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>
33 #include <linux/hugetlb.h>
35 #include <asm/tlbflush.h>
40 * online_page_callback contains pointer to current page onlining function.
41 * Initially it is generic_online_page(). If it is required it could be
42 * changed by calling set_online_page_callback() for callback registration
43 * and restore_online_page_callback() for generic callback restore.
46 static void generic_online_page(struct page *page);
48 static online_page_callback_t online_page_callback = generic_online_page;
50 DEFINE_MUTEX(mem_hotplug_mutex);
52 void lock_memory_hotplug(void)
54 mutex_lock(&mem_hotplug_mutex);
56 /* for exclusive hibernation if CONFIG_HIBERNATION=y */
60 void unlock_memory_hotplug(void)
62 unlock_system_sleep();
63 mutex_unlock(&mem_hotplug_mutex);
67 /* add this memory to iomem resource */
68 static struct resource *register_memory_resource(u64 start, u64 size)
71 res = kzalloc(sizeof(struct resource), GFP_KERNEL);
74 res->name = "System RAM";
76 res->end = start + size - 1;
77 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
78 if (request_resource(&iomem_resource, res) < 0) {
79 pr_debug("System RAM resource %pR cannot be added\n", res);
86 static void release_memory_resource(struct resource *res)
90 release_resource(res);
95 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
96 void get_page_bootmem(unsigned long info, struct page *page,
99 page->lru.next = (struct list_head *) type;
100 SetPagePrivate(page);
101 set_page_private(page, info);
102 atomic_inc(&page->_count);
105 void put_page_bootmem(struct page *page)
109 type = (unsigned long) page->lru.next;
110 BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE ||
111 type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE);
113 if (atomic_dec_return(&page->_count) == 1) {
114 ClearPagePrivate(page);
115 set_page_private(page, 0);
116 INIT_LIST_HEAD(&page->lru);
117 free_reserved_page(page);
121 #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
122 #ifndef CONFIG_SPARSEMEM_VMEMMAP
123 static void register_page_bootmem_info_section(unsigned long start_pfn)
125 unsigned long *usemap, mapsize, section_nr, i;
126 struct mem_section *ms;
127 struct page *page, *memmap;
129 section_nr = pfn_to_section_nr(start_pfn);
130 ms = __nr_to_section(section_nr);
132 /* Get section's memmap address */
133 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
136 * Get page for the memmap's phys address
137 * XXX: need more consideration for sparse_vmemmap...
139 page = virt_to_page(memmap);
140 mapsize = sizeof(struct page) * PAGES_PER_SECTION;
141 mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
143 /* remember memmap's page */
144 for (i = 0; i < mapsize; i++, page++)
145 get_page_bootmem(section_nr, page, SECTION_INFO);
147 usemap = __nr_to_section(section_nr)->pageblock_flags;
148 page = virt_to_page(usemap);
150 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
152 for (i = 0; i < mapsize; i++, page++)
153 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
156 #else /* CONFIG_SPARSEMEM_VMEMMAP */
157 static void register_page_bootmem_info_section(unsigned long start_pfn)
159 unsigned long *usemap, mapsize, section_nr, i;
160 struct mem_section *ms;
161 struct page *page, *memmap;
163 if (!pfn_valid(start_pfn))
166 section_nr = pfn_to_section_nr(start_pfn);
167 ms = __nr_to_section(section_nr);
169 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
171 register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION);
173 usemap = __nr_to_section(section_nr)->pageblock_flags;
174 page = virt_to_page(usemap);
176 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
178 for (i = 0; i < mapsize; i++, page++)
179 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
181 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
183 void register_page_bootmem_info_node(struct pglist_data *pgdat)
185 unsigned long i, pfn, end_pfn, nr_pages;
186 int node = pgdat->node_id;
190 nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
191 page = virt_to_page(pgdat);
193 for (i = 0; i < nr_pages; i++, page++)
194 get_page_bootmem(node, page, NODE_INFO);
196 zone = &pgdat->node_zones[0];
197 for (; zone < pgdat->node_zones + MAX_NR_ZONES - 1; zone++) {
198 if (zone_is_initialized(zone)) {
199 nr_pages = zone->wait_table_hash_nr_entries
200 * sizeof(wait_queue_head_t);
201 nr_pages = PAGE_ALIGN(nr_pages) >> PAGE_SHIFT;
202 page = virt_to_page(zone->wait_table);
204 for (i = 0; i < nr_pages; i++, page++)
205 get_page_bootmem(node, page, NODE_INFO);
209 pfn = pgdat->node_start_pfn;
210 end_pfn = pgdat_end_pfn(pgdat);
212 /* register section info */
213 for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
215 * Some platforms can assign the same pfn to multiple nodes - on
216 * node0 as well as nodeN. To avoid registering a pfn against
217 * multiple nodes we check that this pfn does not already
218 * reside in some other nodes.
