2 * linux/mm/memory_hotplug.c
7 #include <linux/stddef.h>
9 #include <linux/sched/signal.h>
10 #include <linux/swap.h>
11 #include <linux/interrupt.h>
12 #include <linux/pagemap.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/memremap.h>
22 #include <linux/memory_hotplug.h>
23 #include <linux/highmem.h>
24 #include <linux/vmalloc.h>
25 #include <linux/ioport.h>
26 #include <linux/delay.h>
27 #include <linux/migrate.h>
28 #include <linux/page-isolation.h>
29 #include <linux/pfn.h>
30 #include <linux/suspend.h>
31 #include <linux/mm_inline.h>
32 #include <linux/firmware-map.h>
33 #include <linux/stop_machine.h>
34 #include <linux/hugetlb.h>
35 #include <linux/memblock.h>
36 #include <linux/bootmem.h>
37 #include <linux/compaction.h>
38 #include <linux/rmap.h>
40 #include <asm/tlbflush.h>
45 * online_page_callback contains pointer to current page onlining function.
46 * Initially it is generic_online_page(). If it is required it could be
47 * changed by calling set_online_page_callback() for callback registration
48 * and restore_online_page_callback() for generic callback restore.
51 static void generic_online_page(struct page *page);
53 static online_page_callback_t online_page_callback = generic_online_page;
54 static DEFINE_MUTEX(online_page_callback_lock);
56 DEFINE_STATIC_PERCPU_RWSEM(mem_hotplug_lock);
58 void get_online_mems(void)
60 percpu_down_read(&mem_hotplug_lock);
63 void put_online_mems(void)
65 percpu_up_read(&mem_hotplug_lock);
68 bool movable_node_enabled = false;
70 #ifndef CONFIG_MEMORY_HOTPLUG_DEFAULT_ONLINE
71 bool memhp_auto_online;
73 bool memhp_auto_online = true;
75 EXPORT_SYMBOL_GPL(memhp_auto_online);
77 static int __init setup_memhp_default_state(char *str)
79 if (!strcmp(str, "online"))
80 memhp_auto_online = true;
81 else if (!strcmp(str, "offline"))
82 memhp_auto_online = false;
86 __setup("memhp_default_state=", setup_memhp_default_state);
88 void mem_hotplug_begin(void)
91 percpu_down_write(&mem_hotplug_lock);
94 void mem_hotplug_done(void)
96 percpu_up_write(&mem_hotplug_lock);
100 /* add this memory to iomem resource */
101 static struct resource *register_memory_resource(u64 start, u64 size)
103 struct resource *res, *conflict;
104 res = kzalloc(sizeof(struct resource), GFP_KERNEL);
106 return ERR_PTR(-ENOMEM);
108 res->name = "System RAM";
110 res->end = start + size - 1;
111 res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
112 conflict = request_resource_conflict(&iomem_resource, res);
114 if (conflict->desc == IORES_DESC_DEVICE_PRIVATE_MEMORY) {
115 pr_debug("Device unaddressable memory block "
116 "memory hotplug at %#010llx !\n",
117 (unsigned long long)start);
119 pr_debug("System RAM resource %pR cannot be added\n", res);
121 return ERR_PTR(-EEXIST);
126 static void release_memory_resource(struct resource *res)
130 release_resource(res);
135 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
136 void get_page_bootmem(unsigned long info, struct page *page,
139 page->freelist = (void *)type;
140 SetPagePrivate(page);
141 set_page_private(page, info);
145 void put_page_bootmem(struct page *page)
149 type = (unsigned long) page->freelist;
150 BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE ||
151 type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE);
153 if (page_ref_dec_return(page) == 1) {
154 page->freelist = NULL;
155 ClearPagePrivate(page);
156 set_page_private(page, 0);
157 INIT_LIST_HEAD(&page->lru);
158 free_reserved_page(page);
162 #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
163 #ifndef CONFIG_SPARSEMEM_VMEMMAP
164 static void register_page_bootmem_info_section(unsigned long start_pfn)
166 unsigned long *usemap, mapsize, section_nr, i;
167 struct mem_section *ms;
168 struct page *page, *memmap;
170 section_nr = pfn_to_section_nr(start_pfn);
171 ms = __nr_to_section(section_nr);
173 /* Get section's memmap address */
174 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
177 * Get page for the memmap's phys address
178 * XXX: need more consideration for sparse_vmemmap...
180 page = virt_to_page(memmap);
181 mapsize = sizeof(struct page) * PAGES_PER_SECTION;
182 mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
184 /* remember memmap's page */
185 for (i = 0; i < mapsize; i++, page++)
186 get_page_bootmem(section_nr, page, SECTION_INFO);
188 usemap = __nr_to_section(section_nr)->pageblock_flags;
189 page = virt_to_page(usemap);
191 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
193 for (i = 0; i < mapsize; i++, page++)
194 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
197 #else /* CONFIG_SPARSEMEM_VMEMMAP */
198 static void register_page_bootmem_info_section(unsigned long start_pfn)
200 unsigned long *usemap, mapsize, section_nr, i;
201 struct mem_section *ms;
202 struct page *page, *memmap;
204 if (!pfn_valid(start_pfn))
207 section_nr = pfn_to_section_nr(start_pfn);
208 ms = __nr_to_section(section_nr);
210 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
212 register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION);
214 usemap = __nr_to_section(section_nr)->pageblock_flags;
215 page = virt_to_page(usemap);
217 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
219 for (i = 0; i < mapsize; i++, page++)
220 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
222 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
224 void __init register_page_bootmem_info_node(struct pglist_data *pgdat)
226 unsigned long i, pfn, end_pfn, nr_pages;
227 int node = pgdat->node_id;
230 nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
231 page = virt_to_page(pgdat);
233 for (i = 0; i < nr_pages; i++, page++)
234 get_page_bootmem(node, page, NODE_INFO);
236 pfn = pgdat->node_start_pfn;
237 end_pfn = pgdat_end_pfn(pgdat);
239 /* register section info */
240 for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
242 * Some platforms can assign the same pfn to multiple nodes - on
243 * node0 as well as nodeN. To avoid registering a pfn against
244 * multiple nodes we check that this pfn does not already
245 * reside in some other nodes.
