2 * linux/mm/memory_hotplug.c
7 #include <linux/stddef.h>
9 #include <linux/swap.h>
10 #include <linux/interrupt.h>
11 #include <linux/pagemap.h>
12 #include <linux/bootmem.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/memory_hotplug.h>
22 #include <linux/highmem.h>
23 #include <linux/vmalloc.h>
24 #include <linux/ioport.h>
25 #include <linux/delay.h>
26 #include <linux/migrate.h>
27 #include <linux/page-isolation.h>
28 #include <linux/pfn.h>
29 #include <linux/suspend.h>
30 #include <linux/mm_inline.h>
31 #include <linux/firmware-map.h>
32 #include <linux/stop_machine.h>
34 #include <asm/tlbflush.h>
39 * online_page_callback contains pointer to current page onlining function.
40 * Initially it is generic_online_page(). If it is required it could be
41 * changed by calling set_online_page_callback() for callback registration
42 * and restore_online_page_callback() for generic callback restore.
45 static void generic_online_page(struct page *page);
47 static online_page_callback_t online_page_callback = generic_online_page;
49 DEFINE_MUTEX(mem_hotplug_mutex);
51 void lock_memory_hotplug(void)
53 mutex_lock(&mem_hotplug_mutex);
55 /* for exclusive hibernation if CONFIG_HIBERNATION=y */
59 void unlock_memory_hotplug(void)
61 unlock_system_sleep();
62 mutex_unlock(&mem_hotplug_mutex);
66 /* add this memory to iomem resource */
67 static struct resource *register_memory_resource(u64 start, u64 size)
70 res = kzalloc(sizeof(struct resource), GFP_KERNEL);
73 res->name = "System RAM";
75 res->end = start + size - 1;
76 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
77 if (request_resource(&iomem_resource, res) < 0) {
78 printk("System RAM resource %pR cannot be added\n", res);
85 static void release_memory_resource(struct resource *res)
89 release_resource(res);
94 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
95 void get_page_bootmem(unsigned long info, struct page *page,
98 page->lru.next = (struct list_head *) type;
100 set_page_private(page, info);
101 atomic_inc(&page->_count);
104 /* reference to __meminit __free_pages_bootmem is valid
105 * so use __ref to tell modpost not to generate a warning */
106 void __ref put_page_bootmem(struct page *page)
109 static DEFINE_MUTEX(ppb_lock);
111 type = (unsigned long) page->lru.next;
112 BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE ||
113 type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE);
115 if (atomic_dec_return(&page->_count) == 1) {
116 ClearPagePrivate(page);
117 set_page_private(page, 0);
118 INIT_LIST_HEAD(&page->lru);
121 * Please refer to comment for __free_pages_bootmem()
122 * for why we serialize here.
124 mutex_lock(&ppb_lock);
125 __free_pages_bootmem(page, 0);
126 mutex_unlock(&ppb_lock);
132 #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
133 #ifndef CONFIG_SPARSEMEM_VMEMMAP
134 static void register_page_bootmem_info_section(unsigned long start_pfn)
136 unsigned long *usemap, mapsize, section_nr, i;
137 struct mem_section *ms;
138 struct page *page, *memmap;
140 section_nr = pfn_to_section_nr(start_pfn);
141 ms = __nr_to_section(section_nr);
143 /* Get section's memmap address */
144 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
147 * Get page for the memmap's phys address
148 * XXX: need more consideration for sparse_vmemmap...
150 page = virt_to_page(memmap);
151 mapsize = sizeof(struct page) * PAGES_PER_SECTION;
152 mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
154 /* remember memmap's page */
155 for (i = 0; i < mapsize; i++, page++)
156 get_page_bootmem(section_nr, page, SECTION_INFO);
158 usemap = __nr_to_section(section_nr)->pageblock_flags;
159 page = virt_to_page(usemap);
161 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
163 for (i = 0; i < mapsize; i++, page++)
164 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
167 #else /* CONFIG_SPARSEMEM_VMEMMAP */
168 static void register_page_bootmem_info_section(unsigned long start_pfn)
170 unsigned long *usemap, mapsize, section_nr, i;
171 struct mem_section *ms;
172 struct page *page, *memmap;
174 if (!pfn_valid(start_pfn))
177 section_nr = pfn_to_section_nr(start_pfn);
178 ms = __nr_to_section(section_nr);
180 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
182 register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION);
184 usemap = __nr_to_section(section_nr)->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);
192 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
194 void register_page_bootmem_info_node(struct pglist_data *pgdat)
196 unsigned long i, pfn, end_pfn, nr_pages;
197 int node = pgdat->node_id;
201 nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
202 page = virt_to_page(pgdat);
204 for (i = 0; i < nr_pages; i++, page++)
205 get_page_bootmem(node, page, NODE_INFO);
207 zone = &pgdat->node_zones[0];
208 for (; zone < pgdat->node_zones + MAX_NR_ZONES - 1; zone++) {
209 if (zone->wait_table) {
210 nr_pages = zone->wait_table_hash_nr_entries
211 * sizeof(wait_queue_head_t);
212 nr_pages = PAGE_ALIGN(nr_pages) >> PAGE_SHIFT;
213 page = virt_to_page(zone->wait_table);
215 for (i = 0; i < nr_pages; i++, page++)
216 get_page_bootmem(node, page, NODE_INFO);
220 pfn = pgdat->node_start_pfn;
221 end_pfn = pgdat_end_pfn(pgdat);
223 /* register_section info */
224 for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
226 * Some platforms can assign the same pfn to multiple nodes - on
227 * node0 as well as nodeN. To avoid registering a pfn against
228 * multiple nodes we check that this pfn does not already
229 * reside in some other node.
