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);
57 void unlock_memory_hotplug(void)
59 mutex_unlock(&mem_hotplug_mutex);
63 /* add this memory to iomem resource */
64 static struct resource *register_memory_resource(u64 start, u64 size)
67 res = kzalloc(sizeof(struct resource), GFP_KERNEL);
70 res->name = "System RAM";
72 res->end = start + size - 1;
73 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
74 if (request_resource(&iomem_resource, res) < 0) {
75 pr_debug("System RAM resource %pR cannot be added\n", res);
82 static void release_memory_resource(struct resource *res)
86 release_resource(res);
91 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
92 void get_page_bootmem(unsigned long info, struct page *page,
95 page->lru.next = (struct list_head *) type;
97 set_page_private(page, info);
98 atomic_inc(&page->_count);
101 void put_page_bootmem(struct page *page)
105 type = (unsigned long) page->lru.next;
106 BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE ||
107 type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE);
109 if (atomic_dec_return(&page->_count) == 1) {
110 ClearPagePrivate(page);
111 set_page_private(page, 0);
112 INIT_LIST_HEAD(&page->lru);
113 free_reserved_page(page);
117 #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
118 #ifndef CONFIG_SPARSEMEM_VMEMMAP
119 static void register_page_bootmem_info_section(unsigned long start_pfn)
121 unsigned long *usemap, mapsize, section_nr, i;
122 struct mem_section *ms;
123 struct page *page, *memmap;
125 section_nr = pfn_to_section_nr(start_pfn);
126 ms = __nr_to_section(section_nr);
128 /* Get section's memmap address */
129 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
132 * Get page for the memmap's phys address
133 * XXX: need more consideration for sparse_vmemmap...
135 page = virt_to_page(memmap);
136 mapsize = sizeof(struct page) * PAGES_PER_SECTION;
137 mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
139 /* remember memmap's page */
140 for (i = 0; i < mapsize; i++, page++)
141 get_page_bootmem(section_nr, page, SECTION_INFO);
143 usemap = __nr_to_section(section_nr)->pageblock_flags;
144 page = virt_to_page(usemap);
146 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
148 for (i = 0; i < mapsize; i++, page++)
149 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
152 #else /* CONFIG_SPARSEMEM_VMEMMAP */
153 static void register_page_bootmem_info_section(unsigned long start_pfn)
155 unsigned long *usemap, mapsize, section_nr, i;
156 struct mem_section *ms;
157 struct page *page, *memmap;
159 if (!pfn_valid(start_pfn))
162 section_nr = pfn_to_section_nr(start_pfn);
163 ms = __nr_to_section(section_nr);
165 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
167 register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION);
169 usemap = __nr_to_section(section_nr)->pageblock_flags;
170 page = virt_to_page(usemap);
172 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
174 for (i = 0; i < mapsize; i++, page++)
175 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
177 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
179 void register_page_bootmem_info_node(struct pglist_data *pgdat)
181 unsigned long i, pfn, end_pfn, nr_pages;
182 int node = pgdat->node_id;
186 nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
187 page = virt_to_page(pgdat);
189 for (i = 0; i < nr_pages; i++, page++)
190 get_page_bootmem(node, page, NODE_INFO);
192 zone = &pgdat->node_zones[0];
193 for (; zone < pgdat->node_zones + MAX_NR_ZONES - 1; zone++) {
194 if (zone_is_initialized(zone)) {
195 nr_pages = zone->wait_table_hash_nr_entries
196 * sizeof(wait_queue_head_t);
197 nr_pages = PAGE_ALIGN(nr_pages) >> PAGE_SHIFT;
198 page = virt_to_page(zone->wait_table);
200 for (i = 0; i < nr_pages; i++, page++)
201 get_page_bootmem(node, page, NODE_INFO);
205 pfn = pgdat->node_start_pfn;
206 end_pfn = pgdat_end_pfn(pgdat);
208 /* register section info */
209 for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
211 * Some platforms can assign the same pfn to multiple nodes - on
212 * node0 as well as nodeN. To avoid registering a pfn against
213 * multiple nodes we check that this pfn does not already
214 * reside in some other nodes.