247 if (pfn_valid(pfn) && (early_pfn_to_nid(pfn) == node))
248 register_page_bootmem_info_section(pfn);
251 #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
253 static int __meminit __add_section(int nid, unsigned long phys_start_pfn,
259 if (pfn_valid(phys_start_pfn))
262 ret = sparse_add_one_section(NODE_DATA(nid), phys_start_pfn);
267 * Make all the pages reserved so that nobody will stumble over half
269 * FIXME: We also have to associate it with a node because pfn_to_node
270 * relies on having page with the proper node.
272 for (i = 0; i < PAGES_PER_SECTION; i++) {
273 unsigned long pfn = phys_start_pfn + i;
278 page = pfn_to_page(pfn);
279 set_page_node(page, nid);
280 SetPageReserved(page);
286 return register_new_memory(nid, __pfn_to_section(phys_start_pfn));
290 * Reasonably generic function for adding memory. It is
291 * expected that archs that support memory hotplug will
292 * call this function after deciding the zone to which to
295 int __ref __add_pages(int nid, unsigned long phys_start_pfn,
296 unsigned long nr_pages, bool want_memblock)
300 int start_sec, end_sec;
301 struct vmem_altmap *altmap;
303 /* during initialize mem_map, align hot-added range to section */
304 start_sec = pfn_to_section_nr(phys_start_pfn);
305 end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
307 altmap = to_vmem_altmap((unsigned long) pfn_to_page(phys_start_pfn));
310 * Validate altmap is within bounds of the total request
312 if (altmap->base_pfn != phys_start_pfn
313 || vmem_altmap_offset(altmap) > nr_pages) {
314 pr_warn_once("memory add fail, invalid altmap\n");
321 for (i = start_sec; i <= end_sec; i++) {
322 err = __add_section(nid, section_nr_to_pfn(i), want_memblock);
325 * EEXIST is finally dealt with by ioresource collision
326 * check. see add_memory() => register_memory_resource()
327 * Warning will be printed if there is collision.
329 if (err && (err != -EEXIST))
334 vmemmap_populate_print_last();
338 EXPORT_SYMBOL_GPL(__add_pages);
340 #ifdef CONFIG_MEMORY_HOTREMOVE
341 /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
342 static unsigned long find_smallest_section_pfn(int nid, struct zone *zone,
343 unsigned long start_pfn,
344 unsigned long end_pfn)
346 struct mem_section *ms;
348 for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) {
349 ms = __pfn_to_section(start_pfn);
351 if (unlikely(!valid_section(ms)))
354 if (unlikely(pfn_to_nid(start_pfn) != nid))
357 if (zone && zone != page_zone(pfn_to_page(start_pfn)))
366 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
367 static unsigned long find_biggest_section_pfn(int nid, struct zone *zone,
368 unsigned long start_pfn,
369 unsigned long end_pfn)
371 struct mem_section *ms;
374 /* pfn is the end pfn of a memory section. */
376 for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) {
377 ms = __pfn_to_section(pfn);
379 if (unlikely(!valid_section(ms)))
382 if (unlikely(pfn_to_nid(pfn) != nid))
385 if (zone && zone != page_zone(pfn_to_page(pfn)))
394 static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
395 unsigned long end_pfn)
397 unsigned long zone_start_pfn = zone->zone_start_pfn;
398 unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */
399 unsigned long zone_end_pfn = z;
401 struct mem_section *ms;
402 int nid = zone_to_nid(zone);
404 zone_span_writelock(zone);
405 if (zone_start_pfn == start_pfn) {
407 * If the section is smallest section in the zone, it need
408 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
409 * In this case, we find second smallest valid mem_section
410 * for shrinking zone.
412 pfn = find_smallest_section_pfn(nid, zone, end_pfn,
415 zone->zone_start_pfn = pfn;
416 zone->spanned_pages = zone_end_pfn - pfn;
418 } else if (zone_end_pfn == end_pfn) {
420 * If the section is biggest section in the zone, it need
421 * shrink zone->spanned_pages.
422 * In this case, we find second biggest valid mem_section for
425 pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn,
428 zone->spanned_pages = pfn - zone_start_pfn + 1;
432 * The section is not biggest or smallest mem_section in the zone, it
433 * only creates a hole in the zone. So in this case, we need not
434 * change the zone. But perhaps, the zone has only hole data. Thus
435 * it check the zone has only hole or not.
437 pfn = zone_start_pfn;
438 for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) {
439 ms = __pfn_to_section(pfn);
441 if (unlikely(!valid_section(ms)))
444 if (page_zone(pfn_to_page(pfn)) != zone)
447 /* If the section is current section, it continues the loop */
448 if (start_pfn == pfn)
451 /* If we find valid section, we have nothing to do */
452 zone_span_writeunlock(zone);
456 /* The zone has no valid section */
457 zone->zone_start_pfn = 0;
458 zone->spanned_pages = 0;
459 zone_span_writeunlock(zone);
462 static void shrink_pgdat_span(struct pglist_data *pgdat,
463 unsigned long start_pfn, unsigned long end_pfn)
465 unsigned long pgdat_start_pfn = pgdat->node_start_pfn;
466 unsigned long p = pgdat_end_pfn(pgdat); /* pgdat_end_pfn namespace clash */
467 unsigned long pgdat_end_pfn = p;
469 struct mem_section *ms;
470 int nid = pgdat->node_id;
472 if (pgdat_start_pfn == start_pfn) {
474 * If the section is smallest section in the pgdat, it need
475 * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages.
476 * In this case, we find second smallest valid mem_section
477 * for shrinking zone.
479 pfn = find_smallest_section_pfn(nid, NULL, end_pfn,
482 pgdat->node_start_pfn = pfn;
483 pgdat->node_spanned_pages = pgdat_end_pfn - pfn;
485 } else if (pgdat_end_pfn == end_pfn) {
487 * If the section is biggest section in the pgdat, it need
488 * shrink pgdat->node_spanned_pages.
489 * In this case, we find second biggest valid mem_section for
492 pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn,
495 pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1;
499 * If the section is not biggest or smallest mem_section in the pgdat,
500 * it only creates a hole in the pgdat. So in this case, we need not
502 * But perhaps, the pgdat has only hole data. Thus it check the pgdat
503 * has only hole or not.
