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);
137 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
138 void get_page_bootmem(unsigned long info, struct page *page,
141 page->freelist = (void *)type;
142 SetPagePrivate(page);
143 set_page_private(page, info);
147 void put_page_bootmem(struct page *page)
151 type = (unsigned long) page->freelist;
152 BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE ||
153 type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE);
155 if (page_ref_dec_return(page) == 1) {
156 page->freelist = NULL;
157 ClearPagePrivate(page);
158 set_page_private(page, 0);
159 INIT_LIST_HEAD(&page->lru);
160 free_reserved_page(page);
164 #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
165 #ifndef CONFIG_SPARSEMEM_VMEMMAP
166 static void register_page_bootmem_info_section(unsigned long start_pfn)
168 unsigned long mapsize, section_nr, i;
169 struct mem_section *ms;
170 struct page *page, *memmap;
171 struct mem_section_usage *usage;
173 section_nr = pfn_to_section_nr(start_pfn);
174 ms = __nr_to_section(section_nr);
176 /* Get section's memmap address */
177 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
180 * Get page for the memmap's phys address
181 * XXX: need more consideration for sparse_vmemmap...
183 page = virt_to_page(memmap);
184 mapsize = sizeof(struct page) * PAGES_PER_SECTION;
185 mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
187 /* remember memmap's page */
188 for (i = 0; i < mapsize; i++, page++)
189 get_page_bootmem(section_nr, page, SECTION_INFO);
192 page = virt_to_page(usage);
194 mapsize = PAGE_ALIGN(mem_section_usage_size()) >> PAGE_SHIFT;
196 for (i = 0; i < mapsize; i++, page++)
197 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
200 #else /* CONFIG_SPARSEMEM_VMEMMAP */
201 static void register_page_bootmem_info_section(unsigned long start_pfn)
203 unsigned long mapsize, section_nr, i;
204 struct mem_section *ms;
205 struct page *page, *memmap;
206 struct mem_section_usage *usage;
208 section_nr = pfn_to_section_nr(start_pfn);
209 ms = __nr_to_section(section_nr);
211 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
213 register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION);
216 page = virt_to_page(usage);
218 mapsize = PAGE_ALIGN(mem_section_usage_size()) >> PAGE_SHIFT;
220 for (i = 0; i < mapsize; i++, page++)
221 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
223 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
225 void __init register_page_bootmem_info_node(struct pglist_data *pgdat)
227 unsigned long i, pfn, end_pfn, nr_pages;
228 int node = pgdat->node_id;
231 nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
232 page = virt_to_page(pgdat);
234 for (i = 0; i < nr_pages; i++, page++)
235 get_page_bootmem(node, page, NODE_INFO);
237 pfn = pgdat->node_start_pfn;
238 end_pfn = pgdat_end_pfn(pgdat);
240 /* register section info */
241 for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
243 * Some platforms can assign the same pfn to multiple nodes - on
244 * node0 as well as nodeN. To avoid registering a pfn against
245 * multiple nodes we check that this pfn does not already
246 * reside in some other nodes.
248 if (pfn_valid(pfn) && (early_pfn_to_nid(pfn) == node))
249 register_page_bootmem_info_section(pfn);
252 #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
254 static int check_pfn_span(unsigned long pfn, unsigned long nr_pages,
258 * Disallow all operations smaller than a sub-section and only
259 * allow operations smaller than a section for
260 * SPARSEMEM_VMEMMAP. Note that check_hotplug_memory_range()
261 * enforces a larger memory_block_size_bytes() granularity for
262 * memory that will be marked online, so this check should only
263 * fire for direct arch_{add,remove}_memory() users outside of
264 * add_memory_resource().
266 unsigned long min_align;
268 if (IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP))
269 min_align = PAGES_PER_SUBSECTION;
271 min_align = PAGES_PER_SECTION;
272 if (!IS_ALIGNED(pfn, min_align)
273 || !IS_ALIGNED(nr_pages, min_align)) {
274 WARN(1, "Misaligned __%s_pages start: %#lx end: #%lx\n",
275 reason, pfn, pfn + nr_pages - 1);
282 * Reasonably generic function for adding memory. It is
283 * expected that archs that support memory hotplug will
284 * call this function after deciding the zone to which to
287 int __ref __add_pages(int nid, unsigned long pfn, unsigned long nr_pages,
288 struct mhp_restrictions *restrictions)
291 unsigned long nr, start_sec, end_sec;
292 struct vmem_altmap *altmap = restrictions->altmap;
296 * Validate altmap is within bounds of the total request
298 if (altmap->base_pfn != pfn
299 || vmem_altmap_offset(altmap) > nr_pages) {
300 pr_warn_once("memory add fail, invalid altmap\n");
306 err = check_pfn_span(pfn, nr_pages, "add");
310 start_sec = pfn_to_section_nr(pfn);
311 end_sec = pfn_to_section_nr(pfn + nr_pages - 1);
312 for (nr = start_sec; nr <= end_sec; nr++) {
315 pfns = min(nr_pages, PAGES_PER_SECTION
316 - (pfn & ~PAGE_SECTION_MASK));
317 err = sparse_add_section(nid, pfn, pfns, altmap);
324 vmemmap_populate_print_last();
328 /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
329 static unsigned long find_smallest_section_pfn(int nid, struct zone *zone,
330 unsigned long start_pfn,
331 unsigned long end_pfn)
333 for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SUBSECTION) {
334 if (unlikely(!pfn_valid(start_pfn)))
337 if (unlikely(pfn_to_nid(start_pfn) != nid))
340 if (zone && zone != page_zone(pfn_to_page(start_pfn)))
349 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
350 static unsigned long find_biggest_section_pfn(int nid, struct zone *zone,
351 unsigned long start_pfn,
352 unsigned long end_pfn)
356 /* pfn is the end pfn of a memory section. */
358 for (; pfn >= start_pfn; pfn -= PAGES_PER_SUBSECTION) {
359 if (unlikely(!pfn_valid(pfn)))
362 if (unlikely(pfn_to_nid(pfn) != nid))
365 if (zone && zone != page_zone(pfn_to_page(pfn)))
374 static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
375 unsigned long end_pfn)
377 unsigned long zone_start_pfn = zone->zone_start_pfn;
378 unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */
379 unsigned long zone_end_pfn = z;
381 int nid = zone_to_nid(zone);
383 zone_span_writelock(zone);
384 if (zone_start_pfn == start_pfn) {
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.
