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(NODE_DATA(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 struct mem_section *ms;
326 for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) {
327 ms = __pfn_to_section(start_pfn);
329 if (unlikely(!valid_section(ms)))
332 if (unlikely(pfn_to_nid(start_pfn) != nid))
335 if (zone && zone != page_zone(pfn_to_page(start_pfn)))
344 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
345 static unsigned long find_biggest_section_pfn(int nid, struct zone *zone,
346 unsigned long start_pfn,
347 unsigned long end_pfn)
349 struct mem_section *ms;
352 /* pfn is the end pfn of a memory section. */
354 for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) {
355 ms = __pfn_to_section(pfn);
357 if (unlikely(!valid_section(ms)))
360 if (unlikely(pfn_to_nid(pfn) != nid))
363 if (zone && zone != page_zone(pfn_to_page(pfn)))
372 static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
373 unsigned long end_pfn)
375 unsigned long zone_start_pfn = zone->zone_start_pfn;
376 unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */
377 unsigned long zone_end_pfn = z;
379 struct mem_section *ms;
380 int nid = zone_to_nid(zone);
382 zone_span_writelock(zone);
383 if (zone_start_pfn == start_pfn) {
385 * If the section is smallest section in the zone, it need
386 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
387 * In this case, we find second smallest valid mem_section
388 * for shrinking zone.
390 pfn = find_smallest_section_pfn(nid, zone, end_pfn,
393 zone->zone_start_pfn = pfn;
394 zone->spanned_pages = zone_end_pfn - pfn;
396 } else if (zone_end_pfn == end_pfn) {
398 * If the section is biggest section in the zone, it need
399 * shrink zone->spanned_pages.
400 * In this case, we find second biggest valid mem_section for
403 pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn,
406 zone->spanned_pages = pfn - zone_start_pfn + 1;
410 * The section is not biggest or smallest mem_section in the zone, it
411 * only creates a hole in the zone. So in this case, we need not
412 * change the zone. But perhaps, the zone has only hole data. Thus
413 * it check the zone has only hole or not.
415 pfn = zone_start_pfn;
416 for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) {
417 ms = __pfn_to_section(pfn);
419 if (unlikely(!valid_section(ms)))
422 if (page_zone(pfn_to_page(pfn)) != zone)
425 /* If the section is current section, it continues the loop */
426 if (start_pfn == pfn)
429 /* If we find valid section, we have nothing to do */
430 zone_span_writeunlock(zone);
434 /* The zone has no valid section */
435 zone->zone_start_pfn = 0;
436 zone->spanned_pages = 0;
437 zone_span_writeunlock(zone);
440 static void shrink_pgdat_span(struct pglist_data *pgdat,
441 unsigned long start_pfn, unsigned long end_pfn)
443 unsigned long pgdat_start_pfn = pgdat->node_start_pfn;
444 unsigned long p = pgdat_end_pfn(pgdat); /* pgdat_end_pfn namespace clash */
445 unsigned long pgdat_end_pfn = p;
447 struct mem_section *ms;
448 int nid = pgdat->node_id;
450 if (pgdat_start_pfn == start_pfn) {
452 * If the section is smallest section in the pgdat, it need
453 * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages.
454 * In this case, we find second smallest valid mem_section
455 * for shrinking zone.
457 pfn = find_smallest_section_pfn(nid, NULL, end_pfn,
460 pgdat->node_start_pfn = pfn;
461 pgdat->node_spanned_pages = pgdat_end_pfn - pfn;
463 } else if (pgdat_end_pfn == end_pfn) {
465 * If the section is biggest section in the pgdat, it need
466 * shrink pgdat->node_spanned_pages.
467 * In this case, we find second biggest valid mem_section for
470 pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn,
473 pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1;
477 * If the section is not biggest or smallest mem_section in the pgdat,
478 * it only creates a hole in the pgdat. So in this case, we need not
480 * But perhaps, the pgdat has only hole data. Thus it check the pgdat
481 * has only hole or not.
