1
2
3
4
5
6
7
8#include <linux/stddef.h>
9#include <linux/mm.h>
10#include <linux/sched/signal.h>
11#include <linux/swap.h>
12#include <linux/interrupt.h>
13#include <linux/pagemap.h>
14#include <linux/compiler.h>
15#include <linux/export.h>
16#include <linux/pagevec.h>
17#include <linux/writeback.h>
18#include <linux/slab.h>
19#include <linux/sysctl.h>
20#include <linux/cpu.h>
21#include <linux/memory.h>
22#include <linux/memremap.h>
23#include <linux/memory_hotplug.h>
24#include <linux/highmem.h>
25#include <linux/vmalloc.h>
26#include <linux/ioport.h>
27#include <linux/delay.h>
28#include <linux/migrate.h>
29#include <linux/page-isolation.h>
30#include <linux/pfn.h>
31#include <linux/suspend.h>
32#include <linux/mm_inline.h>
33#include <linux/firmware-map.h>
34#include <linux/stop_machine.h>
35#include <linux/hugetlb.h>
36#include <linux/memblock.h>
37#include <linux/compaction.h>
38#include <linux/rmap.h>
39
40#include <asm/tlbflush.h>
41
42#include "internal.h"
43#include "shuffle.h"
44
45
46
47
48
49static bool memmap_on_memory __ro_after_init;
50#ifdef CONFIG_MHP_MEMMAP_ON_MEMORY
51module_param(memmap_on_memory, bool, 0444);
52MODULE_PARM_DESC(memmap_on_memory, "Enable memmap on memory for memory hotplug");
53#endif
54
55
56
57
58
59
60
61
62static online_page_callback_t online_page_callback = generic_online_page;
63static DEFINE_MUTEX(online_page_callback_lock);
64
65DEFINE_STATIC_PERCPU_RWSEM(mem_hotplug_lock);
66
67void get_online_mems(void)
68{
69 percpu_down_read(&mem_hotplug_lock);
70}
71
72void put_online_mems(void)
73{
74 percpu_up_read(&mem_hotplug_lock);
75}
76
77bool movable_node_enabled = false;
78
79#ifndef CONFIG_MEMORY_HOTPLUG_DEFAULT_ONLINE
80int mhp_default_online_type = MMOP_OFFLINE;
81#else
82int mhp_default_online_type = MMOP_ONLINE;
83#endif
84
85static int __init setup_memhp_default_state(char *str)
86{
87 const int online_type = mhp_online_type_from_str(str);
88
89 if (online_type >= 0)
90 mhp_default_online_type = online_type;
91
92 return 1;
93}
94__setup("memhp_default_state=", setup_memhp_default_state);
95
96void mem_hotplug_begin(void)
97{
98 cpus_read_lock();
99 percpu_down_write(&mem_hotplug_lock);
100}
101
102void mem_hotplug_done(void)
103{
104 percpu_up_write(&mem_hotplug_lock);
105 cpus_read_unlock();
106}
107
108u64 max_mem_size = U64_MAX;
109
110
111static struct resource *register_memory_resource(u64 start, u64 size,
112 const char *resource_name)
113{
114 struct resource *res;
115 unsigned long flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
116
117 if (strcmp(resource_name, "System RAM"))
118 flags |= IORESOURCE_SYSRAM_DRIVER_MANAGED;
119
120 if (!mhp_range_allowed(start, size, true))
121 return ERR_PTR(-E2BIG);
122
123
124
125
126
127
128
129 if (start + size > max_mem_size && system_state < SYSTEM_RUNNING)
130 return ERR_PTR(-E2BIG);
131
132
133
134
135
136
137 res = __request_region(&iomem_resource, start, size,
138 resource_name, flags);
139
140 if (!res) {
141 pr_debug("Unable to reserve System RAM region: %016llx->%016llx\n",
142 start, start + size);
143 return ERR_PTR(-EEXIST);
144 }
145 return res;
146}
147
148static void release_memory_resource(struct resource *res)
149{
150 if (!res)
151 return;
152 release_resource(res);
153 kfree(res);
154}
155
156#ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
157static int check_pfn_span(unsigned long pfn, unsigned long nr_pages,
158 const char *reason)
159{
160
161
162
163
164
165
166
167
168
169 unsigned long min_align;
170
171 if (IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP))
172 min_align = PAGES_PER_SUBSECTION;
173 else
174 min_align = PAGES_PER_SECTION;
175 if (!IS_ALIGNED(pfn, min_align)
176 || !IS_ALIGNED(nr_pages, min_align)) {
177 WARN(1, "Misaligned __%s_pages start: %#lx end: #%lx\n",
178 reason, pfn, pfn + nr_pages - 1);
179 return -EINVAL;
180 }
181 return 0;
182}
183
184
185
186
187
188
189struct page *pfn_to_online_page(unsigned long pfn)
190{
191 unsigned long nr = pfn_to_section_nr(pfn);
192 struct dev_pagemap *pgmap;
193 struct mem_section *ms;
194
195 if (nr >= NR_MEM_SECTIONS)
196 return NULL;
197
198 ms = __nr_to_section(nr);
199 if (!online_section(ms))
200 return NULL;
201
202
203
204
205
206 if (IS_ENABLED(CONFIG_HAVE_ARCH_PFN_VALID) && !pfn_valid(pfn))
207 return NULL;
208
209 if (!pfn_section_valid(ms, pfn))
210 return NULL;
211
212 if (!online_device_section(ms))
213 return pfn_to_page(pfn);
214
215
216
217
218
219
220
221 pgmap = get_dev_pagemap(pfn, NULL);
222 put_dev_pagemap(pgmap);
223
224
225 if (pgmap)
226 return NULL;
227
228 return pfn_to_page(pfn);
229}
230EXPORT_SYMBOL_GPL(pfn_to_online_page);
231
232
233
234
235
236
237
238int __ref __add_pages(int nid, unsigned long pfn, unsigned long nr_pages,
239 struct mhp_params *params)
240{
241 const unsigned long end_pfn = pfn + nr_pages;
242 unsigned long cur_nr_pages;
243 int err;
244 struct vmem_altmap *altmap = params->altmap;
245
246 if (WARN_ON_ONCE(!params->pgprot.pgprot))
247 return -EINVAL;
248
249 VM_BUG_ON(!mhp_range_allowed(PFN_PHYS(pfn), nr_pages * PAGE_SIZE, false));
250
251 if (altmap) {
252
253
254
255 if (altmap->base_pfn != pfn
256 || vmem_altmap_offset(altmap) > nr_pages) {
257 pr_warn_once("memory add fail, invalid altmap\n");
258 return -EINVAL;
259 }
260 altmap->alloc = 0;
261 }
262
263 err = check_pfn_span(pfn, nr_pages, "add");
264 if (err)
