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/compiler.h>
13 #include <linux/export.h>
14 #include <linux/pagevec.h>
15 #include <linux/writeback.h>
16 #include <linux/slab.h>
17 #include <linux/sysctl.h>
18 #include <linux/cpu.h>
19 #include <linux/memory.h>
20 #include <linux/memory_hotplug.h>
21 #include <linux/highmem.h>
22 #include <linux/vmalloc.h>
23 #include <linux/ioport.h>
24 #include <linux/delay.h>
25 #include <linux/migrate.h>
26 #include <linux/page-isolation.h>
27 #include <linux/pfn.h>
28 #include <linux/suspend.h>
29 #include <linux/mm_inline.h>
30 #include <linux/firmware-map.h>
31 #include <linux/stop_machine.h>
32 #include <linux/hugetlb.h>
33 #include <linux/memblock.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()/memblock_virt_alloc_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 = pgdat_end_pfn(pgdat);
370 if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
371 pgdat->node_start_pfn = start_pfn;
373 pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
374 pgdat->node_start_pfn;
377 static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn)
379 struct pglist_data *pgdat = zone->zone_pgdat;
380 int nr_pages = PAGES_PER_SECTION;
381 int nid = pgdat->node_id;
386 zone_type = zone - pgdat->node_zones;
387 ret = ensure_zone_is_initialized(zone, phys_start_pfn, nr_pages);
391 pgdat_resize_lock(zone->zone_pgdat, &flags);
392 grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages);
393 grow_pgdat_span(zone->zone_pgdat, phys_start_pfn,
394 phys_start_pfn + nr_pages);
395 pgdat_resize_unlock(zone->zone_pgdat, &flags);
396 memmap_init_zone(nr_pages, nid, zone_type,
397 phys_start_pfn, MEMMAP_HOTPLUG);
401 static int __meminit __add_section(int nid, struct zone *zone,
402 unsigned long phys_start_pfn)
406 if (pfn_valid(phys_start_pfn))
409 ret = sparse_add_one_section(zone, phys_start_pfn);
414 ret = __add_zone(zone, phys_start_pfn);
419 return register_new_memory(nid, __pfn_to_section(phys_start_pfn));
423 * Reasonably generic function for adding memory. It is
424 * expected that archs that support memory hotplug will
425 * call this function after deciding the zone to which to
428 int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn,
429 unsigned long nr_pages)
433 int start_sec, end_sec;
434 /* during initialize mem_map, align hot-added range to section */
435 start_sec = pfn_to_section_nr(phys_start_pfn);
436 end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
438 for (i = start_sec; i <= end_sec; i++) {
439 err = __add_section(nid, zone, i << PFN_SECTION_SHIFT);
442 * EEXIST is finally dealt with by ioresource collision
443 * check. see add_memory() => register_memory_resource()
444 * Warning will be printed if there is collision.
446 if (err && (err != -EEXIST))
453 EXPORT_SYMBOL_GPL(__add_pages);
455 #ifdef CONFIG_MEMORY_HOTREMOVE
456 /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
457 static int find_smallest_section_pfn(int nid, struct zone *zone,
458 unsigned long start_pfn,
459 unsigned long end_pfn)
461 struct mem_section *ms;
463 for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) {
464 ms = __pfn_to_section(start_pfn);
466 if (unlikely(!valid_section(ms)))
469 if (unlikely(pfn_to_nid(start_pfn) != nid))
472 if (zone && zone != page_zone(pfn_to_page(start_pfn)))
481 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
482 static int find_biggest_section_pfn(int nid, struct zone *zone,
483 unsigned long start_pfn,
484 unsigned long end_pfn)
486 struct mem_section *ms;
489 /* pfn is the end pfn of a memory section. */
491 for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) {
492 ms = __pfn_to_section(pfn);
494 if (unlikely(!valid_section(ms)))
497 if (unlikely(pfn_to_nid(pfn) != nid))
500 if (zone && zone != page_zone(pfn_to_page(pfn)))
509 static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
510 unsigned long end_pfn)
512 unsigned long zone_start_pfn = zone->zone_start_pfn;
513 unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */
514 unsigned long zone_end_pfn = z;
516 struct mem_section *ms;
517 int nid = zone_to_nid(zone);
519 zone_span_writelock(zone);
520 if (zone_start_pfn == start_pfn) {
522 * If the section is smallest section in the zone, it need
523 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
524 * In this case, we find second smallest valid mem_section
525 * for shrinking zone.
527 pfn = find_smallest_section_pfn(nid, zone, end_pfn,
530 zone->zone_start_pfn = pfn;
531 zone->spanned_pages = zone_end_pfn - pfn;
533 } else if (zone_end_pfn == end_pfn) {
535 * If the section is biggest section in the zone, it need
536 * shrink zone->spanned_pages.
537 * In this case, we find second biggest valid mem_section for
540 pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn,
543 zone->spanned_pages = pfn - zone_start_pfn + 1;
547 * The section is not biggest or smallest mem_section in the zone, it
548 * only creates a hole in the zone. So in this case, we need not
549 * change the zone. But perhaps, the zone has only hole data. Thus
550 * it check the zone has only hole or not.