505 pfn = pgdat_start_pfn;
506 for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) {
507 ms = __pfn_to_section(pfn);
509 if (unlikely(!valid_section(ms)))
512 if (pfn_to_nid(pfn) != nid)
515 /* If the section is current section, it continues the loop */
516 if (start_pfn == pfn)
519 /* If we find valid section, we have nothing to do */
523 /* The pgdat has no valid section */
524 pgdat->node_start_pfn = 0;
525 pgdat->node_spanned_pages = 0;
528 static void __remove_zone(struct zone *zone, unsigned long start_pfn)
530 struct pglist_data *pgdat = zone->zone_pgdat;
531 int nr_pages = PAGES_PER_SECTION;
534 pgdat_resize_lock(zone->zone_pgdat, &flags);
535 shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
536 shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages);
537 pgdat_resize_unlock(zone->zone_pgdat, &flags);
540 static int __remove_section(struct zone *zone, struct mem_section *ms,
541 unsigned long map_offset)
543 unsigned long start_pfn;
547 if (!valid_section(ms))
550 ret = unregister_memory_section(ms);
554 scn_nr = __section_nr(ms);
555 start_pfn = section_nr_to_pfn((unsigned long)scn_nr);
556 __remove_zone(zone, start_pfn);
558 sparse_remove_one_section(zone, ms, map_offset);
563 * __remove_pages() - remove sections of pages from a zone
564 * @zone: zone from which pages need to be removed
565 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
566 * @nr_pages: number of pages to remove (must be multiple of section size)
568 * Generic helper function to remove section mappings and sysfs entries
569 * for the section of the memory we are removing. Caller needs to make
570 * sure that pages are marked reserved and zones are adjust properly by
571 * calling offline_pages().
573 int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
574 unsigned long nr_pages)
577 unsigned long map_offset = 0;
578 int sections_to_remove, ret = 0;
580 /* In the ZONE_DEVICE case device driver owns the memory region */
581 if (is_dev_zone(zone)) {
582 struct page *page = pfn_to_page(phys_start_pfn);
583 struct vmem_altmap *altmap;
585 altmap = to_vmem_altmap((unsigned long) page);
587 map_offset = vmem_altmap_offset(altmap);
589 resource_size_t start, size;
591 start = phys_start_pfn << PAGE_SHIFT;
592 size = nr_pages * PAGE_SIZE;
594 ret = release_mem_region_adjustable(&iomem_resource, start,
597 resource_size_t endres = start + size - 1;
599 pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
600 &start, &endres, ret);
604 clear_zone_contiguous(zone);
607 * We can only remove entire sections
609 BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
610 BUG_ON(nr_pages % PAGES_PER_SECTION);
612 sections_to_remove = nr_pages / PAGES_PER_SECTION;
613 for (i = 0; i < sections_to_remove; i++) {
614 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
616 ret = __remove_section(zone, __pfn_to_section(pfn), map_offset);
622 set_zone_contiguous(zone);
626 #endif /* CONFIG_MEMORY_HOTREMOVE */
628 int set_online_page_callback(online_page_callback_t callback)
633 mutex_lock(&online_page_callback_lock);
635 if (online_page_callback == generic_online_page) {
636 online_page_callback = callback;
640 mutex_unlock(&online_page_callback_lock);
645 EXPORT_SYMBOL_GPL(set_online_page_callback);
647 int restore_online_page_callback(online_page_callback_t callback)
652 mutex_lock(&online_page_callback_lock);
654 if (online_page_callback == callback) {
655 online_page_callback = generic_online_page;
659 mutex_unlock(&online_page_callback_lock);
664 EXPORT_SYMBOL_GPL(restore_online_page_callback);
666 void __online_page_set_limits(struct page *page)
669 EXPORT_SYMBOL_GPL(__online_page_set_limits);
671 void __online_page_increment_counters(struct page *page)
673 adjust_managed_page_count(page, 1);
675 EXPORT_SYMBOL_GPL(__online_page_increment_counters);
677 void __online_page_free(struct page *page)
679 __free_reserved_page(page);
681 EXPORT_SYMBOL_GPL(__online_page_free);
683 static void generic_online_page(struct page *page)
685 __online_page_set_limits(page);
686 __online_page_increment_counters(page);
687 __online_page_free(page);
690 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
694 unsigned long onlined_pages = *(unsigned long *)arg;
697 if (PageReserved(pfn_to_page(start_pfn)))
698 for (i = 0; i < nr_pages; i++) {
699 page = pfn_to_page(start_pfn + i);
700 (*online_page_callback)(page);
704 online_mem_sections(start_pfn, start_pfn + nr_pages);
706 *(unsigned long *)arg = onlined_pages;
710 /* check which state of node_states will be changed when online memory */
711 static void node_states_check_changes_online(unsigned long nr_pages,
712 struct zone *zone, struct memory_notify *arg)
714 int nid = zone_to_nid(zone);
715 enum zone_type zone_last = ZONE_NORMAL;
718 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
719 * contains nodes which have zones of 0...ZONE_NORMAL,
720 * set zone_last to ZONE_NORMAL.
722 * If we don't have HIGHMEM nor movable node,
723 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
724 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
726 if (N_MEMORY == N_NORMAL_MEMORY)
727 zone_last = ZONE_MOVABLE;
730 * if the memory to be online is in a zone of 0...zone_last, and
731 * the zones of 0...zone_last don't have memory before online, we will
732 * need to set the node to node_states[N_NORMAL_MEMORY] after
733 * the memory is online.
735 if (zone_idx(zone) <= zone_last && !node_state(nid, N_NORMAL_MEMORY))
736 arg->status_change_nid_normal = nid;
738 arg->status_change_nid_normal = -1;
740 #ifdef CONFIG_HIGHMEM
742 * If we have movable node, node_states[N_HIGH_MEMORY]
743 * contains nodes which have zones of 0...ZONE_HIGHMEM,
744 * set zone_last to ZONE_HIGHMEM.
746 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
747 * contains nodes which have zones of 0...ZONE_MOVABLE,
748 * set zone_last to ZONE_MOVABLE.
