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 (zone != page_zone(page))
598 return 0;
599
600 return 1;
601}
602
603
604
605static int __maybe_unused bad_range(struct zone *zone, struct page *page)
606{
607 if (page_outside_zone_boundaries(zone, page))
608 return 1;
609 if (!page_is_consistent(zone, page))
610 return 1;
611
612 return 0;
613}
614#else
615static inline int __maybe_unused bad_range(struct zone *zone, struct page *page)
616{
617 return 0;
618}
619#endif
620
621static void bad_page(struct page *page, const char *reason)
622{
623 static unsigned long resume;
624 static unsigned long nr_shown;
625 static unsigned long nr_unshown;
626
627
628
629
630
631 if (nr_shown == 60) {
632 if (time_before(jiffies, resume)) {
633 nr_unshown++;
634 goto out;
635 }
636 if (nr_unshown) {
637 pr_alert(
638 "BUG: Bad page state: %lu messages suppressed\n",
639 nr_unshown);
640 nr_unshown = 0;
641 }
642 nr_shown = 0;
643 }
644 if (nr_shown++ == 0)
645 resume = jiffies + 60 * HZ;
646
647 pr_alert("BUG: Bad page state in process %s pfn:%05lx\n",
648 current->comm, page_to_pfn(page));
649 dump_page(page, reason);
650
651 print_modules();
652 dump_stack();
653out:
654
655 page_mapcount_reset(page);
656 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
657}
658
659static inline unsigned int order_to_pindex(int migratetype, int order)
660{
661 int base = order;
662
663#ifdef CONFIG_TRANSPARENT_HUGEPAGE
664 if (order > PAGE_ALLOC_COSTLY_ORDER) {
665 VM_BUG_ON(order != pageblock_order);
666 base = PAGE_ALLOC_COSTLY_ORDER + 1;
667 }
668#else
669 VM_BUG_ON(order > PAGE_ALLOC_COSTLY_ORDER);
670#endif
671
672 return (MIGRATE_PCPTYPES * base) + migratetype;
673}
674
675static inline int pindex_to_order(unsigned int pindex)
676{
677 int order = pindex / MIGRATE_PCPTYPES;
678
679#ifdef CONFIG_TRANSPARENT_HUGEPAGE
680 if (order > PAGE_ALLOC_COSTLY_ORDER) {
681 order = pageblock_order;
682 VM_BUG_ON(order != pageblock_order);
683 }
684#else
685 VM_BUG_ON(order > PAGE_ALLOC_COSTLY_ORDER);
686#endif
687
688 return order;
689}
690
691static inline bool pcp_allowed_order(unsigned int order)
692{
693 if (order <= PAGE_ALLOC_COSTLY_ORDER)
694 return true;
695#ifdef CONFIG_TRANSPARENT_HUGEPAGE
696 if (order == pageblock_order)
697 return true;
698#endif
699 return false;
700}
701
702static inline void free_the_page(struct page *page, unsigned int order)
703{
704 if (pcp_allowed_order(order))
705 free_unref_page(page, order);
706 else
707 __free_pages_ok(page, order, FPI_NONE);
708}
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725void free_compound_page(struct page *page)
726{
727 mem_cgroup_uncharge(page);
728 free_the_page(page, compound_order(page));
729}
730
731void prep_compound_page(struct page *page, unsigned int order)
732{
733 int i;
734 int nr_pages = 1 << order;
735
736 __SetPageHead(page);
737 for (i = 1; i < nr_pages; i++) {
738 struct page *p = page + i;
739 p->mapping = TAIL_MAPPING;
740 set_compound_head(p, page);
741 }
742
743 set_compound_page_dtor(page, COMPOUND_PAGE_DTOR);
744 set_compound_order(page, order);
745 atomic_set(compound_mapcount_ptr(page), -1);
746 if (hpage_pincount_available(page))
747 atomic_set(compound_pincount_ptr(page), 0);
748}
749
750#ifdef CONFIG_DEBUG_PAGEALLOC
751unsigned int _debug_guardpage_minorder;
752
753bool _debug_pagealloc_enabled_early __read_mostly
754 = IS_ENABLED(CONFIG_DEBUG_PAGEALLOC_ENABLE_DEFAULT);
755EXPORT_SYMBOL(_debug_pagealloc_enabled_early);
756DEFINE_STATIC_KEY_FALSE(_debug_pagealloc_enabled);
757EXPORT_SYMBOL(_debug_pagealloc_enabled);
758
759DEFINE_STATIC_KEY_FALSE(_debug_guardpage_enabled);
760
761static int __init early_debug_pagealloc(char *buf)
762{
763 return kstrtobool(buf, &_debug_pagealloc_enabled_early);
764}
765early_param("debug_pagealloc", early_debug_pagealloc);
766
767static int __init debug_guardpage_minorder_setup(char *buf)
768{
769 unsigned long res;
770
771 if (kstrtoul(buf, 10, &res) < 0 || res > MAX_ORDER / 2) {
772 pr_err("Bad debug_guardpage_minorder value\n");
773 return 0;
774 }
775 _debug_guardpage_minorder = res;
776 pr_info("Setting debug_guardpage_minorder to %lu\n", res);
777 return 0;
778}
779early_param("debug_guardpage_minorder", debug_guardpage_minorder_setup);
780
781static inline bool set_page_guard(struct zone *zone, struct page *page,
782 unsigned int order, int migratetype)
783{
784 if (!debug_guardpage_enabled())
785 return false;
786
787 if (order >= debug_guardpage_minorder())
788 return false;
789
790 __SetPageGuard(page);
791 INIT_LIST_HEAD(&page->lru);
792 set_page_private(page, order);
793
794 __mod_zone_freepage_state(zone, -(1 << order), migratetype);
795
796 return true;
797}
798
799static inline void clear_page_guard(struct zone *zone, struct page *page,
800 unsigned int order, int migratetype)
801{
802 if (!debug_guardpage_enabled())
803 return;
804
805 __ClearPageGuard(page);
806
807 set_page_private(page, 0);
808 if (!is_migrate_isolate(migratetype))
809 __mod_zone_freepage_state(zone, (1 << order), migratetype);
810}
811#else
812static inline bool set_page_guard(struct zone *zone, struct page *page,
813 unsigned int order, int migratetype) { return false; }
814static inline void clear_page_guard(struct zone *zone, struct page *page,
815 unsigned int order, int migratetype) {}
816#endif
817
818
819
820
821
822
823
824void init_mem_debugging_and_hardening(void)
825{
826 bool page_poisoning_requested = false;
827
828#ifdef CONFIG_PAGE_POISONING
829
830
831
832
833 if (page_poisoning_enabled() ||
834 (!IS_ENABLED(CONFIG_ARCH_SUPPORTS_DEBUG_PAGEALLOC) &&
835 debug_pagealloc_enabled())) {
836 static_branch_enable(&_page_poisoning_enabled);
837 page_poisoning_requested = true;
838 }
839#endif
840
841 if ((_init_on_alloc_enabled_early || _init_on_free_enabled_early) &&
842 page_poisoning_requested) {
843 pr_info("mem auto-init: CONFIG_PAGE_POISONING is on, "
844 "will take precedence over init_on_alloc and init_on_free\n");
845 _init_on_alloc_enabled_early = false;
846 _init_on_free_enabled_early = false;
847 }
848
849 if (_init_on_alloc_enabled_early)
850 static_branch_enable(&init_on_alloc);
851 else
852 static_branch_disable(&init_on_alloc);
853
854 if (_init_on_free_enabled_early)
855 static_branch_enable(&init_on_free);
856 else
857 static_branch_disable(&init_on_free);
858
859#ifdef CONFIG_DEBUG_PAGEALLOC
860 if (!debug_pagealloc_enabled())
861 return;
862
863 static_branch_enable(&_debug_pagealloc_enabled);
864
865 if (!debug_guardpage_minorder())
866 return;
867
868 static_branch_enable(&_debug_guardpage_enabled);
869#endif
870}
871
872static inline void set_buddy_order(struct page *page, unsigned int order)
873{
874 set_page_private(page, order);
875 __SetPageBuddy(page);
876}
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891static inline bool page_is_buddy(struct page *page, struct page *buddy,
892 unsigned int order)
893{
894 if (!page_is_guard(buddy) && !PageBuddy(buddy))
895 return false;
896
897 if (buddy_order(buddy) != order)
898 return false;
899
900
901
902
903
904 if (page_zone_id(page) != page_zone_id(buddy))
905 return false;
906
907 VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
908
909 return true;
910}
911
912#ifdef CONFIG_COMPACTION
913static inline struct capture_control *task_capc(struct zone *zone)
914{
915 struct capture_control *capc = current->capture_control;
916
917 return unlikely(capc) &&
918 !(current->flags & PF_KTHREAD) &&
919 !capc->page &&
920 capc->cc->zone == zone ? capc : NULL;
921}
922
923static inline bool
924compaction_capture(struct capture_control *capc, struct page *page,
925 int order, int migratetype)
926{
927 if (!capc || order != capc->cc->order)
928 return false;
929
930
931 if (is_migrate_cma(migratetype) ||
932 is_migrate_isolate(migratetype))
933 return false;
934
935
936
937
938
939
940
941 if (order < pageblock_order && migratetype == MIGRATE_MOVABLE)
942 return false;
943
944 capc->page = page;
945 return true;
946}
947
948#else
949static inline struct capture_control *task_capc(struct zone *zone)
950{
951 return NULL;
952}
953
954static inline bool
955compaction_capture(struct capture_control *capc, struct page *page,
956 int order, int migratetype)
957{
958 return false;
959}
960#endif
961
962
963static inline void add_to_free_list(struct page *page, struct zone *zone,
964 unsigned int order, int migratetype)
965{
966 struct free_area *area = &zone->free_area[order];
967
968 list_add(&page->lru, &area->free_list[migratetype]);
969 area->nr_free++;
970}
971
972
973static inline void add_to_free_list_tail(struct page *page, struct zone *zone,
974 unsigned int order, int migratetype)
975{
976 struct free_area *area = &zone->free_area[order];
977
978 list_add_tail(&page->lru, &area->free_list[migratetype]);
979 area->nr_free++;
980}
981
982
983
984
985
986
987static inline void move_to_free_list(struct page *page, struct zone *zone,
988 unsigned int order, int migratetype)
989{
990 struct free_area *area = &zone->free_area[order];
991
992 list_move_tail(&page->lru, &area->free_list[migratetype]);
993}
994
995static inline void del_page_from_free_list(struct page *page, struct zone *zone,
996 unsigned int order)
997{
998
999 if (page_reported(page))
1000 __ClearPageReported(page);
1001
1002 list_del(&page->lru);
1003 __ClearPageBuddy(page);
1004 set_page_private(page, 0);
1005 zone->free_area[order].nr_free--;
1006}
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016static inline bool
1017buddy_merge_likely(unsigned long pfn, unsigned long buddy_pfn,
1018 struct page *page, unsigned int order)
1019{
1020 struct page *higher_page, *higher_buddy;
1021 unsigned long combined_pfn;
1022
1023 if (order >= MAX_ORDER - 2)
1024 return false;
1025
1026 combined_pfn = buddy_pfn & pfn;
1027 higher_page = page + (combined_pfn - pfn);
1028 buddy_pfn = __find_buddy_pfn(combined_pfn, order + 1);
1029 higher_buddy = higher_page + (buddy_pfn - combined_pfn);
1030
1031 return page_is_buddy(higher_page, higher_buddy, order + 1);
1032}
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058static inline void __free_one_page(struct page *page,
1059 unsigned long pfn,
1060 struct zone *zone, unsigned int order,
1061 int migratetype, fpi_t fpi_flags)
1062{
1063 struct capture_control *capc = task_capc(zone);
1064 unsigned long buddy_pfn;
1065 unsigned long combined_pfn;
1066 unsigned int max_order;
1067 struct page *buddy;
1068 bool to_tail;
1069
1070 max_order = min_t(unsigned int, MAX_ORDER - 1, pageblock_order);
1071
1072 VM_BUG_ON(!zone_is_initialized(zone));
1073 VM_BUG_ON_PAGE(page->flags & PAGE_FLAGS_CHECK_AT_PREP, page);
1074
1075 VM_BUG_ON(migratetype == -1);
1076 if (likely(!is_migrate_isolate(migratetype)))
1077 __mod_zone_freepage_state(zone, 1 << order, migratetype);
1078
1079 VM_BUG_ON_PAGE(pfn & ((1 << order) - 1), page);
1080 VM_BUG_ON_PAGE(bad_range(zone, page), page);
1081
1082continue_merging:
1083 while (order < max_order) {
1084 if (compaction_capture(capc, page, order, migratetype)) {
1085 __mod_zone_freepage_state(zone, -(1 << order),
1086 migratetype);
1087 return;
1088 }
1089 buddy_pfn = __find_buddy_pfn(pfn, order);
1090 buddy = page + (buddy_pfn - pfn);
1091
1092 if (!page_is_buddy(page, buddy, order))
1093 goto done_merging;
1094
1095
1096
1097
1098 if (page_is_guard(buddy))
1099 clear_page_guard(zone, buddy, order, migratetype);
1100 else
1101 del_page_from_free_list(buddy, zone, order);
1102 combined_pfn = buddy_pfn & pfn;
1103 page = page + (combined_pfn - pfn);
1104 pfn = combined_pfn;
1105 order++;
1106 }
1107 if (order < MAX_ORDER - 1) {
1108
1109
1110
1111
1112
1113
1114
1115
1116 if (unlikely(has_isolate_pageblock(zone))) {
1117 int buddy_mt;
1118
1119 buddy_pfn = __find_buddy_pfn(pfn, order);
1120 buddy = page + (buddy_pfn - pfn);
1121 buddy_mt = get_pageblock_migratetype(buddy);
1122
1123 if (migratetype != buddy_mt
1124 && (is_migrate_isolate(migratetype) ||
1125 is_migrate_isolate(buddy_mt)))
1126 goto done_merging;
1127 }
1128 max_order = order + 1;
1129 goto continue_merging;
1130 }
1131
1132done_merging:
1133 set_buddy_order(page, order);
1134
1135 if (fpi_flags & FPI_TO_TAIL)
1136 to_tail = true;
1137 else if (is_shuffle_order(order))
1138 to_tail = shuffle_pick_tail();
1139 else
1140 to_tail = buddy_merge_likely(pfn, buddy_pfn, page, order);
1141
1142 if (to_tail)
1143 add_to_free_list_tail(page, zone, order, migratetype);
1144 else
1145 add_to_free_list(page, zone, order, migratetype);
1146
1147
1148 if (!(fpi_flags & FPI_SKIP_REPORT_NOTIFY))
1149 page_reporting_notify_free(order);
1150}
1151
1152
1153
1154
1155
1156
1157static inline bool page_expected_state(struct page *page,
1158 unsigned long check_flags)
1159{
1160 if (unlikely(atomic_read(&page->_mapcount) != -1))
1161 return false;
1162
1163 if (unlikely((unsigned long)page->mapping |
1164 page_ref_count(page) |
1165#ifdef CONFIG_MEMCG
1166 page->memcg_data |
1167#endif
1168 (page->flags & check_flags)))
1169 return false;
1170
1171 return true;
1172}
1173
1174static const char *page_bad_reason(struct page *page, unsigned long flags)
1175{
1176 const char *bad_reason = NULL;
1177
1178 if (unlikely(atomic_read(&page->_mapcount) != -1))
1179 bad_reason = "nonzero mapcount";
1180 if (unlikely(page->mapping != NULL))
1181 bad_reason = "non-NULL mapping";
1182 if (unlikely(page_ref_count(page) != 0))
1183 bad_reason = "nonzero _refcount";
1184 if (unlikely(page->flags & flags)) {
1185 if (flags == PAGE_FLAGS_CHECK_AT_PREP)
1186 bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag(s) set";
1187 else
1188 bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set";
1189 }
1190#ifdef CONFIG_MEMCG
1191 if (unlikely(page->memcg_data))
1192 bad_reason = "page still charged to cgroup";
1193#endif
1194 return bad_reason;
1195}
1196
1197static void check_free_page_bad(struct page *page)
1198{
1199 bad_page(page,
1200 page_bad_reason(page, PAGE_FLAGS_CHECK_AT_FREE));
1201}
1202
1203static inline int check_free_page(struct page *page)
1204{
1205 if (likely(page_expected_state(page, PAGE_FLAGS_CHECK_AT_FREE)))
1206 return 0;
1207
1208
1209 check_free_page_bad(page);
1210 return 1;
1211}
1212
1213static int free_tail_pages_check(struct page *head_page, struct page *page)
1214{
1215 int ret = 1;
1216
1217
1218
1219
1220
1221 BUILD_BUG_ON((unsigned long)LIST_POISON1 & 1);
1222
1223 if (!IS_ENABLED(CONFIG_DEBUG_VM)) {
1224 ret = 0;
1225 goto out;
1226 }
1227 switch (page - head_page) {
1228 case 1:
1229
1230 if (unlikely(compound_mapcount(page))) {
1231 bad_page(page, "nonzero compound_mapcount");
1232 goto out;
1233 }
1234 break;
1235 case 2:
1236
1237
1238
1239
1240 break;
1241 default:
1242 if (page->mapping != TAIL_MAPPING) {
1243 bad_page(page, "corrupted mapping in tail page");
1244 goto out;
1245 }
1246 break;
1247 }
1248 if (unlikely(!PageTail(page))) {
1249 bad_page(page, "PageTail not set");
1250 goto out;
1251 }
1252 if (unlikely(compound_head(page) != head_page)) {
1253 bad_page(page, "compound_head not consistent");
1254 goto out;
1255 }
1256 ret = 0;
1257out:
1258 page->mapping = NULL;
1259 clear_compound_head(page);
1260 return ret;
1261}
1262
1263static void kernel_init_free_pages(struct page *page, int numpages, bool zero_tags)
1264{
1265 int i;
1266
1267 if (zero_tags) {
1268 for (i = 0; i < numpages; i++)
1269 tag_clear_highpage(page + i);
1270 return;
1271 }
1272
1273
1274 kasan_disable_current();
1275 for (i = 0; i < numpages; i++) {
1276 u8 tag = page_kasan_tag(page + i);
1277 page_kasan_tag_reset(page + i);
1278 clear_highpage(page + i);
1279 page_kasan_tag_set(page + i, tag);
1280 }
1281 kasan_enable_current();
1282}
1283
1284static __always_inline bool free_pages_prepare(struct page *page,
1285 unsigned int order, bool check_free, fpi_t fpi_flags)
1286{
1287 int bad = 0;
1288 bool skip_kasan_poison = should_skip_kasan_poison(page, fpi_flags);
1289
1290 VM_BUG_ON_PAGE(PageTail(page), page);
1291
1292 trace_mm_page_free(page, order);
1293
1294 if (unlikely(PageHWPoison(page)) && !order) {
1295
1296
1297
1298
1299 if (memcg_kmem_enabled() && PageMemcgKmem(page))
1300 __memcg_kmem_uncharge_page(page, order);
1301 reset_page_owner(page, order);
1302 return false;
1303 }
1304
1305
1306
1307
1308
1309 if (unlikely(order)) {
1310 bool compound = PageCompound(page);
1311 int i;
1312
1313 VM_BUG_ON_PAGE(compound && compound_order(page) != order, page);
1314
1315 if (compound) {
1316 ClearPageDoubleMap(page);
1317 ClearPageHasHWPoisoned(page);
1318 }
1319 for (i = 1; i < (1 << order); i++) {
1320 if (compound)
1321 bad += free_tail_pages_check(page, page + i);
1322 if (unlikely(check_free_page(page + i))) {
1323 bad++;
1324 continue;
1325 }
1326 (page + i)->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
1327 }
1328 }
1329 if (PageMappingFlags(page))
1330 page->mapping = NULL;
1331 if (memcg_kmem_enabled() && PageMemcgKmem(page))
1332 __memcg_kmem_uncharge_page(page, order);
1333 if (check_free)
1334 bad += check_free_page(page);
1335 if (bad)
1336 return false;
1337
1338 page_cpupid_reset_last(page);
1339 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
1340 reset_page_owner(page, order);
1341
1342 if (!PageHighMem(page)) {
1343 debug_check_no_locks_freed(page_address(page),
1344 PAGE_SIZE << order);
1345 debug_check_no_obj_freed(page_address(page),
1346 PAGE_SIZE << order);
1347 }
1348
1349 kernel_poison_pages(page, 1 << order);
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359 if (kasan_has_integrated_init()) {
1360 if (!skip_kasan_poison)
1361 kasan_free_pages(page, order);
1362 } else {
1363 bool init = want_init_on_free();
1364
1365 if (init)
1366 kernel_init_free_pages(page, 1 << order, false);
1367 if (!skip_kasan_poison)
1368 kasan_poison_pages(page, order, init);
1369 }
1370
1371
1372
1373
1374
1375
1376 arch_free_page(page, order);
1377
1378 debug_pagealloc_unmap_pages(page, 1 << order);
1379
1380 return true;
1381}
1382
1383#ifdef CONFIG_DEBUG_VM
1384
1385
1386
1387
1388
1389static bool free_pcp_prepare(struct page *page, unsigned int order)
1390{
1391 return free_pages_prepare(page, order, true, FPI_NONE);
1392}
1393
1394static bool bulkfree_pcp_prepare(struct page *page)
1395{
1396 if (debug_pagealloc_enabled_static())
1397 return check_free_page(page);
1398 else
1399 return false;
1400}
1401#else
1402
1403
1404
1405
1406
1407
1408static bool free_pcp_prepare(struct page *page, unsigned int order)
1409{
1410 if (debug_pagealloc_enabled_static())
1411 return free_pages_prepare(page, order, true, FPI_NONE);
1412 else
1413 return free_pages_prepare(page, order, false, FPI_NONE);
1414}
1415
1416static bool bulkfree_pcp_prepare(struct page *page)
1417{
1418 return check_free_page(page);
1419}
1420#endif
1421
1422static inline void prefetch_buddy(struct page *page)
1423{
1424 unsigned long pfn = page_to_pfn(page);
1425 unsigned long buddy_pfn = __find_buddy_pfn(pfn, 0);
1426 struct page *buddy = page + (buddy_pfn - pfn);
1427
1428 prefetch(buddy);
1429}
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442static void free_pcppages_bulk(struct zone *zone, int count,
1443 struct per_cpu_pages *pcp)
1444{
1445 int pindex = 0;
1446 int batch_free = 0;
1447 int nr_freed = 0;
1448 unsigned int order;
1449 int prefetch_nr = READ_ONCE(pcp->batch);
1450 bool isolated_pageblocks;
1451 struct page *page, *tmp;
1452 LIST_HEAD(head);
1453
1454
1455
1456
1457
1458 count = min(pcp->count, count);
1459 while (count > 0) {
1460 struct list_head *list;
1461
1462
1463
1464
1465
1466
1467
1468
1469 do {
1470 batch_free++;
1471 if (++pindex == NR_PCP_LISTS)
1472 pindex = 0;
1473 list = &pcp->lists[pindex];
1474 } while (list_empty(list));
1475
1476
1477 if (batch_free == NR_PCP_LISTS)
1478 batch_free = count;
1479
1480 order = pindex_to_order(pindex);
1481 BUILD_BUG_ON(MAX_ORDER >= (1<<NR_PCP_ORDER_WIDTH));
1482 do {
1483 page = list_last_entry(list, struct page, lru);
1484
1485 list_del(&page->lru);
1486 nr_freed += 1 << order;
1487 count -= 1 << order;
1488
1489 if (bulkfree_pcp_prepare(page))
1490 continue;
1491
1492
1493 page->index <<= NR_PCP_ORDER_WIDTH;
1494 page->index |= order;
1495
1496 list_add_tail(&page->lru, &head);
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507 if (prefetch_nr) {
1508 prefetch_buddy(page);
1509 prefetch_nr--;
1510 }
1511 } while (count > 0 && --batch_free && !list_empty(list));
1512 }
1513 pcp->count -= nr_freed;
1514
1515
1516
1517
1518
1519 spin_lock(&zone->lock);
1520 isolated_pageblocks = has_isolate_pageblock(zone);
1521
1522
1523
1524
1525
1526 list_for_each_entry_safe(page, tmp, &head, lru) {
1527 int mt = get_pcppage_migratetype(page);
1528
1529
1530 order = mt & NR_PCP_ORDER_MASK;
1531 mt >>= NR_PCP_ORDER_WIDTH;
1532
1533
1534 VM_BUG_ON_PAGE(is_migrate_isolate(mt), page);
1535
1536 if (unlikely(isolated_pageblocks))
1537 mt = get_pageblock_migratetype(page);
1538
1539 __free_one_page(page, page_to_pfn(page), zone, order, mt, FPI_NONE);
1540 trace_mm_page_pcpu_drain(page, order, mt);
1541 }
1542 spin_unlock(&zone->lock);
1543}
1544
1545static void free_one_page(struct zone *zone,
1546 struct page *page, unsigned long pfn,
1547 unsigned int order,
1548 int migratetype, fpi_t fpi_flags)
1549{
1550 unsigned long flags;
1551
1552 spin_lock_irqsave(&zone->lock, flags);
1553 if (unlikely(has_isolate_pageblock(zone) ||
1554 is_migrate_isolate(migratetype))) {
1555 migratetype = get_pfnblock_migratetype(page, pfn);
1556 }
1557 __free_one_page(page, pfn, zone, order, migratetype, fpi_flags);
1558 spin_unlock_irqrestore(&zone->lock, flags);
1559}
1560
1561static void __meminit __init_single_page(struct page *page, unsigned long pfn,
1562 unsigned long zone, int nid)
1563{
1564 mm_zero_struct_page(page);
1565 set_page_links(page, zone, nid, pfn);
1566 init_page_count(page);
1567 page_mapcount_reset(page);
1568 page_cpupid_reset_last(page);
1569 page_kasan_tag_reset(page);
1570
1571 INIT_LIST_HEAD(&page->lru);
1572#ifdef WANT_PAGE_VIRTUAL
1573
1574 if (!is_highmem_idx(zone))
1575 set_page_address(page, __va(pfn << PAGE_SHIFT));
1576#endif
1577}
1578
1579#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1580static void __meminit init_reserved_page(unsigned long pfn)
1581{
1582 pg_data_t *pgdat;
1583 int nid, zid;
1584
1585 if (!early_page_uninitialised(pfn))
1586 return;
1587
1588 nid = early_pfn_to_nid(pfn);
1589 pgdat = NODE_DATA(nid);
1590
1591 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
1592 struct zone *zone = &pgdat->node_zones[zid];
1593
1594 if (pfn >= zone->zone_start_pfn && pfn < zone_end_pfn(zone))
1595 break;
1596 }
1597 __init_single_page(pfn_to_page(pfn), pfn, zid, nid);
1598}
1599#else
1600static inline void init_reserved_page(unsigned long pfn)
1601{
1602}
1603#endif
1604
1605
1606
1607
1608
1609
1610
1611void __meminit reserve_bootmem_region(phys_addr_t start, phys_addr_t end)
1612{
1613 unsigned long start_pfn = PFN_DOWN(start);
1614 unsigned long end_pfn = PFN_UP(end);
1615
1616 for (; start_pfn < end_pfn; start_pfn++) {
1617 if (pfn_valid(start_pfn)) {
1618 struct page *page = pfn_to_page(start_pfn);
1619
1620 init_reserved_page(start_pfn);
1621
1622
1623 INIT_LIST_HEAD(&page->lru);
1624
1625
1626
1627
1628
1629
1630 __SetPageReserved(page);
1631 }
1632 }
1633}
1634
1635static void __free_pages_ok(struct page *page, unsigned int order,
1636 fpi_t fpi_flags)
1637{
1638 unsigned long flags;
1639 int migratetype;
1640 unsigned long pfn = page_to_pfn(page);
1641 struct zone *zone = page_zone(page);
1642
1643 if (!free_pages_prepare(page, order, true, fpi_flags))
1644 return;
1645
1646 migratetype = get_pfnblock_migratetype(page, pfn);
1647
1648 spin_lock_irqsave(&zone->lock, flags);
1649 if (unlikely(has_isolate_pageblock(zone) ||
1650 is_migrate_isolate(migratetype))) {
1651 migratetype = get_pfnblock_migratetype(page, pfn);
1652 }
1653 __free_one_page(page, pfn, zone, order, migratetype, fpi_flags);
1654 spin_unlock_irqrestore(&zone->lock, flags);
1655
1656 __count_vm_events(PGFREE, 1 << order);
1657}
1658
1659void __free_pages_core(struct page *page, unsigned int order)
1660{
1661 unsigned int nr_pages = 1 << order;
1662 struct page *p = page;
1663 unsigned int loop;
1664
1665
1666
1667
1668
1669
1670 prefetchw(p);
1671 for (loop = 0; loop < (nr_pages - 1); loop++, p++) {
1672 prefetchw(p + 1);
1673 __ClearPageReserved(p);
1674 set_page_count(p, 0);
1675 }
1676 __ClearPageReserved(p);
1677 set_page_count(p, 0);
1678
1679 atomic_long_add(nr_pages, &page_zone(page)->managed_pages);
1680
1681
1682
1683
1684
1685 __free_pages_ok(page, order, FPI_TO_TAIL | FPI_SKIP_KASAN_POISON);
1686}
1687
1688#ifdef CONFIG_NUMA
1689
1690
1691
1692
1693
1694
1695struct mminit_pfnnid_cache {
1696 unsigned long last_start;
1697 unsigned long last_end;
1698 int last_nid;
1699};
1700
1701static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata;
1702
1703
1704
1705
1706static int __meminit __early_pfn_to_nid(unsigned long pfn,
1707 struct mminit_pfnnid_cache *state)
1708{
1709 unsigned long start_pfn, end_pfn;
1710 int nid;
1711
1712 if (state->last_start <= pfn && pfn < state->last_end)
1713 return state->last_nid;
1714
1715 nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn);
1716 if (nid != NUMA_NO_NODE) {
1717 state->last_start = start_pfn;
1718 state->last_end = end_pfn;
1719 state->last_nid = nid;
1720 }
1721
1722 return nid;
1723}
1724
1725int __meminit early_pfn_to_nid(unsigned long pfn)
1726{
1727 static DEFINE_SPINLOCK(early_pfn_lock);
1728 int nid;
1729
1730 spin_lock(&early_pfn_lock);
1731 nid = __early_pfn_to_nid(pfn, &early_pfnnid_cache);
1732 if (nid < 0)
1733 nid = first_online_node;
1734 spin_unlock(&early_pfn_lock);
1735
1736 return nid;
1737}
1738#endif
1739
1740void __init memblock_free_pages(struct page *page, unsigned long pfn,
