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 online_page_callback_t online_page_callback = generic_online_page;
53 static DEFINE_MUTEX(online_page_callback_lock);
55 DEFINE_STATIC_PERCPU_RWSEM(mem_hotplug_lock);
57 void get_online_mems(void)
59 percpu_down_read(&mem_hotplug_lock);
62 void put_online_mems(void)
64 percpu_up_read(&mem_hotplug_lock);
67 bool movable_node_enabled = false;
69 #ifndef CONFIG_MEMORY_HOTPLUG_DEFAULT_ONLINE
70 int memhp_default_online_type = MMOP_OFFLINE;
72 int memhp_default_online_type = MMOP_ONLINE;
75 static int __init setup_memhp_default_state(char *str)
77 const int online_type = memhp_online_type_from_str(str);
80 memhp_default_online_type = online_type;
84 __setup("memhp_default_state=", setup_memhp_default_state);
86 void mem_hotplug_begin(void)
89 percpu_down_write(&mem_hotplug_lock);
92 void mem_hotplug_done(void)
94 percpu_up_write(&mem_hotplug_lock);
98 u64 max_mem_size = U64_MAX;
100 /* add this memory to iomem resource */
101 static struct resource *register_memory_resource(u64 start, u64 size)
103 struct resource *res;
104 unsigned long flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
105 char *resource_name = "System RAM";
108 * Make sure value parsed from 'mem=' only restricts memory adding
109 * while booting, so that memory hotplug won't be impacted. Please
110 * refer to document of 'mem=' in kernel-parameters.txt for more
113 if (start + size > max_mem_size && system_state < SYSTEM_RUNNING)
114 return ERR_PTR(-E2BIG);
117 * Request ownership of the new memory range. This might be
118 * a child of an existing resource that was present but
119 * not marked as busy.
121 res = __request_region(&iomem_resource, start, size,
122 resource_name, flags);
125 pr_debug("Unable to reserve System RAM region: %016llx->%016llx\n",
126 start, start + size);
127 return ERR_PTR(-EEXIST);
132 static void release_memory_resource(struct resource *res)
136 release_resource(res);
140 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
141 void get_page_bootmem(unsigned long info, struct page *page,
144 page->freelist = (void *)type;
145 SetPagePrivate(page);
146 set_page_private(page, info);
150 void put_page_bootmem(struct page *page)
154 type = (unsigned long) page->freelist;
155 BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE ||
156 type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE);
158 if (page_ref_dec_return(page) == 1) {
159 page->freelist = NULL;
160 ClearPagePrivate(page);
161 set_page_private(page, 0);
162 INIT_LIST_HEAD(&page->lru);
163 free_reserved_page(page);
167 #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
168 #ifndef CONFIG_SPARSEMEM_VMEMMAP
169 static void register_page_bootmem_info_section(unsigned long start_pfn)
171 unsigned long mapsize, section_nr, i;
172 struct mem_section *ms;
173 struct page *page, *memmap;
174 struct mem_section_usage *usage;
176 section_nr = pfn_to_section_nr(start_pfn);
177 ms = __nr_to_section(section_nr);
179 /* Get section's memmap address */
180 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
183 * Get page for the memmap's phys address
184 * XXX: need more consideration for sparse_vmemmap...
186 page = virt_to_page(memmap);
187 mapsize = sizeof(struct page) * PAGES_PER_SECTION;
188 mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
190 /* remember memmap's page */
191 for (i = 0; i < mapsize; i++, page++)
192 get_page_bootmem(section_nr, page, SECTION_INFO);
195 page = virt_to_page(usage);
197 mapsize = PAGE_ALIGN(mem_section_usage_size()) >> PAGE_SHIFT;
199 for (i = 0; i < mapsize; i++, page++)
200 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
203 #else /* CONFIG_SPARSEMEM_VMEMMAP */
204 static void register_page_bootmem_info_section(unsigned long start_pfn)
206 unsigned long mapsize, section_nr, i;
207 struct mem_section *ms;
208 struct page *page, *memmap;
209 struct mem_section_usage *usage;
211 section_nr = pfn_to_section_nr(start_pfn);
212 ms = __nr_to_section(section_nr);
214 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
216 register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION);
219 page = virt_to_page(usage);
221 mapsize = PAGE_ALIGN(mem_section_usage_size()) >> PAGE_SHIFT;
223 for (i = 0; i < mapsize; i++, page++)
224 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
226 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
228 void __init register_page_bootmem_info_node(struct pglist_data *pgdat)
230 unsigned long i, pfn, end_pfn, nr_pages;
231 int node = pgdat->node_id;
234 nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
235 page = virt_to_page(pgdat);
237 for (i = 0; i < nr_pages; i++, page++)
238 get_page_bootmem(node, page, NODE_INFO);
240 pfn = pgdat->node_start_pfn;
241 end_pfn = pgdat_end_pfn(pgdat);
243 /* register section info */
244 for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
246 * Some platforms can assign the same pfn to multiple nodes - on
247 * node0 as well as nodeN. To avoid registering a pfn against
248 * multiple nodes we check that this pfn does not already
249 * reside in some other nodes.
251 if (pfn_valid(pfn) && (early_pfn_to_nid(pfn) == node))
252 register_page_bootmem_info_section(pfn);
255 #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
257 static int check_pfn_span(unsigned long pfn, unsigned long nr_pages,
261 * Disallow all operations smaller than a sub-section and only
262 * allow operations smaller than a section for
263 * SPARSEMEM_VMEMMAP. Note that check_hotplug_memory_range()
264 * enforces a larger memory_block_size_bytes() granularity for
265 * memory that will be marked online, so this check should only
266 * fire for direct arch_{add,remove}_memory() users outside of
267 * add_memory_resource().
