2 * linux/mm/memory_hotplug.c
7 #include <linux/stddef.h>
9 #include <linux/sched/signal.h>
10 #include <linux/swap.h>
11 #include <linux/interrupt.h>
12 #include <linux/pagemap.h>
13 #include <linux/compiler.h>
14 #include <linux/export.h>
15 #include <linux/pagevec.h>
16 #include <linux/writeback.h>
17 #include <linux/slab.h>
18 #include <linux/sysctl.h>
19 #include <linux/cpu.h>
20 #include <linux/memory.h>
21 #include <linux/memremap.h>
22 #include <linux/memory_hotplug.h>
23 #include <linux/highmem.h>
24 #include <linux/vmalloc.h>
25 #include <linux/ioport.h>
26 #include <linux/delay.h>
27 #include <linux/migrate.h>
28 #include <linux/page-isolation.h>
29 #include <linux/pfn.h>
30 #include <linux/suspend.h>
31 #include <linux/mm_inline.h>
32 #include <linux/firmware-map.h>
33 #include <linux/stop_machine.h>
34 #include <linux/hugetlb.h>
35 #include <linux/memblock.h>
36 #include <linux/compaction.h>
37 #include <linux/rmap.h>
39 #include <asm/tlbflush.h>
44 * online_page_callback contains pointer to current page onlining function.
45 * Initially it is generic_online_page(). If it is required it could be
46 * changed by calling set_online_page_callback() for callback registration
47 * and restore_online_page_callback() for generic callback restore.
50 static void generic_online_page(struct page *page);
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 bool memhp_auto_online;
72 bool memhp_auto_online = true;
74 EXPORT_SYMBOL_GPL(memhp_auto_online);
76 static int __init setup_memhp_default_state(char *str)
78 if (!strcmp(str, "online"))
79 memhp_auto_online = true;
80 else if (!strcmp(str, "offline"))
81 memhp_auto_online = false;
85 __setup("memhp_default_state=", setup_memhp_default_state);
87 void mem_hotplug_begin(void)
90 percpu_down_write(&mem_hotplug_lock);
93 void mem_hotplug_done(void)
95 percpu_up_write(&mem_hotplug_lock);
99 /* add this memory to iomem resource */
100 static struct resource *register_memory_resource(u64 start, u64 size)
102 struct resource *res;
103 unsigned long flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
104 char *resource_name = "System RAM";
107 * Request ownership of the new memory range. This might be
108 * a child of an existing resource that was present but
109 * not marked as busy.
111 res = __request_region(&iomem_resource, start, size,
112 resource_name, flags);
115 pr_debug("Unable to reserve System RAM region: %016llx->%016llx\n",
116 start, start + size);
117 return ERR_PTR(-EEXIST);
122 static void release_memory_resource(struct resource *res)
126 release_resource(res);
131 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
132 void get_page_bootmem(unsigned long info, struct page *page,
135 page->freelist = (void *)type;
136 SetPagePrivate(page);
137 set_page_private(page, info);
141 void put_page_bootmem(struct page *page)
145 type = (unsigned long) page->freelist;
146 BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE ||
147 type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE);
149 if (page_ref_dec_return(page) == 1) {
150 page->freelist = NULL;
151 ClearPagePrivate(page);
152 set_page_private(page, 0);
153 INIT_LIST_HEAD(&page->lru);
154 free_reserved_page(page);
158 #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
159 #ifndef CONFIG_SPARSEMEM_VMEMMAP
160 static void register_page_bootmem_info_section(unsigned long start_pfn)
162 unsigned long *usemap, mapsize, section_nr, i;
163 struct mem_section *ms;
164 struct page *page, *memmap;
166 section_nr = pfn_to_section_nr(start_pfn);
167 ms = __nr_to_section(section_nr);
169 /* Get section's memmap address */
170 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
173 * Get page for the memmap's phys address
174 * XXX: need more consideration for sparse_vmemmap...
176 page = virt_to_page(memmap);
177 mapsize = sizeof(struct page) * PAGES_PER_SECTION;
178 mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
180 /* remember memmap's page */
181 for (i = 0; i < mapsize; i++, page++)
182 get_page_bootmem(section_nr, page, SECTION_INFO);
184 usemap = ms->pageblock_flags;
185 page = virt_to_page(usemap);
187 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
189 for (i = 0; i < mapsize; i++, page++)
190 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
193 #else /* CONFIG_SPARSEMEM_VMEMMAP */
194 static void register_page_bootmem_info_section(unsigned long start_pfn)
196 unsigned long *usemap, mapsize, section_nr, i;
197 struct mem_section *ms;
198 struct page *page, *memmap;
200 section_nr = pfn_to_section_nr(start_pfn);
201 ms = __nr_to_section(section_nr);
203 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
205 register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION);
207 usemap = ms->pageblock_flags;
208 page = virt_to_page(usemap);
210 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
212 for (i = 0; i < mapsize; i++, page++)
213 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
215 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
217 void __init register_page_bootmem_info_node(struct pglist_data *pgdat)
219 unsigned long i, pfn, end_pfn, nr_pages;
220 int node = pgdat->node_id;
223 nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
224 page = virt_to_page(pgdat);
226 for (i = 0; i < nr_pages; i++, page++)
227 get_page_bootmem(node, page, NODE_INFO);
229 pfn = pgdat->node_start_pfn;
230 end_pfn = pgdat_end_pfn(pgdat);
232 /* register section info */
233 for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
235 * Some platforms can assign the same pfn to multiple nodes - on
236 * node0 as well as nodeN. To avoid registering a pfn against
237 * multiple nodes we check that this pfn does not already
238 * reside in some other nodes.
