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)
766 unsigned long pfn = page_to_pfn(page);
768 if (pfn >= num_physpages)
769 num_physpages = pfn + 1;
771 EXPORT_SYMBOL_GPL(__online_page_set_limits);
773 void __online_page_increment_counters(struct page *page)
777 #ifdef CONFIG_HIGHMEM
778 if (PageHighMem(page))
782 EXPORT_SYMBOL_GPL(__online_page_increment_counters);
784 void __online_page_free(struct page *page)
786 ClearPageReserved(page);
787 init_page_count(page);
790 EXPORT_SYMBOL_GPL(__online_page_free);
792 static void generic_online_page(struct page *page)
794 __online_page_set_limits(page);
795 __online_page_increment_counters(page);
796 __online_page_free(page);
799 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
803 unsigned long onlined_pages = *(unsigned long *)arg;
805 if (PageReserved(pfn_to_page(start_pfn)))
806 for (i = 0; i < nr_pages; i++) {
807 page = pfn_to_page(start_pfn + i);
808 (*online_page_callback)(page);
811 *(unsigned long *)arg = onlined_pages;
815 #ifdef CONFIG_MOVABLE_NODE
817 * When CONFIG_MOVABLE_NODE, we permit onlining of a node which doesn't have
820 static bool can_online_high_movable(struct zone *zone)
824 #else /* CONFIG_MOVABLE_NODE */
825 /* ensure every online node has NORMAL memory */
826 static bool can_online_high_movable(struct zone *zone)
828 return node_state(zone_to_nid(zone), N_NORMAL_MEMORY);
830 #endif /* CONFIG_MOVABLE_NODE */
832 /* check which state of node_states will be changed when online memory */
833 static void node_states_check_changes_online(unsigned long nr_pages,
834 struct zone *zone, struct memory_notify *arg)
836 int nid = zone_to_nid(zone);
837 enum zone_type zone_last = ZONE_NORMAL;
840 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
841 * contains nodes which have zones of 0...ZONE_NORMAL,
842 * set zone_last to ZONE_NORMAL.
844 * If we don't have HIGHMEM nor movable node,
845 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
846 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
848 if (N_MEMORY == N_NORMAL_MEMORY)
849 zone_last = ZONE_MOVABLE;
852 * if the memory to be online is in a zone of 0...zone_last, and
853 * the zones of 0...zone_last don't have memory before online, we will
854 * need to set the node to node_states[N_NORMAL_MEMORY] after
855 * the memory is online.
857 if (zone_idx(zone) <= zone_last && !node_state(nid, N_NORMAL_MEMORY))
858 arg->status_change_nid_normal = nid;
860 arg->status_change_nid_normal = -1;
862 #ifdef CONFIG_HIGHMEM
864 * If we have movable node, node_states[N_HIGH_MEMORY]
865 * contains nodes which have zones of 0...ZONE_HIGHMEM,
866 * set zone_last to ZONE_HIGHMEM.
868 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
869 * contains nodes which have zones of 0...ZONE_MOVABLE,
870 * set zone_last to ZONE_MOVABLE.
872 zone_last = ZONE_HIGHMEM;
873 if (N_MEMORY == N_HIGH_MEMORY)
874 zone_last = ZONE_MOVABLE;
876 if (zone_idx(zone) <= zone_last && !node_state(nid, N_HIGH_MEMORY))
877 arg->status_change_nid_high = nid;
879 arg->status_change_nid_high = -1;
881 arg->status_change_nid_high = arg->status_change_nid_normal;
885 * if the node don't have memory befor online, we will need to
886 * set the node to node_states[N_MEMORY] after the memory
889 if (!node_state(nid, N_MEMORY))
890 arg->status_change_nid = nid;
892 arg->status_change_nid = -1;
895 static void node_states_set_node(int node, struct memory_notify *arg)
897 if (arg->status_change_nid_normal >= 0)
898 node_set_state(node, N_NORMAL_MEMORY);
900 if (arg->status_change_nid_high >= 0)
901 node_set_state(node, N_HIGH_MEMORY);
903 node_set_state(node, N_MEMORY);
907 int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type)
910 unsigned long onlined_pages = 0;
912 int need_zonelists_rebuild = 0;
915 struct memory_notify arg;
917 lock_memory_hotplug();
919 * This doesn't need a lock to do pfn_to_page().
