1 // SPDX-License-Identifier: GPL-2.0-only
3 * linux/mm/memory_hotplug.c
8 #include <linux/stddef.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>
40 #include <asm/tlbflush.h>
46 * online_page_callback contains pointer to current page onlining function.
47 * Initially it is generic_online_page(). If it is required it could be
48 * changed by calling set_online_page_callback() for callback registration
49 * and restore_online_page_callback() for generic callback restore.
52 static void generic_online_page(struct page *page, unsigned int order);
54 static online_page_callback_t online_page_callback = generic_online_page;
55 static DEFINE_MUTEX(online_page_callback_lock);
57 DEFINE_STATIC_PERCPU_RWSEM(mem_hotplug_lock);
59 void get_online_mems(void)
61 percpu_down_read(&mem_hotplug_lock);
64 void put_online_mems(void)
66 percpu_up_read(&mem_hotplug_lock);
69 bool movable_node_enabled = false;
71 #ifndef CONFIG_MEMORY_HOTPLUG_DEFAULT_ONLINE
72 bool memhp_auto_online;
74 bool memhp_auto_online = true;
76 EXPORT_SYMBOL_GPL(memhp_auto_online);
78 static int __init setup_memhp_default_state(char *str)
80 if (!strcmp(str, "online"))
81 memhp_auto_online = true;
82 else if (!strcmp(str, "offline"))
83 memhp_auto_online = false;
87 __setup("memhp_default_state=", setup_memhp_default_state);
89 void mem_hotplug_begin(void)
92 percpu_down_write(&mem_hotplug_lock);
95 void mem_hotplug_done(void)
97 percpu_up_write(&mem_hotplug_lock);
101 u64 max_mem_size = U64_MAX;
103 /* add this memory to iomem resource */
104 static struct resource *register_memory_resource(u64 start, u64 size)
106 struct resource *res;
107 unsigned long flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
108 char *resource_name = "System RAM";
110 if (start + size > max_mem_size)
111 return ERR_PTR(-E2BIG);
114 * Request ownership of the new memory range. This might be
115 * a child of an existing resource that was present but
116 * not marked as busy.
118 res = __request_region(&iomem_resource, start, size,
119 resource_name, flags);
122 pr_debug("Unable to reserve System RAM region: %016llx->%016llx\n",
123 start, start + size);
124 return ERR_PTR(-EEXIST);
129 static void release_memory_resource(struct resource *res)
133 release_resource(res);
138 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
139 void get_page_bootmem(unsigned long info, struct page *page,
142 page->freelist = (void *)type;
143 SetPagePrivate(page);
144 set_page_private(page, info);
148 void put_page_bootmem(struct page *page)
152 type = (unsigned long) page->freelist;
153 BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE ||
154 type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE);
156 if (page_ref_dec_return(page) == 1) {
157 page->freelist = NULL;
158 ClearPagePrivate(page);
159 set_page_private(page, 0);
160 INIT_LIST_HEAD(&page->lru);
161 free_reserved_page(page);
165 #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
166 #ifndef CONFIG_SPARSEMEM_VMEMMAP
167 static void register_page_bootmem_info_section(unsigned long start_pfn)
169 unsigned long mapsize, section_nr, i;
170 struct mem_section *ms;
171 struct page *page, *memmap;
172 struct mem_section_usage *usage;
174 section_nr = pfn_to_section_nr(start_pfn);
175 ms = __nr_to_section(section_nr);
177 /* Get section's memmap address */
178 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
181 * Get page for the memmap's phys address
182 * XXX: need more consideration for sparse_vmemmap...
184 page = virt_to_page(memmap);
185 mapsize = sizeof(struct page) * PAGES_PER_SECTION;
186 mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
188 /* remember memmap's page */
189 for (i = 0; i < mapsize; i++, page++)
190 get_page_bootmem(section_nr, page, SECTION_INFO);
193 page = virt_to_page(usage);
195 mapsize = PAGE_ALIGN(mem_section_usage_size()) >> PAGE_SHIFT;
197 for (i = 0; i < mapsize; i++, page++)
198 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
201 #else /* CONFIG_SPARSEMEM_VMEMMAP */
202 static void register_page_bootmem_info_section(unsigned long start_pfn)
204 unsigned long mapsize, section_nr, i;
205 struct mem_section *ms;
206 struct page *page, *memmap;
207 struct mem_section_usage *usage;
209 section_nr = pfn_to_section_nr(start_pfn);
210 ms = __nr_to_section(section_nr);
212 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
214 register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION);
217 page = virt_to_page(usage);
219 mapsize = PAGE_ALIGN(mem_section_usage_size()) >> PAGE_SHIFT;
221 for (i = 0; i < mapsize; i++, page++)
222 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
224 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
226 void __init register_page_bootmem_info_node(struct pglist_data *pgdat)
228 unsigned long i, pfn, end_pfn, nr_pages;
229 int node = pgdat->node_id;
232 nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
233 page = virt_to_page(pgdat);
235 for (i = 0; i < nr_pages; i++, page++)
236 get_page_bootmem(node, page, NODE_INFO);
238 pfn = pgdat->node_start_pfn;
239 end_pfn = pgdat_end_pfn(pgdat);
241 /* register section info */
242 for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
244 * Some platforms can assign the same pfn to multiple nodes - on
245 * node0 as well as nodeN. To avoid registering a pfn against
246 * multiple nodes we check that this pfn does not already
247 * reside in some other nodes.
249 if (pfn_valid(pfn) && (early_pfn_to_nid(pfn) == node))
250 register_page_bootmem_info_section(pfn);
253 #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
255 static int check_pfn_span(unsigned long pfn, unsigned long nr_pages,
259 * Disallow all operations smaller than a sub-section and only
260 * allow operations smaller than a section for
261 * SPARSEMEM_VMEMMAP. Note that check_hotplug_memory_range()
262 * enforces a larger memory_block_size_bytes() granularity for
263 * memory that will be marked online, so this check should only
264 * fire for direct arch_{add,remove}_memory() users outside of
265 * add_memory_resource().