552 pfn = zone_start_pfn;
553 for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) {
554 ms = __pfn_to_section(pfn);
556 if (unlikely(!valid_section(ms)))
559 if (page_zone(pfn_to_page(pfn)) != zone)
562 /* If the section is current section, it continues the loop */
563 if (start_pfn == pfn)
566 /* If we find valid section, we have nothing to do */
567 zone_span_writeunlock(zone);
571 /* The zone has no valid section */
572 zone->zone_start_pfn = 0;
573 zone->spanned_pages = 0;
574 zone_span_writeunlock(zone);
577 static void shrink_pgdat_span(struct pglist_data *pgdat,
578 unsigned long start_pfn, unsigned long end_pfn)
580 unsigned long pgdat_start_pfn = pgdat->node_start_pfn;
581 unsigned long p = pgdat_end_pfn(pgdat); /* pgdat_end_pfn namespace clash */
582 unsigned long pgdat_end_pfn = p;
584 struct mem_section *ms;
585 int nid = pgdat->node_id;
587 if (pgdat_start_pfn == start_pfn) {
589 * If the section is smallest section in the pgdat, it need
590 * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages.
591 * In this case, we find second smallest valid mem_section
592 * for shrinking zone.
594 pfn = find_smallest_section_pfn(nid, NULL, end_pfn,
597 pgdat->node_start_pfn = pfn;
598 pgdat->node_spanned_pages = pgdat_end_pfn - pfn;
600 } else if (pgdat_end_pfn == end_pfn) {
602 * If the section is biggest section in the pgdat, it need
603 * shrink pgdat->node_spanned_pages.
604 * In this case, we find second biggest valid mem_section for
607 pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn,
610 pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1;
614 * If the section is not biggest or smallest mem_section in the pgdat,
615 * it only creates a hole in the pgdat. So in this case, we need not
617 * But perhaps, the pgdat has only hole data. Thus it check the pgdat
618 * has only hole or not.
620 pfn = pgdat_start_pfn;
621 for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) {
622 ms = __pfn_to_section(pfn);
624 if (unlikely(!valid_section(ms)))
627 if (pfn_to_nid(pfn) != nid)
630 /* If the section is current section, it continues the loop */
631 if (start_pfn == pfn)
634 /* If we find valid section, we have nothing to do */
638 /* The pgdat has no valid section */
639 pgdat->node_start_pfn = 0;
640 pgdat->node_spanned_pages = 0;
643 static void __remove_zone(struct zone *zone, unsigned long start_pfn)
645 struct pglist_data *pgdat = zone->zone_pgdat;
646 int nr_pages = PAGES_PER_SECTION;
650 zone_type = zone - pgdat->node_zones;
652 pgdat_resize_lock(zone->zone_pgdat, &flags);
653 shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
654 shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages);
655 pgdat_resize_unlock(zone->zone_pgdat, &flags);
658 static int __remove_section(struct zone *zone, struct mem_section *ms)
660 unsigned long start_pfn;
664 if (!valid_section(ms))
667 ret = unregister_memory_section(ms);
671 scn_nr = __section_nr(ms);
672 start_pfn = section_nr_to_pfn(scn_nr);
673 __remove_zone(zone, start_pfn);
675 sparse_remove_one_section(zone, ms);
680 * __remove_pages() - remove sections of pages from a zone
681 * @zone: zone from which pages need to be removed
682 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
683 * @nr_pages: number of pages to remove (must be multiple of section size)
685 * Generic helper function to remove section mappings and sysfs entries
686 * for the section of the memory we are removing. Caller needs to make
687 * sure that pages are marked reserved and zones are adjust properly by
688 * calling offline_pages().
