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 shrink_pgdat_span(struct pglist_data *pgdat,
440 unsigned long start_pfn, unsigned long end_pfn)
442 unsigned long pgdat_start_pfn = pgdat->node_start_pfn;
443 unsigned long p = pgdat_end_pfn(pgdat); /* pgdat_end_pfn namespace clash */
444 unsigned long pgdat_end_pfn = p;
446 int nid = pgdat->node_id;
448 if (pgdat_start_pfn == start_pfn) {
450 * If the section is smallest section in the pgdat, it need
451 * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages.
452 * In this case, we find second smallest valid mem_section
453 * for shrinking zone.
455 pfn = find_smallest_section_pfn(nid, NULL, end_pfn,
458 pgdat->node_start_pfn = pfn;
459 pgdat->node_spanned_pages = pgdat_end_pfn - pfn;
461 } else if (pgdat_end_pfn == end_pfn) {
463 * If the section is biggest section in the pgdat, it need
464 * shrink pgdat->node_spanned_pages.
465 * In this case, we find second biggest valid mem_section for
468 pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn,
471 pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1;
475 * If the section is not biggest or smallest mem_section in the pgdat,
476 * it only creates a hole in the pgdat. So in this case, we need not
478 * But perhaps, the pgdat has only hole data. Thus it check the pgdat
479 * has only hole or not.
481 pfn = pgdat_start_pfn;
482 for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SUBSECTION) {
483 if (unlikely(!pfn_valid(pfn)))
486 if (pfn_to_nid(pfn) != nid)
489 /* Skip range to be removed */
490 if (pfn >= start_pfn && pfn < end_pfn)
493 /* If we find valid section, we have nothing to do */
497 /* The pgdat has no valid section */
498 pgdat->node_start_pfn = 0;
499 pgdat->node_spanned_pages = 0;
502 static void __remove_zone(struct zone *zone, unsigned long start_pfn,
503 unsigned long nr_pages)
505 struct pglist_data *pgdat = zone->zone_pgdat;
508 pgdat_resize_lock(zone->zone_pgdat, &flags);
509 shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
510 shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages);
511 pgdat_resize_unlock(zone->zone_pgdat, &flags);
514 static void __remove_section(struct zone *zone, unsigned long pfn,
515 unsigned long nr_pages, unsigned long map_offset,
516 struct vmem_altmap *altmap)
518 struct mem_section *ms = __nr_to_section(pfn_to_section_nr(pfn));
520 if (WARN_ON_ONCE(!valid_section(ms)))
523 __remove_zone(zone, pfn, nr_pages);
524 sparse_remove_section(ms, pfn, nr_pages, map_offset, altmap);
528 * __remove_pages() - remove sections of pages from a zone
529 * @zone: zone from which pages need to be removed
530 * @pfn: starting pageframe (must be aligned to start of a section)
531 * @nr_pages: number of pages to remove (must be multiple of section size)
532 * @altmap: alternative device page map or %NULL if default memmap is used
534 * Generic helper function to remove section mappings and sysfs entries
535 * for the section of the memory we are removing. Caller needs to make
536 * sure that pages are marked reserved and zones are adjust properly by
537 * calling offline_pages().
539 void __remove_pages(struct zone *zone, unsigned long pfn,
540 unsigned long nr_pages, struct vmem_altmap *altmap)
542 unsigned long map_offset = 0;
543 unsigned long nr, start_sec, end_sec;
545 map_offset = vmem_altmap_offset(altmap);
547 clear_zone_contiguous(zone);
549 if (check_pfn_span(pfn, nr_pages, "remove"))
552 start_sec = pfn_to_section_nr(pfn);
553 end_sec = pfn_to_section_nr(pfn + nr_pages - 1);
554 for (nr = start_sec; nr <= end_sec; nr++) {
558 pfns = min(nr_pages, PAGES_PER_SECTION
559 - (pfn & ~PAGE_SECTION_MASK));
560 __remove_section(zone, pfn, pfns, map_offset, altmap);
566 set_zone_contiguous(zone);
569 int set_online_page_callback(online_page_callback_t callback)
574 mutex_lock(&online_page_callback_lock);
576 if (online_page_callback == generic_online_page) {
577 online_page_callback = callback;
581 mutex_unlock(&online_page_callback_lock);
586 EXPORT_SYMBOL_GPL(set_online_page_callback);
588 int restore_online_page_callback(online_page_callback_t callback)
593 mutex_lock(&online_page_callback_lock);
595 if (online_page_callback == callback) {
596 online_page_callback = generic_online_page;
600 mutex_unlock(&online_page_callback_lock);
605 EXPORT_SYMBOL_GPL(restore_online_page_callback);
607 void __online_page_set_limits(struct page *page)
610 EXPORT_SYMBOL_GPL(__online_page_set_limits);
612 void __online_page_increment_counters(struct page *page)
614 adjust_managed_page_count(page, 1);
616 EXPORT_SYMBOL_GPL(__online_page_increment_counters);
618 void __online_page_free(struct page *page)
620 __free_reserved_page(page);
622 EXPORT_SYMBOL_GPL(__online_page_free);
624 static void generic_online_page(struct page *page, unsigned int order)
626 kernel_map_pages(page, 1 << order, 1);
627 __free_pages_core(page, order);
628 totalram_pages_add(1UL << order);
629 #ifdef CONFIG_HIGHMEM
630 if (PageHighMem(page))
631 totalhigh_pages_add(1UL << order);
635 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
638 const unsigned long end_pfn = start_pfn + nr_pages;
643 * Online the pages. The callback might decide to keep some pages
644 * PG_reserved (to add them to the buddy later), but we still account
645 * them as being online/belonging to this zone ("present").