1741 unsigned int order)
1742{
1743 if (early_page_uninitialised(pfn))
1744 return;
1745 __free_pages_core(page, order);
1746}
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763struct page *__pageblock_pfn_to_page(unsigned long start_pfn,
1764 unsigned long end_pfn, struct zone *zone)
1765{
1766 struct page *start_page;
1767 struct page *end_page;
1768
1769
1770 end_pfn--;
1771
1772 if (!pfn_valid(start_pfn) || !pfn_valid(end_pfn))
1773 return NULL;
1774
1775 start_page = pfn_to_online_page(start_pfn);
1776 if (!start_page)
1777 return NULL;
1778
1779 if (page_zone(start_page) != zone)
1780 return NULL;
1781
1782 end_page = pfn_to_page(end_pfn);
1783
1784
1785 if (page_zone_id(start_page) != page_zone_id(end_page))
1786 return NULL;
1787
1788 return start_page;
1789}
1790
1791void set_zone_contiguous(struct zone *zone)
1792{
1793 unsigned long block_start_pfn = zone->zone_start_pfn;
1794 unsigned long block_end_pfn;
1795
1796 block_end_pfn = ALIGN(block_start_pfn + 1, pageblock_nr_pages);
1797 for (; block_start_pfn < zone_end_pfn(zone);
1798 block_start_pfn = block_end_pfn,
1799 block_end_pfn += pageblock_nr_pages) {
1800
1801 block_end_pfn = min(block_end_pfn, zone_end_pfn(zone));
1802
1803 if (!__pageblock_pfn_to_page(block_start_pfn,
1804 block_end_pfn, zone))
1805 return;
1806 cond_resched();
1807 }
1808
1809
1810 zone->contiguous = true;
1811}
1812
1813void clear_zone_contiguous(struct zone *zone)
1814{
1815 zone->contiguous = false;
1816}
1817
1818#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1819static void __init deferred_free_range(unsigned long pfn,
1820 unsigned long nr_pages)
1821{
1822 struct page *page;
1823 unsigned long i;
1824
1825 if (!nr_pages)
1826 return;
1827
1828 page = pfn_to_page(pfn);
1829
1830
1831 if (nr_pages == pageblock_nr_pages &&
1832 (pfn & (pageblock_nr_pages - 1)) == 0) {
1833 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
1834 __free_pages_core(page, pageblock_order);
1835 return;
1836 }
1837
1838 for (i = 0; i < nr_pages; i++, page++, pfn++) {
1839 if ((pfn & (pageblock_nr_pages - 1)) == 0)
1840 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
1841 __free_pages_core(page, 0);
1842 }
1843}
1844
1845
1846static atomic_t pgdat_init_n_undone __initdata;
1847static __initdata DECLARE_COMPLETION(pgdat_init_all_done_comp);
1848
1849static inline void __init pgdat_init_report_one_done(void)
1850{
1851 if (atomic_dec_and_test(&pgdat_init_n_undone))
1852 complete(&pgdat_init_all_done_comp);
1853}
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865static inline bool __init deferred_pfn_valid(unsigned long pfn)
1866{
1867 if (!(pfn & (pageblock_nr_pages - 1)) && !pfn_valid(pfn))
1868 return false;
1869 return true;
1870}
1871
1872
1873
1874
1875
1876static void __init deferred_free_pages(unsigned long pfn,
1877 unsigned long end_pfn)
1878{
1879 unsigned long nr_pgmask = pageblock_nr_pages - 1;
1880 unsigned long nr_free = 0;
1881
1882 for (; pfn < end_pfn; pfn++) {
1883 if (!deferred_pfn_valid(pfn)) {
1884 deferred_free_range(pfn - nr_free, nr_free);
1885 nr_free = 0;
1886 } else if (!(pfn & nr_pgmask)) {
1887 deferred_free_range(pfn - nr_free, nr_free);
1888 nr_free = 1;
1889 } else {
1890 nr_free++;
1891 }
1892 }
1893
1894 deferred_free_range(pfn - nr_free, nr_free);
1895}
1896
1897
1898
1899
1900
1901
1902static unsigned long __init deferred_init_pages(struct zone *zone,
1903 unsigned long pfn,
1904 unsigned long end_pfn)
1905{
1906 unsigned long nr_pgmask = pageblock_nr_pages - 1;
1907 int nid = zone_to_nid(zone);
1908 unsigned long nr_pages = 0;
1909 int zid = zone_idx(zone);
1910 struct page *page = NULL;
1911
1912 for (; pfn < end_pfn; pfn++) {
1913 if (!deferred_pfn_valid(pfn)) {
1914 page = NULL;
1915 continue;
1916 } else if (!page || !(pfn & nr_pgmask)) {
1917 page = pfn_to_page(pfn);
1918 } else {
1919 page++;
1920 }
1921 __init_single_page(page, pfn, zid, nid);
1922 nr_pages++;
1923 }
1924 return (nr_pages);
1925}
1926
1927
1928
1929
1930
1931
1932
1933static bool __init
1934deferred_init_mem_pfn_range_in_zone(u64 *i, struct zone *zone,
1935 unsigned long *spfn, unsigned long *epfn,
1936 unsigned long first_init_pfn)
1937{
1938 u64 j;
1939
1940
1941
1942
1943
1944
1945 for_each_free_mem_pfn_range_in_zone(j, zone, spfn, epfn) {
1946 if (*epfn <= first_init_pfn)
1947 continue;
1948 if (*spfn < first_init_pfn)
1949 *spfn = first_init_pfn;
1950 *i = j;
1951 return true;
1952 }
1953
1954 return false;
1955}
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967static unsigned long __init
1968deferred_init_maxorder(u64 *i, struct zone *zone, unsigned long *start_pfn,
1969 unsigned long *end_pfn)
1970{
1971 unsigned long mo_pfn = ALIGN(*start_pfn + 1, MAX_ORDER_NR_PAGES);
1972 unsigned long spfn = *start_pfn, epfn = *end_pfn;
1973 unsigned long nr_pages = 0;
1974 u64 j = *i;
1975
1976
1977 for_each_free_mem_pfn_range_in_zone_from(j, zone, start_pfn, end_pfn) {
1978 unsigned long t;
1979
1980 if (mo_pfn <= *start_pfn)
1981 break;
1982
1983 t = min(mo_pfn, *end_pfn);
1984 nr_pages += deferred_init_pages(zone, *start_pfn, t);
1985
1986 if (mo_pfn < *end_pfn) {
1987 *start_pfn = mo_pfn;
1988 break;
1989 }
1990 }
1991
1992
1993 swap(j, *i);
1994
1995 for_each_free_mem_pfn_range_in_zone_from(j, zone, &spfn, &epfn) {
1996 unsigned long t;
1997
1998 if (mo_pfn <= spfn)
1999 break;
2000
2001 t = min(mo_pfn, epfn);
2002 deferred_free_pages(spfn, t);
2003
2004 if (mo_pfn <= epfn)
2005 break;
2006 }
2007
2008 return nr_pages;
2009}
2010
2011static void __init
2012deferred_init_memmap_chunk(unsigned long start_pfn, unsigned long end_pfn,
2013 void *arg)
2014{
2015 unsigned long spfn, epfn;
2016 struct zone *zone = arg;
2017 u64 i;
2018
2019 deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn, start_pfn);
2020
2021
2022
2023
2024
2025 while (spfn < end_pfn) {
2026 deferred_init_maxorder(&i, zone, &spfn, &epfn);
2027 cond_resched();
2028 }
2029}
2030
2031
2032__weak int __init
2033deferred_page_init_max_threads(const struct cpumask *node_cpumask)
2034{
2035 return 1;
2036}
2037
2038
2039static int __init deferred_init_memmap(void *data)
2040{
2041 pg_data_t *pgdat = data;
2042 const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id);
2043 unsigned long spfn = 0, epfn = 0;
2044 unsigned long first_init_pfn, flags;
2045 unsigned long start = jiffies;
2046 struct zone *zone;
2047 int zid, max_threads;
2048 u64 i;
2049
2050
2051 if (!cpumask_empty(cpumask))
2052 set_cpus_allowed_ptr(current, cpumask);
2053
2054 pgdat_resize_lock(pgdat, &flags);
2055 first_init_pfn = pgdat->first_deferred_pfn;
2056 if (first_init_pfn == ULONG_MAX) {
2057 pgdat_resize_unlock(pgdat, &flags);
2058 pgdat_init_report_one_done();
2059 return 0;
2060 }
2061
2062
2063 BUG_ON(pgdat->first_deferred_pfn < pgdat->node_start_pfn);
2064 BUG_ON(pgdat->first_deferred_pfn > pgdat_end_pfn(pgdat));
2065 pgdat->first_deferred_pfn = ULONG_MAX;
2066
2067
2068
2069
2070
2071
2072 pgdat_resize_unlock(pgdat, &flags);
2073
2074
2075 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
2076 zone = pgdat->node_zones + zid;
2077 if (first_init_pfn < zone_end_pfn(zone))
2078 break;
2079 }
2080
2081
2082 if (!deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn,
2083 first_init_pfn))
2084 goto zone_empty;
2085
2086 max_threads = deferred_page_init_max_threads(cpumask);
2087
2088 while (spfn < epfn) {
2089 unsigned long epfn_align = ALIGN(epfn, PAGES_PER_SECTION);
2090 struct padata_mt_job job = {
2091 .thread_fn = deferred_init_memmap_chunk,
2092 .fn_arg = zone,
2093 .start = spfn,
2094 .size = epfn_align - spfn,
2095 .align = PAGES_PER_SECTION,
2096 .min_chunk = PAGES_PER_SECTION,
2097 .max_threads = max_threads,
2098 };
2099
2100 padata_do_multithreaded(&job);
2101 deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn,
2102 epfn_align);
2103 }
2104zone_empty:
2105
2106 WARN_ON(++zid < MAX_NR_ZONES && populated_zone(++zone));
2107
2108 pr_info("node %d deferred pages initialised in %ums\n",
2109 pgdat->node_id, jiffies_to_msecs(jiffies - start));
2110
2111 pgdat_init_report_one_done();
2112 return 0;
2113}
2114
2115
2116
2117
2118
2119
2120
2121
2122
2123
2124
2125
2126
2127
2128
2129
2130static noinline bool __init
2131deferred_grow_zone(struct zone *zone, unsigned int order)
2132{
2133 unsigned long nr_pages_needed = ALIGN(1 << order, PAGES_PER_SECTION);
2134 pg_data_t *pgdat = zone->zone_pgdat;
2135 unsigned long first_deferred_pfn = pgdat->first_deferred_pfn;
2136 unsigned long spfn, epfn, flags;
2137 unsigned long nr_pages = 0;
2138 u64 i;
2139
2140
2141 if (zone_end_pfn(zone) != pgdat_end_pfn(pgdat))
2142 return false;
2143
2144 pgdat_resize_lock(pgdat, &flags);
2145
2146
2147
2148
2149
2150 if (first_deferred_pfn != pgdat->first_deferred_pfn) {
2151 pgdat_resize_unlock(pgdat, &flags);
2152 return true;
2153 }
2154
2155
2156 if (!deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn,
2157 first_deferred_pfn)) {
2158 pgdat->first_deferred_pfn = ULONG_MAX;
2159 pgdat_resize_unlock(pgdat, &flags);
2160
2161 return first_deferred_pfn != ULONG_MAX;
2162 }
2163
2164
2165
2166
2167
2168
2169 while (spfn < epfn) {
2170
2171 first_deferred_pfn = spfn;
2172
2173 nr_pages += deferred_init_maxorder(&i, zone, &spfn, &epfn);
2174 touch_nmi_watchdog();
2175
2176
2177 if ((first_deferred_pfn ^ spfn) < PAGES_PER_SECTION)
2178 continue;
2179
2180
2181 if (nr_pages >= nr_pages_needed)
2182 break;
2183 }
2184
2185 pgdat->first_deferred_pfn = spfn;
2186 pgdat_resize_unlock(pgdat, &flags);
2187
2188 return nr_pages > 0;
2189}
2190
2191
2192
2193
2194
2195
2196
2197static bool __ref
2198_deferred_grow_zone(struct zone *zone, unsigned int order)
2199{
2200 return deferred_grow_zone(zone, order);
2201}
2202
2203#endif
2204
2205void __init page_alloc_init_late(void)
2206{
2207 struct zone *zone;
2208 int nid;
2209
2210#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
2211
2212
2213 atomic_set(&pgdat_init_n_undone, num_node_state(N_MEMORY));
2214 for_each_node_state(nid, N_MEMORY) {
2215 kthread_run(deferred_init_memmap, NODE_DATA(nid), "pgdatinit%d", nid);
2216 }
2217
2218
2219 wait_for_completion(&pgdat_init_all_done_comp);
2220
2221
2222
2223
2224
2225 static_branch_disable(&deferred_pages);
2226
2227
2228 files_maxfiles_init();
2229#endif
2230
2231 buffer_init();
2232
2233
2234 memblock_discard();
2235
2236 for_each_node_state(nid, N_MEMORY)
2237 shuffle_free_memory(NODE_DATA(nid));
2238
2239 for_each_populated_zone(zone)
2240 set_zone_contiguous(zone);
2241}
2242
2243#ifdef CONFIG_CMA
2244
2245void __init init_cma_reserved_pageblock(struct page *page)
2246{
2247 unsigned i = pageblock_nr_pages;
2248 struct page *p = page;
2249
2250 do {
2251 __ClearPageReserved(p);
2252 set_page_count(p, 0);
2253 } while (++p, --i);
2254
2255 set_pageblock_migratetype(page, MIGRATE_CMA);
2256
2257 if (pageblock_order >= MAX_ORDER) {
2258 i = pageblock_nr_pages;
2259 p = page;
2260 do {
2261 set_page_refcounted(p);
2262 __free_pages(p, MAX_ORDER - 1);
2263 p += MAX_ORDER_NR_PAGES;
2264 } while (i -= MAX_ORDER_NR_PAGES);
2265 } else {
2266 set_page_refcounted(page);
2267 __free_pages(page, pageblock_order);
2268 }
2269
2270 adjust_managed_page_count(page, pageblock_nr_pages);
2271 page_zone(page)->cma_pages += pageblock_nr_pages;
2272}
2273#endif
2274
2275
2276
2277
2278
2279
2280
2281
2282
2283
2284
2285
2286
2287
2288
2289static inline void expand(struct zone *zone, struct page *page,
2290 int low, int high, int migratetype)
2291{
2292 unsigned long size = 1 << high;
2293
2294 while (high > low) {
2295 high--;
2296 size >>= 1;
2297 VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]);
2298
2299
2300
2301
2302
2303
2304
2305 if (set_page_guard(zone, &page[size], high, migratetype))
2306 continue;
2307
2308 add_to_free_list(&page[size], zone, high, migratetype);
2309 set_buddy_order(&page[size], high);
2310 }
2311}
2312
2313static void check_new_page_bad(struct page *page)
2314{
2315 if (unlikely(page->flags & __PG_HWPOISON)) {
2316
2317 page_mapcount_reset(page);
2318 return;
2319 }
2320
2321 bad_page(page,
2322 page_bad_reason(page, PAGE_FLAGS_CHECK_AT_PREP));
2323}
2324
2325
2326
2327
2328static inline int check_new_page(struct page *page)
2329{
2330 if (likely(page_expected_state(page,
2331 PAGE_FLAGS_CHECK_AT_PREP|__PG_HWPOISON)))
2332 return 0;
2333
2334 check_new_page_bad(page);
2335 return 1;
2336}
2337
2338#ifdef CONFIG_DEBUG_VM
2339
2340
2341
2342
2343
2344static inline bool check_pcp_refill(struct page *page)
2345{
2346 if (debug_pagealloc_enabled_static())
2347 return check_new_page(page);
2348 else
2349 return false;
2350}
2351
2352static inline bool check_new_pcp(struct page *page)
2353{
2354 return check_new_page(page);
2355}
2356#else
2357
2358
2359
2360
2361
2362static inline bool check_pcp_refill(struct page *page)
2363{
2364 return check_new_page(page);
2365}
2366static inline bool check_new_pcp(struct page *page)
2367{
2368 if (debug_pagealloc_enabled_static())
2369 return check_new_page(page);
2370 else
2371 return false;
2372}
2373#endif
2374
2375static bool check_new_pages(struct page *page, unsigned int order)
2376{
2377 int i;
2378 for (i = 0; i < (1 << order); i++) {
2379 struct page *p = page + i;
2380
2381 if (unlikely(check_new_page(p)))
2382 return true;
2383 }
2384
2385 return false;
2386}
2387
2388inline void post_alloc_hook(struct page *page, unsigned int order,
2389 gfp_t gfp_flags)
2390{
2391 set_page_private(page, 0);
2392 set_page_refcounted(page);
2393
2394 arch_alloc_page(page, order);
2395 debug_pagealloc_map_pages(page, 1 << order);
2396
2397
2398
2399
2400
2401
2402 kernel_unpoison_pages(page, 1 << order);
2403
2404
2405
2406
2407
2408
2409 if (kasan_has_integrated_init()) {
2410 kasan_alloc_pages(page, order, gfp_flags);
2411 } else {
2412 bool init = !want_init_on_free() && want_init_on_alloc(gfp_flags);
2413
2414 kasan_unpoison_pages(page, order, init);
2415 if (init)
2416 kernel_init_free_pages(page, 1 << order,
2417 gfp_flags & __GFP_ZEROTAGS);
2418 }
2419
2420 set_page_owner(page, order, gfp_flags);
2421}
2422
2423static void prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags,
2424 unsigned int alloc_flags)
2425{
2426 post_alloc_hook(page, order, gfp_flags);
2427
2428 if (order && (gfp_flags & __GFP_COMP))
2429 prep_compound_page(page, order);
2430
2431
2432
2433
2434
2435
2436
2437 if (alloc_flags & ALLOC_NO_WATERMARKS)
2438 set_page_pfmemalloc(page);
2439 else
2440 clear_page_pfmemalloc(page);
2441}
2442
2443
2444
2445
2446
2447static __always_inline
2448struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
2449 int migratetype)
2450{
2451 unsigned int current_order;
2452 struct free_area *area;
2453 struct page *page;
2454
2455
2456 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
2457 area = &(zone->free_area[current_order]);
2458 page = get_page_from_free_area(area, migratetype);
2459 if (!page)
2460 continue;
2461 del_page_from_free_list(page, zone, current_order);
2462 expand(zone, page, order, current_order, migratetype);
2463 set_pcppage_migratetype(page, migratetype);
2464 return page;
2465 }
2466
2467 return NULL;
2468}
2469
2470
2471
2472
2473
2474
2475static int fallbacks[MIGRATE_TYPES][3] = {
2476 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_TYPES },
2477 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_TYPES },
2478 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_TYPES },
2479#ifdef CONFIG_CMA
2480 [MIGRATE_CMA] = { MIGRATE_TYPES },
2481#endif
2482#ifdef CONFIG_MEMORY_ISOLATION
2483 [MIGRATE_ISOLATE] = { MIGRATE_TYPES },
2484#endif
2485};
2486
2487#ifdef CONFIG_CMA
2488static __always_inline struct page *__rmqueue_cma_fallback(struct zone *zone,
2489 unsigned int order)
2490{
2491 return __rmqueue_smallest(zone, order, MIGRATE_CMA);
2492}
2493#else
2494static inline struct page *__rmqueue_cma_fallback(struct zone *zone,
2495 unsigned int order) { return NULL; }
2496#endif
2497
2498
2499
2500
2501
2502
2503static int move_freepages(struct zone *zone,
2504 unsigned long start_pfn, unsigned long end_pfn,
2505 int migratetype, int *num_movable)
2506{
2507 struct page *page;
2508 unsigned long pfn;
2509 unsigned int order;
2510 int pages_moved = 0;
2511
2512 for (pfn = start_pfn; pfn <= end_pfn;) {
2513 page = pfn_to_page(pfn);
2514 if (!PageBuddy(page)) {
2515
2516
2517
2518
2519
2520 if (num_movable &&
2521 (PageLRU(page) || __PageMovable(page)))
2522 (*num_movable)++;
2523 pfn++;
2524 continue;
2525 }
2526
2527
2528 VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page);
2529 VM_BUG_ON_PAGE(page_zone(page) != zone, page);
2530
2531 order = buddy_order(page);
2532 move_to_free_list(page, zone, order, migratetype);
2533 pfn += 1 << order;
2534 pages_moved += 1 << order;
2535 }
2536
2537 return pages_moved;
2538}
2539
2540int move_freepages_block(struct zone *zone, struct page *page,
2541 int migratetype, int *num_movable)
2542{
2543 unsigned long start_pfn, end_pfn, pfn;
2544
2545 if (num_movable)
2546 *num_movable = 0;
2547
2548 pfn = page_to_pfn(page);
2549 start_pfn = pfn & ~(pageblock_nr_pages - 1);
2550 end_pfn = start_pfn + pageblock_nr_pages - 1;
2551
2552
2553 if (!zone_spans_pfn(zone, start_pfn))
2554 start_pfn = pfn;
2555 if (!zone_spans_pfn(zone, end_pfn))
2556 return 0;
2557
2558 return move_freepages(zone, start_pfn, end_pfn, migratetype,
2559 num_movable);
2560}
2561
2562static void change_pageblock_range(struct page *pageblock_page,
2563 int start_order, int migratetype)
2564{
2565 int nr_pageblocks = 1 << (start_order - pageblock_order);
2566
2567 while (nr_pageblocks--) {
2568 set_pageblock_migratetype(pageblock_page, migratetype);
2569 pageblock_page += pageblock_nr_pages;
2570 }
2571}
2572
2573
2574
2575
2576
2577
2578
2579
2580
2581
2582
2583
2584
2585static bool can_steal_fallback(unsigned int order, int start_mt)
2586{
2587
2588
2589
2590
2591
2592
2593
2594 if (order >= pageblock_order)
2595 return true;
2596
2597 if (order >= pageblock_order / 2 ||
2598 start_mt == MIGRATE_RECLAIMABLE ||
2599 start_mt == MIGRATE_UNMOVABLE ||
2600 page_group_by_mobility_disabled)
2601 return true;
2602
2603 return false;
2604}
2605
2606static inline bool boost_watermark(struct zone *zone)
2607{
2608 unsigned long max_boost;
2609
2610 if (!watermark_boost_factor)
2611 return false;
2612
2613
2614
2615
2616
2617
2618 if ((pageblock_nr_pages * 4) > zone_managed_pages(zone))
2619 return false;
2620
2621 max_boost = mult_frac(zone->_watermark[WMARK_HIGH],
2622 watermark_boost_factor, 10000);
2623
2624
2625
2626
2627
2628
2629
2630
2631
2632 if (!max_boost)
2633 return false;
2634
2635 max_boost = max(pageblock_nr_pages, max_boost);
2636
2637 zone->watermark_boost = min(zone->watermark_boost + pageblock_nr_pages,
2638 max_boost);
2639
2640 return true;
2641}
2642
2643
2644
2645
2646
2647
2648
2649
2650
2651static void steal_suitable_fallback(struct zone *zone, struct page *page,
2652 unsigned int alloc_flags, int start_type, bool whole_block)
2653{
2654 unsigned int current_order = buddy_order(page);
2655 int free_pages, movable_pages, alike_pages;
2656 int old_block_type;
2657
2658 old_block_type = get_pageblock_migratetype(page);
2659
2660
2661
2662
2663
2664 if (is_migrate_highatomic(old_block_type))
2665 goto single_page;
2666
2667
2668 if (current_order >= pageblock_order) {
2669 change_pageblock_range(page, current_order, start_type);
2670 goto single_page;
2671 }
2672
2673
2674
2675
2676
2677
2678 if (boost_watermark(zone) && (alloc_flags & ALLOC_KSWAPD))
2679 set_bit(ZONE_BOOSTED_WATERMARK, &zone->flags);
2680
2681
2682 if (!whole_block)
2683 goto single_page;
2684
2685 free_pages = move_freepages_block(zone, page, start_type,
2686 &movable_pages);
2687
2688
2689
2690
2691
2692 if (start_type == MIGRATE_MOVABLE) {
2693 alike_pages = movable_pages;
2694 } else {
2695
2696
2697
2698
2699
2700
2701
2702 if (old_block_type == MIGRATE_MOVABLE)
2703 alike_pages = pageblock_nr_pages
2704 - (free_pages + movable_pages);
2705 else
2706 alike_pages = 0;
2707 }
2708
2709
2710 if (!free_pages)
2711 goto single_page;
2712
2713
2714
2715
2716
2717 if (free_pages + alike_pages >= (1 << (pageblock_order-1)) ||
2718 page_group_by_mobility_disabled)
2719 set_pageblock_migratetype(page, start_type);
2720
2721 return;
2722
2723single_page:
2724 move_to_free_list(page, zone, current_order, start_type);
2725}
2726
2727
2728
2729
2730
2731
2732
2733int find_suitable_fallback(struct free_area *area, unsigned int order,
2734 int migratetype, bool only_stealable, bool *can_steal)
2735{
2736 int i;
2737 int fallback_mt;
2738
2739 if (area->nr_free == 0)
2740 return -1;
2741
2742 *can_steal = false;
2743 for (i = 0;; i++) {
2744 fallback_mt = fallbacks[migratetype][i];
2745 if (fallback_mt == MIGRATE_TYPES)
2746 break;
2747
2748 if (free_area_empty(area, fallback_mt))
2749 continue;
2750
2751 if (can_steal_fallback(order, migratetype))
2752 *can_steal = true;
2753
2754 if (!only_stealable)
2755 return fallback_mt;
2756
2757 if (*can_steal)
2758 return fallback_mt;
2759 }
2760
2761 return -1;
2762}
2763
2764
2765
2766
2767
2768static void reserve_highatomic_pageblock(struct page *page, struct zone *zone,
2769 unsigned int alloc_order)
2770{
2771 int mt;
2772 unsigned long max_managed, flags;
2773
2774
2775
2776
2777
2778 max_managed = (zone_managed_pages(zone) / 100) + pageblock_nr_pages;
2779 if (zone->nr_reserved_highatomic >= max_managed)
2780 return;
2781
2782 spin_lock_irqsave(&zone->lock, flags);
2783
2784
2785 if (zone->nr_reserved_highatomic >= max_managed)
2786 goto out_unlock;
2787
2788
2789 mt = get_pageblock_migratetype(page);
2790 if (!is_migrate_highatomic(mt) && !is_migrate_isolate(mt)
2791 && !is_migrate_cma(mt)) {
2792 zone->nr_reserved_highatomic += pageblock_nr_pages;
2793 set_pageblock_migratetype(page, MIGRATE_HIGHATOMIC);
2794 move_freepages_block(zone, page, MIGRATE_HIGHATOMIC, NULL);
2795 }
2796
2797out_unlock:
2798 spin_unlock_irqrestore(&zone->lock, flags);
2799}
2800
2801
2802
2803
2804
2805
2806
2807
2808
2809
2810static bool unreserve_highatomic_pageblock(const struct alloc_context *ac,
2811 bool force)
2812{
2813 struct zonelist *zonelist = ac->zonelist;
2814 unsigned long flags;
2815 struct zoneref *z;
2816 struct zone *zone;
2817 struct page *page;
2818 int order;
2819 bool ret;
2820
2821 for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->highest_zoneidx,
2822 ac->nodemask) {
2823
2824
2825
2826
2827 if (!force && zone->nr_reserved_highatomic <=
2828 pageblock_nr_pages)
2829 continue;
2830
2831 spin_lock_irqsave(&zone->lock, flags);
2832 for (order = 0; order < MAX_ORDER; order++) {
2833 struct free_area *area = &(zone->free_area[order]);
2834
2835 page = get_page_from_free_area(area, MIGRATE_HIGHATOMIC);
2836 if (!page)
2837 continue;
2838
2839
2840
2841
2842
2843
2844
2845
2846 if (is_migrate_highatomic_page(page)) {
2847
2848
2849
2850
2851
2852
2853
2854 zone->nr_reserved_highatomic -= min(
2855 pageblock_nr_pages,
2856 zone->nr_reserved_highatomic);
2857 }
2858
2859
2860
2861
2862
2863
2864
2865
2866
2867
2868 set_pageblock_migratetype(page, ac->migratetype);
2869 ret = move_freepages_block(zone, page, ac->migratetype,
2870 NULL);
2871 if (ret) {
2872 spin_unlock_irqrestore(&zone->lock, flags);
2873 return ret;
2874 }
2875 }
2876 spin_unlock_irqrestore(&zone->lock, flags);
2877 }
2878
2879 return false;
2880}
2881
2882
2883
2884
2885
2886
2887
2888
2889
2890
2891
2892static __always_inline bool
2893__rmqueue_fallback(struct zone *zone, int order, int start_migratetype,
2894 unsigned int alloc_flags)
2895{
2896 struct free_area *area;
2897 int current_order;
2898 int min_order = order;
2899 struct page *page;
2900 int fallback_mt;
2901 bool can_steal;
2902
2903
2904
2905
2906
2907
2908 if (alloc_flags & ALLOC_NOFRAGMENT)
2909 min_order = pageblock_order;
2910
2911
2912
2913
2914
2915
2916 for (current_order = MAX_ORDER - 1; current_order >= min_order;
2917 --current_order) {
2918 area = &(zone->free_area[current_order]);
2919 fallback_mt = find_suitable_fallback(area, current_order,
2920 start_migratetype, false, &can_steal);
2921 if (fallback_mt == -1)
2922 continue;
2923
2924
2925
2926
2927
2928
2929
2930
2931
2932 if (!can_steal && start_migratetype == MIGRATE_MOVABLE
2933 && current_order > order)
2934 goto find_smallest;
2935
2936 goto do_steal;
2937 }
2938
2939 return false;
2940
2941find_smallest:
2942 for (current_order = order; current_order < MAX_ORDER;
2943 current_order++) {
2944 area = &(zone->free_area[current_order]);
2945 fallback_mt = find_suitable_fallback(area, current_order,
2946 start_migratetype, false, &can_steal);
2947 if (fallback_mt != -1)
2948 break;
2949 }
2950
2951
2952
2953
2954
2955 VM_BUG_ON(current_order == MAX_ORDER);
2956
2957do_steal:
2958 page = get_page_from_free_area(area, fallback_mt);
2959
2960 steal_suitable_fallback(zone, page, alloc_flags, start_migratetype,
2961 can_steal);
2962
2963 trace_mm_page_alloc_extfrag(page, order, current_order,
2964 start_migratetype, fallback_mt);
2965
2966 return true;
2967
2968}
2969
2970
2971
2972
2973
2974static __always_inline struct page *
2975__rmqueue(struct zone *zone, unsigned int order, int migratetype,
2976 unsigned int alloc_flags)
2977{
2978 struct page *page;
2979
2980 if (IS_ENABLED(CONFIG_CMA)) {
2981
2982
2983
2984
2985
2986 if (alloc_flags & ALLOC_CMA &&
2987 zone_page_state(zone, NR_FREE_CMA_PAGES) >
2988 zone_page_state(zone, NR_FREE_PAGES) / 2) {
2989 page = __rmqueue_cma_fallback(zone, order);
2990 if (page)
2991 goto out;
2992 }
2993 }
2994retry:
2995 page = __rmqueue_smallest(zone, order, migratetype);
2996 if (unlikely(!page)) {
2997 if (alloc_flags & ALLOC_CMA)
2998 page = __rmqueue_cma_fallback(zone, order);
2999