240 if (pfn_valid(pfn) && (early_pfn_to_nid(pfn) == node))
241 register_page_bootmem_info_section(pfn);
244 #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
246 static int __meminit __add_section(int nid, unsigned long phys_start_pfn,
247 struct vmem_altmap *altmap, bool want_memblock)
251 if (pfn_valid(phys_start_pfn))
254 ret = sparse_add_one_section(nid, phys_start_pfn, altmap);
261 return hotplug_memory_register(nid, __pfn_to_section(phys_start_pfn));
265 * Reasonably generic function for adding memory. It is
266 * expected that archs that support memory hotplug will
267 * call this function after deciding the zone to which to
270 int __ref __add_pages(int nid, unsigned long phys_start_pfn,
271 unsigned long nr_pages, struct vmem_altmap *altmap,
276 int start_sec, end_sec;
278 /* during initialize mem_map, align hot-added range to section */
279 start_sec = pfn_to_section_nr(phys_start_pfn);
280 end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
284 * Validate altmap is within bounds of the total request
286 if (altmap->base_pfn != phys_start_pfn
287 || vmem_altmap_offset(altmap) > nr_pages) {
288 pr_warn_once("memory add fail, invalid altmap\n");
295 for (i = start_sec; i <= end_sec; i++) {
296 err = __add_section(nid, section_nr_to_pfn(i), altmap,
300 * EEXIST is finally dealt with by ioresource collision
301 * check. see add_memory() => register_memory_resource()
302 * Warning will be printed if there is collision.
304 if (err && (err != -EEXIST))
309 vmemmap_populate_print_last();
314 #ifdef CONFIG_MEMORY_HOTREMOVE
315 /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
316 static unsigned long find_smallest_section_pfn(int nid, struct zone *zone,
317 unsigned long start_pfn,
318 unsigned long end_pfn)
320 struct mem_section *ms;
322 for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) {
323 ms = __pfn_to_section(start_pfn);
325 if (unlikely(!valid_section(ms)))
328 if (unlikely(pfn_to_nid(start_pfn) != nid))
331 if (zone && zone != page_zone(pfn_to_page(start_pfn)))
340 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
341 static unsigned long find_biggest_section_pfn(int nid, struct zone *zone,
342 unsigned long start_pfn,
343 unsigned long end_pfn)
345 struct mem_section *ms;
348 /* pfn is the end pfn of a memory section. */
350 for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) {
351 ms = __pfn_to_section(pfn);
353 if (unlikely(!valid_section(ms)))
356 if (unlikely(pfn_to_nid(pfn) != nid))
359 if (zone && zone != page_zone(pfn_to_page(pfn)))
368 static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
369 unsigned long end_pfn)
371 unsigned long zone_start_pfn = zone->zone_start_pfn;
372 unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */
373 unsigned long zone_end_pfn = z;
375 struct mem_section *ms;
376 int nid = zone_to_nid(zone);
378 zone_span_writelock(zone);
379 if (zone_start_pfn == start_pfn) {
381 * If the section is smallest section in the zone, it need
382 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
383 * In this case, we find second smallest valid mem_section
384 * for shrinking zone.
386 pfn = find_smallest_section_pfn(nid, zone, end_pfn,
389 zone->zone_start_pfn = pfn;
390 zone->spanned_pages = zone_end_pfn - pfn;
392 } else if (zone_end_pfn == end_pfn) {
394 * If the section is biggest section in the zone, it need
395 * shrink zone->spanned_pages.
396 * In this case, we find second biggest valid mem_section for
399 pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn,
402 zone->spanned_pages = pfn - zone_start_pfn + 1;
406 * The section is not biggest or smallest mem_section in the zone, it
407 * only creates a hole in the zone. So in this case, we need not
408 * change the zone. But perhaps, the zone has only hole data. Thus
409 * it check the zone has only hole or not.
411 pfn = zone_start_pfn;
412 for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) {
413 ms = __pfn_to_section(pfn);
415 if (unlikely(!valid_section(ms)))
418 if (page_zone(pfn_to_page(pfn)) != zone)
421 /* If the section is current section, it continues the loop */
422 if (start_pfn == pfn)
425 /* If we find valid section, we have nothing to do */
426 zone_span_writeunlock(zone);
430 /* The zone has no valid section */
431 zone->zone_start_pfn = 0;
432 zone->spanned_pages = 0;
433 zone_span_writeunlock(zone);
436 static void shrink_pgdat_span(struct pglist_data *pgdat,
437 unsigned long start_pfn, unsigned long end_pfn)
439 unsigned long pgdat_start_pfn = pgdat->node_start_pfn;
440 unsigned long p = pgdat_end_pfn(pgdat); /* pgdat_end_pfn namespace clash */
441 unsigned long pgdat_end_pfn = p;
443 struct mem_section *ms;
444 int nid = pgdat->node_id;
446 if (pgdat_start_pfn == start_pfn) {
448 * If the section is smallest section in the pgdat, it need
449 * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages.
450 * In this case, we find second smallest valid mem_section
451 * for shrinking zone.
453 pfn = find_smallest_section_pfn(nid, NULL, end_pfn,
456 pgdat->node_start_pfn = pfn;
457 pgdat->node_spanned_pages = pgdat_end_pfn - pfn;
459 } else if (pgdat_end_pfn == end_pfn) {
461 * If the section is biggest section in the pgdat, it need
462 * shrink pgdat->node_spanned_pages.
463 * In this case, we find second biggest valid mem_section for
466 pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn,
469 pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1;
473 * If the section is not biggest or smallest mem_section in the pgdat,
474 * it only creates a hole in the pgdat. So in this case, we need not
476 * But perhaps, the pgdat has only hole data. Thus it check the pgdat
477 * has only hole or not.
