1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18#include <linux/stddef.h>
19#include <linux/mm.h>
20#include <linux/highmem.h>
21#include <linux/swap.h>
22#include <linux/interrupt.h>
23#include <linux/pagemap.h>
24#include <linux/jiffies.h>
25#include <linux/memblock.h>
26#include <linux/compiler.h>
27#include <linux/kernel.h>
28#include <linux/kasan.h>
29#include <linux/module.h>
30#include <linux/suspend.h>
31#include <linux/pagevec.h>
32#include <linux/blkdev.h>
33#include <linux/slab.h>
34#include <linux/ratelimit.h>
35#include <linux/oom.h>
36#include <linux/topology.h>
37#include <linux/sysctl.h>
38#include <linux/cpu.h>
39#include <linux/cpuset.h>
40#include <linux/memory_hotplug.h>
41#include <linux/nodemask.h>
42#include <linux/vmalloc.h>
43#include <linux/vmstat.h>
44#include <linux/mempolicy.h>
45#include <linux/memremap.h>
46#include <linux/stop_machine.h>
47#include <linux/random.h>
48#include <linux/sort.h>
49#include <linux/pfn.h>
50#include <linux/backing-dev.h>
51#include <linux/fault-inject.h>
52#include <linux/page-isolation.h>
53#include <linux/debugobjects.h>
54#include <linux/kmemleak.h>
55#include <linux/compaction.h>
56#include <trace/events/kmem.h>
57#include <trace/events/oom.h>
58#include <linux/prefetch.h>
59#include <linux/mm_inline.h>
60#include <linux/mmu_notifier.h>
61#include <linux/migrate.h>
62#include <linux/hugetlb.h>
63#include <linux/sched/rt.h>
64#include <linux/sched/mm.h>
65#include <linux/page_owner.h>
66#include <linux/kthread.h>
67#include <linux/memcontrol.h>
68#include <linux/ftrace.h>
69#include <linux/lockdep.h>
70#include <linux/nmi.h>
71#include <linux/psi.h>
72#include <linux/padata.h>
73#include <linux/khugepaged.h>
74#include <linux/buffer_head.h>
75#include <asm/sections.h>
76#include <asm/tlbflush.h>
77#include <asm/div64.h>
78#include "internal.h"
79#include "shuffle.h"
80#include "page_reporting.h"
81
82
83typedef int __bitwise fpi_t;
84
85
86#define FPI_NONE ((__force fpi_t)0)
87
88
89
90
91
92
93
94
95
96#define FPI_SKIP_REPORT_NOTIFY ((__force fpi_t)BIT(0))
97
98
99
100
101
102
103
104
105
106
107
108#define FPI_TO_TAIL ((__force fpi_t)BIT(1))
109
110
111
112
113
114
115
116
117
118
119#define FPI_SKIP_KASAN_POISON ((__force fpi_t)BIT(2))
120
121
122static DEFINE_MUTEX(pcp_batch_high_lock);
123#define MIN_PERCPU_PAGELIST_HIGH_FRACTION (8)
124
125struct pagesets {
126 local_lock_t lock;
127};
128static DEFINE_PER_CPU(struct pagesets, pagesets) = {
129 .lock = INIT_LOCAL_LOCK(lock),
130};
131
132#ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
133DEFINE_PER_CPU(int, numa_node);
134EXPORT_PER_CPU_SYMBOL(numa_node);
135#endif
136
137DEFINE_STATIC_KEY_TRUE(vm_numa_stat_key);
138
139#ifdef CONFIG_HAVE_MEMORYLESS_NODES
140
141
142
143
144
145
146DEFINE_PER_CPU(int, _numa_mem_);
147EXPORT_PER_CPU_SYMBOL(_numa_mem_);
148#endif
149
150
151struct pcpu_drain {
152 struct zone *zone;
153 struct work_struct work;
154};
155static DEFINE_MUTEX(pcpu_drain_mutex);
156static DEFINE_PER_CPU(struct pcpu_drain, pcpu_drain);
157
158#ifdef CONFIG_GCC_PLUGIN_LATENT_ENTROPY
159volatile unsigned long latent_entropy __latent_entropy;
160EXPORT_SYMBOL(latent_entropy);
161#endif
162
163
164
165
166nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
167 [N_POSSIBLE] = NODE_MASK_ALL,
168 [N_ONLINE] = { { [0] = 1UL } },
169#ifndef CONFIG_NUMA
170 [N_NORMAL_MEMORY] = { { [0] = 1UL } },
171#ifdef CONFIG_HIGHMEM
172 [N_HIGH_MEMORY] = { { [0] = 1UL } },
173#endif
174 [N_MEMORY] = { { [0] = 1UL } },
175 [N_CPU] = { { [0] = 1UL } },
176#endif
177};
178EXPORT_SYMBOL(node_states);
179
180atomic_long_t _totalram_pages __read_mostly;
181EXPORT_SYMBOL(_totalram_pages);
182unsigned long totalreserve_pages __read_mostly;
183unsigned long totalcma_pages __read_mostly;
184
185int percpu_pagelist_high_fraction;
186gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK;
187DEFINE_STATIC_KEY_MAYBE(CONFIG_INIT_ON_ALLOC_DEFAULT_ON, init_on_alloc);
188EXPORT_SYMBOL(init_on_alloc);
189
190DEFINE_STATIC_KEY_MAYBE(CONFIG_INIT_ON_FREE_DEFAULT_ON, init_on_free);
191EXPORT_SYMBOL(init_on_free);
192
193static bool _init_on_alloc_enabled_early __read_mostly
194 = IS_ENABLED(CONFIG_INIT_ON_ALLOC_DEFAULT_ON);
195static int __init early_init_on_alloc(char *buf)
196{
197
198 return kstrtobool(buf, &_init_on_alloc_enabled_early);
199}
200early_param("init_on_alloc", early_init_on_alloc);
201
202static bool _init_on_free_enabled_early __read_mostly
203 = IS_ENABLED(CONFIG_INIT_ON_FREE_DEFAULT_ON);
204static int __init early_init_on_free(char *buf)
205{
206 return kstrtobool(buf, &_init_on_free_enabled_early);
207}
208early_param("init_on_free", early_init_on_free);
209
210
211
212
213
214
215
216
217
218static inline int get_pcppage_migratetype(struct page *page)
219{
220 return page->index;
221}
222
223static inline void set_pcppage_migratetype(struct page *page, int migratetype)
224{
225 page->index = migratetype;
226}
227
228#ifdef CONFIG_PM_SLEEP
229
230
231
232
233
234
235
236
237
238
239static gfp_t saved_gfp_mask;
240
241void pm_restore_gfp_mask(void)
242{
243 WARN_ON(!mutex_is_locked(&system_transition_mutex));
244 if (saved_gfp_mask) {
245 gfp_allowed_mask = saved_gfp_mask;
246 saved_gfp_mask = 0;
247 }
248}
249
250void pm_restrict_gfp_mask(void)
251{
252 WARN_ON(!mutex_is_locked(&system_transition_mutex));
253 WARN_ON(saved_gfp_mask);
254 saved_gfp_mask = gfp_allowed_mask;
255 gfp_allowed_mask &= ~(__GFP_IO | __GFP_FS);
256}
257
258bool pm_suspended_storage(void)
259{
260 if ((gfp_allowed_mask & (__GFP_IO | __GFP_FS)) == (__GFP_IO | __GFP_FS))
261 return false;
262 return true;
263}
264#endif
265
266#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
267unsigned int pageblock_order __read_mostly;
268#endif
269
270static void __free_pages_ok(struct page *page, unsigned int order,
271 fpi_t fpi_flags);
272
273
274
275
276
277
278
279
280
281
282
283
284int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES] = {
285#ifdef CONFIG_ZONE_DMA
286 [ZONE_DMA] = 256,
287#endif
288#ifdef CONFIG_ZONE_DMA32
289 [ZONE_DMA32] = 256,
290#endif
291 [ZONE_NORMAL] = 32,
292#ifdef CONFIG_HIGHMEM
293 [ZONE_HIGHMEM] = 0,
294#endif
295 [ZONE_MOVABLE] = 0,
296};
297
298static char * const zone_names[MAX_NR_ZONES] = {
299#ifdef CONFIG_ZONE_DMA
300 "DMA",
301#endif
302#ifdef CONFIG_ZONE_DMA32
303 "DMA32",
304#endif
305 "Normal",
306#ifdef CONFIG_HIGHMEM
307 "HighMem",
308#endif
309 "Movable",
310#ifdef CONFIG_ZONE_DEVICE
311 "Device",
312#endif
313};
314
315const char * const migratetype_names[MIGRATE_TYPES] = {
316 "Unmovable",
317 "Movable",
318 "Reclaimable",
319 "HighAtomic",
320#ifdef CONFIG_CMA
321 "CMA",
322#endif
323#ifdef CONFIG_MEMORY_ISOLATION
324 "Isolate",
325#endif
326};
327
328compound_page_dtor * const compound_page_dtors[NR_COMPOUND_DTORS] = {
329 [NULL_COMPOUND_DTOR] = NULL,
330 [COMPOUND_PAGE_DTOR] = free_compound_page,
331#ifdef CONFIG_HUGETLB_PAGE
332 [HUGETLB_PAGE_DTOR] = free_huge_page,
333#endif
334#ifdef CONFIG_TRANSPARENT_HUGEPAGE
335 [TRANSHUGE_PAGE_DTOR] = free_transhuge_page,
336#endif
337};
338
339int min_free_kbytes = 1024;
340int user_min_free_kbytes = -1;
341int watermark_boost_factor __read_mostly = 15000;
342int watermark_scale_factor = 10;
343
344static unsigned long nr_kernel_pages __initdata;
345static unsigned long nr_all_pages __initdata;
346static unsigned long dma_reserve __initdata;
347
348static unsigned long arch_zone_lowest_possible_pfn[MAX_NR_ZONES] __initdata;
349static unsigned long arch_zone_highest_possible_pfn[MAX_NR_ZONES] __initdata;
350static unsigned long required_kernelcore __initdata;
351static unsigned long required_kernelcore_percent __initdata;
352static unsigned long required_movablecore __initdata;
353static unsigned long required_movablecore_percent __initdata;
354static unsigned long zone_movable_pfn[MAX_NUMNODES] __initdata;
355static bool mirrored_kernelcore __meminitdata;
356
357
358int movable_zone;
359EXPORT_SYMBOL(movable_zone);
360
361#if MAX_NUMNODES > 1
362unsigned int nr_node_ids __read_mostly = MAX_NUMNODES;
363unsigned int nr_online_nodes __read_mostly = 1;
364EXPORT_SYMBOL(nr_node_ids);
365EXPORT_SYMBOL(nr_online_nodes);
366#endif
367
368int page_group_by_mobility_disabled __read_mostly;
369
370#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
371
372
373
374
375
376static DEFINE_STATIC_KEY_TRUE(deferred_pages);
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391static inline bool should_skip_kasan_poison(struct page *page, fpi_t fpi_flags)
392{
393 return static_branch_unlikely(&deferred_pages) ||
394 (!IS_ENABLED(CONFIG_KASAN_GENERIC) &&
395 (fpi_flags & FPI_SKIP_KASAN_POISON)) ||
396 PageSkipKASanPoison(page);
397}
398
399
400static inline bool __meminit early_page_uninitialised(unsigned long pfn)
401{
402 int nid = early_pfn_to_nid(pfn);
403
404 if (node_online(nid) && pfn >= NODE_DATA(nid)->first_deferred_pfn)
405 return true;
406
407 return false;
408}
409
410
411
412
413
414static bool __meminit
415defer_init(int nid, unsigned long pfn, unsigned long end_pfn)
416{
417 static unsigned long prev_end_pfn, nr_initialised;
418
419
420
421
422
423 if (prev_end_pfn != end_pfn) {
424 prev_end_pfn = end_pfn;
425 nr_initialised = 0;
426 }
427
428
429 if (end_pfn < pgdat_end_pfn(NODE_DATA(nid)))
430 return false;
431
432 if (NODE_DATA(nid)->first_deferred_pfn != ULONG_MAX)
433 return true;
434
435
436
437
438 nr_initialised++;
439 if ((nr_initialised > PAGES_PER_SECTION) &&
440 (pfn & (PAGES_PER_SECTION - 1)) == 0) {
441 NODE_DATA(nid)->first_deferred_pfn = pfn;
442 return true;
443 }
444 return false;
445}
446#else
447static inline bool should_skip_kasan_poison(struct page *page, fpi_t fpi_flags)
448{
449 return (!IS_ENABLED(CONFIG_KASAN_GENERIC) &&
450 (fpi_flags & FPI_SKIP_KASAN_POISON)) ||
451 PageSkipKASanPoison(page);
452}
453
454static inline bool early_page_uninitialised(unsigned long pfn)
455{
456 return false;
457}
458
459static inline bool defer_init(int nid, unsigned long pfn, unsigned long end_pfn)
460{
461 return false;
462}
463#endif
464
465
466static inline unsigned long *get_pageblock_bitmap(const struct page *page,
467 unsigned long pfn)
468{
469#ifdef CONFIG_SPARSEMEM
470 return section_to_usemap(__pfn_to_section(pfn));
471#else
472 return page_zone(page)->pageblock_flags;
473#endif
474}
475
476static inline int pfn_to_bitidx(const struct page *page, unsigned long pfn)
477{
478#ifdef CONFIG_SPARSEMEM
479 pfn &= (PAGES_PER_SECTION-1);
480#else
481 pfn = pfn - round_down(page_zone(page)->zone_start_pfn, pageblock_nr_pages);
482#endif
483 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
484}
485
486static __always_inline
487unsigned long __get_pfnblock_flags_mask(const struct page *page,
488 unsigned long pfn,
489 unsigned long mask)
490{
491 unsigned long *bitmap;
492 unsigned long bitidx, word_bitidx;
493 unsigned long word;
494
495 bitmap = get_pageblock_bitmap(page, pfn);
496 bitidx = pfn_to_bitidx(page, pfn);
497 word_bitidx = bitidx / BITS_PER_LONG;
498 bitidx &= (BITS_PER_LONG-1);
499
500 word = bitmap[word_bitidx];
501 return (word >> bitidx) & mask;
502}
503
504
505
506
507
508
509
510
511
512unsigned long get_pfnblock_flags_mask(const struct page *page,
513 unsigned long pfn, unsigned long mask)
514{
515 return __get_pfnblock_flags_mask(page, pfn, mask);
516}
517
518static __always_inline int get_pfnblock_migratetype(const struct page *page,
519 unsigned long pfn)
520{
521 return __get_pfnblock_flags_mask(page, pfn, MIGRATETYPE_MASK);
522}
523
524
525
526
527
528
529
530
531void set_pfnblock_flags_mask(struct page *page, unsigned long flags,
532 unsigned long pfn,
533 unsigned long mask)
534{
535 unsigned long *bitmap;
536 unsigned long bitidx, word_bitidx;
537 unsigned long old_word, word;
538
539 BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4);
540 BUILD_BUG_ON(MIGRATE_TYPES > (1 << PB_migratetype_bits));
541
542 bitmap = get_pageblock_bitmap(page, pfn);
543 bitidx = pfn_to_bitidx(page, pfn);
544 word_bitidx = bitidx / BITS_PER_LONG;
545 bitidx &= (BITS_PER_LONG-1);
546
547 VM_BUG_ON_PAGE(!zone_spans_pfn(page_zone(page), pfn), page);
548
549 mask <<= bitidx;
550 flags <<= bitidx;
551
552 word = READ_ONCE(bitmap[word_bitidx]);
553 for (;;) {
554 old_word = cmpxchg(&bitmap[word_bitidx], word, (word & ~mask) | flags);
555 if (word == old_word)
556 break;
557 word = old_word;
558 }
559}
560
561void set_pageblock_migratetype(struct page *page, int migratetype)
562{
563 if (unlikely(page_group_by_mobility_disabled &&
564 migratetype < MIGRATE_PCPTYPES))
565 migratetype = MIGRATE_UNMOVABLE;
566
567 set_pfnblock_flags_mask(page, (unsigned long)migratetype,
568 page_to_pfn(page), MIGRATETYPE_MASK);
569}
570
571#ifdef CONFIG_DEBUG_VM
572static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
573{
574 int ret = 0;
575 unsigned seq;
576 unsigned long pfn = page_to_pfn(page);
577 unsigned long sp, start_pfn;
578
579 do {
580 seq = zone_span_seqbegin(zone);
581 start_pfn = zone->zone_start_pfn;
582 sp = zone->spanned_pages;
583 if (!zone_spans_pfn(zone, pfn))
584 ret = 1;
585 } while (zone_span_seqretry(zone, seq));
586
587 if (ret)
588 pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n",
589 pfn, zone_to_nid(zone), zone->name,
590 start_pfn, start_pfn + sp);
591
592 return ret;
593}
594
595static int page_is_consistent(struct zone *zone, struct page *page)
596{
597 if (!pfn_valid_within(page_to_pfn(page)))
598 return 0;
599 if (zone != page_zone(page))
600 return 0;
601
602 return 1;
603}
604
605
606
607static int __maybe_unused bad_range(struct zone *zone, struct page *page)
608{
609 if (page_outside_zone_boundaries(zone, page))
610 return 1;
611 if (!page_is_consistent(zone, page))
612 return 1;
613
614 return 0;
615}
616#else
617static inline int __maybe_unused bad_range(struct zone *zone, struct page *page)
618{
619 return 0;
620}
621#endif
622
623static void bad_page(struct page *page, const char *reason)
624{
625 static unsigned long resume;
626 static unsigned long nr_shown;
627 static unsigned long nr_unshown;
628
629
630
631
632
633 if (nr_shown == 60) {
634 if (time_before(jiffies, resume)) {
635 nr_unshown++;
636 goto out;
637 }
638 if (nr_unshown) {
639 pr_alert(
640 "BUG: Bad page state: %lu messages suppressed\n",
641 nr_unshown);
642 nr_unshown = 0;
643 }
644 nr_shown = 0;
645 }
646 if (nr_shown++ == 0)
647 resume = jiffies + 60 * HZ;
648
649 pr_alert("BUG: Bad page state in process %s pfn:%05lx\n",
650 current->comm, page_to_pfn(page));
651 dump_page(page, reason);
652
653 print_modules();
654 dump_stack();
655out:
656
657 page_mapcount_reset(page);
658 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
659}
660
661static inline unsigned int order_to_pindex(int migratetype, int order)
662{
663 int base = order;
664
665#ifdef CONFIG_TRANSPARENT_HUGEPAGE
666 if (order > PAGE_ALLOC_COSTLY_ORDER) {
667 VM_BUG_ON(order != pageblock_order);
668 base = PAGE_ALLOC_COSTLY_ORDER + 1;
669 }
670#else
671 VM_BUG_ON(order > PAGE_ALLOC_COSTLY_ORDER);
672#endif
673
674 return (MIGRATE_PCPTYPES * base) + migratetype;
675}
676
677static inline int pindex_to_order(unsigned int pindex)
678{
679 int order = pindex / MIGRATE_PCPTYPES;
680
681#ifdef CONFIG_TRANSPARENT_HUGEPAGE
682 if (order > PAGE_ALLOC_COSTLY_ORDER) {
683 order = pageblock_order;
684 VM_BUG_ON(order != pageblock_order);
685 }
686#else
687 VM_BUG_ON(order > PAGE_ALLOC_COSTLY_ORDER);
688#endif
689
690 return order;
691}
692
693static inline bool pcp_allowed_order(unsigned int order)
694{
695 if (order <= PAGE_ALLOC_COSTLY_ORDER)
696 return true;
697#ifdef CONFIG_TRANSPARENT_HUGEPAGE
698 if (order == pageblock_order)
699 return true;
700#endif
701 return false;
702}
703
704static inline void free_the_page(struct page *page, unsigned int order)
705{
706 if (pcp_allowed_order(order))
707 free_unref_page(page, order);
708 else
709 __free_pages_ok(page, order, FPI_NONE);
710}
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727void free_compound_page(struct page *page)
728{
729 mem_cgroup_uncharge(page);
730 free_the_page(page, compound_order(page));
731}
732
733void prep_compound_page(struct page *page, unsigned int order)
734{
735 int i;
736 int nr_pages = 1 << order;
737
738 __SetPageHead(page);
739 for (i = 1; i < nr_pages; i++) {
740 struct page *p = page + i;
741 p->mapping = TAIL_MAPPING;
742 set_compound_head(p, page);
743 }
744
745 set_compound_page_dtor(page, COMPOUND_PAGE_DTOR);
746 set_compound_order(page, order);
747 atomic_set(compound_mapcount_ptr(page), -1);
748 if (hpage_pincount_available(page))
749 atomic_set(compound_pincount_ptr(page), 0);
750}
751
752#ifdef CONFIG_DEBUG_PAGEALLOC
753unsigned int _debug_guardpage_minorder;
754
755bool _debug_pagealloc_enabled_early __read_mostly
756 = IS_ENABLED(CONFIG_DEBUG_PAGEALLOC_ENABLE_DEFAULT);
757EXPORT_SYMBOL(_debug_pagealloc_enabled_early);
758DEFINE_STATIC_KEY_FALSE(_debug_pagealloc_enabled);
759EXPORT_SYMBOL(_debug_pagealloc_enabled);
760
761DEFINE_STATIC_KEY_FALSE(_debug_guardpage_enabled);
762
763static int __init early_debug_pagealloc(char *buf)
764{
765 return kstrtobool(buf, &_debug_pagealloc_enabled_early);
766}
767early_param("debug_pagealloc", early_debug_pagealloc);
768
769static int __init debug_guardpage_minorder_setup(char *buf)
770{
771 unsigned long res;
772
773 if (kstrtoul(buf, 10, &res) < 0 || res > MAX_ORDER / 2) {
774 pr_err("Bad debug_guardpage_minorder value\n");
775 return 0;
776 }
777 _debug_guardpage_minorder = res;
778 pr_info("Setting debug_guardpage_minorder to %lu\n", res);
779 return 0;
780}
781early_param("debug_guardpage_minorder", debug_guardpage_minorder_setup);
782
783static inline bool set_page_guard(struct zone *zone, struct page *page,
784 unsigned int order, int migratetype)
785{
786 if (!debug_guardpage_enabled())
787 return false;
788
789 if (order >= debug_guardpage_minorder())
790 return false;
791
792 __SetPageGuard(page);
793 INIT_LIST_HEAD(&page->lru);
794 set_page_private(page, order);
795
796 __mod_zone_freepage_state(zone, -(1 << order), migratetype);
797
798 return true;
799}
800
801static inline void clear_page_guard(struct zone *zone, struct page *page,
802 unsigned int order, int migratetype)
803{
804 if (!debug_guardpage_enabled())
805 return;
806
807 __ClearPageGuard(page);
808
809 set_page_private(page, 0);
810 if (!is_migrate_isolate(migratetype))
811 __mod_zone_freepage_state(zone, (1 << order), migratetype);
812}
813#else
814static inline bool set_page_guard(struct zone *zone, struct page *page,
815 unsigned int order, int migratetype) { return false; }
816static inline void clear_page_guard(struct zone *zone, struct page *page,
817 unsigned int order, int migratetype) {}
818#endif
819
820
821
822
823
824
825
826void init_mem_debugging_and_hardening(void)
827{
828 bool page_poisoning_requested = false;
829
830#ifdef CONFIG_PAGE_POISONING
831
832
833
834
835 if (page_poisoning_enabled() ||
836 (!IS_ENABLED(CONFIG_ARCH_SUPPORTS_DEBUG_PAGEALLOC) &&
837 debug_pagealloc_enabled())) {
838 static_branch_enable(&_page_poisoning_enabled);
839 page_poisoning_requested = true;
840 }
841#endif
842
843 if ((_init_on_alloc_enabled_early || _init_on_free_enabled_early) &&
844 page_poisoning_requested) {
845 pr_info("mem auto-init: CONFIG_PAGE_POISONING is on, "
846 "will take precedence over init_on_alloc and init_on_free\n");
847 _init_on_alloc_enabled_early = false;
848 _init_on_free_enabled_early = false;
849 }
850
851 if (_init_on_alloc_enabled_early)
852 static_branch_enable(&init_on_alloc);
853 else
854 static_branch_disable(&init_on_alloc);
855
856 if (_init_on_free_enabled_early)
857 static_branch_enable(&init_on_free);
858 else
859 static_branch_disable(&init_on_free);
860
861#ifdef CONFIG_DEBUG_PAGEALLOC
862 if (!debug_pagealloc_enabled())
863 return;
864
865 static_branch_enable(&_debug_pagealloc_enabled);
866
867 if (!debug_guardpage_minorder())
868 return;
869
870 static_branch_enable(&_debug_guardpage_enabled);
871#endif
872}
873
874static inline void set_buddy_order(struct page *page, unsigned int order)
875{
876 set_page_private(page, order);
877 __SetPageBuddy(page);
878}
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893static inline bool page_is_buddy(struct page *page, struct page *buddy,
894 unsigned int order)
895{
896 if (!page_is_guard(buddy) && !PageBuddy(buddy))
897 return false;
898
899 if (buddy_order(buddy) != order)
900 return false;
901
902
903
904
905
906 if (page_zone_id(page) != page_zone_id(buddy))
907 return false;
908
909 VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
910
911 return true;
912}
913
914#ifdef CONFIG_COMPACTION
915static inline struct capture_control *task_capc(struct zone *zone)
916{
917 struct capture_control *capc = current->capture_control;
918
919 return unlikely(capc) &&
920 !(current->flags & PF_KTHREAD) &&
921 !capc->page &&
922 capc->cc->zone == zone ? capc : NULL;
923}
924
925static inline bool
926compaction_capture(struct capture_control *capc, struct page *page,
927 int order, int migratetype)
928{
929 if (!capc || order != capc->cc->order)
930 return false;
931
932
933 if (is_migrate_cma(migratetype) ||
934 is_migrate_isolate(migratetype))
935 return false;
936
937
938
939
940
941
942
943 if (order < pageblock_order && migratetype == MIGRATE_MOVABLE)
944 return false;
945
946 capc->page = page;
947 return true;
948}
949
950#else
951static inline struct capture_control *task_capc(struct zone *zone)
952{
953 return NULL;
954}
955
956static inline bool
957compaction_capture(struct capture_control *capc, struct page *page,
958 int order, int migratetype)
959{
960 return false;
961}
962#endif
963
964
965static inline void add_to_free_list(struct page *page, struct zone *zone,
966 unsigned int order, int migratetype)
967{
968 struct free_area *area = &zone->free_area[order];
969
970 list_add(&page->lru, &area->free_list[migratetype]);
971 area->nr_free++;
972}
973
974
975static inline void add_to_free_list_tail(struct page *page, struct zone *zone,
976 unsigned int order, int migratetype)
977{
978 struct free_area *area = &zone->free_area[order];
979
980 list_add_tail(&page->lru, &area->free_list[migratetype]);
981 area->nr_free++;
982}
983
984
985
986
987
988
989static inline void move_to_free_list(struct page *page, struct zone *zone,
990 unsigned int order, int migratetype)
991{
992 struct free_area *area = &zone->free_area[order];
993
994 list_move_tail(&page->lru, &area->free_list[migratetype]);
995}
996
997static inline void del_page_from_free_list(struct page *page, struct zone *zone,
998 unsigned int order)
999{
1000
1001 if (page_reported(page))
1002 __ClearPageReported(page);
1003
1004 list_del(&page->lru);
1005 __ClearPageBuddy(page);
1006 set_page_private(page, 0);
1007 zone->free_area[order].nr_free--;
1008}
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018static inline bool
1019buddy_merge_likely(unsigned long pfn, unsigned long buddy_pfn,
1020 struct page *page, unsigned int order)
1021{
1022 struct page *higher_page, *higher_buddy;
1023 unsigned long combined_pfn;
1024
1025 if (order >= MAX_ORDER - 2)
1026 return false;
1027
1028 if (!pfn_valid_within(buddy_pfn))
1029 return false;
1030
1031 combined_pfn = buddy_pfn & pfn;
1032 higher_page = page + (combined_pfn - pfn);
1033 buddy_pfn = __find_buddy_pfn(combined_pfn, order + 1);
1034 higher_buddy = higher_page + (buddy_pfn - combined_pfn);
1035
1036 return pfn_valid_within(buddy_pfn) &&
1037 page_is_buddy(higher_page, higher_buddy, order + 1);
1038}
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064static inline void __free_one_page(struct page *page,
1065 unsigned long pfn,
1066 struct zone *zone, unsigned int order,
1067 int migratetype, fpi_t fpi_flags)
1068{
1069 struct capture_control *capc = task_capc(zone);
1070 unsigned long buddy_pfn;
1071 unsigned long combined_pfn;
1072 unsigned int max_order;
1073 struct page *buddy;
1074 bool to_tail;
1075
1076 max_order = min_t(unsigned int, MAX_ORDER - 1, pageblock_order);