265 return err;
266
267 for (; pfn < end_pfn; pfn += cur_nr_pages) {
268
269 cur_nr_pages = min(end_pfn - pfn,
270 SECTION_ALIGN_UP(pfn + 1) - pfn);
271 err = sparse_add_section(nid, pfn, cur_nr_pages, altmap);
272 if (err)
273 break;
274 cond_resched();
275 }
276 vmemmap_populate_print_last();
277 return err;
278}
279
280
281static unsigned long find_smallest_section_pfn(int nid, struct zone *zone,
282 unsigned long start_pfn,
283 unsigned long end_pfn)
284{
285 for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SUBSECTION) {
286 if (unlikely(!pfn_to_online_page(start_pfn)))
287 continue;
288
289 if (unlikely(pfn_to_nid(start_pfn) != nid))
290 continue;
291
292 if (zone != page_zone(pfn_to_page(start_pfn)))
293 continue;
294
295 return start_pfn;
296 }
297
298 return 0;
299}
300
301
302static unsigned long find_biggest_section_pfn(int nid, struct zone *zone,
303 unsigned long start_pfn,
304 unsigned long end_pfn)
305{
306 unsigned long pfn;
307
308
309 pfn = end_pfn - 1;
310 for (; pfn >= start_pfn; pfn -= PAGES_PER_SUBSECTION) {
311 if (unlikely(!pfn_to_online_page(pfn)))
312 continue;
313
314 if (unlikely(pfn_to_nid(pfn) != nid))
315 continue;
316
317 if (zone != page_zone(pfn_to_page(pfn)))
318 continue;
319
320 return pfn;
321 }
322
323 return 0;
324}
325
326static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
327 unsigned long end_pfn)
328{
329 unsigned long pfn;
330 int nid = zone_to_nid(zone);
331
332 if (zone->zone_start_pfn == start_pfn) {
333
334
335
336
337
338
339 pfn = find_smallest_section_pfn(nid, zone, end_pfn,
340 zone_end_pfn(zone));
341 if (pfn) {
342 zone->spanned_pages = zone_end_pfn(zone) - pfn;
343 zone->zone_start_pfn = pfn;
344 } else {
345 zone->zone_start_pfn = 0;
346 zone->spanned_pages = 0;
347 }
348 } else if (zone_end_pfn(zone) == end_pfn) {
349
350
351
352
353
354
355 pfn = find_biggest_section_pfn(nid, zone, zone->zone_start_pfn,
356 start_pfn);
357 if (pfn)
358 zone->spanned_pages = pfn - zone->zone_start_pfn + 1;
359 else {
360 zone->zone_start_pfn = 0;
361 zone->spanned_pages = 0;
362 }
363 }
364}
365
366static void update_pgdat_span(struct pglist_data *pgdat)
367{
368 unsigned long node_start_pfn = 0, node_end_pfn = 0;
369 struct zone *zone;
370
371 for (zone = pgdat->node_zones;
372 zone < pgdat->node_zones + MAX_NR_ZONES; zone++) {
373 unsigned long end_pfn = zone_end_pfn(zone);
374
375
376 if (!zone->spanned_pages)
377 continue;
378 if (!node_end_pfn) {
379 node_start_pfn = zone->zone_start_pfn;
380 node_end_pfn = end_pfn;
381 continue;
382 }
383
384 if (end_pfn > node_end_pfn)
385 node_end_pfn = end_pfn;
386 if (zone->zone_start_pfn < node_start_pfn)
387 node_start_pfn = zone->zone_start_pfn;
388 }
389
390 pgdat->node_start_pfn = node_start_pfn;
391 pgdat->node_spanned_pages = node_end_pfn - node_start_pfn;
392}
393
394void __ref remove_pfn_range_from_zone(struct zone *zone,
395 unsigned long start_pfn,
396 unsigned long nr_pages)
397{
398 const unsigned long end_pfn = start_pfn + nr_pages;
399 struct pglist_data *pgdat = zone->zone_pgdat;
400 unsigned long pfn, cur_nr_pages;
401
402
403 for (pfn = start_pfn; pfn < end_pfn; pfn += cur_nr_pages) {
404 cond_resched();
405
406
407 cur_nr_pages =
408 min(end_pfn - pfn, SECTION_ALIGN_UP(pfn + 1) - pfn);
409 page_init_poison(pfn_to_page(pfn),
410 sizeof(struct page) * cur_nr_pages);
411 }
412
413#ifdef CONFIG_ZONE_DEVICE
414
415
416
417
418
419 if (zone_idx(zone) == ZONE_DEVICE)
420 return;
421#endif
422
423 clear_zone_contiguous(zone);
424
425 shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
426 update_pgdat_span(pgdat);
427
428 set_zone_contiguous(zone);
429}
430
431static void __remove_section(unsigned long pfn, unsigned long nr_pages,
432 unsigned long map_offset,
433 struct vmem_altmap *altmap)
434{
435 struct mem_section *ms = __pfn_to_section(pfn);
436
437 if (WARN_ON_ONCE(!valid_section(ms)))
438 return;
439
440 sparse_remove_section(ms, pfn, nr_pages, map_offset, altmap);
441}
442
443
444
445
446
447
448
449
450
451
452
453
454void __remove_pages(unsigned long pfn, unsigned long nr_pages,
455 struct vmem_altmap *altmap)
456{
457 const unsigned long end_pfn = pfn + nr_pages;
458 unsigned long cur_nr_pages;
459 unsigned long map_offset = 0;
460
461 map_offset = vmem_altmap_offset(altmap);
462
463 if (check_pfn_span(pfn, nr_pages, "remove"))
464 return;
465
466 for (; pfn < end_pfn; pfn += cur_nr_pages) {
467 cond_resched();
468
469 cur_nr_pages = min(end_pfn - pfn,
470 SECTION_ALIGN_UP(pfn + 1) - pfn);
471 __remove_section(pfn, cur_nr_pages, map_offset, altmap);
472 map_offset = 0;
473 }
474}
475
476int set_online_page_callback(online_page_callback_t callback)
477{
478 int rc = -EINVAL;
479
480 get_online_mems();
481 mutex_lock(&online_page_callback_lock);
482
483 if (online_page_callback == generic_online_page) {
484 online_page_callback = callback;
485 rc = 0;
486 }
487
488 mutex_unlock(&online_page_callback_lock);
489 put_online_mems();
490
491 return rc;
492}
493EXPORT_SYMBOL_GPL(set_online_page_callback);
494
495int restore_online_page_callback(online_page_callback_t callback)
496{
497 int rc = -EINVAL;
498
499 get_online_mems();
500 mutex_lock(&online_page_callback_lock);
501
502 if (online_page_callback == callback) {
503 online_page_callback = generic_online_page;
504 rc = 0;
505 }
506
507 mutex_unlock(&online_page_callback_lock);
