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 pr_debug("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 void put_page_bootmem(struct page *page)
108 type = (unsigned long) page->lru.next;
109 BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE ||
110 type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE);
112 if (atomic_dec_return(&page->_count) == 1) {
113 ClearPagePrivate(page);
114 set_page_private(page, 0);
115 INIT_LIST_HEAD(&page->lru);
116 free_reserved_page(page);
120 #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
121 #ifndef CONFIG_SPARSEMEM_VMEMMAP
122 static void register_page_bootmem_info_section(unsigned long start_pfn)
124 unsigned long *usemap, mapsize, section_nr, i;
125 struct mem_section *ms;
126 struct page *page, *memmap;
128 section_nr = pfn_to_section_nr(start_pfn);
129 ms = __nr_to_section(section_nr);
131 /* Get section's memmap address */
132 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
135 * Get page for the memmap's phys address
136 * XXX: need more consideration for sparse_vmemmap...
138 page = virt_to_page(memmap);
139 mapsize = sizeof(struct page) * PAGES_PER_SECTION;
140 mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
142 /* remember memmap's page */
143 for (i = 0; i < mapsize; i++, page++)
144 get_page_bootmem(section_nr, page, SECTION_INFO);
146 usemap = __nr_to_section(section_nr)->pageblock_flags;
147 page = virt_to_page(usemap);
149 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
151 for (i = 0; i < mapsize; i++, page++)
152 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
155 #else /* CONFIG_SPARSEMEM_VMEMMAP */
156 static void register_page_bootmem_info_section(unsigned long start_pfn)
158 unsigned long *usemap, mapsize, section_nr, i;
159 struct mem_section *ms;
160 struct page *page, *memmap;
162 if (!pfn_valid(start_pfn))
165 section_nr = pfn_to_section_nr(start_pfn);
166 ms = __nr_to_section(section_nr);
168 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
170 register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION);
172 usemap = __nr_to_section(section_nr)->pageblock_flags;
173 page = virt_to_page(usemap);
175 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
177 for (i = 0; i < mapsize; i++, page++)
178 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
180 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
182 void register_page_bootmem_info_node(struct pglist_data *pgdat)
184 unsigned long i, pfn, end_pfn, nr_pages;
185 int node = pgdat->node_id;
189 nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
190 page = virt_to_page(pgdat);
192 for (i = 0; i < nr_pages; i++, page++)
193 get_page_bootmem(node, page, NODE_INFO);
195 zone = &pgdat->node_zones[0];
196 for (; zone < pgdat->node_zones + MAX_NR_ZONES - 1; zone++) {
197 if (zone->wait_table) {
198 nr_pages = zone->wait_table_hash_nr_entries
199 * sizeof(wait_queue_head_t);
200 nr_pages = PAGE_ALIGN(nr_pages) >> PAGE_SHIFT;
201 page = virt_to_page(zone->wait_table);
203 for (i = 0; i < nr_pages; i++, page++)
204 get_page_bootmem(node, page, NODE_INFO);
208 pfn = pgdat->node_start_pfn;
209 end_pfn = pgdat_end_pfn(pgdat);
211 /* register_section info */
212 for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
214 * Some platforms can assign the same pfn to multiple nodes - on
215 * node0 as well as nodeN. To avoid registering a pfn against
216 * multiple nodes we check that this pfn does not already
217 * reside in some other node.
219 if (pfn_valid(pfn) && (pfn_to_nid(pfn) == node))
220 register_page_bootmem_info_section(pfn);
223 #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
225 static void grow_zone_span(struct zone *zone, unsigned long start_pfn,
226 unsigned long end_pfn)
228 unsigned long old_zone_end_pfn;
230 zone_span_writelock(zone);
232 old_zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages;
233 if (!zone->spanned_pages || start_pfn < zone->zone_start_pfn)
234 zone->zone_start_pfn = start_pfn;
236 zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
237 zone->zone_start_pfn;
239 zone_span_writeunlock(zone);
242 static void resize_zone(struct zone *zone, unsigned long start_pfn,
243 unsigned long end_pfn)
245 zone_span_writelock(zone);
247 if (end_pfn - start_pfn) {
248 zone->zone_start_pfn = start_pfn;
249 zone->spanned_pages = end_pfn - start_pfn;
252 * make it consist as free_area_init_core(),
253 * if spanned_pages = 0, then keep start_pfn = 0
255 zone->zone_start_pfn = 0;
256 zone->spanned_pages = 0;
259 zone_span_writeunlock(zone);
262 static void fix_zone_id(struct zone *zone, unsigned long start_pfn,
263 unsigned long end_pfn)
265 enum zone_type zid = zone_idx(zone);
266 int nid = zone->zone_pgdat->node_id;
269 for (pfn = start_pfn; pfn < end_pfn; pfn++)
270 set_page_links(pfn_to_page(pfn), zid, nid, pfn);
273 /* Can fail with -ENOMEM from allocating a wait table with vmalloc() or
274 * alloc_bootmem_node_nopanic() */
275 static int __ref ensure_zone_is_initialized(struct zone *zone,
276 unsigned long start_pfn, unsigned long num_pages)
278 if (!zone_is_initialized(zone))
279 return init_currently_empty_zone(zone, start_pfn, num_pages,
284 static int __meminit move_pfn_range_left(struct zone *z1, struct zone *z2,
285 unsigned long start_pfn, unsigned long end_pfn)
289 unsigned long z1_start_pfn;
291 ret = ensure_zone_is_initialized(z1, start_pfn, end_pfn - start_pfn);