647 for (pfn = start_pfn; pfn < end_pfn; pfn += 1ul << order) {
648 order = min(MAX_ORDER - 1, get_order(PFN_PHYS(end_pfn - pfn)));
649 (*online_page_callback)(pfn_to_page(pfn), order);
652 /* mark all involved sections as online */
653 online_mem_sections(start_pfn, end_pfn);
655 *(unsigned long *)arg += nr_pages;
659 /* check which state of node_states will be changed when online memory */
660 static void node_states_check_changes_online(unsigned long nr_pages,
661 struct zone *zone, struct memory_notify *arg)
663 int nid = zone_to_nid(zone);
665 arg->status_change_nid = NUMA_NO_NODE;
666 arg->status_change_nid_normal = NUMA_NO_NODE;
667 arg->status_change_nid_high = NUMA_NO_NODE;
669 if (!node_state(nid, N_MEMORY))
670 arg->status_change_nid = nid;
671 if (zone_idx(zone) <= ZONE_NORMAL && !node_state(nid, N_NORMAL_MEMORY))
672 arg->status_change_nid_normal = nid;
673 #ifdef CONFIG_HIGHMEM
674 if (zone_idx(zone) <= ZONE_HIGHMEM && !node_state(nid, N_HIGH_MEMORY))
675 arg->status_change_nid_high = nid;
679 static void node_states_set_node(int node, struct memory_notify *arg)
681 if (arg->status_change_nid_normal >= 0)
682 node_set_state(node, N_NORMAL_MEMORY);
684 if (arg->status_change_nid_high >= 0)
685 node_set_state(node, N_HIGH_MEMORY);
687 if (arg->status_change_nid >= 0)
688 node_set_state(node, N_MEMORY);
691 static void __meminit resize_zone_range(struct zone *zone, unsigned long start_pfn,
692 unsigned long nr_pages)
694 unsigned long old_end_pfn = zone_end_pfn(zone);
696 if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
697 zone->zone_start_pfn = start_pfn;
699 zone->spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - zone->zone_start_pfn;
702 static void __meminit resize_pgdat_range(struct pglist_data *pgdat, unsigned long start_pfn,
703 unsigned long nr_pages)
705 unsigned long old_end_pfn = pgdat_end_pfn(pgdat);
707 if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
708 pgdat->node_start_pfn = start_pfn;
710 pgdat->node_spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - pgdat->node_start_pfn;
714 * Associate the pfn range with the given zone, initializing the memmaps
715 * and resizing the pgdat/zone data to span the added pages. After this
716 * call, all affected pages are PG_reserved.
718 void __ref move_pfn_range_to_zone(struct zone *zone, unsigned long start_pfn,
719 unsigned long nr_pages, struct vmem_altmap *altmap)
721 struct pglist_data *pgdat = zone->zone_pgdat;
722 int nid = pgdat->node_id;
725 clear_zone_contiguous(zone);
727 /* TODO Huh pgdat is irqsave while zone is not. It used to be like that before */
728 pgdat_resize_lock(pgdat, &flags);
729 zone_span_writelock(zone);
730 if (zone_is_empty(zone))
731 init_currently_empty_zone(zone, start_pfn, nr_pages);
732 resize_zone_range(zone, start_pfn, nr_pages);
733 zone_span_writeunlock(zone);
734 resize_pgdat_range(pgdat, start_pfn, nr_pages);
735 pgdat_resize_unlock(pgdat, &flags);
738 * TODO now we have a visible range of pages which are not associated
739 * with their zone properly. Not nice but set_pfnblock_flags_mask
740 * expects the zone spans the pfn range. All the pages in the range
741 * are reserved so nobody should be touching them so we should be safe
743 memmap_init_zone(nr_pages, nid, zone_idx(zone), start_pfn,
744 MEMMAP_HOTPLUG, altmap);
746 set_zone_contiguous(zone);
750 * Returns a default kernel memory zone for the given pfn range.
