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/bootmem.h>
37 #include <linux/compaction.h>
38 #include <linux/rmap.h>
40 #include <asm/tlbflush.h>
45 * online_page_callback contains pointer to current page onlining function.
46 * Initially it is generic_online_page(). If it is required it could be
47 * changed by calling set_online_page_callback() for callback registration
48 * and restore_online_page_callback() for generic callback restore.
51 static void generic_online_page(struct page *page);
53 static online_page_callback_t online_page_callback = generic_online_page;
54 static DEFINE_MUTEX(online_page_callback_lock);
56 DEFINE_STATIC_PERCPU_RWSEM(mem_hotplug_lock);
58 void get_online_mems(void)
60 percpu_down_read(&mem_hotplug_lock);
63 void put_online_mems(void)
65 percpu_up_read(&mem_hotplug_lock);
68 bool movable_node_enabled = false;
70 #ifndef CONFIG_MEMORY_HOTPLUG_DEFAULT_ONLINE
71 bool memhp_auto_online;
73 bool memhp_auto_online = true;
75 EXPORT_SYMBOL_GPL(memhp_auto_online);
77 static int __init setup_memhp_default_state(char *str)
79 if (!strcmp(str, "online"))
80 memhp_auto_online = true;
81 else if (!strcmp(str, "offline"))
82 memhp_auto_online = false;
86 __setup("memhp_default_state=", setup_memhp_default_state);
88 void mem_hotplug_begin(void)
91 percpu_down_write(&mem_hotplug_lock);
94 void mem_hotplug_done(void)
96 percpu_up_write(&mem_hotplug_lock);
100 /* add this memory to iomem resource */
101 static struct resource *register_memory_resource(u64 start, u64 size)
103 struct resource *res, *conflict;
104 res = kzalloc(sizeof(struct resource), GFP_KERNEL);
106 return ERR_PTR(-ENOMEM);
108 res->name = "System RAM";
110 res->end = start + size - 1;
111 res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
112 conflict = request_resource_conflict(&iomem_resource, res);
114 if (conflict->desc == IORES_DESC_DEVICE_PRIVATE_MEMORY) {
115 pr_debug("Device unaddressable memory block "
116 "memory hotplug at %#010llx !\n",
117 (unsigned long long)start);
119 pr_debug("System RAM resource %pR cannot be added\n", res);
121 return ERR_PTR(-EEXIST);
126 static void release_memory_resource(struct resource *res)
130 release_resource(res);
135 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
136 void get_page_bootmem(unsigned long info, struct page *page,
139 page->freelist = (void *)type;
140 SetPagePrivate(page);
141 set_page_private(page, info);
145 void put_page_bootmem(struct page *page)
149 type = (unsigned long) page->freelist;
150 BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE ||
151 type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE);
153 if (page_ref_dec_return(page) == 1) {
154 page->freelist = NULL;
155 ClearPagePrivate(page);
156 set_page_private(page, 0);
157 INIT_LIST_HEAD(&page->lru);
158 free_reserved_page(page);
162 #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
163 #ifndef CONFIG_SPARSEMEM_VMEMMAP
164 static void register_page_bootmem_info_section(unsigned long start_pfn)
166 unsigned long *usemap, mapsize, section_nr, i;
167 struct mem_section *ms;
168 struct page *page, *memmap;
170 section_nr = pfn_to_section_nr(start_pfn);
171 ms = __nr_to_section(section_nr);
173 /* Get section's memmap address */
174 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
177 * Get page for the memmap's phys address
178 * XXX: need more consideration for sparse_vmemmap...
180 page = virt_to_page(memmap);
181 mapsize = sizeof(struct page) * PAGES_PER_SECTION;
182 mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
184 /* remember memmap's page */
185 for (i = 0; i < mapsize; i++, page++)
186 get_page_bootmem(section_nr, page, SECTION_INFO);
188 usemap = ms->pageblock_flags;
189 page = virt_to_page(usemap);
191 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
193 for (i = 0; i < mapsize; i++, page++)
194 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
197 #else /* CONFIG_SPARSEMEM_VMEMMAP */
198 static void register_page_bootmem_info_section(unsigned long start_pfn)
200 unsigned long *usemap, mapsize, section_nr, i;
201 struct mem_section *ms;
202 struct page *page, *memmap;
204 section_nr = pfn_to_section_nr(start_pfn);
205 ms = __nr_to_section(section_nr);
207 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
209 register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION);
211 usemap = ms->pageblock_flags;
212 page = virt_to_page(usemap);
214 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
216 for (i = 0; i < mapsize; i++, page++)
217 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
219 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
221 void __init register_page_bootmem_info_node(struct pglist_data *pgdat)
223 unsigned long i, pfn, end_pfn, nr_pages;
224 int node = pgdat->node_id;
227 nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
228 page = virt_to_page(pgdat);
230 for (i = 0; i < nr_pages; i++, page++)
231 get_page_bootmem(node, page, NODE_INFO);
233 pfn = pgdat->node_start_pfn;
234 end_pfn = pgdat_end_pfn(pgdat);
236 /* register section info */
237 for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
239 * Some platforms can assign the same pfn to multiple nodes - on
240 * node0 as well as nodeN. To avoid registering a pfn against
241 * multiple nodes we check that this pfn does not already
242 * reside in some other nodes.