508 put_online_mems();
509
510 return rc;
511}
512EXPORT_SYMBOL_GPL(restore_online_page_callback);
513
514void generic_online_page(struct page *page, unsigned int order)
515{
516
517
518
519
520
521 debug_pagealloc_map_pages(page, 1 << order);
522 __free_pages_core(page, order);
523 totalram_pages_add(1UL << order);
524#ifdef CONFIG_HIGHMEM
525 if (PageHighMem(page))
526 totalhigh_pages_add(1UL << order);
527#endif
528}
529EXPORT_SYMBOL_GPL(generic_online_page);
530
531static void online_pages_range(unsigned long start_pfn, unsigned long nr_pages)
532{
533 const unsigned long end_pfn = start_pfn + nr_pages;
534 unsigned long pfn;
535
536
537
538
539
540
541
542
543
544
545 for (pfn = start_pfn; pfn < end_pfn;) {
546 int order = min(MAX_ORDER - 1UL, __ffs(pfn));
547
548 (*online_page_callback)(pfn_to_page(pfn), order);
549 pfn += (1UL << order);
550 }
551
552
553 online_mem_sections(start_pfn, end_pfn);
554}
555
556
557static void node_states_check_changes_online(unsigned long nr_pages,
558 struct zone *zone, struct memory_notify *arg)
559{
560 int nid = zone_to_nid(zone);
561
562 arg->status_change_nid = NUMA_NO_NODE;
563 arg->status_change_nid_normal = NUMA_NO_NODE;
564 arg->status_change_nid_high = NUMA_NO_NODE;
565
566 if (!node_state(nid, N_MEMORY))
567 arg->status_change_nid = nid;
568 if (zone_idx(zone) <= ZONE_NORMAL && !node_state(nid, N_NORMAL_MEMORY))
569 arg->status_change_nid_normal = nid;
570#ifdef CONFIG_HIGHMEM
571 if (zone_idx(zone) <= ZONE_HIGHMEM && !node_state(nid, N_HIGH_MEMORY))
572 arg->status_change_nid_high = nid;
573#endif
574}
575
576static void node_states_set_node(int node, struct memory_notify *arg)
577{
578 if (arg->status_change_nid_normal >= 0)
579 node_set_state(node, N_NORMAL_MEMORY);
580
581 if (arg->status_change_nid_high >= 0)
582 node_set_state(node, N_HIGH_MEMORY);
583
584 if (arg->status_change_nid >= 0)
585 node_set_state(node, N_MEMORY);
586}
587
588static void __meminit resize_zone_range(struct zone *zone, unsigned long start_pfn,
589 unsigned long nr_pages)
590{
591 unsigned long old_end_pfn = zone_end_pfn(zone);
592
593 if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
594 zone->zone_start_pfn = start_pfn;
595
596 zone->spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - zone->zone_start_pfn;
597}
598
599static void __meminit resize_pgdat_range(struct pglist_data *pgdat, unsigned long start_pfn,
600 unsigned long nr_pages)
601{
602 unsigned long old_end_pfn = pgdat_end_pfn(pgdat);
603
604 if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
605 pgdat->node_start_pfn = start_pfn;
606
607 pgdat->node_spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - pgdat->node_start_pfn;
608
609}
610
611static void section_taint_zone_device(unsigned long pfn)
612{
613 struct mem_section *ms = __pfn_to_section(pfn);
614
615 ms->section_mem_map |= SECTION_TAINT_ZONE_DEVICE;
616}
617
618
619
620
621
622
623
624
625
626
627void __ref move_pfn_range_to_zone(struct zone *zone, unsigned long start_pfn,
628 unsigned long nr_pages,
629 struct vmem_altmap *altmap, int migratetype)
630{
631 struct pglist_data *pgdat = zone->zone_pgdat;
632 int nid = pgdat->node_id;
633
634 clear_zone_contiguous(zone);
635
636 if (zone_is_empty(zone))
637 init_currently_empty_zone(zone, start_pfn, nr_pages);
638 resize_zone_range(zone, start_pfn, nr_pages);
639 resize_pgdat_range(pgdat, start_pfn, nr_pages);
640
641
642
643
644
645
646
647 if (zone_is_zone_device(zone)) {
648 if (!IS_ALIGNED(start_pfn, PAGES_PER_SECTION))
649 section_taint_zone_device(start_pfn);
650 if (!IS_ALIGNED(start_pfn + nr_pages, PAGES_PER_SECTION))
651 section_taint_zone_device(start_pfn + nr_pages);
652 }
653
654
655
656
657
658
659
660 memmap_init_range(nr_pages, nid, zone_idx(zone), start_pfn, 0,
661 MEMINIT_HOTPLUG, altmap, migratetype);
662
663 set_zone_contiguous(zone);
664}
665
666
667
668
669
670
671static struct zone *default_kernel_zone_for_pfn(int nid, unsigned long start_pfn,
672 unsigned long nr_pages)
673{
674 struct pglist_data *pgdat = NODE_DATA(nid);
675 int zid;
676
677 for (zid = 0; zid <= ZONE_NORMAL; zid++) {
678 struct zone *zone = &pgdat->node_zones[zid];
679
680 if (zone_intersects(zone, start_pfn, nr_pages))
681 return zone;
682 }
683
684 return &pgdat->node_zones[ZONE_NORMAL];
685}
686
687static inline struct zone *default_zone_for_pfn(int nid, unsigned long start_pfn,
688 unsigned long nr_pages)
689{
690 struct zone *kernel_zone = default_kernel_zone_for_pfn(nid, start_pfn,
691 nr_pages);
692 struct zone *movable_zone = &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
693 bool in_kernel = zone_intersects(kernel_zone, start_pfn, nr_pages);
694 bool in_movable = zone_intersects(movable_zone, start_pfn, nr_pages);
695
696
697
698
699
700 if (in_kernel ^ in_movable)
701 return (in_kernel) ? kernel_zone : movable_zone;
702
703
704
705
706
707
708 return movable_node_enabled ? movable_zone : kernel_zone;
709}
710
711struct zone *zone_for_pfn_range(int online_type, int nid, unsigned start_pfn,
712 unsigned long nr_pages)
713{
714 if (online_type == MMOP_ONLINE_KERNEL)
715 return default_kernel_zone_for_pfn(nid, start_pfn, nr_pages);
716
717 if (online_type == MMOP_ONLINE_MOVABLE)
718 return &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
719
720 return default_zone_for_pfn(nid, start_pfn, nr_pages);
721}
722
723
724
725
726