751 * If no kernel zone covers this pfn range it will automatically go
752 * to the ZONE_NORMAL.
754 static struct zone *default_kernel_zone_for_pfn(int nid, unsigned long start_pfn,
755 unsigned long nr_pages)
757 struct pglist_data *pgdat = NODE_DATA(nid);
760 for (zid = 0; zid <= ZONE_NORMAL; zid++) {
761 struct zone *zone = &pgdat->node_zones[zid];
763 if (zone_intersects(zone, start_pfn, nr_pages))
767 return &pgdat->node_zones[ZONE_NORMAL];
770 static inline struct zone *default_zone_for_pfn(int nid, unsigned long start_pfn,
771 unsigned long nr_pages)
773 struct zone *kernel_zone = default_kernel_zone_for_pfn(nid, start_pfn,
775 struct zone *movable_zone = &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
776 bool in_kernel = zone_intersects(kernel_zone, start_pfn, nr_pages);
777 bool in_movable = zone_intersects(movable_zone, start_pfn, nr_pages);
780 * We inherit the existing zone in a simple case where zones do not
781 * overlap in the given range
783 if (in_kernel ^ in_movable)
784 return (in_kernel) ? kernel_zone : movable_zone;
787 * If the range doesn't belong to any zone or two zones overlap in the
788 * given range then we use movable zone only if movable_node is
789 * enabled because we always online to a kernel zone by default.
791 return movable_node_enabled ? movable_zone : kernel_zone;
794 struct zone * zone_for_pfn_range(int online_type, int nid, unsigned start_pfn,
795 unsigned long nr_pages)
797 if (online_type == MMOP_ONLINE_KERNEL)
798 return default_kernel_zone_for_pfn(nid, start_pfn, nr_pages);
800 if (online_type == MMOP_ONLINE_MOVABLE)
801 return &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
803 return default_zone_for_pfn(nid, start_pfn, nr_pages);
806 int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type)
809 unsigned long onlined_pages = 0;
811 int need_zonelists_rebuild = 0;
814 struct memory_notify arg;
815 struct memory_block *mem;
820 * We can't use pfn_to_nid() because nid might be stored in struct page
821 * which is not yet initialized. Instead, we find nid from memory block.
823 mem = find_memory_block(__pfn_to_section(pfn));
825 put_device(&mem->dev);
827 /* associate pfn range with the zone */
828 zone = zone_for_pfn_range(online_type, nid, pfn, nr_pages);
829 move_pfn_range_to_zone(zone, pfn, nr_pages, NULL);
832 arg.nr_pages = nr_pages;
833 node_states_check_changes_online(nr_pages, zone, &arg);
835 ret = memory_notify(MEM_GOING_ONLINE, &arg);
836 ret = notifier_to_errno(ret);
838 goto failed_addition;
841 * If this zone is not populated, then it is not in zonelist.
842 * This means the page allocator ignores this zone.
843 * So, zonelist must be updated after online.
845 if (!populated_zone(zone)) {
846 need_zonelists_rebuild = 1;
847 setup_zone_pageset(zone);
850 ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
853 if (need_zonelists_rebuild)
854 zone_pcp_reset(zone);
855 goto failed_addition;
858 zone->present_pages += onlined_pages;
860 pgdat_resize_lock(zone->zone_pgdat, &flags);
861 zone->zone_pgdat->node_present_pages += onlined_pages;
862 pgdat_resize_unlock(zone->zone_pgdat, &flags);
867 node_states_set_node(nid, &arg);
868 if (need_zonelists_rebuild)
869 build_all_zonelists(NULL);
871 zone_pcp_update(zone);
874 init_per_zone_wmark_min();
881 vm_total_pages = nr_free_pagecache_pages();
883 writeback_set_ratelimit();
886 memory_notify(MEM_ONLINE, &arg);
891 pr_debug("online_pages [mem %#010llx-%#010llx] failed\n",
892 (unsigned long long) pfn << PAGE_SHIFT,
893 (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1);
894 memory_notify(MEM_CANCEL_ONLINE, &arg);
898 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
900 static void reset_node_present_pages(pg_data_t *pgdat)
904 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
905 z->present_pages = 0;
907 pgdat->node_present_pages = 0;
910 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
911 static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
913 struct pglist_data *pgdat;
914 unsigned long start_pfn = PFN_DOWN(start);
916 pgdat = NODE_DATA(nid);
918 pgdat = arch_alloc_nodedata(nid);
922 pgdat->per_cpu_nodestats =
923 alloc_percpu(struct per_cpu_nodestat);
924 arch_refresh_nodedata(nid, pgdat);
928 * Reset the nr_zones, order and classzone_idx before reuse.