216 if (pfn_valid(pfn) && (pfn_to_nid(pfn) == node))
217 register_page_bootmem_info_section(pfn);
220 #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
222 static void grow_zone_span(struct zone *zone, unsigned long start_pfn,
223 unsigned long end_pfn)
225 unsigned long old_zone_end_pfn;
227 zone_span_writelock(zone);
229 old_zone_end_pfn = zone_end_pfn(zone);
230 if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
231 zone->zone_start_pfn = start_pfn;
233 zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
234 zone->zone_start_pfn;
236 zone_span_writeunlock(zone);
239 static void resize_zone(struct zone *zone, unsigned long start_pfn,
240 unsigned long end_pfn)
242 zone_span_writelock(zone);
244 if (end_pfn - start_pfn) {
245 zone->zone_start_pfn = start_pfn;
246 zone->spanned_pages = end_pfn - start_pfn;
249 * make it consist as free_area_init_core(),
250 * if spanned_pages = 0, then keep start_pfn = 0
252 zone->zone_start_pfn = 0;
253 zone->spanned_pages = 0;
256 zone_span_writeunlock(zone);
259 static void fix_zone_id(struct zone *zone, unsigned long start_pfn,
260 unsigned long end_pfn)
262 enum zone_type zid = zone_idx(zone);
263 int nid = zone->zone_pgdat->node_id;
266 for (pfn = start_pfn; pfn < end_pfn; pfn++)
267 set_page_links(pfn_to_page(pfn), zid, nid, pfn);
270 /* Can fail with -ENOMEM from allocating a wait table with vmalloc() or
271 * alloc_bootmem_node_nopanic() */
272 static int __ref ensure_zone_is_initialized(struct zone *zone,
273 unsigned long start_pfn, unsigned long num_pages)
275 if (!zone_is_initialized(zone))
276 return init_currently_empty_zone(zone, start_pfn, num_pages,
281 static int __meminit move_pfn_range_left(struct zone *z1, struct zone *z2,
282 unsigned long start_pfn, unsigned long end_pfn)
286 unsigned long z1_start_pfn;
288 ret = ensure_zone_is_initialized(z1, start_pfn, end_pfn - start_pfn);
292 pgdat_resize_lock(z1->zone_pgdat, &flags);
294 /* can't move pfns which are higher than @z2 */
295 if (end_pfn > zone_end_pfn(z2))
297 /* the move out part must be at the left most of @z2 */
298 if (start_pfn > z2->zone_start_pfn)
300 /* must included/overlap */
301 if (end_pfn <= z2->zone_start_pfn)
304 /* use start_pfn for z1's start_pfn if z1 is empty */
305 if (!zone_is_empty(z1))
306 z1_start_pfn = z1->zone_start_pfn;
308 z1_start_pfn = start_pfn;
310 resize_zone(z1, z1_start_pfn, end_pfn);
311 resize_zone(z2, end_pfn, zone_end_pfn(z2));
313 pgdat_resize_unlock(z1->zone_pgdat, &flags);
315 fix_zone_id(z1, start_pfn, end_pfn);
319 pgdat_resize_unlock(z1->zone_pgdat, &flags);
323 static int __meminit move_pfn_range_right(struct zone *z1, struct zone *z2,
324 unsigned long start_pfn, unsigned long end_pfn)
328 unsigned long z2_end_pfn;
330 ret = ensure_zone_is_initialized(z2, start_pfn, end_pfn - start_pfn);
334 pgdat_resize_lock(z1->zone_pgdat, &flags);
336 /* can't move pfns which are lower than @z1 */
337 if (z1->zone_start_pfn > start_pfn)
339 /* the move out part mast at the right most of @z1 */
340 if (zone_end_pfn(z1) > end_pfn)
342 /* must included/overlap */
343 if (start_pfn >= zone_end_pfn(z1))
346 /* use end_pfn for z2's end_pfn if z2 is empty */
347 if (!zone_is_empty(z2))
348 z2_end_pfn = zone_end_pfn(z2);
350 z2_end_pfn = end_pfn;
352 resize_zone(z1, z1->zone_start_pfn, start_pfn);
353 resize_zone(z2, start_pfn, z2_end_pfn);
355 pgdat_resize_unlock(z1->zone_pgdat, &flags);
357 fix_zone_id(z2, start_pfn, end_pfn);
361 pgdat_resize_unlock(z1->zone_pgdat, &flags);
365 static void grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn,
366 unsigned long end_pfn)
368 unsigned long old_pgdat_end_pfn =
369 pgdat->node_start_pfn + pgdat->node_spanned_pages;
371 if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
372 pgdat->node_start_pfn = start_pfn;
374 pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
375 pgdat->node_start_pfn;
378 static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn)
380 struct pglist_data *pgdat = zone->zone_pgdat;
381 int nr_pages = PAGES_PER_SECTION;
382 int nid = pgdat->node_id;
387 zone_type = zone - pgdat->node_zones;
388 ret = ensure_zone_is_initialized(zone, phys_start_pfn, nr_pages);
392 pgdat_resize_lock(zone->zone_pgdat, &flags);
393 grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages);
394 grow_pgdat_span(zone->zone_pgdat, phys_start_pfn,
395 phys_start_pfn + nr_pages);
396 pgdat_resize_unlock(zone->zone_pgdat, &flags);
397 memmap_init_zone(nr_pages, nid, zone_type,
398 phys_start_pfn, MEMMAP_HOTPLUG);
402 static int __meminit __add_section(int nid, struct zone *zone,
403 unsigned long phys_start_pfn)
405 int nr_pages = PAGES_PER_SECTION;
408 if (pfn_valid(phys_start_pfn))
411 ret = sparse_add_one_section(zone, phys_start_pfn, nr_pages);
416 ret = __add_zone(zone, phys_start_pfn);
421 return register_new_memory(nid, __pfn_to_section(phys_start_pfn));
425 * Reasonably generic function for adding memory. It is
426 * expected that archs that support memory hotplug will
427 * call this function after deciding the zone to which to
430 int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn,
431 unsigned long nr_pages)
435 int start_sec, end_sec;
436 /* during initialize mem_map, align hot-added range to section */
437 start_sec = pfn_to_section_nr(phys_start_pfn);
438 end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
440 for (i = start_sec; i <= end_sec; i++) {
441 err = __add_section(nid, zone, i << PFN_SECTION_SHIFT);
444 * EEXIST is finally dealt with by ioresource collision
445 * check. see add_memory() => register_memory_resource()
446 * Warning will be printed if there is collision.
448 if (err && (err != -EEXIST))
455 EXPORT_SYMBOL_GPL(__add_pages);
457 #ifdef CONFIG_MEMORY_HOTREMOVE
458 /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
459 static int find_smallest_section_pfn(int nid, struct zone *zone,
460 unsigned long start_pfn,
461 unsigned long end_pfn)
463 struct mem_section *ms;
465 for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) {
466 ms = __pfn_to_section(start_pfn);
468 if (unlikely(!valid_section(ms)))
471 if (unlikely(pfn_to_nid(start_pfn) != nid))
474 if (zone && zone != page_zone(pfn_to_page(start_pfn)))
483 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
484 static int find_biggest_section_pfn(int nid, struct zone *zone,
485 unsigned long start_pfn,
486 unsigned long end_pfn)
488 struct mem_section *ms;
491 /* pfn is the end pfn of a memory section. */
493 for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) {
494 ms = __pfn_to_section(pfn);
496 if (unlikely(!valid_section(ms)))
499 if (unlikely(pfn_to_nid(pfn) != nid))
502 if (zone && zone != page_zone(pfn_to_page(pfn)))
511 static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
512 unsigned long end_pfn)
514 unsigned long zone_start_pfn = zone->zone_start_pfn;
515 unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */
516 unsigned long zone_end_pfn = z;
518 struct mem_section *ms;
519 int nid = zone_to_nid(zone);
521 zone_span_writelock(zone);
522 if (zone_start_pfn == start_pfn) {
524 * If the section is smallest section in the zone, it need
525 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
526 * In this case, we find second smallest valid mem_section
527 * for shrinking zone.