483 pfn = pgdat_start_pfn;
484 for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) {
485 ms = __pfn_to_section(pfn);
487 if (unlikely(!valid_section(ms)))
490 if (pfn_to_nid(pfn) != nid)
493 /* If the section is current section, it continues the loop */
494 if (start_pfn == pfn)
497 /* If we find valid section, we have nothing to do */
501 /* The pgdat has no valid section */
502 pgdat->node_start_pfn = 0;
503 pgdat->node_spanned_pages = 0;
506 static void __remove_zone(struct zone *zone, unsigned long start_pfn)
508 struct pglist_data *pgdat = zone->zone_pgdat;
509 int nr_pages = PAGES_PER_SECTION;
512 pgdat_resize_lock(zone->zone_pgdat, &flags);
513 shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
514 shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages);
515 pgdat_resize_unlock(zone->zone_pgdat, &flags);
518 static int __remove_section(struct zone *zone, struct mem_section *ms,
519 unsigned long map_offset, struct vmem_altmap *altmap)
521 unsigned long start_pfn;
525 if (!valid_section(ms))
528 ret = unregister_memory_section(ms);
532 scn_nr = __section_nr(ms);
533 start_pfn = section_nr_to_pfn((unsigned long)scn_nr);
534 __remove_zone(zone, start_pfn);
536 sparse_remove_one_section(zone, ms, map_offset, altmap);
541 * __remove_pages() - remove sections of pages from a zone
542 * @zone: zone from which pages need to be removed
543 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
544 * @nr_pages: number of pages to remove (must be multiple of section size)
545 * @altmap: alternative device page map or %NULL if default memmap is used
547 * Generic helper function to remove section mappings and sysfs entries
548 * for the section of the memory we are removing. Caller needs to make
549 * sure that pages are marked reserved and zones are adjust properly by
550 * calling offline_pages().
552 int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
553 unsigned long nr_pages, struct vmem_altmap *altmap)
556 unsigned long map_offset = 0;
557 int sections_to_remove, ret = 0;
559 /* In the ZONE_DEVICE case device driver owns the memory region */
560 if (is_dev_zone(zone)) {
562 map_offset = vmem_altmap_offset(altmap);
564 resource_size_t start, size;
566 start = phys_start_pfn << PAGE_SHIFT;
567 size = nr_pages * PAGE_SIZE;
569 ret = release_mem_region_adjustable(&iomem_resource, start,
572 resource_size_t endres = start + size - 1;
574 pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
575 &start, &endres, ret);
579 clear_zone_contiguous(zone);
582 * We can only remove entire sections
584 BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
585 BUG_ON(nr_pages % PAGES_PER_SECTION);
587 sections_to_remove = nr_pages / PAGES_PER_SECTION;
588 for (i = 0; i < sections_to_remove; i++) {
589 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
592 ret = __remove_section(zone, __pfn_to_section(pfn), map_offset,
599 set_zone_contiguous(zone);
603 #endif /* CONFIG_MEMORY_HOTREMOVE */
605 int set_online_page_callback(online_page_callback_t callback)
610 mutex_lock(&online_page_callback_lock);
612 if (online_page_callback == generic_online_page) {
613 online_page_callback = callback;
617 mutex_unlock(&online_page_callback_lock);
622 EXPORT_SYMBOL_GPL(set_online_page_callback);
624 int restore_online_page_callback(online_page_callback_t callback)
629 mutex_lock(&online_page_callback_lock);
631 if (online_page_callback == callback) {
632 online_page_callback = generic_online_page;
636 mutex_unlock(&online_page_callback_lock);
641 EXPORT_SYMBOL_GPL(restore_online_page_callback);
643 void __online_page_set_limits(struct page *page)
646 EXPORT_SYMBOL_GPL(__online_page_set_limits);
648 void __online_page_increment_counters(struct page *page)
650 adjust_managed_page_count(page, 1);
652 EXPORT_SYMBOL_GPL(__online_page_increment_counters);
654 void __online_page_free(struct page *page)
656 __free_reserved_page(page);
658 EXPORT_SYMBOL_GPL(__online_page_free);
660 static void generic_online_page(struct page *page)
662 __online_page_set_limits(page);
663 __online_page_increment_counters(page);
664 __online_page_free(page);
667 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
671 unsigned long onlined_pages = *(unsigned long *)arg;
674 if (PageReserved(pfn_to_page(start_pfn)))
675 for (i = 0; i < nr_pages; i++) {
676 page = pfn_to_page(start_pfn + i);
677 (*online_page_callback)(page);
681 online_mem_sections(start_pfn, start_pfn + nr_pages);
683 *(unsigned long *)arg = onlined_pages;
687 /* check which state of node_states will be changed when online memory */
688 static void node_states_check_changes_online(unsigned long nr_pages,
689 struct zone *zone, struct memory_notify *arg)
691 int nid = zone_to_nid(zone);
692 enum zone_type zone_last = ZONE_NORMAL;
695 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
696 * contains nodes which have zones of 0...ZONE_NORMAL,
697 * set zone_last to ZONE_NORMAL.