267 unsigned long min_align;
269 if (IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP))
270 min_align = PAGES_PER_SUBSECTION;
272 min_align = PAGES_PER_SECTION;
273 if (!IS_ALIGNED(pfn, min_align)
274 || !IS_ALIGNED(nr_pages, min_align)) {
275 WARN(1, "Misaligned __%s_pages start: %#lx end: #%lx\n",
276 reason, pfn, pfn + nr_pages - 1);
283 * Reasonably generic function for adding memory. It is
284 * expected that archs that support memory hotplug will
285 * call this function after deciding the zone to which to
288 int __ref __add_pages(int nid, unsigned long pfn, unsigned long nr_pages,
289 struct mhp_restrictions *restrictions)
292 int start_sec, end_sec, err;
293 struct vmem_altmap *altmap = restrictions->altmap;
297 * Validate altmap is within bounds of the total request
299 if (altmap->base_pfn != pfn
300 || vmem_altmap_offset(altmap) > nr_pages) {
301 pr_warn_once("memory add fail, invalid altmap\n");
307 err = check_pfn_span(pfn, nr_pages, "add");
311 start_sec = pfn_to_section_nr(pfn);
312 end_sec = pfn_to_section_nr(pfn + nr_pages - 1);
313 for (i = start_sec; i <= end_sec; i++) {
316 pfns = min(nr_pages, PAGES_PER_SECTION
317 - (pfn & ~PAGE_SECTION_MASK));
318 err = sparse_add_section(nid, pfn, pfns, altmap);
325 vmemmap_populate_print_last();
329 /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
330 static unsigned long find_smallest_section_pfn(int nid, struct zone *zone,
331 unsigned long start_pfn,
332 unsigned long end_pfn)
334 for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SUBSECTION) {
335 if (unlikely(!pfn_valid(start_pfn)))
338 if (unlikely(pfn_to_nid(start_pfn) != nid))
341 if (zone && zone != page_zone(pfn_to_page(start_pfn)))
350 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
351 static unsigned long find_biggest_section_pfn(int nid, struct zone *zone,
352 unsigned long start_pfn,
353 unsigned long end_pfn)
357 /* pfn is the end pfn of a memory section. */
359 for (; pfn >= start_pfn; pfn -= PAGES_PER_SUBSECTION) {
360 if (unlikely(!pfn_valid(pfn)))
363 if (unlikely(pfn_to_nid(pfn) != nid))
366 if (zone && zone != page_zone(pfn_to_page(pfn)))
375 static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
376 unsigned long end_pfn)
378 unsigned long zone_start_pfn = zone->zone_start_pfn;
379 unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */
380 unsigned long zone_end_pfn = z;
382 int nid = zone_to_nid(zone);
384 zone_span_writelock(zone);
385 if (zone_start_pfn == start_pfn) {
387 * If the section is smallest section in the zone, it need
388 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
389 * In this case, we find second smallest valid mem_section
390 * for shrinking zone.
392 pfn = find_smallest_section_pfn(nid, zone, end_pfn,
395 zone->zone_start_pfn = pfn;
396 zone->spanned_pages = zone_end_pfn - pfn;
398 } else if (zone_end_pfn == end_pfn) {
400 * If the section is biggest section in the zone, it need
401 * shrink zone->spanned_pages.
402 * In this case, we find second biggest valid mem_section for
405 pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn,
408 zone->spanned_pages = pfn - zone_start_pfn + 1;
412 * The section is not biggest or smallest mem_section in the zone, it
413 * only creates a hole in the zone. So in this case, we need not
414 * change the zone. But perhaps, the zone has only hole data. Thus
415 * it check the zone has only hole or not.
417 pfn = zone_start_pfn;
418 for (; pfn < zone_end_pfn; pfn += PAGES_PER_SUBSECTION) {
419 if (unlikely(!pfn_valid(pfn)))
422 if (page_zone(pfn_to_page(pfn)) != zone)
425 /* Skip range to be removed */
426 if (pfn >= start_pfn && pfn < end_pfn)
429 /* If we find valid section, we have nothing to do */
430 zone_span_writeunlock(zone);
434 /* The zone has no valid section */
435 zone->zone_start_pfn = 0;
436 zone->spanned_pages = 0;
437 zone_span_writeunlock(zone);
440 static void shrink_pgdat_span(struct pglist_data *pgdat,
441 unsigned long start_pfn, unsigned long end_pfn)
443 unsigned long pgdat_start_pfn = pgdat->node_start_pfn;
444 unsigned long p = pgdat_end_pfn(pgdat); /* pgdat_end_pfn namespace clash */
445 unsigned long pgdat_end_pfn = p;
447 int nid = pgdat->node_id;
449 if (pgdat_start_pfn == start_pfn) {
451 * If the section is smallest section in the pgdat, it need
452 * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages.
453 * In this case, we find second smallest valid mem_section
454 * for shrinking zone.
456 pfn = find_smallest_section_pfn(nid, NULL, end_pfn,
459 pgdat->node_start_pfn = pfn;
460 pgdat->node_spanned_pages = pgdat_end_pfn - pfn;
462 } else if (pgdat_end_pfn == end_pfn) {
464 * If the section is biggest section in the pgdat, it need
465 * shrink pgdat->node_spanned_pages.
466 * In this case, we find second biggest valid mem_section for
469 pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn,
472 pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1;
476 * If the section is not biggest or smallest mem_section in the pgdat,
477 * it only creates a hole in the pgdat. So in this case, we need not
479 * But perhaps, the pgdat has only hole data. Thus it check the pgdat
480 * has only hole or not.