220 if (pfn_valid(pfn) && (pfn_to_nid(pfn) == node))
221 register_page_bootmem_info_section(pfn);
224 #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
226 static void grow_zone_span(struct zone *zone, unsigned long start_pfn,
227 unsigned long end_pfn)
229 unsigned long old_zone_end_pfn;
231 zone_span_writelock(zone);
233 old_zone_end_pfn = zone_end_pfn(zone);
234 if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
235 zone->zone_start_pfn = start_pfn;
237 zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
238 zone->zone_start_pfn;
240 zone_span_writeunlock(zone);
243 static void resize_zone(struct zone *zone, unsigned long start_pfn,
244 unsigned long end_pfn)
246 zone_span_writelock(zone);
248 if (end_pfn - start_pfn) {
249 zone->zone_start_pfn = start_pfn;
250 zone->spanned_pages = end_pfn - start_pfn;
253 * make it consist as free_area_init_core(),
254 * if spanned_pages = 0, then keep start_pfn = 0
256 zone->zone_start_pfn = 0;
257 zone->spanned_pages = 0;
260 zone_span_writeunlock(zone);
263 static void fix_zone_id(struct zone *zone, unsigned long start_pfn,
264 unsigned long end_pfn)
266 enum zone_type zid = zone_idx(zone);
267 int nid = zone->zone_pgdat->node_id;
270 for (pfn = start_pfn; pfn < end_pfn; pfn++)
271 set_page_links(pfn_to_page(pfn), zid, nid, pfn);
274 /* Can fail with -ENOMEM from allocating a wait table with vmalloc() or
275 * alloc_bootmem_node_nopanic() */
276 static int __ref ensure_zone_is_initialized(struct zone *zone,
277 unsigned long start_pfn, unsigned long num_pages)
279 if (!zone_is_initialized(zone))
280 return init_currently_empty_zone(zone, start_pfn, num_pages,
285 static int __meminit move_pfn_range_left(struct zone *z1, struct zone *z2,
286 unsigned long start_pfn, unsigned long end_pfn)
290 unsigned long z1_start_pfn;
292 ret = ensure_zone_is_initialized(z1, start_pfn, end_pfn - start_pfn);
296 pgdat_resize_lock(z1->zone_pgdat, &flags);
298 /* can't move pfns which are higher than @z2 */
299 if (end_pfn > zone_end_pfn(z2))
301 /* the move out part must be at the left most of @z2 */
302 if (start_pfn > z2->zone_start_pfn)
304 /* must included/overlap */
305 if (end_pfn <= z2->zone_start_pfn)
308 /* use start_pfn for z1's start_pfn if z1 is empty */
309 if (!zone_is_empty(z1))
310 z1_start_pfn = z1->zone_start_pfn;
312 z1_start_pfn = start_pfn;
314 resize_zone(z1, z1_start_pfn, end_pfn);
315 resize_zone(z2, end_pfn, zone_end_pfn(z2));
317 pgdat_resize_unlock(z1->zone_pgdat, &flags);
319 fix_zone_id(z1, start_pfn, end_pfn);
323 pgdat_resize_unlock(z1->zone_pgdat, &flags);
327 static int __meminit move_pfn_range_right(struct zone *z1, struct zone *z2,
328 unsigned long start_pfn, unsigned long end_pfn)
332 unsigned long z2_end_pfn;
334 ret = ensure_zone_is_initialized(z2, start_pfn, end_pfn - start_pfn);
338 pgdat_resize_lock(z1->zone_pgdat, &flags);
340 /* can't move pfns which are lower than @z1 */
341 if (z1->zone_start_pfn > start_pfn)
343 /* the move out part mast at the right most of @z1 */
344 if (zone_end_pfn(z1) > end_pfn)
346 /* must included/overlap */
347 if (start_pfn >= zone_end_pfn(z1))
350 /* use end_pfn for z2's end_pfn if z2 is empty */
351 if (!zone_is_empty(z2))
352 z2_end_pfn = zone_end_pfn(z2);
354 z2_end_pfn = end_pfn;
356 resize_zone(z1, z1->zone_start_pfn, start_pfn);
357 resize_zone(z2, start_pfn, z2_end_pfn);
359 pgdat_resize_unlock(z1->zone_pgdat, &flags);
361 fix_zone_id(z2, start_pfn, end_pfn);
365 pgdat_resize_unlock(z1->zone_pgdat, &flags);
369 static void grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn,
370 unsigned long end_pfn)
372 unsigned long old_pgdat_end_pfn =
373 pgdat->node_start_pfn + pgdat->node_spanned_pages;
375 if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
376 pgdat->node_start_pfn = start_pfn;
378 pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
379 pgdat->node_start_pfn;
382 static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn)
384 struct pglist_data *pgdat = zone->zone_pgdat;
385 int nr_pages = PAGES_PER_SECTION;
386 int nid = pgdat->node_id;
391 zone_type = zone - pgdat->node_zones;
392 ret = ensure_zone_is_initialized(zone, phys_start_pfn, nr_pages);
396 pgdat_resize_lock(zone->zone_pgdat, &flags);
397 grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages);
398 grow_pgdat_span(zone->zone_pgdat, phys_start_pfn,
399 phys_start_pfn + nr_pages);
400 pgdat_resize_unlock(zone->zone_pgdat, &flags);
401 memmap_init_zone(nr_pages, nid, zone_type,
402 phys_start_pfn, MEMMAP_HOTPLUG);
406 static int __meminit __add_section(int nid, struct zone *zone,
407 unsigned long phys_start_pfn)
409 int nr_pages = PAGES_PER_SECTION;
412 if (pfn_valid(phys_start_pfn))
415 ret = sparse_add_one_section(zone, phys_start_pfn, nr_pages);
420 ret = __add_zone(zone, phys_start_pfn);
425 return register_new_memory(nid, __pfn_to_section(phys_start_pfn));
429 * Reasonably generic function for adding memory. It is
430 * expected that archs that support memory hotplug will
431 * call this function after deciding the zone to which to
434 int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn,
435 unsigned long nr_pages)
439 int start_sec, end_sec;
440 /* during initialize mem_map, align hot-added range to section */
441 start_sec = pfn_to_section_nr(phys_start_pfn);
442 end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
444 for (i = start_sec; i <= end_sec; i++) {
445 err = __add_section(nid, zone, i << PFN_SECTION_SHIFT);
448 * EEXIST is finally dealt with by ioresource collision
449 * check. see add_memory() => register_memory_resource()
450 * Warning will be printed if there is collision.
