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 = ms->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 section_nr = pfn_to_section_nr(start_pfn);
205 ms = __nr_to_section(section_nr);
207 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
209 register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION);
211 usemap = ms->pageblock_flags;
212 page = virt_to_page(usemap);
214 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
216 for (i = 0; i < mapsize; i++, page++)
217 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
219 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
221 void __init register_page_bootmem_info_node(struct pglist_data *pgdat)
223 unsigned long i, pfn, end_pfn, nr_pages;
224 int node = pgdat->node_id;
227 nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
228 page = virt_to_page(pgdat);
230 for (i = 0; i < nr_pages; i++, page++)
231 get_page_bootmem(node, page, NODE_INFO);
233 pfn = pgdat->node_start_pfn;
234 end_pfn = pgdat_end_pfn(pgdat);
236 /* register section info */
237 for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
239 * Some platforms can assign the same pfn to multiple nodes - on
240 * node0 as well as nodeN. To avoid registering a pfn against
241 * multiple nodes we check that this pfn does not already
242 * reside in some other nodes.
244 if (pfn_valid(pfn) && (early_pfn_to_nid(pfn) == node))
245 register_page_bootmem_info_section(pfn);
248 #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
250 static int __meminit __add_section(int nid, unsigned long phys_start_pfn,
251 struct vmem_altmap *altmap, bool want_memblock)
255 if (pfn_valid(phys_start_pfn))
258 ret = sparse_add_one_section(nid, phys_start_pfn, altmap);
265 return hotplug_memory_register(nid, __pfn_to_section(phys_start_pfn));
269 * Reasonably generic function for adding memory. It is
270 * expected that archs that support memory hotplug will
271 * call this function after deciding the zone to which to
274 int __ref __add_pages(int nid, unsigned long phys_start_pfn,
275 unsigned long nr_pages, struct vmem_altmap *altmap,
280 int start_sec, end_sec;
282 /* during initialize mem_map, align hot-added range to section */
283 start_sec = pfn_to_section_nr(phys_start_pfn);
284 end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
288 * Validate altmap is within bounds of the total request
290 if (altmap->base_pfn != phys_start_pfn
291 || vmem_altmap_offset(altmap) > nr_pages) {
292 pr_warn_once("memory add fail, invalid altmap\n");
299 for (i = start_sec; i <= end_sec; i++) {
300 err = __add_section(nid, section_nr_to_pfn(i), altmap,
304 * EEXIST is finally dealt with by ioresource collision
305 * check. see add_memory() => register_memory_resource()
306 * Warning will be printed if there is collision.
308 if (err && (err != -EEXIST))
313 vmemmap_populate_print_last();
318 #ifdef CONFIG_MEMORY_HOTREMOVE
319 /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
320 static unsigned long find_smallest_section_pfn(int nid, struct zone *zone,
321 unsigned long start_pfn,
322 unsigned long end_pfn)
324 for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) {
325 if (unlikely(!pfn_to_online_page(start_pfn)))
328 if (unlikely(pfn_to_nid(start_pfn) != nid))
331 if (zone && zone != page_zone(pfn_to_page(start_pfn)))
340 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
341 static unsigned long find_biggest_section_pfn(int nid, struct zone *zone,
342 unsigned long start_pfn,
343 unsigned long end_pfn)
347 /* pfn is the end pfn of a memory section. */
349 for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) {
350 if (unlikely(!pfn_to_online_page(pfn)))
353 if (unlikely(pfn_to_nid(pfn) != nid))
356 if (zone && zone != page_zone(pfn_to_page(pfn)))
365 static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
366 unsigned long end_pfn)
368 unsigned long zone_start_pfn = zone->zone_start_pfn;
369 unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */
370 unsigned long zone_end_pfn = z;
372 int nid = zone_to_nid(zone);
374 zone_span_writelock(zone);
375 if (zone_start_pfn == start_pfn) {
377 * If the section is smallest section in the zone, it need
378 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
379 * In this case, we find second smallest valid mem_section
380 * for shrinking zone.
382 pfn = find_smallest_section_pfn(nid, zone, end_pfn,
385 zone->zone_start_pfn = pfn;
386 zone->spanned_pages = zone_end_pfn - pfn;
388 } else if (zone_end_pfn == end_pfn) {
390 * If the section is biggest section in the zone, it need
391 * shrink zone->spanned_pages.
392 * In this case, we find second biggest valid mem_section for
395 pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn,
398 zone->spanned_pages = pfn - zone_start_pfn + 1;
402 * The section is not biggest or smallest mem_section in the zone, it
403 * only creates a hole in the zone. So in this case, we need not
404 * change the zone. But perhaps, the zone has only hole data. Thus
405 * it check the zone has only hole or not.