269 unsigned long min_align;
271 if (IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP))
272 min_align = PAGES_PER_SUBSECTION;
274 min_align = PAGES_PER_SECTION;
275 if (!IS_ALIGNED(pfn, min_align)
276 || !IS_ALIGNED(nr_pages, min_align)) {
277 WARN(1, "Misaligned __%s_pages start: %#lx end: #%lx\n",
278 reason, pfn, pfn + nr_pages - 1);
284 static int check_hotplug_memory_addressable(unsigned long pfn,
285 unsigned long nr_pages)
287 const u64 max_addr = PFN_PHYS(pfn + nr_pages) - 1;
289 if (max_addr >> MAX_PHYSMEM_BITS) {
290 const u64 max_allowed = (1ull << (MAX_PHYSMEM_BITS + 1)) - 1;
292 "Hotplugged memory exceeds maximum addressable address, range=%#llx-%#llx, maximum=%#llx\n",
293 (u64)PFN_PHYS(pfn), max_addr, max_allowed);
301 * Reasonably generic function for adding memory. It is
302 * expected that archs that support memory hotplug will
303 * call this function after deciding the zone to which to
306 int __ref __add_pages(int nid, unsigned long pfn, unsigned long nr_pages,
307 struct mhp_params *params)
309 const unsigned long end_pfn = pfn + nr_pages;
310 unsigned long cur_nr_pages;
312 struct vmem_altmap *altmap = params->altmap;
314 if (WARN_ON_ONCE(!params->pgprot.pgprot))
317 err = check_hotplug_memory_addressable(pfn, nr_pages);
323 * Validate altmap is within bounds of the total request
325 if (altmap->base_pfn != pfn
326 || vmem_altmap_offset(altmap) > nr_pages) {
327 pr_warn_once("memory add fail, invalid altmap\n");
333 err = check_pfn_span(pfn, nr_pages, "add");
337 for (; pfn < end_pfn; pfn += cur_nr_pages) {
338 /* Select all remaining pages up to the next section boundary */
339 cur_nr_pages = min(end_pfn - pfn,
340 SECTION_ALIGN_UP(pfn + 1) - pfn);
341 err = sparse_add_section(nid, pfn, cur_nr_pages, altmap);
346 vmemmap_populate_print_last();
350 /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
351 static unsigned long find_smallest_section_pfn(int nid, struct zone *zone,
352 unsigned long start_pfn,
353 unsigned long end_pfn)
355 for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SUBSECTION) {
356 if (unlikely(!pfn_to_online_page(start_pfn)))
359 if (unlikely(pfn_to_nid(start_pfn) != nid))
362 if (zone != page_zone(pfn_to_page(start_pfn)))
371 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
372 static unsigned long find_biggest_section_pfn(int nid, struct zone *zone,
373 unsigned long start_pfn,
374 unsigned long end_pfn)
378 /* pfn is the end pfn of a memory section. */
380 for (; pfn >= start_pfn; pfn -= PAGES_PER_SUBSECTION) {
381 if (unlikely(!pfn_to_online_page(pfn)))
384 if (unlikely(pfn_to_nid(pfn) != nid))
387 if (zone != page_zone(pfn_to_page(pfn)))
396 static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
397 unsigned long end_pfn)
400 int nid = zone_to_nid(zone);
402 zone_span_writelock(zone);
403 if (zone->zone_start_pfn == start_pfn) {
405 * If the section is smallest section in the zone, it need
406 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
407 * In this case, we find second smallest valid mem_section
408 * for shrinking zone.
410 pfn = find_smallest_section_pfn(nid, zone, end_pfn,
413 zone->spanned_pages = zone_end_pfn(zone) - pfn;
414 zone->zone_start_pfn = pfn;
416 zone->zone_start_pfn = 0;
417 zone->spanned_pages = 0;
419 } else if (zone_end_pfn(zone) == end_pfn) {
421 * If the section is biggest section in the zone, it need
422 * shrink zone->spanned_pages.
423 * In this case, we find second biggest valid mem_section for
426 pfn = find_biggest_section_pfn(nid, zone, zone->zone_start_pfn,
429 zone->spanned_pages = pfn - zone->zone_start_pfn + 1;
431 zone->zone_start_pfn = 0;
432 zone->spanned_pages = 0;
435 zone_span_writeunlock(zone);
438 static void update_pgdat_span(struct pglist_data *pgdat)
440 unsigned long node_start_pfn = 0, node_end_pfn = 0;
443 for (zone = pgdat->node_zones;
444 zone < pgdat->node_zones + MAX_NR_ZONES; zone++) {
445 unsigned long zone_end_pfn = zone->zone_start_pfn +
448 /* No need to lock the zones, they can't change. */
449 if (!zone->spanned_pages)
452 node_start_pfn = zone->zone_start_pfn;
453 node_end_pfn = zone_end_pfn;
457 if (zone_end_pfn > node_end_pfn)
458 node_end_pfn = zone_end_pfn;
459 if (zone->zone_start_pfn < node_start_pfn)
460 node_start_pfn = zone->zone_start_pfn;
463 pgdat->node_start_pfn = node_start_pfn;
464 pgdat->node_spanned_pages = node_end_pfn - node_start_pfn;
467 void __ref remove_pfn_range_from_zone(struct zone *zone,
468 unsigned long start_pfn,
469 unsigned long nr_pages)
471 struct pglist_data *pgdat = zone->zone_pgdat;
474 /* Poison struct pages because they are now uninitialized again. */
475 page_init_poison(pfn_to_page(start_pfn), sizeof(struct page) * nr_pages);
477 #ifdef CONFIG_ZONE_DEVICE
479 * Zone shrinking code cannot properly deal with ZONE_DEVICE. So
480 * we will not try to shrink the zones - which is okay as
481 * set_zone_contiguous() cannot deal with ZONE_DEVICE either way.
