1/*
2 * linux/mm/memory_hotplug.c
3 *
4 * Copyright (C)
5 */
6
7#include <linux/stddef.h>
8#include <linux/mm.h>
9#include <linux/sched/signal.h>
10#include <linux/swap.h>
11#include <linux/interrupt.h>
12#include <linux/pagemap.h>
13#include <linux/compiler.h>
14#include <linux/export.h>
15#include <linux/pagevec.h>
16#include <linux/writeback.h>
17#include <linux/slab.h>
18#include <linux/sysctl.h>
19#include <linux/cpu.h>
20#include <linux/memory.h>
21#include <linux/memremap.h>
22#include <linux/memory_hotplug.h>
23#include <linux/highmem.h>
24#include <linux/vmalloc.h>
25#include <linux/ioport.h>
26#include <linux/delay.h>
27#include <linux/migrate.h>
28#include <linux/page-isolation.h>
29#include <linux/pfn.h>
30#include <linux/suspend.h>
31#include <linux/mm_inline.h>
32#include <linux/firmware-map.h>
33#include <linux/stop_machine.h>
34#include <linux/hugetlb.h>
35#include <linux/memblock.h>
36#include <linux/compaction.h>
37#include <linux/rmap.h>
38
39#include <asm/tlbflush.h>
40
41#include "internal.h"
42
43/*
44 * online_page_callback contains pointer to current page onlining function.
45 * Initially it is generic_online_page(). If it is required it could be
46 * changed by calling set_online_page_callback() for callback registration
47 * and restore_online_page_callback() for generic callback restore.
48 */
49
50static void generic_online_page(struct page *page, unsigned int order);
51
52static online_page_callback_t online_page_callback = generic_online_page;
53static DEFINE_MUTEX(online_page_callback_lock);
54
55DEFINE_STATIC_PERCPU_RWSEM(mem_hotplug_lock);
56
57void get_online_mems(void)
58{
59 percpu_down_read(&mem_hotplug_lock);
60}
61
62void put_online_mems(void)
63{
64 percpu_up_read(&mem_hotplug_lock);
65}
66
67bool movable_node_enabled = false;
68
69#ifndef CONFIG_MEMORY_HOTPLUG_DEFAULT_ONLINE
70bool memhp_auto_online;
71#else
72bool memhp_auto_online = true;
73#endif
74EXPORT_SYMBOL_GPL(memhp_auto_online);
75
76static int __init setup_memhp_default_state(char *str)
77{
78 if (!strcmp(str, "online"))
79 memhp_auto_online = true;
80 else if (!strcmp(str, "offline"))
81 memhp_auto_online = false;
82
83 return 1;
84}
85__setup("memhp_default_state=", setup_memhp_default_state);
86
87void mem_hotplug_begin(void)
88{
89 cpus_read_lock();
90 percpu_down_write(&mem_hotplug_lock);
91}
92
93void mem_hotplug_done(void)
94{
95 percpu_up_write(&mem_hotplug_lock);
96 cpus_read_unlock();
97}
98
99u64 max_mem_size = U64_MAX;
100
101/* add this memory to iomem resource */
102static struct resource *register_memory_resource(u64 start, u64 size)
103{
104 struct resource *res;
105 unsigned long flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
106 char *resource_name = "System RAM";
107
108 if (start + size > max_mem_size)
109 return ERR_PTR(-E2BIG);
110
111 /*
112 * Request ownership of the new memory range. This might be
113 * a child of an existing resource that was present but
114 * not marked as busy.
115 */
116 res = __request_region(&iomem_resource, start, size,
117 resource_name, flags);
118
119 if (!res) {
120 pr_debug("Unable to reserve System RAM region: %016llx->%016llx\n",
121 start, start + size);
122 return ERR_PTR(-EEXIST);
123 }
124 return res;
125}
126
127static void release_memory_resource(struct resource *res)
128{
129 if (!res)
130 return;
131 release_resource(res);
132 kfree(res);
133 return;
134}
135
136#ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
137void get_page_bootmem(unsigned long info, struct page *page,
138 unsigned long type)
139{
140 page->freelist = (void *)type;
141 SetPagePrivate(page);
142 set_page_private(page, info);
143 page_ref_inc(page);
144}
145
146void put_page_bootmem(struct page *page)
147{
148 unsigned long type;
149
150 type = (unsigned long) page->freelist;
151 BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE ||
152 type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE);
153
154 if (page_ref_dec_return(page) == 1) {
155 page->freelist = NULL;
156 ClearPagePrivate(page);
157 set_page_private(page, 0);
158 INIT_LIST_HEAD(&page->lru);
159 free_reserved_page(page);
160 }
161}
162
163#ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
164#ifndef CONFIG_SPARSEMEM_VMEMMAP
165static void register_page_bootmem_info_section(unsigned long start_pfn)
166{
167 unsigned long *usemap, mapsize, section_nr, i;
168 struct mem_section *ms;
169 struct page *page, *memmap;
170
171 section_nr = pfn_to_section_nr(start_pfn);
172 ms = __nr_to_section(section_nr);
173
174 /* Get section's memmap address */
175 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
176
177 /*
178 * Get page for the memmap's phys address
179 * XXX: need more consideration for sparse_vmemmap...
180 */
181 page = virt_to_page(memmap);
182 mapsize = sizeof(struct page) * PAGES_PER_SECTION;
183 mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
184
185 /* remember memmap's page */
186 for (i = 0; i < mapsize; i++, page++)
187 get_page_bootmem(section_nr, page, SECTION_INFO);
188
189 usemap = ms->pageblock_flags;
190 page = virt_to_page(usemap);
191
192 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
193
194 for (i = 0; i < mapsize; i++, page++)
195 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
196
197}
198#else /* CONFIG_SPARSEMEM_VMEMMAP */
199static void register_page_bootmem_info_section(unsigned long start_pfn)
200{
201 unsigned long *usemap, mapsize, section_nr, i;
202 struct mem_section *ms;
203 struct page *page, *memmap;
204
205 section_nr = pfn_to_section_nr(start_pfn);
206 ms = __nr_to_section(section_nr);
207
208 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
209
210 register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION);
211
212 usemap = ms->pageblock_flags;
213 page = virt_to_page(usemap);
214
215 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
216
217 for (i = 0; i < mapsize; i++, page++)
218 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
219}
220#endif /* !CONFIG_SPARSEMEM_VMEMMAP */
221
222void __init register_page_bootmem_info_node(struct pglist_data *pgdat)
223{
224 unsigned long i, pfn, end_pfn, nr_pages;
225 int node = pgdat->node_id;
226 struct page *page;
227
228 nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
229 page = virt_to_page(pgdat);
230
231 for (i = 0; i < nr_pages; i++, page++)
232 get_page_bootmem(node, page, NODE_INFO);
233
234 pfn = pgdat->node_start_pfn;
235 end_pfn = pgdat_end_pfn(pgdat);
236
237 /* register section info */
238 for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
239 /*
240 * Some platforms can assign the same pfn to multiple nodes - on
241 * node0 as well as nodeN. To avoid registering a pfn against
242 * multiple nodes we check that this pfn does not already
243 * reside in some other nodes.