699 * If we don't have HIGHMEM nor movable node,
700 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
701 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
703 if (N_MEMORY == N_NORMAL_MEMORY)
704 zone_last = ZONE_MOVABLE;
707 * if the memory to be online is in a zone of 0...zone_last, and
708 * the zones of 0...zone_last don't have memory before online, we will
709 * need to set the node to node_states[N_NORMAL_MEMORY] after
710 * the memory is online.
712 if (zone_idx(zone) <= zone_last && !node_state(nid, N_NORMAL_MEMORY))
713 arg->status_change_nid_normal = nid;
715 arg->status_change_nid_normal = -1;
717 #ifdef CONFIG_HIGHMEM
719 * If we have movable node, node_states[N_HIGH_MEMORY]
720 * contains nodes which have zones of 0...ZONE_HIGHMEM,
721 * set zone_last to ZONE_HIGHMEM.
723 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
724 * contains nodes which have zones of 0...ZONE_MOVABLE,
725 * set zone_last to ZONE_MOVABLE.
727 zone_last = ZONE_HIGHMEM;
728 if (N_MEMORY == N_HIGH_MEMORY)
729 zone_last = ZONE_MOVABLE;
731 if (zone_idx(zone) <= zone_last && !node_state(nid, N_HIGH_MEMORY))
732 arg->status_change_nid_high = nid;
734 arg->status_change_nid_high = -1;
736 arg->status_change_nid_high = arg->status_change_nid_normal;
740 * if the node don't have memory befor online, we will need to
741 * set the node to node_states[N_MEMORY] after the memory
744 if (!node_state(nid, N_MEMORY))
745 arg->status_change_nid = nid;
747 arg->status_change_nid = -1;
750 static void node_states_set_node(int node, struct memory_notify *arg)
752 if (arg->status_change_nid_normal >= 0)
753 node_set_state(node, N_NORMAL_MEMORY);
755 if (arg->status_change_nid_high >= 0)
756 node_set_state(node, N_HIGH_MEMORY);
758 node_set_state(node, N_MEMORY);
761 static void __meminit resize_zone_range(struct zone *zone, unsigned long start_pfn,
762 unsigned long nr_pages)
764 unsigned long old_end_pfn = zone_end_pfn(zone);
766 if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
767 zone->zone_start_pfn = start_pfn;
769 zone->spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - zone->zone_start_pfn;
772 static void __meminit resize_pgdat_range(struct pglist_data *pgdat, unsigned long start_pfn,
773 unsigned long nr_pages)
775 unsigned long old_end_pfn = pgdat_end_pfn(pgdat);
777 if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
778 pgdat->node_start_pfn = start_pfn;
780 pgdat->node_spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - pgdat->node_start_pfn;
783 void __ref move_pfn_range_to_zone(struct zone *zone, unsigned long start_pfn,
784 unsigned long nr_pages, struct vmem_altmap *altmap)
786 struct pglist_data *pgdat = zone->zone_pgdat;
787 int nid = pgdat->node_id;
790 if (zone_is_empty(zone))
791 init_currently_empty_zone(zone, start_pfn, nr_pages);
793 clear_zone_contiguous(zone);
795 /* TODO Huh pgdat is irqsave while zone is not. It used to be like that before */
796 pgdat_resize_lock(pgdat, &flags);
797 zone_span_writelock(zone);
798 resize_zone_range(zone, start_pfn, nr_pages);
799 zone_span_writeunlock(zone);
800 resize_pgdat_range(pgdat, start_pfn, nr_pages);
801 pgdat_resize_unlock(pgdat, &flags);
804 * TODO now we have a visible range of pages which are not associated
805 * with their zone properly. Not nice but set_pfnblock_flags_mask
806 * expects the zone spans the pfn range. All the pages in the range
807 * are reserved so nobody should be touching them so we should be safe
809 memmap_init_zone(nr_pages, nid, zone_idx(zone), start_pfn,
810 MEMMAP_HOTPLUG, altmap);
812 set_zone_contiguous(zone);
816 * Returns a default kernel memory zone for the given pfn range.
