_bulk(zone, batch, pcp);
2869 }
2870}
2871
2872
2873
2874
2875void free_unref_page(struct page *page)
2876{
2877 unsigned long flags;
2878 unsigned long pfn = page_to_pfn(page);
2879
2880 if (!free_unref_page_prepare(page, pfn))
2881 return;
2882
2883 local_irq_save(flags);
2884 free_unref_page_commit(page, pfn);
2885 local_irq_restore(flags);
2886}
2887
2888
2889
2890
2891void free_unref_page_list(struct list_head *list)
2892{
2893 struct page *page, *next;
2894 unsigned long flags, pfn;
2895 int batch_count = 0;
2896
2897
2898 list_for_each_entry_safe(page, next, list, lru) {
2899 pfn = page_to_pfn(page);
2900 if (!free_unref_page_prepare(page, pfn))
2901 list_del(&page->lru);
2902 set_page_private(page, pfn);
2903 }
2904
2905 local_irq_save(flags);
2906 list_for_each_entry_safe(page, next, list, lru) {
2907 unsigned long pfn = page_private(page);
2908
2909 set_page_private(page, 0);
2910 trace_mm_page_free_batched(page);
2911 free_unref_page_commit(page, pfn);
2912
2913
2914
2915
2916
2917 if (++batch_count == SWAP_CLUSTER_MAX) {
2918 local_irq_restore(flags);
2919 batch_count = 0;
2920 local_irq_save(flags);
2921 }
2922 }
2923 local_irq_restore(flags);
2924}
2925
2926
2927
2928
2929
2930
2931
2932
2933
2934void split_page(struct page *page, unsigned int order)
2935{
2936 int i;
2937
2938 VM_BUG_ON_PAGE(PageCompound(page), page);
2939 VM_BUG_ON_PAGE(!page_count(page), page);
2940
2941 for (i = 1; i < (1 << order); i++)
2942 set_page_refcounted(page + i);
2943 split_page_owner(page, order);
2944}
2945EXPORT_SYMBOL_GPL(split_page);
2946
2947int __isolate_free_page(struct page *page, unsigned int order)
2948{
2949 unsigned long watermark;
2950 struct zone *zone;
2951 int mt;
2952
2953 BUG_ON(!PageBuddy(page));
2954
2955 zone = page_zone(page);
2956 mt = get_pageblock_migratetype(page);
2957
2958 if (!is_migrate_isolate(mt)) {
2959
2960
2961
2962
2963
2964
2965 watermark = min_wmark_pages(zone) + (1UL << order);
2966 if (!zone_watermark_ok(zone, 0, watermark, 0, ALLOC_CMA))
2967 return 0;
2968
2969 __mod_zone_freepage_state(zone, -(1UL << order), mt);
2970 }
2971
2972
2973 list_del(&page->lru);
2974 zone->free_area[order].nr_free--;
2975 rmv_page_order(page);
2976
2977
2978
2979
2980
2981 if (order >= pageblock_order - 1) {
2982 struct page *endpage = page + (1 << order) - 1;
2983 for (; page < endpage; page += pageblock_nr_pages) {
2984 int mt = get_pageblock_migratetype(page);
2985 if (!is_migrate_isolate(mt) && !is_migrate_cma(mt)
2986 && !is_migrate_highatomic(mt))
2987 set_pageblock_migratetype(page,
2988 MIGRATE_MOVABLE);
2989 }
2990 }
2991
2992
2993 return 1UL << order;
2994}
2995
2996
2997
2998
2999
3000
3001static inline void zone_statistics(struct zone *preferred_zone, struct zone *z)
3002{
3003#ifdef CONFIG_NUMA
3004 enum numa_stat_item local_stat = NUMA_LOCAL;
3005
3006
3007 if (!static_branch_likely(&vm_numa_stat_key))
3008 return;
3009
3010 if (zone_to_nid(z) != numa_node_id())
3011 local_stat = NUMA_OTHER;
3012
3013 if (zone_to_nid(z) == zone_to_nid(preferred_zone))
3014 __inc_numa_state(z, NUMA_HIT);
3015 else {
3016 __inc_numa_state(z, NUMA_MISS);
3017 __inc_numa_state(preferred_zone, NUMA_FOREIGN);
3018 }
3019 __inc_numa_state(z, local_stat);
3020#endif
3021}
3022
3023
3024static struct page *__rmqueue_pcplist(struct zone *zone, int migratetype,
3025 unsigned int alloc_flags,
3026 struct per_cpu_pages *pcp,
3027 struct list_head *list)
3028{
3029 struct page *page;
3030
3031 do {
3032 if (list_empty(list)) {
3033 pcp->count += rmqueue_bulk(zone, 0,
3034 pcp->batch, list,
3035 migratetype, alloc_flags);
3036 if (unlikely(list_empty(list)))
3037 return NULL;
3038 }
3039
3040 page = list_first_entry(list, struct page, lru);
3041 list_del(&page->lru);
3042 pcp->count--;
3043 } while (check_new_pcp(page));
3044
3045 return page;
3046}
3047
3048
3049static struct page *rmqueue_pcplist(struct zone *preferred_zone,
3050 struct zone *zone, unsigned int order,
3051 gfp_t gfp_flags, int migratetype,
3052 unsigned int alloc_flags)
3053{
3054 struct per_cpu_pages *pcp;
3055 struct list_head *list;
3056 struct page *page;
3057 unsigned long flags;
3058
3059 local_irq_save(flags);
3060 pcp = &this_cpu_ptr(zone->pageset)->pcp;
3061 list = &pcp->lists[migratetype];
3062 page = __rmqueue_pcplist(zone, migratetype, alloc_flags, pcp, list);
3063 if (page) {
3064 __count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order);
3065 zone_statistics(preferred_zone, zone);
3066 }
3067 local_irq_restore(flags);
3068 return page;
3069}
3070
3071
3072
3073
3074static inline
3075struct page *rmqueue(struct zone *preferred_zone,
3076 struct zone *zone, unsigned int order,
3077 gfp_t gfp_flags, unsigned int alloc_flags,
3078 int migratetype)
3079{
3080 unsigned long flags;
3081 struct page *page;
3082
3083 if (likely(order == 0)) {
3084 page = rmqueue_pcplist(preferred_zone, zone, order,
3085 gfp_flags, migratetype, alloc_flags);
3086 goto out;
3087 }
3088
3089
3090
3091
3092
3093 WARN_ON_ONCE((gfp_flags & __GFP_NOFAIL) && (order > 1));
3094 spin_lock_irqsave(&zone->lock, flags);
3095
3096 do {
3097 page = NULL;
3098 if (alloc_flags & ALLOC_HARDER) {
3099 page = __rmqueue_smallest(zone, order, MIGRATE_HIGHATOMIC);
3100 if (page)
3101 trace_mm_page_alloc_zone_locked(page, order, migratetype);
3102 }
3103 if (!page)
3104 page = __rmqueue(zone, order, migratetype, alloc_flags);
3105 } while (page && check_new_pages(page, order));
3106 spin_unlock(&zone->lock);
3107 if (!page)
3108 goto failed;
3109 __mod_zone_freepage_state(zone, -(1 << order),
3110 get_pcppage_migratetype(page));
3111
3112 __count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order);
3113 zone_statistics(preferred_zone, zone);
3114 local_irq_restore(flags);
3115
3116out:
3117
3118 if (test_bit(ZONE_BOOSTED_WATERMARK, &zone->flags)) {
3119 clear_bit(ZONE_BOOSTED_WATERMARK, &zone->flags);
3120 wakeup_kswapd(zone, 0, 0, zone_idx(zone));
3121 }
3122
3123 VM_BUG_ON_PAGE(page && bad_range(zone, page), page);
3124 return page;
3125
3126failed:
3127 local_irq_restore(flags);
3128 return NULL;
3129}
3130
3131#ifdef CONFIG_FAIL_PAGE_ALLOC
3132
3133static struct {
3134 struct fault_attr attr;
3135
3136 bool ignore_gfp_highmem;
3137 bool ignore_gfp_reclaim;
3138 u32 min_order;
3139} fail_page_alloc = {
3140 .attr = FAULT_ATTR_INITIALIZER,
3141 .ignore_gfp_reclaim = true,
3142 .ignore_gfp_highmem = true,
3143 .min_order = 1,
3144};
3145
3146static int __init setup_fail_page_alloc(char *str)
3147{
3148 return setup_fault_attr(&fail_page_alloc.attr, str);
3149}
3150__setup("fail_page_alloc=", setup_fail_page_alloc);
3151
3152static bool __should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
3153{
3154 if (order < fail_page_alloc.min_order)
3155 return false;
3156 if (gfp_mask & __GFP_NOFAIL)
3157 return false;
3158 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
3159 return false;
3160 if (fail_page_alloc.ignore_gfp_reclaim &&
3161 (gfp_mask & __GFP_DIRECT_RECLAIM))
3162 return false;
3163
3164 return should_fail(&fail_page_alloc.attr, 1 << order);
3165}
3166
3167#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
3168
3169static int __init fail_page_alloc_debugfs(void)
3170{
3171 umode_t mode = S_IFREG | 0600;
3172 struct dentry *dir;
3173
3174 dir = fault_create_debugfs_attr("fail_page_alloc", NULL,
3175 &fail_page_alloc.attr);
3176 if (IS_ERR(dir))
3177 return PTR_ERR(dir);
3178
3179 if (!debugfs_create_bool("ignore-gfp-wait", mode, dir,
3180 &fail_page_alloc.ignore_gfp_reclaim))
3181 goto fail;
3182 if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir,
3183 &fail_page_alloc.ignore_gfp_highmem))
3184 goto fail;
3185 if (!debugfs_create_u32("min-order", mode, dir,
3186 &fail_page_alloc.min_order))
3187 goto fail;
3188
3189 return 0;
3190fail:
3191 debugfs_remove_recursive(dir);
3192
3193 return -ENOMEM;
3194}
3195
3196late_initcall(fail_page_alloc_debugfs);
3197
3198#endif
3199
3200#else
3201
3202static inline bool __should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
3203{
3204 return false;
3205}
3206
3207#endif
3208
3209static noinline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
3210{
3211 return __should_fail_alloc_page(gfp_mask, order);
3212}
3213ALLOW_ERROR_INJECTION(should_fail_alloc_page, TRUE);
3214
3215
3216
3217
3218
3219
3220
3221bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
3222 int classzone_idx, unsigned int alloc_flags,
3223 long free_pages)
3224{
3225 long min = mark;
3226 int o;
3227 const bool alloc_harder = (alloc_flags & (ALLOC_HARDER|ALLOC_OOM));
3228
3229
3230 free_pages -= (1 << order) - 1;
3231
3232 if (alloc_flags & ALLOC_HIGH)
3233 min -= min / 2;
3234
3235
3236
3237
3238
3239
3240 if (likely(!alloc_harder)) {
3241 free_pages -= z->nr_reserved_highatomic;
3242 } else {
3243
3244
3245
3246
3247
3248
3249 if (alloc_flags & ALLOC_OOM)
3250 min -= min / 2;
3251 else
3252 min -= min / 4;
3253 }
3254
3255
3256#ifdef CONFIG_CMA
3257
3258 if (!(alloc_flags & ALLOC_CMA))
3259 free_pages -= zone_page_state(z, NR_FREE_CMA_PAGES);
3260#endif
3261
3262
3263
3264
3265
3266
3267 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
3268 return false;
3269
3270
3271 if (!order)
3272 return true;
3273
3274
3275 for (o = order; o < MAX_ORDER; o++) {
3276 struct free_area *area = &z->free_area[o];
3277 int mt;
3278
3279 if (!area->nr_free)
3280 continue;
3281
3282 for (mt = 0; mt < MIGRATE_PCPTYPES; mt++) {
3283 if (!list_empty(&area->free_list[mt]))
3284 return true;
3285 }
3286
3287#ifdef CONFIG_CMA
3288 if ((alloc_flags & ALLOC_CMA) &&
3289 !list_empty(&area->free_list[MIGRATE_CMA])) {
3290 return true;
3291 }
3292#endif
3293 if (alloc_harder &&
3294 !list_empty(&area->free_list[MIGRATE_HIGHATOMIC]))
3295 return true;
3296 }
3297 return false;
3298}
3299
3300bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
3301 int classzone_idx, unsigned int alloc_flags)
3302{
3303 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
3304 zone_page_state(z, NR_FREE_PAGES));
3305}
3306
3307static inline bool zone_watermark_fast(struct zone *z, unsigned int order,
3308 unsigned long mark, int classzone_idx, unsigned int alloc_flags)
3309{
3310 long free_pages = zone_page_state(z, NR_FREE_PAGES);
3311 long cma_pages = 0;
3312
3313#ifdef CONFIG_CMA
3314
3315 if (!(alloc_flags & ALLOC_CMA))
3316 cma_pages = zone_page_state(z, NR_FREE_CMA_PAGES);
3317#endif
3318
3319
3320
3321
3322
3323
3324
3325
3326 if (!order && (free_pages - cma_pages) > mark + z->lowmem_reserve[classzone_idx])
3327 return true;
3328
3329 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
3330 free_pages);
3331}
3332
3333bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
3334 unsigned long mark, int classzone_idx)
3335{
3336 long free_pages = zone_page_state(z, NR_FREE_PAGES);
3337
3338 if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark)
3339 free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES);
3340
3341 return __zone_watermark_ok(z, order, mark, classzone_idx, 0,
3342 free_pages);
3343}
3344
3345#ifdef CONFIG_NUMA
3346static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
3347{
3348 return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) <=
3349 RECLAIM_DISTANCE;
3350}
3351#else
3352static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
3353{
3354 return true;
3355}
3356#endif
3357
3358
3359
3360
3361
3362
3363
3364
3365
3366static inline unsigned int
3367alloc_flags_nofragment(struct zone *zone, gfp_t gfp_mask)
3368{
3369 unsigned int alloc_flags = 0;
3370
3371 if (gfp_mask & __GFP_KSWAPD_RECLAIM)
3372 alloc_flags |= ALLOC_KSWAPD;
3373
3374#ifdef CONFIG_ZONE_DMA32
3375 if (zone_idx(zone) != ZONE_NORMAL)
3376 goto out;
3377
3378
3379
3380
3381
3382
3383 BUILD_BUG_ON(ZONE_NORMAL - ZONE_DMA32 != 1);
3384 if (nr_online_nodes > 1 && !populated_zone(--zone))
3385 goto out;
3386
3387out:
3388#endif
3389 return alloc_flags;
3390}
3391
3392
3393
3394
3395
3396static struct page *
3397get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags,
3398 const struct alloc_context *ac)
3399{
3400 struct zoneref *z;
3401 struct zone *zone;
3402 struct pglist_data *last_pgdat_dirty_limit = NULL;
3403 bool no_fallback;
3404
3405retry:
3406
3407
3408
3409
3410 no_fallback = alloc_flags & ALLOC_NOFRAGMENT;
3411 z = ac->preferred_zoneref;
3412 for_next_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx,
3413 ac->nodemask) {
3414 struct page *page;
3415 unsigned long mark;
3416
3417 if (cpusets_enabled() &&
3418 (alloc_flags & ALLOC_CPUSET) &&
3419 !__cpuset_zone_allowed(zone, gfp_mask))
3420 continue;
3421
3422
3423
3424
3425
3426
3427
3428
3429
3430
3431
3432
3433
3434
3435
3436
3437
3438
3439
3440 if (ac->spread_dirty_pages) {
3441 if (last_pgdat_dirty_limit == zone->zone_pgdat)
3442 continue;
3443
3444 if (!node_dirty_ok(zone->zone_pgdat)) {
3445 last_pgdat_dirty_limit = zone->zone_pgdat;
3446 continue;
3447 }
3448 }
3449
3450 if (no_fallback && nr_online_nodes > 1 &&
3451 zone != ac->preferred_zoneref->zone) {
3452 int local_nid;
3453
3454
3455
3456
3457
3458
3459 local_nid = zone_to_nid(ac->preferred_zoneref->zone);
3460 if (zone_to_nid(zone) != local_nid) {
3461 alloc_flags &= ~ALLOC_NOFRAGMENT;
3462 goto retry;
3463 }
3464 }
3465
3466 mark = wmark_pages(zone, alloc_flags & ALLOC_WMARK_MASK);
3467 if (!zone_watermark_fast(zone, order, mark,
3468 ac_classzone_idx(ac), alloc_flags)) {
3469 int ret;
3470
3471#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
3472
3473
3474
3475
3476 if (static_branch_unlikely(&deferred_pages)) {
3477 if (_deferred_grow_zone(zone, order))
3478 goto try_this_zone;
3479 }
3480#endif
3481
3482 BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
3483 if (alloc_flags & ALLOC_NO_WATERMARKS)
3484 goto try_this_zone;
3485
3486 if (node_reclaim_mode == 0 ||
3487 !zone_allows_reclaim(ac->preferred_zoneref->zone, zone))
3488 continue;
3489
3490 ret = node_reclaim(zone->zone_pgdat, gfp_mask, order);
3491 switch (ret) {
3492 case NODE_RECLAIM_NOSCAN:
3493
3494 continue;
3495 case NODE_RECLAIM_FULL:
3496
3497 continue;
3498 default:
3499
3500 if (zone_watermark_ok(zone, order, mark,
3501 ac_classzone_idx(ac), alloc_flags))
3502 goto try_this_zone;
3503
3504 continue;
3505 }
3506 }
3507
3508try_this_zone:
3509 page = rmqueue(ac->preferred_zoneref->zone, zone, order,
3510 gfp_mask, alloc_flags, ac->migratetype);
3511 if (page) {
3512 prep_new_page(page, order, gfp_mask, alloc_flags);
3513
3514
3515
3516
3517
3518 if (unlikely(order && (alloc_flags & ALLOC_HARDER)))
3519 reserve_highatomic_pageblock(page, zone, order);
3520
3521 return page;
3522 } else {
3523#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
3524
3525 if (static_branch_unlikely(&deferred_pages)) {
3526 if (_deferred_grow_zone(zone, order))
3527 goto try_this_zone;
3528 }
3529#endif
3530 }
3531 }
3532
3533
3534
3535
3536
3537 if (no_fallback) {
3538 alloc_flags &= ~ALLOC_NOFRAGMENT;
3539 goto retry;
3540 }
3541
3542 return NULL;
3543}
3544
3545static void warn_alloc_show_mem(gfp_t gfp_mask, nodemask_t *nodemask)
3546{
3547 unsigned int filter = SHOW_MEM_FILTER_NODES;
3548 static DEFINE_RATELIMIT_STATE(show_mem_rs, HZ, 1);
3549
3550 if (!__ratelimit(&show_mem_rs))
3551 return;
3552
3553
3554
3555
3556
3557
3558 if (!(gfp_mask & __GFP_NOMEMALLOC))
3559 if (tsk_is_oom_victim(current) ||
3560 (current->flags & (PF_MEMALLOC | PF_EXITING)))
3561 filter &= ~SHOW_MEM_FILTER_NODES;
3562 if (in_interrupt() || !(gfp_mask & __GFP_DIRECT_RECLAIM))
3563 filter &= ~SHOW_MEM_FILTER_NODES;
3564
3565 show_mem(filter, nodemask);
3566}
3567
3568void warn_alloc(gfp_t gfp_mask, nodemask_t *nodemask, const char *fmt, ...)