1077
1078 VM_BUG_ON(!zone_is_initialized(zone));
1079 VM_BUG_ON_PAGE(page->flags & PAGE_FLAGS_CHECK_AT_PREP, page);
1080
1081 VM_BUG_ON(migratetype == -1);
1082 if (likely(!is_migrate_isolate(migratetype)))
1083 __mod_zone_freepage_state(zone, 1 << order, migratetype);
1084
1085 VM_BUG_ON_PAGE(pfn & ((1 << order) - 1), page);
1086 VM_BUG_ON_PAGE(bad_range(zone, page), page);
1087
1088continue_merging:
1089 while (order < max_order) {
1090 if (compaction_capture(capc, page, order, migratetype)) {
1091 __mod_zone_freepage_state(zone, -(1 << order),
1092 migratetype);
1093 return;
1094 }
1095 buddy_pfn = __find_buddy_pfn(pfn, order);
1096 buddy = page + (buddy_pfn - pfn);
1097
1098 if (!pfn_valid_within(buddy_pfn))
1099 goto done_merging;
1100 if (!page_is_buddy(page, buddy, order))
1101 goto done_merging;
1102
1103
1104
1105
1106 if (page_is_guard(buddy))
1107 clear_page_guard(zone, buddy, order, migratetype);
1108 else
1109 del_page_from_free_list(buddy, zone, order);
1110 combined_pfn = buddy_pfn & pfn;
1111 page = page + (combined_pfn - pfn);
1112 pfn = combined_pfn;
1113 order++;
1114 }
1115 if (order < MAX_ORDER - 1) {
1116
1117
1118
1119
1120
1121
1122
1123
1124 if (unlikely(has_isolate_pageblock(zone))) {
1125 int buddy_mt;
1126
1127 buddy_pfn = __find_buddy_pfn(pfn, order);
1128 buddy = page + (buddy_pfn - pfn);
1129 buddy_mt = get_pageblock_migratetype(buddy);
1130
1131 if (migratetype != buddy_mt
1132 && (is_migrate_isolate(migratetype) ||
1133 is_migrate_isolate(buddy_mt)))
1134 goto done_merging;
1135 }
1136 max_order = order + 1;
1137 goto continue_merging;
1138 }
1139
1140done_merging:
1141 set_buddy_order(page, order);
1142
1143 if (fpi_flags & FPI_TO_TAIL)
1144 to_tail = true;
1145 else if (is_shuffle_order(order))
1146 to_tail = shuffle_pick_tail();
1147 else
1148 to_tail = buddy_merge_likely(pfn, buddy_pfn, page, order);
1149
1150 if (to_tail)
1151 add_to_free_list_tail(page, zone, order, migratetype);
1152 else
1153 add_to_free_list(page, zone, order, migratetype);
1154
1155
1156 if (!(fpi_flags & FPI_SKIP_REPORT_NOTIFY))
1157 page_reporting_notify_free(order);
1158}
1159
1160
1161
1162
1163
1164
1165static inline bool page_expected_state(struct page *page,
1166 unsigned long check_flags)
1167{
1168 if (unlikely(atomic_read(&page->_mapcount) != -1))
1169 return false;
1170
1171 if (unlikely((unsigned long)page->mapping |
1172 page_ref_count(page) |
1173#ifdef CONFIG_MEMCG
1174 page->memcg_data |
1175#endif
1176 (page->flags & check_flags)))
1177 return false;
1178
1179 return true;
1180}
1181
1182static const char *page_bad_reason(struct page *page, unsigned long flags)
1183{
1184 const char *bad_reason = NULL;
1185
1186 if (unlikely(atomic_read(&page->_mapcount) != -1))
1187 bad_reason = "nonzero mapcount";
1188 if (unlikely(page->mapping != NULL))
1189 bad_reason = "non-NULL mapping";
1190 if (unlikely(page_ref_count(page) != 0))
1191 bad_reason = "nonzero _refcount";
1192 if (unlikely(page->flags & flags)) {
1193 if (flags == PAGE_FLAGS_CHECK_AT_PREP)
1194 bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag(s) set";
1195 else
1196 bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set";
1197 }
1198#ifdef CONFIG_MEMCG
1199 if (unlikely(page->memcg_data))
1200 bad_reason = "page still charged to cgroup";
1201#endif
1202 return bad_reason;
1203}
1204
1205static void check_free_page_bad(struct page *page)
1206{
1207 bad_page(page,
1208 page_bad_reason(page, PAGE_FLAGS_CHECK_AT_FREE));
1209}
1210
1211static inline int check_free_page(struct page *page)
1212{
1213 if (likely(page_expected_state(page, PAGE_FLAGS_CHECK_AT_FREE)))
1214 return 0;
1215
1216
1217 check_free_page_bad(page);
1218 return 1;
1219}
1220
1221static int free_tail_pages_check(struct page *head_page, struct page *page)
1222{
1223 int ret = 1;
1224
1225
1226
1227
1228
1229 BUILD_BUG_ON((unsigned long)LIST_POISON1 & 1);
1230
1231 if (!IS_ENABLED(CONFIG_DEBUG_VM)) {
1232 ret = 0;
1233 goto out;
1234 }
1235 switch (page - head_page) {
1236 case 1:
1237
1238 if (unlikely(compound_mapcount(page))) {
1239 bad_page(page, "nonzero compound_mapcount");
1240 goto out;
1241 }
1242 break;
1243 case 2:
1244
1245
1246
1247
1248 break;
1249 default:
1250 if (page->mapping != TAIL_MAPPING) {
1251 bad_page(page, "corrupted mapping in tail page");
1252 goto out;
1253 }
1254 break;
1255 }
1256 if (unlikely(!PageTail(page))) {
1257 bad_page(page, "PageTail not set");
1258 goto out;
1259 }
1260 if (unlikely(compound_head(page) != head_page)) {
1261 bad_page(page, "compound_head not consistent");
1262 goto out;
1263 }
1264 ret = 0;
1265out:
1266 page->mapping = NULL;
1267 clear_compound_head(page);
1268 return ret;
1269}
1270
1271static void kernel_init_free_pages(struct page *page, int numpages, bool zero_tags)
1272{
1273 int i;
1274
1275 if (zero_tags) {
1276 for (i = 0; i < numpages; i++)
1277 tag_clear_highpage(page + i);
1278 return;
1279 }
1280
1281
1282 kasan_disable_current();
1283 for (i = 0; i < numpages; i++) {
1284 u8 tag = page_kasan_tag(page + i);
1285 page_kasan_tag_reset(page + i);
1286 clear_highpage(page + i);
1287 page_kasan_tag_set(page + i, tag);
1288 }
1289 kasan_enable_current();
1290}
1291
1292static __always_inline bool free_pages_prepare(struct page *page,
1293 unsigned int order, bool check_free, fpi_t fpi_flags)
1294{
1295 int bad = 0;
1296 bool skip_kasan_poison = should_skip_kasan_poison(page, fpi_flags);
1297
1298 VM_BUG_ON_PAGE(PageTail(page), page);
1299
1300 trace_mm_page_free(page, order);
1301
1302 if (unlikely(PageHWPoison(page)) && !order) {
1303
1304
1305
1306
1307 if (memcg_kmem_enabled() && PageMemcgKmem(page))
1308 __memcg_kmem_uncharge_page(page, order);
1309 reset_page_owner(page, order);
1310 return false;
1311 }
1312
1313
1314
1315
1316
1317 if (unlikely(order)) {
1318 bool compound = PageCompound(page);
1319 int i;
1320
1321 VM_BUG_ON_PAGE(compound && compound_order(page) != order, page);
1322
1323 if (compound)
1324 ClearPageDoubleMap(page);
1325 for (i = 1; i < (1 << order); i++) {
1326 if (compound)
1327 bad += free_tail_pages_check(page, page + i);
1328 if (unlikely(check_free_page(page + i))) {
1329 bad++;
1330 continue;
1331 }
1332 (page + i)->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
1333 }
1334 }
1335 if (PageMappingFlags(page))
1336 page->mapping = NULL;
1337 if (memcg_kmem_enabled() && PageMemcgKmem(page))
1338 __memcg_kmem_uncharge_page(page, order);
1339 if (check_free)
1340 bad += check_free_page(page);
1341 if (bad)
1342 return false;
1343
1344 page_cpupid_reset_last(page);
1345 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
1346 reset_page_owner(page, order);
1347
1348 if (!PageHighMem(page)) {
1349 debug_check_no_locks_freed(page_address(page),
1350 PAGE_SIZE << order);
1351 debug_check_no_obj_freed(page_address(page),
1352 PAGE_SIZE << order);
1353 }
1354
1355 kernel_poison_pages(page, 1 << order);
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365 if (kasan_has_integrated_init()) {
1366 if (!skip_kasan_poison)
1367 kasan_free_pages(page, order);
1368 } else {
1369 bool init = want_init_on_free();
1370
1371 if (init)
1372 kernel_init_free_pages(page, 1 << order, false);
1373 if (!skip_kasan_poison)
1374 kasan_poison_pages(page, order, init);
1375 }
1376
1377
1378
1379
1380
1381
1382 arch_free_page(page, order);
1383
1384 debug_pagealloc_unmap_pages(page, 1 << order);
1385
1386 return true;
1387}
1388
1389#ifdef CONFIG_DEBUG_VM
1390
1391
1392
1393
1394
1395static bool free_pcp_prepare(struct page *page, unsigned int order)
1396{
1397 return free_pages_prepare(page, order, true, FPI_NONE);
1398}
1399
1400static bool bulkfree_pcp_prepare(struct page *page)
1401{
1402 if (debug_pagealloc_enabled_static())
1403 return check_free_page(page);
1404 else
1405 return false;
1406}
1407#else
1408
1409
1410
1411
1412
1413
1414static bool free_pcp_prepare(struct page *page, unsigned int order)
1415{
1416 if (debug_pagealloc_enabled_static())
1417 return free_pages_prepare(page, order, true, FPI_NONE);
1418 else
1419 return free_pages_prepare(page, order, false, FPI_NONE);
1420}
1421
1422static bool bulkfree_pcp_prepare(struct page *page)
1423{
1424 return check_free_page(page);
1425}
1426#endif
1427
1428static inline void prefetch_buddy(struct page *page)
1429{
1430 unsigned long pfn = page_to_pfn(page);
1431 unsigned long buddy_pfn = __find_buddy_pfn(pfn, 0);
1432 struct page *buddy = page + (buddy_pfn - pfn);
1433
1434 prefetch(buddy);
1435}
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448static void free_pcppages_bulk(struct zone *zone, int count,
1449 struct per_cpu_pages *pcp)
1450{
1451 int pindex = 0;
1452 int batch_free = 0;
1453 int nr_freed = 0;
1454 unsigned int order;
1455 int prefetch_nr = READ_ONCE(pcp->batch);
1456 bool isolated_pageblocks;
1457 struct page *page, *tmp;
1458 LIST_HEAD(head);
1459
1460
1461
1462
1463
1464 count = min(pcp->count, count);
1465 while (count > 0) {
1466 struct list_head *list;
1467
1468
1469
1470
1471
1472
1473
1474
1475 do {
1476 batch_free++;
1477 if (++pindex == NR_PCP_LISTS)
1478 pindex = 0;
1479 list = &pcp->lists[pindex];
1480 } while (list_empty(list));
1481
1482
1483 if (batch_free == NR_PCP_LISTS)
1484 batch_free = count;
1485
1486 order = pindex_to_order(pindex);
1487 BUILD_BUG_ON(MAX_ORDER >= (1<<NR_PCP_ORDER_WIDTH));
1488 do {
1489 page = list_last_entry(list, struct page, lru);
1490
1491 list_del(&page->lru);
1492 nr_freed += 1 << order;
1493 count -= 1 << order;
1494
1495 if (bulkfree_pcp_prepare(page))
1496 continue;
1497
1498
1499 page->index <<= NR_PCP_ORDER_WIDTH;
1500 page->index |= order;
1501
1502 list_add_tail(&page->lru, &head);
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513 if (prefetch_nr) {
1514 prefetch_buddy(page);
1515 prefetch_nr--;
1516 }
1517 } while (count > 0 && --batch_free && !list_empty(list));
1518 }
1519 pcp->count -= nr_freed;
1520
1521
1522
1523
1524
1525 spin_lock(&zone->lock);
1526 isolated_pageblocks = has_isolate_pageblock(zone);
1527
1528
1529
1530
1531
1532 list_for_each_entry_safe(page, tmp, &head, lru) {
1533 int mt = get_pcppage_migratetype(page);
1534
1535
1536 order = mt & NR_PCP_ORDER_MASK;
1537 mt >>= NR_PCP_ORDER_WIDTH;
1538
1539
1540 VM_BUG_ON_PAGE(is_migrate_isolate(mt), page);
1541
1542 if (unlikely(isolated_pageblocks))
1543 mt = get_pageblock_migratetype(page);
1544
1545 __free_one_page(page, page_to_pfn(page), zone, order, mt, FPI_NONE);
1546 trace_mm_page_pcpu_drain(page, order, mt);
1547 }
1548 spin_unlock(&zone->lock);
1549}
1550
1551static void free_one_page(struct zone *zone,
1552 struct page *page, unsigned long pfn,
1553 unsigned int order,
1554 int migratetype, fpi_t fpi_flags)
1555{
1556 unsigned long flags;
1557
1558 spin_lock_irqsave(&zone->lock, flags);
1559 if (unlikely(has_isolate_pageblock(zone) ||
1560 is_migrate_isolate(migratetype))) {
1561 migratetype = get_pfnblock_migratetype(page, pfn);
1562 }
1563 __free_one_page(page, pfn, zone, order, migratetype, fpi_flags);
1564 spin_unlock_irqrestore(&zone->lock, flags);
1565}
1566
1567static void __meminit __init_single_page(struct page *page, unsigned long pfn,
1568 unsigned long zone, int nid)
1569{
1570 mm_zero_struct_page(page);
1571 set_page_links(page, zone, nid, pfn);
1572 init_page_count(page);
1573 page_mapcount_reset(page);
1574 page_cpupid_reset_last(page);
1575 page_kasan_tag_reset(page);
1576
1577 INIT_LIST_HEAD(&page->lru);
1578#ifdef WANT_PAGE_VIRTUAL
1579
1580 if (!is_highmem_idx(zone))
1581 set_page_address(page, __va(pfn << PAGE_SHIFT));
1582#endif
1583}
1584
1585#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1586static void __meminit init_reserved_page(unsigned long pfn)
1587{
1588 pg_data_t *pgdat;
1589 int nid, zid;
1590
1591 if (!early_page_uninitialised(pfn))
1592 return;
1593
1594 nid = early_pfn_to_nid(pfn);
1595 pgdat = NODE_DATA(nid);
1596
1597 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
1598 struct zone *zone = &pgdat->node_zones[zid];
1599
1600 if (pfn >= zone->zone_start_pfn && pfn < zone_end_pfn(zone))
1601 break;
1602 }
1603 __init_single_page(pfn_to_page(pfn), pfn, zid, nid);
1604}
1605#else
1606static inline void init_reserved_page(unsigned long pfn)
1607{
1608}
1609#endif
1610
1611
1612
1613
1614
1615
1616
1617void __meminit reserve_bootmem_region(phys_addr_t start, phys_addr_t end)
1618{
1619 unsigned long start_pfn = PFN_DOWN(start);
1620 unsigned long end_pfn = PFN_UP(end);
1621
1622 for (; start_pfn < end_pfn; start_pfn++) {
1623 if (pfn_valid(start_pfn)) {
1624 struct page *page = pfn_to_page(start_pfn);
1625
1626 init_reserved_page(start_pfn);
1627
1628
1629 INIT_LIST_HEAD(&page->lru);
1630
1631
1632
1633
1634
1635
1636 __SetPageReserved(page);
1637 }
1638 }
1639}
1640
1641static void __free_pages_ok(struct page *page, unsigned int order,
1642 fpi_t fpi_flags)
1643{
1644 unsigned long flags;
1645 int migratetype;
1646 unsigned long pfn = page_to_pfn(page);
1647 struct zone *zone = page_zone(page);
1648
1649 if (!free_pages_prepare(page, order, true, fpi_flags))
1650 return;
1651
1652 migratetype = get_pfnblock_migratetype(page, pfn);
1653
1654 spin_lock_irqsave(&zone->lock, flags);
1655 if (unlikely(has_isolate_pageblock(zone) ||
1656 is_migrate_isolate(migratetype))) {
1657 migratetype = get_pfnblock_migratetype(page, pfn);
1658 }
1659 __free_one_page(page, pfn, zone, order, migratetype, fpi_flags);
1660 spin_unlock_irqrestore(&zone->lock, flags);
1661
1662 __count_vm_events(PGFREE, 1 << order);
1663}
1664
1665void __free_pages_core(struct page *page, unsigned int order)
1666{
1667 unsigned int nr_pages = 1 << order;
1668 struct page *p = page;
1669 unsigned int loop;
1670
1671
1672
1673
1674
1675
1676 prefetchw(p);
1677 for (loop = 0; loop < (nr_pages - 1); loop++, p++) {
1678 prefetchw(p + 1);
1679 __ClearPageReserved(p);
1680 set_page_count(p, 0);
1681 }
1682 __ClearPageReserved(p);
1683 set_page_count(p, 0);
1684
1685 atomic_long_add(nr_pages, &page_zone(page)->managed_pages);
1686
1687
1688
1689
1690
1691 __free_pages_ok(page, order, FPI_TO_TAIL | FPI_SKIP_KASAN_POISON);
1692}
1693
1694#ifdef CONFIG_NUMA
1695
1696
1697
1698
1699
1700
1701struct mminit_pfnnid_cache {
1702 unsigned long last_start;
1703 unsigned long last_end;
1704 int last_nid;
1705};
1706
1707static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata;
1708
1709
1710
1711
1712static int __meminit __early_pfn_to_nid(unsigned long pfn,
1713 struct mminit_pfnnid_cache *state)
1714{
1715 unsigned long start_pfn, end_pfn;
1716 int nid;
1717
1718 if (state->last_start <= pfn && pfn < state->last_end)
1719 return state->last_nid;
1720
1721 nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn);
1722 if (nid != NUMA_NO_NODE) {
1723 state->last_start = start_pfn;
1724 state->last_end = end_pfn;
1725 state->last_nid = nid;
1726 }
1727
1728 return nid;
1729}
1730
1731int __meminit early_pfn_to_nid(unsigned long pfn)
1732{
1733 static DEFINE_SPINLOCK(early_pfn_lock);
1734 int nid;
1735
1736 spin_lock(&early_pfn_lock);
1737 nid = __early_pfn_to_nid(pfn, &early_pfnnid_cache);
1738 if (nid < 0)
1739 nid = first_online_node;
1740 spin_unlock(&early_pfn_lock);
1741
1742 return nid;
1743}
1744#endif
1745
1746void __init memblock_free_pages(struct page *page, unsigned long pfn,
1747 unsigned int order)
1748{
1749 if (early_page_uninitialised(pfn))
1750 return;
1751 __free_pages_core(page, order);
1752}
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771struct page *__pageblock_pfn_to_page(unsigned long start_pfn,
1772 unsigned long end_pfn, struct zone *zone)
1773{
1774 struct page *start_page;
1775 struct page *end_page;
1776
1777
1778 end_pfn--;
1779
1780 if (!pfn_valid(start_pfn) || !pfn_valid(end_pfn))
1781 return NULL;
1782
1783 start_page = pfn_to_online_page(start_pfn);
1784 if (!start_page)
1785 return NULL;
1786
1787 if (page_zone(start_page) != zone)
1788 return NULL;
1789
1790 end_page = pfn_to_page(end_pfn);
1791
1792
1793 if (page_zone_id(start_page) != page_zone_id(end_page))
1794 return NULL;
1795
1796 return start_page;
1797}
1798
1799void set_zone_contiguous(struct zone *zone)
1800{
1801 unsigned long block_start_pfn = zone->zone_start_pfn;
1802 unsigned long block_end_pfn;
1803
1804 block_end_pfn = ALIGN(block_start_pfn + 1, pageblock_nr_pages);
1805 for (; block_start_pfn < zone_end_pfn(zone);
1806 block_start_pfn = block_end_pfn,
1807 block_end_pfn += pageblock_nr_pages) {
1808
1809 block_end_pfn = min(block_end_pfn, zone_end_pfn(zone));
1810
1811 if (!__pageblock_pfn_to_page(block_start_pfn,
1812 block_end_pfn, zone))
1813 return;
1814 cond_resched();
1815 }
1816
1817
1818 zone->contiguous = true;
1819}
1820
1821void clear_zone_contiguous(struct zone *zone)
1822{
1823 zone->contiguous = false;
1824}
1825
1826#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1827static void __init deferred_free_range(unsigned long pfn,
1828 unsigned long nr_pages)
1829{
1830 struct page *page;
1831 unsigned long i;
1832
1833 if (!nr_pages)
1834 return;
1835
1836 page = pfn_to_page(pfn);
1837
1838
1839 if (nr_pages == pageblock_nr_pages &&
1840 (pfn & (pageblock_nr_pages - 1)) == 0) {
1841 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
1842 __free_pages_core(page, pageblock_order);
1843 return;
1844 }
1845
1846 for (i = 0; i < nr_pages; i++, page++, pfn++) {
1847 if ((pfn & (pageblock_nr_pages - 1)) == 0)
1848 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
1849 __free_pages_core(page, 0);
1850 }
1851}
1852
1853
1854static atomic_t pgdat_init_n_undone __initdata;
1855static __initdata DECLARE_COMPLETION(pgdat_init_all_done_comp);
1856
1857static inline void __init pgdat_init_report_one_done(void)
1858{
1859 if (atomic_dec_and_test(&pgdat_init_n_undone))
1860 complete(&pgdat_init_all_done_comp);
1861}
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873static inline bool __init deferred_pfn_valid(unsigned long pfn)
1874{
1875 if (!pfn_valid_within(pfn))
1876 return false;
1877 if (!(pfn & (pageblock_nr_pages - 1)) && !pfn_valid(pfn))
1878 return false;
1879 return true;
1880}
1881
1882
1883
1884
1885
1886static void __init deferred_free_pages(unsigned long pfn,
1887 unsigned long end_pfn)
1888{
1889 unsigned long nr_pgmask = pageblock_nr_pages - 1;
1890 unsigned long nr_free = 0;
1891
1892 for (; pfn < end_pfn; pfn++) {
1893 if (!deferred_pfn_valid(pfn)) {
1894 deferred_free_range(pfn - nr_free, nr_free);
1895 nr_free = 0;
1896 } else if (!(pfn & nr_pgmask)) {
1897 deferred_free_range(pfn - nr_free, nr_free);
1898 nr_free = 1;
1899 } else {
1900 nr_free++;
1901 }
1902 }
1903
1904 deferred_free_range(pfn - nr_free, nr_free);
1905}
1906
1907
1908
1909
1910
1911
1912static unsigned long __init deferred_init_pages(struct zone *zone,
1913 unsigned long pfn,
1914 unsigned long end_pfn)
1915{
1916 unsigned long nr_pgmask = pageblock_nr_pages - 1;
1917 int nid = zone_to_nid(zone);
1918 unsigned long nr_pages = 0;
1919 int zid = zone_idx(zone);
1920 struct page *page = NULL;
1921
1922 for (; pfn < end_pfn; pfn++) {
1923 if (!deferred_pfn_valid(pfn)) {
1924 page = NULL;
1925 continue;
1926 } else if (!page || !(pfn & nr_pgmask)) {
1927 page = pfn_to_page(pfn);
1928 } else {
1929 page++;
1930 }
1931 __init_single_page(page, pfn, zid, nid);
1932 nr_pages++;
1933 }
1934 return (nr_pages);
1935}
1936
1937
1938
1939
1940
1941
1942
1943static bool __init
1944deferred_init_mem_pfn_range_in_zone(u64 *i, struct zone *zone,
1945 unsigned long *spfn, unsigned long *epfn,
1946 unsigned long first_init_pfn)
1947{
1948 u64 j;
1949
1950
1951
1952
1953
1954
1955 for_each_free_mem_pfn_range_in_zone(j, zone, spfn, epfn) {
1956 if (*epfn <= first_init_pfn)
1957 continue;
1958 if (*spfn < first_init_pfn)
1959 *spfn = first_init_pfn;
1960 *i = j;
1961 return true;
1962 }
1963
1964 return false;
1965}
1966
1967
1968
1969
1970
1971
1972
1973
1974
1975
1976
1977static unsigned long __init
1978deferred_init_maxorder(u64 *i, struct zone *zone, unsigned long *start_pfn,
1979 unsigned long *end_pfn)
1980{
1981 unsigned long mo_pfn = ALIGN(*start_pfn + 1, MAX_ORDER_NR_PAGES);
1982 unsigned long spfn = *start_pfn, epfn = *end_pfn;
1983 unsigned long nr_pages = 0;
1984 u64 j = *i;
1985
1986
1987 for_each_free_mem_pfn_range_in_zone_from(j, zone, start_pfn, end_pfn) {
1988 unsigned long t;
1989
1990 if (mo_pfn <= *start_pfn)
1991 break;
1992
1993 t = min(mo_pfn, *end_pfn);
1994 nr_pages += deferred_init_pages(zone, *start_pfn, t);
1995
1996 if (mo_pfn < *end_pfn) {
1997 *start_pfn = mo_pfn;
1998 break;
1999 }
2000 }
2001
2002
2003 swap(j, *i);
2004
2005 for_each_free_mem_pfn_range_in_zone_from(j, zone, &spfn, &epfn) {
2006 unsigned long t;
2007
2008 if (mo_pfn <= spfn)
2009 break;
2010
2011 t = min(mo_pfn, epfn);
2012 deferred_free_pages(spfn, t);
2013
2014 if (mo_pfn <= epfn)
2015 break;
2016 }
2017
2018 return nr_pages;
2019}
2020
2021static void __init
2022deferred_init_memmap_chunk(unsigned long start_pfn, unsigned long end_pfn,
2023 void *arg)
2024{
2025 unsigned long spfn, epfn;
2026 struct zone *zone = arg;
2027 u64 i;
2028
2029 deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn, start_pfn);
2030
2031
2032
2033
2034
2035 while (spfn < end_pfn) {
2036 deferred_init_maxorder(&i, zone, &spfn, &epfn);
2037 cond_resched();
2038 }
2039}
2040
2041
2042__weak int __init
2043deferred_page_init_max_threads(const struct cpumask *node_cpumask)
2044{
2045 return 1;
2046}
2047
2048
2049static int __init deferred_init_memmap(void *data)
2050{
2051 pg_data_t *pgdat = data;
2052 const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id);
2053 unsigned long spfn = 0, epfn = 0;
2054 unsigned long first_init_pfn, flags;
2055 unsigned long start = jiffies;
2056 struct zone *zone;
2057 int zid, max_threads;
2058 u64 i;
2059
2060
2061 if (!cpumask_empty(cpumask))
2062 set_cpus_allowed_ptr(current, cpumask);
2063
2064 pgdat_resize_lock(pgdat, &flags);
2065 first_init_pfn = pgdat->first_deferred_pfn;
2066 if (first_init_pfn == ULONG_MAX) {
2067 pgdat_resize_unlock(pgdat, &flags);
2068 pgdat_init_report_one_done();
2069 return 0;
2070 }
2071
2072
2073 BUG_ON(pgdat->first_deferred_pfn < pgdat->node_start_pfn);
2074 BUG_ON(pgdat->first_deferred_pfn > pgdat_end_pfn(pgdat));
2075 pgdat->first_deferred_pfn = ULONG_MAX;
2076
2077
2078
2079
2080
2081
2082 pgdat_resize_unlock(pgdat, &flags);
2083
2084
2085 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
2086 zone = pgdat->node_zones + zid;
2087 if (first_init_pfn < zone_end_pfn(zone))
2088 break;
2089 }
2090
2091
2092 if (!deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn,
2093 first_init_pfn))
2094 goto zone_empty;
2095
2096 max_threads = deferred_page_init_max_threads(cpumask);
2097
2098 while (spfn < epfn) {
2099 unsigned long epfn_align = ALIGN(epfn, PAGES_PER_SECTION);
2100 struct padata_mt_job job = {
2101 .thread_fn = deferred_init_memmap_chunk,
2102 .fn_arg = zone,
2103 .start = spfn,
2104 .size = epfn_align - spfn,
2105 .align = PAGES_PER_SECTION,
2106 .min_chunk = PAGES_PER_SECTION,
2107 .max_threads = max_threads,
2108 };
2109
2110 padata_do_multithreaded(&job);
2111 deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn,
2112 epfn_align);
2113 }
2114zone_empty:
2115
2116 WARN_ON(++zid < MAX_NR_ZONES && populated_zone(++zone));
2117
2118 pr_info("node %d deferred pages initialised in %ums\n",
2119 pgdat->node_id, jiffies_to_msecs(jiffies - start));
2120
2121 pgdat_init_report_one_done();
2122 return 0;
2123}
2124
2125
2126
2127
2128
2129
2130
2131
2132
2133
2134
2135
2136
2137
2138
2139
2140static noinline bool __init
2141deferred_grow_zone(struct zone *zone, unsigned int order)
2142{
2143 unsigned long nr_pages_needed = ALIGN(1 << order, PAGES_PER_SECTION);
2144 pg_data_t *pgdat = zone->zone_pgdat;
2145 unsigned long first_deferred_pfn = pgdat->first_deferred_pfn;
2146 unsigned long spfn, epfn, flags;
2147 unsigned long nr_pages = 0;
2148 u64 i;
2149
2150
2151 if (zone_end_pfn(zone) != pgdat_end_pfn(pgdat))
2152 return false;
2153
2154 pgdat_resize_lock(pgdat, &flags);
2155
2156
2157