407 pfn = zone_start_pfn;
408 for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) {
409 if (unlikely(!pfn_to_online_page(pfn)))
412 if (page_zone(pfn_to_page(pfn)) != zone)
415 /* If the section is current section, it continues the loop */
416 if (start_pfn == pfn)
419 /* If we find valid section, we have nothing to do */
420 zone_span_writeunlock(zone);
424 /* The zone has no valid section */
425 zone->zone_start_pfn = 0;
426 zone->spanned_pages = 0;
427 zone_span_writeunlock(zone);
430 static void update_pgdat_span(struct pglist_data *pgdat)
432 unsigned long node_start_pfn = 0, node_end_pfn = 0;
435 for (zone = pgdat->node_zones;
436 zone < pgdat->node_zones + MAX_NR_ZONES; zone++) {
437 unsigned long zone_end_pfn = zone->zone_start_pfn +
440 /* No need to lock the zones, they can't change. */
441 if (!zone->spanned_pages)
444 node_start_pfn = zone->zone_start_pfn;
445 node_end_pfn = zone_end_pfn;
449 if (zone_end_pfn > node_end_pfn)
450 node_end_pfn = zone_end_pfn;
451 if (zone->zone_start_pfn < node_start_pfn)
452 node_start_pfn = zone->zone_start_pfn;
455 pgdat->node_start_pfn = node_start_pfn;
456 pgdat->node_spanned_pages = node_end_pfn - node_start_pfn;
459 static void __remove_zone(struct zone *zone, unsigned long start_pfn)
461 struct pglist_data *pgdat = zone->zone_pgdat;
462 int nr_pages = PAGES_PER_SECTION;
465 #ifdef CONFIG_ZONE_DEVICE
467 * Zone shrinking code cannot properly deal with ZONE_DEVICE. So
468 * we will not try to shrink the zones - which is okay as
469 * set_zone_contiguous() cannot deal with ZONE_DEVICE either way.
471 if (zone_idx(zone) == ZONE_DEVICE)
475 pgdat_resize_lock(zone->zone_pgdat, &flags);
476 shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
477 update_pgdat_span(pgdat);
478 pgdat_resize_unlock(zone->zone_pgdat, &flags);
481 static int __remove_section(struct zone *zone, struct mem_section *ms,
482 unsigned long map_offset, struct vmem_altmap *altmap)
484 unsigned long start_pfn;
488 if (!valid_section(ms))
491 ret = unregister_memory_section(ms);
495 scn_nr = __section_nr(ms);
496 start_pfn = section_nr_to_pfn((unsigned long)scn_nr);
497 __remove_zone(zone, start_pfn);
499 sparse_remove_one_section(zone, ms, map_offset, altmap);
504 * __remove_pages() - remove sections of pages from a zone
505 * @zone: zone from which pages need to be removed
506 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
507 * @nr_pages: number of pages to remove (must be multiple of section size)
508 * @altmap: alternative device page map or %NULL if default memmap is used
510 * Generic helper function to remove section mappings and sysfs entries
511 * for the section of the memory we are removing. Caller needs to make
512 * sure that pages are marked reserved and zones are adjust properly by
513 * calling offline_pages().
515 int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
516 unsigned long nr_pages, struct vmem_altmap *altmap)
519 unsigned long map_offset = 0;
520 int sections_to_remove, ret = 0;
522 /* In the ZONE_DEVICE case device driver owns the memory region */
523 if (is_dev_zone(zone)) {
525 map_offset = vmem_altmap_offset(altmap);
527 resource_size_t start, size;
529 start = phys_start_pfn << PAGE_SHIFT;
530 size = nr_pages * PAGE_SIZE;
532 ret = release_mem_region_adjustable(&iomem_resource, start,
535 resource_size_t endres = start + size - 1;
537 pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
538 &start, &endres, ret);
542 clear_zone_contiguous(zone);
545 * We can only remove entire sections
547 BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
548 BUG_ON(nr_pages % PAGES_PER_SECTION);
550 sections_to_remove = nr_pages / PAGES_PER_SECTION;
551 for (i = 0; i < sections_to_remove; i++) {
552 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
555 ret = __remove_section(zone, __pfn_to_section(pfn), map_offset,
562 set_zone_contiguous(zone);
566 #endif /* CONFIG_MEMORY_HOTREMOVE */
568 int set_online_page_callback(online_page_callback_t callback)
573 mutex_lock(&online_page_callback_lock);
575 if (online_page_callback == generic_online_page) {
576 online_page_callback = callback;
580 mutex_unlock(&online_page_callback_lock);
585 EXPORT_SYMBOL_GPL(set_online_page_callback);
587 int restore_online_page_callback(online_page_callback_t callback)
592 mutex_lock(&online_page_callback_lock);
594 if (online_page_callback == callback) {
595 online_page_callback = generic_online_page;
599 mutex_unlock(&online_page_callback_lock);
604 EXPORT_SYMBOL_GPL(restore_online_page_callback);
606 void __online_page_set_limits(struct page *page)
609 EXPORT_SYMBOL_GPL(__online_page_set_limits);
611 void __online_page_increment_counters(struct page *page)
613 adjust_managed_page_count(page, 1);
615 EXPORT_SYMBOL_GPL(__online_page_increment_counters);
617 void __online_page_free(struct page *page)
619 __free_reserved_page(page);
621 EXPORT_SYMBOL_GPL(__online_page_free);
623 static void generic_online_page(struct page *page)
625 __online_page_set_limits(page);
626 __online_page_increment_counters(page);
627 __online_page_free(page);
630 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
634 unsigned long onlined_pages = *(unsigned long *)arg;
637 if (PageReserved(pfn_to_page(start_pfn)))
638 for (i = 0; i < nr_pages; i++) {
639 page = pfn_to_page(start_pfn + i);
640 (*online_page_callback)(page);
644 online_mem_sections(start_pfn, start_pfn + nr_pages);
646 *(unsigned long *)arg = onlined_pages;
650 /* check which state of node_states will be changed when online memory */
651 static void node_states_check_changes_online(unsigned long nr_pages,
652 struct zone *zone, struct memory_notify *arg)
654 int nid = zone_to_nid(zone);
655 enum zone_type zone_last = ZONE_NORMAL;
658 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
659 * contains nodes which have zones of 0...ZONE_NORMAL,
660 * set zone_last to ZONE_NORMAL.