295 pgdat_resize_lock(z1->zone_pgdat, &flags);
297 /* can't move pfns which are higher than @z2 */
298 if (end_pfn > zone_end_pfn(z2))
300 /* the move out part must be at the left most of @z2 */
301 if (start_pfn > z2->zone_start_pfn)
303 /* must included/overlap */
304 if (end_pfn <= z2->zone_start_pfn)
307 /* use start_pfn for z1's start_pfn if z1 is empty */
308 if (z1->spanned_pages)
309 z1_start_pfn = z1->zone_start_pfn;
311 z1_start_pfn = start_pfn;
313 resize_zone(z1, z1_start_pfn, end_pfn);
314 resize_zone(z2, end_pfn, zone_end_pfn(z2));
316 pgdat_resize_unlock(z1->zone_pgdat, &flags);
318 fix_zone_id(z1, start_pfn, end_pfn);
322 pgdat_resize_unlock(z1->zone_pgdat, &flags);
326 static int __meminit move_pfn_range_right(struct zone *z1, struct zone *z2,
327 unsigned long start_pfn, unsigned long end_pfn)
331 unsigned long z2_end_pfn;
333 ret = ensure_zone_is_initialized(z2, start_pfn, end_pfn - start_pfn);
337 pgdat_resize_lock(z1->zone_pgdat, &flags);
339 /* can't move pfns which are lower than @z1 */
340 if (z1->zone_start_pfn > start_pfn)
342 /* the move out part mast at the right most of @z1 */
343 if (zone_end_pfn(z1) > end_pfn)
345 /* must included/overlap */
346 if (start_pfn >= zone_end_pfn(z1))
349 /* use end_pfn for z2's end_pfn if z2 is empty */
350 if (z2->spanned_pages)
351 z2_end_pfn = zone_end_pfn(z2);
353 z2_end_pfn = end_pfn;
355 resize_zone(z1, z1->zone_start_pfn, start_pfn);
356 resize_zone(z2, start_pfn, z2_end_pfn);
358 pgdat_resize_unlock(z1->zone_pgdat, &flags);
360 fix_zone_id(z2, start_pfn, end_pfn);
364 pgdat_resize_unlock(z1->zone_pgdat, &flags);
368 static void grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn,
369 unsigned long end_pfn)
371 unsigned long old_pgdat_end_pfn =
372 pgdat->node_start_pfn + pgdat->node_spanned_pages;
374 if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
375 pgdat->node_start_pfn = start_pfn;
377 pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
378 pgdat->node_start_pfn;
381 static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn)
383 struct pglist_data *pgdat = zone->zone_pgdat;
384 int nr_pages = PAGES_PER_SECTION;
385 int nid = pgdat->node_id;
390 zone_type = zone - pgdat->node_zones;
391 ret = ensure_zone_is_initialized(zone, phys_start_pfn, nr_pages);
395 pgdat_resize_lock(zone->zone_pgdat, &flags);
396 grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages);
397 grow_pgdat_span(zone->zone_pgdat, phys_start_pfn,
398 phys_start_pfn + nr_pages);
399 pgdat_resize_unlock(zone->zone_pgdat, &flags);
400 memmap_init_zone(nr_pages, nid, zone_type,
401 phys_start_pfn, MEMMAP_HOTPLUG);
405 static int __meminit __add_section(int nid, struct zone *zone,
406 unsigned long phys_start_pfn)
408 int nr_pages = PAGES_PER_SECTION;
411 if (pfn_valid(phys_start_pfn))
414 ret = sparse_add_one_section(zone, phys_start_pfn, nr_pages);
419 ret = __add_zone(zone, phys_start_pfn);
424 return register_new_memory(nid, __pfn_to_section(phys_start_pfn));
428 * Reasonably generic function for adding memory. It is
429 * expected that archs that support memory hotplug will
430 * call this function after deciding the zone to which to
433 int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn,
434 unsigned long nr_pages)
438 int start_sec, end_sec;
439 /* during initialize mem_map, align hot-added range to section */
440 start_sec = pfn_to_section_nr(phys_start_pfn);
441 end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
443 for (i = start_sec; i <= end_sec; i++) {
444 err = __add_section(nid, zone, i << PFN_SECTION_SHIFT);
447 * EEXIST is finally dealt with by ioresource collision
448 * check. see add_memory() => register_memory_resource()
449 * Warning will be printed if there is collision.
451 if (err && (err != -EEXIST))
458 EXPORT_SYMBOL_GPL(__add_pages);
460 #ifdef CONFIG_MEMORY_HOTREMOVE
461 /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
462 static int find_smallest_section_pfn(int nid, struct zone *zone,
463 unsigned long start_pfn,
464 unsigned long end_pfn)
466 struct mem_section *ms;
468 for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) {
469 ms = __pfn_to_section(start_pfn);
471 if (unlikely(!valid_section(ms)))
474 if (unlikely(pfn_to_nid(start_pfn) != nid))
477 if (zone && zone != page_zone(pfn_to_page(start_pfn)))
486 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
487 static int find_biggest_section_pfn(int nid, struct zone *zone,
488 unsigned long start_pfn,
489 unsigned long end_pfn)
491 struct mem_section *ms;
494 /* pfn is the end pfn of a memory section. */
496 for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) {
497 ms = __pfn_to_section(pfn);
499 if (unlikely(!valid_section(ms)))
502 if (unlikely(pfn_to_nid(pfn) != nid))
505 if (zone && zone != page_zone(pfn_to_page(pfn)))
514 static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
515 unsigned long end_pfn)
517 unsigned long zone_start_pfn = zone->zone_start_pfn;
518 unsigned long zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages;
520 struct mem_section *ms;
521 int nid = zone_to_nid(zone);
523 zone_span_writelock(zone);
524 if (zone_start_pfn == start_pfn) {
526 * If the section is smallest section in the zone, it need
527 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
528 * In this case, we find second smallest valid mem_section
529 * for shrinking zone.