2158
2159
2160 if (first_deferred_pfn != pgdat->first_deferred_pfn) {
2161 pgdat_resize_unlock(pgdat, &flags);
2162 return true;
2163 }
2164
2165
2166 if (!deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn,
2167 first_deferred_pfn)) {
2168 pgdat->first_deferred_pfn = ULONG_MAX;
2169 pgdat_resize_unlock(pgdat, &flags);
2170
2171 return first_deferred_pfn != ULONG_MAX;
2172 }
2173
2174
2175
2176
2177
2178
2179 while (spfn < epfn) {
2180
2181 first_deferred_pfn = spfn;
2182
2183 nr_pages += deferred_init_maxorder(&i, zone, &spfn, &epfn);
2184 touch_nmi_watchdog();
2185
2186
2187 if ((first_deferred_pfn ^ spfn) < PAGES_PER_SECTION)
2188 continue;
2189
2190
2191 if (nr_pages >= nr_pages_needed)
2192 break;
2193 }
2194
2195 pgdat->first_deferred_pfn = spfn;
2196 pgdat_resize_unlock(pgdat, &flags);
2197
2198 return nr_pages > 0;
2199}
2200
2201
2202
2203
2204
2205
2206
2207static bool __ref
2208_deferred_grow_zone(struct zone *zone, unsigned int order)
2209{
2210 return deferred_grow_zone(zone, order);
2211}
2212
2213#endif
2214
2215void __init page_alloc_init_late(void)
2216{
2217 struct zone *zone;
2218 int nid;
2219
2220#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
2221
2222
2223 atomic_set(&pgdat_init_n_undone, num_node_state(N_MEMORY));
2224 for_each_node_state(nid, N_MEMORY) {
2225 kthread_run(deferred_init_memmap, NODE_DATA(nid), "pgdatinit%d", nid);
2226 }
2227
2228
2229 wait_for_completion(&pgdat_init_all_done_comp);
2230
2231
2232
2233
2234
2235 static_branch_disable(&deferred_pages);
2236
2237
2238 files_maxfiles_init();
2239#endif
2240
2241 buffer_init();
2242
2243
2244 memblock_discard();
2245
2246 for_each_node_state(nid, N_MEMORY)
2247 shuffle_free_memory(NODE_DATA(nid));
2248
2249 for_each_populated_zone(zone)
2250 set_zone_contiguous(zone);
2251}
2252
2253#ifdef CONFIG_CMA
2254
2255void __init init_cma_reserved_pageblock(struct page *page)
2256{
2257 unsigned i = pageblock_nr_pages;
2258 struct page *p = page;
2259
2260 do {
2261 __ClearPageReserved(p);
2262 set_page_count(p, 0);
2263 } while (++p, --i);
2264
2265 set_pageblock_migratetype(page, MIGRATE_CMA);
2266
2267 if (pageblock_order >= MAX_ORDER) {
2268 i = pageblock_nr_pages;
2269 p = page;
2270 do {
2271 set_page_refcounted(p);
2272 __free_pages(p, MAX_ORDER - 1);
2273 p += MAX_ORDER_NR_PAGES;
2274 } while (i -= MAX_ORDER_NR_PAGES);
2275 } else {
2276 set_page_refcounted(page);
2277 __free_pages(page, pageblock_order);
2278 }
2279
2280 adjust_managed_page_count(page, pageblock_nr_pages);
2281 page_zone(page)->cma_pages += pageblock_nr_pages;
2282}
2283#endif
2284
2285
2286
2287
2288
2289
2290
2291
2292
2293
2294
2295
2296
2297
2298
2299static inline void expand(struct zone *zone, struct page *page,
2300 int low, int high, int migratetype)
2301{
2302 unsigned long size = 1 << high;
2303
2304 while (high > low) {
2305 high--;
2306 size >>= 1;
2307 VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]);
2308
2309
2310
2311
2312
2313
2314
2315 if (set_page_guard(zone, &page[size], high, migratetype))
2316 continue;
2317
2318 add_to_free_list(&page[size], zone, high, migratetype);
2319 set_buddy_order(&page[size], high);
2320 }
2321}
2322
2323static void check_new_page_bad(struct page *page)
2324{
2325 if (unlikely(page->flags & __PG_HWPOISON)) {
2326
2327 page_mapcount_reset(page);
2328 return;
2329 }
2330
2331 bad_page(page,
2332 page_bad_reason(page, PAGE_FLAGS_CHECK_AT_PREP));
2333}
2334
2335
2336
2337
2338static inline int check_new_page(struct page *page)
2339{
2340 if (likely(page_expected_state(page,
2341 PAGE_FLAGS_CHECK_AT_PREP|__PG_HWPOISON)))
2342 return 0;
2343
2344 check_new_page_bad(page);
2345 return 1;
2346}
2347
2348#ifdef CONFIG_DEBUG_VM
2349
2350
2351
2352
2353
2354static inline bool check_pcp_refill(struct page *page)
2355{
2356 if (debug_pagealloc_enabled_static())
2357 return check_new_page(page);
2358 else
2359 return false;
2360}
2361
2362static inline bool check_new_pcp(struct page *page)
2363{
2364 return check_new_page(page);
2365}
2366#else
2367
2368
2369
2370
2371
2372static inline bool check_pcp_refill(struct page *page)
2373{
2374 return check_new_page(page);
2375}
2376static inline bool check_new_pcp(struct page *page)
2377{
2378 if (debug_pagealloc_enabled_static())
2379 return check_new_page(page);
2380 else
2381 return false;
2382}
2383#endif
2384
2385static bool check_new_pages(struct page *page, unsigned int order)
2386{
2387 int i;
2388 for (i = 0; i < (1 << order); i++) {
2389 struct page *p = page + i;
2390
2391 if (unlikely(check_new_page(p)))
2392 return true;
2393 }
2394
2395 return false;
2396}
2397
2398inline void post_alloc_hook(struct page *page, unsigned int order,
2399 gfp_t gfp_flags)
2400{
2401 set_page_private(page, 0);
2402 set_page_refcounted(page);
2403
2404 arch_alloc_page(page, order);
2405 debug_pagealloc_map_pages(page, 1 << order);
2406
2407
2408
2409
2410
2411
2412 kernel_unpoison_pages(page, 1 << order);
2413
2414
2415
2416
2417
2418
2419 if (kasan_has_integrated_init()) {
2420 kasan_alloc_pages(page, order, gfp_flags);
2421 } else {
2422 bool init = !want_init_on_free() && want_init_on_alloc(gfp_flags);
2423
2424 kasan_unpoison_pages(page, order, init);
2425 if (init)
2426 kernel_init_free_pages(page, 1 << order,
2427 gfp_flags & __GFP_ZEROTAGS);
2428 }
2429
2430 set_page_owner(page, order, gfp_flags);
2431}
2432
2433static void prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags,
2434 unsigned int alloc_flags)
2435{
2436 post_alloc_hook(page, order, gfp_flags);
2437
2438 if (order && (gfp_flags & __GFP_COMP))
2439 prep_compound_page(page, order);
2440
2441
2442
2443
2444
2445
2446
2447 if (alloc_flags & ALLOC_NO_WATERMARKS)
2448 set_page_pfmemalloc(page);
2449 else
2450 clear_page_pfmemalloc(page);
2451}
2452
2453
2454
2455
2456
2457static __always_inline
2458struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
2459 int migratetype)
2460{
2461 unsigned int current_order;
2462 struct free_area *area;
2463 struct page *page;
2464
2465
2466 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
2467 area = &(zone->free_area[current_order]);
2468 page = get_page_from_free_area(area, migratetype);
2469 if (!page)
2470 continue;
2471 del_page_from_free_list(page, zone, current_order);
2472 expand(zone, page, order, current_order, migratetype);
2473 set_pcppage_migratetype(page, migratetype);
2474 return page;
2475 }
2476
2477 return NULL;
2478}
2479
2480
2481
2482
2483
2484
2485static int fallbacks[MIGRATE_TYPES][3] = {
2486 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_TYPES },
2487 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_TYPES },
2488 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_TYPES },
2489#ifdef CONFIG_CMA
2490 [MIGRATE_CMA] = { MIGRATE_TYPES },
2491#endif
2492#ifdef CONFIG_MEMORY_ISOLATION
2493 [MIGRATE_ISOLATE] = { MIGRATE_TYPES },
2494#endif
2495};
2496
2497#ifdef CONFIG_CMA
2498static __always_inline struct page *__rmqueue_cma_fallback(struct zone *zone,
2499 unsigned int order)
2500{
2501 return __rmqueue_smallest(zone, order, MIGRATE_CMA);
2502}
2503#else
2504static inline struct page *__rmqueue_cma_fallback(struct zone *zone,
2505 unsigned int order) { return NULL; }
2506#endif
2507
2508
2509
2510
2511
2512
2513static int move_freepages(struct zone *zone,
2514 unsigned long start_pfn, unsigned long end_pfn,
2515 int migratetype, int *num_movable)
2516{
2517 struct page *page;
2518 unsigned long pfn;
2519 unsigned int order;
2520 int pages_moved = 0;
2521
2522 for (pfn = start_pfn; pfn <= end_pfn;) {
2523 if (!pfn_valid_within(pfn)) {
2524 pfn++;
2525 continue;
2526 }
2527
2528 page = pfn_to_page(pfn);
2529 if (!PageBuddy(page)) {
2530
2531
2532
2533
2534
2535 if (num_movable &&
2536 (PageLRU(page) || __PageMovable(page)))
2537 (*num_movable)++;
2538 pfn++;
2539 continue;
2540 }
2541
2542
2543 VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page);
2544 VM_BUG_ON_PAGE(page_zone(page) != zone, page);
2545
2546 order = buddy_order(page);
2547 move_to_free_list(page, zone, order, migratetype);
2548 pfn += 1 << order;
2549 pages_moved += 1 << order;
2550 }
2551
2552 return pages_moved;
2553}
2554
2555int move_freepages_block(struct zone *zone, struct page *page,
2556 int migratetype, int *num_movable)
2557{
2558 unsigned long start_pfn, end_pfn, pfn;
2559
2560 if (num_movable)
2561 *num_movable = 0;
2562
2563 pfn = page_to_pfn(page);
2564 start_pfn = pfn & ~(pageblock_nr_pages - 1);
2565 end_pfn = start_pfn + pageblock_nr_pages - 1;
2566
2567
2568 if (!zone_spans_pfn(zone, start_pfn))
2569 start_pfn = pfn;
2570 if (!zone_spans_pfn(zone, end_pfn))
2571 return 0;
2572
2573 return move_freepages(zone, start_pfn, end_pfn, migratetype,
2574 num_movable);
2575}
2576
2577static void change_pageblock_range(struct page *pageblock_page,
2578 int start_order, int migratetype)
2579{
2580 int nr_pageblocks = 1 << (start_order - pageblock_order);
2581
2582 while (nr_pageblocks--) {
2583 set_pageblock_migratetype(pageblock_page, migratetype);
2584 pageblock_page += pageblock_nr_pages;
2585 }
2586}
2587
2588
2589
2590
2591
2592
2593
2594
2595
2596
2597
2598
2599
2600static bool can_steal_fallback(unsigned int order, int start_mt)
2601{
2602
2603
2604
2605
2606
2607
2608
2609 if (order >= pageblock_order)
2610 return true;
2611
2612 if (order >= pageblock_order / 2 ||
2613 start_mt == MIGRATE_RECLAIMABLE ||
2614 start_mt == MIGRATE_UNMOVABLE ||
2615 page_group_by_mobility_disabled)
2616 return true;
2617
2618 return false;
2619}
2620
2621static inline bool boost_watermark(struct zone *zone)
2622{
2623 unsigned long max_boost;
2624
2625 if (!watermark_boost_factor)
2626 return false;
2627
2628
2629
2630
2631
2632
2633 if ((pageblock_nr_pages * 4) > zone_managed_pages(zone))
2634 return false;
2635
2636 max_boost = mult_frac(zone->_watermark[WMARK_HIGH],
2637 watermark_boost_factor, 10000);
2638
2639
2640
2641
2642
2643
2644
2645
2646
2647 if (!max_boost)
2648 return false;
2649
2650 max_boost = max(pageblock_nr_pages, max_boost);
2651
2652 zone->watermark_boost = min(zone->watermark_boost + pageblock_nr_pages,
2653 max_boost);
2654
2655 return true;
2656}
2657
2658
2659
2660
2661
2662
2663
2664
2665
2666static void steal_suitable_fallback(struct zone *zone, struct page *page,
2667 unsigned int alloc_flags, int start_type, bool whole_block)
2668{
2669 unsigned int current_order = buddy_order(page);
2670 int free_pages, movable_pages, alike_pages;
2671 int old_block_type;
2672
2673 old_block_type = get_pageblock_migratetype(page);
2674
2675
2676
2677
2678
2679 if (is_migrate_highatomic(old_block_type))
2680 goto single_page;
2681
2682
2683 if (current_order >= pageblock_order) {
2684 change_pageblock_range(page, current_order, start_type);
2685 goto single_page;
2686 }
2687
2688
2689
2690
2691
2692
2693 if (boost_watermark(zone) && (alloc_flags & ALLOC_KSWAPD))
2694 set_bit(ZONE_BOOSTED_WATERMARK, &zone->flags);
2695
2696
2697 if (!whole_block)
2698 goto single_page;
2699
2700 free_pages = move_freepages_block(zone, page, start_type,
2701 &movable_pages);
2702
2703
2704
2705
2706
2707 if (start_type == MIGRATE_MOVABLE) {
2708 alike_pages = movable_pages;
2709 } else {
2710
2711
2712
2713
2714
2715
2716
2717 if (old_block_type == MIGRATE_MOVABLE)
2718 alike_pages = pageblock_nr_pages
2719 - (free_pages + movable_pages);
2720 else
2721 alike_pages = 0;
2722 }
2723
2724
2725 if (!free_pages)
2726 goto single_page;
2727
2728
2729
2730
2731
2732 if (free_pages + alike_pages >= (1 << (pageblock_order-1)) ||
2733 page_group_by_mobility_disabled)
2734 set_pageblock_migratetype(page, start_type);
2735
2736 return;
2737
2738single_page:
2739 move_to_free_list(page, zone, current_order, start_type);
2740}
2741
2742
2743
2744
2745
2746
2747
2748int find_suitable_fallback(struct free_area *area, unsigned int order,
2749 int migratetype, bool only_stealable, bool *can_steal)
2750{
2751 int i;
2752 int fallback_mt;
2753
2754 if (area->nr_free == 0)
2755 return -1;
2756
2757 *can_steal = false;
2758 for (i = 0;; i++) {
2759 fallback_mt = fallbacks[migratetype][i];
2760 if (fallback_mt == MIGRATE_TYPES)
2761 break;
2762
2763 if (free_area_empty(area, fallback_mt))
2764 continue;
2765
2766 if (can_steal_fallback(order, migratetype))
2767 *can_steal = true;
2768
2769 if (!only_stealable)
2770 return fallback_mt;
2771
2772 if (*can_steal)
2773 return fallback_mt;
2774 }
2775
2776 return -1;
2777}
2778
2779
2780
2781
2782
2783static void reserve_highatomic_pageblock(struct page *page, struct zone *zone,
2784 unsigned int alloc_order)
2785{
2786 int mt;
2787 unsigned long max_managed, flags;
2788
2789
2790
2791
2792
2793 max_managed = (zone_managed_pages(zone) / 100) + pageblock_nr_pages;
2794 if (zone->nr_reserved_highatomic >= max_managed)
2795 return;
2796
2797 spin_lock_irqsave(&zone->lock, flags);
2798
2799
2800 if (zone->nr_reserved_highatomic >= max_managed)
2801 goto out_unlock;
2802
2803
2804 mt = get_pageblock_migratetype(page);
2805 if (!is_migrate_highatomic(mt) && !is_migrate_isolate(mt)
2806 && !is_migrate_cma(mt)) {
2807 zone->nr_reserved_highatomic += pageblock_nr_pages;
2808 set_pageblock_migratetype(page, MIGRATE_HIGHATOMIC);
2809 move_freepages_block(zone, page, MIGRATE_HIGHATOMIC, NULL);
2810 }
2811
2812out_unlock:
2813 spin_unlock_irqrestore(&zone->lock, flags);
2814}
2815
2816
2817
2818
2819
2820
2821
2822
2823
2824
2825static bool unreserve_highatomic_pageblock(const struct alloc_context *ac,
2826 bool force)
2827{
2828 struct zonelist *zonelist = ac->zonelist;
2829 unsigned long flags;
2830 struct zoneref *z;
2831 struct zone *zone;
2832 struct page *page;
2833 int order;
2834 bool ret;
2835
2836 for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->highest_zoneidx,
2837 ac->nodemask) {
2838
2839
2840
2841
2842 if (!force && zone->nr_reserved_highatomic <=
2843 pageblock_nr_pages)
2844 continue;
2845
2846 spin_lock_irqsave(&zone->lock, flags);
2847 for (order = 0; order < MAX_ORDER; order++) {
2848 struct free_area *area = &(zone->free_area[order]);
2849
2850 page = get_page_from_free_area(area, MIGRATE_HIGHATOMIC);
2851 if (!page)
2852 continue;
2853
2854
2855
2856
2857
2858
2859
2860
2861 if (is_migrate_highatomic_page(page)) {
2862
2863
2864
2865
2866
2867
2868
2869 zone->nr_reserved_highatomic -= min(
2870 pageblock_nr_pages,
2871 zone->nr_reserved_highatomic);
2872 }
2873
2874
2875
2876
2877
2878
2879
2880
2881
2882
2883 set_pageblock_migratetype(page, ac->migratetype);
2884 ret = move_freepages_block(zone, page, ac->migratetype,
2885 NULL);
2886 if (ret) {
2887 spin_unlock_irqrestore(&zone->lock, flags);
2888 return ret;
2889 }
2890 }
2891 spin_unlock_irqrestore(&zone->lock, flags);
2892 }
2893
2894 return false;
2895}
2896
2897
2898
2899
2900
2901
2902
2903
2904
2905
2906
2907static __always_inline bool
2908__rmqueue_fallback(struct zone *zone, int order, int start_migratetype,
2909 unsigned int alloc_flags)
2910{
2911 struct free_area *area;
2912 int current_order;
2913 int min_order = order;
2914 struct page *page;
2915 int fallback_mt;
2916 bool can_steal;
2917
2918
2919
2920
2921
2922
2923 if (alloc_flags & ALLOC_NOFRAGMENT)
2924 min_order = pageblock_order;
2925
2926
2927
2928
2929
2930
2931 for (current_order = MAX_ORDER - 1; current_order >= min_order;
2932 --current_order) {
2933 area = &(zone->free_area[current_order]);
2934 fallback_mt = find_suitable_fallback(area, current_order,
2935 start_migratetype, false, &can_steal);
2936 if (fallback_mt == -1)
2937 continue;
2938
2939
2940
2941
2942
2943
2944
2945
2946
2947 if (!can_steal && start_migratetype == MIGRATE_MOVABLE
2948 && current_order > order)
2949 goto find_smallest;
2950
2951 goto do_steal;
2952 }
2953
2954 return false;
2955
2956find_smallest:
2957 for (current_order = order; current_order < MAX_ORDER;
2958 current_order++) {
2959 area = &(zone->free_area[current_order]);
2960 fallback_mt = find_suitable_fallback(area, current_order,
2961 start_migratetype, false, &can_steal);
2962 if (fallback_mt != -1)
2963 break;
2964 }
2965
2966
2967
2968
2969
2970 VM_BUG_ON(current_order == MAX_ORDER);
2971
2972do_steal:
2973 page = get_page_from_free_area(area, fallback_mt);
2974
2975 steal_suitable_fallback(zone, page, alloc_flags, start_migratetype,
2976 can_steal);
2977
2978 trace_mm_page_alloc_extfrag(page, order, current_order,
2979 start_migratetype, fallback_mt);
2980
2981 return true;
2982
2983}
2984
2985
2986
2987
2988
2989static __always_inline struct page *
2990__rmqueue(struct zone *zone, unsigned int order, int migratetype,
2991 unsigned int alloc_flags)
2992{
2993 struct page *page;
2994
2995 if (IS_ENABLED(CONFIG_CMA)) {
2996
2997
2998
2999
3000
3001 if (alloc_flags & ALLOC_CMA &&
3002 zone_page_state(zone, NR_FREE_CMA_PAGES) >
3003 zone_page_state(zone, NR_FREE_PAGES) / 2) {
3004 page = __rmqueue_cma_fallback(zone, order);
3005 if (page)
3006 goto out;
3007 }
3008 }
3009retry:
3010 page = __rmqueue_smallest(zone, order, migratetype);
3011 if (unlikely(!page)) {
3012 if (alloc_flags & ALLOC_CMA)
3013 page = __rmqueue_cma_fallback(zone, order);
3014
3015 if (!page && __rmqueue_fallback(zone, order, migratetype,
3016 alloc_flags))
3017 goto retry;
3018 }
3019out:
3020 if (page)
3021 trace_mm_page_alloc_zone_locked(page, order, migratetype);
3022 return page;
3023}
3024
3025
3026
3027
3028
3029
3030static int rmqueue_bulk(struct zone *zone, unsigned int order,
3031 unsigned long count, struct list_head *list,
3032 int migratetype, unsigned int alloc_flags)
3033{
3034 int i, allocated = 0;
3035
3036
3037
3038
3039
3040 spin_lock(&zone->lock);
3041 for (i = 0; i < count; ++i) {
3042 struct page *page = __rmqueue(zone, order, migratetype,
3043 alloc_flags);
3044 if (unlikely(page == NULL))
3045 break;
3046
3047 if (unlikely(check_pcp_refill(page)))
3048 continue;
3049
3050
3051
3052
3053
3054
3055
3056
3057
3058
3059
3060 list_add_tail(&page->lru, list);
3061 allocated++;
3062 if (is_migrate_cma(get_pcppage_migratetype(page)))
3063 __mod_zone_page_state(zone, NR_FREE_CMA_PAGES,
3064 -(1 << order));
3065 }
3066
3067
3068
3069
3070
3071
3072
3073 __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
3074 spin_unlock(&zone->lock);
3075 return allocated;
3076}
3077
3078#ifdef CONFIG_NUMA
3079
3080
3081
3082
3083
3084
3085
3086
3087void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
3088{
3089 unsigned long flags;
3090 int to_drain, batch;
3091
3092 local_lock_irqsave(&pagesets.lock, flags);
3093 batch = READ_ONCE(pcp->batch);
3094 to_drain = min(pcp->count, batch);
3095 if (to_drain > 0)
3096 free_pcppages_bulk(zone, to_drain, pcp);
3097 local_unlock_irqrestore(&pagesets.lock, flags);
3098}
3099#endif
3100
3101
3102
3103
3104
3105
3106
3107
3108static void drain_pages_zone(unsigned int cpu, struct zone *zone)
3109{
3110 unsigned long flags;
3111 struct per_cpu_pages *pcp;
3112
3113 local_lock_irqsave(&pagesets.lock, flags);
3114
3115 pcp = per_cpu_ptr(zone->per_cpu_pageset, cpu);
3116 if (pcp->count)
3117 free_pcppages_bulk(zone, pcp->count, pcp);
3118
3119 local_unlock_irqrestore(&pagesets.lock, flags);
3120}
3121
3122
3123
3124
3125
3126
3127
3128
3129static void drain_pages(unsigned int cpu)
3130{
3131 struct zone *zone;
3132
3133 for_each_populated_zone(zone) {
3134 drain_pages_zone(cpu, zone);
3135 }
3136}
3137
3138
3139
3140
3141
3142
3143
3144void drain_local_pages(struct zone *zone)
3145{
3146 int cpu = smp_processor_id();
3147
3148 if (zone)
3149 drain_pages_zone(cpu, zone);
3150 else
3151 drain_pages(cpu);
3152}
3153
3154static void drain_local_pages_wq(struct work_struct *work)
3155{
3156 struct pcpu_drain *drain;
3157
3158 drain = container_of(work, struct pcpu_drain, work);
3159
3160
3161
3162
3163
3164
3165
3166
3167 preempt_disable();
3168 drain_local_pages(drain->zone);
3169 preempt_enable();
3170}
3171
3172
3173
3174
3175
3176
3177
3178
3179
3180
3181
3182static void __drain_all_pages(struct zone *zone, bool force_all_cpus)
3183{
3184 int cpu;
3185
3186
3187
3188
3189
3190 static cpumask_t cpus_with_pcps;
3191
3192
3193
3194
3195
3196 if (WARN_ON_ONCE(!mm_percpu_wq))
3197 return;
3198
3199
3200
3201
3202
3203
3204 if (unlikely(!mutex_trylock(&pcpu_drain_mutex))) {
3205 if (!zone)
3206 return;
3207 mutex_lock(&pcpu_drain_mutex);
3208 }
3209
3210
3211
3212
3213
3214
3215
3216 for_each_online_cpu(cpu) {
3217 struct per_cpu_pages *pcp;
3218 struct zone *z;
3219 bool has_pcps = false;
3220
3221 if (force_all_cpus) {
3222
3223
3224
3225
3226 has_pcps = true;
3227 } else if (zone) {
3228 pcp = per_cpu_ptr(zone->per_cpu_pageset, cpu);
3229 if (pcp->count)
3230 has_pcps = true;
3231 } else {
3232 for_each_populated_zone(z) {
3233 pcp = per_cpu_ptr(z->per_cpu_pageset, cpu);
3234 if (pcp->count) {
3235 has_pcps = true;
3236 break;
3237 }
3238 }
3239 }
3240
3241 if (has_pcps)
3242 cpumask_set_cpu(cpu, &cpus_with_pcps);
3243 else
3244 cpumask_clear_cpu(cpu, &cpus_with_pcps);
3245 }
3246
3247 for_each_cpu(cpu, &cpus_with_pcps) {
3248 struct pcpu_drain *drain = per_cpu_ptr(&pcpu_drain, cpu);
3249
3250 drain->zone = zone;
3251 INIT_WORK(&drain->work, drain_local_pages_wq);
3252 queue_work_on(cpu, mm_percpu_wq, &drain->work);
3253 }
3254 for_each_cpu(cpu, &cpus_with_pcps)
3255 flush_work(&per_cpu_ptr(&pcpu_drain, cpu)->work);
3256
3257 mutex_unlock(&pcpu_drain_mutex);
3258}
3259
3260
3261
3262
3263
3264
3265
3266
3267void drain_all_pages(struct zone *zone)
3268{
3269 __drain_all_pages(zone, false);
3270}
3271
3272#ifdef CONFIG_HIBERNATION
3273
3274
3275
3276
3277#define WD_PAGE_COUNT (128*1024)
3278
3279void mark_free_pages(struct zone *zone)
3280{
3281 unsigned long pfn, max_zone_pfn, page_count = WD_PAGE_COUNT;
3282 unsigned long flags;
3283 unsigned int order, t;
3284 struct page *page;
3285
3286 if (zone_is_empty(zone))
3287 return;
3288
3289 spin_lock_irqsave(&zone->lock, flags);
3290
3291 max_zone_pfn = zone_end_pfn(zone);
3292 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
3293 if (pfn_valid(pfn)) {
3294 page = pfn_to_page(pfn);
3295
3296 if (!--page_count) {
3297 touch_nmi_watchdog();
3298 page_count = WD_PAGE_COUNT;
3299 }
3300
3301 if (page_zone(page) != zone)
3302 continue;
3303
3304 if (!swsusp_page_is_forbidden(page))
3305 swsusp_unset_page_free(page);
3306 }
3307
3308 for_each_migratetype_order(order, t) {
3309 list_for_each_entry(page,
3310 &zone->free_area[order].free_list[t], lru) {
3311 unsigned long i;
3312
3313 pfn = page_to_pfn(page);
3314 for (i = 0; i < (1UL << order); i++) {
3315 if (!--page_count) {
3316 touch_nmi_watchdog();
3317 page_count = WD_PAGE_COUNT;
3318 }
3319 swsusp_set_page_free(pfn_to_page(pfn + i));
3320 }
3321 }
3322 }
3323 spin_unlock_irqrestore(&zone->lock, flags);
3324}
3325#endif
3326
3327static bool free_unref_page_prepare(struct page *page, unsigned long pfn,
3328 unsigned int order)
3329{
3330 int migratetype;
3331
3332 if (!free_pcp_prepare(page, order))
3333 return false;
3334
3335 migratetype = get_pfnblock_migratetype(page, pfn);
3336 set_pcppage_migratetype(page, migratetype);
3337 return true;
3338}
3339
3340static int nr_pcp_free(struct per_cpu_pages *pcp, int high, int batch)
3341{
3342 int min_nr_free, max_nr_free;
3343
3344
3345 if (unlikely(high < batch))
3346 return 1;
3347
3348
3349 min_nr_free = batch;
3350 max_nr_free = high - batch;
3351
3352
3353
3354
3355
3356 batch <<= pcp->free_factor;
3357 if (batch < max_nr_free)
3358 pcp->free_factor++;
3359 batch = clamp(batch, min_nr_free, max_nr_free);
3360
3361 return batch;
3362}
3363
3364static int nr_pcp_high(struct per_cpu_pages *pcp, struct zone *zone)
3365{
3366 int high = READ_ONCE(pcp->high);
3367
3368 if (unlikely(!high))
3369 return 0;
3370
3371 if (!test_bit(ZONE_RECLAIM_ACTIVE, &zone->flags))
3372 return high;
3373
3374
3375
3376
3377
3378 return min(READ_ONCE(pcp->batch) << 2, high);
3379}
3380
3381static void free_unref_page_commit(struct page *page, unsigned long pfn,
3382 int migratetype, unsigned int order)
3383{
3384 struct zone *zone = page_zone(page);
3385 struct per_cpu_pages *pcp;