662 * If we don't have HIGHMEM nor movable node,
663 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
664 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
666 if (N_MEMORY == N_NORMAL_MEMORY)
667 zone_last = ZONE_MOVABLE;
670 * if the memory to be online is in a zone of 0...zone_last, and
671 * the zones of 0...zone_last don't have memory before online, we will
672 * need to set the node to node_states[N_NORMAL_MEMORY] after
673 * the memory is online.
675 if (zone_idx(zone) <= zone_last && !node_state(nid, N_NORMAL_MEMORY))
676 arg->status_change_nid_normal = nid;
678 arg->status_change_nid_normal = -1;
680 #ifdef CONFIG_HIGHMEM
682 * If we have movable node, node_states[N_HIGH_MEMORY]
683 * contains nodes which have zones of 0...ZONE_HIGHMEM,
684 * set zone_last to ZONE_HIGHMEM.
686 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
687 * contains nodes which have zones of 0...ZONE_MOVABLE,
688 * set zone_last to ZONE_MOVABLE.
690 zone_last = ZONE_HIGHMEM;
691 if (N_MEMORY == N_HIGH_MEMORY)
692 zone_last = ZONE_MOVABLE;
694 if (zone_idx(zone) <= zone_last && !node_state(nid, N_HIGH_MEMORY))
695 arg->status_change_nid_high = nid;
697 arg->status_change_nid_high = -1;
699 arg->status_change_nid_high = arg->status_change_nid_normal;
703 * if the node don't have memory befor online, we will need to
704 * set the node to node_states[N_MEMORY] after the memory
707 if (!node_state(nid, N_MEMORY))
708 arg->status_change_nid = nid;
710 arg->status_change_nid = -1;
713 static void node_states_set_node(int node, struct memory_notify *arg)
715 if (arg->status_change_nid_normal >= 0)
716 node_set_state(node, N_NORMAL_MEMORY);
718 if (arg->status_change_nid_high >= 0)
719 node_set_state(node, N_HIGH_MEMORY);
721 node_set_state(node, N_MEMORY);
724 static void __meminit resize_zone_range(struct zone *zone, unsigned long start_pfn,
725 unsigned long nr_pages)
727 unsigned long old_end_pfn = zone_end_pfn(zone);
729 if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
730 zone->zone_start_pfn = start_pfn;
732 zone->spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - zone->zone_start_pfn;
735 static void __meminit resize_pgdat_range(struct pglist_data *pgdat, unsigned long start_pfn,
736 unsigned long nr_pages)
738 unsigned long old_end_pfn = pgdat_end_pfn(pgdat);
740 if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
741 pgdat->node_start_pfn = start_pfn;
743 pgdat->node_spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - pgdat->node_start_pfn;
746 void __ref move_pfn_range_to_zone(struct zone *zone, unsigned long start_pfn,
747 unsigned long nr_pages, struct vmem_altmap *altmap)
749 struct pglist_data *pgdat = zone->zone_pgdat;
750 int nid = pgdat->node_id;
753 if (zone_is_empty(zone))
754 init_currently_empty_zone(zone, start_pfn, nr_pages);
756 clear_zone_contiguous(zone);
758 /* TODO Huh pgdat is irqsave while zone is not. It used to be like that before */
759 pgdat_resize_lock(pgdat, &flags);
760 zone_span_writelock(zone);
761 resize_zone_range(zone, start_pfn, nr_pages);
762 zone_span_writeunlock(zone);
763 resize_pgdat_range(pgdat, start_pfn, nr_pages);
764 pgdat_resize_unlock(pgdat, &flags);
767 * TODO now we have a visible range of pages which are not associated
768 * with their zone properly. Not nice but set_pfnblock_flags_mask
769 * expects the zone spans the pfn range. All the pages in the range
770 * are reserved so nobody should be touching them so we should be safe
772 memmap_init_zone(nr_pages, nid, zone_idx(zone), start_pfn,
773 MEMMAP_HOTPLUG, altmap);
775 set_zone_contiguous(zone);
779 * Returns a default kernel memory zone for the given pfn range.
780 * If no kernel zone covers this pfn range it will automatically go
781 * to the ZONE_NORMAL.
783 static struct zone *default_kernel_zone_for_pfn(int nid, unsigned long start_pfn,
784 unsigned long nr_pages)
786 struct pglist_data *pgdat = NODE_DATA(nid);
789 for (zid = 0; zid <= ZONE_NORMAL; zid++) {
790 struct zone *zone = &pgdat->node_zones[zid];
792 if (zone_intersects(zone, start_pfn, nr_pages))
796 return &pgdat->node_zones[ZONE_NORMAL];
799 static inline struct zone *default_zone_for_pfn(int nid, unsigned long start_pfn,
800 unsigned long nr_pages)
802 struct zone *kernel_zone = default_kernel_zone_for_pfn(nid, start_pfn,
804 struct zone *movable_zone = &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
805 bool in_kernel = zone_intersects(kernel_zone, start_pfn, nr_pages);
806 bool in_movable = zone_intersects(movable_zone, start_pfn, nr_pages);
809 * We inherit the existing zone in a simple case where zones do not
810 * overlap in the given range
812 if (in_kernel ^ in_movable)
813 return (in_kernel) ? kernel_zone : movable_zone;
816 * If the range doesn't belong to any zone or two zones overlap in the
817 * given range then we use movable zone only if movable_node is
818 * enabled because we always online to a kernel zone by default.