529 pfn = find_smallest_section_pfn(nid, zone, end_pfn,
532 zone->zone_start_pfn = pfn;
533 zone->spanned_pages = zone_end_pfn - pfn;
535 } else if (zone_end_pfn == end_pfn) {
537 * If the section is biggest section in the zone, it need
538 * shrink zone->spanned_pages.
539 * In this case, we find second biggest valid mem_section for
542 pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn,
545 zone->spanned_pages = pfn - zone_start_pfn + 1;
549 * The section is not biggest or smallest mem_section in the zone, it
550 * only creates a hole in the zone. So in this case, we need not
551 * change the zone. But perhaps, the zone has only hole data. Thus
552 * it check the zone has only hole or not.
554 pfn = zone_start_pfn;
555 for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) {
556 ms = __pfn_to_section(pfn);
558 if (unlikely(!valid_section(ms)))
561 if (page_zone(pfn_to_page(pfn)) != zone)
564 /* If the section is current section, it continues the loop */
565 if (start_pfn == pfn)
568 /* If we find valid section, we have nothing to do */
569 zone_span_writeunlock(zone);
573 /* The zone has no valid section */
574 zone->zone_start_pfn = 0;
575 zone->spanned_pages = 0;
576 zone_span_writeunlock(zone);
579 static void shrink_pgdat_span(struct pglist_data *pgdat,
580 unsigned long start_pfn, unsigned long end_pfn)
582 unsigned long pgdat_start_pfn = pgdat->node_start_pfn;
583 unsigned long pgdat_end_pfn =
584 pgdat->node_start_pfn + pgdat->node_spanned_pages;
586 struct mem_section *ms;
587 int nid = pgdat->node_id;
589 if (pgdat_start_pfn == start_pfn) {
591 * If the section is smallest section in the pgdat, it need
592 * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages.
593 * In this case, we find second smallest valid mem_section
594 * for shrinking zone.
596 pfn = find_smallest_section_pfn(nid, NULL, end_pfn,
599 pgdat->node_start_pfn = pfn;
600 pgdat->node_spanned_pages = pgdat_end_pfn - pfn;
602 } else if (pgdat_end_pfn == end_pfn) {
604 * If the section is biggest section in the pgdat, it need
605 * shrink pgdat->node_spanned_pages.
606 * In this case, we find second biggest valid mem_section for
609 pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn,
612 pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1;
616 * If the section is not biggest or smallest mem_section in the pgdat,
617 * it only creates a hole in the pgdat. So in this case, we need not
619 * But perhaps, the pgdat has only hole data. Thus it check the pgdat
620 * has only hole or not.
622 pfn = pgdat_start_pfn;
623 for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) {
624 ms = __pfn_to_section(pfn);
626 if (unlikely(!valid_section(ms)))
629 if (pfn_to_nid(pfn) != nid)
632 /* If the section is current section, it continues the loop */
633 if (start_pfn == pfn)
636 /* If we find valid section, we have nothing to do */
640 /* The pgdat has no valid section */
641 pgdat->node_start_pfn = 0;
642 pgdat->node_spanned_pages = 0;
645 static void __remove_zone(struct zone *zone, unsigned long start_pfn)
647 struct pglist_data *pgdat = zone->zone_pgdat;
648 int nr_pages = PAGES_PER_SECTION;
652 zone_type = zone - pgdat->node_zones;
654 pgdat_resize_lock(zone->zone_pgdat, &flags);
655 shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
656 shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages);
657 pgdat_resize_unlock(zone->zone_pgdat, &flags);
660 static int __remove_section(struct zone *zone, struct mem_section *ms)
662 unsigned long start_pfn;
666 if (!valid_section(ms))
669 ret = unregister_memory_section(ms);
673 scn_nr = __section_nr(ms);
674 start_pfn = section_nr_to_pfn(scn_nr);
675 __remove_zone(zone, start_pfn);
677 sparse_remove_one_section(zone, ms);
682 * __remove_pages() - remove sections of pages from a zone
683 * @zone: zone from which pages need to be removed
684 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
685 * @nr_pages: number of pages to remove (must be multiple of section size)
687 * Generic helper function to remove section mappings and sysfs entries
688 * for the section of the memory we are removing. Caller needs to make
689 * sure that pages are marked reserved and zones are adjust properly by
690 * calling offline_pages().