531 pfn = find_smallest_section_pfn(nid, zone, end_pfn,
534 zone->zone_start_pfn = pfn;
535 zone->spanned_pages = zone_end_pfn - pfn;
537 } else if (zone_end_pfn == end_pfn) {
539 * If the section is biggest section in the zone, it need
540 * shrink zone->spanned_pages.
541 * In this case, we find second biggest valid mem_section for
544 pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn,
547 zone->spanned_pages = pfn - zone_start_pfn + 1;
551 * The section is not biggest or smallest mem_section in the zone, it
552 * only creates a hole in the zone. So in this case, we need not
553 * change the zone. But perhaps, the zone has only hole data. Thus
554 * it check the zone has only hole or not.
556 pfn = zone_start_pfn;
557 for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) {
558 ms = __pfn_to_section(pfn);
560 if (unlikely(!valid_section(ms)))
563 if (page_zone(pfn_to_page(pfn)) != zone)
566 /* If the section is current section, it continues the loop */
567 if (start_pfn == pfn)
570 /* If we find valid section, we have nothing to do */
571 zone_span_writeunlock(zone);
575 /* The zone has no valid section */
576 zone->zone_start_pfn = 0;
577 zone->spanned_pages = 0;
578 zone_span_writeunlock(zone);
581 static void shrink_pgdat_span(struct pglist_data *pgdat,
582 unsigned long start_pfn, unsigned long end_pfn)
584 unsigned long pgdat_start_pfn = pgdat->node_start_pfn;
585 unsigned long pgdat_end_pfn =
586 pgdat->node_start_pfn + pgdat->node_spanned_pages;
588 struct mem_section *ms;
589 int nid = pgdat->node_id;
591 if (pgdat_start_pfn == start_pfn) {
593 * If the section is smallest section in the pgdat, it need
594 * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages.
595 * In this case, we find second smallest valid mem_section
596 * for shrinking zone.
598 pfn = find_smallest_section_pfn(nid, NULL, end_pfn,
601 pgdat->node_start_pfn = pfn;
602 pgdat->node_spanned_pages = pgdat_end_pfn - pfn;
604 } else if (pgdat_end_pfn == end_pfn) {
606 * If the section is biggest section in the pgdat, it need
607 * shrink pgdat->node_spanned_pages.
608 * In this case, we find second biggest valid mem_section for
611 pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn,
614 pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1;
618 * If the section is not biggest or smallest mem_section in the pgdat,
619 * it only creates a hole in the pgdat. So in this case, we need not
621 * But perhaps, the pgdat has only hole data. Thus it check the pgdat
622 * has only hole or not.
624 pfn = pgdat_start_pfn;
625 for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) {
626 ms = __pfn_to_section(pfn);
628 if (unlikely(!valid_section(ms)))
631 if (pfn_to_nid(pfn) != nid)
634 /* If the section is current section, it continues the loop */
635 if (start_pfn == pfn)
638 /* If we find valid section, we have nothing to do */
642 /* The pgdat has no valid section */
643 pgdat->node_start_pfn = 0;
644 pgdat->node_spanned_pages = 0;
647 static void __remove_zone(struct zone *zone, unsigned long start_pfn)
649 struct pglist_data *pgdat = zone->zone_pgdat;
650 int nr_pages = PAGES_PER_SECTION;
654 zone_type = zone - pgdat->node_zones;
656 pgdat_resize_lock(zone->zone_pgdat, &flags);
657 shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
658 shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages);
659 pgdat_resize_unlock(zone->zone_pgdat, &flags);
662 static int __remove_section(struct zone *zone, struct mem_section *ms)
664 unsigned long start_pfn;
668 if (!valid_section(ms))
671 ret = unregister_memory_section(ms);
675 scn_nr = __section_nr(ms);
676 start_pfn = section_nr_to_pfn(scn_nr);
677 __remove_zone(zone, start_pfn);
679 sparse_remove_one_section(zone, ms);
684 * __remove_pages() - remove sections of pages from a zone
685 * @zone: zone from which pages need to be removed
686 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
687 * @nr_pages: number of pages to remove (must be multiple of section size)
689 * Generic helper function to remove section mappings and sysfs entries
690 * for the section of the memory we are removing. Caller needs to make
691 * sure that pages are marked reserved and zones are adjust properly by
692 * calling offline_pages().