3386 int high;
3387 int pindex;
3388
3389 __count_vm_event(PGFREE);
3390 pcp = this_cpu_ptr(zone->per_cpu_pageset);
3391 pindex = order_to_pindex(migratetype, order);
3392 list_add(&page->lru, &pcp->lists[pindex]);
3393 pcp->count += 1 << order;
3394 high = nr_pcp_high(pcp, zone);
3395 if (pcp->count >= high) {
3396 int batch = READ_ONCE(pcp->batch);
3397
3398 free_pcppages_bulk(zone, nr_pcp_free(pcp, high, batch), pcp);
3399 }
3400}
3401
3402
3403
3404
3405void free_unref_page(struct page *page, unsigned int order)
3406{
3407 unsigned long flags;
3408 unsigned long pfn = page_to_pfn(page);
3409 int migratetype;
3410
3411 if (!free_unref_page_prepare(page, pfn, order))
3412 return;
3413
3414
3415
3416
3417
3418
3419
3420
3421 migratetype = get_pcppage_migratetype(page);
3422 if (unlikely(migratetype >= MIGRATE_PCPTYPES)) {
3423 if (unlikely(is_migrate_isolate(migratetype))) {
3424 free_one_page(page_zone(page), page, pfn, order, migratetype, FPI_NONE);
3425 return;
3426 }
3427 migratetype = MIGRATE_MOVABLE;
3428 }
3429
3430 local_lock_irqsave(&pagesets.lock, flags);
3431 free_unref_page_commit(page, pfn, migratetype, order);
3432 local_unlock_irqrestore(&pagesets.lock, flags);
3433}
3434
3435
3436
3437
3438void free_unref_page_list(struct list_head *list)
3439{
3440 struct page *page, *next;
3441 unsigned long flags, pfn;
3442 int batch_count = 0;
3443 int migratetype;
3444
3445
3446 list_for_each_entry_safe(page, next, list, lru) {
3447 pfn = page_to_pfn(page);
3448 if (!free_unref_page_prepare(page, pfn, 0))
3449 list_del(&page->lru);
3450
3451
3452
3453
3454
3455 migratetype = get_pcppage_migratetype(page);
3456 if (unlikely(is_migrate_isolate(migratetype))) {
3457 list_del(&page->lru);
3458 free_one_page(page_zone(page), page, pfn, 0, migratetype, FPI_NONE);
3459 continue;
3460 }
3461
3462 set_page_private(page, pfn);
3463 }
3464
3465 local_lock_irqsave(&pagesets.lock, flags);
3466 list_for_each_entry_safe(page, next, list, lru) {
3467 pfn = page_private(page);
3468 set_page_private(page, 0);
3469
3470
3471
3472
3473
3474 migratetype = get_pcppage_migratetype(page);
3475 if (unlikely(migratetype >= MIGRATE_PCPTYPES))
3476 migratetype = MIGRATE_MOVABLE;
3477
3478 trace_mm_page_free_batched(page);
3479 free_unref_page_commit(page, pfn, migratetype, 0);
3480
3481
3482
3483
3484
3485 if (++batch_count == SWAP_CLUSTER_MAX) {
3486 local_unlock_irqrestore(&pagesets.lock, flags);
3487 batch_count = 0;
3488 local_lock_irqsave(&pagesets.lock, flags);
3489 }
3490 }
3491 local_unlock_irqrestore(&pagesets.lock, flags);
3492}
3493
3494
3495
3496
3497
3498
3499
3500
3501
3502void split_page(struct page *page, unsigned int order)
3503{
3504 int i;
3505
3506 VM_BUG_ON_PAGE(PageCompound(page), page);
3507 VM_BUG_ON_PAGE(!page_count(page), page);
3508
3509 for (i = 1; i < (1 << order); i++)
3510 set_page_refcounted(page + i);
3511 split_page_owner(page, 1 << order);
3512 split_page_memcg(page, 1 << order);
3513}
3514EXPORT_SYMBOL_GPL(split_page);
3515
3516int __isolate_free_page(struct page *page, unsigned int order)
3517{
3518 unsigned long watermark;
3519 struct zone *zone;
3520 int mt;
3521
3522 BUG_ON(!PageBuddy(page));
3523
3524 zone = page_zone(page);
3525 mt = get_pageblock_migratetype(page);
3526
3527 if (!is_migrate_isolate(mt)) {
3528
3529
3530
3531
3532
3533
3534 watermark = zone->_watermark[WMARK_MIN] + (1UL << order);
3535 if (!zone_watermark_ok(zone, 0, watermark, 0, ALLOC_CMA))
3536 return 0;
3537
3538 __mod_zone_freepage_state(zone, -(1UL << order), mt);
3539 }
3540
3541
3542
3543 del_page_from_free_list(page, zone, order);
3544
3545
3546
3547
3548
3549 if (order >= pageblock_order - 1) {
3550 struct page *endpage = page + (1 << order) - 1;
3551 for (; page < endpage; page += pageblock_nr_pages) {
3552 int mt = get_pageblock_migratetype(page);
3553 if (!is_migrate_isolate(mt) && !is_migrate_cma(mt)
3554 && !is_migrate_highatomic(mt))
3555 set_pageblock_migratetype(page,
3556 MIGRATE_MOVABLE);
3557 }
3558 }
3559
3560
3561 return 1UL << order;
3562}
3563
3564
3565
3566
3567
3568
3569
3570
3571
3572
3573void __putback_isolated_page(struct page *page, unsigned int order, int mt)
3574{
3575 struct zone *zone = page_zone(page);
3576
3577
3578 lockdep_assert_held(&zone->lock);
3579
3580
3581 __free_one_page(page, page_to_pfn(page), zone, order, mt,
3582 FPI_SKIP_REPORT_NOTIFY | FPI_TO_TAIL);
3583}
3584
3585
3586
3587
3588
3589
3590static inline void zone_statistics(struct zone *preferred_zone, struct zone *z,
3591 long nr_account)
3592{
3593#ifdef CONFIG_NUMA
3594 enum numa_stat_item local_stat = NUMA_LOCAL;
3595
3596
3597 if (!static_branch_likely(&vm_numa_stat_key))
3598 return;
3599
3600 if (zone_to_nid(z) != numa_node_id())
3601 local_stat = NUMA_OTHER;
3602
3603 if (zone_to_nid(z) == zone_to_nid(preferred_zone))
3604 __count_numa_events(z, NUMA_HIT, nr_account);
3605 else {
3606 __count_numa_events(z, NUMA_MISS, nr_account);
3607 __count_numa_events(preferred_zone, NUMA_FOREIGN, nr_account);
3608 }
3609 __count_numa_events(z, local_stat, nr_account);
3610#endif
3611}
3612
3613
3614static inline
3615struct page *__rmqueue_pcplist(struct zone *zone, unsigned int order,
3616 int migratetype,
3617 unsigned int alloc_flags,
3618 struct per_cpu_pages *pcp,
3619 struct list_head *list)
3620{
3621 struct page *page;
3622
3623 do {
3624 if (list_empty(list)) {
3625 int batch = READ_ONCE(pcp->batch);
3626 int alloced;
3627
3628
3629
3630
3631
3632
3633
3634
3635 if (batch > 1)
3636 batch = max(batch >> order, 2);
3637 alloced = rmqueue_bulk(zone, order,
3638 batch, list,
3639 migratetype, alloc_flags);
3640
3641 pcp->count += alloced << order;
3642 if (unlikely(list_empty(list)))
3643 return NULL;
3644 }
3645
3646 page = list_first_entry(list, struct page, lru);
3647 list_del(&page->lru);
3648 pcp->count -= 1 << order;
3649 } while (check_new_pcp(page));
3650
3651 return page;
3652}
3653
3654
3655static struct page *rmqueue_pcplist(struct zone *preferred_zone,
3656 struct zone *zone, unsigned int order,
3657 gfp_t gfp_flags, int migratetype,
3658 unsigned int alloc_flags)
3659{
3660 struct per_cpu_pages *pcp;
3661 struct list_head *list;
3662 struct page *page;
3663 unsigned long flags;
3664
3665 local_lock_irqsave(&pagesets.lock, flags);
3666
3667
3668
3669
3670
3671
3672 pcp = this_cpu_ptr(zone->per_cpu_pageset);
3673 pcp->free_factor >>= 1;
3674 list = &pcp->lists[order_to_pindex(migratetype, order)];
3675 page = __rmqueue_pcplist(zone, order, migratetype, alloc_flags, pcp, list);
3676 local_unlock_irqrestore(&pagesets.lock, flags);
3677 if (page) {
3678 __count_zid_vm_events(PGALLOC, page_zonenum(page), 1);
3679 zone_statistics(preferred_zone, zone, 1);
3680 }
3681 return page;
3682}
3683
3684
3685
3686
3687static inline
3688struct page *rmqueue(struct zone *preferred_zone,
3689 struct zone *zone, unsigned int order,
3690 gfp_t gfp_flags, unsigned int alloc_flags,
3691 int migratetype)
3692{
3693 unsigned long flags;
3694 struct page *page;
3695
3696 if (likely(pcp_allowed_order(order))) {
3697
3698
3699
3700
3701 if (!IS_ENABLED(CONFIG_CMA) || alloc_flags & ALLOC_CMA ||
3702 migratetype != MIGRATE_MOVABLE) {
3703 page = rmqueue_pcplist(preferred_zone, zone, order,
3704 gfp_flags, migratetype, alloc_flags);
3705 goto out;
3706 }
3707 }
3708
3709
3710
3711
3712
3713 WARN_ON_ONCE((gfp_flags & __GFP_NOFAIL) && (order > 1));
3714 spin_lock_irqsave(&zone->lock, flags);
3715
3716 do {
3717 page = NULL;
3718
3719
3720
3721
3722
3723
3724 if (order > 0 && alloc_flags & ALLOC_HARDER) {
3725 page = __rmqueue_smallest(zone, order, MIGRATE_HIGHATOMIC);
3726 if (page)
3727 trace_mm_page_alloc_zone_locked(page, order, migratetype);
3728 }
3729 if (!page)
3730 page = __rmqueue(zone, order, migratetype, alloc_flags);
3731 } while (page && check_new_pages(page, order));
3732 if (!page)
3733 goto failed;
3734
3735 __mod_zone_freepage_state(zone, -(1 << order),
3736 get_pcppage_migratetype(page));
3737 spin_unlock_irqrestore(&zone->lock, flags);
3738
3739 __count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order);
3740 zone_statistics(preferred_zone, zone, 1);
3741
3742out:
3743
3744 if (test_bit(ZONE_BOOSTED_WATERMARK, &zone->flags)) {
3745 clear_bit(ZONE_BOOSTED_WATERMARK, &zone->flags);
3746 wakeup_kswapd(zone, 0, 0, zone_idx(zone));
3747 }
3748
3749 VM_BUG_ON_PAGE(page && bad_range(zone, page), page);
3750 return page;
3751
3752failed:
3753 spin_unlock_irqrestore(&zone->lock, flags);
3754 return NULL;
3755}
3756
3757#ifdef CONFIG_FAIL_PAGE_ALLOC
3758
3759static struct {
3760 struct fault_attr attr;
3761
3762 bool ignore_gfp_highmem;
3763 bool ignore_gfp_reclaim;
3764 u32 min_order;
3765} fail_page_alloc = {
3766 .attr = FAULT_ATTR_INITIALIZER,
3767 .ignore_gfp_reclaim = true,
3768 .ignore_gfp_highmem = true,
3769 .min_order = 1,
3770};
3771
3772static int __init setup_fail_page_alloc(char *str)
3773{
3774 return setup_fault_attr(&fail_page_alloc.attr, str);
3775}
3776__setup("fail_page_alloc=", setup_fail_page_alloc);
3777
3778static bool __should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
3779{
3780 if (order < fail_page_alloc.min_order)
3781 return false;
3782 if (gfp_mask & __GFP_NOFAIL)
3783 return false;
3784 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
3785 return false;
3786 if (fail_page_alloc.ignore_gfp_reclaim &&
3787 (gfp_mask & __GFP_DIRECT_RECLAIM))
3788 return false;
3789
3790 return should_fail(&fail_page_alloc.attr, 1 << order);
3791}
3792
3793#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
3794
3795static int __init fail_page_alloc_debugfs(void)
3796{
3797 umode_t mode = S_IFREG | 0600;
3798 struct dentry *dir;
3799
3800 dir = fault_create_debugfs_attr("fail_page_alloc", NULL,
3801 &fail_page_alloc.attr);
3802
3803 debugfs_create_bool("ignore-gfp-wait", mode, dir,
3804 &fail_page_alloc.ignore_gfp_reclaim);
3805 debugfs_create_bool("ignore-gfp-highmem", mode, dir,
3806 &fail_page_alloc.ignore_gfp_highmem);
3807 debugfs_create_u32("min-order", mode, dir, &fail_page_alloc.min_order);
3808
3809 return 0;
3810}
3811
3812late_initcall(fail_page_alloc_debugfs);
3813
3814#endif
3815
3816#else
3817
3818static inline bool __should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
3819{
3820 return false;
3821}
3822
3823#endif
3824
3825noinline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
3826{
3827 return __should_fail_alloc_page(gfp_mask, order);
3828}
3829ALLOW_ERROR_INJECTION(should_fail_alloc_page, TRUE);
3830
3831static inline long __zone_watermark_unusable_free(struct zone *z,
3832 unsigned int order, unsigned int alloc_flags)
3833{
3834 const bool alloc_harder = (alloc_flags & (ALLOC_HARDER|ALLOC_OOM));
3835 long unusable_free = (1 << order) - 1;
3836
3837
3838
3839
3840
3841
3842 if (likely(!alloc_harder))
3843 unusable_free += z->nr_reserved_highatomic;
3844
3845#ifdef CONFIG_CMA
3846
3847 if (!(alloc_flags & ALLOC_CMA))
3848 unusable_free += zone_page_state(z, NR_FREE_CMA_PAGES);
3849#endif
3850
3851 return unusable_free;
3852}
3853
3854
3855
3856
3857
3858
3859
3860bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
3861 int highest_zoneidx, unsigned int alloc_flags,
3862 long free_pages)
3863{
3864 long min = mark;
3865 int o;
3866 const bool alloc_harder = (alloc_flags & (ALLOC_HARDER|ALLOC_OOM));
3867
3868
3869 free_pages -= __zone_watermark_unusable_free(z, order, alloc_flags);
3870
3871 if (alloc_flags & ALLOC_HIGH)
3872 min -= min / 2;
3873
3874 if (unlikely(alloc_harder)) {
3875
3876
3877
3878
3879
3880
3881 if (alloc_flags & ALLOC_OOM)
3882 min -= min / 2;
3883 else
3884 min -= min / 4;
3885 }
3886
3887
3888
3889
3890
3891
3892 if (free_pages <= min + z->lowmem_reserve[highest_zoneidx])
3893 return false;
3894
3895
3896 if (!order)
3897 return true;
3898
3899
3900 for (o = order; o < MAX_ORDER; o++) {
3901 struct free_area *area = &z->free_area[o];
3902 int mt;
3903
3904 if (!area->nr_free)
3905 continue;
3906
3907 for (mt = 0; mt < MIGRATE_PCPTYPES; mt++) {
3908 if (!free_area_empty(area, mt))
3909 return true;
3910 }
3911
3912#ifdef CONFIG_CMA
3913 if ((alloc_flags & ALLOC_CMA) &&
3914 !free_area_empty(area, MIGRATE_CMA)) {
3915 return true;
3916 }
3917#endif
3918 if (alloc_harder && !free_area_empty(area, MIGRATE_HIGHATOMIC))
3919 return true;
3920 }
3921 return false;
3922}
3923
3924bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
3925 int highest_zoneidx, unsigned int alloc_flags)
3926{
3927 return __zone_watermark_ok(z, order, mark, highest_zoneidx, alloc_flags,
3928 zone_page_state(z, NR_FREE_PAGES));
3929}
3930
3931static inline bool zone_watermark_fast(struct zone *z, unsigned int order,
3932 unsigned long mark, int highest_zoneidx,
3933 unsigned int alloc_flags, gfp_t gfp_mask)
3934{
3935 long free_pages;
3936
3937 free_pages = zone_page_state(z, NR_FREE_PAGES);
3938
3939
3940
3941
3942
3943 if (!order) {
3944 long fast_free;
3945
3946 fast_free = free_pages;
3947 fast_free -= __zone_watermark_unusable_free(z, 0, alloc_flags);
3948 if (fast_free > mark + z->lowmem_reserve[highest_zoneidx])
3949 return true;
3950 }
3951
3952 if (__zone_watermark_ok(z, order, mark, highest_zoneidx, alloc_flags,
3953 free_pages))
3954 return true;
3955
3956
3957
3958
3959
3960
3961 if (unlikely(!order && (gfp_mask & __GFP_ATOMIC) && z->watermark_boost
3962 && ((alloc_flags & ALLOC_WMARK_MASK) == WMARK_MIN))) {
3963 mark = z->_watermark[WMARK_MIN];
3964 return __zone_watermark_ok(z, order, mark, highest_zoneidx,
3965 alloc_flags, free_pages);
3966 }
3967
3968 return false;
3969}
3970
3971bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
3972 unsigned long mark, int highest_zoneidx)
3973{
3974 long free_pages = zone_page_state(z, NR_FREE_PAGES);
3975
3976 if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark)
3977 free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES);
3978
3979 return __zone_watermark_ok(z, order, mark, highest_zoneidx, 0,
3980 free_pages);
3981}
3982
3983#ifdef CONFIG_NUMA
3984static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
3985{
3986 return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) <=
3987 node_reclaim_distance;
3988}
3989#else
3990static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
3991{
3992 return true;
3993}
3994#endif
3995
3996
3997
3998
3999
4000
4001
4002
4003
4004static inline unsigned int
4005alloc_flags_nofragment(struct zone *zone, gfp_t gfp_mask)
4006{
4007 unsigned int alloc_flags;
4008
4009
4010
4011
4012
4013 alloc_flags = (__force int) (gfp_mask & __GFP_KSWAPD_RECLAIM);
4014
4015#ifdef CONFIG_ZONE_DMA32
4016 if (!zone)
4017 return alloc_flags;
4018
4019 if (zone_idx(zone) != ZONE_NORMAL)
4020 return alloc_flags;
4021
4022
4023
4024
4025
4026
4027 BUILD_BUG_ON(ZONE_NORMAL - ZONE_DMA32 != 1);
4028 if (nr_online_nodes > 1 && !populated_zone(--zone))
4029 return alloc_flags;
4030
4031 alloc_flags |= ALLOC_NOFRAGMENT;
4032#endif
4033 return alloc_flags;
4034}
4035
4036
4037static inline unsigned int gfp_to_alloc_flags_cma(gfp_t gfp_mask,
4038 unsigned int alloc_flags)
4039{
4040#ifdef CONFIG_CMA
4041 if (gfp_migratetype(gfp_mask) == MIGRATE_MOVABLE)
4042 alloc_flags |= ALLOC_CMA;
4043#endif
4044 return alloc_flags;
4045}
4046
4047
4048
4049
4050
4051static struct page *
4052get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags,
4053 const struct alloc_context *ac)
4054{
4055 struct zoneref *z;
4056 struct zone *zone;
4057 struct pglist_data *last_pgdat_dirty_limit = NULL;
4058 bool no_fallback;
4059
4060retry:
4061
4062
4063
4064
4065 no_fallback = alloc_flags & ALLOC_NOFRAGMENT;
4066 z = ac->preferred_zoneref;
4067 for_next_zone_zonelist_nodemask(zone, z, ac->highest_zoneidx,
4068 ac->nodemask) {
4069 struct page *page;
4070 unsigned long mark;
4071
4072 if (cpusets_enabled() &&
4073 (alloc_flags & ALLOC_CPUSET) &&
4074 !__cpuset_zone_allowed(zone, gfp_mask))
4075 continue;
4076
4077
4078
4079
4080
4081
4082
4083
4084
4085
4086
4087
4088
4089
4090
4091
4092
4093
4094
4095 if (ac->spread_dirty_pages) {
4096 if (last_pgdat_dirty_limit == zone->zone_pgdat)
4097 continue;
4098
4099 if (!node_dirty_ok(zone->zone_pgdat)) {
4100 last_pgdat_dirty_limit = zone->zone_pgdat;
4101 continue;
4102 }
4103 }
4104
4105 if (no_fallback && nr_online_nodes > 1 &&
4106 zone != ac->preferred_zoneref->zone) {
4107 int local_nid;
4108
4109
4110
4111
4112
4113
4114 local_nid = zone_to_nid(ac->preferred_zoneref->zone);
4115 if (zone_to_nid(zone) != local_nid) {
4116 alloc_flags &= ~ALLOC_NOFRAGMENT;
4117 goto retry;
4118 }
4119 }
4120
4121 mark = wmark_pages(zone, alloc_flags & ALLOC_WMARK_MASK);
4122 if (!zone_watermark_fast(zone, order, mark,
4123 ac->highest_zoneidx, alloc_flags,
4124 gfp_mask)) {
4125 int ret;
4126
4127#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
4128
4129
4130
4131
4132 if (static_branch_unlikely(&deferred_pages)) {
4133 if (_deferred_grow_zone(zone, order))
4134 goto try_this_zone;
4135 }
4136#endif
4137
4138 BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
4139 if (alloc_flags & ALLOC_NO_WATERMARKS)
4140 goto try_this_zone;
4141
4142 if (!node_reclaim_enabled() ||
4143 !zone_allows_reclaim(ac->preferred_zoneref->zone, zone))
4144 continue;
4145
4146 ret = node_reclaim(zone->zone_pgdat, gfp_mask, order);
4147 switch (ret) {
4148 case NODE_RECLAIM_NOSCAN:
4149
4150 continue;
4151 case NODE_RECLAIM_FULL:
4152
4153 continue;
4154 default:
4155
4156 if (zone_watermark_ok(zone, order, mark,
4157 ac->highest_zoneidx, alloc_flags))
4158 goto try_this_zone;
4159
4160 continue;
4161 }
4162 }
4163
4164try_this_zone:
4165 page = rmqueue(ac->preferred_zoneref->zone, zone, order,
4166 gfp_mask, alloc_flags, ac->migratetype);
4167 if (page) {
4168 prep_new_page(page, order, gfp_mask, alloc_flags);
4169
4170
4171
4172
4173
4174 if (unlikely(order && (alloc_flags & ALLOC_HARDER)))
4175 reserve_highatomic_pageblock(page, zone, order);
4176
4177 return page;
4178 } else {
4179#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
4180
4181 if (static_branch_unlikely(&deferred_pages)) {
4182 if (_deferred_grow_zone(zone, order))
4183 goto try_this_zone;
4184 }
4185#endif
4186 }
4187 }
4188
4189
4190
4191
4192
4193 if (no_fallback) {
4194 alloc_flags &= ~ALLOC_NOFRAGMENT;
4195 goto retry;
4196 }
4197
4198 return NULL;
4199}
4200
4201static void warn_alloc_show_mem(gfp_t gfp_mask, nodemask_t *nodemask)
4202{
4203 unsigned int filter = SHOW_MEM_FILTER_NODES;
4204
4205
4206
4207
4208
4209
4210 if (!(gfp_mask & __GFP_NOMEMALLOC))
4211 if (tsk_is_oom_victim(current) ||
4212 (current->flags & (PF_MEMALLOC | PF_EXITING)))
4213 filter &= ~SHOW_MEM_FILTER_NODES;
4214 if (in_interrupt() || !(gfp_mask & __GFP_DIRECT_RECLAIM))
4215 filter &= ~SHOW_MEM_FILTER_NODES;
4216
4217 show_mem(filter, nodemask);
4218}
4219
4220void warn_alloc(gfp_t gfp_mask, nodemask_t *nodemask, const char *fmt, ...)
4221{
4222 struct va_format vaf;
4223 va_list args;
4224 static DEFINE_RATELIMIT_STATE(nopage_rs, 10*HZ, 1);
4225
4226 if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs))
4227 return;
4228
4229 va_start(args, fmt);
4230 vaf.fmt = fmt;
4231 vaf.va = &args;
4232 pr_warn("%s: %pV, mode:%#x(%pGg), nodemask=%*pbl",
4233 current->comm, &vaf, gfp_mask, &gfp_mask,
4234 nodemask_pr_args(nodemask));
4235 va_end(args);
4236
4237 cpuset_print_current_mems_allowed();
4238 pr_cont("\n");
4239 dump_stack();
4240 warn_alloc_show_mem(gfp_mask, nodemask);
4241}
4242
4243static inline struct page *
4244__alloc_pages_cpuset_fallback(gfp_t gfp_mask, unsigned int order,
4245 unsigned int alloc_flags,
4246 const struct alloc_context *ac)
4247{
4248 struct page *page;
4249
4250 page = get_page_from_freelist(gfp_mask, order,
4251 alloc_flags|ALLOC_CPUSET, ac);
4252
4253
4254
4255
4256 if (!page)
4257 page = get_page_from_freelist(gfp_mask, order,
4258 alloc_flags, ac);
4259
4260 return page;
4261}
4262
4263static inline struct page *
4264__alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
4265 const struct alloc_context *ac, unsigned long *did_some_progress)
4266{
4267 struct oom_control oc = {
4268 .zonelist = ac->zonelist,
4269 .nodemask = ac->nodemask,
4270 .memcg = NULL,
4271 .gfp_mask = gfp_mask,
4272 .order = order,
4273 };
4274 struct page *page;
4275
4276 *did_some_progress = 0;
4277
4278
4279
4280
4281
4282 if (!mutex_trylock(&oom_lock)) {
4283 *did_some_progress = 1;
4284 schedule_timeout_uninterruptible(1);
4285 return NULL;
4286 }
4287
4288
4289
4290
4291
4292
4293
4294
4295 page = get_page_from_freelist((gfp_mask | __GFP_HARDWALL) &
4296 ~__GFP_DIRECT_RECLAIM, order,
4297 ALLOC_WMARK_HIGH|ALLOC_CPUSET, ac);
4298 if (page)
4299 goto out;
4300
4301
4302 if (current->flags & PF_DUMPCORE)
4303 goto out;
4304
4305 if (order > PAGE_ALLOC_COSTLY_ORDER)
4306 goto out;
4307
4308
4309
4310
4311
4312
4313
4314
4315 if (gfp_mask & (__GFP_RETRY_MAYFAIL | __GFP_THISNODE))
4316 goto out;
4317
4318 if (ac->highest_zoneidx < ZONE_NORMAL)
4319 goto out;
4320 if (pm_suspended_storage())
4321 goto out;
4322
4323
4324
4325
4326
4327
4328
4329
4330
4331
4332
4333 if (out_of_memory(&oc) || WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) {
4334 *did_some_progress = 1;
4335
4336
4337
4338
4339
4340 if (gfp_mask & __GFP_NOFAIL)
4341 page = __alloc_pages_cpuset_fallback(gfp_mask, order,
4342 ALLOC_NO_WATERMARKS, ac);
4343 }
4344out:
4345 mutex_unlock(&oom_lock);
4346 return page;
4347}
4348
4349
4350
4351
4352
4353#define MAX_COMPACT_RETRIES 16
4354
4355#ifdef CONFIG_COMPACTION
4356
4357static struct page *
4358__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
4359 unsigned int alloc_flags, const struct alloc_context *ac,
4360 enum compact_priority prio, enum compact_result *compact_result)
4361{
4362 struct page *page = NULL;
4363 unsigned long pflags;
4364 unsigned int noreclaim_flag;
4365
4366 if (!order)
4367 return NULL;
4368
4369 psi_memstall_enter(&pflags);
4370 noreclaim_flag = memalloc_noreclaim_save();
4371
4372 *compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac,
4373 prio, &page);
4374
4375 memalloc_noreclaim_restore(noreclaim_flag);
4376 psi_memstall_leave(&pflags);
4377
4378 if (*compact_result == COMPACT_SKIPPED)
4379 return NULL;
4380
4381
4382
4383
4384 count_vm_event(COMPACTSTALL);
4385
4386
4387 if (page)
4388 prep_new_page(page, order, gfp_mask, alloc_flags);
4389
4390
4391 if (!page)
4392 page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
4393
4394 if (page) {
4395 struct zone *zone = page_zone(page);
4396
4397 zone->compact_blockskip_flush = false;
4398 compaction_defer_reset(zone, order, true);
4399 count_vm_event(COMPACTSUCCESS);
4400 return page;
4401 }
4402
4403
4404
4405
4406
4407 count_vm_event(COMPACTFAIL);
4408
4409 cond_resched();
4410
4411 return NULL;
4412}
4413
4414static inline bool
4415should_compact_retry(struct alloc_context *ac, int order, int alloc_flags,
4416 enum compact_result compact_result,
4417 enum compact_priority *compact_priority,
4418 int *compaction_retries)
4419{
4420 int max_retries = MAX_COMPACT_RETRIES;
4421 int min_priority;
4422 bool ret = false;
4423 int retries = *compaction_retries;
4424 enum compact_priority priority = *compact_priority;
4425
4426 if (!order)
4427 return false;
4428
4429 if (fatal_signal_pending(current))
4430 return false;
4431
4432 if (compaction_made_progress(compact_result))
4433 (*compaction_retries)++;
4434
4435
4436
4437
4438
4439
4440 if (compaction_failed(compact_result))
4441 goto check_priority;
4442
4443
4444
4445
4446
4447 if (compaction_needs_reclaim(compact_result)) {
4448 ret = compaction_zonelist_suitable(ac, order, alloc_flags);
4449 goto out;
4450 }
4451
4452
4453
4454
4455
4456
4457
4458 if (compaction_withdrawn(compact_result)) {
4459 goto check_priority;
4460 }
4461
4462
4463
4464
4465
4466
4467
4468
4469
4470 if (order > PAGE_ALLOC_COSTLY_ORDER)
4471 max_retries /= 4;
4472 if (*compaction_retries <= max_retries) {
4473 ret = true;
4474 goto out;
4475 }
4476
4477
4478
4479
4480
4481check_priority:
4482 min_priority = (order > PAGE_ALLOC_COSTLY_ORDER) ?