692 int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
693 unsigned long nr_pages)
696 int sections_to_remove;
697 resource_size_t start, size;
701 * We can only remove entire sections
703 BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
704 BUG_ON(nr_pages % PAGES_PER_SECTION);
706 start = phys_start_pfn << PAGE_SHIFT;
707 size = nr_pages * PAGE_SIZE;
708 ret = release_mem_region_adjustable(&iomem_resource, start, size);
710 resource_size_t endres = start + size - 1;
712 pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
713 &start, &endres, ret);
716 sections_to_remove = nr_pages / PAGES_PER_SECTION;
717 for (i = 0; i < sections_to_remove; i++) {
718 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
719 ret = __remove_section(zone, __pfn_to_section(pfn));
725 EXPORT_SYMBOL_GPL(__remove_pages);
726 #endif /* CONFIG_MEMORY_HOTREMOVE */
728 int set_online_page_callback(online_page_callback_t callback)
732 lock_memory_hotplug();
734 if (online_page_callback == generic_online_page) {
735 online_page_callback = callback;
739 unlock_memory_hotplug();
743 EXPORT_SYMBOL_GPL(set_online_page_callback);
745 int restore_online_page_callback(online_page_callback_t callback)
749 lock_memory_hotplug();
751 if (online_page_callback == callback) {
752 online_page_callback = generic_online_page;
756 unlock_memory_hotplug();
760 EXPORT_SYMBOL_GPL(restore_online_page_callback);
762 void __online_page_set_limits(struct page *page)
765 EXPORT_SYMBOL_GPL(__online_page_set_limits);
767 void __online_page_increment_counters(struct page *page)
769 adjust_managed_page_count(page, 1);
771 EXPORT_SYMBOL_GPL(__online_page_increment_counters);
773 void __online_page_free(struct page *page)
775 __free_reserved_page(page);
777 EXPORT_SYMBOL_GPL(__online_page_free);
779 static void generic_online_page(struct page *page)
781 __online_page_set_limits(page);
782 __online_page_increment_counters(page);
783 __online_page_free(page);
786 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
790 unsigned long onlined_pages = *(unsigned long *)arg;
792 if (PageReserved(pfn_to_page(start_pfn)))
793 for (i = 0; i < nr_pages; i++) {
794 page = pfn_to_page(start_pfn + i);
795 (*online_page_callback)(page);
798 *(unsigned long *)arg = onlined_pages;
802 #ifdef CONFIG_MOVABLE_NODE
804 * When CONFIG_MOVABLE_NODE, we permit onlining of a node which doesn't have
807 static bool can_online_high_movable(struct zone *zone)
811 #else /* CONFIG_MOVABLE_NODE */
812 /* ensure every online node has NORMAL memory */
813 static bool can_online_high_movable(struct zone *zone)
815 return node_state(zone_to_nid(zone), N_NORMAL_MEMORY);
817 #endif /* CONFIG_MOVABLE_NODE */
819 /* check which state of node_states will be changed when online memory */
820 static void node_states_check_changes_online(unsigned long nr_pages,
821 struct zone *zone, struct memory_notify *arg)
823 int nid = zone_to_nid(zone);
824 enum zone_type zone_last = ZONE_NORMAL;
827 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
828 * contains nodes which have zones of 0...ZONE_NORMAL,
829 * set zone_last to ZONE_NORMAL.
831 * If we don't have HIGHMEM nor movable node,
832 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
833 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
835 if (N_MEMORY == N_NORMAL_MEMORY)
836 zone_last = ZONE_MOVABLE;
839 * if the memory to be online is in a zone of 0...zone_last, and
840 * the zones of 0...zone_last don't have memory before online, we will
841 * need to set the node to node_states[N_NORMAL_MEMORY] after
842 * the memory is online.
844 if (zone_idx(zone) <= zone_last && !node_state(nid, N_NORMAL_MEMORY))
845 arg->status_change_nid_normal = nid;
847 arg->status_change_nid_normal = -1;
849 #ifdef CONFIG_HIGHMEM
851 * If we have movable node, node_states[N_HIGH_MEMORY]
852 * contains nodes which have zones of 0...ZONE_HIGHMEM,
853 * set zone_last to ZONE_HIGHMEM.
855 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
856 * contains nodes which have zones of 0...ZONE_MOVABLE,
857 * set zone_last to ZONE_MOVABLE.
859 zone_last = ZONE_HIGHMEM;
860 if (N_MEMORY == N_HIGH_MEMORY)
861 zone_last = ZONE_MOVABLE;
863 if (zone_idx(zone) <= zone_last && !node_state(nid, N_HIGH_MEMORY))
864 arg->status_change_nid_high = nid;
866 arg->status_change_nid_high = -1;
868 arg->status_change_nid_high = arg->status_change_nid_normal;
872 * if the node don't have memory befor online, we will need to
873 * set the node to node_states[N_MEMORY] after the memory
876 if (!node_state(nid, N_MEMORY))
877 arg->status_change_nid = nid;
879 arg->status_change_nid = -1;
882 static void node_states_set_node(int node, struct memory_notify *arg)
884 if (arg->status_change_nid_normal >= 0)
885 node_set_state(node, N_NORMAL_MEMORY);
887 if (arg->status_change_nid_high >= 0)
888 node_set_state(node, N_HIGH_MEMORY);
890 node_set_state(node, N_MEMORY);
894 int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type)
897 unsigned long onlined_pages = 0;
899 int need_zonelists_rebuild = 0;
902 struct memory_notify arg;
904 lock_memory_hotplug();
906 * This doesn't need a lock to do pfn_to_page().
907 * The section can't be removed here because of the
908 * memory_block->state_mutex.
910 zone = page_zone(pfn_to_page(pfn));
912 if ((zone_idx(zone) > ZONE_NORMAL || online_type == ONLINE_MOVABLE) &&
913 !can_online_high_movable(zone)) {
914 unlock_memory_hotplug();
918 if (online_type == ONLINE_KERNEL && zone_idx(zone) == ZONE_MOVABLE) {
919 if (move_pfn_range_left(zone - 1, zone, pfn, pfn + nr_pages)) {
920 unlock_memory_hotplug();
924 if (online_type == ONLINE_MOVABLE && zone_idx(zone) == ZONE_MOVABLE - 1) {
925 if (move_pfn_range_right(zone, zone + 1, pfn, pfn + nr_pages)) {
926 unlock_memory_hotplug();
931 /* Previous code may changed the zone of the pfn range */
932 zone = page_zone(pfn_to_page(pfn));
935 arg.nr_pages = nr_pages;
936 node_states_check_changes_online(nr_pages, zone, &arg);
938 nid = page_to_nid(pfn_to_page(pfn));
940 ret = memory_notify(MEM_GOING_ONLINE, &arg);
941 ret = notifier_to_errno(ret);
943 memory_notify(MEM_CANCEL_ONLINE, &arg);
944 unlock_memory_hotplug();
948 * If this zone is not populated, then it is not in zonelist.