483 if (zone_idx(zone) == ZONE_DEVICE)
487 clear_zone_contiguous(zone);
489 pgdat_resize_lock(zone->zone_pgdat, &flags);
490 shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
491 update_pgdat_span(pgdat);
492 pgdat_resize_unlock(zone->zone_pgdat, &flags);
494 set_zone_contiguous(zone);
497 static void __remove_section(unsigned long pfn, unsigned long nr_pages,
498 unsigned long map_offset,
499 struct vmem_altmap *altmap)
501 struct mem_section *ms = __pfn_to_section(pfn);
503 if (WARN_ON_ONCE(!valid_section(ms)))
506 sparse_remove_section(ms, pfn, nr_pages, map_offset, altmap);
510 * __remove_pages() - remove sections of pages
511 * @pfn: starting pageframe (must be aligned to start of a section)
512 * @nr_pages: number of pages to remove (must be multiple of section size)
513 * @altmap: alternative device page map or %NULL if default memmap is used
515 * Generic helper function to remove section mappings and sysfs entries
516 * for the section of the memory we are removing. Caller needs to make
517 * sure that pages are marked reserved and zones are adjust properly by
518 * calling offline_pages().
520 void __remove_pages(unsigned long pfn, unsigned long nr_pages,
521 struct vmem_altmap *altmap)
523 const unsigned long end_pfn = pfn + nr_pages;
524 unsigned long cur_nr_pages;
525 unsigned long map_offset = 0;
527 map_offset = vmem_altmap_offset(altmap);
529 if (check_pfn_span(pfn, nr_pages, "remove"))
532 for (; pfn < end_pfn; pfn += cur_nr_pages) {
534 /* Select all remaining pages up to the next section boundary */
535 cur_nr_pages = min(end_pfn - pfn,
536 SECTION_ALIGN_UP(pfn + 1) - pfn);
537 __remove_section(pfn, cur_nr_pages, map_offset, altmap);
542 int set_online_page_callback(online_page_callback_t callback)
547 mutex_lock(&online_page_callback_lock);
549 if (online_page_callback == generic_online_page) {
550 online_page_callback = callback;
554 mutex_unlock(&online_page_callback_lock);
559 EXPORT_SYMBOL_GPL(set_online_page_callback);
561 int restore_online_page_callback(online_page_callback_t callback)
566 mutex_lock(&online_page_callback_lock);
568 if (online_page_callback == callback) {
569 online_page_callback = generic_online_page;
573 mutex_unlock(&online_page_callback_lock);
578 EXPORT_SYMBOL_GPL(restore_online_page_callback);
580 void generic_online_page(struct page *page, unsigned int order)
583 * Freeing the page with debug_pagealloc enabled will try to unmap it,
584 * so we should map it first. This is better than introducing a special
585 * case in page freeing fast path.
587 if (debug_pagealloc_enabled_static())
588 kernel_map_pages(page, 1 << order, 1);
589 __free_pages_core(page, order);
590 totalram_pages_add(1UL << order);
591 #ifdef CONFIG_HIGHMEM
592 if (PageHighMem(page))
593 totalhigh_pages_add(1UL << order);
596 EXPORT_SYMBOL_GPL(generic_online_page);
598 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
601 const unsigned long end_pfn = start_pfn + nr_pages;
606 * Online the pages. The callback might decide to keep some pages
607 * PG_reserved (to add them to the buddy later), but we still account
608 * them as being online/belonging to this zone ("present").
610 for (pfn = start_pfn; pfn < end_pfn; pfn += 1ul << order) {
611 order = min(MAX_ORDER - 1, get_order(PFN_PHYS(end_pfn - pfn)));
612 /* __free_pages_core() wants pfns to be aligned to the order */
613 if (WARN_ON_ONCE(!IS_ALIGNED(pfn, 1ul << order)))
615 (*online_page_callback)(pfn_to_page(pfn), order);
618 /* mark all involved sections as online */
619 online_mem_sections(start_pfn, end_pfn);
621 *(unsigned long *)arg += nr_pages;
625 /* check which state of node_states will be changed when online memory */
626 static void node_states_check_changes_online(unsigned long nr_pages,
627 struct zone *zone, struct memory_notify *arg)
629 int nid = zone_to_nid(zone);
631 arg->status_change_nid = NUMA_NO_NODE;
632 arg->status_change_nid_normal = NUMA_NO_NODE;
633 arg->status_change_nid_high = NUMA_NO_NODE;
635 if (!node_state(nid, N_MEMORY))
636 arg->status_change_nid = nid;
637 if (zone_idx(zone) <= ZONE_NORMAL && !node_state(nid, N_NORMAL_MEMORY))
638 arg->status_change_nid_normal = nid;
639 #ifdef CONFIG_HIGHMEM
640 if (zone_idx(zone) <= ZONE_HIGHMEM && !node_state(nid, N_HIGH_MEMORY))
641 arg->status_change_nid_high = nid;
645 static void node_states_set_node(int node, struct memory_notify *arg)
647 if (arg->status_change_nid_normal >= 0)
648 node_set_state(node, N_NORMAL_MEMORY);
650 if (arg->status_change_nid_high >= 0)
651 node_set_state(node, N_HIGH_MEMORY);
653 if (arg->status_change_nid >= 0)
654 node_set_state(node, N_MEMORY);
657 static void __meminit resize_zone_range(struct zone *zone, unsigned long start_pfn,
658 unsigned long nr_pages)
660 unsigned long old_end_pfn = zone_end_pfn(zone);
662 if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
663 zone->zone_start_pfn = start_pfn;
665 zone->spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - zone->zone_start_pfn;
668 static void __meminit resize_pgdat_range(struct pglist_data *pgdat, unsigned long start_pfn,
669 unsigned long nr_pages)
671 unsigned long old_end_pfn = pgdat_end_pfn(pgdat);
673 if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
674 pgdat->node_start_pfn = start_pfn;
676 pgdat->node_spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - pgdat->node_start_pfn;
680 * Associate the pfn range with the given zone, initializing the memmaps
681 * and resizing the pgdat/zone data to span the added pages. After this
682 * call, all affected pages are PG_reserved.