750 zone_last = ZONE_HIGHMEM;
751 if (N_MEMORY == N_HIGH_MEMORY)
752 zone_last = ZONE_MOVABLE;
754 if (zone_idx(zone) <= zone_last && !node_state(nid, N_HIGH_MEMORY))
755 arg->status_change_nid_high = nid;
757 arg->status_change_nid_high = -1;
759 arg->status_change_nid_high = arg->status_change_nid_normal;
763 * if the node don't have memory befor online, we will need to
764 * set the node to node_states[N_MEMORY] after the memory
767 if (!node_state(nid, N_MEMORY))
768 arg->status_change_nid = nid;
770 arg->status_change_nid = -1;
773 static void node_states_set_node(int node, struct memory_notify *arg)
775 if (arg->status_change_nid_normal >= 0)
776 node_set_state(node, N_NORMAL_MEMORY);
778 if (arg->status_change_nid_high >= 0)
779 node_set_state(node, N_HIGH_MEMORY);
781 node_set_state(node, N_MEMORY);
784 static void __meminit resize_zone_range(struct zone *zone, unsigned long start_pfn,
785 unsigned long nr_pages)
787 unsigned long old_end_pfn = zone_end_pfn(zone);
789 if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
790 zone->zone_start_pfn = start_pfn;
792 zone->spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - zone->zone_start_pfn;
795 static void __meminit resize_pgdat_range(struct pglist_data *pgdat, unsigned long start_pfn,
796 unsigned long nr_pages)
798 unsigned long old_end_pfn = pgdat_end_pfn(pgdat);
800 if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
801 pgdat->node_start_pfn = start_pfn;
803 pgdat->node_spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - pgdat->node_start_pfn;
806 void __ref move_pfn_range_to_zone(struct zone *zone,
807 unsigned long start_pfn, unsigned long nr_pages)
809 struct pglist_data *pgdat = zone->zone_pgdat;
810 int nid = pgdat->node_id;
813 if (zone_is_empty(zone))
814 init_currently_empty_zone(zone, start_pfn, nr_pages);
816 clear_zone_contiguous(zone);
818 /* TODO Huh pgdat is irqsave while zone is not. It used to be like that before */
819 pgdat_resize_lock(pgdat, &flags);
820 zone_span_writelock(zone);
821 resize_zone_range(zone, start_pfn, nr_pages);
822 zone_span_writeunlock(zone);
823 resize_pgdat_range(pgdat, start_pfn, nr_pages);
824 pgdat_resize_unlock(pgdat, &flags);
827 * TODO now we have a visible range of pages which are not associated
828 * with their zone properly. Not nice but set_pfnblock_flags_mask
829 * expects the zone spans the pfn range. All the pages in the range
830 * are reserved so nobody should be touching them so we should be safe
832 memmap_init_zone(nr_pages, nid, zone_idx(zone), start_pfn, MEMMAP_HOTPLUG);
834 set_zone_contiguous(zone);
838 * Returns a default kernel memory zone for the given pfn range.
839 * If no kernel zone covers this pfn range it will automatically go
840 * to the ZONE_NORMAL.
842 static struct zone *default_kernel_zone_for_pfn(int nid, unsigned long start_pfn,
843 unsigned long nr_pages)
845 struct pglist_data *pgdat = NODE_DATA(nid);
848 for (zid = 0; zid <= ZONE_NORMAL; zid++) {
849 struct zone *zone = &pgdat->node_zones[zid];
851 if (zone_intersects(zone, start_pfn, nr_pages))
855 return &pgdat->node_zones[ZONE_NORMAL];
858 static inline struct zone *default_zone_for_pfn(int nid, unsigned long start_pfn,
859 unsigned long nr_pages)
861 struct zone *kernel_zone = default_kernel_zone_for_pfn(nid, start_pfn,
863 struct zone *movable_zone = &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
864 bool in_kernel = zone_intersects(kernel_zone, start_pfn, nr_pages);
865 bool in_movable = zone_intersects(movable_zone, start_pfn, nr_pages);
868 * We inherit the existing zone in a simple case where zones do not
869 * overlap in the given range
871 if (in_kernel ^ in_movable)
872 return (in_kernel) ? kernel_zone : movable_zone;
875 * If the range doesn't belong to any zone or two zones overlap in the
876 * given range then we use movable zone only if movable_node is
877 * enabled because we always online to a kernel zone by default.
879 return movable_node_enabled ? movable_zone : kernel_zone;
882 struct zone * zone_for_pfn_range(int online_type, int nid, unsigned start_pfn,
883 unsigned long nr_pages)
885 if (online_type == MMOP_ONLINE_KERNEL)
886 return default_kernel_zone_for_pfn(nid, start_pfn, nr_pages);
888 if (online_type == MMOP_ONLINE_MOVABLE)
889 return &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
891 return default_zone_for_pfn(nid, start_pfn, nr_pages);
895 * Associates the given pfn range with the given node and the zone appropriate
896 * for the given online type.
898 static struct zone * __meminit move_pfn_range(int online_type, int nid,
899 unsigned long start_pfn, unsigned long nr_pages)
903 zone = zone_for_pfn_range(online_type, nid, start_pfn, nr_pages);
904 move_pfn_range_to_zone(zone, start_pfn, nr_pages);
908 /* Must be protected by mem_hotplug_begin() or a device_lock */
909 int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type)
912 unsigned long onlined_pages = 0;
914 int need_zonelists_rebuild = 0;
917 struct memory_notify arg;
919 nid = pfn_to_nid(pfn);
920 /* associate pfn range with the zone */
921 zone = move_pfn_range(online_type, nid, pfn, nr_pages);
924 arg.nr_pages = nr_pages;
925 node_states_check_changes_online(nr_pages, zone, &arg);
927 ret = memory_notify(MEM_GOING_ONLINE, &arg);
928 ret = notifier_to_errno(ret);
930 goto failed_addition;
933 * If this zone is not populated, then it is not in zonelist.
934 * This means the page allocator ignores this zone.
935 * So, zonelist must be updated after online.