3000 if (!page && __rmqueue_fallback(zone, order, migratetype,
3001 alloc_flags))
3002 goto retry;
3003 }
3004out:
3005 if (page)
3006 trace_mm_page_alloc_zone_locked(page, order, migratetype);
3007 return page;
3008}
3009
3010
3011
3012
3013
3014
3015static int rmqueue_bulk(struct zone *zone, unsigned int order,
3016 unsigned long count, struct list_head *list,
3017 int migratetype, unsigned int alloc_flags)
3018{
3019 int i, allocated = 0;
3020
3021
3022
3023
3024
3025 spin_lock(&zone->lock);
3026 for (i = 0; i < count; ++i) {
3027 struct page *page = __rmqueue(zone, order, migratetype,
3028 alloc_flags);
3029 if (unlikely(page == NULL))
3030 break;
3031
3032 if (unlikely(check_pcp_refill(page)))
3033 continue;
3034
3035
3036
3037
3038
3039
3040
3041
3042
3043
3044
3045 list_add_tail(&page->lru, list);
3046 allocated++;
3047 if (is_migrate_cma(get_pcppage_migratetype(page)))
3048 __mod_zone_page_state(zone, NR_FREE_CMA_PAGES,
3049 -(1 << order));
3050 }
3051
3052
3053
3054
3055
3056
3057
3058 __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
3059 spin_unlock(&zone->lock);
3060 return allocated;
3061}
3062
3063#ifdef CONFIG_NUMA
3064
3065
3066
3067
3068
3069
3070
3071
3072void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
3073{
3074 unsigned long flags;
3075 int to_drain, batch;
3076
3077 local_lock_irqsave(&pagesets.lock, flags);
3078 batch = READ_ONCE(pcp->batch);
3079 to_drain = min(pcp->count, batch);
3080 if (to_drain > 0)
3081 free_pcppages_bulk(zone, to_drain, pcp);
3082 local_unlock_irqrestore(&pagesets.lock, flags);
3083}
3084#endif
3085
3086
3087
3088
3089
3090
3091
3092
3093static void drain_pages_zone(unsigned int cpu, struct zone *zone)
3094{
3095 unsigned long flags;
3096 struct per_cpu_pages *pcp;
3097
3098 local_lock_irqsave(&pagesets.lock, flags);
3099
3100 pcp = per_cpu_ptr(zone->per_cpu_pageset, cpu);
3101 if (pcp->count)
3102 free_pcppages_bulk(zone, pcp->count, pcp);
3103
3104 local_unlock_irqrestore(&pagesets.lock, flags);
3105}
3106
3107
3108
3109
3110
3111
3112
3113
3114static void drain_pages(unsigned int cpu)
3115{
3116 struct zone *zone;
3117
3118 for_each_populated_zone(zone) {
3119 drain_pages_zone(cpu, zone);
3120 }
3121}
3122
3123
3124
3125
3126
3127
3128
3129void drain_local_pages(struct zone *zone)
3130{
3131 int cpu = smp_processor_id();
3132
3133 if (zone)
3134 drain_pages_zone(cpu, zone);
3135 else
3136 drain_pages(cpu);
3137}
3138
3139static void drain_local_pages_wq(struct work_struct *work)
3140{
3141 struct pcpu_drain *drain;
3142
3143 drain = container_of(work, struct pcpu_drain, work);
3144
3145
3146
3147
3148
3149
3150
3151
3152 preempt_disable();
3153 drain_local_pages(drain->zone);
3154 preempt_enable();
3155}
3156
3157
3158
3159
3160
3161
3162
3163
3164
3165
3166
3167static void __drain_all_pages(struct zone *zone, bool force_all_cpus)
3168{
3169 int cpu;
3170
3171
3172
3173
3174
3175 static cpumask_t cpus_with_pcps;
3176
3177
3178
3179
3180
3181 if (WARN_ON_ONCE(!mm_percpu_wq))
3182 return;
3183
3184
3185
3186
3187
3188
3189 if (unlikely(!mutex_trylock(&pcpu_drain_mutex))) {
3190 if (!zone)
3191 return;
3192 mutex_lock(&pcpu_drain_mutex);
3193 }
3194
3195
3196
3197
3198
3199
3200
3201 for_each_online_cpu(cpu) {
3202 struct per_cpu_pages *pcp;
3203 struct zone *z;
3204 bool has_pcps = false;
3205
3206 if (force_all_cpus) {
3207
3208
3209
3210
3211 has_pcps = true;
3212 } else if (zone) {
3213 pcp = per_cpu_ptr(zone->per_cpu_pageset, cpu);
3214 if (pcp->count)
3215 has_pcps = true;
3216 } else {
3217 for_each_populated_zone(z) {
3218 pcp = per_cpu_ptr(z->per_cpu_pageset, cpu);
3219 if (pcp->count) {
3220 has_pcps = true;
3221 break;
3222 }
3223 }
3224 }
3225
3226 if (has_pcps)
3227 cpumask_set_cpu(cpu, &cpus_with_pcps);
3228 else
3229 cpumask_clear_cpu(cpu, &cpus_with_pcps);
3230 }
3231
3232 for_each_cpu(cpu, &cpus_with_pcps) {
3233 struct pcpu_drain *drain = per_cpu_ptr(&pcpu_drain, cpu);
3234
3235 drain->zone = zone;
3236 INIT_WORK(&drain->work, drain_local_pages_wq);
3237 queue_work_on(cpu, mm_percpu_wq, &drain->work);
3238 }
3239 for_each_cpu(cpu, &cpus_with_pcps)
3240 flush_work(&per_cpu_ptr(&pcpu_drain, cpu)->work);
3241
3242 mutex_unlock(&pcpu_drain_mutex);
3243}
3244
3245
3246
3247
3248
3249
3250
3251
3252void drain_all_pages(struct zone *zone)
3253{
3254 __drain_all_pages(zone, false);
3255}
3256
3257#ifdef CONFIG_HIBERNATION
3258
3259
3260
3261
3262#define WD_PAGE_COUNT (128*1024)
3263
3264void mark_free_pages(struct zone *zone)
3265{
3266 unsigned long pfn, max_zone_pfn, page_count = WD_PAGE_COUNT;
3267 unsigned long flags;
3268 unsigned int order, t;
3269 struct page *page;
3270
3271 if (zone_is_empty(zone))
3272 return;
3273
3274 spin_lock_irqsave(&zone->lock, flags);
3275
3276 max_zone_pfn = zone_end_pfn(zone);
3277 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
3278 if (pfn_valid(pfn)) {
3279 page = pfn_to_page(pfn);
3280
3281 if (!--page_count) {
3282 touch_nmi_watchdog();
3283 page_count = WD_PAGE_COUNT;
3284 }
3285
3286 if (page_zone(page) != zone)
3287 continue;
3288
3289 if (!swsusp_page_is_forbidden(page))
3290 swsusp_unset_page_free(page);
3291 }
3292
3293 for_each_migratetype_order(order, t) {
3294 list_for_each_entry(page,
3295 &zone->free_area[order].free_list[t], lru) {
3296 unsigned long i;
3297
3298 pfn = page_to_pfn(page);
3299 for (i = 0; i < (1UL << order); i++) {
3300 if (!--page_count) {
3301 touch_nmi_watchdog();
3302 page_count = WD_PAGE_COUNT;
3303 }
3304 swsusp_set_page_free(pfn_to_page(pfn + i));
3305 }
3306 }
3307 }
3308 spin_unlock_irqrestore(&zone->lock, flags);
3309}
3310#endif
3311
3312static bool free_unref_page_prepare(struct page *page, unsigned long pfn,
3313 unsigned int order)
3314{
3315 int migratetype;
3316
3317 if (!free_pcp_prepare(page, order))
3318 return false;
3319
3320 migratetype = get_pfnblock_migratetype(page, pfn);
3321 set_pcppage_migratetype(page, migratetype);
3322 return true;
3323}
3324
3325static int nr_pcp_free(struct per_cpu_pages *pcp, int high, int batch)
3326{
3327 int min_nr_free, max_nr_free;
3328
3329
3330 if (unlikely(high < batch))
3331 return 1;
3332
3333
3334 min_nr_free = batch;
3335 max_nr_free = high - batch;
3336
3337
3338
3339
3340
3341 batch <<= pcp->free_factor;
3342 if (batch < max_nr_free)
3343 pcp->free_factor++;
3344 batch = clamp(batch, min_nr_free, max_nr_free);
3345
3346 return batch;
3347}
3348
3349static int nr_pcp_high(struct per_cpu_pages *pcp, struct zone *zone)
3350{
3351 int high = READ_ONCE(pcp->high);
3352
3353 if (unlikely(!high))
3354 return 0;
3355
3356 if (!test_bit(ZONE_RECLAIM_ACTIVE, &zone->flags))
3357 return high;
3358
3359
3360
3361
3362
3363 return min(READ_ONCE(pcp->batch) << 2, high);
3364}
3365
3366static void free_unref_page_commit(struct page *page, unsigned long pfn,
3367 int migratetype, unsigned int order)
3368{
3369 struct zone *zone = page_zone(page);
3370 struct per_cpu_pages *pcp;
3371 int high;
3372 int pindex;
3373
3374 __count_vm_event(PGFREE);
3375 pcp = this_cpu_ptr(zone->per_cpu_pageset);
3376 pindex = order_to_pindex(migratetype, order);
3377 list_add(&page->lru, &pcp->lists[pindex]);
3378 pcp->count += 1 << order;
3379 high = nr_pcp_high(pcp, zone);
3380 if (pcp->count >= high) {
3381 int batch = READ_ONCE(pcp->batch);
3382
3383 free_pcppages_bulk(zone, nr_pcp_free(pcp, high, batch), pcp);
3384 }
3385}
3386
3387
3388
3389
3390void free_unref_page(struct page *page, unsigned int order)
3391{
3392 unsigned long flags;
3393 unsigned long pfn = page_to_pfn(page);
3394 int migratetype;
3395
3396 if (!free_unref_page_prepare(page, pfn, order))
3397 return;
3398
3399
3400
3401
3402
3403
3404
3405
3406 migratetype = get_pcppage_migratetype(page);
3407 if (unlikely(migratetype >= MIGRATE_PCPTYPES)) {
3408 if (unlikely(is_migrate_isolate(migratetype))) {
3409 free_one_page(page_zone(page), page, pfn, order, migratetype, FPI_NONE);
3410 return;
3411 }
3412 migratetype = MIGRATE_MOVABLE;
3413 }
3414
3415 local_lock_irqsave(&pagesets.lock, flags);
3416 free_unref_page_commit(page, pfn, migratetype, order);
3417 local_unlock_irqrestore(&pagesets.lock, flags);
3418}
3419
3420
3421
3422
3423void free_unref_page_list(struct list_head *list)
3424{
3425 struct page *page, *next;
3426 unsigned long flags, pfn;
3427 int batch_count = 0;
3428 int migratetype;
3429
3430
3431 list_for_each_entry_safe(page, next, list, lru) {
3432 pfn = page_to_pfn(page);
3433 if (!free_unref_page_prepare(page, pfn, 0)) {
3434 list_del(&page->lru);
3435 continue;
3436 }
3437
3438
3439
3440
3441
3442 migratetype = get_pcppage_migratetype(page);
3443 if (unlikely(is_migrate_isolate(migratetype))) {
3444 list_del(&page->lru);
3445 free_one_page(page_zone(page), page, pfn, 0, migratetype, FPI_NONE);
3446 continue;
3447 }
3448
3449 set_page_private(page, pfn);
3450 }
3451
3452 local_lock_irqsave(&pagesets.lock, flags);
3453 list_for_each_entry_safe(page, next, list, lru) {
3454 pfn = page_private(page);
3455 set_page_private(page, 0);
3456
3457
3458
3459
3460
3461 migratetype = get_pcppage_migratetype(page);
3462 if (unlikely(migratetype >= MIGRATE_PCPTYPES))
3463 migratetype = MIGRATE_MOVABLE;
3464
3465 trace_mm_page_free_batched(page);
3466 free_unref_page_commit(page, pfn, migratetype, 0);
3467
3468
3469
3470
3471
3472 if (++batch_count == SWAP_CLUSTER_MAX) {
3473 local_unlock_irqrestore(&pagesets.lock, flags);
3474 batch_count = 0;
3475 local_lock_irqsave(&pagesets.lock, flags);
3476 }
3477 }
3478 local_unlock_irqrestore(&pagesets.lock, flags);
3479}
3480
3481
3482
3483
3484
3485
3486
3487
3488
3489void split_page(struct page *page, unsigned int order)
3490{
3491 int i;
3492
3493 VM_BUG_ON_PAGE(PageCompound(page), page);
3494 VM_BUG_ON_PAGE(!page_count(page), page);
3495
3496 for (i = 1; i < (1 << order); i++)
3497 set_page_refcounted(page + i);
3498 split_page_owner(page, 1 << order);
3499 split_page_memcg(page, 1 << order);
3500}
3501EXPORT_SYMBOL_GPL(split_page);
3502
3503int __isolate_free_page(struct page *page, unsigned int order)
3504{
3505 unsigned long watermark;
3506 struct zone *zone;
3507 int mt;
3508
3509 BUG_ON(!PageBuddy(page));
3510
3511 zone = page_zone(page);
3512 mt = get_pageblock_migratetype(page);
3513
3514 if (!is_migrate_isolate(mt)) {
3515
3516
3517
3518
3519
3520
3521 watermark = zone->_watermark[WMARK_MIN] + (1UL << order);
3522 if (!zone_watermark_ok(zone, 0, watermark, 0, ALLOC_CMA))
3523 return 0;
3524
3525 __mod_zone_freepage_state(zone, -(1UL << order), mt);
3526 }
3527
3528
3529
3530 del_page_from_free_list(page, zone, order);
3531
3532
3533
3534
3535
3536 if (order >= pageblock_order - 1) {
3537 struct page *endpage = page + (1 << order) - 1;
3538 for (; page < endpage; page += pageblock_nr_pages) {
3539 int mt = get_pageblock_migratetype(page);
3540 if (!is_migrate_isolate(mt) && !is_migrate_cma(mt)
3541 && !is_migrate_highatomic(mt))
3542 set_pageblock_migratetype(page,
3543 MIGRATE_MOVABLE);
3544 }
3545 }
3546
3547
3548 return 1UL << order;
3549}
3550
3551
3552
3553
3554
3555
3556
3557
3558
3559
3560void __putback_isolated_page(struct page *page, unsigned int order, int mt)
3561{
3562 struct zone *zone = page_zone(page);
3563
3564
3565 lockdep_assert_held(&zone->lock);
3566
3567
3568 __free_one_page(page, page_to_pfn(page), zone, order, mt,
3569 FPI_SKIP_REPORT_NOTIFY | FPI_TO_TAIL);
3570}
3571
3572
3573
3574
3575
3576
3577static inline void zone_statistics(struct zone *preferred_zone, struct zone *z,
3578 long nr_account)
3579{
3580#ifdef CONFIG_NUMA
3581 enum numa_stat_item local_stat = NUMA_LOCAL;
3582
3583
3584 if (!static_branch_likely(&vm_numa_stat_key))
3585 return;
3586
3587 if (zone_to_nid(z) != numa_node_id())
3588 local_stat = NUMA_OTHER;
3589
3590 if (zone_to_nid(z) == zone_to_nid(preferred_zone))
3591 __count_numa_events(z, NUMA_HIT, nr_account);
3592 else {
3593 __count_numa_events(z, NUMA_MISS, nr_account);
3594 __count_numa_events(preferred_zone, NUMA_FOREIGN, nr_account);
3595 }
3596 __count_numa_events(z, local_stat, nr_account);
3597#endif
3598}
3599
3600
3601static inline
3602struct page *__rmqueue_pcplist(struct zone *zone, unsigned int order,
3603 int migratetype,
3604 unsigned int alloc_flags,
3605 struct per_cpu_pages *pcp,
3606 struct list_head *list)
3607{
3608 struct page *page;
3609
3610 do {
3611 if (list_empty(list)) {
3612 int batch = READ_ONCE(pcp->batch);
3613 int alloced;
3614
3615
3616
3617
3618
3619
3620
3621
3622 if (batch > 1)
3623 batch = max(batch >> order, 2);
3624 alloced = rmqueue_bulk(zone, order,
3625 batch, list,
3626 migratetype, alloc_flags);
3627
3628 pcp->count += alloced << order;
3629 if (unlikely(list_empty(list)))
3630 return NULL;
3631 }
3632
3633 page = list_first_entry(list, struct page, lru);
3634 list_del(&page->lru);
3635 pcp->count -= 1 << order;
3636 } while (check_new_pcp(page));
3637
3638 return page;
3639}
3640
3641
3642static struct page *rmqueue_pcplist(struct zone *preferred_zone,
3643 struct zone *zone, unsigned int order,
3644 gfp_t gfp_flags, int migratetype,
3645 unsigned int alloc_flags)
3646{
3647 struct per_cpu_pages *pcp;
3648 struct list_head *list;
3649 struct page *page;
3650 unsigned long flags;
3651
3652 local_lock_irqsave(&pagesets.lock, flags);
3653
3654
3655
3656
3657
3658
3659 pcp = this_cpu_ptr(zone->per_cpu_pageset);
3660 pcp->free_factor >>= 1;
3661 list = &pcp->lists[order_to_pindex(migratetype, order)];
3662 page = __rmqueue_pcplist(zone, order, migratetype, alloc_flags, pcp, list);
3663 local_unlock_irqrestore(&pagesets.lock, flags);
3664 if (page) {
3665 __count_zid_vm_events(PGALLOC, page_zonenum(page), 1);
3666 zone_statistics(preferred_zone, zone, 1);
3667 }
3668 return page;
3669}
3670
3671
3672
3673
3674static inline
3675struct page *rmqueue(struct zone *preferred_zone,
3676 struct zone *zone, unsigned int order,
3677 gfp_t gfp_flags, unsigned int alloc_flags,
3678 int migratetype)
3679{
3680 unsigned long flags;
3681 struct page *page;
3682
3683 if (likely(pcp_allowed_order(order))) {
3684
3685
3686
3687
3688 if (!IS_ENABLED(CONFIG_CMA) || alloc_flags & ALLOC_CMA ||
3689 migratetype != MIGRATE_MOVABLE) {
3690 page = rmqueue_pcplist(preferred_zone, zone, order,
3691 gfp_flags, migratetype, alloc_flags);
3692 goto out;
3693 }
3694 }
3695
3696
3697
3698
3699
3700 WARN_ON_ONCE((gfp_flags & __GFP_NOFAIL) && (order > 1));
3701 spin_lock_irqsave(&zone->lock, flags);
3702
3703 do {
3704 page = NULL;
3705
3706
3707
3708
3709
3710
3711 if (order > 0 && alloc_flags & ALLOC_HARDER) {
3712 page = __rmqueue_smallest(zone, order, MIGRATE_HIGHATOMIC);
3713 if (page)
3714 trace_mm_page_alloc_zone_locked(page, order, migratetype);
3715 }
3716 if (!page)
3717 page = __rmqueue(zone, order, migratetype, alloc_flags);
3718 } while (page && check_new_pages(page, order));
3719 if (!page)
3720 goto failed;
3721
3722 __mod_zone_freepage_state(zone, -(1 << order),
3723 get_pcppage_migratetype(page));
3724 spin_unlock_irqrestore(&zone->lock, flags);
3725
3726 __count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order);
3727 zone_statistics(preferred_zone, zone, 1);
3728
3729out:
3730
3731 if (test_bit(ZONE_BOOSTED_WATERMARK, &zone->flags)) {
3732 clear_bit(ZONE_BOOSTED_WATERMARK, &zone->flags);
3733 wakeup_kswapd(zone, 0, 0, zone_idx(zone));
3734 }
3735
3736 VM_BUG_ON_PAGE(page && bad_range(zone, page), page);
3737 return page;
3738
3739failed:
3740 spin_unlock_irqrestore(&zone->lock, flags);
3741 return NULL;
3742}
3743
3744#ifdef CONFIG_FAIL_PAGE_ALLOC
3745
3746static struct {
3747 struct fault_attr attr;
3748
3749 bool ignore_gfp_highmem;
3750 bool ignore_gfp_reclaim;
3751 u32 min_order;
3752} fail_page_alloc = {
3753 .attr = FAULT_ATTR_INITIALIZER,
3754 .ignore_gfp_reclaim = true,
3755 .ignore_gfp_highmem = true,
3756 .min_order = 1,
3757};
3758
3759static int __init setup_fail_page_alloc(char *str)
3760{
3761 return setup_fault_attr(&fail_page_alloc.attr, str);
3762}
3763__setup("fail_page_alloc=", setup_fail_page_alloc);
3764
3765static bool __should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
3766{
3767 if (order < fail_page_alloc.min_order)
3768 return false;
3769 if (gfp_mask & __GFP_NOFAIL)
3770 return false;
3771 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
3772 return false;
3773 if (fail_page_alloc.ignore_gfp_reclaim &&
3774 (gfp_mask & __GFP_DIRECT_RECLAIM))
3775 return false;
3776
3777 return should_fail(&fail_page_alloc.attr, 1 << order);
3778}
3779
3780#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
3781
3782static int __init fail_page_alloc_debugfs(void)
3783{
3784 umode_t mode = S_IFREG | 0600;
3785 struct dentry *dir;
3786
3787 dir = fault_create_debugfs_attr("fail_page_alloc", NULL,
3788 &fail_page_alloc.attr);
3789
3790 debugfs_create_bool("ignore-gfp-wait", mode, dir,
3791 &fail_page_alloc.ignore_gfp_reclaim);
3792 debugfs_create_bool("ignore-gfp-highmem", mode, dir,
3793 &fail_page_alloc.ignore_gfp_highmem);
3794 debugfs_create_u32("min-order", mode, dir, &fail_page_alloc.min_order);
3795
3796 return 0;
3797}
3798
3799late_initcall(fail_page_alloc_debugfs);
3800
3801#endif
3802
3803#else
3804
3805static inline bool __should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
3806{
3807 return false;
3808}
3809
3810#endif
3811
3812noinline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
3813{
3814 return __should_fail_alloc_page(gfp_mask, order);
3815}
3816ALLOW_ERROR_INJECTION(should_fail_alloc_page, TRUE);
3817
3818static inline long __zone_watermark_unusable_free(struct zone *z,
3819 unsigned int order, unsigned int alloc_flags)
3820{
3821 const bool alloc_harder = (alloc_flags & (ALLOC_HARDER|ALLOC_OOM));
3822 long unusable_free = (1 << order) - 1;
3823
3824
3825
3826
3827
3828
3829 if (likely(!alloc_harder))
3830 unusable_free += z->nr_reserved_highatomic;
3831
3832#ifdef CONFIG_CMA
3833
3834 if (!(alloc_flags & ALLOC_CMA))
3835 unusable_free += zone_page_state(z, NR_FREE_CMA_PAGES);
3836#endif
3837
3838 return unusable_free;
3839}
3840
3841
3842
3843
3844
3845
3846
3847bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
3848 int highest_zoneidx, unsigned int alloc_flags,
3849 long free_pages)
3850{
3851 long min = mark;
3852 int o;
3853 const bool alloc_harder = (alloc_flags & (ALLOC_HARDER|ALLOC_OOM));
3854
3855
3856 free_pages -= __zone_watermark_unusable_free(z, order, alloc_flags);
3857
3858 if (alloc_flags & ALLOC_HIGH)
3859 min -= min / 2;
3860
3861 if (unlikely(alloc_harder)) {
3862
3863
3864
3865
3866
3867
3868 if (alloc_flags & ALLOC_OOM)
3869 min -= min / 2;
3870 else
3871 min -= min / 4;
3872 }
3873
3874
3875
3876
3877
3878
3879 if (free_pages <= min + z->lowmem_reserve[highest_zoneidx])
3880 return false;
3881
3882
3883 if (!order)
3884 return true;
3885
3886
3887 for (o = order; o < MAX_ORDER; o++) {
3888 struct free_area *area = &z->free_area[o];
3889 int mt;
3890
3891 if (!area->nr_free)
3892 continue;
3893
3894 for (mt = 0; mt < MIGRATE_PCPTYPES; mt++) {
3895 if (!free_area_empty(area, mt))
3896 return true;
3897 }
3898
3899#ifdef CONFIG_CMA
3900 if ((alloc_flags & ALLOC_CMA) &&
3901 !free_area_empty(area, MIGRATE_CMA)) {
3902 return true;
3903 }
3904#endif
3905 if (alloc_harder && !free_area_empty(area, MIGRATE_HIGHATOMIC))
3906 return true;
3907 }
3908 return false;
3909}
3910
3911bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
3912 int highest_zoneidx, unsigned int alloc_flags)
3913{
3914 return __zone_watermark_ok(z, order, mark, highest_zoneidx, alloc_flags,
3915 zone_page_state(z, NR_FREE_PAGES));
3916}
3917
3918static inline bool zone_watermark_fast(struct zone *z, unsigned int order,
3919 unsigned long mark, int highest_zoneidx,
3920 unsigned int alloc_flags, gfp_t gfp_mask)
3921{
3922 long free_pages;
3923
3924 free_pages = zone_page_state(z, NR_FREE_PAGES);
3925
3926
3927
3928
3929
3930 if (!order) {
3931 long fast_free;
3932
3933 fast_free = free_pages;
3934 fast_free -= __zone_watermark_unusable_free(z, 0, alloc_flags);
3935 if (fast_free > mark + z->lowmem_reserve[highest_zoneidx])
3936 return true;
3937 }
3938
3939 if (__zone_watermark_ok(z, order, mark, highest_zoneidx, alloc_flags,
3940 free_pages))
3941 return true;
3942
3943
3944
3945
3946
3947
3948 if (unlikely(!order && (gfp_mask & __GFP_ATOMIC) && z->watermark_boost
3949 && ((alloc_flags & ALLOC_WMARK_MASK) == WMARK_MIN))) {
3950 mark = z->_watermark[WMARK_MIN];
3951 return __zone_watermark_ok(z, order, mark, highest_zoneidx,
3952 alloc_flags, free_pages);
3953 }
3954
3955 return false;
3956}
3957
3958bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
3959 unsigned long mark, int highest_zoneidx)
3960{
3961 long free_pages = zone_page_state(z, NR_FREE_PAGES);
3962
3963 if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark)
3964 free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES);
3965
3966 return __zone_watermark_ok(z, order, mark, highest_zoneidx, 0,
3967 free_pages);
3968}
3969
3970#ifdef CONFIG_NUMA
3971static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
3972{
3973 return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) <=
3974 node_reclaim_distance;
3975}
3976#else
3977static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
3978{
3979 return true;
3980}
3981#endif
3982
3983
3984
3985
3986
3987
3988
3989
3990
3991static inline unsigned int
3992alloc_flags_nofragment(struct zone *zone, gfp_t gfp_mask)
3993{
3994 unsigned int alloc_flags;
3995
3996
3997
3998
3999
4000 alloc_flags = (__force int) (gfp_mask & __GFP_KSWAPD_RECLAIM);
4001
4002#ifdef CONFIG_ZONE_DMA32
4003 if (!zone)
4004 return alloc_flags;
4005
4006 if (zone_idx(zone) != ZONE_NORMAL)
4007 return alloc_flags;
4008
4009
4010
4011
4012
4013
4014 BUILD_BUG_ON(ZONE_NORMAL - ZONE_DMA32 != 1);
4015 if (nr_online_nodes > 1 && !populated_zone(--zone))
4016 return alloc_flags;
4017
4018 alloc_flags |= ALLOC_NOFRAGMENT;
4019#endif
4020 return alloc_flags;
4021}
4022
4023
4024static inline unsigned int gfp_to_alloc_flags_cma(gfp_t gfp_mask,
4025 unsigned int alloc_flags)
4026{
4027#ifdef CONFIG_CMA
4028 if (gfp_migratetype(gfp_mask) == MIGRATE_MOVABLE)
4029 alloc_flags |= ALLOC_CMA;
4030#endif
4031 return alloc_flags;
4032}
4033
4034
4035
4036
4037
4038static struct page *
4039get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags,
4040 const struct alloc_context *ac)
4041{
4042 struct zoneref *z;
4043 struct zone *zone;
4044 struct pglist_data *last_pgdat_dirty_limit = NULL;
4045 bool no_fallback;
4046
4047retry:
4048
4049
4050
4051
4052 no_fallback = alloc_flags & ALLOC_NOFRAGMENT;
4053 z = ac->preferred_zoneref;
4054 for_next_zone_zonelist_nodemask(zone, z, ac->highest_zoneidx,
4055 ac->nodemask) {
4056 struct page *page;
4057 unsigned long mark;
4058
4059 if (cpusets_enabled() &&
4060 (alloc_flags & ALLOC_CPUSET) &&
4061 !__cpuset_zone_allowed(zone, gfp_mask))
4062 continue;
4063
4064
4065
4066
4067
4068
4069
4070
4071
4072
4073
4074
4075
4076
4077
4078
4079
4080
4081
4082 if (ac->spread_dirty_pages) {
4083 if (last_pgdat_dirty_limit == zone->zone_pgdat)
4084 continue;
4085
4086 if (!node_dirty_ok(zone->zone_pgdat)) {
4087 last_pgdat_dirty_limit = zone->zone_pgdat;
4088 continue;
4089 }
4090 }
4091
4092 if (no_fallback && nr_online_nodes > 1 &&
4093 zone != ac->preferred_zoneref->zone) {
4094 int local_nid;
4095
4096
4097
4098
4099
4100
4101 local_nid = zone_to_nid(ac->preferred_zoneref->zone);
4102 if (zone_to_nid(zone) != local_nid) {
4103 alloc_flags &= ~ALLOC_NOFRAGMENT;
4104 goto retry;
4105 }
4106 }
4107
4108 mark = wmark_pages(zone, alloc_flags & ALLOC_WMARK_MASK);
4109 if (!zone_watermark_fast(zone, order, mark,
4110 ac->highest_zoneidx, alloc_flags,
4111 gfp_mask)) {
4112 int ret;
4113
4114#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
4115
4116
4117
4118
4119 if (static_branch_unlikely(&deferred_pages)) {
4120 if (_deferred_grow_zone(zone, order))
4121 goto try_this_zone;
4122 }
4123#endif
4124
4125 BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
4126 if (alloc_flags & ALLOC_NO_WATERMARKS)
4127 goto try_this_zone;
4128
4129 if (!node_reclaim_enabled() ||
4130 !zone_allows_reclaim(ac->preferred_zoneref->zone, zone))
4131 continue;
4132
4133 ret = node_reclaim(zone->zone_pgdat, gfp_mask, order);
4134 switch (ret) {
4135 case NODE_RECLAIM_NOSCAN:
4136
4137 continue;
4138 case NODE_RECLAIM_FULL:
4139
4140 continue;
4141 default:
4142
4143 if (zone_watermark_ok(zone, order, mark,
4144 ac->highest_zoneidx, alloc_flags))
4145 goto try_this_zone;
4146
4147 continue;
4148 }
4149 }
4150
4151try_this_zone:
4152 page = rmqueue(ac->preferred_zoneref->zone, zone, order,
4153 gfp_mask, alloc_flags, ac->migratetype);
4154 if (page) {
4155 prep_new_page(page, order, gfp_mask, alloc_flags);
4156
4157
4158
4159
4160
4161 if (unlikely(order && (alloc_flags & ALLOC_HARDER)))
4162 reserve_highatomic_pageblock(page, zone, order);
4163
4164 return page;
4165 } else {
4166#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
4167
4168 if (static_branch_unlikely(&deferred_pages)) {
4169 if (_deferred_grow_zone(zone, order))
4170 goto try_this_zone;
4171 }
4172#endif
4173 }
4174 }
4175
4176
4177
4178
4179
4180 if (no_fallback) {
4181 alloc_flags &= ~ALLOC_NOFRAGMENT;
4182 goto retry;
4183 }
4184
4185 return NULL;
4186}
4187
4188static void warn_alloc_show_mem(gfp_t gfp_mask, nodemask_t *nodemask)
4189{
4190 unsigned int filter = SHOW_MEM_FILTER_NODES;
4191
4192
4193
4194
4195
4196
4197 if (!(gfp_mask & __GFP_NOMEMALLOC))
4198 if (tsk_is_oom_victim(current) ||
4199 (current->flags & (PF_MEMALLOC | PF_EXITING)))
4200 filter &= ~SHOW_MEM_FILTER_NODES;
4201 if (!in_task() || !(gfp_mask & __GFP_DIRECT_RECLAIM))
4202 filter &= ~SHOW_MEM_FILTER_NODES;
4203
4204 show_mem(filter, nodemask);
4205}
4206
4207void warn_alloc(gfp_t gfp_mask, nodemask_t *nodemask, const char *fmt, ...)