690 int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
691 unsigned long nr_pages)
694 int sections_to_remove;
695 resource_size_t start, size;
699 * We can only remove entire sections
701 BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
702 BUG_ON(nr_pages % PAGES_PER_SECTION);
704 start = phys_start_pfn << PAGE_SHIFT;
705 size = nr_pages * PAGE_SIZE;
706 ret = release_mem_region_adjustable(&iomem_resource, start, size);
708 resource_size_t endres = start + size - 1;
710 pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
711 &start, &endres, ret);
714 sections_to_remove = nr_pages / PAGES_PER_SECTION;
715 for (i = 0; i < sections_to_remove; i++) {
716 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
717 ret = __remove_section(zone, __pfn_to_section(pfn));
723 EXPORT_SYMBOL_GPL(__remove_pages);
724 #endif /* CONFIG_MEMORY_HOTREMOVE */
726 int set_online_page_callback(online_page_callback_t callback)
730 lock_memory_hotplug();
732 if (online_page_callback == generic_online_page) {
733 online_page_callback = callback;
737 unlock_memory_hotplug();
741 EXPORT_SYMBOL_GPL(set_online_page_callback);
743 int restore_online_page_callback(online_page_callback_t callback)
747 lock_memory_hotplug();
749 if (online_page_callback == callback) {
750 online_page_callback = generic_online_page;
754 unlock_memory_hotplug();
758 EXPORT_SYMBOL_GPL(restore_online_page_callback);
760 void __online_page_set_limits(struct page *page)
763 EXPORT_SYMBOL_GPL(__online_page_set_limits);
765 void __online_page_increment_counters(struct page *page)
767 adjust_managed_page_count(page, 1);
769 EXPORT_SYMBOL_GPL(__online_page_increment_counters);
771 void __online_page_free(struct page *page)
773 __free_reserved_page(page);
775 EXPORT_SYMBOL_GPL(__online_page_free);
777 static void generic_online_page(struct page *page)
779 __online_page_set_limits(page);
780 __online_page_increment_counters(page);
781 __online_page_free(page);
784 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
788 unsigned long onlined_pages = *(unsigned long *)arg;
790 if (PageReserved(pfn_to_page(start_pfn)))
791 for (i = 0; i < nr_pages; i++) {
792 page = pfn_to_page(start_pfn + i);
793 (*online_page_callback)(page);
796 *(unsigned long *)arg = onlined_pages;
800 #ifdef CONFIG_MOVABLE_NODE
802 * When CONFIG_MOVABLE_NODE, we permit onlining of a node which doesn't have
805 static bool can_online_high_movable(struct zone *zone)
809 #else /* CONFIG_MOVABLE_NODE */
810 /* ensure every online node has NORMAL memory */
811 static bool can_online_high_movable(struct zone *zone)
813 return node_state(zone_to_nid(zone), N_NORMAL_MEMORY);
815 #endif /* CONFIG_MOVABLE_NODE */
817 /* check which state of node_states will be changed when online memory */
818 static void node_states_check_changes_online(unsigned long nr_pages,
819 struct zone *zone, struct memory_notify *arg)
821 int nid = zone_to_nid(zone);
822 enum zone_type zone_last = ZONE_NORMAL;
825 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
826 * contains nodes which have zones of 0...ZONE_NORMAL,
827 * set zone_last to ZONE_NORMAL.
829 * If we don't have HIGHMEM nor movable node,
830 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
831 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
833 if (N_MEMORY == N_NORMAL_MEMORY)
834 zone_last = ZONE_MOVABLE;
837 * if the memory to be online is in a zone of 0...zone_last, and
838 * the zones of 0...zone_last don't have memory before online, we will
839 * need to set the node to node_states[N_NORMAL_MEMORY] after
840 * the memory is online.
842 if (zone_idx(zone) <= zone_last && !node_state(nid, N_NORMAL_MEMORY))
843 arg->status_change_nid_normal = nid;
845 arg->status_change_nid_normal = -1;
847 #ifdef CONFIG_HIGHMEM
849 * If we have movable node, node_states[N_HIGH_MEMORY]
850 * contains nodes which have zones of 0...ZONE_HIGHMEM,
851 * set zone_last to ZONE_HIGHMEM.
853 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
854 * contains nodes which have zones of 0...ZONE_MOVABLE,
855 * set zone_last to ZONE_MOVABLE.
857 zone_last = ZONE_HIGHMEM;
858 if (N_MEMORY == N_HIGH_MEMORY)
859 zone_last = ZONE_MOVABLE;
861 if (zone_idx(zone) <= zone_last && !node_state(nid, N_HIGH_MEMORY))
862 arg->status_change_nid_high = nid;
864 arg->status_change_nid_high = -1;
866 arg->status_change_nid_high = arg->status_change_nid_normal;
870 * if the node don't have memory befor online, we will need to
871 * set the node to node_states[N_MEMORY] after the memory
874 if (!node_state(nid, N_MEMORY))
875 arg->status_change_nid = nid;
877 arg->status_change_nid = -1;
880 static void node_states_set_node(int node, struct memory_notify *arg)
882 if (arg->status_change_nid_normal >= 0)
883 node_set_state(node, N_NORMAL_MEMORY);
885 if (arg->status_change_nid_high >= 0)
886 node_set_state(node, N_HIGH_MEMORY);
888 node_set_state(node, N_MEMORY);
892 int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type)
895 unsigned long onlined_pages = 0;
897 int need_zonelists_rebuild = 0;
900 struct memory_notify arg;
902 lock_memory_hotplug();
904 * This doesn't need a lock to do pfn_to_page().
905 * The section can't be removed here because of the
906 * memory_block->state_mutex.