244 */
245 if (pfn_valid(pfn) && (early_pfn_to_nid(pfn) == node))
246 register_page_bootmem_info_section(pfn);
247 }
248}
249#endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
250
251static int __meminit __add_section(int nid, unsigned long phys_start_pfn,
252 struct vmem_altmap *altmap, bool want_memblock)
253{
254 int ret;
255
256 if (pfn_valid(phys_start_pfn))
257 return -EEXIST;
258
259 ret = sparse_add_one_section(nid, phys_start_pfn, altmap);
260 if (ret < 0)
261 return ret;
262
263 if (!want_memblock)
264 return 0;
265
266 return hotplug_memory_register(nid, __pfn_to_section(phys_start_pfn));
267}
268
269/*
270 * Reasonably generic function for adding memory. It is
271 * expected that archs that support memory hotplug will
272 * call this function after deciding the zone to which to
273 * add the new pages.
274 */
275int __ref __add_pages(int nid, unsigned long phys_start_pfn,
276 unsigned long nr_pages, struct vmem_altmap *altmap,
277 bool want_memblock)
278{
279 unsigned long i;
280 int err = 0;
281 int start_sec, end_sec;
282
283 /* during initialize mem_map, align hot-added range to section */
284 start_sec = pfn_to_section_nr(phys_start_pfn);
285 end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
286
287 if (altmap) {
288 /*
289 * Validate altmap is within bounds of the total request
290 */
291 if (altmap->base_pfn != phys_start_pfn
292 || vmem_altmap_offset(altmap) > nr_pages) {
293 pr_warn_once("memory add fail, invalid altmap\n");
294 err = -EINVAL;
295 goto out;
296 }
297 altmap->alloc = 0;
298 }
299
300 for (i = start_sec; i <= end_sec; i++) {
301 err = __add_section(nid, section_nr_to_pfn(i), altmap,
302 want_memblock);
303
304 /*
305 * EEXIST is finally dealt with by ioresource collision
306 * check. see add_memory() => register_memory_resource()
307 * Warning will be printed if there is collision.
308 */
309 if (err && (err != -EEXIST))
310 break;
311 err = 0;
312 cond_resched();
313 }
314 vmemmap_populate_print_last();
315out:
316 return err;
317}
318
319#ifdef CONFIG_MEMORY_HOTREMOVE
320/* find the smallest valid pfn in the range [start_pfn, end_pfn) */
321static unsigned long find_smallest_section_pfn(int nid, struct zone *zone,
322 unsigned long start_pfn,
323 unsigned long end_pfn)
324{
325 struct mem_section *ms;
326
327 for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) {
328 ms = __pfn_to_section(start_pfn);
329
330 if (unlikely(!valid_section(ms)))
331 continue;
332
333 if (unlikely(pfn_to_nid(start_pfn) != nid))
334 continue;
335
336 if (zone && zone != page_zone(pfn_to_page(start_pfn)))
337 continue;
338
339 return start_pfn;
340 }
341
342 return 0;
343}
344
345/* find the biggest valid pfn in the range [start_pfn, end_pfn). */
346static unsigned long find_biggest_section_pfn(int nid, struct zone *zone,
347 unsigned long start_pfn,
348 unsigned long end_pfn)
349{
350 struct mem_section *ms;
351 unsigned long pfn;
352
353 /* pfn is the end pfn of a memory section. */
354 pfn = end_pfn - 1;
355 for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) {
356 ms = __pfn_to_section(pfn);
357
358 if (unlikely(!valid_section(ms)))
359 continue;
360
361 if (unlikely(pfn_to_nid(pfn) != nid))
362 continue;
363
364 if (zone && zone != page_zone(pfn_to_page(pfn)))
365 continue;
366
367 return pfn;
368 }
369
370 return 0;
371}
372
373static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
374 unsigned long end_pfn)
375{
376 unsigned long zone_start_pfn = zone->zone_start_pfn;
377 unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */
378 unsigned long zone_end_pfn = z;
379 unsigned long pfn;
380 struct mem_section *ms;
381 int nid = zone_to_nid(zone);
382
383 zone_span_writelock(zone);
384 if (zone_start_pfn == start_pfn) {
385 /*
386 * If the section is smallest section in the zone, it need
387 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
388 * In this case, we find second smallest valid mem_section
389 * for shrinking zone.
390 */
391 pfn = find_smallest_section_pfn(nid, zone, end_pfn,
392 zone_end_pfn);
393 if (pfn) {
394 zone->zone_start_pfn = pfn;
395 zone->spanned_pages = zone_end_pfn - pfn;
396 }
397 } else if (zone_end_pfn == end_pfn) {
398 /*
399 * If the section is biggest section in the zone, it need
400 * shrink zone->spanned_pages.
401 * In this case, we find second biggest valid mem_section for
402 * shrinking zone.
403 */
404 pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn,
405 start_pfn);
406 if (pfn)
407 zone->spanned_pages = pfn - zone_start_pfn + 1;
408 }
409
410 /*
411 * The section is not biggest or smallest mem_section in the zone, it
412 * only creates a hole in the zone. So in this case, we need not
413 * change the zone. But perhaps, the zone has only hole data. Thus
414 * it check the zone has only hole or not.
415 */
416 pfn = zone_start_pfn;
417 for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) {
418 ms = __pfn_to_section(pfn);
419
420 if (unlikely(!valid_section(ms)))
421 continue;
422
423 if (page_zone(pfn_to_page(pfn)) != zone)
424 continue;
425
426 /* If the section is current section, it continues the loop */
427 if (start_pfn == pfn)
428 continue;
429
430 /* If we find valid section, we have nothing to do */
431 zone_span_writeunlock(zone);
432 return;
433 }
434
435 /* The zone has no valid section */
436 zone->zone_start_pfn = 0;
437 zone->spanned_pages = 0;
438 zone_span_writeunlock(zone);
439}
440
441static void shrink_pgdat_span(struct pglist_data *pgdat,
442 unsigned long start_pfn, unsigned long end_pfn)
443{
444 unsigned long pgdat_start_pfn = pgdat->node_start_pfn;
445 unsigned long p = pgdat_end_pfn(pgdat); /* pgdat_end_pfn namespace clash */
446 unsigned long pgdat_end_pfn = p;
447 unsigned long pfn;
448 struct mem_section *ms;
449 int nid = pgdat->node_id;
450
451 if (pgdat_start_pfn == start_pfn) {
452 /*
453 * If the section is smallest section in the pgdat, it need
454 * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages.
455 * In this case, we find second smallest valid mem_section
456 * for shrinking zone.
457 */
458 pfn = find_smallest_section_pfn(nid, NULL, end_pfn,
459 pgdat_end_pfn);
460 if (pfn) {
461 pgdat->node_start_pfn = pfn;
462 pgdat->node_spanned_pages = pgdat_end_pfn - pfn;
463 }
464 } else if (pgdat_end_pfn == end_pfn) {
465 /*
466 * If the section is biggest section in the pgdat, it need
467 * shrink pgdat->node_spanned_pages.
468 * In this case, we find second biggest valid mem_section for
469 * shrinking zone.
470 */
471 pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn,
472 start_pfn);
473 if (pfn)
474 pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1;
475 }
476
477 /*
478 * If the section is not biggest or smallest mem_section in the pgdat,
479 * it only creates a hole in the pgdat. So in this case, we need not
480 * change the pgdat.
481 * But perhaps, the pgdat has only hole data. Thus it check the pgdat
482 * has only hole or not.
