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));
439 /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
440 static int find_smallest_section_pfn(int nid, struct zone *zone,
441 unsigned long start_pfn,
442 unsigned long end_pfn)
444 struct mem_section *ms;
446 for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) {
447 ms = __pfn_to_section(start_pfn);
449 if (unlikely(!valid_section(ms)))
452 if (unlikely(pfn_to_nid(start_pfn) != nid))
455 if (zone && zone != page_zone(pfn_to_page(start_pfn)))
464 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
465 static int find_biggest_section_pfn(int nid, struct zone *zone,
466 unsigned long start_pfn,
467 unsigned long end_pfn)
469 struct mem_section *ms;
472 /* pfn is the end pfn of a memory section. */
474 for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) {
475 ms = __pfn_to_section(pfn);
477 if (unlikely(!valid_section(ms)))
480 if (unlikely(pfn_to_nid(pfn) != nid))
483 if (zone && zone != page_zone(pfn_to_page(pfn)))
492 static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
493 unsigned long end_pfn)
495 unsigned long zone_start_pfn = zone->zone_start_pfn;
496 unsigned long zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages;
498 struct mem_section *ms;
499 int nid = zone_to_nid(zone);
501 zone_span_writelock(zone);
502 if (zone_start_pfn == start_pfn) {
504 * If the section is smallest section in the zone, it need
505 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
506 * In this case, we find second smallest valid mem_section
507 * for shrinking zone.
509 pfn = find_smallest_section_pfn(nid, zone, end_pfn,
512 zone->zone_start_pfn = pfn;
513 zone->spanned_pages = zone_end_pfn - pfn;
515 } else if (zone_end_pfn == end_pfn) {
517 * If the section is biggest section in the zone, it need
518 * shrink zone->spanned_pages.
519 * In this case, we find second biggest valid mem_section for
522 pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn,
525 zone->spanned_pages = pfn - zone_start_pfn + 1;
529 * The section is not biggest or smallest mem_section in the zone, it
530 * only creates a hole in the zone. So in this case, we need not
531 * change the zone. But perhaps, the zone has only hole data. Thus
532 * it check the zone has only hole or not.
534 pfn = zone_start_pfn;
535 for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) {
536 ms = __pfn_to_section(pfn);
538 if (unlikely(!valid_section(ms)))
541 if (page_zone(pfn_to_page(pfn)) != zone)
544 /* If the section is current section, it continues the loop */
545 if (start_pfn == pfn)
548 /* If we find valid section, we have nothing to do */
549 zone_span_writeunlock(zone);
553 /* The zone has no valid section */
554 zone->zone_start_pfn = 0;
555 zone->spanned_pages = 0;
556 zone_span_writeunlock(zone);
559 static void shrink_pgdat_span(struct pglist_data *pgdat,
560 unsigned long start_pfn, unsigned long end_pfn)
562 unsigned long pgdat_start_pfn = pgdat->node_start_pfn;
563 unsigned long pgdat_end_pfn =
564 pgdat->node_start_pfn + pgdat->node_spanned_pages;
566 struct mem_section *ms;
567 int nid = pgdat->node_id;
569 if (pgdat_start_pfn == start_pfn) {
571 * If the section is smallest section in the pgdat, it need
572 * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages.
573 * In this case, we find second smallest valid mem_section
574 * for shrinking zone.
576 pfn = find_smallest_section_pfn(nid, NULL, end_pfn,
579 pgdat->node_start_pfn = pfn;
580 pgdat->node_spanned_pages = pgdat_end_pfn - pfn;
582 } else if (pgdat_end_pfn == end_pfn) {
584 * If the section is biggest section in the pgdat, it need
585 * shrink pgdat->node_spanned_pages.
586 * In this case, we find second biggest valid mem_section for
589 pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn,
592 pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1;
596 * If the section is not biggest or smallest mem_section in the pgdat,
597 * it only creates a hole in the pgdat. So in this case, we need not
599 * But perhaps, the pgdat has only hole data. Thus it check the pgdat
600 * has only hole or not.