920 * The section can't be removed here because of the
921 * memory_block->state_mutex.
923 zone = page_zone(pfn_to_page(pfn));
925 if ((zone_idx(zone) > ZONE_NORMAL || online_type == ONLINE_MOVABLE) &&
926 !can_online_high_movable(zone)) {
927 unlock_memory_hotplug();
931 if (online_type == ONLINE_KERNEL && zone_idx(zone) == ZONE_MOVABLE) {
932 if (move_pfn_range_left(zone - 1, zone, pfn, pfn + nr_pages)) {
933 unlock_memory_hotplug();
937 if (online_type == ONLINE_MOVABLE && zone_idx(zone) == ZONE_MOVABLE - 1) {
938 if (move_pfn_range_right(zone, zone + 1, pfn, pfn + nr_pages)) {
939 unlock_memory_hotplug();
944 /* Previous code may changed the zone of the pfn range */
945 zone = page_zone(pfn_to_page(pfn));
948 arg.nr_pages = nr_pages;
949 node_states_check_changes_online(nr_pages, zone, &arg);
951 nid = page_to_nid(pfn_to_page(pfn));
953 ret = memory_notify(MEM_GOING_ONLINE, &arg);
954 ret = notifier_to_errno(ret);
956 memory_notify(MEM_CANCEL_ONLINE, &arg);
957 unlock_memory_hotplug();
961 * If this zone is not populated, then it is not in zonelist.
962 * This means the page allocator ignores this zone.
963 * So, zonelist must be updated after online.
965 mutex_lock(&zonelists_mutex);
966 if (!populated_zone(zone)) {
967 need_zonelists_rebuild = 1;
968 build_all_zonelists(NULL, zone);
971 ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
974 if (need_zonelists_rebuild)
975 zone_pcp_reset(zone);
976 mutex_unlock(&zonelists_mutex);
977 printk(KERN_DEBUG "online_pages [mem %#010llx-%#010llx] failed\n",
978 (unsigned long long) pfn << PAGE_SHIFT,
979 (((unsigned long long) pfn + nr_pages)
981 memory_notify(MEM_CANCEL_ONLINE, &arg);
982 unlock_memory_hotplug();
986 zone->managed_pages += onlined_pages;
987 zone->present_pages += onlined_pages;
989 pgdat_resize_lock(zone->zone_pgdat, &flags);
990 zone->zone_pgdat->node_present_pages += onlined_pages;
991 pgdat_resize_unlock(zone->zone_pgdat, &flags);
994 node_states_set_node(zone_to_nid(zone), &arg);
995 if (need_zonelists_rebuild)
996 build_all_zonelists(NULL, NULL);
998 zone_pcp_update(zone);
1001 mutex_unlock(&zonelists_mutex);
1003 init_per_zone_wmark_min();
1006 kswapd_run(zone_to_nid(zone));
1008 vm_total_pages = nr_free_pagecache_pages();
1010 writeback_set_ratelimit();
1013 memory_notify(MEM_ONLINE, &arg);
1014 unlock_memory_hotplug();
1018 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
1020 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1021 static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
1023 struct pglist_data *pgdat;
1024 unsigned long zones_size[MAX_NR_ZONES] = {0};
1025 unsigned long zholes_size[MAX_NR_ZONES] = {0};
1026 unsigned long start_pfn = start >> PAGE_SHIFT;
1028 pgdat = NODE_DATA(nid);
1030 pgdat = arch_alloc_nodedata(nid);
1034 arch_refresh_nodedata(nid, pgdat);
1037 /* we can use NODE_DATA(nid) from here */
1039 /* init node's zones as empty zones, we don't have any present pages.*/
1040 free_area_init_node(nid, zones_size, start_pfn, zholes_size);
1043 * The node we allocated has no zone fallback lists. For avoiding
1044 * to access not-initialized zonelist, build here.