694 int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
695 unsigned long nr_pages)
698 int sections_to_remove;
699 resource_size_t start, size;
703 * We can only remove entire sections
705 BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
706 BUG_ON(nr_pages % PAGES_PER_SECTION);
708 start = phys_start_pfn << PAGE_SHIFT;
709 size = nr_pages * PAGE_SIZE;
710 ret = release_mem_region_adjustable(&iomem_resource, start, size);
712 resource_size_t endres = start + size - 1;
714 pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
715 &start, &endres, ret);
718 sections_to_remove = nr_pages / PAGES_PER_SECTION;
719 for (i = 0; i < sections_to_remove; i++) {
720 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
721 ret = __remove_section(zone, __pfn_to_section(pfn));
727 EXPORT_SYMBOL_GPL(__remove_pages);
728 #endif /* CONFIG_MEMORY_HOTREMOVE */
730 int set_online_page_callback(online_page_callback_t callback)
734 lock_memory_hotplug();
736 if (online_page_callback == generic_online_page) {
737 online_page_callback = callback;
741 unlock_memory_hotplug();
745 EXPORT_SYMBOL_GPL(set_online_page_callback);
747 int restore_online_page_callback(online_page_callback_t callback)
751 lock_memory_hotplug();
753 if (online_page_callback == callback) {
754 online_page_callback = generic_online_page;
758 unlock_memory_hotplug();
762 EXPORT_SYMBOL_GPL(restore_online_page_callback);
764 void __online_page_set_limits(struct page *page)
767 EXPORT_SYMBOL_GPL(__online_page_set_limits);
769 void __online_page_increment_counters(struct page *page)
771 adjust_managed_page_count(page, 1);
773 EXPORT_SYMBOL_GPL(__online_page_increment_counters);
775 void __online_page_free(struct page *page)
777 __free_reserved_page(page);
779 EXPORT_SYMBOL_GPL(__online_page_free);
781 static void generic_online_page(struct page *page)
783 __online_page_set_limits(page);
784 __online_page_increment_counters(page);
785 __online_page_free(page);
788 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
792 unsigned long onlined_pages = *(unsigned long *)arg;
794 if (PageReserved(pfn_to_page(start_pfn)))
795 for (i = 0; i < nr_pages; i++) {
796 page = pfn_to_page(start_pfn + i);
797 (*online_page_callback)(page);
800 *(unsigned long *)arg = onlined_pages;
804 #ifdef CONFIG_MOVABLE_NODE
806 * When CONFIG_MOVABLE_NODE, we permit onlining of a node which doesn't have
809 static bool can_online_high_movable(struct zone *zone)
813 #else /* CONFIG_MOVABLE_NODE */
814 /* ensure every online node has NORMAL memory */
815 static bool can_online_high_movable(struct zone *zone)
817 return node_state(zone_to_nid(zone), N_NORMAL_MEMORY);
819 #endif /* CONFIG_MOVABLE_NODE */
821 /* check which state of node_states will be changed when online memory */
822 static void node_states_check_changes_online(unsigned long nr_pages,
823 struct zone *zone, struct memory_notify *arg)
825 int nid = zone_to_nid(zone);
826 enum zone_type zone_last = ZONE_NORMAL;
829 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
830 * contains nodes which have zones of 0...ZONE_NORMAL,
831 * set zone_last to ZONE_NORMAL.
833 * If we don't have HIGHMEM nor movable node,
834 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
835 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
837 if (N_MEMORY == N_NORMAL_MEMORY)
838 zone_last = ZONE_MOVABLE;
841 * if the memory to be online is in a zone of 0...zone_last, and
842 * the zones of 0...zone_last don't have memory before online, we will
843 * need to set the node to node_states[N_NORMAL_MEMORY] after
844 * the memory is online.
846 if (zone_idx(zone) <= zone_last && !node_state(nid, N_NORMAL_MEMORY))
847 arg->status_change_nid_normal = nid;
849 arg->status_change_nid_normal = -1;
851 #ifdef CONFIG_HIGHMEM
853 * If we have movable node, node_states[N_HIGH_MEMORY]
854 * contains nodes which have zones of 0...ZONE_HIGHMEM,
855 * set zone_last to ZONE_HIGHMEM.
857 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
858 * contains nodes which have zones of 0...ZONE_MOVABLE,
859 * set zone_last to ZONE_MOVABLE.
861 zone_last = ZONE_HIGHMEM;
862 if (N_MEMORY == N_HIGH_MEMORY)
863 zone_last = ZONE_MOVABLE;
865 if (zone_idx(zone) <= zone_last && !node_state(nid, N_HIGH_MEMORY))
866 arg->status_change_nid_high = nid;
868 arg->status_change_nid_high = -1;
870 arg->status_change_nid_high = arg->status_change_nid_normal;
874 * if the node don't have memory befor online, we will need to
875 * set the node to node_states[N_MEMORY] after the memory
878 if (!node_state(nid, N_MEMORY))
879 arg->status_change_nid = nid;
881 arg->status_change_nid = -1;
884 static void node_states_set_node(int node, struct memory_notify *arg)
886 if (arg->status_change_nid_normal >= 0)
887 node_set_state(node, N_NORMAL_MEMORY);
889 if (arg->status_change_nid_high >= 0)
890 node_set_state(node, N_HIGH_MEMORY);
892 node_set_state(node, N_MEMORY);
896 int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type)
899 unsigned long onlined_pages = 0;
901 int need_zonelists_rebuild = 0;
904 struct memory_notify arg;
906 lock_memory_hotplug();
908 * This doesn't need a lock to do pfn_to_page().
909 * The section can't be removed here because of the
910 * memory_block->state_mutex.