482 pfn = pgdat_start_pfn;
483 for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SUBSECTION) {
484 if (unlikely(!pfn_valid(pfn)))
487 if (pfn_to_nid(pfn) != nid)
490 /* Skip range to be removed */
491 if (pfn >= start_pfn && pfn < end_pfn)
494 /* If we find valid section, we have nothing to do */
498 /* The pgdat has no valid section */
499 pgdat->node_start_pfn = 0;
500 pgdat->node_spanned_pages = 0;
503 static void __remove_zone(struct zone *zone, unsigned long start_pfn,
504 unsigned long nr_pages)
506 struct pglist_data *pgdat = zone->zone_pgdat;
509 pgdat_resize_lock(zone->zone_pgdat, &flags);
510 shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
511 shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages);
512 pgdat_resize_unlock(zone->zone_pgdat, &flags);
515 static void __remove_section(struct zone *zone, unsigned long pfn,
516 unsigned long nr_pages, unsigned long map_offset,
517 struct vmem_altmap *altmap)
519 struct mem_section *ms = __nr_to_section(pfn_to_section_nr(pfn));
521 if (WARN_ON_ONCE(!valid_section(ms)))
524 __remove_zone(zone, pfn, nr_pages);
525 sparse_remove_section(ms, pfn, nr_pages, map_offset, altmap);
529 * __remove_pages() - remove sections of pages from a zone
530 * @zone: zone from which pages need to be removed
531 * @pfn: starting pageframe (must be aligned to start of a section)
532 * @nr_pages: number of pages to remove (must be multiple of section size)
533 * @altmap: alternative device page map or %NULL if default memmap is used
535 * Generic helper function to remove section mappings and sysfs entries
536 * for the section of the memory we are removing. Caller needs to make
537 * sure that pages are marked reserved and zones are adjust properly by
538 * calling offline_pages().
540 void __remove_pages(struct zone *zone, unsigned long pfn,
541 unsigned long nr_pages, struct vmem_altmap *altmap)
543 unsigned long map_offset = 0;
544 int i, start_sec, end_sec;
546 map_offset = vmem_altmap_offset(altmap);
548 clear_zone_contiguous(zone);
550 if (check_pfn_span(pfn, nr_pages, "remove"))
553 start_sec = pfn_to_section_nr(pfn);
554 end_sec = pfn_to_section_nr(pfn + nr_pages - 1);
555 for (i = start_sec; i <= end_sec; i++) {
559 pfns = min(nr_pages, PAGES_PER_SECTION
560 - (pfn & ~PAGE_SECTION_MASK));
561 __remove_section(zone, pfn, pfns, map_offset, altmap);
567 set_zone_contiguous(zone);
570 int set_online_page_callback(online_page_callback_t callback)
575 mutex_lock(&online_page_callback_lock);
577 if (online_page_callback == generic_online_page) {
578 online_page_callback = callback;
582 mutex_unlock(&online_page_callback_lock);
587 EXPORT_SYMBOL_GPL(set_online_page_callback);
589 int restore_online_page_callback(online_page_callback_t callback)
594 mutex_lock(&online_page_callback_lock);
596 if (online_page_callback == callback) {
597 online_page_callback = generic_online_page;
601 mutex_unlock(&online_page_callback_lock);
606 EXPORT_SYMBOL_GPL(restore_online_page_callback);
608 void __online_page_set_limits(struct page *page)
611 EXPORT_SYMBOL_GPL(__online_page_set_limits);
613 void __online_page_increment_counters(struct page *page)
615 adjust_managed_page_count(page, 1);
617 EXPORT_SYMBOL_GPL(__online_page_increment_counters);
619 void __online_page_free(struct page *page)
621 __free_reserved_page(page);
623 EXPORT_SYMBOL_GPL(__online_page_free);
625 static void generic_online_page(struct page *page, unsigned int order)
627 kernel_map_pages(page, 1 << order, 1);
628 __free_pages_core(page, order);
629 totalram_pages_add(1UL << order);
630 #ifdef CONFIG_HIGHMEM
631 if (PageHighMem(page))
632 totalhigh_pages_add(1UL << order);
636 static int online_pages_blocks(unsigned long start, unsigned long nr_pages)
638 unsigned long end = start + nr_pages;
639 int order, onlined_pages = 0;
641 while (start < end) {
642 order = min(MAX_ORDER - 1,
643 get_order(PFN_PHYS(end) - PFN_PHYS(start)));
644 (*online_page_callback)(pfn_to_page(start), order);
646 onlined_pages += (1UL << order);
647 start += (1UL << order);
649 return onlined_pages;
652 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
655 unsigned long onlined_pages = *(unsigned long *)arg;
657 if (PageReserved(pfn_to_page(start_pfn)))
658 onlined_pages += online_pages_blocks(start_pfn, nr_pages);
660 online_mem_sections(start_pfn, start_pfn + nr_pages);
662 *(unsigned long *)arg = onlined_pages;
666 /* check which state of node_states will be changed when online memory */
667 static void node_states_check_changes_online(unsigned long nr_pages,
668 struct zone *zone, struct memory_notify *arg)
670 int nid = zone_to_nid(zone);
672 arg->status_change_nid = NUMA_NO_NODE;
673 arg->status_change_nid_normal = NUMA_NO_NODE;
674 arg->status_change_nid_high = NUMA_NO_NODE;
676 if (!node_state(nid, N_MEMORY))
677 arg->status_change_nid = nid;
678 if (zone_idx(zone) <= ZONE_NORMAL && !node_state(nid, N_NORMAL_MEMORY))
679 arg->status_change_nid_normal = nid;
680 #ifdef CONFIG_HIGHMEM
681 if (zone_idx(zone) <= ZONE_HIGHMEM && !node_state(nid, N_HIGH_MEMORY))
682 arg->status_change_nid_high = nid;
686 static void node_states_set_node(int node, struct memory_notify *arg)