4208{
4209 struct va_format vaf;
4210 va_list args;
4211 static DEFINE_RATELIMIT_STATE(nopage_rs, 10*HZ, 1);
4212
4213 if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs))
4214 return;
4215
4216 va_start(args, fmt);
4217 vaf.fmt = fmt;
4218 vaf.va = &args;
4219 pr_warn("%s: %pV, mode:%#x(%pGg), nodemask=%*pbl",
4220 current->comm, &vaf, gfp_mask, &gfp_mask,
4221 nodemask_pr_args(nodemask));
4222 va_end(args);
4223
4224 cpuset_print_current_mems_allowed();
4225 pr_cont("\n");
4226 dump_stack();
4227 warn_alloc_show_mem(gfp_mask, nodemask);
4228}
4229
4230static inline struct page *
4231__alloc_pages_cpuset_fallback(gfp_t gfp_mask, unsigned int order,
4232 unsigned int alloc_flags,
4233 const struct alloc_context *ac)
4234{
4235 struct page *page;
4236
4237 page = get_page_from_freelist(gfp_mask, order,
4238 alloc_flags|ALLOC_CPUSET, ac);
4239
4240
4241
4242
4243 if (!page)
4244 page = get_page_from_freelist(gfp_mask, order,
4245 alloc_flags, ac);
4246
4247 return page;
4248}
4249
4250static inline struct page *
4251__alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
4252 const struct alloc_context *ac, unsigned long *did_some_progress)
4253{
4254 struct oom_control oc = {
4255 .zonelist = ac->zonelist,
4256 .nodemask = ac->nodemask,
4257 .memcg = NULL,
4258 .gfp_mask = gfp_mask,
4259 .order = order,
4260 };
4261 struct page *page;
4262
4263 *did_some_progress = 0;
4264
4265
4266
4267
4268
4269 if (!mutex_trylock(&oom_lock)) {
4270 *did_some_progress = 1;
4271 schedule_timeout_uninterruptible(1);
4272 return NULL;
4273 }
4274
4275
4276
4277
4278
4279
4280
4281
4282 page = get_page_from_freelist((gfp_mask | __GFP_HARDWALL) &
4283 ~__GFP_DIRECT_RECLAIM, order,
4284 ALLOC_WMARK_HIGH|ALLOC_CPUSET, ac);
4285 if (page)
4286 goto out;
4287
4288
4289 if (current->flags & PF_DUMPCORE)
4290 goto out;
4291
4292 if (order > PAGE_ALLOC_COSTLY_ORDER)
4293 goto out;
4294
4295
4296
4297
4298
4299
4300
4301
4302 if (gfp_mask & (__GFP_RETRY_MAYFAIL | __GFP_THISNODE))
4303 goto out;
4304
4305 if (ac->highest_zoneidx < ZONE_NORMAL)
4306 goto out;
4307 if (pm_suspended_storage())
4308 goto out;
4309
4310
4311
4312
4313
4314
4315
4316
4317
4318
4319
4320 if (out_of_memory(&oc) || WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) {
4321 *did_some_progress = 1;
4322
4323
4324
4325
4326
4327 if (gfp_mask & __GFP_NOFAIL)
4328 page = __alloc_pages_cpuset_fallback(gfp_mask, order,
4329 ALLOC_NO_WATERMARKS, ac);
4330 }
4331out:
4332 mutex_unlock(&oom_lock);
4333 return page;
4334}
4335
4336
4337
4338
4339
4340#define MAX_COMPACT_RETRIES 16
4341
4342#ifdef CONFIG_COMPACTION
4343
4344static struct page *
4345__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
4346 unsigned int alloc_flags, const struct alloc_context *ac,
4347 enum compact_priority prio, enum compact_result *compact_result)
4348{
4349 struct page *page = NULL;
4350 unsigned long pflags;
4351 unsigned int noreclaim_flag;
4352
4353 if (!order)
4354 return NULL;
4355
4356 psi_memstall_enter(&pflags);
4357 noreclaim_flag = memalloc_noreclaim_save();
4358
4359 *compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac,
4360 prio, &page);
4361
4362 memalloc_noreclaim_restore(noreclaim_flag);
4363 psi_memstall_leave(&pflags);
4364
4365 if (*compact_result == COMPACT_SKIPPED)
4366 return NULL;
4367
4368
4369
4370
4371 count_vm_event(COMPACTSTALL);
4372
4373
4374 if (page)
4375 prep_new_page(page, order, gfp_mask, alloc_flags);
4376
4377
4378 if (!page)
4379 page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
4380
4381 if (page) {
4382 struct zone *zone = page_zone(page);
4383
4384 zone->compact_blockskip_flush = false;
4385 compaction_defer_reset(zone, order, true);
4386 count_vm_event(COMPACTSUCCESS);
4387 return page;
4388 }
4389
4390
4391
4392
4393
4394 count_vm_event(COMPACTFAIL);
4395
4396 cond_resched();
4397
4398 return NULL;
4399}
4400
4401static inline bool
4402should_compact_retry(struct alloc_context *ac, int order, int alloc_flags,
4403 enum compact_result compact_result,
4404 enum compact_priority *compact_priority,
4405 int *compaction_retries)
4406{
4407 int max_retries = MAX_COMPACT_RETRIES;
4408 int min_priority;
4409 bool ret = false;
4410 int retries = *compaction_retries;
4411 enum compact_priority priority = *compact_priority;
4412
4413 if (!order)
4414 return false;
4415
4416 if (fatal_signal_pending(current))
4417 return false;
4418
4419 if (compaction_made_progress(compact_result))
4420 (*compaction_retries)++;
4421
4422
4423
4424
4425
4426
4427 if (compaction_failed(compact_result))
4428 goto check_priority;
4429
4430
4431
4432
4433
4434 if (compaction_needs_reclaim(compact_result)) {
4435 ret = compaction_zonelist_suitable(ac, order, alloc_flags);
4436 goto out;
4437 }
4438
4439
4440
4441
4442
4443
4444
4445 if (compaction_withdrawn(compact_result)) {
4446 goto check_priority;
4447 }
4448
4449
4450
4451
4452
4453
4454
4455
4456
4457 if (order > PAGE_ALLOC_COSTLY_ORDER)
4458 max_retries /= 4;
4459 if (*compaction_retries <= max_retries) {
4460 ret = true;
4461 goto out;
4462 }
4463
4464
4465
4466
4467
4468check_priority:
4469 min_priority = (order > PAGE_ALLOC_COSTLY_ORDER) ?
4470 MIN_COMPACT_COSTLY_PRIORITY : MIN_COMPACT_PRIORITY;
4471
4472 if (*compact_priority > min_priority) {
4473 (*compact_priority)--;
4474 *compaction_retries = 0;
4475 ret = true;
4476 }
4477out:
4478 trace_compact_retry(order, priority, compact_result, retries, max_retries, ret);
4479 return ret;
4480}
4481#else
4482static inline struct page *
4483__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
4484 unsigned int alloc_flags, const struct alloc_context *ac,
4485 enum compact_priority prio, enum compact_result *compact_result)
4486{
4487 *compact_result = COMPACT_SKIPPED;
4488 return NULL;
4489}
4490
4491static inline bool
4492should_compact_retry(struct alloc_context *ac, unsigned int order, int alloc_flags,
4493 enum compact_result compact_result,
4494 enum compact_priority *compact_priority,
4495 int *compaction_retries)
4496{
4497 struct zone *zone;
4498 struct zoneref *z;
4499
4500 if (!order || order > PAGE_ALLOC_COSTLY_ORDER)
4501 return false;
4502
4503
4504
4505
4506
4507
4508
4509 for_each_zone_zonelist_nodemask(zone, z, ac->zonelist,
4510 ac->highest_zoneidx, ac->nodemask) {
4511 if (zone_watermark_ok(zone, 0, min_wmark_pages(zone),
4512 ac->highest_zoneidx, alloc_flags))
4513 return true;
4514 }
4515 return false;
4516}
4517#endif
4518
4519#ifdef CONFIG_LOCKDEP
4520static struct lockdep_map __fs_reclaim_map =
4521 STATIC_LOCKDEP_MAP_INIT("fs_reclaim", &__fs_reclaim_map);
4522
4523static bool __need_reclaim(gfp_t gfp_mask)
4524{
4525
4526 if (!(gfp_mask & __GFP_DIRECT_RECLAIM))
4527 return false;
4528
4529
4530 if (current->flags & PF_MEMALLOC)
4531 return false;
4532
4533 if (gfp_mask & __GFP_NOLOCKDEP)
4534 return false;
4535
4536 return true;
4537}
4538
4539void __fs_reclaim_acquire(unsigned long ip)
4540{
4541 lock_acquire_exclusive(&__fs_reclaim_map, 0, 0, NULL, ip);
4542}
4543
4544void __fs_reclaim_release(unsigned long ip)
4545{
4546 lock_release(&__fs_reclaim_map, ip);
4547}
4548
4549void fs_reclaim_acquire(gfp_t gfp_mask)
4550{
4551 gfp_mask = current_gfp_context(gfp_mask);
4552
4553 if (__need_reclaim(gfp_mask)) {
4554 if (gfp_mask & __GFP_FS)
4555 __fs_reclaim_acquire(_RET_IP_);
4556
4557#ifdef CONFIG_MMU_NOTIFIER
4558 lock_map_acquire(&__mmu_notifier_invalidate_range_start_map);
4559 lock_map_release(&__mmu_notifier_invalidate_range_start_map);
4560#endif
4561
4562 }
4563}
4564EXPORT_SYMBOL_GPL(fs_reclaim_acquire);
4565
4566void fs_reclaim_release(gfp_t gfp_mask)
4567{
4568 gfp_mask = current_gfp_context(gfp_mask);
4569
4570 if (__need_reclaim(gfp_mask)) {
4571 if (gfp_mask & __GFP_FS)
4572 __fs_reclaim_release(_RET_IP_);
4573 }
4574}
4575EXPORT_SYMBOL_GPL(fs_reclaim_release);
4576#endif
4577
4578
4579static unsigned long
4580__perform_reclaim(gfp_t gfp_mask, unsigned int order,
4581 const struct alloc_context *ac)
4582{
4583 unsigned int noreclaim_flag;
4584 unsigned long pflags, progress;
4585
4586 cond_resched();
4587
4588
4589 cpuset_memory_pressure_bump();
4590 psi_memstall_enter(&pflags);
4591 fs_reclaim_acquire(gfp_mask);
4592 noreclaim_flag = memalloc_noreclaim_save();
4593
4594 progress = try_to_free_pages(ac->zonelist, order, gfp_mask,
4595 ac->nodemask);
4596
4597 memalloc_noreclaim_restore(noreclaim_flag);
4598 fs_reclaim_release(gfp_mask);
4599 psi_memstall_leave(&pflags);
4600
4601 cond_resched();
4602
4603 return progress;
4604}
4605
4606
4607static inline struct page *
4608__alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
4609 unsigned int alloc_flags, const struct alloc_context *ac,
4610 unsigned long *did_some_progress)
4611{
4612 struct page *page = NULL;
4613 bool drained = false;
4614
4615 *did_some_progress = __perform_reclaim(gfp_mask, order, ac);
4616 if (unlikely(!(*did_some_progress)))
4617 return NULL;
4618
4619retry:
4620 page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
4621
4622
4623
4624
4625
4626
4627 if (!page && !drained) {
4628 unreserve_highatomic_pageblock(ac, false);
4629 drain_all_pages(NULL);
4630 drained = true;
4631 goto retry;
4632 }
4633
4634 return page;
4635}
4636
4637static void wake_all_kswapds(unsigned int order, gfp_t gfp_mask,
4638 const struct alloc_context *ac)
4639{
4640 struct zoneref *z;
4641 struct zone *zone;
4642 pg_data_t *last_pgdat = NULL;
4643 enum zone_type highest_zoneidx = ac->highest_zoneidx;
4644
4645 for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, highest_zoneidx,
4646 ac->nodemask) {
4647 if (last_pgdat != zone->zone_pgdat)
4648 wakeup_kswapd(zone, gfp_mask, order, highest_zoneidx);
4649 last_pgdat = zone->zone_pgdat;
4650 }
4651}
4652
4653static inline unsigned int
4654gfp_to_alloc_flags(gfp_t gfp_mask)
4655{
4656 unsigned int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
4657
4658
4659
4660
4661
4662
4663 BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
4664 BUILD_BUG_ON(__GFP_KSWAPD_RECLAIM != (__force gfp_t) ALLOC_KSWAPD);
4665
4666
4667
4668
4669
4670
4671
4672 alloc_flags |= (__force int)
4673 (gfp_mask & (__GFP_HIGH | __GFP_KSWAPD_RECLAIM));
4674
4675 if (gfp_mask & __GFP_ATOMIC) {
4676
4677
4678
4679
4680 if (!(gfp_mask & __GFP_NOMEMALLOC))
4681 alloc_flags |= ALLOC_HARDER;
4682
4683
4684
4685
4686 alloc_flags &= ~ALLOC_CPUSET;
4687 } else if (unlikely(rt_task(current)) && in_task())
4688 alloc_flags |= ALLOC_HARDER;
4689
4690 alloc_flags = gfp_to_alloc_flags_cma(gfp_mask, alloc_flags);
4691
4692 return alloc_flags;
4693}
4694
4695static bool oom_reserves_allowed(struct task_struct *tsk)
4696{
4697 if (!tsk_is_oom_victim(tsk))
4698 return false;
4699
4700
4701
4702
4703
4704 if (!IS_ENABLED(CONFIG_MMU) && !test_thread_flag(TIF_MEMDIE))
4705 return false;
4706
4707 return true;
4708}
4709
4710
4711
4712
4713
4714static inline int __gfp_pfmemalloc_flags(gfp_t gfp_mask)
4715{
4716 if (unlikely(gfp_mask & __GFP_NOMEMALLOC))
4717 return 0;
4718 if (gfp_mask & __GFP_MEMALLOC)
4719 return ALLOC_NO_WATERMARKS;
4720 if (in_serving_softirq() && (current->flags & PF_MEMALLOC))
4721 return ALLOC_NO_WATERMARKS;
4722 if (!in_interrupt()) {
4723 if (current->flags & PF_MEMALLOC)
4724 return ALLOC_NO_WATERMARKS;
4725 else if (oom_reserves_allowed(current))
4726 return ALLOC_OOM;
4727 }
4728
4729 return 0;
4730}
4731
4732bool gfp_pfmemalloc_allowed(gfp_t gfp_mask)
4733{
4734 return !!__gfp_pfmemalloc_flags(gfp_mask);
4735}
4736
4737
4738
4739
4740
4741
4742
4743
4744
4745
4746
4747static inline bool
4748should_reclaim_retry(gfp_t gfp_mask, unsigned order,
4749 struct alloc_context *ac, int alloc_flags,
4750 bool did_some_progress, int *no_progress_loops)
4751{
4752 struct zone *zone;
4753 struct zoneref *z;
4754 bool ret = false;
4755
4756
4757
4758
4759
4760
4761 if (did_some_progress && order <= PAGE_ALLOC_COSTLY_ORDER)
4762 *no_progress_loops = 0;
4763 else
4764 (*no_progress_loops)++;
4765
4766
4767
4768
4769
4770 if (*no_progress_loops > MAX_RECLAIM_RETRIES) {
4771
4772 return unreserve_highatomic_pageblock(ac, true);
4773 }
4774
4775
4776
4777
4778
4779
4780
4781 for_each_zone_zonelist_nodemask(zone, z, ac->zonelist,
4782 ac->highest_zoneidx, ac->nodemask) {
4783 unsigned long available;
4784 unsigned long reclaimable;
4785 unsigned long min_wmark = min_wmark_pages(zone);
4786 bool wmark;
4787
4788 available = reclaimable = zone_reclaimable_pages(zone);
4789 available += zone_page_state_snapshot(zone, NR_FREE_PAGES);
4790
4791
4792
4793
4794
4795 wmark = __zone_watermark_ok(zone, order, min_wmark,
4796 ac->highest_zoneidx, alloc_flags, available);
4797 trace_reclaim_retry_zone(z, order, reclaimable,
4798 available, min_wmark, *no_progress_loops, wmark);
4799 if (wmark) {
4800
4801
4802
4803
4804
4805
4806 if (!did_some_progress) {
4807 unsigned long write_pending;
4808
4809 write_pending = zone_page_state_snapshot(zone,
4810 NR_ZONE_WRITE_PENDING);
4811
4812 if (2 * write_pending > reclaimable) {
4813 congestion_wait(BLK_RW_ASYNC, HZ/10);
4814 return true;
4815 }
4816 }
4817
4818 ret = true;
4819 goto out;
4820 }
4821 }
4822
4823out:
4824
4825
4826
4827
4828
4829
4830
4831 if (current->flags & PF_WQ_WORKER)
4832 schedule_timeout_uninterruptible(1);
4833 else
4834 cond_resched();
4835 return ret;
4836}
4837
4838static inline bool
4839check_retry_cpuset(int cpuset_mems_cookie, struct alloc_context *ac)
4840{
4841
4842
4843
4844
4845
4846
4847
4848
4849
4850
4851
4852 if (cpusets_enabled() && ac->nodemask &&
4853 !cpuset_nodemask_valid_mems_allowed(ac->nodemask)) {
4854 ac->nodemask = NULL;
4855 return true;
4856 }
4857
4858
4859
4860
4861
4862
4863
4864
4865 if (read_mems_allowed_retry(cpuset_mems_cookie))
4866 return true;
4867
4868 return false;
4869}
4870
4871static inline struct page *
4872__alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
4873 struct alloc_context *ac)
4874{
4875 bool can_direct_reclaim = gfp_mask & __GFP_DIRECT_RECLAIM;
4876 const bool costly_order = order > PAGE_ALLOC_COSTLY_ORDER;
4877 struct page *page = NULL;
4878 unsigned int alloc_flags;
4879 unsigned long did_some_progress;
4880 enum compact_priority compact_priority;
4881 enum compact_result compact_result;
4882 int compaction_retries;
4883 int no_progress_loops;
4884 unsigned int cpuset_mems_cookie;
4885 int reserve_flags;
4886
4887
4888
4889
4890
4891 if (WARN_ON_ONCE((gfp_mask & (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)) ==
4892 (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)))
4893 gfp_mask &= ~__GFP_ATOMIC;
4894
4895retry_cpuset:
4896 compaction_retries = 0;
4897 no_progress_loops = 0;
4898 compact_priority = DEF_COMPACT_PRIORITY;
4899 cpuset_mems_cookie = read_mems_allowed_begin();
4900
4901
4902
4903
4904
4905
4906 alloc_flags = gfp_to_alloc_flags(gfp_mask);
4907
4908
4909
4910
4911
4912
4913
4914 ac->preferred_zoneref = first_zones_zonelist(ac->zonelist,
4915 ac->highest_zoneidx, ac->nodemask);
4916 if (!ac->preferred_zoneref->zone)
4917 goto nopage;
4918
4919 if (alloc_flags & ALLOC_KSWAPD)
4920 wake_all_kswapds(order, gfp_mask, ac);
4921
4922
4923
4924
4925
4926 page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
4927 if (page)
4928 goto got_pg;
4929
4930
4931
4932
4933
4934
4935
4936
4937
4938
4939 if (can_direct_reclaim &&
4940 (costly_order ||
4941 (order > 0 && ac->migratetype != MIGRATE_MOVABLE))
4942 && !gfp_pfmemalloc_allowed(gfp_mask)) {
4943 page = __alloc_pages_direct_compact(gfp_mask, order,
4944 alloc_flags, ac,
4945 INIT_COMPACT_PRIORITY,
4946 &compact_result);
4947 if (page)
4948 goto got_pg;
4949
4950
4951
4952
4953
4954 if (costly_order && (gfp_mask & __GFP_NORETRY)) {
4955
4956
4957
4958
4959
4960
4961
4962
4963
4964
4965
4966
4967
4968
4969
4970
4971
4972 if (compact_result == COMPACT_SKIPPED ||
4973 compact_result == COMPACT_DEFERRED)
4974 goto nopage;
4975
4976
4977
4978
4979
4980
4981 compact_priority = INIT_COMPACT_PRIORITY;
4982 }
4983 }
4984
4985retry:
4986
4987 if (alloc_flags & ALLOC_KSWAPD)
4988 wake_all_kswapds(order, gfp_mask, ac);
4989
4990 reserve_flags = __gfp_pfmemalloc_flags(gfp_mask);
4991 if (reserve_flags)
4992 alloc_flags = gfp_to_alloc_flags_cma(gfp_mask, reserve_flags);
4993
4994
4995
4996
4997
4998
4999 if (!(alloc_flags & ALLOC_CPUSET) || reserve_flags) {
5000 ac->nodemask = NULL;
5001 ac->preferred_zoneref = first_zones_zonelist(ac->zonelist,
5002 ac->highest_zoneidx, ac->nodemask);
5003 }
5004
5005
5006 page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
5007 if (page)
5008 goto got_pg;
5009
5010
5011 if (!can_direct_reclaim)
5012 goto nopage;
5013
5014
5015 if (current->flags & PF_MEMALLOC)
5016 goto nopage;
5017
5018
5019 page = __alloc_pages_direct_reclaim(gfp_mask, order, alloc_flags, ac,
5020 &did_some_progress);
5021 if (page)
5022 goto got_pg;
5023
5024
5025 page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, ac,
5026 compact_priority, &compact_result);
5027 if (page)
5028 goto got_pg;
5029
5030
5031 if (gfp_mask & __GFP_NORETRY)
5032 goto nopage;
5033
5034
5035
5036
5037
5038 if (costly_order && !(gfp_mask & __GFP_RETRY_MAYFAIL))
5039 goto nopage;
5040
5041 if (should_reclaim_retry(gfp_mask, order, ac, alloc_flags,
5042 did_some_progress > 0, &no_progress_loops))
5043 goto retry;
5044
5045
5046
5047
5048
5049
5050
5051 if (did_some_progress > 0 &&
5052 should_compact_retry(ac, order, alloc_flags,
5053 compact_result, &compact_priority,
5054 &compaction_retries))
5055 goto retry;
5056
5057
5058
5059 if (check_retry_cpuset(cpuset_mems_cookie, ac))
5060 goto retry_cpuset;
5061
5062
5063 page = __alloc_pages_may_oom(gfp_mask, order, ac, &did_some_progress);
5064 if (page)
5065 goto got_pg;
5066
5067
5068 if (tsk_is_oom_victim(current) &&
5069 (alloc_flags & ALLOC_OOM ||
5070 (gfp_mask & __GFP_NOMEMALLOC)))
5071 goto nopage;
5072
5073
5074 if (did_some_progress) {
5075 no_progress_loops = 0;
5076 goto retry;
5077 }
5078
5079nopage:
5080
5081 if (check_retry_cpuset(cpuset_mems_cookie, ac))
5082 goto retry_cpuset;
5083
5084
5085
5086
5087
5088 if (gfp_mask & __GFP_NOFAIL) {
5089
5090
5091
5092
5093 if (WARN_ON_ONCE(!can_direct_reclaim))
5094 goto fail;
5095
5096
5097
5098
5099
5100
5101 WARN_ON_ONCE(current->flags & PF_MEMALLOC);
5102
5103
5104
5105
5106
5107
5108
5109 WARN_ON_ONCE(order > PAGE_ALLOC_COSTLY_ORDER);
5110
5111
5112
5113
5114
5115
5116
5117 page = __alloc_pages_cpuset_fallback(gfp_mask, order, ALLOC_HARDER, ac);
5118 if (page)
5119 goto got_pg;
5120
5121 cond_resched();
5122 goto retry;
5123 }
5124fail:
5125 warn_alloc(gfp_mask, ac->nodemask,
5126 "page allocation failure: order:%u", order);
5127got_pg:
5128 return page;
5129}
5130
5131static inline bool prepare_alloc_pages(gfp_t gfp_mask, unsigned int order,
5132 int preferred_nid, nodemask_t *nodemask,
5133 struct alloc_context *ac, gfp_t *alloc_gfp,
5134 unsigned int *alloc_flags)
5135{
5136 ac->highest_zoneidx = gfp_zone(gfp_mask);
5137 ac->zonelist = node_zonelist(preferred_nid, gfp_mask);