4483 MIN_COMPACT_COSTLY_PRIORITY : MIN_COMPACT_PRIORITY;
4484
4485 if (*compact_priority > min_priority) {
4486 (*compact_priority)--;
4487 *compaction_retries = 0;
4488 ret = true;
4489 }
4490out:
4491 trace_compact_retry(order, priority, compact_result, retries, max_retries, ret);
4492 return ret;
4493}
4494#else
4495static inline struct page *
4496__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
4497 unsigned int alloc_flags, const struct alloc_context *ac,
4498 enum compact_priority prio, enum compact_result *compact_result)
4499{
4500 *compact_result = COMPACT_SKIPPED;
4501 return NULL;
4502}
4503
4504static inline bool
4505should_compact_retry(struct alloc_context *ac, unsigned int order, int alloc_flags,
4506 enum compact_result compact_result,
4507 enum compact_priority *compact_priority,
4508 int *compaction_retries)
4509{
4510 struct zone *zone;
4511 struct zoneref *z;
4512
4513 if (!order || order > PAGE_ALLOC_COSTLY_ORDER)
4514 return false;
4515
4516
4517
4518
4519
4520
4521
4522 for_each_zone_zonelist_nodemask(zone, z, ac->zonelist,
4523 ac->highest_zoneidx, ac->nodemask) {
4524 if (zone_watermark_ok(zone, 0, min_wmark_pages(zone),
4525 ac->highest_zoneidx, alloc_flags))
4526 return true;
4527 }
4528 return false;
4529}
4530#endif
4531
4532#ifdef CONFIG_LOCKDEP
4533static struct lockdep_map __fs_reclaim_map =
4534 STATIC_LOCKDEP_MAP_INIT("fs_reclaim", &__fs_reclaim_map);
4535
4536static bool __need_reclaim(gfp_t gfp_mask)
4537{
4538
4539 if (!(gfp_mask & __GFP_DIRECT_RECLAIM))
4540 return false;
4541
4542
4543 if (current->flags & PF_MEMALLOC)
4544 return false;
4545
4546 if (gfp_mask & __GFP_NOLOCKDEP)
4547 return false;
4548
4549 return true;
4550}
4551
4552void __fs_reclaim_acquire(void)
4553{
4554 lock_map_acquire(&__fs_reclaim_map);
4555}
4556
4557void __fs_reclaim_release(void)
4558{
4559 lock_map_release(&__fs_reclaim_map);
4560}
4561
4562void fs_reclaim_acquire(gfp_t gfp_mask)
4563{
4564 gfp_mask = current_gfp_context(gfp_mask);
4565
4566 if (__need_reclaim(gfp_mask)) {
4567 if (gfp_mask & __GFP_FS)
4568 __fs_reclaim_acquire();
4569
4570#ifdef CONFIG_MMU_NOTIFIER
4571 lock_map_acquire(&__mmu_notifier_invalidate_range_start_map);
4572 lock_map_release(&__mmu_notifier_invalidate_range_start_map);
4573#endif
4574
4575 }
4576}
4577EXPORT_SYMBOL_GPL(fs_reclaim_acquire);
4578
4579void fs_reclaim_release(gfp_t gfp_mask)
4580{
4581 gfp_mask = current_gfp_context(gfp_mask);
4582
4583 if (__need_reclaim(gfp_mask)) {
4584 if (gfp_mask & __GFP_FS)
4585 __fs_reclaim_release();
4586 }
4587}
4588EXPORT_SYMBOL_GPL(fs_reclaim_release);
4589#endif
4590
4591
4592static unsigned long
4593__perform_reclaim(gfp_t gfp_mask, unsigned int order,
4594 const struct alloc_context *ac)
4595{
4596 unsigned int noreclaim_flag;
4597 unsigned long pflags, progress;
4598
4599 cond_resched();
4600
4601
4602 cpuset_memory_pressure_bump();
4603 psi_memstall_enter(&pflags);
4604 fs_reclaim_acquire(gfp_mask);
4605 noreclaim_flag = memalloc_noreclaim_save();
4606
4607 progress = try_to_free_pages(ac->zonelist, order, gfp_mask,
4608 ac->nodemask);
4609
4610 memalloc_noreclaim_restore(noreclaim_flag);
4611 fs_reclaim_release(gfp_mask);
4612 psi_memstall_leave(&pflags);
4613
4614 cond_resched();
4615
4616 return progress;
4617}
4618
4619
4620static inline struct page *
4621__alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
4622 unsigned int alloc_flags, const struct alloc_context *ac,
4623 unsigned long *did_some_progress)
4624{
4625 struct page *page = NULL;
4626 bool drained = false;
4627
4628 *did_some_progress = __perform_reclaim(gfp_mask, order, ac);
4629 if (unlikely(!(*did_some_progress)))
4630 return NULL;
4631
4632retry:
4633 page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
4634
4635
4636
4637
4638
4639
4640 if (!page && !drained) {
4641 unreserve_highatomic_pageblock(ac, false);
4642 drain_all_pages(NULL);
4643 drained = true;
4644 goto retry;
4645 }
4646
4647 return page;
4648}
4649
4650static void wake_all_kswapds(unsigned int order, gfp_t gfp_mask,
4651 const struct alloc_context *ac)
4652{
4653 struct zoneref *z;
4654 struct zone *zone;
4655 pg_data_t *last_pgdat = NULL;
4656 enum zone_type highest_zoneidx = ac->highest_zoneidx;
4657
4658 for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, highest_zoneidx,
4659 ac->nodemask) {
4660 if (last_pgdat != zone->zone_pgdat)
4661 wakeup_kswapd(zone, gfp_mask, order, highest_zoneidx);
4662 last_pgdat = zone->zone_pgdat;
4663 }
4664}
4665
4666static inline unsigned int
4667gfp_to_alloc_flags(gfp_t gfp_mask)
4668{
4669 unsigned int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
4670
4671
4672
4673
4674
4675
4676 BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
4677 BUILD_BUG_ON(__GFP_KSWAPD_RECLAIM != (__force gfp_t) ALLOC_KSWAPD);
4678
4679
4680
4681
4682
4683
4684
4685 alloc_flags |= (__force int)
4686 (gfp_mask & (__GFP_HIGH | __GFP_KSWAPD_RECLAIM));
4687
4688 if (gfp_mask & __GFP_ATOMIC) {
4689
4690
4691
4692
4693 if (!(gfp_mask & __GFP_NOMEMALLOC))
4694 alloc_flags |= ALLOC_HARDER;
4695
4696
4697
4698
4699 alloc_flags &= ~ALLOC_CPUSET;
4700 } else if (unlikely(rt_task(current)) && !in_interrupt())
4701 alloc_flags |= ALLOC_HARDER;
4702
4703 alloc_flags = gfp_to_alloc_flags_cma(gfp_mask, alloc_flags);
4704
4705 return alloc_flags;
4706}
4707
4708static bool oom_reserves_allowed(struct task_struct *tsk)
4709{
4710 if (!tsk_is_oom_victim(tsk))
4711 return false;
4712
4713
4714
4715
4716
4717 if (!IS_ENABLED(CONFIG_MMU) && !test_thread_flag(TIF_MEMDIE))
4718 return false;
4719
4720 return true;
4721}
4722
4723
4724
4725
4726
4727static inline int __gfp_pfmemalloc_flags(gfp_t gfp_mask)
4728{
4729 if (unlikely(gfp_mask & __GFP_NOMEMALLOC))
4730 return 0;
4731 if (gfp_mask & __GFP_MEMALLOC)
4732 return ALLOC_NO_WATERMARKS;
4733 if (in_serving_softirq() && (current->flags & PF_MEMALLOC))
4734 return ALLOC_NO_WATERMARKS;
4735 if (!in_interrupt()) {
4736 if (current->flags & PF_MEMALLOC)
4737 return ALLOC_NO_WATERMARKS;
4738 else if (oom_reserves_allowed(current))
4739 return ALLOC_OOM;
4740 }
4741
4742 return 0;
4743}
4744
4745bool gfp_pfmemalloc_allowed(gfp_t gfp_mask)
4746{
4747 return !!__gfp_pfmemalloc_flags(gfp_mask);
4748}
4749
4750
4751
4752
4753
4754
4755
4756
4757
4758
4759
4760static inline bool
4761should_reclaim_retry(gfp_t gfp_mask, unsigned order,
4762 struct alloc_context *ac, int alloc_flags,
4763 bool did_some_progress, int *no_progress_loops)
4764{
4765 struct zone *zone;
4766 struct zoneref *z;
4767 bool ret = false;
4768
4769
4770
4771
4772
4773
4774 if (did_some_progress && order <= PAGE_ALLOC_COSTLY_ORDER)
4775 *no_progress_loops = 0;
4776 else
4777 (*no_progress_loops)++;
4778
4779
4780
4781
4782
4783 if (*no_progress_loops > MAX_RECLAIM_RETRIES) {
4784
4785 return unreserve_highatomic_pageblock(ac, true);
4786 }
4787
4788
4789
4790
4791
4792
4793
4794 for_each_zone_zonelist_nodemask(zone, z, ac->zonelist,
4795 ac->highest_zoneidx, ac->nodemask) {
4796 unsigned long available;
4797 unsigned long reclaimable;
4798 unsigned long min_wmark = min_wmark_pages(zone);
4799 bool wmark;
4800
4801 available = reclaimable = zone_reclaimable_pages(zone);
4802 available += zone_page_state_snapshot(zone, NR_FREE_PAGES);
4803
4804
4805
4806
4807
4808 wmark = __zone_watermark_ok(zone, order, min_wmark,
4809 ac->highest_zoneidx, alloc_flags, available);
4810 trace_reclaim_retry_zone(z, order, reclaimable,
4811 available, min_wmark, *no_progress_loops, wmark);
4812 if (wmark) {
4813
4814
4815
4816
4817
4818
4819 if (!did_some_progress) {
4820 unsigned long write_pending;
4821
4822 write_pending = zone_page_state_snapshot(zone,
4823 NR_ZONE_WRITE_PENDING);
4824
4825 if (2 * write_pending > reclaimable) {
4826 congestion_wait(BLK_RW_ASYNC, HZ/10);
4827 return true;
4828 }
4829 }
4830
4831 ret = true;
4832 goto out;
4833 }
4834 }
4835
4836out:
4837
4838
4839
4840
4841
4842
4843
4844 if (current->flags & PF_WQ_WORKER)
4845 schedule_timeout_uninterruptible(1);
4846 else
4847 cond_resched();
4848 return ret;
4849}
4850
4851static inline bool
4852check_retry_cpuset(int cpuset_mems_cookie, struct alloc_context *ac)
4853{
4854
4855
4856
4857
4858
4859
4860
4861
4862
4863
4864
4865 if (cpusets_enabled() && ac->nodemask &&
4866 !cpuset_nodemask_valid_mems_allowed(ac->nodemask)) {
4867 ac->nodemask = NULL;
4868 return true;
4869 }
4870
4871
4872
4873
4874
4875
4876
4877
4878 if (read_mems_allowed_retry(cpuset_mems_cookie))
4879 return true;
4880
4881 return false;
4882}
4883
4884static inline struct page *
4885__alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
4886 struct alloc_context *ac)
4887{
4888 bool can_direct_reclaim = gfp_mask & __GFP_DIRECT_RECLAIM;
4889 const bool costly_order = order > PAGE_ALLOC_COSTLY_ORDER;
4890 struct page *page = NULL;
4891 unsigned int alloc_flags;
4892 unsigned long did_some_progress;
4893 enum compact_priority compact_priority;
4894 enum compact_result compact_result;
4895 int compaction_retries;
4896 int no_progress_loops;
4897 unsigned int cpuset_mems_cookie;
4898 int reserve_flags;
4899
4900
4901
4902
4903
4904 if (WARN_ON_ONCE((gfp_mask & (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)) ==
4905 (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)))
4906 gfp_mask &= ~__GFP_ATOMIC;
4907
4908retry_cpuset:
4909 compaction_retries = 0;
4910 no_progress_loops = 0;
4911 compact_priority = DEF_COMPACT_PRIORITY;
4912 cpuset_mems_cookie = read_mems_allowed_begin();
4913
4914
4915
4916
4917
4918
4919 alloc_flags = gfp_to_alloc_flags(gfp_mask);
4920
4921
4922
4923
4924
4925
4926
4927 ac->preferred_zoneref = first_zones_zonelist(ac->zonelist,
4928 ac->highest_zoneidx, ac->nodemask);
4929 if (!ac->preferred_zoneref->zone)
4930 goto nopage;
4931
4932 if (alloc_flags & ALLOC_KSWAPD)
4933 wake_all_kswapds(order, gfp_mask, ac);
4934
4935
4936
4937
4938
4939 page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
4940 if (page)
4941 goto got_pg;
4942
4943
4944
4945
4946
4947
4948
4949
4950
4951
4952 if (can_direct_reclaim &&
4953 (costly_order ||
4954 (order > 0 && ac->migratetype != MIGRATE_MOVABLE))
4955 && !gfp_pfmemalloc_allowed(gfp_mask)) {
4956 page = __alloc_pages_direct_compact(gfp_mask, order,
4957 alloc_flags, ac,
4958 INIT_COMPACT_PRIORITY,
4959 &compact_result);
4960 if (page)
4961 goto got_pg;
4962
4963
4964
4965
4966
4967 if (costly_order && (gfp_mask & __GFP_NORETRY)) {
4968
4969
4970
4971
4972
4973
4974
4975
4976
4977
4978
4979
4980
4981
4982
4983
4984
4985 if (compact_result == COMPACT_SKIPPED ||
4986 compact_result == COMPACT_DEFERRED)
4987 goto nopage;
4988
4989
4990
4991
4992
4993
4994 compact_priority = INIT_COMPACT_PRIORITY;
4995 }
4996 }
4997
4998retry:
4999
5000 if (alloc_flags & ALLOC_KSWAPD)
5001 wake_all_kswapds(order, gfp_mask, ac);
5002
5003 reserve_flags = __gfp_pfmemalloc_flags(gfp_mask);
5004 if (reserve_flags)
5005 alloc_flags = gfp_to_alloc_flags_cma(gfp_mask, reserve_flags);
5006
5007
5008
5009
5010
5011
5012 if (!(alloc_flags & ALLOC_CPUSET) || reserve_flags) {
5013 ac->nodemask = NULL;
5014 ac->preferred_zoneref = first_zones_zonelist(ac->zonelist,
5015 ac->highest_zoneidx, ac->nodemask);
5016 }
5017
5018
5019 page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
5020 if (page)
5021 goto got_pg;
5022
5023
5024 if (!can_direct_reclaim)
5025 goto nopage;
5026
5027
5028 if (current->flags & PF_MEMALLOC)
5029 goto nopage;
5030
5031
5032 page = __alloc_pages_direct_reclaim(gfp_mask, order, alloc_flags, ac,
5033 &did_some_progress);
5034 if (page)
5035 goto got_pg;
5036
5037
5038 page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, ac,
5039 compact_priority, &compact_result);
5040 if (page)
5041 goto got_pg;
5042
5043
5044 if (gfp_mask & __GFP_NORETRY)
5045 goto nopage;
5046
5047
5048
5049
5050
5051 if (costly_order && !(gfp_mask & __GFP_RETRY_MAYFAIL))
5052 goto nopage;
5053
5054 if (should_reclaim_retry(gfp_mask, order, ac, alloc_flags,
5055 did_some_progress > 0, &no_progress_loops))
5056 goto retry;
5057
5058
5059
5060
5061
5062
5063
5064 if (did_some_progress > 0 &&
5065 should_compact_retry(ac, order, alloc_flags,
5066 compact_result, &compact_priority,
5067 &compaction_retries))
5068 goto retry;
5069
5070
5071
5072 if (check_retry_cpuset(cpuset_mems_cookie, ac))
5073 goto retry_cpuset;
5074
5075
5076 page = __alloc_pages_may_oom(gfp_mask, order, ac, &did_some_progress);
5077 if (page)
5078 goto got_pg;
5079
5080
5081 if (tsk_is_oom_victim(current) &&
5082 (alloc_flags & ALLOC_OOM ||
5083 (gfp_mask & __GFP_NOMEMALLOC)))
5084 goto nopage;
5085
5086
5087 if (did_some_progress) {
5088 no_progress_loops = 0;
5089 goto retry;
5090 }
5091
5092nopage:
5093
5094 if (check_retry_cpuset(cpuset_mems_cookie, ac))
5095 goto retry_cpuset;
5096
5097
5098
5099
5100
5101 if (gfp_mask & __GFP_NOFAIL) {
5102
5103
5104
5105
5106 if (WARN_ON_ONCE(!can_direct_reclaim))
5107 goto fail;
5108
5109
5110
5111
5112
5113
5114 WARN_ON_ONCE(current->flags & PF_MEMALLOC);
5115
5116
5117
5118
5119
5120
5121
5122 WARN_ON_ONCE(order > PAGE_ALLOC_COSTLY_ORDER);
5123
5124
5125
5126
5127
5128
5129
5130 page = __alloc_pages_cpuset_fallback(gfp_mask, order, ALLOC_HARDER, ac);
5131 if (page)
5132 goto got_pg;
5133
5134 cond_resched();
5135 goto retry;
5136 }
5137fail:
5138 warn_alloc(gfp_mask, ac->nodemask,
5139 "page allocation failure: order:%u", order);
5140got_pg:
5141 return page;
5142}
5143
5144static inline bool prepare_alloc_pages(gfp_t gfp_mask, unsigned int order,
5145 int preferred_nid, nodemask_t *nodemask,
5146 struct alloc_context *ac, gfp_t *alloc_gfp,
5147 unsigned int *alloc_flags)
5148{
5149 ac->highest_zoneidx = gfp_zone(gfp_mask);
5150 ac->zonelist = node_zonelist(preferred_nid, gfp_mask);
5151 ac->nodemask = nodemask;
5152 ac->migratetype = gfp_migratetype(gfp_mask);
5153
5154 if (cpusets_enabled()) {
5155 *alloc_gfp |= __GFP_HARDWALL;
5156
5157
5158
5159
5160 if (!in_interrupt() && !ac->nodemask)
5161 ac->nodemask = &cpuset_current_mems_allowed;
5162 else
5163 *alloc_flags |= ALLOC_CPUSET;
5164 }
5165
5166 fs_reclaim_acquire(gfp_mask);
5167 fs_reclaim_release(gfp_mask);
5168
5169 might_sleep_if(gfp_mask & __GFP_DIRECT_RECLAIM);
5170
5171 if (should_fail_alloc_page(gfp_mask, order))
5172 return false;
5173
5174 *alloc_flags = gfp_to_alloc_flags_cma(gfp_mask, *alloc_flags);
5175
5176
5177 ac->spread_dirty_pages = (gfp_mask & __GFP_WRITE);
5178
5179
5180
5181
5182
5183
5184 ac->preferred_zoneref = first_zones_zonelist(ac->zonelist,
5185 ac->highest_zoneidx, ac->nodemask);
5186
5187 return true;
5188}
5189
5190
5191
5192
5193
5194
5195
5196
5197
5198
5199
5200
5201
5202
5203
5204
5205
5206
5207
5208
5209
5210unsigned long __alloc_pages_bulk(gfp_t gfp, int preferred_nid,
5211 nodemask_t *nodemask, int nr_pages,
5212 struct list_head *page_list,
5213 struct page **page_array)
5214{
5215 struct page *page;
5216 unsigned long flags;
5217 struct zone *zone;
5218 struct zoneref *z;
5219 struct per_cpu_pages *pcp;
5220 struct list_head *pcp_list;
5221 struct alloc_context ac;
5222 gfp_t alloc_gfp;
5223 unsigned int alloc_flags = ALLOC_WMARK_LOW;
5224 int nr_populated = 0, nr_account = 0;
5225
5226
5227
5228
5229
5230 while (page_array && nr_populated < nr_pages && page_array[nr_populated])
5231 nr_populated++;
5232
5233
5234 if (unlikely(nr_pages <= 0))
5235 goto out;
5236
5237
5238 if (unlikely(page_array && nr_pages - nr_populated == 0))
5239 goto out;
5240
5241
5242 if (nr_pages - nr_populated == 1)
5243 goto failed;
5244
5245#ifdef CONFIG_PAGE_OWNER
5246
5247
5248
5249
5250
5251
5252
5253 if (static_branch_unlikely(&page_owner_inited))
5254 goto failed;
5255#endif
5256
5257
5258 gfp &= gfp_allowed_mask;
5259 alloc_gfp = gfp;
5260 if (!prepare_alloc_pages(gfp, 0, preferred_nid, nodemask, &ac, &alloc_gfp, &alloc_flags))
5261 goto out;
5262 gfp = alloc_gfp;
5263
5264
5265 for_each_zone_zonelist_nodemask(zone, z, ac.zonelist, ac.highest_zoneidx, ac.nodemask) {
5266 unsigned long mark;
5267
5268 if (cpusets_enabled() && (alloc_flags & ALLOC_CPUSET) &&
5269 !__cpuset_zone_allowed(zone, gfp)) {
5270 continue;
5271 }
5272
5273 if (nr_online_nodes > 1 && zone != ac.preferred_zoneref->zone &&
5274 zone_to_nid(zone) != zone_to_nid(ac.preferred_zoneref->zone)) {
5275 goto failed;
5276 }
5277
5278 mark = wmark_pages(zone, alloc_flags & ALLOC_WMARK_MASK) + nr_pages;
5279 if (zone_watermark_fast(zone, 0, mark,
5280 zonelist_zone_idx(ac.preferred_zoneref),
5281 alloc_flags, gfp)) {
5282 break;
5283 }
5284 }
5285
5286
5287
5288
5289
5290 if (unlikely(!zone))
5291 goto failed;
5292
5293
5294 local_lock_irqsave(&pagesets.lock, flags);
5295 pcp = this_cpu_ptr(zone->per_cpu_pageset);
5296 pcp_list = &pcp->lists[order_to_pindex(ac.migratetype, 0)];
5297
5298 while (nr_populated < nr_pages) {
5299
5300
5301 if (page_array && page_array[nr_populated]) {
5302 nr_populated++;
5303 continue;
5304 }
5305
5306 page = __rmqueue_pcplist(zone, 0, ac.migratetype, alloc_flags,
5307 pcp, pcp_list);
5308 if (unlikely(!page)) {
5309
5310 if (!nr_populated)
5311 goto failed_irq;
5312 break;
5313 }
5314 nr_account++;
5315
5316 prep_new_page(page, 0, gfp, 0);
5317 if (page_list)
5318 list_add(&page->lru, page_list);
5319 else
5320 page_array[nr_populated] = page;
5321 nr_populated++;
5322 }
5323
5324 local_unlock_irqrestore(&pagesets.lock, flags);
5325
5326 __count_zid_vm_events(PGALLOC, zone_idx(zone), nr_account);
5327 zone_statistics(ac.preferred_zoneref->zone, zone, nr_account);
5328
5329out:
5330 return nr_populated;
5331
5332failed_irq:
5333 local_unlock_irqrestore(&pagesets.lock, flags);
5334
5335failed:
5336 page = __alloc_pages(gfp, 0, preferred_nid, nodemask);
5337 if (page) {
5338 if (page_list)
5339 list_add(&page->lru, page_list);
5340 else
5341 page_array[nr_populated] = page;
5342 nr_populated++;
5343 }
5344
5345 goto out;
5346}
5347EXPORT_SYMBOL_GPL(__alloc_pages_bulk);
5348
5349
5350
5351
5352struct page *__alloc_pages(gfp_t gfp, unsigned int order, int preferred_nid,
5353 nodemask_t *nodemask)
5354{
5355 struct page *page;
5356 unsigned int alloc_flags = ALLOC_WMARK_LOW;
5357 gfp_t alloc_gfp;
5358 struct alloc_context ac = { };
5359
5360
5361
5362
5363
5364 if (unlikely(order >= MAX_ORDER)) {
5365 WARN_ON_ONCE(!(gfp & __GFP_NOWARN));
5366 return NULL;
5367 }
5368
5369 gfp &= gfp_allowed_mask;
5370
5371
5372
5373
5374
5375
5376
5377 gfp = current_gfp_context(gfp);
5378 alloc_gfp = gfp;
5379 if (!prepare_alloc_pages(gfp, order, preferred_nid, nodemask, &ac,
5380 &alloc_gfp, &alloc_flags))
5381 return NULL;
5382
5383
5384
5385
5386
5387 alloc_flags |= alloc_flags_nofragment(ac.preferred_zoneref->zone, gfp);
5388
5389
5390 page = get_page_from_freelist(alloc_gfp, order, alloc_flags, &ac);
5391 if (likely(page))
5392 goto out;
5393
5394 alloc_gfp = gfp;
5395 ac.spread_dirty_pages = false;
5396
5397
5398
5399
5400
5401 ac.nodemask = nodemask;
5402
5403 page = __alloc_pages_slowpath(alloc_gfp, order, &ac);
5404
5405out:
5406 if (memcg_kmem_enabled() && (gfp & __GFP_ACCOUNT) && page &&
5407 unlikely(__memcg_kmem_charge_page(page, gfp, order) != 0)) {
5408 __free_pages(page, order);
5409 page = NULL;
5410 }
5411
5412 trace_mm_page_alloc(page, order, alloc_gfp, ac.migratetype);
5413
5414 return page;
5415}
5416EXPORT_SYMBOL(__alloc_pages);
5417
5418
5419
5420
5421
5422
5423unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
5424{
5425 struct page *page;
5426
5427 page = alloc_pages(gfp_mask & ~__GFP_HIGHMEM, order);
5428 if (!page)
5429 return 0;
5430 return (unsigned long) page_address(page);
5431}
5432EXPORT_SYMBOL(__get_free_pages);
5433
5434unsigned long get_zeroed_page(gfp_t gfp_mask)
5435{
5436 return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
5437}
5438EXPORT_SYMBOL(get_zeroed_page);
5439
5440
5441
5442
5443
5444
5445
5446
5447
5448
5449
5450
5451
5452
5453
5454
5455
5456
5457
5458
5459
5460void __free_pages(struct page *page, unsigned int order)
5461{
5462 if (put_page_testzero(page))
5463 free_the_page(page, order);
5464 else if (!PageHead(page))
5465 while (order-- > 0)
5466 free_the_page(page + (1 << order), order);
5467}
5468EXPORT_SYMBOL(__free_pages);
5469
5470void free_pages(unsigned long addr, unsigned int order)
5471{
5472 if (addr != 0) {
5473 VM_BUG_ON(!virt_addr_valid((void *)addr));
5474 __free_pages(virt_to_page((void *)addr), order);
5475 }
5476}
5477
5478EXPORT_SYMBOL(free_pages);
5479
5480
5481
5482
5483
5484
5485
5486
5487
5488
5489
5490
5491static struct page *__page_frag_cache_refill(struct page_frag_cache *nc,
5492 gfp_t gfp_mask)
5493{
5494 struct page *page = NULL;
5495 gfp_t gfp = gfp_mask;
5496
5497#if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
5498 gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY |
5499 __GFP_NOMEMALLOC;
5500 page = alloc_pages_node(NUMA_NO_NODE, gfp_mask,
5501 PAGE_FRAG_CACHE_MAX_ORDER);
5502 nc->size = page ? PAGE_FRAG_CACHE_MAX_SIZE : PAGE_SIZE;
5503#endif
5504 if (unlikely(!page))
5505 page = alloc_pages_node(NUMA_NO_NODE, gfp, 0);
5506
5507 nc->va = page ? page_address(page) : NULL;
5508
5509 return page;
5510}
5511
5512void __page_frag_cache_drain(struct page *page, unsigned int count)
5513{
5514 VM_BUG_ON_PAGE(page_ref_count(page) == 0, page);
5515
5516 if (page_ref_sub_and_test(page, count))
5517 free_the_page(page, compound_order(page));
5518}
5519EXPORT_SYMBOL(__page_frag_cache_drain);
5520
5521void *page_frag_alloc_align(struct page_frag_cache *nc,
5522 unsigned int fragsz, gfp_t gfp_mask,
5523 unsigned int align_mask)
5524{
5525 unsigned int size = PAGE_SIZE;
5526 struct page *page;
5527 int offset;
5528
5529 if (unlikely(!nc->va)) {
5530refill:
5531 page = __page_frag_cache_refill(nc, gfp_mask);
5532 if (!page)
5533 return NULL;
5534
5535#if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
5536
5537 size = nc->size;
5538#endif
5539
5540
5541
5542 page_ref_add(page, PAGE_FRAG_CACHE_MAX_SIZE);
5543
5544
5545 nc->pfmemalloc = page_is_pfmemalloc(page);
5546 nc->pagecnt_bias = PAGE_FRAG_CACHE_MAX_SIZE + 1;
5547 nc->offset = size;
5548 }
5549
5550 offset = nc->offset - fragsz;
5551 if (unlikely(offset < 0)) {
5552 page = virt_to_page(nc->va);
5553
5554 if (!page_ref_sub_and_test(page, nc->pagecnt_bias))
5555 goto refill;
5556
5557 if (unlikely(nc->pfmemalloc)) {
5558 free_the_page(page, compound_order(page));
5559 goto refill;
5560 }
5561
5562#if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
5563
5564 size = nc->size;
5565#endif
5566
5567 set_page_count(page, PAGE_FRAG_CACHE_MAX_SIZE + 1);
5568
5569
5570 nc->pagecnt_bias = PAGE_FRAG_CACHE_MAX_SIZE + 1;
5571 offset = size - fragsz;
5572 }
5573
5574 nc->pagecnt_bias--;
5575 offset &= align_mask;
5576 nc->offset = offset;
5577
5578 return nc->va + offset;
5579}
5580EXPORT_SYMBOL(page_frag_alloc_align);
5581
5582
5583
5584
5585void page_frag_free(void *addr)
5586{
5587 struct page *page = virt_to_head_page(addr);
5588
5589 if (unlikely(put_page_testzero(page)))
5590 free_the_page(page, compound_order(page));
5591}
5592EXPORT_SYMBOL(page_frag_free);
5593
5594static void *make_alloc_exact(unsigned long addr, unsigned int order,
5595 size_t size)
5596{
5597 if (addr) {
5598 unsigned long alloc_end = addr + (PAGE_SIZE << order);
5599 unsigned long used = addr + PAGE_ALIGN(size);
5600
5601 split_page(virt_to_page((void *)addr), order);
5602 while (used < alloc_end) {
5603 free_page(used);
5604 used += PAGE_SIZE;
5605 }
5606 }
5607 return (void *)addr;
5608}
5609
5610
5611
5612
5613
5614
5615
5616
5617
5618
5619
5620
5621
5622
5623
5624
5625void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
5626{
5627 unsigned int order = get_order(size);
5628 unsigned long addr;
5629
5630 if (WARN_ON_ONCE(gfp_mask & __GFP_COMP))
5631 gfp_mask &= ~__GFP_COMP;
5632
5633 addr = __get_free_pages(gfp_mask, order);
5634 return make_alloc_exact(addr, order, size);
5635}
5636EXPORT_SYMBOL(alloc_pages_exact);
5637
5638
5639
5640
5641
5642
5643
5644
5645
5646
5647
5648
5649
5650void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask)
5651{
5652 unsigned int order = get_order(size);
5653 struct page *p;
5654
5655 if (WARN_ON_ONCE(gfp_mask & __GFP_COMP))
5656 gfp_mask &= ~__GFP_COMP;
5657
5658 p = alloc_pages_node(nid, gfp_mask, order);
5659 if (!p)
5660 return NULL;
5661 return make_alloc_exact((unsigned long)page_address(p), order, size);
5662}
5663
5664
5665
5666
5667
5668
5669
5670
5671void free_pages_exact(void *virt, size_t size)
5672{
5673 unsigned long addr = (unsigned long)virt;
5674 unsigned long end = addr + PAGE_ALIGN(size);
5675
5676 while (addr < end) {
5677 free_page(addr);
5678 addr += PAGE_SIZE;
5679 }
5680}
5681EXPORT_SYMBOL(free_pages_exact);
5682
5683
5684
5685
5686
5687
5688
5689
5690
5691
5692
5693
5694
5695static unsigned long nr_free_zone_pages(int offset)
5696{
5697 struct zoneref *z;
5698 struct zone *zone;
5699
5700
5701 unsigned long sum = 0;
5702
5703 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
5704
5705 for_each_zone_zonelist(zone, z, zonelist, offset) {
5706 unsigned long size = zone_managed_pages(zone);
5707 unsigned long high = high_wmark_pages(zone);
5708 if (size > high)
5709 sum += size - high;
5710 }
5711
5712 return sum;
5713}
5714
5715
5716
5717
5718
5719
5720
5721
5722
5723
5724unsigned long nr_free_buffer_pages(void)
5725{
5726 return nr_free_zone_pages(gfp_zone(GFP_USER));
5727}
5728EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
5729
5730static inline void show_node(struct zone *zone)
5731{
5732 if (IS_ENABLED(CONFIG_NUMA))
5733 printk("Node %d ", zone_to_nid(zone));
5734}
5735
5736long si_mem_available(void)
5737{
5738 long available;
5739 unsigned long pagecache;
5740 unsigned long wmark_low = 0;
5741 unsigned long pages[NR_LRU_LISTS];
5742 unsigned long reclaimable;
5743 struct zone *zone;
5744 int lru;
5745
5746 for (lru = LRU_BASE; lru < NR_LRU_LISTS; lru++)
5747 pages[lru] = global_node_page_state(NR_LRU_BASE + lru);
5748
5749 for_each_zone(zone)
5750 wmark_low += low_wmark_pages(zone);
5751
5752
5753
5754
5755
5756 available = global_zone_page_state(NR_FREE_PAGES) - totalreserve_pages;
5757
5758
5759
5760
5761
5762
5763 pagecache = pages[LRU_ACTIVE_FILE] + pages[LRU_INACTIVE_FILE];
5764 pagecache -= min(pagecache / 2, wmark_low);
5765 available += pagecache;
5766
5767
5768
5769
5770
5771
5772 reclaimable = global_node_page_state_pages(NR_SLAB_RECLAIMABLE_B) +
5773 global_node_page_state(NR_KERNEL_MISC_RECLAIMABLE);
5774 available += reclaimable - min(reclaimable / 2, wmark_low);
5775
5776 if (available < 0)
5777 available = 0;
5778 return available;
5779}
5780EXPORT_SYMBOL_GPL(si_mem_available);
5781
5782void si_meminfo(struct sysinfo *val)
5783{
5784 val->totalram = totalram_pages();
5785 val->sharedram = global_node_page_state(NR_SHMEM);
5786 val->freeram = global_zone_page_state(NR_FREE_PAGES);
5787 val->bufferram = nr_blockdev_pages();
5788 val->totalhigh = totalhigh_pages();
5789 val->freehigh = nr_free_highpages();
5790 val->mem_unit = PAGE_SIZE;
5791}
5792
5793EXPORT_SYMBOL(si_meminfo);
5794
5795#ifdef CONFIG_NUMA
5796void si_meminfo_node(struct sysinfo *val, int nid)
5797{
5798 int zone_type;
5799 unsigned long managed_pages = 0;
5800 unsigned long managed_highpages = 0;
5801 unsigned long free_highpages = 0;
5802 pg_data_t *pgdat = NODE_DATA(nid);
5803
5804 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++)
5805 managed_pages += zone_managed_pages(&pgdat->node_zones[zone_type]);
5806 val->totalram = managed_pages;
5807 val->sharedram = node_page_state(pgdat, NR_SHMEM);
5808 val->freeram = sum_zone_node_page_state(nid, NR_FREE_PAGES);
5809#ifdef CONFIG_HIGHMEM
5810 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
5811 struct zone *zone = &pgdat->node_zones[zone_type];
5812
5813 if (is_highmem(zone)) {
5814 managed_highpages += zone_managed_pages(zone);
5815 free_highpages += zone_page_state(zone, NR_FREE_PAGES);
5816 }
5817 }
5818 val->totalhigh = managed_highpages;
5819 val->freehigh = free_highpages;
5820#else
5821 val->totalhigh = managed_highpages;
5822 val->freehigh = free_highpages;
5823#endif
5824 val->mem_unit = PAGE_SIZE;
5825}
5826#endif
5827
5828
5829
5830
5831
5832static bool show_mem_node_skip(unsigned int flags, int nid, nodemask_t *nodemask)
5833{
5834 if (!(flags & SHOW_MEM_FILTER_NODES))
5835 return false;
5836
5837
5838
5839
5840
5841
5842 if (!nodemask)
5843 nodemask = &cpuset_current_mems_allowed;
5844
5845 return !node_isset(nid, *nodemask);
5846}
5847
5848#define K(x) ((x) << (PAGE_SHIFT-10))
5849
5850static void show_migration_types(unsigned char type)
5851{
5852 static const char types[MIGRATE_TYPES] = {
5853 [MIGRATE_UNMOVABLE] = 'U',
5854 [MIGRATE_MOVABLE] = 'M',
5855 [MIGRATE_RECLAIMABLE] = 'E',
5856 [MIGRATE_HIGHATOMIC] = 'H',
5857#ifdef CONFIG_CMA
5858 [MIGRATE_CMA] = 'C',
5859#endif
5860#ifdef CONFIG_MEMORY_ISOLATION
5861 [MIGRATE_ISOLATE] = 'I',
5862#endif
5863 };
5864 char tmp[MIGRATE_TYPES + 1];
5865 char *p = tmp;
5866 int i;
5867
5868 for (i = 0; i < MIGRATE_TYPES; i++) {
5869 if (type & (1 << i))
5870 *p++ = types[i];
5871 }
5872
5873 *p = '\0';
5874 printk(KERN_CONT "(%s) ", tmp);
5875}
5876
5877
5878
5879
5880
5881
5882
5883
5884
5885
5886void show_free_areas(unsigned int filter, nodemask_t *nodemask)
5887{
5888 unsigned long free_pcp = 0;
5889 int cpu;
5890 struct zone *zone;
5891 pg_data_t *pgdat;
5892
5893 for_each_populated_zone(zone) {
5894 if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask))
5895 continue;
5896
5897 for_each_online_cpu(cpu)
5898 free_pcp += per_cpu_ptr(zone->per_cpu_pageset, cpu)->count;
5899 }
5900
5901 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
5902 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
5903 " unevictable:%lu dirty:%lu writeback:%lu\n"
5904 " slab_reclaimable:%lu slab_unreclaimable:%lu\n"
5905 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
5906 " free:%lu free_pcp:%lu free_cma:%lu\n",
5907 global_node_page_state(NR_ACTIVE_ANON),
5908 global_node_page_state(NR_INACTIVE_ANON),
5909 global_node_page_state(NR_ISOLATED_ANON),
5910 global_node_page_state(NR_ACTIVE_FILE),
5911 global_node_page_state(NR_INACTIVE_FILE),
5912 global_node_page_state(NR_ISOLATED_FILE),
5913 global_node_page_state(NR_UNEVICTABLE),
5914 global_node_page_state(NR_FILE_DIRTY),
5915 global_node_page_state(NR_WRITEBACK),
5916 global_node_page_state_pages(NR_SLAB_RECLAIMABLE_B),
5917 global_node_page_state_pages(NR_SLAB_UNRECLAIMABLE_B),
5918 global_node_page_state(NR_FILE_MAPPED),
5919 global_node_page_state(NR_SHMEM),
5920 global_node_page_state(NR_PAGETABLE),
5921 global_zone_page_state(NR_BOUNCE),
5922 global_zone_page_state(NR_FREE_PAGES),
5923 free_pcp,
5924 global_zone_page_state(NR_FREE_CMA_PAGES));
5925
5926 for_each_online_pgdat(pgdat) {
5927 if (show_mem_node_skip(filter, pgdat->node_id, nodemask))
5928 continue;
5929
5930 printk("Node %d"
5931 " active_anon:%lukB"
5932 " inactive_anon:%lukB"
5933 " active_file:%lukB"
5934 " inactive_file:%lukB"
5935 " unevictable:%lukB"
5936 " isolated(anon):%lukB"
5937 " isolated(file):%lukB"
5938 " mapped:%lukB"
5939 " dirty:%lukB"
5940 " writeback:%lukB"
5941 " shmem:%lukB"
5942#ifdef CONFIG_TRANSPARENT_HUGEPAGE
5943 " shmem_thp: %lukB"
5944 " shmem_pmdmapped: %lukB"
5945 " anon_thp: %lukB"
5946#endif
5947 " writeback_tmp:%lukB"
5948 " kernel_stack:%lukB"
5949#ifdef CONFIG_SHADOW_CALL_STACK
5950 " shadow_call_stack:%lukB"
5951#endif
5952 " pagetables:%lukB"
5953 " all_unreclaimable? %s"
5954 "\n",
5955 pgdat->node_id,
5956 K(node_page_state(pgdat, NR_ACTIVE_ANON)),
5957 K(node_page_state(pgdat, NR_INACTIVE_ANON)),
5958 K(node_page_state(pgdat, NR_ACTIVE_FILE)),
5959 K(node_page_state(pgdat, NR_INACTIVE_FILE)),
5960 K(node_page_state(pgdat, NR_UNEVICTABLE)),
5961 K(node_page_state(pgdat, NR_ISOLATED_ANON)),
5962 K(node_page_state(pgdat, NR_ISOLATED_FILE)),
5963 K(node_page_state(pgdat, NR_FILE_MAPPED)),
5964 K(node_page_state(pgdat, NR_FILE_DIRTY)),
5965 K(node_page_state(pgdat, NR_WRITEBACK)),
5966 K(node_page_state(pgdat, NR_SHMEM)),
5967#ifdef CONFIG_TRANSPARENT_HUGEPAGE
5968 K(node_page_state(pgdat, NR_SHMEM_THPS)),
5969 K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED)),
5970 K(node_page_state(pgdat, NR_ANON_THPS)),
5971#endif
5972 K(node_page_state(pgdat, NR_WRITEBACK_TEMP)),
5973 node_page_state(pgdat, NR_KERNEL_STACK_KB),
5974#ifdef CONFIG_SHADOW_CALL_STACK
5975 node_page_state(pgdat, NR_KERNEL_SCS_KB),
5976#endif
5977 K(node_page_state(pgdat, NR_PAGETABLE)),
5978 pgdat->kswapd_failures >= MAX_RECLAIM_RETRIES ?