817 * If no kernel zone covers this pfn range it will automatically go
818 * to the ZONE_NORMAL.
820 static struct zone *default_kernel_zone_for_pfn(int nid, unsigned long start_pfn,
821 unsigned long nr_pages)
823 struct pglist_data *pgdat = NODE_DATA(nid);
826 for (zid = 0; zid <= ZONE_NORMAL; zid++) {
827 struct zone *zone = &pgdat->node_zones[zid];
829 if (zone_intersects(zone, start_pfn, nr_pages))
833 return &pgdat->node_zones[ZONE_NORMAL];
836 static inline struct zone *default_zone_for_pfn(int nid, unsigned long start_pfn,
837 unsigned long nr_pages)
839 struct zone *kernel_zone = default_kernel_zone_for_pfn(nid, start_pfn,
841 struct zone *movable_zone = &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
842 bool in_kernel = zone_intersects(kernel_zone, start_pfn, nr_pages);
843 bool in_movable = zone_intersects(movable_zone, start_pfn, nr_pages);
846 * We inherit the existing zone in a simple case where zones do not
847 * overlap in the given range
849 if (in_kernel ^ in_movable)
850 return (in_kernel) ? kernel_zone : movable_zone;
853 * If the range doesn't belong to any zone or two zones overlap in the
854 * given range then we use movable zone only if movable_node is
855 * enabled because we always online to a kernel zone by default.
857 return movable_node_enabled ? movable_zone : kernel_zone;
860 struct zone * zone_for_pfn_range(int online_type, int nid, unsigned start_pfn,
861 unsigned long nr_pages)
863 if (online_type == MMOP_ONLINE_KERNEL)
864 return default_kernel_zone_for_pfn(nid, start_pfn, nr_pages);
866 if (online_type == MMOP_ONLINE_MOVABLE)
867 return &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
869 return default_zone_for_pfn(nid, start_pfn, nr_pages);
873 * Associates the given pfn range with the given node and the zone appropriate
874 * for the given online type.
876 static struct zone * __meminit move_pfn_range(int online_type, int nid,
877 unsigned long start_pfn, unsigned long nr_pages)
881 zone = zone_for_pfn_range(online_type, nid, start_pfn, nr_pages);
882 move_pfn_range_to_zone(zone, start_pfn, nr_pages, NULL);
886 /* Must be protected by mem_hotplug_begin() or a device_lock */
887 int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type)
890 unsigned long onlined_pages = 0;
892 int need_zonelists_rebuild = 0;
895 struct memory_notify arg;
896 struct memory_block *mem;
899 * We can't use pfn_to_nid() because nid might be stored in struct page
900 * which is not yet initialized. Instead, we find nid from memory block.
902 mem = find_memory_block(__pfn_to_section(pfn));
905 /* associate pfn range with the zone */
906 zone = move_pfn_range(online_type, nid, pfn, nr_pages);
909 arg.nr_pages = nr_pages;
910 node_states_check_changes_online(nr_pages, zone, &arg);
912 ret = memory_notify(MEM_GOING_ONLINE, &arg);
913 ret = notifier_to_errno(ret);
915 goto failed_addition;
918 * If this zone is not populated, then it is not in zonelist.
919 * This means the page allocator ignores this zone.
920 * So, zonelist must be updated after online.