3569{
3570 struct va_format vaf;
3571 va_list args;
3572 static DEFINE_RATELIMIT_STATE(nopage_rs, DEFAULT_RATELIMIT_INTERVAL,
3573 DEFAULT_RATELIMIT_BURST);
3574
3575 if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs))
3576 return;
3577
3578 va_start(args, fmt);
3579 vaf.fmt = fmt;
3580 vaf.va = &args;
3581 pr_warn("%s: %pV, mode:%#x(%pGg), nodemask=%*pbl",
3582 current->comm, &vaf, gfp_mask, &gfp_mask,
3583 nodemask_pr_args(nodemask));
3584 va_end(args);
3585
3586 cpuset_print_current_mems_allowed();
3587 pr_cont("\n");
3588 dump_stack();
3589 warn_alloc_show_mem(gfp_mask, nodemask);
3590}
3591
3592static inline struct page *
3593__alloc_pages_cpuset_fallback(gfp_t gfp_mask, unsigned int order,
3594 unsigned int alloc_flags,
3595 const struct alloc_context *ac)
3596{
3597 struct page *page;
3598
3599 page = get_page_from_freelist(gfp_mask, order,
3600 alloc_flags|ALLOC_CPUSET, ac);
3601
3602
3603
3604
3605 if (!page)
3606 page = get_page_from_freelist(gfp_mask, order,
3607 alloc_flags, ac);
3608
3609 return page;
3610}
3611
3612static inline struct page *
3613__alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
3614 const struct alloc_context *ac, unsigned long *did_some_progress)
3615{
3616 struct oom_control oc = {
3617 .zonelist = ac->zonelist,
3618 .nodemask = ac->nodemask,
3619 .memcg = NULL,
3620 .gfp_mask = gfp_mask,
3621 .order = order,
3622 };
3623 struct page *page;
3624
3625 *did_some_progress = 0;
3626
3627
3628
3629
3630
3631 if (!mutex_trylock(&oom_lock)) {
3632 *did_some_progress = 1;
3633 schedule_timeout_uninterruptible(1);
3634 return NULL;
3635 }
3636
3637
3638
3639
3640
3641
3642
3643
3644 page = get_page_from_freelist((gfp_mask | __GFP_HARDWALL) &
3645 ~__GFP_DIRECT_RECLAIM, order,
3646 ALLOC_WMARK_HIGH|ALLOC_CPUSET, ac);
3647 if (page)
3648 goto out;
3649
3650
3651 if (current->flags & PF_DUMPCORE)
3652 goto out;
3653
3654 if (order > PAGE_ALLOC_COSTLY_ORDER)
3655 goto out;
3656
3657
3658
3659
3660
3661
3662 if (gfp_mask & __GFP_RETRY_MAYFAIL)
3663 goto out;
3664
3665 if (ac->high_zoneidx < ZONE_NORMAL)
3666 goto out;
3667 if (pm_suspended_storage())
3668 goto out;
3669
3670
3671
3672
3673
3674
3675
3676
3677
3678
3679
3680 if (gfp_mask & __GFP_THISNODE)
3681 goto out;
3682
3683
3684 if (out_of_memory(&oc) || WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) {
3685 *did_some_progress = 1;
3686
3687
3688
3689
3690
3691 if (gfp_mask & __GFP_NOFAIL)
3692 page = __alloc_pages_cpuset_fallback(gfp_mask, order,
3693 ALLOC_NO_WATERMARKS, ac);
3694 }
3695out:
3696 mutex_unlock(&oom_lock);
3697 return page;
3698}
3699
3700
3701
3702
3703
3704#define MAX_COMPACT_RETRIES 16
3705
3706#ifdef CONFIG_COMPACTION
3707
3708static struct page *
3709__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
3710 unsigned int alloc_flags, const struct alloc_context *ac,
3711 enum compact_priority prio, enum compact_result *compact_result)
3712{
3713 struct page *page;
3714 unsigned long pflags;
3715 unsigned int noreclaim_flag;
3716
3717 if (!order)
3718 return NULL;
3719
3720 psi_memstall_enter(&pflags);
3721 noreclaim_flag = memalloc_noreclaim_save();
3722
3723 *compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac,
3724 prio);
3725
3726 memalloc_noreclaim_restore(noreclaim_flag);
3727 psi_memstall_leave(&pflags);
3728
3729 if (*compact_result <= COMPACT_INACTIVE)
3730 return NULL;
3731
3732
3733
3734
3735
3736 count_vm_event(COMPACTSTALL);
3737
3738 page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
3739
3740 if (page) {
3741 struct zone *zone = page_zone(page);
3742
3743 zone->compact_blockskip_flush = false;
3744 compaction_defer_reset(zone, order, true);
3745 count_vm_event(COMPACTSUCCESS);
3746 return page;
3747 }
3748
3749
3750
3751
3752
3753 count_vm_event(COMPACTFAIL);
3754
3755 cond_resched();
3756
3757 return NULL;
3758}
3759
3760static inline bool
3761should_compact_retry(struct alloc_context *ac, int order, int alloc_flags,
3762 enum compact_result compact_result,
3763 enum compact_priority *compact_priority,
3764 int *compaction_retries)
3765{
3766 int max_retries = MAX_COMPACT_RETRIES;
3767 int min_priority;
3768 bool ret = false;
3769 int retries = *compaction_retries;
3770 enum compact_priority priority = *compact_priority;
3771
3772 if (!order)
3773 return false;
3774
3775 if (compaction_made_progress(compact_result))
3776 (*compaction_retries)++;
3777
3778
3779
3780
3781
3782
3783 if (compaction_failed(compact_result))
3784 goto check_priority;
3785
3786
3787
3788
3789
3790
3791
3792 if (compaction_withdrawn(compact_result)) {
3793 ret = compaction_zonelist_suitable(ac, order, alloc_flags);
3794 goto out;
3795 }
3796
3797
3798
3799
3800
3801
3802
3803
3804
3805 if (order > PAGE_ALLOC_COSTLY_ORDER)
3806 max_retries /= 4;
3807 if (*compaction_retries <= max_retries) {
3808 ret = true;
3809 goto out;
3810 }
3811
3812
3813
3814
3815
3816check_priority:
3817 min_priority = (order > PAGE_ALLOC_COSTLY_ORDER) ?
3818 MIN_COMPACT_COSTLY_PRIORITY : MIN_COMPACT_PRIORITY;
3819
3820 if (*compact_priority > min_priority) {
3821 (*compact_priority)--;
3822 *compaction_retries = 0;
3823 ret = true;
3824 }
3825out:
3826 trace_compact_retry(order, priority, compact_result, retries, max_retries, ret);
3827 return ret;
3828}
3829#else
3830static inline struct page *
3831__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
3832 unsigned int alloc_flags, const struct alloc_context *ac,
3833 enum compact_priority prio, enum compact_result *compact_result)
3834{
3835 *compact_result = COMPACT_SKIPPED;
3836 return NULL;
3837}
3838
3839static inline bool
3840should_compact_retry(struct alloc_context *ac, unsigned int order, int alloc_flags,
3841 enum compact_result compact_result,
3842 enum compact_priority *compact_priority,
3843 int *compaction_retries)
3844{
3845 struct zone *zone;
3846 struct zoneref *z;
3847
3848 if (!order || order > PAGE_ALLOC_COSTLY_ORDER)
3849 return false;
3850
3851
3852
3853
3854
3855
3856
3857 for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx,
3858 ac->nodemask) {
3859 if (zone_watermark_ok(zone, 0, min_wmark_pages(zone),
3860 ac_classzone_idx(ac), alloc_flags))
3861 return true;
3862 }
3863 return false;
3864}
3865#endif
3866
3867#ifdef CONFIG_LOCKDEP
3868static struct lockdep_map __fs_reclaim_map =
3869 STATIC_LOCKDEP_MAP_INIT("fs_reclaim", &__fs_reclaim_map);
3870
3871static bool __need_fs_reclaim(gfp_t gfp_mask)
3872{
3873 gfp_mask = current_gfp_context(gfp_mask);
3874
3875
3876 if (!(gfp_mask & __GFP_DIRECT_RECLAIM))
3877 return false;
3878
3879
3880 if (current->flags & PF_MEMALLOC)
3881 return false;
3882
3883
3884 if (!(gfp_mask & __GFP_FS))
3885 return false;
3886
3887 if (gfp_mask & __GFP_NOLOCKDEP)
3888 return false;
3889
3890 return true;
3891}
3892
3893void __fs_reclaim_acquire(void)
3894{
3895 lock_map_acquire(&__fs_reclaim_map);
3896}
3897
3898void __fs_reclaim_release(void)
3899{
3900 lock_map_release(&__fs_reclaim_map);
3901}
3902
3903void fs_reclaim_acquire(gfp_t gfp_mask)
3904{
3905 if (__need_fs_reclaim(gfp_mask))
3906 __fs_reclaim_acquire();
3907}
3908EXPORT_SYMBOL_GPL(fs_reclaim_acquire);
3909
3910void fs_reclaim_release(gfp_t gfp_mask)
3911{
3912 if (__need_fs_reclaim(gfp_mask))
3913 __fs_reclaim_release();
3914}
3915EXPORT_SYMBOL_GPL(fs_reclaim_release);
3916#endif
3917
3918
3919static int
3920__perform_reclaim(gfp_t gfp_mask, unsigned int order,
3921 const struct alloc_context *ac)
3922{
3923 struct reclaim_state reclaim_state;
3924 int progress;
3925 unsigned int noreclaim_flag;
3926 unsigned long pflags;
3927
3928 cond_resched();
3929
3930
3931 cpuset_memory_pressure_bump();
3932 psi_memstall_enter(&pflags);
3933 fs_reclaim_acquire(gfp_mask);
3934 noreclaim_flag = memalloc_noreclaim_save();
3935 reclaim_state.reclaimed_slab = 0;
3936 current->reclaim_state = &reclaim_state;
3937
3938 progress = try_to_free_pages(ac->zonelist, order, gfp_mask,
3939 ac->nodemask);
3940
3941 current->reclaim_state = NULL;
3942 memalloc_noreclaim_restore(noreclaim_flag);
3943 fs_reclaim_release(gfp_mask);
3944 psi_memstall_leave(&pflags);
3945
3946 cond_resched();
3947
3948 return progress;
3949}
3950
3951
3952static inline struct page *
3953__alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
3954 unsigned int alloc_flags, const struct alloc_context *ac,
3955 unsigned long *did_some_progress)
3956{
3957 struct page *page = NULL;
3958 bool drained = false;
3959
3960 *did_some_progress = __perform_reclaim(gfp_mask, order, ac);
3961 if (unlikely(!(*did_some_progress)))
3962 return NULL;
3963
3964retry:
3965 page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
3966
3967
3968
3969
3970
3971
3972 if (!page && !drained) {
3973 unreserve_highatomic_pageblock(ac, false);
3974 drain_all_pages(NULL);
3975 drained = true;
3976 goto retry;
3977 }
3978
3979 return page;
3980}
3981
3982static void wake_all_kswapds(unsigned int order, gfp_t gfp_mask,
3983 const struct alloc_context *ac)
3984{
3985 struct zoneref *z;
3986 struct zone *zone;
3987 pg_data_t *last_pgdat = NULL;
3988 enum zone_type high_zoneidx = ac->high_zoneidx;
3989
3990 for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, high_zoneidx,
3991 ac->nodemask) {
3992 if (last_pgdat != zone->zone_pgdat)
3993 wakeup_kswapd(zone, gfp_mask, order, high_zoneidx);
3994 last_pgdat = zone->zone_pgdat;
3995 }
3996}
3997
3998static inline unsigned int
3999gfp_to_alloc_flags(gfp_t gfp_mask)
4000{
4001 unsigned int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
4002
4003
4004 BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
4005
4006
4007
4008
4009
4010
4011
4012 alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH);
4013
4014 if (gfp_mask & __GFP_ATOMIC) {
4015
4016
4017
4018
4019 if (!(gfp_mask & __GFP_NOMEMALLOC))
4020 alloc_flags |= ALLOC_HARDER;
4021
4022
4023
4024
4025 alloc_flags &= ~ALLOC_CPUSET;
4026 } else if (unlikely(rt_task(current)) && !in_interrupt())
4027 alloc_flags |= ALLOC_HARDER;
4028
4029 if (gfp_mask & __GFP_KSWAPD_RECLAIM)
4030 alloc_flags |= ALLOC_KSWAPD;
4031
4032#ifdef CONFIG_CMA
4033 if (gfpflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE)
4034 alloc_flags |= ALLOC_CMA;
4035#endif
4036 return alloc_flags;
4037}
4038
4039static bool oom_reserves_allowed(struct task_struct *tsk)
4040{
4041 if (!tsk_is_oom_victim(tsk))
4042 return false;
4043
4044
4045
4046
4047
4048 if (!IS_ENABLED(CONFIG_MMU) && !test_thread_flag(TIF_MEMDIE))
4049 return false;
4050
4051 return true;
4052}
4053
4054
4055
4056
4057
4058static inline int __gfp_pfmemalloc_flags(gfp_t gfp_mask)
4059{
4060 if (unlikely(gfp_mask & __GFP_NOMEMALLOC))
4061 return 0;
4062 if (gfp_mask & __GFP_MEMALLOC)
4063 return ALLOC_NO_WATERMARKS;
4064 if (in_serving_softirq() && (current->flags & PF_MEMALLOC))
4065 return ALLOC_NO_WATERMARKS;
4066 if (!in_interrupt()) {
4067 if (current->flags & PF_MEMALLOC)
4068 return ALLOC_NO_WATERMARKS;
4069 else if (oom_reserves_allowed(current))
4070 return ALLOC_OOM;
4071 }
4072
4073 return 0;
4074}
4075
4076bool gfp_pfmemalloc_allowed(gfp_t gfp_mask)
4077{
4078 return !!__gfp_pfmemalloc_flags(gfp_mask);
4079}
4080
4081
4082
4083
4084
4085
4086
4087
4088
4089
4090
4091static inline bool
4092should_reclaim_retry(gfp_t gfp_mask, unsigned order,
4093 struct alloc_context *ac, int alloc_flags,
4094 bool did_some_progress, int *no_progress_loops)
4095{
4096 struct zone *zone;
4097 struct zoneref *z;
4098 bool ret = false;
4099
4100
4101
4102
4103
4104
4105 if (did_some_progress && order <= PAGE_ALLOC_COSTLY_ORDER)
4106 *no_progress_loops = 0;
4107 else
4108 (*no_progress_loops)++;
4109
4110
4111
4112
4113
4114 if (*no_progress_loops > MAX_RECLAIM_RETRIES) {
4115
4116 return unreserve_highatomic_pageblock(ac, true);
4117 }
4118
4119
4120
4121
4122
4123
4124
4125 for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx,
4126 ac->nodemask) {
4127 unsigned long available;
4128 unsigned long reclaimable;
4129 unsigned long min_wmark = min_wmark_pages(zone);
4130 bool wmark;
4131
4132 available = reclaimable = zone_reclaimable_pages(zone);
4133 available += zone_page_state_snapshot(zone, NR_FREE_PAGES);
4134
4135
4136
4137
4138
4139 wmark = __zone_watermark_ok(zone, order, min_wmark,
4140 ac_classzone_idx(ac), alloc_flags, available);
4141 trace_reclaim_retry_zone(z, order, reclaimable,
4142 available, min_wmark, *no_progress_loops, wmark);
4143 if (wmark) {
4144
4145
4146
4147
4148
4149
4150 if (!did_some_progress) {
4151 unsigned long write_pending;
4152
4153 write_pending = zone_page_state_snapshot(zone,
4154 NR_ZONE_WRITE_PENDING);
4155
4156 if (2 * write_pending > reclaimable) {
4157 congestion_wait(BLK_RW_ASYNC, HZ/10);
4158 return true;
4159 }
4160 }
4161
4162 ret = true;
4163 goto out;
4164 }
4165 }
4166
4167out:
4168
4169
4170
4171
4172
4173
4174
4175 if (current->flags & PF_WQ_WORKER)
4176 schedule_timeout_uninterruptible(1);
4177 else
4178 cond_resched();
4179 return ret;
4180}
4181
4182static inline bool
4183check_retry_cpuset(int cpuset_mems_cookie, struct alloc_context *ac)
4184{
4185
4186
4187
4188
4189
4190
4191
4192
4193
4194
4195
4196 if (cpusets_enabled() && ac->nodemask &&
4197 !cpuset_nodemask_valid_mems_allowed(ac->nodemask)) {
4198 ac->nodemask = NULL;
4199 return true;
4200 }
4201
4202
4203
4204
4205
4206
4207
4208
4209 if (read_mems_allowed_retry(cpuset_mems_cookie))
4210 return true;
4211
4212 return false;
4213}
4214
4215static inline struct page *
4216__alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
4217 struct alloc_context *ac)
4218{
4219 bool can_direct_reclaim = gfp_mask & __GFP_DIRECT_RECLAIM;
4220 const bool costly_order = order > PAGE_ALLOC_COSTLY_ORDER;
4221 struct page *page = NULL;
4222 unsigned int alloc_flags;
4223 unsigned long did_some_progress;
4224 enum compact_priority compact_priority;
4225 enum compact_result compact_result;
4226 int compaction_retries;
4227 int no_progress_loops;
4228 unsigned int cpuset_mems_cookie;
4229 int reserve_flags;
4230
4231
4232
4233
4234
4235 if (WARN_ON_ONCE((gfp_mask & (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)) ==
4236 (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)))
4237 gfp_mask &= ~__GFP_ATOMIC;
4238
4239retry_cpuset:
4240 compaction_retries = 0;
4241 no_progress_loops = 0;
4242 compact_priority = DEF_COMPACT_PRIORITY;
4243 cpuset_mems_cookie = read_mems_allowed_begin();
4244
4245
4246
4247
4248
4249
4250 alloc_flags = gfp_to_alloc_flags(gfp_mask);
4251
4252
4253
4254
4255
4256
4257
4258 ac->preferred_zoneref = first_zones_zonelist(ac->zonelist,
4259 ac->high_zoneidx, ac->nodemask);
4260 if (!ac->preferred_zoneref->zone)
4261 goto nopage;
4262
4263 if (alloc_flags & ALLOC_KSWAPD)
4264 wake_all_kswapds(order, gfp_mask, ac);
4265
4266
4267
4268
4269
4270 page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
4271 if (page)
4272 goto got_pg;
4273
4274
4275
4276
4277
4278
4279
4280
4281
4282