949 * This means the page allocator ignores this zone.
950 * So, zonelist must be updated after online.
952 mutex_lock(&zonelists_mutex);
953 if (!populated_zone(zone)) {
954 need_zonelists_rebuild = 1;
955 build_all_zonelists(NULL, zone);
958 ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
961 if (need_zonelists_rebuild)
962 zone_pcp_reset(zone);
963 mutex_unlock(&zonelists_mutex);
964 printk(KERN_DEBUG "online_pages [mem %#010llx-%#010llx] failed\n",
965 (unsigned long long) pfn << PAGE_SHIFT,
966 (((unsigned long long) pfn + nr_pages)
968 memory_notify(MEM_CANCEL_ONLINE, &arg);
969 unlock_memory_hotplug();
973 zone->present_pages += onlined_pages;
975 pgdat_resize_lock(zone->zone_pgdat, &flags);
976 zone->zone_pgdat->node_present_pages += onlined_pages;
977 pgdat_resize_unlock(zone->zone_pgdat, &flags);
980 node_states_set_node(zone_to_nid(zone), &arg);
981 if (need_zonelists_rebuild)
982 build_all_zonelists(NULL, NULL);
984 zone_pcp_update(zone);
987 mutex_unlock(&zonelists_mutex);
989 init_per_zone_wmark_min();
992 kswapd_run(zone_to_nid(zone));
994 vm_total_pages = nr_free_pagecache_pages();
996 writeback_set_ratelimit();
999 memory_notify(MEM_ONLINE, &arg);
1000 unlock_memory_hotplug();
1004 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
1006 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1007 static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
1009 struct pglist_data *pgdat;
1010 unsigned long zones_size[MAX_NR_ZONES] = {0};
1011 unsigned long zholes_size[MAX_NR_ZONES] = {0};
1012 unsigned long start_pfn = start >> PAGE_SHIFT;
1014 pgdat = NODE_DATA(nid);
1016 pgdat = arch_alloc_nodedata(nid);
1020 arch_refresh_nodedata(nid, pgdat);
1023 /* we can use NODE_DATA(nid) from here */
1025 /* init node's zones as empty zones, we don't have any present pages.*/
1026 free_area_init_node(nid, zones_size, start_pfn, zholes_size);
1029 * The node we allocated has no zone fallback lists. For avoiding
1030 * to access not-initialized zonelist, build here.
1032 mutex_lock(&zonelists_mutex);
1033 build_all_zonelists(pgdat, NULL);
1034 mutex_unlock(&zonelists_mutex);
1039 static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
1041 arch_refresh_nodedata(nid, NULL);
1042 arch_free_nodedata(pgdat);
1048 * called by cpu_up() to online a node without onlined memory.
1050 int mem_online_node(int nid)
1055 lock_memory_hotplug();
1056 pgdat = hotadd_new_pgdat(nid, 0);
1061 node_set_online(nid);
1062 ret = register_one_node(nid);
1066 unlock_memory_hotplug();
1070 static int check_hotplug_memory_range(u64 start, u64 size)
1072 u64 start_pfn = start >> PAGE_SHIFT;
1073 u64 nr_pages = size >> PAGE_SHIFT;
1075 /* Memory range must be aligned with section */
1076 if ((start_pfn & ~PAGE_SECTION_MASK) ||
1077 (nr_pages % PAGES_PER_SECTION) || (!nr_pages)) {
1078 pr_err("Section-unaligned hotplug range: start 0x%llx, size 0x%llx\n",
1079 (unsigned long long)start,
1080 (unsigned long long)size);
1087 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1088 int __ref add_memory(int nid, u64 start, u64 size)
1090 pg_data_t *pgdat = NULL;
1093 struct resource *res;
1096 ret = check_hotplug_memory_range(start, size);
1100 lock_memory_hotplug();
1102 res = register_memory_resource(start, size);
1107 { /* Stupid hack to suppress address-never-null warning */
1108 void *p = NODE_DATA(nid);
1111 new_node = !node_online(nid);
1113 pgdat = hotadd_new_pgdat(nid, start);
1119 /* call arch's memory hotadd */
1120 ret = arch_add_memory(nid, start, size);
1125 /* we online node here. we can't roll back from here. */
1126 node_set_online(nid);
1129 ret = register_one_node(nid);
1131 * If sysfs file of new node can't create, cpu on the node
1132 * can't be hot-added. There is no rollback way now.
1133 * So, check by BUG_ON() to catch it reluctantly..
1138 /* create new memmap entry */
1139 firmware_map_add_hotplug(start, start + size, "System RAM");
1144 /* rollback pgdat allocation and others */
1146 rollback_node_hotadd(nid, pgdat);
1147 release_memory_resource(res);
1150 unlock_memory_hotplug();
1153 EXPORT_SYMBOL_GPL(add_memory);
1155 #ifdef CONFIG_MEMORY_HOTREMOVE
1157 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1158 * set and the size of the free page is given by page_order(). Using this,
1159 * the function determines if the pageblock contains only free pages.