479 pfn = pgdat_start_pfn;
480 for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) {
481 ms = __pfn_to_section(pfn);
483 if (unlikely(!valid_section(ms)))
486 if (pfn_to_nid(pfn) != nid)
489 /* If the section is current section, it continues the loop */
490 if (start_pfn == 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)
504 struct pglist_data *pgdat = zone->zone_pgdat;
505 int nr_pages = PAGES_PER_SECTION;
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 int __remove_section(struct zone *zone, struct mem_section *ms,
515 unsigned long map_offset, struct vmem_altmap *altmap)
517 unsigned long start_pfn;
521 if (!valid_section(ms))
524 ret = unregister_memory_section(ms);
528 scn_nr = __section_nr(ms);
529 start_pfn = section_nr_to_pfn((unsigned long)scn_nr);
530 __remove_zone(zone, start_pfn);
532 sparse_remove_one_section(zone, ms, map_offset, altmap);
537 * __remove_pages() - remove sections of pages from a zone
538 * @zone: zone from which pages need to be removed
539 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
540 * @nr_pages: number of pages to remove (must be multiple of section size)
541 * @altmap: alternative device page map or %NULL if default memmap is used
543 * Generic helper function to remove section mappings and sysfs entries
544 * for the section of the memory we are removing. Caller needs to make
545 * sure that pages are marked reserved and zones are adjust properly by
546 * calling offline_pages().
548 int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
549 unsigned long nr_pages, struct vmem_altmap *altmap)
552 unsigned long map_offset = 0;
553 int sections_to_remove, ret = 0;
555 /* In the ZONE_DEVICE case device driver owns the memory region */
556 if (is_dev_zone(zone)) {
558 map_offset = vmem_altmap_offset(altmap);
560 resource_size_t start, size;
562 start = phys_start_pfn << PAGE_SHIFT;
563 size = nr_pages * PAGE_SIZE;
565 ret = release_mem_region_adjustable(&iomem_resource, start,
568 resource_size_t endres = start + size - 1;
570 pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
571 &start, &endres, ret);
575 clear_zone_contiguous(zone);
578 * We can only remove entire sections
580 BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
581 BUG_ON(nr_pages % PAGES_PER_SECTION);
583 sections_to_remove = nr_pages / PAGES_PER_SECTION;
584 for (i = 0; i < sections_to_remove; i++) {
585 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
588 ret = __remove_section(zone, __pfn_to_section(pfn), map_offset,
595 set_zone_contiguous(zone);
599 #endif /* CONFIG_MEMORY_HOTREMOVE */
601 int set_online_page_callback(online_page_callback_t callback)
606 mutex_lock(&online_page_callback_lock);
608 if (online_page_callback == generic_online_page) {
609 online_page_callback = callback;
613 mutex_unlock(&online_page_callback_lock);
618 EXPORT_SYMBOL_GPL(set_online_page_callback);
620 int restore_online_page_callback(online_page_callback_t callback)
625 mutex_lock(&online_page_callback_lock);
627 if (online_page_callback == callback) {
628 online_page_callback = generic_online_page;
632 mutex_unlock(&online_page_callback_lock);
637 EXPORT_SYMBOL_GPL(restore_online_page_callback);
639 void __online_page_set_limits(struct page *page)
642 EXPORT_SYMBOL_GPL(__online_page_set_limits);
644 void __online_page_increment_counters(struct page *page)
646 adjust_managed_page_count(page, 1);
648 EXPORT_SYMBOL_GPL(__online_page_increment_counters);
650 void __online_page_free(struct page *page)
652 __free_reserved_page(page);
654 EXPORT_SYMBOL_GPL(__online_page_free);
656 static void generic_online_page(struct page *page)
658 __online_page_set_limits(page);
659 __online_page_increment_counters(page);
660 __online_page_free(page);
663 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
667 unsigned long onlined_pages = *(unsigned long *)arg;
670 if (PageReserved(pfn_to_page(start_pfn)))
671 for (i = 0; i < nr_pages; i++) {
672 page = pfn_to_page(start_pfn + i);
673 (*online_page_callback)(page);
677 online_mem_sections(start_pfn, start_pfn + nr_pages);
679 *(unsigned long *)arg = onlined_pages;
683 /* check which state of node_states will be changed when online memory */
684 static void node_states_check_changes_online(unsigned long nr_pages,
685 struct zone *zone, struct memory_notify *arg)
687 int nid = zone_to_nid(zone);
689 arg->status_change_nid = -1;
690 arg->status_change_nid_normal = -1;
691 arg->status_change_nid_high = -1;
693 if (!node_state(nid, N_MEMORY))
694 arg->status_change_nid = nid;
695 if (zone_idx(zone) <= ZONE_NORMAL && !node_state(nid, N_NORMAL_MEMORY))
696 arg->status_change_nid_normal = nid;
697 #ifdef CONFIG_HIGHMEM
698 if (zone_idx(zone) <= N_HIGH_MEMORY && !node_state(nid, N_HIGH_MEMORY))
699 arg->status_change_nid_high = nid;
703 static void node_states_set_node(int node, struct memory_notify *arg)
705 if (arg->status_change_nid_normal >= 0)
706 node_set_state(node, N_NORMAL_MEMORY);
708 if (arg->status_change_nid_high >= 0)
709 node_set_state(node, N_HIGH_MEMORY);
711 if (arg->status_change_nid >= 0)
712 node_set_state(node, N_MEMORY);
715 static void __meminit resize_zone_range(struct zone *zone, unsigned long start_pfn,
716 unsigned long nr_pages)
718 unsigned long old_end_pfn = zone_end_pfn(zone);
720 if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
721 zone->zone_start_pfn = start_pfn;
723 zone->spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - zone->zone_start_pfn;
726 static void __meminit resize_pgdat_range(struct pglist_data *pgdat, unsigned long start_pfn,
727 unsigned long nr_pages)
729 unsigned long old_end_pfn = pgdat_end_pfn(pgdat);
731 if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
732 pgdat->node_start_pfn = start_pfn;
734 pgdat->node_spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - pgdat->node_start_pfn;
737 void __ref move_pfn_range_to_zone(struct zone *zone, unsigned long start_pfn,
738 unsigned long nr_pages, struct vmem_altmap *altmap)
740 struct pglist_data *pgdat = zone->zone_pgdat;
741 int nid = pgdat->node_id;
744 clear_zone_contiguous(zone);
746 /* TODO Huh pgdat is irqsave while zone is not. It used to be like that before */
747 pgdat_resize_lock(pgdat, &flags);
748 zone_span_writelock(zone);
749 if (zone_is_empty(zone))
750 init_currently_empty_zone(zone, start_pfn, nr_pages);
751 resize_zone_range(zone, start_pfn, nr_pages);
752 zone_span_writeunlock(zone);
753 resize_pgdat_range(pgdat, start_pfn, nr_pages);
754 pgdat_resize_unlock(pgdat, &flags);
757 * TODO now we have a visible range of pages which are not associated
758 * with their zone properly. Not nice but set_pfnblock_flags_mask
759 * expects the zone spans the pfn range. All the pages in the range
760 * are reserved so nobody should be touching them so we should be safe
762 memmap_init_zone(nr_pages, nid, zone_idx(zone), start_pfn,
763 MEMMAP_HOTPLUG, altmap);
765 set_zone_contiguous(zone);
769 * Returns a default kernel memory zone for the given pfn range.