4283 if (can_direct_reclaim &&
4284 (costly_order ||
4285 (order > 0 && ac->migratetype != MIGRATE_MOVABLE))
4286 && !gfp_pfmemalloc_allowed(gfp_mask)) {
4287 page = __alloc_pages_direct_compact(gfp_mask, order,
4288 alloc_flags, ac,
4289 INIT_COMPACT_PRIORITY,
4290 &compact_result);
4291 if (page)
4292 goto got_pg;
4293
4294
4295
4296
4297
4298 if (costly_order && (gfp_mask & __GFP_NORETRY)) {
4299
4300
4301
4302
4303
4304
4305
4306
4307 if (compact_result == COMPACT_DEFERRED)
4308 goto nopage;
4309
4310
4311
4312
4313
4314
4315 compact_priority = INIT_COMPACT_PRIORITY;
4316 }
4317 }
4318
4319retry:
4320
4321 if (alloc_flags & ALLOC_KSWAPD)
4322 wake_all_kswapds(order, gfp_mask, ac);
4323
4324 reserve_flags = __gfp_pfmemalloc_flags(gfp_mask);
4325 if (reserve_flags)
4326 alloc_flags = reserve_flags;
4327
4328
4329
4330
4331
4332
4333 if (!(alloc_flags & ALLOC_CPUSET) || reserve_flags) {
4334 ac->nodemask = NULL;
4335 ac->preferred_zoneref = first_zones_zonelist(ac->zonelist,
4336 ac->high_zoneidx, ac->nodemask);
4337 }
4338
4339
4340 page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
4341 if (page)
4342 goto got_pg;
4343
4344
4345 if (!can_direct_reclaim)
4346 goto nopage;
4347
4348
4349 if (current->flags & PF_MEMALLOC)
4350 goto nopage;
4351
4352
4353 page = __alloc_pages_direct_reclaim(gfp_mask, order, alloc_flags, ac,
4354 &did_some_progress);
4355 if (page)
4356 goto got_pg;
4357
4358
4359 page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, ac,
4360 compact_priority, &compact_result);
4361 if (page)
4362 goto got_pg;
4363
4364
4365 if (gfp_mask & __GFP_NORETRY)
4366 goto nopage;
4367
4368
4369
4370
4371
4372 if (costly_order && !(gfp_mask & __GFP_RETRY_MAYFAIL))
4373 goto nopage;
4374
4375 if (should_reclaim_retry(gfp_mask, order, ac, alloc_flags,
4376 did_some_progress > 0, &no_progress_loops))
4377 goto retry;
4378
4379
4380
4381
4382
4383
4384
4385 if (did_some_progress > 0 &&
4386 should_compact_retry(ac, order, alloc_flags,
4387 compact_result, &compact_priority,
4388 &compaction_retries))
4389 goto retry;
4390
4391
4392
4393 if (check_retry_cpuset(cpuset_mems_cookie, ac))
4394 goto retry_cpuset;
4395
4396
4397 page = __alloc_pages_may_oom(gfp_mask, order, ac, &did_some_progress);
4398 if (page)
4399 goto got_pg;
4400
4401
4402 if (tsk_is_oom_victim(current) &&
4403 (alloc_flags == ALLOC_OOM ||
4404 (gfp_mask & __GFP_NOMEMALLOC)))
4405 goto nopage;
4406
4407
4408 if (did_some_progress) {
4409 no_progress_loops = 0;
4410 goto retry;
4411 }
4412
4413nopage:
4414
4415 if (check_retry_cpuset(cpuset_mems_cookie, ac))
4416 goto retry_cpuset;
4417
4418
4419
4420
4421
4422 if (gfp_mask & __GFP_NOFAIL) {
4423
4424
4425
4426
4427 if (WARN_ON_ONCE(!can_direct_reclaim))
4428 goto fail;
4429
4430
4431
4432
4433
4434
4435 WARN_ON_ONCE(current->flags & PF_MEMALLOC);
4436
4437
4438
4439
4440
4441
4442
4443 WARN_ON_ONCE(order > PAGE_ALLOC_COSTLY_ORDER);
4444
4445
4446
4447
4448
4449
4450
4451 page = __alloc_pages_cpuset_fallback(gfp_mask, order, ALLOC_HARDER, ac);
4452 if (page)
4453 goto got_pg;
4454
4455 cond_resched();
4456 goto retry;
4457 }
4458fail:
4459 warn_alloc(gfp_mask, ac->nodemask,
4460 "page allocation failure: order:%u", order);
4461got_pg:
4462 return page;
4463}
4464
4465static inline bool prepare_alloc_pages(gfp_t gfp_mask, unsigned int order,
4466 int preferred_nid, nodemask_t *nodemask,
4467 struct alloc_context *ac, gfp_t *alloc_mask,
4468 unsigned int *alloc_flags)
4469{
4470 ac->high_zoneidx = gfp_zone(gfp_mask);
4471 ac->zonelist = node_zonelist(preferred_nid, gfp_mask);
4472 ac->nodemask = nodemask;
4473 ac->migratetype = gfpflags_to_migratetype(gfp_mask);
4474
4475 if (cpusets_enabled()) {
4476 *alloc_mask |= __GFP_HARDWALL;
4477 if (!ac->nodemask)
4478 ac->nodemask = &cpuset_current_mems_allowed;
4479 else
4480 *alloc_flags |= ALLOC_CPUSET;
4481 }
4482
4483 fs_reclaim_acquire(gfp_mask);
4484 fs_reclaim_release(gfp_mask);
4485
4486 might_sleep_if(gfp_mask & __GFP_DIRECT_RECLAIM);
4487
4488 if (should_fail_alloc_page(gfp_mask, order))
4489 return false;
4490
4491 if (IS_ENABLED(CONFIG_CMA) && ac->migratetype == MIGRATE_MOVABLE)
4492 *alloc_flags |= ALLOC_CMA;
4493
4494 return true;
4495}
4496
4497
4498static inline void finalise_ac(gfp_t gfp_mask, struct alloc_context *ac)
4499{
4500
4501 ac->spread_dirty_pages = (gfp_mask & __GFP_WRITE);
4502
4503
4504
4505
4506
4507
4508 ac->preferred_zoneref = first_zones_zonelist(ac->zonelist,
4509 ac->high_zoneidx, ac->nodemask);
4510}
4511
4512
4513
4514
4515struct page *
4516__alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order, int preferred_nid,
4517 nodemask_t *nodemask)
4518{
4519 struct page *page;
4520 unsigned int alloc_flags = ALLOC_WMARK_LOW;
4521 gfp_t alloc_mask;
4522 struct alloc_context ac = { };
4523
4524
4525
4526
4527
4528 if (unlikely(order >= MAX_ORDER)) {
4529 WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
4530 return NULL;
4531 }
4532
4533 gfp_mask &= gfp_allowed_mask;
4534 alloc_mask = gfp_mask;
4535 if (!prepare_alloc_pages(gfp_mask, order, preferred_nid, nodemask, &ac, &alloc_mask, &alloc_flags))
4536 return NULL;
4537
4538 finalise_ac(gfp_mask, &ac);
4539
4540
4541
4542
4543
4544 alloc_flags |= alloc_flags_nofragment(ac.preferred_zoneref->zone, gfp_mask);
4545
4546
4547 page = get_page_from_freelist(alloc_mask, order, alloc_flags, &ac);
4548 if (likely(page))
4549 goto out;
4550
4551
4552
4553
4554
4555
4556
4557 alloc_mask = current_gfp_context(gfp_mask);
4558 ac.spread_dirty_pages = false;
4559
4560
4561
4562
4563
4564 if (unlikely(ac.nodemask != nodemask))
4565 ac.nodemask = nodemask;
4566
4567 page = __alloc_pages_slowpath(alloc_mask, order, &ac);
4568
4569out:
4570 if (memcg_kmem_enabled() && (gfp_mask & __GFP_ACCOUNT) && page &&
4571 unlikely(memcg_kmem_charge(page, gfp_mask, order) != 0)) {
4572 __free_pages(page, order);
4573 page = NULL;
4574 }
4575
4576 trace_mm_page_alloc(page, order, alloc_mask, ac.migratetype);
4577
4578 return page;
4579}
4580EXPORT_SYMBOL(__alloc_pages_nodemask);
4581
4582
4583
4584
4585
4586
4587unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
4588{
4589 struct page *page;
4590
4591 page = alloc_pages(gfp_mask & ~__GFP_HIGHMEM, order);
4592 if (!page)
4593 return 0;
4594 return (unsigned long) page_address(page);
4595}
4596EXPORT_SYMBOL(__get_free_pages);
4597
4598unsigned long get_zeroed_page(gfp_t gfp_mask)
4599{
4600 return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
4601}
4602EXPORT_SYMBOL(get_zeroed_page);
4603
4604static inline void free_the_page(struct page *page, unsigned int order)
4605{
4606 if (order == 0)
4607 free_unref_page(page);
4608 else
4609 __free_pages_ok(page, order);
4610}
4611
4612void __free_pages(struct page *page, unsigned int order)
4613{
4614 if (put_page_testzero(page))
4615 free_the_page(page, order);
4616}
4617EXPORT_SYMBOL(__free_pages);
4618
4619void free_pages(unsigned long addr, unsigned int order)
4620{
4621 if (addr != 0) {
4622 VM_BUG_ON(!virt_addr_valid((void *)addr));
4623 __free_pages(virt_to_page((void *)addr), order);
4624 }
4625}
4626
4627EXPORT_SYMBOL(free_pages);
4628
4629
4630
4631
4632
4633
4634
4635
4636
4637
4638
4639
4640static struct page *__page_frag_cache_refill(struct page_frag_cache *nc,
4641 gfp_t gfp_mask)
4642{
4643 struct page *page = NULL;
4644 gfp_t gfp = gfp_mask;
4645
4646#if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
4647 gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY |
4648 __GFP_NOMEMALLOC;
4649 page = alloc_pages_node(NUMA_NO_NODE, gfp_mask,
4650 PAGE_FRAG_CACHE_MAX_ORDER);
4651 nc->size = page ? PAGE_FRAG_CACHE_MAX_SIZE : PAGE_SIZE;
4652#endif
4653 if (unlikely(!page))
4654 page = alloc_pages_node(NUMA_NO_NODE, gfp, 0);
4655
4656 nc->va = page ? page_address(page) : NULL;
4657
4658 return page;
4659}
4660
4661void __page_frag_cache_drain(struct page *page, unsigned int count)
4662{
4663 VM_BUG_ON_PAGE(page_ref_count(page) == 0, page);
4664
4665 if (page_ref_sub_and_test(page, count))
4666 free_the_page(page, compound_order(page));
4667}
4668EXPORT_SYMBOL(__page_frag_cache_drain);
4669
4670void *page_frag_alloc(struct page_frag_cache *nc,
4671 unsigned int fragsz, gfp_t gfp_mask)
4672{
4673 unsigned int size = PAGE_SIZE;
4674 struct page *page;
4675 int offset;
4676
4677 if (unlikely(!nc->va)) {
4678refill:
4679 page = __page_frag_cache_refill(nc, gfp_mask);
4680 if (!page)
4681 return NULL;
4682
4683#if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
4684
4685 size = nc->size;
4686#endif
4687
4688
4689
4690 page_ref_add(page, PAGE_FRAG_CACHE_MAX_SIZE);
4691
4692
4693 nc->pfmemalloc = page_is_pfmemalloc(page);
4694 nc->pagecnt_bias = PAGE_FRAG_CACHE_MAX_SIZE + 1;
4695 nc->offset = size;
4696 }
4697
4698 offset = nc->offset - fragsz;
4699 if (unlikely(offset < 0)) {
4700 page = virt_to_page(nc->va);
4701
4702 if (!page_ref_sub_and_test(page, nc->pagecnt_bias))
4703 goto refill;
4704
4705#if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
4706
4707 size = nc->size;
4708#endif
4709
4710 set_page_count(page, PAGE_FRAG_CACHE_MAX_SIZE + 1);
4711
4712
4713 nc->pagecnt_bias = PAGE_FRAG_CACHE_MAX_SIZE + 1;
4714 offset = size - fragsz;
4715 }
4716
4717 nc->pagecnt_bias--;
4718 nc->offset = offset;
4719
4720 return nc->va + offset;
4721}
4722EXPORT_SYMBOL(page_frag_alloc);
4723
4724
4725
4726
4727void page_frag_free(void *addr)
4728{
4729 struct page *page = virt_to_head_page(addr);
4730
4731 if (unlikely(put_page_testzero(page)))
4732 free_the_page(page, compound_order(page));
4733}
4734EXPORT_SYMBOL(page_frag_free);
4735
4736static void *make_alloc_exact(unsigned long addr, unsigned int order,
4737 size_t size)
4738{
4739 if (addr) {
4740 unsigned long alloc_end = addr + (PAGE_SIZE << order);
4741 unsigned long used = addr + PAGE_ALIGN(size);
4742
4743 split_page(virt_to_page((void *)addr), order);
4744 while (used < alloc_end) {
4745 free_page(used);
4746 used += PAGE_SIZE;
4747 }
4748 }
4749 return (void *)addr;
4750}
4751
4752
4753
4754
4755
4756
4757
4758
4759
4760
4761
4762
4763
4764
4765void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
4766{
4767 unsigned int order = get_order(size);
4768 unsigned long addr;
4769
4770 addr = __get_free_pages(gfp_mask, order);
4771 return make_alloc_exact(addr, order, size);
4772}
4773EXPORT_SYMBOL(alloc_pages_exact);
4774
4775
4776
4777
4778
4779
4780
4781
4782
4783
4784
4785void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask)
4786{
4787 unsigned int order = get_order(size);
4788 struct page *p = alloc_pages_node(nid, gfp_mask, order);
4789 if (!p)
4790 return NULL;
4791 return make_alloc_exact((unsigned long)page_address(p), order, size);
4792}
4793
4794
4795
4796
4797
4798
4799
4800
4801void free_pages_exact(void *virt, size_t size)
4802{
4803 unsigned long addr = (unsigned long)virt;
4804 unsigned long end = addr + PAGE_ALIGN(size);
4805
4806 while (addr < end) {
4807 free_page(addr);
4808 addr += PAGE_SIZE;
4809 }
4810}
4811EXPORT_SYMBOL(free_pages_exact);
4812
4813
4814
4815
4816
4817
4818
4819
4820
4821
4822
4823static unsigned long nr_free_zone_pages(int offset)
4824{
4825 struct zoneref *z;
4826 struct zone *zone;
4827
4828
4829 unsigned long sum = 0;
4830
4831 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
4832
4833 for_each_zone_zonelist(zone, z, zonelist, offset) {
4834 unsigned long size = zone_managed_pages(zone);
4835 unsigned long high = high_wmark_pages(zone);
4836 if (size > high)
4837 sum += size - high;
4838 }
4839
4840 return sum;
4841}
4842
4843
4844
4845
4846
4847
4848
4849unsigned long nr_free_buffer_pages(void)
4850{
4851 return nr_free_zone_pages(gfp_zone(GFP_USER));
4852}
4853EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
4854
4855
4856
4857
4858
4859
4860
4861unsigned long nr_free_pagecache_pages(void)
4862{
4863 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
4864}
4865
4866static inline void show_node(struct zone *zone)
4867{
4868 if (IS_ENABLED(CONFIG_NUMA))
4869 printk("Node %d ", zone_to_nid(zone));
4870}
4871
4872long si_mem_available(void)
4873{
4874 long available;
4875 unsigned long pagecache;
4876 unsigned long wmark_low = 0;
4877 unsigned long pages[NR_LRU_LISTS];
4878 unsigned long reclaimable;
4879 struct zone *zone;
4880 int lru;
4881
4882 for (lru = LRU_BASE; lru < NR_LRU_LISTS; lru++)
4883 pages[lru] = global_node_page_state(NR_LRU_BASE + lru);
4884
4885 for_each_zone(zone)
4886 wmark_low += low_wmark_pages(zone);
4887
4888
4889
4890
4891
4892 available = global_zone_page_state(NR_FREE_PAGES) - totalreserve_pages;
4893
4894
4895
4896
4897
4898
4899 pagecache = pages[LRU_ACTIVE_FILE] + pages[LRU_INACTIVE_FILE];
4900 pagecache -= min(pagecache / 2, wmark_low);
4901 available += pagecache;
4902
4903
4904
4905
4906
4907
4908 reclaimable = global_node_page_state(NR_SLAB_RECLAIMABLE) +
4909 global_node_page_state(NR_KERNEL_MISC_RECLAIMABLE);
4910 available += reclaimable - min(reclaimable / 2, wmark_low);
4911
4912 if (available < 0)
4913 available = 0;
4914 return available;
4915}
4916EXPORT_SYMBOL_GPL(si_mem_available);
4917
4918void si_meminfo(struct sysinfo *val)
4919{
4920 val->totalram = totalram_pages();
4921 val->sharedram = global_node_page_state(NR_SHMEM);
4922 val->freeram = global_zone_page_state(NR_FREE_PAGES);
4923 val->bufferram = nr_blockdev_pages();
4924 val->totalhigh = totalhigh_pages();
4925 val->freehigh = nr_free_highpages();
4926 val->mem_unit = PAGE_SIZE;
4927}
4928
4929EXPORT_SYMBOL(si_meminfo);
4930
4931#ifdef CONFIG_NUMA
4932void si_meminfo_node(struct sysinfo *val, int nid)
4933{
4934 int zone_type;
4935 unsigned long managed_pages = 0;
4936 unsigned long managed_highpages = 0;
4937 unsigned long free_highpages = 0;
4938 pg_data_t *pgdat = NODE_DATA(nid);
4939
4940 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++)
4941 managed_pages += zone_managed_pages(&pgdat->node_zones[zone_type]);
4942 val->totalram = managed_pages;
4943 val->sharedram = node_page_state(pgdat, NR_SHMEM);
4944 val->freeram = sum_zone_node_page_state(nid, NR_FREE_PAGES);
4945#ifdef CONFIG_HIGHMEM
4946 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
4947 struct zone *zone = &pgdat->node_zones[zone_type];
4948
4949 if (is_highmem(zone)) {
4950 managed_highpages += zone_managed_pages(zone);
4951 free_highpages += zone_page_state(zone, NR_FREE_PAGES);
4952 }
4953 }
4954 val->totalhigh = managed_highpages;
4955 val->freehigh = free_highpages;
4956#else
4957 val->totalhigh = managed_highpages;
4958 val->freehigh = free_highpages;
4959#endif
4960 val->mem_unit = PAGE_SIZE;
4961}
4962#endif
4963
4964
4965
4966
4967
4968static bool show_mem_node_skip(unsigned int flags, int nid, nodemask_t *nodemask)
4969{
4970 if (!(flags & SHOW_MEM_FILTER_NODES))
4971 return false;
4972
4973
4974
4975
4976
4977
4978 if (!nodemask)
4979 nodemask = &cpuset_current_mems_allowed;
4980
4981 return !node_isset(nid, *nodemask);