1046 mutex_lock(&zonelists_mutex);
1047 build_all_zonelists(pgdat, NULL);
1048 mutex_unlock(&zonelists_mutex);
1053 static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
1055 arch_refresh_nodedata(nid, NULL);
1056 arch_free_nodedata(pgdat);
1062 * called by cpu_up() to online a node without onlined memory.
1064 int mem_online_node(int nid)
1069 lock_memory_hotplug();
1070 pgdat = hotadd_new_pgdat(nid, 0);
1075 node_set_online(nid);
1076 ret = register_one_node(nid);
1080 unlock_memory_hotplug();
1084 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1085 int __ref add_memory(int nid, u64 start, u64 size)
1087 pg_data_t *pgdat = NULL;
1090 struct resource *res;
1093 lock_memory_hotplug();
1095 res = register_memory_resource(start, size);
1100 { /* Stupid hack to suppress address-never-null warning */
1101 void *p = NODE_DATA(nid);
1104 new_node = !node_online(nid);
1106 pgdat = hotadd_new_pgdat(nid, start);
1112 /* call arch's memory hotadd */
1113 ret = arch_add_memory(nid, start, size);
1118 /* we online node here. we can't roll back from here. */
1119 node_set_online(nid);
1122 ret = register_one_node(nid);
1124 * If sysfs file of new node can't create, cpu on the node
1125 * can't be hot-added. There is no rollback way now.
1126 * So, check by BUG_ON() to catch it reluctantly..
1131 /* create new memmap entry */
1132 firmware_map_add_hotplug(start, start + size, "System RAM");
1137 /* rollback pgdat allocation and others */
1139 rollback_node_hotadd(nid, pgdat);
1140 release_memory_resource(res);
1143 unlock_memory_hotplug();
1146 EXPORT_SYMBOL_GPL(add_memory);
1148 #ifdef CONFIG_MEMORY_HOTREMOVE
1150 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1151 * set and the size of the free page is given by page_order(). Using this,
1152 * the function determines if the pageblock contains only free pages.
1153 * Due to buddy contraints, a free page at least the size of a pageblock will
1154 * be located at the start of the pageblock
1156 static inline int pageblock_free(struct page *page)
1158 return PageBuddy(page) && page_order(page) >= pageblock_order;
1161 /* Return the start of the next active pageblock after a given page */
1162 static struct page *next_active_pageblock(struct page *page)
1164 /* Ensure the starting page is pageblock-aligned */
1165 BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
1167 /* If the entire pageblock is free, move to the end of free page */
1168 if (pageblock_free(page)) {
1170 /* be careful. we don't have locks, page_order can be changed.*/
1171 order = page_order(page);
1172 if ((order < MAX_ORDER) && (order >= pageblock_order))
1173 return page + (1 << order);
1176 return page + pageblock_nr_pages;
1179 /* Checks if this range of memory is likely to be hot-removable. */
1180 int is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
1182 struct page *page = pfn_to_page(start_pfn);
1183 struct page *end_page = page + nr_pages;
1185 /* Check the starting page of each pageblock within the range */
1186 for (; page < end_page; page = next_active_pageblock(page)) {
1187 if (!is_pageblock_removable_nolock(page))
1192 /* All pageblocks in the memory block are likely to be hot-removable */
1197 * Confirm all pages in a range [start, end) is belongs to the same zone.
1199 static int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn)
1202 struct zone *zone = NULL;
1205 for (pfn = start_pfn;
1207 pfn += MAX_ORDER_NR_PAGES) {
1209 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1210 while ((i < MAX_ORDER_NR_PAGES) && !pfn_valid_within(pfn + i))
1212 if (i == MAX_ORDER_NR_PAGES)
1214 page = pfn_to_page(pfn + i);
1215 if (zone && page_zone(page) != zone)
1217 zone = page_zone(page);
1223 * Scanning pfn is much easier than scanning lru list.
1224 * Scan pfn from start to end and Find LRU page.