688 if (arg->status_change_nid_normal >= 0)
689 node_set_state(node, N_NORMAL_MEMORY);
691 if (arg->status_change_nid_high >= 0)
692 node_set_state(node, N_HIGH_MEMORY);
694 if (arg->status_change_nid >= 0)
695 node_set_state(node, N_MEMORY);
698 static void __meminit resize_zone_range(struct zone *zone, unsigned long start_pfn,
699 unsigned long nr_pages)
701 unsigned long old_end_pfn = zone_end_pfn(zone);
703 if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
704 zone->zone_start_pfn = start_pfn;
706 zone->spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - zone->zone_start_pfn;
709 static void __meminit resize_pgdat_range(struct pglist_data *pgdat, unsigned long start_pfn,
710 unsigned long nr_pages)
712 unsigned long old_end_pfn = pgdat_end_pfn(pgdat);
714 if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
715 pgdat->node_start_pfn = start_pfn;
717 pgdat->node_spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - pgdat->node_start_pfn;
720 void __ref move_pfn_range_to_zone(struct zone *zone, unsigned long start_pfn,
721 unsigned long nr_pages, struct vmem_altmap *altmap)
723 struct pglist_data *pgdat = zone->zone_pgdat;
724 int nid = pgdat->node_id;
727 clear_zone_contiguous(zone);
729 /* TODO Huh pgdat is irqsave while zone is not. It used to be like that before */
730 pgdat_resize_lock(pgdat, &flags);
731 zone_span_writelock(zone);
732 if (zone_is_empty(zone))
733 init_currently_empty_zone(zone, start_pfn, nr_pages);
734 resize_zone_range(zone, start_pfn, nr_pages);
735 zone_span_writeunlock(zone);
736 resize_pgdat_range(pgdat, start_pfn, nr_pages);
737 pgdat_resize_unlock(pgdat, &flags);
740 * TODO now we have a visible range of pages which are not associated
741 * with their zone properly. Not nice but set_pfnblock_flags_mask
742 * expects the zone spans the pfn range. All the pages in the range
743 * are reserved so nobody should be touching them so we should be safe
745 memmap_init_zone(nr_pages, nid, zone_idx(zone), start_pfn,
746 MEMMAP_HOTPLUG, altmap);
748 set_zone_contiguous(zone);
752 * Returns a default kernel memory zone for the given pfn range.
753 * If no kernel zone covers this pfn range it will automatically go
754 * to the ZONE_NORMAL.
756 static struct zone *default_kernel_zone_for_pfn(int nid, unsigned long start_pfn,
757 unsigned long nr_pages)
759 struct pglist_data *pgdat = NODE_DATA(nid);
762 for (zid = 0; zid <= ZONE_NORMAL; zid++) {
763 struct zone *zone = &pgdat->node_zones[zid];
765 if (zone_intersects(zone, start_pfn, nr_pages))
769 return &pgdat->node_zones[ZONE_NORMAL];
772 static inline struct zone *default_zone_for_pfn(int nid, unsigned long start_pfn,
773 unsigned long nr_pages)
775 struct zone *kernel_zone = default_kernel_zone_for_pfn(nid, start_pfn,
777 struct zone *movable_zone = &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
778 bool in_kernel = zone_intersects(kernel_zone, start_pfn, nr_pages);
779 bool in_movable = zone_intersects(movable_zone, start_pfn, nr_pages);
782 * We inherit the existing zone in a simple case where zones do not
783 * overlap in the given range
785 if (in_kernel ^ in_movable)
786 return (in_kernel) ? kernel_zone : movable_zone;
789 * If the range doesn't belong to any zone or two zones overlap in the
790 * given range then we use movable zone only if movable_node is
791 * enabled because we always online to a kernel zone by default.
793 return movable_node_enabled ? movable_zone : kernel_zone;
796 struct zone * zone_for_pfn_range(int online_type, int nid, unsigned start_pfn,
797 unsigned long nr_pages)
799 if (online_type == MMOP_ONLINE_KERNEL)
800 return default_kernel_zone_for_pfn(nid, start_pfn, nr_pages);
802 if (online_type == MMOP_ONLINE_MOVABLE)
803 return &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
805 return default_zone_for_pfn(nid, start_pfn, nr_pages);
809 * Associates the given pfn range with the given node and the zone appropriate
810 * for the given online type.
812 static struct zone * __meminit move_pfn_range(int online_type, int nid,
813 unsigned long start_pfn, unsigned long nr_pages)
817 zone = zone_for_pfn_range(online_type, nid, start_pfn, nr_pages);
818 move_pfn_range_to_zone(zone, start_pfn, nr_pages, NULL);
822 int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type)
825 unsigned long onlined_pages = 0;
827 int need_zonelists_rebuild = 0;
830 struct memory_notify arg;
831 struct memory_block *mem;
836 * We can't use pfn_to_nid() because nid might be stored in struct page
837 * which is not yet initialized. Instead, we find nid from memory block.
839 mem = find_memory_block(__pfn_to_section(pfn));
841 put_device(&mem->dev);
843 /* associate pfn range with the zone */
844 zone = move_pfn_range(online_type, nid, pfn, nr_pages);
847 arg.nr_pages = nr_pages;
848 node_states_check_changes_online(nr_pages, zone, &arg);
850 ret = memory_notify(MEM_GOING_ONLINE, &arg);
851 ret = notifier_to_errno(ret);
853 goto failed_addition;
856 * If this zone is not populated, then it is not in zonelist.
857 * This means the page allocator ignores this zone.
858 * So, zonelist must be updated after online.
