1 // SPDX-License-Identifier: GPL-2.0-only
3 * linux/mm/memory_hotplug.c
8 #include <linux/stddef.h>
10 #include <linux/sched/signal.h>
11 #include <linux/swap.h>
12 #include <linux/interrupt.h>
13 #include <linux/pagemap.h>
14 #include <linux/compiler.h>
15 #include <linux/export.h>
16 #include <linux/pagevec.h>
17 #include <linux/writeback.h>
18 #include <linux/slab.h>
19 #include <linux/sysctl.h>
20 #include <linux/cpu.h>
21 #include <linux/memory.h>
22 #include <linux/memremap.h>
23 #include <linux/memory_hotplug.h>
24 #include <linux/highmem.h>
25 #include <linux/vmalloc.h>
26 #include <linux/ioport.h>
27 #include <linux/delay.h>
28 #include <linux/migrate.h>
29 #include <linux/page-isolation.h>
30 #include <linux/pfn.h>
31 #include <linux/suspend.h>
32 #include <linux/mm_inline.h>
33 #include <linux/firmware-map.h>
34 #include <linux/stop_machine.h>
35 #include <linux/hugetlb.h>
36 #include <linux/memblock.h>
37 #include <linux/compaction.h>
38 #include <linux/rmap.h>
40 #include <asm/tlbflush.h>
46 * online_page_callback contains pointer to current page onlining function.
47 * Initially it is generic_online_page(). If it is required it could be
48 * changed by calling set_online_page_callback() for callback registration
49 * and restore_online_page_callback() for generic callback restore.
52 static void generic_online_page(struct page *page, unsigned int order);
54 static online_page_callback_t online_page_callback = generic_online_page;
55 static DEFINE_MUTEX(online_page_callback_lock);
57 DEFINE_STATIC_PERCPU_RWSEM(mem_hotplug_lock);
59 void get_online_mems(void)
61 percpu_down_read(&mem_hotplug_lock);
64 void put_online_mems(void)
66 percpu_up_read(&mem_hotplug_lock);
69 bool movable_node_enabled = false;
71 #ifndef CONFIG_MEMORY_HOTPLUG_DEFAULT_ONLINE
72 bool memhp_auto_online;
74 bool memhp_auto_online = true;
76 EXPORT_SYMBOL_GPL(memhp_auto_online);
78 static int __init setup_memhp_default_state(char *str)
80 if (!strcmp(str, "online"))
81 memhp_auto_online = true;
82 else if (!strcmp(str, "offline"))
83 memhp_auto_online = false;
87 __setup("memhp_default_state=", setup_memhp_default_state);
89 void mem_hotplug_begin(void)
92 percpu_down_write(&mem_hotplug_lock);
95 void mem_hotplug_done(void)
97 percpu_up_write(&mem_hotplug_lock);
101 u64 max_mem_size = U64_MAX;
103 /* add this memory to iomem resource */
104 static struct resource *register_memory_resource(u64 start, u64 size)
106 struct resource *res;
107 unsigned long flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
108 char *resource_name = "System RAM";
110 if (start + size > max_mem_size)
111 return ERR_PTR(-E2BIG);
114 * Request ownership of the new memory range. This might be
115 * a child of an existing resource that was present but
116 * not marked as busy.
118 res = __request_region(&iomem_resource, start, size,
119 resource_name, flags);
122 pr_debug("Unable to reserve System RAM region: %016llx->%016llx\n",
123 start, start + size);
124 return ERR_PTR(-EEXIST);
129 static void release_memory_resource(struct resource *res)
133 release_resource(res);
138 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
139 void get_page_bootmem(unsigned long info, struct page *page,
142 page->freelist = (void *)type;
143 SetPagePrivate(page);
144 set_page_private(page, info);
148 void put_page_bootmem(struct page *page)
152 type = (unsigned long) page->freelist;
153 BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE ||
154 type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE);
156 if (page_ref_dec_return(page) == 1) {
157 page->freelist = NULL;
158 ClearPagePrivate(page);
159 set_page_private(page, 0);
160 INIT_LIST_HEAD(&page->lru);
161 free_reserved_page(page);
165 #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
166 #ifndef CONFIG_SPARSEMEM_VMEMMAP
167 static void register_page_bootmem_info_section(unsigned long start_pfn)
169 unsigned long mapsize, section_nr, i;
170 struct mem_section *ms;
171 struct page *page, *memmap;
172 struct mem_section_usage *usage;
174 section_nr = pfn_to_section_nr(start_pfn);
175 ms = __nr_to_section(section_nr);
177 /* Get section's memmap address */
178 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
181 * Get page for the memmap's phys address
182 * XXX: need more consideration for sparse_vmemmap...
184 page = virt_to_page(memmap);
185 mapsize = sizeof(struct page) * PAGES_PER_SECTION;
186 mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
188 /* remember memmap's page */
189 for (i = 0; i < mapsize; i++, page++)
190 get_page_bootmem(section_nr, page, SECTION_INFO);
193 page = virt_to_page(usage);
195 mapsize = PAGE_ALIGN(mem_section_usage_size()) >> PAGE_SHIFT;
197 for (i = 0; i < mapsize; i++, page++)
198 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
201 #else /* CONFIG_SPARSEMEM_VMEMMAP */
202 static void register_page_bootmem_info_section(unsigned long start_pfn)
204 unsigned long mapsize, section_nr, i;
205 struct mem_section *ms;
206 struct page *page, *memmap;
207 struct mem_section_usage *usage;
209 section_nr = pfn_to_section_nr(start_pfn);
210 ms = __nr_to_section(section_nr);
212 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
214 register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION);
217 page = virt_to_page(usage);
219 mapsize = PAGE_ALIGN(mem_section_usage_size()) >> PAGE_SHIFT;
221 for (i = 0; i < mapsize; i++, page++)
222 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
224 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
226 void __init register_page_bootmem_info_node(struct pglist_data *pgdat)
228 unsigned long i, pfn, end_pfn, nr_pages;
229 int node = pgdat->node_id;
232 nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
233 page = virt_to_page(pgdat);
235 for (i = 0; i < nr_pages; i++, page++)
236 get_page_bootmem(node, page, NODE_INFO);
238 pfn = pgdat->node_start_pfn;
239 end_pfn = pgdat_end_pfn(pgdat);
241 /* register section info */
242 for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
244 * Some platforms can assign the same pfn to multiple nodes - on
245 * node0 as well as nodeN. To avoid registering a pfn against
246 * multiple nodes we check that this pfn does not already
247 * reside in some other nodes.