912 zone = page_zone(pfn_to_page(pfn));
914 if ((zone_idx(zone) > ZONE_NORMAL || online_type == ONLINE_MOVABLE) &&
915 !can_online_high_movable(zone)) {
916 unlock_memory_hotplug();
920 if (online_type == ONLINE_KERNEL && zone_idx(zone) == ZONE_MOVABLE) {
921 if (move_pfn_range_left(zone - 1, zone, pfn, pfn + nr_pages)) {
922 unlock_memory_hotplug();
926 if (online_type == ONLINE_MOVABLE && zone_idx(zone) == ZONE_MOVABLE - 1) {
927 if (move_pfn_range_right(zone, zone + 1, pfn, pfn + nr_pages)) {
928 unlock_memory_hotplug();
933 /* Previous code may changed the zone of the pfn range */
934 zone = page_zone(pfn_to_page(pfn));
937 arg.nr_pages = nr_pages;
938 node_states_check_changes_online(nr_pages, zone, &arg);
940 nid = page_to_nid(pfn_to_page(pfn));
942 ret = memory_notify(MEM_GOING_ONLINE, &arg);
943 ret = notifier_to_errno(ret);
945 memory_notify(MEM_CANCEL_ONLINE, &arg);
946 unlock_memory_hotplug();
950 * If this zone is not populated, then it is not in zonelist.
951 * This means the page allocator ignores this zone.
952 * So, zonelist must be updated after online.
954 mutex_lock(&zonelists_mutex);
955 if (!populated_zone(zone)) {
956 need_zonelists_rebuild = 1;
957 build_all_zonelists(NULL, zone);
960 ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
963 if (need_zonelists_rebuild)
964 zone_pcp_reset(zone);
965 mutex_unlock(&zonelists_mutex);
966 printk(KERN_DEBUG "online_pages [mem %#010llx-%#010llx] failed\n",
967 (unsigned long long) pfn << PAGE_SHIFT,
968 (((unsigned long long) pfn + nr_pages)
970 memory_notify(MEM_CANCEL_ONLINE, &arg);
971 unlock_memory_hotplug();
975 zone->present_pages += onlined_pages;
977 pgdat_resize_lock(zone->zone_pgdat, &flags);
978 zone->zone_pgdat->node_present_pages += onlined_pages;
979 pgdat_resize_unlock(zone->zone_pgdat, &flags);
982 node_states_set_node(zone_to_nid(zone), &arg);
983 if (need_zonelists_rebuild)
984 build_all_zonelists(NULL, NULL);
986 zone_pcp_update(zone);
989 mutex_unlock(&zonelists_mutex);
991 init_per_zone_wmark_min();
994 kswapd_run(zone_to_nid(zone));
996 vm_total_pages = nr_free_pagecache_pages();
998 writeback_set_ratelimit();
1001 memory_notify(MEM_ONLINE, &arg);
1002 unlock_memory_hotplug();
1006 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
1008 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1009 static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
1011 struct pglist_data *pgdat;
1012 unsigned long zones_size[MAX_NR_ZONES] = {0};
1013 unsigned long zholes_size[MAX_NR_ZONES] = {0};
1014 unsigned long start_pfn = start >> PAGE_SHIFT;
1016 pgdat = NODE_DATA(nid);
1018 pgdat = arch_alloc_nodedata(nid);
1022 arch_refresh_nodedata(nid, pgdat);
1025 /* we can use NODE_DATA(nid) from here */
1027 /* init node's zones as empty zones, we don't have any present pages.*/
1028 free_area_init_node(nid, zones_size, start_pfn, zholes_size);
1031 * The node we allocated has no zone fallback lists. For avoiding
1032 * to access not-initialized zonelist, build here.
1034 mutex_lock(&zonelists_mutex);
1035 build_all_zonelists(pgdat, NULL);
1036 mutex_unlock(&zonelists_mutex);
1041 static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
1043 arch_refresh_nodedata(nid, NULL);
1044 arch_free_nodedata(pgdat);
1050 * called by cpu_up() to online a node without onlined memory.
1052 int mem_online_node(int nid)
1057 lock_memory_hotplug();
1058 pgdat = hotadd_new_pgdat(nid, 0);
1063 node_set_online(nid);
1064 ret = register_one_node(nid);
1068 unlock_memory_hotplug();
1072 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1073 int __ref add_memory(int nid, u64 start, u64 size)
1075 pg_data_t *pgdat = NULL;
1078 struct resource *res;
1081 lock_memory_hotplug();
1083 res = register_memory_resource(start, size);
1088 { /* Stupid hack to suppress address-never-null warning */
1089 void *p = NODE_DATA(nid);
1092 new_node = !node_online(nid);
1094 pgdat = hotadd_new_pgdat(nid, start);
1100 /* call arch's memory hotadd */
1101 ret = arch_add_memory(nid, start, size);
1106 /* we online node here. we can't roll back from here. */
1107 node_set_online(nid);
1110 ret = register_one_node(nid);
1112 * If sysfs file of new node can't create, cpu on the node
1113 * can't be hot-added. There is no rollback way now.
1114 * So, check by BUG_ON() to catch it reluctantly..
1119 /* create new memmap entry */
1120 firmware_map_add_hotplug(start, start + size, "System RAM");
1125 /* rollback pgdat allocation and others */
1127 rollback_node_hotadd(nid, pgdat);
1128 release_memory_resource(res);
1131 unlock_memory_hotplug();
1134 EXPORT_SYMBOL_GPL(add_memory);
1136 #ifdef CONFIG_MEMORY_HOTREMOVE
1138 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1139 * set and the size of the free page is given by page_order(). Using this,
1140 * the function determines if the pageblock contains only free pages.