727void adjust_present_page_count(struct zone *zone, long nr_pages)
728{
729 zone->present_pages += nr_pages;
730 zone->zone_pgdat->node_present_pages += nr_pages;
731}
732
733int mhp_init_memmap_on_memory(unsigned long pfn, unsigned long nr_pages,
734 struct zone *zone)
735{
736 unsigned long end_pfn = pfn + nr_pages;
737 int ret;
738
739 ret = kasan_add_zero_shadow(__va(PFN_PHYS(pfn)), PFN_PHYS(nr_pages));
740 if (ret)
741 return ret;
742
743 move_pfn_range_to_zone(zone, pfn, nr_pages, NULL, MIGRATE_UNMOVABLE);
744
745
746
747
748
749
750 if (nr_pages >= PAGES_PER_SECTION)
751 online_mem_sections(pfn, ALIGN_DOWN(end_pfn, PAGES_PER_SECTION));
752
753 return ret;
754}
755
756void mhp_deinit_memmap_on_memory(unsigned long pfn, unsigned long nr_pages)
757{
758 unsigned long end_pfn = pfn + nr_pages;
759
760
761
762
763
764
765 if (nr_pages >= PAGES_PER_SECTION)
766 offline_mem_sections(pfn, ALIGN_DOWN(end_pfn, PAGES_PER_SECTION));
767
768
769
770
771
772 remove_pfn_range_from_zone(page_zone(pfn_to_page(pfn)), pfn, nr_pages);
773 kasan_remove_zero_shadow(__va(PFN_PHYS(pfn)), PFN_PHYS(nr_pages));
774}
775
776int __ref online_pages(unsigned long pfn, unsigned long nr_pages, struct zone *zone)
777{
778 unsigned long flags;
779 int need_zonelists_rebuild = 0;
780 const int nid = zone_to_nid(zone);
781 int ret;
782 struct memory_notify arg;
783
784
785
786
787
788
789
790
791 if (WARN_ON_ONCE(!nr_pages ||
792 !IS_ALIGNED(pfn, pageblock_nr_pages) ||
793 !IS_ALIGNED(pfn + nr_pages, PAGES_PER_SECTION)))
794 return -EINVAL;
795
796 mem_hotplug_begin();
797
798
799 move_pfn_range_to_zone(zone, pfn, nr_pages, NULL, MIGRATE_ISOLATE);
800
801 arg.start_pfn = pfn;
802 arg.nr_pages = nr_pages;
803 node_states_check_changes_online(nr_pages, zone, &arg);
804
805 ret = memory_notify(MEM_GOING_ONLINE, &arg);
806 ret = notifier_to_errno(ret);
807 if (ret)
808 goto failed_addition;
809
810
811
812
813
814 spin_lock_irqsave(&zone->lock, flags);
815 zone->nr_isolate_pageblock += nr_pages / pageblock_nr_pages;
816 spin_unlock_irqrestore(&zone->lock, flags);
817
818
819
820
821
822
823 if (!populated_zone(zone)) {
824 need_zonelists_rebuild = 1;
825 setup_zone_pageset(zone);
826 }
827
828 online_pages_range(pfn, nr_pages);
829 adjust_present_page_count(zone, nr_pages);
830
831 node_states_set_node(nid, &arg);
832 if (need_zonelists_rebuild)
833 build_all_zonelists(NULL);
834
835
836 undo_isolate_page_range(pfn, pfn + nr_pages, MIGRATE_MOVABLE);
837
838
839
840
841
842
843
844 shuffle_zone(zone);
845
846
847 init_per_zone_wmark_min();
848
849 kswapd_run(nid);
850 kcompactd_run(nid);
851
852 writeback_set_ratelimit();
853
854 memory_notify(MEM_ONLINE, &arg);
855 mem_hotplug_done();
856 return 0;
857
858failed_addition:
859 pr_debug("online_pages [mem %#010llx-%#010llx] failed\n",
860 (unsigned long long) pfn << PAGE_SHIFT,
861 (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1);
862 memory_notify(MEM_CANCEL_ONLINE, &arg);
863 remove_pfn_range_from_zone(zone, pfn, nr_pages);
864 mem_hotplug_done();
865 return ret;
866}
867#endif
868
869static void reset_node_present_pages(pg_data_t *pgdat)
870{
871 struct zone *z;
872
873 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
874 z->present_pages = 0;
875
876 pgdat->node_present_pages = 0;
877}
878
879
880static pg_data_t __ref *hotadd_new_pgdat(int nid)
881{
882 struct pglist_data *pgdat;
883
884 pgdat = NODE_DATA(nid);
885 if (!pgdat) {
886 pgdat = arch_alloc_nodedata(nid);
887 if (!pgdat)
888 return NULL;
889
890 pgdat->per_cpu_nodestats =
891 alloc_percpu(struct per_cpu_nodestat);
892 arch_refresh_nodedata(nid, pgdat);
893 } else {
894 int cpu;
895
896
897
898
899
900 pgdat->nr_zones = 0;
901 pgdat->kswapd_order = 0;
902 pgdat->kswapd_highest_zoneidx = 0;
903 for_each_online_cpu(cpu) {
904 struct per_cpu_nodestat *p;
905
906 p = per_cpu_ptr(pgdat->per_cpu_nodestats, cpu);
907 memset(p, 0, sizeof(*p));
908 }
909 }
910
911
912 pgdat->node_id = nid;
913 pgdat->node_start_pfn = 0;
914
915
916 free_area_init_core_hotplug(nid);
917
918
919
920
921
922 build_all_zonelists(pgdat);
923
924
925
926
927
928
929 reset_node_managed_pages(pgdat);
930 reset_node_present_pages(pgdat);
931
932 return pgdat;
933}
934
935static void rollback_node_hotadd(int nid)
936{
937 pg_data_t *pgdat = NODE_DATA(nid);
938
939 arch_refresh_nodedata(nid, NULL);
940 free_percpu(pgdat->per_cpu_nodestats);
941 arch_free_nodedata(pgdat);
942}
943
944
945
946
947
948
949
950
951
952
953
954
955
956static int __try_online_node(int nid, bool set_node_online)
957{
958 pg_data_t *pgdat;
959 int ret = 1;
960
961 if (node_online(nid))
962 return 0;
963
964 pgdat = hotadd_new_pgdat(nid);
965 if (!pgdat) {
966 pr_err("Cannot online node %d due to NULL pgdat\n", nid);
967 ret = -ENOMEM;
968 goto out;
969 }
970
971 if (set_node_online) {
972 node_set_online(nid);
973 ret = register_one_node(nid);
974 BUG_ON(ret);
975 }
976out:
977 return ret;
978}
979
980
981
982
983int try_online_node(int nid)
984{
985 int ret;
986
987 mem_hotplug_begin();
988 ret = __try_online_node(nid, true);
989 mem_hotplug_done();
990 return ret;
991}
992
993static int check_hotplug_memory_range(u64 start, u64 size)
994{
995
996 if (!size || !IS_ALIGNED(start, memory_block_size_bytes()) ||
997 !IS_ALIGNED(size, memory_block_size_bytes())) {
998 pr_err("Block size [%#lx] unaligned hotplug range: start %#llx, size %#llx",