1160 * Due to buddy contraints, a free page at least the size of a pageblock will
1161 * be located at the start of the pageblock
1163 static inline int pageblock_free(struct page *page)
1165 return PageBuddy(page) && page_order(page) >= pageblock_order;
1168 /* Return the start of the next active pageblock after a given page */
1169 static struct page *next_active_pageblock(struct page *page)
1171 /* Ensure the starting page is pageblock-aligned */
1172 BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
1174 /* If the entire pageblock is free, move to the end of free page */
1175 if (pageblock_free(page)) {
1177 /* be careful. we don't have locks, page_order can be changed.*/
1178 order = page_order(page);
1179 if ((order < MAX_ORDER) && (order >= pageblock_order))
1180 return page + (1 << order);
1183 return page + pageblock_nr_pages;
1186 /* Checks if this range of memory is likely to be hot-removable. */
1187 int is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
1189 struct page *page = pfn_to_page(start_pfn);
1190 struct page *end_page = page + nr_pages;
1192 /* Check the starting page of each pageblock within the range */
1193 for (; page < end_page; page = next_active_pageblock(page)) {
1194 if (!is_pageblock_removable_nolock(page))
1199 /* All pageblocks in the memory block are likely to be hot-removable */
1204 * Confirm all pages in a range [start, end) is belongs to the same zone.
1206 static int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn)
1209 struct zone *zone = NULL;
1212 for (pfn = start_pfn;
1214 pfn += MAX_ORDER_NR_PAGES) {
1216 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1217 while ((i < MAX_ORDER_NR_PAGES) && !pfn_valid_within(pfn + i))
1219 if (i == MAX_ORDER_NR_PAGES)
1221 page = pfn_to_page(pfn + i);
1222 if (zone && page_zone(page) != zone)
1224 zone = page_zone(page);
1230 * Scan pfn range [start,end) to find movable/migratable pages (LRU pages
1231 * and hugepages). We scan pfn because it's much easier than scanning over
1232 * linked list. This function returns the pfn of the first found movable
1233 * page if it's found, otherwise 0.
1235 static unsigned long scan_movable_pages(unsigned long start, unsigned long end)
1239 for (pfn = start; pfn < end; pfn++) {
1240 if (pfn_valid(pfn)) {
1241 page = pfn_to_page(pfn);
1244 if (PageHuge(page)) {
1245 if (is_hugepage_active(page))
1248 pfn = round_up(pfn + 1,
1249 1 << compound_order(page)) - 1;
1256 #define NR_OFFLINE_AT_ONCE_PAGES (256)
1258 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1262 int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
1263 int not_managed = 0;
1267 for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
1268 if (!pfn_valid(pfn))
1270 page = pfn_to_page(pfn);
1272 if (PageHuge(page)) {
1273 struct page *head = compound_head(page);
1274 pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1;
1275 if (compound_order(head) > PFN_SECTION_SHIFT) {
1279 if (isolate_huge_page(page, &source))
1280 move_pages -= 1 << compound_order(head);
1284 if (!get_page_unless_zero(page))
1287 * We can skip free pages. And we can only deal with pages on
1290 ret = isolate_lru_page(page);
1291 if (!ret) { /* Success */
1293 list_add_tail(&page->lru, &source);
1295 inc_zone_page_state(page, NR_ISOLATED_ANON +
1296 page_is_file_cache(page));
1299 #ifdef CONFIG_DEBUG_VM
1300 printk(KERN_ALERT "removing pfn %lx from LRU failed\n",
1305 /* Because we don't have big zone->lock. we should
1306 check this again here. */
1307 if (page_count(page)) {
1314 if (!list_empty(&source)) {
1316 putback_movable_pages(&source);
1321 * alloc_migrate_target should be improooooved!!
1322 * migrate_pages returns # of failed pages.
1324 ret = migrate_pages(&source, alloc_migrate_target, 0,
1325 MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
1327 putback_movable_pages(&source);
1334 * remove from free_area[] and mark all as Reserved.
1337 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1340 __offline_isolated_pages(start, start + nr_pages);
1345 offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
1347 walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
1348 offline_isolated_pages_cb);
1352 * Check all pages in range, recoreded as memory resource, are isolated.
1355 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1359 long offlined = *(long *)data;
1360 ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true);
1361 offlined = nr_pages;
1363 *(long *)data += offlined;
1368 check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
1373 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
1374 check_pages_isolated_cb);
1376 offlined = (long)ret;
1380 #ifdef CONFIG_MOVABLE_NODE
1382 * When CONFIG_MOVABLE_NODE, we permit offlining of a node which doesn't have
1385 static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1389 #else /* CONFIG_MOVABLE_NODE */
1390 /* ensure the node has NORMAL memory if it is still online */
1391 static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1393 struct pglist_data *pgdat = zone->zone_pgdat;
1394 unsigned long present_pages = 0;
1397 for (zt = 0; zt <= ZONE_NORMAL; zt++)
1398 present_pages += pgdat->node_zones[zt].present_pages;
1400 if (present_pages > nr_pages)
1404 for (; zt <= ZONE_MOVABLE; zt++)
1405 present_pages += pgdat->node_zones[zt].present_pages;
1408 * we can't offline the last normal memory until all
1409 * higher memory is offlined.
1411 return present_pages == 0;
1413 #endif /* CONFIG_MOVABLE_NODE */
1415 /* check which state of node_states will be changed when offline memory */
1416 static void node_states_check_changes_offline(unsigned long nr_pages,
1417 struct zone *zone, struct memory_notify *arg)
1419 struct pglist_data *pgdat = zone->zone_pgdat;
1420 unsigned long present_pages = 0;
1421 enum zone_type zt, zone_last = ZONE_NORMAL;
1424 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
1425 * contains nodes which have zones of 0...ZONE_NORMAL,
1426 * set zone_last to ZONE_NORMAL.
1428 * If we don't have HIGHMEM nor movable node,
1429 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
1430 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
1432 if (N_MEMORY == N_NORMAL_MEMORY)
1433 zone_last = ZONE_MOVABLE;
1436 * check whether node_states[N_NORMAL_MEMORY] will be changed.