684 void __ref move_pfn_range_to_zone(struct zone *zone, unsigned long start_pfn,
685 unsigned long nr_pages, struct vmem_altmap *altmap)
687 struct pglist_data *pgdat = zone->zone_pgdat;
688 int nid = pgdat->node_id;
691 clear_zone_contiguous(zone);
693 /* TODO Huh pgdat is irqsave while zone is not. It used to be like that before */
694 pgdat_resize_lock(pgdat, &flags);
695 zone_span_writelock(zone);
696 if (zone_is_empty(zone))
697 init_currently_empty_zone(zone, start_pfn, nr_pages);
698 resize_zone_range(zone, start_pfn, nr_pages);
699 zone_span_writeunlock(zone);
700 resize_pgdat_range(pgdat, start_pfn, nr_pages);
701 pgdat_resize_unlock(pgdat, &flags);
704 * TODO now we have a visible range of pages which are not associated
705 * with their zone properly. Not nice but set_pfnblock_flags_mask
706 * expects the zone spans the pfn range. All the pages in the range
707 * are reserved so nobody should be touching them so we should be safe
709 memmap_init_zone(nr_pages, nid, zone_idx(zone), start_pfn,
710 MEMMAP_HOTPLUG, altmap);
712 set_zone_contiguous(zone);
716 * Returns a default kernel memory zone for the given pfn range.
717 * If no kernel zone covers this pfn range it will automatically go
718 * to the ZONE_NORMAL.
720 static struct zone *default_kernel_zone_for_pfn(int nid, unsigned long start_pfn,
721 unsigned long nr_pages)
723 struct pglist_data *pgdat = NODE_DATA(nid);
726 for (zid = 0; zid <= ZONE_NORMAL; zid++) {
727 struct zone *zone = &pgdat->node_zones[zid];
729 if (zone_intersects(zone, start_pfn, nr_pages))
733 return &pgdat->node_zones[ZONE_NORMAL];
736 static inline struct zone *default_zone_for_pfn(int nid, unsigned long start_pfn,
737 unsigned long nr_pages)
739 struct zone *kernel_zone = default_kernel_zone_for_pfn(nid, start_pfn,
741 struct zone *movable_zone = &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
742 bool in_kernel = zone_intersects(kernel_zone, start_pfn, nr_pages);
743 bool in_movable = zone_intersects(movable_zone, start_pfn, nr_pages);
746 * We inherit the existing zone in a simple case where zones do not
747 * overlap in the given range
749 if (in_kernel ^ in_movable)
750 return (in_kernel) ? kernel_zone : movable_zone;
753 * If the range doesn't belong to any zone or two zones overlap in the
754 * given range then we use movable zone only if movable_node is
755 * enabled because we always online to a kernel zone by default.
757 return movable_node_enabled ? movable_zone : kernel_zone;
760 struct zone * zone_for_pfn_range(int online_type, int nid, unsigned start_pfn,
761 unsigned long nr_pages)
763 if (online_type == MMOP_ONLINE_KERNEL)
764 return default_kernel_zone_for_pfn(nid, start_pfn, nr_pages);
766 if (online_type == MMOP_ONLINE_MOVABLE)
767 return &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
769 return default_zone_for_pfn(nid, start_pfn, nr_pages);
772 int __ref online_pages(unsigned long pfn, unsigned long nr_pages,
773 int online_type, int nid)
776 unsigned long onlined_pages = 0;
778 int need_zonelists_rebuild = 0;
780 struct memory_notify arg;
784 /* associate pfn range with the zone */
785 zone = zone_for_pfn_range(online_type, nid, pfn, nr_pages);
786 move_pfn_range_to_zone(zone, pfn, nr_pages, NULL);
789 arg.nr_pages = nr_pages;
790 node_states_check_changes_online(nr_pages, zone, &arg);
792 ret = memory_notify(MEM_GOING_ONLINE, &arg);
793 ret = notifier_to_errno(ret);
795 goto failed_addition;
798 * If this zone is not populated, then it is not in zonelist.
799 * This means the page allocator ignores this zone.
800 * So, zonelist must be updated after online.
802 if (!populated_zone(zone)) {
803 need_zonelists_rebuild = 1;
804 setup_zone_pageset(zone);
807 ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
810 /* not a single memory resource was applicable */
811 if (need_zonelists_rebuild)
812 zone_pcp_reset(zone);
813 goto failed_addition;
816 zone->present_pages += onlined_pages;
818 pgdat_resize_lock(zone->zone_pgdat, &flags);
819 zone->zone_pgdat->node_present_pages += onlined_pages;
820 pgdat_resize_unlock(zone->zone_pgdat, &flags);
824 node_states_set_node(nid, &arg);
825 if (need_zonelists_rebuild)
826 build_all_zonelists(NULL);
828 zone_pcp_update(zone);
830 init_per_zone_wmark_min();
835 vm_total_pages = nr_free_pagecache_pages();
837 writeback_set_ratelimit();
839 memory_notify(MEM_ONLINE, &arg);
844 pr_debug("online_pages [mem %#010llx-%#010llx] failed\n",
845 (unsigned long long) pfn << PAGE_SHIFT,
846 (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1);
847 memory_notify(MEM_CANCEL_ONLINE, &arg);
848 remove_pfn_range_from_zone(zone, pfn, nr_pages);
852 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
854 static void reset_node_present_pages(pg_data_t *pgdat)
858 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
859 z->present_pages = 0;
861 pgdat->node_present_pages = 0;
864 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
865 static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
867 struct pglist_data *pgdat;
868 unsigned long start_pfn = PFN_DOWN(start);
870 pgdat = NODE_DATA(nid);
872 pgdat = arch_alloc_nodedata(nid);
876 pgdat->per_cpu_nodestats =
877 alloc_percpu(struct per_cpu_nodestat);
878 arch_refresh_nodedata(nid, pgdat);
882 * Reset the nr_zones, order and highest_zoneidx before reuse.
883 * Note that kswapd will init kswapd_highest_zoneidx properly
884 * when it starts in the near future.