5979 "yes" : "no");
5980 }
5981
5982 for_each_populated_zone(zone) {
5983 int i;
5984
5985 if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask))
5986 continue;
5987
5988 free_pcp = 0;
5989 for_each_online_cpu(cpu)
5990 free_pcp += per_cpu_ptr(zone->per_cpu_pageset, cpu)->count;
5991
5992 show_node(zone);
5993 printk(KERN_CONT
5994 "%s"
5995 " free:%lukB"
5996 " min:%lukB"
5997 " low:%lukB"
5998 " high:%lukB"
5999 " reserved_highatomic:%luKB"
6000 " active_anon:%lukB"
6001 " inactive_anon:%lukB"
6002 " active_file:%lukB"
6003 " inactive_file:%lukB"
6004 " unevictable:%lukB"
6005 " writepending:%lukB"
6006 " present:%lukB"
6007 " managed:%lukB"
6008 " mlocked:%lukB"
6009 " bounce:%lukB"
6010 " free_pcp:%lukB"
6011 " local_pcp:%ukB"
6012 " free_cma:%lukB"
6013 "\n",
6014 zone->name,
6015 K(zone_page_state(zone, NR_FREE_PAGES)),
6016 K(min_wmark_pages(zone)),
6017 K(low_wmark_pages(zone)),
6018 K(high_wmark_pages(zone)),
6019 K(zone->nr_reserved_highatomic),
6020 K(zone_page_state(zone, NR_ZONE_ACTIVE_ANON)),
6021 K(zone_page_state(zone, NR_ZONE_INACTIVE_ANON)),
6022 K(zone_page_state(zone, NR_ZONE_ACTIVE_FILE)),
6023 K(zone_page_state(zone, NR_ZONE_INACTIVE_FILE)),
6024 K(zone_page_state(zone, NR_ZONE_UNEVICTABLE)),
6025 K(zone_page_state(zone, NR_ZONE_WRITE_PENDING)),
6026 K(zone->present_pages),
6027 K(zone_managed_pages(zone)),
6028 K(zone_page_state(zone, NR_MLOCK)),
6029 K(zone_page_state(zone, NR_BOUNCE)),
6030 K(free_pcp),
6031 K(this_cpu_read(zone->per_cpu_pageset->count)),
6032 K(zone_page_state(zone, NR_FREE_CMA_PAGES)));
6033 printk("lowmem_reserve[]:");
6034 for (i = 0; i < MAX_NR_ZONES; i++)
6035 printk(KERN_CONT " %ld", zone->lowmem_reserve[i]);
6036 printk(KERN_CONT "\n");
6037 }
6038
6039 for_each_populated_zone(zone) {
6040 unsigned int order;
6041 unsigned long nr[MAX_ORDER], flags, total = 0;
6042 unsigned char types[MAX_ORDER];
6043
6044 if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask))
6045 continue;
6046 show_node(zone);
6047 printk(KERN_CONT "%s: ", zone->name);
6048
6049 spin_lock_irqsave(&zone->lock, flags);
6050 for (order = 0; order < MAX_ORDER; order++) {
6051 struct free_area *area = &zone->free_area[order];
6052 int type;
6053
6054 nr[order] = area->nr_free;
6055 total += nr[order] << order;
6056
6057 types[order] = 0;
6058 for (type = 0; type < MIGRATE_TYPES; type++) {
6059 if (!free_area_empty(area, type))
6060 types[order] |= 1 << type;
6061 }
6062 }
6063 spin_unlock_irqrestore(&zone->lock, flags);
6064 for (order = 0; order < MAX_ORDER; order++) {
6065 printk(KERN_CONT "%lu*%lukB ",
6066 nr[order], K(1UL) << order);
6067 if (nr[order])
6068 show_migration_types(types[order]);
6069 }
6070 printk(KERN_CONT "= %lukB\n", K(total));
6071 }
6072
6073 hugetlb_show_meminfo();
6074
6075 printk("%ld total pagecache pages\n", global_node_page_state(NR_FILE_PAGES));
6076
6077 show_swap_cache_info();
6078}
6079
6080static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
6081{
6082 zoneref->zone = zone;
6083 zoneref->zone_idx = zone_idx(zone);
6084}
6085
6086
6087
6088
6089
6090
6091static int build_zonerefs_node(pg_data_t *pgdat, struct zoneref *zonerefs)
6092{
6093 struct zone *zone;
6094 enum zone_type zone_type = MAX_NR_ZONES;
6095 int nr_zones = 0;
6096
6097 do {
6098 zone_type--;
6099 zone = pgdat->node_zones + zone_type;
6100 if (managed_zone(zone)) {
6101 zoneref_set_zone(zone, &zonerefs[nr_zones++]);
6102 check_highest_zone(zone_type);
6103 }
6104 } while (zone_type);
6105
6106 return nr_zones;
6107}
6108
6109#ifdef CONFIG_NUMA
6110
6111static int __parse_numa_zonelist_order(char *s)
6112{
6113
6114
6115
6116
6117
6118
6119 if (!(*s == 'd' || *s == 'D' || *s == 'n' || *s == 'N')) {
6120 pr_warn("Ignoring unsupported numa_zonelist_order value: %s\n", s);
6121 return -EINVAL;
6122 }
6123 return 0;
6124}
6125
6126char numa_zonelist_order[] = "Node";
6127
6128
6129
6130
6131int numa_zonelist_order_handler(struct ctl_table *table, int write,
6132 void *buffer, size_t *length, loff_t *ppos)
6133{
6134 if (write)
6135 return __parse_numa_zonelist_order(buffer);
6136 return proc_dostring(table, write, buffer, length, ppos);
6137}
6138
6139
6140#define MAX_NODE_LOAD (nr_online_nodes)
6141static int node_load[MAX_NUMNODES];
6142
6143
6144
6145
6146
6147
6148
6149
6150
6151
6152
6153
6154
6155
6156
6157
6158static int find_next_best_node(int node, nodemask_t *used_node_mask)
6159{
6160 int n, val;
6161 int min_val = INT_MAX;
6162 int best_node = NUMA_NO_NODE;
6163
6164
6165 if (!node_isset(node, *used_node_mask)) {
6166 node_set(node, *used_node_mask);
6167 return node;
6168 }
6169
6170 for_each_node_state(n, N_MEMORY) {
6171
6172
6173 if (node_isset(n, *used_node_mask))
6174 continue;
6175
6176
6177 val = node_distance(node, n);
6178
6179
6180 val += (n < node);
6181
6182
6183 if (!cpumask_empty(cpumask_of_node(n)))
6184 val += PENALTY_FOR_NODE_WITH_CPUS;
6185
6186
6187 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
6188 val += node_load[n];
6189
6190 if (val < min_val) {
6191 min_val = val;
6192 best_node = n;
6193 }
6194 }
6195
6196 if (best_node >= 0)
6197 node_set(best_node, *used_node_mask);
6198
6199 return best_node;
6200}
6201
6202
6203
6204
6205
6206
6207
6208static void build_zonelists_in_node_order(pg_data_t *pgdat, int *node_order,
6209 unsigned nr_nodes)
6210{
6211 struct zoneref *zonerefs;
6212 int i;
6213
6214 zonerefs = pgdat->node_zonelists[ZONELIST_FALLBACK]._zonerefs;
6215
6216 for (i = 0; i < nr_nodes; i++) {
6217 int nr_zones;
6218
6219 pg_data_t *node = NODE_DATA(node_order[i]);
6220
6221 nr_zones = build_zonerefs_node(node, zonerefs);
6222 zonerefs += nr_zones;
6223 }
6224 zonerefs->zone = NULL;
6225 zonerefs->zone_idx = 0;
6226}
6227
6228
6229
6230
6231static void build_thisnode_zonelists(pg_data_t *pgdat)
6232{
6233 struct zoneref *zonerefs;
6234 int nr_zones;
6235
6236 zonerefs = pgdat->node_zonelists[ZONELIST_NOFALLBACK]._zonerefs;
6237 nr_zones = build_zonerefs_node(pgdat, zonerefs);
6238 zonerefs += nr_zones;
6239 zonerefs->zone = NULL;
6240 zonerefs->zone_idx = 0;
6241}
6242
6243
6244
6245
6246
6247
6248
6249
6250static void build_zonelists(pg_data_t *pgdat)
6251{
6252 static int node_order[MAX_NUMNODES];
6253 int node, load, nr_nodes = 0;
6254 nodemask_t used_mask = NODE_MASK_NONE;
6255 int local_node, prev_node;
6256
6257
6258 local_node = pgdat->node_id;
6259 load = nr_online_nodes;
6260 prev_node = local_node;
6261
6262 memset(node_order, 0, sizeof(node_order));
6263 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
6264
6265
6266
6267
6268
6269 if (node_distance(local_node, node) !=
6270 node_distance(local_node, prev_node))
6271 node_load[node] = load;
6272
6273 node_order[nr_nodes++] = node;
6274 prev_node = node;
6275 load--;
6276 }
6277
6278 build_zonelists_in_node_order(pgdat, node_order, nr_nodes);
6279 build_thisnode_zonelists(pgdat);
6280}
6281
6282#ifdef CONFIG_HAVE_MEMORYLESS_NODES
6283
6284
6285
6286
6287
6288
6289int local_memory_node(int node)
6290{
6291 struct zoneref *z;
6292
6293 z = first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
6294 gfp_zone(GFP_KERNEL),
6295 NULL);
6296 return zone_to_nid(z->zone);
6297}
6298#endif
6299
6300static void setup_min_unmapped_ratio(void);
6301static void setup_min_slab_ratio(void);
6302#else
6303
6304static void build_zonelists(pg_data_t *pgdat)
6305{
6306 int node, local_node;
6307 struct zoneref *zonerefs;
6308 int nr_zones;
6309
6310 local_node = pgdat->node_id;
6311
6312 zonerefs = pgdat->node_zonelists[ZONELIST_FALLBACK]._zonerefs;
6313 nr_zones = build_zonerefs_node(pgdat, zonerefs);
6314 zonerefs += nr_zones;
6315
6316
6317
6318
6319
6320
6321
6322
6323
6324 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
6325 if (!node_online(node))
6326 continue;
6327 nr_zones = build_zonerefs_node(NODE_DATA(node), zonerefs);
6328 zonerefs += nr_zones;
6329 }
6330 for (node = 0; node < local_node; node++) {
6331 if (!node_online(node))
6332 continue;
6333 nr_zones = build_zonerefs_node(NODE_DATA(node), zonerefs);
6334 zonerefs += nr_zones;
6335 }
6336
6337 zonerefs->zone = NULL;
6338 zonerefs->zone_idx = 0;
6339}
6340
6341#endif
6342
6343
6344
6345
6346
6347
6348
6349
6350
6351
6352
6353
6354
6355
6356
6357
6358static void per_cpu_pages_init(struct per_cpu_pages *pcp, struct per_cpu_zonestat *pzstats);
6359
6360#define BOOT_PAGESET_HIGH 0
6361#define BOOT_PAGESET_BATCH 1
6362static DEFINE_PER_CPU(struct per_cpu_pages, boot_pageset);
6363static DEFINE_PER_CPU(struct per_cpu_zonestat, boot_zonestats);
6364static DEFINE_PER_CPU(struct per_cpu_nodestat, boot_nodestats);
6365
6366static void __build_all_zonelists(void *data)
6367{
6368 int nid;
6369 int __maybe_unused cpu;
6370 pg_data_t *self = data;
6371 static DEFINE_SPINLOCK(lock);
6372
6373 spin_lock(&lock);
6374
6375#ifdef CONFIG_NUMA
6376 memset(node_load, 0, sizeof(node_load));
6377#endif
6378
6379
6380
6381
6382
6383 if (self && !node_online(self->node_id)) {
6384 build_zonelists(self);
6385 } else {
6386 for_each_online_node(nid) {
6387 pg_data_t *pgdat = NODE_DATA(nid);
6388
6389 build_zonelists(pgdat);
6390 }
6391
6392#ifdef CONFIG_HAVE_MEMORYLESS_NODES
6393
6394
6395
6396
6397
6398
6399
6400
6401 for_each_online_cpu(cpu)
6402 set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
6403#endif
6404 }
6405
6406 spin_unlock(&lock);
6407}
6408
6409static noinline void __init
6410build_all_zonelists_init(void)
6411{
6412 int cpu;
6413
6414 __build_all_zonelists(NULL);
6415
6416
6417
6418
6419
6420
6421
6422
6423
6424
6425
6426
6427
6428
6429 for_each_possible_cpu(cpu)
6430 per_cpu_pages_init(&per_cpu(boot_pageset, cpu), &per_cpu(boot_zonestats, cpu));
6431
6432 mminit_verify_zonelist();
6433 cpuset_init_current_mems_allowed();
6434}
6435
6436
6437
6438
6439
6440
6441
6442void __ref build_all_zonelists(pg_data_t *pgdat)
6443{
6444 unsigned long vm_total_pages;
6445
6446 if (system_state == SYSTEM_BOOTING) {
6447 build_all_zonelists_init();
6448 } else {
6449 __build_all_zonelists(pgdat);
6450
6451 }
6452
6453 vm_total_pages = nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
6454
6455
6456
6457
6458
6459
6460
6461 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
6462 page_group_by_mobility_disabled = 1;
6463 else
6464 page_group_by_mobility_disabled = 0;
6465
6466 pr_info("Built %u zonelists, mobility grouping %s. Total pages: %ld\n",
6467 nr_online_nodes,
6468 page_group_by_mobility_disabled ? "off" : "on",
6469 vm_total_pages);
6470#ifdef CONFIG_NUMA
6471 pr_info("Policy zone: %s\n", zone_names[policy_zone]);
6472#endif
6473}
6474
6475
6476static bool __meminit
6477overlap_memmap_init(unsigned long zone, unsigned long *pfn)
6478{
6479 static struct memblock_region *r;
6480
6481 if (mirrored_kernelcore && zone == ZONE_MOVABLE) {
6482 if (!r || *pfn >= memblock_region_memory_end_pfn(r)) {
6483 for_each_mem_region(r) {
6484 if (*pfn < memblock_region_memory_end_pfn(r))
6485 break;
6486 }
6487 }
6488 if (*pfn >= memblock_region_memory_base_pfn(r) &&
6489 memblock_is_mirror(r)) {
6490 *pfn = memblock_region_memory_end_pfn(r);
6491 return true;
6492 }
6493 }
6494 return false;
6495}
6496
6497
6498
6499
6500
6501
6502
6503
6504
6505
6506void __meminit memmap_init_range(unsigned long size, int nid, unsigned long zone,
6507 unsigned long start_pfn, unsigned long zone_end_pfn,
6508 enum meminit_context context,
6509 struct vmem_altmap *altmap, int migratetype)
6510{
6511 unsigned long pfn, end_pfn = start_pfn + size;
6512 struct page *page;
6513
6514 if (highest_memmap_pfn < end_pfn - 1)
6515 highest_memmap_pfn = end_pfn - 1;
6516
6517#ifdef CONFIG_ZONE_DEVICE
6518
6519
6520
6521
6522
6523
6524
6525 if (zone == ZONE_DEVICE) {
6526 if (!altmap)
6527 return;
6528
6529 if (start_pfn == altmap->base_pfn)
6530 start_pfn += altmap->reserve;
6531 end_pfn = altmap->base_pfn + vmem_altmap_offset(altmap);
6532 }
6533#endif
6534
6535 for (pfn = start_pfn; pfn < end_pfn; ) {
6536
6537
6538
6539
6540 if (context == MEMINIT_EARLY) {
6541 if (overlap_memmap_init(zone, &pfn))
6542 continue;
6543 if (defer_init(nid, pfn, zone_end_pfn))
6544 break;
6545 }
6546
6547 page = pfn_to_page(pfn);
6548 __init_single_page(page, pfn, zone, nid);
6549 if (context == MEMINIT_HOTPLUG)
6550 __SetPageReserved(page);
6551
6552
6553
6554
6555
6556
6557 if (IS_ALIGNED(pfn, pageblock_nr_pages)) {
6558 set_pageblock_migratetype(page, migratetype);
6559 cond_resched();
6560 }
6561 pfn++;
6562 }
6563}
6564
6565#ifdef CONFIG_ZONE_DEVICE
6566void __ref memmap_init_zone_device(struct zone *zone,
6567 unsigned long start_pfn,
6568 unsigned long nr_pages,
6569 struct dev_pagemap *pgmap)
6570{
6571 unsigned long pfn, end_pfn = start_pfn + nr_pages;
6572 struct pglist_data *pgdat = zone->zone_pgdat;
6573 struct vmem_altmap *altmap = pgmap_altmap(pgmap);
6574 unsigned long zone_idx = zone_idx(zone);
6575 unsigned long start = jiffies;
6576 int nid = pgdat->node_id;
6577
6578 if (WARN_ON_ONCE(!pgmap || zone_idx(zone) != ZONE_DEVICE))
6579 return;
6580
6581
6582
6583
6584
6585
6586 if (altmap) {
6587 start_pfn = altmap->base_pfn + vmem_altmap_offset(altmap);
6588 nr_pages = end_pfn - start_pfn;
6589 }
6590
6591 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
6592 struct page *page = pfn_to_page(pfn);
6593
6594 __init_single_page(page, pfn, zone_idx, nid);
6595
6596
6597
6598
6599
6600
6601
6602
6603 __SetPageReserved(page);
6604
6605
6606
6607
6608
6609
6610 page->pgmap = pgmap;
6611 page->zone_device_data = NULL;
6612
6613
6614
6615
6616
6617
6618
6619
6620
6621
6622
6623 if (IS_ALIGNED(pfn, pageblock_nr_pages)) {
6624 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
6625 cond_resched();
6626 }
6627 }
6628
6629 pr_info("%s initialised %lu pages in %ums\n", __func__,
6630 nr_pages, jiffies_to_msecs(jiffies - start));
6631}
6632
6633#endif
6634static void __meminit zone_init_free_lists(struct zone *zone)
6635{
6636 unsigned int order, t;
6637 for_each_migratetype_order(order, t) {
6638 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
6639 zone->free_area[order].nr_free = 0;
6640 }
6641}
6642
6643#if !defined(CONFIG_FLATMEM)
6644
6645
6646
6647
6648
6649
6650
6651
6652
6653
6654
6655
6656
6657
6658
6659
6660
6661
6662
6663
6664
6665
6666static void __init init_unavailable_range(unsigned long spfn,
6667 unsigned long epfn,
6668 int zone, int node)
6669{
6670 unsigned long pfn;
6671 u64 pgcnt = 0;
6672
6673 for (pfn = spfn; pfn < epfn; pfn++) {
6674 if (!pfn_valid(ALIGN_DOWN(pfn, pageblock_nr_pages))) {
6675 pfn = ALIGN_DOWN(pfn, pageblock_nr_pages)
6676 + pageblock_nr_pages - 1;
6677 continue;
6678 }
6679 __init_single_page(pfn_to_page(pfn), pfn, zone, node);
6680 __SetPageReserved(pfn_to_page(pfn));
6681 pgcnt++;
6682 }
6683
6684 if (pgcnt)
6685 pr_info("On node %d, zone %s: %lld pages in unavailable ranges",
6686 node, zone_names[zone], pgcnt);
6687}
6688#else
6689static inline void init_unavailable_range(unsigned long spfn,
6690 unsigned long epfn,
6691 int zone, int node)
6692{
6693}
6694#endif
6695
6696static void __init memmap_init_zone_range(struct zone *zone,
6697 unsigned long start_pfn,
6698 unsigned long end_pfn,
6699 unsigned long *hole_pfn)
6700{
6701 unsigned long zone_start_pfn = zone->zone_start_pfn;
6702 unsigned long zone_end_pfn = zone_start_pfn + zone->spanned_pages;
6703 int nid = zone_to_nid(zone), zone_id = zone_idx(zone);
6704
6705 start_pfn = clamp(start_pfn, zone_start_pfn, zone_end_pfn);
6706 end_pfn = clamp(end_pfn, zone_start_pfn, zone_end_pfn);
6707
6708 if (start_pfn >= end_pfn)
6709 return;
6710
6711 memmap_init_range(end_pfn - start_pfn, nid, zone_id, start_pfn,
6712 zone_end_pfn, MEMINIT_EARLY, NULL, MIGRATE_MOVABLE);
6713
6714 if (*hole_pfn < start_pfn)
6715 init_unavailable_range(*hole_pfn, start_pfn, zone_id, nid);
6716
6717 *hole_pfn = end_pfn;
6718}
6719
6720static void __init memmap_init(void)
6721{
6722 unsigned long start_pfn, end_pfn;
6723 unsigned long hole_pfn = 0;
6724 int i, j, zone_id, nid;
6725
6726 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
6727 struct pglist_data *node = NODE_DATA(nid);
6728
6729 for (j = 0; j < MAX_NR_ZONES; j++) {
6730 struct zone *zone = node->node_zones + j;
6731
6732 if (!populated_zone(zone))
6733 continue;
6734
6735 memmap_init_zone_range(zone, start_pfn, end_pfn,
6736 &hole_pfn);
6737 zone_id = j;
6738 }
6739 }
6740
6741#ifdef CONFIG_SPARSEMEM
6742
6743
6744
6745
6746
6747
6748
6749
6750
6751 end_pfn = round_up(end_pfn, PAGES_PER_SECTION);
6752 if (hole_pfn < end_pfn)
6753#endif
6754 init_unavailable_range(hole_pfn, end_pfn, zone_id, nid);
6755}
6756
6757static int zone_batchsize(struct zone *zone)
6758{
6759#ifdef CONFIG_MMU
6760 int batch;
6761
6762
6763
6764
6765
6766
6767
6768 batch = min(zone_managed_pages(zone) >> 10, (1024 * 1024) / PAGE_SIZE);
6769 batch /= 4;
6770 if (batch < 1)
6771 batch = 1;
6772
6773
6774
6775
6776
6777
6778
6779
6780
6781
6782
6783 batch = rounddown_pow_of_two(batch + batch/2) - 1;
6784
6785 return batch;
6786
6787#else
6788
6789
6790
6791
6792
6793
6794
6795
6796
6797
6798
6799
6800
6801 return 0;
6802#endif
6803}
6804
6805static int zone_highsize(struct zone *zone, int batch, int cpu_online)
6806{
6807#ifdef CONFIG_MMU
6808 int high;
6809 int nr_split_cpus;
6810 unsigned long total_pages;
6811
6812 if (!percpu_pagelist_high_fraction) {
6813
6814
6815
6816
6817
6818 total_pages = low_wmark_pages(zone);
6819 } else {
6820
6821
6822
6823
6824
6825 total_pages = zone_managed_pages(zone) / percpu_pagelist_high_fraction;
6826 }
6827
6828
6829
6830
6831
6832
6833
6834
6835
6836 nr_split_cpus = cpumask_weight(cpumask_of_node(zone_to_nid(zone))) + cpu_online;
6837 if (!nr_split_cpus)
6838 nr_split_cpus = num_online_cpus();
6839 high = total_pages / nr_split_cpus;
6840
6841
6842
6843
6844
6845 high = max(high, batch << 2);
6846
6847 return high;
6848#else
6849 return 0;
6850#endif
6851}
6852
6853
6854
6855
6856
6857
6858
6859
6860
6861
6862
6863
6864
6865
6866
6867
6868
6869static void pageset_update(struct per_cpu_pages *pcp, unsigned long high,
6870 unsigned long batch)
6871{
6872 WRITE_ONCE(pcp->batch, batch);
6873 WRITE_ONCE(pcp->high, high);
6874}
6875
6876static void per_cpu_pages_init(struct per_cpu_pages *pcp, struct per_cpu_zonestat *pzstats)
6877{
6878 int pindex;
6879
6880 memset(pcp, 0, sizeof(*pcp));
6881 memset(pzstats, 0, sizeof(*pzstats));
6882
6883 for (pindex = 0; pindex < NR_PCP_LISTS; pindex++)
6884 INIT_LIST_HEAD(&pcp->lists[pindex]);
6885
6886
6887
6888
6889
6890
6891
6892 pcp->high = BOOT_PAGESET_HIGH;
6893 pcp->batch = BOOT_PAGESET_BATCH;
6894 pcp->free_factor = 0;
6895}
6896
6897static void __zone_set_pageset_high_and_batch(struct zone *zone, unsigned long high,
6898 unsigned long batch)
6899{
6900 struct per_cpu_pages *pcp;
6901 int cpu;
6902
6903 for_each_possible_cpu(cpu) {
6904 pcp = per_cpu_ptr(zone->per_cpu_pageset, cpu);
6905 pageset_update(pcp, high, batch);
6906 }
6907}
6908
6909
6910
6911
6912
6913static void zone_set_pageset_high_and_batch(struct zone *zone, int cpu_online)
6914{
6915 int new_high, new_batch;
6916
6917 new_batch = max(1, zone_batchsize(zone));
6918 new_high = zone_highsize(zone, new_batch, cpu_online);
6919
6920 if (zone->pageset_high == new_high &&
6921 zone->pageset_batch == new_batch)
6922 return;
6923
6924 zone->pageset_high = new_high;
6925 zone->pageset_batch = new_batch;
6926
6927 __zone_set_pageset_high_and_batch(zone, new_high, new_batch);
6928}
6929
6930void __meminit setup_zone_pageset(struct zone *zone)
6931{
6932 int cpu;
6933
6934
6935 if (sizeof(struct per_cpu_zonestat) > 0)
6936 zone->per_cpu_zonestats = alloc_percpu(struct per_cpu_zonestat);
6937
6938 zone->per_cpu_pageset = alloc_percpu(struct per_cpu_pages);
6939 for_each_possible_cpu(cpu) {
6940 struct per_cpu_pages *pcp;
6941 struct per_cpu_zonestat *pzstats;
6942
6943 pcp = per_cpu_ptr(zone->per_cpu_pageset, cpu);
6944 pzstats = per_cpu_ptr(zone->per_cpu_zonestats, cpu);
6945 per_cpu_pages_init(pcp, pzstats);
6946 }
6947
6948 zone_set_pageset_high_and_batch(zone, 0);
6949}
6950
6951
6952
6953
6954
6955void __init setup_per_cpu_pageset(void)
6956{
6957 struct pglist_data *pgdat;
6958 struct zone *zone;
6959 int __maybe_unused cpu;
6960
6961 for_each_populated_zone(zone)
6962 setup_zone_pageset(zone);
6963
6964#ifdef CONFIG_NUMA
6965
6966
6967
6968
6969
6970
6971 for_each_possible_cpu(cpu) {
6972 struct per_cpu_zonestat *pzstats = &per_cpu(boot_zonestats, cpu);
6973 memset(pzstats->vm_numa_event, 0,
6974 sizeof(pzstats->vm_numa_event));
6975 }
6976#endif
6977
6978 for_each_online_pgdat(pgdat)
6979 pgdat->per_cpu_nodestats =
6980 alloc_percpu(struct per_cpu_nodestat);
6981}
6982
6983static __meminit void zone_pcp_init(struct zone *zone)
6984{
6985
6986
6987
6988
6989
6990 zone->per_cpu_pageset = &boot_pageset;
6991 zone->per_cpu_zonestats = &boot_zonestats;
6992 zone->pageset_high = BOOT_PAGESET_HIGH;
6993 zone->pageset_batch = BOOT_PAGESET_BATCH;
6994
6995 if (populated_zone(zone))
6996 pr_debug(" %s zone: %lu pages, LIFO batch:%u\n", zone->name,
6997 zone->present_pages, zone_batchsize(zone));
6998}
6999
7000void __meminit init_currently_empty_zone(struct zone *zone,
7001 unsigned long zone_start_pfn,
7002 unsigned long size)
7003{
7004 struct pglist_data *pgdat = zone->zone_pgdat;
7005 int zone_idx = zone_idx(zone) + 1;
7006
7007 if (zone_idx > pgdat->nr_zones)
7008 pgdat->nr_zones = zone_idx;
7009
7010 zone->zone_start_pfn = zone_start_pfn;
7011
7012 mminit_dprintk(MMINIT_TRACE, "memmap_init",
7013 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
7014 pgdat->node_id,
7015 (unsigned long)zone_idx(zone),
7016 zone_start_pfn, (zone_start_pfn + size));
7017
7018 zone_init_free_lists(zone);
7019 zone->initialized = 1;
7020}
7021
7022
7023
7024
7025
7026
7027
7028
7029
7030
7031
7032
7033void __init get_pfn_range_for_nid(unsigned int nid,
7034 unsigned long *start_pfn, unsigned long *end_pfn)
7035{
7036 unsigned long this_start_pfn, this_end_pfn;
7037 int i;
7038
7039 *start_pfn = -1UL;
7040 *end_pfn = 0;
7041
7042 for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) {
7043 *start_pfn = min(*start_pfn, this_start_pfn);
7044 *end_pfn = max(*end_pfn, this_end_pfn);
7045 }
7046
7047 if (*start_pfn == -1UL)
7048 *start_pfn = 0;
7049}
7050
7051
7052
7053
7054
7055
7056static void __init find_usable_zone_for_movable(void)
7057{
7058 int zone_index;
7059 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
7060 if (zone_index == ZONE_MOVABLE)
7061 continue;
7062
7063 if (arch_zone_highest_possible_pfn[zone_index] >
7064 arch_zone_lowest_possible_pfn[zone_index])
7065 break;
7066 }
7067
7068 VM_BUG_ON(zone_index == -1);
7069 movable_zone = zone_index;
7070}
7071
7072
7073
7074
7075
7076
7077
7078
7079
7080
7081
7082static void __init adjust_zone_range_for_zone_movable(int nid,
7083 unsigned long zone_type,
7084 unsigned long node_start_pfn,
7085 unsigned long node_end_pfn,
7086 unsigned long *zone_start_pfn,
7087 unsigned long *zone_end_pfn)
7088{
7089
7090 if (zone_movable_pfn[nid]) {
7091
7092 if (zone_type == ZONE_MOVABLE) {
7093 *zone_start_pfn = zone_movable_pfn[nid];
7094 *zone_end_pfn = min(node_end_pfn,
7095 arch_zone_highest_possible_pfn[movable_zone]);
7096
7097
7098 } else if (!mirrored_kernelcore &&
7099 *zone_start_pfn < zone_movable_pfn[nid] &&
7100 *zone_end_pfn > zone_movable_pfn[nid]) {
7101 *zone_end_pfn = zone_movable_pfn[nid];
7102
7103
7104 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
7105 *zone_start_pfn = *zone_end_pfn;
7106 }
7107}
7108
7109
7110
7111
7112
7113static unsigned long __init zone_spanned_pages_in_node(int nid,
7114 unsigned long zone_type,
7115 unsigned long node_start_pfn,
7116 unsigned long node_end_pfn,
7117 unsigned long *zone_start_pfn,
7118 unsigned long *zone_end_pfn)
7119{
7120 unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type];
7121 unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type];
7122
7123 if (!node_start_pfn && !node_end_pfn)
7124 return 0;
7125
7126
7127 *zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high);
7128 *zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high);
7129 adjust_zone_range_for_zone_movable(nid, zone_type,
7130 node_start_pfn, node_end_pfn,
7131 zone_start_pfn, zone_end_pfn);
7132
7133
7134 if (*zone_end_pfn < node_start_pfn || *zone_start_pfn > node_end_pfn)
7135 return 0;
7136
7137
7138 *zone_end_pfn = min(*zone_end_pfn, node_end_pfn);
7139 *zone_start_pfn = max(*zone_start_pfn, node_start_pfn);
7140
7141
7142 return *zone_end_pfn - *zone_start_pfn;
7143}
7144
7145
7146
7147
7148
7149unsigned long __init __absent_pages_in_range(int nid,
7150 unsigned long range_start_pfn,
7151 unsigned long range_end_pfn)
7152{