860 if (!populated_zone(zone)) {
861 need_zonelists_rebuild = 1;
862 setup_zone_pageset(zone);
865 ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
868 if (need_zonelists_rebuild)
869 zone_pcp_reset(zone);
870 goto failed_addition;
873 zone->present_pages += onlined_pages;
875 pgdat_resize_lock(zone->zone_pgdat, &flags);
876 zone->zone_pgdat->node_present_pages += onlined_pages;
877 pgdat_resize_unlock(zone->zone_pgdat, &flags);
882 node_states_set_node(nid, &arg);
883 if (need_zonelists_rebuild)
884 build_all_zonelists(NULL);
886 zone_pcp_update(zone);
889 init_per_zone_wmark_min();
896 vm_total_pages = nr_free_pagecache_pages();
898 writeback_set_ratelimit();
901 memory_notify(MEM_ONLINE, &arg);
906 pr_debug("online_pages [mem %#010llx-%#010llx] failed\n",
907 (unsigned long long) pfn << PAGE_SHIFT,
908 (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1);
909 memory_notify(MEM_CANCEL_ONLINE, &arg);
913 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
915 static void reset_node_present_pages(pg_data_t *pgdat)
919 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
920 z->present_pages = 0;
922 pgdat->node_present_pages = 0;
925 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
926 static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
928 struct pglist_data *pgdat;
929 unsigned long start_pfn = PFN_DOWN(start);
931 pgdat = NODE_DATA(nid);
933 pgdat = arch_alloc_nodedata(nid);
937 arch_refresh_nodedata(nid, pgdat);
940 * Reset the nr_zones, order and classzone_idx before reuse.
941 * Note that kswapd will init kswapd_classzone_idx properly
942 * when it starts in the near future.
945 pgdat->kswapd_order = 0;
946 pgdat->kswapd_classzone_idx = 0;
949 /* we can use NODE_DATA(nid) from here */
951 pgdat->node_id = nid;
952 pgdat->node_start_pfn = start_pfn;
954 /* init node's zones as empty zones, we don't have any present pages.*/
955 free_area_init_core_hotplug(nid);
956 pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat);
959 * The node we allocated has no zone fallback lists. For avoiding
960 * to access not-initialized zonelist, build here.
962 build_all_zonelists(pgdat);
965 * When memory is hot-added, all the memory is in offline state. So
966 * clear all zones' present_pages because they will be updated in
967 * online_pages() and offline_pages().
969 reset_node_managed_pages(pgdat);
970 reset_node_present_pages(pgdat);
975 static void rollback_node_hotadd(int nid)
977 pg_data_t *pgdat = NODE_DATA(nid);
979 arch_refresh_nodedata(nid, NULL);
980 free_percpu(pgdat->per_cpu_nodestats);
981 arch_free_nodedata(pgdat);
987 * try_online_node - online a node if offlined
989 * @start: start addr of the node
990 * @set_node_online: Whether we want to online the node
991 * called by cpu_up() to online a node without onlined memory.
994 * 1 -> a new node has been allocated
995 * 0 -> the node is already online
996 * -ENOMEM -> the node could not be allocated
998 static int __try_online_node(int nid, u64 start, bool set_node_online)
1003 if (node_online(nid))
1006 pgdat = hotadd_new_pgdat(nid, start);
1008 pr_err("Cannot online node %d due to NULL pgdat\n", nid);
1013 if (set_node_online) {
1014 node_set_online(nid);
1015 ret = register_one_node(nid);
1023 * Users of this function always want to online/register the node
1025 int try_online_node(int nid)
1029 mem_hotplug_begin();
1030 ret = __try_online_node(nid, 0, true);
1035 static int check_hotplug_memory_range(u64 start, u64 size)
1037 /* memory range must be block size aligned */
1038 if (!size || !IS_ALIGNED(start, memory_block_size_bytes()) ||
1039 !IS_ALIGNED(size, memory_block_size_bytes())) {
1040 pr_err("Block size [%#lx] unaligned hotplug range: start %#llx, size %#llx",
1041 memory_block_size_bytes(), start, size);
1048 static int online_memory_block(struct memory_block *mem, void *arg)
1050 return device_online(&mem->dev);
1054 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1055 * and online/offline operations (triggered e.g. by sysfs).
1057 * we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG
1059 int __ref add_memory_resource(int nid, struct resource *res)
1061 struct mhp_restrictions restrictions = {};
1063 bool new_node = false;
1067 size = resource_size(res);
1069 ret = check_hotplug_memory_range(start, size);
1073 mem_hotplug_begin();
1076 * Add new range to memblock so that when hotadd_new_pgdat() is called
1077 * to allocate new pgdat, get_pfn_range_for_nid() will be able to find
1078 * this new range and calculate total pages correctly. The range will
1079 * be removed at hot-remove time.
1081 memblock_add_node(start, size, nid);
1083 ret = __try_online_node(nid, start, false);
1088 /* call arch's memory hotadd */
1089 ret = arch_add_memory(nid, start, size, &restrictions);
1093 /* create memory block devices after memory was added */
1094 ret = create_memory_block_devices(start, size);
1096 arch_remove_memory(nid, start, size, NULL);
1101 /* If sysfs file of new node can't be created, cpu on the node
1102 * can't be hot-added. There is no rollback way now.
1103 * So, check by BUG_ON() to catch it reluctantly..
1104 * We online node here. We can't roll back from here.
1106 node_set_online(nid);
1107 ret = __register_one_node(nid);
1111 /* link memory sections under this node.*/
1112 ret = link_mem_sections(nid, PFN_DOWN(start), PFN_UP(start + size - 1));
1115 /* create new memmap entry */
1116 firmware_map_add_hotplug(start, start + size, "System RAM");
1118 /* device_online() will take the lock when calling online_pages() */
1121 /* online pages if requested */
1122 if (memhp_auto_online)
1123 walk_memory_blocks(start, size, NULL, online_memory_block);
1127 /* rollback pgdat allocation and others */
1129 rollback_node_hotadd(nid);
1130 memblock_remove(start, size);
1135 /* requires device_hotplug_lock, see add_memory_resource() */
1136 int __ref __add_memory(int nid, u64 start, u64 size)
1138 struct resource *res;
1141 res = register_memory_resource(start, size);
1143 return PTR_ERR(res);
1145 ret = add_memory_resource(nid, res);
1147 release_memory_resource(res);
1151 int add_memory(int nid, u64 start, u64 size)
1155 lock_device_hotplug();
1156 rc = __add_memory(nid, start, size);
1157 unlock_device_hotplug();
1161 EXPORT_SYMBOL_GPL(add_memory);
1163 #ifdef CONFIG_MEMORY_HOTREMOVE
1165 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1166 * set and the size of the free page is given by page_order(). Using this,
1167 * the function determines if the pageblock contains only free pages.