602 pfn = pgdat_start_pfn;
603 for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) {
604 ms = __pfn_to_section(pfn);
606 if (unlikely(!valid_section(ms)))
609 if (pfn_to_nid(pfn) != nid)
612 /* If the section is current section, it continues the loop */
613 if (start_pfn == pfn)
616 /* If we find valid section, we have nothing to do */
620 /* The pgdat has no valid section */
621 pgdat->node_start_pfn = 0;
622 pgdat->node_spanned_pages = 0;
625 static void __remove_zone(struct zone *zone, unsigned long start_pfn)
627 struct pglist_data *pgdat = zone->zone_pgdat;
628 int nr_pages = PAGES_PER_SECTION;
632 zone_type = zone - pgdat->node_zones;
634 pgdat_resize_lock(zone->zone_pgdat, &flags);
635 shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
636 shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages);
637 pgdat_resize_unlock(zone->zone_pgdat, &flags);
640 static int __remove_section(struct zone *zone, struct mem_section *ms)
642 unsigned long start_pfn;
646 if (!valid_section(ms))
649 ret = unregister_memory_section(ms);
653 scn_nr = __section_nr(ms);
654 start_pfn = section_nr_to_pfn(scn_nr);
655 __remove_zone(zone, start_pfn);
657 sparse_remove_one_section(zone, ms);
662 * Reasonably generic function for adding memory. It is
663 * expected that archs that support memory hotplug will
664 * call this function after deciding the zone to which to
667 int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn,
668 unsigned long nr_pages)
672 int start_sec, end_sec;
673 /* during initialize mem_map, align hot-added range to section */
674 start_sec = pfn_to_section_nr(phys_start_pfn);
675 end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
677 for (i = start_sec; i <= end_sec; i++) {
678 err = __add_section(nid, zone, i << PFN_SECTION_SHIFT);
681 * EEXIST is finally dealt with by ioresource collision
682 * check. see add_memory() => register_memory_resource()
683 * Warning will be printed if there is collision.
685 if (err && (err != -EEXIST))
692 EXPORT_SYMBOL_GPL(__add_pages);
695 * __remove_pages() - remove sections of pages from a zone
696 * @zone: zone from which pages need to be removed
697 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
698 * @nr_pages: number of pages to remove (must be multiple of section size)
700 * Generic helper function to remove section mappings and sysfs entries
701 * for the section of the memory we are removing. Caller needs to make
702 * sure that pages are marked reserved and zones are adjust properly by
703 * calling offline_pages().
705 int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
706 unsigned long nr_pages)
709 int sections_to_remove;
710 resource_size_t start, size;
714 * We can only remove entire sections
716 BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
717 BUG_ON(nr_pages % PAGES_PER_SECTION);
719 start = phys_start_pfn << PAGE_SHIFT;
720 size = nr_pages * PAGE_SIZE;
721 ret = release_mem_region_adjustable(&iomem_resource, start, size);
723 pr_warn("Unable to release resource <%016llx-%016llx> (%d)\n",
724 start, start + size - 1, ret);
726 sections_to_remove = nr_pages / PAGES_PER_SECTION;
727 for (i = 0; i < sections_to_remove; i++) {
728 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
729 ret = __remove_section(zone, __pfn_to_section(pfn));
735 EXPORT_SYMBOL_GPL(__remove_pages);
737 int set_online_page_callback(online_page_callback_t callback)
741 lock_memory_hotplug();
743 if (online_page_callback == generic_online_page) {
744 online_page_callback = callback;
748 unlock_memory_hotplug();
752 EXPORT_SYMBOL_GPL(set_online_page_callback);
754 int restore_online_page_callback(online_page_callback_t callback)
758 lock_memory_hotplug();
760 if (online_page_callback == callback) {
761 online_page_callback = generic_online_page;
765 unlock_memory_hotplug();
769 EXPORT_SYMBOL_GPL(restore_online_page_callback);
771 void __online_page_set_limits(struct page *page)
773 unsigned long pfn = page_to_pfn(page);
775 if (pfn >= num_physpages)
776 num_physpages = pfn + 1;
778 EXPORT_SYMBOL_GPL(__online_page_set_limits);
780 void __online_page_increment_counters(struct page *page)
784 #ifdef CONFIG_HIGHMEM
785 if (PageHighMem(page))
789 EXPORT_SYMBOL_GPL(__online_page_increment_counters);
791 void __online_page_free(struct page *page)
793 ClearPageReserved(page);
794 init_page_count(page);
797 EXPORT_SYMBOL_GPL(__online_page_free);
799 static void generic_online_page(struct page *page)
801 __online_page_set_limits(page);
802 __online_page_increment_counters(page);
803 __online_page_free(page);
806 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
810 unsigned long onlined_pages = *(unsigned long *)arg;
812 if (PageReserved(pfn_to_page(start_pfn)))
813 for (i = 0; i < nr_pages; i++) {
814 page = pfn_to_page(start_pfn + i);
815 (*online_page_callback)(page);
818 *(unsigned long *)arg = onlined_pages;
822 #ifdef CONFIG_MOVABLE_NODE
824 * When CONFIG_MOVABLE_NODE, we permit onlining of a node which doesn't have
827 static bool can_online_high_movable(struct zone *zone)
831 #else /* CONFIG_MOVABLE_NODE */
832 /* ensure every online node has NORMAL memory */
833 static bool can_online_high_movable(struct zone *zone)
835 return node_state(zone_to_nid(zone), N_NORMAL_MEMORY);
837 #endif /* CONFIG_MOVABLE_NODE */
839 /* check which state of node_states will be changed when online memory */
840 static void node_states_check_changes_online(unsigned long nr_pages,
841 struct zone *zone, struct memory_notify *arg)
843 int nid = zone_to_nid(zone);
844 enum zone_type zone_last = ZONE_NORMAL;
847 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
848 * contains nodes which have zones of 0...ZONE_NORMAL,
849 * set zone_last to ZONE_NORMAL.