1141 * Due to buddy contraints, a free page at least the size of a pageblock will
1142 * be located at the start of the pageblock
1144 static inline int pageblock_free(struct page *page)
1146 return PageBuddy(page) && page_order(page) >= pageblock_order;
1149 /* Return the start of the next active pageblock after a given page */
1150 static struct page *next_active_pageblock(struct page *page)
1152 /* Ensure the starting page is pageblock-aligned */
1153 BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
1155 /* If the entire pageblock is free, move to the end of free page */
1156 if (pageblock_free(page)) {
1158 /* be careful. we don't have locks, page_order can be changed.*/
1159 order = page_order(page);
1160 if ((order < MAX_ORDER) && (order >= pageblock_order))
1161 return page + (1 << order);
1164 return page + pageblock_nr_pages;
1167 /* Checks if this range of memory is likely to be hot-removable. */
1168 int is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
1170 struct page *page = pfn_to_page(start_pfn);
1171 struct page *end_page = page + nr_pages;
1173 /* Check the starting page of each pageblock within the range */
1174 for (; page < end_page; page = next_active_pageblock(page)) {
1175 if (!is_pageblock_removable_nolock(page))
1180 /* All pageblocks in the memory block are likely to be hot-removable */
1185 * Confirm all pages in a range [start, end) is belongs to the same zone.
1187 static int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn)
1190 struct zone *zone = NULL;
1193 for (pfn = start_pfn;
1195 pfn += MAX_ORDER_NR_PAGES) {
1197 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1198 while ((i < MAX_ORDER_NR_PAGES) && !pfn_valid_within(pfn + i))
1200 if (i == MAX_ORDER_NR_PAGES)
1202 page = pfn_to_page(pfn + i);
1203 if (zone && page_zone(page) != zone)
1205 zone = page_zone(page);
1211 * Scanning pfn is much easier than scanning lru list.
1212 * Scan pfn from start to end and Find LRU page.
1214 static unsigned long scan_lru_pages(unsigned long start, unsigned long end)
1218 for (pfn = start; pfn < end; pfn++) {
1219 if (pfn_valid(pfn)) {
1220 page = pfn_to_page(pfn);
1228 #define NR_OFFLINE_AT_ONCE_PAGES (256)
1230 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1234 int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
1235 int not_managed = 0;
1239 for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
1240 if (!pfn_valid(pfn))
1242 page = pfn_to_page(pfn);
1243 if (!get_page_unless_zero(page))
1246 * We can skip free pages. And we can only deal with pages on
1249 ret = isolate_lru_page(page);
1250 if (!ret) { /* Success */
1252 list_add_tail(&page->lru, &source);
1254 inc_zone_page_state(page, NR_ISOLATED_ANON +
1255 page_is_file_cache(page));
1258 #ifdef CONFIG_DEBUG_VM
1259 printk(KERN_ALERT "removing pfn %lx from LRU failed\n",
1264 /* Because we don't have big zone->lock. we should
1265 check this again here. */
1266 if (page_count(page)) {
1273 if (!list_empty(&source)) {
1275 putback_lru_pages(&source);
1280 * alloc_migrate_target should be improooooved!!
1281 * migrate_pages returns # of failed pages.
1283 ret = migrate_pages(&source, alloc_migrate_target, 0,
1284 MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
1286 putback_lru_pages(&source);
1293 * remove from free_area[] and mark all as Reserved.
1296 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1299 __offline_isolated_pages(start, start + nr_pages);
1304 offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
1306 walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
1307 offline_isolated_pages_cb);
1311 * Check all pages in range, recoreded as memory resource, are isolated.
1314 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1318 long offlined = *(long *)data;
1319 ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true);
1320 offlined = nr_pages;
1322 *(long *)data += offlined;
1327 check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
1332 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
1333 check_pages_isolated_cb);
1335 offlined = (long)ret;
1339 #ifdef CONFIG_MOVABLE_NODE
1341 * When CONFIG_MOVABLE_NODE, we permit offlining of a node which doesn't have
1344 static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1348 #else /* CONFIG_MOVABLE_NODE */
1349 /* ensure the node has NORMAL memory if it is still online */
1350 static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1352 struct pglist_data *pgdat = zone->zone_pgdat;
1353 unsigned long present_pages = 0;
1356 for (zt = 0; zt <= ZONE_NORMAL; zt++)
1357 present_pages += pgdat->node_zones[zt].present_pages;
1359 if (present_pages > nr_pages)
1363 for (; zt <= ZONE_MOVABLE; zt++)
1364 present_pages += pgdat->node_zones[zt].present_pages;
1367 * we can't offline the last normal memory until all
1368 * higher memory is offlined.
1370 return present_pages == 0;
1372 #endif /* CONFIG_MOVABLE_NODE */
1374 /* check which state of node_states will be changed when offline memory */
1375 static void node_states_check_changes_offline(unsigned long nr_pages,
1376 struct zone *zone, struct memory_notify *arg)
1378 struct pglist_data *pgdat = zone->zone_pgdat;
1379 unsigned long present_pages = 0;
1380 enum zone_type zt, zone_last = ZONE_NORMAL;
1383 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
1384 * contains nodes which have zones of 0...ZONE_NORMAL,
1385 * set zone_last to ZONE_NORMAL.
1387 * If we don't have HIGHMEM nor movable node,
1388 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
1389 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
1391 if (N_MEMORY == N_NORMAL_MEMORY)
1392 zone_last = ZONE_MOVABLE;
1395 * check whether node_states[N_NORMAL_MEMORY] will be changed.
1396 * If the memory to be offline is in a zone of 0...zone_last,
1397 * and it is the last present memory, 0...zone_last will
1398 * become empty after offline , thus we can determind we will
1399 * need to clear the node from node_states[N_NORMAL_MEMORY].