452 if (err && (err != -EEXIST))
459 EXPORT_SYMBOL_GPL(__add_pages);
461 #ifdef CONFIG_MEMORY_HOTREMOVE
462 /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
463 static int find_smallest_section_pfn(int nid, struct zone *zone,
464 unsigned long start_pfn,
465 unsigned long end_pfn)
467 struct mem_section *ms;
469 for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) {
470 ms = __pfn_to_section(start_pfn);
472 if (unlikely(!valid_section(ms)))
475 if (unlikely(pfn_to_nid(start_pfn) != nid))
478 if (zone && zone != page_zone(pfn_to_page(start_pfn)))
487 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
488 static int find_biggest_section_pfn(int nid, struct zone *zone,
489 unsigned long start_pfn,
490 unsigned long end_pfn)
492 struct mem_section *ms;
495 /* pfn is the end pfn of a memory section. */
497 for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) {
498 ms = __pfn_to_section(pfn);
500 if (unlikely(!valid_section(ms)))
503 if (unlikely(pfn_to_nid(pfn) != nid))
506 if (zone && zone != page_zone(pfn_to_page(pfn)))
515 static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
516 unsigned long end_pfn)
518 unsigned long zone_start_pfn = zone->zone_start_pfn;
519 unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */
520 unsigned long zone_end_pfn = z;
522 struct mem_section *ms;
523 int nid = zone_to_nid(zone);
525 zone_span_writelock(zone);
526 if (zone_start_pfn == start_pfn) {
528 * If the section is smallest section in the zone, it need
529 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
530 * In this case, we find second smallest valid mem_section
531 * for shrinking zone.
533 pfn = find_smallest_section_pfn(nid, zone, end_pfn,
536 zone->zone_start_pfn = pfn;
537 zone->spanned_pages = zone_end_pfn - pfn;
539 } else if (zone_end_pfn == end_pfn) {
541 * If the section is biggest section in the zone, it need
542 * shrink zone->spanned_pages.
543 * In this case, we find second biggest valid mem_section for
546 pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn,
549 zone->spanned_pages = pfn - zone_start_pfn + 1;
553 * The section is not biggest or smallest mem_section in the zone, it
554 * only creates a hole in the zone. So in this case, we need not
555 * change the zone. But perhaps, the zone has only hole data. Thus
556 * it check the zone has only hole or not.
558 pfn = zone_start_pfn;
559 for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) {
560 ms = __pfn_to_section(pfn);
562 if (unlikely(!valid_section(ms)))
565 if (page_zone(pfn_to_page(pfn)) != zone)
568 /* If the section is current section, it continues the loop */
569 if (start_pfn == pfn)
572 /* If we find valid section, we have nothing to do */
573 zone_span_writeunlock(zone);
577 /* The zone has no valid section */
578 zone->zone_start_pfn = 0;
579 zone->spanned_pages = 0;
580 zone_span_writeunlock(zone);
583 static void shrink_pgdat_span(struct pglist_data *pgdat,
584 unsigned long start_pfn, unsigned long end_pfn)
586 unsigned long pgdat_start_pfn = pgdat->node_start_pfn;
587 unsigned long pgdat_end_pfn =
588 pgdat->node_start_pfn + pgdat->node_spanned_pages;
590 struct mem_section *ms;
591 int nid = pgdat->node_id;
593 if (pgdat_start_pfn == start_pfn) {
595 * If the section is smallest section in the pgdat, it need
596 * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages.
597 * In this case, we find second smallest valid mem_section
598 * for shrinking zone.
600 pfn = find_smallest_section_pfn(nid, NULL, end_pfn,
603 pgdat->node_start_pfn = pfn;
604 pgdat->node_spanned_pages = pgdat_end_pfn - pfn;
606 } else if (pgdat_end_pfn == end_pfn) {
608 * If the section is biggest section in the pgdat, it need
609 * shrink pgdat->node_spanned_pages.
610 * In this case, we find second biggest valid mem_section for
613 pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn,
616 pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1;
620 * If the section is not biggest or smallest mem_section in the pgdat,
621 * it only creates a hole in the pgdat. So in this case, we need not
623 * But perhaps, the pgdat has only hole data. Thus it check the pgdat
624 * has only hole or not.
626 pfn = pgdat_start_pfn;
627 for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) {
628 ms = __pfn_to_section(pfn);
630 if (unlikely(!valid_section(ms)))
633 if (pfn_to_nid(pfn) != nid)
636 /* If the section is current section, it continues the loop */
637 if (start_pfn == pfn)
640 /* If we find valid section, we have nothing to do */
644 /* The pgdat has no valid section */
645 pgdat->node_start_pfn = 0;
646 pgdat->node_spanned_pages = 0;
649 static void __remove_zone(struct zone *zone, unsigned long start_pfn)
651 struct pglist_data *pgdat = zone->zone_pgdat;
652 int nr_pages = PAGES_PER_SECTION;
656 zone_type = zone - pgdat->node_zones;
658 pgdat_resize_lock(zone->zone_pgdat, &flags);
659 shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
660 shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages);
661 pgdat_resize_unlock(zone->zone_pgdat, &flags);
664 static int __remove_section(struct zone *zone, struct mem_section *ms)
666 unsigned long start_pfn;
670 if (!valid_section(ms))
673 ret = unregister_memory_section(ms);
677 scn_nr = __section_nr(ms);
678 start_pfn = section_nr_to_pfn(scn_nr);
679 __remove_zone(zone, start_pfn);
681 sparse_remove_one_section(zone, ms);
686 * __remove_pages() - remove sections of pages from a zone
687 * @zone: zone from which pages need to be removed
688 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
689 * @nr_pages: number of pages to remove (must be multiple of section size)
691 * Generic helper function to remove section mappings and sysfs entries
692 * for the section of the memory we are removing. Caller needs to make
693 * sure that pages are marked reserved and zones are adjust properly by
694 * calling offline_pages().