391 pfn = find_smallest_section_pfn(nid, zone, end_pfn,
394 zone->zone_start_pfn = pfn;
395 zone->spanned_pages = zone_end_pfn - pfn;
397 } else if (zone_end_pfn == end_pfn) {
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
404 pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn,
407 zone->spanned_pages = pfn - zone_start_pfn + 1;
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.
416 pfn = zone_start_pfn;
417 for (; pfn < zone_end_pfn; pfn += PAGES_PER_SUBSECTION) {
418 if (unlikely(!pfn_valid(pfn)))
421 if (page_zone(pfn_to_page(pfn)) != zone)
424 /* Skip range to be removed */
425 if (pfn >= start_pfn && pfn < end_pfn)
428 /* If we find valid section, we have nothing to do */
429 zone_span_writeunlock(zone);
433 /* The zone has no valid section */
434 zone->zone_start_pfn = 0;
435 zone->spanned_pages = 0;
436 zone_span_writeunlock(zone);
439 static void update_pgdat_span(struct pglist_data *pgdat)
441 unsigned long node_start_pfn = 0, node_end_pfn = 0;
444 for (zone = pgdat->node_zones;
445 zone < pgdat->node_zones + MAX_NR_ZONES; zone++) {
446 unsigned long zone_end_pfn = zone->zone_start_pfn +
449 /* No need to lock the zones, they can't change. */
450 if (zone_end_pfn > node_end_pfn)
451 node_end_pfn = zone_end_pfn;
452 if (zone->zone_start_pfn < node_start_pfn)
453 node_start_pfn = zone->zone_start_pfn;
456 pgdat->node_start_pfn = node_start_pfn;
457 pgdat->node_spanned_pages = node_end_pfn - node_start_pfn;
460 static void __remove_zone(struct zone *zone, unsigned long start_pfn,
461 unsigned long nr_pages)
463 struct pglist_data *pgdat = zone->zone_pgdat;
466 pgdat_resize_lock(zone->zone_pgdat, &flags);
467 shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
468 update_pgdat_span(pgdat);
469 pgdat_resize_unlock(zone->zone_pgdat, &flags);
472 static void __remove_section(struct zone *zone, unsigned long pfn,
473 unsigned long nr_pages, unsigned long map_offset,
474 struct vmem_altmap *altmap)
476 struct mem_section *ms = __nr_to_section(pfn_to_section_nr(pfn));
478 if (WARN_ON_ONCE(!valid_section(ms)))
481 __remove_zone(zone, pfn, nr_pages);
482 sparse_remove_section(ms, pfn, nr_pages, map_offset, altmap);
486 * __remove_pages() - remove sections of pages from a zone
487 * @zone: zone from which pages need to be removed
488 * @pfn: starting pageframe (must be aligned to start of a section)
489 * @nr_pages: number of pages to remove (must be multiple of section size)
490 * @altmap: alternative device page map or %NULL if default memmap is used
492 * Generic helper function to remove section mappings and sysfs entries
493 * for the section of the memory we are removing. Caller needs to make
494 * sure that pages are marked reserved and zones are adjust properly by
495 * calling offline_pages().
497 void __remove_pages(struct zone *zone, unsigned long pfn,
498 unsigned long nr_pages, struct vmem_altmap *altmap)
500 unsigned long map_offset = 0;
501 unsigned long nr, start_sec, end_sec;
503 map_offset = vmem_altmap_offset(altmap);
505 clear_zone_contiguous(zone);
507 if (check_pfn_span(pfn, nr_pages, "remove"))
510 start_sec = pfn_to_section_nr(pfn);
511 end_sec = pfn_to_section_nr(pfn + nr_pages - 1);
512 for (nr = start_sec; nr <= end_sec; nr++) {
516 pfns = min(nr_pages, PAGES_PER_SECTION
517 - (pfn & ~PAGE_SECTION_MASK));
518 __remove_section(zone, pfn, pfns, map_offset, altmap);
524 set_zone_contiguous(zone);
527 int set_online_page_callback(online_page_callback_t callback)
532 mutex_lock(&online_page_callback_lock);
534 if (online_page_callback == generic_online_page) {
535 online_page_callback = callback;
539 mutex_unlock(&online_page_callback_lock);
544 EXPORT_SYMBOL_GPL(set_online_page_callback);
546 int restore_online_page_callback(online_page_callback_t callback)
551 mutex_lock(&online_page_callback_lock);
553 if (online_page_callback == callback) {
554 online_page_callback = generic_online_page;
558 mutex_unlock(&online_page_callback_lock);
563 EXPORT_SYMBOL_GPL(restore_online_page_callback);
565 void __online_page_set_limits(struct page *page)
568 EXPORT_SYMBOL_GPL(__online_page_set_limits);
570 void __online_page_increment_counters(struct page *page)
572 adjust_managed_page_count(page, 1);
574 EXPORT_SYMBOL_GPL(__online_page_increment_counters);
576 void __online_page_free(struct page *page)
578 __free_reserved_page(page);
580 EXPORT_SYMBOL_GPL(__online_page_free);
582 static void generic_online_page(struct page *page, unsigned int order)
584 kernel_map_pages(page, 1 << order, 1);
585 __free_pages_core(page, order);
586 totalram_pages_add(1UL << order);
587 #ifdef CONFIG_HIGHMEM
588 if (PageHighMem(page))
589 totalhigh_pages_add(1UL << order);
593 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
596 const unsigned long end_pfn = start_pfn + nr_pages;
601 * Online the pages. The callback might decide to keep some pages
602 * PG_reserved (to add them to the buddy later), but we still account
603 * them as being online/belonging to this zone ("present").