231 if (pfn_valid(pfn) && (pfn_to_nid(pfn) == node))
232 register_page_bootmem_info_section(pfn);
235 #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
237 static void grow_zone_span(struct zone *zone, unsigned long start_pfn,
238 unsigned long end_pfn)
240 unsigned long old_zone_end_pfn;
242 zone_span_writelock(zone);
244 old_zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages;
245 if (!zone->spanned_pages || start_pfn < zone->zone_start_pfn)
246 zone->zone_start_pfn = start_pfn;
248 zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
249 zone->zone_start_pfn;
251 zone_span_writeunlock(zone);
254 static void resize_zone(struct zone *zone, unsigned long start_pfn,
255 unsigned long end_pfn)
257 zone_span_writelock(zone);
259 if (end_pfn - start_pfn) {
260 zone->zone_start_pfn = start_pfn;
261 zone->spanned_pages = end_pfn - start_pfn;
264 * make it consist as free_area_init_core(),
265 * if spanned_pages = 0, then keep start_pfn = 0
267 zone->zone_start_pfn = 0;
268 zone->spanned_pages = 0;
271 zone_span_writeunlock(zone);
274 static void fix_zone_id(struct zone *zone, unsigned long start_pfn,
275 unsigned long end_pfn)
277 enum zone_type zid = zone_idx(zone);
278 int nid = zone->zone_pgdat->node_id;
281 for (pfn = start_pfn; pfn < end_pfn; pfn++)
282 set_page_links(pfn_to_page(pfn), zid, nid, pfn);
285 /* Can fail with -ENOMEM from allocating a wait table with vmalloc() or
286 * alloc_bootmem_node_nopanic() */
287 static int __ref ensure_zone_is_initialized(struct zone *zone,
288 unsigned long start_pfn, unsigned long num_pages)
290 if (!zone_is_initialized(zone))
291 return init_currently_empty_zone(zone, start_pfn, num_pages,
296 static int __meminit move_pfn_range_left(struct zone *z1, struct zone *z2,
297 unsigned long start_pfn, unsigned long end_pfn)
301 unsigned long z1_start_pfn;
303 ret = ensure_zone_is_initialized(z1, start_pfn, end_pfn - start_pfn);
307 pgdat_resize_lock(z1->zone_pgdat, &flags);
309 /* can't move pfns which are higher than @z2 */
310 if (end_pfn > zone_end_pfn(z2))
312 /* the move out part mast at the left most of @z2 */
313 if (start_pfn > z2->zone_start_pfn)
315 /* must included/overlap */
316 if (end_pfn <= z2->zone_start_pfn)
319 /* use start_pfn for z1's start_pfn if z1 is empty */
320 if (z1->spanned_pages)
321 z1_start_pfn = z1->zone_start_pfn;
323 z1_start_pfn = start_pfn;
325 resize_zone(z1, z1_start_pfn, end_pfn);
326 resize_zone(z2, end_pfn, zone_end_pfn(z2));
328 pgdat_resize_unlock(z1->zone_pgdat, &flags);
330 fix_zone_id(z1, start_pfn, end_pfn);
334 pgdat_resize_unlock(z1->zone_pgdat, &flags);
338 static int __meminit move_pfn_range_right(struct zone *z1, struct zone *z2,
339 unsigned long start_pfn, unsigned long end_pfn)
343 unsigned long z2_end_pfn;
345 ret = ensure_zone_is_initialized(z2, start_pfn, end_pfn - start_pfn);
349 pgdat_resize_lock(z1->zone_pgdat, &flags);
351 /* can't move pfns which are lower than @z1 */
352 if (z1->zone_start_pfn > start_pfn)
354 /* the move out part mast at the right most of @z1 */
355 if (zone_end_pfn(z1) > end_pfn)
357 /* must included/overlap */
358 if (start_pfn >= zone_end_pfn(z1))
361 /* use end_pfn for z2's end_pfn if z2 is empty */
362 if (z2->spanned_pages)
363 z2_end_pfn = zone_end_pfn(z2);
365 z2_end_pfn = end_pfn;
367 resize_zone(z1, z1->zone_start_pfn, start_pfn);
368 resize_zone(z2, start_pfn, z2_end_pfn);
370 pgdat_resize_unlock(z1->zone_pgdat, &flags);
372 fix_zone_id(z2, start_pfn, end_pfn);
376 pgdat_resize_unlock(z1->zone_pgdat, &flags);
380 static void grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn,
381 unsigned long end_pfn)
383 unsigned long old_pgdat_end_pfn =
384 pgdat->node_start_pfn + pgdat->node_spanned_pages;
386 if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
387 pgdat->node_start_pfn = start_pfn;
389 pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
390 pgdat->node_start_pfn;
393 static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn)
395 struct pglist_data *pgdat = zone->zone_pgdat;
396 int nr_pages = PAGES_PER_SECTION;
397 int nid = pgdat->node_id;
402 zone_type = zone - pgdat->node_zones;
403 ret = ensure_zone_is_initialized(zone, phys_start_pfn, nr_pages);
407 pgdat_resize_lock(zone->zone_pgdat, &flags);
408 grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages);
409 grow_pgdat_span(zone->zone_pgdat, phys_start_pfn,
410 phys_start_pfn + nr_pages);
411 pgdat_resize_unlock(zone->zone_pgdat, &flags);
412 memmap_init_zone(nr_pages, nid, zone_type,
413 phys_start_pfn, MEMMAP_HOTPLUG);
417 static int __meminit __add_section(int nid, struct zone *zone,
418 unsigned long phys_start_pfn)
420 int nr_pages = PAGES_PER_SECTION;
423 if (pfn_valid(phys_start_pfn))
426 ret = sparse_add_one_section(zone, phys_start_pfn, nr_pages);
431 ret = __add_zone(zone, phys_start_pfn);
436 return register_new_memory(nid, __pfn_to_section(phys_start_pfn));
440 * Reasonably generic function for adding memory. It is
441 * expected that archs that support memory hotplug will
442 * call this function after deciding the zone to which to
445 int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn,
446 unsigned long nr_pages)
450 int start_sec, end_sec;
451 /* during initialize mem_map, align hot-added range to section */
452 start_sec = pfn_to_section_nr(phys_start_pfn);
453 end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
455 for (i = start_sec; i <= end_sec; i++) {
456 err = __add_section(nid, zone, i << PFN_SECTION_SHIFT);
459 * EEXIST is finally dealt with by ioresource collision
460 * check. see add_memory() => register_memory_resource()
461 * Warning will be printed if there is collision.