851 * If we don't have HIGHMEM nor movable node,
852 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
853 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
855 if (N_MEMORY == N_NORMAL_MEMORY)
856 zone_last = ZONE_MOVABLE;
859 * if the memory to be online is in a zone of 0...zone_last, and
860 * the zones of 0...zone_last don't have memory before online, we will
861 * need to set the node to node_states[N_NORMAL_MEMORY] after
862 * the memory is online.
864 if (zone_idx(zone) <= zone_last && !node_state(nid, N_NORMAL_MEMORY))
865 arg->status_change_nid_normal = nid;
867 arg->status_change_nid_normal = -1;
869 #ifdef CONFIG_HIGHMEM
871 * If we have movable node, node_states[N_HIGH_MEMORY]
872 * contains nodes which have zones of 0...ZONE_HIGHMEM,
873 * set zone_last to ZONE_HIGHMEM.
875 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
876 * contains nodes which have zones of 0...ZONE_MOVABLE,
877 * set zone_last to ZONE_MOVABLE.
879 zone_last = ZONE_HIGHMEM;
880 if (N_MEMORY == N_HIGH_MEMORY)
881 zone_last = ZONE_MOVABLE;
883 if (zone_idx(zone) <= zone_last && !node_state(nid, N_HIGH_MEMORY))
884 arg->status_change_nid_high = nid;
886 arg->status_change_nid_high = -1;
888 arg->status_change_nid_high = arg->status_change_nid_normal;
892 * if the node don't have memory befor online, we will need to
893 * set the node to node_states[N_MEMORY] after the memory
896 if (!node_state(nid, N_MEMORY))
897 arg->status_change_nid = nid;
899 arg->status_change_nid = -1;
902 static void node_states_set_node(int node, struct memory_notify *arg)
904 if (arg->status_change_nid_normal >= 0)
905 node_set_state(node, N_NORMAL_MEMORY);
907 if (arg->status_change_nid_high >= 0)
908 node_set_state(node, N_HIGH_MEMORY);
910 node_set_state(node, N_MEMORY);
914 int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type)
916 unsigned long onlined_pages = 0;
918 int need_zonelists_rebuild = 0;
921 struct memory_notify arg;
923 lock_memory_hotplug();
925 * This doesn't need a lock to do pfn_to_page().
926 * The section can't be removed here because of the
927 * memory_block->state_mutex.
929 zone = page_zone(pfn_to_page(pfn));
931 if ((zone_idx(zone) > ZONE_NORMAL || online_type == ONLINE_MOVABLE) &&
932 !can_online_high_movable(zone)) {
933 unlock_memory_hotplug();
937 if (online_type == ONLINE_KERNEL && zone_idx(zone) == ZONE_MOVABLE) {
938 if (move_pfn_range_left(zone - 1, zone, pfn, pfn + nr_pages)) {
939 unlock_memory_hotplug();
943 if (online_type == ONLINE_MOVABLE && zone_idx(zone) == ZONE_MOVABLE - 1) {
944 if (move_pfn_range_right(zone, zone + 1, pfn, pfn + nr_pages)) {
945 unlock_memory_hotplug();
950 /* Previous code may changed the zone of the pfn range */
951 zone = page_zone(pfn_to_page(pfn));
954 arg.nr_pages = nr_pages;
955 node_states_check_changes_online(nr_pages, zone, &arg);
957 nid = page_to_nid(pfn_to_page(pfn));
959 ret = memory_notify(MEM_GOING_ONLINE, &arg);
960 ret = notifier_to_errno(ret);
962 memory_notify(MEM_CANCEL_ONLINE, &arg);
963 unlock_memory_hotplug();
967 * If this zone is not populated, then it is not in zonelist.