4982}
4983
4984#define K(x) ((x) << (PAGE_SHIFT-10))
4985
4986static void show_migration_types(unsigned char type)
4987{
4988 static const char types[MIGRATE_TYPES] = {
4989 [MIGRATE_UNMOVABLE] = 'U',
4990 [MIGRATE_MOVABLE] = 'M',
4991 [MIGRATE_RECLAIMABLE] = 'E',
4992 [MIGRATE_HIGHATOMIC] = 'H',
4993#ifdef CONFIG_CMA
4994 [MIGRATE_CMA] = 'C',
4995#endif
4996#ifdef CONFIG_MEMORY_ISOLATION
4997 [MIGRATE_ISOLATE] = 'I',
4998#endif
4999 };
5000 char tmp[MIGRATE_TYPES + 1];
5001 char *p = tmp;
5002 int i;
5003
5004 for (i = 0; i < MIGRATE_TYPES; i++) {
5005 if (type & (1 << i))
5006 *p++ = types[i];
5007 }
5008
5009 *p = '\0';
5010 printk(KERN_CONT "(%s) ", tmp);
5011}
5012
5013
5014
5015
5016
5017
5018
5019
5020
5021
5022void show_free_areas(unsigned int filter, nodemask_t *nodemask)
5023{
5024 unsigned long free_pcp = 0;
5025 int cpu;
5026 struct zone *zone;
5027 pg_data_t *pgdat;
5028
5029 for_each_populated_zone(zone) {
5030 if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask))
5031 continue;
5032
5033 for_each_online_cpu(cpu)
5034 free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count;
5035 }
5036
5037 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
5038 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
5039 " unevictable:%lu dirty:%lu writeback:%lu unstable:%lu\n"
5040 " slab_reclaimable:%lu slab_unreclaimable:%lu\n"
5041 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
5042 " free:%lu free_pcp:%lu free_cma:%lu\n",
5043 global_node_page_state(NR_ACTIVE_ANON),
5044 global_node_page_state(NR_INACTIVE_ANON),
5045 global_node_page_state(NR_ISOLATED_ANON),
5046 global_node_page_state(NR_ACTIVE_FILE),
5047 global_node_page_state(NR_INACTIVE_FILE),
5048 global_node_page_state(NR_ISOLATED_FILE),
5049 global_node_page_state(NR_UNEVICTABLE),
5050 global_node_page_state(NR_FILE_DIRTY),
5051 global_node_page_state(NR_WRITEBACK),
5052 global_node_page_state(NR_UNSTABLE_NFS),
5053 global_node_page_state(NR_SLAB_RECLAIMABLE),
5054 global_node_page_state(NR_SLAB_UNRECLAIMABLE),
5055 global_node_page_state(NR_FILE_MAPPED),
5056 global_node_page_state(NR_SHMEM),
5057 global_zone_page_state(NR_PAGETABLE),
5058 global_zone_page_state(NR_BOUNCE),
5059 global_zone_page_state(NR_FREE_PAGES),
5060 free_pcp,
5061 global_zone_page_state(NR_FREE_CMA_PAGES));
5062
5063 for_each_online_pgdat(pgdat) {
5064 if (show_mem_node_skip(filter, pgdat->node_id, nodemask))
5065 continue;
5066
5067 printk("Node %d"
5068 " active_anon:%lukB"
5069 " inactive_anon:%lukB"
5070 " active_file:%lukB"
5071 " inactive_file:%lukB"
5072 " unevictable:%lukB"
5073 " isolated(anon):%lukB"
5074 " isolated(file):%lukB"
5075 " mapped:%lukB"
5076 " dirty:%lukB"
5077 " writeback:%lukB"
5078 " shmem:%lukB"
5079#ifdef CONFIG_TRANSPARENT_HUGEPAGE
5080 " shmem_thp: %lukB"
5081 " shmem_pmdmapped: %lukB"
5082 " anon_thp: %lukB"
5083#endif
5084 " writeback_tmp:%lukB"
5085 " unstable:%lukB"
5086 " all_unreclaimable? %s"
5087 "\n",
5088 pgdat->node_id,
5089 K(node_page_state(pgdat, NR_ACTIVE_ANON)),
5090 K(node_page_state(pgdat, NR_INACTIVE_ANON)),
5091 K(node_page_state(pgdat, NR_ACTIVE_FILE)),
5092 K(node_page_state(pgdat, NR_INACTIVE_FILE)),
5093 K(node_page_state(pgdat, NR_UNEVICTABLE)),
5094 K(node_page_state(pgdat, NR_ISOLATED_ANON)),
5095 K(node_page_state(pgdat, NR_ISOLATED_FILE)),
5096 K(node_page_state(pgdat, NR_FILE_MAPPED)),
5097 K(node_page_state(pgdat, NR_FILE_DIRTY)),
5098 K(node_page_state(pgdat, NR_WRITEBACK)),
5099 K(node_page_state(pgdat, NR_SHMEM)),
5100#ifdef CONFIG_TRANSPARENT_HUGEPAGE
5101 K(node_page_state(pgdat, NR_SHMEM_THPS) * HPAGE_PMD_NR),
5102 K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED)
5103 * HPAGE_PMD_NR),
5104 K(node_page_state(pgdat, NR_ANON_THPS) * HPAGE_PMD_NR),
5105#endif
5106 K(node_page_state(pgdat, NR_WRITEBACK_TEMP)),
5107 K(node_page_state(pgdat, NR_UNSTABLE_NFS)),
5108 pgdat->kswapd_failures >= MAX_RECLAIM_RETRIES ?
5109 "yes" : "no");
5110 }
5111
5112 for_each_populated_zone(zone) {
5113 int i;
5114
5115 if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask))
5116 continue;
5117
5118 free_pcp = 0;
5119 for_each_online_cpu(cpu)
5120 free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count;
5121
5122 show_node(zone);
5123 printk(KERN_CONT
5124 "%s"
5125 " free:%lukB"
5126 " min:%lukB"
5127 " low:%lukB"
5128 " high:%lukB"
5129 " active_anon:%lukB"
5130 " inactive_anon:%lukB"
5131 " active_file:%lukB"
5132 " inactive_file:%lukB"
5133 " unevictable:%lukB"
5134 " writepending:%lukB"
5135 " present:%lukB"
5136 " managed:%lukB"
5137 " mlocked:%lukB"
5138 " kernel_stack:%lukB"
5139 " pagetables:%lukB"
5140 " bounce:%lukB"
5141 " free_pcp:%lukB"
5142 " local_pcp:%ukB"
5143 " free_cma:%lukB"
5144 "\n",
5145 zone->name,
5146 K(zone_page_state(zone, NR_FREE_PAGES)),
5147 K(min_wmark_pages(zone)),
5148 K(low_wmark_pages(zone)),
5149 K(high_wmark_pages(zone)),
5150 K(zone_page_state(zone, NR_ZONE_ACTIVE_ANON)),
5151 K(zone_page_state(zone, NR_ZONE_INACTIVE_ANON)),
5152 K(zone_page_state(zone, NR_ZONE_ACTIVE_FILE)),
5153 K(zone_page_state(zone, NR_ZONE_INACTIVE_FILE)),
5154 K(zone_page_state(zone, NR_ZONE_UNEVICTABLE)),
5155 K(zone_page_state(zone, NR_ZONE_WRITE_PENDING)),
5156 K(zone->present_pages),
5157 K(zone_managed_pages(zone)),
5158 K(zone_page_state(zone, NR_MLOCK)),
5159 zone_page_state(zone, NR_KERNEL_STACK_KB),
5160 K(zone_page_state(zone, NR_PAGETABLE)),
5161 K(zone_page_state(zone, NR_BOUNCE)),
5162 K(free_pcp),
5163 K(this_cpu_read(zone->pageset->pcp.count)),
5164 K(zone_page_state(zone, NR_FREE_CMA_PAGES)));
5165 printk("lowmem_reserve[]:");
5166 for (i = 0; i < MAX_NR_ZONES; i++)
5167 printk(KERN_CONT " %ld", zone->lowmem_reserve[i]);
5168 printk(KERN_CONT "\n");
5169 }
5170
5171 for_each_populated_zone(zone) {
5172 unsigned int order;
5173 unsigned long nr[MAX_ORDER], flags, total = 0;
5174 unsigned char types[MAX_ORDER];
5175
5176 if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask))
5177 continue;
5178 show_node(zone);
5179 printk(KERN_CONT "%s: ", zone->name);
5180
5181 spin_lock_irqsave(&zone->lock, flags);
5182 for (order = 0; order < MAX_ORDER; order++) {
5183 struct free_area *area = &zone->free_area[order];
5184 int type;
5185
5186 nr[order] = area->nr_free;
5187 total += nr[order] << order;
5188
5189 types[order] = 0;
5190 for (type = 0; type < MIGRATE_TYPES; type++) {
5191 if (!list_empty(&area->free_list[type]))
5192 types[order] |= 1 << type;
5193 }
5194 }
5195 spin_unlock_irqrestore(&zone->lock, flags);
5196 for (order = 0; order < MAX_ORDER; order++) {
5197 printk(KERN_CONT "%lu*%lukB ",
5198 nr[order], K(1UL) << order);
5199 if (nr[order])
5200 show_migration_types(types[order]);
5201 }
5202 printk(KERN_CONT "= %lukB\n", K(total));
5203 }
5204
5205 hugetlb_show_meminfo();
5206
5207 printk("%ld total pagecache pages\n", global_node_page_state(NR_FILE_PAGES));
5208
5209 show_swap_cache_info();
5210}
5211
5212static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
5213{
5214 zoneref->zone = zone;
5215 zoneref->zone_idx = zone_idx(zone);
5216}
5217
5218
5219
5220
5221
5222
5223static int build_zonerefs_node(pg_data_t *pgdat, struct zoneref *zonerefs)
5224{
5225 struct zone *zone;
5226 enum zone_type zone_type = MAX_NR_ZONES;
5227 int nr_zones = 0;
5228
5229 do {
5230 zone_type--;
5231 zone = pgdat->node_zones + zone_type;
5232 if (managed_zone(zone)) {
5233 zoneref_set_zone(zone, &zonerefs[nr_zones++]);
5234 check_highest_zone(zone_type);
5235 }
5236 } while (zone_type);
5237
5238 return nr_zones;
5239}
5240
5241#ifdef CONFIG_NUMA
5242
5243static int __parse_numa_zonelist_order(char *s)
5244{
5245
5246
5247
5248
5249
5250
5251 if (!(*s == 'd' || *s == 'D' || *s == 'n' || *s == 'N')) {
5252 pr_warn("Ignoring unsupported numa_zonelist_order value: %s\n", s);
5253 return -EINVAL;
5254 }
5255 return 0;
5256}
5257
5258static __init int setup_numa_zonelist_order(char *s)
5259{
5260 if (!s)
5261 return 0;
5262
5263 return __parse_numa_zonelist_order(s);
5264}
5265early_param("numa_zonelist_order", setup_numa_zonelist_order);
5266
5267char numa_zonelist_order[] = "Node";
5268
5269
5270
5271
5272int numa_zonelist_order_handler(struct ctl_table *table, int write,
5273 void __user *buffer, size_t *length,
5274 loff_t *ppos)
5275{
5276 char *str;
5277 int ret;
5278
5279 if (!write)
5280 return proc_dostring(table, write, buffer, length, ppos);
5281 str = memdup_user_nul(buffer, 16);
5282 if (IS_ERR(str))
5283 return PTR_ERR(str);
5284
5285 ret = __parse_numa_zonelist_order(str);
5286 kfree(str);
5287 return ret;
5288}
5289
5290
5291#define MAX_NODE_LOAD (nr_online_nodes)
5292static int node_load[MAX_NUMNODES];
5293
5294
5295
5296
5297
5298
5299
5300
5301
5302
5303
5304
5305
5306
5307
5308static int find_next_best_node(int node, nodemask_t *used_node_mask)
5309{
5310 int n, val;
5311 int min_val = INT_MAX;
5312 int best_node = NUMA_NO_NODE;
5313 const struct cpumask *tmp = cpumask_of_node(0);
5314
5315
5316 if (!node_isset(node, *used_node_mask)) {
5317 node_set(node, *used_node_mask);
5318 return node;
5319 }
5320
5321 for_each_node_state(n, N_MEMORY) {
5322
5323
5324 if (node_isset(n, *used_node_mask))
5325 continue;
5326
5327
5328 val = node_distance(node, n);
5329
5330
5331 val += (n < node);
5332
5333
5334 tmp = cpumask_of_node(n);
5335 if (!cpumask_empty(tmp))
5336 val += PENALTY_FOR_NODE_WITH_CPUS;
5337
5338
5339 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
5340 val += node_load[n];
5341
5342 if (val < min_val) {
5343 min_val = val;
5344 best_node = n;
5345 }
5346 }
5347
5348 if (best_node >= 0)
5349 node_set(best_node, *used_node_mask);
5350
5351 return best_node;
5352}
5353
5354
5355
5356
5357
5358
5359
5360static void build_zonelists_in_node_order(pg_data_t *pgdat, int *node_order,
5361 unsigned nr_nodes)
5362{
5363 struct zoneref *zonerefs;
5364 int i;
5365
5366 zonerefs = pgdat->node_zonelists[ZONELIST_FALLBACK]._zonerefs;
5367
5368 for (i = 0; i < nr_nodes; i++) {
5369 int nr_zones;
5370
5371 pg_data_t *node = NODE_DATA(node_order[i]);
5372
5373 nr_zones = build_zonerefs_node(node, zonerefs);
5374 zonerefs += nr_zones;
5375 }
5376 zonerefs->zone = NULL;
5377 zonerefs->zone_idx = 0;
5378}
5379
5380
5381
5382
5383static void build_thisnode_zonelists(pg_data_t *pgdat)
5384{
5385 struct zoneref *zonerefs;
5386 int nr_zones;
5387
5388 zonerefs = pgdat->node_zonelists[ZONELIST_NOFALLBACK]._zonerefs;
5389 nr_zones = build_zonerefs_node(pgdat, zonerefs);
5390 zonerefs += nr_zones;
5391 zonerefs->zone = NULL;
5392 zonerefs->zone_idx = 0;
5393}
5394
5395
5396
5397
5398
5399
5400
5401
5402static void build_zonelists(pg_data_t *pgdat)
5403{
5404 static int node_order[MAX_NUMNODES];
5405 int node, load, nr_nodes = 0;
5406 nodemask_t used_mask;
5407 int local_node, prev_node;
5408
5409
5410 local_node = pgdat->node_id;
5411 load = nr_online_nodes;
5412 prev_node = local_node;
5413 nodes_clear(used_mask);
5414
5415 memset(node_order, 0, sizeof(node_order));
5416 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
5417
5418
5419
5420
5421
5422 if (node_distance(local_node, node) !=
5423 node_distance(local_node, prev_node))
5424 node_load[node] = load;
5425
5426 node_order[nr_nodes++] = node;
5427 prev_node = node;
5428 load--;
5429 }
5430
5431 build_zonelists_in_node_order(pgdat, node_order, nr_nodes);
5432 build_thisnode_zonelists(pgdat);
5433}
5434
5435#ifdef CONFIG_HAVE_MEMORYLESS_NODES
5436
5437
5438
5439
5440
5441
5442int local_memory_node(int node)
5443{
5444 struct zoneref *z;
5445
5446 z = first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
5447 gfp_zone(GFP_KERNEL),
5448 NULL);
5449 return zone_to_nid(z->zone);
5450}
5451#endif
5452
5453static void setup_min_unmapped_ratio(void);
5454static void setup_min_slab_ratio(void);
5455#else
5456
5457static void build_zonelists(pg_data_t *pgdat)
5458{
5459 int node, local_node;
5460 struct zoneref *zonerefs;
5461 int nr_zones;
5462
5463 local_node = pgdat->node_id;
5464
5465 zonerefs = pgdat->node_zonelists[ZONELIST_FALLBACK]._zonerefs;
5466 nr_zones = build_zonerefs_node(pgdat, zonerefs);
5467 zonerefs += nr_zones;
5468
5469
5470
5471
5472
5473
5474
5475
5476
5477 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
5478 if (!node_online(node))
5479 continue;
5480 nr_zones = build_zonerefs_node(NODE_DATA(node), zonerefs);
5481 zonerefs += nr_zones;
5482 }
5483 for (node = 0; node < local_node; node++) {
5484 if (!node_online(node))
5485 continue;
5486 nr_zones = build_zonerefs_node(NODE_DATA(node), zonerefs);
5487 zonerefs += nr_zones;
5488 }
5489
5490 zonerefs->zone = NULL;
5491 zonerefs->zone_idx = 0;
5492}
5493
5494#endif
5495
5496
5497
5498
5499
5500
5501
5502
5503
5504
5505
5506
5507
5508
5509
5510
5511static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch);
5512static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset);
5513static DEFINE_PER_CPU(struct per_cpu_nodestat, boot_nodestats);
5514
5515static void __build_all_zonelists(void *data)
5516{
5517 int nid;
5518 int __maybe_unused cpu;
5519 pg_data_t *self = data;
5520 static DEFINE_SPINLOCK(lock);
5521
5522 spin_lock(&lock);
5523
5524#ifdef CONFIG_NUMA
5525 memset(node_load, 0, sizeof(node_load));
5526#endif
5527
5528
5529
5530
5531
5532 if (self && !node_online(self->node_id)) {
5533 build_zonelists(self);
5534 } else {
5535 for_each_online_node(nid) {
5536 pg_data_t *pgdat = NODE_DATA(nid);
5537
5538 build_zonelists(pgdat);
5539 }
5540
5541#ifdef CONFIG_HAVE_MEMORYLESS_NODES
5542
5543
5544
5545
5546
5547
5548
5549
5550 for_each_online_cpu(cpu)
5551 set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
5552#endif
5553 }
5554
5555 spin_unlock(&lock);
5556}
5557
5558static noinline void __init
5559build_all_zonelists_init(void)
5560{
5561 int cpu;
5562
5563 __build_all_zonelists(NULL);
5564
5565
5566
5567
5568
5569
5570
5571
5572
5573
5574
5575
5576
5577
5578 for_each_possible_cpu(cpu)
5579 setup_pageset(&per_cpu(boot_pageset, cpu), 0);
5580
5581 mminit_verify_zonelist();
5582 cpuset_init_current_mems_allowed();
5583}
5584
5585
5586
5587
5588
5589
5590
5591void __ref build_all_zonelists(pg_data_t *pgdat)
5592{
5593 if (system_state == SYSTEM_BOOTING) {
5594 build_all_zonelists_init();
5595 } else {
5596 __build_all_zonelists(pgdat);
5597
5598 }
5599 vm_total_pages = nr_free_pagecache_pages();
5600
5601
5602
5603
5604
5605
5606
5607 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
5608 page_group_by_mobility_disabled = 1;
5609 else
5610 page_group_by_mobility_disabled = 0;
5611
5612 pr_info("Built %i zonelists, mobility grouping %s. Total pages: %ld\n",
5613 nr_online_nodes,
5614 page_group_by_mobility_disabled ? "off" : "on",
5615 vm_total_pages);
5616#ifdef CONFIG_NUMA
5617 pr_info("Policy zone: %s\n", zone_names[policy_zone]);
5618#endif
5619}
5620
5621
5622static bool __meminit
5623overlap_memmap_init(unsigned long zone, unsigned long *pfn)
5624{