483 */
484 pfn = pgdat_start_pfn;
485 for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) {
486 ms = __pfn_to_section(pfn);
487
488 if (unlikely(!valid_section(ms)))
489 continue;
490
491 if (pfn_to_nid(pfn) != nid)
492 continue;
493
494 /* If the section is current section, it continues the loop */
495 if (start_pfn == pfn)
496 continue;
497
498 /* If we find valid section, we have nothing to do */
499 return;
500 }
501
502 /* The pgdat has no valid section */
503 pgdat->node_start_pfn = 0;
504 pgdat->node_spanned_pages = 0;
505}
506
507static void __remove_zone(struct zone *zone, unsigned long start_pfn)
508{
509 struct pglist_data *pgdat = zone->zone_pgdat;
510 int nr_pages = PAGES_PER_SECTION;
511 unsigned long flags;
512
513 pgdat_resize_lock(zone->zone_pgdat, &flags);
514 shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
515 shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages);
516 pgdat_resize_unlock(zone->zone_pgdat, &flags);
517}
518
519static int __remove_section(struct zone *zone, struct mem_section *ms,
520 unsigned long map_offset, struct vmem_altmap *altmap)
521{
522 unsigned long start_pfn;
523 int scn_nr;
524 int ret = -EINVAL;
525
526 if (!valid_section(ms))
527 return ret;
528
529 ret = unregister_memory_section(ms);
530 if (ret)
531 return ret;
532
533 scn_nr = __section_nr(ms);
534 start_pfn = section_nr_to_pfn((unsigned long)scn_nr);
535 __remove_zone(zone, start_pfn);
536
537 sparse_remove_one_section(zone, ms, map_offset, altmap);
538 return 0;
539}
540
541/**
542 * __remove_pages() - remove sections of pages from a zone
543 * @zone: zone from which pages need to be removed
544 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
545 * @nr_pages: number of pages to remove (must be multiple of section size)
546 * @altmap: alternative device page map or %NULL if default memmap is used
547 *
548 * Generic helper function to remove section mappings and sysfs entries
549 * for the section of the memory we are removing. Caller needs to make
550 * sure that pages are marked reserved and zones are adjust properly by
551 * calling offline_pages().
552 */
553int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
554 unsigned long nr_pages, struct vmem_altmap *altmap)
555{
556 unsigned long i;
557 unsigned long map_offset = 0;
558 int sections_to_remove, ret = 0;
559
560 /* In the ZONE_DEVICE case device driver owns the memory region */
561 if (is_dev_zone(zone)) {
562 if (altmap)
563 map_offset = vmem_altmap_offset(altmap);
564 } else {
565 resource_size_t start, size;
566
567 start = phys_start_pfn << PAGE_SHIFT;
568 size = nr_pages * PAGE_SIZE;
569
570 ret = release_mem_region_adjustable(&iomem_resource, start,
571 size);
572 if (ret) {
573 resource_size_t endres = start + size - 1;
574
575 pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
576 &start, &endres, ret);
577 }
578 }
579
580 clear_zone_contiguous(zone);
581
582 /*
583 * We can only remove entire sections
584 */
585 BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
586 BUG_ON(nr_pages % PAGES_PER_SECTION);
587
588 sections_to_remove = nr_pages / PAGES_PER_SECTION;
589 for (i = 0; i < sections_to_remove; i++) {
590 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
591
592 cond_resched();
593 ret = __remove_section(zone, __pfn_to_section(pfn), map_offset,
594 altmap);
595 map_offset = 0;
596 if (ret)
597 break;
598 }
599
600 set_zone_contiguous(zone);
601
602 return ret;
603}
604#endif /* CONFIG_MEMORY_HOTREMOVE */
605
606int set_online_page_callback(online_page_callback_t callback)
607{
608 int rc = -EINVAL;
609
610 get_online_mems();
611 mutex_lock(&online_page_callback_lock);
612
613 if (online_page_callback == generic_online_page) {
614 online_page_callback = callback;
615 rc = 0;
616 }
617
618 mutex_unlock(&online_page_callback_lock);
619 put_online_mems();
620
621 return rc;
622}
623EXPORT_SYMBOL_GPL(set_online_page_callback);
624
625int restore_online_page_callback(online_page_callback_t callback)
626{
627 int rc = -EINVAL;
628
629 get_online_mems();
630 mutex_lock(&online_page_callback_lock);
631
632 if (online_page_callback == callback) {
633 online_page_callback = generic_online_page;
634 rc = 0;
635 }
636
637 mutex_unlock(&online_page_callback_lock);
638 put_online_mems();
639
640 return rc;
641}
642EXPORT_SYMBOL_GPL(restore_online_page_callback);
643
644void __online_page_set_limits(struct page *page)
645{
646}
647EXPORT_SYMBOL_GPL(__online_page_set_limits);
648
649void __online_page_increment_counters(struct page *page)
650{
651 adjust_managed_page_count(page, 1);
652}
653EXPORT_SYMBOL_GPL(__online_page_increment_counters);
654
655void __online_page_free(struct page *page)
656{
657 __free_reserved_page(page);
658}
659EXPORT_SYMBOL_GPL(__online_page_free);
660
661static void generic_online_page(struct page *page, unsigned int order)
662{
663 kernel_map_pages(page, 1 << order, 1);
664 __free_pages_core(page, order);
665 totalram_pages_add(1UL << order);
666#ifdef CONFIG_HIGHMEM
667 if (PageHighMem(page))
668 totalhigh_pages_add(1UL << order);
669#endif
670}
671
672static int online_pages_blocks(unsigned long start, unsigned long nr_pages)
673{
674 unsigned long end = start + nr_pages;
675 int order, onlined_pages = 0;
676
677 while (start < end) {
678 order = min(MAX_ORDER - 1,
679 get_order(PFN_PHYS(end) - PFN_PHYS(start)));
680 (*online_page_callback)(pfn_to_page(start), order);
681
682 onlined_pages += (1UL << order);
683 start += (1UL << order);
684 }
685 return onlined_pages;
686}
687
688static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
689 void *arg)
690{
691 unsigned long onlined_pages = *(unsigned long *)arg;
692
693 if (PageReserved(pfn_to_page(start_pfn)))
694 onlined_pages += online_pages_blocks(start_pfn, nr_pages);
695
696 online_mem_sections(start_pfn, start_pfn + nr_pages);
697
698 *(unsigned long *)arg = onlined_pages;
699 return 0;
700}
701
702/* check which state of node_states will be changed when online memory */
703static void node_states_check_changes_online(unsigned long nr_pages,
704 struct zone *zone, struct memory_notify *arg)
705{
706 int nid = zone_to_nid(zone);
707
708 arg->status_change_nid = NUMA_NO_NODE;
709 arg->status_change_nid_normal = NUMA_NO_NODE;
710 arg->status_change_nid_high = NUMA_NO_NODE;
711
712 if (!node_state(nid, N_MEMORY))
713 arg->status_change_nid = nid;
714 if (zone_idx(zone) <= ZONE_NORMAL && !node_state(nid, N_NORMAL_MEMORY))
715 arg->status_change_nid_normal = nid;
716#ifdef CONFIG_HIGHMEM
717 if (zone_idx(zone) <= N_HIGH_MEMORY && !node_state(nid, N_HIGH_MEMORY))
718 arg->status_change_nid_high = nid;
719#endif
720}
721
722static void node_states_set_node(int node, struct memory_notify *arg)
723{
724 if (arg->status_change_nid_normal >= 0)
725 node_set_state(node, N_NORMAL_MEMORY);
726
727 if (arg->status_change_nid_high >= 0)
728 node_set_state(node, N_HIGH_MEMORY);
729
730 if (arg->status_change_nid >= 0)
731 node_set_state(node, N_MEMORY);
732}
733
734static void __meminit resize_zone_range(struct zone *zone, unsigned long start_pfn,
735 unsigned long nr_pages)
736{
737 unsigned long old_end_pfn = zone_end_pfn(zone);
738
739 if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
740 zone->zone_start_pfn = start_pfn;
741
742 zone->spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - zone->zone_start_pfn;
743}
744
745static void __meminit resize_pgdat_range(struct pglist_data *pgdat, unsigned long start_pfn,
746 unsigned long nr_pages)
747{
748 unsigned long old_end_pfn = pgdat_end_pfn(pgdat);
749
750 if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
751 pgdat->node_start_pfn = start_pfn;
752
753 pgdat->node_spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - pgdat->node_start_pfn;
754}
755
756void __ref move_pfn_range_to_zone(struct zone *zone, unsigned long start_pfn,
757 unsigned long nr_pages, struct vmem_altmap *altmap)
758{
759 struct pglist_data *pgdat = zone->zone_pgdat;
760 int nid = pgdat->node_id;
761 unsigned long flags;
762
763 clear_zone_contiguous(zone);
764
765 /* TODO Huh pgdat is irqsave while zone is not. It used to be like that before */
766 pgdat_resize_lock(pgdat, &flags);
767 zone_span_writelock(zone);
768 if (zone_is_empty(zone))
769 init_currently_empty_zone(zone, start_pfn, nr_pages);
770 resize_zone_range(zone, start_pfn, nr_pages);
771 zone_span_writeunlock(zone);
772 resize_pgdat_range(pgdat, start_pfn, nr_pages);
773 pgdat_resize_unlock(pgdat, &flags);
774
775 /*
776 * TODO now we have a visible range of pages which are not associated
777 * with their zone properly. Not nice but set_pfnblock_flags_mask
778 * expects the zone spans the pfn range. All the pages in the range
779 * are reserved so nobody should be touching them so we should be safe
780 */
781 memmap_init_zone(nr_pages, nid, zone_idx(zone), start_pfn,
782 MEMMAP_HOTPLUG, altmap);
783
784 set_zone_contiguous(zone);
785}
786
787/*
788 * Returns a default kernel memory zone for the given pfn range.