249 if (pfn_valid(pfn) && (early_pfn_to_nid(pfn) == node))
250 register_page_bootmem_info_section(pfn);
253 #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
255 static int __meminit __add_section(int nid, unsigned long phys_start_pfn,
256 struct vmem_altmap *altmap)
260 if (pfn_valid(phys_start_pfn))
263 ret = sparse_add_one_section(nid, phys_start_pfn, altmap);
264 return ret < 0 ? ret : 0;
268 * Reasonably generic function for adding memory. It is
269 * expected that archs that support memory hotplug will
270 * call this function after deciding the zone to which to
273 int __ref __add_pages(int nid, unsigned long phys_start_pfn,
274 unsigned long nr_pages, struct mhp_restrictions *restrictions)
278 int start_sec, end_sec;
279 struct vmem_altmap *altmap = restrictions->altmap;
281 /* during initialize mem_map, align hot-added range to section */
282 start_sec = pfn_to_section_nr(phys_start_pfn);
283 end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
287 * Validate altmap is within bounds of the total request
289 if (altmap->base_pfn != phys_start_pfn
290 || vmem_altmap_offset(altmap) > nr_pages) {
291 pr_warn_once("memory add fail, invalid altmap\n");
298 for (i = start_sec; i <= end_sec; i++) {
299 err = __add_section(nid, section_nr_to_pfn(i), altmap);
302 * EEXIST is finally dealt with by ioresource collision
303 * check. see add_memory() => register_memory_resource()
304 * Warning will be printed if there is collision.
306 if (err && (err != -EEXIST))
311 vmemmap_populate_print_last();
316 /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
317 static unsigned long find_smallest_section_pfn(int nid, struct zone *zone,
318 unsigned long start_pfn,
319 unsigned long end_pfn)
321 for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SUBSECTION) {
322 if (unlikely(!pfn_valid(start_pfn)))
325 if (unlikely(pfn_to_nid(start_pfn) != nid))
328 if (zone && zone != page_zone(pfn_to_page(start_pfn)))
337 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
338 static unsigned long find_biggest_section_pfn(int nid, struct zone *zone,
339 unsigned long start_pfn,
340 unsigned long end_pfn)
344 /* pfn is the end pfn of a memory section. */
346 for (; pfn >= start_pfn; pfn -= PAGES_PER_SUBSECTION) {
347 if (unlikely(!pfn_valid(pfn)))
350 if (unlikely(pfn_to_nid(pfn) != nid))
353 if (zone && zone != page_zone(pfn_to_page(pfn)))
362 static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
363 unsigned long end_pfn)
365 unsigned long zone_start_pfn = zone->zone_start_pfn;
366 unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */
367 unsigned long zone_end_pfn = z;
369 int nid = zone_to_nid(zone);
371 zone_span_writelock(zone);
372 if (zone_start_pfn == start_pfn) {
374 * If the section is smallest section in the zone, it need
375 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
376 * In this case, we find second smallest valid mem_section
377 * for shrinking zone.
379 pfn = find_smallest_section_pfn(nid, zone, end_pfn,
382 zone->zone_start_pfn = pfn;
383 zone->spanned_pages = zone_end_pfn - pfn;
385 } else if (zone_end_pfn == end_pfn) {
387 * If the section is biggest section in the zone, it need
388 * shrink zone->spanned_pages.
389 * In this case, we find second biggest valid mem_section for
392 pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn,
395 zone->spanned_pages = pfn - zone_start_pfn + 1;
399 * The section is not biggest or smallest mem_section in the zone, it
400 * only creates a hole in the zone. So in this case, we need not
401 * change the zone. But perhaps, the zone has only hole data. Thus
402 * it check the zone has only hole or not.
404 pfn = zone_start_pfn;
405 for (; pfn < zone_end_pfn; pfn += PAGES_PER_SUBSECTION) {
406 if (unlikely(!pfn_valid(pfn)))
409 if (page_zone(pfn_to_page(pfn)) != zone)
412 /* Skip range to be removed */
413 if (pfn >= start_pfn && pfn < end_pfn)
416 /* If we find valid section, we have nothing to do */
417 zone_span_writeunlock(zone);
421 /* The zone has no valid section */
422 zone->zone_start_pfn = 0;
423 zone->spanned_pages = 0;
424 zone_span_writeunlock(zone);
427 static void shrink_pgdat_span(struct pglist_data *pgdat,
428 unsigned long start_pfn, unsigned long end_pfn)
430 unsigned long pgdat_start_pfn = pgdat->node_start_pfn;
431 unsigned long p = pgdat_end_pfn(pgdat); /* pgdat_end_pfn namespace clash */
432 unsigned long pgdat_end_pfn = p;
434 int nid = pgdat->node_id;
436 if (pgdat_start_pfn == start_pfn) {
438 * If the section is smallest section in the pgdat, it need
439 * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages.