1437 * If the memory to be offline is in a zone of 0...zone_last,
1438 * and it is the last present memory, 0...zone_last will
1439 * become empty after offline , thus we can determind we will
1440 * need to clear the node from node_states[N_NORMAL_MEMORY].
1442 for (zt = 0; zt <= zone_last; zt++)
1443 present_pages += pgdat->node_zones[zt].present_pages;
1444 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1445 arg->status_change_nid_normal = zone_to_nid(zone);
1447 arg->status_change_nid_normal = -1;
1449 #ifdef CONFIG_HIGHMEM
1451 * If we have movable node, node_states[N_HIGH_MEMORY]
1452 * contains nodes which have zones of 0...ZONE_HIGHMEM,
1453 * set zone_last to ZONE_HIGHMEM.
1455 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
1456 * contains nodes which have zones of 0...ZONE_MOVABLE,
1457 * set zone_last to ZONE_MOVABLE.
1459 zone_last = ZONE_HIGHMEM;
1460 if (N_MEMORY == N_HIGH_MEMORY)
1461 zone_last = ZONE_MOVABLE;
1463 for (; zt <= zone_last; zt++)
1464 present_pages += pgdat->node_zones[zt].present_pages;
1465 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1466 arg->status_change_nid_high = zone_to_nid(zone);
1468 arg->status_change_nid_high = -1;
1470 arg->status_change_nid_high = arg->status_change_nid_normal;
1474 * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE
1476 zone_last = ZONE_MOVABLE;
1479 * check whether node_states[N_HIGH_MEMORY] will be changed
1480 * If we try to offline the last present @nr_pages from the node,
1481 * we can determind we will need to clear the node from
1482 * node_states[N_HIGH_MEMORY].
1484 for (; zt <= zone_last; zt++)
1485 present_pages += pgdat->node_zones[zt].present_pages;
1486 if (nr_pages >= present_pages)
1487 arg->status_change_nid = zone_to_nid(zone);
1489 arg->status_change_nid = -1;
1492 static void node_states_clear_node(int node, struct memory_notify *arg)
1494 if (arg->status_change_nid_normal >= 0)
1495 node_clear_state(node, N_NORMAL_MEMORY);
1497 if ((N_MEMORY != N_NORMAL_MEMORY) &&
1498 (arg->status_change_nid_high >= 0))
1499 node_clear_state(node, N_HIGH_MEMORY);
1501 if ((N_MEMORY != N_HIGH_MEMORY) &&
1502 (arg->status_change_nid >= 0))
1503 node_clear_state(node, N_MEMORY);
1506 static int __ref __offline_pages(unsigned long start_pfn,
1507 unsigned long end_pfn, unsigned long timeout)
1509 unsigned long pfn, nr_pages, expire;
1510 long offlined_pages;
1511 int ret, drain, retry_max, node;
1512 unsigned long flags;
1514 struct memory_notify arg;
1516 /* at least, alignment against pageblock is necessary */
1517 if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
1519 if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
1521 /* This makes hotplug much easier...and readable.
1522 we assume this for now. .*/
1523 if (!test_pages_in_a_zone(start_pfn, end_pfn))
1526 lock_memory_hotplug();
1528 zone = page_zone(pfn_to_page(start_pfn));
1529 node = zone_to_nid(zone);
1530 nr_pages = end_pfn - start_pfn;
1533 if (zone_idx(zone) <= ZONE_NORMAL && !can_offline_normal(zone, nr_pages))
1536 /* set above range as isolated */
1537 ret = start_isolate_page_range(start_pfn, end_pfn,
1538 MIGRATE_MOVABLE, true);
1542 arg.start_pfn = start_pfn;
1543 arg.nr_pages = nr_pages;
1544 node_states_check_changes_offline(nr_pages, zone, &arg);
1546 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1547 ret = notifier_to_errno(ret);
1549 goto failed_removal;
1552 expire = jiffies + timeout;
1556 /* start memory hot removal */
1558 if (time_after(jiffies, expire))
1559 goto failed_removal;
1561 if (signal_pending(current))
1562 goto failed_removal;
1565 lru_add_drain_all();
1570 pfn = scan_movable_pages(start_pfn, end_pfn);
1571 if (pfn) { /* We have movable pages */
1572 ret = do_migrate_range(pfn, end_pfn);
1578 if (--retry_max == 0)
1579 goto failed_removal;
1585 /* drain all zone's lru pagevec, this is asynchronous... */
1586 lru_add_drain_all();
1588 /* drain pcp pages, this is synchronous. */
1591 * dissolve free hugepages in the memory block before doing offlining
1592 * actually in order to make hugetlbfs's object counting consistent.
1594 dissolve_free_huge_pages(start_pfn, end_pfn);
1596 offlined_pages = check_pages_isolated(start_pfn, end_pfn);
1597 if (offlined_pages < 0) {
1599 goto failed_removal;
1601 printk(KERN_INFO "Offlined Pages %ld\n", offlined_pages);
1602 /* Ok, all of our target is isolated.