968 * This means the page allocator ignores this zone.
969 * So, zonelist must be updated after online.
971 mutex_lock(&zonelists_mutex);
972 if (!populated_zone(zone)) {
973 need_zonelists_rebuild = 1;
974 build_all_zonelists(NULL, zone);
977 ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
980 if (need_zonelists_rebuild)
981 zone_pcp_reset(zone);
982 mutex_unlock(&zonelists_mutex);
983 printk(KERN_DEBUG "online_pages [mem %#010llx-%#010llx] failed\n",
984 (unsigned long long) pfn << PAGE_SHIFT,
985 (((unsigned long long) pfn + nr_pages)
987 memory_notify(MEM_CANCEL_ONLINE, &arg);
988 unlock_memory_hotplug();
992 zone->managed_pages += onlined_pages;
993 zone->present_pages += onlined_pages;
994 zone->zone_pgdat->node_present_pages += onlined_pages;
996 node_states_set_node(zone_to_nid(zone), &arg);
997 if (need_zonelists_rebuild)
998 build_all_zonelists(NULL, NULL);
1000 zone_pcp_update(zone);
1003 mutex_unlock(&zonelists_mutex);
1005 init_per_zone_wmark_min();
1008 kswapd_run(zone_to_nid(zone));
1010 vm_total_pages = nr_free_pagecache_pages();
1012 writeback_set_ratelimit();
1015 memory_notify(MEM_ONLINE, &arg);
1016 unlock_memory_hotplug();
1020 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
1022 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1023 static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
1025 struct pglist_data *pgdat;
1026 unsigned long zones_size[MAX_NR_ZONES] = {0};
1027 unsigned long zholes_size[MAX_NR_ZONES] = {0};
1028 unsigned long start_pfn = start >> PAGE_SHIFT;
1030 pgdat = NODE_DATA(nid);
1032 pgdat = arch_alloc_nodedata(nid);
1036 arch_refresh_nodedata(nid, pgdat);
1039 /* we can use NODE_DATA(nid) from here */
1041 /* init node's zones as empty zones, we don't have any present pages.*/
1042 free_area_init_node(nid, zones_size, start_pfn, zholes_size);
1045 * The node we allocated has no zone fallback lists. For avoiding
1046 * to access not-initialized zonelist, build here.
1048 mutex_lock(&zonelists_mutex);
1049 build_all_zonelists(pgdat, NULL);
1050 mutex_unlock(&zonelists_mutex);
1055 static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
1057 arch_refresh_nodedata(nid, NULL);
1058 arch_free_nodedata(pgdat);
1064 * called by cpu_up() to online a node without onlined memory.
1066 int mem_online_node(int nid)
1071 lock_memory_hotplug();
1072 pgdat = hotadd_new_pgdat(nid, 0);
1077 node_set_online(nid);
1078 ret = register_one_node(nid);
1082 unlock_memory_hotplug();
1086 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1087 int __ref add_memory(int nid, u64 start, u64 size)
1089 pg_data_t *pgdat = NULL;
1092 struct resource *res;
1095 lock_memory_hotplug();
1097 res = register_memory_resource(start, size);
1102 { /* Stupid hack to suppress address-never-null warning */
1103 void *p = NODE_DATA(nid);
1106 new_node = !node_online(nid);
1108 pgdat = hotadd_new_pgdat(nid, start);
1114 /* call arch's memory hotadd */
1115 ret = arch_add_memory(nid, start, size);
1120 /* we online node here. we can't roll back from here. */
1121 node_set_online(nid);
1124 ret = register_one_node(nid);
1126 * If sysfs file of new node can't create, cpu on the node
1127 * can't be hot-added. There is no rollback way now.
1128 * So, check by BUG_ON() to catch it reluctantly..
1133 /* create new memmap entry */
1134 firmware_map_add_hotplug(start, start + size, "System RAM");
1139 /* rollback pgdat allocation and others */
1141 rollback_node_hotadd(nid, pgdat);
1142 release_memory_resource(res);
1145 unlock_memory_hotplug();
1148 EXPORT_SYMBOL_GPL(add_memory);
1150 #ifdef CONFIG_MEMORY_HOTREMOVE
1152 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1153 * set and the size of the free page is given by page_order(). Using this,
1154 * the function determines if the pageblock contains only free pages.