789 * If no kernel zone covers this pfn range it will automatically go
790 * to the ZONE_NORMAL.
791 */
792static struct zone *default_kernel_zone_for_pfn(int nid, unsigned long start_pfn,
793 unsigned long nr_pages)
794{
795 struct pglist_data *pgdat = NODE_DATA(nid);
796 int zid;
797
798 for (zid = 0; zid <= ZONE_NORMAL; zid++) {
799 struct zone *zone = &pgdat->node_zones[zid];
800
801 if (zone_intersects(zone, start_pfn, nr_pages))
802 return zone;
803 }
804
805 return &pgdat->node_zones[ZONE_NORMAL];
806}
807
808static inline struct zone *default_zone_for_pfn(int nid, unsigned long start_pfn,
809 unsigned long nr_pages)
810{
811 struct zone *kernel_zone = default_kernel_zone_for_pfn(nid, start_pfn,
812 nr_pages);
813 struct zone *movable_zone = &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
814 bool in_kernel = zone_intersects(kernel_zone, start_pfn, nr_pages);
815 bool in_movable = zone_intersects(movable_zone, start_pfn, nr_pages);
816
817 /*
818 * We inherit the existing zone in a simple case where zones do not
819 * overlap in the given range
820 */
821 if (in_kernel ^ in_movable)
822 return (in_kernel) ? kernel_zone : movable_zone;
823
824 /*
825 * If the range doesn't belong to any zone or two zones overlap in the
826 * given range then we use movable zone only if movable_node is
827 * enabled because we always online to a kernel zone by default.
828 */
829 return movable_node_enabled ? movable_zone : kernel_zone;
830}
831
832struct zone * zone_for_pfn_range(int online_type, int nid, unsigned start_pfn,
833 unsigned long nr_pages)
834{
835 if (online_type == MMOP_ONLINE_KERNEL)
836 return default_kernel_zone_for_pfn(nid, start_pfn, nr_pages);
837
838 if (online_type == MMOP_ONLINE_MOVABLE)
839 return &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
840
841 return default_zone_for_pfn(nid, start_pfn, nr_pages);
842}
843
844/*
845 * Associates the given pfn range with the given node and the zone appropriate
846 * for the given online type.
847 */
848static struct zone * __meminit move_pfn_range(int online_type, int nid,
849 unsigned long start_pfn, unsigned long nr_pages)
850{
851 struct zone *zone;
852
853 zone = zone_for_pfn_range(online_type, nid, start_pfn, nr_pages);
854 move_pfn_range_to_zone(zone, start_pfn, nr_pages, NULL);
855 return zone;
856}
857
858int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type)
859{
860 unsigned long flags;
861 unsigned long onlined_pages = 0;
862 struct zone *zone;
863 int need_zonelists_rebuild = 0;
864 int nid;
865 int ret;
866 struct memory_notify arg;
867 struct memory_block *mem;
868
869 mem_hotplug_begin();
870
871 /*
872 * We can't use pfn_to_nid() because nid might be stored in struct page
873 * which is not yet initialized. Instead, we find nid from memory block.
874 */
875 mem = find_memory_block(__pfn_to_section(pfn));
876 nid = mem->nid;
877
878 /* associate pfn range with the zone */
879 zone = move_pfn_range(online_type, nid, pfn, nr_pages);
880
881 arg.start_pfn = pfn;
882 arg.nr_pages = nr_pages;
883 node_states_check_changes_online(nr_pages, zone, &arg);
884
885 ret = memory_notify(MEM_GOING_ONLINE, &arg);
886 ret = notifier_to_errno(ret);
887 if (ret)
888 goto failed_addition;
889
890 /*
891 * If this zone is not populated, then it is not in zonelist.
892 * This means the page allocator ignores this zone.
893 * So, zonelist must be updated after online.
894 */
895 if (!populated_zone(zone)) {
896 need_zonelists_rebuild = 1;
897 setup_zone_pageset(zone);
898 }
899
900 ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
901 online_pages_range);
902 if (ret) {
903 if (need_zonelists_rebuild)
904 zone_pcp_reset(zone);
905 goto failed_addition;
906 }
907
908 zone->present_pages += onlined_pages;
909
910 pgdat_resize_lock(zone->zone_pgdat, &flags);
911 zone->zone_pgdat->node_present_pages += onlined_pages;
912 pgdat_resize_unlock(zone->zone_pgdat, &flags);
913
914 if (onlined_pages) {
915 node_states_set_node(nid, &arg);
916 if (need_zonelists_rebuild)
917 build_all_zonelists(NULL);
918 else
919 zone_pcp_update(zone);
920 }
921
922 init_per_zone_wmark_min();
923
924 if (onlined_pages) {
925 kswapd_run(nid);
926 kcompactd_run(nid);
927 }
928
929 vm_total_pages = nr_free_pagecache_pages();
930
931 writeback_set_ratelimit();
932
933 if (onlined_pages)
934 memory_notify(MEM_ONLINE, &arg);
935 mem_hotplug_done();
936 return 0;
937
938failed_addition:
939 pr_debug("online_pages [mem %#010llx-%#010llx] failed\n",
940 (unsigned long long) pfn << PAGE_SHIFT,
941 (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1);
942 memory_notify(MEM_CANCEL_ONLINE, &arg);
943 mem_hotplug_done();
944 return ret;
945}
946#endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
947
948static void reset_node_present_pages(pg_data_t *pgdat)
949{
950 struct zone *z;
951
952 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
953 z->present_pages = 0;
954
955 pgdat->node_present_pages = 0;
956}
957
958/* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
959static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
960{
961 struct pglist_data *pgdat;
962 unsigned long start_pfn = PFN_DOWN(start);
963
964 pgdat = NODE_DATA(nid);
965 if (!pgdat) {
966 pgdat = arch_alloc_nodedata(nid);
967 if (!pgdat)
968 return NULL;
969
970 arch_refresh_nodedata(nid, pgdat);
971 } else {
972 /*
973 * Reset the nr_zones, order and classzone_idx before reuse.