770 * If no kernel zone covers this pfn range it will automatically go
771 * to the ZONE_NORMAL.
773 static struct zone *default_kernel_zone_for_pfn(int nid, unsigned long start_pfn,
774 unsigned long nr_pages)
776 struct pglist_data *pgdat = NODE_DATA(nid);
779 for (zid = 0; zid <= ZONE_NORMAL; zid++) {
780 struct zone *zone = &pgdat->node_zones[zid];
782 if (zone_intersects(zone, start_pfn, nr_pages))
786 return &pgdat->node_zones[ZONE_NORMAL];
789 static inline struct zone *default_zone_for_pfn(int nid, unsigned long start_pfn,
790 unsigned long nr_pages)
792 struct zone *kernel_zone = default_kernel_zone_for_pfn(nid, start_pfn,
794 struct zone *movable_zone = &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
795 bool in_kernel = zone_intersects(kernel_zone, start_pfn, nr_pages);
796 bool in_movable = zone_intersects(movable_zone, start_pfn, nr_pages);
799 * We inherit the existing zone in a simple case where zones do not
800 * overlap in the given range
802 if (in_kernel ^ in_movable)
803 return (in_kernel) ? kernel_zone : movable_zone;
806 * If the range doesn't belong to any zone or two zones overlap in the
807 * given range then we use movable zone only if movable_node is
808 * enabled because we always online to a kernel zone by default.
810 return movable_node_enabled ? movable_zone : kernel_zone;
813 struct zone * zone_for_pfn_range(int online_type, int nid, unsigned start_pfn,
814 unsigned long nr_pages)
816 if (online_type == MMOP_ONLINE_KERNEL)
817 return default_kernel_zone_for_pfn(nid, start_pfn, nr_pages);
819 if (online_type == MMOP_ONLINE_MOVABLE)
820 return &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
822 return default_zone_for_pfn(nid, start_pfn, nr_pages);
826 * Associates the given pfn range with the given node and the zone appropriate
827 * for the given online type.
829 static struct zone * __meminit move_pfn_range(int online_type, int nid,
830 unsigned long start_pfn, unsigned long nr_pages)
834 zone = zone_for_pfn_range(online_type, nid, start_pfn, nr_pages);
835 move_pfn_range_to_zone(zone, start_pfn, nr_pages, NULL);
839 int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type)
842 unsigned long onlined_pages = 0;
844 int need_zonelists_rebuild = 0;
847 struct memory_notify arg;
848 struct memory_block *mem;
853 * We can't use pfn_to_nid() because nid might be stored in struct page
854 * which is not yet initialized. Instead, we find nid from memory block.
856 mem = find_memory_block(__pfn_to_section(pfn));
859 /* associate pfn range with the zone */
860 zone = move_pfn_range(online_type, nid, pfn, nr_pages);
863 arg.nr_pages = nr_pages;
864 node_states_check_changes_online(nr_pages, zone, &arg);
866 ret = memory_notify(MEM_GOING_ONLINE, &arg);
867 ret = notifier_to_errno(ret);
869 goto failed_addition;
872 * If this zone is not populated, then it is not in zonelist.
873 * This means the page allocator ignores this zone.
874 * So, zonelist must be updated after online.
876 if (!populated_zone(zone)) {
877 need_zonelists_rebuild = 1;
878 setup_zone_pageset(zone);
881 ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
884 if (need_zonelists_rebuild)
885 zone_pcp_reset(zone);
886 goto failed_addition;
889 zone->present_pages += onlined_pages;
891 pgdat_resize_lock(zone->zone_pgdat, &flags);
892 zone->zone_pgdat->node_present_pages += onlined_pages;
893 pgdat_resize_unlock(zone->zone_pgdat, &flags);
896 node_states_set_node(nid, &arg);
897 if (need_zonelists_rebuild)
898 build_all_zonelists(NULL);
900 zone_pcp_update(zone);
903 init_per_zone_wmark_min();
910 vm_total_pages = nr_free_pagecache_pages();
912 writeback_set_ratelimit();
915 memory_notify(MEM_ONLINE, &arg);
920 pr_debug("online_pages [mem %#010llx-%#010llx] failed\n",
921 (unsigned long long) pfn << PAGE_SHIFT,
922 (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1);
923 memory_notify(MEM_CANCEL_ONLINE, &arg);
927 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
929 static void reset_node_present_pages(pg_data_t *pgdat)
933 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
934 z->present_pages = 0;
936 pgdat->node_present_pages = 0;
939 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
940 static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
942 struct pglist_data *pgdat;
943 unsigned long start_pfn = PFN_DOWN(start);
945 pgdat = NODE_DATA(nid);
947 pgdat = arch_alloc_nodedata(nid);
951 arch_refresh_nodedata(nid, pgdat);
954 * Reset the nr_zones, order and classzone_idx before reuse.