5138 ac->nodemask = nodemask;
5139 ac->migratetype = gfp_migratetype(gfp_mask);
5140
5141 if (cpusets_enabled()) {
5142 *alloc_gfp |= __GFP_HARDWALL;
5143
5144
5145
5146
5147 if (in_task() && !ac->nodemask)
5148 ac->nodemask = &cpuset_current_mems_allowed;
5149 else
5150 *alloc_flags |= ALLOC_CPUSET;
5151 }
5152
5153 fs_reclaim_acquire(gfp_mask);
5154 fs_reclaim_release(gfp_mask);
5155
5156 might_sleep_if(gfp_mask & __GFP_DIRECT_RECLAIM);
5157
5158 if (should_fail_alloc_page(gfp_mask, order))
5159 return false;
5160
5161 *alloc_flags = gfp_to_alloc_flags_cma(gfp_mask, *alloc_flags);
5162
5163
5164 ac->spread_dirty_pages = (gfp_mask & __GFP_WRITE);
5165
5166
5167
5168
5169
5170
5171 ac->preferred_zoneref = first_zones_zonelist(ac->zonelist,
5172 ac->highest_zoneidx, ac->nodemask);
5173
5174 return true;
5175}
5176
5177
5178
5179
5180
5181
5182
5183
5184
5185
5186
5187
5188
5189
5190
5191
5192
5193
5194
5195
5196
5197unsigned long __alloc_pages_bulk(gfp_t gfp, int preferred_nid,
5198 nodemask_t *nodemask, int nr_pages,
5199 struct list_head *page_list,
5200 struct page **page_array)
5201{
5202 struct page *page;
5203 unsigned long flags;
5204 struct zone *zone;
5205 struct zoneref *z;
5206 struct per_cpu_pages *pcp;
5207 struct list_head *pcp_list;
5208 struct alloc_context ac;
5209 gfp_t alloc_gfp;
5210 unsigned int alloc_flags = ALLOC_WMARK_LOW;
5211 int nr_populated = 0, nr_account = 0;
5212
5213
5214
5215
5216
5217 while (page_array && nr_populated < nr_pages && page_array[nr_populated])
5218 nr_populated++;
5219
5220
5221 if (unlikely(nr_pages <= 0))
5222 goto out;
5223
5224
5225 if (unlikely(page_array && nr_pages - nr_populated == 0))
5226 goto out;
5227
5228
5229 if (memcg_kmem_enabled() && (gfp & __GFP_ACCOUNT))
5230 goto failed;
5231
5232
5233 if (nr_pages - nr_populated == 1)
5234 goto failed;
5235
5236#ifdef CONFIG_PAGE_OWNER
5237
5238
5239
5240
5241
5242
5243
5244 if (static_branch_unlikely(&page_owner_inited))
5245 goto failed;
5246#endif
5247
5248
5249 gfp &= gfp_allowed_mask;
5250 alloc_gfp = gfp;
5251 if (!prepare_alloc_pages(gfp, 0, preferred_nid, nodemask, &ac, &alloc_gfp, &alloc_flags))
5252 goto out;
5253 gfp = alloc_gfp;
5254
5255
5256 for_each_zone_zonelist_nodemask(zone, z, ac.zonelist, ac.highest_zoneidx, ac.nodemask) {
5257 unsigned long mark;
5258
5259 if (cpusets_enabled() && (alloc_flags & ALLOC_CPUSET) &&
5260 !__cpuset_zone_allowed(zone, gfp)) {
5261 continue;
5262 }
5263
5264 if (nr_online_nodes > 1 && zone != ac.preferred_zoneref->zone &&
5265 zone_to_nid(zone) != zone_to_nid(ac.preferred_zoneref->zone)) {
5266 goto failed;
5267 }
5268
5269 mark = wmark_pages(zone, alloc_flags & ALLOC_WMARK_MASK) + nr_pages;
5270 if (zone_watermark_fast(zone, 0, mark,
5271 zonelist_zone_idx(ac.preferred_zoneref),
5272 alloc_flags, gfp)) {
5273 break;
5274 }
5275 }
5276
5277
5278
5279
5280
5281 if (unlikely(!zone))
5282 goto failed;
5283
5284
5285 local_lock_irqsave(&pagesets.lock, flags);
5286 pcp = this_cpu_ptr(zone->per_cpu_pageset);
5287 pcp_list = &pcp->lists[order_to_pindex(ac.migratetype, 0)];
5288
5289 while (nr_populated < nr_pages) {
5290
5291
5292 if (page_array && page_array[nr_populated]) {
5293 nr_populated++;
5294 continue;
5295 }
5296
5297 page = __rmqueue_pcplist(zone, 0, ac.migratetype, alloc_flags,
5298 pcp, pcp_list);
5299 if (unlikely(!page)) {
5300
5301 if (!nr_populated)
5302 goto failed_irq;
5303 break;
5304 }
5305 nr_account++;
5306
5307 prep_new_page(page, 0, gfp, 0);
5308 if (page_list)
5309 list_add(&page->lru, page_list);
5310 else
5311 page_array[nr_populated] = page;
5312 nr_populated++;
5313 }
5314
5315 local_unlock_irqrestore(&pagesets.lock, flags);
5316
5317 __count_zid_vm_events(PGALLOC, zone_idx(zone), nr_account);
5318 zone_statistics(ac.preferred_zoneref->zone, zone, nr_account);
5319
5320out:
5321 return nr_populated;
5322
5323failed_irq:
5324 local_unlock_irqrestore(&pagesets.lock, flags);
5325
5326failed:
5327 page = __alloc_pages(gfp, 0, preferred_nid, nodemask);
5328 if (page) {
5329 if (page_list)
5330 list_add(&page->lru, page_list);
5331 else
5332 page_array[nr_populated] = page;
5333 nr_populated++;
5334 }
5335
5336 goto out;
5337}
5338EXPORT_SYMBOL_GPL(__alloc_pages_bulk);
5339
5340
5341
5342
5343struct page *__alloc_pages(gfp_t gfp, unsigned int order, int preferred_nid,
5344 nodemask_t *nodemask)
5345{
5346 struct page *page;
5347 unsigned int alloc_flags = ALLOC_WMARK_LOW;
5348 gfp_t alloc_gfp;
5349 struct alloc_context ac = { };
5350
5351
5352
5353
5354
5355 if (unlikely(order >= MAX_ORDER)) {
5356 WARN_ON_ONCE(!(gfp & __GFP_NOWARN));
5357 return NULL;
5358 }
5359
5360 gfp &= gfp_allowed_mask;
5361
5362
5363
5364
5365
5366
5367
5368 gfp = current_gfp_context(gfp);
5369 alloc_gfp = gfp;
5370 if (!prepare_alloc_pages(gfp, order, preferred_nid, nodemask, &ac,
5371 &alloc_gfp, &alloc_flags))
5372 return NULL;
5373
5374
5375
5376
5377
5378 alloc_flags |= alloc_flags_nofragment(ac.preferred_zoneref->zone, gfp);
5379
5380
5381 page = get_page_from_freelist(alloc_gfp, order, alloc_flags, &ac);
5382 if (likely(page))
5383 goto out;
5384
5385 alloc_gfp = gfp;
5386 ac.spread_dirty_pages = false;
5387
5388
5389
5390
5391
5392 ac.nodemask = nodemask;
5393
5394 page = __alloc_pages_slowpath(alloc_gfp, order, &ac);
5395
5396out:
5397 if (memcg_kmem_enabled() && (gfp & __GFP_ACCOUNT) && page &&
5398 unlikely(__memcg_kmem_charge_page(page, gfp, order) != 0)) {
5399 __free_pages(page, order);
5400 page = NULL;
5401 }
5402
5403 trace_mm_page_alloc(page, order, alloc_gfp, ac.migratetype);
5404
5405 return page;
5406}
5407EXPORT_SYMBOL(__alloc_pages);
5408
5409
5410
5411
5412
5413
5414unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
5415{
5416 struct page *page;
5417
5418 page = alloc_pages(gfp_mask & ~__GFP_HIGHMEM, order);
5419 if (!page)
5420 return 0;
5421 return (unsigned long) page_address(page);
5422}
5423EXPORT_SYMBOL(__get_free_pages);
5424
5425unsigned long get_zeroed_page(gfp_t gfp_mask)
5426{
5427 return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
5428}
5429EXPORT_SYMBOL(get_zeroed_page);
5430
5431
5432
5433
5434
5435
5436
5437
5438
5439
5440
5441
5442
5443
5444
5445
5446
5447
5448
5449
5450
5451void __free_pages(struct page *page, unsigned int order)
5452{
5453 if (put_page_testzero(page))
5454 free_the_page(page, order);
5455 else if (!PageHead(page))
5456 while (order-- > 0)
5457 free_the_page(page + (1 << order), order);
5458}
5459EXPORT_SYMBOL(__free_pages);
5460
5461void free_pages(unsigned long addr, unsigned int order)
5462{
5463 if (addr != 0) {
5464 VM_BUG_ON(!virt_addr_valid((void *)addr));
5465 __free_pages(virt_to_page((void *)addr), order);
5466 }
5467}
5468
5469EXPORT_SYMBOL(free_pages);
5470
5471
5472
5473
5474
5475
5476
5477
5478
5479
5480
5481
5482static struct page *__page_frag_cache_refill(struct page_frag_cache *nc,
5483 gfp_t gfp_mask)
5484{
5485 struct page *page = NULL;
5486 gfp_t gfp = gfp_mask;
5487
5488#if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
5489 gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY |
5490 __GFP_NOMEMALLOC;
5491 page = alloc_pages_node(NUMA_NO_NODE, gfp_mask,
5492 PAGE_FRAG_CACHE_MAX_ORDER);
5493 nc->size = page ? PAGE_FRAG_CACHE_MAX_SIZE : PAGE_SIZE;
5494#endif
5495 if (unlikely(!page))
5496 page = alloc_pages_node(NUMA_NO_NODE, gfp, 0);
5497
5498 nc->va = page ? page_address(page) : NULL;
5499
5500 return page;
5501}
5502
5503void __page_frag_cache_drain(struct page *page, unsigned int count)
5504{
5505 VM_BUG_ON_PAGE(page_ref_count(page) == 0, page);
5506
5507 if (page_ref_sub_and_test(page, count))
5508 free_the_page(page, compound_order(page));
5509}
5510EXPORT_SYMBOL(__page_frag_cache_drain);
5511
5512void *page_frag_alloc_align(struct page_frag_cache *nc,
5513 unsigned int fragsz, gfp_t gfp_mask,
5514 unsigned int align_mask)
5515{
5516 unsigned int size = PAGE_SIZE;
5517 struct page *page;
5518 int offset;
5519
5520 if (unlikely(!nc->va)) {
5521refill:
5522 page = __page_frag_cache_refill(nc, gfp_mask);
5523 if (!page)
5524 return NULL;
5525
5526#if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
5527
5528 size = nc->size;
5529#endif
5530
5531
5532
5533 page_ref_add(page, PAGE_FRAG_CACHE_MAX_SIZE);
5534
5535
5536 nc->pfmemalloc = page_is_pfmemalloc(page);
5537 nc->pagecnt_bias = PAGE_FRAG_CACHE_MAX_SIZE + 1;
5538 nc->offset = size;
5539 }
5540
5541 offset = nc->offset - fragsz;
5542 if (unlikely(offset < 0)) {
5543 page = virt_to_page(nc->va);
5544
5545 if (!page_ref_sub_and_test(page, nc->pagecnt_bias))
5546 goto refill;
5547
5548 if (unlikely(nc->pfmemalloc)) {
5549 free_the_page(page, compound_order(page));
5550 goto refill;
5551 }
5552
5553#if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
5554
5555 size = nc->size;
5556#endif
5557
5558 set_page_count(page, PAGE_FRAG_CACHE_MAX_SIZE + 1);
5559
5560
5561 nc->pagecnt_bias = PAGE_FRAG_CACHE_MAX_SIZE + 1;
5562 offset = size - fragsz;
5563 }
5564
5565 nc->pagecnt_bias--;
5566 offset &= align_mask;
5567 nc->offset = offset;
5568
5569 return nc->va + offset;
5570}
5571EXPORT_SYMBOL(page_frag_alloc_align);
5572
5573
5574
5575
5576void page_frag_free(void *addr)
5577{
5578 struct page *page = virt_to_head_page(addr);
5579
5580 if (unlikely(put_page_testzero(page)))
5581 free_the_page(page, compound_order(page));
5582}
5583EXPORT_SYMBOL(page_frag_free);
5584
5585static void *make_alloc_exact(unsigned long addr, unsigned int order,
5586 size_t size)
5587{
5588 if (addr) {
5589 unsigned long alloc_end = addr + (PAGE_SIZE << order);
5590 unsigned long used = addr + PAGE_ALIGN(size);
5591
5592 split_page(virt_to_page((void *)addr), order);
5593 while (used < alloc_end) {
5594 free_page(used);
5595 used += PAGE_SIZE;
5596 }
5597 }
5598 return (void *)addr;
5599}
5600
5601
5602
5603
5604
5605
5606
5607
5608
5609
5610
5611
5612
5613
5614
5615
5616void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
5617{
5618 unsigned int order = get_order(size);
5619 unsigned long addr;
5620
5621 if (WARN_ON_ONCE(gfp_mask & __GFP_COMP))
5622 gfp_mask &= ~__GFP_COMP;
5623
5624 addr = __get_free_pages(gfp_mask, order);
5625 return make_alloc_exact(addr, order, size);
5626}
5627EXPORT_SYMBOL(alloc_pages_exact);
5628
5629
5630
5631
5632
5633
5634
5635
5636
5637
5638
5639
5640
5641void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask)
5642{
5643 unsigned int order = get_order(size);
5644 struct page *p;
5645
5646 if (WARN_ON_ONCE(gfp_mask & __GFP_COMP))
5647 gfp_mask &= ~__GFP_COMP;
5648
5649 p = alloc_pages_node(nid, gfp_mask, order);
5650 if (!p)
5651 return NULL;
5652 return make_alloc_exact((unsigned long)page_address(p), order, size);
5653}
5654
5655
5656
5657
5658
5659
5660
5661
5662void free_pages_exact(void *virt, size_t size)
5663{
5664 unsigned long addr = (unsigned long)virt;
5665 unsigned long end = addr + PAGE_ALIGN(size);
5666
5667 while (addr < end) {
5668 free_page(addr);
5669 addr += PAGE_SIZE;
5670 }
5671}
5672EXPORT_SYMBOL(free_pages_exact);
5673
5674
5675
5676
5677
5678
5679
5680
5681
5682
5683
5684
5685
5686static unsigned long nr_free_zone_pages(int offset)
5687{
5688 struct zoneref *z;
5689 struct zone *zone;
5690
5691
5692 unsigned long sum = 0;
5693
5694 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
5695
5696 for_each_zone_zonelist(zone, z, zonelist, offset) {
5697 unsigned long size = zone_managed_pages(zone);
5698 unsigned long high = high_wmark_pages(zone);
5699 if (size > high)
5700 sum += size - high;
5701 }
5702
5703 return sum;
5704}
5705
5706
5707
5708
5709
5710
5711
5712
5713
5714
5715unsigned long nr_free_buffer_pages(void)
5716{
5717 return nr_free_zone_pages(gfp_zone(GFP_USER));
5718}
5719EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
5720
5721static inline void show_node(struct zone *zone)
5722{
5723 if (IS_ENABLED(CONFIG_NUMA))
5724 printk("Node %d ", zone_to_nid(zone));
5725}
5726
5727long si_mem_available(void)
5728{
5729 long available;
5730 unsigned long pagecache;
5731 unsigned long wmark_low = 0;
5732 unsigned long pages[NR_LRU_LISTS];
5733 unsigned long reclaimable;
5734 struct zone *zone;
5735 int lru;
5736
5737 for (lru = LRU_BASE; lru < NR_LRU_LISTS; lru++)
5738 pages[lru] = global_node_page_state(NR_LRU_BASE + lru);
5739
5740 for_each_zone(zone)
5741 wmark_low += low_wmark_pages(zone);
5742
5743
5744
5745
5746
5747 available = global_zone_page_state(NR_FREE_PAGES) - totalreserve_pages;
5748
5749
5750
5751
5752
5753
5754 pagecache = pages[LRU_ACTIVE_FILE] + pages[LRU_INACTIVE_FILE];
5755 pagecache -= min(pagecache / 2, wmark_low);
5756 available += pagecache;
5757
5758
5759
5760
5761
5762
5763 reclaimable = global_node_page_state_pages(NR_SLAB_RECLAIMABLE_B) +
5764 global_node_page_state(NR_KERNEL_MISC_RECLAIMABLE);
5765 available += reclaimable - min(reclaimable / 2, wmark_low);
5766
5767 if (available < 0)
5768 available = 0;
5769 return available;
5770}
5771EXPORT_SYMBOL_GPL(si_mem_available);
5772
5773void si_meminfo(struct sysinfo *val)
5774{
5775 val->totalram = totalram_pages();
5776 val->sharedram = global_node_page_state(NR_SHMEM);
5777 val->freeram = global_zone_page_state(NR_FREE_PAGES);
5778 val->bufferram = nr_blockdev_pages();
5779 val->totalhigh = totalhigh_pages();
5780 val->freehigh = nr_free_highpages();
5781 val->mem_unit = PAGE_SIZE;
5782}
5783
5784EXPORT_SYMBOL(si_meminfo);
5785
5786#ifdef CONFIG_NUMA
5787void si_meminfo_node(struct sysinfo *val, int nid)
5788{
5789 int zone_type;
5790 unsigned long managed_pages = 0;
5791 unsigned long managed_highpages = 0;
5792 unsigned long free_highpages = 0;
5793 pg_data_t *pgdat = NODE_DATA(nid);
5794
5795 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++)
5796 managed_pages += zone_managed_pages(&pgdat->node_zones[zone_type]);
5797 val->totalram = managed_pages;
5798 val->sharedram = node_page_state(pgdat, NR_SHMEM);
5799 val->freeram = sum_zone_node_page_state(nid, NR_FREE_PAGES);
5800#ifdef CONFIG_HIGHMEM
5801 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
5802 struct zone *zone = &pgdat->node_zones[zone_type];
5803
5804 if (is_highmem(zone)) {
5805 managed_highpages += zone_managed_pages(zone);
5806 free_highpages += zone_page_state(zone, NR_FREE_PAGES);
5807 }
5808 }
5809 val->totalhigh = managed_highpages;
5810 val->freehigh = free_highpages;
5811#else
5812 val->totalhigh = managed_highpages;
5813 val->freehigh = free_highpages;
5814#endif
5815 val->mem_unit = PAGE_SIZE;
5816}
5817#endif
5818
5819
5820
5821
5822
5823static bool show_mem_node_skip(unsigned int flags, int nid, nodemask_t *nodemask)
5824{
5825 if (!(flags & SHOW_MEM_FILTER_NODES))
5826 return false;
5827
5828
5829
5830
5831
5832
5833 if (!nodemask)
5834 nodemask = &cpuset_current_mems_allowed;
5835
5836 return !node_isset(nid, *nodemask);
5837}
5838
5839#define K(x) ((x) << (PAGE_SHIFT-10))
5840
5841static void show_migration_types(unsigned char type)
5842{
5843 static const char types[MIGRATE_TYPES] = {
5844 [MIGRATE_UNMOVABLE] = 'U',
5845 [MIGRATE_MOVABLE] = 'M',
5846 [MIGRATE_RECLAIMABLE] = 'E',
5847 [MIGRATE_HIGHATOMIC] = 'H',
5848#ifdef CONFIG_CMA
5849 [MIGRATE_CMA] = 'C',
5850#endif
5851#ifdef CONFIG_MEMORY_ISOLATION
5852 [MIGRATE_ISOLATE] = 'I',
5853#endif
5854 };
5855 char tmp[MIGRATE_TYPES + 1];
5856 char *p = tmp;
5857 int i;
5858
5859 for (i = 0; i < MIGRATE_TYPES; i++) {
5860 if (type & (1 << i))
5861 *p++ = types[i];
5862 }
5863
5864 *p = '\0';
5865 printk(KERN_CONT "(%s) ", tmp);
5866}
5867
5868
5869
5870
5871
5872
5873
5874
5875
5876
5877void show_free_areas(unsigned int filter, nodemask_t *nodemask)
5878{
5879 unsigned long free_pcp = 0;
5880 int cpu;
5881 struct zone *zone;
5882 pg_data_t *pgdat;
5883
5884 for_each_populated_zone(zone) {
5885 if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask))
5886 continue;
5887
5888 for_each_online_cpu(cpu)
5889 free_pcp += per_cpu_ptr(zone->per_cpu_pageset, cpu)->count;
5890 }
5891
5892 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
5893 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
5894 " unevictable:%lu dirty:%lu writeback:%lu\n"
5895 " slab_reclaimable:%lu slab_unreclaimable:%lu\n"
5896 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
5897 " kernel_misc_reclaimable:%lu\n"
5898 " free:%lu free_pcp:%lu free_cma:%lu\n",
5899 global_node_page_state(NR_ACTIVE_ANON),
5900 global_node_page_state(NR_INACTIVE_ANON),
5901 global_node_page_state(NR_ISOLATED_ANON),
5902 global_node_page_state(NR_ACTIVE_FILE),
5903 global_node_page_state(NR_INACTIVE_FILE),
5904 global_node_page_state(NR_ISOLATED_FILE),
5905 global_node_page_state(NR_UNEVICTABLE),
5906 global_node_page_state(NR_FILE_DIRTY),
5907 global_node_page_state(NR_WRITEBACK),
5908 global_node_page_state_pages(NR_SLAB_RECLAIMABLE_B),
5909 global_node_page_state_pages(NR_SLAB_UNRECLAIMABLE_B),
5910 global_node_page_state(NR_FILE_MAPPED),
5911 global_node_page_state(NR_SHMEM),
5912 global_node_page_state(NR_PAGETABLE),
5913 global_zone_page_state(NR_BOUNCE),
5914 global_node_page_state(NR_KERNEL_MISC_RECLAIMABLE),
5915 global_zone_page_state(NR_FREE_PAGES),
5916 free_pcp,
5917 global_zone_page_state(NR_FREE_CMA_PAGES));
5918
5919 for_each_online_pgdat(pgdat) {
5920 if (show_mem_node_skip(filter, pgdat->node_id, nodemask))
5921 continue;
5922
5923 printk("Node %d"
5924 " active_anon:%lukB"
5925 " inactive_anon:%lukB"
5926 " active_file:%lukB"
5927 " inactive_file:%lukB"
5928 " unevictable:%lukB"
5929 " isolated(anon):%lukB"
5930 " isolated(file):%lukB"
5931 " mapped:%lukB"
5932 " dirty:%lukB"
5933 " writeback:%lukB"
5934 " shmem:%lukB"
5935#ifdef CONFIG_TRANSPARENT_HUGEPAGE
5936 " shmem_thp: %lukB"
5937 " shmem_pmdmapped: %lukB"
5938 " anon_thp: %lukB"
5939#endif
5940 " writeback_tmp:%lukB"
5941 " kernel_stack:%lukB"
5942#ifdef CONFIG_SHADOW_CALL_STACK
5943 " shadow_call_stack:%lukB"
5944#endif
5945 " pagetables:%lukB"
5946 " all_unreclaimable? %s"
5947 "\n",
5948 pgdat->node_id,
5949 K(node_page_state(pgdat, NR_ACTIVE_ANON)),
5950 K(node_page_state(pgdat, NR_INACTIVE_ANON)),
5951 K(node_page_state(pgdat, NR_ACTIVE_FILE)),
5952 K(node_page_state(pgdat, NR_INACTIVE_FILE)),
5953 K(node_page_state(pgdat, NR_UNEVICTABLE)),
5954 K(node_page_state(pgdat, NR_ISOLATED_ANON)),
5955 K(node_page_state(pgdat, NR_ISOLATED_FILE)),
5956 K(node_page_state(pgdat, NR_FILE_MAPPED)),
5957 K(node_page_state(pgdat, NR_FILE_DIRTY)),
5958 K(node_page_state(pgdat, NR_WRITEBACK)),
5959 K(node_page_state(pgdat, NR_SHMEM)),
5960#ifdef CONFIG_TRANSPARENT_HUGEPAGE
5961 K(node_page_state(pgdat, NR_SHMEM_THPS)),
5962 K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED)),
5963 K(node_page_state(pgdat, NR_ANON_THPS)),
5964#endif
5965 K(node_page_state(pgdat, NR_WRITEBACK_TEMP)),
5966 node_page_state(pgdat, NR_KERNEL_STACK_KB),
5967#ifdef CONFIG_SHADOW_CALL_STACK
5968 node_page_state(pgdat, NR_KERNEL_SCS_KB),
5969#endif
5970 K(node_page_state(pgdat, NR_PAGETABLE)),
5971 pgdat->kswapd_failures >= MAX_RECLAIM_RETRIES ?