605 for (pfn = start_pfn; pfn < end_pfn; pfn += 1ul << order) {
606 order = min(MAX_ORDER - 1, get_order(PFN_PHYS(end_pfn - pfn)));
607 /* __free_pages_core() wants pfns to be aligned to the order */
608 if (WARN_ON_ONCE(!IS_ALIGNED(pfn, 1ul << order)))
610 (*online_page_callback)(pfn_to_page(pfn), order);
613 /* mark all involved sections as online */
614 online_mem_sections(start_pfn, end_pfn);
616 *(unsigned long *)arg += nr_pages;
620 /* check which state of node_states will be changed when online memory */
621 static void node_states_check_changes_online(unsigned long nr_pages,
622 struct zone *zone, struct memory_notify *arg)
624 int nid = zone_to_nid(zone);
626 arg->status_change_nid = NUMA_NO_NODE;
627 arg->status_change_nid_normal = NUMA_NO_NODE;
628 arg->status_change_nid_high = NUMA_NO_NODE;
630 if (!node_state(nid, N_MEMORY))
631 arg->status_change_nid = nid;
632 if (zone_idx(zone) <= ZONE_NORMAL && !node_state(nid, N_NORMAL_MEMORY))
633 arg->status_change_nid_normal = nid;
634 #ifdef CONFIG_HIGHMEM
635 if (zone_idx(zone) <= ZONE_HIGHMEM && !node_state(nid, N_HIGH_MEMORY))
636 arg->status_change_nid_high = nid;
640 static void node_states_set_node(int node, struct memory_notify *arg)
642 if (arg->status_change_nid_normal >= 0)
643 node_set_state(node, N_NORMAL_MEMORY);
645 if (arg->status_change_nid_high >= 0)
646 node_set_state(node, N_HIGH_MEMORY);
648 if (arg->status_change_nid >= 0)
649 node_set_state(node, N_MEMORY);
652 static void __meminit resize_zone_range(struct zone *zone, unsigned long start_pfn,
653 unsigned long nr_pages)
655 unsigned long old_end_pfn = zone_end_pfn(zone);
657 if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
658 zone->zone_start_pfn = start_pfn;
660 zone->spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - zone->zone_start_pfn;
663 static void __meminit resize_pgdat_range(struct pglist_data *pgdat, unsigned long start_pfn,
664 unsigned long nr_pages)
666 unsigned long old_end_pfn = pgdat_end_pfn(pgdat);
668 if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
669 pgdat->node_start_pfn = start_pfn;
671 pgdat->node_spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - pgdat->node_start_pfn;
675 * Associate the pfn range with the given zone, initializing the memmaps
676 * and resizing the pgdat/zone data to span the added pages. After this
677 * call, all affected pages are PG_reserved.
679 void __ref move_pfn_range_to_zone(struct zone *zone, unsigned long start_pfn,
680 unsigned long nr_pages, struct vmem_altmap *altmap)
682 struct pglist_data *pgdat = zone->zone_pgdat;
683 int nid = pgdat->node_id;
686 clear_zone_contiguous(zone);
688 /* TODO Huh pgdat is irqsave while zone is not. It used to be like that before */
689 pgdat_resize_lock(pgdat, &flags);
690 zone_span_writelock(zone);
691 if (zone_is_empty(zone))
692 init_currently_empty_zone(zone, start_pfn, nr_pages);
693 resize_zone_range(zone, start_pfn, nr_pages);
694 zone_span_writeunlock(zone);
695 resize_pgdat_range(pgdat, start_pfn, nr_pages);
696 pgdat_resize_unlock(pgdat, &flags);
699 * TODO now we have a visible range of pages which are not associated
700 * with their zone properly. Not nice but set_pfnblock_flags_mask
701 * expects the zone spans the pfn range. All the pages in the range
702 * are reserved so nobody should be touching them so we should be safe
704 memmap_init_zone(nr_pages, nid, zone_idx(zone), start_pfn,
705 MEMMAP_HOTPLUG, altmap);
707 set_zone_contiguous(zone);
711 * Returns a default kernel memory zone for the given pfn range.
712 * If no kernel zone covers this pfn range it will automatically go
713 * to the ZONE_NORMAL.