5625#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5626 static struct memblock_region *r;
5627
5628 if (mirrored_kernelcore && zone == ZONE_MOVABLE) {
5629 if (!r || *pfn >= memblock_region_memory_end_pfn(r)) {
5630 for_each_memblock(memory, r) {
5631 if (*pfn < memblock_region_memory_end_pfn(r))
5632 break;
5633 }
5634 }
5635 if (*pfn >= memblock_region_memory_base_pfn(r) &&
5636 memblock_is_mirror(r)) {
5637 *pfn = memblock_region_memory_end_pfn(r);
5638 return true;
5639 }
5640 }
5641#endif
5642 return false;
5643}
5644
5645
5646
5647
5648
5649
5650void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
5651 unsigned long start_pfn, enum memmap_context context,
5652 struct vmem_altmap *altmap)
5653{
5654 unsigned long pfn, end_pfn = start_pfn + size;
5655 struct page *page;
5656
5657 if (highest_memmap_pfn < end_pfn - 1)
5658 highest_memmap_pfn = end_pfn - 1;
5659
5660#ifdef CONFIG_ZONE_DEVICE
5661
5662
5663
5664
5665
5666
5667
5668 if (zone == ZONE_DEVICE) {
5669 if (!altmap)
5670 return;
5671
5672 if (start_pfn == altmap->base_pfn)
5673 start_pfn += altmap->reserve;
5674 end_pfn = altmap->base_pfn + vmem_altmap_offset(altmap);
5675 }
5676#endif
5677
5678 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
5679
5680
5681
5682
5683 if (context == MEMMAP_EARLY) {
5684 if (!early_pfn_valid(pfn))
5685 continue;
5686 if (!early_pfn_in_nid(pfn, nid))
5687 continue;
5688 if (overlap_memmap_init(zone, &pfn))
5689 continue;
5690 if (defer_init(nid, pfn, end_pfn))
5691 break;
5692 }
5693
5694 page = pfn_to_page(pfn);
5695 __init_single_page(page, pfn, zone, nid);
5696 if (context == MEMMAP_HOTPLUG)
5697 __SetPageReserved(page);
5698
5699
5700
5701
5702
5703
5704
5705
5706
5707
5708
5709
5710
5711 if (!(pfn & (pageblock_nr_pages - 1))) {
5712 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
5713 cond_resched();
5714 }
5715 }
5716}
5717
5718#ifdef CONFIG_ZONE_DEVICE
5719void __ref memmap_init_zone_device(struct zone *zone,
5720 unsigned long start_pfn,
5721 unsigned long size,
5722 struct dev_pagemap *pgmap)
5723{
5724 unsigned long pfn, end_pfn = start_pfn + size;
5725 struct pglist_data *pgdat = zone->zone_pgdat;
5726 unsigned long zone_idx = zone_idx(zone);
5727 unsigned long start = jiffies;
5728 int nid = pgdat->node_id;
5729
5730 if (WARN_ON_ONCE(!pgmap || !is_dev_zone(zone)))
5731 return;
5732
5733
5734
5735
5736
5737
5738 if (pgmap->altmap_valid) {
5739 struct vmem_altmap *altmap = &pgmap->altmap;
5740
5741 start_pfn = altmap->base_pfn + vmem_altmap_offset(altmap);
5742 size = end_pfn - start_pfn;
5743 }
5744
5745 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
5746 struct page *page = pfn_to_page(pfn);
5747
5748 __init_single_page(page, pfn, zone_idx, nid);
5749
5750
5751
5752
5753
5754
5755
5756
5757 __SetPageReserved(page);
5758
5759
5760
5761
5762
5763
5764 page->pgmap = pgmap;
5765 page->hmm_data = 0;
5766
5767
5768
5769
5770
5771
5772
5773
5774
5775
5776
5777
5778
5779
5780
5781
5782 if (!(pfn & (pageblock_nr_pages - 1))) {
5783 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
5784 cond_resched();
5785 }
5786 }
5787
5788 pr_info("%s initialised, %lu pages in %ums\n", dev_name(pgmap->dev),
5789 size, jiffies_to_msecs(jiffies - start));
5790}
5791
5792#endif
5793static void __meminit zone_init_free_lists(struct zone *zone)
5794{
5795 unsigned int order, t;
5796 for_each_migratetype_order(order, t) {
5797 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
5798 zone->free_area[order].nr_free = 0;
5799 }
5800}
5801
5802void __meminit __weak memmap_init(unsigned long size, int nid,
5803 unsigned long zone, unsigned long start_pfn)
5804{
5805 memmap_init_zone(size, nid, zone, start_pfn, MEMMAP_EARLY, NULL);
5806}
5807
5808static int zone_batchsize(struct zone *zone)
5809{
5810#ifdef CONFIG_MMU
5811 int batch;
5812
5813
5814
5815
5816
5817 batch = zone_managed_pages(zone) / 1024;
5818
5819 if (batch * PAGE_SIZE > 1024 * 1024)
5820 batch = (1024 * 1024) / PAGE_SIZE;
5821 batch /= 4;
5822 if (batch < 1)
5823 batch = 1;
5824
5825
5826
5827
5828
5829
5830
5831
5832
5833
5834
5835 batch = rounddown_pow_of_two(batch + batch/2) - 1;
5836
5837 return batch;
5838
5839#else
5840
5841
5842
5843
5844
5845
5846
5847
5848
5849
5850
5851
5852
5853 return 0;
5854#endif
5855}
5856
5857
5858
5859
5860
5861
5862
5863
5864
5865
5866
5867
5868
5869
5870static void pageset_update(struct per_cpu_pages *pcp, unsigned long high,
5871 unsigned long batch)
5872{
5873
5874 pcp->batch = 1;
5875 smp_wmb();
5876
5877
5878 pcp->high = high;
5879 smp_wmb();
5880
5881 pcp->batch = batch;
5882}
5883
5884
5885static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch)
5886{
5887 pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch));
5888}
5889
5890static void pageset_init(struct per_cpu_pageset *p)
5891{
5892 struct per_cpu_pages *pcp;
5893 int migratetype;
5894
5895 memset(p, 0, sizeof(*p));
5896
5897 pcp = &p->pcp;
5898 for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
5899 INIT_LIST_HEAD(&pcp->lists[migratetype]);
5900}
5901
5902static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
5903{
5904 pageset_init(p);
5905 pageset_set_batch(p, batch);
5906}
5907
5908
5909
5910
5911
5912static void pageset_set_high(struct per_cpu_pageset *p,
5913 unsigned long high)
5914{
5915 unsigned long batch = max(1UL, high / 4);
5916 if ((high / 4) > (PAGE_SHIFT * 8))
5917 batch = PAGE_SHIFT * 8;
5918
5919 pageset_update(&p->pcp, high, batch);
5920}
5921
5922static void pageset_set_high_and_batch(struct zone *zone,
5923 struct per_cpu_pageset *pcp)
5924{
5925 if (percpu_pagelist_fraction)
5926 pageset_set_high(pcp,
5927 (zone_managed_pages(zone) /
5928 percpu_pagelist_fraction));
5929 else
5930 pageset_set_batch(pcp, zone_batchsize(zone));
5931}
5932
5933static void __meminit zone_pageset_init(struct zone *zone, int cpu)
5934{
5935 struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu);
5936
5937 pageset_init(pcp);
5938 pageset_set_high_and_batch(zone, pcp);
5939}
5940
5941void __meminit setup_zone_pageset(struct zone *zone)
5942{
5943 int cpu;
5944 zone->pageset = alloc_percpu(struct per_cpu_pageset);
5945 for_each_possible_cpu(cpu)
5946 zone_pageset_init(zone, cpu);
5947}
5948
5949
5950
5951
5952
5953void __init setup_per_cpu_pageset(void)
5954{
5955 struct pglist_data *pgdat;
5956 struct zone *zone;
5957
5958 for_each_populated_zone(zone)
5959 setup_zone_pageset(zone);
5960
5961 for_each_online_pgdat(pgdat)
5962 pgdat->per_cpu_nodestats =
5963 alloc_percpu(struct per_cpu_nodestat);
5964}
5965
5966static __meminit void zone_pcp_init(struct zone *zone)
5967{
5968
5969
5970
5971
5972
5973 zone->pageset = &boot_pageset;
5974
5975 if (populated_zone(zone))
5976 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n",
5977 zone->name, zone->present_pages,
5978 zone_batchsize(zone));
5979}
5980
5981void __meminit init_currently_empty_zone(struct zone *zone,
5982 unsigned long zone_start_pfn,
5983 unsigned long size)
5984{
5985 struct pglist_data *pgdat = zone->zone_pgdat;
5986 int zone_idx = zone_idx(zone) + 1;
5987
5988 if (zone_idx > pgdat->nr_zones)
5989 pgdat->nr_zones = zone_idx;
5990
5991 zone->zone_start_pfn = zone_start_pfn;
5992
5993 mminit_dprintk(MMINIT_TRACE, "memmap_init",
5994 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
5995 pgdat->node_id,
5996 (unsigned long)zone_idx(zone),
5997 zone_start_pfn, (zone_start_pfn + size));
5998
5999 zone_init_free_lists(zone);
6000 zone->initialized = 1;
6001}
6002
6003#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
6004#ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
6005
6006
6007
6008
6009int __meminit __early_pfn_to_nid(unsigned long pfn,
6010 struct mminit_pfnnid_cache *state)
6011{
6012 unsigned long start_pfn, end_pfn;
6013 int nid;
6014
6015 if (state->last_start <= pfn && pfn < state->last_end)
6016 return state->last_nid;
6017
6018 nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn);
6019 if (nid != -1) {
6020 state->last_start = start_pfn;
6021 state->last_end = end_pfn;
6022 state->last_nid = nid;
6023 }
6024
6025 return nid;
6026}
6027#endif
6028
6029
6030
6031
6032
6033
6034
6035
6036
6037
6038void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn)
6039{
6040 unsigned long start_pfn, end_pfn;
6041 int i, this_nid;
6042
6043 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) {
6044 start_pfn = min(start_pfn, max_low_pfn);
6045 end_pfn = min(end_pfn, max_low_pfn);
6046
6047 if (start_pfn < end_pfn)
6048 memblock_free_early_nid(PFN_PHYS(start_pfn),
6049 (end_pfn - start_pfn) << PAGE_SHIFT,
6050 this_nid);
6051 }
6052}
6053
6054
6055
6056
6057
6058
6059
6060
6061void __init sparse_memory_present_with_active_regions(int nid)
6062{
6063 unsigned long start_pfn, end_pfn;
6064 int i, this_nid;
6065
6066 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid)
6067 memory_present(this_nid, start_pfn, end_pfn);
6068}
6069
6070
6071
6072
6073
6074
6075
6076
6077
6078
6079
6080
6081void __init get_pfn_range_for_nid(unsigned int nid,
6082 unsigned long *start_pfn, unsigned long *end_pfn)
6083{
6084 unsigned long this_start_pfn, this_end_pfn;
6085 int i;
6086
6087 *start_pfn = -1UL;
6088 *end_pfn = 0;
6089
6090 for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) {
6091 *start_pfn = min(*start_pfn, this_start_pfn);
6092 *end_pfn = max(*end_pfn, this_end_pfn);
6093 }
6094
6095 if (*start_pfn == -1UL)
6096 *start_pfn = 0;
6097}
6098
6099
6100
6101
6102
6103
6104static void __init find_usable_zone_for_movable(void)
6105{
6106 int zone_index;
6107 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
6108 if (zone_index == ZONE_MOVABLE)
6109 continue;
6110
6111 if (arch_zone_highest_possible_pfn[zone_index] >
6112 arch_zone_lowest_possible_pfn[zone_index])
6113 break;
6114 }
6115
6116 VM_BUG_ON(zone_index == -1);
6117 movable_zone = zone_index;
6118}
6119
6120
6121
6122
6123
6124
6125
6126
6127
6128
6129
6130static void __init adjust_zone_range_for_zone_movable(int nid,
6131 unsigned long zone_type,
6132 unsigned long node_start_pfn,
6133 unsigned long node_end_pfn,
6134 unsigned long *zone_start_pfn,
6135 unsigned long *zone_end_pfn)
6136{
6137
6138 if (zone_movable_pfn[nid]) {
6139
6140 if (zone_type == ZONE_MOVABLE) {
6141 *zone_start_pfn = zone_movable_pfn[nid];
6142 *zone_end_pfn = min(node_end_pfn,
6143 arch_zone_highest_possible_pfn[movable_zone]);
6144
6145
6146 } else if (!mirrored_kernelcore &&
6147 *zone_start_pfn < zone_movable_pfn[nid] &&
6148 *zone_end_pfn > zone_movable_pfn[nid]) {
6149 *zone_end_pfn = zone_movable_pfn[nid];
6150
6151
6152 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
6153 *zone_start_pfn = *zone_end_pfn;
6154 }
6155}
6156
6157
6158
6159
6160
6161static unsigned long __init zone_spanned_pages_in_node(int nid,
6162 unsigned long zone_type,
6163 unsigned long node_start_pfn,
6164 unsigned long node_end_pfn,
6165 unsigned long *zone_start_pfn,
6166 unsigned long *zone_end_pfn,
6167 unsigned long *ignored)
6168{
6169
6170 if (!node_start_pfn && !node_end_pfn)
6171 return 0;
6172
6173
6174 *zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
6175 *zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
6176 adjust_zone_range_for_zone_movable(nid, zone_type,
6177 node_start_pfn, node_end_pfn,
6178 zone_start_pfn, zone_end_pfn);
6179
6180
6181 if (*zone_end_pfn < node_start_pfn || *zone_start_pfn > node_end_pfn)
6182 return 0;
6183
6184
6185 *zone_end_pfn = min(*zone_end_pfn, node_end_pfn);
6186 *zone_start_pfn = max(*zone_start_pfn, node_start_pfn);
6187
6188
6189 return *zone_end_pfn - *zone_start_pfn;
6190}
6191
6192
6193
6194
6195
6196unsigned long __init __absent_pages_in_range(int nid,
6197 unsigned long range_start_pfn,
6198 unsigned long range_end_pfn)
6199{
6200 unsigned long nr_absent = range_end_pfn - range_start_pfn;
6201 unsigned long start_pfn, end_pfn;
6202 int i;
6203
6204 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
6205 start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn);
6206 end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn);
6207 nr_absent -= end_pfn - start_pfn;
6208 }
6209 return nr_absent;
6210}
6211
6212
6213
6214
6215
6216
6217
6218
6219unsigned long __init absent_pages_in_range(unsigned long start_pfn,
6220 unsigned long end_pfn)
6221{
6222 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
6223}
6224
6225
6226static unsigned long __init zone_absent_pages_in_node(int nid,
6227 unsigned long zone_type,
6228 unsigned long node_start_pfn,
6229 unsigned long node_end_pfn,
6230 unsigned long *ignored)
6231{
6232 unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type];
6233 unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type];
6234 unsigned long zone_start_pfn, zone_end_pfn;
6235 unsigned long nr_absent;
6236
6237
6238 if (!node_start_pfn && !node_end_pfn)
6239 return 0;
6240
6241 zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high);
6242 zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high);
6243
6244 adjust_zone_range_for_zone_movable(nid, zone_type,
6245 node_start_pfn, node_end_pfn,
6246 &zone_start_pfn, &zone_end_pfn);
6247 nr_absent = __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
6248
6249
6250
6251
6252
6253
6254 if (mirrored_kernelcore && zone_movable_pfn[nid]) {
6255 unsigned long start_pfn, end_pfn;
6256 struct memblock_region *r;
6257
6258 for_each_memblock(memory, r) {
6259 start_pfn = clamp(memblock_region_memory_base_pfn(r),
6260 zone_start_pfn, zone_end_pfn);
6261 end_pfn = clamp(memblock_region_memory_end_pfn(r),
6262 zone_start_pfn, zone_end_pfn);
6263
6264 if (zone_type == ZONE_MOVABLE &&
6265 memblock_is_mirror(r))
6266 nr_absent += end_pfn - start_pfn;
6267
6268 if (zone_type == ZONE_NORMAL &&
6269 !memblock_is_mirror(r))
6270 nr_absent += end_pfn - start_pfn;
6271 }
6272 }
6273
6274 return nr_absent;
6275}
6276
6277#else
6278static inline unsigned long __init zone_spanned_pages_in_node(int nid,
6279 unsigned long zone_type,
6280 unsigned long node_start_pfn,
6281 unsigned long node_end_pfn,
6282 unsigned long *zone_start_pfn,
6283 unsigned long *zone_end_pfn,
6284 unsigned long *zones_size)
6285{
6286 unsigned int zone;
6287
6288 *zone_start_pfn = node_start_pfn;
6289 for (zone = 0; zone < zone_type; zone++)
6290 *zone_start_pfn += zones_size[zone];
6291
6292 *zone_end_pfn = *zone_start_pfn + zones_size[zone_type];
6293
6294 return zones_size[zone_type];
6295}
6296
6297static inline unsigned long __init zone_absent_pages_in_node(int nid,
6298 unsigned long zone_type,
6299 unsigned long node_start_pfn,
6300 unsigned long node_end_pfn,
6301 unsigned long *zholes_size)
6302{
6303 if (!zholes_size)
6304 return 0;
6305
6306 return zholes_size[zone_type];
6307}
6308
6309#endif
6310
6311static void __init calculate_node_totalpages(struct pglist_data *pgdat,
6312 unsigned long node_start_pfn,
6313 unsigned long node_end_pfn,
6314 unsigned long *zones_size,