5972 "yes" : "no");
5973 }
5974
5975 for_each_populated_zone(zone) {
5976 int i;
5977
5978 if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask))
5979 continue;
5980
5981 free_pcp = 0;
5982 for_each_online_cpu(cpu)
5983 free_pcp += per_cpu_ptr(zone->per_cpu_pageset, cpu)->count;
5984
5985 show_node(zone);
5986 printk(KERN_CONT
5987 "%s"
5988 " free:%lukB"
5989 " min:%lukB"
5990 " low:%lukB"
5991 " high:%lukB"
5992 " reserved_highatomic:%luKB"
5993 " active_anon:%lukB"
5994 " inactive_anon:%lukB"
5995 " active_file:%lukB"
5996 " inactive_file:%lukB"
5997 " unevictable:%lukB"
5998 " writepending:%lukB"
5999 " present:%lukB"
6000 " managed:%lukB"
6001 " mlocked:%lukB"
6002 " bounce:%lukB"
6003 " free_pcp:%lukB"
6004 " local_pcp:%ukB"
6005 " free_cma:%lukB"
6006 "\n",
6007 zone->name,
6008 K(zone_page_state(zone, NR_FREE_PAGES)),
6009 K(min_wmark_pages(zone)),
6010 K(low_wmark_pages(zone)),
6011 K(high_wmark_pages(zone)),
6012 K(zone->nr_reserved_highatomic),
6013 K(zone_page_state(zone, NR_ZONE_ACTIVE_ANON)),
6014 K(zone_page_state(zone, NR_ZONE_INACTIVE_ANON)),
6015 K(zone_page_state(zone, NR_ZONE_ACTIVE_FILE)),
6016 K(zone_page_state(zone, NR_ZONE_INACTIVE_FILE)),
6017 K(zone_page_state(zone, NR_ZONE_UNEVICTABLE)),
6018 K(zone_page_state(zone, NR_ZONE_WRITE_PENDING)),
6019 K(zone->present_pages),
6020 K(zone_managed_pages(zone)),
6021 K(zone_page_state(zone, NR_MLOCK)),
6022 K(zone_page_state(zone, NR_BOUNCE)),
6023 K(free_pcp),
6024 K(this_cpu_read(zone->per_cpu_pageset->count)),
6025 K(zone_page_state(zone, NR_FREE_CMA_PAGES)));
6026 printk("lowmem_reserve[]:");
6027 for (i = 0; i < MAX_NR_ZONES; i++)
6028 printk(KERN_CONT " %ld", zone->lowmem_reserve[i]);
6029 printk(KERN_CONT "\n");
6030 }
6031
6032 for_each_populated_zone(zone) {
6033 unsigned int order;
6034 unsigned long nr[MAX_ORDER], flags, total = 0;
6035 unsigned char types[MAX_ORDER];
6036
6037 if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask))
6038 continue;
6039 show_node(zone);
6040 printk(KERN_CONT "%s: ", zone->name);
6041
6042 spin_lock_irqsave(&zone->lock, flags);
6043 for (order = 0; order < MAX_ORDER; order++) {
6044 struct free_area *area = &zone->free_area[order];
6045 int type;
6046
6047 nr[order] = area->nr_free;
6048 total += nr[order] << order;
6049
6050 types[order] = 0;
6051 for (type = 0; type < MIGRATE_TYPES; type++) {
6052 if (!free_area_empty(area, type))
6053 types[order] |= 1 << type;
6054 }
6055 }
6056 spin_unlock_irqrestore(&zone->lock, flags);
6057 for (order = 0; order < MAX_ORDER; order++) {
6058 printk(KERN_CONT "%lu*%lukB ",
6059 nr[order], K(1UL) << order);
6060 if (nr[order])
6061 show_migration_types(types[order]);
6062 }
6063 printk(KERN_CONT "= %lukB\n", K(total));
6064 }
6065
6066 hugetlb_show_meminfo();
6067
6068 printk("%ld total pagecache pages\n", global_node_page_state(NR_FILE_PAGES));
6069
6070 show_swap_cache_info();
6071}
6072
6073static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
6074{
6075 zoneref->zone = zone;
6076 zoneref->zone_idx = zone_idx(zone);
6077}
6078
6079
6080
6081
6082
6083
6084static int build_zonerefs_node(pg_data_t *pgdat, struct zoneref *zonerefs)
6085{
6086 struct zone *zone;
6087 enum zone_type zone_type = MAX_NR_ZONES;
6088 int nr_zones = 0;
6089
6090 do {
6091 zone_type--;
6092 zone = pgdat->node_zones + zone_type;
6093 if (managed_zone(zone)) {
6094 zoneref_set_zone(zone, &zonerefs[nr_zones++]);
6095 check_highest_zone(zone_type);
6096 }
6097 } while (zone_type);
6098
6099 return nr_zones;
6100}
6101
6102#ifdef CONFIG_NUMA
6103
6104static int __parse_numa_zonelist_order(char *s)
6105{
6106
6107
6108
6109
6110
6111
6112 if (!(*s == 'd' || *s == 'D' || *s == 'n' || *s == 'N')) {
6113 pr_warn("Ignoring unsupported numa_zonelist_order value: %s\n", s);
6114 return -EINVAL;
6115 }
6116 return 0;
6117}
6118
6119char numa_zonelist_order[] = "Node";
6120
6121
6122
6123
6124int numa_zonelist_order_handler(struct ctl_table *table, int write,
6125 void *buffer, size_t *length, loff_t *ppos)
6126{
6127 if (write)
6128 return __parse_numa_zonelist_order(buffer);
6129 return proc_dostring(table, write, buffer, length, ppos);
6130}
6131
6132
6133#define MAX_NODE_LOAD (nr_online_nodes)
6134static int node_load[MAX_NUMNODES];
6135
6136
6137
6138
6139
6140
6141
6142
6143
6144
6145
6146
6147
6148
6149
6150
6151int find_next_best_node(int node, nodemask_t *used_node_mask)
6152{
6153 int n, val;
6154 int min_val = INT_MAX;
6155 int best_node = NUMA_NO_NODE;
6156
6157
6158 if (!node_isset(node, *used_node_mask)) {
6159 node_set(node, *used_node_mask);
6160 return node;
6161 }
6162
6163 for_each_node_state(n, N_MEMORY) {
6164
6165
6166 if (node_isset(n, *used_node_mask))
6167 continue;
6168
6169
6170 val = node_distance(node, n);
6171
6172
6173 val += (n < node);
6174
6175
6176 if (!cpumask_empty(cpumask_of_node(n)))
6177 val += PENALTY_FOR_NODE_WITH_CPUS;
6178
6179
6180 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
6181 val += node_load[n];
6182
6183 if (val < min_val) {
6184 min_val = val;
6185 best_node = n;
6186 }
6187 }
6188
6189 if (best_node >= 0)
6190 node_set(best_node, *used_node_mask);
6191
6192 return best_node;
6193}
6194
6195
6196
6197
6198
6199
6200
6201static void build_zonelists_in_node_order(pg_data_t *pgdat, int *node_order,
6202 unsigned nr_nodes)
6203{
6204 struct zoneref *zonerefs;
6205 int i;
6206
6207 zonerefs = pgdat->node_zonelists[ZONELIST_FALLBACK]._zonerefs;
6208
6209 for (i = 0; i < nr_nodes; i++) {
6210 int nr_zones;
6211
6212 pg_data_t *node = NODE_DATA(node_order[i]);
6213
6214 nr_zones = build_zonerefs_node(node, zonerefs);
6215 zonerefs += nr_zones;
6216 }
6217 zonerefs->zone = NULL;
6218 zonerefs->zone_idx = 0;
6219}
6220
6221
6222
6223
6224static void build_thisnode_zonelists(pg_data_t *pgdat)
6225{
6226 struct zoneref *zonerefs;
6227 int nr_zones;
6228
6229 zonerefs = pgdat->node_zonelists[ZONELIST_NOFALLBACK]._zonerefs;
6230 nr_zones = build_zonerefs_node(pgdat, zonerefs);
6231 zonerefs += nr_zones;
6232 zonerefs->zone = NULL;
6233 zonerefs->zone_idx = 0;
6234}
6235
6236
6237
6238
6239
6240
6241
6242
6243static void build_zonelists(pg_data_t *pgdat)
6244{
6245 static int node_order[MAX_NUMNODES];
6246 int node, load, nr_nodes = 0;
6247 nodemask_t used_mask = NODE_MASK_NONE;
6248 int local_node, prev_node;
6249
6250
6251 local_node = pgdat->node_id;
6252 load = nr_online_nodes;
6253 prev_node = local_node;
6254
6255 memset(node_order, 0, sizeof(node_order));
6256 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
6257
6258
6259
6260
6261
6262 if (node_distance(local_node, node) !=
6263 node_distance(local_node, prev_node))
6264 node_load[node] = load;
6265
6266 node_order[nr_nodes++] = node;
6267 prev_node = node;
6268 load--;
6269 }
6270
6271 build_zonelists_in_node_order(pgdat, node_order, nr_nodes);
6272 build_thisnode_zonelists(pgdat);
6273}
6274
6275#ifdef CONFIG_HAVE_MEMORYLESS_NODES
6276
6277
6278
6279
6280
6281
6282int local_memory_node(int node)
6283{
6284 struct zoneref *z;
6285
6286 z = first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
6287 gfp_zone(GFP_KERNEL),
6288 NULL);
6289 return zone_to_nid(z->zone);
6290}
6291#endif
6292
6293static void setup_min_unmapped_ratio(void);
6294static void setup_min_slab_ratio(void);
6295#else
6296
6297static void build_zonelists(pg_data_t *pgdat)
6298{
6299 int node, local_node;
6300 struct zoneref *zonerefs;
6301 int nr_zones;
6302
6303 local_node = pgdat->node_id;
6304
6305 zonerefs = pgdat->node_zonelists[ZONELIST_FALLBACK]._zonerefs;
6306 nr_zones = build_zonerefs_node(pgdat, zonerefs);
6307 zonerefs += nr_zones;
6308
6309
6310
6311
6312
6313
6314
6315
6316
6317 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
6318 if (!node_online(node))
6319 continue;
6320 nr_zones = build_zonerefs_node(NODE_DATA(node), zonerefs);
6321 zonerefs += nr_zones;
6322 }
6323 for (node = 0; node < local_node; node++) {
6324 if (!node_online(node))
6325 continue;
6326 nr_zones = build_zonerefs_node(NODE_DATA(node), zonerefs);
6327 zonerefs += nr_zones;
6328 }
6329
6330 zonerefs->zone = NULL;
6331 zonerefs->zone_idx = 0;
6332}
6333
6334#endif
6335
6336
6337
6338
6339
6340
6341
6342
6343
6344
6345
6346
6347
6348
6349
6350
6351static void per_cpu_pages_init(struct per_cpu_pages *pcp, struct per_cpu_zonestat *pzstats);
6352
6353#define BOOT_PAGESET_HIGH 0
6354#define BOOT_PAGESET_BATCH 1
6355static DEFINE_PER_CPU(struct per_cpu_pages, boot_pageset);
6356static DEFINE_PER_CPU(struct per_cpu_zonestat, boot_zonestats);
6357static DEFINE_PER_CPU(struct per_cpu_nodestat, boot_nodestats);
6358
6359static void __build_all_zonelists(void *data)
6360{
6361 int nid;
6362 int __maybe_unused cpu;
6363 pg_data_t *self = data;
6364 static DEFINE_SPINLOCK(lock);
6365
6366 spin_lock(&lock);
6367
6368#ifdef CONFIG_NUMA
6369 memset(node_load, 0, sizeof(node_load));
6370#endif
6371
6372
6373
6374
6375
6376 if (self && !node_online(self->node_id)) {
6377 build_zonelists(self);
6378 } else {
6379 for_each_online_node(nid) {
6380 pg_data_t *pgdat = NODE_DATA(nid);
6381
6382 build_zonelists(pgdat);
6383 }
6384
6385#ifdef CONFIG_HAVE_MEMORYLESS_NODES
6386
6387
6388
6389
6390
6391
6392
6393
6394 for_each_online_cpu(cpu)
6395 set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
6396#endif
6397 }
6398
6399 spin_unlock(&lock);
6400}
6401
6402static noinline void __init
6403build_all_zonelists_init(void)
6404{
6405 int cpu;
6406
6407 __build_all_zonelists(NULL);
6408
6409
6410
6411
6412
6413
6414
6415
6416
6417
6418
6419
6420
6421
6422 for_each_possible_cpu(cpu)
6423 per_cpu_pages_init(&per_cpu(boot_pageset, cpu), &per_cpu(boot_zonestats, cpu));
6424
6425 mminit_verify_zonelist();
6426 cpuset_init_current_mems_allowed();
6427}
6428
6429
6430
6431
6432
6433
6434
6435void __ref build_all_zonelists(pg_data_t *pgdat)
6436{
6437 unsigned long vm_total_pages;
6438
6439 if (system_state == SYSTEM_BOOTING) {
6440 build_all_zonelists_init();
6441 } else {
6442 __build_all_zonelists(pgdat);
6443
6444 }
6445
6446 vm_total_pages = nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
6447
6448
6449
6450
6451
6452
6453
6454 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
6455 page_group_by_mobility_disabled = 1;
6456 else
6457 page_group_by_mobility_disabled = 0;
6458
6459 pr_info("Built %u zonelists, mobility grouping %s. Total pages: %ld\n",
6460 nr_online_nodes,
6461 page_group_by_mobility_disabled ? "off" : "on",
6462 vm_total_pages);
6463#ifdef CONFIG_NUMA
6464 pr_info("Policy zone: %s\n", zone_names[policy_zone]);
6465#endif
6466}
6467
6468
6469static bool __meminit
6470overlap_memmap_init(unsigned long zone, unsigned long *pfn)
6471{
6472 static struct memblock_region *r;
6473
6474 if (mirrored_kernelcore && zone == ZONE_MOVABLE) {
6475 if (!r || *pfn >= memblock_region_memory_end_pfn(r)) {
6476 for_each_mem_region(r) {
6477 if (*pfn < memblock_region_memory_end_pfn(r))
6478 break;
6479 }
6480 }
6481 if (*pfn >= memblock_region_memory_base_pfn(r) &&
6482 memblock_is_mirror(r)) {
6483 *pfn = memblock_region_memory_end_pfn(r);
6484 return true;
6485 }
6486 }
6487 return false;
6488}
6489
6490
6491
6492
6493
6494
6495
6496
6497
6498
6499void __meminit memmap_init_range(unsigned long size, int nid, unsigned long zone,
6500 unsigned long start_pfn, unsigned long zone_end_pfn,
6501 enum meminit_context context,
6502 struct vmem_altmap *altmap, int migratetype)
6503{
6504 unsigned long pfn, end_pfn = start_pfn + size;
6505 struct page *page;
6506
6507 if (highest_memmap_pfn < end_pfn - 1)
6508 highest_memmap_pfn = end_pfn - 1;
6509
6510#ifdef CONFIG_ZONE_DEVICE
6511
6512
6513
6514
6515
6516
6517
6518 if (zone == ZONE_DEVICE) {
6519 if (!altmap)
6520 return;
6521
6522 if (start_pfn == altmap->base_pfn)
6523 start_pfn += altmap->reserve;
6524 end_pfn = altmap->base_pfn + vmem_altmap_offset(altmap);
6525 }
6526#endif
6527
6528 for (pfn = start_pfn; pfn < end_pfn; ) {
6529
6530
6531
6532
6533 if (context == MEMINIT_EARLY) {
6534 if (overlap_memmap_init(zone, &pfn))
6535 continue;
6536 if (defer_init(nid, pfn, zone_end_pfn))
6537 break;
6538 }
6539
6540 page = pfn_to_page(pfn);
6541 __init_single_page(page, pfn, zone, nid);
6542 if (context == MEMINIT_HOTPLUG)
6543 __SetPageReserved(page);
6544
6545
6546
6547
6548
6549
6550 if (IS_ALIGNED(pfn, pageblock_nr_pages)) {
6551 set_pageblock_migratetype(page, migratetype);
6552 cond_resched();
6553 }
6554 pfn++;
6555 }
6556}
6557
6558#ifdef CONFIG_ZONE_DEVICE
6559void __ref memmap_init_zone_device(struct zone *zone,
6560 unsigned long start_pfn,
6561 unsigned long nr_pages,
6562 struct dev_pagemap *pgmap)
6563{
6564 unsigned long pfn, end_pfn = start_pfn + nr_pages;
6565 struct pglist_data *pgdat = zone->zone_pgdat;
6566 struct vmem_altmap *altmap = pgmap_altmap(pgmap);
6567 unsigned long zone_idx = zone_idx(zone);
6568 unsigned long start = jiffies;
6569 int nid = pgdat->node_id;
6570
6571 if (WARN_ON_ONCE(!pgmap || zone_idx(zone) != ZONE_DEVICE))
6572 return;
6573
6574
6575
6576
6577
6578
6579 if (altmap) {
6580 start_pfn = altmap->base_pfn + vmem_altmap_offset(altmap);
6581 nr_pages = end_pfn - start_pfn;
6582 }
6583
6584 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
6585 struct page *page = pfn_to_page(pfn);
6586
6587 __init_single_page(page, pfn, zone_idx, nid);
6588
6589
6590
6591
6592
6593
6594
6595
6596 __SetPageReserved(page);
6597
6598
6599
6600
6601
6602
6603 page->pgmap = pgmap;
6604 page->zone_device_data = NULL;
6605
6606
6607
6608
6609
6610
6611
6612
6613
6614
6615
6616 if (IS_ALIGNED(pfn, pageblock_nr_pages)) {
6617 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
6618 cond_resched();
6619 }
6620 }
6621
6622 pr_info("%s initialised %lu pages in %ums\n", __func__,
6623 nr_pages, jiffies_to_msecs(jiffies - start));
6624}
6625
6626#endif
6627static void __meminit zone_init_free_lists(struct zone *zone)
6628{
6629 unsigned int order, t;
6630 for_each_migratetype_order(order, t) {
6631 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
6632 zone->free_area[order].nr_free = 0;
6633 }
6634}
6635
6636
6637
6638
6639
6640
6641
6642
6643
6644
6645
6646
6647
6648
6649
6650
6651
6652
6653
6654
6655
6656
6657
6658static void __init init_unavailable_range(unsigned long spfn,
6659 unsigned long epfn,
6660 int zone, int node)
6661{
6662 unsigned long pfn;
6663 u64 pgcnt = 0;
6664
6665 for (pfn = spfn; pfn < epfn; pfn++) {
6666 if (!pfn_valid(ALIGN_DOWN(pfn, pageblock_nr_pages))) {
6667 pfn = ALIGN_DOWN(pfn, pageblock_nr_pages)
6668 + pageblock_nr_pages - 1;
6669 continue;
6670 }
6671 __init_single_page(pfn_to_page(pfn), pfn, zone, node);
6672 __SetPageReserved(pfn_to_page(pfn));
6673 pgcnt++;
6674 }
6675
6676 if (pgcnt)
6677 pr_info("On node %d, zone %s: %lld pages in unavailable ranges",
6678 node, zone_names[zone], pgcnt);
6679}
6680
6681static void __init memmap_init_zone_range(struct zone *zone,
6682 unsigned long start_pfn,
6683 unsigned long end_pfn,
6684 unsigned long *hole_pfn)
6685{
6686 unsigned long zone_start_pfn = zone->zone_start_pfn;
6687 unsigned long zone_end_pfn = zone_start_pfn + zone->spanned_pages;
6688 int nid = zone_to_nid(zone), zone_id = zone_idx(zone);
6689
6690 start_pfn = clamp(start_pfn, zone_start_pfn, zone_end_pfn);
6691 end_pfn = clamp(end_pfn, zone_start_pfn, zone_end_pfn);
6692
6693 if (start_pfn >= end_pfn)
6694 return;
6695
6696 memmap_init_range(end_pfn - start_pfn, nid, zone_id, start_pfn,
6697 zone_end_pfn, MEMINIT_EARLY, NULL, MIGRATE_MOVABLE);
6698
6699 if (*hole_pfn < start_pfn)
6700 init_unavailable_range(*hole_pfn, start_pfn, zone_id, nid);
6701
6702 *hole_pfn = end_pfn;
6703}
6704
6705static void __init memmap_init(void)
6706{
6707 unsigned long start_pfn, end_pfn;
6708 unsigned long hole_pfn = 0;
6709 int i, j, zone_id = 0, nid;
6710
6711 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
6712 struct pglist_data *node = NODE_DATA(nid);
6713
6714 for (j = 0; j < MAX_NR_ZONES; j++) {
6715 struct zone *zone = node->node_zones + j;
6716
6717 if (!populated_zone(zone))
6718 continue;
6719
6720 memmap_init_zone_range(zone, start_pfn, end_pfn,
6721 &hole_pfn);
6722 zone_id = j;
6723 }
6724 }
6725
6726#ifdef CONFIG_SPARSEMEM
6727
6728
6729
6730
6731
6732
6733
6734
6735
6736 end_pfn = round_up(end_pfn, PAGES_PER_SECTION);
6737 if (hole_pfn < end_pfn)
6738#endif
6739 init_unavailable_range(hole_pfn, end_pfn, zone_id, nid);
6740}
6741
6742void __init *memmap_alloc(phys_addr_t size, phys_addr_t align,
6743 phys_addr_t min_addr, int nid, bool exact_nid)
6744{
6745 void *ptr;
6746
6747 if (exact_nid)
6748 ptr = memblock_alloc_exact_nid_raw(size, align, min_addr,
6749 MEMBLOCK_ALLOC_ACCESSIBLE,
6750 nid);
6751 else
6752 ptr = memblock_alloc_try_nid_raw(size, align, min_addr,
6753 MEMBLOCK_ALLOC_ACCESSIBLE,
6754 nid);
6755
6756 if (ptr && size > 0)
6757 page_init_poison(ptr, size);
6758
6759 return ptr;
6760}
6761
6762static int zone_batchsize(struct zone *zone)
6763{
6764#ifdef CONFIG_MMU
6765 int batch;
6766
6767
6768
6769
6770
6771
6772
6773 batch = min(zone_managed_pages(zone) >> 10, (1024 * 1024) / PAGE_SIZE);
6774 batch /= 4;
6775 if (batch < 1)
6776 batch = 1;
6777
6778
6779
6780
6781
6782
6783
6784
6785
6786
6787
6788 batch = rounddown_pow_of_two(batch + batch/2) - 1;
6789
6790 return batch;
6791
6792#else
6793
6794
6795
6796
6797
6798
6799
6800
6801
6802
6803
6804
6805
6806 return 0;
6807#endif
6808}
6809
6810static int zone_highsize(struct zone *zone, int batch, int cpu_online)
6811{
6812#ifdef CONFIG_MMU
6813 int high;
6814 int nr_split_cpus;
6815 unsigned long total_pages;
6816
6817 if (!percpu_pagelist_high_fraction) {
6818
6819
6820
6821
6822
6823 total_pages = low_wmark_pages(zone);
6824 } else {
6825
6826
6827
6828
6829
6830 total_pages = zone_managed_pages(zone) / percpu_pagelist_high_fraction;
6831 }
6832
6833
6834
6835
6836
6837
6838
6839
6840
6841 nr_split_cpus = cpumask_weight(cpumask_of_node(zone_to_nid(zone))) + cpu_online;
6842 if (!nr_split_cpus)
6843 nr_split_cpus = num_online_cpus();
6844 high = total_pages / nr_split_cpus;
6845
6846
6847
6848
6849
6850 high = max(high, batch << 2);
6851
6852 return high;
6853#else
6854 return 0;
6855#endif
6856}
6857
6858
6859
6860
6861
6862
6863
6864
6865
6866
6867
6868
6869
6870
6871
6872
6873
6874static void pageset_update(struct per_cpu_pages *pcp, unsigned long high,
6875 unsigned long batch)
6876{
6877 WRITE_ONCE(pcp->batch, batch);
6878 WRITE_ONCE(pcp->high, high);
6879}
6880
6881static void per_cpu_pages_init(struct per_cpu_pages *pcp, struct per_cpu_zonestat *pzstats)
6882{
6883 int pindex;
6884
6885 memset(pcp, 0, sizeof(*pcp));
6886 memset(pzstats, 0, sizeof(*pzstats));
6887
6888 for (pindex = 0; pindex < NR_PCP_LISTS; pindex++)
6889 INIT_LIST_HEAD(&pcp->lists[pindex]);
6890
6891
6892
6893
6894
6895
6896
6897 pcp->high = BOOT_PAGESET_HIGH;
6898 pcp->batch = BOOT_PAGESET_BATCH;
6899 pcp->free_factor = 0;
6900}
6901
6902static void __zone_set_pageset_high_and_batch(struct zone *zone, unsigned long high,
6903 unsigned long batch)
6904{
6905 struct per_cpu_pages *pcp;
6906 int cpu;
6907
6908 for_each_possible_cpu(cpu) {
6909 pcp = per_cpu_ptr(zone->per_cpu_pageset, cpu);
6910 pageset_update(pcp, high, batch);
6911 }
6912}
6913
6914
6915
6916
6917
6918static void zone_set_pageset_high_and_batch(struct zone *zone, int cpu_online)
6919{
6920 int new_high, new_batch;
6921
6922 new_batch = max(1, zone_batchsize(zone));
6923 new_high = zone_highsize(zone, new_batch, cpu_online);
6924
6925 if (zone->pageset_high == new_high &&
6926 zone->pageset_batch == new_batch)
6927 return;
6928
6929 zone->pageset_high = new_high;
6930 zone->pageset_batch = new_batch;
6931
6932 __zone_set_pageset_high_and_batch(zone, new_high, new_batch);
6933}
6934
6935void __meminit setup_zone_pageset(struct zone *zone)
6936{
6937 int cpu;
6938
6939
6940 if (sizeof(struct per_cpu_zonestat) > 0)
6941 zone->per_cpu_zonestats = alloc_percpu(struct per_cpu_zonestat);
6942
6943 zone->per_cpu_pageset = alloc_percpu(struct per_cpu_pages);
6944 for_each_possible_cpu(cpu) {
6945 struct per_cpu_pages *pcp;
6946 struct per_cpu_zonestat *pzstats;
6947
6948 pcp = per_cpu_ptr(zone->per_cpu_pageset, cpu);
6949 pzstats = per_cpu_ptr(zone->per_cpu_zonestats, cpu);
6950 per_cpu_pages_init(pcp, pzstats);
6951 }
6952
6953 zone_set_pageset_high_and_batch(zone, 0);
6954}
6955
6956
6957
6958
6959
6960void __init setup_per_cpu_pageset(void)
6961{
6962 struct pglist_data *pgdat;
6963 struct zone *zone;
6964 int __maybe_unused cpu;
6965
6966 for_each_populated_zone(zone)
6967 setup_zone_pageset(zone);
6968
6969#ifdef CONFIG_NUMA
6970
6971
6972
6973
6974
6975
6976 for_each_possible_cpu(cpu) {
6977 struct per_cpu_zonestat *pzstats = &per_cpu(boot_zonestats, cpu);
6978 memset(pzstats->vm_numa_event, 0,
6979 sizeof(pzstats->vm_numa_event));
6980 }
6981#endif
6982
6983 for_each_online_pgdat(pgdat)
6984 pgdat->per_cpu_nodestats =
6985 alloc_percpu(struct per_cpu_nodestat);
6986}
6987
6988static __meminit void zone_pcp_init(struct zone *zone)
6989{
6990
6991
6992
6993
6994
6995 zone->per_cpu_pageset = &boot_pageset;
6996 zone->per_cpu_zonestats = &boot_zonestats;
6997 zone->pageset_high = BOOT_PAGESET_HIGH;
6998 zone->pageset_batch = BOOT_PAGESET_BATCH;
6999
7000 if (populated_zone(zone))
7001 pr_debug(" %s zone: %lu pages, LIFO batch:%u\n", zone->name,
7002 zone->present_pages, zone_batchsize(zone));
7003}
7004
7005void __meminit init_currently_empty_zone(struct zone *zone,
7006 unsigned long zone_start_pfn,
7007 unsigned long size)
7008{
7009 struct pglist_data *pgdat = zone->zone_pgdat;
7010 int zone_idx = zone_idx(zone) + 1;
7011
7012 if (zone_idx > pgdat->nr_zones)
7013 pgdat->nr_zones = zone_idx;
7014
7015 zone->zone_start_pfn = zone_start_pfn;
7016
7017 mminit_dprintk(MMINIT_TRACE, "memmap_init",
7018 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
7019 pgdat->node_id,
7020 (unsigned long)zone_idx(zone),
7021 zone_start_pfn, (zone_start_pfn + size));
7022
7023 zone_init_free_lists(zone);
7024 zone->initialized = 1;
7025}
7026
7027
7028
7029
7030
7031
7032
7033
7034
7035
7036
7037
7038void __init get_pfn_range_for_nid(unsigned int nid,
7039 unsigned long *start_pfn, unsigned long *end_pfn)
7040{
7041 unsigned long this_start_pfn, this_end_pfn;
7042 int i;
7043
7044 *start_pfn = -1UL;
7045 *end_pfn = 0;
7046
7047 for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) {
7048 *start_pfn = min(*start_pfn, this_start_pfn);
7049 *end_pfn = max(*end_pfn, this_end_pfn);
7050 }
7051
7052 if (*start_pfn == -1UL)
7053 *start_pfn = 0;
7054}
7055
7056
7057
7058
7059
7060
7061static void __init find_usable_zone_for_movable(void)
7062{
7063 int zone_index;
7064 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
7065 if (zone_index == ZONE_MOVABLE)
7066 continue;
7067
7068 if (arch_zone_highest_possible_pfn[zone_index] >
7069 arch_zone_lowest_possible_pfn[zone_index])
7070 break;
7071 }
7072
7073 VM_BUG_ON(zone_index == -1);
7074 movable_zone = zone_index;
7075}
7076
7077
7078
7079
7080
7081
7082
7083
7084
7085
7086
7087static void __init adjust_zone_range_for_zone_movable(int nid,
7088 unsigned long zone_type,
7089 unsigned long node_start_pfn,
7090 unsigned long node_end_pfn,
7091 unsigned long *zone_start_pfn,
7092 unsigned long *zone_end_pfn)
7093{
7094
7095 if (zone_movable_pfn[nid]) {
7096
7097 if (zone_type == ZONE_MOVABLE) {
7098 *zone_start_pfn = zone_movable_pfn[nid];
7099 *zone_end_pfn = min(node_end_pfn,
7100 arch_zone_highest_possible_pfn[movable_zone]);
7101
7102
7103 } else if (!mirrored_kernelcore &&
7104 *zone_start_pfn < zone_movable_pfn[nid] &&
7105 *zone_end_pfn > zone_movable_pfn[nid]) {
7106 *zone_end_pfn = zone_movable_pfn[nid];
7107
7108
7109 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
7110 *zone_start_pfn = *zone_end_pfn;
7111 }
7112}
7113
7114
7115
7116
7117
7118static unsigned long __init zone_spanned_pages_in_node(int nid,
7119 unsigned long zone_type,
7120 unsigned long node_start_pfn,
7121 unsigned long node_end_pfn,
7122 unsigned long *zone_start_pfn,
7123 unsigned long *zone_end_pfn)
7124{
7125 unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type];
7126 unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type];
7127
7128 if (!node_start_pfn && !node_end_pfn)
7129 return 0;
7130
7131
7132 *zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high);
7133 *zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high);
7134 adjust_zone_range_for_zone_movable(nid, zone_type,
7135 node_start_pfn, node_end_pfn,
7136 zone_start_pfn, zone_end_pfn);
7137
7138
7139 if (*zone_end_pfn < node_start_pfn || *zone_start_pfn > node_end_pfn)
7140 return 0;
7141
7142
7143 *zone_end_pfn = min(*zone_end_pfn, node_end_pfn);
7144 *zone_start_pfn = max(*zone_start_pfn, node_start_pfn);
7145
7146
7147 return *zone_end_pfn - *zone_start_pfn;
7148}
7149
7150