715 static struct zone *default_kernel_zone_for_pfn(int nid, unsigned long start_pfn,
716 unsigned long nr_pages)
718 struct pglist_data *pgdat = NODE_DATA(nid);
721 for (zid = 0; zid <= ZONE_NORMAL; zid++) {
722 struct zone *zone = &pgdat->node_zones[zid];
724 if (zone_intersects(zone, start_pfn, nr_pages))
728 return &pgdat->node_zones[ZONE_NORMAL];
731 static inline struct zone *default_zone_for_pfn(int nid, unsigned long start_pfn,
732 unsigned long nr_pages)
734 struct zone *kernel_zone = default_kernel_zone_for_pfn(nid, start_pfn,
736 struct zone *movable_zone = &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
737 bool in_kernel = zone_intersects(kernel_zone, start_pfn, nr_pages);
738 bool in_movable = zone_intersects(movable_zone, start_pfn, nr_pages);
741 * We inherit the existing zone in a simple case where zones do not
742 * overlap in the given range
744 if (in_kernel ^ in_movable)
745 return (in_kernel) ? kernel_zone : movable_zone;
748 * If the range doesn't belong to any zone or two zones overlap in the
749 * given range then we use movable zone only if movable_node is
750 * enabled because we always online to a kernel zone by default.
752 return movable_node_enabled ? movable_zone : kernel_zone;
755 struct zone * zone_for_pfn_range(int online_type, int nid, unsigned start_pfn,
756 unsigned long nr_pages)
758 if (online_type == MMOP_ONLINE_KERNEL)
759 return default_kernel_zone_for_pfn(nid, start_pfn, nr_pages);
761 if (online_type == MMOP_ONLINE_MOVABLE)
762 return &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
764 return default_zone_for_pfn(nid, start_pfn, nr_pages);
767 int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type)
770 unsigned long onlined_pages = 0;
772 int need_zonelists_rebuild = 0;
775 struct memory_notify arg;
776 struct memory_block *mem;
781 * We can't use pfn_to_nid() because nid might be stored in struct page
782 * which is not yet initialized. Instead, we find nid from memory block.
784 mem = find_memory_block(__pfn_to_section(pfn));
786 put_device(&mem->dev);
788 /* associate pfn range with the zone */
789 zone = zone_for_pfn_range(online_type, nid, pfn, nr_pages);
790 move_pfn_range_to_zone(zone, pfn, nr_pages, NULL);
793 arg.nr_pages = nr_pages;
794 node_states_check_changes_online(nr_pages, zone, &arg);
796 ret = memory_notify(MEM_GOING_ONLINE, &arg);
797 ret = notifier_to_errno(ret);
799 goto failed_addition;
802 * If this zone is not populated, then it is not in zonelist.
803 * This means the page allocator ignores this zone.
804 * So, zonelist must be updated after online.
806 if (!populated_zone(zone)) {
807 need_zonelists_rebuild = 1;
808 setup_zone_pageset(zone);
811 ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
814 /* not a single memory resource was applicable */
815 if (need_zonelists_rebuild)
816 zone_pcp_reset(zone);
817 goto failed_addition;
820 zone->present_pages += onlined_pages;
822 pgdat_resize_lock(zone->zone_pgdat, &flags);
823 zone->zone_pgdat->node_present_pages += onlined_pages;
824 pgdat_resize_unlock(zone->zone_pgdat, &flags);
828 node_states_set_node(nid, &arg);
829 if (need_zonelists_rebuild)
830 build_all_zonelists(NULL);
832 zone_pcp_update(zone);
834 init_per_zone_wmark_min();
839 vm_total_pages = nr_free_pagecache_pages();
841 writeback_set_ratelimit();
843 memory_notify(MEM_ONLINE, &arg);
848 pr_debug("online_pages [mem %#010llx-%#010llx] failed\n",
849 (unsigned long long) pfn << PAGE_SHIFT,
850 (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1);
851 memory_notify(MEM_CANCEL_ONLINE, &arg);
855 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
857 static void reset_node_present_pages(pg_data_t *pgdat)
861 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
862 z->present_pages = 0;
864 pgdat->node_present_pages = 0;
867 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
868 static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
870 struct pglist_data *pgdat;
871 unsigned long start_pfn = PFN_DOWN(start);
873 pgdat = NODE_DATA(nid);
875 pgdat = arch_alloc_nodedata(nid);
879 pgdat->per_cpu_nodestats =
880 alloc_percpu(struct per_cpu_nodestat);
881 arch_refresh_nodedata(nid, pgdat);
885 * Reset the nr_zones, order and classzone_idx before reuse.
886 * Note that kswapd will init kswapd_classzone_idx properly
887 * when it starts in the near future.
890 pgdat->kswapd_order = 0;
891 pgdat->kswapd_classzone_idx = 0;
892 for_each_online_cpu(cpu) {
893 struct per_cpu_nodestat *p;
895 p = per_cpu_ptr(pgdat->per_cpu_nodestats, cpu);
896 memset(p, 0, sizeof(*p));
900 /* we can use NODE_DATA(nid) from here */
902 pgdat->node_id = nid;
903 pgdat->node_start_pfn = start_pfn;
905 /* init node's zones as empty zones, we don't have any present pages.*/
906 free_area_init_core_hotplug(nid);
909 * The node we allocated has no zone fallback lists. For avoiding
910 * to access not-initialized zonelist, build here.
912 build_all_zonelists(pgdat);
915 * When memory is hot-added, all the memory is in offline state. So
916 * clear all zones' present_pages because they will be updated in
917 * online_pages() and offline_pages().