244 if (pfn_valid(pfn) && (early_pfn_to_nid(pfn) == node))
245 register_page_bootmem_info_section(pfn);
248 #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
250 static int __meminit __add_section(int nid, unsigned long phys_start_pfn,
251 struct vmem_altmap *altmap, bool want_memblock)
255 if (pfn_valid(phys_start_pfn))
258 ret = sparse_add_one_section(NODE_DATA(nid), phys_start_pfn, altmap);
265 return hotplug_memory_register(nid, __pfn_to_section(phys_start_pfn));
269 * Reasonably generic function for adding memory. It is
270 * expected that archs that support memory hotplug will
271 * call this function after deciding the zone to which to
274 int __ref __add_pages(int nid, unsigned long phys_start_pfn,
275 unsigned long nr_pages, struct vmem_altmap *altmap,
280 int start_sec, end_sec;
282 /* during initialize mem_map, align hot-added range to section */
283 start_sec = pfn_to_section_nr(phys_start_pfn);
284 end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
288 * Validate altmap is within bounds of the total request
290 if (altmap->base_pfn != phys_start_pfn
291 || vmem_altmap_offset(altmap) > nr_pages) {
292 pr_warn_once("memory add fail, invalid altmap\n");
299 for (i = start_sec; i <= end_sec; i++) {
300 err = __add_section(nid, section_nr_to_pfn(i), altmap,
304 * EEXIST is finally dealt with by ioresource collision
305 * check. see add_memory() => register_memory_resource()
306 * Warning will be printed if there is collision.
308 if (err && (err != -EEXIST))
313 vmemmap_populate_print_last();
318 #ifdef CONFIG_MEMORY_HOTREMOVE
319 /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
320 static unsigned long find_smallest_section_pfn(int nid, struct zone *zone,
321 unsigned long start_pfn,
322 unsigned long end_pfn)
324 struct mem_section *ms;
326 for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) {
327 ms = __pfn_to_section(start_pfn);
329 if (unlikely(!valid_section(ms)))
332 if (unlikely(pfn_to_nid(start_pfn) != nid))
335 if (zone && zone != page_zone(pfn_to_page(start_pfn)))
344 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
345 static unsigned long find_biggest_section_pfn(int nid, struct zone *zone,
346 unsigned long start_pfn,
347 unsigned long end_pfn)
349 struct mem_section *ms;
352 /* pfn is the end pfn of a memory section. */
354 for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) {
355 ms = __pfn_to_section(pfn);
357 if (unlikely(!valid_section(ms)))
360 if (unlikely(pfn_to_nid(pfn) != nid))
363 if (zone && zone != page_zone(pfn_to_page(pfn)))
372 static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
373 unsigned long end_pfn)
375 unsigned long zone_start_pfn = zone->zone_start_pfn;
376 unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */
377 unsigned long zone_end_pfn = z;
379 struct mem_section *ms;
380 int nid = zone_to_nid(zone);
382 zone_span_writelock(zone);
383 if (zone_start_pfn == start_pfn) {
385 * If the section is smallest section in the zone, it need
386 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
387 * In this case, we find second smallest valid mem_section
388 * for shrinking zone.
390 pfn = find_smallest_section_pfn(nid, zone, end_pfn,
393 zone->zone_start_pfn = pfn;
394 zone->spanned_pages = zone_end_pfn - pfn;
396 } else if (zone_end_pfn == end_pfn) {
398 * If the section is biggest section in the zone, it need
399 * shrink zone->spanned_pages.
400 * In this case, we find second biggest valid mem_section for
403 pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn,
406 zone->spanned_pages = pfn - zone_start_pfn + 1;
410 * The section is not biggest or smallest mem_section in the zone, it
411 * only creates a hole in the zone. So in this case, we need not
412 * change the zone. But perhaps, the zone has only hole data. Thus
413 * it check the zone has only hole or not.
415 pfn = zone_start_pfn;
416 for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) {
417 ms = __pfn_to_section(pfn);
419 if (unlikely(!valid_section(ms)))
422 if (page_zone(pfn_to_page(pfn)) != zone)
425 /* If the section is current section, it continues the loop */
426 if (start_pfn == pfn)
429 /* If we find valid section, we have nothing to do */
430 zone_span_writeunlock(zone);
434 /* The zone has no valid section */
435 zone->zone_start_pfn = 0;
436 zone->spanned_pages = 0;
437 zone_span_writeunlock(zone);
440 static void update_pgdat_span(struct pglist_data *pgdat)
442 unsigned long node_start_pfn = 0, node_end_pfn = 0;
445 for (zone = pgdat->node_zones;
446 zone < pgdat->node_zones + MAX_NR_ZONES; zone++) {
447 unsigned long zone_end_pfn = zone->zone_start_pfn +
450 /* No need to lock the zones, they can't change. */
451 if (!zone->spanned_pages)
454 node_start_pfn = zone->zone_start_pfn;
455 node_end_pfn = zone_end_pfn;
459 if (zone_end_pfn > node_end_pfn)
460 node_end_pfn = zone_end_pfn;
461 if (zone->zone_start_pfn < node_start_pfn)
462 node_start_pfn = zone->zone_start_pfn;
465 pgdat->node_start_pfn = node_start_pfn;
466 pgdat->node_spanned_pages = node_end_pfn - node_start_pfn;
469 static void __remove_zone(struct zone *zone, unsigned long start_pfn)
471 struct pglist_data *pgdat = zone->zone_pgdat;
472 int nr_pages = PAGES_PER_SECTION;
475 pgdat_resize_lock(zone->zone_pgdat, &flags);
476 shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
477 update_pgdat_span(pgdat);
478 pgdat_resize_unlock(zone->zone_pgdat, &flags);
481 static int __remove_section(struct zone *zone, struct mem_section *ms,
482 unsigned long map_offset, struct vmem_altmap *altmap)
484 unsigned long start_pfn;
488 if (!valid_section(ms))
491 ret = unregister_memory_section(ms);
495 scn_nr = __section_nr(ms);
496 start_pfn = section_nr_to_pfn((unsigned long)scn_nr);
497 __remove_zone(zone, start_pfn);
499 sparse_remove_one_section(zone, ms, map_offset, altmap);
504 * __remove_pages() - remove sections of pages from a zone
505 * @zone: zone from which pages need to be removed
506 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
507 * @nr_pages: number of pages to remove (must be multiple of section size)
508 * @altmap: alternative device page map or %NULL if default memmap is used
510 * Generic helper function to remove section mappings and sysfs entries
511 * for the section of the memory we are removing. Caller needs to make
512 * sure that pages are marked reserved and zones are adjust properly by
513 * calling offline_pages().