1168 * Due to buddy contraints, a free page at least the size of a pageblock will
1169 * be located at the start of the pageblock
1171 static inline int pageblock_free(struct page *page)
1173 return PageBuddy(page) && page_order(page) >= pageblock_order;
1176 /* Return the pfn of the start of the next active pageblock after a given pfn */
1177 static unsigned long next_active_pageblock(unsigned long pfn)
1179 struct page *page = pfn_to_page(pfn);
1181 /* Ensure the starting page is pageblock-aligned */
1182 BUG_ON(pfn & (pageblock_nr_pages - 1));
1184 /* If the entire pageblock is free, move to the end of free page */
1185 if (pageblock_free(page)) {
1187 /* be careful. we don't have locks, page_order can be changed.*/
1188 order = page_order(page);
1189 if ((order < MAX_ORDER) && (order >= pageblock_order))
1190 return pfn + (1 << order);
1193 return pfn + pageblock_nr_pages;
1196 static bool is_pageblock_removable_nolock(unsigned long pfn)
1198 struct page *page = pfn_to_page(pfn);
1202 * We have to be careful here because we are iterating over memory
1203 * sections which are not zone aware so we might end up outside of
1204 * the zone but still within the section.
1205 * We have to take care about the node as well. If the node is offline
1206 * its NODE_DATA will be NULL - see page_zone.
1208 if (!node_online(page_to_nid(page)))
1211 zone = page_zone(page);
1212 pfn = page_to_pfn(page);
1213 if (!zone_spans_pfn(zone, pfn))
1216 return !has_unmovable_pages(zone, page, 0, MIGRATE_MOVABLE, SKIP_HWPOISON);
1219 /* Checks if this range of memory is likely to be hot-removable. */
1220 bool is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
1222 unsigned long end_pfn, pfn;
1224 end_pfn = min(start_pfn + nr_pages,
1225 zone_end_pfn(page_zone(pfn_to_page(start_pfn))));
1227 /* Check the starting page of each pageblock within the range */
1228 for (pfn = start_pfn; pfn < end_pfn; pfn = next_active_pageblock(pfn)) {
1229 if (!is_pageblock_removable_nolock(pfn))
1234 /* All pageblocks in the memory block are likely to be hot-removable */
1239 * Confirm all pages in a range [start, end) belong to the same zone.
1240 * When true, return its valid [start, end).
1242 int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn,
1243 unsigned long *valid_start, unsigned long *valid_end)
1245 unsigned long pfn, sec_end_pfn;
1246 unsigned long start, end;
1247 struct zone *zone = NULL;
1250 for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1);
1252 pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) {
1253 /* Make sure the memory section is present first */
1254 if (!present_section_nr(pfn_to_section_nr(pfn)))
1256 for (; pfn < sec_end_pfn && pfn < end_pfn;
1257 pfn += MAX_ORDER_NR_PAGES) {
1259 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1260 while ((i < MAX_ORDER_NR_PAGES) &&
1261 !pfn_valid_within(pfn + i))
1263 if (i == MAX_ORDER_NR_PAGES || pfn + i >= end_pfn)
1265 /* Check if we got outside of the zone */
1266 if (zone && !zone_spans_pfn(zone, pfn + i))
1268 page = pfn_to_page(pfn + i);
1269 if (zone && page_zone(page) != zone)
1273 zone = page_zone(page);
1274 end = pfn + MAX_ORDER_NR_PAGES;
1279 *valid_start = start;
1280 *valid_end = min(end, end_pfn);
1288 * Scan pfn range [start,end) to find movable/migratable pages (LRU pages,
1289 * non-lru movable pages and hugepages). We scan pfn because it's much
1290 * easier than scanning over linked list. This function returns the pfn
1291 * of the first found movable page if it's found, otherwise 0.
1293 static unsigned long scan_movable_pages(unsigned long start, unsigned long end)
1297 for (pfn = start; pfn < end; pfn++) {
1298 struct page *page, *head;
1301 if (!pfn_valid(pfn))
1303 page = pfn_to_page(pfn);
1306 if (__PageMovable(page))
1309 if (!PageHuge(page))
1311 head = compound_head(page);
1312 if (page_huge_active(head))
1314 skip = (1 << compound_order(head)) - (page - head);
1320 static struct page *new_node_page(struct page *page, unsigned long private)
1322 int nid = page_to_nid(page);
1323 nodemask_t nmask = node_states[N_MEMORY];
1326 * try to allocate from a different node but reuse this node if there
1327 * are no other online nodes to be used (e.g. we are offlining a part
1328 * of the only existing node)
1330 node_clear(nid, nmask);
1331 if (nodes_empty(nmask))
1332 node_set(nid, nmask);
1334 return new_page_nodemask(page, nid, &nmask);
1338 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1345 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1346 if (!pfn_valid(pfn))
1348 page = pfn_to_page(pfn);
1350 if (PageHuge(page)) {
1351 struct page *head = compound_head(page);
1352 pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1;
1353 isolate_huge_page(head, &source);
1355 } else if (PageTransHuge(page))
1356 pfn = page_to_pfn(compound_head(page))
1357 + hpage_nr_pages(page) - 1;
1360 * HWPoison pages have elevated reference counts so the migration would
1361 * fail on them. It also doesn't make any sense to migrate them in the
1362 * first place. Still try to unmap such a page in case it is still mapped
1363 * (e.g. current hwpoison implementation doesn't unmap KSM pages but keep
1364 * the unmap as the catch all safety net).
1366 if (PageHWPoison(page)) {
1367 if (WARN_ON(PageLRU(page)))
1368 isolate_lru_page(page);
1369 if (page_mapped(page))
1370 try_to_unmap(page, TTU_IGNORE_MLOCK | TTU_IGNORE_ACCESS);
1374 if (!get_page_unless_zero(page))
1377 * We can skip free pages. And we can deal with pages on
1378 * LRU and non-lru movable pages.