929 * Note that kswapd will init kswapd_classzone_idx properly
930 * when it starts in the near future.
933 pgdat->kswapd_order = 0;
934 pgdat->kswapd_classzone_idx = 0;
935 for_each_online_cpu(cpu) {
936 struct per_cpu_nodestat *p;
938 p = per_cpu_ptr(pgdat->per_cpu_nodestats, cpu);
939 memset(p, 0, sizeof(*p));
943 /* we can use NODE_DATA(nid) from here */
945 pgdat->node_id = nid;
946 pgdat->node_start_pfn = start_pfn;
948 /* init node's zones as empty zones, we don't have any present pages.*/
949 free_area_init_core_hotplug(nid);
952 * The node we allocated has no zone fallback lists. For avoiding
953 * to access not-initialized zonelist, build here.
955 build_all_zonelists(pgdat);
958 * When memory is hot-added, all the memory is in offline state. So
959 * clear all zones' present_pages because they will be updated in
960 * online_pages() and offline_pages().
962 reset_node_managed_pages(pgdat);
963 reset_node_present_pages(pgdat);
968 static void rollback_node_hotadd(int nid)
970 pg_data_t *pgdat = NODE_DATA(nid);
972 arch_refresh_nodedata(nid, NULL);
973 free_percpu(pgdat->per_cpu_nodestats);
974 arch_free_nodedata(pgdat);
979 * try_online_node - online a node if offlined
981 * @start: start addr of the node
982 * @set_node_online: Whether we want to online the node
983 * called by cpu_up() to online a node without onlined memory.
986 * 1 -> a new node has been allocated
987 * 0 -> the node is already online
988 * -ENOMEM -> the node could not be allocated
990 static int __try_online_node(int nid, u64 start, bool set_node_online)
995 if (node_online(nid))
998 pgdat = hotadd_new_pgdat(nid, start);
1000 pr_err("Cannot online node %d due to NULL pgdat\n", nid);
1005 if (set_node_online) {
1006 node_set_online(nid);
1007 ret = register_one_node(nid);
1015 * Users of this function always want to online/register the node
1017 int try_online_node(int nid)
1021 mem_hotplug_begin();
1022 ret = __try_online_node(nid, 0, true);
1027 static int check_hotplug_memory_range(u64 start, u64 size)
1029 /* memory range must be block size aligned */
1030 if (!size || !IS_ALIGNED(start, memory_block_size_bytes()) ||
1031 !IS_ALIGNED(size, memory_block_size_bytes())) {
1032 pr_err("Block size [%#lx] unaligned hotplug range: start %#llx, size %#llx",
1033 memory_block_size_bytes(), start, size);
1040 static int online_memory_block(struct memory_block *mem, void *arg)
1042 return device_online(&mem->dev);
1046 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1047 * and online/offline operations (triggered e.g. by sysfs).
1049 * we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG
1051 int __ref add_memory_resource(int nid, struct resource *res)
1053 struct mhp_restrictions restrictions = {};
1055 bool new_node = false;
1059 size = resource_size(res);
1061 ret = check_hotplug_memory_range(start, size);
1065 mem_hotplug_begin();
1068 * Add new range to memblock so that when hotadd_new_pgdat() is called
1069 * to allocate new pgdat, get_pfn_range_for_nid() will be able to find
1070 * this new range and calculate total pages correctly. The range will
1071 * be removed at hot-remove time.
1073 memblock_add_node(start, size, nid);
1075 ret = __try_online_node(nid, start, false);
1080 /* call arch's memory hotadd */
1081 ret = arch_add_memory(nid, start, size, &restrictions);
1085 /* create memory block devices after memory was added */
1086 ret = create_memory_block_devices(start, size);
1088 arch_remove_memory(nid, start, size, NULL);
1093 /* If sysfs file of new node can't be created, cpu on the node
1094 * can't be hot-added. There is no rollback way now.
1095 * So, check by BUG_ON() to catch it reluctantly..
1096 * We online node here. We can't roll back from here.