955 * Note that kswapd will init kswapd_classzone_idx properly
956 * when it starts in the near future.
959 pgdat->kswapd_order = 0;
960 pgdat->kswapd_classzone_idx = 0;
963 /* we can use NODE_DATA(nid) from here */
965 pgdat->node_id = nid;
966 pgdat->node_start_pfn = start_pfn;
968 /* init node's zones as empty zones, we don't have any present pages.*/
969 free_area_init_core_hotplug(nid);
970 pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat);
973 * The node we allocated has no zone fallback lists. For avoiding
974 * to access not-initialized zonelist, build here.
976 build_all_zonelists(pgdat);
979 * When memory is hot-added, all the memory is in offline state. So
980 * clear all zones' present_pages because they will be updated in
981 * online_pages() and offline_pages().
983 reset_node_managed_pages(pgdat);
984 reset_node_present_pages(pgdat);
989 static void rollback_node_hotadd(int nid)
991 pg_data_t *pgdat = NODE_DATA(nid);
993 arch_refresh_nodedata(nid, NULL);
994 free_percpu(pgdat->per_cpu_nodestats);
995 arch_free_nodedata(pgdat);
1001 * try_online_node - online a node if offlined
1003 * @start: start addr of the node
1004 * @set_node_online: Whether we want to online the node
1005 * called by cpu_up() to online a node without onlined memory.
1008 * 1 -> a new node has been allocated
1009 * 0 -> the node is already online
1010 * -ENOMEM -> the node could not be allocated
1012 static int __try_online_node(int nid, u64 start, bool set_node_online)
1017 if (node_online(nid))
1020 pgdat = hotadd_new_pgdat(nid, start);
1022 pr_err("Cannot online node %d due to NULL pgdat\n", nid);
1027 if (set_node_online) {
1028 node_set_online(nid);
1029 ret = register_one_node(nid);
1037 * Users of this function always want to online/register the node
1039 int try_online_node(int nid)
1043 mem_hotplug_begin();
1044 ret = __try_online_node(nid, 0, true);
1049 static int check_hotplug_memory_range(u64 start, u64 size)
1051 unsigned long block_sz = memory_block_size_bytes();
1052 u64 block_nr_pages = block_sz >> PAGE_SHIFT;
1053 u64 nr_pages = size >> PAGE_SHIFT;
1054 u64 start_pfn = PFN_DOWN(start);
1056 /* memory range must be block size aligned */
1057 if (!nr_pages || !IS_ALIGNED(start_pfn, block_nr_pages) ||
1058 !IS_ALIGNED(nr_pages, block_nr_pages)) {
1059 pr_err("Block size [%#lx] unaligned hotplug range: start %#llx, size %#llx",
1060 block_sz, start, size);
1067 static int online_memory_block(struct memory_block *mem, void *arg)
1069 return device_online(&mem->dev);
1073 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1074 * and online/offline operations (triggered e.g. by sysfs).
1076 * we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG
1078 int __ref add_memory_resource(int nid, struct resource *res)
1081 bool new_node = false;
1085 size = resource_size(res);
1087 ret = check_hotplug_memory_range(start, size);
1091 mem_hotplug_begin();
1094 * Add new range to memblock so that when hotadd_new_pgdat() is called
1095 * to allocate new pgdat, get_pfn_range_for_nid() will be able to find
1096 * this new range and calculate total pages correctly. The range will
1097 * be removed at hot-remove time.
1099 memblock_add_node(start, size, nid);
1101 ret = __try_online_node(nid, start, false);
1106 /* call arch's memory hotadd */
1107 ret = arch_add_memory(nid, start, size, NULL, true);
1112 /* If sysfs file of new node can't be created, cpu on the node
1113 * can't be hot-added. There is no rollback way now.
1114 * So, check by BUG_ON() to catch it reluctantly..
1115 * We online node here. We can't roll back from here.
1117 node_set_online(nid);
1118 ret = __register_one_node(nid);
1122 /* link memory sections under this node.*/
1123 ret = link_mem_sections(nid, PFN_DOWN(start), PFN_UP(start + size - 1));
1126 /* create new memmap entry */
1127 firmware_map_add_hotplug(start, start + size, "System RAM");
1129 /* device_online() will take the lock when calling online_pages() */
1132 /* online pages if requested */
1133 if (memhp_auto_online)
1134 walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1),
1135 NULL, online_memory_block);
1139 /* rollback pgdat allocation and others */
1141 rollback_node_hotadd(nid);
1142 memblock_remove(start, size);
1147 /* requires device_hotplug_lock, see add_memory_resource() */
1148 int __ref __add_memory(int nid, u64 start, u64 size)
1150 struct resource *res;
1153 res = register_memory_resource(start, size);
1155 return PTR_ERR(res);
1157 ret = add_memory_resource(nid, res);
1159 release_memory_resource(res);
1163 int add_memory(int nid, u64 start, u64 size)
1167 lock_device_hotplug();
1168 rc = __add_memory(nid, start, size);
1169 unlock_device_hotplug();
1173 EXPORT_SYMBOL_GPL(add_memory);
1175 #ifdef CONFIG_MEMORY_HOTREMOVE
1177 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1178 * set and the size of the free page is given by page_order(). Using this,
1179 * the function determines if the pageblock contains only free pages.