1226 static unsigned long scan_lru_pages(unsigned long start, unsigned long end)
1230 for (pfn = start; pfn < end; pfn++) {
1231 if (pfn_valid(pfn)) {
1232 page = pfn_to_page(pfn);
1240 #define NR_OFFLINE_AT_ONCE_PAGES (256)
1242 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1246 int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
1247 int not_managed = 0;
1251 for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
1252 if (!pfn_valid(pfn))
1254 page = pfn_to_page(pfn);
1255 if (!get_page_unless_zero(page))
1258 * We can skip free pages. And we can only deal with pages on
1261 ret = isolate_lru_page(page);
1262 if (!ret) { /* Success */
1264 list_add_tail(&page->lru, &source);
1266 inc_zone_page_state(page, NR_ISOLATED_ANON +
1267 page_is_file_cache(page));
1270 #ifdef CONFIG_DEBUG_VM
1271 printk(KERN_ALERT "removing pfn %lx from LRU failed\n",
1276 /* Because we don't have big zone->lock. we should
1277 check this again here. */
1278 if (page_count(page)) {
1285 if (!list_empty(&source)) {
1287 putback_lru_pages(&source);
1292 * alloc_migrate_target should be improooooved!!
1293 * migrate_pages returns # of failed pages.
1295 ret = migrate_pages(&source, alloc_migrate_target, 0,
1296 MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
1298 putback_lru_pages(&source);
1305 * remove from free_area[] and mark all as Reserved.
1308 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1311 __offline_isolated_pages(start, start + nr_pages);
1316 offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
1318 walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
1319 offline_isolated_pages_cb);
1323 * Check all pages in range, recoreded as memory resource, are isolated.
1326 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1330 long offlined = *(long *)data;
1331 ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true);
1332 offlined = nr_pages;
1334 *(long *)data += offlined;
1339 check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
1344 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
1345 check_pages_isolated_cb);
1347 offlined = (long)ret;
1351 #ifdef CONFIG_MOVABLE_NODE
1353 * When CONFIG_MOVABLE_NODE, we permit offlining of a node which doesn't have
1356 static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1360 #else /* CONFIG_MOVABLE_NODE */
1361 /* ensure the node has NORMAL memory if it is still online */
1362 static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1364 struct pglist_data *pgdat = zone->zone_pgdat;
1365 unsigned long present_pages = 0;
1368 for (zt = 0; zt <= ZONE_NORMAL; zt++)
1369 present_pages += pgdat->node_zones[zt].present_pages;
1371 if (present_pages > nr_pages)
1375 for (; zt <= ZONE_MOVABLE; zt++)
1376 present_pages += pgdat->node_zones[zt].present_pages;
1379 * we can't offline the last normal memory until all
1380 * higher memory is offlined.
1382 return present_pages == 0;
1384 #endif /* CONFIG_MOVABLE_NODE */
1386 /* check which state of node_states will be changed when offline memory */
1387 static void node_states_check_changes_offline(unsigned long nr_pages,
1388 struct zone *zone, struct memory_notify *arg)
1390 struct pglist_data *pgdat = zone->zone_pgdat;
1391 unsigned long present_pages = 0;
1392 enum zone_type zt, zone_last = ZONE_NORMAL;
1395 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
1396 * contains nodes which have zones of 0...ZONE_NORMAL,
1397 * set zone_last to ZONE_NORMAL.
1399 * If we don't have HIGHMEM nor movable node,
1400 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
1401 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
1403 if (N_MEMORY == N_NORMAL_MEMORY)
1404 zone_last = ZONE_MOVABLE;
1407 * check whether node_states[N_NORMAL_MEMORY] will be changed.
1408 * If the memory to be offline is in a zone of 0...zone_last,
1409 * and it is the last present memory, 0...zone_last will
1410 * become empty after offline , thus we can determind we will
1411 * need to clear the node from node_states[N_NORMAL_MEMORY].
1413 for (zt = 0; zt <= zone_last; zt++)
1414 present_pages += pgdat->node_zones[zt].present_pages;
1415 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1416 arg->status_change_nid_normal = zone_to_nid(zone);
1418 arg->status_change_nid_normal = -1;
1420 #ifdef CONFIG_HIGHMEM
1422 * If we have movable node, node_states[N_HIGH_MEMORY]
1423 * contains nodes which have zones of 0...ZONE_HIGHMEM,
1424 * set zone_last to ZONE_HIGHMEM.