515 int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
516 unsigned long nr_pages, struct vmem_altmap *altmap)
519 unsigned long map_offset = 0;
520 int sections_to_remove, ret = 0;
522 /* In the ZONE_DEVICE case device driver owns the memory region */
523 if (is_dev_zone(zone)) {
525 map_offset = vmem_altmap_offset(altmap);
527 resource_size_t start, size;
529 start = phys_start_pfn << PAGE_SHIFT;
530 size = nr_pages * PAGE_SIZE;
532 ret = release_mem_region_adjustable(&iomem_resource, start,
535 resource_size_t endres = start + size - 1;
537 pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
538 &start, &endres, ret);
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 ret = __remove_section(zone, __pfn_to_section(pfn), map_offset,
562 set_zone_contiguous(zone);
566 #endif /* CONFIG_MEMORY_HOTREMOVE */
568 int set_online_page_callback(online_page_callback_t callback)
573 mutex_lock(&online_page_callback_lock);
575 if (online_page_callback == generic_online_page) {
576 online_page_callback = callback;
580 mutex_unlock(&online_page_callback_lock);
585 EXPORT_SYMBOL_GPL(set_online_page_callback);
587 int restore_online_page_callback(online_page_callback_t callback)
592 mutex_lock(&online_page_callback_lock);
594 if (online_page_callback == callback) {
595 online_page_callback = generic_online_page;
599 mutex_unlock(&online_page_callback_lock);
604 EXPORT_SYMBOL_GPL(restore_online_page_callback);
606 void __online_page_set_limits(struct page *page)
609 EXPORT_SYMBOL_GPL(__online_page_set_limits);
611 void __online_page_increment_counters(struct page *page)
613 adjust_managed_page_count(page, 1);
615 EXPORT_SYMBOL_GPL(__online_page_increment_counters);
617 void __online_page_free(struct page *page)
619 __free_reserved_page(page);
621 EXPORT_SYMBOL_GPL(__online_page_free);
623 static void generic_online_page(struct page *page)
625 __online_page_set_limits(page);
626 __online_page_increment_counters(page);
627 __online_page_free(page);
630 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
634 unsigned long onlined_pages = *(unsigned long *)arg;
637 if (PageReserved(pfn_to_page(start_pfn)))
638 for (i = 0; i < nr_pages; i++) {
639 page = pfn_to_page(start_pfn + i);
640 (*online_page_callback)(page);
644 online_mem_sections(start_pfn, start_pfn + nr_pages);
646 *(unsigned long *)arg = onlined_pages;
650 /* check which state of node_states will be changed when online memory */
651 static void node_states_check_changes_online(unsigned long nr_pages,
652 struct zone *zone, struct memory_notify *arg)
654 int nid = zone_to_nid(zone);
655 enum zone_type zone_last = ZONE_NORMAL;
658 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
659 * contains nodes which have zones of 0...ZONE_NORMAL,
660 * set zone_last to ZONE_NORMAL.
662 * If we don't have HIGHMEM nor movable node,
663 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
664 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
666 if (N_MEMORY == N_NORMAL_MEMORY)
667 zone_last = ZONE_MOVABLE;
670 * if the memory to be online is in a zone of 0...zone_last, and
671 * the zones of 0...zone_last don't have memory before online, we will
672 * need to set the node to node_states[N_NORMAL_MEMORY] after
673 * the memory is online.
675 if (zone_idx(zone) <= zone_last && !node_state(nid, N_NORMAL_MEMORY))
676 arg->status_change_nid_normal = nid;
678 arg->status_change_nid_normal = -1;
680 #ifdef CONFIG_HIGHMEM
682 * If we have movable node, node_states[N_HIGH_MEMORY]
683 * contains nodes which have zones of 0...ZONE_HIGHMEM,
684 * set zone_last to ZONE_HIGHMEM.
686 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
687 * contains nodes which have zones of 0...ZONE_MOVABLE,
688 * set zone_last to ZONE_MOVABLE.
690 zone_last = ZONE_HIGHMEM;
691 if (N_MEMORY == N_HIGH_MEMORY)
692 zone_last = ZONE_MOVABLE;
694 if (zone_idx(zone) <= zone_last && !node_state(nid, N_HIGH_MEMORY))
695 arg->status_change_nid_high = nid;
697 arg->status_change_nid_high = -1;
699 arg->status_change_nid_high = arg->status_change_nid_normal;
703 * if the node don't have memory befor online, we will need to
704 * set the node to node_states[N_MEMORY] after the memory
707 if (!node_state(nid, N_MEMORY))
708 arg->status_change_nid = nid;
710 arg->status_change_nid = -1;
713 static void node_states_set_node(int node, struct memory_notify *arg)
715 if (arg->status_change_nid_normal >= 0)
716 node_set_state(node, N_NORMAL_MEMORY);
718 if (arg->status_change_nid_high >= 0)
719 node_set_state(node, N_HIGH_MEMORY);
721 node_set_state(node, N_MEMORY);
724 static void __meminit resize_zone_range(struct zone *zone, unsigned long start_pfn,
725 unsigned long nr_pages)
727 unsigned long old_end_pfn = zone_end_pfn(zone);
729 if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
730 zone->zone_start_pfn = start_pfn;
732 zone->spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - zone->zone_start_pfn;
735 static void __meminit resize_pgdat_range(struct pglist_data *pgdat, unsigned long start_pfn,
736 unsigned long nr_pages)
738 unsigned long old_end_pfn = pgdat_end_pfn(pgdat);
740 if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
741 pgdat->node_start_pfn = start_pfn;
743 pgdat->node_spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - pgdat->node_start_pfn;
746 void __ref move_pfn_range_to_zone(struct zone *zone, unsigned long start_pfn,
747 unsigned long nr_pages, struct vmem_altmap *altmap)
749 struct pglist_data *pgdat = zone->zone_pgdat;
750 int nid = pgdat->node_id;
753 if (zone_is_empty(zone))
754 init_currently_empty_zone(zone, start_pfn, nr_pages);
756 clear_zone_contiguous(zone);
758 /* TODO Huh pgdat is irqsave while zone is not. It used to be like that before */
759 pgdat_resize_lock(pgdat, &flags);
760 zone_span_writelock(zone);
761 resize_zone_range(zone, start_pfn, nr_pages);
762 zone_span_writeunlock(zone);
763 resize_pgdat_range(pgdat, start_pfn, nr_pages);
764 pgdat_resize_unlock(pgdat, &flags);
767 * TODO now we have a visible range of pages which are not associated
768 * with their zone properly. Not nice but set_pfnblock_flags_mask
769 * expects the zone spans the pfn range. All the pages in the range
770 * are reserved so nobody should be touching them so we should be safe
772 memmap_init_zone(nr_pages, nid, zone_idx(zone), start_pfn,
773 MEMMAP_HOTPLUG, altmap);
775 set_zone_contiguous(zone);
779 * Returns a default kernel memory zone for the given pfn range.