440 * In this case, we find second smallest valid mem_section
441 * for shrinking zone.
443 pfn = find_smallest_section_pfn(nid, NULL, end_pfn,
446 pgdat->node_start_pfn = pfn;
447 pgdat->node_spanned_pages = pgdat_end_pfn - pfn;
449 } else if (pgdat_end_pfn == end_pfn) {
451 * If the section is biggest section in the pgdat, it need
452 * shrink pgdat->node_spanned_pages.
453 * In this case, we find second biggest valid mem_section for
456 pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn,
459 pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1;
463 * If the section is not biggest or smallest mem_section in the pgdat,
464 * it only creates a hole in the pgdat. So in this case, we need not
466 * But perhaps, the pgdat has only hole data. Thus it check the pgdat
467 * has only hole or not.
469 pfn = pgdat_start_pfn;
470 for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SUBSECTION) {
471 if (unlikely(!pfn_valid(pfn)))
474 if (pfn_to_nid(pfn) != nid)
477 /* Skip range to be removed */
478 if (pfn >= start_pfn && pfn < end_pfn)
481 /* If we find valid section, we have nothing to do */
485 /* The pgdat has no valid section */
486 pgdat->node_start_pfn = 0;
487 pgdat->node_spanned_pages = 0;
490 static void __remove_zone(struct zone *zone, unsigned long start_pfn)
492 struct pglist_data *pgdat = zone->zone_pgdat;
493 int nr_pages = PAGES_PER_SECTION;
496 pgdat_resize_lock(zone->zone_pgdat, &flags);
497 shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
498 shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages);
499 pgdat_resize_unlock(zone->zone_pgdat, &flags);
502 static void __remove_section(struct zone *zone, struct mem_section *ms,
503 unsigned long map_offset,
504 struct vmem_altmap *altmap)
506 unsigned long start_pfn;
509 if (WARN_ON_ONCE(!valid_section(ms)))
512 scn_nr = __section_nr(ms);
513 start_pfn = section_nr_to_pfn((unsigned long)scn_nr);
514 __remove_zone(zone, start_pfn);
516 sparse_remove_one_section(ms, map_offset, altmap);
520 * __remove_pages() - remove sections of pages from a zone
521 * @zone: zone from which pages need to be removed
522 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
523 * @nr_pages: number of pages to remove (must be multiple of section size)
524 * @altmap: alternative device page map or %NULL if default memmap is used
526 * Generic helper function to remove section mappings and sysfs entries
527 * for the section of the memory we are removing. Caller needs to make
528 * sure that pages are marked reserved and zones are adjust properly by
529 * calling offline_pages().
531 void __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
532 unsigned long nr_pages, struct vmem_altmap *altmap)
535 unsigned long map_offset = 0;
536 int sections_to_remove;
538 /* In the ZONE_DEVICE case device driver owns the memory region */
539 if (is_dev_zone(zone))
540 map_offset = vmem_altmap_offset(altmap);
542 clear_zone_contiguous(zone);
545 * We can only remove entire sections
547 BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
548 BUG_ON(nr_pages % PAGES_PER_SECTION);
550 sections_to_remove = nr_pages / PAGES_PER_SECTION;
551 for (i = 0; i < sections_to_remove; i++) {
552 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
555 __remove_section(zone, __pfn_to_section(pfn), map_offset,
560 set_zone_contiguous(zone);
563 int set_online_page_callback(online_page_callback_t callback)
568 mutex_lock(&online_page_callback_lock);
570 if (online_page_callback == generic_online_page) {
571 online_page_callback = callback;
575 mutex_unlock(&online_page_callback_lock);
580 EXPORT_SYMBOL_GPL(set_online_page_callback);
582 int restore_online_page_callback(online_page_callback_t callback)
587 mutex_lock(&online_page_callback_lock);
589 if (online_page_callback == callback) {
590 online_page_callback = generic_online_page;
594 mutex_unlock(&online_page_callback_lock);
599 EXPORT_SYMBOL_GPL(restore_online_page_callback);
601 void __online_page_set_limits(struct page *page)
604 EXPORT_SYMBOL_GPL(__online_page_set_limits);
606 void __online_page_increment_counters(struct page *page)
608 adjust_managed_page_count(page, 1);
610 EXPORT_SYMBOL_GPL(__online_page_increment_counters);
612 void __online_page_free(struct page *page)
614 __free_reserved_page(page);
616 EXPORT_SYMBOL_GPL(__online_page_free);
618 static void generic_online_page(struct page *page, unsigned int order)
620 kernel_map_pages(page, 1 << order, 1);
621 __free_pages_core(page, order);
622 totalram_pages_add(1UL << order);
623 #ifdef CONFIG_HIGHMEM
624 if (PageHighMem(page))
625 totalhigh_pages_add(1UL << order);
629 static int online_pages_blocks(unsigned long start, unsigned long nr_pages)
631 unsigned long end = start + nr_pages;
632 int order, onlined_pages = 0;
634 while (start < end) {
635 order = min(MAX_ORDER - 1,
636 get_order(PFN_PHYS(end) - PFN_PHYS(start)));
637 (*online_page_callback)(pfn_to_page(start), order);
639 onlined_pages += (1UL << order);
640 start += (1UL << order);
642 return onlined_pages;
645 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
648 unsigned long onlined_pages = *(unsigned long *)arg;
650 if (PageReserved(pfn_to_page(start_pfn)))
651 onlined_pages += online_pages_blocks(start_pfn, nr_pages);
653 online_mem_sections(start_pfn, start_pfn + nr_pages);
655 *(unsigned long *)arg = onlined_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;
713 void __ref move_pfn_range_to_zone(struct zone *zone, unsigned long start_pfn,
714 unsigned long nr_pages, struct vmem_altmap *altmap)
716 struct pglist_data *pgdat = zone->zone_pgdat;
717 int nid = pgdat->node_id;
720 clear_zone_contiguous(zone);
722 /* TODO Huh pgdat is irqsave while zone is not. It used to be like that before */
723 pgdat_resize_lock(pgdat, &flags);
724 zone_span_writelock(zone);
725 if (zone_is_empty(zone))
726 init_currently_empty_zone(zone, start_pfn, nr_pages);
727 resize_zone_range(zone, start_pfn, nr_pages);
728 zone_span_writeunlock(zone);
729 resize_pgdat_range(pgdat, start_pfn, nr_pages);
730 pgdat_resize_unlock(pgdat, &flags);
733 * TODO now we have a visible range of pages which are not associated
734 * with their zone properly. Not nice but set_pfnblock_flags_mask
735 * expects the zone spans the pfn range. All the pages in the range
736 * are reserved so nobody should be touching them so we should be safe
738 memmap_init_zone(nr_pages, nid, zone_idx(zone), start_pfn,
739 MEMMAP_HOTPLUG, altmap);
741 set_zone_contiguous(zone);
745 * Returns a default kernel memory zone for the given pfn range.