922 if (!populated_zone(zone)) {
923 need_zonelists_rebuild = 1;
924 setup_zone_pageset(zone);
927 ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
930 if (need_zonelists_rebuild)
931 zone_pcp_reset(zone);
932 goto failed_addition;
935 zone->present_pages += onlined_pages;
937 pgdat_resize_lock(zone->zone_pgdat, &flags);
938 zone->zone_pgdat->node_present_pages += onlined_pages;
939 pgdat_resize_unlock(zone->zone_pgdat, &flags);
942 node_states_set_node(nid, &arg);
943 if (need_zonelists_rebuild)
944 build_all_zonelists(NULL);
946 zone_pcp_update(zone);
949 init_per_zone_wmark_min();
956 vm_total_pages = nr_free_pagecache_pages();
958 writeback_set_ratelimit();
961 memory_notify(MEM_ONLINE, &arg);
965 pr_debug("online_pages [mem %#010llx-%#010llx] failed\n",
966 (unsigned long long) pfn << PAGE_SHIFT,
967 (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1);
968 memory_notify(MEM_CANCEL_ONLINE, &arg);
971 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
973 static void reset_node_present_pages(pg_data_t *pgdat)
977 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
978 z->present_pages = 0;
980 pgdat->node_present_pages = 0;
983 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
984 static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
986 struct pglist_data *pgdat;
987 unsigned long start_pfn = PFN_DOWN(start);
989 pgdat = NODE_DATA(nid);
991 pgdat = arch_alloc_nodedata(nid);
995 arch_refresh_nodedata(nid, pgdat);
998 * Reset the nr_zones, order and classzone_idx before reuse.
999 * Note that kswapd will init kswapd_classzone_idx properly
1000 * when it starts in the near future.
1002 pgdat->nr_zones = 0;
1003 pgdat->kswapd_order = 0;
1004 pgdat->kswapd_classzone_idx = 0;
1007 /* we can use NODE_DATA(nid) from here */
1009 pgdat->node_id = nid;
1010 pgdat->node_start_pfn = start_pfn;
1012 /* init node's zones as empty zones, we don't have any present pages.*/
1013 free_area_init_core_hotplug(nid);
1014 pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat);
1017 * The node we allocated has no zone fallback lists. For avoiding
1018 * to access not-initialized zonelist, build here.
1020 build_all_zonelists(pgdat);
1023 * When memory is hot-added, all the memory is in offline state. So
1024 * clear all zones' present_pages because they will be updated in
1025 * online_pages() and offline_pages().
1027 reset_node_managed_pages(pgdat);
1028 reset_node_present_pages(pgdat);
1033 static void rollback_node_hotadd(int nid)
1035 pg_data_t *pgdat = NODE_DATA(nid);
1037 arch_refresh_nodedata(nid, NULL);
1038 free_percpu(pgdat->per_cpu_nodestats);
1039 arch_free_nodedata(pgdat);
1045 * try_online_node - online a node if offlined
1047 * @start: start addr of the node
1048 * @set_node_online: Whether we want to online the node
1049 * called by cpu_up() to online a node without onlined memory.
1052 * 1 -> a new node has been allocated
1053 * 0 -> the node is already online
1054 * -ENOMEM -> the node could not be allocated
1056 static int __try_online_node(int nid, u64 start, bool set_node_online)
1061 if (node_online(nid))
1064 pgdat = hotadd_new_pgdat(nid, start);
1066 pr_err("Cannot online node %d due to NULL pgdat\n", nid);
1071 if (set_node_online) {
1072 node_set_online(nid);
1073 ret = register_one_node(nid);
1081 * Users of this function always want to online/register the node
1083 int try_online_node(int nid)
1087 mem_hotplug_begin();
1088 ret = __try_online_node(nid, 0, true);
1093 static int check_hotplug_memory_range(u64 start, u64 size)
1095 unsigned long block_sz = memory_block_size_bytes();
1096 u64 block_nr_pages = block_sz >> PAGE_SHIFT;
1097 u64 nr_pages = size >> PAGE_SHIFT;
1098 u64 start_pfn = PFN_DOWN(start);
1100 /* memory range must be block size aligned */
1101 if (!nr_pages || !IS_ALIGNED(start_pfn, block_nr_pages) ||
1102 !IS_ALIGNED(nr_pages, block_nr_pages)) {
1103 pr_err("Block size [%#lx] unaligned hotplug range: start %#llx, size %#llx",
1104 block_sz, start, size);
1111 static int online_memory_block(struct memory_block *mem, void *arg)
1113 return device_online(&mem->dev);
1116 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1117 int __ref add_memory_resource(int nid, struct resource *res, bool online)
1120 bool new_node = false;
1124 size = resource_size(res);
1126 ret = check_hotplug_memory_range(start, size);
1130 mem_hotplug_begin();
1133 * Add new range to memblock so that when hotadd_new_pgdat() is called
1134 * to allocate new pgdat, get_pfn_range_for_nid() will be able to find
1135 * this new range and calculate total pages correctly. The range will
1136 * be removed at hot-remove time.