1155 * Due to buddy contraints, a free page at least the size of a pageblock will
1156 * be located at the start of the pageblock
1158 static inline int pageblock_free(struct page *page)
1160 return PageBuddy(page) && page_order(page) >= pageblock_order;
1163 /* Return the start of the next active pageblock after a given page */
1164 static struct page *next_active_pageblock(struct page *page)
1166 /* Ensure the starting page is pageblock-aligned */
1167 BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
1169 /* If the entire pageblock is free, move to the end of free page */
1170 if (pageblock_free(page)) {
1172 /* be careful. we don't have locks, page_order can be changed.*/
1173 order = page_order(page);
1174 if ((order < MAX_ORDER) && (order >= pageblock_order))
1175 return page + (1 << order);
1178 return page + pageblock_nr_pages;
1181 /* Checks if this range of memory is likely to be hot-removable. */
1182 int is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
1184 struct page *page = pfn_to_page(start_pfn);
1185 struct page *end_page = page + nr_pages;
1187 /* Check the starting page of each pageblock within the range */
1188 for (; page < end_page; page = next_active_pageblock(page)) {
1189 if (!is_pageblock_removable_nolock(page))
1194 /* All pageblocks in the memory block are likely to be hot-removable */
1199 * Confirm all pages in a range [start, end) is belongs to the same zone.
1201 static int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn)
1204 struct zone *zone = NULL;
1207 for (pfn = start_pfn;
1209 pfn += MAX_ORDER_NR_PAGES) {
1211 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1212 while ((i < MAX_ORDER_NR_PAGES) && !pfn_valid_within(pfn + i))
1214 if (i == MAX_ORDER_NR_PAGES)
1216 page = pfn_to_page(pfn + i);
1217 if (zone && page_zone(page) != zone)
1219 zone = page_zone(page);
1225 * Scanning pfn is much easier than scanning lru list.
1226 * Scan pfn from start to end and Find LRU page.
1228 static unsigned long scan_lru_pages(unsigned long start, unsigned long end)
1232 for (pfn = start; pfn < end; pfn++) {
1233 if (pfn_valid(pfn)) {
1234 page = pfn_to_page(pfn);
1242 #define NR_OFFLINE_AT_ONCE_PAGES (256)
1244 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1248 int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
1249 int not_managed = 0;
1253 for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
1254 if (!pfn_valid(pfn))
1256 page = pfn_to_page(pfn);
1257 if (!get_page_unless_zero(page))
1260 * We can skip free pages. And we can only deal with pages on
1263 ret = isolate_lru_page(page);
1264 if (!ret) { /* Success */
1266 list_add_tail(&page->lru, &source);
1268 inc_zone_page_state(page, NR_ISOLATED_ANON +
1269 page_is_file_cache(page));
1272 #ifdef CONFIG_DEBUG_VM
1273 printk(KERN_ALERT "removing pfn %lx from LRU failed\n",
1278 /* Because we don't have big zone->lock. we should
1279 check this again here. */
1280 if (page_count(page)) {
1287 if (!list_empty(&source)) {
1289 putback_lru_pages(&source);
1294 * alloc_migrate_target should be improooooved!!
1295 * migrate_pages returns # of failed pages.
1297 ret = migrate_pages(&source, alloc_migrate_target, 0,
1298 MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
1300 putback_lru_pages(&source);
1307 * remove from free_area[] and mark all as Reserved.
1310 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1313 __offline_isolated_pages(start, start + nr_pages);
1318 offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
1320 walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
1321 offline_isolated_pages_cb);
1325 * Check all pages in range, recoreded as memory resource, are isolated.
1328 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1332 long offlined = *(long *)data;
1333 ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true);
1334 offlined = nr_pages;
1336 *(long *)data += offlined;
1341 check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
1346 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
1347 check_pages_isolated_cb);
1349 offlined = (long)ret;
1353 #ifdef CONFIG_MOVABLE_NODE
1355 * When CONFIG_MOVABLE_NODE, we permit offlining of a node which doesn't have
1358 static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1362 #else /* CONFIG_MOVABLE_NODE */
1363 /* ensure the node has NORMAL memory if it is still online */
1364 static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1366 struct pglist_data *pgdat = zone->zone_pgdat;
1367 unsigned long present_pages = 0;
1370 for (zt = 0; zt <= ZONE_NORMAL; zt++)
1371 present_pages += pgdat->node_zones[zt].present_pages;
1373 if (present_pages > nr_pages)
1377 for (; zt <= ZONE_MOVABLE; zt++)
1378 present_pages += pgdat->node_zones[zt].present_pages;
1381 * we can't offline the last normal memory until all
1382 * higher memory is offlined.