1381 ret = isolate_lru_page(page);
1383 ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE);
1384 if (!ret) { /* Success */
1385 list_add_tail(&page->lru, &source);
1386 if (!__PageMovable(page))
1387 inc_node_page_state(page, NR_ISOLATED_ANON +
1388 page_is_file_cache(page));
1391 pr_warn("failed to isolate pfn %lx\n", pfn);
1392 dump_page(page, "isolation failed");
1396 if (!list_empty(&source)) {
1397 /* Allocate a new page from the nearest neighbor node */
1398 ret = migrate_pages(&source, new_node_page, NULL, 0,
1399 MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
1401 list_for_each_entry(page, &source, lru) {
1402 pr_warn("migrating pfn %lx failed ret:%d ",
1403 page_to_pfn(page), ret);
1404 dump_page(page, "migration failure");
1406 putback_movable_pages(&source);
1414 * remove from free_area[] and mark all as Reserved.
1417 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1420 unsigned long *offlined_pages = (unsigned long *)data;
1422 *offlined_pages += __offline_isolated_pages(start, start + nr_pages);
1427 * Check all pages in range, recoreded as memory resource, are isolated.
1430 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1433 return test_pages_isolated(start_pfn, start_pfn + nr_pages, true);
1436 static int __init cmdline_parse_movable_node(char *p)
1438 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1439 movable_node_enabled = true;
1441 pr_warn("movable_node parameter depends on CONFIG_HAVE_MEMBLOCK_NODE_MAP to work properly\n");
1445 early_param("movable_node", cmdline_parse_movable_node);
1447 /* check which state of node_states will be changed when offline memory */
1448 static void node_states_check_changes_offline(unsigned long nr_pages,
1449 struct zone *zone, struct memory_notify *arg)
1451 struct pglist_data *pgdat = zone->zone_pgdat;
1452 unsigned long present_pages = 0;
1455 arg->status_change_nid = NUMA_NO_NODE;
1456 arg->status_change_nid_normal = NUMA_NO_NODE;
1457 arg->status_change_nid_high = NUMA_NO_NODE;
1460 * Check whether node_states[N_NORMAL_MEMORY] will be changed.
1461 * If the memory to be offline is within the range
1462 * [0..ZONE_NORMAL], and it is the last present memory there,
1463 * the zones in that range will become empty after the offlining,
1464 * thus we can determine that we need to clear the node from
1465 * node_states[N_NORMAL_MEMORY].
1467 for (zt = 0; zt <= ZONE_NORMAL; zt++)
1468 present_pages += pgdat->node_zones[zt].present_pages;
1469 if (zone_idx(zone) <= ZONE_NORMAL && nr_pages >= present_pages)
1470 arg->status_change_nid_normal = zone_to_nid(zone);
1472 #ifdef CONFIG_HIGHMEM
1474 * node_states[N_HIGH_MEMORY] contains nodes which
1475 * have normal memory or high memory.
1476 * Here we add the present_pages belonging to ZONE_HIGHMEM.
1477 * If the zone is within the range of [0..ZONE_HIGHMEM), and
1478 * we determine that the zones in that range become empty,
1479 * we need to clear the node for N_HIGH_MEMORY.
1481 present_pages += pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1482 if (zone_idx(zone) <= ZONE_HIGHMEM && nr_pages >= present_pages)
1483 arg->status_change_nid_high = zone_to_nid(zone);
1487 * We have accounted the pages from [0..ZONE_NORMAL), and
1488 * in case of CONFIG_HIGHMEM the pages from ZONE_HIGHMEM
1490 * Here we count the possible pages from ZONE_MOVABLE.
1491 * If after having accounted all the pages, we see that the nr_pages
1492 * to be offlined is over or equal to the accounted pages,
1493 * we know that the node will become empty, and so, we can clear
1494 * it for N_MEMORY as well.
1496 present_pages += pgdat->node_zones[ZONE_MOVABLE].present_pages;
1498 if (nr_pages >= present_pages)
1499 arg->status_change_nid = zone_to_nid(zone);
1502 static void node_states_clear_node(int node, struct memory_notify *arg)
1504 if (arg->status_change_nid_normal >= 0)
1505 node_clear_state(node, N_NORMAL_MEMORY);
1507 if (arg->status_change_nid_high >= 0)
1508 node_clear_state(node, N_HIGH_MEMORY);
1510 if (arg->status_change_nid >= 0)
1511 node_clear_state(node, N_MEMORY);
1514 static int __ref __offline_pages(unsigned long start_pfn,
1515 unsigned long end_pfn)
1517 unsigned long pfn, nr_pages;
1518 unsigned long offlined_pages = 0;
1519 int ret, node, nr_isolate_pageblock;
1520 unsigned long flags;
1521 unsigned long valid_start, valid_end;
1523 struct memory_notify arg;
1526 mem_hotplug_begin();
1528 /* This makes hotplug much easier...and readable.
1529 we assume this for now. .*/
1530 if (!test_pages_in_a_zone(start_pfn, end_pfn, &valid_start,
1533 reason = "multizone range";
1534 goto failed_removal;
1537 zone = page_zone(pfn_to_page(valid_start));
1538 node = zone_to_nid(zone);
1539 nr_pages = end_pfn - start_pfn;
1541 /* set above range as isolated */
1542 ret = start_isolate_page_range(start_pfn, end_pfn,
1544 SKIP_HWPOISON | REPORT_FAILURE);
1546 reason = "failure to isolate range";
1547 goto failed_removal;
1549 nr_isolate_pageblock = ret;
1551 arg.start_pfn = start_pfn;
1552 arg.nr_pages = nr_pages;
1553 node_states_check_changes_offline(nr_pages, zone, &arg);
1555 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1556 ret = notifier_to_errno(ret);
1558 reason = "notifier failure";
1559 goto failed_removal_isolated;
1563 for (pfn = start_pfn; pfn;) {
1564 if (signal_pending(current)) {
1566 reason = "signal backoff";
1567 goto failed_removal_isolated;
1571 lru_add_drain_all();
1573 pfn = scan_movable_pages(pfn, end_pfn);
1576 * TODO: fatal migration failures should bail
1579 do_migrate_range(pfn, end_pfn);
1584 * Dissolve free hugepages in the memory block before doing
1585 * offlining actually in order to make hugetlbfs's object
1586 * counting consistent.