463 if (err && (err != -EEXIST))
470 EXPORT_SYMBOL_GPL(__add_pages);
472 #ifdef CONFIG_MEMORY_HOTREMOVE
473 /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
474 static int find_smallest_section_pfn(int nid, struct zone *zone,
475 unsigned long start_pfn,
476 unsigned long end_pfn)
478 struct mem_section *ms;
480 for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) {
481 ms = __pfn_to_section(start_pfn);
483 if (unlikely(!valid_section(ms)))
486 if (unlikely(pfn_to_nid(start_pfn) != nid))
489 if (zone && zone != page_zone(pfn_to_page(start_pfn)))
498 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
499 static int find_biggest_section_pfn(int nid, struct zone *zone,
500 unsigned long start_pfn,
501 unsigned long end_pfn)
503 struct mem_section *ms;
506 /* pfn is the end pfn of a memory section. */
508 for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) {
509 ms = __pfn_to_section(pfn);
511 if (unlikely(!valid_section(ms)))
514 if (unlikely(pfn_to_nid(pfn) != nid))
517 if (zone && zone != page_zone(pfn_to_page(pfn)))
526 static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
527 unsigned long end_pfn)
529 unsigned long zone_start_pfn = zone->zone_start_pfn;
530 unsigned long zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages;
532 struct mem_section *ms;
533 int nid = zone_to_nid(zone);
535 zone_span_writelock(zone);
536 if (zone_start_pfn == start_pfn) {
538 * If the section is smallest section in the zone, it need
539 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
540 * In this case, we find second smallest valid mem_section
541 * for shrinking zone.
543 pfn = find_smallest_section_pfn(nid, zone, end_pfn,
546 zone->zone_start_pfn = pfn;
547 zone->spanned_pages = zone_end_pfn - pfn;
549 } else if (zone_end_pfn == end_pfn) {
551 * If the section is biggest section in the zone, it need
552 * shrink zone->spanned_pages.
553 * In this case, we find second biggest valid mem_section for
556 pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn,
559 zone->spanned_pages = pfn - zone_start_pfn + 1;
563 * The section is not biggest or smallest mem_section in the zone, it
564 * only creates a hole in the zone. So in this case, we need not
565 * change the zone. But perhaps, the zone has only hole data. Thus
566 * it check the zone has only hole or not.
568 pfn = zone_start_pfn;
569 for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) {
570 ms = __pfn_to_section(pfn);
572 if (unlikely(!valid_section(ms)))
575 if (page_zone(pfn_to_page(pfn)) != zone)
578 /* If the section is current section, it continues the loop */
579 if (start_pfn == pfn)
582 /* If we find valid section, we have nothing to do */
583 zone_span_writeunlock(zone);
587 /* The zone has no valid section */
588 zone->zone_start_pfn = 0;
589 zone->spanned_pages = 0;
590 zone_span_writeunlock(zone);
593 static void shrink_pgdat_span(struct pglist_data *pgdat,
594 unsigned long start_pfn, unsigned long end_pfn)
596 unsigned long pgdat_start_pfn = pgdat->node_start_pfn;
597 unsigned long pgdat_end_pfn =
598 pgdat->node_start_pfn + pgdat->node_spanned_pages;
600 struct mem_section *ms;
601 int nid = pgdat->node_id;
603 if (pgdat_start_pfn == start_pfn) {
605 * If the section is smallest section in the pgdat, it need
606 * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages.
607 * In this case, we find second smallest valid mem_section
608 * for shrinking zone.
610 pfn = find_smallest_section_pfn(nid, NULL, end_pfn,
613 pgdat->node_start_pfn = pfn;
614 pgdat->node_spanned_pages = pgdat_end_pfn - pfn;
616 } else if (pgdat_end_pfn == end_pfn) {
618 * If the section is biggest section in the pgdat, it need
619 * shrink pgdat->node_spanned_pages.
620 * In this case, we find second biggest valid mem_section for
623 pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn,
626 pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1;
630 * If the section is not biggest or smallest mem_section in the pgdat,
631 * it only creates a hole in the pgdat. So in this case, we need not
633 * But perhaps, the pgdat has only hole data. Thus it check the pgdat
634 * has only hole or not.
636 pfn = pgdat_start_pfn;
637 for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) {
638 ms = __pfn_to_section(pfn);
640 if (unlikely(!valid_section(ms)))
643 if (pfn_to_nid(pfn) != nid)
646 /* If the section is current section, it continues the loop */
647 if (start_pfn == pfn)
650 /* If we find valid section, we have nothing to do */
654 /* The pgdat has no valid section */
655 pgdat->node_start_pfn = 0;
656 pgdat->node_spanned_pages = 0;
659 static void __remove_zone(struct zone *zone, unsigned long start_pfn)
661 struct pglist_data *pgdat = zone->zone_pgdat;
662 int nr_pages = PAGES_PER_SECTION;
666 zone_type = zone - pgdat->node_zones;
668 pgdat_resize_lock(zone->zone_pgdat, &flags);
669 shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
670 shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages);
671 pgdat_resize_unlock(zone->zone_pgdat, &flags);
674 static int __remove_section(struct zone *zone, struct mem_section *ms)
676 unsigned long start_pfn;
680 if (!valid_section(ms))
683 ret = unregister_memory_section(ms);
687 scn_nr = __section_nr(ms);
688 start_pfn = section_nr_to_pfn(scn_nr);
689 __remove_zone(zone, start_pfn);
691 sparse_remove_one_section(zone, ms);
696 * __remove_pages() - remove sections of pages from a zone
697 * @zone: zone from which pages need to be removed
698 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
699 * @nr_pages: number of pages to remove (must be multiple of section size)
701 * Generic helper function to remove section mappings and sysfs entries
702 * for the section of the memory we are removing. Caller needs to make
703 * sure that pages are marked reserved and zones are adjust properly by
704 * calling offline_pages().