1426 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
1427 * contains nodes which have zones of 0...ZONE_MOVABLE,
1428 * set zone_last to ZONE_MOVABLE.
1430 zone_last = ZONE_HIGHMEM;
1431 if (N_MEMORY == N_HIGH_MEMORY)
1432 zone_last = ZONE_MOVABLE;
1434 for (; zt <= zone_last; zt++)
1435 present_pages += pgdat->node_zones[zt].present_pages;
1436 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1437 arg->status_change_nid_high = zone_to_nid(zone);
1439 arg->status_change_nid_high = -1;
1441 arg->status_change_nid_high = arg->status_change_nid_normal;
1445 * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE
1447 zone_last = ZONE_MOVABLE;
1450 * check whether node_states[N_HIGH_MEMORY] will be changed
1451 * If we try to offline the last present @nr_pages from the node,
1452 * we can determind we will need to clear the node from
1453 * node_states[N_HIGH_MEMORY].
1455 for (; zt <= zone_last; zt++)
1456 present_pages += pgdat->node_zones[zt].present_pages;
1457 if (nr_pages >= present_pages)
1458 arg->status_change_nid = zone_to_nid(zone);
1460 arg->status_change_nid = -1;
1463 static void node_states_clear_node(int node, struct memory_notify *arg)
1465 if (arg->status_change_nid_normal >= 0)
1466 node_clear_state(node, N_NORMAL_MEMORY);
1468 if ((N_MEMORY != N_NORMAL_MEMORY) &&
1469 (arg->status_change_nid_high >= 0))
1470 node_clear_state(node, N_HIGH_MEMORY);
1472 if ((N_MEMORY != N_HIGH_MEMORY) &&
1473 (arg->status_change_nid >= 0))
1474 node_clear_state(node, N_MEMORY);
1477 static int __ref __offline_pages(unsigned long start_pfn,
1478 unsigned long end_pfn, unsigned long timeout)
1480 unsigned long pfn, nr_pages, expire;
1481 long offlined_pages;
1482 int ret, drain, retry_max, node;
1483 unsigned long flags;
1485 struct memory_notify arg;
1487 BUG_ON(start_pfn >= end_pfn);
1488 /* at least, alignment against pageblock is necessary */
1489 if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
1491 if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
1493 /* This makes hotplug much easier...and readable.
1494 we assume this for now. .*/
1495 if (!test_pages_in_a_zone(start_pfn, end_pfn))
1498 lock_memory_hotplug();
1500 zone = page_zone(pfn_to_page(start_pfn));
1501 node = zone_to_nid(zone);
1502 nr_pages = end_pfn - start_pfn;
1505 if (zone_idx(zone) <= ZONE_NORMAL && !can_offline_normal(zone, nr_pages))
1508 /* set above range as isolated */
1509 ret = start_isolate_page_range(start_pfn, end_pfn,
1510 MIGRATE_MOVABLE, true);
1514 arg.start_pfn = start_pfn;
1515 arg.nr_pages = nr_pages;
1516 node_states_check_changes_offline(nr_pages, zone, &arg);
1518 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1519 ret = notifier_to_errno(ret);
1521 goto failed_removal;
1524 expire = jiffies + timeout;
1528 /* start memory hot removal */
1530 if (time_after(jiffies, expire))
1531 goto failed_removal;
1533 if (signal_pending(current))
1534 goto failed_removal;
1537 lru_add_drain_all();
1542 pfn = scan_lru_pages(start_pfn, end_pfn);
1543 if (pfn) { /* We have page on LRU */
1544 ret = do_migrate_range(pfn, end_pfn);
1550 if (--retry_max == 0)
1551 goto failed_removal;
1557 /* drain all zone's lru pagevec, this is asynchronous... */
1558 lru_add_drain_all();
1560 /* drain pcp pages, this is synchronous. */
1563 offlined_pages = check_pages_isolated(start_pfn, end_pfn);
1564 if (offlined_pages < 0) {
1566 goto failed_removal;
1568 printk(KERN_INFO "Offlined Pages %ld\n", offlined_pages);
1569 /* Ok, all of our target is isolated.