908 zone = page_zone(pfn_to_page(pfn));
910 if ((zone_idx(zone) > ZONE_NORMAL || online_type == ONLINE_MOVABLE) &&
911 !can_online_high_movable(zone)) {
912 unlock_memory_hotplug();
916 if (online_type == ONLINE_KERNEL && zone_idx(zone) == ZONE_MOVABLE) {
917 if (move_pfn_range_left(zone - 1, zone, pfn, pfn + nr_pages)) {
918 unlock_memory_hotplug();
922 if (online_type == ONLINE_MOVABLE && zone_idx(zone) == ZONE_MOVABLE - 1) {
923 if (move_pfn_range_right(zone, zone + 1, pfn, pfn + nr_pages)) {
924 unlock_memory_hotplug();
929 /* Previous code may changed the zone of the pfn range */
930 zone = page_zone(pfn_to_page(pfn));
933 arg.nr_pages = nr_pages;
934 node_states_check_changes_online(nr_pages, zone, &arg);
936 nid = pfn_to_nid(pfn);
938 ret = memory_notify(MEM_GOING_ONLINE, &arg);
939 ret = notifier_to_errno(ret);
941 memory_notify(MEM_CANCEL_ONLINE, &arg);
942 unlock_memory_hotplug();
946 * If this zone is not populated, then it is not in zonelist.
947 * This means the page allocator ignores this zone.
948 * So, zonelist must be updated after online.
950 mutex_lock(&zonelists_mutex);
951 if (!populated_zone(zone)) {
952 need_zonelists_rebuild = 1;
953 build_all_zonelists(NULL, zone);
956 ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
959 if (need_zonelists_rebuild)
960 zone_pcp_reset(zone);
961 mutex_unlock(&zonelists_mutex);
962 printk(KERN_DEBUG "online_pages [mem %#010llx-%#010llx] failed\n",
963 (unsigned long long) pfn << PAGE_SHIFT,
964 (((unsigned long long) pfn + nr_pages)
966 memory_notify(MEM_CANCEL_ONLINE, &arg);
967 unlock_memory_hotplug();
971 zone->present_pages += onlined_pages;
973 pgdat_resize_lock(zone->zone_pgdat, &flags);
974 zone->zone_pgdat->node_present_pages += onlined_pages;
975 pgdat_resize_unlock(zone->zone_pgdat, &flags);
978 node_states_set_node(zone_to_nid(zone), &arg);
979 if (need_zonelists_rebuild)
980 build_all_zonelists(NULL, NULL);
982 zone_pcp_update(zone);
985 mutex_unlock(&zonelists_mutex);
987 init_per_zone_wmark_min();
990 kswapd_run(zone_to_nid(zone));
992 vm_total_pages = nr_free_pagecache_pages();
994 writeback_set_ratelimit();
997 memory_notify(MEM_ONLINE, &arg);
998 unlock_memory_hotplug();
1002 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
1004 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1005 static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
1007 struct pglist_data *pgdat;
1008 unsigned long zones_size[MAX_NR_ZONES] = {0};
1009 unsigned long zholes_size[MAX_NR_ZONES] = {0};
1010 unsigned long start_pfn = start >> PAGE_SHIFT;
1012 pgdat = NODE_DATA(nid);
1014 pgdat = arch_alloc_nodedata(nid);
1018 arch_refresh_nodedata(nid, pgdat);
1021 /* we can use NODE_DATA(nid) from here */
1023 /* init node's zones as empty zones, we don't have any present pages.*/
1024 free_area_init_node(nid, zones_size, start_pfn, zholes_size);
1027 * The node we allocated has no zone fallback lists. For avoiding
1028 * to access not-initialized zonelist, build here.
1030 mutex_lock(&zonelists_mutex);
1031 build_all_zonelists(pgdat, NULL);
1032 mutex_unlock(&zonelists_mutex);
1037 static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
1039 arch_refresh_nodedata(nid, NULL);
1040 arch_free_nodedata(pgdat);
1046 * try_online_node - online a node if offlined
1048 * called by cpu_up() to online a node without onlined memory.
1050 int try_online_node(int nid)
1055 if (node_online(nid))
1058 lock_memory_hotplug();
1059 pgdat = hotadd_new_pgdat(nid, 0);
1061 pr_err("Cannot online node %d due to NULL pgdat\n", nid);
1065 node_set_online(nid);
1066 ret = register_one_node(nid);
1069 if (pgdat->node_zonelists->_zonerefs->zone == NULL) {
1070 mutex_lock(&zonelists_mutex);
1071 build_all_zonelists(NULL, NULL);
1072 mutex_unlock(&zonelists_mutex);
1076 unlock_memory_hotplug();
1080 static int check_hotplug_memory_range(u64 start, u64 size)
1082 u64 start_pfn = start >> PAGE_SHIFT;
1083 u64 nr_pages = size >> PAGE_SHIFT;
1085 /* Memory range must be aligned with section */
1086 if ((start_pfn & ~PAGE_SECTION_MASK) ||
1087 (nr_pages % PAGES_PER_SECTION) || (!nr_pages)) {
1088 pr_err("Section-unaligned hotplug range: start 0x%llx, size 0x%llx\n",
1089 (unsigned long long)start,
1090 (unsigned long long)size);
1097 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1098 int __ref add_memory(int nid, u64 start, u64 size)
1100 pg_data_t *pgdat = NULL;
1103 struct resource *res;
1106 ret = check_hotplug_memory_range(start, size);
1110 res = register_memory_resource(start, size);
1115 { /* Stupid hack to suppress address-never-null warning */
1116 void *p = NODE_DATA(nid);
1120 lock_memory_hotplug();
1122 new_node = !node_online(nid);
1124 pgdat = hotadd_new_pgdat(nid, start);
1130 /* call arch's memory hotadd */
1131 ret = arch_add_memory(nid, start, size);
1136 /* we online node here. we can't roll back from here. */
1137 node_set_online(nid);
1140 ret = register_one_node(nid);
1142 * If sysfs file of new node can't create, cpu on the node
1143 * can't be hot-added. There is no rollback way now.