887 pgdat->kswapd_order = 0;
888 pgdat->kswapd_highest_zoneidx = 0;
889 for_each_online_cpu(cpu) {
890 struct per_cpu_nodestat *p;
892 p = per_cpu_ptr(pgdat->per_cpu_nodestats, cpu);
893 memset(p, 0, sizeof(*p));
897 /* we can use NODE_DATA(nid) from here */
899 pgdat->node_id = nid;
900 pgdat->node_start_pfn = start_pfn;
902 /* init node's zones as empty zones, we don't have any present pages.*/
903 free_area_init_core_hotplug(nid);
906 * The node we allocated has no zone fallback lists. For avoiding
907 * to access not-initialized zonelist, build here.
909 build_all_zonelists(pgdat);
912 * When memory is hot-added, all the memory is in offline state. So
913 * clear all zones' present_pages because they will be updated in
914 * online_pages() and offline_pages().
916 reset_node_managed_pages(pgdat);
917 reset_node_present_pages(pgdat);
922 static void rollback_node_hotadd(int nid)
924 pg_data_t *pgdat = NODE_DATA(nid);
926 arch_refresh_nodedata(nid, NULL);
927 free_percpu(pgdat->per_cpu_nodestats);
928 arch_free_nodedata(pgdat);
933 * try_online_node - online a node if offlined
935 * @start: start addr of the node
936 * @set_node_online: Whether we want to online the node
937 * called by cpu_up() to online a node without onlined memory.
940 * 1 -> a new node has been allocated
941 * 0 -> the node is already online
942 * -ENOMEM -> the node could not be allocated
944 static int __try_online_node(int nid, u64 start, bool set_node_online)
949 if (node_online(nid))
952 pgdat = hotadd_new_pgdat(nid, start);
954 pr_err("Cannot online node %d due to NULL pgdat\n", nid);
959 if (set_node_online) {
960 node_set_online(nid);
961 ret = register_one_node(nid);
969 * Users of this function always want to online/register the node
971 int try_online_node(int nid)
976 ret = __try_online_node(nid, 0, true);
981 static int check_hotplug_memory_range(u64 start, u64 size)
983 /* memory range must be block size aligned */
984 if (!size || !IS_ALIGNED(start, memory_block_size_bytes()) ||
985 !IS_ALIGNED(size, memory_block_size_bytes())) {
986 pr_err("Block size [%#lx] unaligned hotplug range: start %#llx, size %#llx",
987 memory_block_size_bytes(), start, size);
994 static int online_memory_block(struct memory_block *mem, void *arg)
996 mem->online_type = memhp_default_online_type;
997 return device_online(&mem->dev);
1001 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1002 * and online/offline operations (triggered e.g. by sysfs).
1004 * we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG
1006 int __ref add_memory_resource(int nid, struct resource *res)
1008 struct mhp_params params = { .pgprot = PAGE_KERNEL };
1010 bool new_node = false;
1014 size = resource_size(res);
1016 ret = check_hotplug_memory_range(start, size);
1020 mem_hotplug_begin();
1023 * Add new range to memblock so that when hotadd_new_pgdat() is called
1024 * to allocate new pgdat, get_pfn_range_for_nid() will be able to find
1025 * this new range and calculate total pages correctly. The range will
1026 * be removed at hot-remove time.
1028 memblock_add_node(start, size, nid);
1030 ret = __try_online_node(nid, start, false);
1035 /* call arch's memory hotadd */
1036 ret = arch_add_memory(nid, start, size, ¶ms);
1040 /* create memory block devices after memory was added */
1041 ret = create_memory_block_devices(start, size);
1043 arch_remove_memory(nid, start, size, NULL);
1048 /* If sysfs file of new node can't be created, cpu on the node
1049 * can't be hot-added. There is no rollback way now.
1050 * So, check by BUG_ON() to catch it reluctantly..
1051 * We online node here. We can't roll back from here.
1053 node_set_online(nid);
1054 ret = __register_one_node(nid);
1058 /* link memory sections under this node.*/
1059 ret = link_mem_sections(nid, PFN_DOWN(start), PFN_UP(start + size - 1));
1062 /* create new memmap entry */
1063 firmware_map_add_hotplug(start, start + size, "System RAM");
1065 /* device_online() will take the lock when calling online_pages() */
1068 /* online pages if requested */
1069 if (memhp_default_online_type != MMOP_OFFLINE)
1070 walk_memory_blocks(start, size, NULL, online_memory_block);
1074 /* rollback pgdat allocation and others */
1076 rollback_node_hotadd(nid);
1077 memblock_remove(start, size);
1082 /* requires device_hotplug_lock, see add_memory_resource() */
1083 int __ref __add_memory(int nid, u64 start, u64 size)
1085 struct resource *res;
1088 res = register_memory_resource(start, size);
1090 return PTR_ERR(res);
1092 ret = add_memory_resource(nid, res);
1094 release_memory_resource(res);
1098 int add_memory(int nid, u64 start, u64 size)
1102 lock_device_hotplug();
1103 rc = __add_memory(nid, start, size);
1104 unlock_device_hotplug();
1108 EXPORT_SYMBOL_GPL(add_memory);
1110 #ifdef CONFIG_MEMORY_HOTREMOVE
1112 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1113 * set and the size of the free page is given by page_order(). Using this,
1114 * the function determines if the pageblock contains only free pages.
1115 * Due to buddy contraints, a free page at least the size of a pageblock will
1116 * be located at the start of the pageblock
1118 static inline int pageblock_free(struct page *page)
1120 return PageBuddy(page) && page_order(page) >= pageblock_order;
1123 /* Return the pfn of the start of the next active pageblock after a given pfn */
1124 static unsigned long next_active_pageblock(unsigned long pfn)
1126 struct page *page = pfn_to_page(pfn);
1128 /* Ensure the starting page is pageblock-aligned */
1129 BUG_ON(pfn & (pageblock_nr_pages - 1));
1131 /* If the entire pageblock is free, move to the end of free page */
1132 if (pageblock_free(page)) {
1134 /* be careful. we don't have locks, page_order can be changed.*/
1135 order = page_order(page);
1136 if ((order < MAX_ORDER) && (order >= pageblock_order))
1137 return pfn + (1 << order);
1140 return pfn + pageblock_nr_pages;
1143 static bool is_pageblock_removable_nolock(unsigned long pfn)
1145 struct page *page = pfn_to_page(pfn);
1149 * We have to be careful here because we are iterating over memory
1150 * sections which are not zone aware so we might end up outside of
1151 * the zone but still within the section.