1588 ret = dissolve_free_huge_pages(start_pfn, end_pfn);
1590 reason = "failure to dissolve huge pages";
1591 goto failed_removal_isolated;
1594 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn,
1595 NULL, check_pages_isolated_cb);
1598 /* Ok, all of our target is isolated.
1599 We cannot do rollback at this point. */
1600 walk_system_ram_range(start_pfn, end_pfn - start_pfn,
1601 &offlined_pages, offline_isolated_pages_cb);
1602 pr_info("Offlined Pages %ld\n", offlined_pages);
1604 * Onlining will reset pagetype flags and makes migrate type
1605 * MOVABLE, so just need to decrease the number of isolated
1606 * pageblocks zone counter here.
1608 spin_lock_irqsave(&zone->lock, flags);
1609 zone->nr_isolate_pageblock -= nr_isolate_pageblock;
1610 spin_unlock_irqrestore(&zone->lock, flags);
1612 /* removal success */
1613 adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages);
1614 zone->present_pages -= offlined_pages;
1616 pgdat_resize_lock(zone->zone_pgdat, &flags);
1617 zone->zone_pgdat->node_present_pages -= offlined_pages;
1618 pgdat_resize_unlock(zone->zone_pgdat, &flags);
1620 init_per_zone_wmark_min();
1622 if (!populated_zone(zone)) {
1623 zone_pcp_reset(zone);
1624 build_all_zonelists(NULL);
1626 zone_pcp_update(zone);
1628 node_states_clear_node(node, &arg);
1629 if (arg.status_change_nid >= 0) {
1631 kcompactd_stop(node);
1634 vm_total_pages = nr_free_pagecache_pages();
1635 writeback_set_ratelimit();
1637 memory_notify(MEM_OFFLINE, &arg);
1641 failed_removal_isolated:
1642 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1643 memory_notify(MEM_CANCEL_OFFLINE, &arg);
1645 pr_debug("memory offlining [mem %#010llx-%#010llx] failed due to %s\n",
1646 (unsigned long long) start_pfn << PAGE_SHIFT,
1647 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1,
1649 /* pushback to free area */
1654 int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1656 return __offline_pages(start_pfn, start_pfn + nr_pages);
1659 static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
1661 int ret = !is_memblock_offlined(mem);
1663 if (unlikely(ret)) {
1664 phys_addr_t beginpa, endpa;
1666 beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
1667 endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1;
1668 pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n",
1676 static int check_cpu_on_node(pg_data_t *pgdat)
1680 for_each_present_cpu(cpu) {
1681 if (cpu_to_node(cpu) == pgdat->node_id)
1683 * the cpu on this node isn't removed, and we can't
1684 * offline this node.
1696 * Offline a node if all memory sections and cpus of the node are removed.
1698 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1699 * and online/offline operations before this call.
1701 void try_offline_node(int nid)
1703 pg_data_t *pgdat = NODE_DATA(nid);
1704 unsigned long start_pfn = pgdat->node_start_pfn;
1705 unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
1708 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1709 unsigned long section_nr = pfn_to_section_nr(pfn);
1711 if (!present_section_nr(section_nr))
1714 if (pfn_to_nid(pfn) != nid)
1718 * some memory sections of this node are not removed, and we
1719 * can't offline node now.
1724 if (check_cpu_on_node(pgdat))
1728 * all memory/cpu of this node are removed, we can offline this
1731 node_set_offline(nid);
1732 unregister_one_node(nid);
1734 EXPORT_SYMBOL(try_offline_node);
1736 static void __release_memory_resource(resource_size_t start,
1737 resource_size_t size)
1742 * When removing memory in the same granularity as it was added,
1743 * this function never fails. It might only fail if resources
1744 * have to be adjusted or split. We'll ignore the error, as
1745 * removing of memory cannot fail.
1747 ret = release_mem_region_adjustable(&iomem_resource, start, size);
1749 resource_size_t endres = start + size - 1;
1751 pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
1752 &start, &endres, ret);
1756 static int __ref try_remove_memory(int nid, u64 start, u64 size)
1760 BUG_ON(check_hotplug_memory_range(start, size));
1762 mem_hotplug_begin();
1765 * All memory blocks must be offlined before removing memory. Check
1766 * whether all memory blocks in question are offline and return error
1767 * if this is not the case.
1769 rc = walk_memory_blocks(start, size, NULL, check_memblock_offlined_cb);
1773 /* remove memmap entry */
1774 firmware_map_remove(start, start + size, "System RAM");
1775 memblock_free(start, size);
1776 memblock_remove(start, size);
1778 /* remove memory block devices before removing memory */
1779 remove_memory_block_devices(start, size);
1781 arch_remove_memory(nid, start, size, NULL);
1782 __release_memory_resource(start, size);
1784 try_offline_node(nid);
1794 * @start: physical address of the region to remove
1795 * @size: size of the region to remove
1797 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1798 * and online/offline operations before this call, as required by
1799 * try_offline_node().
1801 void __remove_memory(int nid, u64 start, u64 size)
1805 * trigger BUG() is some memory is not offlined prior to calling this
1808 if (try_remove_memory(nid, start, size))
1813 * Remove memory if every memory block is offline, otherwise return -EBUSY is
1814 * some memory is not offline
1816 int remove_memory(int nid, u64 start, u64 size)
1820 lock_device_hotplug();
1821 rc = try_remove_memory(nid, start, size);
1822 unlock_device_hotplug();
1826 EXPORT_SYMBOL_GPL(remove_memory);
1827 #endif /* CONFIG_MEMORY_HOTREMOVE */