820 return movable_node_enabled ? movable_zone : kernel_zone;
823 struct zone * zone_for_pfn_range(int online_type, int nid, unsigned start_pfn,
824 unsigned long nr_pages)
826 if (online_type == MMOP_ONLINE_KERNEL)
827 return default_kernel_zone_for_pfn(nid, start_pfn, nr_pages);
829 if (online_type == MMOP_ONLINE_MOVABLE)
830 return &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
832 return default_zone_for_pfn(nid, start_pfn, nr_pages);
836 * Associates the given pfn range with the given node and the zone appropriate
837 * for the given online type.
839 static struct zone * __meminit move_pfn_range(int online_type, int nid,
840 unsigned long start_pfn, unsigned long nr_pages)
844 zone = zone_for_pfn_range(online_type, nid, start_pfn, nr_pages);
845 move_pfn_range_to_zone(zone, start_pfn, nr_pages, NULL);
849 int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type)
852 unsigned long onlined_pages = 0;
854 int need_zonelists_rebuild = 0;
857 struct memory_notify arg;
858 struct memory_block *mem;
863 * We can't use pfn_to_nid() because nid might be stored in struct page
864 * which is not yet initialized. Instead, we find nid from memory block.
866 mem = find_memory_block(__pfn_to_section(pfn));
868 put_device(&mem->dev);
870 /* associate pfn range with the zone */
871 zone = move_pfn_range(online_type, nid, pfn, nr_pages);
874 arg.nr_pages = nr_pages;
875 node_states_check_changes_online(nr_pages, zone, &arg);
877 ret = memory_notify(MEM_GOING_ONLINE, &arg);
878 ret = notifier_to_errno(ret);
880 goto failed_addition;
883 * If this zone is not populated, then it is not in zonelist.
884 * This means the page allocator ignores this zone.
885 * So, zonelist must be updated after online.
887 if (!populated_zone(zone)) {
888 need_zonelists_rebuild = 1;
889 setup_zone_pageset(zone);
892 ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
895 if (need_zonelists_rebuild)
896 zone_pcp_reset(zone);
897 goto failed_addition;
900 zone->present_pages += onlined_pages;
902 pgdat_resize_lock(zone->zone_pgdat, &flags);
903 zone->zone_pgdat->node_present_pages += onlined_pages;
904 pgdat_resize_unlock(zone->zone_pgdat, &flags);
907 node_states_set_node(nid, &arg);
908 if (need_zonelists_rebuild)
909 build_all_zonelists(NULL);
911 zone_pcp_update(zone);
914 init_per_zone_wmark_min();
921 vm_total_pages = nr_free_pagecache_pages();
923 writeback_set_ratelimit();
926 memory_notify(MEM_ONLINE, &arg);
931 pr_debug("online_pages [mem %#010llx-%#010llx] failed\n",
932 (unsigned long long) pfn << PAGE_SHIFT,
933 (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1);
934 memory_notify(MEM_CANCEL_ONLINE, &arg);
938 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
940 static void reset_node_present_pages(pg_data_t *pgdat)
944 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
945 z->present_pages = 0;
947 pgdat->node_present_pages = 0;
950 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
951 static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
953 struct pglist_data *pgdat;
954 unsigned long start_pfn = PFN_DOWN(start);
956 pgdat = NODE_DATA(nid);
958 pgdat = arch_alloc_nodedata(nid);
962 arch_refresh_nodedata(nid, pgdat);
965 * Reset the nr_zones, order and classzone_idx before reuse.
966 * Note that kswapd will init kswapd_classzone_idx properly
967 * when it starts in the near future.
970 pgdat->kswapd_order = 0;
971 pgdat->kswapd_classzone_idx = 0;
974 /* we can use NODE_DATA(nid) from here */
976 pgdat->node_id = nid;
977 pgdat->node_start_pfn = start_pfn;
979 /* init node's zones as empty zones, we don't have any present pages.*/
980 free_area_init_core_hotplug(nid);
981 pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat);
984 * The node we allocated has no zone fallback lists. For avoiding
985 * to access not-initialized zonelist, build here.
987 build_all_zonelists(pgdat);
990 * When memory is hot-added, all the memory is in offline state. So
991 * clear all zones' present_pages because they will be updated in
992 * online_pages() and offline_pages().
994 reset_node_managed_pages(pgdat);
995 reset_node_present_pages(pgdat);
1000 static void rollback_node_hotadd(int nid)
1002 pg_data_t *pgdat = NODE_DATA(nid);
1004 arch_refresh_nodedata(nid, NULL);
1005 free_percpu(pgdat->per_cpu_nodestats);
1006 arch_free_nodedata(pgdat);
1012 * try_online_node - online a node if offlined
1014 * @start: start addr of the node
1015 * @set_node_online: Whether we want to online the node
1016 * called by cpu_up() to online a node without onlined memory.