746 * If no kernel zone covers this pfn range it will automatically go
747 * to the ZONE_NORMAL.
749 static struct zone *default_kernel_zone_for_pfn(int nid, unsigned long start_pfn,
750 unsigned long nr_pages)
752 struct pglist_data *pgdat = NODE_DATA(nid);
755 for (zid = 0; zid <= ZONE_NORMAL; zid++) {
756 struct zone *zone = &pgdat->node_zones[zid];
758 if (zone_intersects(zone, start_pfn, nr_pages))
762 return &pgdat->node_zones[ZONE_NORMAL];
765 static inline struct zone *default_zone_for_pfn(int nid, unsigned long start_pfn,
766 unsigned long nr_pages)
768 struct zone *kernel_zone = default_kernel_zone_for_pfn(nid, start_pfn,
770 struct zone *movable_zone = &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
771 bool in_kernel = zone_intersects(kernel_zone, start_pfn, nr_pages);
772 bool in_movable = zone_intersects(movable_zone, start_pfn, nr_pages);
775 * We inherit the existing zone in a simple case where zones do not
776 * overlap in the given range
778 if (in_kernel ^ in_movable)
779 return (in_kernel) ? kernel_zone : movable_zone;
782 * If the range doesn't belong to any zone or two zones overlap in the
783 * given range then we use movable zone only if movable_node is
784 * enabled because we always online to a kernel zone by default.
786 return movable_node_enabled ? movable_zone : kernel_zone;
789 struct zone * zone_for_pfn_range(int online_type, int nid, unsigned start_pfn,
790 unsigned long nr_pages)
792 if (online_type == MMOP_ONLINE_KERNEL)
793 return default_kernel_zone_for_pfn(nid, start_pfn, nr_pages);
795 if (online_type == MMOP_ONLINE_MOVABLE)
796 return &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
798 return default_zone_for_pfn(nid, start_pfn, nr_pages);
802 * Associates the given pfn range with the given node and the zone appropriate
803 * for the given online type.
805 static struct zone * __meminit move_pfn_range(int online_type, int nid,
806 unsigned long start_pfn, unsigned long nr_pages)
810 zone = zone_for_pfn_range(online_type, nid, start_pfn, nr_pages);
811 move_pfn_range_to_zone(zone, start_pfn, nr_pages, NULL);
815 int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type)
818 unsigned long onlined_pages = 0;
820 int need_zonelists_rebuild = 0;
823 struct memory_notify arg;
824 struct memory_block *mem;
829 * We can't use pfn_to_nid() because nid might be stored in struct page
830 * which is not yet initialized. Instead, we find nid from memory block.
832 mem = find_memory_block(__pfn_to_section(pfn));
834 put_device(&mem->dev);
836 /* associate pfn range with the zone */
837 zone = move_pfn_range(online_type, nid, pfn, nr_pages);
840 arg.nr_pages = nr_pages;
841 node_states_check_changes_online(nr_pages, zone, &arg);
843 ret = memory_notify(MEM_GOING_ONLINE, &arg);
844 ret = notifier_to_errno(ret);
846 goto failed_addition;
849 * If this zone is not populated, then it is not in zonelist.
850 * This means the page allocator ignores this zone.
851 * So, zonelist must be updated after online.
853 if (!populated_zone(zone)) {
854 need_zonelists_rebuild = 1;
855 setup_zone_pageset(zone);
858 ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
861 if (need_zonelists_rebuild)
862 zone_pcp_reset(zone);
863 goto failed_addition;
866 zone->present_pages += onlined_pages;
868 pgdat_resize_lock(zone->zone_pgdat, &flags);
869 zone->zone_pgdat->node_present_pages += onlined_pages;
870 pgdat_resize_unlock(zone->zone_pgdat, &flags);
875 node_states_set_node(nid, &arg);
876 if (need_zonelists_rebuild)
877 build_all_zonelists(NULL);
879 zone_pcp_update(zone);
882 init_per_zone_wmark_min();
889 vm_total_pages = nr_free_pagecache_pages();
891 writeback_set_ratelimit();
894 memory_notify(MEM_ONLINE, &arg);
899 pr_debug("online_pages [mem %#010llx-%#010llx] failed\n",
900 (unsigned long long) pfn << PAGE_SHIFT,
901 (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1);
902 memory_notify(MEM_CANCEL_ONLINE, &arg);
906 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
908 static void reset_node_present_pages(pg_data_t *pgdat)
912 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
913 z->present_pages = 0;
915 pgdat->node_present_pages = 0;
918 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
919 static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
921 struct pglist_data *pgdat;
922 unsigned long start_pfn = PFN_DOWN(start);
924 pgdat = NODE_DATA(nid);
926 pgdat = arch_alloc_nodedata(nid);
930 arch_refresh_nodedata(nid, pgdat);
933 * Reset the nr_zones, order and classzone_idx before reuse.