6315 unsigned long *zholes_size)
6316{
6317 unsigned long realtotalpages = 0, totalpages = 0;
6318 enum zone_type i;
6319
6320 for (i = 0; i < MAX_NR_ZONES; i++) {
6321 struct zone *zone = pgdat->node_zones + i;
6322 unsigned long zone_start_pfn, zone_end_pfn;
6323 unsigned long size, real_size;
6324
6325 size = zone_spanned_pages_in_node(pgdat->node_id, i,
6326 node_start_pfn,
6327 node_end_pfn,
6328 &zone_start_pfn,
6329 &zone_end_pfn,
6330 zones_size);
6331 real_size = size - zone_absent_pages_in_node(pgdat->node_id, i,
6332 node_start_pfn, node_end_pfn,
6333 zholes_size);
6334 if (size)
6335 zone->zone_start_pfn = zone_start_pfn;
6336 else
6337 zone->zone_start_pfn = 0;
6338 zone->spanned_pages = size;
6339 zone->present_pages = real_size;
6340
6341 totalpages += size;
6342 realtotalpages += real_size;
6343 }
6344
6345 pgdat->node_spanned_pages = totalpages;
6346 pgdat->node_present_pages = realtotalpages;
6347 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
6348 realtotalpages);
6349}
6350
6351#ifndef CONFIG_SPARSEMEM
6352
6353
6354
6355
6356
6357
6358
6359static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize)
6360{
6361 unsigned long usemapsize;
6362
6363 zonesize += zone_start_pfn & (pageblock_nr_pages-1);
6364 usemapsize = roundup(zonesize, pageblock_nr_pages);
6365 usemapsize = usemapsize >> pageblock_order;
6366 usemapsize *= NR_PAGEBLOCK_BITS;
6367 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
6368
6369 return usemapsize / 8;
6370}
6371
6372static void __ref setup_usemap(struct pglist_data *pgdat,
6373 struct zone *zone,
6374 unsigned long zone_start_pfn,
6375 unsigned long zonesize)
6376{
6377 unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize);
6378 zone->pageblock_flags = NULL;
6379 if (usemapsize)
6380 zone->pageblock_flags =
6381 memblock_alloc_node_nopanic(usemapsize,
6382 pgdat->node_id);
6383}
6384#else
6385static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone,
6386 unsigned long zone_start_pfn, unsigned long zonesize) {}
6387#endif
6388
6389#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
6390
6391
6392void __init set_pageblock_order(void)
6393{
6394 unsigned int order;
6395
6396
6397 if (pageblock_order)
6398 return;
6399
6400 if (HPAGE_SHIFT > PAGE_SHIFT)
6401 order = HUGETLB_PAGE_ORDER;
6402 else
6403 order = MAX_ORDER - 1;
6404
6405
6406
6407
6408
6409
6410 pageblock_order = order;
6411}
6412#else
6413
6414
6415
6416
6417
6418
6419
6420void __init set_pageblock_order(void)
6421{
6422}
6423
6424#endif
6425
6426static unsigned long __init calc_memmap_size(unsigned long spanned_pages,
6427 unsigned long present_pages)
6428{
6429 unsigned long pages = spanned_pages;
6430
6431
6432
6433
6434
6435
6436
6437
6438
6439 if (spanned_pages > present_pages + (present_pages >> 4) &&
6440 IS_ENABLED(CONFIG_SPARSEMEM))
6441 pages = present_pages;
6442
6443 return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT;
6444}
6445
6446#ifdef CONFIG_TRANSPARENT_HUGEPAGE
6447static void pgdat_init_split_queue(struct pglist_data *pgdat)
6448{
6449 spin_lock_init(&pgdat->split_queue_lock);
6450 INIT_LIST_HEAD(&pgdat->split_queue);
6451 pgdat->split_queue_len = 0;
6452}
6453#else
6454static void pgdat_init_split_queue(struct pglist_data *pgdat) {}
6455#endif
6456
6457#ifdef CONFIG_COMPACTION
6458static void pgdat_init_kcompactd(struct pglist_data *pgdat)
6459{
6460 init_waitqueue_head(&pgdat->kcompactd_wait);
6461}
6462#else
6463static void pgdat_init_kcompactd(struct pglist_data *pgdat) {}
6464#endif
6465
6466static void __meminit pgdat_init_internals(struct pglist_data *pgdat)
6467{
6468 pgdat_resize_init(pgdat);
6469
6470 pgdat_init_split_queue(pgdat);
6471 pgdat_init_kcompactd(pgdat);
6472
6473 init_waitqueue_head(&pgdat->kswapd_wait);
6474 init_waitqueue_head(&pgdat->pfmemalloc_wait);
6475
6476 pgdat_page_ext_init(pgdat);
6477 spin_lock_init(&pgdat->lru_lock);
6478 lruvec_init(node_lruvec(pgdat));
6479}
6480
6481static void __meminit zone_init_internals(struct zone *zone, enum zone_type idx, int nid,
6482 unsigned long remaining_pages)
6483{
6484 atomic_long_set(&zone->managed_pages, remaining_pages);
6485 zone_set_nid(zone, nid);
6486 zone->name = zone_names[idx];
6487 zone->zone_pgdat = NODE_DATA(nid);
6488 spin_lock_init(&zone->lock);
6489 zone_seqlock_init(zone);
6490 zone_pcp_init(zone);
6491}
6492
6493
6494
6495
6496
6497
6498
6499
6500#ifdef CONFIG_MEMORY_HOTPLUG
6501void __ref free_area_init_core_hotplug(int nid)
6502{
6503 enum zone_type z;
6504 pg_data_t *pgdat = NODE_DATA(nid);
6505
6506 pgdat_init_internals(pgdat);
6507 for (z = 0; z < MAX_NR_ZONES; z++)
6508 zone_init_internals(&pgdat->node_zones[z], z, nid, 0);
6509}
6510#endif
6511
6512
6513
6514
6515
6516
6517
6518
6519
6520
6521static void __init free_area_init_core(struct pglist_data *pgdat)
6522{
6523 enum zone_type j;
6524 int nid = pgdat->node_id;
6525
6526 pgdat_init_internals(pgdat);
6527 pgdat->per_cpu_nodestats = &boot_nodestats;
6528
6529 for (j = 0; j < MAX_NR_ZONES; j++) {
6530 struct zone *zone = pgdat->node_zones + j;
6531 unsigned long size, freesize, memmap_pages;
6532 unsigned long zone_start_pfn = zone->zone_start_pfn;
6533
6534 size = zone->spanned_pages;
6535 freesize = zone->present_pages;
6536
6537
6538
6539
6540
6541
6542 memmap_pages = calc_memmap_size(size, freesize);
6543 if (!is_highmem_idx(j)) {
6544 if (freesize >= memmap_pages) {
6545 freesize -= memmap_pages;
6546 if (memmap_pages)
6547 printk(KERN_DEBUG
6548 " %s zone: %lu pages used for memmap\n",
6549 zone_names[j], memmap_pages);
6550 } else
6551 pr_warn(" %s zone: %lu pages exceeds freesize %lu\n",
6552 zone_names[j], memmap_pages, freesize);
6553 }
6554
6555
6556 if (j == 0 && freesize > dma_reserve) {
6557 freesize -= dma_reserve;
6558 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
6559 zone_names[0], dma_reserve);
6560 }
6561
6562 if (!is_highmem_idx(j))
6563 nr_kernel_pages += freesize;
6564
6565 else if (nr_kernel_pages > memmap_pages * 2)
6566 nr_kernel_pages -= memmap_pages;
6567 nr_all_pages += freesize;
6568
6569
6570
6571
6572
6573
6574 zone_init_internals(zone, j, nid, freesize);
6575
6576 if (!size)
6577 continue;
6578
6579 set_pageblock_order();
6580 setup_usemap(pgdat, zone, zone_start_pfn, size);
6581 init_currently_empty_zone(zone, zone_start_pfn, size);
6582 memmap_init(size, nid, j, zone_start_pfn);
6583 }
6584}
6585
6586#ifdef CONFIG_FLAT_NODE_MEM_MAP
6587static void __ref alloc_node_mem_map(struct pglist_data *pgdat)
6588{
6589 unsigned long __maybe_unused start = 0;
6590 unsigned long __maybe_unused offset = 0;
6591
6592
6593 if (!pgdat->node_spanned_pages)
6594 return;
6595
6596 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
6597 offset = pgdat->node_start_pfn - start;
6598
6599 if (!pgdat->node_mem_map) {
6600 unsigned long size, end;
6601 struct page *map;
6602
6603
6604
6605
6606
6607
6608 end = pgdat_end_pfn(pgdat);
6609 end = ALIGN(end, MAX_ORDER_NR_PAGES);
6610 size = (end - start) * sizeof(struct page);
6611 map = memblock_alloc_node_nopanic(size, pgdat->node_id);
6612 pgdat->node_mem_map = map + offset;
6613 }
6614 pr_debug("%s: node %d, pgdat %08lx, node_mem_map %08lx\n",
6615 __func__, pgdat->node_id, (unsigned long)pgdat,
6616 (unsigned long)pgdat->node_mem_map);
6617#ifndef CONFIG_NEED_MULTIPLE_NODES
6618
6619
6620
6621 if (pgdat == NODE_DATA(0)) {
6622 mem_map = NODE_DATA(0)->node_mem_map;
6623#if defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) || defined(CONFIG_FLATMEM)
6624 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
6625 mem_map -= offset;
6626#endif
6627 }
6628#endif
6629}
6630#else
6631static void __ref alloc_node_mem_map(struct pglist_data *pgdat) { }
6632#endif
6633
6634#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
6635static inline void pgdat_set_deferred_range(pg_data_t *pgdat)
6636{
6637 pgdat->first_deferred_pfn = ULONG_MAX;
6638}
6639#else
6640static inline void pgdat_set_deferred_range(pg_data_t *pgdat) {}
6641#endif
6642
6643void __init free_area_init_node(int nid, unsigned long *zones_size,
6644 unsigned long node_start_pfn,
6645 unsigned long *zholes_size)
6646{
6647 pg_data_t *pgdat = NODE_DATA(nid);
6648 unsigned long start_pfn = 0;
6649 unsigned long end_pfn = 0;
6650
6651
6652 WARN_ON(pgdat->nr_zones || pgdat->kswapd_classzone_idx);
6653
6654 pgdat->node_id = nid;
6655 pgdat->node_start_pfn = node_start_pfn;
6656 pgdat->per_cpu_nodestats = NULL;
6657#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
6658 get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
6659 pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid,
6660 (u64)start_pfn << PAGE_SHIFT,
6661 end_pfn ? ((u64)end_pfn << PAGE_SHIFT) - 1 : 0);
6662#else
6663 start_pfn = node_start_pfn;
6664#endif
6665 calculate_node_totalpages(pgdat, start_pfn, end_pfn,
6666 zones_size, zholes_size);
6667
6668 alloc_node_mem_map(pgdat);
6669 pgdat_set_deferred_range(pgdat);
6670
6671 free_area_init_core(pgdat);
6672}
6673
6674#if !defined(CONFIG_FLAT_NODE_MEM_MAP)
6675
6676
6677
6678
6679static u64 zero_pfn_range(unsigned long spfn, unsigned long epfn)
6680{
6681 unsigned long pfn;
6682 u64 pgcnt = 0;
6683
6684 for (pfn = spfn; pfn < epfn; pfn++) {
6685 if (!pfn_valid(ALIGN_DOWN(pfn, pageblock_nr_pages))) {
6686 pfn = ALIGN_DOWN(pfn, pageblock_nr_pages)
6687 + pageblock_nr_pages - 1;
6688 continue;
6689 }
6690 mm_zero_struct_page(pfn_to_page(pfn));
6691 pgcnt++;
6692 }
6693
6694 return pgcnt;
6695}
6696
6697
6698
6699
6700
6701
6702
6703
6704
6705
6706
6707
6708
6709void __init zero_resv_unavail(void)
6710{
6711 phys_addr_t start, end;
6712 u64 i, pgcnt;
6713 phys_addr_t next = 0;
6714
6715
6716
6717
6718 pgcnt = 0;
6719 for_each_mem_range(i, &memblock.memory, NULL,
6720 NUMA_NO_NODE, MEMBLOCK_NONE, &start, &end, NULL) {
6721 if (next < start)
6722 pgcnt += zero_pfn_range(PFN_DOWN(next), PFN_UP(start));
6723 next = end;
6724 }
6725 pgcnt += zero_pfn_range(PFN_DOWN(next), max_pfn);
6726
6727
6728
6729
6730
6731 if (pgcnt)
6732 pr_info("Zeroed struct page in unavailable ranges: %lld pages", pgcnt);
6733}
6734#endif
6735
6736#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
6737
6738#if MAX_NUMNODES > 1
6739
6740
6741
6742void __init setup_nr_node_ids(void)
6743{
6744 unsigned int highest;
6745
6746 highest = find_last_bit(node_possible_map.bits, MAX_NUMNODES);
6747 nr_node_ids = highest + 1;
6748}
6749#endif
6750
6751
6752
6753
6754
6755
6756
6757
6758
6759
6760
6761
6762
6763
6764
6765
6766
6767
6768
6769
6770unsigned long __init node_map_pfn_alignment(void)
6771{
6772 unsigned long accl_mask = 0, last_end = 0;
6773 unsigned long start, end, mask;
6774 int last_nid = -1;
6775 int i, nid;
6776
6777 for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) {
6778 if (!start || last_nid < 0 || last_nid == nid) {
6779 last_nid = nid;
6780 last_end = end;
6781 continue;
6782 }
6783
6784
6785
6786
6787
6788
6789 mask = ~((1 << __ffs(start)) - 1);
6790 while (mask && last_end <= (start & (mask << 1)))
6791 mask <<= 1;
6792
6793
6794 accl_mask |= mask;
6795 }
6796
6797
6798 return ~accl_mask + 1;
6799}
6800
6801
6802static unsigned long __init find_min_pfn_for_node(int nid)
6803{
6804 unsigned long min_pfn = ULONG_MAX;
6805 unsigned long start_pfn;
6806 int i;
6807
6808 for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL)
6809 min_pfn = min(min_pfn, start_pfn);
6810
6811 if (min_pfn == ULONG_MAX) {
6812 pr_warn("Could not find start_pfn for node %d\n", nid);
6813 return 0;
6814 }
6815
6816 return min_pfn;
6817}
6818
6819
6820
6821
6822
6823
6824
6825unsigned long __init find_min_pfn_with_active_regions(void)
6826{
6827 return find_min_pfn_for_node(MAX_NUMNODES);
6828}
6829
6830
6831
6832
6833
6834
6835static unsigned long __init early_calculate_totalpages(void)
6836{
6837 unsigned long totalpages = 0;
6838 unsigned long start_pfn, end_pfn;
6839 int i, nid;
6840
6841 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
6842 unsigned long pages = end_pfn - start_pfn;
6843
6844 totalpages += pages;
6845 if (pages)
6846 node_set_state(nid, N_MEMORY);
6847 }
6848 return totalpages;
6849}
6850
6851
6852
6853
6854
6855
6856
6857static void __init find_zone_movable_pfns_for_nodes(void)
6858{
6859 int i, nid;
6860 unsigned long usable_startpfn;
6861 unsigned long kernelcore_node, kernelcore_remaining;
6862
6863 nodemask_t saved_node_state = node_states[N_MEMORY];
6864 unsigned long totalpages = early_calculate_totalpages();
6865 int usable_nodes = nodes_weight(node_states[N_MEMORY]);
6866 struct memblock_region *r;
6867
6868
6869 find_usable_zone_for_movable();
6870
6871
6872
6873
6874
6875 if (movable_node_is_enabled()) {
6876 for_each_memblock(memory, r) {
6877 if (!memblock_is_hotpluggable(r))
6878 continue;
6879
6880 nid = r->nid;
6881
6882 usable_startpfn = PFN_DOWN(r->base);
6883 zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
6884 min(usable_startpfn, zone_movable_pfn[nid]) :
6885 usable_startpfn;
6886 }
6887
6888 goto out2;
6889 }
6890
6891
6892
6893
6894 if (mirrored_kernelcore) {
6895 bool mem_below_4gb_not_mirrored = false;
6896
6897 for_each_memblock(memory, r) {
6898 if (memblock_is_mirror(r))
6899 continue;
6900
6901 nid = r->nid;
6902
6903 usable_startpfn = memblock_region_memory_base_pfn(r);
6904
6905 if (usable_startpfn < 0x100000) {
6906 mem_below_4gb_not_mirrored = true;
6907 continue;
6908 }
6909
6910 zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
6911 min(usable_startpfn, zone_movable_pfn[nid]) :
6912 usable_startpfn;
6913 }
6914
6915 if (mem_below_4gb_not_mirrored)
6916 pr_warn("This configuration results in unmirrored kernel memory.");
6917
6918 goto out2;
6919 }
6920
6921
6922
6923
6924
6925 if (required_kernelcore_percent)
6926 required_kernelcore = (totalpages * 100 * required_kernelcore_percent) /
6927 10000UL;
6928 if (required_movablecore_percent)
6929 required_movablecore = (totalpages * 100 * required_movablecore_percent) /
6930 10000UL;
6931
6932
6933
6934
6935
6936
6937
6938
6939
6940 if (required_movablecore) {
6941 unsigned long corepages;
6942
6943
6944
6945
6946
6947 required_movablecore =
6948 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
6949 required_movablecore = min(totalpages, required_movablecore);
6950 corepages = totalpages - required_movablecore;
6951
6952 required_kernelcore = max(required_kernelcore, corepages);
6953 }
6954
6955
6956
6957
6958
6959 if (!required_kernelcore || required_kernelcore >= totalpages)
6960 goto out;
6961
6962
6963 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
6964
6965restart:
6966
6967 kernelcore_node = required_kernelcore / usable_nodes;
6968 for_each_node_state(nid, N_MEMORY) {
6969 unsigned long start_pfn, end_pfn;
6970
6971
6972
6973
6974
6975
6976 if (required_kernelcore < kernelcore_node)
6977 kernelcore_node = required_kernelcore / usable_nodes;
6978
6979
6980
6981
6982
6983