1098 node_set_online(nid);
1099 ret = __register_one_node(nid);
1103 /* link memory sections under this node.*/
1104 ret = link_mem_sections(nid, PFN_DOWN(start), PFN_UP(start + size - 1));
1107 /* create new memmap entry */
1108 firmware_map_add_hotplug(start, start + size, "System RAM");
1110 /* device_online() will take the lock when calling online_pages() */
1113 /* online pages if requested */
1114 if (memhp_auto_online)
1115 walk_memory_blocks(start, size, NULL, online_memory_block);
1119 /* rollback pgdat allocation and others */
1121 rollback_node_hotadd(nid);
1122 memblock_remove(start, size);
1127 /* requires device_hotplug_lock, see add_memory_resource() */
1128 int __ref __add_memory(int nid, u64 start, u64 size)
1130 struct resource *res;
1133 res = register_memory_resource(start, size);
1135 return PTR_ERR(res);
1137 ret = add_memory_resource(nid, res);
1139 release_memory_resource(res);
1143 int add_memory(int nid, u64 start, u64 size)
1147 lock_device_hotplug();
1148 rc = __add_memory(nid, start, size);
1149 unlock_device_hotplug();
1153 EXPORT_SYMBOL_GPL(add_memory);
1155 #ifdef CONFIG_MEMORY_HOTREMOVE
1157 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1158 * set and the size of the free page is given by page_order(). Using this,
1159 * the function determines if the pageblock contains only free pages.
1160 * Due to buddy contraints, a free page at least the size of a pageblock will
1161 * be located at the start of the pageblock
1163 static inline int pageblock_free(struct page *page)
1165 return PageBuddy(page) && page_order(page) >= pageblock_order;
1168 /* Return the pfn of the start of the next active pageblock after a given pfn */
1169 static unsigned long next_active_pageblock(unsigned long pfn)
1171 struct page *page = pfn_to_page(pfn);
1173 /* Ensure the starting page is pageblock-aligned */
1174 BUG_ON(pfn & (pageblock_nr_pages - 1));
1176 /* If the entire pageblock is free, move to the end of free page */
1177 if (pageblock_free(page)) {
1179 /* be careful. we don't have locks, page_order can be changed.*/
1180 order = page_order(page);
1181 if ((order < MAX_ORDER) && (order >= pageblock_order))
1182 return pfn + (1 << order);
1185 return pfn + pageblock_nr_pages;
1188 static bool is_pageblock_removable_nolock(unsigned long pfn)
1190 struct page *page = pfn_to_page(pfn);
1194 * We have to be careful here because we are iterating over memory
1195 * sections which are not zone aware so we might end up outside of
1196 * the zone but still within the section.
1197 * We have to take care about the node as well. If the node is offline
1198 * its NODE_DATA will be NULL - see page_zone.
1200 if (!node_online(page_to_nid(page)))
1203 zone = page_zone(page);
1204 pfn = page_to_pfn(page);
1205 if (!zone_spans_pfn(zone, pfn))
1208 return !has_unmovable_pages(zone, page, 0, MIGRATE_MOVABLE, SKIP_HWPOISON);
1211 /* Checks if this range of memory is likely to be hot-removable. */
1212 bool is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
1214 unsigned long end_pfn, pfn;
1216 end_pfn = min(start_pfn + nr_pages,
1217 zone_end_pfn(page_zone(pfn_to_page(start_pfn))));
1219 /* Check the starting page of each pageblock within the range */
1220 for (pfn = start_pfn; pfn < end_pfn; pfn = next_active_pageblock(pfn)) {
1221 if (!is_pageblock_removable_nolock(pfn))
1226 /* All pageblocks in the memory block are likely to be hot-removable */
1231 * Confirm all pages in a range [start, end) belong to the same zone.
1232 * When true, return its valid [start, end).
1234 int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn,
1235 unsigned long *valid_start, unsigned long *valid_end)
1237 unsigned long pfn, sec_end_pfn;
1238 unsigned long start, end;
1239 struct zone *zone = NULL;
1242 for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1);
1244 pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) {
1245 /* Make sure the memory section is present first */
1246 if (!present_section_nr(pfn_to_section_nr(pfn)))
1248 for (; pfn < sec_end_pfn && pfn < end_pfn;
1249 pfn += MAX_ORDER_NR_PAGES) {
1251 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1252 while ((i < MAX_ORDER_NR_PAGES) &&
1253 !pfn_valid_within(pfn + i))
1255 if (i == MAX_ORDER_NR_PAGES || pfn + i >= end_pfn)
1257 /* Check if we got outside of the zone */
1258 if (zone && !zone_spans_pfn(zone, pfn + i))
1260 page = pfn_to_page(pfn + i);
1261 if (zone && page_zone(page) != zone)
1265 zone = page_zone(page);
1266 end = pfn + MAX_ORDER_NR_PAGES;
1271 *valid_start = start;
1272 *valid_end = min(end, end_pfn);
1280 * Scan pfn range [start,end) to find movable/migratable pages (LRU pages,
1281 * non-lru movable pages and hugepages). We scan pfn because it's much
1282 * easier than scanning over linked list. This function returns the pfn
1283 * of the first found movable page if it's found, otherwise 0.