1019 * 1 -> a new node has been allocated
1020 * 0 -> the node is already online
1021 * -ENOMEM -> the node could not be allocated
1023 static int __try_online_node(int nid, u64 start, bool set_node_online)
1028 if (node_online(nid))
1031 pgdat = hotadd_new_pgdat(nid, start);
1033 pr_err("Cannot online node %d due to NULL pgdat\n", nid);
1038 if (set_node_online) {
1039 node_set_online(nid);
1040 ret = register_one_node(nid);
1048 * Users of this function always want to online/register the node
1050 int try_online_node(int nid)
1054 mem_hotplug_begin();
1055 ret = __try_online_node(nid, 0, true);
1060 static int check_hotplug_memory_range(u64 start, u64 size)
1062 unsigned long block_sz = memory_block_size_bytes();
1063 u64 block_nr_pages = block_sz >> PAGE_SHIFT;
1064 u64 nr_pages = size >> PAGE_SHIFT;
1065 u64 start_pfn = PFN_DOWN(start);
1067 /* memory range must be block size aligned */
1068 if (!nr_pages || !IS_ALIGNED(start_pfn, block_nr_pages) ||
1069 !IS_ALIGNED(nr_pages, block_nr_pages)) {
1070 pr_err("Block size [%#lx] unaligned hotplug range: start %#llx, size %#llx",
1071 block_sz, start, size);
1078 static int online_memory_block(struct memory_block *mem, void *arg)
1080 return device_online(&mem->dev);
1084 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1085 * and online/offline operations (triggered e.g. by sysfs).
1087 * we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG
1089 int __ref add_memory_resource(int nid, struct resource *res, bool online)
1092 bool new_node = false;
1096 size = resource_size(res);
1098 ret = check_hotplug_memory_range(start, size);
1102 mem_hotplug_begin();
1105 * Add new range to memblock so that when hotadd_new_pgdat() is called
1106 * to allocate new pgdat, get_pfn_range_for_nid() will be able to find
1107 * this new range and calculate total pages correctly. The range will
1108 * be removed at hot-remove time.
1110 memblock_add_node(start, size, nid);
1112 ret = __try_online_node(nid, start, false);
1117 /* call arch's memory hotadd */
1118 ret = arch_add_memory(nid, start, size, NULL, true);
1123 /* If sysfs file of new node can't be created, cpu on the node
1124 * can't be hot-added. There is no rollback way now.
1125 * So, check by BUG_ON() to catch it reluctantly..
1126 * We online node here. We can't roll back from here.
1128 node_set_online(nid);
1129 ret = __register_one_node(nid);
1133 /* link memory sections under this node.*/
1134 ret = link_mem_sections(nid, PFN_DOWN(start), PFN_UP(start + size - 1));
1137 /* create new memmap entry */
1138 firmware_map_add_hotplug(start, start + size, "System RAM");
1140 /* device_online() will take the lock when calling online_pages() */
1143 /* online pages if requested */
1145 walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1),
1146 NULL, online_memory_block);
1150 /* rollback pgdat allocation and others */
1152 rollback_node_hotadd(nid);
1153 memblock_remove(start, size);
1158 /* requires device_hotplug_lock, see add_memory_resource() */
1159 int __ref __add_memory(int nid, u64 start, u64 size)
1161 struct resource *res;
1164 res = register_memory_resource(start, size);
1166 return PTR_ERR(res);
1168 ret = add_memory_resource(nid, res, memhp_auto_online);
1170 release_memory_resource(res);
1174 int add_memory(int nid, u64 start, u64 size)
1178 lock_device_hotplug();
1179 rc = __add_memory(nid, start, size);
1180 unlock_device_hotplug();
1184 EXPORT_SYMBOL_GPL(add_memory);
1186 #ifdef CONFIG_MEMORY_HOTREMOVE
1188 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1189 * set and the size of the free page is given by page_order(). Using this,
1190 * the function determines if the pageblock contains only free pages.
1191 * Due to buddy contraints, a free page at least the size of a pageblock will
1192 * be located at the start of the pageblock
1194 static inline int pageblock_free(struct page *page)
1196 return PageBuddy(page) && page_order(page) >= pageblock_order;
1199 /* Return the pfn of the start of the next active pageblock after a given pfn */
1200 static unsigned long next_active_pageblock(unsigned long pfn)
1202 struct page *page = pfn_to_page(pfn);
1204 /* Ensure the starting page is pageblock-aligned */
1205 BUG_ON(pfn & (pageblock_nr_pages - 1));
1207 /* If the entire pageblock is free, move to the end of free page */
1208 if (pageblock_free(page)) {
1210 /* be careful. we don't have locks, page_order can be changed.*/
1211 order = page_order(page);
1212 if ((order < MAX_ORDER) && (order >= pageblock_order))
1213 return pfn + (1 << order);
1216 return pfn + pageblock_nr_pages;
1219 static bool is_pageblock_removable_nolock(unsigned long pfn)
1221 struct page *page = pfn_to_page(pfn);
1225 * We have to be careful here because we are iterating over memory
1226 * sections which are not zone aware so we might end up outside of
1227 * the zone but still within the section.
1228 * We have to take care about the node as well. If the node is offline
1229 * its NODE_DATA will be NULL - see page_zone.
1231 if (!node_online(page_to_nid(page)))
1234 zone = page_zone(page);
1235 pfn = page_to_pfn(page);
1236 if (!zone_spans_pfn(zone, pfn))
1239 return !has_unmovable_pages(zone, page, 0, MIGRATE_MOVABLE, true);
1242 /* Checks if this range of memory is likely to be hot-removable. */
1243 bool is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
1245 unsigned long end_pfn, pfn;
1247 end_pfn = min(start_pfn + nr_pages,
1248 zone_end_pfn(page_zone(pfn_to_page(start_pfn))));
1250 /* Check the starting page of each pageblock within the range */
1251 for (pfn = start_pfn; pfn < end_pfn; pfn = next_active_pageblock(pfn)) {
1252 if (!is_pageblock_removable_nolock(pfn))
1257 /* All pageblocks in the memory block are likely to be hot-removable */
1262 * Confirm all pages in a range [start, end) belong to the same zone.
1263 * When true, return its valid [start, end).