974 * Note that kswapd will init kswapd_classzone_idx properly
975 * when it starts in the near future.
976 */
977 pgdat->nr_zones = 0;
978 pgdat->kswapd_order = 0;
979 pgdat->kswapd_classzone_idx = 0;
980 }
981
982 /* we can use NODE_DATA(nid) from here */
983
984 pgdat->node_id = nid;
985 pgdat->node_start_pfn = start_pfn;
986
987 /* init node's zones as empty zones, we don't have any present pages.*/
988 free_area_init_core_hotplug(nid);
989 pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat);
990
991 /*
992 * The node we allocated has no zone fallback lists. For avoiding
993 * to access not-initialized zonelist, build here.
994 */
995 build_all_zonelists(pgdat);
996
997 /*
998 * When memory is hot-added, all the memory is in offline state. So
999 * clear all zones' present_pages because they will be updated in
1000 * online_pages() and offline_pages().
1001 */
1002 reset_node_managed_pages(pgdat);
1003 reset_node_present_pages(pgdat);
1004
1005 return pgdat;
1006}
1007
1008static void rollback_node_hotadd(int nid)
1009{
1010 pg_data_t *pgdat = NODE_DATA(nid);
1011
1012 arch_refresh_nodedata(nid, NULL);
1013 free_percpu(pgdat->per_cpu_nodestats);
1014 arch_free_nodedata(pgdat);
1015 return;
1016}
1017
1018
1019/**
1020 * try_online_node - online a node if offlined
1021 * @nid: the node ID
1022 * @start: start addr of the node
1023 * @set_node_online: Whether we want to online the node
1024 * called by cpu_up() to online a node without onlined memory.
1025 *
1026 * Returns:
1027 * 1 -> a new node has been allocated
1028 * 0 -> the node is already online
1029 * -ENOMEM -> the node could not be allocated
1030 */
1031static int __try_online_node(int nid, u64 start, bool set_node_online)
1032{
1033 pg_data_t *pgdat;
1034 int ret = 1;
1035
1036 if (node_online(nid))
1037 return 0;
1038
1039 pgdat = hotadd_new_pgdat(nid, start);
1040 if (!pgdat) {
1041 pr_err("Cannot online node %d due to NULL pgdat\n", nid);
1042 ret = -ENOMEM;
1043 goto out;
1044 }
1045
1046 if (set_node_online) {
1047 node_set_online(nid);
1048 ret = register_one_node(nid);
1049 BUG_ON(ret);
1050 }
1051out:
1052 return ret;
1053}
1054
1055/*
1056 * Users of this function always want to online/register the node
1057 */
1058int try_online_node(int nid)
1059{
1060 int ret;
1061
1062 mem_hotplug_begin();
1063 ret = __try_online_node(nid, 0, true);
1064 mem_hotplug_done();
1065 return ret;
1066}
1067
1068static int check_hotplug_memory_range(u64 start, u64 size)
1069{
1070 unsigned long block_sz = memory_block_size_bytes();
1071 u64 block_nr_pages = block_sz >> PAGE_SHIFT;
1072 u64 nr_pages = size >> PAGE_SHIFT;
1073 u64 start_pfn = PFN_DOWN(start);
1074
1075 /* memory range must be block size aligned */
1076 if (!nr_pages || !IS_ALIGNED(start_pfn, block_nr_pages) ||
1077 !IS_ALIGNED(nr_pages, block_nr_pages)) {
1078 pr_err("Block size [%#lx] unaligned hotplug range: start %#llx, size %#llx",
1079 block_sz, start, size);
1080 return -EINVAL;
1081 }
1082
1083 return 0;
1084}
1085
1086static int online_memory_block(struct memory_block *mem, void *arg)
1087{
1088 return device_online(&mem->dev);
1089}
1090
1091/*
1092 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1093 * and online/offline operations (triggered e.g. by sysfs).
1094 *
1095 * we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG
1096 */
1097int __ref add_memory_resource(int nid, struct resource *res)
1098{
1099 u64 start, size;
1100 bool new_node = false;
1101 int ret;
1102
1103 start = res->start;
1104 size = resource_size(res);
1105
1106 ret = check_hotplug_memory_range(start, size);
1107 if (ret)
1108 return ret;
1109
1110 mem_hotplug_begin();
1111
1112 /*
1113 * Add new range to memblock so that when hotadd_new_pgdat() is called
1114 * to allocate new pgdat, get_pfn_range_for_nid() will be able to find
1115 * this new range and calculate total pages correctly. The range will
1116 * be removed at hot-remove time.
1117 */
1118 memblock_add_node(start, size, nid);
1119
1120 ret = __try_online_node(nid, start, false);
1121 if (ret < 0)
1122 goto error;
1123 new_node = ret;
1124
1125 /* call arch's memory hotadd */
1126 ret = arch_add_memory(nid, start, size, NULL, true);
1127 if (ret < 0)
1128 goto error;
1129
1130 if (new_node) {
1131 /* If sysfs file of new node can't be created, cpu on the node
1132 * can't be hot-added. There is no rollback way now.
1133 * So, check by BUG_ON() to catch it reluctantly..
1134 * We online node here. We can't roll back from here.
1135 */
1136 node_set_online(nid);
1137 ret = __register_one_node(nid);
1138 BUG_ON(ret);
1139 }
1140
1141 /* link memory sections under this node.*/
1142 ret = link_mem_sections(nid, PFN_DOWN(start), PFN_UP(start + size - 1));
1143 BUG_ON(ret);
1144
1145 /* create new memmap entry */
1146 firmware_map_add_hotplug(start, start + size, "System RAM");
1147
1148 /* device_online() will take the lock when calling online_pages() */
1149 mem_hotplug_done();
1150
1151 /* online pages if requested */
1152 if (memhp_auto_online)
1153 walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1),
1154 NULL, online_memory_block);
1155
1156 return ret;
1157error:
1158 /* rollback pgdat allocation and others */
1159 if (new_node)
1160 rollback_node_hotadd(nid);
1161 memblock_remove(start, size);
1162 mem_hotplug_done();
1163 return ret;
1164}
1165
1166/* requires device_hotplug_lock, see add_memory_resource() */
1167int __ref __add_memory(int nid, u64 start, u64 size)
1168{
1169 struct resource *res;
1170 int ret;
1171
1172 res = register_memory_resource(start, size);
1173 if (IS_ERR(res))
1174 return PTR_ERR(res);
1175
1176 ret = add_memory_resource(nid, res);
1177 if (ret < 0)
1178 release_memory_resource(res);
1179 return ret;
1180}
1181
1182int add_memory(int nid, u64 start, u64 size)
1183{
1184 int rc;
1185
1186 lock_device_hotplug();
1187 rc = __add_memory(nid, start, size);
1188 unlock_device_hotplug();
1189
1190 return rc;
1191}
1192EXPORT_SYMBOL_GPL(add_memory);
1193
1194#ifdef CONFIG_MEMORY_HOTREMOVE
1195/*
1196 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1197 * set and the size of the free page is given by page_order(). Using this,
1198 * the function determines if the pageblock contains only free pages.