696 int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
697 unsigned long nr_pages)
700 int sections_to_remove;
701 resource_size_t start, size;
705 * We can only remove entire sections
707 BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
708 BUG_ON(nr_pages % PAGES_PER_SECTION);
710 start = phys_start_pfn << PAGE_SHIFT;
711 size = nr_pages * PAGE_SIZE;
712 ret = release_mem_region_adjustable(&iomem_resource, start, size);
714 resource_size_t endres = start + size - 1;
716 pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
717 &start, &endres, ret);
720 sections_to_remove = nr_pages / PAGES_PER_SECTION;
721 for (i = 0; i < sections_to_remove; i++) {
722 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
723 ret = __remove_section(zone, __pfn_to_section(pfn));
729 EXPORT_SYMBOL_GPL(__remove_pages);
730 #endif /* CONFIG_MEMORY_HOTREMOVE */
732 int set_online_page_callback(online_page_callback_t callback)
736 lock_memory_hotplug();
738 if (online_page_callback == generic_online_page) {
739 online_page_callback = callback;
743 unlock_memory_hotplug();
747 EXPORT_SYMBOL_GPL(set_online_page_callback);
749 int restore_online_page_callback(online_page_callback_t callback)
753 lock_memory_hotplug();
755 if (online_page_callback == callback) {
756 online_page_callback = generic_online_page;
760 unlock_memory_hotplug();
764 EXPORT_SYMBOL_GPL(restore_online_page_callback);
766 void __online_page_set_limits(struct page *page)
769 EXPORT_SYMBOL_GPL(__online_page_set_limits);
771 void __online_page_increment_counters(struct page *page)
773 adjust_managed_page_count(page, 1);
775 EXPORT_SYMBOL_GPL(__online_page_increment_counters);
777 void __online_page_free(struct page *page)
779 __free_reserved_page(page);
781 EXPORT_SYMBOL_GPL(__online_page_free);
783 static void generic_online_page(struct page *page)
785 __online_page_set_limits(page);
786 __online_page_increment_counters(page);
787 __online_page_free(page);
790 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
794 unsigned long onlined_pages = *(unsigned long *)arg;
796 if (PageReserved(pfn_to_page(start_pfn)))
797 for (i = 0; i < nr_pages; i++) {
798 page = pfn_to_page(start_pfn + i);
799 (*online_page_callback)(page);
802 *(unsigned long *)arg = onlined_pages;
806 #ifdef CONFIG_MOVABLE_NODE
808 * When CONFIG_MOVABLE_NODE, we permit onlining of a node which doesn't have
811 static bool can_online_high_movable(struct zone *zone)
815 #else /* CONFIG_MOVABLE_NODE */
816 /* ensure every online node has NORMAL memory */
817 static bool can_online_high_movable(struct zone *zone)
819 return node_state(zone_to_nid(zone), N_NORMAL_MEMORY);
821 #endif /* CONFIG_MOVABLE_NODE */
823 /* check which state of node_states will be changed when online memory */
824 static void node_states_check_changes_online(unsigned long nr_pages,
825 struct zone *zone, struct memory_notify *arg)
827 int nid = zone_to_nid(zone);
828 enum zone_type zone_last = ZONE_NORMAL;
831 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
832 * contains nodes which have zones of 0...ZONE_NORMAL,
833 * set zone_last to ZONE_NORMAL.
835 * If we don't have HIGHMEM nor movable node,
836 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
837 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
839 if (N_MEMORY == N_NORMAL_MEMORY)
840 zone_last = ZONE_MOVABLE;
843 * if the memory to be online is in a zone of 0...zone_last, and
844 * the zones of 0...zone_last don't have memory before online, we will
845 * need to set the node to node_states[N_NORMAL_MEMORY] after
846 * the memory is online.
848 if (zone_idx(zone) <= zone_last && !node_state(nid, N_NORMAL_MEMORY))
849 arg->status_change_nid_normal = nid;
851 arg->status_change_nid_normal = -1;
853 #ifdef CONFIG_HIGHMEM
855 * If we have movable node, node_states[N_HIGH_MEMORY]
856 * contains nodes which have zones of 0...ZONE_HIGHMEM,
857 * set zone_last to ZONE_HIGHMEM.
859 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
860 * contains nodes which have zones of 0...ZONE_MOVABLE,
861 * set zone_last to ZONE_MOVABLE.
863 zone_last = ZONE_HIGHMEM;
864 if (N_MEMORY == N_HIGH_MEMORY)
865 zone_last = ZONE_MOVABLE;
867 if (zone_idx(zone) <= zone_last && !node_state(nid, N_HIGH_MEMORY))
868 arg->status_change_nid_high = nid;
870 arg->status_change_nid_high = -1;
872 arg->status_change_nid_high = arg->status_change_nid_normal;
876 * if the node don't have memory befor online, we will need to
877 * set the node to node_states[N_MEMORY] after the memory
880 if (!node_state(nid, N_MEMORY))
881 arg->status_change_nid = nid;
883 arg->status_change_nid = -1;
886 static void node_states_set_node(int node, struct memory_notify *arg)
888 if (arg->status_change_nid_normal >= 0)
889 node_set_state(node, N_NORMAL_MEMORY);
891 if (arg->status_change_nid_high >= 0)
892 node_set_state(node, N_HIGH_MEMORY);
894 node_set_state(node, N_MEMORY);
898 int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type)
901 unsigned long onlined_pages = 0;
903 int need_zonelists_rebuild = 0;
906 struct memory_notify arg;
908 lock_memory_hotplug();
910 * This doesn't need a lock to do pfn_to_page().
911 * The section can't be removed here because of the
912 * memory_block->state_mutex.