1603 We cannot do rollback at this point. */
1604 offline_isolated_pages(start_pfn, end_pfn);
1605 /* reset pagetype flags and makes migrate type to be MOVABLE */
1606 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1607 /* removal success */
1608 adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages);
1609 zone->present_pages -= offlined_pages;
1611 pgdat_resize_lock(zone->zone_pgdat, &flags);
1612 zone->zone_pgdat->node_present_pages -= offlined_pages;
1613 pgdat_resize_unlock(zone->zone_pgdat, &flags);
1615 init_per_zone_wmark_min();
1617 if (!populated_zone(zone)) {
1618 zone_pcp_reset(zone);
1619 mutex_lock(&zonelists_mutex);
1620 build_all_zonelists(NULL, NULL);
1621 mutex_unlock(&zonelists_mutex);
1623 zone_pcp_update(zone);
1625 node_states_clear_node(node, &arg);
1626 if (arg.status_change_nid >= 0)
1629 vm_total_pages = nr_free_pagecache_pages();
1630 writeback_set_ratelimit();
1632 memory_notify(MEM_OFFLINE, &arg);
1633 unlock_memory_hotplug();
1637 printk(KERN_INFO "memory offlining [mem %#010llx-%#010llx] failed\n",
1638 (unsigned long long) start_pfn << PAGE_SHIFT,
1639 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
1640 memory_notify(MEM_CANCEL_OFFLINE, &arg);
1641 /* pushback to free area */
1642 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1645 unlock_memory_hotplug();
1649 int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1651 return __offline_pages(start_pfn, start_pfn + nr_pages, 120 * HZ);
1653 #endif /* CONFIG_MEMORY_HOTREMOVE */
1656 * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
1657 * @start_pfn: start pfn of the memory range
1658 * @end_pfn: end pfn of the memory range
1659 * @arg: argument passed to func
1660 * @func: callback for each memory section walked
1662 * This function walks through all present mem sections in range
1663 * [start_pfn, end_pfn) and call func on each mem section.
1665 * Returns the return value of func.
1667 int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn,
1668 void *arg, int (*func)(struct memory_block *, void *))
1670 struct memory_block *mem = NULL;
1671 struct mem_section *section;
1672 unsigned long pfn, section_nr;
1675 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1676 section_nr = pfn_to_section_nr(pfn);
1677 if (!present_section_nr(section_nr))
1680 section = __nr_to_section(section_nr);
1681 /* same memblock? */
1683 if ((section_nr >= mem->start_section_nr) &&
1684 (section_nr <= mem->end_section_nr))
1687 mem = find_memory_block_hinted(section, mem);
1691 ret = func(mem, arg);
1693 kobject_put(&mem->dev.kobj);
1699 kobject_put(&mem->dev.kobj);
1704 #ifdef CONFIG_MEMORY_HOTREMOVE
1705 static int is_memblock_offlined_cb(struct memory_block *mem, void *arg)
1707 int ret = !is_memblock_offlined(mem);
1709 if (unlikely(ret)) {
1710 phys_addr_t beginpa, endpa;
1712 beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
1713 endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1;
1714 pr_warn("removing memory fails, because memory "
1715 "[%pa-%pa] is onlined\n",
1722 static int check_cpu_on_node(pg_data_t *pgdat)
1726 for_each_present_cpu(cpu) {
1727 if (cpu_to_node(cpu) == pgdat->node_id)
1729 * the cpu on this node isn't removed, and we can't
1730 * offline this node.
1738 static void unmap_cpu_on_node(pg_data_t *pgdat)
1740 #ifdef CONFIG_ACPI_NUMA
1743 for_each_possible_cpu(cpu)
1744 if (cpu_to_node(cpu) == pgdat->node_id)
1745 numa_clear_node(cpu);
1749 static int check_and_unmap_cpu_on_node(pg_data_t *pgdat)
1753 ret = check_cpu_on_node(pgdat);
1758 * the node will be offlined when we come here, so we can clear
1759 * the cpu_to_node() now.
1762 unmap_cpu_on_node(pgdat);
1769 * Offline a node if all memory sections and cpus of the node are removed.
1771 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1772 * and online/offline operations before this call.
1774 void try_offline_node(int nid)
1776 pg_data_t *pgdat = NODE_DATA(nid);
1777 unsigned long start_pfn = pgdat->node_start_pfn;
1778 unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
1780 struct page *pgdat_page = virt_to_page(pgdat);
1783 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1784 unsigned long section_nr = pfn_to_section_nr(pfn);
1786 if (!present_section_nr(section_nr))
1789 if (pfn_to_nid(pfn) != nid)
1793 * some memory sections of this node are not removed, and we
1794 * can't offline node now.
1799 if (check_and_unmap_cpu_on_node(pgdat))
1803 * all memory/cpu of this node are removed, we can offline this
1806 node_set_offline(nid);
1807 unregister_one_node(nid);
1809 if (!PageSlab(pgdat_page) && !PageCompound(pgdat_page))
1810 /* node data is allocated from boot memory */
1813 /* free waittable in each zone */
1814 for (i = 0; i < MAX_NR_ZONES; i++) {
1815 struct zone *zone = pgdat->node_zones + i;
1818 * wait_table may be allocated from boot memory,
1819 * here only free if it's allocated by vmalloc.
1821 if (is_vmalloc_addr(zone->wait_table))
1822 vfree(zone->wait_table);
1826 * Since there is no way to guarentee the address of pgdat/zone is not
1827 * on stack of any kernel threads or used by other kernel objects
1828 * without reference counting or other symchronizing method, do not
1829 * reset node_data and free pgdat here. Just reset it to 0 and reuse
1830 * the memory when the node is online again.
1832 memset(pgdat, 0, sizeof(*pgdat));
1834 EXPORT_SYMBOL(try_offline_node);
1839 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1840 * and online/offline operations before this call, as required by
1841 * try_offline_node().
1843 void __ref remove_memory(int nid, u64 start, u64 size)
1847 BUG_ON(check_hotplug_memory_range(start, size));
1849 lock_memory_hotplug();
1852 * All memory blocks must be offlined before removing memory. Check
1853 * whether all memory blocks in question are offline and trigger a BUG()
1854 * if this is not the case.
1856 ret = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL,
1857 is_memblock_offlined_cb);
1859 unlock_memory_hotplug();
1863 /* remove memmap entry */
1864 firmware_map_remove(start, start + size, "System RAM");
1866 arch_remove_memory(start, size);
1868 try_offline_node(nid);
1870 unlock_memory_hotplug();
1872 EXPORT_SYMBOL_GPL(remove_memory);
1873 #endif /* CONFIG_MEMORY_HOTREMOVE */