934 * Note that kswapd will init kswapd_classzone_idx properly
935 * when it starts in the near future.
938 pgdat->kswapd_order = 0;
939 pgdat->kswapd_classzone_idx = 0;
942 /* we can use NODE_DATA(nid) from here */
944 pgdat->node_id = nid;
945 pgdat->node_start_pfn = start_pfn;
947 /* init node's zones as empty zones, we don't have any present pages.*/
948 free_area_init_core_hotplug(nid);
949 pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat);
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);
980 * try_online_node - online a node if offlined
982 * @start: start addr of the node
983 * @set_node_online: Whether we want to online the node
984 * called by cpu_up() to online a node without onlined memory.
987 * 1 -> a new node has been allocated
988 * 0 -> the node is already online
989 * -ENOMEM -> the node could not be allocated
991 static int __try_online_node(int nid, u64 start, bool set_node_online)
996 if (node_online(nid))
999 pgdat = hotadd_new_pgdat(nid, start);
1001 pr_err("Cannot online node %d due to NULL pgdat\n", nid);
1006 if (set_node_online) {
1007 node_set_online(nid);
1008 ret = register_one_node(nid);
1016 * Users of this function always want to online/register the node
1018 int try_online_node(int nid)
1022 mem_hotplug_begin();
1023 ret = __try_online_node(nid, 0, true);
1028 static int check_hotplug_memory_range(u64 start, u64 size)
1030 /* memory range must be block size aligned */
1031 if (!size || !IS_ALIGNED(start, memory_block_size_bytes()) ||
1032 !IS_ALIGNED(size, memory_block_size_bytes())) {
1033 pr_err("Block size [%#lx] unaligned hotplug range: start %#llx, size %#llx",
1034 memory_block_size_bytes(), start, size);
1041 static int online_memory_block(struct memory_block *mem, void *arg)
1043 return device_online(&mem->dev);
1047 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1048 * and online/offline operations (triggered e.g. by sysfs).
1050 * we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG
1052 int __ref add_memory_resource(int nid, struct resource *res)
1054 struct mhp_restrictions restrictions = {};
1056 bool new_node = false;
1060 size = resource_size(res);
1062 ret = check_hotplug_memory_range(start, size);
1066 mem_hotplug_begin();
1069 * Add new range to memblock so that when hotadd_new_pgdat() is called
1070 * to allocate new pgdat, get_pfn_range_for_nid() will be able to find
1071 * this new range and calculate total pages correctly. The range will
1072 * be removed at hot-remove time.
1074 memblock_add_node(start, size, nid);
1076 ret = __try_online_node(nid, start, false);
1081 /* call arch's memory hotadd */
1082 ret = arch_add_memory(nid, start, size, &restrictions);
1086 /* create memory block devices after memory was added */
1087 ret = create_memory_block_devices(start, size);
1089 arch_remove_memory(nid, start, size, NULL);
1094 /* If sysfs file of new node can't be created, cpu on the node
1095 * can't be hot-added. There is no rollback way now.
1096 * So, check by BUG_ON() to catch it reluctantly..
1097 * We online node here. We can't roll back from here.
1099 node_set_online(nid);
1100 ret = __register_one_node(nid);
1104 /* link memory sections under this node.*/
1105 ret = link_mem_sections(nid, PFN_DOWN(start), PFN_UP(start + size - 1));
1108 /* create new memmap entry */
1109 firmware_map_add_hotplug(start, start + size, "System RAM");
1111 /* device_online() will take the lock when calling online_pages() */
1114 /* online pages if requested */
1115 if (memhp_auto_online)
1116 walk_memory_blocks(start, size, NULL, online_memory_block);
1120 /* rollback pgdat allocation and others */
1122 rollback_node_hotadd(nid);
1123 memblock_remove(start, size);
1128 /* requires device_hotplug_lock, see add_memory_resource() */
1129 int __ref __add_memory(int nid, u64 start, u64 size)
1131 struct resource *res;
1134 res = register_memory_resource(start, size);
1136 return PTR_ERR(res);
1138 ret = add_memory_resource(nid, res);
1140 release_memory_resource(res);
1144 int add_memory(int nid, u64 start, u64 size)
1148 lock_device_hotplug();
1149 rc = __add_memory(nid, start, size);
1150 unlock_device_hotplug();
1154 EXPORT_SYMBOL_GPL(add_memory);
1156 #ifdef CONFIG_MEMORY_HOTREMOVE
1158 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1159 * set and the size of the free page is given by page_order(). Using this,
1160 * the function determines if the pageblock contains only free pages.
1161 * Due to buddy contraints, a free page at least the size of a pageblock will
1162 * be located at the start of the pageblock
1164 static inline int pageblock_free(struct page *page)
1166 return PageBuddy(page) && page_order(page) >= pageblock_order;
1169 /* Return the pfn of the start of the next active pageblock after a given pfn */
1170 static unsigned long next_active_pageblock(unsigned long pfn)
1172 struct page *page = pfn_to_page(pfn);
1174 /* Ensure the starting page is pageblock-aligned */
1175 BUG_ON(pfn & (pageblock_nr_pages - 1));
1177 /* If the entire pageblock is free, move to the end of free page */
1178 if (pageblock_free(page)) {
1180 /* be careful. we don't have locks, page_order can be changed.*/
1181 order = page_order(page);
1182 if ((order < MAX_ORDER) && (order >= pageblock_order))
1183 return pfn + (1 << order);
1186 return pfn + pageblock_nr_pages;
1189 static bool is_pageblock_removable_nolock(unsigned long pfn)
1191 struct page *page = pfn_to_page(pfn);
1195 * We have to be careful here because we are iterating over memory
1196 * sections which are not zone aware so we might end up outside of
1197 * the zone but still within the section.