7151
7152
7153
7154unsigned long __init __absent_pages_in_range(int nid,
7155 unsigned long range_start_pfn,
7156 unsigned long range_end_pfn)
7157{
7158 unsigned long nr_absent = range_end_pfn - range_start_pfn;
7159 unsigned long start_pfn, end_pfn;
7160 int i;
7161
7162 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
7163 start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn);
7164 end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn);
7165 nr_absent -= end_pfn - start_pfn;
7166 }
7167 return nr_absent;
7168}
7169
7170
7171
7172
7173
7174
7175
7176
7177unsigned long __init absent_pages_in_range(unsigned long start_pfn,
7178 unsigned long end_pfn)
7179{
7180 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
7181}
7182
7183
7184static unsigned long __init zone_absent_pages_in_node(int nid,
7185 unsigned long zone_type,
7186 unsigned long node_start_pfn,
7187 unsigned long node_end_pfn)
7188{
7189 unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type];
7190 unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type];
7191 unsigned long zone_start_pfn, zone_end_pfn;
7192 unsigned long nr_absent;
7193
7194
7195 if (!node_start_pfn && !node_end_pfn)
7196 return 0;
7197
7198 zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high);
7199 zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high);
7200
7201 adjust_zone_range_for_zone_movable(nid, zone_type,
7202 node_start_pfn, node_end_pfn,
7203 &zone_start_pfn, &zone_end_pfn);
7204 nr_absent = __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
7205
7206
7207
7208
7209
7210
7211 if (mirrored_kernelcore && zone_movable_pfn[nid]) {
7212 unsigned long start_pfn, end_pfn;
7213 struct memblock_region *r;
7214
7215 for_each_mem_region(r) {
7216 start_pfn = clamp(memblock_region_memory_base_pfn(r),
7217 zone_start_pfn, zone_end_pfn);
7218 end_pfn = clamp(memblock_region_memory_end_pfn(r),
7219 zone_start_pfn, zone_end_pfn);
7220
7221 if (zone_type == ZONE_MOVABLE &&
7222 memblock_is_mirror(r))
7223 nr_absent += end_pfn - start_pfn;
7224
7225 if (zone_type == ZONE_NORMAL &&
7226 !memblock_is_mirror(r))
7227 nr_absent += end_pfn - start_pfn;
7228 }
7229 }
7230
7231 return nr_absent;
7232}
7233
7234static void __init calculate_node_totalpages(struct pglist_data *pgdat,
7235 unsigned long node_start_pfn,
7236 unsigned long node_end_pfn)
7237{
7238 unsigned long realtotalpages = 0, totalpages = 0;
7239 enum zone_type i;
7240
7241 for (i = 0; i < MAX_NR_ZONES; i++) {
7242 struct zone *zone = pgdat->node_zones + i;
7243 unsigned long zone_start_pfn, zone_end_pfn;
7244 unsigned long spanned, absent;
7245 unsigned long size, real_size;
7246
7247 spanned = zone_spanned_pages_in_node(pgdat->node_id, i,
7248 node_start_pfn,
7249 node_end_pfn,
7250 &zone_start_pfn,
7251 &zone_end_pfn);
7252 absent = zone_absent_pages_in_node(pgdat->node_id, i,
7253 node_start_pfn,
7254 node_end_pfn);
7255
7256 size = spanned;
7257 real_size = size - absent;
7258
7259 if (size)
7260 zone->zone_start_pfn = zone_start_pfn;
7261 else
7262 zone->zone_start_pfn = 0;
7263 zone->spanned_pages = size;
7264 zone->present_pages = real_size;
7265#if defined(CONFIG_MEMORY_HOTPLUG)
7266 zone->present_early_pages = real_size;
7267#endif
7268
7269 totalpages += size;
7270 realtotalpages += real_size;
7271 }
7272
7273 pgdat->node_spanned_pages = totalpages;
7274 pgdat->node_present_pages = realtotalpages;
7275 pr_debug("On node %d totalpages: %lu\n", pgdat->node_id, realtotalpages);
7276}
7277
7278#ifndef CONFIG_SPARSEMEM
7279
7280
7281
7282
7283
7284
7285
7286static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize)
7287{
7288 unsigned long usemapsize;
7289
7290 zonesize += zone_start_pfn & (pageblock_nr_pages-1);
7291 usemapsize = roundup(zonesize, pageblock_nr_pages);
7292 usemapsize = usemapsize >> pageblock_order;
7293 usemapsize *= NR_PAGEBLOCK_BITS;
7294 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
7295
7296 return usemapsize / 8;
7297}
7298
7299static void __ref setup_usemap(struct zone *zone)
7300{
7301 unsigned long usemapsize = usemap_size(zone->zone_start_pfn,
7302 zone->spanned_pages);
7303 zone->pageblock_flags = NULL;
7304 if (usemapsize) {
7305 zone->pageblock_flags =
7306 memblock_alloc_node(usemapsize, SMP_CACHE_BYTES,
7307 zone_to_nid(zone));
7308 if (!zone->pageblock_flags)
7309 panic("Failed to allocate %ld bytes for zone %s pageblock flags on node %d\n",
7310 usemapsize, zone->name, zone_to_nid(zone));
7311 }
7312}
7313#else
7314static inline void setup_usemap(struct zone *zone) {}
7315#endif
7316
7317#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
7318
7319
7320void __init set_pageblock_order(void)
7321{
7322 unsigned int order;
7323
7324
7325 if (pageblock_order)
7326 return;
7327
7328 if (HPAGE_SHIFT > PAGE_SHIFT)
7329 order = HUGETLB_PAGE_ORDER;
7330 else
7331 order = MAX_ORDER - 1;
7332
7333
7334
7335
7336
7337
7338 pageblock_order = order;
7339}
7340#else
7341
7342
7343
7344
7345
7346
7347
7348void __init set_pageblock_order(void)
7349{
7350}
7351
7352#endif
7353
7354static unsigned long __init calc_memmap_size(unsigned long spanned_pages,
7355 unsigned long present_pages)
7356{
7357 unsigned long pages = spanned_pages;
7358
7359
7360
7361
7362
7363
7364
7365
7366
7367 if (spanned_pages > present_pages + (present_pages >> 4) &&
7368 IS_ENABLED(CONFIG_SPARSEMEM))
7369 pages = present_pages;
7370
7371 return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT;
7372}
7373
7374#ifdef CONFIG_TRANSPARENT_HUGEPAGE
7375static void pgdat_init_split_queue(struct pglist_data *pgdat)
7376{
7377 struct deferred_split *ds_queue = &pgdat->deferred_split_queue;
7378
7379 spin_lock_init(&ds_queue->split_queue_lock);
7380 INIT_LIST_HEAD(&ds_queue->split_queue);
7381 ds_queue->split_queue_len = 0;
7382}
7383#else
7384static void pgdat_init_split_queue(struct pglist_data *pgdat) {}
7385#endif
7386
7387#ifdef CONFIG_COMPACTION
7388static void pgdat_init_kcompactd(struct pglist_data *pgdat)
7389{
7390 init_waitqueue_head(&pgdat->kcompactd_wait);
7391}
7392#else
7393static void pgdat_init_kcompactd(struct pglist_data *pgdat) {}
7394#endif
7395
7396static void __meminit pgdat_init_internals(struct pglist_data *pgdat)
7397{
7398 pgdat_resize_init(pgdat);
7399
7400 pgdat_init_split_queue(pgdat);
7401 pgdat_init_kcompactd(pgdat);
7402
7403 init_waitqueue_head(&pgdat->kswapd_wait);
7404 init_waitqueue_head(&pgdat->pfmemalloc_wait);
7405
7406 pgdat_page_ext_init(pgdat);
7407 lruvec_init(&pgdat->__lruvec);
7408}
7409
7410static void __meminit zone_init_internals(struct zone *zone, enum zone_type idx, int nid,
7411 unsigned long remaining_pages)
7412{
7413 atomic_long_set(&zone->managed_pages, remaining_pages);
7414 zone_set_nid(zone, nid);
7415 zone->name = zone_names[idx];
7416 zone->zone_pgdat = NODE_DATA(nid);
7417 spin_lock_init(&zone->lock);
7418 zone_seqlock_init(zone);
7419 zone_pcp_init(zone);
7420}
7421
7422
7423
7424
7425
7426
7427
7428
7429#ifdef CONFIG_MEMORY_HOTPLUG
7430void __ref free_area_init_core_hotplug(int nid)
7431{
7432 enum zone_type z;
7433 pg_data_t *pgdat = NODE_DATA(nid);
7434
7435 pgdat_init_internals(pgdat);
7436 for (z = 0; z < MAX_NR_ZONES; z++)
7437 zone_init_internals(&pgdat->node_zones[z], z, nid, 0);
7438}
7439#endif
7440
7441
7442
7443
7444
7445
7446
7447
7448
7449
7450static void __init free_area_init_core(struct pglist_data *pgdat)
7451{
7452 enum zone_type j;
7453 int nid = pgdat->node_id;
7454
7455 pgdat_init_internals(pgdat);
7456 pgdat->per_cpu_nodestats = &boot_nodestats;
7457
7458 for (j = 0; j < MAX_NR_ZONES; j++) {
7459 struct zone *zone = pgdat->node_zones + j;
7460 unsigned long size, freesize, memmap_pages;
7461
7462 size = zone->spanned_pages;
7463 freesize = zone->present_pages;
7464
7465
7466
7467
7468
7469
7470 memmap_pages = calc_memmap_size(size, freesize);
7471 if (!is_highmem_idx(j)) {
7472 if (freesize >= memmap_pages) {
7473 freesize -= memmap_pages;
7474 if (memmap_pages)
7475 pr_debug(" %s zone: %lu pages used for memmap\n",
7476 zone_names[j], memmap_pages);
7477 } else
7478 pr_warn(" %s zone: %lu memmap pages exceeds freesize %lu\n",
7479 zone_names[j], memmap_pages, freesize);
7480 }
7481
7482
7483 if (j == 0 && freesize > dma_reserve) {
7484 freesize -= dma_reserve;
7485 pr_debug(" %s zone: %lu pages reserved\n", zone_names[0], dma_reserve);
7486 }
7487
7488 if (!is_highmem_idx(j))
7489 nr_kernel_pages += freesize;
7490
7491 else if (nr_kernel_pages > memmap_pages * 2)
7492 nr_kernel_pages -= memmap_pages;
7493 nr_all_pages += freesize;
7494
7495
7496
7497
7498
7499
7500 zone_init_internals(zone, j, nid, freesize);
7501
7502 if (!size)
7503 continue;
7504
7505 set_pageblock_order();
7506 setup_usemap(zone);
7507 init_currently_empty_zone(zone, zone->zone_start_pfn, size);
7508 }
7509}
7510
7511#ifdef CONFIG_FLATMEM
7512static void __init alloc_node_mem_map(struct pglist_data *pgdat)
7513{
7514 unsigned long __maybe_unused start = 0;
7515 unsigned long __maybe_unused offset = 0;
7516
7517
7518 if (!pgdat->node_spanned_pages)
7519 return;
7520
7521 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
7522 offset = pgdat->node_start_pfn - start;
7523
7524 if (!pgdat->node_mem_map) {
7525 unsigned long size, end;
7526 struct page *map;
7527
7528
7529
7530
7531
7532
7533 end = pgdat_end_pfn(pgdat);
7534 end = ALIGN(end, MAX_ORDER_NR_PAGES);
7535 size = (end - start) * sizeof(struct page);
7536 map = memmap_alloc(size, SMP_CACHE_BYTES, MEMBLOCK_LOW_LIMIT,
7537 pgdat->node_id, false);
7538 if (!map)
7539 panic("Failed to allocate %ld bytes for node %d memory map\n",
7540 size, pgdat->node_id);
7541 pgdat->node_mem_map = map + offset;
7542 }
7543 pr_debug("%s: node %d, pgdat %08lx, node_mem_map %08lx\n",
7544 __func__, pgdat->node_id, (unsigned long)pgdat,
7545 (unsigned long)pgdat->node_mem_map);
7546#ifndef CONFIG_NUMA
7547
7548
7549
7550 if (pgdat == NODE_DATA(0)) {
7551 mem_map = NODE_DATA(0)->node_mem_map;
7552 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
7553 mem_map -= offset;
7554 }
7555#endif
7556}
7557#else
7558static inline void alloc_node_mem_map(struct pglist_data *pgdat) { }
7559#endif
7560
7561#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
7562static inline void pgdat_set_deferred_range(pg_data_t *pgdat)
7563{
7564 pgdat->first_deferred_pfn = ULONG_MAX;
7565}
7566#else
7567static inline void pgdat_set_deferred_range(pg_data_t *pgdat) {}
7568#endif
7569
7570static void __init free_area_init_node(int nid)
7571{
7572 pg_data_t *pgdat = NODE_DATA(nid);
7573 unsigned long start_pfn = 0;
7574 unsigned long end_pfn = 0;
7575
7576
7577 WARN_ON(pgdat->nr_zones || pgdat->kswapd_highest_zoneidx);
7578
7579 get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
7580
7581 pgdat->node_id = nid;
7582 pgdat->node_start_pfn = start_pfn;
7583 pgdat->per_cpu_nodestats = NULL;
7584
7585 pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid,
7586 (u64)start_pfn << PAGE_SHIFT,
7587 end_pfn ? ((u64)end_pfn << PAGE_SHIFT) - 1 : 0);
7588 calculate_node_totalpages(pgdat, start_pfn, end_pfn);
7589
7590 alloc_node_mem_map(pgdat);
7591 pgdat_set_deferred_range(pgdat);
7592
7593 free_area_init_core(pgdat);
7594}
7595
7596void __init free_area_init_memoryless_node(int nid)
7597{
7598 free_area_init_node(nid);
7599}
7600
7601#if MAX_NUMNODES > 1
7602
7603
7604
7605void __init setup_nr_node_ids(void)
7606{
7607 unsigned int highest;
7608
7609 highest = find_last_bit(node_possible_map.bits, MAX_NUMNODES);
7610 nr_node_ids = highest + 1;
7611}
7612#endif
7613
7614
7615
7616
7617
7618
7619
7620
7621
7622
7623
7624
7625
7626
7627
7628
7629
7630
7631
7632
7633unsigned long __init node_map_pfn_alignment(void)
7634{
7635 unsigned long accl_mask = 0, last_end = 0;
7636 unsigned long start, end, mask;
7637 int last_nid = NUMA_NO_NODE;
7638 int i, nid;
7639
7640 for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) {
7641 if (!start || last_nid < 0 || last_nid == nid) {
7642 last_nid = nid;
7643 last_end = end;
7644 continue;
7645 }
7646
7647
7648
7649
7650
7651
7652 mask = ~((1 << __ffs(start)) - 1);
7653 while (mask && last_end <= (start & (mask << 1)))
7654 mask <<= 1;
7655
7656
7657 accl_mask |= mask;
7658 }
7659
7660
7661 return ~accl_mask + 1;
7662}
7663
7664
7665
7666
7667
7668
7669
7670unsigned long __init find_min_pfn_with_active_regions(void)
7671{
7672 return PHYS_PFN(memblock_start_of_DRAM());
7673}
7674
7675
7676
7677
7678
7679
7680static unsigned long __init early_calculate_totalpages(void)
7681{
7682 unsigned long totalpages = 0;
7683 unsigned long start_pfn, end_pfn;
7684 int i, nid;
7685
7686 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
7687 unsigned long pages = end_pfn - start_pfn;
7688
7689 totalpages += pages;
7690 if (pages)
7691 node_set_state(nid, N_MEMORY);
7692 }
7693 return totalpages;
7694}
7695
7696
7697
7698
7699
7700
7701
7702static void __init find_zone_movable_pfns_for_nodes(void)
7703{
7704 int i, nid;
7705 unsigned long usable_startpfn;
7706 unsigned long kernelcore_node, kernelcore_remaining;
7707
7708 nodemask_t saved_node_state = node_states[N_MEMORY];
7709 unsigned long totalpages = early_calculate_totalpages();
7710 int usable_nodes = nodes_weight(node_states[N_MEMORY]);
7711 struct memblock_region *r;
7712
7713
7714 find_usable_zone_for_movable();
7715
7716
7717
7718
7719
7720 if (movable_node_is_enabled()) {
7721 for_each_mem_region(r) {
7722 if (!memblock_is_hotpluggable(r))
7723 continue;
7724
7725 nid = memblock_get_region_node(r);
7726
7727 usable_startpfn = PFN_DOWN(r->base);
7728 zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
7729 min(usable_startpfn, zone_movable_pfn[nid]) :
7730 usable_startpfn;
7731 }
7732
7733 goto out2;
7734 }
7735
7736
7737
7738
7739 if (mirrored_kernelcore) {
7740 bool mem_below_4gb_not_mirrored = false;
7741
7742 for_each_mem_region(r) {
7743 if (memblock_is_mirror(r))
7744 continue;
7745
7746 nid = memblock_get_region_node(r);
7747
7748 usable_startpfn = memblock_region_memory_base_pfn(r);
7749
7750 if (usable_startpfn < 0x100000) {
7751 mem_below_4gb_not_mirrored = true;
7752 continue;
7753 }
7754
7755 zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
7756 min(usable_startpfn, zone_movable_pfn[nid]) :
7757 usable_startpfn;
7758 }
7759
7760 if (mem_below_4gb_not_mirrored)
7761 pr_warn("This configuration results in unmirrored kernel memory.\n");
7762
7763 goto out2;
7764 }
7765
7766
7767
7768
7769
7770 if (required_kernelcore_percent)
7771 required_kernelcore = (totalpages * 100 * required_kernelcore_percent) /
7772 10000UL;
7773 if (required_movablecore_percent)
7774 required_movablecore = (totalpages * 100 * required_movablecore_percent) /
7775 10000UL;
7776
7777
7778
7779
7780
7781
7782
7783
7784
7785 if (required_movablecore) {
7786 unsigned long corepages;
7787
7788
7789
7790
7791
7792 required_movablecore =
7793 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
7794 required_movablecore = min(totalpages, required_movablecore);
7795 corepages = totalpages - required_movablecore;
7796
7797 required_kernelcore = max(required_kernelcore, corepages);
7798 }
7799
7800
7801
7802
7803
7804 if (!required_kernelcore || required_kernelcore >= totalpages)
7805 goto out;
7806
7807
7808 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
7809
7810restart:
7811
7812 kernelcore_node = required_kernelcore / usable_nodes;
7813 for_each_node_state(nid, N_MEMORY) {
7814 unsigned long start_pfn, end_pfn;
7815
7816
7817
7818
7819
7820
7821 if (required_kernelcore < kernelcore_node)
7822 kernelcore_node = required_kernelcore / usable_nodes;
7823
7824
7825
7826
7827
7828
7829 kernelcore_remaining = kernelcore_node;
7830
7831
7832 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
7833 unsigned long size_pages;
7834
7835 start_pfn = max(start_pfn, zone_movable_pfn[nid]);
7836 if (start_pfn >= end_pfn)
7837 continue;
7838
7839
7840 if (start_pfn < usable_startpfn) {
7841 unsigned long kernel_pages;
7842 kernel_pages = min(end_pfn, usable_startpfn)
7843 - start_pfn;
7844
7845 kernelcore_remaining -= min(kernel_pages,
7846 kernelcore_remaining);
7847 required_kernelcore -= min(kernel_pages,
7848 required_kernelcore);
7849
7850
7851 if (end_pfn <= usable_startpfn) {
7852
7853
7854
7855
7856
7857
7858
7859 zone_movable_pfn[nid] = end_pfn;
7860 continue;
7861 }
7862 start_pfn = usable_startpfn;
7863 }
7864
7865
7866
7867
7868
7869
7870 size_pages = end_pfn - start_pfn;
7871 if (size_pages > kernelcore_remaining)
7872 size_pages = kernelcore_remaining;
7873 zone_movable_pfn[nid] = start_pfn + size_pages;
7874
7875
7876
7877
7878
7879
7880 required_kernelcore -= min(required_kernelcore,
7881 size_pages);
7882 kernelcore_remaining -= size_pages;
7883 if (!kernelcore_remaining)
7884 break;
7885 }
7886 }
7887
7888
7889
7890
7891
7892
7893
7894 usable_nodes--;
7895 if (usable_nodes && required_kernelcore > usable_nodes)
7896 goto restart;
7897
7898out2:
7899
7900 for (nid = 0; nid < MAX_NUMNODES; nid++)
7901 zone_movable_pfn[nid] =
7902 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
7903
7904out:
7905
7906 node_states[N_MEMORY] = saved_node_state;
7907}
7908
7909
7910static void check_for_memory(pg_data_t *pgdat, int nid)
7911{
7912 enum zone_type zone_type;
7913
7914 for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) {
7915 struct zone *zone = &pgdat->node_zones[zone_type];
7916 if (populated_zone(zone)) {
7917 if (IS_ENABLED(CONFIG_HIGHMEM))
7918 node_set_state(nid, N_HIGH_MEMORY);
7919 if (zone_type <= ZONE_NORMAL)
7920 node_set_state(nid, N_NORMAL_MEMORY);
7921 break;
7922 }
7923 }
7924}
7925
7926
7927
7928
7929
7930bool __weak arch_has_descending_max_zone_pfns(void)
7931{
7932 return false;
7933}
7934
7935
7936
7937
7938
7939
7940
7941
7942
7943
7944
7945
7946
7947
7948void __init free_area_init(unsigned long *max_zone_pfn)
7949{
7950 unsigned long start_pfn, end_pfn;
7951 int i, nid, zone;
7952 bool descending;
7953
7954
7955 memset(arch_zone_lowest_possible_pfn, 0,
7956 sizeof(arch_zone_lowest_possible_pfn));
7957 memset(arch_zone_highest_possible_pfn, 0,
7958 sizeof(arch_zone_highest_possible_pfn));
7959
7960 start_pfn = find_min_pfn_with_active_regions();
7961 descending = arch_has_descending_max_zone_pfns();
7962
7963 for (i = 0; i < MAX_NR_ZONES; i++) {
7964 if (descending)
7965 zone = MAX_NR_ZONES - i - 1;
7966 else
7967 zone = i;
7968
7969 if (zone == ZONE_MOVABLE)
7970 continue;
7971
7972 end_pfn = max(max_zone_pfn[zone], start_pfn);
7973 arch_zone_lowest_possible_pfn[zone] = start_pfn;
7974 arch_zone_highest_possible_pfn[zone] = end_pfn;
7975
7976 start_pfn = end_pfn;
7977 }
7978
7979
7980 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
7981 find_zone_movable_pfns_for_nodes();
7982
7983
7984 pr_info("Zone ranges:\n");
7985 for (i = 0; i < MAX_NR_ZONES; i++) {
7986 if (i == ZONE_MOVABLE)
7987 continue;
7988 pr_info(" %-8s ", zone_names[i]);
7989 if (arch_zone_lowest_possible_pfn[i] ==
7990 arch_zone_highest_possible_pfn[i])
7991 pr_cont("empty\n");
7992 else
7993 pr_cont("[mem %#018Lx-%#018Lx]\n",
7994 (u64)arch_zone_lowest_possible_pfn[i]
7995 << PAGE_SHIFT,
7996 ((u64)arch_zone_highest_possible_pfn[i]
7997 << PAGE_SHIFT) - 1);
7998 }
7999
8000
8001 pr_info("Movable zone start for each node\n");
8002 for (i = 0; i < MAX_NUMNODES; i++) {
8003 if (zone_movable_pfn[i])
8004 pr_info(" Node %d: %#018Lx\n", i,
8005 (u64)zone_movable_pfn[i] << PAGE_SHIFT);
8006 }
8007
8008
8009
8010
8011
8012
8013 pr_info("Early memory node ranges\n");
8014 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
8015 pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid,
8016 (u64)start_pfn << PAGE_SHIFT,
8017 ((u64)end_pfn << PAGE_SHIFT) - 1);
8018 subsection_map_init(start_pfn, end_pfn - start_pfn);
8019 }
8020
8021
8022 mminit_verify_pageflags_layout();
8023 setup_nr_node_ids();
8024 for_each_online_node(nid) {
8025 pg_data_t *pgdat = NODE_DATA(nid);
8026 free_area_init_node(nid);
8027
8028
8029 if (pgdat->node_present_pages)
8030 node_set_state(nid, N_MEMORY);
8031 check_for_memory(pgdat, nid);
8032 }
8033
8034 memmap_init();
8035}
8036
8037static int __init cmdline_parse_core(char *p, unsigned long *core,
8038 unsigned long *percent)
8039{
8040 unsigned long long coremem;
8041 char *endptr;
8042
8043 if (!p)
8044 return -EINVAL;
8045
8046
8047 coremem = simple_strtoull(p, &endptr, 0);
8048 if (*endptr == '%') {
8049
8050 WARN_ON(coremem > 100);
8051
8052 *percent = coremem;
8053 } else {
8054 coremem = memparse(p, &p);
8055
8056 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
8057
8058 *core = coremem >> PAGE_SHIFT;
8059 *percent = 0UL;
8060 }
8061 return 0;
8062}
8063
8064
8065
8066
8067
8068static int __init cmdline_parse_kernelcore(char *p)
8069{
8070
8071 if (parse_option_str(p, "mirror")) {
8072 mirrored_kernelcore = true;
8073 return 0;
8074 }
8075
8076 return cmdline_parse_core(p, &required_kernelcore,
8077 &required_kernelcore_percent);
8078}
8079
8080
8081
8082
8083
8084static int __init cmdline_parse_movablecore(char *p)
8085{
8086 return cmdline_parse_core(p, &required_movablecore,
8087 &required_movablecore_percent);
8088}
8089
8090early_param("kernelcore", cmdline_parse_kernelcore);
8091early_param("movablecore", cmdline_parse_movablecore);
8092
8093void adjust_managed_page_count(struct page *page, long count)
8094{
8095 atomic_long_add(count, &page_zone(page)->managed_pages);
8096 totalram_pages_add(count);
8097#ifdef CONFIG_HIGHMEM
8098 if (PageHighMem(page))
8099 totalhigh_pages_add(count);
8100#endif
8101}
8102EXPORT_SYMBOL(adjust_managed_page_count);
8103
8104unsigned long free_reserved_area(void *start, void *end, int poison, const char *s)
8105{
8106 void *pos;
8107 unsigned long pages = 0;
8108
8109 start = (void *)PAGE_ALIGN((unsigned long)start);
8110 end = (void *)((unsigned long)end & PAGE_MASK);
8111 for (pos = start; pos < end; pos += PAGE_SIZE, pages++) {
8112 struct page *page = virt_to_page(pos);
8113 void *direct_map_addr;
8114
8115
8116
8117
8118
8119
8120
8121
8122 direct_map_addr = page_address(page);
8123
8124
8125
8126
8127 direct_map_addr = kasan_reset_tag(direct_map_addr);
8128 if ((unsigned int)poison <= 0xFF)
8129 memset(direct_map_addr, poison, PAGE_SIZE);
8130
8131 free_reserved_page(page);
8132 }
8133
8134 if (pages && s)
8135 pr_info("Freeing %s memory: %ldK\n",
8136 s, pages << (PAGE_SHIFT - 10));
8137
8138 return pages;
8139}
8140
8141void __init mem_init_print_info(void)
8142{
8143 unsigned long physpages, codesize, datasize, rosize, bss_size;
8144 unsigned long init_code_size, init_data_size;
8145
8146 physpages = get_num_physpages();
8147 codesize = _etext - _stext;
8148 datasize = _edata - _sdata;
8149 rosize = __end_rodata - __start_rodata;
8150 bss_size = __bss_stop - __bss_start;
8151 init_data_size = __init_end - __init_begin;
8152 init_code_size = _einittext - _sinittext;
8153
8154
8155
8156
8157
8158
8159
8160
8161#define adj_init_size(start, end, size, pos, adj) \
8162 do { \
8163 if (start <= pos && pos < end && size > adj) \
8164 size -= adj; \
8165 } while (0)
8166
8167 adj_init_size(__init_begin, __init_end, init_data_size,
8168 _sinittext, init_code_size);
8169 adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size);
8170 adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size);
8171 adj_init_size(_stext, _etext, codesize, __start_rodata, rosize);
8172 adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize);
8173
8174#undef adj_init_size
8175
8176 pr_info("Memory: %luK/%luK available (%luK kernel code, %luK rwdata, %luK rodata, %luK init, %luK bss, %luK reserved, %luK cma-reserved"
8177#ifdef CONFIG_HIGHMEM
8178 ", %luK highmem"
8179#endif
8180 ")\n",
8181 nr_free_pages() << (PAGE_SHIFT - 10),
8182 physpages << (PAGE_SHIFT - 10),
8183 codesize >> 10, datasize >> 10, rosize >> 10,
8184 (init_data_size + init_code_size) >> 10, bss_size >> 10,
8185 (physpages - totalram_pages() - totalcma_pages) << (PAGE_SHIFT - 10),
8186 totalcma_pages << (PAGE_SHIFT - 10)
8187#ifdef CONFIG_HIGHMEM
8188 , totalhigh_pages() << (PAGE_SHIFT - 10)
8189#endif
8190 );
8191}
8192
8193
8194
8195
8196
8197
8198
8199
8200
8201
8202
8203
8204void __init set_dma_reserve(unsigned long new_dma_reserve)
8205{
8206 dma_reserve = new_dma_reserve;
8207}
8208
8209static int page_alloc_cpu_dead(unsigned int cpu)
8210{
8211 struct zone *zone;
8212
8213 lru_add_drain_cpu(cpu);
8214 drain_pages(cpu);
8215
8216
8217
8218
8219
8220
8221
8222 vm_events_fold_cpu(cpu);
8223
8224
8225
8226
8227
8228
8229
8230
8231 cpu_vm_stats_fold(cpu);
8232
8233 for_each_populated_zone(zone)
8234 zone_pcp_update(zone, 0);
8235
8236 return 0;
8237}
8238
8239static int page_alloc_cpu_online(unsigned int cpu)
8240{
8241 struct zone *zone;
8242
8243 for_each_populated_zone(zone)
8244 zone_pcp_update(zone, 1);
8245 return 0;
8246}
8247
8248#ifdef CONFIG_NUMA