706 int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
707 unsigned long nr_pages)
710 int sections_to_remove;
711 resource_size_t start, size;
715 * We can only remove entire sections
717 BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
718 BUG_ON(nr_pages % PAGES_PER_SECTION);
720 start = phys_start_pfn << PAGE_SHIFT;
721 size = nr_pages * PAGE_SIZE;
722 ret = release_mem_region_adjustable(&iomem_resource, start, size);
724 pr_warn("Unable to release resource <%016llx-%016llx> (%d)\n",
725 start, start + size - 1, ret);
727 sections_to_remove = nr_pages / PAGES_PER_SECTION;
728 for (i = 0; i < sections_to_remove; i++) {
729 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
730 ret = __remove_section(zone, __pfn_to_section(pfn));
736 EXPORT_SYMBOL_GPL(__remove_pages);
737 #endif /* CONFIG_MEMORY_HOTREMOVE */
739 int set_online_page_callback(online_page_callback_t callback)
743 lock_memory_hotplug();
745 if (online_page_callback == generic_online_page) {
746 online_page_callback = callback;
750 unlock_memory_hotplug();
754 EXPORT_SYMBOL_GPL(set_online_page_callback);
756 int restore_online_page_callback(online_page_callback_t callback)
760 lock_memory_hotplug();
762 if (online_page_callback == callback) {
763 online_page_callback = generic_online_page;
767 unlock_memory_hotplug();
771 EXPORT_SYMBOL_GPL(restore_online_page_callback);
773 void __online_page_set_limits(struct page *page)
775 unsigned long pfn = page_to_pfn(page);
777 if (pfn >= num_physpages)
778 num_physpages = pfn + 1;
780 EXPORT_SYMBOL_GPL(__online_page_set_limits);
782 void __online_page_increment_counters(struct page *page)
786 #ifdef CONFIG_HIGHMEM
787 if (PageHighMem(page))
791 EXPORT_SYMBOL_GPL(__online_page_increment_counters);
793 void __online_page_free(struct page *page)
795 ClearPageReserved(page);
796 init_page_count(page);
799 EXPORT_SYMBOL_GPL(__online_page_free);
801 static void generic_online_page(struct page *page)
803 __online_page_set_limits(page);
804 __online_page_increment_counters(page);
805 __online_page_free(page);
808 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
812 unsigned long onlined_pages = *(unsigned long *)arg;
814 if (PageReserved(pfn_to_page(start_pfn)))
815 for (i = 0; i < nr_pages; i++) {
816 page = pfn_to_page(start_pfn + i);
817 (*online_page_callback)(page);
820 *(unsigned long *)arg = onlined_pages;
824 #ifdef CONFIG_MOVABLE_NODE
826 * When CONFIG_MOVABLE_NODE, we permit onlining of a node which doesn't have
829 static bool can_online_high_movable(struct zone *zone)
833 #else /* CONFIG_MOVABLE_NODE */
834 /* ensure every online node has NORMAL memory */
835 static bool can_online_high_movable(struct zone *zone)
837 return node_state(zone_to_nid(zone), N_NORMAL_MEMORY);
839 #endif /* CONFIG_MOVABLE_NODE */
841 /* check which state of node_states will be changed when online memory */
842 static void node_states_check_changes_online(unsigned long nr_pages,
843 struct zone *zone, struct memory_notify *arg)
845 int nid = zone_to_nid(zone);
846 enum zone_type zone_last = ZONE_NORMAL;
849 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
850 * contains nodes which have zones of 0...ZONE_NORMAL,
851 * set zone_last to ZONE_NORMAL.
853 * If we don't have HIGHMEM nor movable node,
854 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
855 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
857 if (N_MEMORY == N_NORMAL_MEMORY)
858 zone_last = ZONE_MOVABLE;
861 * if the memory to be online is in a zone of 0...zone_last, and
862 * the zones of 0...zone_last don't have memory before online, we will
863 * need to set the node to node_states[N_NORMAL_MEMORY] after
864 * the memory is online.
866 if (zone_idx(zone) <= zone_last && !node_state(nid, N_NORMAL_MEMORY))
867 arg->status_change_nid_normal = nid;
869 arg->status_change_nid_normal = -1;
871 #ifdef CONFIG_HIGHMEM
873 * If we have movable node, node_states[N_HIGH_MEMORY]
874 * contains nodes which have zones of 0...ZONE_HIGHMEM,
875 * set zone_last to ZONE_HIGHMEM.
877 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
878 * contains nodes which have zones of 0...ZONE_MOVABLE,
879 * set zone_last to ZONE_MOVABLE.
881 zone_last = ZONE_HIGHMEM;
882 if (N_MEMORY == N_HIGH_MEMORY)
883 zone_last = ZONE_MOVABLE;
885 if (zone_idx(zone) <= zone_last && !node_state(nid, N_HIGH_MEMORY))
886 arg->status_change_nid_high = nid;
888 arg->status_change_nid_high = -1;
890 arg->status_change_nid_high = arg->status_change_nid_normal;
894 * if the node don't have memory befor online, we will need to
895 * set the node to node_states[N_MEMORY] after the memory
898 if (!node_state(nid, N_MEMORY))
899 arg->status_change_nid = nid;
901 arg->status_change_nid = -1;
904 static void node_states_set_node(int node, struct memory_notify *arg)
906 if (arg->status_change_nid_normal >= 0)
907 node_set_state(node, N_NORMAL_MEMORY);
909 if (arg->status_change_nid_high >= 0)
910 node_set_state(node, N_HIGH_MEMORY);
912 node_set_state(node, N_MEMORY);
916 int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type)
918 unsigned long onlined_pages = 0;
920 int need_zonelists_rebuild = 0;
923 struct memory_notify arg;
925 lock_memory_hotplug();
927 * This doesn't need a lock to do pfn_to_page().