1199 * Due to buddy contraints, a free page at least the size of a pageblock will
1200 * be located at the start of the pageblock
1201 */
1202static inline int pageblock_free(struct page *page)
1203{
1204 return PageBuddy(page) && page_order(page) >= pageblock_order;
1205}
1206
1207/* Return the pfn of the start of the next active pageblock after a given pfn */
1208static unsigned long next_active_pageblock(unsigned long pfn)
1209{
1210 struct page *page = pfn_to_page(pfn);
1211
1212 /* Ensure the starting page is pageblock-aligned */
1213 BUG_ON(pfn & (pageblock_nr_pages - 1));
1214
1215 /* If the entire pageblock is free, move to the end of free page */
1216 if (pageblock_free(page)) {
1217 int order;
1218 /* be careful. we don't have locks, page_order can be changed.*/
1219 order = page_order(page);
1220 if ((order < MAX_ORDER) && (order >= pageblock_order))
1221 return pfn + (1 << order);
1222 }
1223
1224 return pfn + pageblock_nr_pages;
1225}
1226
1227static bool is_pageblock_removable_nolock(unsigned long pfn)
1228{
1229 struct page *page = pfn_to_page(pfn);
1230 struct zone *zone;
1231
1232 /*
1233 * We have to be careful here because we are iterating over memory
1234 * sections which are not zone aware so we might end up outside of
1235 * the zone but still within the section.
1236 * We have to take care about the node as well. If the node is offline
1237 * its NODE_DATA will be NULL - see page_zone.
1238 */
1239 if (!node_online(page_to_nid(page)))
1240 return false;
1241
1242 zone = page_zone(page);
1243 pfn = page_to_pfn(page);
1244 if (!zone_spans_pfn(zone, pfn))
1245 return false;
1246
1247 return !has_unmovable_pages(zone, page, 0, MIGRATE_MOVABLE, SKIP_HWPOISON);
1248}
1249
1250/* Checks if this range of memory is likely to be hot-removable. */
1251bool is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
1252{
1253 unsigned long end_pfn, pfn;
1254
1255 end_pfn = min(start_pfn + nr_pages,
1256 zone_end_pfn(page_zone(pfn_to_page(start_pfn))));
1257
1258 /* Check the starting page of each pageblock within the range */
1259 for (pfn = start_pfn; pfn < end_pfn; pfn = next_active_pageblock(pfn)) {
1260 if (!is_pageblock_removable_nolock(pfn))
1261 return false;
1262 cond_resched();
1263 }
1264
1265 /* All pageblocks in the memory block are likely to be hot-removable */
1266 return true;
1267}
1268
1269/*
1270 * Confirm all pages in a range [start, end) belong to the same zone.
1271 * When true, return its valid [start, end).
1272 */
1273int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn,
1274 unsigned long *valid_start, unsigned long *valid_end)
1275{
1276 unsigned long pfn, sec_end_pfn;
1277 unsigned long start, end;
1278 struct zone *zone = NULL;
1279 struct page *page;
1280 int i;
1281 for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1);
1282 pfn < end_pfn;
1283 pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) {
1284 /* Make sure the memory section is present first */
1285 if (!present_section_nr(pfn_to_section_nr(pfn)))
1286 continue;
1287 for (; pfn < sec_end_pfn && pfn < end_pfn;
1288 pfn += MAX_ORDER_NR_PAGES) {
1289 i = 0;
1290 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1291 while ((i < MAX_ORDER_NR_PAGES) &&
1292 !pfn_valid_within(pfn + i))
1293 i++;
1294 if (i == MAX_ORDER_NR_PAGES || pfn + i >= end_pfn)
1295 continue;
1296 /* Check if we got outside of the zone */
1297 if (zone && !zone_spans_pfn(zone, pfn + i))
1298 return 0;
1299 page = pfn_to_page(pfn + i);
1300 if (zone && page_zone(page) != zone)
1301 return 0;
1302 if (!zone)
1303 start = pfn + i;
1304 zone = page_zone(page);
1305 end = pfn + MAX_ORDER_NR_PAGES;
1306 }
1307 }
1308
1309 if (zone) {
1310 *valid_start = start;
1311 *valid_end = min(end, end_pfn);
1312 return 1;
1313 } else {
1314 return 0;
1315 }
1316}
1317
1318/*
1319 * Scan pfn range [start,end) to find movable/migratable pages (LRU pages,
1320 * non-lru movable pages and hugepages). We scan pfn because it's much
1321 * easier than scanning over linked list. This function returns the pfn
1322 * of the first found movable page if it's found, otherwise 0.
1323 */
1324static unsigned long scan_movable_pages(unsigned long start, unsigned long end)
1325{
1326 unsigned long pfn;
1327
1328 for (pfn = start; pfn < end; pfn++) {
1329 struct page *page, *head;
1330 unsigned long skip;
1331
1332 if (!pfn_valid(pfn))
1333 continue;
1334 page = pfn_to_page(pfn);
1335 if (PageLRU(page))
1336 return pfn;
1337 if (__PageMovable(page))
1338 return pfn;
1339
1340 if (!PageHuge(page))
1341 continue;
1342 head = compound_head(page);
1343 if (hugepage_migration_supported(page_hstate(head)) &&
1344 page_huge_active(head))
1345 return pfn;
1346 skip = (1 << compound_order(head)) - (page - head);
1347 pfn += skip - 1;
1348 }
1349 return 0;
1350}
1351
1352static struct page *new_node_page(struct page *page, unsigned long private)
1353{
1354 int nid = page_to_nid(page);
1355 nodemask_t nmask = node_states[N_MEMORY];
1356
1357 /*
1358 * try to allocate from a different node but reuse this node if there
1359 * are no other online nodes to be used (e.g. we are offlining a part
1360 * of the only existing node)
1361 */
1362 node_clear(nid, nmask);
1363 if (nodes_empty(nmask))
1364 node_set(nid, nmask);
1365
1366 return new_page_nodemask(page, nid, &nmask);
1367}
1368
1369static int
1370do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1371{
1372 unsigned long pfn;
1373 struct page *page;
1374 int ret = 0;
1375 LIST_HEAD(source);
1376
1377 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1378 if (!pfn_valid(pfn))
1379 continue;
1380 page = pfn_to_page(pfn);
1381
1382 if (PageHuge(page)) {
1383 struct page *head = compound_head(page);
1384 if (compound_order(head) > PFN_SECTION_SHIFT) {
1385 ret = -EBUSY;
1386 break;
1387 }
1388 pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1;
1389 isolate_huge_page(head, &source);
1390 continue;
1391 } else if (PageTransHuge(page))
1392 pfn = page_to_pfn(compound_head(page))
1393 + hpage_nr_pages(page) - 1;
1394
1395 /*
1396 * HWPoison pages have elevated reference counts so the migration would
1397 * fail on them. It also doesn't make any sense to migrate them in the
1398 * first place. Still try to unmap such a page in case it is still mapped
1399 * (e.g. current hwpoison implementation doesn't unmap KSM pages but keep
1400 * the unmap as the catch all safety net).
1401 */
1402 if (PageHWPoison(page)) {
1403 if (WARN_ON(PageLRU(page)))
1404 isolate_lru_page(page);
1405 if (page_mapped(page))
1406 try_to_unmap(page, TTU_IGNORE_MLOCK | TTU_IGNORE_ACCESS);
1407 continue;
1408 }
1409
1410 if (!get_page_unless_zero(page))
1411 continue;
1412 /*
1413 * We can skip free pages. And we can deal with pages on
1414 * LRU and non-lru movable pages.