8249int hashdist = HASHDIST_DEFAULT;
8250
8251static int __init set_hashdist(char *str)
8252{
8253 if (!str)
8254 return 0;
8255 hashdist = simple_strtoul(str, &str, 0);
8256 return 1;
8257}
8258__setup("hashdist=", set_hashdist);
8259#endif
8260
8261void __init page_alloc_init(void)
8262{
8263 int ret;
8264
8265#ifdef CONFIG_NUMA
8266 if (num_node_state(N_MEMORY) == 1)
8267 hashdist = 0;
8268#endif
8269
8270 ret = cpuhp_setup_state_nocalls(CPUHP_PAGE_ALLOC,
8271 "mm/page_alloc:pcp",
8272 page_alloc_cpu_online,
8273 page_alloc_cpu_dead);
8274 WARN_ON(ret < 0);
8275}
8276
8277
8278
8279
8280
8281static void calculate_totalreserve_pages(void)
8282{
8283 struct pglist_data *pgdat;
8284 unsigned long reserve_pages = 0;
8285 enum zone_type i, j;
8286
8287 for_each_online_pgdat(pgdat) {
8288
8289 pgdat->totalreserve_pages = 0;
8290
8291 for (i = 0; i < MAX_NR_ZONES; i++) {
8292 struct zone *zone = pgdat->node_zones + i;
8293 long max = 0;
8294 unsigned long managed_pages = zone_managed_pages(zone);
8295
8296
8297 for (j = i; j < MAX_NR_ZONES; j++) {
8298 if (zone->lowmem_reserve[j] > max)
8299 max = zone->lowmem_reserve[j];
8300 }
8301
8302
8303 max += high_wmark_pages(zone);
8304
8305 if (max > managed_pages)
8306 max = managed_pages;
8307
8308 pgdat->totalreserve_pages += max;
8309
8310 reserve_pages += max;
8311 }
8312 }
8313 totalreserve_pages = reserve_pages;
8314}
8315
8316
8317
8318
8319
8320
8321
8322static void setup_per_zone_lowmem_reserve(void)
8323{
8324 struct pglist_data *pgdat;
8325 enum zone_type i, j;
8326
8327 for_each_online_pgdat(pgdat) {
8328 for (i = 0; i < MAX_NR_ZONES - 1; i++) {
8329 struct zone *zone = &pgdat->node_zones[i];
8330 int ratio = sysctl_lowmem_reserve_ratio[i];
8331 bool clear = !ratio || !zone_managed_pages(zone);
8332 unsigned long managed_pages = 0;
8333
8334 for (j = i + 1; j < MAX_NR_ZONES; j++) {
8335 struct zone *upper_zone = &pgdat->node_zones[j];
8336
8337 managed_pages += zone_managed_pages(upper_zone);
8338
8339 if (clear)
8340 zone->lowmem_reserve[j] = 0;
8341 else
8342 zone->lowmem_reserve[j] = managed_pages / ratio;
8343 }
8344 }
8345 }
8346
8347
8348 calculate_totalreserve_pages();
8349}
8350
8351static void __setup_per_zone_wmarks(void)
8352{
8353 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
8354 unsigned long lowmem_pages = 0;
8355 struct zone *zone;
8356 unsigned long flags;
8357
8358
8359 for_each_zone(zone) {
8360 if (!is_highmem(zone))
8361 lowmem_pages += zone_managed_pages(zone);
8362 }
8363
8364 for_each_zone(zone) {
8365 u64 tmp;
8366
8367 spin_lock_irqsave(&zone->lock, flags);
8368 tmp = (u64)pages_min * zone_managed_pages(zone);
8369 do_div(tmp, lowmem_pages);
8370 if (is_highmem(zone)) {
8371
8372
8373
8374
8375
8376
8377
8378
8379
8380 unsigned long min_pages;
8381
8382 min_pages = zone_managed_pages(zone) / 1024;
8383 min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL);
8384 zone->_watermark[WMARK_MIN] = min_pages;
8385 } else {
8386
8387
8388
8389
8390 zone->_watermark[WMARK_MIN] = tmp;
8391 }
8392
8393
8394
8395
8396
8397
8398 tmp = max_t(u64, tmp >> 2,
8399 mult_frac(zone_managed_pages(zone),
8400 watermark_scale_factor, 10000));
8401
8402 zone->watermark_boost = 0;
8403 zone->_watermark[WMARK_LOW] = min_wmark_pages(zone) + tmp;
8404 zone->_watermark[WMARK_HIGH] = min_wmark_pages(zone) + tmp * 2;
8405
8406 spin_unlock_irqrestore(&zone->lock, flags);
8407 }
8408
8409
8410 calculate_totalreserve_pages();
8411}
8412
8413
8414
8415
8416
8417
8418
8419
8420void setup_per_zone_wmarks(void)
8421{
8422 struct zone *zone;
8423 static DEFINE_SPINLOCK(lock);
8424
8425 spin_lock(&lock);
8426 __setup_per_zone_wmarks();
8427 spin_unlock(&lock);
8428
8429
8430
8431
8432
8433 for_each_zone(zone)
8434 zone_pcp_update(zone, 0);
8435}
8436
8437
8438
8439
8440
8441
8442
8443
8444
8445
8446
8447
8448
8449
8450
8451
8452
8453
8454
8455
8456
8457
8458
8459
8460
8461int __meminit init_per_zone_wmark_min(void)
8462{
8463 unsigned long lowmem_kbytes;
8464 int new_min_free_kbytes;
8465
8466 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
8467 new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
8468
8469 if (new_min_free_kbytes > user_min_free_kbytes) {
8470 min_free_kbytes = new_min_free_kbytes;
8471 if (min_free_kbytes < 128)
8472 min_free_kbytes = 128;
8473 if (min_free_kbytes > 262144)
8474 min_free_kbytes = 262144;
8475 } else {
8476 pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n",
8477 new_min_free_kbytes, user_min_free_kbytes);
8478 }
8479 setup_per_zone_wmarks();
8480 refresh_zone_stat_thresholds();
8481 setup_per_zone_lowmem_reserve();
8482
8483#ifdef CONFIG_NUMA
8484 setup_min_unmapped_ratio();
8485 setup_min_slab_ratio();
8486#endif
8487
8488 khugepaged_min_free_kbytes_update();
8489
8490 return 0;
8491}
8492postcore_initcall(init_per_zone_wmark_min)
8493
8494
8495
8496
8497
8498
8499int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write,
8500 void *buffer, size_t *length, loff_t *ppos)
8501{
8502 int rc;
8503
8504 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
8505 if (rc)
8506 return rc;
8507
8508 if (write) {
8509 user_min_free_kbytes = min_free_kbytes;
8510 setup_per_zone_wmarks();
8511 }
8512 return 0;
8513}
8514
8515int watermark_scale_factor_sysctl_handler(struct ctl_table *table, int write,
8516 void *buffer, size_t *length, loff_t *ppos)
8517{
8518 int rc;
8519
8520 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
8521 if (rc)
8522 return rc;
8523
8524 if (write)
8525 setup_per_zone_wmarks();
8526
8527 return 0;
8528}
8529
8530#ifdef CONFIG_NUMA
8531static void setup_min_unmapped_ratio(void)
8532{
8533 pg_data_t *pgdat;
8534 struct zone *zone;
8535
8536 for_each_online_pgdat(pgdat)
8537 pgdat->min_unmapped_pages = 0;
8538
8539 for_each_zone(zone)
8540 zone->zone_pgdat->min_unmapped_pages += (zone_managed_pages(zone) *
8541 sysctl_min_unmapped_ratio) / 100;
8542}
8543
8544
8545int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write,
8546 void *buffer, size_t *length, loff_t *ppos)
8547{
8548 int rc;
8549
8550 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
8551 if (rc)
8552 return rc;
8553
8554 setup_min_unmapped_ratio();
8555
8556 return 0;
8557}
8558
8559static void setup_min_slab_ratio(void)
8560{
8561 pg_data_t *pgdat;
8562 struct zone *zone;
8563
8564 for_each_online_pgdat(pgdat)
8565 pgdat->min_slab_pages = 0;
8566
8567 for_each_zone(zone)
8568 zone->zone_pgdat->min_slab_pages += (zone_managed_pages(zone) *
8569 sysctl_min_slab_ratio) / 100;
8570}
8571
8572int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write,
8573 void *buffer, size_t *length, loff_t *ppos)
8574{
8575 int rc;
8576
8577 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
8578 if (rc)
8579 return rc;
8580
8581 setup_min_slab_ratio();
8582
8583 return 0;
8584}
8585#endif
8586
8587
8588
8589
8590
8591
8592
8593
8594
8595
8596int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write,
8597 void *buffer, size_t *length, loff_t *ppos)
8598{
8599 int i;
8600
8601 proc_dointvec_minmax(table, write, buffer, length, ppos);
8602
8603 for (i = 0; i < MAX_NR_ZONES; i++) {
8604 if (sysctl_lowmem_reserve_ratio[i] < 1)
8605 sysctl_lowmem_reserve_ratio[i] = 0;
8606 }
8607
8608 setup_per_zone_lowmem_reserve();
8609 return 0;
8610}
8611
8612
8613
8614
8615
8616
8617int percpu_pagelist_high_fraction_sysctl_handler(struct ctl_table *table,
8618 int write, void *buffer, size_t *length, loff_t *ppos)
8619{
8620 struct zone *zone;
8621 int old_percpu_pagelist_high_fraction;
8622 int ret;
8623
8624 mutex_lock(&pcp_batch_high_lock);
8625 old_percpu_pagelist_high_fraction = percpu_pagelist_high_fraction;
8626
8627 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
8628 if (!write || ret < 0)
8629 goto out;
8630
8631
8632 if (percpu_pagelist_high_fraction &&
8633 percpu_pagelist_high_fraction < MIN_PERCPU_PAGELIST_HIGH_FRACTION) {
8634 percpu_pagelist_high_fraction = old_percpu_pagelist_high_fraction;
8635 ret = -EINVAL;
8636 goto out;
8637 }
8638
8639
8640 if (percpu_pagelist_high_fraction == old_percpu_pagelist_high_fraction)
8641 goto out;
8642
8643 for_each_populated_zone(zone)
8644 zone_set_pageset_high_and_batch(zone, 0);
8645out:
8646 mutex_unlock(&pcp_batch_high_lock);
8647 return ret;
8648}
8649
8650#ifndef __HAVE_ARCH_RESERVED_KERNEL_PAGES
8651
8652
8653
8654
8655static unsigned long __init arch_reserved_kernel_pages(void)
8656{
8657 return 0;
8658}
8659#endif
8660
8661
8662
8663
8664
8665
8666
8667
8668
8669
8670#if __BITS_PER_LONG > 32
8671#define ADAPT_SCALE_BASE (64ul << 30)
8672#define ADAPT_SCALE_SHIFT 2
8673#define ADAPT_SCALE_NPAGES (ADAPT_SCALE_BASE >> PAGE_SHIFT)
8674#endif
8675
8676
8677
8678
8679
8680
8681
8682void *__init alloc_large_system_hash(const char *tablename,
8683 unsigned long bucketsize,
8684 unsigned long numentries,
8685 int scale,
8686 int flags,
8687 unsigned int *_hash_shift,
8688 unsigned int *_hash_mask,
8689 unsigned long low_limit,
8690 unsigned long high_limit)
8691{
8692 unsigned long long max = high_limit;
8693 unsigned long log2qty, size;
8694 void *table = NULL;
8695 gfp_t gfp_flags;
8696 bool virt;
8697 bool huge;
8698
8699
8700 if (!numentries) {
8701
8702 numentries = nr_kernel_pages;
8703 numentries -= arch_reserved_kernel_pages();
8704
8705
8706 if (PAGE_SHIFT < 20)
8707 numentries = round_up(numentries, (1<<20)/PAGE_SIZE);
8708
8709#if __BITS_PER_LONG > 32
8710 if (!high_limit) {
8711 unsigned long adapt;
8712
8713 for (adapt = ADAPT_SCALE_NPAGES; adapt < numentries;
8714 adapt <<= ADAPT_SCALE_SHIFT)
8715 scale++;
8716 }
8717#endif
8718
8719
8720 if (scale > PAGE_SHIFT)
8721 numentries >>= (scale - PAGE_SHIFT);
8722 else
8723 numentries <<= (PAGE_SHIFT - scale);
8724
8725
8726 if (unlikely(flags & HASH_SMALL)) {
8727
8728 WARN_ON(!(flags & HASH_EARLY));
8729 if (!(numentries >> *_hash_shift)) {
8730 numentries = 1UL << *_hash_shift;
8731 BUG_ON(!numentries);
8732 }
8733 } else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
8734 numentries = PAGE_SIZE / bucketsize;
8735 }
8736 numentries = roundup_pow_of_two(numentries);
8737
8738
8739 if (max == 0) {
8740 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
8741 do_div(max, bucketsize);
8742 }
8743 max = min(max, 0x80000000ULL);
8744
8745 if (numentries < low_limit)
8746 numentries = low_limit;
8747 if (numentries > max)
8748 numentries = max;
8749
8750 log2qty = ilog2(numentries);
8751
8752 gfp_flags = (flags & HASH_ZERO) ? GFP_ATOMIC | __GFP_ZERO : GFP_ATOMIC;
8753 do {
8754 virt = false;
8755 size = bucketsize << log2qty;
8756 if (flags & HASH_EARLY) {
8757 if (flags & HASH_ZERO)
8758 table = memblock_alloc(size, SMP_CACHE_BYTES);
8759 else
8760 table = memblock_alloc_raw(size,
8761 SMP_CACHE_BYTES);
8762 } else if (get_order(size) >= MAX_ORDER || hashdist) {
8763 table = __vmalloc(size, gfp_flags);
8764 virt = true;
8765 huge = is_vm_area_hugepages(table);
8766 } else {
8767
8768
8769
8770
8771
8772 table = alloc_pages_exact(size, gfp_flags);
8773 kmemleak_alloc(table, size, 1, gfp_flags);
8774 }
8775 } while (!table && size > PAGE_SIZE && --log2qty);
8776
8777 if (!table)
8778 panic("Failed to allocate %s hash table\n", tablename);
8779
8780 pr_info("%s hash table entries: %ld (order: %d, %lu bytes, %s)\n",
8781 tablename, 1UL << log2qty, ilog2(size) - PAGE_SHIFT, size,
8782 virt ? (huge ? "vmalloc hugepage" : "vmalloc") : "linear");
8783
8784 if (_hash_shift)
8785 *_hash_shift = log2qty;
8786 if (_hash_mask)
8787 *_hash_mask = (1 << log2qty) - 1;
8788
8789 return table;
8790}
8791
8792
8793
8794
8795
8796
8797
8798
8799
8800
8801
8802
8803
8804
8805struct page *has_unmovable_pages(struct zone *zone, struct page *page,
8806 int migratetype, int flags)
8807{
8808 unsigned long iter = 0;
8809 unsigned long pfn = page_to_pfn(page);
8810 unsigned long offset = pfn % pageblock_nr_pages;
8811
8812 if (is_migrate_cma_page(page)) {
8813
8814
8815
8816
8817
8818 if (is_migrate_cma(migratetype))
8819 return NULL;
8820
8821 return page;
8822 }
8823
8824 for (; iter < pageblock_nr_pages - offset; iter++) {
8825 page = pfn_to_page(pfn + iter);
8826
8827
8828
8829
8830
8831
8832
8833 if (PageReserved(page))
8834 return page;
8835
8836
8837
8838
8839
8840
8841 if (zone_idx(zone) == ZONE_MOVABLE)
8842 continue;
8843
8844
8845
8846
8847
8848
8849
8850 if (PageHuge(page) || PageTransCompound(page)) {
8851 struct page *head = compound_head(page);
8852 unsigned int skip_pages;
8853
8854 if (PageHuge(page)) {
8855 if (!hugepage_migration_supported(page_hstate(head)))
8856 return page;
8857 } else if (!PageLRU(head) && !__PageMovable(head)) {
8858 return page;
8859 }
8860
8861 skip_pages = compound_nr(head) - (page - head);
8862 iter += skip_pages - 1;
8863 continue;
8864 }
8865
8866
8867
8868
8869
8870
8871
8872 if (!page_ref_count(page)) {
8873 if (PageBuddy(page))
8874 iter += (1 << buddy_order(page)) - 1;
8875 continue;
8876 }
8877
8878
8879
8880
8881
8882 if ((flags & MEMORY_OFFLINE) && PageHWPoison(page))
8883 continue;
8884
8885
8886
8887
8888
8889
8890
8891
8892
8893
8894
8895 if ((flags & MEMORY_OFFLINE) && PageOffline(page))
8896 continue;
8897
8898 if (__PageMovable(page) || PageLRU(page))
8899 continue;
8900
8901
8902
8903
8904
8905
8906 return page;
8907 }
8908 return NULL;
8909}
8910
8911#ifdef CONFIG_CONTIG_ALLOC
8912static unsigned long pfn_max_align_down(unsigned long pfn)
8913{
8914 return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES,
8915 pageblock_nr_pages) - 1);
8916}
8917
8918static unsigned long pfn_max_align_up(unsigned long pfn)
8919{
8920 return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES,
8921 pageblock_nr_pages));
8922}
8923
8924#if defined(CONFIG_DYNAMIC_DEBUG) || \
8925 (defined(CONFIG_DYNAMIC_DEBUG_CORE) && defined(DYNAMIC_DEBUG_MODULE))
8926
8927static void alloc_contig_dump_pages(struct list_head *page_list)
8928{
8929 DEFINE_DYNAMIC_DEBUG_METADATA(descriptor, "migrate failure");
8930
8931 if (DYNAMIC_DEBUG_BRANCH(descriptor)) {
8932 struct page *page;
8933
8934 dump_stack();
8935 list_for_each_entry(page, page_list, lru)
8936 dump_page(page, "migration failure");
8937 }
8938}
8939#else
8940static inline void alloc_contig_dump_pages(struct list_head *page_list)
8941{
8942}
8943#endif
8944
8945
8946static int __alloc_contig_migrate_range(struct compact_control *cc,
8947 unsigned long start, unsigned long end)
8948{
8949
8950 unsigned int nr_reclaimed;
8951 unsigned long pfn = start;
8952 unsigned int tries = 0;
8953 int ret = 0;
8954 struct migration_target_control mtc = {
8955 .nid = zone_to_nid(cc->zone),
8956 .gfp_mask = GFP_USER | __GFP_MOVABLE | __GFP_RETRY_MAYFAIL,
8957 };
8958
8959 lru_cache_disable();
8960
8961 while (pfn < end || !list_empty(&cc->migratepages)) {
8962 if (fatal_signal_pending(current)) {
8963 ret = -EINTR;
8964 break;
8965 }
8966
8967 if (list_empty(&cc->migratepages)) {
8968 cc->nr_migratepages = 0;
8969 ret = isolate_migratepages_range(cc, pfn, end);
8970 if (ret && ret != -EAGAIN)
8971 break;
8972 pfn = cc->migrate_pfn;
8973 tries = 0;
8974 } else if (++tries == 5) {
8975 ret = -EBUSY;
8976 break;
8977 }
8978
8979 nr_reclaimed = reclaim_clean_pages_from_list(cc->zone,
8980 &cc->migratepages);
8981 cc->nr_migratepages -= nr_reclaimed;
8982
8983 ret = migrate_pages(&cc->migratepages, alloc_migration_target,
8984 NULL, (unsigned long)&mtc, cc->mode, MR_CONTIG_RANGE, NULL);
8985
8986
8987
8988
8989
8990 if (ret == -ENOMEM)
8991 break;
8992 }
8993
8994 lru_cache_enable();
8995 if (ret < 0) {
8996 if (ret == -EBUSY)
8997 alloc_contig_dump_pages(&cc->migratepages);
8998 putback_movable_pages(&cc->migratepages);
8999 return ret;
9000 }
9001 return 0;
9002}
9003
9004
9005
9006
9007
9008
9009
9010
9011
9012
9013
9014
9015
9016
9017
9018
9019
9020
9021
9022
9023
9024
9025int alloc_contig_range(unsigned long start, unsigned long end,
9026 unsigned migratetype, gfp_t gfp_mask)
9027{
9028 unsigned long outer_start, outer_end;
9029 unsigned int order;
9030 int ret = 0;
9031
9032 struct compact_control cc = {
9033 .nr_migratepages = 0,
9034 .order = -1,
9035 .zone = page_zone(pfn_to_page(start)),
9036 .mode = MIGRATE_SYNC,
9037 .ignore_skip_hint = true,
9038 .no_set_skip_hint = true,
9039 .gfp_mask = current_gfp_context(gfp_mask),
9040 .alloc_contig = true,
9041 };
9042 INIT_LIST_HEAD(&cc.migratepages);
9043
9044
9045
9046
9047
9048
9049
9050
9051
9052
9053
9054
9055
9056
9057
9058
9059
9060
9061
9062
9063
9064
9065
9066
9067
9068 ret = start_isolate_page_range(pfn_max_align_down(start),
9069 pfn_max_align_up(end), migratetype, 0);
9070 if (ret)
9071 return ret;
9072
9073 drain_all_pages(cc.zone);
9074
9075
9076
9077
9078
9079
9080
9081
9082
9083
9084
9085 ret = __alloc_contig_migrate_range(&cc, start, end);
9086 if (ret && ret != -EBUSY)
9087 goto done;
9088 ret = 0;
9089
9090
9091
9092
9093
9094
9095
9096
9097
9098
9099
9100
9101
9102
9103
9104
9105
9106
9107 order = 0;
9108 outer_start = start;
9109 while (!PageBuddy(pfn_to_page(outer_start))) {
9110 if (++order >= MAX_ORDER) {
9111 outer_start = start;
9112 break;
9113 }
9114 outer_start &= ~0UL << order;
9115 }
9116
9117 if (outer_start != start) {
9118 order = buddy_order(pfn_to_page(outer_start));
9119
9120
9121
9122
9123
9124
9125
9126 if (outer_start + (1UL << order) <= start)
9127 outer_start = start;
9128 }
9129
9130
9131 if (test_pages_isolated(outer_start, end, 0)) {
9132 ret = -EBUSY;
9133 goto done;
9134 }
9135
9136
9137 outer_end = isolate_freepages_range(&cc, outer_start, end);
9138 if (!outer_end) {
9139 ret = -EBUSY;
9140 goto done;
9141 }
9142
9143
9144 if (start != outer_start)
9145 free_contig_range(outer_start, start - outer_start);
9146 if (end != outer_end)
9147 free_contig_range(end, outer_end - end);
9148
9149done:
9150 undo_isolate_page_range(pfn_max_align_down(start),
9151 pfn_max_align_up(end), migratetype);
9152 return ret;
9153}
9154EXPORT_SYMBOL(alloc_contig_range);
9155
9156static int __alloc_contig_pages(unsigned long start_pfn,
9157 unsigned long nr_pages, gfp_t gfp_mask)
9158{
9159 unsigned long end_pfn = start_pfn + nr_pages;
9160
9161 return alloc_contig_range(start_pfn, end_pfn, MIGRATE_MOVABLE,
9162 gfp_mask);
9163}
9164
9165static bool pfn_range_valid_contig(struct zone *z, unsigned long start_pfn,
9166 unsigned long nr_pages)
9167{
9168 unsigned long i, end_pfn = start_pfn + nr_pages;
9169 struct page *page;
9170
9171 for (i = start_pfn; i < end_pfn; i++) {
9172 page = pfn_to_online_page(i);
9173 if (!page)
9174 return false;
9175
9176 if (page_zone(page) != z)
9177 return false;
9178
9179 if (PageReserved(page))
9180 return false;
9181 }
9182 return true;
9183}
9184
9185static bool zone_spans_last_pfn(const struct zone *zone,
9186 unsigned long start_pfn, unsigned long nr_pages)
9187{
9188 unsigned long last_pfn = start_pfn + nr_pages - 1;
9189
9190 return zone_spans_pfn(zone, last_pfn);
9191}
9192
9193
9194
9195
9196
9197
9198
9199
9200
9201
9202
9203
9204
9205
9206
9207
9208
9209
9210
9211
9212
9213
9214struct page *alloc_contig_pages(unsigned long nr_pages, gfp_t gfp_mask,
9215 int nid, nodemask_t *nodemask)
9216{
9217 unsigned long ret, pfn, flags;
9218 struct zonelist *zonelist;
9219 struct zone *zone;
9220 struct zoneref *z;
9221
9222 zonelist = node_zonelist(nid, gfp_mask);
9223 for_each_zone_zonelist_nodemask(zone, z, zonelist,
9224 gfp_zone(gfp_mask), nodemask) {
9225 spin_lock_irqsave(&zone->lock, flags);
9226
9227 pfn = ALIGN(zone->zone_start_pfn, nr_pages);
9228 while (zone_spans_last_pfn(zone, pfn, nr_pages)) {
9229 if (pfn_range_valid_contig(zone, pfn, nr_pages)) {
9230
9231
9232
9233
9234
9235
9236
9237 spin_unlock_irqrestore(&zone->lock, flags);
9238 ret = __alloc_contig_pages(pfn, nr_pages,
9239 gfp_mask);
9240 if (!ret)
9241 return pfn_to_page(pfn);
9242 spin_lock_irqsave(&zone->lock, flags);
9243 }
9244 pfn += nr_pages;
9245 }
9246 spin_unlock_irqrestore(&zone->lock, flags);
9247 }
9248 return NULL;
9249}
9250#endif
9251
9252void free_contig_range(unsigned long pfn, unsigned long nr_pages)
9253{
9254 unsigned long count = 0;
9255
9256 for (; nr_pages--; pfn++) {
9257 struct page *page = pfn_to_page(pfn);
9258
9259 count += page_count(page) != 1;
9260 __free_page(page);
9261 }
9262 WARN(count != 0, "%lu pages are still in use!\n", count);
9263}
9264EXPORT_SYMBOL(free_contig_range);
9265
9266
9267
9268
9269
9270void zone_pcp_update(struct zone *zone, int cpu_online)
9271{
9272 mutex_lock(&pcp_batch_high_lock);
9273 zone_set_pageset_high_and_batch(zone, cpu_online);
9274 mutex_unlock(&pcp_batch_high_lock);
9275}
9276
9277
9278
9279
9280
9281
9282
9283
9284
9285void zone_pcp_disable(struct zone *zone)
9286{
9287 mutex_lock(&pcp_batch_high_lock);
9288 __zone_set_pageset_high_and_batch(zone, 0, 1);
9289 __drain_all_pages(zone, true);
9290}
9291
9292void zone_pcp_enable(struct zone *zone)
9293{
9294 __zone_set_pageset_high_and_batch(zone, zone->pageset_high, zone->pageset_batch);
9295 mutex_unlock(&pcp_batch_high_lock);
9296}
9297
9298void zone_pcp_reset(struct zone *zone)
9299{
9300 int cpu;
9301 struct per_cpu_zonestat *pzstats;
9302
9303 if (zone->per_cpu_pageset != &boot_pageset) {
9304 for_each_online_cpu(cpu) {
9305 pzstats = per_cpu_ptr(zone->per_cpu_zonestats, cpu);
9306 drain_zonestat(zone, pzstats);
9307 }
9308 free_percpu(zone->per_cpu_pageset);
9309 free_percpu(zone->per_cpu_zonestats);
9310 zone->per_cpu_pageset = &boot_pageset;
9311 zone->per_cpu_zonestats = &boot_zonestats;
9312 }
9313}
9314
9315#ifdef CONFIG_MEMORY_HOTREMOVE
9316
9317
9318
9319
9320void __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
9321{
9322 unsigned long pfn = start_pfn;
9323 struct page *page;
9324 struct zone *zone;
9325 unsigned int order;
9326 unsigned long flags;
9327
9328 offline_mem_sections(pfn, end_pfn);
9329 zone = page_zone(pfn_to_page(pfn));
9330 spin_lock_irqsave(&zone->lock, flags);
9331 while (pfn < end_pfn) {
9332 page = pfn_to_page(pfn);
9333
9334
9335
9336
9337 if (unlikely(!PageBuddy(page) && PageHWPoison(page))) {
9338 pfn++;
9339 continue;
9340 }
9341
9342
9343
9344
9345 if (PageOffline(page)) {
9346 BUG_ON(page_count(page));
9347 BUG_ON(PageBuddy(page));
9348 pfn++;
9349 continue;
9350 }
9351
9352 BUG_ON(page_count(page));
9353 BUG_ON(!PageBuddy(page));
9354 order = buddy_order(page);
9355 del_page_from_free_list(page, zone, order);
9356 pfn += (1 << order);
9357 }
9358 spin_unlock_irqrestore(&zone->lock, flags);
9359}
9360#endif
9361
9362bool is_free_buddy_page(struct page *page)
9363{
9364 struct zone *zone = page_zone(page);
9365 unsigned long pfn = page_to_pfn(page);
9366 unsigned long flags;
9367 unsigned int order;
9368
9369 spin_lock_irqsave(&zone->lock, flags);
9370 for (order = 0; order < MAX_ORDER; order++) {
9371 struct page *page_head = page - (pfn & ((1 << order) - 1));
9372
9373 if (PageBuddy(page_head) && buddy_order(page_head) >= order)
9374 break;
9375 }
9376 spin_unlock_irqrestore(&zone->lock, flags);
9377
9378 return order < MAX_ORDER;
9379}
9380
9381#ifdef CONFIG_MEMORY_FAILURE
9382
9383
9384
9385
9386static void break_down_buddy_pages(struct zone *zone, struct page *page,
9387 struct page *target, int low, int high,
9388 int migratetype)
9389{
9390 unsigned long size = 1 << high;
9391 struct page *current_buddy, *next_page;
9392
9393 while (high > low) {
9394 high--;
9395 size >>= 1;
9396
9397 if (target >= &page[size]) {
9398 next_page = page + size;
9399 current_buddy = page;
9400 } else {
9401 next_page = page;
9402 current_buddy = page + size;
9403 }
9404
9405 if (set_page_guard(zone, current_buddy, high, migratetype))
9406 continue;
9407
9408 if (current_buddy != target) {
9409 add_to_free_list(current_buddy, zone, high, migratetype);
9410 set_buddy_order(current_buddy, high);
9411 page = next_page;
9412 }
9413 }
9414}
9415
9416
9417
9418
9419bool take_page_off_buddy(struct page *page)
9420{
9421 struct zone *zone = page_zone(page);
9422 unsigned long pfn = page_to_pfn(page);
9423 unsigned long flags;
9424 unsigned int order;
9425 bool ret = false;
9426
9427 spin_lock_irqsave(&zone->lock, flags);
9428 for (order = 0; order < MAX_ORDER; order++) {
9429 struct page *page_head = page - (pfn & ((1 << order) - 1));
9430 int page_order = buddy_order(page_head);
9431
9432 if (PageBuddy(page_head) && page_order >= order) {
9433 unsigned long pfn_head = page_to_pfn(page_head);
9434 int migratetype = get_pfnblock_migratetype(page_head,
9435 pfn_head);
9436
9437 del_page_from_free_list(page_head, zone, page_order);
9438 break_down_buddy_pages(zone, page_head, page, 0,
9439 page_order, migratetype);
9440 if (!is_migrate_isolate(migratetype))
9441 __mod_zone_freepage_state(zone, -1, migratetype);
9442 ret = true;
9443 break;
9444 }
9445 if (page_count(page_head) > 0)
9446 break;
9447 }
9448 spin_unlock_irqrestore(&zone->lock, flags);
9449 return ret;
9450}
9451#endif
9452