928 * The section can't be removed here because of the
929 * memory_block->state_mutex.
931 zone = page_zone(pfn_to_page(pfn));
933 if ((zone_idx(zone) > ZONE_NORMAL || online_type == ONLINE_MOVABLE) &&
934 !can_online_high_movable(zone)) {
935 unlock_memory_hotplug();
939 if (online_type == ONLINE_KERNEL && zone_idx(zone) == ZONE_MOVABLE) {
940 if (move_pfn_range_left(zone - 1, zone, pfn, pfn + nr_pages)) {
941 unlock_memory_hotplug();
945 if (online_type == ONLINE_MOVABLE && zone_idx(zone) == ZONE_MOVABLE - 1) {
946 if (move_pfn_range_right(zone, zone + 1, pfn, pfn + nr_pages)) {
947 unlock_memory_hotplug();
952 /* Previous code may changed the zone of the pfn range */
953 zone = page_zone(pfn_to_page(pfn));
956 arg.nr_pages = nr_pages;
957 node_states_check_changes_online(nr_pages, zone, &arg);
959 nid = page_to_nid(pfn_to_page(pfn));
961 ret = memory_notify(MEM_GOING_ONLINE, &arg);
962 ret = notifier_to_errno(ret);
964 memory_notify(MEM_CANCEL_ONLINE, &arg);
965 unlock_memory_hotplug();
969 * If this zone is not populated, then it is not in zonelist.
970 * This means the page allocator ignores this zone.
971 * So, zonelist must be updated after online.
973 mutex_lock(&zonelists_mutex);
974 if (!populated_zone(zone)) {
975 need_zonelists_rebuild = 1;
976 build_all_zonelists(NULL, zone);
979 ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
982 if (need_zonelists_rebuild)
983 zone_pcp_reset(zone);
984 mutex_unlock(&zonelists_mutex);
985 printk(KERN_DEBUG "online_pages [mem %#010llx-%#010llx] failed\n",
986 (unsigned long long) pfn << PAGE_SHIFT,
987 (((unsigned long long) pfn + nr_pages)
989 memory_notify(MEM_CANCEL_ONLINE, &arg);
990 unlock_memory_hotplug();
994 zone->managed_pages += onlined_pages;
995 zone->present_pages += onlined_pages;
996 zone->zone_pgdat->node_present_pages += onlined_pages;
998 node_states_set_node(zone_to_nid(zone), &arg);
999 if (need_zonelists_rebuild)
1000 build_all_zonelists(NULL, NULL);
1002 zone_pcp_update(zone);
1005 mutex_unlock(&zonelists_mutex);
1007 init_per_zone_wmark_min();
1010 kswapd_run(zone_to_nid(zone));
1012 vm_total_pages = nr_free_pagecache_pages();
1014 writeback_set_ratelimit();
1017 memory_notify(MEM_ONLINE, &arg);
1018 unlock_memory_hotplug();
1022 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
1024 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1025 static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
1027 struct pglist_data *pgdat;
1028 unsigned long zones_size[MAX_NR_ZONES] = {0};
1029 unsigned long zholes_size[MAX_NR_ZONES] = {0};
1030 unsigned long start_pfn = start >> PAGE_SHIFT;
1032 pgdat = NODE_DATA(nid);
1034 pgdat = arch_alloc_nodedata(nid);
1038 arch_refresh_nodedata(nid, pgdat);
1041 /* we can use NODE_DATA(nid) from here */
1043 /* init node's zones as empty zones, we don't have any present pages.*/
1044 free_area_init_node(nid, zones_size, start_pfn, zholes_size);
1047 * The node we allocated has no zone fallback lists. For avoiding
1048 * to access not-initialized zonelist, build here.
1050 mutex_lock(&zonelists_mutex);
1051 build_all_zonelists(pgdat, NULL);
1052 mutex_unlock(&zonelists_mutex);
1057 static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
1059 arch_refresh_nodedata(nid, NULL);
1060 arch_free_nodedata(pgdat);
1066 * called by cpu_up() to online a node without onlined memory.
1068 int mem_online_node(int nid)
1073 lock_memory_hotplug();
1074 pgdat = hotadd_new_pgdat(nid, 0);
1079 node_set_online(nid);
1080 ret = register_one_node(nid);
1084 unlock_memory_hotplug();
1088 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1089 int __ref add_memory(int nid, u64 start, u64 size)
1091 pg_data_t *pgdat = NULL;
1094 struct resource *res;
1097 lock_memory_hotplug();
1099 res = register_memory_resource(start, size);
1104 { /* Stupid hack to suppress address-never-null warning */
1105 void *p = NODE_DATA(nid);
1108 new_node = !node_online(nid);
1110 pgdat = hotadd_new_pgdat(nid, start);
1116 /* call arch's memory hotadd */
1117 ret = arch_add_memory(nid, start, size);
1122 /* we online node here. we can't roll back from here. */
1123 node_set_online(nid);
1126 ret = register_one_node(nid);
1128 * If sysfs file of new node can't create, cpu on the node
1129 * can't be hot-added. There is no rollback way now.
1130 * So, check by BUG_ON() to catch it reluctantly..
1135 /* create new memmap entry */
1136 firmware_map_add_hotplug(start, start + size, "System RAM");
1141 /* rollback pgdat allocation and others */
1143 rollback_node_hotadd(nid, pgdat);
1144 release_memory_resource(res);
1147 unlock_memory_hotplug();
1150 EXPORT_SYMBOL_GPL(add_memory);
1152 #ifdef CONFIG_MEMORY_HOTREMOVE
1154 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1155 * set and the size of the free page is given by page_order(). Using this,
1156 * the function determines if the pageblock contains only free pages.