6984 kernelcore_remaining = kernelcore_node;
6985
6986
6987 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
6988 unsigned long size_pages;
6989
6990 start_pfn = max(start_pfn, zone_movable_pfn[nid]);
6991 if (start_pfn >= end_pfn)
6992 continue;
6993
6994
6995 if (start_pfn < usable_startpfn) {
6996 unsigned long kernel_pages;
6997 kernel_pages = min(end_pfn, usable_startpfn)
6998 - start_pfn;
6999
7000 kernelcore_remaining -= min(kernel_pages,
7001 kernelcore_remaining);
7002 required_kernelcore -= min(kernel_pages,
7003 required_kernelcore);
7004
7005
7006 if (end_pfn <= usable_startpfn) {
7007
7008
7009
7010
7011
7012
7013
7014 zone_movable_pfn[nid] = end_pfn;
7015 continue;
7016 }
7017 start_pfn = usable_startpfn;
7018 }
7019
7020
7021
7022
7023
7024
7025 size_pages = end_pfn - start_pfn;
7026 if (size_pages > kernelcore_remaining)
7027 size_pages = kernelcore_remaining;
7028 zone_movable_pfn[nid] = start_pfn + size_pages;
7029
7030
7031
7032
7033
7034
7035 required_kernelcore -= min(required_kernelcore,
7036 size_pages);
7037 kernelcore_remaining -= size_pages;
7038 if (!kernelcore_remaining)
7039 break;
7040 }
7041 }
7042
7043
7044
7045
7046
7047
7048
7049 usable_nodes--;
7050 if (usable_nodes && required_kernelcore > usable_nodes)
7051 goto restart;
7052
7053out2:
7054
7055 for (nid = 0; nid < MAX_NUMNODES; nid++)
7056 zone_movable_pfn[nid] =
7057 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
7058
7059out:
7060
7061 node_states[N_MEMORY] = saved_node_state;
7062}
7063
7064
7065static void check_for_memory(pg_data_t *pgdat, int nid)
7066{
7067 enum zone_type zone_type;
7068
7069 for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) {
7070 struct zone *zone = &pgdat->node_zones[zone_type];
7071 if (populated_zone(zone)) {
7072 if (IS_ENABLED(CONFIG_HIGHMEM))
7073 node_set_state(nid, N_HIGH_MEMORY);
7074 if (zone_type <= ZONE_NORMAL)
7075 node_set_state(nid, N_NORMAL_MEMORY);
7076 break;
7077 }
7078 }
7079}
7080
7081
7082
7083
7084
7085
7086
7087
7088
7089
7090
7091
7092
7093
7094void __init free_area_init_nodes(unsigned long *max_zone_pfn)
7095{
7096 unsigned long start_pfn, end_pfn;
7097 int i, nid;
7098
7099
7100 memset(arch_zone_lowest_possible_pfn, 0,
7101 sizeof(arch_zone_lowest_possible_pfn));
7102 memset(arch_zone_highest_possible_pfn, 0,
7103 sizeof(arch_zone_highest_possible_pfn));
7104
7105 start_pfn = find_min_pfn_with_active_regions();
7106
7107 for (i = 0; i < MAX_NR_ZONES; i++) {
7108 if (i == ZONE_MOVABLE)
7109 continue;
7110
7111 end_pfn = max(max_zone_pfn[i], start_pfn);
7112 arch_zone_lowest_possible_pfn[i] = start_pfn;
7113 arch_zone_highest_possible_pfn[i] = end_pfn;
7114
7115 start_pfn = end_pfn;
7116 }
7117
7118
7119 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
7120 find_zone_movable_pfns_for_nodes();
7121
7122
7123 pr_info("Zone ranges:\n");
7124 for (i = 0; i < MAX_NR_ZONES; i++) {
7125 if (i == ZONE_MOVABLE)
7126 continue;
7127 pr_info(" %-8s ", zone_names[i]);
7128 if (arch_zone_lowest_possible_pfn[i] ==
7129 arch_zone_highest_possible_pfn[i])
7130 pr_cont("empty\n");
7131 else
7132 pr_cont("[mem %#018Lx-%#018Lx]\n",
7133 (u64)arch_zone_lowest_possible_pfn[i]
7134 << PAGE_SHIFT,
7135 ((u64)arch_zone_highest_possible_pfn[i]
7136 << PAGE_SHIFT) - 1);
7137 }
7138
7139
7140 pr_info("Movable zone start for each node\n");
7141 for (i = 0; i < MAX_NUMNODES; i++) {
7142 if (zone_movable_pfn[i])
7143 pr_info(" Node %d: %#018Lx\n", i,
7144 (u64)zone_movable_pfn[i] << PAGE_SHIFT);
7145 }
7146
7147
7148 pr_info("Early memory node ranges\n");
7149 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid)
7150 pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid,
7151 (u64)start_pfn << PAGE_SHIFT,
7152 ((u64)end_pfn << PAGE_SHIFT) - 1);
7153
7154
7155 mminit_verify_pageflags_layout();
7156 setup_nr_node_ids();
7157 zero_resv_unavail();
7158 for_each_online_node(nid) {
7159 pg_data_t *pgdat = NODE_DATA(nid);
7160 free_area_init_node(nid, NULL,
7161 find_min_pfn_for_node(nid), NULL);
7162
7163
7164 if (pgdat->node_present_pages)
7165 node_set_state(nid, N_MEMORY);
7166 check_for_memory(pgdat, nid);
7167 }
7168}
7169
7170static int __init cmdline_parse_core(char *p, unsigned long *core,
7171 unsigned long *percent)
7172{
7173 unsigned long long coremem;
7174 char *endptr;
7175
7176 if (!p)
7177 return -EINVAL;
7178
7179
7180 coremem = simple_strtoull(p, &endptr, 0);
7181 if (*endptr == '%') {
7182
7183 WARN_ON(coremem > 100);
7184
7185 *percent = coremem;
7186 } else {
7187 coremem = memparse(p, &p);
7188
7189 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
7190
7191 *core = coremem >> PAGE_SHIFT;
7192 *percent = 0UL;
7193 }
7194 return 0;
7195}
7196
7197
7198
7199
7200
7201static int __init cmdline_parse_kernelcore(char *p)
7202{
7203
7204 if (parse_option_str(p, "mirror")) {
7205 mirrored_kernelcore = true;
7206 return 0;
7207 }
7208
7209 return cmdline_parse_core(p, &required_kernelcore,
7210 &required_kernelcore_percent);
7211}
7212
7213
7214
7215
7216
7217static int __init cmdline_parse_movablecore(char *p)
7218{
7219 return cmdline_parse_core(p, &required_movablecore,
7220 &required_movablecore_percent);
7221}
7222
7223early_param("kernelcore", cmdline_parse_kernelcore);
7224early_param("movablecore", cmdline_parse_movablecore);
7225
7226#endif
7227
7228void adjust_managed_page_count(struct page *page, long count)
7229{
7230 atomic_long_add(count, &page_zone(page)->managed_pages);
7231 totalram_pages_add(count);
7232#ifdef CONFIG_HIGHMEM
7233 if (PageHighMem(page))
7234 totalhigh_pages_add(count);
7235#endif
7236}
7237EXPORT_SYMBOL(adjust_managed_page_count);
7238
7239unsigned long free_reserved_area(void *start, void *end, int poison, const char *s)
7240{
7241 void *pos;
7242 unsigned long pages = 0;
7243
7244 start = (void *)PAGE_ALIGN((unsigned long)start);
7245 end = (void *)((unsigned long)end & PAGE_MASK);
7246 for (pos = start; pos < end; pos += PAGE_SIZE, pages++) {
7247 struct page *page = virt_to_page(pos);
7248 void *direct_map_addr;
7249
7250
7251
7252
7253
7254
7255
7256
7257 direct_map_addr = page_address(page);
7258 if ((unsigned int)poison <= 0xFF)
7259 memset(direct_map_addr, poison, PAGE_SIZE);
7260
7261 free_reserved_page(page);
7262 }
7263
7264 if (pages && s)
7265 pr_info("Freeing %s memory: %ldK\n",
7266 s, pages << (PAGE_SHIFT - 10));
7267
7268 return pages;
7269}
7270EXPORT_SYMBOL(free_reserved_area);
7271
7272#ifdef CONFIG_HIGHMEM
7273void free_highmem_page(struct page *page)
7274{
7275 __free_reserved_page(page);
7276 totalram_pages_inc();
7277 atomic_long_inc(&page_zone(page)->managed_pages);
7278 totalhigh_pages_inc();
7279}
7280#endif
7281
7282
7283void __init mem_init_print_info(const char *str)
7284{
7285 unsigned long physpages, codesize, datasize, rosize, bss_size;
7286 unsigned long init_code_size, init_data_size;
7287
7288 physpages = get_num_physpages();
7289 codesize = _etext - _stext;
7290 datasize = _edata - _sdata;
7291 rosize = __end_rodata - __start_rodata;
7292 bss_size = __bss_stop - __bss_start;
7293 init_data_size = __init_end - __init_begin;
7294 init_code_size = _einittext - _sinittext;
7295
7296
7297
7298
7299
7300
7301
7302
7303#define adj_init_size(start, end, size, pos, adj) \
7304 do { \
7305 if (start <= pos && pos < end && size > adj) \
7306 size -= adj; \
7307 } while (0)
7308
7309 adj_init_size(__init_begin, __init_end, init_data_size,
7310 _sinittext, init_code_size);
7311 adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size);
7312 adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size);
7313 adj_init_size(_stext, _etext, codesize, __start_rodata, rosize);
7314 adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize);
7315
7316#undef adj_init_size
7317
7318 pr_info("Memory: %luK/%luK available (%luK kernel code, %luK rwdata, %luK rodata, %luK init, %luK bss, %luK reserved, %luK cma-reserved"
7319#ifdef CONFIG_HIGHMEM
7320 ", %luK highmem"
7321#endif
7322 "%s%s)\n",
7323 nr_free_pages() << (PAGE_SHIFT - 10),
7324 physpages << (PAGE_SHIFT - 10),
7325 codesize >> 10, datasize >> 10, rosize >> 10,
7326 (init_data_size + init_code_size) >> 10, bss_size >> 10,
7327 (physpages - totalram_pages() - totalcma_pages) << (PAGE_SHIFT - 10),
7328 totalcma_pages << (PAGE_SHIFT - 10),
7329#ifdef CONFIG_HIGHMEM
7330 totalhigh_pages() << (PAGE_SHIFT - 10),
7331#endif
7332 str ? ", " : "", str ? str : "");
7333}
7334
7335
7336
7337
7338
7339
7340
7341
7342
7343
7344
7345
7346void __init set_dma_reserve(unsigned long new_dma_reserve)
7347{
7348 dma_reserve = new_dma_reserve;
7349}
7350
7351void __init free_area_init(unsigned long *zones_size)
7352{
7353 zero_resv_unavail();
7354 free_area_init_node(0, zones_size,
7355 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
7356}
7357
7358static int page_alloc_cpu_dead(unsigned int cpu)
7359{
7360
7361 lru_add_drain_cpu(cpu);
7362 drain_pages(cpu);
7363
7364
7365
7366
7367
7368
7369
7370 vm_events_fold_cpu(cpu);
7371
7372
7373
7374
7375
7376
7377
7378
7379 cpu_vm_stats_fold(cpu);
7380 return 0;
7381}
7382
7383void __init page_alloc_init(void)
7384{
7385 int ret;
7386
7387 ret = cpuhp_setup_state_nocalls(CPUHP_PAGE_ALLOC_DEAD,
7388 "mm/page_alloc:dead", NULL,
7389 page_alloc_cpu_dead);
7390 WARN_ON(ret < 0);
7391}
7392
7393
7394
7395
7396
7397static void calculate_totalreserve_pages(void)
7398{
7399 struct pglist_data *pgdat;
7400 unsigned long reserve_pages = 0;
7401 enum zone_type i, j;
7402
7403 for_each_online_pgdat(pgdat) {
7404
7405 pgdat->totalreserve_pages = 0;
7406
7407 for (i = 0; i < MAX_NR_ZONES; i++) {
7408 struct zone *zone = pgdat->node_zones + i;
7409 long max = 0;
7410 unsigned long managed_pages = zone_managed_pages(zone);
7411
7412
7413 for (j = i; j < MAX_NR_ZONES; j++) {
7414 if (zone->lowmem_reserve[j] > max)
7415 max = zone->lowmem_reserve[j];
7416 }
7417
7418
7419 max += high_wmark_pages(zone);
7420
7421 if (max > managed_pages)
7422 max = managed_pages;
7423
7424 pgdat->totalreserve_pages += max;
7425
7426 reserve_pages += max;
7427 }
7428 }
7429 totalreserve_pages = reserve_pages;
7430}
7431
7432
7433
7434
7435
7436
7437
7438static void setup_per_zone_lowmem_reserve(void)
7439{
7440 struct pglist_data *pgdat;
7441 enum zone_type j, idx;
7442
7443 for_each_online_pgdat(pgdat) {
7444 for (j = 0; j < MAX_NR_ZONES; j++) {
7445 struct zone *zone = pgdat->node_zones + j;
7446 unsigned long managed_pages = zone_managed_pages(zone);
7447
7448 zone->lowmem_reserve[j] = 0;
7449
7450 idx = j;
7451 while (idx) {
7452 struct zone *lower_zone;
7453
7454 idx--;
7455 lower_zone = pgdat->node_zones + idx;
7456
7457 if (sysctl_lowmem_reserve_ratio[idx] < 1) {
7458 sysctl_lowmem_reserve_ratio[idx] = 0;
7459 lower_zone->lowmem_reserve[j] = 0;
7460 } else {
7461 lower_zone->lowmem_reserve[j] =
7462 managed_pages / sysctl_lowmem_reserve_ratio[idx];
7463 }
7464 managed_pages += zone_managed_pages(lower_zone);
7465 }
7466 }
7467 }
7468
7469
7470 calculate_totalreserve_pages();
7471}
7472
7473static void __setup_per_zone_wmarks(void)
7474{
7475 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
7476 unsigned long lowmem_pages = 0;
7477 struct zone *zone;
7478 unsigned long flags;
7479
7480
7481 for_each_zone(zone) {
7482 if (!is_highmem(zone))
7483 lowmem_pages += zone_managed_pages(zone);
7484 }
7485
7486 for_each_zone(zone) {
7487 u64 tmp;
7488
7489 spin_lock_irqsave(&zone->lock, flags);
7490 tmp = (u64)pages_min * zone_managed_pages(zone);
7491 do_div(tmp, lowmem_pages);
7492 if (is_highmem(zone)) {
7493
7494
7495
7496
7497
7498
7499
7500
7501
7502 unsigned long min_pages;
7503
7504 min_pages = zone_managed_pages(zone) / 1024;
7505 min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL);
7506 zone->_watermark[WMARK_MIN] = min_pages;
7507 } else {
7508
7509
7510
7511
7512 zone->_watermark[WMARK_MIN] = tmp;
7513 }
7514
7515
7516
7517
7518
7519
7520 tmp = max_t(u64, tmp >> 2,
7521 mult_frac(zone_managed_pages(zone),
7522 watermark_scale_factor, 10000));
7523
7524 zone->_watermark[WMARK_LOW] = min_wmark_pages(zone) + tmp;
7525 zone->_watermark[WMARK_HIGH] = min_wmark_pages(zone) + tmp * 2;
7526 zone->watermark_boost = 0;
7527
7528 spin_unlock_irqrestore(&zone->lock, flags);
7529 }
7530
7531
7532 calculate_totalreserve_pages();
7533}
7534
7535
7536
7537
7538
7539
7540
7541
7542void setup_per_zone_wmarks(void)
7543{
7544 static DEFINE_SPINLOCK(lock);
7545
7546 spin_lock(&lock);
7547 __setup_per_zone_wmarks();
7548 spin_unlock(&lock);
7549}
7550
7551
7552
7553
7554
7555
7556
7557
7558
7559
7560
7561
7562
7563
7564
7565
7566
7567
7568
7569
7570
7571
7572
7573
7574
7575int __meminit init_per_zone_wmark_min(void)
7576{
7577 unsigned long lowmem_kbytes;
7578 int new_min_free_kbytes;
7579
7580 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
7581 new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
7582
7583 if (new_min_free_kbytes > user_min_free_kbytes) {
7584 min_free_kbytes = new_min_free_kbytes;
7585 if (min_free_kbytes < 128)
7586 min_free_kbytes = 128;
7587 if (min_free_kbytes > 65536)
7588 min_free_kbytes = 65536;
7589 } else {
7590 pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n",
7591 new_min_free_kbytes, user_min_free_kbytes);
7592 }
7593 setup_per_zone_wmarks();
7594 refresh_zone_stat_thresholds();
7595 setup_per_zone_lowmem_reserve();
7596
7597#ifdef CONFIG_NUMA
7598 setup_min_unmapped_ratio();
7599 setup_min_slab_ratio();
7600#endif
7601
7602 return 0;
7603}
7604core_initcall(init_per_zone_wmark_min)
7605
7606
7607
7608
7609
7610
7611int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write,
7612 void __user *buffer, size_t *length, loff_t *ppos)
7613{
7614 int rc;
7615
7616 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
7617 if (rc)
7618 return rc;
7619
7620 if (write) {
7621 user_min_free_kbytes = min_free_kbytes;
7622 setup_per_zone_wmarks();
7623 }
7624 return 0;
7625}
7626
7627int watermark_boost_factor_sysctl_handler(struct ctl_table *table, int write,
7628 void __user *buffer, size_t *length, loff_t *ppos)
7629{
7630 int rc;
7631
7632 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
7633 if (rc)
7634 return rc;
7635
7636 return 0;
7637}
7638
7639int watermark_scale_factor_sysctl_handler(struct ctl_table *table, int write,
7640 void __user *buffer, size_t *length, loff_t *ppos)
7641{
7642 int rc;
7643
7644 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
7645 if (rc)
7646 return rc;
7647
7648 if (write)
7649 setup_per_zone_wmarks();
7650
7651 return 0;
7652}
7653
7654#ifdef CONFIG_NUMA
7655static void setup_min_unmapped_ratio(void)
7656{
7657 pg_data_t *pgdat;
7658 struct zone *zone;
7659
7660 for_each_online_pgdat(pgdat)
7661 pgdat->min_unmapped_pages = 0;
7662
7663 for_each_zone(zone)
7664 zone->zone_pgdat->min_unmapped_pages += (zone_managed_pages(zone) *
7665 sysctl_min_unmapped_ratio) / 100;
7666}
7667
7668