1180 * Due to buddy contraints, a free page at least the size of a pageblock will
1181 * be located at the start of the pageblock
1183 static inline int pageblock_free(struct page *page)
1185 return PageBuddy(page) && page_order(page) >= pageblock_order;
1188 /* Return the start of the next active pageblock after a given page */
1189 static struct page *next_active_pageblock(struct page *page)
1191 /* Ensure the starting page is pageblock-aligned */
1192 BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
1194 /* If the entire pageblock is free, move to the end of free page */
1195 if (pageblock_free(page)) {
1197 /* be careful. we don't have locks, page_order can be changed.*/
1198 order = page_order(page);
1199 if ((order < MAX_ORDER) && (order >= pageblock_order))
1200 return page + (1 << order);
1203 return page + pageblock_nr_pages;
1206 static bool is_pageblock_removable_nolock(struct page *page)
1212 * We have to be careful here because we are iterating over memory
1213 * sections which are not zone aware so we might end up outside of
1214 * the zone but still within the section.
1215 * We have to take care about the node as well. If the node is offline
1216 * its NODE_DATA will be NULL - see page_zone.
1218 if (!node_online(page_to_nid(page)))
1221 zone = page_zone(page);
1222 pfn = page_to_pfn(page);
1223 if (!zone_spans_pfn(zone, pfn))
1226 return !has_unmovable_pages(zone, page, 0, MIGRATE_MOVABLE, SKIP_HWPOISON);
1229 /* Checks if this range of memory is likely to be hot-removable. */
1230 bool is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
1232 struct page *page = pfn_to_page(start_pfn);
1233 struct page *end_page = page + nr_pages;
1235 /* Check the starting page of each pageblock within the range */
1236 for (; page < end_page; page = next_active_pageblock(page)) {
1237 if (!is_pageblock_removable_nolock(page))
1242 /* All pageblocks in the memory block are likely to be hot-removable */
1247 * Confirm all pages in a range [start, end) belong to the same zone.
1248 * When true, return its valid [start, end).
1250 int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn,
1251 unsigned long *valid_start, unsigned long *valid_end)
1253 unsigned long pfn, sec_end_pfn;
1254 unsigned long start, end;
1255 struct zone *zone = NULL;
1258 for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1);
1260 pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) {
1261 /* Make sure the memory section is present first */
1262 if (!present_section_nr(pfn_to_section_nr(pfn)))
1264 for (; pfn < sec_end_pfn && pfn < end_pfn;
1265 pfn += MAX_ORDER_NR_PAGES) {
1267 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1268 while ((i < MAX_ORDER_NR_PAGES) &&
1269 !pfn_valid_within(pfn + i))
1271 if (i == MAX_ORDER_NR_PAGES || pfn + i >= end_pfn)
1273 page = pfn_to_page(pfn + i);
1274 if (zone && page_zone(page) != zone)
1278 zone = page_zone(page);
1279 end = pfn + MAX_ORDER_NR_PAGES;
1284 *valid_start = start;
1285 *valid_end = min(end, end_pfn);
1293 * Scan pfn range [start,end) to find movable/migratable pages (LRU pages,
1294 * non-lru movable pages and hugepages). We scan pfn because it's much
1295 * easier than scanning over linked list. This function returns the pfn
1296 * of the first found movable page if it's found, otherwise 0.
1298 static unsigned long scan_movable_pages(unsigned long start, unsigned long end)
1302 for (pfn = start; pfn < end; pfn++) {
1303 if (pfn_valid(pfn)) {
1304 page = pfn_to_page(pfn);
1307 if (__PageMovable(page))
1309 if (PageHuge(page)) {
1310 if (hugepage_migration_supported(page_hstate(page)) &&
1311 page_huge_active(page))
1314 pfn = round_up(pfn + 1,
1315 1 << compound_order(page)) - 1;
1322 static struct page *new_node_page(struct page *page, unsigned long private)
1324 int nid = page_to_nid(page);
1325 nodemask_t nmask = node_states[N_MEMORY];
1328 * try to allocate from a different node but reuse this node if there
1329 * are no other online nodes to be used (e.g. we are offlining a part
1330 * of the only existing node)
1332 node_clear(nid, nmask);
1333 if (nodes_empty(nmask))
1334 node_set(nid, nmask);
1336 return new_page_nodemask(page, nid, &nmask);
1340 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1344 int not_managed = 0;
1348 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1349 if (!pfn_valid(pfn))
1351 page = pfn_to_page(pfn);
1353 if (PageHuge(page)) {
1354 struct page *head = compound_head(page);
1355 pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1;
1356 if (compound_order(head) > PFN_SECTION_SHIFT) {
1360 isolate_huge_page(page, &source);
1362 } else if (PageTransHuge(page))
1363 pfn = page_to_pfn(compound_head(page))
1364 + hpage_nr_pages(page) - 1;
1367 * HWPoison pages have elevated reference counts so the migration would
1368 * fail on them. It also doesn't make any sense to migrate them in the
1369 * first place. Still try to unmap such a page in case it is still mapped
1370 * (e.g. current hwpoison implementation doesn't unmap KSM pages but keep
1371 * the unmap as the catch all safety net).
1373 if (PageHWPoison(page)) {
1374 if (WARN_ON(PageLRU(page)))
1375 isolate_lru_page(page);
1376 if (page_mapped(page))
1377 try_to_unmap(page, TTU_IGNORE_MLOCK | TTU_IGNORE_ACCESS);
1381 if (!get_page_unless_zero(page))
1384 * We can skip free pages. And we can deal with pages on
1385 * LRU and non-lru movable pages.