1401 for (zt = 0; zt <= zone_last; zt++)
1402 present_pages += pgdat->node_zones[zt].present_pages;
1403 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1404 arg->status_change_nid_normal = zone_to_nid(zone);
1406 arg->status_change_nid_normal = -1;
1408 #ifdef CONFIG_HIGHMEM
1410 * If we have movable node, node_states[N_HIGH_MEMORY]
1411 * contains nodes which have zones of 0...ZONE_HIGHMEM,
1412 * set zone_last to ZONE_HIGHMEM.
1414 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
1415 * contains nodes which have zones of 0...ZONE_MOVABLE,
1416 * set zone_last to ZONE_MOVABLE.
1418 zone_last = ZONE_HIGHMEM;
1419 if (N_MEMORY == N_HIGH_MEMORY)
1420 zone_last = ZONE_MOVABLE;
1422 for (; zt <= zone_last; zt++)
1423 present_pages += pgdat->node_zones[zt].present_pages;
1424 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1425 arg->status_change_nid_high = zone_to_nid(zone);
1427 arg->status_change_nid_high = -1;
1429 arg->status_change_nid_high = arg->status_change_nid_normal;
1433 * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE
1435 zone_last = ZONE_MOVABLE;
1438 * check whether node_states[N_HIGH_MEMORY] will be changed
1439 * If we try to offline the last present @nr_pages from the node,
1440 * we can determind we will need to clear the node from
1441 * node_states[N_HIGH_MEMORY].
1443 for (; zt <= zone_last; zt++)
1444 present_pages += pgdat->node_zones[zt].present_pages;
1445 if (nr_pages >= present_pages)
1446 arg->status_change_nid = zone_to_nid(zone);
1448 arg->status_change_nid = -1;
1451 static void node_states_clear_node(int node, struct memory_notify *arg)
1453 if (arg->status_change_nid_normal >= 0)
1454 node_clear_state(node, N_NORMAL_MEMORY);
1456 if ((N_MEMORY != N_NORMAL_MEMORY) &&
1457 (arg->status_change_nid_high >= 0))
1458 node_clear_state(node, N_HIGH_MEMORY);
1460 if ((N_MEMORY != N_HIGH_MEMORY) &&
1461 (arg->status_change_nid >= 0))
1462 node_clear_state(node, N_MEMORY);
1465 static int __ref __offline_pages(unsigned long start_pfn,
1466 unsigned long end_pfn, unsigned long timeout)
1468 unsigned long pfn, nr_pages, expire;
1469 long offlined_pages;
1470 int ret, drain, retry_max, node;
1471 unsigned long flags;
1473 struct memory_notify arg;
1475 BUG_ON(start_pfn >= end_pfn);
1476 /* at least, alignment against pageblock is necessary */
1477 if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
1479 if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
1481 /* This makes hotplug much easier...and readable.
1482 we assume this for now. .*/
1483 if (!test_pages_in_a_zone(start_pfn, end_pfn))
1486 lock_memory_hotplug();
1488 zone = page_zone(pfn_to_page(start_pfn));
1489 node = zone_to_nid(zone);
1490 nr_pages = end_pfn - start_pfn;
1493 if (zone_idx(zone) <= ZONE_NORMAL && !can_offline_normal(zone, nr_pages))
1496 /* set above range as isolated */
1497 ret = start_isolate_page_range(start_pfn, end_pfn,
1498 MIGRATE_MOVABLE, true);
1502 arg.start_pfn = start_pfn;
1503 arg.nr_pages = nr_pages;
1504 node_states_check_changes_offline(nr_pages, zone, &arg);
1506 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1507 ret = notifier_to_errno(ret);
1509 goto failed_removal;
1512 expire = jiffies + timeout;
1516 /* start memory hot removal */
1518 if (time_after(jiffies, expire))
1519 goto failed_removal;
1521 if (signal_pending(current))
1522 goto failed_removal;
1525 lru_add_drain_all();
1530 pfn = scan_lru_pages(start_pfn, end_pfn);
1531 if (pfn) { /* We have page on LRU */
1532 ret = do_migrate_range(pfn, end_pfn);
1538 if (--retry_max == 0)
1539 goto failed_removal;
1545 /* drain all zone's lru pagevec, this is asynchronous... */
1546 lru_add_drain_all();
1548 /* drain pcp pages, this is synchronous. */
1551 offlined_pages = check_pages_isolated(start_pfn, end_pfn);
1552 if (offlined_pages < 0) {
1554 goto failed_removal;
1556 printk(KERN_INFO "Offlined Pages %ld\n", offlined_pages);
1557 /* Ok, all of our target is isolated.
1558 We cannot do rollback at this point. */
1559 offline_isolated_pages(start_pfn, end_pfn);
1560 /* reset pagetype flags and makes migrate type to be MOVABLE */
1561 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1562 /* removal success */
1563 adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages);
1564 zone->present_pages -= offlined_pages;
1566 pgdat_resize_lock(zone->zone_pgdat, &flags);
1567 zone->zone_pgdat->node_present_pages -= offlined_pages;
1568 pgdat_resize_unlock(zone->zone_pgdat, &flags);
1570 init_per_zone_wmark_min();
1572 if (!populated_zone(zone)) {
1573 zone_pcp_reset(zone);
1574 mutex_lock(&zonelists_mutex);
1575 build_all_zonelists(NULL, NULL);
1576 mutex_unlock(&zonelists_mutex);
1578 zone_pcp_update(zone);
1580 node_states_clear_node(node, &arg);
1581 if (arg.status_change_nid >= 0)
1584 vm_total_pages = nr_free_pagecache_pages();
1585 writeback_set_ratelimit();
1587 memory_notify(MEM_OFFLINE, &arg);
1588 unlock_memory_hotplug();
1592 printk(KERN_INFO "memory offlining [mem %#010llx-%#010llx] failed\n",
1593 (unsigned long long) start_pfn << PAGE_SHIFT,
1594 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
1595 memory_notify(MEM_CANCEL_OFFLINE, &arg);
1596 /* pushback to free area */
1597 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1600 unlock_memory_hotplug();
1604 int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1606 return __offline_pages(start_pfn, start_pfn + nr_pages, 120 * HZ);
1608 #endif /* CONFIG_MEMORY_HOTREMOVE */
1611 * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
1612 * @start_pfn: start pfn of the memory range
1613 * @end_pfn: end pfn of the memory range
1614 * @arg: argument passed to func
1615 * @func: callback for each memory section walked
1617 * This function walks through all present mem sections in range
1618 * [start_pfn, end_pfn) and call func on each mem section.