1265 int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn,
1266 unsigned long *valid_start, unsigned long *valid_end)
1268 unsigned long pfn, sec_end_pfn;
1269 unsigned long start, end;
1270 struct zone *zone = NULL;
1273 for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1);
1275 pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) {
1276 /* Make sure the memory section is present first */
1277 if (!present_section_nr(pfn_to_section_nr(pfn)))
1279 for (; pfn < sec_end_pfn && pfn < end_pfn;
1280 pfn += MAX_ORDER_NR_PAGES) {
1282 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1283 while ((i < MAX_ORDER_NR_PAGES) &&
1284 !pfn_valid_within(pfn + i))
1286 if (i == MAX_ORDER_NR_PAGES || pfn + i >= end_pfn)
1288 /* Check if we got outside of the zone */
1289 if (zone && !zone_spans_pfn(zone, pfn + i))
1291 page = pfn_to_page(pfn + i);
1292 if (zone && page_zone(page) != zone)
1296 zone = page_zone(page);
1297 end = pfn + MAX_ORDER_NR_PAGES;
1302 *valid_start = start;
1303 *valid_end = min(end, end_pfn);
1311 * Scan pfn range [start,end) to find movable/migratable pages (LRU pages,
1312 * non-lru movable pages and hugepages). We scan pfn because it's much
1313 * easier than scanning over linked list. This function returns the pfn
1314 * of the first found movable page if it's found, otherwise 0.
1316 static unsigned long scan_movable_pages(unsigned long start, unsigned long end)
1320 for (pfn = start; pfn < end; pfn++) {
1321 struct page *page, *head;
1324 if (!pfn_valid(pfn))
1326 page = pfn_to_page(pfn);
1329 if (__PageMovable(page))
1332 if (!PageHuge(page))
1334 head = compound_head(page);
1335 if (hugepage_migration_supported(page_hstate(head)) &&
1336 page_huge_active(head))
1338 skip = (1 << compound_order(head)) - (page - head);
1344 static struct page *new_node_page(struct page *page, unsigned long private)
1346 int nid = page_to_nid(page);
1347 nodemask_t nmask = node_states[N_MEMORY];
1350 * try to allocate from a different node but reuse this node if there
1351 * are no other online nodes to be used (e.g. we are offlining a part
1352 * of the only existing node)
1354 node_clear(nid, nmask);
1355 if (nodes_empty(nmask))
1356 node_set(nid, nmask);
1358 return new_page_nodemask(page, nid, &nmask);
1361 #define NR_OFFLINE_AT_ONCE_PAGES (256)
1363 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1367 int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
1368 int not_managed = 0;
1372 for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
1373 if (!pfn_valid(pfn))
1375 page = pfn_to_page(pfn);
1377 if (PageHuge(page)) {
1378 struct page *head = compound_head(page);
1379 pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1;
1380 if (compound_order(head) > PFN_SECTION_SHIFT) {
1384 if (isolate_huge_page(page, &source))
1385 move_pages -= 1 << compound_order(head);
1387 } else if (PageTransHuge(page))
1388 pfn = page_to_pfn(compound_head(page))
1389 + hpage_nr_pages(page) - 1;
1392 * HWPoison pages have elevated reference counts so the migration would
1393 * fail on them. It also doesn't make any sense to migrate them in the
1394 * first place. Still try to unmap such a page in case it is still mapped
1395 * (e.g. current hwpoison implementation doesn't unmap KSM pages but keep
1396 * the unmap as the catch all safety net).
1398 if (PageHWPoison(page)) {
1399 if (WARN_ON(PageLRU(page)))
1400 isolate_lru_page(page);
1401 if (page_mapped(page))
1402 try_to_unmap(page, TTU_IGNORE_MLOCK | TTU_IGNORE_ACCESS);
1406 if (!get_page_unless_zero(page))
1409 * We can skip free pages. And we can deal with pages on
1410 * LRU and non-lru movable pages.
1413 ret = isolate_lru_page(page);
1415 ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE);
1416 if (!ret) { /* Success */
1418 list_add_tail(&page->lru, &source);
1420 if (!__PageMovable(page))
1421 inc_node_page_state(page, NR_ISOLATED_ANON +
1422 page_is_file_cache(page));
1425 #ifdef CONFIG_DEBUG_VM
1426 pr_alert("failed to isolate pfn %lx\n", pfn);
1427 dump_page(page, "isolation failed");
1430 /* Because we don't have big zone->lock. we should
1431 check this again here. */
1432 if (page_count(page)) {
1439 if (!list_empty(&source)) {
1441 putback_movable_pages(&source);
1445 /* Allocate a new page from the nearest neighbor node */
1446 ret = migrate_pages(&source, new_node_page, NULL, 0,
1447 MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
1449 putback_movable_pages(&source);
1456 * remove from free_area[] and mark all as Reserved.
1459 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1462 __offline_isolated_pages(start, start + nr_pages);
1467 offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
1469 walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
1470 offline_isolated_pages_cb);
1474 * Check all pages in range, recoreded as memory resource, are isolated.