1388 ret = isolate_lru_page(page);
1390 ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE);
1391 if (!ret) { /* Success */
1393 list_add_tail(&page->lru, &source);
1394 if (!__PageMovable(page))
1395 inc_node_page_state(page, NR_ISOLATED_ANON +
1396 page_is_file_cache(page));
1399 pr_warn("failed to isolate pfn %lx\n", pfn);
1400 dump_page(page, "isolation failed");
1402 /* Because we don't have big zone->lock. we should
1403 check this again here. */
1404 if (page_count(page)) {
1411 if (!list_empty(&source)) {
1413 putback_movable_pages(&source);
1417 /* Allocate a new page from the nearest neighbor node */
1418 ret = migrate_pages(&source, new_node_page, NULL, 0,
1419 MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
1421 list_for_each_entry(page, &source, lru) {
1422 pr_warn("migrating pfn %lx failed ret:%d ",
1423 page_to_pfn(page), ret);
1424 dump_page(page, "migration failure");
1426 putback_movable_pages(&source);
1434 * remove from free_area[] and mark all as Reserved.
1437 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1440 __offline_isolated_pages(start, start + nr_pages);
1445 offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
1447 walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
1448 offline_isolated_pages_cb);
1452 * Check all pages in range, recoreded as memory resource, are isolated.
1455 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1459 long offlined = *(long *)data;
1460 ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true);
1461 offlined = nr_pages;
1463 *(long *)data += offlined;
1468 check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
1473 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
1474 check_pages_isolated_cb);
1476 offlined = (long)ret;
1480 static int __init cmdline_parse_movable_node(char *p)
1482 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1483 movable_node_enabled = true;
1485 pr_warn("movable_node parameter depends on CONFIG_HAVE_MEMBLOCK_NODE_MAP to work properly\n");
1489 early_param("movable_node", cmdline_parse_movable_node);
1491 /* check which state of node_states will be changed when offline memory */
1492 static void node_states_check_changes_offline(unsigned long nr_pages,
1493 struct zone *zone, struct memory_notify *arg)
1495 struct pglist_data *pgdat = zone->zone_pgdat;
1496 unsigned long present_pages = 0;
1499 arg->status_change_nid = -1;
1500 arg->status_change_nid_normal = -1;
1501 arg->status_change_nid_high = -1;
1504 * Check whether node_states[N_NORMAL_MEMORY] will be changed.
1505 * If the memory to be offline is within the range
1506 * [0..ZONE_NORMAL], and it is the last present memory there,
1507 * the zones in that range will become empty after the offlining,
1508 * thus we can determine that we need to clear the node from
1509 * node_states[N_NORMAL_MEMORY].
1511 for (zt = 0; zt <= ZONE_NORMAL; zt++)
1512 present_pages += pgdat->node_zones[zt].present_pages;
1513 if (zone_idx(zone) <= ZONE_NORMAL && nr_pages >= present_pages)
1514 arg->status_change_nid_normal = zone_to_nid(zone);
1516 #ifdef CONFIG_HIGHMEM
1518 * node_states[N_HIGH_MEMORY] contains nodes which
1519 * have normal memory or high memory.
1520 * Here we add the present_pages belonging to ZONE_HIGHMEM.
1521 * If the zone is within the range of [0..ZONE_HIGHMEM), and
1522 * we determine that the zones in that range become empty,
1523 * we need to clear the node for N_HIGH_MEMORY.
1525 present_pages += pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1526 if (zone_idx(zone) <= ZONE_HIGHMEM && nr_pages >= present_pages)
1527 arg->status_change_nid_high = zone_to_nid(zone);
1531 * We have accounted the pages from [0..ZONE_NORMAL), and
1532 * in case of CONFIG_HIGHMEM the pages from ZONE_HIGHMEM
1534 * Here we count the possible pages from ZONE_MOVABLE.
1535 * If after having accounted all the pages, we see that the nr_pages
1536 * to be offlined is over or equal to the accounted pages,
1537 * we know that the node will become empty, and so, we can clear
1538 * it for N_MEMORY as well.
1540 present_pages += pgdat->node_zones[ZONE_MOVABLE].present_pages;
1542 if (nr_pages >= present_pages)
1543 arg->status_change_nid = zone_to_nid(zone);
1546 static void node_states_clear_node(int node, struct memory_notify *arg)
1548 if (arg->status_change_nid_normal >= 0)
1549 node_clear_state(node, N_NORMAL_MEMORY);
1551 if (arg->status_change_nid_high >= 0)
1552 node_clear_state(node, N_HIGH_MEMORY);
1554 if (arg->status_change_nid >= 0)
1555 node_clear_state(node, N_MEMORY);
1558 static int __ref __offline_pages(unsigned long start_pfn,
1559 unsigned long end_pfn)
1561 unsigned long pfn, nr_pages;
1562 long offlined_pages;
1564 unsigned long flags;
1565 unsigned long valid_start, valid_end;
1567 struct memory_notify arg;
1570 mem_hotplug_begin();
1572 /* This makes hotplug much easier...and readable.
1573 we assume this for now. .*/
1574 if (!test_pages_in_a_zone(start_pfn, end_pfn, &valid_start,
1578 reason = "multizone range";
1579 goto failed_removal;
1582 zone = page_zone(pfn_to_page(valid_start));
1583 node = zone_to_nid(zone);
1584 nr_pages = end_pfn - start_pfn;
1586 /* set above range as isolated */
1587 ret = start_isolate_page_range(start_pfn, end_pfn,
1589 SKIP_HWPOISON | REPORT_FAILURE);
1592 reason = "failure to isolate range";
1593 goto failed_removal;
1596 arg.start_pfn = start_pfn;
1597 arg.nr_pages = nr_pages;
1598 node_states_check_changes_offline(nr_pages, zone, &arg);
1600 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1601 ret = notifier_to_errno(ret);
1603 reason = "notifier failure";
1604 goto failed_removal_isolated;
1608 for (pfn = start_pfn; pfn;) {
1609 if (signal_pending(current)) {
1611 reason = "signal backoff";
1612 goto failed_removal_isolated;
1616 lru_add_drain_all();
1617 drain_all_pages(zone);
1619 pfn = scan_movable_pages(pfn, end_pfn);
1622 * TODO: fatal migration failures should bail
1625 do_migrate_range(pfn, end_pfn);
1630 * Dissolve free hugepages in the memory block before doing
1631 * offlining actually in order to make hugetlbfs's object
1632 * counting consistent.