914 zone = page_zone(pfn_to_page(pfn));
916 if ((zone_idx(zone) > ZONE_NORMAL || online_type == ONLINE_MOVABLE) &&
917 !can_online_high_movable(zone)) {
918 unlock_memory_hotplug();
922 if (online_type == ONLINE_KERNEL && zone_idx(zone) == ZONE_MOVABLE) {
923 if (move_pfn_range_left(zone - 1, zone, pfn, pfn + nr_pages)) {
924 unlock_memory_hotplug();
928 if (online_type == ONLINE_MOVABLE && zone_idx(zone) == ZONE_MOVABLE - 1) {
929 if (move_pfn_range_right(zone, zone + 1, pfn, pfn + nr_pages)) {
930 unlock_memory_hotplug();
935 /* Previous code may changed the zone of the pfn range */
936 zone = page_zone(pfn_to_page(pfn));
939 arg.nr_pages = nr_pages;
940 node_states_check_changes_online(nr_pages, zone, &arg);
942 nid = page_to_nid(pfn_to_page(pfn));
944 ret = memory_notify(MEM_GOING_ONLINE, &arg);
945 ret = notifier_to_errno(ret);
947 memory_notify(MEM_CANCEL_ONLINE, &arg);
948 unlock_memory_hotplug();
952 * If this zone is not populated, then it is not in zonelist.
953 * This means the page allocator ignores this zone.
954 * So, zonelist must be updated after online.
956 mutex_lock(&zonelists_mutex);
957 if (!populated_zone(zone)) {
958 need_zonelists_rebuild = 1;
959 build_all_zonelists(NULL, zone);
962 ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
965 if (need_zonelists_rebuild)
966 zone_pcp_reset(zone);
967 mutex_unlock(&zonelists_mutex);
968 printk(KERN_DEBUG "online_pages [mem %#010llx-%#010llx] failed\n",
969 (unsigned long long) pfn << PAGE_SHIFT,
970 (((unsigned long long) pfn + nr_pages)
972 memory_notify(MEM_CANCEL_ONLINE, &arg);
973 unlock_memory_hotplug();
977 zone->present_pages += onlined_pages;
979 pgdat_resize_lock(zone->zone_pgdat, &flags);
980 zone->zone_pgdat->node_present_pages += onlined_pages;
981 pgdat_resize_unlock(zone->zone_pgdat, &flags);
984 node_states_set_node(zone_to_nid(zone), &arg);
985 if (need_zonelists_rebuild)
986 build_all_zonelists(NULL, NULL);
988 zone_pcp_update(zone);
991 mutex_unlock(&zonelists_mutex);
993 init_per_zone_wmark_min();
996 kswapd_run(zone_to_nid(zone));
998 vm_total_pages = nr_free_pagecache_pages();
1000 writeback_set_ratelimit();
1003 memory_notify(MEM_ONLINE, &arg);
1004 unlock_memory_hotplug();
1008 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
1010 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1011 static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
1013 struct pglist_data *pgdat;
1014 unsigned long zones_size[MAX_NR_ZONES] = {0};
1015 unsigned long zholes_size[MAX_NR_ZONES] = {0};
1016 unsigned long start_pfn = start >> PAGE_SHIFT;
1018 pgdat = NODE_DATA(nid);
1020 pgdat = arch_alloc_nodedata(nid);
1024 arch_refresh_nodedata(nid, pgdat);
1027 /* we can use NODE_DATA(nid) from here */
1029 /* init node's zones as empty zones, we don't have any present pages.*/
1030 free_area_init_node(nid, zones_size, start_pfn, zholes_size);
1033 * The node we allocated has no zone fallback lists. For avoiding
1034 * to access not-initialized zonelist, build here.
1036 mutex_lock(&zonelists_mutex);
1037 build_all_zonelists(pgdat, NULL);
1038 mutex_unlock(&zonelists_mutex);
1043 static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
1045 arch_refresh_nodedata(nid, NULL);
1046 arch_free_nodedata(pgdat);
1052 * called by cpu_up() to online a node without onlined memory.
1054 int mem_online_node(int nid)
1059 lock_memory_hotplug();
1060 pgdat = hotadd_new_pgdat(nid, 0);
1065 node_set_online(nid);
1066 ret = register_one_node(nid);
1070 unlock_memory_hotplug();
1074 static int check_hotplug_memory_range(u64 start, u64 size)
1076 u64 start_pfn = start >> PAGE_SHIFT;
1077 u64 nr_pages = size >> PAGE_SHIFT;
1079 /* Memory range must be aligned with section */
1080 if ((start_pfn & ~PAGE_SECTION_MASK) ||
1081 (nr_pages % PAGES_PER_SECTION) || (!nr_pages)) {
1082 pr_err("Section-unaligned hotplug range: start 0x%llx, size 0x%llx\n",
1083 (unsigned long long)start,
1084 (unsigned long long)size);
1091 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1092 int __ref add_memory(int nid, u64 start, u64 size)
1094 pg_data_t *pgdat = NULL;
1097 struct resource *res;
1100 ret = check_hotplug_memory_range(start, size);
1104 lock_memory_hotplug();
1106 res = register_memory_resource(start, size);
1111 { /* Stupid hack to suppress address-never-null warning */
1112 void *p = NODE_DATA(nid);
1115 new_node = !node_online(nid);
1117 pgdat = hotadd_new_pgdat(nid, start);
1123 /* call arch's memory hotadd */
1124 ret = arch_add_memory(nid, start, size);
1129 /* we online node here. we can't roll back from here. */
1130 node_set_online(nid);
1133 ret = register_one_node(nid);
1135 * If sysfs file of new node can't create, cpu on the node
1136 * can't be hot-added. There is no rollback way now.
1137 * So, check by BUG_ON() to catch it reluctantly..
1142 /* create new memmap entry */
1143 firmware_map_add_hotplug(start, start + size, "System RAM");
1148 /* rollback pgdat allocation and others */
1150 rollback_node_hotadd(nid, pgdat);
1151 release_memory_resource(res);
1154 unlock_memory_hotplug();
1157 EXPORT_SYMBOL_GPL(add_memory);
1159 #ifdef CONFIG_MEMORY_HOTREMOVE
1161 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1162 * set and the size of the free page is given by page_order(). Using this,
1163 * the function determines if the pageblock contains only free pages.