999 memory_block_size_bytes(), start, size);
1000 return -EINVAL;
1001 }
1002
1003 return 0;
1004}
1005
1006static int online_memory_block(struct memory_block *mem, void *arg)
1007{
1008 mem->online_type = mhp_default_online_type;
1009 return device_online(&mem->dev);
1010}
1011
1012bool mhp_supports_memmap_on_memory(unsigned long size)
1013{
1014 unsigned long nr_vmemmap_pages = size / PAGE_SIZE;
1015 unsigned long vmemmap_size = nr_vmemmap_pages * sizeof(struct page);
1016 unsigned long remaining_size = size - vmemmap_size;
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044 return memmap_on_memory &&
1045 !hugetlb_free_vmemmap_enabled &&
1046 IS_ENABLED(CONFIG_MHP_MEMMAP_ON_MEMORY) &&
1047 size == memory_block_size_bytes() &&
1048 IS_ALIGNED(vmemmap_size, PMD_SIZE) &&
1049 IS_ALIGNED(remaining_size, (pageblock_nr_pages << PAGE_SHIFT));
1050}
1051
1052
1053
1054
1055
1056
1057
1058int __ref add_memory_resource(int nid, struct resource *res, mhp_t mhp_flags)
1059{
1060 struct mhp_params params = { .pgprot = pgprot_mhp(PAGE_KERNEL) };
1061 struct vmem_altmap mhp_altmap = {};
1062 u64 start, size;
1063 bool new_node = false;
1064 int ret;
1065
1066 start = res->start;
1067 size = resource_size(res);
1068
1069 ret = check_hotplug_memory_range(start, size);
1070 if (ret)
1071 return ret;
1072
1073 if (!node_possible(nid)) {
1074 WARN(1, "node %d was absent from the node_possible_map\n", nid);
1075 return -EINVAL;
1076 }
1077
1078 mem_hotplug_begin();
1079
1080 if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK))
1081 memblock_add_node(start, size, nid);
1082
1083 ret = __try_online_node(nid, false);
1084 if (ret < 0)
1085 goto error;
1086 new_node = ret;
1087
1088
1089
1090
1091 if (mhp_flags & MHP_MEMMAP_ON_MEMORY) {
1092 if (!mhp_supports_memmap_on_memory(size)) {
1093 ret = -EINVAL;
1094 goto error;
1095 }
1096 mhp_altmap.free = PHYS_PFN(size);
1097 mhp_altmap.base_pfn = PHYS_PFN(start);
1098 params.altmap = &mhp_altmap;
1099 }
1100
1101
1102 ret = arch_add_memory(nid, start, size, ¶ms);
1103 if (ret < 0)
1104 goto error;
1105
1106
1107 ret = create_memory_block_devices(start, size, mhp_altmap.alloc);
1108 if (ret) {
1109 arch_remove_memory(nid, start, size, NULL);
1110 goto error;
1111 }
1112
1113 if (new_node) {
1114
1115
1116
1117
1118
1119 node_set_online(nid);
1120 ret = __register_one_node(nid);
1121 BUG_ON(ret);
1122 }
1123
1124
1125 link_mem_sections(nid, PFN_DOWN(start), PFN_UP(start + size - 1),
1126 MEMINIT_HOTPLUG);
1127
1128
1129 if (!strcmp(res->name, "System RAM"))
1130 firmware_map_add_hotplug(start, start + size, "System RAM");
1131
1132
1133 mem_hotplug_done();
1134
1135
1136
1137
1138
1139 if (mhp_flags & MHP_MERGE_RESOURCE)
1140 merge_system_ram_resource(res);
1141
1142
1143 if (mhp_default_online_type != MMOP_OFFLINE)
1144 walk_memory_blocks(start, size, NULL, online_memory_block);
1145
1146 return ret;
1147error:
1148
1149 if (new_node)
1150 rollback_node_hotadd(nid);
1151 if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK))
1152 memblock_remove(start, size);
1153 mem_hotplug_done();
1154 return ret;
1155}
1156
1157
1158int __ref __add_memory(int nid, u64 start, u64 size, mhp_t mhp_flags)
1159{
1160 struct resource *res;
1161 int ret;
1162
1163 res = register_memory_resource(start, size, "System RAM");
1164 if (IS_ERR(res))
1165 return PTR_ERR(res);
1166
1167 ret = add_memory_resource(nid, res, mhp_flags);
1168 if (ret < 0)
1169 release_memory_resource(res);
1170 return ret;
1171}
1172
1173int add_memory(int nid, u64 start, u64 size, mhp_t mhp_flags)
1174{
1175 int rc;
1176
1177 lock_device_hotplug();
1178 rc = __add_memory(nid, start, size, mhp_flags);
1179 unlock_device_hotplug();
1180
1181 return rc;
1182}
1183EXPORT_SYMBOL_GPL(add_memory);
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206int add_memory_driver_managed(int nid, u64 start, u64 size,
1207 const char *resource_name, mhp_t mhp_flags)
1208{
1209 struct resource *res;
1210 int rc;
1211
1212 if (!resource_name ||
1213 strstr(resource_name, "System RAM (") != resource_name ||
1214 resource_name[strlen(resource_name) - 1] != ')')
1215 return -EINVAL;
1216
1217 lock_device_hotplug();
1218
1219 res = register_memory_resource(start, size, resource_name);
1220 if (IS_ERR(res)) {
1221 rc = PTR_ERR(res);
1222 goto out_unlock;
1223 }
1224
1225 rc = add_memory_resource(nid, res, mhp_flags);
1226 if (rc < 0)
1227 release_memory_resource(res);
1228
1229out_unlock:
1230 unlock_device_hotplug();
1231 return rc;
1232}
1233EXPORT_SYMBOL_GPL(add_memory_driver_managed);
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248struct range __weak arch_get_mappable_range(void)
1249{
1250 struct range mhp_range = {
1251 .start = 0UL,
1252 .end = -1ULL,
1253 };
1254 return mhp_range;
1255}
1256
1257struct range mhp_get_pluggable_range(bool need_mapping)
1258{
1259 const u64 max_phys = (1ULL << MAX_PHYSMEM_BITS) - 1;
1260 struct range mhp_range;
1261
1262 if (need_mapping) {
1263 mhp_range = arch_get_mappable_range();
1264 if (mhp_range.start > max_phys) {
1265 mhp_range.start = 0;
1266 mhp_range.end = 0;
1267 }
1268 mhp_range.end = min_t(u64, mhp_range.end, max_phys);
1269 } else {
1270 mhp_range.start = 0;
1271 mhp_range.end = max_phys;
1272 }
1273 return mhp_range;
1274}
1275EXPORT_SYMBOL_GPL(mhp_get_pluggable_range);
1276
1277bool mhp_range_allowed(u64 start, u64 size, bool need_mapping)