1138 memblock_add_node(start, size, nid);
1140 ret = __try_online_node(nid, start, false);
1145 /* call arch's memory hotadd */
1146 ret = arch_add_memory(nid, start, size, NULL, true);
1151 /* If sysfs file of new node can't be created, cpu on the node
1152 * can't be hot-added. There is no rollback way now.
1153 * So, check by BUG_ON() to catch it reluctantly..
1154 * We online node here. We can't roll back from here.
1156 node_set_online(nid);
1157 ret = __register_one_node(nid);
1161 /* link memory sections under this node.*/
1162 ret = link_mem_sections(nid, PFN_DOWN(start), PFN_UP(start + size - 1));
1165 /* create new memmap entry */
1166 firmware_map_add_hotplug(start, start + size, "System RAM");
1168 /* online pages if requested */
1170 walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1),
1171 NULL, online_memory_block);
1176 /* rollback pgdat allocation and others */
1178 rollback_node_hotadd(nid);
1179 memblock_remove(start, size);
1185 EXPORT_SYMBOL_GPL(add_memory_resource);
1187 int __ref add_memory(int nid, u64 start, u64 size)
1189 struct resource *res;
1192 res = register_memory_resource(start, size);
1194 return PTR_ERR(res);
1196 ret = add_memory_resource(nid, res, memhp_auto_online);
1198 release_memory_resource(res);
1201 EXPORT_SYMBOL_GPL(add_memory);
1203 #ifdef CONFIG_MEMORY_HOTREMOVE
1205 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1206 * set and the size of the free page is given by page_order(). Using this,
1207 * the function determines if the pageblock contains only free pages.
1208 * Due to buddy contraints, a free page at least the size of a pageblock will
1209 * be located at the start of the pageblock
1211 static inline int pageblock_free(struct page *page)
1213 return PageBuddy(page) && page_order(page) >= pageblock_order;
1216 /* Return the start of the next active pageblock after a given page */
1217 static struct page *next_active_pageblock(struct page *page)
1219 /* Ensure the starting page is pageblock-aligned */
1220 BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
1222 /* If the entire pageblock is free, move to the end of free page */
1223 if (pageblock_free(page)) {
1225 /* be careful. we don't have locks, page_order can be changed.*/
1226 order = page_order(page);
1227 if ((order < MAX_ORDER) && (order >= pageblock_order))
1228 return page + (1 << order);
1231 return page + pageblock_nr_pages;
1234 static bool is_pageblock_removable_nolock(struct page *page)
1240 * We have to be careful here because we are iterating over memory
1241 * sections which are not zone aware so we might end up outside of
1242 * the zone but still within the section.
1243 * We have to take care about the node as well. If the node is offline
1244 * its NODE_DATA will be NULL - see page_zone.
1246 if (!node_online(page_to_nid(page)))
1249 zone = page_zone(page);
1250 pfn = page_to_pfn(page);
1251 if (!zone_spans_pfn(zone, pfn))
1254 return !has_unmovable_pages(zone, page, 0, MIGRATE_MOVABLE, true);
1257 /* Checks if this range of memory is likely to be hot-removable. */
1258 bool is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
1260 struct page *page = pfn_to_page(start_pfn);
1261 unsigned long end_pfn = min(start_pfn + nr_pages, zone_end_pfn(page_zone(page)));
1262 struct page *end_page = pfn_to_page(end_pfn);
1264 /* Check the starting page of each pageblock within the range */
1265 for (; page < end_page; page = next_active_pageblock(page)) {
1266 if (!is_pageblock_removable_nolock(page))
1271 /* All pageblocks in the memory block are likely to be hot-removable */
1276 * Confirm all pages in a range [start, end) belong to the same zone.
1277 * When true, return its valid [start, end).