1570 We cannot do rollback at this point. */
1571 offline_isolated_pages(start_pfn, end_pfn);
1572 /* reset pagetype flags and makes migrate type to be MOVABLE */
1573 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1574 /* removal success */
1575 zone->managed_pages -= offlined_pages;
1576 zone->present_pages -= offlined_pages;
1578 pgdat_resize_lock(zone->zone_pgdat, &flags);
1579 zone->zone_pgdat->node_present_pages -= offlined_pages;
1580 pgdat_resize_unlock(zone->zone_pgdat, &flags);
1582 totalram_pages -= offlined_pages;
1584 init_per_zone_wmark_min();
1586 if (!populated_zone(zone)) {
1587 zone_pcp_reset(zone);
1588 mutex_lock(&zonelists_mutex);
1589 build_all_zonelists(NULL, NULL);
1590 mutex_unlock(&zonelists_mutex);
1592 zone_pcp_update(zone);
1594 node_states_clear_node(node, &arg);
1595 if (arg.status_change_nid >= 0)
1598 vm_total_pages = nr_free_pagecache_pages();
1599 writeback_set_ratelimit();
1601 memory_notify(MEM_OFFLINE, &arg);
1602 unlock_memory_hotplug();
1606 printk(KERN_INFO "memory offlining [mem %#010llx-%#010llx] failed\n",
1607 (unsigned long long) start_pfn << PAGE_SHIFT,
1608 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
1609 memory_notify(MEM_CANCEL_OFFLINE, &arg);
1610 /* pushback to free area */
1611 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1614 unlock_memory_hotplug();
1618 int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1620 return __offline_pages(start_pfn, start_pfn + nr_pages, 120 * HZ);
1624 * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
1625 * @start_pfn: start pfn of the memory range
1626 * @end_pfn: end pfn of the memory range
1627 * @arg: argument passed to func
1628 * @func: callback for each memory section walked
1630 * This function walks through all present mem sections in range
1631 * [start_pfn, end_pfn) and call func on each mem section.
1633 * Returns the return value of func.
1635 static int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn,
1636 void *arg, int (*func)(struct memory_block *, void *))
1638 struct memory_block *mem = NULL;
1639 struct mem_section *section;
1640 unsigned long pfn, section_nr;
1643 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1644 section_nr = pfn_to_section_nr(pfn);
1645 if (!present_section_nr(section_nr))
1648 section = __nr_to_section(section_nr);
1649 /* same memblock? */
1651 if ((section_nr >= mem->start_section_nr) &&
1652 (section_nr <= mem->end_section_nr))
1655 mem = find_memory_block_hinted(section, mem);
1659 ret = func(mem, arg);
1661 kobject_put(&mem->dev.kobj);
1667 kobject_put(&mem->dev.kobj);
1673 * offline_memory_block_cb - callback function for offlining memory block
1674 * @mem: the memory block to be offlined
1675 * @arg: buffer to hold error msg
1677 * Always return 0, and put the error msg in arg if any.
1679 static int offline_memory_block_cb(struct memory_block *mem, void *arg)
1682 int error = offline_memory_block(mem);
1684 if (error != 0 && *ret == 0)
1690 static int is_memblock_offlined_cb(struct memory_block *mem, void *arg)
1692 int ret = !is_memblock_offlined(mem);
1694 if (unlikely(ret)) {
1695 phys_addr_t beginpa, endpa;
1697 beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
1698 endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1;
1699 pr_warn("removing memory fails, because memory "
1700 "[%pa-%pa] is onlined\n",
1707 static int check_cpu_on_node(void *data)
1709 struct pglist_data *pgdat = data;
1712 for_each_present_cpu(cpu) {
1713 if (cpu_to_node(cpu) == pgdat->node_id)
1715 * the cpu on this node isn't removed, and we can't
1716 * offline this node.