1278{
1279 struct range mhp_range = mhp_get_pluggable_range(need_mapping);
1280 u64 end = start + size;
1281
1282 if (start < end && start >= mhp_range.start && (end - 1) <= mhp_range.end)
1283 return true;
1284
1285 pr_warn("Hotplug memory [%#llx-%#llx] exceeds maximum addressable range [%#llx-%#llx]\n",
1286 start, end, mhp_range.start, mhp_range.end);
1287 return false;
1288}
1289
1290#ifdef CONFIG_MEMORY_HOTREMOVE
1291
1292
1293
1294
1295struct zone *test_pages_in_a_zone(unsigned long start_pfn,
1296 unsigned long end_pfn)
1297{
1298 unsigned long pfn, sec_end_pfn;
1299 struct zone *zone = NULL;
1300 struct page *page;
1301 int i;
1302 for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1);
1303 pfn < end_pfn;
1304 pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) {
1305
1306 if (!present_section_nr(pfn_to_section_nr(pfn)))
1307 continue;
1308 for (; pfn < sec_end_pfn && pfn < end_pfn;
1309 pfn += MAX_ORDER_NR_PAGES) {
1310 i = 0;
1311
1312 while ((i < MAX_ORDER_NR_PAGES) &&
1313 !pfn_valid_within(pfn + i))
1314 i++;
1315 if (i == MAX_ORDER_NR_PAGES || pfn + i >= end_pfn)
1316 continue;
1317
1318 if (zone && !zone_spans_pfn(zone, pfn + i))
1319 return NULL;
1320 page = pfn_to_page(pfn + i);
1321 if (zone && page_zone(page) != zone)
1322 return NULL;
1323 zone = page_zone(page);
1324 }
1325 }
1326
1327 return zone;
1328}
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341static int scan_movable_pages(unsigned long start, unsigned long end,
1342 unsigned long *movable_pfn)
1343{
1344 unsigned long pfn;
1345
1346 for (pfn = start; pfn < end; pfn++) {
1347 struct page *page, *head;
1348 unsigned long skip;
1349
1350 if (!pfn_valid(pfn))
1351 continue;
1352 page = pfn_to_page(pfn);
1353 if (PageLRU(page))
1354 goto found;
1355 if (__PageMovable(page))
1356 goto found;
1357
1358
1359
1360
1361
1362
1363
1364 if (PageOffline(page) && page_count(page))
1365 return -EBUSY;
1366
1367 if (!PageHuge(page))
1368 continue;
1369 head = compound_head(page);
1370
1371
1372
1373
1374
1375
1376
1377 if (HPageMigratable(head))
1378 goto found;
1379 skip = compound_nr(head) - (page - head);
1380 pfn += skip - 1;
1381 }
1382 return -ENOENT;
1383found:
1384 *movable_pfn = pfn;
1385 return 0;
1386}
1387
1388static int
1389do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1390{
1391 unsigned long pfn;
1392 struct page *page, *head;
1393 int ret = 0;
1394 LIST_HEAD(source);
1395 static DEFINE_RATELIMIT_STATE(migrate_rs, DEFAULT_RATELIMIT_INTERVAL,
1396 DEFAULT_RATELIMIT_BURST);
1397
1398 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1399 if (!pfn_valid(pfn))
1400 continue;
1401 page = pfn_to_page(pfn);
1402 head = compound_head(page);
1403
1404 if (PageHuge(page)) {
1405 pfn = page_to_pfn(head) + compound_nr(head) - 1;
1406 isolate_huge_page(head, &source);
1407 continue;
1408 } else if (PageTransHuge(page))
1409 pfn = page_to_pfn(head) + thp_nr_pages(page) - 1;
1410
1411
1412
1413
1414
1415
1416
1417
1418 if (PageHWPoison(page)) {
1419 if (WARN_ON(PageLRU(page)))
1420 isolate_lru_page(page);
1421 if (page_mapped(page))
1422 try_to_unmap(page, TTU_IGNORE_MLOCK);
1423 continue;
1424 }
1425
1426 if (!get_page_unless_zero(page))
1427 continue;
1428
1429
1430
1431
1432 if (PageLRU(page))
1433 ret = isolate_lru_page(page);
1434 else
1435 ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE);
1436 if (!ret) {
1437 list_add_tail(&page->lru, &source);
1438 if (!__PageMovable(page))
1439 inc_node_page_state(page, NR_ISOLATED_ANON +
1440 page_is_file_lru(page));
1441
1442 } else {
1443 if (__ratelimit(&migrate_rs)) {
1444 pr_warn("failed to isolate pfn %lx\n", pfn);
1445 dump_page(page, "isolation failed");
1446 }
1447 }
1448 put_page(page);
1449 }
1450 if (!list_empty(&source)) {
1451 nodemask_t nmask = node_states[N_MEMORY];
1452 struct migration_target_control mtc = {
1453 .nmask = &nmask,
1454 .gfp_mask = GFP_USER | __GFP_MOVABLE | __GFP_RETRY_MAYFAIL,
1455 };
1456
1457
1458
1459
1460
1461 mtc.nid = page_to_nid(list_first_entry(&source, struct page, lru));
1462
1463
1464
1465
1466
1467
1468 node_clear(mtc.nid, nmask);
1469 if (nodes_empty(nmask))
1470 node_set(mtc.nid, nmask);
1471 ret = migrate_pages(&source, alloc_migration_target, NULL,
1472 (unsigned long)&mtc, MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
1473 if (ret) {
1474 list_for_each_entry(page, &source, lru) {
1475 if (__ratelimit(&migrate_rs)) {
1476 pr_warn("migrating pfn %lx failed ret:%d\n",
1477 page_to_pfn(page), ret);
1478 dump_page(page, "migration failure");
1479 }
1480 }
1481 putback_movable_pages(&source);
1482 }
1483 }
1484
1485 return ret;
1486}
1487
1488static int __init cmdline_parse_movable_node(char *p)
1489{
1490 movable_node_enabled = true;
1491 return 0;
1492}
1493early_param("movable_node", cmdline_parse_movable_node);
1494
1495
1496static void node_states_check_changes_offline(unsigned long nr_pages,
1497 struct zone *zone, struct memory_notify *arg)
1498{
1499 struct pglist_data *pgdat = zone->zone_pgdat;
1500 unsigned long present_pages = 0;
1501 enum zone_type zt;
1502
1503 arg->status_change_nid = NUMA_NO_NODE;
1504 arg->status_change_nid_normal = NUMA_NO_NODE;
1505 arg->status_change_nid_high = NUMA_NO_NODE;
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515 for (zt = 0; zt <= ZONE_NORMAL; zt++)
1516 present_pages += pgdat->node_zones[zt].present_pages;