919 reset_node_managed_pages(pgdat);
920 reset_node_present_pages(pgdat);
925 static void rollback_node_hotadd(int nid)
927 pg_data_t *pgdat = NODE_DATA(nid);
929 arch_refresh_nodedata(nid, NULL);
930 free_percpu(pgdat->per_cpu_nodestats);
931 arch_free_nodedata(pgdat);
936 * try_online_node - online a node if offlined
938 * @start: start addr of the node
939 * @set_node_online: Whether we want to online the node
940 * called by cpu_up() to online a node without onlined memory.
943 * 1 -> a new node has been allocated
944 * 0 -> the node is already online
945 * -ENOMEM -> the node could not be allocated
947 static int __try_online_node(int nid, u64 start, bool set_node_online)
952 if (node_online(nid))
955 pgdat = hotadd_new_pgdat(nid, start);
957 pr_err("Cannot online node %d due to NULL pgdat\n", nid);
962 if (set_node_online) {
963 node_set_online(nid);
964 ret = register_one_node(nid);
972 * Users of this function always want to online/register the node
974 int try_online_node(int nid)
979 ret = __try_online_node(nid, 0, true);
984 static int check_hotplug_memory_range(u64 start, u64 size)
986 /* memory range must be block size aligned */
987 if (!size || !IS_ALIGNED(start, memory_block_size_bytes()) ||
988 !IS_ALIGNED(size, memory_block_size_bytes())) {
989 pr_err("Block size [%#lx] unaligned hotplug range: start %#llx, size %#llx",
990 memory_block_size_bytes(), start, size);
997 static int online_memory_block(struct memory_block *mem, void *arg)
999 return device_online(&mem->dev);
1003 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1004 * and online/offline operations (triggered e.g. by sysfs).
1006 * we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG
1008 int __ref add_memory_resource(int nid, struct resource *res)
1010 struct mhp_restrictions restrictions = {};
1012 bool new_node = false;
1016 size = resource_size(res);
1018 ret = check_hotplug_memory_range(start, size);
1022 mem_hotplug_begin();
1025 * Add new range to memblock so that when hotadd_new_pgdat() is called
1026 * to allocate new pgdat, get_pfn_range_for_nid() will be able to find
1027 * this new range and calculate total pages correctly. The range will
1028 * be removed at hot-remove time.
1030 memblock_add_node(start, size, nid);
1032 ret = __try_online_node(nid, start, false);
1037 /* call arch's memory hotadd */
1038 ret = arch_add_memory(nid, start, size, &restrictions);
1042 /* create memory block devices after memory was added */
1043 ret = create_memory_block_devices(start, size);
1045 arch_remove_memory(nid, start, size, NULL);
1050 /* If sysfs file of new node can't be created, cpu on the node
1051 * can't be hot-added. There is no rollback way now.
1052 * So, check by BUG_ON() to catch it reluctantly..
1053 * We online node here. We can't roll back from here.
1055 node_set_online(nid);
1056 ret = __register_one_node(nid);
1060 /* link memory sections under this node.*/
1061 ret = link_mem_sections(nid, PFN_DOWN(start), PFN_UP(start + size - 1));
1064 /* create new memmap entry */
1065 firmware_map_add_hotplug(start, start + size, "System RAM");
1067 /* device_online() will take the lock when calling online_pages() */
1070 /* online pages if requested */
1071 if (memhp_auto_online)
1072 walk_memory_blocks(start, size, NULL, online_memory_block);
1076 /* rollback pgdat allocation and others */
1078 rollback_node_hotadd(nid);
1079 memblock_remove(start, size);
1084 /* requires device_hotplug_lock, see add_memory_resource() */
1085 int __ref __add_memory(int nid, u64 start, u64 size)
1087 struct resource *res;
1090 res = register_memory_resource(start, size);
1092 return PTR_ERR(res);
1094 ret = add_memory_resource(nid, res);
1096 release_memory_resource(res);
1100 int add_memory(int nid, u64 start, u64 size)
1104 lock_device_hotplug();
1105 rc = __add_memory(nid, start, size);
1106 unlock_device_hotplug();
1110 EXPORT_SYMBOL_GPL(add_memory);
1112 #ifdef CONFIG_MEMORY_HOTREMOVE
1114 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1115 * set and the size of the free page is given by page_order(). Using this,
1116 * the function determines if the pageblock contains only free pages.
1117 * Due to buddy contraints, a free page at least the size of a pageblock will
1118 * be located at the start of the pageblock
1120 static inline int pageblock_free(struct page *page)
1122 return PageBuddy(page) && page_order(page) >= pageblock_order;
1125 /* Return the pfn of the start of the next active pageblock after a given pfn */
1126 static unsigned long next_active_pageblock(unsigned long pfn)
1128 struct page *page = pfn_to_page(pfn);
1130 /* Ensure the starting page is pageblock-aligned */
1131 BUG_ON(pfn & (pageblock_nr_pages - 1));
1133 /* If the entire pageblock is free, move to the end of free page */
1134 if (pageblock_free(page)) {
1136 /* be careful. we don't have locks, page_order can be changed.*/
1137 order = page_order(page);
1138 if ((order < MAX_ORDER) && (order >= pageblock_order))
1139 return pfn + (1 << order);
1142 return pfn + pageblock_nr_pages;
1145 static bool is_pageblock_removable_nolock(unsigned long pfn)
1147 struct page *page = pfn_to_page(pfn);
1151 * We have to be careful here because we are iterating over memory
1152 * sections which are not zone aware so we might end up outside of
1153 * the zone but still within the section.
1154 * We have to take care about the node as well. If the node is offline
1155 * its NODE_DATA will be NULL - see page_zone.