1152 * We have to take care about the node as well. If the node is offline
1153 * its NODE_DATA will be NULL - see page_zone.
1155 if (!node_online(page_to_nid(page)))
1158 zone = page_zone(page);
1159 pfn = page_to_pfn(page);
1160 if (!zone_spans_pfn(zone, pfn))
1163 return !has_unmovable_pages(zone, page, MIGRATE_MOVABLE,
1167 /* Checks if this range of memory is likely to be hot-removable. */
1168 bool is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
1170 unsigned long end_pfn, pfn;
1172 end_pfn = min(start_pfn + nr_pages,
1173 zone_end_pfn(page_zone(pfn_to_page(start_pfn))));
1175 /* Check the starting page of each pageblock within the range */
1176 for (pfn = start_pfn; pfn < end_pfn; pfn = next_active_pageblock(pfn)) {
1177 if (!is_pageblock_removable_nolock(pfn))
1182 /* All pageblocks in the memory block are likely to be hot-removable */
1187 * Confirm all pages in a range [start, end) belong to the same zone (skipping
1188 * memory holes). When true, return the zone.
1190 struct zone *test_pages_in_a_zone(unsigned long start_pfn,
1191 unsigned long end_pfn)
1193 unsigned long pfn, sec_end_pfn;
1194 struct zone *zone = NULL;
1197 for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1);
1199 pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) {
1200 /* Make sure the memory section is present first */
1201 if (!present_section_nr(pfn_to_section_nr(pfn)))
1203 for (; pfn < sec_end_pfn && pfn < end_pfn;
1204 pfn += MAX_ORDER_NR_PAGES) {
1206 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1207 while ((i < MAX_ORDER_NR_PAGES) &&
1208 !pfn_valid_within(pfn + i))
1210 if (i == MAX_ORDER_NR_PAGES || pfn + i >= end_pfn)
1212 /* Check if we got outside of the zone */
1213 if (zone && !zone_spans_pfn(zone, pfn + i))
1215 page = pfn_to_page(pfn + i);
1216 if (zone && page_zone(page) != zone)
1218 zone = page_zone(page);
1226 * Scan pfn range [start,end) to find movable/migratable pages (LRU pages,
1227 * non-lru movable pages and hugepages). We scan pfn because it's much
1228 * easier than scanning over linked list. This function returns the pfn
1229 * of the first found movable page if it's found, otherwise 0.
1231 static unsigned long scan_movable_pages(unsigned long start, unsigned long end)
1235 for (pfn = start; pfn < end; pfn++) {
1236 struct page *page, *head;
1239 if (!pfn_valid(pfn))
1241 page = pfn_to_page(pfn);
1244 if (__PageMovable(page))
1247 if (!PageHuge(page))
1249 head = compound_head(page);
1250 if (page_huge_active(head))
1252 skip = compound_nr(head) - (page - head);
1258 static struct page *new_node_page(struct page *page, unsigned long private)
1260 int nid = page_to_nid(page);
1261 nodemask_t nmask = node_states[N_MEMORY];
1264 * try to allocate from a different node but reuse this node if there
1265 * are no other online nodes to be used (e.g. we are offlining a part
1266 * of the only existing node)
1268 node_clear(nid, nmask);
1269 if (nodes_empty(nmask))
1270 node_set(nid, nmask);
1272 return new_page_nodemask(page, nid, &nmask);
1276 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1283 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1284 if (!pfn_valid(pfn))
1286 page = pfn_to_page(pfn);
1288 if (PageHuge(page)) {
1289 struct page *head = compound_head(page);
1290 pfn = page_to_pfn(head) + compound_nr(head) - 1;
1291 isolate_huge_page(head, &source);
1293 } else if (PageTransHuge(page))
1294 pfn = page_to_pfn(compound_head(page))
1295 + hpage_nr_pages(page) - 1;
1298 * HWPoison pages have elevated reference counts so the migration would
1299 * fail on them. It also doesn't make any sense to migrate them in the
1300 * first place. Still try to unmap such a page in case it is still mapped
1301 * (e.g. current hwpoison implementation doesn't unmap KSM pages but keep
1302 * the unmap as the catch all safety net).
1304 if (PageHWPoison(page)) {
1305 if (WARN_ON(PageLRU(page)))
1306 isolate_lru_page(page);
1307 if (page_mapped(page))
1308 try_to_unmap(page, TTU_IGNORE_MLOCK | TTU_IGNORE_ACCESS);
1312 if (!get_page_unless_zero(page))
1315 * We can skip free pages. And we can deal with pages on
1316 * LRU and non-lru movable pages.
1319 ret = isolate_lru_page(page);
1321 ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE);
1322 if (!ret) { /* Success */
1323 list_add_tail(&page->lru, &source);
1324 if (!__PageMovable(page))
1325 inc_node_page_state(page, NR_ISOLATED_ANON +
1326 page_is_file_lru(page));
1329 pr_warn("failed to isolate pfn %lx\n", pfn);
1330 dump_page(page, "isolation failed");
1334 if (!list_empty(&source)) {
1335 /* Allocate a new page from the nearest neighbor node */
1336 ret = migrate_pages(&source, new_node_page, NULL, 0,
1337 MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
1339 list_for_each_entry(page, &source, lru) {
1340 pr_warn("migrating pfn %lx failed ret:%d ",
1341 page_to_pfn(page), ret);
1342 dump_page(page, "migration failure");
1344 putback_movable_pages(&source);
1351 /* Mark all sections offline and remove all free pages from the buddy. */
1353 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1356 unsigned long *offlined_pages = (unsigned long *)data;
1358 *offlined_pages += __offline_isolated_pages(start, start + nr_pages);
1363 * Check all pages in range, recoreded as memory resource, are isolated.