1517 if (zone_idx(zone) <= ZONE_NORMAL && nr_pages >= present_pages)
1518 arg->status_change_nid_normal = zone_to_nid(zone);
1519
1520#ifdef CONFIG_HIGHMEM
1521
1522
1523
1524
1525
1526
1527
1528
1529 present_pages += pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1530 if (zone_idx(zone) <= ZONE_HIGHMEM && nr_pages >= present_pages)
1531 arg->status_change_nid_high = zone_to_nid(zone);
1532#endif
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544 present_pages += pgdat->node_zones[ZONE_MOVABLE].present_pages;
1545
1546 if (nr_pages >= present_pages)
1547 arg->status_change_nid = zone_to_nid(zone);
1548}
1549
1550static void node_states_clear_node(int node, struct memory_notify *arg)
1551{
1552 if (arg->status_change_nid_normal >= 0)
1553 node_clear_state(node, N_NORMAL_MEMORY);
1554
1555 if (arg->status_change_nid_high >= 0)
1556 node_clear_state(node, N_HIGH_MEMORY);
1557
1558 if (arg->status_change_nid >= 0)
1559 node_clear_state(node, N_MEMORY);
1560}
1561
1562static int count_system_ram_pages_cb(unsigned long start_pfn,
1563 unsigned long nr_pages, void *data)
1564{
1565 unsigned long *nr_system_ram_pages = data;
1566
1567 *nr_system_ram_pages += nr_pages;
1568 return 0;
1569}
1570
1571int __ref offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1572{
1573 const unsigned long end_pfn = start_pfn + nr_pages;
1574 unsigned long pfn, system_ram_pages = 0;
1575 unsigned long flags;
1576 struct zone *zone;
1577 struct memory_notify arg;
1578 int ret, node;
1579 char *reason;
1580
1581
1582
1583
1584
1585
1586
1587
1588 if (WARN_ON_ONCE(!nr_pages ||
1589 !IS_ALIGNED(start_pfn, pageblock_nr_pages) ||
1590 !IS_ALIGNED(start_pfn + nr_pages, PAGES_PER_SECTION)))
1591 return -EINVAL;
1592
1593 mem_hotplug_begin();
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603 walk_system_ram_range(start_pfn, nr_pages, &system_ram_pages,
1604 count_system_ram_pages_cb);
1605 if (system_ram_pages != nr_pages) {
1606 ret = -EINVAL;
1607 reason = "memory holes";
1608 goto failed_removal;
1609 }
1610
1611
1612
1613 zone = test_pages_in_a_zone(start_pfn, end_pfn);
1614 if (!zone) {
1615 ret = -EINVAL;
1616 reason = "multizone range";
1617 goto failed_removal;
1618 }
1619 node = zone_to_nid(zone);
1620
1621
1622
1623
1624
1625 zone_pcp_disable(zone);
1626 lru_cache_disable();
1627
1628
1629 ret = start_isolate_page_range(start_pfn, end_pfn,
1630 MIGRATE_MOVABLE,
1631 MEMORY_OFFLINE | REPORT_FAILURE);
1632 if (ret) {
1633 reason = "failure to isolate range";
1634 goto failed_removal_pcplists_disabled;
1635 }
1636
1637 arg.start_pfn = start_pfn;
1638 arg.nr_pages = nr_pages;
1639 node_states_check_changes_offline(nr_pages, zone, &arg);
1640
1641 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1642 ret = notifier_to_errno(ret);
1643 if (ret) {
1644 reason = "notifier failure";
1645 goto failed_removal_isolated;
1646 }
1647
1648 do {
1649 pfn = start_pfn;
1650 do {
1651 if (signal_pending(current)) {
1652 ret = -EINTR;
1653 reason = "signal backoff";
1654 goto failed_removal_isolated;
1655 }
1656
1657 cond_resched();
1658
1659 ret = scan_movable_pages(pfn, end_pfn, &pfn);
1660 if (!ret) {
1661
1662
1663
1664
1665 do_migrate_range(pfn, end_pfn);
1666 }
1667 } while (!ret);
1668
1669 if (ret != -ENOENT) {
1670 reason = "unmovable page";
1671 goto failed_removal_isolated;
1672 }
1673
1674
1675
1676
1677
1678
1679 ret = dissolve_free_huge_pages(start_pfn, end_pfn);
1680 if (ret) {
1681 reason = "failure to dissolve huge pages";
1682 goto failed_removal_isolated;
1683 }
1684
1685 ret = test_pages_isolated(start_pfn, end_pfn, MEMORY_OFFLINE);
1686
1687 } while (ret);
1688
1689
1690 __offline_isolated_pages(start_pfn, end_pfn);
1691 pr_debug("Offlined Pages %ld\n", nr_pages);
1692
1693
1694
1695
1696
1697
1698 spin_lock_irqsave(&zone->lock, flags);
1699 zone->nr_isolate_pageblock -= nr_pages / pageblock_nr_pages;
1700 spin_unlock_irqrestore(&zone->lock, flags);
1701
1702 lru_cache_enable();
1703 zone_pcp_enable(zone);
1704
1705
1706 adjust_managed_page_count(pfn_to_page(start_pfn), -nr_pages);
1707 adjust_present_page_count(zone, -nr_pages);
1708
1709
1710 init_per_zone_wmark_min();
1711
1712 if (!populated_zone(zone)) {
1713 zone_pcp_reset(zone);
1714 build_all_zonelists(NULL);
1715 }
1716
1717 node_states_clear_node(node, &arg);
1718 if (arg.status_change_nid >= 0) {
1719 kswapd_stop(node);
1720 kcompactd_stop(node);
1721 }
1722
1723 writeback_set_ratelimit();
1724
1725 memory_notify(MEM_OFFLINE, &arg);
1726 remove_pfn_range_from_zone(zone, start_pfn, nr_pages);
1727 mem_hotplug_done();
1728 return 0;
1729
1730failed_removal_isolated:
1731 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1732 memory_notify(MEM_CANCEL_OFFLINE, &arg);
1733failed_removal_pcplists_disabled:
1734 lru_cache_enable();
1735 zone_pcp_enable(zone);
1736failed_removal:
1737 pr_debug("memory offlining [mem %#010llx-%#010llx] failed due to %s\n",
1738 (unsigned long long) start_pfn << PAGE_SHIFT,
1739 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1,
1740 reason);
1741
1742 mem_hotplug_done();
1743 return ret;
1744}
1745
1746static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
1747{
1748 int ret = !is_memblock_offlined(mem);
1749
1750 if (unlikely(ret)) {
1751 phys_addr_t beginpa, endpa;
1752
1753 beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