1477 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1481 long offlined = *(long *)data;
1482 ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true);
1483 offlined = nr_pages;
1485 *(long *)data += offlined;
1490 check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
1495 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
1496 check_pages_isolated_cb);
1498 offlined = (long)ret;
1502 static int __init cmdline_parse_movable_node(char *p)
1504 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1505 movable_node_enabled = true;
1507 pr_warn("movable_node parameter depends on CONFIG_HAVE_MEMBLOCK_NODE_MAP to work properly\n");
1511 early_param("movable_node", cmdline_parse_movable_node);
1513 /* check which state of node_states will be changed when offline memory */
1514 static void node_states_check_changes_offline(unsigned long nr_pages,
1515 struct zone *zone, struct memory_notify *arg)
1517 struct pglist_data *pgdat = zone->zone_pgdat;
1518 unsigned long present_pages = 0;
1519 enum zone_type zt, zone_last = ZONE_NORMAL;
1522 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
1523 * contains nodes which have zones of 0...ZONE_NORMAL,
1524 * set zone_last to ZONE_NORMAL.
1526 * If we don't have HIGHMEM nor movable node,
1527 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
1528 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
1530 if (N_MEMORY == N_NORMAL_MEMORY)
1531 zone_last = ZONE_MOVABLE;
1534 * check whether node_states[N_NORMAL_MEMORY] will be changed.
1535 * If the memory to be offline is in a zone of 0...zone_last,
1536 * and it is the last present memory, 0...zone_last will
1537 * become empty after offline , thus we can determind we will
1538 * need to clear the node from node_states[N_NORMAL_MEMORY].
1540 for (zt = 0; zt <= zone_last; zt++)
1541 present_pages += pgdat->node_zones[zt].present_pages;
1542 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1543 arg->status_change_nid_normal = zone_to_nid(zone);
1545 arg->status_change_nid_normal = -1;
1547 #ifdef CONFIG_HIGHMEM
1549 * If we have movable node, node_states[N_HIGH_MEMORY]
1550 * contains nodes which have zones of 0...ZONE_HIGHMEM,
1551 * set zone_last to ZONE_HIGHMEM.
1553 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
1554 * contains nodes which have zones of 0...ZONE_MOVABLE,
1555 * set zone_last to ZONE_MOVABLE.
1557 zone_last = ZONE_HIGHMEM;
1558 if (N_MEMORY == N_HIGH_MEMORY)
1559 zone_last = ZONE_MOVABLE;
1561 for (; zt <= zone_last; zt++)
1562 present_pages += pgdat->node_zones[zt].present_pages;
1563 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1564 arg->status_change_nid_high = zone_to_nid(zone);
1566 arg->status_change_nid_high = -1;
1568 arg->status_change_nid_high = arg->status_change_nid_normal;
1572 * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE
1574 zone_last = ZONE_MOVABLE;
1577 * check whether node_states[N_HIGH_MEMORY] will be changed
1578 * If we try to offline the last present @nr_pages from the node,
1579 * we can determind we will need to clear the node from
1580 * node_states[N_HIGH_MEMORY].
1582 for (; zt <= zone_last; zt++)
1583 present_pages += pgdat->node_zones[zt].present_pages;
1584 if (nr_pages >= present_pages)
1585 arg->status_change_nid = zone_to_nid(zone);
1587 arg->status_change_nid = -1;
1590 static void node_states_clear_node(int node, struct memory_notify *arg)
1592 if (arg->status_change_nid_normal >= 0)
1593 node_clear_state(node, N_NORMAL_MEMORY);
1595 if ((N_MEMORY != N_NORMAL_MEMORY) &&
1596 (arg->status_change_nid_high >= 0))
1597 node_clear_state(node, N_HIGH_MEMORY);
1599 if ((N_MEMORY != N_HIGH_MEMORY) &&
1600 (arg->status_change_nid >= 0))
1601 node_clear_state(node, N_MEMORY);
1604 static int __ref __offline_pages(unsigned long start_pfn,
1605 unsigned long end_pfn)
1607 unsigned long pfn, nr_pages;
1608 long offlined_pages;
1610 unsigned long flags;
1611 unsigned long valid_start, valid_end;
1613 struct memory_notify arg;
1615 /* at least, alignment against pageblock is necessary */
1616 if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
1618 if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
1621 mem_hotplug_begin();
1623 /* This makes hotplug much easier...and readable.
1624 we assume this for now. .*/
1625 if (!test_pages_in_a_zone(start_pfn, end_pfn, &valid_start,
1631 zone = page_zone(pfn_to_page(valid_start));
1632 node = zone_to_nid(zone);
1633 nr_pages = end_pfn - start_pfn;
1635 /* set above range as isolated */
1636 ret = start_isolate_page_range(start_pfn, end_pfn,
1637 MIGRATE_MOVABLE, true);
1643 arg.start_pfn = start_pfn;
1644 arg.nr_pages = nr_pages;
1645 node_states_check_changes_offline(nr_pages, zone, &arg);
1647 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1648 ret = notifier_to_errno(ret);
1650 goto failed_removal;
1654 /* start memory hot removal */
1656 if (signal_pending(current))
1657 goto failed_removal;
1660 lru_add_drain_all();
1661 drain_all_pages(zone);
1663 pfn = scan_movable_pages(start_pfn, end_pfn);
1664 if (pfn) { /* We have movable pages */
1665 ret = do_migrate_range(pfn, end_pfn);
1670 * dissolve free hugepages in the memory block before doing offlining
1671 * actually in order to make hugetlbfs's object counting consistent.
1673 ret = dissolve_free_huge_pages(start_pfn, end_pfn);
1675 goto failed_removal;
1677 offlined_pages = check_pages_isolated(start_pfn, end_pfn);
1678 if (offlined_pages < 0)
1680 pr_info("Offlined Pages %ld\n", offlined_pages);
1681 /* Ok, all of our target is isolated.