1157 * Due to buddy contraints, a free page at least the size of a pageblock will
1158 * be located at the start of the pageblock
1160 static inline int pageblock_free(struct page *page)
1162 return PageBuddy(page) && page_order(page) >= pageblock_order;
1165 /* Return the start of the next active pageblock after a given page */
1166 static struct page *next_active_pageblock(struct page *page)
1168 /* Ensure the starting page is pageblock-aligned */
1169 BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
1171 /* If the entire pageblock is free, move to the end of free page */
1172 if (pageblock_free(page)) {
1174 /* be careful. we don't have locks, page_order can be changed.*/
1175 order = page_order(page);
1176 if ((order < MAX_ORDER) && (order >= pageblock_order))
1177 return page + (1 << order);
1180 return page + pageblock_nr_pages;
1183 /* Checks if this range of memory is likely to be hot-removable. */
1184 int is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
1186 struct page *page = pfn_to_page(start_pfn);
1187 struct page *end_page = page + nr_pages;
1189 /* Check the starting page of each pageblock within the range */
1190 for (; page < end_page; page = next_active_pageblock(page)) {
1191 if (!is_pageblock_removable_nolock(page))
1196 /* All pageblocks in the memory block are likely to be hot-removable */
1201 * Confirm all pages in a range [start, end) is belongs to the same zone.
1203 static int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn)
1206 struct zone *zone = NULL;
1209 for (pfn = start_pfn;
1211 pfn += MAX_ORDER_NR_PAGES) {
1213 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1214 while ((i < MAX_ORDER_NR_PAGES) && !pfn_valid_within(pfn + i))
1216 if (i == MAX_ORDER_NR_PAGES)
1218 page = pfn_to_page(pfn + i);
1219 if (zone && page_zone(page) != zone)
1221 zone = page_zone(page);
1227 * Scanning pfn is much easier than scanning lru list.
1228 * Scan pfn from start to end and Find LRU page.
1230 static unsigned long scan_lru_pages(unsigned long start, unsigned long end)
1234 for (pfn = start; pfn < end; pfn++) {
1235 if (pfn_valid(pfn)) {
1236 page = pfn_to_page(pfn);
1244 #define NR_OFFLINE_AT_ONCE_PAGES (256)
1246 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1250 int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
1251 int not_managed = 0;
1255 for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
1256 if (!pfn_valid(pfn))
1258 page = pfn_to_page(pfn);
1259 if (!get_page_unless_zero(page))
1262 * We can skip free pages. And we can only deal with pages on
1265 ret = isolate_lru_page(page);
1266 if (!ret) { /* Success */
1268 list_add_tail(&page->lru, &source);
1270 inc_zone_page_state(page, NR_ISOLATED_ANON +
1271 page_is_file_cache(page));
1274 #ifdef CONFIG_DEBUG_VM
1275 printk(KERN_ALERT "removing pfn %lx from LRU failed\n",
1280 /* Because we don't have big zone->lock. we should
1281 check this again here. */
1282 if (page_count(page)) {
1289 if (!list_empty(&source)) {
1291 putback_lru_pages(&source);
1296 * alloc_migrate_target should be improooooved!!
1297 * migrate_pages returns # of failed pages.
1299 ret = migrate_pages(&source, alloc_migrate_target, 0,
1300 MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
1302 putback_lru_pages(&source);
1309 * remove from free_area[] and mark all as Reserved.
1312 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1315 __offline_isolated_pages(start, start + nr_pages);
1320 offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
1322 walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
1323 offline_isolated_pages_cb);
1327 * Check all pages in range, recoreded as memory resource, are isolated.
1330 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1334 long offlined = *(long *)data;
1335 ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true);
1336 offlined = nr_pages;
1338 *(long *)data += offlined;
1343 check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
1348 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
1349 check_pages_isolated_cb);
1351 offlined = (long)ret;
1355 #ifdef CONFIG_MOVABLE_NODE
1357 * When CONFIG_MOVABLE_NODE, we permit offlining of a node which doesn't have
1360 static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1364 #else /* CONFIG_MOVABLE_NODE */
1365 /* ensure the node has NORMAL memory if it is still online */
1366 static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1368 struct pglist_data *pgdat = zone->zone_pgdat;
1369 unsigned long present_pages = 0;
1372 for (zt = 0; zt <= ZONE_NORMAL; zt++)
1373 present_pages += pgdat->node_zones[zt].present_pages;
1375 if (present_pages > nr_pages)
1379 for (; zt <= ZONE_MOVABLE; zt++)
1380 present_pages += pgdat->node_zones[zt].present_pages;
1383 * we can't offline the last normal memory until all
1384 * higher memory is offlined.
1386 return present_pages == 0;
1388 #endif /* CONFIG_MOVABLE_NODE */
1390 /* check which state of node_states will be changed when offline memory */
1391 static void node_states_check_changes_offline(unsigned long nr_pages,
1392 struct zone *zone, struct memory_notify *arg)
1394 struct pglist_data *pgdat = zone->zone_pgdat;
1395 unsigned long present_pages = 0;
1396 enum zone_type zt, zone_last = ZONE_NORMAL;
1399 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
1400 * contains nodes which have zones of 0...ZONE_NORMAL,
1401 * set zone_last to ZONE_NORMAL.
1403 * If we don't have HIGHMEM nor movable node,
1404 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
1405 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
1407 if (N_MEMORY == N_NORMAL_MEMORY)
1408 zone_last = ZONE_MOVABLE;
1411 * check whether node_states[N_NORMAL_MEMORY] will be changed.
1412 * If the memory to be offline is in a zone of 0...zone_last,
1413 * and it is the last present memory, 0...zone_last will
1414 * become empty after offline , thus we can determind we will
1415 * need to clear the node from node_states[N_NORMAL_MEMORY].
1417 for (zt = 0; zt <= zone_last; zt++)
1418 present_pages += pgdat->node_zones[zt].present_pages;
1419 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1420 arg->status_change_nid_normal = zone_to_nid(zone);
1422 arg->status_change_nid_normal = -1;
1424 #ifdef CONFIG_HIGHMEM
1426 * If we have movable node, node_states[N_HIGH_MEMORY]
1427 * contains nodes which have zones of 0...ZONE_HIGHMEM,
1428 * set zone_last to ZONE_HIGHMEM.