1366 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1369 return test_pages_isolated(start_pfn, start_pfn + nr_pages,
1373 static int __init cmdline_parse_movable_node(char *p)
1375 movable_node_enabled = true;
1378 early_param("movable_node", cmdline_parse_movable_node);
1380 /* check which state of node_states will be changed when offline memory */
1381 static void node_states_check_changes_offline(unsigned long nr_pages,
1382 struct zone *zone, struct memory_notify *arg)
1384 struct pglist_data *pgdat = zone->zone_pgdat;
1385 unsigned long present_pages = 0;
1388 arg->status_change_nid = NUMA_NO_NODE;
1389 arg->status_change_nid_normal = NUMA_NO_NODE;
1390 arg->status_change_nid_high = NUMA_NO_NODE;
1393 * Check whether node_states[N_NORMAL_MEMORY] will be changed.
1394 * If the memory to be offline is within the range
1395 * [0..ZONE_NORMAL], and it is the last present memory there,
1396 * the zones in that range will become empty after the offlining,
1397 * thus we can determine that we need to clear the node from
1398 * node_states[N_NORMAL_MEMORY].
1400 for (zt = 0; zt <= ZONE_NORMAL; zt++)
1401 present_pages += pgdat->node_zones[zt].present_pages;
1402 if (zone_idx(zone) <= ZONE_NORMAL && nr_pages >= present_pages)
1403 arg->status_change_nid_normal = zone_to_nid(zone);
1405 #ifdef CONFIG_HIGHMEM
1407 * node_states[N_HIGH_MEMORY] contains nodes which
1408 * have normal memory or high memory.
1409 * Here we add the present_pages belonging to ZONE_HIGHMEM.
1410 * If the zone is within the range of [0..ZONE_HIGHMEM), and
1411 * we determine that the zones in that range become empty,
1412 * we need to clear the node for N_HIGH_MEMORY.
1414 present_pages += pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1415 if (zone_idx(zone) <= ZONE_HIGHMEM && nr_pages >= present_pages)
1416 arg->status_change_nid_high = zone_to_nid(zone);
1420 * We have accounted the pages from [0..ZONE_NORMAL), and
1421 * in case of CONFIG_HIGHMEM the pages from ZONE_HIGHMEM
1423 * Here we count the possible pages from ZONE_MOVABLE.
1424 * If after having accounted all the pages, we see that the nr_pages
1425 * to be offlined is over or equal to the accounted pages,
1426 * we know that the node will become empty, and so, we can clear
1427 * it for N_MEMORY as well.
1429 present_pages += pgdat->node_zones[ZONE_MOVABLE].present_pages;
1431 if (nr_pages >= present_pages)
1432 arg->status_change_nid = zone_to_nid(zone);
1435 static void node_states_clear_node(int node, struct memory_notify *arg)
1437 if (arg->status_change_nid_normal >= 0)
1438 node_clear_state(node, N_NORMAL_MEMORY);
1440 if (arg->status_change_nid_high >= 0)
1441 node_clear_state(node, N_HIGH_MEMORY);
1443 if (arg->status_change_nid >= 0)
1444 node_clear_state(node, N_MEMORY);
1447 static int count_system_ram_pages_cb(unsigned long start_pfn,
1448 unsigned long nr_pages, void *data)
1450 unsigned long *nr_system_ram_pages = data;
1452 *nr_system_ram_pages += nr_pages;
1456 static int __ref __offline_pages(unsigned long start_pfn,
1457 unsigned long end_pfn)
1459 unsigned long pfn, nr_pages = 0;
1460 unsigned long offlined_pages = 0;
1461 int ret, node, nr_isolate_pageblock;
1462 unsigned long flags;
1464 struct memory_notify arg;
1467 mem_hotplug_begin();
1470 * Don't allow to offline memory blocks that contain holes.
1471 * Consequently, memory blocks with holes can never get onlined
1472 * via the hotplug path - online_pages() - as hotplugged memory has
1473 * no holes. This way, we e.g., don't have to worry about marking
1474 * memory holes PG_reserved, don't need pfn_valid() checks, and can
1475 * avoid using walk_system_ram_range() later.
1477 walk_system_ram_range(start_pfn, end_pfn - start_pfn, &nr_pages,
1478 count_system_ram_pages_cb);
1479 if (nr_pages != end_pfn - start_pfn) {
1481 reason = "memory holes";
1482 goto failed_removal;
1485 /* This makes hotplug much easier...and readable.
1486 we assume this for now. .*/
1487 zone = test_pages_in_a_zone(start_pfn, end_pfn);
1490 reason = "multizone range";
1491 goto failed_removal;
1493 node = zone_to_nid(zone);
1495 /* set above range as isolated */
1496 ret = start_isolate_page_range(start_pfn, end_pfn,
1498 MEMORY_OFFLINE | REPORT_FAILURE);
1500 reason = "failure to isolate range";
1501 goto failed_removal;
1503 nr_isolate_pageblock = ret;
1505 arg.start_pfn = start_pfn;
1506 arg.nr_pages = nr_pages;
1507 node_states_check_changes_offline(nr_pages, zone, &arg);
1509 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1510 ret = notifier_to_errno(ret);
1512 reason = "notifier failure";
1513 goto failed_removal_isolated;
1517 for (pfn = start_pfn; pfn;) {
1518 if (signal_pending(current)) {
1520 reason = "signal backoff";
1521 goto failed_removal_isolated;
1525 lru_add_drain_all();
1527 pfn = scan_movable_pages(pfn, end_pfn);
1530 * TODO: fatal migration failures should bail
1533 do_migrate_range(pfn, end_pfn);
1538 * Dissolve free hugepages in the memory block before doing
1539 * offlining actually in order to make hugetlbfs's object
1540 * counting consistent.
1542 ret = dissolve_free_huge_pages(start_pfn, end_pfn);
1544 reason = "failure to dissolve huge pages";
1545 goto failed_removal_isolated;
1548 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn,
1549 NULL, check_pages_isolated_cb);
1552 /* Ok, all of our target is isolated.