780 * If no kernel zone covers this pfn range it will automatically go
781 * to the ZONE_NORMAL.
783 static struct zone *default_kernel_zone_for_pfn(int nid, unsigned long start_pfn,
784 unsigned long nr_pages)
786 struct pglist_data *pgdat = NODE_DATA(nid);
789 for (zid = 0; zid <= ZONE_NORMAL; zid++) {
790 struct zone *zone = &pgdat->node_zones[zid];
792 if (zone_intersects(zone, start_pfn, nr_pages))
796 return &pgdat->node_zones[ZONE_NORMAL];
799 static inline struct zone *default_zone_for_pfn(int nid, unsigned long start_pfn,
800 unsigned long nr_pages)
802 struct zone *kernel_zone = default_kernel_zone_for_pfn(nid, start_pfn,
804 struct zone *movable_zone = &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
805 bool in_kernel = zone_intersects(kernel_zone, start_pfn, nr_pages);
806 bool in_movable = zone_intersects(movable_zone, start_pfn, nr_pages);
809 * We inherit the existing zone in a simple case where zones do not
810 * overlap in the given range
812 if (in_kernel ^ in_movable)
813 return (in_kernel) ? kernel_zone : movable_zone;
816 * If the range doesn't belong to any zone or two zones overlap in the
817 * given range then we use movable zone only if movable_node is
818 * enabled because we always online to a kernel zone by default.
820 return movable_node_enabled ? movable_zone : kernel_zone;
823 struct zone * zone_for_pfn_range(int online_type, int nid, unsigned start_pfn,
824 unsigned long nr_pages)
826 if (online_type == MMOP_ONLINE_KERNEL)
827 return default_kernel_zone_for_pfn(nid, start_pfn, nr_pages);
829 if (online_type == MMOP_ONLINE_MOVABLE)
830 return &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
832 return default_zone_for_pfn(nid, start_pfn, nr_pages);
836 * Associates the given pfn range with the given node and the zone appropriate
837 * for the given online type.
839 static struct zone * __meminit move_pfn_range(int online_type, int nid,
840 unsigned long start_pfn, unsigned long nr_pages)
844 zone = zone_for_pfn_range(online_type, nid, start_pfn, nr_pages);
845 move_pfn_range_to_zone(zone, start_pfn, nr_pages, NULL);
849 /* Must be protected by mem_hotplug_begin() or a device_lock */
850 int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type)
853 unsigned long onlined_pages = 0;
855 int need_zonelists_rebuild = 0;
858 struct memory_notify arg;
859 struct memory_block *mem;
862 * We can't use pfn_to_nid() because nid might be stored in struct page
863 * which is not yet initialized. Instead, we find nid from memory block.
865 mem = find_memory_block(__pfn_to_section(pfn));
867 put_device(&mem->dev);
869 /* associate pfn range with the zone */
870 zone = move_pfn_range(online_type, nid, pfn, nr_pages);
873 arg.nr_pages = nr_pages;
874 node_states_check_changes_online(nr_pages, zone, &arg);
876 ret = memory_notify(MEM_GOING_ONLINE, &arg);
877 ret = notifier_to_errno(ret);
879 goto failed_addition;
882 * If this zone is not populated, then it is not in zonelist.
883 * This means the page allocator ignores this zone.
884 * So, zonelist must be updated after online.
886 if (!populated_zone(zone)) {
887 need_zonelists_rebuild = 1;
888 setup_zone_pageset(zone);
891 ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
894 if (need_zonelists_rebuild)
895 zone_pcp_reset(zone);
896 goto failed_addition;
899 zone->present_pages += onlined_pages;
901 pgdat_resize_lock(zone->zone_pgdat, &flags);
902 zone->zone_pgdat->node_present_pages += onlined_pages;
903 pgdat_resize_unlock(zone->zone_pgdat, &flags);
906 node_states_set_node(nid, &arg);
907 if (need_zonelists_rebuild)
908 build_all_zonelists(NULL);
910 zone_pcp_update(zone);
913 init_per_zone_wmark_min();
920 vm_total_pages = nr_free_pagecache_pages();
922 writeback_set_ratelimit();
925 memory_notify(MEM_ONLINE, &arg);
929 pr_debug("online_pages [mem %#010llx-%#010llx] failed\n",
930 (unsigned long long) pfn << PAGE_SHIFT,
931 (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1);
932 memory_notify(MEM_CANCEL_ONLINE, &arg);
935 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
937 static void reset_node_present_pages(pg_data_t *pgdat)
941 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
942 z->present_pages = 0;
944 pgdat->node_present_pages = 0;
947 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
948 static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
950 struct pglist_data *pgdat;
951 unsigned long start_pfn = PFN_DOWN(start);
953 pgdat = NODE_DATA(nid);
955 pgdat = arch_alloc_nodedata(nid);
959 arch_refresh_nodedata(nid, pgdat);
962 * Reset the nr_zones, order and classzone_idx before reuse.
963 * Note that kswapd will init kswapd_classzone_idx properly
964 * when it starts in the near future.