1430 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
1431 * contains nodes which have zones of 0...ZONE_MOVABLE,
1432 * set zone_last to ZONE_MOVABLE.
1434 zone_last = ZONE_HIGHMEM;
1435 if (N_MEMORY == N_HIGH_MEMORY)
1436 zone_last = ZONE_MOVABLE;
1438 for (; zt <= zone_last; zt++)
1439 present_pages += pgdat->node_zones[zt].present_pages;
1440 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1441 arg->status_change_nid_high = zone_to_nid(zone);
1443 arg->status_change_nid_high = -1;
1445 arg->status_change_nid_high = arg->status_change_nid_normal;
1449 * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE
1451 zone_last = ZONE_MOVABLE;
1454 * check whether node_states[N_HIGH_MEMORY] will be changed
1455 * If we try to offline the last present @nr_pages from the node,
1456 * we can determind we will need to clear the node from
1457 * node_states[N_HIGH_MEMORY].
1459 for (; zt <= zone_last; zt++)
1460 present_pages += pgdat->node_zones[zt].present_pages;
1461 if (nr_pages >= present_pages)
1462 arg->status_change_nid = zone_to_nid(zone);
1464 arg->status_change_nid = -1;
1467 static void node_states_clear_node(int node, struct memory_notify *arg)
1469 if (arg->status_change_nid_normal >= 0)
1470 node_clear_state(node, N_NORMAL_MEMORY);
1472 if ((N_MEMORY != N_NORMAL_MEMORY) &&
1473 (arg->status_change_nid_high >= 0))
1474 node_clear_state(node, N_HIGH_MEMORY);
1476 if ((N_MEMORY != N_HIGH_MEMORY) &&
1477 (arg->status_change_nid >= 0))
1478 node_clear_state(node, N_MEMORY);
1481 static int __ref __offline_pages(unsigned long start_pfn,
1482 unsigned long end_pfn, unsigned long timeout)
1484 unsigned long pfn, nr_pages, expire;
1485 long offlined_pages;
1486 int ret, drain, retry_max, node;
1488 struct memory_notify arg;
1490 BUG_ON(start_pfn >= end_pfn);
1491 /* at least, alignment against pageblock is necessary */
1492 if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
1494 if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
1496 /* This makes hotplug much easier...and readable.
1497 we assume this for now. .*/
1498 if (!test_pages_in_a_zone(start_pfn, end_pfn))
1501 lock_memory_hotplug();
1503 zone = page_zone(pfn_to_page(start_pfn));
1504 node = zone_to_nid(zone);
1505 nr_pages = end_pfn - start_pfn;
1508 if (zone_idx(zone) <= ZONE_NORMAL && !can_offline_normal(zone, nr_pages))
1511 /* set above range as isolated */
1512 ret = start_isolate_page_range(start_pfn, end_pfn,
1513 MIGRATE_MOVABLE, true);
1517 arg.start_pfn = start_pfn;
1518 arg.nr_pages = nr_pages;
1519 node_states_check_changes_offline(nr_pages, zone, &arg);
1521 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1522 ret = notifier_to_errno(ret);
1524 goto failed_removal;
1527 expire = jiffies + timeout;
1531 /* start memory hot removal */
1533 if (time_after(jiffies, expire))
1534 goto failed_removal;
1536 if (signal_pending(current))
1537 goto failed_removal;
1540 lru_add_drain_all();
1545 pfn = scan_lru_pages(start_pfn, end_pfn);
1546 if (pfn) { /* We have page on LRU */
1547 ret = do_migrate_range(pfn, end_pfn);
1553 if (--retry_max == 0)
1554 goto failed_removal;
1560 /* drain all zone's lru pagevec, this is asynchronous... */
1561 lru_add_drain_all();
1563 /* drain pcp pages, this is synchronous. */
1566 offlined_pages = check_pages_isolated(start_pfn, end_pfn);
1567 if (offlined_pages < 0) {
1569 goto failed_removal;
1571 printk(KERN_INFO "Offlined Pages %ld\n", offlined_pages);
1572 /* Ok, all of our target is isolated.
1573 We cannot do rollback at this point. */
1574 offline_isolated_pages(start_pfn, end_pfn);
1575 /* reset pagetype flags and makes migrate type to be MOVABLE */
1576 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1577 /* removal success */
1578 zone->managed_pages -= offlined_pages;
1579 zone->present_pages -= offlined_pages;
1580 zone->zone_pgdat->node_present_pages -= offlined_pages;
1581 totalram_pages -= offlined_pages;
1583 init_per_zone_wmark_min();
1585 if (!populated_zone(zone)) {
1586 zone_pcp_reset(zone);
1587 mutex_lock(&zonelists_mutex);
1588 build_all_zonelists(NULL, NULL);
1589 mutex_unlock(&zonelists_mutex);
1591 zone_pcp_update(zone);
1593 node_states_clear_node(node, &arg);
1594 if (arg.status_change_nid >= 0)
1597 vm_total_pages = nr_free_pagecache_pages();
1598 writeback_set_ratelimit();
1600 memory_notify(MEM_OFFLINE, &arg);
1601 unlock_memory_hotplug();
1605 printk(KERN_INFO "memory offlining [mem %#010llx-%#010llx] failed\n",
1606 (unsigned long long) start_pfn << PAGE_SHIFT,
1607 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
1608 memory_notify(MEM_CANCEL_OFFLINE, &arg);
1609 /* pushback to free area */
1610 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1613 unlock_memory_hotplug();
1617 int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1619 return __offline_pages(start_pfn, start_pfn + nr_pages, 120 * HZ);
1623 * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
1624 * @start_pfn: start pfn of the memory range
1625 * @end_pfn: end pfn of the memory range
1626 * @arg: argument passed to func
1627 * @func: callback for each memory section walked
1629 * This function walks through all present mem sections in range
1630 * [start_pfn, end_pfn) and call func on each mem section.