1144 * So, check by BUG_ON() to catch it reluctantly..
1149 /* create new memmap entry */
1150 firmware_map_add_hotplug(start, start + size, "System RAM");
1155 /* rollback pgdat allocation and others */
1157 rollback_node_hotadd(nid, pgdat);
1158 release_memory_resource(res);
1161 unlock_memory_hotplug();
1164 EXPORT_SYMBOL_GPL(add_memory);
1166 #ifdef CONFIG_MEMORY_HOTREMOVE
1168 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1169 * set and the size of the free page is given by page_order(). Using this,
1170 * the function determines if the pageblock contains only free pages.
1171 * Due to buddy contraints, a free page at least the size of a pageblock will
1172 * be located at the start of the pageblock
1174 static inline int pageblock_free(struct page *page)
1176 return PageBuddy(page) && page_order(page) >= pageblock_order;
1179 /* Return the start of the next active pageblock after a given page */
1180 static struct page *next_active_pageblock(struct page *page)
1182 /* Ensure the starting page is pageblock-aligned */
1183 BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
1185 /* If the entire pageblock is free, move to the end of free page */
1186 if (pageblock_free(page)) {
1188 /* be careful. we don't have locks, page_order can be changed.*/
1189 order = page_order(page);
1190 if ((order < MAX_ORDER) && (order >= pageblock_order))
1191 return page + (1 << order);
1194 return page + pageblock_nr_pages;
1197 /* Checks if this range of memory is likely to be hot-removable. */
1198 int is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
1200 struct page *page = pfn_to_page(start_pfn);
1201 struct page *end_page = page + nr_pages;
1203 /* Check the starting page of each pageblock within the range */
1204 for (; page < end_page; page = next_active_pageblock(page)) {
1205 if (!is_pageblock_removable_nolock(page))
1210 /* All pageblocks in the memory block are likely to be hot-removable */
1215 * Confirm all pages in a range [start, end) is belongs to the same zone.
1217 static int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn)
1220 struct zone *zone = NULL;
1223 for (pfn = start_pfn;
1225 pfn += MAX_ORDER_NR_PAGES) {
1227 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1228 while ((i < MAX_ORDER_NR_PAGES) && !pfn_valid_within(pfn + i))
1230 if (i == MAX_ORDER_NR_PAGES)
1232 page = pfn_to_page(pfn + i);
1233 if (zone && page_zone(page) != zone)
1235 zone = page_zone(page);
1241 * Scan pfn range [start,end) to find movable/migratable pages (LRU pages
1242 * and hugepages). We scan pfn because it's much easier than scanning over
1243 * linked list. This function returns the pfn of the first found movable
1244 * page if it's found, otherwise 0.
1246 static unsigned long scan_movable_pages(unsigned long start, unsigned long end)
1250 for (pfn = start; pfn < end; pfn++) {
1251 if (pfn_valid(pfn)) {
1252 page = pfn_to_page(pfn);
1255 if (PageHuge(page)) {
1256 if (is_hugepage_active(page))
1259 pfn = round_up(pfn + 1,
1260 1 << compound_order(page)) - 1;
1267 #define NR_OFFLINE_AT_ONCE_PAGES (256)
1269 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1273 int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
1274 int not_managed = 0;
1278 for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
1279 if (!pfn_valid(pfn))
1281 page = pfn_to_page(pfn);
1283 if (PageHuge(page)) {
1284 struct page *head = compound_head(page);
1285 pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1;
1286 if (compound_order(head) > PFN_SECTION_SHIFT) {
1290 if (isolate_huge_page(page, &source))
1291 move_pages -= 1 << compound_order(head);
1295 if (!get_page_unless_zero(page))
1298 * We can skip free pages. And we can only deal with pages on
1301 ret = isolate_lru_page(page);
1302 if (!ret) { /* Success */
1304 list_add_tail(&page->lru, &source);
1306 inc_zone_page_state(page, NR_ISOLATED_ANON +
1307 page_is_file_cache(page));
1310 #ifdef CONFIG_DEBUG_VM
1311 printk(KERN_ALERT "removing pfn %lx from LRU failed\n",
1313 dump_page(page, "failed to remove from LRU");
1316 /* Because we don't have big zone->lock. we should
1317 check this again here. */
1318 if (page_count(page)) {
1325 if (!list_empty(&source)) {
1327 putback_movable_pages(&source);
1332 * alloc_migrate_target should be improooooved!!
1333 * migrate_pages returns # of failed pages.