1157 if (!node_online(page_to_nid(page)))
1160 zone = page_zone(page);
1161 pfn = page_to_pfn(page);
1162 if (!zone_spans_pfn(zone, pfn))
1165 return !has_unmovable_pages(zone, page, 0, MIGRATE_MOVABLE, SKIP_HWPOISON);
1168 /* Checks if this range of memory is likely to be hot-removable. */
1169 bool is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
1171 unsigned long end_pfn, pfn;
1173 end_pfn = min(start_pfn + nr_pages,
1174 zone_end_pfn(page_zone(pfn_to_page(start_pfn))));
1176 /* Check the starting page of each pageblock within the range */
1177 for (pfn = start_pfn; pfn < end_pfn; pfn = next_active_pageblock(pfn)) {
1178 if (!is_pageblock_removable_nolock(pfn))
1183 /* All pageblocks in the memory block are likely to be hot-removable */
1188 * Confirm all pages in a range [start, end) belong to the same zone.
1189 * When true, return its valid [start, end).
1191 int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn,
1192 unsigned long *valid_start, unsigned long *valid_end)
1194 unsigned long pfn, sec_end_pfn;
1195 unsigned long start, end;
1196 struct zone *zone = NULL;
1199 for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1);
1201 pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) {
1202 /* Make sure the memory section is present first */
1203 if (!present_section_nr(pfn_to_section_nr(pfn)))
1205 for (; pfn < sec_end_pfn && pfn < end_pfn;
1206 pfn += MAX_ORDER_NR_PAGES) {
1208 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1209 while ((i < MAX_ORDER_NR_PAGES) &&
1210 !pfn_valid_within(pfn + i))
1212 if (i == MAX_ORDER_NR_PAGES || pfn + i >= end_pfn)
1214 /* Check if we got outside of the zone */
1215 if (zone && !zone_spans_pfn(zone, pfn + i))
1217 page = pfn_to_page(pfn + i);
1218 if (zone && page_zone(page) != zone)
1222 zone = page_zone(page);
1223 end = pfn + MAX_ORDER_NR_PAGES;
1228 *valid_start = start;
1229 *valid_end = min(end, end_pfn);
1237 * Scan pfn range [start,end) to find movable/migratable pages (LRU pages,
1238 * non-lru movable pages and hugepages). We scan pfn because it's much
1239 * easier than scanning over linked list. This function returns the pfn
1240 * of the first found movable page if it's found, otherwise 0.
1242 static unsigned long scan_movable_pages(unsigned long start, unsigned long end)
1246 for (pfn = start; pfn < end; pfn++) {
1247 struct page *page, *head;
1250 if (!pfn_valid(pfn))
1252 page = pfn_to_page(pfn);
1255 if (__PageMovable(page))
1258 if (!PageHuge(page))
1260 head = compound_head(page);
1261 if (page_huge_active(head))
1263 skip = compound_nr(head) - (page - head);
1269 static struct page *new_node_page(struct page *page, unsigned long private)
1271 int nid = page_to_nid(page);
1272 nodemask_t nmask = node_states[N_MEMORY];
1275 * try to allocate from a different node but reuse this node if there
1276 * are no other online nodes to be used (e.g. we are offlining a part
1277 * of the only existing node)
1279 node_clear(nid, nmask);
1280 if (nodes_empty(nmask))
1281 node_set(nid, nmask);
1283 return new_page_nodemask(page, nid, &nmask);
1287 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1294 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1295 if (!pfn_valid(pfn))
1297 page = pfn_to_page(pfn);
1299 if (PageHuge(page)) {
1300 struct page *head = compound_head(page);
1301 pfn = page_to_pfn(head) + compound_nr(head) - 1;
1302 isolate_huge_page(head, &source);
1304 } else if (PageTransHuge(page))
1305 pfn = page_to_pfn(compound_head(page))
1306 + hpage_nr_pages(page) - 1;
1309 * HWPoison pages have elevated reference counts so the migration would
1310 * fail on them. It also doesn't make any sense to migrate them in the
1311 * first place. Still try to unmap such a page in case it is still mapped
1312 * (e.g. current hwpoison implementation doesn't unmap KSM pages but keep
1313 * the unmap as the catch all safety net).
1315 if (PageHWPoison(page)) {
1316 if (WARN_ON(PageLRU(page)))
1317 isolate_lru_page(page);
1318 if (page_mapped(page))
1319 try_to_unmap(page, TTU_IGNORE_MLOCK | TTU_IGNORE_ACCESS);
1323 if (!get_page_unless_zero(page))
1326 * We can skip free pages. And we can deal with pages on
1327 * LRU and non-lru movable pages.
1330 ret = isolate_lru_page(page);
1332 ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE);
1333 if (!ret) { /* Success */
1334 list_add_tail(&page->lru, &source);
1335 if (!__PageMovable(page))
1336 inc_node_page_state(page, NR_ISOLATED_ANON +
1337 page_is_file_cache(page));
1340 pr_warn("failed to isolate pfn %lx\n", pfn);
1341 dump_page(page, "isolation failed");
1345 if (!list_empty(&source)) {
1346 /* Allocate a new page from the nearest neighbor node */
1347 ret = migrate_pages(&source, new_node_page, NULL, 0,
1348 MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
1350 list_for_each_entry(page, &source, lru) {
1351 pr_warn("migrating pfn %lx failed ret:%d ",
1352 page_to_pfn(page), ret);
1353 dump_page(page, "migration failure");
1355 putback_movable_pages(&source);
1363 * remove from free_area[] and mark all as Reserved.