967 pgdat->kswapd_order = 0;
968 pgdat->kswapd_classzone_idx = 0;
971 /* we can use NODE_DATA(nid) from here */
973 pgdat->node_id = nid;
974 pgdat->node_start_pfn = start_pfn;
976 /* init node's zones as empty zones, we don't have any present pages.*/
977 free_area_init_core_hotplug(nid);
978 pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat);
981 * The node we allocated has no zone fallback lists. For avoiding
982 * to access not-initialized zonelist, build here.
984 build_all_zonelists(pgdat);
987 * When memory is hot-added, all the memory is in offline state. So
988 * clear all zones' present_pages because they will be updated in
989 * online_pages() and offline_pages().
991 reset_node_managed_pages(pgdat);
992 reset_node_present_pages(pgdat);
997 static void rollback_node_hotadd(int nid)
999 pg_data_t *pgdat = NODE_DATA(nid);
1001 arch_refresh_nodedata(nid, NULL);
1002 free_percpu(pgdat->per_cpu_nodestats);
1003 arch_free_nodedata(pgdat);
1009 * try_online_node - online a node if offlined
1011 * @start: start addr of the node
1012 * @set_node_online: Whether we want to online the node
1013 * called by cpu_up() to online a node without onlined memory.
1016 * 1 -> a new node has been allocated
1017 * 0 -> the node is already online
1018 * -ENOMEM -> the node could not be allocated
1020 static int __try_online_node(int nid, u64 start, bool set_node_online)
1025 if (node_online(nid))
1028 pgdat = hotadd_new_pgdat(nid, start);
1030 pr_err("Cannot online node %d due to NULL pgdat\n", nid);
1035 if (set_node_online) {
1036 node_set_online(nid);
1037 ret = register_one_node(nid);
1045 * Users of this function always want to online/register the node
1047 int try_online_node(int nid)
1051 mem_hotplug_begin();
1052 ret = __try_online_node(nid, 0, true);
1057 static int check_hotplug_memory_range(u64 start, u64 size)
1059 unsigned long block_sz = memory_block_size_bytes();
1060 u64 block_nr_pages = block_sz >> PAGE_SHIFT;
1061 u64 nr_pages = size >> PAGE_SHIFT;
1062 u64 start_pfn = PFN_DOWN(start);
1064 /* memory range must be block size aligned */
1065 if (!nr_pages || !IS_ALIGNED(start_pfn, block_nr_pages) ||
1066 !IS_ALIGNED(nr_pages, block_nr_pages)) {
1067 pr_err("Block size [%#lx] unaligned hotplug range: start %#llx, size %#llx",
1068 block_sz, start, size);
1075 static int online_memory_block(struct memory_block *mem, void *arg)
1077 return device_online(&mem->dev);
1081 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1082 * and online/offline operations (triggered e.g. by sysfs).
1084 * we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG
1086 int __ref add_memory_resource(int nid, struct resource *res, bool online)
1089 bool new_node = false;
1093 size = resource_size(res);
1095 ret = check_hotplug_memory_range(start, size);
1099 mem_hotplug_begin();
1102 * Add new range to memblock so that when hotadd_new_pgdat() is called
1103 * to allocate new pgdat, get_pfn_range_for_nid() will be able to find
1104 * this new range and calculate total pages correctly. The range will
1105 * be removed at hot-remove time.
1107 memblock_add_node(start, size, nid);
1109 ret = __try_online_node(nid, start, false);
1114 /* call arch's memory hotadd */
1115 ret = arch_add_memory(nid, start, size, NULL, true);
1120 /* If sysfs file of new node can't be created, cpu on the node
1121 * can't be hot-added. There is no rollback way now.
1122 * So, check by BUG_ON() to catch it reluctantly..
1123 * We online node here. We can't roll back from here.
1125 node_set_online(nid);
1126 ret = __register_one_node(nid);
1130 /* link memory sections under this node.*/
1131 ret = link_mem_sections(nid, PFN_DOWN(start), PFN_UP(start + size - 1));
1134 /* create new memmap entry */
1135 firmware_map_add_hotplug(start, start + size, "System RAM");
1137 /* online pages if requested */
1139 walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1),
1140 NULL, online_memory_block);
1145 /* rollback pgdat allocation and others */
1147 rollback_node_hotadd(nid);
1148 memblock_remove(start, size);
1155 /* requires device_hotplug_lock, see add_memory_resource() */
1156 int __ref __add_memory(int nid, u64 start, u64 size)
1158 struct resource *res;
1161 res = register_memory_resource(start, size);
1163 return PTR_ERR(res);
1165 ret = add_memory_resource(nid, res, memhp_auto_online);
1167 release_memory_resource(res);
1171 int add_memory(int nid, u64 start, u64 size)
1175 lock_device_hotplug();
1176 rc = __add_memory(nid, start, size);
1177 unlock_device_hotplug();
1181 EXPORT_SYMBOL_GPL(add_memory);
1183 #ifdef CONFIG_MEMORY_HOTREMOVE
1185 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1186 * set and the size of the free page is given by page_order(). Using this,
1187 * the function determines if the pageblock contains only free pages.
1188 * Due to buddy contraints, a free page at least the size of a pageblock will
1189 * be located at the start of the pageblock
1191 static inline int pageblock_free(struct page *page)
1193 return PageBuddy(page) && page_order(page) >= pageblock_order;
1196 /* Return the pfn of the start of the next active pageblock after a given pfn */
1197 static unsigned long next_active_pageblock(unsigned long pfn)
1199 struct page *page = pfn_to_page(pfn);
1201 /* Ensure the starting page is pageblock-aligned */
1202 BUG_ON(pfn & (pageblock_nr_pages - 1));
1204 /* If the entire pageblock is free, move to the end of free page */
1205 if (pageblock_free(page)) {
1207 /* be careful. we don't have locks, page_order can be changed.*/
1208 order = page_order(page);
1209 if ((order < MAX_ORDER) && (order >= pageblock_order))
1210 return pfn + (1 << order);
1213 return pfn + pageblock_nr_pages;
1216 static bool is_pageblock_removable_nolock(unsigned long pfn)
1218 struct page *page = pfn_to_page(pfn);
1222 * We have to be careful here because we are iterating over memory
1223 * sections which are not zone aware so we might end up outside of
1224 * the zone but still within the section.