1335 ret = migrate_pages(&source, alloc_migrate_target, 0,
1336 MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
1338 putback_movable_pages(&source);
1345 * remove from free_area[] and mark all as Reserved.
1348 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1351 __offline_isolated_pages(start, start + nr_pages);
1356 offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
1358 walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
1359 offline_isolated_pages_cb);
1363 * Check all pages in range, recoreded as memory resource, are isolated.
1366 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1370 long offlined = *(long *)data;
1371 ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true);
1372 offlined = nr_pages;
1374 *(long *)data += offlined;
1379 check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
1384 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
1385 check_pages_isolated_cb);
1387 offlined = (long)ret;
1391 #ifdef CONFIG_MOVABLE_NODE
1393 * When CONFIG_MOVABLE_NODE, we permit offlining of a node which doesn't have
1396 static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1400 #else /* CONFIG_MOVABLE_NODE */
1401 /* ensure the node has NORMAL memory if it is still online */
1402 static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1404 struct pglist_data *pgdat = zone->zone_pgdat;
1405 unsigned long present_pages = 0;
1408 for (zt = 0; zt <= ZONE_NORMAL; zt++)
1409 present_pages += pgdat->node_zones[zt].present_pages;
1411 if (present_pages > nr_pages)
1415 for (; zt <= ZONE_MOVABLE; zt++)
1416 present_pages += pgdat->node_zones[zt].present_pages;
1419 * we can't offline the last normal memory until all
1420 * higher memory is offlined.
1422 return present_pages == 0;
1424 #endif /* CONFIG_MOVABLE_NODE */
1426 static int __init cmdline_parse_movable_node(char *p)
1428 #ifdef CONFIG_MOVABLE_NODE
1430 * Memory used by the kernel cannot be hot-removed because Linux
1431 * cannot migrate the kernel pages. When memory hotplug is
1432 * enabled, we should prevent memblock from allocating memory
1435 * ACPI SRAT records all hotpluggable memory ranges. But before
1436 * SRAT is parsed, we don't know about it.
1438 * The kernel image is loaded into memory at very early time. We
1439 * cannot prevent this anyway. So on NUMA system, we set any
1440 * node the kernel resides in as un-hotpluggable.
1442 * Since on modern servers, one node could have double-digit
1443 * gigabytes memory, we can assume the memory around the kernel
1444 * image is also un-hotpluggable. So before SRAT is parsed, just
1445 * allocate memory near the kernel image to try the best to keep
1446 * the kernel away from hotpluggable memory.
1448 memblock_set_bottom_up(true);
1449 movable_node_enabled = true;
1451 pr_warn("movable_node option not supported\n");
1455 early_param("movable_node", cmdline_parse_movable_node);
1457 /* check which state of node_states will be changed when offline memory */
1458 static void node_states_check_changes_offline(unsigned long nr_pages,
1459 struct zone *zone, struct memory_notify *arg)
1461 struct pglist_data *pgdat = zone->zone_pgdat;
1462 unsigned long present_pages = 0;
1463 enum zone_type zt, zone_last = ZONE_NORMAL;
1466 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
1467 * contains nodes which have zones of 0...ZONE_NORMAL,
1468 * set zone_last to ZONE_NORMAL.
1470 * If we don't have HIGHMEM nor movable node,
1471 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
1472 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
1474 if (N_MEMORY == N_NORMAL_MEMORY)
1475 zone_last = ZONE_MOVABLE;
1478 * check whether node_states[N_NORMAL_MEMORY] will be changed.
1479 * If the memory to be offline is in a zone of 0...zone_last,
1480 * and it is the last present memory, 0...zone_last will
1481 * become empty after offline , thus we can determind we will
1482 * need to clear the node from node_states[N_NORMAL_MEMORY].
1484 for (zt = 0; zt <= zone_last; zt++)
1485 present_pages += pgdat->node_zones[zt].present_pages;
1486 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1487 arg->status_change_nid_normal = zone_to_nid(zone);
1489 arg->status_change_nid_normal = -1;
1491 #ifdef CONFIG_HIGHMEM
1493 * If we have movable node, node_states[N_HIGH_MEMORY]
1494 * contains nodes which have zones of 0...ZONE_HIGHMEM,
1495 * set zone_last to ZONE_HIGHMEM.
1497 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
1498 * contains nodes which have zones of 0...ZONE_MOVABLE,
1499 * set zone_last to ZONE_MOVABLE.
1501 zone_last = ZONE_HIGHMEM;
1502 if (N_MEMORY == N_HIGH_MEMORY)
1503 zone_last = ZONE_MOVABLE;
1505 for (; zt <= zone_last; zt++)
1506 present_pages += pgdat->node_zones[zt].present_pages;
1507 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1508 arg->status_change_nid_high = zone_to_nid(zone);
1510 arg->status_change_nid_high = -1;
1512 arg->status_change_nid_high = arg->status_change_nid_normal;
1516 * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE
1518 zone_last = ZONE_MOVABLE;
1521 * check whether node_states[N_HIGH_MEMORY] will be changed
1522 * If we try to offline the last present @nr_pages from the node,
1523 * we can determind we will need to clear the node from
1524 * node_states[N_HIGH_MEMORY].