1164 * Due to buddy contraints, a free page at least the size of a pageblock will
1165 * be located at the start of the pageblock
1167 static inline int pageblock_free(struct page *page)
1169 return PageBuddy(page) && page_order(page) >= pageblock_order;
1172 /* Return the start of the next active pageblock after a given page */
1173 static struct page *next_active_pageblock(struct page *page)
1175 /* Ensure the starting page is pageblock-aligned */
1176 BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
1178 /* If the entire pageblock is free, move to the end of free page */
1179 if (pageblock_free(page)) {
1181 /* be careful. we don't have locks, page_order can be changed.*/
1182 order = page_order(page);
1183 if ((order < MAX_ORDER) && (order >= pageblock_order))
1184 return page + (1 << order);
1187 return page + pageblock_nr_pages;
1190 /* Checks if this range of memory is likely to be hot-removable. */
1191 int is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
1193 struct page *page = pfn_to_page(start_pfn);
1194 struct page *end_page = page + nr_pages;
1196 /* Check the starting page of each pageblock within the range */
1197 for (; page < end_page; page = next_active_pageblock(page)) {
1198 if (!is_pageblock_removable_nolock(page))
1203 /* All pageblocks in the memory block are likely to be hot-removable */
1208 * Confirm all pages in a range [start, end) is belongs to the same zone.
1210 static int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn)
1213 struct zone *zone = NULL;
1216 for (pfn = start_pfn;
1218 pfn += MAX_ORDER_NR_PAGES) {
1220 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1221 while ((i < MAX_ORDER_NR_PAGES) && !pfn_valid_within(pfn + i))
1223 if (i == MAX_ORDER_NR_PAGES)
1225 page = pfn_to_page(pfn + i);
1226 if (zone && page_zone(page) != zone)
1228 zone = page_zone(page);
1234 * Scan pfn range [start,end) to find movable/migratable pages (LRU pages
1235 * and hugepages). We scan pfn because it's much easier than scanning over
1236 * linked list. This function returns the pfn of the first found movable
1237 * page if it's found, otherwise 0.
1239 static unsigned long scan_movable_pages(unsigned long start, unsigned long end)
1243 for (pfn = start; pfn < end; pfn++) {
1244 if (pfn_valid(pfn)) {
1245 page = pfn_to_page(pfn);
1248 if (PageHuge(page)) {
1249 if (is_hugepage_active(page))
1252 pfn = round_up(pfn + 1,
1253 1 << compound_order(page)) - 1;
1260 #define NR_OFFLINE_AT_ONCE_PAGES (256)
1262 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1266 int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
1267 int not_managed = 0;
1271 for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
1272 if (!pfn_valid(pfn))
1274 page = pfn_to_page(pfn);
1276 if (PageHuge(page)) {
1277 struct page *head = compound_head(page);
1278 pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1;
1279 if (compound_order(head) > PFN_SECTION_SHIFT) {
1283 if (isolate_huge_page(page, &source))
1284 move_pages -= 1 << compound_order(head);
1288 if (!get_page_unless_zero(page))
1291 * We can skip free pages. And we can only deal with pages on
1294 ret = isolate_lru_page(page);
1295 if (!ret) { /* Success */
1297 list_add_tail(&page->lru, &source);
1299 inc_zone_page_state(page, NR_ISOLATED_ANON +
1300 page_is_file_cache(page));
1303 #ifdef CONFIG_DEBUG_VM
1304 printk(KERN_ALERT "removing pfn %lx from LRU failed\n",
1309 /* Because we don't have big zone->lock. we should
1310 check this again here. */
1311 if (page_count(page)) {
1318 if (!list_empty(&source)) {
1320 putback_movable_pages(&source);
1325 * alloc_migrate_target should be improooooved!!
1326 * migrate_pages returns # of failed pages.
1328 ret = migrate_pages(&source, alloc_migrate_target, 0,
1329 MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
1331 putback_movable_pages(&source);
1338 * remove from free_area[] and mark all as Reserved.
1341 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1344 __offline_isolated_pages(start, start + nr_pages);
1349 offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
1351 walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
1352 offline_isolated_pages_cb);
1356 * Check all pages in range, recoreded as memory resource, are isolated.
1359 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1363 long offlined = *(long *)data;
1364 ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true);
1365 offlined = nr_pages;
1367 *(long *)data += offlined;
1372 check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
1377 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
1378 check_pages_isolated_cb);
1380 offlined = (long)ret;
1384 #ifdef CONFIG_MOVABLE_NODE
1386 * When CONFIG_MOVABLE_NODE, we permit offlining of a node which doesn't have
1389 static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1393 #else /* CONFIG_MOVABLE_NODE */
1394 /* ensure the node has NORMAL memory if it is still online */
1395 static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1397 struct pglist_data *pgdat = zone->zone_pgdat;
1398 unsigned long present_pages = 0;
1401 for (zt = 0; zt <= ZONE_NORMAL; zt++)
1402 present_pages += pgdat->node_zones[zt].present_pages;
1404 if (present_pages > nr_pages)
1408 for (; zt <= ZONE_MOVABLE; zt++)
1409 present_pages += pgdat->node_zones[zt].present_pages;
1412 * we can't offline the last normal memory until all
1413 * higher memory is offlined.
1415 return present_pages == 0;
1417 #endif /* CONFIG_MOVABLE_NODE */
1419 /* check which state of node_states will be changed when offline memory */
1420 static void node_states_check_changes_offline(unsigned long nr_pages,
1421 struct zone *zone, struct memory_notify *arg)
1423 struct pglist_data *pgdat = zone->zone_pgdat;
1424 unsigned long present_pages = 0;
1425 enum zone_type zt, zone_last = ZONE_NORMAL;
1428 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
1429 * contains nodes which have zones of 0...ZONE_NORMAL,
1430 * set zone_last to ZONE_NORMAL.
1432 * If we don't have HIGHMEM nor movable node,
1433 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
1434 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
1436 if (N_MEMORY == N_NORMAL_MEMORY)
1437 zone_last = ZONE_MOVABLE;
1440 * check whether node_states[N_NORMAL_MEMORY] will be changed.