1620 * Returns the return value of func.
1622 int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn,
1623 void *arg, int (*func)(struct memory_block *, void *))
1625 struct memory_block *mem = NULL;
1626 struct mem_section *section;
1627 unsigned long pfn, section_nr;
1630 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1631 section_nr = pfn_to_section_nr(pfn);
1632 if (!present_section_nr(section_nr))
1635 section = __nr_to_section(section_nr);
1636 /* same memblock? */
1638 if ((section_nr >= mem->start_section_nr) &&
1639 (section_nr <= mem->end_section_nr))
1642 mem = find_memory_block_hinted(section, mem);
1646 ret = func(mem, arg);
1648 kobject_put(&mem->dev.kobj);
1654 kobject_put(&mem->dev.kobj);
1659 #ifdef CONFIG_MEMORY_HOTREMOVE
1660 static int is_memblock_offlined_cb(struct memory_block *mem, void *arg)
1662 int ret = !is_memblock_offlined(mem);
1664 if (unlikely(ret)) {
1665 phys_addr_t beginpa, endpa;
1667 beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
1668 endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1;
1669 pr_warn("removing memory fails, because memory "
1670 "[%pa-%pa] is onlined\n",
1677 static int check_cpu_on_node(void *data)
1679 struct pglist_data *pgdat = data;
1682 for_each_present_cpu(cpu) {
1683 if (cpu_to_node(cpu) == pgdat->node_id)
1685 * the cpu on this node isn't removed, and we can't
1686 * offline this node.
1694 static void unmap_cpu_on_node(void *data)
1696 #ifdef CONFIG_ACPI_NUMA
1697 struct pglist_data *pgdat = data;
1700 for_each_possible_cpu(cpu)
1701 if (cpu_to_node(cpu) == pgdat->node_id)
1702 numa_clear_node(cpu);
1706 static int check_and_unmap_cpu_on_node(void *data)
1708 int ret = check_cpu_on_node(data);
1714 * the node will be offlined when we come here, so we can clear
1715 * the cpu_to_node() now.
1718 unmap_cpu_on_node(data);
1722 /* offline the node if all memory sections of this node are removed */
1723 void try_offline_node(int nid)
1725 pg_data_t *pgdat = NODE_DATA(nid);
1726 unsigned long start_pfn = pgdat->node_start_pfn;
1727 unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
1729 struct page *pgdat_page = virt_to_page(pgdat);
1732 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1733 unsigned long section_nr = pfn_to_section_nr(pfn);
1735 if (!present_section_nr(section_nr))
1738 if (pfn_to_nid(pfn) != nid)
1742 * some memory sections of this node are not removed, and we
1743 * can't offline node now.
1748 if (stop_machine(check_and_unmap_cpu_on_node, pgdat, NULL))
1752 * all memory/cpu of this node are removed, we can offline this
1755 node_set_offline(nid);
1756 unregister_one_node(nid);
1758 if (!PageSlab(pgdat_page) && !PageCompound(pgdat_page))
1759 /* node data is allocated from boot memory */
1762 /* free waittable in each zone */
1763 for (i = 0; i < MAX_NR_ZONES; i++) {
1764 struct zone *zone = pgdat->node_zones + i;
1767 * wait_table may be allocated from boot memory,
1768 * here only free if it's allocated by vmalloc.
1770 if (is_vmalloc_addr(zone->wait_table))
1771 vfree(zone->wait_table);
1775 * Since there is no way to guarentee the address of pgdat/zone is not
1776 * on stack of any kernel threads or used by other kernel objects
1777 * without reference counting or other symchronizing method, do not
1778 * reset node_data and free pgdat here. Just reset it to 0 and reuse
1779 * the memory when the node is online again.
1781 memset(pgdat, 0, sizeof(*pgdat));
1783 EXPORT_SYMBOL(try_offline_node);
1785 void __ref remove_memory(int nid, u64 start, u64 size)
1789 lock_memory_hotplug();
1792 * All memory blocks must be offlined before removing memory. Check
1793 * whether all memory blocks in question are offline and trigger a BUG()
1794 * if this is not the case.
1796 ret = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL,
1797 is_memblock_offlined_cb);
1799 unlock_memory_hotplug();
1803 /* remove memmap entry */
1804 firmware_map_remove(start, start + size, "System RAM");
1806 arch_remove_memory(start, size);
1808 try_offline_node(nid);
1810 unlock_memory_hotplug();
1812 EXPORT_SYMBOL_GPL(remove_memory);
1813 #endif /* CONFIG_MEMORY_HOTREMOVE */