1526 for (; zt <= zone_last; zt++)
1527 present_pages += pgdat->node_zones[zt].present_pages;
1528 if (nr_pages >= present_pages)
1529 arg->status_change_nid = zone_to_nid(zone);
1531 arg->status_change_nid = -1;
1534 static void node_states_clear_node(int node, struct memory_notify *arg)
1536 if (arg->status_change_nid_normal >= 0)
1537 node_clear_state(node, N_NORMAL_MEMORY);
1539 if ((N_MEMORY != N_NORMAL_MEMORY) &&
1540 (arg->status_change_nid_high >= 0))
1541 node_clear_state(node, N_HIGH_MEMORY);
1543 if ((N_MEMORY != N_HIGH_MEMORY) &&
1544 (arg->status_change_nid >= 0))
1545 node_clear_state(node, N_MEMORY);
1548 static int __ref __offline_pages(unsigned long start_pfn,
1549 unsigned long end_pfn, unsigned long timeout)
1551 unsigned long pfn, nr_pages, expire;
1552 long offlined_pages;
1553 int ret, drain, retry_max, node;
1554 unsigned long flags;
1556 struct memory_notify arg;
1558 /* at least, alignment against pageblock is necessary */
1559 if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
1561 if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
1563 /* This makes hotplug much easier...and readable.
1564 we assume this for now. .*/
1565 if (!test_pages_in_a_zone(start_pfn, end_pfn))
1568 lock_memory_hotplug();
1570 zone = page_zone(pfn_to_page(start_pfn));
1571 node = zone_to_nid(zone);
1572 nr_pages = end_pfn - start_pfn;
1575 if (zone_idx(zone) <= ZONE_NORMAL && !can_offline_normal(zone, nr_pages))
1578 /* set above range as isolated */
1579 ret = start_isolate_page_range(start_pfn, end_pfn,
1580 MIGRATE_MOVABLE, true);
1584 arg.start_pfn = start_pfn;
1585 arg.nr_pages = nr_pages;
1586 node_states_check_changes_offline(nr_pages, zone, &arg);
1588 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1589 ret = notifier_to_errno(ret);
1591 goto failed_removal;
1594 expire = jiffies + timeout;
1598 /* start memory hot removal */
1600 if (time_after(jiffies, expire))
1601 goto failed_removal;
1603 if (signal_pending(current))
1604 goto failed_removal;
1607 lru_add_drain_all();
1612 pfn = scan_movable_pages(start_pfn, end_pfn);
1613 if (pfn) { /* We have movable pages */
1614 ret = do_migrate_range(pfn, end_pfn);
1620 if (--retry_max == 0)
1621 goto failed_removal;
1627 /* drain all zone's lru pagevec, this is asynchronous... */
1628 lru_add_drain_all();
1630 /* drain pcp pages, this is synchronous. */
1633 * dissolve free hugepages in the memory block before doing offlining
1634 * actually in order to make hugetlbfs's object counting consistent.
1636 dissolve_free_huge_pages(start_pfn, end_pfn);
1638 offlined_pages = check_pages_isolated(start_pfn, end_pfn);
1639 if (offlined_pages < 0) {
1641 goto failed_removal;
1643 printk(KERN_INFO "Offlined Pages %ld\n", offlined_pages);
1644 /* Ok, all of our target is isolated.
1645 We cannot do rollback at this point. */
1646 offline_isolated_pages(start_pfn, end_pfn);
1647 /* reset pagetype flags and makes migrate type to be MOVABLE */
1648 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1649 /* removal success */
1650 adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages);
1651 zone->present_pages -= offlined_pages;
1653 pgdat_resize_lock(zone->zone_pgdat, &flags);
1654 zone->zone_pgdat->node_present_pages -= offlined_pages;
1655 pgdat_resize_unlock(zone->zone_pgdat, &flags);
1657 init_per_zone_wmark_min();
1659 if (!populated_zone(zone)) {
1660 zone_pcp_reset(zone);
1661 mutex_lock(&zonelists_mutex);
1662 build_all_zonelists(NULL, NULL);
1663 mutex_unlock(&zonelists_mutex);
1665 zone_pcp_update(zone);
1667 node_states_clear_node(node, &arg);
1668 if (arg.status_change_nid >= 0)
1671 vm_total_pages = nr_free_pagecache_pages();
1672 writeback_set_ratelimit();
1674 memory_notify(MEM_OFFLINE, &arg);
1675 unlock_memory_hotplug();
1679 printk(KERN_INFO "memory offlining [mem %#010llx-%#010llx] failed\n",
1680 (unsigned long long) start_pfn << PAGE_SHIFT,
1681 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
1682 memory_notify(MEM_CANCEL_OFFLINE, &arg);
1683 /* pushback to free area */
1684 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1687 unlock_memory_hotplug();
1691 int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1693 return __offline_pages(start_pfn, start_pfn + nr_pages, 120 * HZ);
1695 #endif /* CONFIG_MEMORY_HOTREMOVE */
1698 * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
1699 * @start_pfn: start pfn of the memory range
1700 * @end_pfn: end pfn of the memory range
1701 * @arg: argument passed to func
1702 * @func: callback for each memory section walked
1704 * This function walks through all present mem sections in range
1705 * [start_pfn, end_pfn) and call func on each mem section.