937 if (!populated_zone(zone)) {
938 need_zonelists_rebuild = 1;
939 setup_zone_pageset(zone);
942 ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
945 if (need_zonelists_rebuild)
946 zone_pcp_reset(zone);
947 goto failed_addition;
950 zone->present_pages += onlined_pages;
952 pgdat_resize_lock(zone->zone_pgdat, &flags);
953 zone->zone_pgdat->node_present_pages += onlined_pages;
954 pgdat_resize_unlock(zone->zone_pgdat, &flags);
957 node_states_set_node(nid, &arg);
958 if (need_zonelists_rebuild)
959 build_all_zonelists(NULL);
961 zone_pcp_update(zone);
964 init_per_zone_wmark_min();
971 vm_total_pages = nr_free_pagecache_pages();
973 writeback_set_ratelimit();
976 memory_notify(MEM_ONLINE, &arg);
980 pr_debug("online_pages [mem %#010llx-%#010llx] failed\n",
981 (unsigned long long) pfn << PAGE_SHIFT,
982 (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1);
983 memory_notify(MEM_CANCEL_ONLINE, &arg);
986 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
988 static void reset_node_present_pages(pg_data_t *pgdat)
992 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
993 z->present_pages = 0;
995 pgdat->node_present_pages = 0;
998 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
999 static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
1001 struct pglist_data *pgdat;
1002 unsigned long zones_size[MAX_NR_ZONES] = {0};
1003 unsigned long zholes_size[MAX_NR_ZONES] = {0};
1004 unsigned long start_pfn = PFN_DOWN(start);
1006 pgdat = NODE_DATA(nid);
1008 pgdat = arch_alloc_nodedata(nid);
1012 arch_refresh_nodedata(nid, pgdat);
1015 * Reset the nr_zones, order and classzone_idx before reuse.
1016 * Note that kswapd will init kswapd_classzone_idx properly
1017 * when it starts in the near future.
1019 pgdat->nr_zones = 0;
1020 pgdat->kswapd_order = 0;
1021 pgdat->kswapd_classzone_idx = 0;
1024 /* we can use NODE_DATA(nid) from here */
1026 /* init node's zones as empty zones, we don't have any present pages.*/
1027 free_area_init_node(nid, zones_size, start_pfn, zholes_size);
1028 pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat);
1031 * The node we allocated has no zone fallback lists. For avoiding
1032 * to access not-initialized zonelist, build here.
1034 build_all_zonelists(pgdat);
1037 * zone->managed_pages is set to an approximate value in
1038 * free_area_init_core(), which will cause
1039 * /sys/device/system/node/nodeX/meminfo has wrong data.
1040 * So reset it to 0 before any memory is onlined.
1042 reset_node_managed_pages(pgdat);
1045 * When memory is hot-added, all the memory is in offline state. So
1046 * clear all zones' present_pages because they will be updated in
1047 * online_pages() and offline_pages().
1049 reset_node_present_pages(pgdat);
1054 static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
1056 arch_refresh_nodedata(nid, NULL);
1057 free_percpu(pgdat->per_cpu_nodestats);
1058 arch_free_nodedata(pgdat);
1064 * try_online_node - online a node if offlined
1066 * called by cpu_up() to online a node without onlined memory.
1068 int try_online_node(int nid)
1073 if (node_online(nid))
1076 mem_hotplug_begin();
1077 pgdat = hotadd_new_pgdat(nid, 0);
1079 pr_err("Cannot online node %d due to NULL pgdat\n", nid);
1083 node_set_online(nid);
1084 ret = register_one_node(nid);
1091 static int check_hotplug_memory_range(u64 start, u64 size)
1093 u64 start_pfn = PFN_DOWN(start);
1094 u64 nr_pages = size >> PAGE_SHIFT;
1096 /* Memory range must be aligned with section */
1097 if ((start_pfn & ~PAGE_SECTION_MASK) ||
1098 (nr_pages % PAGES_PER_SECTION) || (!nr_pages)) {
1099 pr_err("Section-unaligned hotplug range: start 0x%llx, size 0x%llx\n",
1100 (unsigned long long)start,
1101 (unsigned long long)size);
1108 static int online_memory_block(struct memory_block *mem, void *arg)
1110 return device_online(&mem->dev);
1113 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1114 int __ref add_memory_resource(int nid, struct resource *res, bool online)
1117 pg_data_t *pgdat = NULL;
1123 size = resource_size(res);
1125 ret = check_hotplug_memory_range(start, size);
1129 { /* Stupid hack to suppress address-never-null warning */
1130 void *p = NODE_DATA(nid);
1134 mem_hotplug_begin();
1137 * Add new range to memblock so that when hotadd_new_pgdat() is called
1138 * to allocate new pgdat, get_pfn_range_for_nid() will be able to find
1139 * this new range and calculate total pages correctly. The range will
1140 * be removed at hot-remove time.
1142 memblock_add_node(start, size, nid);
1144 new_node = !node_online(nid);
1146 pgdat = hotadd_new_pgdat(nid, start);
1152 /* call arch's memory hotadd */
1153 ret = arch_add_memory(nid, start, size, true);
1158 /* we online node here. we can't roll back from here. */
1159 node_set_online(nid);
1162 unsigned long start_pfn = start >> PAGE_SHIFT;
1163 unsigned long nr_pages = size >> PAGE_SHIFT;
1165 ret = __register_one_node(nid);
1170 * link memory sections under this node. This is already
1171 * done when creatig memory section in register_new_memory
1172 * but that depends to have the node registered so offline
1173 * nodes have to go through register_node.
1174 * TODO clean up this mess.
1176 ret = link_mem_sections(nid, start_pfn, nr_pages);
1179 * If sysfs file of new node can't create, cpu on the node
1180 * can't be hot-added. There is no rollback way now.
1181 * So, check by BUG_ON() to catch it reluctantly..