1682 We cannot do rollback at this point. */
1683 offline_isolated_pages(start_pfn, end_pfn);
1684 /* reset pagetype flags and makes migrate type to be MOVABLE */
1685 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1686 /* removal success */
1687 adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages);
1688 zone->present_pages -= offlined_pages;
1690 pgdat_resize_lock(zone->zone_pgdat, &flags);
1691 zone->zone_pgdat->node_present_pages -= offlined_pages;
1692 pgdat_resize_unlock(zone->zone_pgdat, &flags);
1694 init_per_zone_wmark_min();
1696 if (!populated_zone(zone)) {
1697 zone_pcp_reset(zone);
1698 build_all_zonelists(NULL);
1700 zone_pcp_update(zone);
1702 node_states_clear_node(node, &arg);
1703 if (arg.status_change_nid >= 0) {
1705 kcompactd_stop(node);
1708 vm_total_pages = nr_free_pagecache_pages();
1709 writeback_set_ratelimit();
1711 memory_notify(MEM_OFFLINE, &arg);
1716 pr_debug("memory offlining [mem %#010llx-%#010llx] failed\n",
1717 (unsigned long long) start_pfn << PAGE_SHIFT,
1718 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
1719 memory_notify(MEM_CANCEL_OFFLINE, &arg);
1720 /* pushback to free area */
1721 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1726 int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1728 return __offline_pages(start_pfn, start_pfn + nr_pages);
1730 #endif /* CONFIG_MEMORY_HOTREMOVE */
1733 * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
1734 * @start_pfn: start pfn of the memory range
1735 * @end_pfn: end pfn of the memory range
1736 * @arg: argument passed to func
1737 * @func: callback for each memory section walked
1739 * This function walks through all present mem sections in range
1740 * [start_pfn, end_pfn) and call func on each mem section.
1742 * Returns the return value of func.
1744 int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn,
1745 void *arg, int (*func)(struct memory_block *, void *))
1747 struct memory_block *mem = NULL;
1748 struct mem_section *section;
1749 unsigned long pfn, section_nr;
1752 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1753 section_nr = pfn_to_section_nr(pfn);
1754 if (!present_section_nr(section_nr))
1757 section = __nr_to_section(section_nr);
1758 /* same memblock? */
1760 if ((section_nr >= mem->start_section_nr) &&
1761 (section_nr <= mem->end_section_nr))
1764 mem = find_memory_block_hinted(section, mem);
1768 ret = func(mem, arg);
1770 kobject_put(&mem->dev.kobj);
1776 kobject_put(&mem->dev.kobj);
1781 #ifdef CONFIG_MEMORY_HOTREMOVE
1782 static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
1784 int ret = !is_memblock_offlined(mem);
1786 if (unlikely(ret)) {
1787 phys_addr_t beginpa, endpa;
1789 beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
1790 endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1;
1791 pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n",
1798 static int check_cpu_on_node(pg_data_t *pgdat)
1802 for_each_present_cpu(cpu) {
1803 if (cpu_to_node(cpu) == pgdat->node_id)
1805 * the cpu on this node isn't removed, and we can't
1806 * offline this node.
1814 static void unmap_cpu_on_node(pg_data_t *pgdat)
1816 #ifdef CONFIG_ACPI_NUMA
1819 for_each_possible_cpu(cpu)
1820 if (cpu_to_node(cpu) == pgdat->node_id)
1821 numa_clear_node(cpu);
1825 static int check_and_unmap_cpu_on_node(pg_data_t *pgdat)
1829 ret = check_cpu_on_node(pgdat);
1834 * the node will be offlined when we come here, so we can clear
1835 * the cpu_to_node() now.
1838 unmap_cpu_on_node(pgdat);
1846 * Offline a node if all memory sections and cpus of the node are removed.
1848 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1849 * and online/offline operations before this call.
1851 void try_offline_node(int nid)
1853 pg_data_t *pgdat = NODE_DATA(nid);
1854 unsigned long start_pfn = pgdat->node_start_pfn;
1855 unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
1858 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1859 unsigned long section_nr = pfn_to_section_nr(pfn);
1861 if (!present_section_nr(section_nr))
1864 if (pfn_to_nid(pfn) != nid)
1868 * some memory sections of this node are not removed, and we
1869 * can't offline node now.
1874 if (check_and_unmap_cpu_on_node(pgdat))
1878 * all memory/cpu of this node are removed, we can offline this
1881 node_set_offline(nid);
1882 unregister_one_node(nid);
1884 EXPORT_SYMBOL(try_offline_node);
1889 * @start: physical address of the region to remove
1890 * @size: size of the region to remove
1892 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1893 * and online/offline operations before this call, as required by
1894 * try_offline_node().
1896 void __ref __remove_memory(int nid, u64 start, u64 size)
1900 BUG_ON(check_hotplug_memory_range(start, size));
1902 mem_hotplug_begin();
1905 * All memory blocks must be offlined before removing memory. Check
1906 * whether all memory blocks in question are offline and trigger a BUG()
1907 * if this is not the case.
1909 ret = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL,
1910 check_memblock_offlined_cb);
1914 /* remove memmap entry */
1915 firmware_map_remove(start, start + size, "System RAM");
1916 memblock_free(start, size);
1917 memblock_remove(start, size);
1919 arch_remove_memory(start, size, NULL);
1921 try_offline_node(nid);
1926 void remove_memory(int nid, u64 start, u64 size)
1928 lock_device_hotplug();
1929 __remove_memory(nid, start, size);
1930 unlock_device_hotplug();
1932 EXPORT_SYMBOL_GPL(remove_memory);
1933 #endif /* CONFIG_MEMORY_HOTREMOVE */