1441 * If the memory to be offline is in a zone of 0...zone_last,
1442 * and it is the last present memory, 0...zone_last will
1443 * become empty after offline , thus we can determind we will
1444 * need to clear the node from node_states[N_NORMAL_MEMORY].
1446 for (zt = 0; zt <= zone_last; zt++)
1447 present_pages += pgdat->node_zones[zt].present_pages;
1448 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1449 arg->status_change_nid_normal = zone_to_nid(zone);
1451 arg->status_change_nid_normal = -1;
1453 #ifdef CONFIG_HIGHMEM
1455 * If we have movable node, node_states[N_HIGH_MEMORY]
1456 * contains nodes which have zones of 0...ZONE_HIGHMEM,
1457 * set zone_last to ZONE_HIGHMEM.
1459 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
1460 * contains nodes which have zones of 0...ZONE_MOVABLE,
1461 * set zone_last to ZONE_MOVABLE.
1463 zone_last = ZONE_HIGHMEM;
1464 if (N_MEMORY == N_HIGH_MEMORY)
1465 zone_last = ZONE_MOVABLE;
1467 for (; zt <= zone_last; zt++)
1468 present_pages += pgdat->node_zones[zt].present_pages;
1469 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1470 arg->status_change_nid_high = zone_to_nid(zone);
1472 arg->status_change_nid_high = -1;
1474 arg->status_change_nid_high = arg->status_change_nid_normal;
1478 * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE
1480 zone_last = ZONE_MOVABLE;
1483 * check whether node_states[N_HIGH_MEMORY] will be changed
1484 * If we try to offline the last present @nr_pages from the node,
1485 * we can determind we will need to clear the node from
1486 * node_states[N_HIGH_MEMORY].
1488 for (; zt <= zone_last; zt++)
1489 present_pages += pgdat->node_zones[zt].present_pages;
1490 if (nr_pages >= present_pages)
1491 arg->status_change_nid = zone_to_nid(zone);
1493 arg->status_change_nid = -1;
1496 static void node_states_clear_node(int node, struct memory_notify *arg)
1498 if (arg->status_change_nid_normal >= 0)
1499 node_clear_state(node, N_NORMAL_MEMORY);
1501 if ((N_MEMORY != N_NORMAL_MEMORY) &&
1502 (arg->status_change_nid_high >= 0))
1503 node_clear_state(node, N_HIGH_MEMORY);
1505 if ((N_MEMORY != N_HIGH_MEMORY) &&
1506 (arg->status_change_nid >= 0))
1507 node_clear_state(node, N_MEMORY);
1510 static int __ref __offline_pages(unsigned long start_pfn,
1511 unsigned long end_pfn, unsigned long timeout)
1513 unsigned long pfn, nr_pages, expire;
1514 long offlined_pages;
1515 int ret, drain, retry_max, node;
1516 unsigned long flags;
1518 struct memory_notify arg;
1520 /* at least, alignment against pageblock is necessary */
1521 if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
1523 if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
1525 /* This makes hotplug much easier...and readable.
1526 we assume this for now. .*/
1527 if (!test_pages_in_a_zone(start_pfn, end_pfn))
1530 lock_memory_hotplug();
1532 zone = page_zone(pfn_to_page(start_pfn));
1533 node = zone_to_nid(zone);
1534 nr_pages = end_pfn - start_pfn;
1537 if (zone_idx(zone) <= ZONE_NORMAL && !can_offline_normal(zone, nr_pages))
1540 /* set above range as isolated */
1541 ret = start_isolate_page_range(start_pfn, end_pfn,
1542 MIGRATE_MOVABLE, true);
1546 arg.start_pfn = start_pfn;
1547 arg.nr_pages = nr_pages;
1548 node_states_check_changes_offline(nr_pages, zone, &arg);
1550 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1551 ret = notifier_to_errno(ret);
1553 goto failed_removal;
1556 expire = jiffies + timeout;
1560 /* start memory hot removal */
1562 if (time_after(jiffies, expire))
1563 goto failed_removal;
1565 if (signal_pending(current))
1566 goto failed_removal;
1569 lru_add_drain_all();
1574 pfn = scan_movable_pages(start_pfn, end_pfn);
1575 if (pfn) { /* We have movable pages */
1576 ret = do_migrate_range(pfn, end_pfn);
1582 if (--retry_max == 0)
1583 goto failed_removal;
1589 /* drain all zone's lru pagevec, this is asynchronous... */
1590 lru_add_drain_all();
1592 /* drain pcp pages, this is synchronous. */
1595 * dissolve free hugepages in the memory block before doing offlining
1596 * actually in order to make hugetlbfs's object counting consistent.
1598 dissolve_free_huge_pages(start_pfn, end_pfn);
1600 offlined_pages = check_pages_isolated(start_pfn, end_pfn);
1601 if (offlined_pages < 0) {
1603 goto failed_removal;
1605 printk(KERN_INFO "Offlined Pages %ld\n", offlined_pages);
1606 /* Ok, all of our target is isolated.