1632 * Returns the return value of func.
1634 static int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn,
1635 void *arg, int (*func)(struct memory_block *, void *))
1637 struct memory_block *mem = NULL;
1638 struct mem_section *section;
1639 unsigned long pfn, section_nr;
1642 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1643 section_nr = pfn_to_section_nr(pfn);
1644 if (!present_section_nr(section_nr))
1647 section = __nr_to_section(section_nr);
1648 /* same memblock? */
1650 if ((section_nr >= mem->start_section_nr) &&
1651 (section_nr <= mem->end_section_nr))
1654 mem = find_memory_block_hinted(section, mem);
1658 ret = func(mem, arg);
1660 kobject_put(&mem->dev.kobj);
1666 kobject_put(&mem->dev.kobj);
1672 * offline_memory_block_cb - callback function for offlining memory block
1673 * @mem: the memory block to be offlined
1674 * @arg: buffer to hold error msg
1676 * Always return 0, and put the error msg in arg if any.
1678 static int offline_memory_block_cb(struct memory_block *mem, void *arg)
1681 int error = offline_memory_block(mem);
1683 if (error != 0 && *ret == 0)
1689 static int is_memblock_offlined_cb(struct memory_block *mem, void *arg)
1691 int ret = !is_memblock_offlined(mem);
1693 if (unlikely(ret)) {
1694 phys_addr_t beginpa, endpa;
1696 beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
1697 endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1;
1698 pr_warn("removing memory fails, because memory "
1699 "[%pa-%pa] is onlined\n",
1706 static int check_cpu_on_node(void *data)
1708 struct pglist_data *pgdat = data;
1711 for_each_present_cpu(cpu) {
1712 if (cpu_to_node(cpu) == pgdat->node_id)
1714 * the cpu on this node isn't removed, and we can't
1715 * offline this node.
1723 static void unmap_cpu_on_node(void *data)
1725 #ifdef CONFIG_ACPI_NUMA
1726 struct pglist_data *pgdat = data;
1729 for_each_possible_cpu(cpu)
1730 if (cpu_to_node(cpu) == pgdat->node_id)
1731 numa_clear_node(cpu);
1735 static int check_and_unmap_cpu_on_node(void *data)
1737 int ret = check_cpu_on_node(data);
1743 * the node will be offlined when we come here, so we can clear
1744 * the cpu_to_node() now.
1747 unmap_cpu_on_node(data);
1751 /* offline the node if all memory sections of this node are removed */
1752 void try_offline_node(int nid)
1754 pg_data_t *pgdat = NODE_DATA(nid);
1755 unsigned long start_pfn = pgdat->node_start_pfn;
1756 unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
1758 struct page *pgdat_page = virt_to_page(pgdat);
1761 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1762 unsigned long section_nr = pfn_to_section_nr(pfn);
1764 if (!present_section_nr(section_nr))
1767 if (pfn_to_nid(pfn) != nid)
1771 * some memory sections of this node are not removed, and we
1772 * can't offline node now.
1777 if (stop_machine(check_and_unmap_cpu_on_node, pgdat, NULL))
1781 * all memory/cpu of this node are removed, we can offline this
1784 node_set_offline(nid);
1785 unregister_one_node(nid);
1787 if (!PageSlab(pgdat_page) && !PageCompound(pgdat_page))
1788 /* node data is allocated from boot memory */
1791 /* free waittable in each zone */
1792 for (i = 0; i < MAX_NR_ZONES; i++) {
1793 struct zone *zone = pgdat->node_zones + i;
1796 * wait_table may be allocated from boot memory,
1797 * here only free if it's allocated by vmalloc.
1799 if (is_vmalloc_addr(zone->wait_table))
1800 vfree(zone->wait_table);
1804 * Since there is no way to guarentee the address of pgdat/zone is not
1805 * on stack of any kernel threads or used by other kernel objects
1806 * without reference counting or other symchronizing method, do not
1807 * reset node_data and free pgdat here. Just reset it to 0 and reuse
1808 * the memory when the node is online again.
1810 memset(pgdat, 0, sizeof(*pgdat));
1812 EXPORT_SYMBOL(try_offline_node);
1814 int __ref remove_memory(int nid, u64 start, u64 size)
1816 unsigned long start_pfn, end_pfn;
1820 start_pfn = PFN_DOWN(start);
1821 end_pfn = PFN_UP(start + size - 1);
1824 * When CONFIG_MEMCG is on, one memory block may be used by other
1825 * blocks to store page cgroup when onlining pages. But we don't know
1826 * in what order pages are onlined. So we iterate twice to offline
1828 * 1st iterate: offline every non primary memory block.
1829 * 2nd iterate: offline primary (i.e. first added) memory block.
1832 walk_memory_range(start_pfn, end_pfn, &ret,
1833 offline_memory_block_cb);
1843 lock_memory_hotplug();
1846 * we have offlined all memory blocks like this:
1847 * 1. lock memory hotplug
1848 * 2. offline a memory block
1849 * 3. unlock memory hotplug
1851 * repeat step1-3 to offline the memory block. All memory blocks
1852 * must be offlined before removing memory. But we don't hold the
1853 * lock in the whole operation. So we should check whether all
1854 * memory blocks are offlined.
1857 ret = walk_memory_range(start_pfn, end_pfn, NULL,
1858 is_memblock_offlined_cb);
1860 unlock_memory_hotplug();
1864 /* remove memmap entry */
1865 firmware_map_remove(start, start + size, "System RAM");
1867 arch_remove_memory(start, size);
1869 try_offline_node(nid);
1871 unlock_memory_hotplug();
1876 int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1880 int remove_memory(int nid, u64 start, u64 size)
1884 #endif /* CONFIG_MEMORY_HOTREMOVE */
1885 EXPORT_SYMBOL_GPL(remove_memory);