1285 static unsigned long scan_movable_pages(unsigned long start, unsigned long end)
1289 for (pfn = start; pfn < end; pfn++) {
1290 struct page *page, *head;
1293 if (!pfn_valid(pfn))
1295 page = pfn_to_page(pfn);
1298 if (__PageMovable(page))
1301 if (!PageHuge(page))
1303 head = compound_head(page);
1304 if (page_huge_active(head))
1306 skip = compound_nr(head) - (page - head);
1312 static struct page *new_node_page(struct page *page, unsigned long private)
1314 int nid = page_to_nid(page);
1315 nodemask_t nmask = node_states[N_MEMORY];
1318 * try to allocate from a different node but reuse this node if there
1319 * are no other online nodes to be used (e.g. we are offlining a part
1320 * of the only existing node)
1322 node_clear(nid, nmask);
1323 if (nodes_empty(nmask))
1324 node_set(nid, nmask);
1326 return new_page_nodemask(page, nid, &nmask);
1330 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1337 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1338 if (!pfn_valid(pfn))
1340 page = pfn_to_page(pfn);
1342 if (PageHuge(page)) {
1343 struct page *head = compound_head(page);
1344 pfn = page_to_pfn(head) + compound_nr(head) - 1;
1345 isolate_huge_page(head, &source);
1347 } else if (PageTransHuge(page))
1348 pfn = page_to_pfn(compound_head(page))
1349 + hpage_nr_pages(page) - 1;
1352 * HWPoison pages have elevated reference counts so the migration would
1353 * fail on them. It also doesn't make any sense to migrate them in the
1354 * first place. Still try to unmap such a page in case it is still mapped
1355 * (e.g. current hwpoison implementation doesn't unmap KSM pages but keep
1356 * the unmap as the catch all safety net).
1358 if (PageHWPoison(page)) {
1359 if (WARN_ON(PageLRU(page)))
1360 isolate_lru_page(page);
1361 if (page_mapped(page))
1362 try_to_unmap(page, TTU_IGNORE_MLOCK | TTU_IGNORE_ACCESS);
1366 if (!get_page_unless_zero(page))
1369 * We can skip free pages. And we can deal with pages on
1370 * LRU and non-lru movable pages.
1373 ret = isolate_lru_page(page);
1375 ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE);
1376 if (!ret) { /* Success */
1377 list_add_tail(&page->lru, &source);
1378 if (!__PageMovable(page))
1379 inc_node_page_state(page, NR_ISOLATED_ANON +
1380 page_is_file_cache(page));
1383 pr_warn("failed to isolate pfn %lx\n", pfn);
1384 dump_page(page, "isolation failed");
1388 if (!list_empty(&source)) {
1389 /* Allocate a new page from the nearest neighbor node */
1390 ret = migrate_pages(&source, new_node_page, NULL, 0,
1391 MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
1393 list_for_each_entry(page, &source, lru) {
1394 pr_warn("migrating pfn %lx failed ret:%d ",
1395 page_to_pfn(page), ret);
1396 dump_page(page, "migration failure");
1398 putback_movable_pages(&source);
1406 * remove from free_area[] and mark all as Reserved.
1409 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1412 unsigned long *offlined_pages = (unsigned long *)data;
1414 *offlined_pages += __offline_isolated_pages(start, start + nr_pages);
1419 * Check all pages in range, recoreded as memory resource, are isolated.
1422 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1425 return test_pages_isolated(start_pfn, start_pfn + nr_pages, true);
1428 static int __init cmdline_parse_movable_node(char *p)
1430 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1431 movable_node_enabled = true;
1433 pr_warn("movable_node parameter depends on CONFIG_HAVE_MEMBLOCK_NODE_MAP to work properly\n");
1437 early_param("movable_node", cmdline_parse_movable_node);
1439 /* check which state of node_states will be changed when offline memory */
1440 static void node_states_check_changes_offline(unsigned long nr_pages,
1441 struct zone *zone, struct memory_notify *arg)
1443 struct pglist_data *pgdat = zone->zone_pgdat;
1444 unsigned long present_pages = 0;
1447 arg->status_change_nid = NUMA_NO_NODE;
1448 arg->status_change_nid_normal = NUMA_NO_NODE;
1449 arg->status_change_nid_high = NUMA_NO_NODE;
1452 * Check whether node_states[N_NORMAL_MEMORY] will be changed.
1453 * If the memory to be offline is within the range
1454 * [0..ZONE_NORMAL], and it is the last present memory there,
1455 * the zones in that range will become empty after the offlining,
1456 * thus we can determine that we need to clear the node from
1457 * node_states[N_NORMAL_MEMORY].