1186 /* create new memmap entry */
1187 firmware_map_add_hotplug(start, start + size, "System RAM");
1189 /* online pages if requested */
1191 walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1),
1192 NULL, online_memory_block);
1197 /* rollback pgdat allocation and others */
1198 if (new_pgdat && pgdat)
1199 rollback_node_hotadd(nid, pgdat);
1200 memblock_remove(start, size);
1206 EXPORT_SYMBOL_GPL(add_memory_resource);
1208 int __ref add_memory(int nid, u64 start, u64 size)
1210 struct resource *res;
1213 res = register_memory_resource(start, size);
1215 return PTR_ERR(res);
1217 ret = add_memory_resource(nid, res, memhp_auto_online);
1219 release_memory_resource(res);
1222 EXPORT_SYMBOL_GPL(add_memory);
1224 #ifdef CONFIG_MEMORY_HOTREMOVE
1226 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1227 * set and the size of the free page is given by page_order(). Using this,
1228 * the function determines if the pageblock contains only free pages.
1229 * Due to buddy contraints, a free page at least the size of a pageblock will
1230 * be located at the start of the pageblock
1232 static inline int pageblock_free(struct page *page)
1234 return PageBuddy(page) && page_order(page) >= pageblock_order;
1237 /* Return the start of the next active pageblock after a given page */
1238 static struct page *next_active_pageblock(struct page *page)
1240 /* Ensure the starting page is pageblock-aligned */
1241 BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
1243 /* If the entire pageblock is free, move to the end of free page */
1244 if (pageblock_free(page)) {
1246 /* be careful. we don't have locks, page_order can be changed.*/
1247 order = page_order(page);
1248 if ((order < MAX_ORDER) && (order >= pageblock_order))
1249 return page + (1 << order);
1252 return page + pageblock_nr_pages;
1255 /* Checks if this range of memory is likely to be hot-removable. */
1256 bool is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
1258 struct page *page = pfn_to_page(start_pfn);
1259 struct page *end_page = page + nr_pages;
1261 /* Check the starting page of each pageblock within the range */
1262 for (; page < end_page; page = next_active_pageblock(page)) {
1263 if (!is_pageblock_removable_nolock(page))
1268 /* All pageblocks in the memory block are likely to be hot-removable */
1273 * Confirm all pages in a range [start, end) belong to the same zone.
1274 * When true, return its valid [start, end).
1276 int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn,
1277 unsigned long *valid_start, unsigned long *valid_end)
1279 unsigned long pfn, sec_end_pfn;
1280 unsigned long start, end;
1281 struct zone *zone = NULL;
1284 for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1);
1286 pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) {
1287 /* Make sure the memory section is present first */
1288 if (!present_section_nr(pfn_to_section_nr(pfn)))
1290 for (; pfn < sec_end_pfn && pfn < end_pfn;
1291 pfn += MAX_ORDER_NR_PAGES) {
1293 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1294 while ((i < MAX_ORDER_NR_PAGES) &&
1295 !pfn_valid_within(pfn + i))
1297 if (i == MAX_ORDER_NR_PAGES || pfn + i >= end_pfn)
1299 page = pfn_to_page(pfn + i);
1300 if (zone && page_zone(page) != zone)
1304 zone = page_zone(page);
1305 end = pfn + MAX_ORDER_NR_PAGES;
1310 *valid_start = start;
1311 *valid_end = min(end, end_pfn);
1319 * Scan pfn range [start,end) to find movable/migratable pages (LRU pages,
1320 * non-lru movable pages and hugepages). We scan pfn because it's much
1321 * easier than scanning over linked list. This function returns the pfn
1322 * of the first found movable page if it's found, otherwise 0.
1324 static unsigned long scan_movable_pages(unsigned long start, unsigned long end)
1328 for (pfn = start; pfn < end; pfn++) {
1329 if (pfn_valid(pfn)) {
1330 page = pfn_to_page(pfn);
1333 if (__PageMovable(page))
1335 if (PageHuge(page)) {
1336 if (page_huge_active(page))
1339 pfn = round_up(pfn + 1,
1340 1 << compound_order(page)) - 1;
1347 static struct page *new_node_page(struct page *page, unsigned long private,
1350 int nid = page_to_nid(page);
1351 nodemask_t nmask = node_states[N_MEMORY];
1354 * try to allocate from a different node but reuse this node if there
1355 * are no other online nodes to be used (e.g. we are offlining a part
1356 * of the only existing node)
1358 node_clear(nid, nmask);
1359 if (nodes_empty(nmask))
1360 node_set(nid, nmask);
1362 return new_page_nodemask(page, nid, &nmask);
1365 #define NR_OFFLINE_AT_ONCE_PAGES (256)
1367 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1371 int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
1372 int not_managed = 0;
1376 for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
1377 if (!pfn_valid(pfn))
1379 page = pfn_to_page(pfn);
1381 if (PageHuge(page)) {
1382 struct page *head = compound_head(page);
1383 pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1;
1384 if (compound_order(head) > PFN_SECTION_SHIFT) {
1388 if (isolate_huge_page(page, &source))
1389 move_pages -= 1 << compound_order(head);
1391 } else if (thp_migration_supported() && PageTransHuge(page))
1392 pfn = page_to_pfn(compound_head(page))
1393 + hpage_nr_pages(page) - 1;
1396 * HWPoison pages have elevated reference counts so the migration would
1397 * fail on them. It also doesn't make any sense to migrate them in the
1398 * first place. Still try to unmap such a page in case it is still mapped
1399 * (e.g. current hwpoison implementation doesn't unmap KSM pages but keep
1400 * the unmap as the catch all safety net).
1402 if (PageHWPoison(page)) {
1403 if (WARN_ON(PageLRU(page)))
1404 isolate_lru_page(page);
1405 if (page_mapped(page))
1406 try_to_unmap(page, TTU_IGNORE_MLOCK | TTU_IGNORE_ACCESS);
1410 if (!get_page_unless_zero(page))
1413 * We can skip free pages. And we can deal with pages on
1414 * LRU and non-lru movable pages.
1417 ret = isolate_lru_page(page);
1419 ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE);
1420 if (!ret) { /* Success */
1422 list_add_tail(&page->lru, &source);
1424 if (!__PageMovable(page))
1425 inc_node_page_state(page, NR_ISOLATED_ANON +
1426 page_is_file_cache(page));
1429 #ifdef CONFIG_DEBUG_VM
1430 pr_alert("failed to isolate pfn %lx\n", pfn);
1431 dump_page(page, "isolation failed");
1434 /* Because we don't have big zone->lock. we should
1435 check this again here. */
1436 if (page_count(page)) {
1443 if (!list_empty(&source)) {
1445 putback_movable_pages(&source);
1449 /* Allocate a new page from the nearest neighbor node */
1450 ret = migrate_pages(&source, new_node_page, NULL, 0,
1451 MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
1453 putback_movable_pages(&source);
1460 * remove from free_area[] and mark all as Reserved.