7153 unsigned long nr_absent = range_end_pfn - range_start_pfn;
7154 unsigned long start_pfn, end_pfn;
7155 int i;
7156
7157 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
7158 start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn);
7159 end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn);
7160 nr_absent -= end_pfn - start_pfn;
7161 }
7162 return nr_absent;
7163}
7164
7165
7166
7167
7168
7169
7170
7171
7172unsigned long __init absent_pages_in_range(unsigned long start_pfn,
7173 unsigned long end_pfn)
7174{
7175 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
7176}
7177
7178
7179static unsigned long __init zone_absent_pages_in_node(int nid,
7180 unsigned long zone_type,
7181 unsigned long node_start_pfn,
7182 unsigned long node_end_pfn)
7183{
7184 unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type];
7185 unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type];
7186 unsigned long zone_start_pfn, zone_end_pfn;
7187 unsigned long nr_absent;
7188
7189
7190 if (!node_start_pfn && !node_end_pfn)
7191 return 0;
7192
7193 zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high);
7194 zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high);
7195
7196 adjust_zone_range_for_zone_movable(nid, zone_type,
7197 node_start_pfn, node_end_pfn,
7198 &zone_start_pfn, &zone_end_pfn);
7199 nr_absent = __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
7200
7201
7202
7203
7204
7205
7206 if (mirrored_kernelcore && zone_movable_pfn[nid]) {
7207 unsigned long start_pfn, end_pfn;
7208 struct memblock_region *r;
7209
7210 for_each_mem_region(r) {
7211 start_pfn = clamp(memblock_region_memory_base_pfn(r),
7212 zone_start_pfn, zone_end_pfn);
7213 end_pfn = clamp(memblock_region_memory_end_pfn(r),
7214 zone_start_pfn, zone_end_pfn);
7215
7216 if (zone_type == ZONE_MOVABLE &&
7217 memblock_is_mirror(r))
7218 nr_absent += end_pfn - start_pfn;
7219
7220 if (zone_type == ZONE_NORMAL &&
7221 !memblock_is_mirror(r))
7222 nr_absent += end_pfn - start_pfn;
7223 }
7224 }
7225
7226 return nr_absent;
7227}
7228
7229static void __init calculate_node_totalpages(struct pglist_data *pgdat,
7230 unsigned long node_start_pfn,
7231 unsigned long node_end_pfn)
7232{
7233 unsigned long realtotalpages = 0, totalpages = 0;
7234 enum zone_type i;
7235
7236 for (i = 0; i < MAX_NR_ZONES; i++) {
7237 struct zone *zone = pgdat->node_zones + i;
7238 unsigned long zone_start_pfn, zone_end_pfn;
7239 unsigned long spanned, absent;
7240 unsigned long size, real_size;
7241
7242 spanned = zone_spanned_pages_in_node(pgdat->node_id, i,
7243 node_start_pfn,
7244 node_end_pfn,
7245 &zone_start_pfn,
7246 &zone_end_pfn);
7247 absent = zone_absent_pages_in_node(pgdat->node_id, i,
7248 node_start_pfn,
7249 node_end_pfn);
7250
7251 size = spanned;
7252 real_size = size - absent;
7253
7254 if (size)
7255 zone->zone_start_pfn = zone_start_pfn;
7256 else
7257 zone->zone_start_pfn = 0;
7258 zone->spanned_pages = size;
7259 zone->present_pages = real_size;
7260
7261 totalpages += size;
7262 realtotalpages += real_size;
7263 }
7264
7265 pgdat->node_spanned_pages = totalpages;
7266 pgdat->node_present_pages = realtotalpages;
7267 pr_debug("On node %d totalpages: %lu\n", pgdat->node_id, realtotalpages);
7268}
7269
7270#ifndef CONFIG_SPARSEMEM
7271
7272
7273
7274
7275
7276
7277
7278static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize)
7279{
7280 unsigned long usemapsize;
7281
7282 zonesize += zone_start_pfn & (pageblock_nr_pages-1);
7283 usemapsize = roundup(zonesize, pageblock_nr_pages);
7284 usemapsize = usemapsize >> pageblock_order;
7285 usemapsize *= NR_PAGEBLOCK_BITS;
7286 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
7287
7288 return usemapsize / 8;
7289}
7290
7291static void __ref setup_usemap(struct zone *zone)
7292{
7293 unsigned long usemapsize = usemap_size(zone->zone_start_pfn,
7294 zone->spanned_pages);
7295 zone->pageblock_flags = NULL;
7296 if (usemapsize) {
7297 zone->pageblock_flags =
7298 memblock_alloc_node(usemapsize, SMP_CACHE_BYTES,
7299 zone_to_nid(zone));
7300 if (!zone->pageblock_flags)
7301 panic("Failed to allocate %ld bytes for zone %s pageblock flags on node %d\n",
7302 usemapsize, zone->name, zone_to_nid(zone));
7303 }
7304}
7305#else
7306static inline void setup_usemap(struct zone *zone) {}
7307#endif
7308
7309#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
7310
7311
7312void __init set_pageblock_order(void)
7313{
7314 unsigned int order;
7315
7316
7317 if (pageblock_order)
7318 return;
7319
7320 if (HPAGE_SHIFT > PAGE_SHIFT)
7321 order = HUGETLB_PAGE_ORDER;
7322 else
7323 order = MAX_ORDER - 1;
7324
7325
7326
7327
7328
7329
7330 pageblock_order = order;
7331}
7332#else
7333
7334
7335
7336
7337
7338
7339
7340void __init set_pageblock_order(void)
7341{
7342}
7343
7344#endif
7345
7346static unsigned long __init calc_memmap_size(unsigned long spanned_pages,
7347 unsigned long present_pages)
7348{
7349 unsigned long pages = spanned_pages;
7350
7351
7352
7353
7354
7355
7356
7357
7358
7359 if (spanned_pages > present_pages + (present_pages >> 4) &&
7360 IS_ENABLED(CONFIG_SPARSEMEM))
7361 pages = present_pages;
7362
7363 return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT;
7364}
7365
7366#ifdef CONFIG_TRANSPARENT_HUGEPAGE
7367static void pgdat_init_split_queue(struct pglist_data *pgdat)
7368{
7369 struct deferred_split *ds_queue = &pgdat->deferred_split_queue;
7370
7371 spin_lock_init(&ds_queue->split_queue_lock);
7372 INIT_LIST_HEAD(&ds_queue->split_queue);
7373 ds_queue->split_queue_len = 0;
7374}
7375#else
7376static void pgdat_init_split_queue(struct pglist_data *pgdat) {}
7377#endif
7378
7379#ifdef CONFIG_COMPACTION
7380static void pgdat_init_kcompactd(struct pglist_data *pgdat)
7381{
7382 init_waitqueue_head(&pgdat->kcompactd_wait);
7383}
7384#else
7385static void pgdat_init_kcompactd(struct pglist_data *pgdat) {}
7386#endif
7387
7388static void __meminit pgdat_init_internals(struct pglist_data *pgdat)
7389{
7390 pgdat_resize_init(pgdat);
7391
7392 pgdat_init_split_queue(pgdat);
7393 pgdat_init_kcompactd(pgdat);
7394
7395 init_waitqueue_head(&pgdat->kswapd_wait);
7396 init_waitqueue_head(&pgdat->pfmemalloc_wait);
7397
7398 pgdat_page_ext_init(pgdat);
7399 lruvec_init(&pgdat->__lruvec);
7400}
7401
7402static void __meminit zone_init_internals(struct zone *zone, enum zone_type idx, int nid,
7403 unsigned long remaining_pages)
7404{
7405 atomic_long_set(&zone->managed_pages, remaining_pages);
7406 zone_set_nid(zone, nid);
7407 zone->name = zone_names[idx];
7408 zone->zone_pgdat = NODE_DATA(nid);
7409 spin_lock_init(&zone->lock);
7410 zone_seqlock_init(zone);
7411 zone_pcp_init(zone);
7412}
7413
7414
7415
7416
7417
7418
7419
7420
7421#ifdef CONFIG_MEMORY_HOTPLUG
7422void __ref free_area_init_core_hotplug(int nid)
7423{
7424 enum zone_type z;
7425 pg_data_t *pgdat = NODE_DATA(nid);
7426
7427 pgdat_init_internals(pgdat);
7428 for (z = 0; z < MAX_NR_ZONES; z++)
7429 zone_init_internals(&pgdat->node_zones[z], z, nid, 0);
7430}
7431#endif
7432
7433
7434
7435
7436
7437
7438
7439
7440
7441
7442static void __init free_area_init_core(struct pglist_data *pgdat)
7443{
7444 enum zone_type j;
7445 int nid = pgdat->node_id;
7446
7447 pgdat_init_internals(pgdat);
7448 pgdat->per_cpu_nodestats = &boot_nodestats;
7449
7450 for (j = 0; j < MAX_NR_ZONES; j++) {
7451 struct zone *zone = pgdat->node_zones + j;
7452 unsigned long size, freesize, memmap_pages;
7453
7454 size = zone->spanned_pages;
7455 freesize = zone->present_pages;
7456
7457
7458
7459
7460
7461
7462 memmap_pages = calc_memmap_size(size, freesize);
7463 if (!is_highmem_idx(j)) {
7464 if (freesize >= memmap_pages) {
7465 freesize -= memmap_pages;
7466 if (memmap_pages)
7467 pr_debug(" %s zone: %lu pages used for memmap\n",
7468 zone_names[j], memmap_pages);
7469 } else
7470 pr_warn(" %s zone: %lu memmap pages exceeds freesize %lu\n",
7471 zone_names[j], memmap_pages, freesize);
7472 }
7473
7474
7475 if (j == 0 && freesize > dma_reserve) {
7476 freesize -= dma_reserve;
7477 pr_debug(" %s zone: %lu pages reserved\n", zone_names[0], dma_reserve);
7478 }
7479
7480 if (!is_highmem_idx(j))
7481 nr_kernel_pages += freesize;
7482
7483 else if (nr_kernel_pages > memmap_pages * 2)
7484 nr_kernel_pages -= memmap_pages;
7485 nr_all_pages += freesize;
7486
7487
7488
7489
7490
7491
7492 zone_init_internals(zone, j, nid, freesize);
7493
7494 if (!size)
7495 continue;
7496
7497 set_pageblock_order();
7498 setup_usemap(zone);
7499 init_currently_empty_zone(zone, zone->zone_start_pfn, size);
7500 }
7501}
7502
7503#ifdef CONFIG_FLATMEM
7504static void __ref alloc_node_mem_map(struct pglist_data *pgdat)
7505{
7506 unsigned long __maybe_unused start = 0;
7507 unsigned long __maybe_unused offset = 0;
7508
7509
7510 if (!pgdat->node_spanned_pages)
7511 return;
7512
7513 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
7514 offset = pgdat->node_start_pfn - start;
7515
7516 if (!pgdat->node_mem_map) {
7517 unsigned long size, end;
7518 struct page *map;
7519
7520
7521
7522
7523
7524
7525 end = pgdat_end_pfn(pgdat);
7526 end = ALIGN(end, MAX_ORDER_NR_PAGES);
7527 size = (end - start) * sizeof(struct page);
7528 map = memblock_alloc_node(size, SMP_CACHE_BYTES,
7529 pgdat->node_id);
7530 if (!map)
7531 panic("Failed to allocate %ld bytes for node %d memory map\n",
7532 size, pgdat->node_id);
7533 pgdat->node_mem_map = map + offset;
7534 }
7535 pr_debug("%s: node %d, pgdat %08lx, node_mem_map %08lx\n",
7536 __func__, pgdat->node_id, (unsigned long)pgdat,
7537 (unsigned long)pgdat->node_mem_map);
7538#ifndef CONFIG_NUMA
7539
7540
7541
7542 if (pgdat == NODE_DATA(0)) {
7543 mem_map = NODE_DATA(0)->node_mem_map;
7544 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
7545 mem_map -= offset;
7546 }
7547#endif
7548}
7549#else
7550static void __ref alloc_node_mem_map(struct pglist_data *pgdat) { }
7551#endif
7552
7553#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
7554static inline void pgdat_set_deferred_range(pg_data_t *pgdat)
7555{
7556 pgdat->first_deferred_pfn = ULONG_MAX;
7557}
7558#else
7559static inline void pgdat_set_deferred_range(pg_data_t *pgdat) {}
7560#endif
7561
7562static void __init free_area_init_node(int nid)
7563{
7564 pg_data_t *pgdat = NODE_DATA(nid);
7565 unsigned long start_pfn = 0;
7566 unsigned long end_pfn = 0;
7567
7568
7569 WARN_ON(pgdat->nr_zones || pgdat->kswapd_highest_zoneidx);
7570
7571 get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
7572
7573 pgdat->node_id = nid;
7574 pgdat->node_start_pfn = start_pfn;
7575 pgdat->per_cpu_nodestats = NULL;
7576
7577 pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid,
7578 (u64)start_pfn << PAGE_SHIFT,
7579 end_pfn ? ((u64)end_pfn << PAGE_SHIFT) - 1 : 0);
7580 calculate_node_totalpages(pgdat, start_pfn, end_pfn);
7581
7582 alloc_node_mem_map(pgdat);
7583 pgdat_set_deferred_range(pgdat);
7584
7585 free_area_init_core(pgdat);
7586}
7587
7588void __init free_area_init_memoryless_node(int nid)
7589{
7590 free_area_init_node(nid);
7591}
7592
7593#if MAX_NUMNODES > 1
7594
7595
7596
7597void __init setup_nr_node_ids(void)
7598{
7599 unsigned int highest;
7600
7601 highest = find_last_bit(node_possible_map.bits, MAX_NUMNODES);
7602 nr_node_ids = highest + 1;
7603}
7604#endif
7605
7606
7607
7608
7609
7610
7611
7612
7613
7614
7615
7616
7617
7618
7619
7620
7621
7622
7623
7624
7625unsigned long __init node_map_pfn_alignment(void)
7626{
7627 unsigned long accl_mask = 0, last_end = 0;
7628 unsigned long start, end, mask;
7629 int last_nid = NUMA_NO_NODE;
7630 int i, nid;
7631
7632 for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) {
7633 if (!start || last_nid < 0 || last_nid == nid) {
7634 last_nid = nid;
7635 last_end = end;
7636 continue;
7637 }
7638
7639
7640
7641
7642
7643
7644 mask = ~((1 << __ffs(start)) - 1);
7645 while (mask && last_end <= (start & (mask << 1)))
7646 mask <<= 1;
7647
7648
7649 accl_mask |= mask;
7650 }
7651
7652
7653 return ~accl_mask + 1;
7654}
7655
7656
7657
7658
7659
7660
7661
7662unsigned long __init find_min_pfn_with_active_regions(void)
7663{
7664 return PHYS_PFN(memblock_start_of_DRAM());
7665}
7666
7667
7668
7669
7670
7671
7672static unsigned long __init early_calculate_totalpages(void)
7673{
7674 unsigned long totalpages = 0;
7675 unsigned long start_pfn, end_pfn;
7676 int i, nid;
7677
7678 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
7679 unsigned long pages = end_pfn - start_pfn;
7680
7681 totalpages += pages;
7682 if (pages)
7683 node_set_state(nid, N_MEMORY);
7684 }
7685 return totalpages;
7686}
7687
7688
7689
7690
7691
7692
7693
7694static void __init find_zone_movable_pfns_for_nodes(void)
7695{
7696 int i, nid;
7697 unsigned long usable_startpfn;
7698 unsigned long kernelcore_node, kernelcore_remaining;
7699
7700 nodemask_t saved_node_state = node_states[N_MEMORY];
7701 unsigned long totalpages = early_calculate_totalpages();
7702 int usable_nodes = nodes_weight(node_states[N_MEMORY]);
7703 struct memblock_region *r;
7704
7705
7706 find_usable_zone_for_movable();
7707
7708
7709
7710
7711
7712 if (movable_node_is_enabled()) {
7713 for_each_mem_region(r) {
7714 if (!memblock_is_hotpluggable(r))
7715 continue;
7716
7717 nid = memblock_get_region_node(r);
7718
7719 usable_startpfn = PFN_DOWN(r->base);
7720 zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
7721 min(usable_startpfn, zone_movable_pfn[nid]) :
7722 usable_startpfn;
7723 }
7724
7725 goto out2;
7726 }
7727
7728
7729
7730
7731 if (mirrored_kernelcore) {
7732 bool mem_below_4gb_not_mirrored = false;
7733
7734 for_each_mem_region(r) {
7735 if (memblock_is_mirror(r))
7736 continue;
7737
7738 nid = memblock_get_region_node(r);
7739
7740 usable_startpfn = memblock_region_memory_base_pfn(r);
7741
7742 if (usable_startpfn < 0x100000) {
7743 mem_below_4gb_not_mirrored = true;
7744 continue;
7745 }
7746
7747 zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
7748 min(usable_startpfn, zone_movable_pfn[nid]) :
7749 usable_startpfn;
7750 }
7751
7752 if (mem_below_4gb_not_mirrored)
7753 pr_warn("This configuration results in unmirrored kernel memory.\n");
7754
7755 goto out2;
7756 }
7757
7758
7759
7760
7761
7762 if (required_kernelcore_percent)
7763 required_kernelcore = (totalpages * 100 * required_kernelcore_percent) /
7764 10000UL;
7765 if (required_movablecore_percent)
7766 required_movablecore = (totalpages * 100 * required_movablecore_percent) /
7767 10000UL;
7768
7769
7770
7771
7772
7773
7774
7775
7776
7777 if (required_movablecore) {
7778 unsigned long corepages;
7779
7780
7781
7782
7783
7784 required_movablecore =
7785 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
7786 required_movablecore = min(totalpages, required_movablecore);
7787 corepages = totalpages - required_movablecore;
7788
7789 required_kernelcore = max(required_kernelcore, corepages);
7790 }
7791
7792
7793
7794
7795
7796 if (!required_kernelcore || required_kernelcore >= totalpages)
7797 goto out;
7798
7799
7800 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
7801
7802restart:
7803
7804 kernelcore_node = required_kernelcore / usable_nodes;
7805 for_each_node_state(nid, N_MEMORY) {
7806 unsigned long start_pfn, end_pfn;
7807
7808
7809
7810
7811
7812
7813 if (required_kernelcore < kernelcore_node)
7814 kernelcore_node = required_kernelcore / usable_nodes;
7815
7816
7817
7818
7819
7820
7821 kernelcore_remaining = kernelcore_node;
7822
7823
7824 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
7825 unsigned long size_pages;
7826
7827 start_pfn = max(start_pfn, zone_movable_pfn[nid]);
7828 if (start_pfn >= end_pfn)
7829 continue;
7830
7831
7832 if (start_pfn < usable_startpfn) {
7833 unsigned long kernel_pages;
7834 kernel_pages = min(end_pfn, usable_startpfn)
7835 - start_pfn;
7836
7837 kernelcore_remaining -= min(kernel_pages,
7838 kernelcore_remaining);
7839 required_kernelcore -= min(kernel_pages,
7840 required_kernelcore);
7841
7842
7843 if (end_pfn <= usable_startpfn) {
7844
7845
7846
7847
7848
7849
7850
7851 zone_movable_pfn[nid] = end_pfn;
7852 continue;
7853 }
7854 start_pfn = usable_startpfn;
7855 }
7856
7857
7858
7859
7860
7861
7862 size_pages = end_pfn - start_pfn;
7863 if (size_pages > kernelcore_remaining)
7864 size_pages = kernelcore_remaining;
7865 zone_movable_pfn[nid] = start_pfn + size_pages;
7866
7867
7868
7869
7870
7871
7872 required_kernelcore -= min(required_kernelcore,
7873 size_pages);
7874 kernelcore_remaining -= size_pages;
7875 if (!kernelcore_remaining)
7876 break;
7877 }
7878 }
7879
7880
7881
7882
7883
7884
7885
7886 usable_nodes--;
7887 if (usable_nodes && required_kernelcore > usable_nodes)
7888 goto restart;
7889
7890out2:
7891
7892 for (nid = 0; nid < MAX_NUMNODES; nid++)
7893 zone_movable_pfn[nid] =
7894 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
7895
7896out:
7897
7898 node_states[N_MEMORY] = saved_node_state;
7899}
7900
7901
7902static void check_for_memory(pg_data_t *pgdat, int nid)
7903{
7904 enum zone_type zone_type;
7905
7906 for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) {
7907 struct zone *zone = &pgdat->node_zones[zone_type];
7908 if (populated_zone(zone)) {
7909 if (IS_ENABLED(CONFIG_HIGHMEM))
7910 node_set_state(nid, N_HIGH_MEMORY);
7911 if (zone_type <= ZONE_NORMAL)
7912 node_set_state(nid, N_NORMAL_MEMORY);
7913 break;
7914 }
7915 }
7916}
7917
7918
7919
7920
7921
7922bool __weak arch_has_descending_max_zone_pfns(void)
7923{
7924 return false;
7925}
7926
7927
7928
7929
7930
7931
7932
7933
7934
7935
7936
7937
7938
7939
7940void __init free_area_init(unsigned long *max_zone_pfn)
7941{
7942 unsigned long start_pfn, end_pfn;
7943 int i, nid, zone;
7944 bool descending;
7945
7946
7947 memset(arch_zone_lowest_possible_pfn, 0,
7948 sizeof(arch_zone_lowest_possible_pfn));
7949 memset(arch_zone_highest_possible_pfn, 0,
7950 sizeof(arch_zone_highest_possible_pfn));
7951
7952 start_pfn = find_min_pfn_with_active_regions();
7953 descending = arch_has_descending_max_zone_pfns();
7954
7955 for (i = 0; i < MAX_NR_ZONES; i++) {
7956 if (descending)
7957 zone = MAX_NR_ZONES - i - 1;
7958 else
7959 zone = i;
7960
7961 if (zone == ZONE_MOVABLE)
7962 continue;
7963
7964 end_pfn = max(max_zone_pfn[zone], start_pfn);
7965 arch_zone_lowest_possible_pfn[zone] = start_pfn;
7966 arch_zone_highest_possible_pfn[zone] = end_pfn;
7967
7968 start_pfn = end_pfn;
7969 }
7970
7971
7972 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
7973 find_zone_movable_pfns_for_nodes();
7974
7975
7976 pr_info("Zone ranges:\n");
7977 for (i = 0; i < MAX_NR_ZONES; i++) {
7978 if (i == ZONE_MOVABLE)
7979 continue;
7980 pr_info(" %-8s ", zone_names[i]);
7981 if (arch_zone_lowest_possible_pfn[i] ==
7982 arch_zone_highest_possible_pfn[i])
7983 pr_cont("empty\n");
7984 else
7985 pr_cont("[mem %#018Lx-%#018Lx]\n",
7986 (u64)arch_zone_lowest_possible_pfn[i]
7987 << PAGE_SHIFT,
7988 ((u64)arch_zone_highest_possible_pfn[i]
7989 << PAGE_SHIFT) - 1);
7990 }
7991
7992
7993 pr_info("Movable zone start for each node\n");
7994 for (i = 0; i < MAX_NUMNODES; i++) {
7995 if (zone_movable_pfn[i])
7996 pr_info(" Node %d: %#018Lx\n", i,
7997 (u64)zone_movable_pfn[i] << PAGE_SHIFT);
7998 }
7999
8000
8001
8002
8003
8004
8005 pr_info("Early memory node ranges\n");
8006 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
8007 pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid,
8008 (u64)start_pfn << PAGE_SHIFT,
8009 ((u64)end_pfn << PAGE_SHIFT) - 1);
8010 subsection_map_init(start_pfn, end_pfn - start_pfn);
8011 }
8012
8013
8014 mminit_verify_pageflags_layout();
8015 setup_nr_node_ids();
8016 for_each_online_node(nid) {
8017 pg_data_t *pgdat = NODE_DATA(nid);
8018 free_area_init_node(nid);
8019
8020
8021 if (pgdat->node_present_pages)
8022 node_set_state(nid, N_MEMORY);
8023 check_for_memory(pgdat, nid);
8024 }
8025
8026 memmap_init();
8027}
8028
8029static int __init cmdline_parse_core(char *p, unsigned long *core,
8030 unsigned long *percent)
8031{
8032 unsigned long long coremem;
8033 char *endptr;
8034
8035 if (!p)
8036 return -EINVAL;
8037
8038
8039 coremem = simple_strtoull(p, &endptr, 0);
8040 if (*endptr == '%') {
8041
8042 WARN_ON(coremem > 100);
8043
8044 *percent = coremem;
8045 } else {
8046 coremem = memparse(p, &p);
8047
8048 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
8049
8050 *core = coremem >> PAGE_SHIFT;
8051 *percent = 0UL;
8052 }
8053 return 0;
8054}
8055
8056
8057
8058
8059
8060static int __init cmdline_parse_kernelcore(char *p)
8061{
8062
8063 if (parse_option_str(p, "mirror")) {
8064 mirrored_kernelcore = true;
8065 return 0;
8066 }
8067
8068 return cmdline_parse_core(p, &required_kernelcore,
8069 &required_kernelcore_percent);
8070}
8071
8072
8073
8074
8075
8076static int __init cmdline_parse_movablecore(char *p)
8077{
8078 return cmdline_parse_core(p, &required_movablecore,
8079 &required_movablecore_percent);
8080}
8081
8082early_param("kernelcore", cmdline_parse_kernelcore);
8083early_param("movablecore", cmdline_parse_movablecore);
8084
8085void adjust_managed_page_count(struct page *page, long count)
8086{
8087 atomic_long_add(count, &page_zone(page)->managed_pages);
8088 totalram_pages_add(count);
8089#ifdef CONFIG_HIGHMEM
8090 if (PageHighMem(page))
8091 totalhigh_pages_add(count);
8092#endif
8093}
8094EXPORT_SYMBOL(adjust_managed_page_count);
8095
8096unsigned long free_reserved_area(void *start, void *end, int poison, const char *s)
8097{
8098 void *pos;
8099 unsigned long pages = 0;
8100
8101 start = (void *)PAGE_ALIGN((unsigned long)start);
8102 end = (void *)((unsigned long)end & PAGE_MASK);
8103 for (pos = start; pos < end; pos += PAGE_SIZE, pages++) {
8104 struct page *page = virt_to_page(pos);
8105 void *direct_map_addr;
8106
8107
8108
8109
8110
8111
8112
8113
8114 direct_map_addr = page_address(page);
8115
8116
8117
8118
8119 direct_map_addr = kasan_reset_tag(direct_map_addr);
8120 if ((unsigned int)poison <= 0xFF)
8121 memset(direct_map_addr, poison, PAGE_SIZE);
8122
8123 free_reserved_page(page);
8124 }
8125
8126 if (pages && s)
8127 pr_info("Freeing %s memory: %ldK\n",
8128 s, pages << (PAGE_SHIFT - 10));
8129
8130 return pages;
8131}
8132
8133void __init mem_init_print_info(void)
8134{
8135 unsigned long physpages, codesize, datasize, rosize, bss_size;
8136 unsigned long init_code_size, init_data_size;
8137
8138 physpages = get_num_physpages();
8139 codesize = _etext - _stext;
8140 datasize = _edata - _sdata;
8141 rosize = __end_rodata - __start_rodata;
8142 bss_size = __bss_stop - __bss_start;
8143 init_data_size = __init_end - __init_begin;
8144 init_code_size = _einittext - _sinittext;
8145
8146
8147
8148
8149
8150
8151
8152
8153#define adj_init_size(start, end, size, pos, adj) \
8154 do { \
8155 if (start <= pos && pos < end && size > adj) \
8156 size -= adj; \
8157 } while (0)
8158
8159 adj_init_size(__init_begin, __init_end, init_data_size,
8160 _sinittext, init_code_size);
8161 adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size);
8162 adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size);
8163 adj_init_size(_stext, _etext, codesize, __start_rodata, rosize);
8164 adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize);
8165
8166#undef adj_init_size
8167
8168 pr_info("Memory: %luK/%luK available (%luK kernel code, %luK rwdata, %luK rodata, %luK init, %luK bss, %luK reserved, %luK cma-reserved"
8169#ifdef CONFIG_HIGHMEM
8170 ", %luK highmem"
8171#endif
8172 ")\n",
8173 nr_free_pages() << (PAGE_SHIFT - 10),
8174 physpages << (PAGE_SHIFT - 10),
8175 codesize >> 10, datasize >> 10, rosize >> 10,
8176 (init_data_size + init_code_size) >> 10, bss_size >> 10,
8177 (physpages - totalram_pages() - totalcma_pages) << (PAGE_SHIFT - 10),
8178 totalcma_pages << (PAGE_SHIFT - 10)
8179#ifdef CONFIG_HIGHMEM
8180 , totalhigh_pages() << (PAGE_SHIFT - 10)
8181#endif
8182 );
8183}
8184
8185
8186
8187
8188
8189
8190
8191
8192
8193
8194
8195
8196void __init set_dma_reserve(unsigned long new_dma_reserve)
8197{
8198 dma_reserve = new_dma_reserve;
8199}
8200
8201static int page_alloc_cpu_dead(unsigned int cpu)
8202{
8203 struct zone *zone;
8204
8205 lru_add_drain_cpu(cpu);
8206 drain_pages(cpu);
8207
8208
8209
8210
8211
8212
8213
8214 vm_events_fold_cpu(cpu);
8215
8216
8217
8218
8219
8220
8221
8222
8223 cpu_vm_stats_fold(cpu);
8224
8225 for_each_populated_zone(zone)
8226 zone_pcp_update(zone, 0);
8227
8228 return 0;
8229}
8230
8231static int page_alloc_cpu_online(unsigned int cpu)
8232{
8233 struct zone *zone;
8234
8235 for_each_populated_zone(zone)
8236 zone_pcp_update(zone, 1);
8237 return 0;
8238}
8239
8240#ifdef CONFIG_NUMA
8241int hashdist = HASHDIST_DEFAULT;
8242
8243static int __init set_hashdist(char *str)
8244{
8245 if (!str)
8246 return 0;
8247 hashdist = simple_strtoul(str, &str, 0);