1459 for (zt = 0; zt <= ZONE_NORMAL; zt++)
1460 present_pages += pgdat->node_zones[zt].present_pages;
1461 if (zone_idx(zone) <= ZONE_NORMAL && nr_pages >= present_pages)
1462 arg->status_change_nid_normal = zone_to_nid(zone);
1464 #ifdef CONFIG_HIGHMEM
1466 * node_states[N_HIGH_MEMORY] contains nodes which
1467 * have normal memory or high memory.
1468 * Here we add the present_pages belonging to ZONE_HIGHMEM.
1469 * If the zone is within the range of [0..ZONE_HIGHMEM), and
1470 * we determine that the zones in that range become empty,
1471 * we need to clear the node for N_HIGH_MEMORY.
1473 present_pages += pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1474 if (zone_idx(zone) <= ZONE_HIGHMEM && nr_pages >= present_pages)
1475 arg->status_change_nid_high = zone_to_nid(zone);
1479 * We have accounted the pages from [0..ZONE_NORMAL), and
1480 * in case of CONFIG_HIGHMEM the pages from ZONE_HIGHMEM
1482 * Here we count the possible pages from ZONE_MOVABLE.
1483 * If after having accounted all the pages, we see that the nr_pages
1484 * to be offlined is over or equal to the accounted pages,
1485 * we know that the node will become empty, and so, we can clear
1486 * it for N_MEMORY as well.
1488 present_pages += pgdat->node_zones[ZONE_MOVABLE].present_pages;
1490 if (nr_pages >= present_pages)
1491 arg->status_change_nid = zone_to_nid(zone);
1494 static void node_states_clear_node(int node, struct memory_notify *arg)
1496 if (arg->status_change_nid_normal >= 0)
1497 node_clear_state(node, N_NORMAL_MEMORY);
1499 if (arg->status_change_nid_high >= 0)
1500 node_clear_state(node, N_HIGH_MEMORY);
1502 if (arg->status_change_nid >= 0)
1503 node_clear_state(node, N_MEMORY);
1506 static int __ref __offline_pages(unsigned long start_pfn,
1507 unsigned long end_pfn)
1509 unsigned long pfn, nr_pages;
1510 unsigned long offlined_pages = 0;
1511 int ret, node, nr_isolate_pageblock;
1512 unsigned long flags;
1513 unsigned long valid_start, valid_end;
1515 struct memory_notify arg;
1518 mem_hotplug_begin();
1520 /* This makes hotplug much easier...and readable.
1521 we assume this for now. .*/
1522 if (!test_pages_in_a_zone(start_pfn, end_pfn, &valid_start,
1525 reason = "multizone range";
1526 goto failed_removal;
1529 zone = page_zone(pfn_to_page(valid_start));
1530 node = zone_to_nid(zone);
1531 nr_pages = end_pfn - start_pfn;
1533 /* set above range as isolated */
1534 ret = start_isolate_page_range(start_pfn, end_pfn,
1536 SKIP_HWPOISON | REPORT_FAILURE);
1538 reason = "failure to isolate range";
1539 goto failed_removal;
1541 nr_isolate_pageblock = ret;
1543 arg.start_pfn = start_pfn;
1544 arg.nr_pages = nr_pages;
1545 node_states_check_changes_offline(nr_pages, zone, &arg);
1547 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1548 ret = notifier_to_errno(ret);
1550 reason = "notifier failure";
1551 goto failed_removal_isolated;
1555 for (pfn = start_pfn; pfn;) {
1556 if (signal_pending(current)) {
1558 reason = "signal backoff";
1559 goto failed_removal_isolated;
1563 lru_add_drain_all();
1565 pfn = scan_movable_pages(pfn, end_pfn);
1568 * TODO: fatal migration failures should bail
1571 do_migrate_range(pfn, end_pfn);
1576 * Dissolve free hugepages in the memory block before doing
1577 * offlining actually in order to make hugetlbfs's object
1578 * counting consistent.
1580 ret = dissolve_free_huge_pages(start_pfn, end_pfn);
1582 reason = "failure to dissolve huge pages";
1583 goto failed_removal_isolated;
1586 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn,
1587 NULL, check_pages_isolated_cb);
1590 /* Ok, all of our target is isolated.
1591 We cannot do rollback at this point. */
1592 walk_system_ram_range(start_pfn, end_pfn - start_pfn,
1593 &offlined_pages, offline_isolated_pages_cb);
1594 pr_info("Offlined Pages %ld\n", offlined_pages);
1596 * Onlining will reset pagetype flags and makes migrate type
1597 * MOVABLE, so just need to decrease the number of isolated
1598 * pageblocks zone counter here.