1415 */
1416 if (PageLRU(page))
1417 ret = isolate_lru_page(page);
1418 else
1419 ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE);
1420 if (!ret) { /* Success */
1421 list_add_tail(&page->lru, &source);
1422 if (!__PageMovable(page))
1423 inc_node_page_state(page, NR_ISOLATED_ANON +
1424 page_is_file_cache(page));
1425
1426 } else {
1427 pr_warn("failed to isolate pfn %lx\n", pfn);
1428 dump_page(page, "isolation failed");
1429 }
1430 put_page(page);
1431 }
1432 if (!list_empty(&source)) {
1433 /* Allocate a new page from the nearest neighbor node */
1434 ret = migrate_pages(&source, new_node_page, NULL, 0,
1435 MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
1436 if (ret) {
1437 list_for_each_entry(page, &source, lru) {
1438 pr_warn("migrating pfn %lx failed ret:%d ",
1439 page_to_pfn(page), ret);
1440 dump_page(page, "migration failure");
1441 }
1442 putback_movable_pages(&source);
1443 }
1444 }
1445
1446 return ret;
1447}
1448
1449/*
1450 * remove from free_area[] and mark all as Reserved.
1451 */
1452static int
1453offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1454 void *data)
1455{
1456 __offline_isolated_pages(start, start + nr_pages);
1457 return 0;
1458}
1459
1460static void
1461offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
1462{
1463 walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
1464 offline_isolated_pages_cb);
1465}
1466
1467/*
1468 * Check all pages in range, recoreded as memory resource, are isolated.
1469 */
1470static int
1471check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1472 void *data)
1473{
1474 int ret;
1475 long offlined = *(long *)data;
1476 ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true);
1477 offlined = nr_pages;
1478 if (!ret)
1479 *(long *)data += offlined;
1480 return ret;
1481}
1482
1483static long
1484check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
1485{
1486 long offlined = 0;
1487 int ret;
1488
1489 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
1490 check_pages_isolated_cb);
1491 if (ret < 0)
1492 offlined = (long)ret;
1493 return offlined;
1494}
1495
1496static int __init cmdline_parse_movable_node(char *p)
1497{
1498#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1499 movable_node_enabled = true;
1500#else
1501 pr_warn("movable_node parameter depends on CONFIG_HAVE_MEMBLOCK_NODE_MAP to work properly\n");
1502#endif
1503 return 0;
1504}
1505early_param("movable_node", cmdline_parse_movable_node);
1506
1507/* check which state of node_states will be changed when offline memory */
1508static void node_states_check_changes_offline(unsigned long nr_pages,
1509 struct zone *zone, struct memory_notify *arg)
1510{
1511 struct pglist_data *pgdat = zone->zone_pgdat;
1512 unsigned long present_pages = 0;
1513 enum zone_type zt;
1514
1515 arg->status_change_nid = NUMA_NO_NODE;
1516 arg->status_change_nid_normal = NUMA_NO_NODE;
1517 arg->status_change_nid_high = NUMA_NO_NODE;
1518
1519 /*
1520 * Check whether node_states[N_NORMAL_MEMORY] will be changed.
1521 * If the memory to be offline is within the range
1522 * [0..ZONE_NORMAL], and it is the last present memory there,
1523 * the zones in that range will become empty after the offlining,
1524 * thus we can determine that we need to clear the node from
1525 * node_states[N_NORMAL_MEMORY].
1526 */
1527 for (zt = 0; zt <= ZONE_NORMAL; zt++)
1528 present_pages += pgdat->node_zones[zt].present_pages;
1529 if (zone_idx(zone) <= ZONE_NORMAL && nr_pages >= present_pages)
1530 arg->status_change_nid_normal = zone_to_nid(zone);
1531
1532#ifdef CONFIG_HIGHMEM
1533 /*
1534 * node_states[N_HIGH_MEMORY] contains nodes which
1535 * have normal memory or high memory.
1536 * Here we add the present_pages belonging to ZONE_HIGHMEM.
1537 * If the zone is within the range of [0..ZONE_HIGHMEM), and
1538 * we determine that the zones in that range become empty,
1539 * we need to clear the node for N_HIGH_MEMORY.
1540 */
1541 present_pages += pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1542 if (zone_idx(zone) <= ZONE_HIGHMEM && nr_pages >= present_pages)
1543 arg->status_change_nid_high = zone_to_nid(zone);
1544#endif
1545
1546 /*
1547 * We have accounted the pages from [0..ZONE_NORMAL), and
1548 * in case of CONFIG_HIGHMEM the pages from ZONE_HIGHMEM
1549 * as well.
1550 * Here we count the possible pages from ZONE_MOVABLE.
1551 * If after having accounted all the pages, we see that the nr_pages
1552 * to be offlined is over or equal to the accounted pages,
1553 * we know that the node will become empty, and so, we can clear
1554 * it for N_MEMORY as well.
1555 */
1556 present_pages += pgdat->node_zones[ZONE_MOVABLE].present_pages;
1557
1558 if (nr_pages >= present_pages)
1559 arg->status_change_nid = zone_to_nid(zone);
1560}
1561
1562static void node_states_clear_node(int node, struct memory_notify *arg)
1563{
1564 if (arg->status_change_nid_normal >= 0)
1565 node_clear_state(node, N_NORMAL_MEMORY);
1566
1567 if (arg->status_change_nid_high >= 0)
1568 node_clear_state(node, N_HIGH_MEMORY);
1569
1570 if (arg->status_change_nid >= 0)
1571 node_clear_state(node, N_MEMORY);
1572}
1573
1574static int __ref __offline_pages(unsigned long start_pfn,
1575 unsigned long end_pfn)
1576{
1577 unsigned long pfn, nr_pages;
1578 long offlined_pages;
1579 int ret, node;
1580 unsigned long flags;
1581 unsigned long valid_start, valid_end;
1582 struct zone *zone;
1583 struct memory_notify arg;
1584 char *reason;
1585
1586 mem_hotplug_begin();
1587
1588 /* This makes hotplug much easier...and readable.
1589 we assume this for now. .*/
1590 if (!test_pages_in_a_zone(start_pfn, end_pfn, &valid_start,
1591 &valid_end)) {
1592 ret = -EINVAL;
1593 reason = "multizone range";
1594 goto failed_removal;
1595 }
1596
1597 zone = page_zone(pfn_to_page(valid_start));
1598 node = zone_to_nid(zone);
1599 nr_pages = end_pfn - start_pfn;
1600
1601 /* set above range as isolated */
1602 ret = start_isolate_page_range(start_pfn, end_pfn,
1603 MIGRATE_MOVABLE,
1604 SKIP_HWPOISON | REPORT_FAILURE);
1605 if (ret) {
1606 reason = "failure to isolate range";
1607 goto failed_removal;
1608 }
1609
1610 arg.start_pfn = start_pfn;
1611 arg.nr_pages = nr_pages;
1612 node_states_check_changes_offline(nr_pages, zone, &arg);
1613
1614 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1615 ret = notifier_to_errno(ret);
1616 if (ret) {
1617 reason = "notifier failure";
1618 goto failed_removal_isolated;
1619 }
1620
1621 do {
1622 for (pfn = start_pfn; pfn;) {
1623 if (signal_pending(current)) {
1624 ret = -EINTR;
1625 reason = "signal backoff";
1626 goto failed_removal_isolated;
1627 }
1628
1629 cond_resched();
1630 lru_add_drain_all();
1631
1632 pfn = scan_movable_pages(pfn, end_pfn);
1633 if (pfn) {
1634 /*
1635 * TODO: fatal migration failures should bail
1636 * out
1637 */
1638 do_migrate_range(pfn, end_pfn);
1639 }
1640 }
1641
1642 /*
1643 * Dissolve free hugepages in the memory block before doing
1644 * offlining actually in order to make hugetlbfs's object
1645 * counting consistent.