1366 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1369 unsigned long *offlined_pages = (unsigned long *)data;
1371 *offlined_pages += __offline_isolated_pages(start, start + nr_pages);
1376 * Check all pages in range, recoreded as memory resource, are isolated.
1379 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1382 return test_pages_isolated(start_pfn, start_pfn + nr_pages, true);
1385 static int __init cmdline_parse_movable_node(char *p)
1387 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1388 movable_node_enabled = true;
1390 pr_warn("movable_node parameter depends on CONFIG_HAVE_MEMBLOCK_NODE_MAP to work properly\n");
1394 early_param("movable_node", cmdline_parse_movable_node);
1396 /* check which state of node_states will be changed when offline memory */
1397 static void node_states_check_changes_offline(unsigned long nr_pages,
1398 struct zone *zone, struct memory_notify *arg)
1400 struct pglist_data *pgdat = zone->zone_pgdat;
1401 unsigned long present_pages = 0;
1404 arg->status_change_nid = NUMA_NO_NODE;
1405 arg->status_change_nid_normal = NUMA_NO_NODE;
1406 arg->status_change_nid_high = NUMA_NO_NODE;
1409 * Check whether node_states[N_NORMAL_MEMORY] will be changed.
1410 * If the memory to be offline is within the range
1411 * [0..ZONE_NORMAL], and it is the last present memory there,
1412 * the zones in that range will become empty after the offlining,
1413 * thus we can determine that we need to clear the node from
1414 * node_states[N_NORMAL_MEMORY].
1416 for (zt = 0; zt <= ZONE_NORMAL; zt++)
1417 present_pages += pgdat->node_zones[zt].present_pages;
1418 if (zone_idx(zone) <= ZONE_NORMAL && nr_pages >= present_pages)
1419 arg->status_change_nid_normal = zone_to_nid(zone);
1421 #ifdef CONFIG_HIGHMEM
1423 * node_states[N_HIGH_MEMORY] contains nodes which
1424 * have normal memory or high memory.
1425 * Here we add the present_pages belonging to ZONE_HIGHMEM.
1426 * If the zone is within the range of [0..ZONE_HIGHMEM), and
1427 * we determine that the zones in that range become empty,
1428 * we need to clear the node for N_HIGH_MEMORY.
1430 present_pages += pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1431 if (zone_idx(zone) <= ZONE_HIGHMEM && nr_pages >= present_pages)
1432 arg->status_change_nid_high = zone_to_nid(zone);
1436 * We have accounted the pages from [0..ZONE_NORMAL), and
1437 * in case of CONFIG_HIGHMEM the pages from ZONE_HIGHMEM
1439 * Here we count the possible pages from ZONE_MOVABLE.
1440 * If after having accounted all the pages, we see that the nr_pages
1441 * to be offlined is over or equal to the accounted pages,
1442 * we know that the node will become empty, and so, we can clear
1443 * it for N_MEMORY as well.
1445 present_pages += pgdat->node_zones[ZONE_MOVABLE].present_pages;
1447 if (nr_pages >= present_pages)
1448 arg->status_change_nid = zone_to_nid(zone);
1451 static void node_states_clear_node(int node, struct memory_notify *arg)
1453 if (arg->status_change_nid_normal >= 0)
1454 node_clear_state(node, N_NORMAL_MEMORY);
1456 if (arg->status_change_nid_high >= 0)
1457 node_clear_state(node, N_HIGH_MEMORY);
1459 if (arg->status_change_nid >= 0)
1460 node_clear_state(node, N_MEMORY);
1463 static int __ref __offline_pages(unsigned long start_pfn,
1464 unsigned long end_pfn)
1466 unsigned long pfn, nr_pages;
1467 unsigned long offlined_pages = 0;
1468 int ret, node, nr_isolate_pageblock;
1469 unsigned long flags;
1470 unsigned long valid_start, valid_end;
1472 struct memory_notify arg;
1475 mem_hotplug_begin();
1477 /* This makes hotplug much easier...and readable.
1478 we assume this for now. .*/
1479 if (!test_pages_in_a_zone(start_pfn, end_pfn, &valid_start,
1482 reason = "multizone range";
1483 goto failed_removal;
1486 zone = page_zone(pfn_to_page(valid_start));
1487 node = zone_to_nid(zone);
1488 nr_pages = end_pfn - start_pfn;
1490 /* set above range as isolated */
1491 ret = start_isolate_page_range(start_pfn, end_pfn,
1493 SKIP_HWPOISON | REPORT_FAILURE);
1495 reason = "failure to isolate range";
1496 goto failed_removal;
1498 nr_isolate_pageblock = ret;
1500 arg.start_pfn = start_pfn;
1501 arg.nr_pages = nr_pages;
1502 node_states_check_changes_offline(nr_pages, zone, &arg);
1504 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1505 ret = notifier_to_errno(ret);
1507 reason = "notifier failure";
1508 goto failed_removal_isolated;
1512 for (pfn = start_pfn; pfn;) {
1513 if (signal_pending(current)) {
1515 reason = "signal backoff";
1516 goto failed_removal_isolated;
1520 lru_add_drain_all();
1522 pfn = scan_movable_pages(pfn, end_pfn);
1525 * TODO: fatal migration failures should bail
1528 do_migrate_range(pfn, end_pfn);
1533 * Dissolve free hugepages in the memory block before doing
1534 * offlining actually in order to make hugetlbfs's object
1535 * counting consistent.