1600 spin_lock_irqsave(&zone->lock, flags);
1601 zone->nr_isolate_pageblock -= nr_isolate_pageblock;
1602 spin_unlock_irqrestore(&zone->lock, flags);
1604 /* removal success */
1605 adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages);
1606 zone->present_pages -= offlined_pages;
1608 pgdat_resize_lock(zone->zone_pgdat, &flags);
1609 zone->zone_pgdat->node_present_pages -= offlined_pages;
1610 pgdat_resize_unlock(zone->zone_pgdat, &flags);
1612 init_per_zone_wmark_min();
1614 if (!populated_zone(zone)) {
1615 zone_pcp_reset(zone);
1616 build_all_zonelists(NULL);
1618 zone_pcp_update(zone);
1620 node_states_clear_node(node, &arg);
1621 if (arg.status_change_nid >= 0) {
1623 kcompactd_stop(node);
1626 vm_total_pages = nr_free_pagecache_pages();
1627 writeback_set_ratelimit();
1629 memory_notify(MEM_OFFLINE, &arg);
1633 failed_removal_isolated:
1634 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1635 memory_notify(MEM_CANCEL_OFFLINE, &arg);
1637 pr_debug("memory offlining [mem %#010llx-%#010llx] failed due to %s\n",
1638 (unsigned long long) start_pfn << PAGE_SHIFT,
1639 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1,
1641 /* pushback to free area */
1646 int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1648 return __offline_pages(start_pfn, start_pfn + nr_pages);
1651 static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
1653 int ret = !is_memblock_offlined(mem);
1655 if (unlikely(ret)) {
1656 phys_addr_t beginpa, endpa;
1658 beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
1659 endpa = beginpa + memory_block_size_bytes() - 1;
1660 pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n",
1668 static int check_cpu_on_node(pg_data_t *pgdat)
1672 for_each_present_cpu(cpu) {
1673 if (cpu_to_node(cpu) == pgdat->node_id)
1675 * the cpu on this node isn't removed, and we can't
1676 * offline this node.
1688 * Offline a node if all memory sections and cpus of the node are removed.
1690 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1691 * and online/offline operations before this call.
1693 void try_offline_node(int nid)
1695 pg_data_t *pgdat = NODE_DATA(nid);
1696 unsigned long start_pfn = pgdat->node_start_pfn;
1697 unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
1700 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1701 unsigned long section_nr = pfn_to_section_nr(pfn);
1703 if (!present_section_nr(section_nr))
1706 if (pfn_to_nid(pfn) != nid)
1710 * some memory sections of this node are not removed, and we
1711 * can't offline node now.
1716 if (check_cpu_on_node(pgdat))
1720 * all memory/cpu of this node are removed, we can offline this
1723 node_set_offline(nid);
1724 unregister_one_node(nid);
1726 EXPORT_SYMBOL(try_offline_node);
1728 static void __release_memory_resource(resource_size_t start,
1729 resource_size_t size)
1734 * When removing memory in the same granularity as it was added,
1735 * this function never fails. It might only fail if resources
1736 * have to be adjusted or split. We'll ignore the error, as
1737 * removing of memory cannot fail.
1739 ret = release_mem_region_adjustable(&iomem_resource, start, size);
1741 resource_size_t endres = start + size - 1;
1743 pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
1744 &start, &endres, ret);
1748 static int __ref try_remove_memory(int nid, u64 start, u64 size)
1752 BUG_ON(check_hotplug_memory_range(start, size));
1754 mem_hotplug_begin();
1757 * All memory blocks must be offlined before removing memory. Check
1758 * whether all memory blocks in question are offline and return error
1759 * if this is not the case.
1761 rc = walk_memory_blocks(start, size, NULL, check_memblock_offlined_cb);
1765 /* remove memmap entry */
1766 firmware_map_remove(start, start + size, "System RAM");
1767 memblock_free(start, size);
1768 memblock_remove(start, size);
1770 /* remove memory block devices before removing memory */
1771 remove_memory_block_devices(start, size);
1773 arch_remove_memory(nid, start, size, NULL);
1774 __release_memory_resource(start, size);
1776 try_offline_node(nid);
1786 * @start: physical address of the region to remove
1787 * @size: size of the region to remove
1789 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1790 * and online/offline operations before this call, as required by
1791 * try_offline_node().
1793 void __remove_memory(int nid, u64 start, u64 size)
1797 * trigger BUG() is some memory is not offlined prior to calling this
1800 if (try_remove_memory(nid, start, size))
1805 * Remove memory if every memory block is offline, otherwise return -EBUSY is
1806 * some memory is not offline
1808 int remove_memory(int nid, u64 start, u64 size)
1812 lock_device_hotplug();
1813 rc = try_remove_memory(nid, start, size);
1814 unlock_device_hotplug();
1818 EXPORT_SYMBOL_GPL(remove_memory);
1819 #endif /* CONFIG_MEMORY_HOTREMOVE */