8248 return 1;
8249}
8250__setup("hashdist=", set_hashdist);
8251#endif
8252
8253void __init page_alloc_init(void)
8254{
8255 int ret;
8256
8257#ifdef CONFIG_NUMA
8258 if (num_node_state(N_MEMORY) == 1)
8259 hashdist = 0;
8260#endif
8261
8262 ret = cpuhp_setup_state_nocalls(CPUHP_PAGE_ALLOC,
8263 "mm/page_alloc:pcp",
8264 page_alloc_cpu_online,
8265 page_alloc_cpu_dead);
8266 WARN_ON(ret < 0);
8267}
8268
8269
8270
8271
8272
8273static void calculate_totalreserve_pages(void)
8274{
8275 struct pglist_data *pgdat;
8276 unsigned long reserve_pages = 0;
8277 enum zone_type i, j;
8278
8279 for_each_online_pgdat(pgdat) {
8280
8281 pgdat->totalreserve_pages = 0;
8282
8283 for (i = 0; i < MAX_NR_ZONES; i++) {
8284 struct zone *zone = pgdat->node_zones + i;
8285 long max = 0;
8286 unsigned long managed_pages = zone_managed_pages(zone);
8287
8288
8289 for (j = i; j < MAX_NR_ZONES; j++) {
8290 if (zone->lowmem_reserve[j] > max)
8291 max = zone->lowmem_reserve[j];
8292 }
8293
8294
8295 max += high_wmark_pages(zone);
8296
8297 if (max > managed_pages)
8298 max = managed_pages;
8299
8300 pgdat->totalreserve_pages += max;
8301
8302 reserve_pages += max;
8303 }
8304 }
8305 totalreserve_pages = reserve_pages;
8306}
8307
8308
8309
8310
8311
8312
8313
8314static void setup_per_zone_lowmem_reserve(void)
8315{
8316 struct pglist_data *pgdat;
8317 enum zone_type i, j;
8318
8319 for_each_online_pgdat(pgdat) {
8320 for (i = 0; i < MAX_NR_ZONES - 1; i++) {
8321 struct zone *zone = &pgdat->node_zones[i];
8322 int ratio = sysctl_lowmem_reserve_ratio[i];
8323 bool clear = !ratio || !zone_managed_pages(zone);
8324 unsigned long managed_pages = 0;
8325
8326 for (j = i + 1; j < MAX_NR_ZONES; j++) {
8327 struct zone *upper_zone = &pgdat->node_zones[j];
8328
8329 managed_pages += zone_managed_pages(upper_zone);
8330
8331 if (clear)
8332 zone->lowmem_reserve[j] = 0;
8333 else
8334 zone->lowmem_reserve[j] = managed_pages / ratio;
8335 }
8336 }
8337 }
8338
8339
8340 calculate_totalreserve_pages();
8341}
8342
8343static void __setup_per_zone_wmarks(void)
8344{
8345 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
8346 unsigned long lowmem_pages = 0;
8347 struct zone *zone;
8348 unsigned long flags;
8349
8350
8351 for_each_zone(zone) {
8352 if (!is_highmem(zone))
8353 lowmem_pages += zone_managed_pages(zone);
8354 }
8355
8356 for_each_zone(zone) {
8357 u64 tmp;
8358
8359 spin_lock_irqsave(&zone->lock, flags);
8360 tmp = (u64)pages_min * zone_managed_pages(zone);
8361 do_div(tmp, lowmem_pages);
8362 if (is_highmem(zone)) {
8363
8364
8365
8366
8367
8368
8369
8370
8371
8372 unsigned long min_pages;
8373
8374 min_pages = zone_managed_pages(zone) / 1024;
8375 min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL);
8376 zone->_watermark[WMARK_MIN] = min_pages;
8377 } else {
8378
8379
8380
8381
8382 zone->_watermark[WMARK_MIN] = tmp;
8383 }
8384
8385
8386
8387
8388
8389
8390 tmp = max_t(u64, tmp >> 2,
8391 mult_frac(zone_managed_pages(zone),
8392 watermark_scale_factor, 10000));
8393
8394 zone->watermark_boost = 0;
8395 zone->_watermark[WMARK_LOW] = min_wmark_pages(zone) + tmp;
8396 zone->_watermark[WMARK_HIGH] = min_wmark_pages(zone) + tmp * 2;
8397
8398 spin_unlock_irqrestore(&zone->lock, flags);
8399 }
8400
8401
8402 calculate_totalreserve_pages();
8403}
8404
8405
8406
8407
8408
8409
8410
8411
8412void setup_per_zone_wmarks(void)
8413{
8414 struct zone *zone;
8415 static DEFINE_SPINLOCK(lock);
8416
8417 spin_lock(&lock);
8418 __setup_per_zone_wmarks();
8419 spin_unlock(&lock);
8420
8421
8422
8423
8424
8425 for_each_zone(zone)
8426 zone_pcp_update(zone, 0);
8427}
8428
8429
8430
8431
8432
8433
8434
8435
8436
8437
8438
8439
8440
8441
8442
8443
8444
8445
8446
8447
8448
8449
8450
8451
8452
8453int __meminit init_per_zone_wmark_min(void)
8454{
8455 unsigned long lowmem_kbytes;
8456 int new_min_free_kbytes;
8457
8458 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
8459 new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
8460
8461 if (new_min_free_kbytes > user_min_free_kbytes) {
8462 min_free_kbytes = new_min_free_kbytes;
8463 if (min_free_kbytes < 128)
8464 min_free_kbytes = 128;
8465 if (min_free_kbytes > 262144)
8466 min_free_kbytes = 262144;
8467 } else {
8468 pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n",
8469 new_min_free_kbytes, user_min_free_kbytes);
8470 }
8471 setup_per_zone_wmarks();
8472 refresh_zone_stat_thresholds();
8473 setup_per_zone_lowmem_reserve();
8474
8475#ifdef CONFIG_NUMA
8476 setup_min_unmapped_ratio();
8477 setup_min_slab_ratio();
8478#endif
8479
8480 khugepaged_min_free_kbytes_update();
8481
8482 return 0;
8483}
8484postcore_initcall(init_per_zone_wmark_min)
8485
8486
8487
8488
8489
8490
8491int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write,
8492 void *buffer, size_t *length, loff_t *ppos)
8493{
8494 int rc;
8495
8496 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
8497 if (rc)
8498 return rc;
8499
8500 if (write) {
8501 user_min_free_kbytes = min_free_kbytes;
8502 setup_per_zone_wmarks();
8503 }
8504 return 0;
8505}
8506
8507int watermark_scale_factor_sysctl_handler(struct ctl_table *table, int write,
8508 void *buffer, size_t *length, loff_t *ppos)
8509{
8510 int rc;
8511
8512 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
8513 if (rc)
8514 return rc;
8515
8516 if (write)
8517 setup_per_zone_wmarks();
8518
8519 return 0;
8520}
8521
8522#ifdef CONFIG_NUMA
8523static void setup_min_unmapped_ratio(void)
8524{
8525 pg_data_t *pgdat;
8526 struct zone *zone;
8527
8528 for_each_online_pgdat(pgdat)
8529 pgdat->min_unmapped_pages = 0;
8530
8531 for_each_zone(zone)
8532 zone->zone_pgdat->min_unmapped_pages += (zone_managed_pages(zone) *
8533 sysctl_min_unmapped_ratio) / 100;
8534}
8535
8536
8537int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write,
8538 void *buffer, size_t *length, loff_t *ppos)
8539{
8540 int rc;
8541
8542 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
8543 if (rc)
8544 return rc;
8545
8546 setup_min_unmapped_ratio();
8547
8548 return 0;
8549}
8550
8551static void setup_min_slab_ratio(void)
8552{
8553 pg_data_t *pgdat;
8554 struct zone *zone;
8555
8556 for_each_online_pgdat(pgdat)
8557 pgdat->min_slab_pages = 0;
8558
8559 for_each_zone(zone)
8560 zone->zone_pgdat->min_slab_pages += (zone_managed_pages(zone) *
8561 sysctl_min_slab_ratio) / 100;
8562}
8563
8564int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write,
8565 void *buffer, size_t *length, loff_t *ppos)
8566{
8567 int rc;
8568
8569 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
8570 if (rc)
8571 return rc;
8572
8573 setup_min_slab_ratio();
8574
8575 return 0;
8576}
8577#endif
8578
8579
8580
8581
8582
8583
8584
8585
8586
8587
8588int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write,
8589 void *buffer, size_t *length, loff_t *ppos)
8590{
8591 int i;
8592
8593 proc_dointvec_minmax(table, write, buffer, length, ppos);
8594
8595 for (i = 0; i < MAX_NR_ZONES; i++) {
8596 if (sysctl_lowmem_reserve_ratio[i] < 1)
8597 sysctl_lowmem_reserve_ratio[i] = 0;
8598 }
8599
8600 setup_per_zone_lowmem_reserve();
8601 return 0;
8602}
8603
8604
8605
8606
8607
8608
8609int percpu_pagelist_high_fraction_sysctl_handler(struct ctl_table *table,
8610 int write, void *buffer, size_t *length, loff_t *ppos)
8611{
8612 struct zone *zone;
8613 int old_percpu_pagelist_high_fraction;
8614 int ret;
8615
8616 mutex_lock(&pcp_batch_high_lock);
8617 old_percpu_pagelist_high_fraction = percpu_pagelist_high_fraction;
8618
8619 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
8620 if (!write || ret < 0)
8621 goto out;
8622
8623
8624 if (percpu_pagelist_high_fraction &&
8625 percpu_pagelist_high_fraction < MIN_PERCPU_PAGELIST_HIGH_FRACTION) {
8626 percpu_pagelist_high_fraction = old_percpu_pagelist_high_fraction;
8627 ret = -EINVAL;
8628 goto out;
8629 }
8630
8631
8632 if (percpu_pagelist_high_fraction == old_percpu_pagelist_high_fraction)
8633 goto out;
8634
8635 for_each_populated_zone(zone)
8636 zone_set_pageset_high_and_batch(zone, 0);
8637out:
8638 mutex_unlock(&pcp_batch_high_lock);
8639 return ret;
8640}
8641
8642#ifndef __HAVE_ARCH_RESERVED_KERNEL_PAGES
8643
8644
8645
8646
8647static unsigned long __init arch_reserved_kernel_pages(void)
8648{
8649 return 0;
8650}
8651#endif
8652
8653
8654
8655
8656
8657
8658
8659
8660
8661
8662#if __BITS_PER_LONG > 32
8663#define ADAPT_SCALE_BASE (64ul << 30)
8664#define ADAPT_SCALE_SHIFT 2
8665#define ADAPT_SCALE_NPAGES (ADAPT_SCALE_BASE >> PAGE_SHIFT)
8666#endif
8667
8668
8669
8670
8671
8672
8673
8674void *__init alloc_large_system_hash(const char *tablename,
8675 unsigned long bucketsize,
8676 unsigned long numentries,
8677 int scale,
8678 int flags,
8679 unsigned int *_hash_shift,
8680 unsigned int *_hash_mask,
8681 unsigned long low_limit,
8682 unsigned long high_limit)
8683{
8684 unsigned long long max = high_limit;
8685 unsigned long log2qty, size;
8686 void *table = NULL;
8687 gfp_t gfp_flags;
8688 bool virt;
8689 bool huge;
8690
8691
8692 if (!numentries) {
8693
8694 numentries = nr_kernel_pages;
8695 numentries -= arch_reserved_kernel_pages();
8696
8697
8698 if (PAGE_SHIFT < 20)
8699 numentries = round_up(numentries, (1<<20)/PAGE_SIZE);
8700
8701#if __BITS_PER_LONG > 32
8702 if (!high_limit) {
8703 unsigned long adapt;
8704
8705 for (adapt = ADAPT_SCALE_NPAGES; adapt < numentries;
8706 adapt <<= ADAPT_SCALE_SHIFT)
8707 scale++;
8708 }
8709#endif
8710
8711
8712 if (scale > PAGE_SHIFT)
8713 numentries >>= (scale - PAGE_SHIFT);
8714 else
8715 numentries <<= (PAGE_SHIFT - scale);
8716
8717
8718 if (unlikely(flags & HASH_SMALL)) {
8719
8720 WARN_ON(!(flags & HASH_EARLY));
8721 if (!(numentries >> *_hash_shift)) {
8722 numentries = 1UL << *_hash_shift;
8723 BUG_ON(!numentries);
8724 }
8725 } else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
8726 numentries = PAGE_SIZE / bucketsize;
8727 }
8728 numentries = roundup_pow_of_two(numentries);
8729
8730
8731 if (max == 0) {
8732 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
8733 do_div(max, bucketsize);
8734 }
8735 max = min(max, 0x80000000ULL);
8736
8737 if (numentries < low_limit)
8738 numentries = low_limit;
8739 if (numentries > max)
8740 numentries = max;
8741
8742 log2qty = ilog2(numentries);
8743
8744 gfp_flags = (flags & HASH_ZERO) ? GFP_ATOMIC | __GFP_ZERO : GFP_ATOMIC;
8745 do {
8746 virt = false;
8747 size = bucketsize << log2qty;
8748 if (flags & HASH_EARLY) {
8749 if (flags & HASH_ZERO)
8750 table = memblock_alloc(size, SMP_CACHE_BYTES);
8751 else
8752 table = memblock_alloc_raw(size,
8753 SMP_CACHE_BYTES);
8754 } else if (get_order(size) >= MAX_ORDER || hashdist) {
8755 table = __vmalloc(size, gfp_flags);
8756 virt = true;
8757 huge = is_vm_area_hugepages(table);
8758 } else {
8759
8760
8761
8762
8763
8764 table = alloc_pages_exact(size, gfp_flags);
8765 kmemleak_alloc(table, size, 1, gfp_flags);
8766 }
8767 } while (!table && size > PAGE_SIZE && --log2qty);
8768
8769 if (!table)
8770 panic("Failed to allocate %s hash table\n", tablename);
8771
8772 pr_info("%s hash table entries: %ld (order: %d, %lu bytes, %s)\n",
8773 tablename, 1UL << log2qty, ilog2(size) - PAGE_SHIFT, size,
8774 virt ? (huge ? "vmalloc hugepage" : "vmalloc") : "linear");
8775
8776 if (_hash_shift)
8777 *_hash_shift = log2qty;
8778 if (_hash_mask)
8779 *_hash_mask = (1 << log2qty) - 1;
8780
8781 return table;
8782}
8783
8784
8785
8786
8787
8788
8789
8790
8791
8792
8793
8794
8795
8796
8797struct page *has_unmovable_pages(struct zone *zone, struct page *page,
8798 int migratetype, int flags)
8799{
8800 unsigned long iter = 0;
8801 unsigned long pfn = page_to_pfn(page);
8802 unsigned long offset = pfn % pageblock_nr_pages;
8803
8804 if (is_migrate_cma_page(page)) {
8805
8806
8807
8808
8809
8810 if (is_migrate_cma(migratetype))
8811 return NULL;
8812
8813 return page;
8814 }
8815
8816 for (; iter < pageblock_nr_pages - offset; iter++) {
8817 if (!pfn_valid_within(pfn + iter))
8818 continue;
8819
8820 page = pfn_to_page(pfn + iter);
8821
8822
8823
8824
8825
8826
8827
8828 if (PageReserved(page))
8829 return page;
8830
8831
8832
8833
8834
8835
8836 if (zone_idx(zone) == ZONE_MOVABLE)
8837 continue;
8838
8839
8840
8841
8842
8843
8844
8845 if (PageHuge(page) || PageTransCompound(page)) {
8846 struct page *head = compound_head(page);
8847 unsigned int skip_pages;
8848
8849 if (PageHuge(page)) {
8850 if (!hugepage_migration_supported(page_hstate(head)))
8851 return page;
8852 } else if (!PageLRU(head) && !__PageMovable(head)) {
8853 return page;
8854 }
8855
8856 skip_pages = compound_nr(head) - (page - head);
8857 iter += skip_pages - 1;
8858 continue;
8859 }
8860
8861
8862
8863
8864
8865
8866
8867 if (!page_ref_count(page)) {
8868 if (PageBuddy(page))
8869 iter += (1 << buddy_order(page)) - 1;
8870 continue;
8871 }
8872
8873
8874
8875
8876
8877 if ((flags & MEMORY_OFFLINE) && PageHWPoison(page))
8878 continue;
8879
8880
8881
8882
8883
8884
8885
8886
8887
8888
8889
8890 if ((flags & MEMORY_OFFLINE) && PageOffline(page))
8891 continue;
8892
8893 if (__PageMovable(page) || PageLRU(page))
8894 continue;
8895
8896
8897
8898
8899
8900
8901 return page;
8902 }
8903 return NULL;
8904}
8905
8906#ifdef CONFIG_CONTIG_ALLOC
8907static unsigned long pfn_max_align_down(unsigned long pfn)
8908{
8909 return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES,
8910 pageblock_nr_pages) - 1);
8911}
8912
8913static unsigned long pfn_max_align_up(unsigned long pfn)
8914{
8915 return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES,
8916 pageblock_nr_pages));
8917}
8918
8919#if defined(CONFIG_DYNAMIC_DEBUG) || \
8920 (defined(CONFIG_DYNAMIC_DEBUG_CORE) && defined(DYNAMIC_DEBUG_MODULE))
8921
8922static void alloc_contig_dump_pages(struct list_head *page_list)
8923{
8924 DEFINE_DYNAMIC_DEBUG_METADATA(descriptor, "migrate failure");
8925
8926 if (DYNAMIC_DEBUG_BRANCH(descriptor)) {
8927 struct page *page;
8928
8929 dump_stack();
8930 list_for_each_entry(page, page_list, lru)
8931 dump_page(page, "migration failure");
8932 }
8933}
8934#else
8935static inline void alloc_contig_dump_pages(struct list_head *page_list)
8936{
8937}
8938#endif
8939
8940
8941static int __alloc_contig_migrate_range(struct compact_control *cc,
8942 unsigned long start, unsigned long end)
8943{
8944
8945 unsigned int nr_reclaimed;
8946 unsigned long pfn = start;
8947 unsigned int tries = 0;
8948 int ret = 0;
8949 struct migration_target_control mtc = {
8950 .nid = zone_to_nid(cc->zone),
8951 .gfp_mask = GFP_USER | __GFP_MOVABLE | __GFP_RETRY_MAYFAIL,
8952 };
8953
8954 lru_cache_disable();
8955
8956 while (pfn < end || !list_empty(&cc->migratepages)) {
8957 if (fatal_signal_pending(current)) {
8958 ret = -EINTR;
8959 break;
8960 }
8961
8962 if (list_empty(&cc->migratepages)) {
8963 cc->nr_migratepages = 0;
8964 ret = isolate_migratepages_range(cc, pfn, end);
8965 if (ret && ret != -EAGAIN)
8966 break;
8967 pfn = cc->migrate_pfn;
8968 tries = 0;
8969 } else if (++tries == 5) {
8970 ret = -EBUSY;
8971 break;
8972 }
8973
8974 nr_reclaimed = reclaim_clean_pages_from_list(cc->zone,
8975 &cc->migratepages);
8976 cc->nr_migratepages -= nr_reclaimed;
8977
8978 ret = migrate_pages(&cc->migratepages, alloc_migration_target,
8979 NULL, (unsigned long)&mtc, cc->mode, MR_CONTIG_RANGE);
8980
8981
8982
8983
8984
8985 if (ret == -ENOMEM)
8986 break;
8987 }
8988
8989 lru_cache_enable();
8990 if (ret < 0) {
8991 if (ret == -EBUSY)
8992 alloc_contig_dump_pages(&cc->migratepages);
8993 putback_movable_pages(&cc->migratepages);
8994 return ret;
8995 }
8996 return 0;
8997}
8998
8999
9000
9001
9002
9003
9004
9005
9006
9007
9008
9009
9010
9011
9012
9013
9014
9015
9016
9017
9018
9019
9020int alloc_contig_range(unsigned long start, unsigned long end,
9021 unsigned migratetype, gfp_t gfp_mask)
9022{
9023 unsigned long outer_start, outer_end;
9024 unsigned int order;
9025 int ret = 0;
9026
9027 struct compact_control cc = {
9028 .nr_migratepages = 0,
9029 .order = -1,
9030 .zone = page_zone(pfn_to_page(start)),
9031 .mode = MIGRATE_SYNC,
9032 .ignore_skip_hint = true,
9033 .no_set_skip_hint = true,
9034 .gfp_mask = current_gfp_context(gfp_mask),
9035 .alloc_contig = true,
9036 };
9037 INIT_LIST_HEAD(&cc.migratepages);
9038
9039
9040
9041
9042
9043
9044
9045
9046
9047
9048
9049
9050
9051
9052
9053
9054
9055
9056
9057
9058
9059
9060
9061
9062
9063 ret = start_isolate_page_range(pfn_max_align_down(start),
9064 pfn_max_align_up(end), migratetype, 0);
9065 if (ret)
9066 return ret;
9067
9068 drain_all_pages(cc.zone);
9069
9070
9071
9072
9073
9074
9075
9076
9077
9078
9079
9080 ret = __alloc_contig_migrate_range(&cc, start, end);
9081 if (ret && ret != -EBUSY)
9082 goto done;
9083 ret = 0;
9084
9085
9086
9087
9088
9089
9090
9091
9092
9093
9094
9095
9096
9097
9098
9099
9100
9101
9102 order = 0;
9103 outer_start = start;
9104 while (!PageBuddy(pfn_to_page(outer_start))) {
9105 if (++order >= MAX_ORDER) {
9106 outer_start = start;
9107 break;
9108 }
9109 outer_start &= ~0UL << order;
9110 }
9111
9112 if (outer_start != start) {
9113 order = buddy_order(pfn_to_page(outer_start));
9114
9115
9116
9117
9118
9119
9120
9121 if (outer_start + (1UL << order) <= start)
9122 outer_start = start;
9123 }
9124
9125
9126 if (test_pages_isolated(outer_start, end, 0)) {
9127 ret = -EBUSY;
9128 goto done;
9129 }
9130
9131
9132 outer_end = isolate_freepages_range(&cc, outer_start, end);
9133 if (!outer_end) {
9134 ret = -EBUSY;
9135 goto done;
9136 }
9137
9138
9139 if (start != outer_start)
9140 free_contig_range(outer_start, start - outer_start);
9141 if (end != outer_end)
9142 free_contig_range(end, outer_end - end);
9143
9144done:
9145 undo_isolate_page_range(pfn_max_align_down(start),
9146 pfn_max_align_up(end), migratetype);
9147 return ret;
9148}
9149EXPORT_SYMBOL(alloc_contig_range);
9150
9151static int __alloc_contig_pages(unsigned long start_pfn,
9152 unsigned long nr_pages, gfp_t gfp_mask)
9153{
9154 unsigned long end_pfn = start_pfn + nr_pages;
9155
9156 return alloc_contig_range(start_pfn, end_pfn, MIGRATE_MOVABLE,
9157 gfp_mask);
9158}
9159
9160static bool pfn_range_valid_contig(struct zone *z, unsigned long start_pfn,
9161 unsigned long nr_pages)
9162{
9163 unsigned long i, end_pfn = start_pfn + nr_pages;
9164 struct page *page;
9165
9166 for (i = start_pfn; i < end_pfn; i++) {
9167 page = pfn_to_online_page(i);
9168 if (!page)
9169 return false;
9170
9171 if (page_zone(page) != z)
9172 return false;
9173
9174 if (PageReserved(page))
9175 return false;
9176 }
9177 return true;
9178}
9179
9180static bool zone_spans_last_pfn(const struct zone *zone,
9181 unsigned long start_pfn, unsigned long nr_pages)
9182{
9183 unsigned long last_pfn = start_pfn + nr_pages - 1;
9184
9185 return zone_spans_pfn(zone, last_pfn);
9186}
9187
9188
9189
9190
9191
9192
9193
9194
9195
9196
9197
9198
9199
9200
9201
9202
9203
9204
9205
9206
9207
9208
9209struct page *alloc_contig_pages(unsigned long nr_pages, gfp_t gfp_mask,
9210 int nid, nodemask_t *nodemask)
9211{
9212 unsigned long ret, pfn, flags;
9213 struct zonelist *zonelist;
9214 struct zone *zone;
9215 struct zoneref *z;
9216
9217 zonelist = node_zonelist(nid, gfp_mask);
9218 for_each_zone_zonelist_nodemask(zone, z, zonelist,
9219 gfp_zone(gfp_mask), nodemask) {
9220 spin_lock_irqsave(&zone->lock, flags);
9221
9222 pfn = ALIGN(zone->zone_start_pfn, nr_pages);
9223 while (zone_spans_last_pfn(zone, pfn, nr_pages)) {
9224 if (pfn_range_valid_contig(zone, pfn, nr_pages)) {
9225
9226
9227
9228
9229
9230
9231
9232 spin_unlock_irqrestore(&zone->lock, flags);
9233 ret = __alloc_contig_pages(pfn, nr_pages,
9234 gfp_mask);
9235 if (!ret)
9236 return pfn_to_page(pfn);
9237 spin_lock_irqsave(&zone->lock, flags);
9238 }
9239 pfn += nr_pages;
9240 }
9241 spin_unlock_irqrestore(&zone->lock, flags);
9242 }
9243 return NULL;
9244}
9245#endif
9246
9247void free_contig_range(unsigned long pfn, unsigned long nr_pages)
9248{
9249 unsigned long count = 0;
9250
9251 for (; nr_pages--; pfn++) {
9252 struct page *page = pfn_to_page(pfn);
9253
9254 count += page_count(page) != 1;
9255 __free_page(page);
9256 }
9257 WARN(count != 0, "%lu pages are still in use!\n", count);
9258}
9259EXPORT_SYMBOL(free_contig_range);
9260
9261
9262
9263
9264
9265void zone_pcp_update(struct zone *zone, int cpu_online)
9266{
9267 mutex_lock(&pcp_batch_high_lock);
9268 zone_set_pageset_high_and_batch(zone, cpu_online);
9269 mutex_unlock(&pcp_batch_high_lock);
9270}
9271
9272
9273
9274
9275
9276
9277
9278
9279
9280void zone_pcp_disable(struct zone *zone)
9281{
9282 mutex_lock(&pcp_batch_high_lock);
9283 __zone_set_pageset_high_and_batch(zone, 0, 1);
9284 __drain_all_pages(zone, true);
9285}
9286
9287void zone_pcp_enable(struct zone *zone)
9288{
9289 __zone_set_pageset_high_and_batch(zone, zone->pageset_high, zone->pageset_batch);
9290 mutex_unlock(&pcp_batch_high_lock);
9291}
9292
9293void zone_pcp_reset(struct zone *zone)
9294{
9295 int cpu;
9296 struct per_cpu_zonestat *pzstats;
9297
9298 if (zone->per_cpu_pageset != &boot_pageset) {
9299 for_each_online_cpu(cpu) {
9300 pzstats = per_cpu_ptr(zone->per_cpu_zonestats, cpu);
9301 drain_zonestat(zone, pzstats);
9302 }
9303 free_percpu(zone->per_cpu_pageset);
9304 free_percpu(zone->per_cpu_zonestats);
9305 zone->per_cpu_pageset = &boot_pageset;
9306 zone->per_cpu_zonestats = &boot_zonestats;
9307 }
9308}
9309
9310#ifdef CONFIG_MEMORY_HOTREMOVE
9311
9312
9313
9314
9315void __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
9316{
9317 unsigned long pfn = start_pfn;
9318 struct page *page;
9319 struct zone *zone;
9320 unsigned int order;
9321 unsigned long flags;
9322
9323 offline_mem_sections(pfn, end_pfn);
9324 zone = page_zone(pfn_to_page(pfn));
9325 spin_lock_irqsave(&zone->lock, flags);
9326 while (pfn < end_pfn) {
9327 page = pfn_to_page(pfn);
9328
9329
9330
9331
9332 if (unlikely(!PageBuddy(page) && PageHWPoison(page))) {
9333 pfn++;
9334 continue;
9335 }
9336
9337
9338
9339
9340 if (PageOffline(page)) {
9341 BUG_ON(page_count(page));
9342 BUG_ON(PageBuddy(page));
9343 pfn++;
9344 continue;
9345 }
9346
9347 BUG_ON(page_count(page));
9348 BUG_ON(!PageBuddy(page));
9349 order = buddy_order(page);
9350 del_page_from_free_list(page, zone, order);
9351 pfn += (1 << order);
9352 }
9353 spin_unlock_irqrestore(&zone->lock, flags);
9354}
9355#endif
9356
9357bool is_free_buddy_page(struct page *page)
9358{
9359 struct zone *zone = page_zone(page);
9360 unsigned long pfn = page_to_pfn(page);
9361 unsigned long flags;
9362 unsigned int order;
9363
9364 spin_lock_irqsave(&zone->lock, flags);
9365 for (order = 0; order < MAX_ORDER; order++) {
9366 struct page *page_head = page - (pfn & ((1 << order) - 1));
9367
9368 if (PageBuddy(page_head) && buddy_order(page_head) >= order)
9369 break;
9370 }
9371 spin_unlock_irqrestore(&zone->lock, flags);
9372
9373 return order < MAX_ORDER;
9374}
9375
9376#ifdef CONFIG_MEMORY_FAILURE
9377
9378
9379
9380
9381static void break_down_buddy_pages(struct zone *zone, struct page *page,
9382 struct page *target, int low, int high,
9383 int migratetype)
9384{
9385 unsigned long size = 1 << high;
9386 struct page *current_buddy, *next_page;
9387
9388 while (high > low) {
9389 high--;
9390 size >>= 1;
9391
9392 if (target >= &page[size]) {
9393 next_page = page + size;
9394 current_buddy = page;
9395 } else {
9396 next_page = page;
9397 current_buddy = page + size;
9398 }
9399
9400 if (set_page_guard(zone, current_buddy, high, migratetype))
9401 continue;
9402
9403 if (current_buddy != target) {
9404 add_to_free_list(current_buddy, zone, high, migratetype);
9405 set_buddy_order(current_buddy, high);
9406 page = next_page;
9407 }
9408 }
9409}
9410
9411
9412
9413
9414bool take_page_off_buddy(struct page *page)
9415{
9416 struct zone *zone = page_zone(page);
9417 unsigned long pfn = page_to_pfn(page);
9418 unsigned long flags;
9419 unsigned int order;
9420 bool ret = false;
9421
9422 spin_lock_irqsave(&zone->lock, flags);
9423 for (order = 0; order < MAX_ORDER; order++) {
9424 struct page *page_head = page - (pfn & ((1 << order) - 1));
9425 int page_order = buddy_order(page_head);
9426
9427 if (PageBuddy(page_head) && page_order >= order) {
9428 unsigned long pfn_head = page_to_pfn(page_head);
9429 int migratetype = get_pfnblock_migratetype(page_head,
9430 pfn_head);
9431
9432 del_page_from_free_list(page_head, zone, page_order);
9433 break_down_buddy_pages(zone, page_head, page, 0,
9434 page_order, migratetype);
9435 if (!is_migrate_isolate(migratetype))
9436 __mod_zone_freepage_state(zone, -1, migratetype);
9437 ret = true;
9438 break;
9439 }
9440 if (page_count(page_head) > 0)
9441 break;
9442 }
9443 spin_unlock_irqrestore(&zone->lock, flags);
9444 return ret;
9445}
9446#endif
9447