1463 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1466 __offline_isolated_pages(start, start + nr_pages);
1471 offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
1473 walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
1474 offline_isolated_pages_cb);
1478 * Check all pages in range, recoreded as memory resource, are isolated.
1481 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1485 long offlined = *(long *)data;
1486 ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true);
1487 offlined = nr_pages;
1489 *(long *)data += offlined;
1494 check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
1499 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
1500 check_pages_isolated_cb);
1502 offlined = (long)ret;
1506 static int __init cmdline_parse_movable_node(char *p)
1508 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1509 movable_node_enabled = true;
1511 pr_warn("movable_node parameter depends on CONFIG_HAVE_MEMBLOCK_NODE_MAP to work properly\n");
1515 early_param("movable_node", cmdline_parse_movable_node);
1517 /* check which state of node_states will be changed when offline memory */
1518 static void node_states_check_changes_offline(unsigned long nr_pages,
1519 struct zone *zone, struct memory_notify *arg)
1521 struct pglist_data *pgdat = zone->zone_pgdat;
1522 unsigned long present_pages = 0;
1523 enum zone_type zt, zone_last = ZONE_NORMAL;
1526 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
1527 * contains nodes which have zones of 0...ZONE_NORMAL,
1528 * set zone_last to ZONE_NORMAL.
1530 * If we don't have HIGHMEM nor movable node,
1531 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
1532 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
1534 if (N_MEMORY == N_NORMAL_MEMORY)
1535 zone_last = ZONE_MOVABLE;
1538 * check whether node_states[N_NORMAL_MEMORY] will be changed.
1539 * If the memory to be offline is in a zone of 0...zone_last,
1540 * and it is the last present memory, 0...zone_last will
1541 * become empty after offline , thus we can determind we will
1542 * need to clear the node from node_states[N_NORMAL_MEMORY].
1544 for (zt = 0; zt <= zone_last; zt++)
1545 present_pages += pgdat->node_zones[zt].present_pages;
1546 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1547 arg->status_change_nid_normal = zone_to_nid(zone);
1549 arg->status_change_nid_normal = -1;
1551 #ifdef CONFIG_HIGHMEM
1553 * If we have movable node, node_states[N_HIGH_MEMORY]
1554 * contains nodes which have zones of 0...ZONE_HIGHMEM,
1555 * set zone_last to ZONE_HIGHMEM.
1557 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
1558 * contains nodes which have zones of 0...ZONE_MOVABLE,
1559 * set zone_last to ZONE_MOVABLE.
1561 zone_last = ZONE_HIGHMEM;
1562 if (N_MEMORY == N_HIGH_MEMORY)
1563 zone_last = ZONE_MOVABLE;
1565 for (; zt <= zone_last; zt++)
1566 present_pages += pgdat->node_zones[zt].present_pages;
1567 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1568 arg->status_change_nid_high = zone_to_nid(zone);
1570 arg->status_change_nid_high = -1;
1572 arg->status_change_nid_high = arg->status_change_nid_normal;
1576 * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE
1578 zone_last = ZONE_MOVABLE;
1581 * check whether node_states[N_HIGH_MEMORY] will be changed
1582 * If we try to offline the last present @nr_pages from the node,
1583 * we can determind we will need to clear the node from
1584 * node_states[N_HIGH_MEMORY].
1586 for (; zt <= zone_last; zt++)
1587 present_pages += pgdat->node_zones[zt].present_pages;
1588 if (nr_pages >= present_pages)
1589 arg->status_change_nid = zone_to_nid(zone);
1591 arg->status_change_nid = -1;
1594 static void node_states_clear_node(int node, struct memory_notify *arg)
1596 if (arg->status_change_nid_normal >= 0)
1597 node_clear_state(node, N_NORMAL_MEMORY);
1599 if ((N_MEMORY != N_NORMAL_MEMORY) &&
1600 (arg->status_change_nid_high >= 0))
1601 node_clear_state(node, N_HIGH_MEMORY);
1603 if ((N_MEMORY != N_HIGH_MEMORY) &&
1604 (arg->status_change_nid >= 0))
1605 node_clear_state(node, N_MEMORY);
1608 static int __ref __offline_pages(unsigned long start_pfn,
1609 unsigned long end_pfn, unsigned long timeout)
1611 unsigned long pfn, nr_pages, expire;
1612 long offlined_pages;
1613 int ret, drain, retry_max, node;
1614 unsigned long flags;
1615 unsigned long valid_start, valid_end;
1617 struct memory_notify arg;
1619 /* at least, alignment against pageblock is necessary */
1620 if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
1622 if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
1624 /* This makes hotplug much easier...and readable.
1625 we assume this for now. .*/
1626 if (!test_pages_in_a_zone(start_pfn, end_pfn, &valid_start, &valid_end))
1629 zone = page_zone(pfn_to_page(valid_start));
1630 node = zone_to_nid(zone);
1631 nr_pages = end_pfn - start_pfn;
1633 /* set above range as isolated */
1634 ret = start_isolate_page_range(start_pfn, end_pfn,
1635 MIGRATE_MOVABLE, true);
1639 arg.start_pfn = start_pfn;
1640 arg.nr_pages = nr_pages;
1641 node_states_check_changes_offline(nr_pages, zone, &arg);
1643 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1644 ret = notifier_to_errno(ret);
1646 goto failed_removal;
1649 expire = jiffies + timeout;
1653 /* start memory hot removal */
1655 if (time_after(jiffies, expire))
1656 goto failed_removal;
1658 if (signal_pending(current))
1659 goto failed_removal;
1662 lru_add_drain_all_cpuslocked();
1664 drain_all_pages(zone);
1667 pfn = scan_movable_pages(start_pfn, end_pfn);
1668 if (pfn) { /* We have movable pages */
1669 ret = do_migrate_range(pfn, end_pfn);
1675 if (--retry_max == 0)
1676 goto failed_removal;
1682 /* drain all zone's lru pagevec, this is asynchronous... */
1683 lru_add_drain_all_cpuslocked();
1685 /* drain pcp pages, this is synchronous. */
1686 drain_all_pages(zone);
1688 * dissolve free hugepages in the memory block before doing offlining
1689 * actually in order to make hugetlbfs's object counting consistent.