1198 * We have to take care about the node as well. If the node is offline
1199 * its NODE_DATA will be NULL - see page_zone.
1201 if (!node_online(page_to_nid(page)))
1204 zone = page_zone(page);
1205 pfn = page_to_pfn(page);
1206 if (!zone_spans_pfn(zone, pfn))
1209 return !has_unmovable_pages(zone, page, 0, MIGRATE_MOVABLE, SKIP_HWPOISON);
1212 /* Checks if this range of memory is likely to be hot-removable. */
1213 bool is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
1215 unsigned long end_pfn, pfn;
1217 end_pfn = min(start_pfn + nr_pages,
1218 zone_end_pfn(page_zone(pfn_to_page(start_pfn))));
1220 /* Check the starting page of each pageblock within the range */
1221 for (pfn = start_pfn; pfn < end_pfn; pfn = next_active_pageblock(pfn)) {
1222 if (!is_pageblock_removable_nolock(pfn))
1227 /* All pageblocks in the memory block are likely to be hot-removable */
1232 * Confirm all pages in a range [start, end) belong to the same zone.
1233 * When true, return its valid [start, end).
1235 int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn,
1236 unsigned long *valid_start, unsigned long *valid_end)
1238 unsigned long pfn, sec_end_pfn;
1239 unsigned long start, end;
1240 struct zone *zone = NULL;
1243 for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1);
1245 pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) {
1246 /* Make sure the memory section is present first */
1247 if (!present_section_nr(pfn_to_section_nr(pfn)))
1249 for (; pfn < sec_end_pfn && pfn < end_pfn;
1250 pfn += MAX_ORDER_NR_PAGES) {
1252 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1253 while ((i < MAX_ORDER_NR_PAGES) &&
1254 !pfn_valid_within(pfn + i))
1256 if (i == MAX_ORDER_NR_PAGES || pfn + i >= end_pfn)
1258 /* Check if we got outside of the zone */
1259 if (zone && !zone_spans_pfn(zone, pfn + i))
1261 page = pfn_to_page(pfn + i);
1262 if (zone && page_zone(page) != zone)
1266 zone = page_zone(page);
1267 end = pfn + MAX_ORDER_NR_PAGES;
1272 *valid_start = start;
1273 *valid_end = min(end, end_pfn);
1281 * Scan pfn range [start,end) to find movable/migratable pages (LRU pages,
1282 * non-lru movable pages and hugepages). We scan pfn because it's much
1283 * easier than scanning over linked list. This function returns the pfn
1284 * of the first found movable page if it's found, otherwise 0.
1286 static unsigned long scan_movable_pages(unsigned long start, unsigned long end)
1290 for (pfn = start; pfn < end; pfn++) {
1291 struct page *page, *head;
1294 if (!pfn_valid(pfn))
1296 page = pfn_to_page(pfn);
1299 if (__PageMovable(page))
1302 if (!PageHuge(page))
1304 head = compound_head(page);
1305 if (page_huge_active(head))
1307 skip = (1 << compound_order(head)) - (page - head);
1313 static struct page *new_node_page(struct page *page, unsigned long private)
1315 int nid = page_to_nid(page);
1316 nodemask_t nmask = node_states[N_MEMORY];
1319 * try to allocate from a different node but reuse this node if there
1320 * are no other online nodes to be used (e.g. we are offlining a part
1321 * of the only existing node)
1323 node_clear(nid, nmask);
1324 if (nodes_empty(nmask))
1325 node_set(nid, nmask);
1327 return new_page_nodemask(page, nid, &nmask);
1331 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1338 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1339 if (!pfn_valid(pfn))
1341 page = pfn_to_page(pfn);
1343 if (PageHuge(page)) {
1344 struct page *head = compound_head(page);
1345 pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1;
1346 isolate_huge_page(head, &source);
1348 } else if (PageTransHuge(page))
1349 pfn = page_to_pfn(compound_head(page))
1350 + hpage_nr_pages(page) - 1;
1353 * HWPoison pages have elevated reference counts so the migration would
1354 * fail on them. It also doesn't make any sense to migrate them in the
1355 * first place. Still try to unmap such a page in case it is still mapped
1356 * (e.g. current hwpoison implementation doesn't unmap KSM pages but keep
1357 * the unmap as the catch all safety net).
1359 if (PageHWPoison(page)) {
1360 if (WARN_ON(PageLRU(page)))
1361 isolate_lru_page(page);
1362 if (page_mapped(page))
1363 try_to_unmap(page, TTU_IGNORE_MLOCK | TTU_IGNORE_ACCESS);
1367 if (!get_page_unless_zero(page))
1370 * We can skip free pages. And we can deal with pages on
1371 * LRU and non-lru movable pages.
1374 ret = isolate_lru_page(page);
1376 ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE);
1377 if (!ret) { /* Success */
1378 list_add_tail(&page->lru, &source);
1379 if (!__PageMovable(page))
1380 inc_node_page_state(page, NR_ISOLATED_ANON +
1381 page_is_file_cache(page));
1384 pr_warn("failed to isolate pfn %lx\n", pfn);
1385 dump_page(page, "isolation failed");
1389 if (!list_empty(&source)) {
1390 /* Allocate a new page from the nearest neighbor node */
1391 ret = migrate_pages(&source, new_node_page, NULL, 0,
1392 MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
1394 list_for_each_entry(page, &source, lru) {
1395 pr_warn("migrating pfn %lx failed ret:%d ",
1396 page_to_pfn(page), ret);
1397 dump_page(page, "migration failure");
1399 putback_movable_pages(&source);
1407 * remove from free_area[] and mark all as Reserved.