1384 return present_pages == 0;
1386 #endif /* CONFIG_MOVABLE_NODE */
1388 /* check which state of node_states will be changed when offline memory */
1389 static void node_states_check_changes_offline(unsigned long nr_pages,
1390 struct zone *zone, struct memory_notify *arg)
1392 struct pglist_data *pgdat = zone->zone_pgdat;
1393 unsigned long present_pages = 0;
1394 enum zone_type zt, zone_last = ZONE_NORMAL;
1397 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
1398 * contains nodes which have zones of 0...ZONE_NORMAL,
1399 * set zone_last to ZONE_NORMAL.
1401 * If we don't have HIGHMEM nor movable node,
1402 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
1403 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
1405 if (N_MEMORY == N_NORMAL_MEMORY)
1406 zone_last = ZONE_MOVABLE;
1409 * check whether node_states[N_NORMAL_MEMORY] will be changed.
1410 * If the memory to be offline is in a zone of 0...zone_last,
1411 * and it is the last present memory, 0...zone_last will
1412 * become empty after offline , thus we can determind we will
1413 * need to clear the node from node_states[N_NORMAL_MEMORY].
1415 for (zt = 0; zt <= zone_last; zt++)
1416 present_pages += pgdat->node_zones[zt].present_pages;
1417 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1418 arg->status_change_nid_normal = zone_to_nid(zone);
1420 arg->status_change_nid_normal = -1;
1422 #ifdef CONFIG_HIGHMEM
1424 * If we have movable node, node_states[N_HIGH_MEMORY]
1425 * contains nodes which have zones of 0...ZONE_HIGHMEM,
1426 * set zone_last to ZONE_HIGHMEM.
1428 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
1429 * contains nodes which have zones of 0...ZONE_MOVABLE,
1430 * set zone_last to ZONE_MOVABLE.
1432 zone_last = ZONE_HIGHMEM;
1433 if (N_MEMORY == N_HIGH_MEMORY)
1434 zone_last = ZONE_MOVABLE;
1436 for (; zt <= zone_last; zt++)
1437 present_pages += pgdat->node_zones[zt].present_pages;
1438 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1439 arg->status_change_nid_high = zone_to_nid(zone);
1441 arg->status_change_nid_high = -1;
1443 arg->status_change_nid_high = arg->status_change_nid_normal;
1447 * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE
1449 zone_last = ZONE_MOVABLE;
1452 * check whether node_states[N_HIGH_MEMORY] will be changed
1453 * If we try to offline the last present @nr_pages from the node,
1454 * we can determind we will need to clear the node from
1455 * node_states[N_HIGH_MEMORY].
1457 for (; zt <= zone_last; zt++)
1458 present_pages += pgdat->node_zones[zt].present_pages;
1459 if (nr_pages >= present_pages)
1460 arg->status_change_nid = zone_to_nid(zone);
1462 arg->status_change_nid = -1;
1465 static void node_states_clear_node(int node, struct memory_notify *arg)
1467 if (arg->status_change_nid_normal >= 0)
1468 node_clear_state(node, N_NORMAL_MEMORY);
1470 if ((N_MEMORY != N_NORMAL_MEMORY) &&
1471 (arg->status_change_nid_high >= 0))
1472 node_clear_state(node, N_HIGH_MEMORY);
1474 if ((N_MEMORY != N_HIGH_MEMORY) &&
1475 (arg->status_change_nid >= 0))
1476 node_clear_state(node, N_MEMORY);
1479 static int __ref __offline_pages(unsigned long start_pfn,
1480 unsigned long end_pfn, unsigned long timeout)
1482 unsigned long pfn, nr_pages, expire;
1483 long offlined_pages;
1484 int ret, drain, retry_max, node;
1486 struct memory_notify arg;
1488 BUG_ON(start_pfn >= end_pfn);
1489 /* at least, alignment against pageblock is necessary */
1490 if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
1492 if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
1494 /* This makes hotplug much easier...and readable.