1754 endpa = beginpa + memory_block_size_bytes() - 1;
1755 pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n",
1756 &beginpa, &endpa);
1757
1758 return -EBUSY;
1759 }
1760 return 0;
1761}
1762
1763static int get_nr_vmemmap_pages_cb(struct memory_block *mem, void *arg)
1764{
1765
1766
1767
1768 return mem->nr_vmemmap_pages;
1769}
1770
1771static int check_cpu_on_node(pg_data_t *pgdat)
1772{
1773 int cpu;
1774
1775 for_each_present_cpu(cpu) {
1776 if (cpu_to_node(cpu) == pgdat->node_id)
1777
1778
1779
1780
1781 return -EBUSY;
1782 }
1783
1784 return 0;
1785}
1786
1787static int check_no_memblock_for_node_cb(struct memory_block *mem, void *arg)
1788{
1789 int nid = *(int *)arg;
1790
1791
1792
1793
1794
1795
1796 return mem->nid == nid ? -EEXIST : 0;
1797}
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808void try_offline_node(int nid)
1809{
1810 pg_data_t *pgdat = NODE_DATA(nid);
1811 int rc;
1812
1813
1814
1815
1816
1817
1818 if (pgdat->node_spanned_pages)
1819 return;
1820
1821
1822
1823
1824
1825
1826 rc = for_each_memory_block(&nid, check_no_memblock_for_node_cb);
1827 if (rc)
1828 return;
1829
1830 if (check_cpu_on_node(pgdat))
1831 return;
1832
1833
1834
1835
1836
1837 node_set_offline(nid);
1838 unregister_one_node(nid);
1839}
1840EXPORT_SYMBOL(try_offline_node);
1841
1842static int __ref try_remove_memory(int nid, u64 start, u64 size)
1843{
1844 int rc = 0;
1845 struct vmem_altmap mhp_altmap = {};
1846 struct vmem_altmap *altmap = NULL;
1847 unsigned long nr_vmemmap_pages;
1848
1849 BUG_ON(check_hotplug_memory_range(start, size));
1850
1851
1852
1853
1854
1855
1856 rc = walk_memory_blocks(start, size, NULL, check_memblock_offlined_cb);
1857 if (rc)
1858 return rc;
1859
1860
1861
1862
1863
1864 if (memmap_on_memory) {
1865 nr_vmemmap_pages = walk_memory_blocks(start, size, NULL,
1866 get_nr_vmemmap_pages_cb);
1867 if (nr_vmemmap_pages) {
1868 if (size != memory_block_size_bytes()) {
1869 pr_warn("Refuse to remove %#llx - %#llx,"
1870 "wrong granularity\n",
1871 start, start + size);
1872 return -EINVAL;
1873 }
1874
1875
1876
1877
1878
1879
1880 mhp_altmap.alloc = nr_vmemmap_pages;
1881 altmap = &mhp_altmap;
1882 }
1883 }
1884
1885
1886 firmware_map_remove(start, start + size, "System RAM");
1887
1888
1889
1890
1891
1892 remove_memory_block_devices(start, size);
1893
1894 mem_hotplug_begin();
1895
1896 arch_remove_memory(nid, start, size, altmap);
1897
1898 if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK)) {
1899 memblock_free(start, size);
1900 memblock_remove(start, size);
1901 }
1902
1903 release_mem_region_adjustable(start, size);
1904
1905 try_offline_node(nid);
1906
1907 mem_hotplug_done();
1908 return 0;
1909}
1910
1911
1912
1913
1914
1915
1916
1917
1918
1919
1920
1921void __remove_memory(int nid, u64 start, u64 size)
1922{
1923
1924
1925
1926
1927
1928 if (try_remove_memory(nid, start, size))
1929 BUG();
1930}
1931
1932
1933
1934
1935
1936int remove_memory(int nid, u64 start, u64 size)
1937{
1938 int rc;
1939
1940 lock_device_hotplug();
1941 rc = try_remove_memory(nid, start, size);
1942 unlock_device_hotplug();
1943
1944 return rc;
1945}
1946EXPORT_SYMBOL_GPL(remove_memory);
1947
1948static int try_offline_memory_block(struct memory_block *mem, void *arg)
1949{
1950 uint8_t online_type = MMOP_ONLINE_KERNEL;
1951 uint8_t **online_types = arg;
1952 struct page *page;
1953 int rc;
1954
1955
1956
1957
1958
1959
1960 page = pfn_to_online_page(section_nr_to_pfn(mem->start_section_nr));
1961 if (page && zone_idx(page_zone(page)) == ZONE_MOVABLE)
1962 online_type = MMOP_ONLINE_MOVABLE;
1963
1964 rc = device_offline(&mem->dev);
1965
1966
1967
1968
1969 if (!rc)
1970 **online_types = online_type;
1971
1972 (*online_types)++;
1973
1974 return rc < 0 ? rc : 0;
1975}
1976
1977static int try_reonline_memory_block(struct memory_block *mem, void *arg)
1978{
1979 uint8_t **online_types = arg;
1980 int rc;
1981
1982 if (**online_types != MMOP_OFFLINE) {
1983 mem->online_type = **online_types;
1984 rc = device_online(&mem->dev);
1985 if (rc < 0)
1986 pr_warn("%s: Failed to re-online memory: %d",
1987 __func__, rc);
1988 }
1989
1990
1991 (*online_types)++;
1992 return 0;
1993}
1994
1995
1996
1997
1998
1999
2000
2001int offline_and_remove_memory(int nid, u64 start, u64 size)
2002{
2003 const unsigned long mb_count = size / memory_block_size_bytes();
2004 uint8_t *online_types, *tmp;
2005 int rc;
2006
2007 if (!IS_ALIGNED(start, memory_block_size_bytes()) ||
2008 !IS_ALIGNED(size, memory_block_size_bytes()) || !size)
2009 return -EINVAL;
2010
2011
2012
2013
2014
2015
2016 online_types = kmalloc_array(mb_count, sizeof(*online_types),
2017 GFP_KERNEL);
2018 if (!online_types)
2019 return -ENOMEM;
2020
2021
2022
2023
2024
2025 memset(online_types, MMOP_OFFLINE, mb_count);
2026
2027 lock_device_hotplug();
2028
2029 tmp = online_types;
2030 rc = walk_memory_blocks(start, size, &tmp, try_offline_memory_block);
2031
2032
2033
2034
2035
2036 if (!rc) {
2037 rc = try_remove_memory(nid, start, size);
2038 if (rc)
2039 pr_err("%s: Failed to remove memory: %d", __func__, rc);
2040 }
2041
2042
2043
2044
2045
2046 if (rc) {
2047 tmp = online_types;
2048 walk_memory_blocks(start, size, &tmp,
2049 try_reonline_memory_block);
2050 }
2051 unlock_device_hotplug();
2052
2053 kfree(online_types);
2054 return rc;
2055}
2056EXPORT_SYMBOL_GPL(offline_and_remove_memory);
2057#endif
2058