1225 * We have to take care about the node as well. If the node is offline
1226 * its NODE_DATA will be NULL - see page_zone.
1228 if (!node_online(page_to_nid(page)))
1231 zone = page_zone(page);
1232 pfn = page_to_pfn(page);
1233 if (!zone_spans_pfn(zone, pfn))
1236 return !has_unmovable_pages(zone, page, 0, MIGRATE_MOVABLE, true);
1239 /* Checks if this range of memory is likely to be hot-removable. */
1240 bool is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
1242 unsigned long end_pfn, pfn;
1244 end_pfn = min(start_pfn + nr_pages,
1245 zone_end_pfn(page_zone(pfn_to_page(start_pfn))));
1247 /* Check the starting page of each pageblock within the range */
1248 for (pfn = start_pfn; pfn < end_pfn; pfn = next_active_pageblock(pfn)) {
1249 if (!is_pageblock_removable_nolock(pfn))
1254 /* All pageblocks in the memory block are likely to be hot-removable */
1259 * Confirm all pages in a range [start, end) belong to the same zone.
1260 * When true, return its valid [start, end).
1262 int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn,
1263 unsigned long *valid_start, unsigned long *valid_end)
1265 unsigned long pfn, sec_end_pfn;
1266 unsigned long start, end;
1267 struct zone *zone = NULL;
1270 for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1);
1272 pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) {
1273 /* Make sure the memory section is present first */
1274 if (!present_section_nr(pfn_to_section_nr(pfn)))
1276 for (; pfn < sec_end_pfn && pfn < end_pfn;
1277 pfn += MAX_ORDER_NR_PAGES) {
1279 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1280 while ((i < MAX_ORDER_NR_PAGES) &&
1281 !pfn_valid_within(pfn + i))
1283 if (i == MAX_ORDER_NR_PAGES || pfn + i >= end_pfn)
1285 /* Check if we got outside of the zone */
1286 if (zone && !zone_spans_pfn(zone, pfn + i))
1288 page = pfn_to_page(pfn + i);
1289 if (zone && page_zone(page) != zone)
1293 zone = page_zone(page);
1294 end = pfn + MAX_ORDER_NR_PAGES;
1299 *valid_start = start;
1300 *valid_end = min(end, end_pfn);
1308 * Scan pfn range [start,end) to find movable/migratable pages (LRU pages,
1309 * non-lru movable pages and hugepages). We scan pfn because it's much
1310 * easier than scanning over linked list. This function returns the pfn
1311 * of the first found movable page if it's found, otherwise 0.
1313 static unsigned long scan_movable_pages(unsigned long start, unsigned long end)
1317 for (pfn = start; pfn < end; pfn++) {
1318 struct page *page, *head;
1321 if (!pfn_valid(pfn))
1323 page = pfn_to_page(pfn);
1326 if (__PageMovable(page))
1329 if (!PageHuge(page))
1331 head = compound_head(page);
1332 if (hugepage_migration_supported(page_hstate(head)) &&
1333 page_huge_active(head))
1335 skip = (1 << compound_order(head)) - (page - head);
1341 static struct page *new_node_page(struct page *page, unsigned long private)
1343 int nid = page_to_nid(page);
1344 nodemask_t nmask = node_states[N_MEMORY];
1347 * try to allocate from a different node but reuse this node if there
1348 * are no other online nodes to be used (e.g. we are offlining a part
1349 * of the only existing node)
1351 node_clear(nid, nmask);
1352 if (nodes_empty(nmask))
1353 node_set(nid, nmask);
1355 return new_page_nodemask(page, nid, &nmask);
1358 #define NR_OFFLINE_AT_ONCE_PAGES (256)
1360 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1364 int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
1365 int not_managed = 0;
1369 for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
1370 if (!pfn_valid(pfn))
1372 page = pfn_to_page(pfn);
1374 if (PageHuge(page)) {
1375 struct page *head = compound_head(page);
1376 pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1;
1377 if (compound_order(head) > PFN_SECTION_SHIFT) {
1381 if (isolate_huge_page(page, &source))
1382 move_pages -= 1 << compound_order(head);
1384 } else if (PageTransHuge(page))
1385 pfn = page_to_pfn(compound_head(page))
1386 + hpage_nr_pages(page) - 1;
1389 * HWPoison pages have elevated reference counts so the migration would
1390 * fail on them. It also doesn't make any sense to migrate them in the
1391 * first place. Still try to unmap such a page in case it is still mapped
1392 * (e.g. current hwpoison implementation doesn't unmap KSM pages but keep
1393 * the unmap as the catch all safety net).
1395 if (PageHWPoison(page)) {
1396 if (WARN_ON(PageLRU(page)))
1397 isolate_lru_page(page);
1398 if (page_mapped(page))
1399 try_to_unmap(page, TTU_IGNORE_MLOCK | TTU_IGNORE_ACCESS);
1403 if (!get_page_unless_zero(page))
1406 * We can skip free pages. And we can deal with pages on
1407 * LRU and non-lru movable pages.
1410 ret = isolate_lru_page(page);
1412 ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE);
1413 if (!ret) { /* Success */
1415 list_add_tail(&page->lru, &source);
1417 if (!__PageMovable(page))
1418 inc_node_page_state(page, NR_ISOLATED_ANON +
1419 page_is_file_cache(page));
1422 #ifdef CONFIG_DEBUG_VM
1423 pr_alert("failed to isolate pfn %lx\n", pfn);
1424 dump_page(page, "isolation failed");
1427 /* Because we don't have big zone->lock. we should
1428 check this again here. */
1429 if (page_count(page)) {
1436 if (!list_empty(&source)) {
1438 putback_movable_pages(&source);
1442 /* Allocate a new page from the nearest neighbor node */
1443 ret = migrate_pages(&source, new_node_page, NULL, 0,
1444 MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
1446 putback_movable_pages(&source);
1453 * remove from free_area[] and mark all as Reserved.