1607 We cannot do rollback at this point. */
1608 offline_isolated_pages(start_pfn, end_pfn);
1609 /* reset pagetype flags and makes migrate type to be MOVABLE */
1610 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1611 /* removal success */
1612 adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages);
1613 zone->present_pages -= offlined_pages;
1615 pgdat_resize_lock(zone->zone_pgdat, &flags);
1616 zone->zone_pgdat->node_present_pages -= offlined_pages;
1617 pgdat_resize_unlock(zone->zone_pgdat, &flags);
1619 init_per_zone_wmark_min();
1621 if (!populated_zone(zone)) {
1622 zone_pcp_reset(zone);
1623 mutex_lock(&zonelists_mutex);
1624 build_all_zonelists(NULL, NULL);
1625 mutex_unlock(&zonelists_mutex);
1627 zone_pcp_update(zone);
1629 node_states_clear_node(node, &arg);
1630 if (arg.status_change_nid >= 0)
1633 vm_total_pages = nr_free_pagecache_pages();
1634 writeback_set_ratelimit();
1636 memory_notify(MEM_OFFLINE, &arg);
1637 unlock_memory_hotplug();
1641 printk(KERN_INFO "memory offlining [mem %#010llx-%#010llx] failed\n",
1642 (unsigned long long) start_pfn << PAGE_SHIFT,
1643 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
1644 memory_notify(MEM_CANCEL_OFFLINE, &arg);
1645 /* pushback to free area */
1646 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1649 unlock_memory_hotplug();
1653 int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1655 return __offline_pages(start_pfn, start_pfn + nr_pages, 120 * HZ);
1657 #endif /* CONFIG_MEMORY_HOTREMOVE */
1660 * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
1661 * @start_pfn: start pfn of the memory range
1662 * @end_pfn: end pfn of the memory range
1663 * @arg: argument passed to func
1664 * @func: callback for each memory section walked
1666 * This function walks through all present mem sections in range
1667 * [start_pfn, end_pfn) and call func on each mem section.
1669 * Returns the return value of func.
1671 int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn,
1672 void *arg, int (*func)(struct memory_block *, void *))
1674 struct memory_block *mem = NULL;
1675 struct mem_section *section;
1676 unsigned long pfn, section_nr;
1679 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1680 section_nr = pfn_to_section_nr(pfn);
1681 if (!present_section_nr(section_nr))
1684 section = __nr_to_section(section_nr);
1685 /* same memblock? */
1687 if ((section_nr >= mem->start_section_nr) &&
1688 (section_nr <= mem->end_section_nr))
1691 mem = find_memory_block_hinted(section, mem);
1695 ret = func(mem, arg);
1697 kobject_put(&mem->dev.kobj);
1703 kobject_put(&mem->dev.kobj);
1708 #ifdef CONFIG_MEMORY_HOTREMOVE
1709 static int is_memblock_offlined_cb(struct memory_block *mem, void *arg)
1711 int ret = !is_memblock_offlined(mem);
1713 if (unlikely(ret)) {
1714 phys_addr_t beginpa, endpa;
1716 beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
1717 endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1;
1718 pr_warn("removing memory fails, because memory "
1719 "[%pa-%pa] is onlined\n",
1726 static int check_cpu_on_node(pg_data_t *pgdat)
1730 for_each_present_cpu(cpu) {
1731 if (cpu_to_node(cpu) == pgdat->node_id)
1733 * the cpu on this node isn't removed, and we can't
1734 * offline this node.
1742 static void unmap_cpu_on_node(pg_data_t *pgdat)
1744 #ifdef CONFIG_ACPI_NUMA
1747 for_each_possible_cpu(cpu)
1748 if (cpu_to_node(cpu) == pgdat->node_id)
1749 numa_clear_node(cpu);
1753 static int check_and_unmap_cpu_on_node(pg_data_t *pgdat)
1757 ret = check_cpu_on_node(pgdat);
1762 * the node will be offlined when we come here, so we can clear
1763 * the cpu_to_node() now.
1766 unmap_cpu_on_node(pgdat);
1773 * Offline a node if all memory sections and cpus of the node are removed.
1775 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1776 * and online/offline operations before this call.
1778 void try_offline_node(int nid)
1780 pg_data_t *pgdat = NODE_DATA(nid);
1781 unsigned long start_pfn = pgdat->node_start_pfn;
1782 unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
1784 struct page *pgdat_page = virt_to_page(pgdat);
1787 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1788 unsigned long section_nr = pfn_to_section_nr(pfn);
1790 if (!present_section_nr(section_nr))
1793 if (pfn_to_nid(pfn) != nid)
1797 * some memory sections of this node are not removed, and we
1798 * can't offline node now.
1803 if (check_and_unmap_cpu_on_node(pgdat))
1807 * all memory/cpu of this node are removed, we can offline this
1810 node_set_offline(nid);
1811 unregister_one_node(nid);
1813 if (!PageSlab(pgdat_page) && !PageCompound(pgdat_page))
1814 /* node data is allocated from boot memory */
1817 /* free waittable in each zone */
1818 for (i = 0; i < MAX_NR_ZONES; i++) {
1819 struct zone *zone = pgdat->node_zones + i;
1822 * wait_table may be allocated from boot memory,
1823 * here only free if it's allocated by vmalloc.
1825 if (is_vmalloc_addr(zone->wait_table))
1826 vfree(zone->wait_table);
1830 * Since there is no way to guarentee the address of pgdat/zone is not
1831 * on stack of any kernel threads or used by other kernel objects
1832 * without reference counting or other symchronizing method, do not
1833 * reset node_data and free pgdat here. Just reset it to 0 and reuse
1834 * the memory when the node is online again.
1836 memset(pgdat, 0, sizeof(*pgdat));
1838 EXPORT_SYMBOL(try_offline_node);
1843 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1844 * and online/offline operations before this call, as required by
1845 * try_offline_node().
1847 void __ref remove_memory(int nid, u64 start, u64 size)
1851 BUG_ON(check_hotplug_memory_range(start, size));
1853 lock_memory_hotplug();
1856 * All memory blocks must be offlined before removing memory. Check
1857 * whether all memory blocks in question are offline and trigger a BUG()
1858 * if this is not the case.
1860 ret = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL,
1861 is_memblock_offlined_cb);
1863 unlock_memory_hotplug();
1867 /* remove memmap entry */
1868 firmware_map_remove(start, start + size, "System RAM");
1870 arch_remove_memory(start, size);
1872 try_offline_node(nid);
1874 unlock_memory_hotplug();
1876 EXPORT_SYMBOL_GPL(remove_memory);
1877 #endif /* CONFIG_MEMORY_HOTREMOVE */