1495 we assume this for now. .*/
1496 if (!test_pages_in_a_zone(start_pfn, end_pfn))
1499 lock_memory_hotplug();
1501 zone = page_zone(pfn_to_page(start_pfn));
1502 node = zone_to_nid(zone);
1503 nr_pages = end_pfn - start_pfn;
1506 if (zone_idx(zone) <= ZONE_NORMAL && !can_offline_normal(zone, nr_pages))
1509 /* set above range as isolated */
1510 ret = start_isolate_page_range(start_pfn, end_pfn,
1511 MIGRATE_MOVABLE, true);
1515 arg.start_pfn = start_pfn;
1516 arg.nr_pages = nr_pages;
1517 node_states_check_changes_offline(nr_pages, zone, &arg);
1519 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1520 ret = notifier_to_errno(ret);
1522 goto failed_removal;
1525 expire = jiffies + timeout;
1529 /* start memory hot removal */
1531 if (time_after(jiffies, expire))
1532 goto failed_removal;
1534 if (signal_pending(current))
1535 goto failed_removal;
1538 lru_add_drain_all();
1543 pfn = scan_lru_pages(start_pfn, end_pfn);
1544 if (pfn) { /* We have page on LRU */
1545 ret = do_migrate_range(pfn, end_pfn);
1551 if (--retry_max == 0)
1552 goto failed_removal;
1558 /* drain all zone's lru pagevec, this is asynchronous... */
1559 lru_add_drain_all();
1561 /* drain pcp pages, this is synchronous. */
1564 offlined_pages = check_pages_isolated(start_pfn, end_pfn);
1565 if (offlined_pages < 0) {
1567 goto failed_removal;
1569 printk(KERN_INFO "Offlined Pages %ld\n", offlined_pages);
1570 /* Ok, all of our target is isolated.
1571 We cannot do rollback at this point. */
1572 offline_isolated_pages(start_pfn, end_pfn);
1573 /* reset pagetype flags and makes migrate type to be MOVABLE */
1574 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1575 /* removal success */
1576 zone->managed_pages -= offlined_pages;
1577 zone->present_pages -= offlined_pages;
1578 zone->zone_pgdat->node_present_pages -= offlined_pages;
1579 totalram_pages -= offlined_pages;
1581 init_per_zone_wmark_min();
1583 if (!populated_zone(zone)) {
1584 zone_pcp_reset(zone);
1585 mutex_lock(&zonelists_mutex);
1586 build_all_zonelists(NULL, NULL);
1587 mutex_unlock(&zonelists_mutex);
1589 zone_pcp_update(zone);
1591 node_states_clear_node(node, &arg);
1592 if (arg.status_change_nid >= 0)
1595 vm_total_pages = nr_free_pagecache_pages();
1596 writeback_set_ratelimit();
1598 memory_notify(MEM_OFFLINE, &arg);
1599 unlock_memory_hotplug();
1603 printk(KERN_INFO "memory offlining [mem %#010llx-%#010llx] failed\n",
1604 (unsigned long long) start_pfn << PAGE_SHIFT,
1605 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
1606 memory_notify(MEM_CANCEL_OFFLINE, &arg);
1607 /* pushback to free area */
1608 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1611 unlock_memory_hotplug();
1615 int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1617 return __offline_pages(start_pfn, start_pfn + nr_pages, 120 * HZ);
1621 * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
1622 * @start_pfn: start pfn of the memory range
1623 * @end_pfn: end pfn of the memory range
1624 * @arg: argument passed to func
1625 * @func: callback for each memory section walked
1627 * This function walks through all present mem sections in range
1628 * [start_pfn, end_pfn) and call func on each mem section.
1630 * Returns the return value of func.
1632 static int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn,
1633 void *arg, int (*func)(struct memory_block *, void *))
1635 struct memory_block *mem = NULL;
1636 struct mem_section *section;
1637 unsigned long pfn, section_nr;
1640 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1641 section_nr = pfn_to_section_nr(pfn);
1642 if (!present_section_nr(section_nr))
1645 section = __nr_to_section(section_nr);
1646 /* same memblock? */
1648 if ((section_nr >= mem->start_section_nr) &&
1649 (section_nr <= mem->end_section_nr))
1652 mem = find_memory_block_hinted(section, mem);
1656 ret = func(mem, arg);
1658 kobject_put(&mem->dev.kobj);
1664 kobject_put(&mem->dev.kobj);
1670 * offline_memory_block_cb - callback function for offlining memory block
1671 * @mem: the memory block to be offlined
1672 * @arg: buffer to hold error msg
1674 * Always return 0, and put the error msg in arg if any.