1553 We cannot do rollback at this point. */
1554 walk_system_ram_range(start_pfn, end_pfn - start_pfn,
1555 &offlined_pages, offline_isolated_pages_cb);
1556 pr_info("Offlined Pages %ld\n", offlined_pages);
1558 * Onlining will reset pagetype flags and makes migrate type
1559 * MOVABLE, so just need to decrease the number of isolated
1560 * pageblocks zone counter here.
1562 spin_lock_irqsave(&zone->lock, flags);
1563 zone->nr_isolate_pageblock -= nr_isolate_pageblock;
1564 spin_unlock_irqrestore(&zone->lock, flags);
1566 /* removal success */
1567 adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages);
1568 zone->present_pages -= offlined_pages;
1570 pgdat_resize_lock(zone->zone_pgdat, &flags);
1571 zone->zone_pgdat->node_present_pages -= offlined_pages;
1572 pgdat_resize_unlock(zone->zone_pgdat, &flags);
1574 init_per_zone_wmark_min();
1576 if (!populated_zone(zone)) {
1577 zone_pcp_reset(zone);
1578 build_all_zonelists(NULL);
1580 zone_pcp_update(zone);
1582 node_states_clear_node(node, &arg);
1583 if (arg.status_change_nid >= 0) {
1585 kcompactd_stop(node);
1588 vm_total_pages = nr_free_pagecache_pages();
1589 writeback_set_ratelimit();
1591 memory_notify(MEM_OFFLINE, &arg);
1592 remove_pfn_range_from_zone(zone, start_pfn, nr_pages);
1596 failed_removal_isolated:
1597 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1598 memory_notify(MEM_CANCEL_OFFLINE, &arg);
1600 pr_debug("memory offlining [mem %#010llx-%#010llx] failed due to %s\n",
1601 (unsigned long long) start_pfn << PAGE_SHIFT,
1602 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1,
1604 /* pushback to free area */
1609 int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1611 return __offline_pages(start_pfn, start_pfn + nr_pages);
1614 static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
1616 int ret = !is_memblock_offlined(mem);
1618 if (unlikely(ret)) {
1619 phys_addr_t beginpa, endpa;
1621 beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
1622 endpa = beginpa + memory_block_size_bytes() - 1;
1623 pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n",
1631 static int check_cpu_on_node(pg_data_t *pgdat)
1635 for_each_present_cpu(cpu) {
1636 if (cpu_to_node(cpu) == pgdat->node_id)
1638 * the cpu on this node isn't removed, and we can't
1639 * offline this node.
1647 static int check_no_memblock_for_node_cb(struct memory_block *mem, void *arg)
1649 int nid = *(int *)arg;
1652 * If a memory block belongs to multiple nodes, the stored nid is not
1653 * reliable. However, such blocks are always online (e.g., cannot get
1654 * offlined) and, therefore, are still spanned by the node.
1656 return mem->nid == nid ? -EEXIST : 0;
1663 * Offline a node if all memory sections and cpus of the node are removed.
1665 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1666 * and online/offline operations before this call.
1668 void try_offline_node(int nid)
1670 pg_data_t *pgdat = NODE_DATA(nid);
1674 * If the node still spans pages (especially ZONE_DEVICE), don't
1675 * offline it. A node spans memory after move_pfn_range_to_zone(),
1676 * e.g., after the memory block was onlined.
1678 if (pgdat->node_spanned_pages)
1682 * Especially offline memory blocks might not be spanned by the
1683 * node. They will get spanned by the node once they get onlined.
1684 * However, they link to the node in sysfs and can get onlined later.
1686 rc = for_each_memory_block(&nid, check_no_memblock_for_node_cb);
1690 if (check_cpu_on_node(pgdat))
1694 * all memory/cpu of this node are removed, we can offline this
1697 node_set_offline(nid);
1698 unregister_one_node(nid);
1700 EXPORT_SYMBOL(try_offline_node);
1702 static void __release_memory_resource(resource_size_t start,
1703 resource_size_t size)
1708 * When removing memory in the same granularity as it was added,
1709 * this function never fails. It might only fail if resources
1710 * have to be adjusted or split. We'll ignore the error, as
1711 * removing of memory cannot fail.
1713 ret = release_mem_region_adjustable(&iomem_resource, start, size);
1715 resource_size_t endres = start + size - 1;
1717 pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
1718 &start, &endres, ret);
1722 static int __ref try_remove_memory(int nid, u64 start, u64 size)
1726 BUG_ON(check_hotplug_memory_range(start, size));
1729 * All memory blocks must be offlined before removing memory. Check
1730 * whether all memory blocks in question are offline and return error
1731 * if this is not the case.
1733 rc = walk_memory_blocks(start, size, NULL, check_memblock_offlined_cb);
1737 /* remove memmap entry */
1738 firmware_map_remove(start, start + size, "System RAM");
1741 * Memory block device removal under the device_hotplug_lock is
1742 * a barrier against racing online attempts.
1744 remove_memory_block_devices(start, size);
1746 mem_hotplug_begin();
1748 arch_remove_memory(nid, start, size, NULL);
1749 memblock_free(start, size);
1750 memblock_remove(start, size);
1751 __release_memory_resource(start, size);
1753 try_offline_node(nid);
1763 * @start: physical address of the region to remove
1764 * @size: size of the region to remove
1766 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1767 * and online/offline operations before this call, as required by
1768 * try_offline_node().
1770 void __remove_memory(int nid, u64 start, u64 size)
1774 * trigger BUG() if some memory is not offlined prior to calling this
1777 if (try_remove_memory(nid, start, size))
1782 * Remove memory if every memory block is offline, otherwise return -EBUSY is
1783 * some memory is not offline
1785 int remove_memory(int nid, u64 start, u64 size)
1789 lock_device_hotplug();
1790 rc = try_remove_memory(nid, start, size);
1791 unlock_device_hotplug();
1795 EXPORT_SYMBOL_GPL(remove_memory);
1796 #endif /* CONFIG_MEMORY_HOTREMOVE */