1691 ret = dissolve_free_huge_pages(start_pfn, end_pfn);
1693 goto failed_removal;
1695 offlined_pages = check_pages_isolated(start_pfn, end_pfn);
1696 if (offlined_pages < 0) {
1698 goto failed_removal;
1700 pr_info("Offlined Pages %ld\n", offlined_pages);
1701 /* Ok, all of our target is isolated.
1702 We cannot do rollback at this point. */
1703 offline_isolated_pages(start_pfn, end_pfn);
1704 /* reset pagetype flags and makes migrate type to be MOVABLE */
1705 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1706 /* removal success */
1707 adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages);
1708 zone->present_pages -= offlined_pages;
1710 pgdat_resize_lock(zone->zone_pgdat, &flags);
1711 zone->zone_pgdat->node_present_pages -= offlined_pages;
1712 pgdat_resize_unlock(zone->zone_pgdat, &flags);
1714 init_per_zone_wmark_min();
1716 if (!populated_zone(zone)) {
1717 zone_pcp_reset(zone);
1718 build_all_zonelists(NULL);
1720 zone_pcp_update(zone);
1722 node_states_clear_node(node, &arg);
1723 if (arg.status_change_nid >= 0) {
1725 kcompactd_stop(node);
1728 vm_total_pages = nr_free_pagecache_pages();
1729 writeback_set_ratelimit();
1731 memory_notify(MEM_OFFLINE, &arg);
1735 pr_debug("memory offlining [mem %#010llx-%#010llx] failed\n",
1736 (unsigned long long) start_pfn << PAGE_SHIFT,
1737 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
1738 memory_notify(MEM_CANCEL_OFFLINE, &arg);
1739 /* pushback to free area */
1740 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1744 /* Must be protected by mem_hotplug_begin() or a device_lock */
1745 int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1747 return __offline_pages(start_pfn, start_pfn + nr_pages, 120 * HZ);
1749 #endif /* CONFIG_MEMORY_HOTREMOVE */
1752 * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
1753 * @start_pfn: start pfn of the memory range
1754 * @end_pfn: end pfn of the memory range
1755 * @arg: argument passed to func
1756 * @func: callback for each memory section walked
1758 * This function walks through all present mem sections in range
1759 * [start_pfn, end_pfn) and call func on each mem section.
1761 * Returns the return value of func.
1763 int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn,
1764 void *arg, int (*func)(struct memory_block *, void *))
1766 struct memory_block *mem = NULL;
1767 struct mem_section *section;
1768 unsigned long pfn, section_nr;
1771 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1772 section_nr = pfn_to_section_nr(pfn);
1773 if (!present_section_nr(section_nr))
1776 section = __nr_to_section(section_nr);
1777 /* same memblock? */
1779 if ((section_nr >= mem->start_section_nr) &&
1780 (section_nr <= mem->end_section_nr))
1783 mem = find_memory_block_hinted(section, mem);
1787 ret = func(mem, arg);
1789 kobject_put(&mem->dev.kobj);
1795 kobject_put(&mem->dev.kobj);
1800 #ifdef CONFIG_MEMORY_HOTREMOVE
1801 static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
1803 int ret = !is_memblock_offlined(mem);
1805 if (unlikely(ret)) {
1806 phys_addr_t beginpa, endpa;
1808 beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
1809 endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1;
1810 pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n",
1817 static int check_cpu_on_node(pg_data_t *pgdat)
1821 for_each_present_cpu(cpu) {
1822 if (cpu_to_node(cpu) == pgdat->node_id)
1824 * the cpu on this node isn't removed, and we can't
1825 * offline this node.
1833 static void unmap_cpu_on_node(pg_data_t *pgdat)
1835 #ifdef CONFIG_ACPI_NUMA
1838 for_each_possible_cpu(cpu)
1839 if (cpu_to_node(cpu) == pgdat->node_id)
1840 numa_clear_node(cpu);
1844 static int check_and_unmap_cpu_on_node(pg_data_t *pgdat)
1848 ret = check_cpu_on_node(pgdat);
1853 * the node will be offlined when we come here, so we can clear
1854 * the cpu_to_node() now.
1857 unmap_cpu_on_node(pgdat);
1864 * Offline a node if all memory sections and cpus of the node are removed.
1866 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1867 * and online/offline operations before this call.
1869 void try_offline_node(int nid)
1871 pg_data_t *pgdat = NODE_DATA(nid);
1872 unsigned long start_pfn = pgdat->node_start_pfn;
1873 unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
1876 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1877 unsigned long section_nr = pfn_to_section_nr(pfn);
1879 if (!present_section_nr(section_nr))
1882 if (pfn_to_nid(pfn) != nid)
1886 * some memory sections of this node are not removed, and we
1887 * can't offline node now.
1892 if (check_and_unmap_cpu_on_node(pgdat))
1896 * all memory/cpu of this node are removed, we can offline this
1899 node_set_offline(nid);
1900 unregister_one_node(nid);
1902 EXPORT_SYMBOL(try_offline_node);
1907 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1908 * and online/offline operations before this call, as required by
1909 * try_offline_node().
1911 void __ref remove_memory(int nid, u64 start, u64 size)
1915 BUG_ON(check_hotplug_memory_range(start, size));
1917 mem_hotplug_begin();
1920 * All memory blocks must be offlined before removing memory. Check
1921 * whether all memory blocks in question are offline and trigger a BUG()
1922 * if this is not the case.
1924 ret = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL,
1925 check_memblock_offlined_cb);
1929 /* remove memmap entry */
1930 firmware_map_remove(start, start + size, "System RAM");
1931 memblock_free(start, size);
1932 memblock_remove(start, size);
1934 arch_remove_memory(start, size);
1936 try_offline_node(nid);
1940 EXPORT_SYMBOL_GPL(remove_memory);
1941 #endif /* CONFIG_MEMORY_HOTREMOVE */