1724 static void unmap_cpu_on_node(void *data)
1726 #ifdef CONFIG_ACPI_NUMA
1727 struct pglist_data *pgdat = data;
1730 for_each_possible_cpu(cpu)
1731 if (cpu_to_node(cpu) == pgdat->node_id)
1732 numa_clear_node(cpu);
1736 static int check_and_unmap_cpu_on_node(void *data)
1738 int ret = check_cpu_on_node(data);
1744 * the node will be offlined when we come here, so we can clear
1745 * the cpu_to_node() now.
1748 unmap_cpu_on_node(data);
1752 /* offline the node if all memory sections of this node are removed */
1753 void try_offline_node(int nid)
1755 pg_data_t *pgdat = NODE_DATA(nid);
1756 unsigned long start_pfn = pgdat->node_start_pfn;
1757 unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
1759 struct page *pgdat_page = virt_to_page(pgdat);
1762 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1763 unsigned long section_nr = pfn_to_section_nr(pfn);
1765 if (!present_section_nr(section_nr))
1768 if (pfn_to_nid(pfn) != nid)
1772 * some memory sections of this node are not removed, and we
1773 * can't offline node now.
1778 if (stop_machine(check_and_unmap_cpu_on_node, pgdat, NULL))
1782 * all memory/cpu of this node are removed, we can offline this
1785 node_set_offline(nid);
1786 unregister_one_node(nid);
1788 if (!PageSlab(pgdat_page) && !PageCompound(pgdat_page))
1789 /* node data is allocated from boot memory */
1792 /* free waittable in each zone */
1793 for (i = 0; i < MAX_NR_ZONES; i++) {
1794 struct zone *zone = pgdat->node_zones + i;
1797 * wait_table may be allocated from boot memory,
1798 * here only free if it's allocated by vmalloc.
1800 if (is_vmalloc_addr(zone->wait_table))
1801 vfree(zone->wait_table);
1805 * Since there is no way to guarentee the address of pgdat/zone is not
1806 * on stack of any kernel threads or used by other kernel objects
1807 * without reference counting or other symchronizing method, do not
1808 * reset node_data and free pgdat here. Just reset it to 0 and reuse
1809 * the memory when the node is online again.
1811 memset(pgdat, 0, sizeof(*pgdat));
1813 EXPORT_SYMBOL(try_offline_node);
1815 int __ref remove_memory(int nid, u64 start, u64 size)
1817 unsigned long start_pfn, end_pfn;
1821 start_pfn = PFN_DOWN(start);
1822 end_pfn = PFN_UP(start + size - 1);
1825 * When CONFIG_MEMCG is on, one memory block may be used by other
1826 * blocks to store page cgroup when onlining pages. But we don't know
1827 * in what order pages are onlined. So we iterate twice to offline
1829 * 1st iterate: offline every non primary memory block.
1830 * 2nd iterate: offline primary (i.e. first added) memory block.
1833 walk_memory_range(start_pfn, end_pfn, &ret,
1834 offline_memory_block_cb);
1844 lock_memory_hotplug();
1847 * we have offlined all memory blocks like this:
1848 * 1. lock memory hotplug
1849 * 2. offline a memory block
1850 * 3. unlock memory hotplug
1852 * repeat step1-3 to offline the memory block. All memory blocks
1853 * must be offlined before removing memory. But we don't hold the
1854 * lock in the whole operation. So we should check whether all
1855 * memory blocks are offlined.
1858 ret = walk_memory_range(start_pfn, end_pfn, NULL,
1859 is_memblock_offlined_cb);
1861 unlock_memory_hotplug();
1865 /* remove memmap entry */
1866 firmware_map_remove(start, start + size, "System RAM");
1868 arch_remove_memory(start, size);
1870 try_offline_node(nid);
1872 unlock_memory_hotplug();
1877 int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1881 int remove_memory(int nid, u64 start, u64 size)
1885 #endif /* CONFIG_MEMORY_HOTREMOVE */
1886 EXPORT_SYMBOL_GPL(remove_memory);