1676 static int offline_memory_block_cb(struct memory_block *mem, void *arg)
1679 int error = offline_memory_block(mem);
1681 if (error != 0 && *ret == 0)
1687 static int is_memblock_offlined_cb(struct memory_block *mem, void *arg)
1689 int ret = !is_memblock_offlined(mem);
1692 pr_warn("removing memory fails, because memory "
1693 "[%#010llx-%#010llx] is onlined\n",
1694 PFN_PHYS(section_nr_to_pfn(mem->start_section_nr)),
1695 PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1);
1700 static int check_cpu_on_node(void *data)
1702 struct pglist_data *pgdat = data;
1705 for_each_present_cpu(cpu) {
1706 if (cpu_to_node(cpu) == pgdat->node_id)
1708 * the cpu on this node isn't removed, and we can't
1709 * offline this node.
1717 static void unmap_cpu_on_node(void *data)
1719 #ifdef CONFIG_ACPI_NUMA
1720 struct pglist_data *pgdat = data;
1723 for_each_possible_cpu(cpu)
1724 if (cpu_to_node(cpu) == pgdat->node_id)
1725 numa_clear_node(cpu);
1729 static int check_and_unmap_cpu_on_node(void *data)
1731 int ret = check_cpu_on_node(data);
1737 * the node will be offlined when we come here, so we can clear
1738 * the cpu_to_node() now.
1741 unmap_cpu_on_node(data);
1745 /* offline the node if all memory sections of this node are removed */
1746 void try_offline_node(int nid)
1748 pg_data_t *pgdat = NODE_DATA(nid);
1749 unsigned long start_pfn = pgdat->node_start_pfn;
1750 unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
1752 struct page *pgdat_page = virt_to_page(pgdat);
1755 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1756 unsigned long section_nr = pfn_to_section_nr(pfn);
1758 if (!present_section_nr(section_nr))
1761 if (pfn_to_nid(pfn) != nid)
1765 * some memory sections of this node are not removed, and we
1766 * can't offline node now.
1771 if (stop_machine(check_and_unmap_cpu_on_node, pgdat, NULL))
1775 * all memory/cpu of this node are removed, we can offline this
1778 node_set_offline(nid);
1779 unregister_one_node(nid);
1781 if (!PageSlab(pgdat_page) && !PageCompound(pgdat_page))
1782 /* node data is allocated from boot memory */
1785 /* free waittable in each zone */
1786 for (i = 0; i < MAX_NR_ZONES; i++) {
1787 struct zone *zone = pgdat->node_zones + i;
1790 * wait_table may be allocated from boot memory,
1791 * here only free if it's allocated by vmalloc.
1793 if (is_vmalloc_addr(zone->wait_table))
1794 vfree(zone->wait_table);
1798 * Since there is no way to guarentee the address of pgdat/zone is not
1799 * on stack of any kernel threads or used by other kernel objects
1800 * without reference counting or other symchronizing method, do not
1801 * reset node_data and free pgdat here. Just reset it to 0 and reuse
1802 * the memory when the node is online again.
1804 memset(pgdat, 0, sizeof(*pgdat));
1806 EXPORT_SYMBOL(try_offline_node);
1808 int __ref remove_memory(int nid, u64 start, u64 size)
1810 unsigned long start_pfn, end_pfn;
1814 start_pfn = PFN_DOWN(start);
1815 end_pfn = PFN_UP(start + size - 1);
1818 * When CONFIG_MEMCG is on, one memory block may be used by other
1819 * blocks to store page cgroup when onlining pages. But we don't know
1820 * in what order pages are onlined. So we iterate twice to offline
1822 * 1st iterate: offline every non primary memory block.
1823 * 2nd iterate: offline primary (i.e. first added) memory block.
1826 walk_memory_range(start_pfn, end_pfn, &ret,
1827 offline_memory_block_cb);
1837 lock_memory_hotplug();
1840 * we have offlined all memory blocks like this:
1841 * 1. lock memory hotplug
1842 * 2. offline a memory block
1843 * 3. unlock memory hotplug
1845 * repeat step1-3 to offline the memory block. All memory blocks
1846 * must be offlined before removing memory. But we don't hold the
1847 * lock in the whole operation. So we should check whether all
1848 * memory blocks are offlined.
1851 ret = walk_memory_range(start_pfn, end_pfn, NULL,
1852 is_memblock_offlined_cb);
1854 unlock_memory_hotplug();
1858 /* remove memmap entry */
1859 firmware_map_remove(start, start + size, "System RAM");
1861 arch_remove_memory(start, size);
1863 try_offline_node(nid);
1865 unlock_memory_hotplug();
1870 int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1874 int remove_memory(int nid, u64 start, u64 size)
1878 #endif /* CONFIG_MEMORY_HOTREMOVE */
1879 EXPORT_SYMBOL_GPL(remove_memory);