1537 ret = dissolve_free_huge_pages(start_pfn, end_pfn);
1539 reason = "failure to dissolve huge pages";
1540 goto failed_removal_isolated;
1543 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn,
1544 NULL, check_pages_isolated_cb);
1547 /* Ok, all of our target is isolated.
1548 We cannot do rollback at this point. */
1549 walk_system_ram_range(start_pfn, end_pfn - start_pfn,
1550 &offlined_pages, offline_isolated_pages_cb);
1551 pr_info("Offlined Pages %ld\n", offlined_pages);
1553 * Onlining will reset pagetype flags and makes migrate type
1554 * MOVABLE, so just need to decrease the number of isolated
1555 * pageblocks zone counter here.
1557 spin_lock_irqsave(&zone->lock, flags);
1558 zone->nr_isolate_pageblock -= nr_isolate_pageblock;
1559 spin_unlock_irqrestore(&zone->lock, flags);
1561 /* removal success */
1562 adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages);
1563 zone->present_pages -= offlined_pages;
1565 pgdat_resize_lock(zone->zone_pgdat, &flags);
1566 zone->zone_pgdat->node_present_pages -= offlined_pages;
1567 pgdat_resize_unlock(zone->zone_pgdat, &flags);
1569 init_per_zone_wmark_min();
1571 if (!populated_zone(zone)) {
1572 zone_pcp_reset(zone);
1573 build_all_zonelists(NULL);
1575 zone_pcp_update(zone);
1577 node_states_clear_node(node, &arg);
1578 if (arg.status_change_nid >= 0) {
1580 kcompactd_stop(node);
1583 vm_total_pages = nr_free_pagecache_pages();
1584 writeback_set_ratelimit();
1586 memory_notify(MEM_OFFLINE, &arg);
1590 failed_removal_isolated:
1591 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1592 memory_notify(MEM_CANCEL_OFFLINE, &arg);
1594 pr_debug("memory offlining [mem %#010llx-%#010llx] failed due to %s\n",
1595 (unsigned long long) start_pfn << PAGE_SHIFT,
1596 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1,
1598 /* pushback to free area */
1603 int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1605 return __offline_pages(start_pfn, start_pfn + nr_pages);
1608 static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
1610 int ret = !is_memblock_offlined(mem);
1612 if (unlikely(ret)) {
1613 phys_addr_t beginpa, endpa;
1615 beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
1616 endpa = beginpa + memory_block_size_bytes() - 1;
1617 pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n",
1625 static int check_cpu_on_node(pg_data_t *pgdat)
1629 for_each_present_cpu(cpu) {
1630 if (cpu_to_node(cpu) == pgdat->node_id)
1632 * the cpu on this node isn't removed, and we can't
1633 * offline this node.
1645 * Offline a node if all memory sections and cpus of the node are removed.
1647 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1648 * and online/offline operations before this call.
1650 void try_offline_node(int nid)
1652 pg_data_t *pgdat = NODE_DATA(nid);
1653 unsigned long start_pfn = pgdat->node_start_pfn;
1654 unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
1657 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1658 unsigned long section_nr = pfn_to_section_nr(pfn);
1660 if (!present_section_nr(section_nr))
1663 if (pfn_to_nid(pfn) != nid)
1667 * some memory sections of this node are not removed, and we
1668 * can't offline node now.
1673 if (check_cpu_on_node(pgdat))
1677 * all memory/cpu of this node are removed, we can offline this
1680 node_set_offline(nid);
1681 unregister_one_node(nid);
1683 EXPORT_SYMBOL(try_offline_node);
1685 static void __release_memory_resource(resource_size_t start,
1686 resource_size_t size)
1691 * When removing memory in the same granularity as it was added,
1692 * this function never fails. It might only fail if resources
1693 * have to be adjusted or split. We'll ignore the error, as
1694 * removing of memory cannot fail.
1696 ret = release_mem_region_adjustable(&iomem_resource, start, size);
1698 resource_size_t endres = start + size - 1;
1700 pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
1701 &start, &endres, ret);
1705 static int __ref try_remove_memory(int nid, u64 start, u64 size)
1709 BUG_ON(check_hotplug_memory_range(start, size));
1711 mem_hotplug_begin();
1714 * All memory blocks must be offlined before removing memory. Check
1715 * whether all memory blocks in question are offline and return error
1716 * if this is not the case.
1718 rc = walk_memory_blocks(start, size, NULL, check_memblock_offlined_cb);
1722 /* remove memmap entry */
1723 firmware_map_remove(start, start + size, "System RAM");
1724 memblock_free(start, size);
1725 memblock_remove(start, size);
1727 /* remove memory block devices before removing memory */
1728 remove_memory_block_devices(start, size);
1730 arch_remove_memory(nid, start, size, NULL);
1731 __release_memory_resource(start, size);
1733 try_offline_node(nid);
1743 * @start: physical address of the region to remove
1744 * @size: size of the region to remove
1746 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1747 * and online/offline operations before this call, as required by
1748 * try_offline_node().
1750 void __remove_memory(int nid, u64 start, u64 size)
1754 * trigger BUG() if some memory is not offlined prior to calling this
1757 if (try_remove_memory(nid, start, size))
1762 * Remove memory if every memory block is offline, otherwise return -EBUSY is
1763 * some memory is not offline
1765 int remove_memory(int nid, u64 start, u64 size)
1769 lock_device_hotplug();
1770 rc = try_remove_memory(nid, start, size);
1771 unlock_device_hotplug();
1775 EXPORT_SYMBOL_GPL(remove_memory);
1776 #endif /* CONFIG_MEMORY_HOTREMOVE */