1410 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1413 unsigned long *offlined_pages = (unsigned long *)data;
1415 *offlined_pages += __offline_isolated_pages(start, start + nr_pages);
1420 * Check all pages in range, recoreded as memory resource, are isolated.
1423 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1426 return test_pages_isolated(start_pfn, start_pfn + nr_pages, true);
1429 static int __init cmdline_parse_movable_node(char *p)
1431 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1432 movable_node_enabled = true;
1434 pr_warn("movable_node parameter depends on CONFIG_HAVE_MEMBLOCK_NODE_MAP to work properly\n");
1438 early_param("movable_node", cmdline_parse_movable_node);
1440 /* check which state of node_states will be changed when offline memory */
1441 static void node_states_check_changes_offline(unsigned long nr_pages,
1442 struct zone *zone, struct memory_notify *arg)
1444 struct pglist_data *pgdat = zone->zone_pgdat;
1445 unsigned long present_pages = 0;
1448 arg->status_change_nid = NUMA_NO_NODE;
1449 arg->status_change_nid_normal = NUMA_NO_NODE;
1450 arg->status_change_nid_high = NUMA_NO_NODE;
1453 * Check whether node_states[N_NORMAL_MEMORY] will be changed.
1454 * If the memory to be offline is within the range
1455 * [0..ZONE_NORMAL], and it is the last present memory there,
1456 * the zones in that range will become empty after the offlining,
1457 * thus we can determine that we need to clear the node from
1458 * node_states[N_NORMAL_MEMORY].
1460 for (zt = 0; zt <= ZONE_NORMAL; zt++)
1461 present_pages += pgdat->node_zones[zt].present_pages;
1462 if (zone_idx(zone) <= ZONE_NORMAL && nr_pages >= present_pages)
1463 arg->status_change_nid_normal = zone_to_nid(zone);
1465 #ifdef CONFIG_HIGHMEM
1467 * node_states[N_HIGH_MEMORY] contains nodes which
1468 * have normal memory or high memory.
1469 * Here we add the present_pages belonging to ZONE_HIGHMEM.
1470 * If the zone is within the range of [0..ZONE_HIGHMEM), and
1471 * we determine that the zones in that range become empty,
1472 * we need to clear the node for N_HIGH_MEMORY.
1474 present_pages += pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1475 if (zone_idx(zone) <= ZONE_HIGHMEM && nr_pages >= present_pages)
1476 arg->status_change_nid_high = zone_to_nid(zone);
1480 * We have accounted the pages from [0..ZONE_NORMAL), and
1481 * in case of CONFIG_HIGHMEM the pages from ZONE_HIGHMEM
1483 * Here we count the possible pages from ZONE_MOVABLE.
1484 * If after having accounted all the pages, we see that the nr_pages
1485 * to be offlined is over or equal to the accounted pages,
1486 * we know that the node will become empty, and so, we can clear
1487 * it for N_MEMORY as well.
1489 present_pages += pgdat->node_zones[ZONE_MOVABLE].present_pages;
1491 if (nr_pages >= present_pages)
1492 arg->status_change_nid = zone_to_nid(zone);
1495 static void node_states_clear_node(int node, struct memory_notify *arg)
1497 if (arg->status_change_nid_normal >= 0)
1498 node_clear_state(node, N_NORMAL_MEMORY);
1500 if (arg->status_change_nid_high >= 0)
1501 node_clear_state(node, N_HIGH_MEMORY);
1503 if (arg->status_change_nid >= 0)
1504 node_clear_state(node, N_MEMORY);
1507 static int __ref __offline_pages(unsigned long start_pfn,
1508 unsigned long end_pfn)
1510 unsigned long pfn, nr_pages;
1511 unsigned long offlined_pages = 0;
1512 int ret, node, nr_isolate_pageblock;
1513 unsigned long flags;
1514 unsigned long valid_start, valid_end;
1516 struct memory_notify arg;
1519 mem_hotplug_begin();
1521 /* This makes hotplug much easier...and readable.
1522 we assume this for now. .*/
1523 if (!test_pages_in_a_zone(start_pfn, end_pfn, &valid_start,
1526 reason = "multizone range";
1527 goto failed_removal;
1530 zone = page_zone(pfn_to_page(valid_start));
1531 node = zone_to_nid(zone);
1532 nr_pages = end_pfn - start_pfn;
1534 /* set above range as isolated */
1535 ret = start_isolate_page_range(start_pfn, end_pfn,
1537 SKIP_HWPOISON | REPORT_FAILURE);
1539 reason = "failure to isolate range";
1540 goto failed_removal;
1542 nr_isolate_pageblock = ret;
1544 arg.start_pfn = start_pfn;
1545 arg.nr_pages = nr_pages;
1546 node_states_check_changes_offline(nr_pages, zone, &arg);
1548 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1549 ret = notifier_to_errno(ret);
1551 reason = "notifier failure";
1552 goto failed_removal_isolated;
1556 for (pfn = start_pfn; pfn;) {
1557 if (signal_pending(current)) {
1559 reason = "signal backoff";
1560 goto failed_removal_isolated;
1564 lru_add_drain_all();
1566 pfn = scan_movable_pages(pfn, end_pfn);
1569 * TODO: fatal migration failures should bail
1572 do_migrate_range(pfn, end_pfn);
1577 * Dissolve free hugepages in the memory block before doing
1578 * offlining actually in order to make hugetlbfs's object
1579 * counting consistent.