1456 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1459 __offline_isolated_pages(start, start + nr_pages);
1464 offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
1466 walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
1467 offline_isolated_pages_cb);
1471 * Check all pages in range, recoreded as memory resource, are isolated.
1474 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1478 long offlined = *(long *)data;
1479 ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true);
1480 offlined = nr_pages;
1482 *(long *)data += offlined;
1487 check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
1492 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
1493 check_pages_isolated_cb);
1495 offlined = (long)ret;
1499 static int __init cmdline_parse_movable_node(char *p)
1501 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1502 movable_node_enabled = true;
1504 pr_warn("movable_node parameter depends on CONFIG_HAVE_MEMBLOCK_NODE_MAP to work properly\n");
1508 early_param("movable_node", cmdline_parse_movable_node);
1510 /* check which state of node_states will be changed when offline memory */
1511 static void node_states_check_changes_offline(unsigned long nr_pages,
1512 struct zone *zone, struct memory_notify *arg)
1514 struct pglist_data *pgdat = zone->zone_pgdat;
1515 unsigned long present_pages = 0;
1516 enum zone_type zt, zone_last = ZONE_NORMAL;
1519 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
1520 * contains nodes which have zones of 0...ZONE_NORMAL,
1521 * set zone_last to ZONE_NORMAL.
1523 * If we don't have HIGHMEM nor movable node,
1524 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
1525 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
1527 if (N_MEMORY == N_NORMAL_MEMORY)
1528 zone_last = ZONE_MOVABLE;
1531 * check whether node_states[N_NORMAL_MEMORY] will be changed.
1532 * If the memory to be offline is in a zone of 0...zone_last,
1533 * and it is the last present memory, 0...zone_last will
1534 * become empty after offline , thus we can determind we will
1535 * need to clear the node from node_states[N_NORMAL_MEMORY].
1537 for (zt = 0; zt <= zone_last; zt++)
1538 present_pages += pgdat->node_zones[zt].present_pages;
1539 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1540 arg->status_change_nid_normal = zone_to_nid(zone);
1542 arg->status_change_nid_normal = -1;
1544 #ifdef CONFIG_HIGHMEM
1546 * If we have movable node, node_states[N_HIGH_MEMORY]
1547 * contains nodes which have zones of 0...ZONE_HIGHMEM,
1548 * set zone_last to ZONE_HIGHMEM.
1550 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
1551 * contains nodes which have zones of 0...ZONE_MOVABLE,
1552 * set zone_last to ZONE_MOVABLE.
1554 zone_last = ZONE_HIGHMEM;
1555 if (N_MEMORY == N_HIGH_MEMORY)
1556 zone_last = ZONE_MOVABLE;
1558 for (; zt <= zone_last; zt++)
1559 present_pages += pgdat->node_zones[zt].present_pages;
1560 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1561 arg->status_change_nid_high = zone_to_nid(zone);
1563 arg->status_change_nid_high = -1;
1565 arg->status_change_nid_high = arg->status_change_nid_normal;
1569 * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE
1571 zone_last = ZONE_MOVABLE;
1574 * check whether node_states[N_HIGH_MEMORY] will be changed
1575 * If we try to offline the last present @nr_pages from the node,
1576 * we can determind we will need to clear the node from
1577 * node_states[N_HIGH_MEMORY].
1579 for (; zt <= zone_last; zt++)
1580 present_pages += pgdat->node_zones[zt].present_pages;
1581 if (nr_pages >= present_pages)
1582 arg->status_change_nid = zone_to_nid(zone);
1584 arg->status_change_nid = -1;
1587 static void node_states_clear_node(int node, struct memory_notify *arg)
1589 if (arg->status_change_nid_normal >= 0)
1590 node_clear_state(node, N_NORMAL_MEMORY);
1592 if ((N_MEMORY != N_NORMAL_MEMORY) &&
1593 (arg->status_change_nid_high >= 0))
1594 node_clear_state(node, N_HIGH_MEMORY);
1596 if ((N_MEMORY != N_HIGH_MEMORY) &&
1597 (arg->status_change_nid >= 0))
1598 node_clear_state(node, N_MEMORY);
1601 static int __ref __offline_pages(unsigned long start_pfn,
1602 unsigned long end_pfn)
1604 unsigned long pfn, nr_pages;
1605 long offlined_pages;
1607 unsigned long flags;
1608 unsigned long valid_start, valid_end;
1610 struct memory_notify arg;
1612 /* at least, alignment against pageblock is necessary */
1613 if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
1615 if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
1617 /* This makes hotplug much easier...and readable.
1618 we assume this for now. .*/
1619 if (!test_pages_in_a_zone(start_pfn, end_pfn, &valid_start, &valid_end))
1622 zone = page_zone(pfn_to_page(valid_start));
1623 node = zone_to_nid(zone);
1624 nr_pages = end_pfn - start_pfn;
1626 /* set above range as isolated */
1627 ret = start_isolate_page_range(start_pfn, end_pfn,
1628 MIGRATE_MOVABLE, true);
1632 arg.start_pfn = start_pfn;
1633 arg.nr_pages = nr_pages;
1634 node_states_check_changes_offline(nr_pages, zone, &arg);
1636 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1637 ret = notifier_to_errno(ret);
1639 goto failed_removal;
1643 /* start memory hot removal */
1645 if (signal_pending(current))
1646 goto failed_removal;
1649 lru_add_drain_all();
1650 drain_all_pages(zone);
1652 pfn = scan_movable_pages(start_pfn, end_pfn);
1653 if (pfn) { /* We have movable pages */
1654 ret = do_migrate_range(pfn, end_pfn);
1659 * dissolve free hugepages in the memory block before doing offlining
1660 * actually in order to make hugetlbfs's object counting consistent.