1634 ret = dissolve_free_huge_pages(start_pfn, end_pfn);
1636 reason = "failure to dissolve huge pages";
1637 goto failed_removal_isolated;
1640 offlined_pages = check_pages_isolated(start_pfn, end_pfn);
1641 } while (offlined_pages < 0);
1643 pr_info("Offlined Pages %ld\n", offlined_pages);
1644 /* Ok, all of our target is isolated.
1645 We cannot do rollback at this point. */
1646 offline_isolated_pages(start_pfn, end_pfn);
1647 /* reset pagetype flags and makes migrate type to be MOVABLE */
1648 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1649 /* removal success */
1650 adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages);
1651 zone->present_pages -= offlined_pages;
1653 pgdat_resize_lock(zone->zone_pgdat, &flags);
1654 zone->zone_pgdat->node_present_pages -= offlined_pages;
1655 pgdat_resize_unlock(zone->zone_pgdat, &flags);
1657 init_per_zone_wmark_min();
1659 if (!populated_zone(zone)) {
1660 zone_pcp_reset(zone);
1661 build_all_zonelists(NULL);
1663 zone_pcp_update(zone);
1665 node_states_clear_node(node, &arg);
1666 if (arg.status_change_nid >= 0) {
1668 kcompactd_stop(node);
1671 vm_total_pages = nr_free_pagecache_pages();
1672 writeback_set_ratelimit();
1674 memory_notify(MEM_OFFLINE, &arg);
1678 failed_removal_isolated:
1679 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1681 pr_debug("memory offlining [mem %#010llx-%#010llx] failed due to %s\n",
1682 (unsigned long long) start_pfn << PAGE_SHIFT,
1683 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1,
1685 memory_notify(MEM_CANCEL_OFFLINE, &arg);
1686 /* pushback to free area */
1691 int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1693 return __offline_pages(start_pfn, start_pfn + nr_pages);
1695 #endif /* CONFIG_MEMORY_HOTREMOVE */
1698 * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
1699 * @start_pfn: start pfn of the memory range
1700 * @end_pfn: end pfn of the memory range
1701 * @arg: argument passed to func
1702 * @func: callback for each memory section walked
1704 * This function walks through all present mem sections in range
1705 * [start_pfn, end_pfn) and call func on each mem section.
1707 * Returns the return value of func.
1709 int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn,
1710 void *arg, int (*func)(struct memory_block *, void *))
1712 struct memory_block *mem = NULL;
1713 struct mem_section *section;
1714 unsigned long pfn, section_nr;
1717 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1718 section_nr = pfn_to_section_nr(pfn);
1719 if (!present_section_nr(section_nr))
1722 section = __nr_to_section(section_nr);
1723 /* same memblock? */
1725 if ((section_nr >= mem->start_section_nr) &&
1726 (section_nr <= mem->end_section_nr))
1729 mem = find_memory_block_hinted(section, mem);
1733 ret = func(mem, arg);
1735 kobject_put(&mem->dev.kobj);
1741 kobject_put(&mem->dev.kobj);
1746 #ifdef CONFIG_MEMORY_HOTREMOVE
1747 static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
1749 int ret = !is_memblock_offlined(mem);
1751 if (unlikely(ret)) {
1752 phys_addr_t beginpa, endpa;
1754 beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
1755 endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1;
1756 pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n",
1763 static int check_cpu_on_node(pg_data_t *pgdat)
1767 for_each_present_cpu(cpu) {
1768 if (cpu_to_node(cpu) == pgdat->node_id)
1770 * the cpu on this node isn't removed, and we can't
1771 * offline this node.
1783 * Offline a node if all memory sections and cpus of the node are removed.
1785 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1786 * and online/offline operations before this call.
1788 void try_offline_node(int nid)
1790 pg_data_t *pgdat = NODE_DATA(nid);
1791 unsigned long start_pfn = pgdat->node_start_pfn;
1792 unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
1795 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1796 unsigned long section_nr = pfn_to_section_nr(pfn);
1798 if (!present_section_nr(section_nr))
1801 if (pfn_to_nid(pfn) != nid)
1805 * some memory sections of this node are not removed, and we
1806 * can't offline node now.
1811 if (check_cpu_on_node(pgdat))
1815 * all memory/cpu of this node are removed, we can offline this
1818 node_set_offline(nid);
1819 unregister_one_node(nid);
1821 EXPORT_SYMBOL(try_offline_node);
1826 * @start: physical address of the region to remove
1827 * @size: size of the region to remove
1829 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1830 * and online/offline operations before this call, as required by
1831 * try_offline_node().
1833 void __ref __remove_memory(int nid, u64 start, u64 size)
1837 BUG_ON(check_hotplug_memory_range(start, size));
1839 mem_hotplug_begin();
1842 * All memory blocks must be offlined before removing memory. Check
1843 * whether all memory blocks in question are offline and trigger a BUG()
1844 * if this is not the case.
1846 ret = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL,
1847 check_memblock_offlined_cb);
1851 /* remove memmap entry */
1852 firmware_map_remove(start, start + size, "System RAM");
1853 memblock_free(start, size);
1854 memblock_remove(start, size);
1856 arch_remove_memory(nid, start, size, NULL);
1858 try_offline_node(nid);
1863 void remove_memory(int nid, u64 start, u64 size)
1865 lock_device_hotplug();
1866 __remove_memory(nid, start, size);
1867 unlock_device_hotplug();
1869 EXPORT_SYMBOL_GPL(remove_memory);
1870 #endif /* CONFIG_MEMORY_HOTREMOVE */