1279 int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn,
1280 unsigned long *valid_start, unsigned long *valid_end)
1282 unsigned long pfn, sec_end_pfn;
1283 unsigned long start, end;
1284 struct zone *zone = NULL;
1287 for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1);
1289 pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) {
1290 /* Make sure the memory section is present first */
1291 if (!present_section_nr(pfn_to_section_nr(pfn)))
1293 for (; pfn < sec_end_pfn && pfn < end_pfn;
1294 pfn += MAX_ORDER_NR_PAGES) {
1296 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1297 while ((i < MAX_ORDER_NR_PAGES) &&
1298 !pfn_valid_within(pfn + i))
1300 if (i == MAX_ORDER_NR_PAGES || pfn + i >= end_pfn)
1302 /* Check if we got outside of the zone */
1303 if (zone && !zone_spans_pfn(zone, pfn + i))
1305 page = pfn_to_page(pfn + i);
1306 if (zone && page_zone(page) != zone)
1310 zone = page_zone(page);
1311 end = pfn + MAX_ORDER_NR_PAGES;
1316 *valid_start = start;
1317 *valid_end = min(end, end_pfn);
1325 * Scan pfn range [start,end) to find movable/migratable pages (LRU pages,
1326 * non-lru movable pages and hugepages). We scan pfn because it's much
1327 * easier than scanning over linked list. This function returns the pfn
1328 * of the first found movable page if it's found, otherwise 0.
1330 static unsigned long scan_movable_pages(unsigned long start, unsigned long end)
1334 for (pfn = start; pfn < end; pfn++) {
1335 struct page *page, *head;
1338 if (!pfn_valid(pfn))
1340 page = pfn_to_page(pfn);
1343 if (__PageMovable(page))
1346 if (!PageHuge(page))
1348 head = compound_head(page);
1349 if (hugepage_migration_supported(page_hstate(head)) &&
1350 page_huge_active(head))
1352 skip = (1 << compound_order(head)) - (page - head);
1358 static struct page *new_node_page(struct page *page, unsigned long private)
1360 int nid = page_to_nid(page);
1361 nodemask_t nmask = node_states[N_MEMORY];
1364 * try to allocate from a different node but reuse this node if there
1365 * are no other online nodes to be used (e.g. we are offlining a part
1366 * of the only existing node)
1368 node_clear(nid, nmask);
1369 if (nodes_empty(nmask))
1370 node_set(nid, nmask);
1372 return new_page_nodemask(page, nid, &nmask);
1375 #define NR_OFFLINE_AT_ONCE_PAGES (256)
1377 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1381 int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
1382 int not_managed = 0;
1386 for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
1387 if (!pfn_valid(pfn))
1389 page = pfn_to_page(pfn);
1391 if (PageHuge(page)) {
1392 struct page *head = compound_head(page);
1393 pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1;
1394 if (compound_order(head) > PFN_SECTION_SHIFT) {
1398 if (isolate_huge_page(page, &source))
1399 move_pages -= 1 << compound_order(head);
1401 } else if (PageTransHuge(page))
1402 pfn = page_to_pfn(compound_head(page))
1403 + hpage_nr_pages(page) - 1;
1406 * HWPoison pages have elevated reference counts so the migration would
1407 * fail on them. It also doesn't make any sense to migrate them in the
1408 * first place. Still try to unmap such a page in case it is still mapped
1409 * (e.g. current hwpoison implementation doesn't unmap KSM pages but keep
1410 * the unmap as the catch all safety net).
1412 if (PageHWPoison(page)) {
1413 if (WARN_ON(PageLRU(page)))
1414 isolate_lru_page(page);
1415 if (page_mapped(page))
1416 try_to_unmap(page, TTU_IGNORE_MLOCK | TTU_IGNORE_ACCESS);
1420 if (!get_page_unless_zero(page))
1423 * We can skip free pages. And we can deal with pages on
1424 * LRU and non-lru movable pages.
1427 ret = isolate_lru_page(page);
1429 ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE);
1430 if (!ret) { /* Success */
1432 list_add_tail(&page->lru, &source);
1434 if (!__PageMovable(page))
1435 inc_node_page_state(page, NR_ISOLATED_ANON +
1436 page_is_file_cache(page));
1439 #ifdef CONFIG_DEBUG_VM
1440 pr_alert("failed to isolate pfn %lx\n", pfn);
1441 dump_page(page, "isolation failed");
1444 /* Because we don't have big zone->lock. we should
1445 check this again here. */
1446 if (page_count(page)) {
1453 if (!list_empty(&source)) {
1455 putback_movable_pages(&source);
1459 /* Allocate a new page from the nearest neighbor node */
1460 ret = migrate_pages(&source, new_node_page, NULL, 0,
1461 MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
1463 putback_movable_pages(&source);
1470 * remove from free_area[] and mark all as Reserved.