1646 */
1647 ret = dissolve_free_huge_pages(start_pfn, end_pfn);
1648 if (ret) {
1649 reason = "failure to dissolve huge pages";
1650 goto failed_removal_isolated;
1651 }
1652 /* check again */
1653 offlined_pages = check_pages_isolated(start_pfn, end_pfn);
1654 } while (offlined_pages < 0);
1655
1656 pr_info("Offlined Pages %ld\n", offlined_pages);
1657 /* Ok, all of our target is isolated.
1658 We cannot do rollback at this point. */
1659 offline_isolated_pages(start_pfn, end_pfn);
1660 /* reset pagetype flags and makes migrate type to be MOVABLE */
1661 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1662 /* removal success */
1663 adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages);
1664 zone->present_pages -= offlined_pages;
1665
1666 pgdat_resize_lock(zone->zone_pgdat, &flags);
1667 zone->zone_pgdat->node_present_pages -= offlined_pages;
1668 pgdat_resize_unlock(zone->zone_pgdat, &flags);
1669
1670 init_per_zone_wmark_min();
1671
1672 if (!populated_zone(zone)) {
1673 zone_pcp_reset(zone);
1674 build_all_zonelists(NULL);
1675 } else
1676 zone_pcp_update(zone);
1677
1678 node_states_clear_node(node, &arg);
1679 if (arg.status_change_nid >= 0) {
1680 kswapd_stop(node);
1681 kcompactd_stop(node);
1682 }
1683
1684 vm_total_pages = nr_free_pagecache_pages();
1685 writeback_set_ratelimit();
1686
1687 memory_notify(MEM_OFFLINE, &arg);
1688 mem_hotplug_done();
1689 return 0;
1690
1691failed_removal_isolated:
1692 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1693failed_removal:
1694 pr_debug("memory offlining [mem %#010llx-%#010llx] failed due to %s\n",
1695 (unsigned long long) start_pfn << PAGE_SHIFT,
1696 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1,
1697 reason);
1698 memory_notify(MEM_CANCEL_OFFLINE, &arg);
1699 /* pushback to free area */
1700 mem_hotplug_done();
1701 return ret;
1702}
1703
1704int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1705{
1706 return __offline_pages(start_pfn, start_pfn + nr_pages);
1707}
1708#endif /* CONFIG_MEMORY_HOTREMOVE */
1709
1710/**
1711 * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
1712 * @start_pfn: start pfn of the memory range
1713 * @end_pfn: end pfn of the memory range
1714 * @arg: argument passed to func
1715 * @func: callback for each memory section walked
1716 *
1717 * This function walks through all present mem sections in range
1718 * [start_pfn, end_pfn) and call func on each mem section.
1719 *
1720 * Returns the return value of func.
1721 */
1722int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn,
1723 void *arg, int (*func)(struct memory_block *, void *))
1724{
1725 struct memory_block *mem = NULL;
1726 struct mem_section *section;
1727 unsigned long pfn, section_nr;
1728 int ret;
1729
1730 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1731 section_nr = pfn_to_section_nr(pfn);
1732 if (!present_section_nr(section_nr))
1733 continue;
1734
1735 section = __nr_to_section(section_nr);
1736 /* same memblock? */
1737 if (mem)
1738 if ((section_nr >= mem->start_section_nr) &&
1739 (section_nr <= mem->end_section_nr))
1740 continue;
1741
1742 mem = find_memory_block_hinted(section, mem);
1743 if (!mem)
1744 continue;
1745
1746 ret = func(mem, arg);
1747 if (ret) {
1748 kobject_put(&mem->dev.kobj);
1749 return ret;
1750 }
1751 }
1752
1753 if (mem)
1754 kobject_put(&mem->dev.kobj);
1755
1756 return 0;
1757}
1758
1759#ifdef CONFIG_MEMORY_HOTREMOVE
1760static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
1761{
1762 int ret = !is_memblock_offlined(mem);
1763
1764 if (unlikely(ret)) {
1765 phys_addr_t beginpa, endpa;
1766
1767 beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
1768 endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1;
1769 pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n",
1770 &beginpa, &endpa);
1771 }
1772
1773 return ret;
1774}
1775
1776static int check_cpu_on_node(pg_data_t *pgdat)
1777{
1778 int cpu;
1779
1780 for_each_present_cpu(cpu) {
1781 if (cpu_to_node(cpu) == pgdat->node_id)
1782 /*
1783 * the cpu on this node isn't removed, and we can't
1784 * offline this node.
1785 */
1786 return -EBUSY;
1787 }
1788
1789 return 0;
1790}
1791
1792/**
1793 * try_offline_node
1794 * @nid: the node ID
1795 *
1796 * Offline a node if all memory sections and cpus of the node are removed.
1797 *
1798 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1799 * and online/offline operations before this call.
1800 */
1801void try_offline_node(int nid)
1802{
1803 pg_data_t *pgdat = NODE_DATA(nid);
1804 unsigned long start_pfn = pgdat->node_start_pfn;
1805 unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
1806 unsigned long pfn;
1807
1808 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1809 unsigned long section_nr = pfn_to_section_nr(pfn);
1810
1811 if (!present_section_nr(section_nr))
1812 continue;
1813
1814 if (pfn_to_nid(pfn) != nid)
1815 continue;
1816
1817 /*
1818 * some memory sections of this node are not removed, and we
1819 * can't offline node now.
1820 */
1821 return;
1822 }
1823
1824 if (check_cpu_on_node(pgdat))
1825 return;
1826
1827 /*
1828 * all memory/cpu of this node are removed, we can offline this
1829 * node now.
1830 */
1831 node_set_offline(nid);
1832 unregister_one_node(nid);
1833}
1834EXPORT_SYMBOL(try_offline_node);
1835
1836/**
1837 * remove_memory
1838 * @nid: the node ID
1839 * @start: physical address of the region to remove
1840 * @size: size of the region to remove
1841 *
1842 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1843 * and online/offline operations before this call, as required by
1844 * try_offline_node().
1845 */
1846void __ref __remove_memory(int nid, u64 start, u64 size)
1847{
1848 int ret;
1849
1850 BUG_ON(check_hotplug_memory_range(start, size));
1851
1852 mem_hotplug_begin();
1853
1854 /*
1855 * All memory blocks must be offlined before removing memory. Check
1856 * whether all memory blocks in question are offline and trigger a BUG()
1857 * if this is not the case.
1858 */
1859 ret = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL,
1860 check_memblock_offlined_cb);
1861 if (ret)
1862 BUG();
1863
1864 /* remove memmap entry */
1865 firmware_map_remove(start, start + size, "System RAM");
1866 memblock_free(start, size);
1867 memblock_remove(start, size);
1868
1869 arch_remove_memory(nid, start, size, NULL);
1870
1871 try_offline_node(nid);
1872
1873 mem_hotplug_done();
1874}
1875
1876void remove_memory(int nid, u64 start, u64 size)
1877{
1878 lock_device_hotplug();
1879 __remove_memory(nid, start, size);
1880 unlock_device_hotplug();
1881}
1882EXPORT_SYMBOL_GPL(remove_memory);
1883#endif /* CONFIG_MEMORY_HOTREMOVE */
1884