1707 * Returns the return value of func.
1709 int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn,
1710 void *arg, int (*func)(struct memory_block *, void *))
1712 struct memory_block *mem = NULL;
1713 struct mem_section *section;
1714 unsigned long pfn, section_nr;
1717 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1718 section_nr = pfn_to_section_nr(pfn);
1719 if (!present_section_nr(section_nr))
1722 section = __nr_to_section(section_nr);
1723 /* same memblock? */
1725 if ((section_nr >= mem->start_section_nr) &&
1726 (section_nr <= mem->end_section_nr))
1729 mem = find_memory_block_hinted(section, mem);
1733 ret = func(mem, arg);
1735 kobject_put(&mem->dev.kobj);
1741 kobject_put(&mem->dev.kobj);
1746 #ifdef CONFIG_MEMORY_HOTREMOVE
1747 static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
1749 int ret = !is_memblock_offlined(mem);
1751 if (unlikely(ret)) {
1752 phys_addr_t beginpa, endpa;
1754 beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
1755 endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1;
1756 pr_warn("removing memory fails, because memory "
1757 "[%pa-%pa] is onlined\n",
1764 static int check_cpu_on_node(pg_data_t *pgdat)
1768 for_each_present_cpu(cpu) {
1769 if (cpu_to_node(cpu) == pgdat->node_id)
1771 * the cpu on this node isn't removed, and we can't
1772 * offline this node.
1780 static void unmap_cpu_on_node(pg_data_t *pgdat)
1782 #ifdef CONFIG_ACPI_NUMA
1785 for_each_possible_cpu(cpu)
1786 if (cpu_to_node(cpu) == pgdat->node_id)
1787 numa_clear_node(cpu);
1791 static int check_and_unmap_cpu_on_node(pg_data_t *pgdat)
1795 ret = check_cpu_on_node(pgdat);
1800 * the node will be offlined when we come here, so we can clear
1801 * the cpu_to_node() now.
1804 unmap_cpu_on_node(pgdat);
1811 * Offline a node if all memory sections and cpus of the node are removed.
1813 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1814 * and online/offline operations before this call.
1816 void try_offline_node(int nid)
1818 pg_data_t *pgdat = NODE_DATA(nid);
1819 unsigned long start_pfn = pgdat->node_start_pfn;
1820 unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
1822 struct page *pgdat_page = virt_to_page(pgdat);
1825 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1826 unsigned long section_nr = pfn_to_section_nr(pfn);
1828 if (!present_section_nr(section_nr))
1831 if (pfn_to_nid(pfn) != nid)
1835 * some memory sections of this node are not removed, and we
1836 * can't offline node now.
1841 if (check_and_unmap_cpu_on_node(pgdat))
1845 * all memory/cpu of this node are removed, we can offline this
1848 node_set_offline(nid);
1849 unregister_one_node(nid);
1851 if (!PageSlab(pgdat_page) && !PageCompound(pgdat_page))
1852 /* node data is allocated from boot memory */
1855 /* free waittable in each zone */
1856 for (i = 0; i < MAX_NR_ZONES; i++) {
1857 struct zone *zone = pgdat->node_zones + i;
1860 * wait_table may be allocated from boot memory,
1861 * here only free if it's allocated by vmalloc.
1863 if (is_vmalloc_addr(zone->wait_table))
1864 vfree(zone->wait_table);
1868 * Since there is no way to guarentee the address of pgdat/zone is not
1869 * on stack of any kernel threads or used by other kernel objects
1870 * without reference counting or other symchronizing method, do not
1871 * reset node_data and free pgdat here. Just reset it to 0 and reuse
1872 * the memory when the node is online again.
1874 memset(pgdat, 0, sizeof(*pgdat));
1876 EXPORT_SYMBOL(try_offline_node);
1881 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1882 * and online/offline operations before this call, as required by
1883 * try_offline_node().
1885 void __ref remove_memory(int nid, u64 start, u64 size)
1889 BUG_ON(check_hotplug_memory_range(start, size));
1891 lock_memory_hotplug();
1894 * All memory blocks must be offlined before removing memory. Check
1895 * whether all memory blocks in question are offline and trigger a BUG()
1896 * if this is not the case.
1898 ret = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL,
1899 check_memblock_offlined_cb);
1901 unlock_memory_hotplug();
1905 /* remove memmap entry */
1906 firmware_map_remove(start, start + size, "System RAM");
1908 arch_remove_memory(start, size);
1910 try_offline_node(nid);
1912 unlock_memory_hotplug();
1914 EXPORT_SYMBOL_GPL(remove_memory);
1915 #endif /* CONFIG_MEMORY_HOTREMOVE */