7669int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write,
7670 void __user *buffer, size_t *length, loff_t *ppos)
7671{
7672 int rc;
7673
7674 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
7675 if (rc)
7676 return rc;
7677
7678 setup_min_unmapped_ratio();
7679
7680 return 0;
7681}
7682
7683static void setup_min_slab_ratio(void)
7684{
7685 pg_data_t *pgdat;
7686 struct zone *zone;
7687
7688 for_each_online_pgdat(pgdat)
7689 pgdat->min_slab_pages = 0;
7690
7691 for_each_zone(zone)
7692 zone->zone_pgdat->min_slab_pages += (zone_managed_pages(zone) *
7693 sysctl_min_slab_ratio) / 100;
7694}
7695
7696int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write,
7697 void __user *buffer, size_t *length, loff_t *ppos)
7698{
7699 int rc;
7700
7701 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
7702 if (rc)
7703 return rc;
7704
7705 setup_min_slab_ratio();
7706
7707 return 0;
7708}
7709#endif
7710
7711
7712
7713
7714
7715
7716
7717
7718
7719
7720int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write,
7721 void __user *buffer, size_t *length, loff_t *ppos)
7722{
7723 proc_dointvec_minmax(table, write, buffer, length, ppos);
7724 setup_per_zone_lowmem_reserve();
7725 return 0;
7726}
7727
7728
7729
7730
7731
7732
7733int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *table, int write,
7734 void __user *buffer, size_t *length, loff_t *ppos)
7735{
7736 struct zone *zone;
7737 int old_percpu_pagelist_fraction;
7738 int ret;
7739
7740 mutex_lock(&pcp_batch_high_lock);
7741 old_percpu_pagelist_fraction = percpu_pagelist_fraction;
7742
7743 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
7744 if (!write || ret < 0)
7745 goto out;
7746
7747
7748 if (percpu_pagelist_fraction &&
7749 percpu_pagelist_fraction < MIN_PERCPU_PAGELIST_FRACTION) {
7750 percpu_pagelist_fraction = old_percpu_pagelist_fraction;
7751 ret = -EINVAL;
7752 goto out;
7753 }
7754
7755
7756 if (percpu_pagelist_fraction == old_percpu_pagelist_fraction)
7757 goto out;
7758
7759 for_each_populated_zone(zone) {
7760 unsigned int cpu;
7761
7762 for_each_possible_cpu(cpu)
7763 pageset_set_high_and_batch(zone,
7764 per_cpu_ptr(zone->pageset, cpu));
7765 }
7766out:
7767 mutex_unlock(&pcp_batch_high_lock);
7768 return ret;
7769}
7770
7771#ifdef CONFIG_NUMA
7772int hashdist = HASHDIST_DEFAULT;
7773
7774static int __init set_hashdist(char *str)
7775{
7776 if (!str)
7777 return 0;
7778 hashdist = simple_strtoul(str, &str, 0);
7779 return 1;
7780}
7781__setup("hashdist=", set_hashdist);
7782#endif
7783
7784#ifndef __HAVE_ARCH_RESERVED_KERNEL_PAGES
7785
7786
7787
7788
7789static unsigned long __init arch_reserved_kernel_pages(void)
7790{
7791 return 0;
7792}
7793#endif
7794
7795
7796
7797
7798
7799
7800
7801
7802
7803
7804#if __BITS_PER_LONG > 32
7805#define ADAPT_SCALE_BASE (64ul << 30)
7806#define ADAPT_SCALE_SHIFT 2
7807#define ADAPT_SCALE_NPAGES (ADAPT_SCALE_BASE >> PAGE_SHIFT)
7808#endif
7809
7810
7811
7812
7813
7814
7815
7816void *__init alloc_large_system_hash(const char *tablename,
7817 unsigned long bucketsize,
7818 unsigned long numentries,
7819 int scale,
7820 int flags,
7821 unsigned int *_hash_shift,
7822 unsigned int *_hash_mask,
7823 unsigned long low_limit,
7824 unsigned long high_limit)
7825{
7826 unsigned long long max = high_limit;
7827 unsigned long log2qty, size;
7828 void *table = NULL;
7829 gfp_t gfp_flags;
7830
7831
7832 if (!numentries) {
7833
7834 numentries = nr_kernel_pages;
7835 numentries -= arch_reserved_kernel_pages();
7836
7837
7838 if (PAGE_SHIFT < 20)
7839 numentries = round_up(numentries, (1<<20)/PAGE_SIZE);
7840
7841#if __BITS_PER_LONG > 32
7842 if (!high_limit) {
7843 unsigned long adapt;
7844
7845 for (adapt = ADAPT_SCALE_NPAGES; adapt < numentries;
7846 adapt <<= ADAPT_SCALE_SHIFT)
7847 scale++;
7848 }
7849#endif
7850
7851
7852 if (scale > PAGE_SHIFT)
7853 numentries >>= (scale - PAGE_SHIFT);
7854 else
7855 numentries <<= (PAGE_SHIFT - scale);
7856
7857
7858 if (unlikely(flags & HASH_SMALL)) {
7859
7860 WARN_ON(!(flags & HASH_EARLY));
7861 if (!(numentries >> *_hash_shift)) {
7862 numentries = 1UL << *_hash_shift;
7863 BUG_ON(!numentries);
7864 }
7865 } else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
7866 numentries = PAGE_SIZE / bucketsize;
7867 }
7868 numentries = roundup_pow_of_two(numentries);
7869
7870
7871 if (max == 0) {
7872 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
7873 do_div(max, bucketsize);
7874 }
7875 max = min(max, 0x80000000ULL);
7876
7877 if (numentries < low_limit)
7878 numentries = low_limit;
7879 if (numentries > max)
7880 numentries = max;
7881
7882 log2qty = ilog2(numentries);
7883
7884 gfp_flags = (flags & HASH_ZERO) ? GFP_ATOMIC | __GFP_ZERO : GFP_ATOMIC;
7885 do {
7886 size = bucketsize << log2qty;
7887 if (flags & HASH_EARLY) {
7888 if (flags & HASH_ZERO)
7889 table = memblock_alloc_nopanic(size,
7890 SMP_CACHE_BYTES);
7891 else
7892 table = memblock_alloc_raw(size,
7893 SMP_CACHE_BYTES);
7894 } else if (hashdist) {
7895 table = __vmalloc(size, gfp_flags, PAGE_KERNEL);
7896 } else {
7897
7898
7899
7900
7901
7902 if (get_order(size) < MAX_ORDER) {
7903 table = alloc_pages_exact(size, gfp_flags);
7904 kmemleak_alloc(table, size, 1, gfp_flags);
7905 }
7906 }
7907 } while (!table && size > PAGE_SIZE && --log2qty);
7908
7909 if (!table)
7910 panic("Failed to allocate %s hash table\n", tablename);
7911
7912 pr_info("%s hash table entries: %ld (order: %d, %lu bytes)\n",
7913 tablename, 1UL << log2qty, ilog2(size) - PAGE_SHIFT, size);
7914
7915 if (_hash_shift)
7916 *_hash_shift = log2qty;
7917 if (_hash_mask)
7918 *_hash_mask = (1 << log2qty) - 1;
7919
7920 return table;
7921}
7922
7923
7924
7925
7926
7927
7928
7929
7930
7931
7932bool has_unmovable_pages(struct zone *zone, struct page *page, int count,
7933 int migratetype, int flags)
7934{
7935 unsigned long pfn, iter, found;
7936
7937
7938
7939
7940
7941
7942
7943
7944
7945
7946
7947
7948
7949
7950 if (is_migrate_cma(migratetype) &&
7951 is_migrate_cma(get_pageblock_migratetype(page)))
7952 return false;
7953
7954 pfn = page_to_pfn(page);
7955 for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) {
7956 unsigned long check = pfn + iter;
7957
7958 if (!pfn_valid_within(check))
7959 continue;
7960
7961 page = pfn_to_page(check);
7962
7963 if (PageReserved(page))
7964 goto unmovable;
7965
7966
7967
7968
7969
7970
7971 if (zone_idx(zone) == ZONE_MOVABLE)
7972 continue;
7973
7974
7975
7976
7977
7978
7979 if (PageHuge(page)) {
7980 struct page *head = compound_head(page);
7981 unsigned int skip_pages;
7982
7983 if (!hugepage_migration_supported(page_hstate(head)))
7984 goto unmovable;
7985
7986 skip_pages = (1 << compound_order(head)) - (page - head);
7987 iter += skip_pages - 1;
7988 continue;
7989 }
7990
7991
7992
7993
7994
7995
7996
7997 if (!page_ref_count(page)) {
7998 if (PageBuddy(page))
7999 iter += (1 << page_order(page)) - 1;
8000 continue;
8001 }
8002
8003
8004
8005
8006
8007 if ((flags & SKIP_HWPOISON) && PageHWPoison(page))
8008 continue;
8009
8010 if (__PageMovable(page))
8011 continue;
8012
8013 if (!PageLRU(page))
8014 found++;
8015
8016
8017
8018
8019
8020
8021
8022
8023
8024
8025
8026
8027
8028 if (found > count)
8029 goto unmovable;
8030 }
8031 return false;
8032unmovable:
8033 WARN_ON_ONCE(zone_idx(zone) == ZONE_MOVABLE);
8034 if (flags & REPORT_FAILURE)
8035 dump_page(pfn_to_page(pfn+iter), "unmovable page");
8036 return true;
8037}
8038
8039#if (defined(CONFIG_MEMORY_ISOLATION) && defined(CONFIG_COMPACTION)) || defined(CONFIG_CMA)
8040
8041static unsigned long pfn_max_align_down(unsigned long pfn)
8042{
8043 return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES,
8044 pageblock_nr_pages) - 1);
8045}
8046
8047static unsigned long pfn_max_align_up(unsigned long pfn)
8048{
8049 return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES,
8050 pageblock_nr_pages));
8051}
8052
8053
8054static int __alloc_contig_migrate_range(struct compact_control *cc,
8055 unsigned long start, unsigned long end)
8056{
8057
8058 unsigned long nr_reclaimed;
8059 unsigned long pfn = start;
8060 unsigned int tries = 0;
8061 int ret = 0;
8062
8063 migrate_prep();
8064
8065 while (pfn < end || !list_empty(&cc->migratepages)) {
8066 if (fatal_signal_pending(current)) {
8067 ret = -EINTR;
8068 break;
8069 }
8070
8071 if (list_empty(&cc->migratepages)) {
8072 cc->nr_migratepages = 0;
8073 pfn = isolate_migratepages_range(cc, pfn, end);
8074 if (!pfn) {
8075 ret = -EINTR;
8076 break;
8077 }
8078 tries = 0;
8079 } else if (++tries == 5) {
8080 ret = ret < 0 ? ret : -EBUSY;
8081 break;
8082 }
8083
8084 nr_reclaimed = reclaim_clean_pages_from_list(cc->zone,
8085 &cc->migratepages);
8086 cc->nr_migratepages -= nr_reclaimed;
8087
8088 ret = migrate_pages(&cc->migratepages, alloc_migrate_target,
8089 NULL, 0, cc->mode, MR_CONTIG_RANGE);
8090 }
8091 if (ret < 0) {
8092 putback_movable_pages(&cc->migratepages);
8093 return ret;
8094 }
8095 return 0;
8096}
8097
8098
8099
8100
8101
8102
8103
8104
8105
8106
8107
8108
8109
8110
8111
8112
8113
8114
8115
8116
8117
8118
8119int alloc_contig_range(unsigned long start, unsigned long end,
8120 unsigned migratetype, gfp_t gfp_mask)
8121{
8122 unsigned long outer_start, outer_end;
8123 unsigned int order;
8124 int ret = 0;
8125
8126 struct compact_control cc = {
8127 .nr_migratepages = 0,
8128 .order = -1,
8129 .zone = page_zone(pfn_to_page(start)),
8130 .mode = MIGRATE_SYNC,
8131 .ignore_skip_hint = true,
8132 .no_set_skip_hint = true,
8133 .gfp_mask = current_gfp_context(gfp_mask),
8134 };
8135 INIT_LIST_HEAD(&cc.migratepages);
8136
8137
8138
8139
8140
8141
8142
8143
8144
8145
8146
8147
8148
8149
8150
8151
8152
8153
8154
8155
8156
8157
8158
8159
8160
8161 ret = start_isolate_page_range(pfn_max_align_down(start),
8162 pfn_max_align_up(end), migratetype, 0);
8163 if (ret)
8164 return ret;
8165
8166
8167
8168
8169
8170
8171
8172
8173
8174
8175
8176 ret = __alloc_contig_migrate_range(&cc, start, end);
8177 if (ret && ret != -EBUSY)
8178 goto done;
8179 ret =0;
8180
8181
8182
8183
8184
8185
8186
8187
8188
8189
8190
8191
8192
8193
8194
8195
8196
8197
8198 lru_add_drain_all();
8199 drain_all_pages(cc.zone);
8200
8201 order = 0;
8202 outer_start = start;
8203 while (!PageBuddy(pfn_to_page(outer_start))) {
8204 if (++order >= MAX_ORDER) {
8205 outer_start = start;
8206 break;
8207 }
8208 outer_start &= ~0UL << order;
8209 }
8210
8211 if (outer_start != start) {
8212 order = page_order(pfn_to_page(outer_start));
8213
8214
8215
8216
8217
8218
8219
8220 if (outer_start + (1UL << order) <= start)
8221 outer_start = start;
8222 }
8223
8224
8225 if (test_pages_isolated(outer_start, end, false)) {
8226 pr_info_ratelimited("%s: [%lx, %lx) PFNs busy\n",
8227 __func__, outer_start, end);
8228 ret = -EBUSY;
8229 goto done;
8230 }
8231
8232
8233 outer_end = isolate_freepages_range(&cc, outer_start, end);
8234 if (!outer_end) {
8235 ret = -EBUSY;
8236 goto done;
8237 }
8238
8239
8240 if (start != outer_start)
8241 free_contig_range(outer_start, start - outer_start);
8242 if (end != outer_end)
8243 free_contig_range(end, outer_end - end);
8244
8245done:
8246 undo_isolate_page_range(pfn_max_align_down(start),
8247 pfn_max_align_up(end), migratetype);
8248 return ret;
8249}
8250
8251void free_contig_range(unsigned long pfn, unsigned nr_pages)
8252{
8253 unsigned int count = 0;
8254
8255 for (; nr_pages--; pfn++) {
8256 struct page *page = pfn_to_page(pfn);
8257
8258 count += page_count(page) != 1;
8259 __free_page(page);
8260 }
8261 WARN(count != 0, "%d pages are still in use!\n", count);
8262}
8263#endif
8264
8265#ifdef CONFIG_MEMORY_HOTPLUG
8266
8267
8268
8269
8270void __meminit zone_pcp_update(struct zone *zone)
8271{
8272 unsigned cpu;
8273 mutex_lock(&pcp_batch_high_lock);
8274 for_each_possible_cpu(cpu)
8275 pageset_set_high_and_batch(zone,
8276 per_cpu_ptr(zone->pageset, cpu));
8277 mutex_unlock(&pcp_batch_high_lock);
8278}
8279#endif
8280
8281void zone_pcp_reset(struct zone *zone)
8282{
8283 unsigned long flags;
8284 int cpu;
8285 struct per_cpu_pageset *pset;
8286
8287
8288 local_irq_save(flags);
8289 if (zone->pageset != &boot_pageset) {
8290 for_each_online_cpu(cpu) {
8291 pset = per_cpu_ptr(zone->pageset, cpu);
8292 drain_zonestat(zone, pset);
8293 }
8294 free_percpu(zone->pageset);
8295 zone->pageset = &boot_pageset;
8296 }
8297 local_irq_restore(flags);
8298}
8299
8300#ifdef CONFIG_MEMORY_HOTREMOVE
8301
8302
8303
8304
8305void
8306__offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
8307{
8308 struct page *page;
8309 struct zone *zone;
8310 unsigned int order, i;
8311 unsigned long pfn;
8312 unsigned long flags;
8313
8314 for (pfn = start_pfn; pfn < end_pfn; pfn++)
8315 if (pfn_valid(pfn))
8316 break;
8317 if (pfn == end_pfn)
8318 return;
8319 offline_mem_sections(pfn, end_pfn);
8320 zone = page_zone(pfn_to_page(pfn));
8321 spin_lock_irqsave(&zone->lock, flags);
8322 pfn = start_pfn;
8323 while (pfn < end_pfn) {
8324 if (!pfn_valid(pfn)) {
8325 pfn++;
8326 continue;
8327 }
8328 page = pfn_to_page(pfn);
8329
8330
8331
8332
8333 if (unlikely(!PageBuddy(page) && PageHWPoison(page))) {
8334 pfn++;
8335 SetPageReserved(page);
8336 continue;
8337 }
8338
8339 BUG_ON(page_count(page));
8340 BUG_ON(!PageBuddy(page));
8341 order = page_order(page);
8342#ifdef CONFIG_DEBUG_VM
8343 pr_info("remove from free list %lx %d %lx\n",
8344 pfn, 1 << order, end_pfn);
8345#endif
8346 list_del(&page->lru);
8347 rmv_page_order(page);
8348 zone->free_area[order].nr_free--;
8349 for (i = 0; i < (1 << order); i++)
8350 SetPageReserved((page+i));
8351 pfn += (1 << order);
8352 }
8353 spin_unlock_irqrestore(&zone->lock, flags);
8354}
8355#endif
8356
8357bool is_free_buddy_page(struct page *page)
8358{
8359 struct zone *zone = page_zone(page);
8360 unsigned long pfn = page_to_pfn(page);
8361 unsigned long flags;
8362 unsigned int order;
8363
8364 spin_lock_irqsave(&zone->lock, flags);
8365 for (order = 0; order < MAX_ORDER; order++) {
8366 struct page *page_head = page - (pfn & ((1 << order) - 1));
8367
8368 if (PageBuddy(page_head) && page_order(page_head) >= order)
8369 break;
8370 }
8371 spin_unlock_irqrestore(&zone->lock, flags);
8372
8373 return order < MAX_ORDER;
8374}
8375
8376#ifdef CONFIG_MEMORY_FAILURE
8377
8378
8379
8380
8381
8382bool set_hwpoison_free_buddy_page(struct page *page)
8383{
8384 struct zone *zone = page_zone(page);
8385 unsigned long pfn = page_to_pfn(page);
8386 unsigned long flags;
8387 unsigned int order;
8388 bool hwpoisoned = false;
8389
8390 spin_lock_irqsave(&zone->lock, flags);
8391 for (order = 0; order < MAX_ORDER; order++) {
8392 struct page *page_head = page - (pfn & ((1 << order) - 1));
8393
8394 if (PageBuddy(page_head) && page_order(page_head) >= order) {
8395 if (!TestSetPageHWPoison(page))
8396 hwpoisoned = true;
8397 break;
8398 }
8399 }
8400 spin_unlock_irqrestore(&zone->lock, flags);
8401
8402 return hwpoisoned;
8403}
8404#endif
8405