1662 ret = dissolve_free_huge_pages(start_pfn, end_pfn);
1664 goto failed_removal;
1666 offlined_pages = check_pages_isolated(start_pfn, end_pfn);
1667 if (offlined_pages < 0)
1669 pr_info("Offlined Pages %ld\n", offlined_pages);
1670 /* Ok, all of our target is isolated.
1671 We cannot do rollback at this point. */
1672 offline_isolated_pages(start_pfn, end_pfn);
1673 /* reset pagetype flags and makes migrate type to be MOVABLE */
1674 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1675 /* removal success */
1676 adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages);
1677 zone->present_pages -= offlined_pages;
1679 pgdat_resize_lock(zone->zone_pgdat, &flags);
1680 zone->zone_pgdat->node_present_pages -= offlined_pages;
1681 pgdat_resize_unlock(zone->zone_pgdat, &flags);
1683 init_per_zone_wmark_min();
1685 if (!populated_zone(zone)) {
1686 zone_pcp_reset(zone);
1687 build_all_zonelists(NULL);
1689 zone_pcp_update(zone);
1691 node_states_clear_node(node, &arg);
1692 if (arg.status_change_nid >= 0) {
1694 kcompactd_stop(node);
1697 vm_total_pages = nr_free_pagecache_pages();
1698 writeback_set_ratelimit();
1700 memory_notify(MEM_OFFLINE, &arg);
1704 pr_debug("memory offlining [mem %#010llx-%#010llx] failed\n",
1705 (unsigned long long) start_pfn << PAGE_SHIFT,
1706 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
1707 memory_notify(MEM_CANCEL_OFFLINE, &arg);
1708 /* pushback to free area */
1709 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1713 /* Must be protected by mem_hotplug_begin() or a device_lock */
1714 int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1716 return __offline_pages(start_pfn, start_pfn + nr_pages);
1718 #endif /* CONFIG_MEMORY_HOTREMOVE */
1721 * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
1722 * @start_pfn: start pfn of the memory range
1723 * @end_pfn: end pfn of the memory range
1724 * @arg: argument passed to func
1725 * @func: callback for each memory section walked
1727 * This function walks through all present mem sections in range
1728 * [start_pfn, end_pfn) and call func on each mem section.
1730 * Returns the return value of func.
1732 int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn,
1733 void *arg, int (*func)(struct memory_block *, void *))
1735 struct memory_block *mem = NULL;
1736 struct mem_section *section;
1737 unsigned long pfn, section_nr;
1740 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1741 section_nr = pfn_to_section_nr(pfn);
1742 if (!present_section_nr(section_nr))
1745 section = __nr_to_section(section_nr);
1746 /* same memblock? */
1748 if ((section_nr >= mem->start_section_nr) &&
1749 (section_nr <= mem->end_section_nr))
1752 mem = find_memory_block_hinted(section, mem);
1756 ret = func(mem, arg);
1758 kobject_put(&mem->dev.kobj);
1764 kobject_put(&mem->dev.kobj);
1769 #ifdef CONFIG_MEMORY_HOTREMOVE
1770 static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
1772 int ret = !is_memblock_offlined(mem);
1774 if (unlikely(ret)) {
1775 phys_addr_t beginpa, endpa;
1777 beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
1778 endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1;
1779 pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n",
1786 static int check_cpu_on_node(pg_data_t *pgdat)
1790 for_each_present_cpu(cpu) {
1791 if (cpu_to_node(cpu) == pgdat->node_id)
1793 * the cpu on this node isn't removed, and we can't
1794 * offline this node.
1802 static void unmap_cpu_on_node(pg_data_t *pgdat)
1804 #ifdef CONFIG_ACPI_NUMA
1807 for_each_possible_cpu(cpu)
1808 if (cpu_to_node(cpu) == pgdat->node_id)
1809 numa_clear_node(cpu);
1813 static int check_and_unmap_cpu_on_node(pg_data_t *pgdat)
1817 ret = check_cpu_on_node(pgdat);
1822 * the node will be offlined when we come here, so we can clear
1823 * the cpu_to_node() now.
1826 unmap_cpu_on_node(pgdat);
1834 * Offline a node if all memory sections and cpus of the node are removed.
1836 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1837 * and online/offline operations before this call.
1839 void try_offline_node(int nid)
1841 pg_data_t *pgdat = NODE_DATA(nid);
1842 unsigned long start_pfn = pgdat->node_start_pfn;
1843 unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
1846 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1847 unsigned long section_nr = pfn_to_section_nr(pfn);
1849 if (!present_section_nr(section_nr))
1852 if (pfn_to_nid(pfn) != nid)
1856 * some memory sections of this node are not removed, and we
1857 * can't offline node now.
1862 if (check_and_unmap_cpu_on_node(pgdat))
1866 * all memory/cpu of this node are removed, we can offline this
1869 node_set_offline(nid);
1870 unregister_one_node(nid);
1872 EXPORT_SYMBOL(try_offline_node);
1877 * @start: physical address of the region to remove
1878 * @size: size of the region to remove
1880 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1881 * and online/offline operations before this call, as required by
1882 * try_offline_node().
1884 void __ref remove_memory(int nid, u64 start, u64 size)
1888 BUG_ON(check_hotplug_memory_range(start, size));
1890 mem_hotplug_begin();
1893 * All memory blocks must be offlined before removing memory. Check
1894 * whether all memory blocks in question are offline and trigger a BUG()
1895 * if this is not the case.
1897 ret = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL,
1898 check_memblock_offlined_cb);
1902 /* remove memmap entry */
1903 firmware_map_remove(start, start + size, "System RAM");
1904 memblock_free(start, size);
1905 memblock_remove(start, size);
1907 arch_remove_memory(start, size, NULL);
1909 try_offline_node(nid);
1913 EXPORT_SYMBOL_GPL(remove_memory);
1914 #endif /* CONFIG_MEMORY_HOTREMOVE */