1581 ret = dissolve_free_huge_pages(start_pfn, end_pfn);
1583 reason = "failure to dissolve huge pages";
1584 goto failed_removal_isolated;
1587 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn,
1588 NULL, check_pages_isolated_cb);
1591 /* Ok, all of our target is isolated.
1592 We cannot do rollback at this point. */
1593 walk_system_ram_range(start_pfn, end_pfn - start_pfn,
1594 &offlined_pages, offline_isolated_pages_cb);
1595 pr_info("Offlined Pages %ld\n", offlined_pages);
1597 * Onlining will reset pagetype flags and makes migrate type
1598 * MOVABLE, so just need to decrease the number of isolated
1599 * pageblocks zone counter here.
1601 spin_lock_irqsave(&zone->lock, flags);
1602 zone->nr_isolate_pageblock -= nr_isolate_pageblock;
1603 spin_unlock_irqrestore(&zone->lock, flags);
1605 /* removal success */
1606 adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages);
1607 zone->present_pages -= offlined_pages;
1609 pgdat_resize_lock(zone->zone_pgdat, &flags);
1610 zone->zone_pgdat->node_present_pages -= offlined_pages;
1611 pgdat_resize_unlock(zone->zone_pgdat, &flags);
1613 init_per_zone_wmark_min();
1615 if (!populated_zone(zone)) {
1616 zone_pcp_reset(zone);
1617 build_all_zonelists(NULL);
1619 zone_pcp_update(zone);
1621 node_states_clear_node(node, &arg);
1622 if (arg.status_change_nid >= 0) {
1624 kcompactd_stop(node);
1627 vm_total_pages = nr_free_pagecache_pages();
1628 writeback_set_ratelimit();
1630 memory_notify(MEM_OFFLINE, &arg);
1634 failed_removal_isolated:
1635 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1636 memory_notify(MEM_CANCEL_OFFLINE, &arg);
1638 pr_debug("memory offlining [mem %#010llx-%#010llx] failed due to %s\n",
1639 (unsigned long long) start_pfn << PAGE_SHIFT,
1640 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1,
1642 /* pushback to free area */
1647 int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1649 return __offline_pages(start_pfn, start_pfn + nr_pages);
1652 static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
1654 int ret = !is_memblock_offlined(mem);
1656 if (unlikely(ret)) {
1657 phys_addr_t beginpa, endpa;
1659 beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
1660 endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1;
1661 pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n",
1669 static int check_cpu_on_node(pg_data_t *pgdat)
1673 for_each_present_cpu(cpu) {
1674 if (cpu_to_node(cpu) == pgdat->node_id)
1676 * the cpu on this node isn't removed, and we can't
1677 * offline this node.
1689 * Offline a node if all memory sections and cpus of the node are removed.
1691 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1692 * and online/offline operations before this call.
1694 void try_offline_node(int nid)
1696 pg_data_t *pgdat = NODE_DATA(nid);
1697 unsigned long start_pfn = pgdat->node_start_pfn;
1698 unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
1701 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1702 unsigned long section_nr = pfn_to_section_nr(pfn);
1704 if (!present_section_nr(section_nr))
1707 if (pfn_to_nid(pfn) != nid)
1711 * some memory sections of this node are not removed, and we
1712 * can't offline node now.
1717 if (check_cpu_on_node(pgdat))
1721 * all memory/cpu of this node are removed, we can offline this
1724 node_set_offline(nid);
1725 unregister_one_node(nid);
1727 EXPORT_SYMBOL(try_offline_node);
1729 static void __release_memory_resource(resource_size_t start,
1730 resource_size_t size)
1735 * When removing memory in the same granularity as it was added,
1736 * this function never fails. It might only fail if resources
1737 * have to be adjusted or split. We'll ignore the error, as
1738 * removing of memory cannot fail.
1740 ret = release_mem_region_adjustable(&iomem_resource, start, size);
1742 resource_size_t endres = start + size - 1;
1744 pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
1745 &start, &endres, ret);
1749 static int __ref try_remove_memory(int nid, u64 start, u64 size)
1753 BUG_ON(check_hotplug_memory_range(start, size));
1755 mem_hotplug_begin();
1758 * All memory blocks must be offlined before removing memory. Check
1759 * whether all memory blocks in question are offline and return error
1760 * if this is not the case.
1762 rc = walk_memory_blocks(start, size, NULL, check_memblock_offlined_cb);
1766 /* remove memmap entry */
1767 firmware_map_remove(start, start + size, "System RAM");
1768 memblock_free(start, size);
1769 memblock_remove(start, size);
1771 /* remove memory block devices before removing memory */
1772 remove_memory_block_devices(start, size);
1774 arch_remove_memory(nid, start, size, NULL);
1775 __release_memory_resource(start, size);
1777 try_offline_node(nid);
1787 * @start: physical address of the region to remove
1788 * @size: size of the region to remove
1790 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1791 * and online/offline operations before this call, as required by
1792 * try_offline_node().
1794 void __remove_memory(int nid, u64 start, u64 size)
1798 * trigger BUG() is some memory is not offlined prior to calling this
1801 if (try_remove_memory(nid, start, size))
1806 * Remove memory if every memory block is offline, otherwise return -EBUSY is
1807 * some memory is not offline
1809 int remove_memory(int nid, u64 start, u64 size)
1813 lock_device_hotplug();
1814 rc = try_remove_memory(nid, start, size);
1815 unlock_device_hotplug();
1819 EXPORT_SYMBOL_GPL(remove_memory);
1820 #endif /* CONFIG_MEMORY_HOTREMOVE */