1473 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1476 __offline_isolated_pages(start, start + nr_pages);
1481 offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
1483 walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
1484 offline_isolated_pages_cb);
1488 * Check all pages in range, recoreded as memory resource, are isolated.
1491 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1495 long offlined = *(long *)data;
1496 ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true);
1497 offlined = nr_pages;
1499 *(long *)data += offlined;
1504 check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
1509 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
1510 check_pages_isolated_cb);
1512 offlined = (long)ret;
1516 static int __init cmdline_parse_movable_node(char *p)
1518 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1519 movable_node_enabled = true;
1521 pr_warn("movable_node parameter depends on CONFIG_HAVE_MEMBLOCK_NODE_MAP to work properly\n");
1525 early_param("movable_node", cmdline_parse_movable_node);
1527 /* check which state of node_states will be changed when offline memory */
1528 static void node_states_check_changes_offline(unsigned long nr_pages,
1529 struct zone *zone, struct memory_notify *arg)
1531 struct pglist_data *pgdat = zone->zone_pgdat;
1532 unsigned long present_pages = 0;
1533 enum zone_type zt, zone_last = ZONE_NORMAL;
1536 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
1537 * contains nodes which have zones of 0...ZONE_NORMAL,
1538 * set zone_last to ZONE_NORMAL.
1540 * If we don't have HIGHMEM nor movable node,
1541 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
1542 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
1544 if (N_MEMORY == N_NORMAL_MEMORY)
1545 zone_last = ZONE_MOVABLE;
1548 * check whether node_states[N_NORMAL_MEMORY] will be changed.
1549 * If the memory to be offline is in a zone of 0...zone_last,
1550 * and it is the last present memory, 0...zone_last will
1551 * become empty after offline , thus we can determind we will
1552 * need to clear the node from node_states[N_NORMAL_MEMORY].
1554 for (zt = 0; zt <= zone_last; zt++)
1555 present_pages += pgdat->node_zones[zt].present_pages;
1556 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1557 arg->status_change_nid_normal = zone_to_nid(zone);
1559 arg->status_change_nid_normal = -1;
1561 #ifdef CONFIG_HIGHMEM
1563 * If we have movable node, node_states[N_HIGH_MEMORY]
1564 * contains nodes which have zones of 0...ZONE_HIGHMEM,
1565 * set zone_last to ZONE_HIGHMEM.
1567 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
1568 * contains nodes which have zones of 0...ZONE_MOVABLE,
1569 * set zone_last to ZONE_MOVABLE.
1571 zone_last = ZONE_HIGHMEM;
1572 if (N_MEMORY == N_HIGH_MEMORY)
1573 zone_last = ZONE_MOVABLE;
1575 for (; zt <= zone_last; zt++)
1576 present_pages += pgdat->node_zones[zt].present_pages;
1577 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1578 arg->status_change_nid_high = zone_to_nid(zone);
1580 arg->status_change_nid_high = -1;
1582 arg->status_change_nid_high = arg->status_change_nid_normal;
1586 * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE
1588 zone_last = ZONE_MOVABLE;
1591 * check whether node_states[N_HIGH_MEMORY] will be changed
1592 * If we try to offline the last present @nr_pages from the node,
1593 * we can determind we will need to clear the node from
1594 * node_states[N_HIGH_MEMORY].
1596 for (; zt <= zone_last; zt++)
1597 present_pages += pgdat->node_zones[zt].present_pages;
1598 if (nr_pages >= present_pages)
1599 arg->status_change_nid = zone_to_nid(zone);
1601 arg->status_change_nid = -1;
1604 static void node_states_clear_node(int node, struct memory_notify *arg)
1606 if (arg->status_change_nid_normal >= 0)
1607 node_clear_state(node, N_NORMAL_MEMORY);
1609 if ((N_MEMORY != N_NORMAL_MEMORY) &&
1610 (arg->status_change_nid_high >= 0))
1611 node_clear_state(node, N_HIGH_MEMORY);
1613 if ((N_MEMORY != N_HIGH_MEMORY) &&
1614 (arg->status_change_nid >= 0))
1615 node_clear_state(node, N_MEMORY);
1618 static int __ref __offline_pages(unsigned long start_pfn,
1619 unsigned long end_pfn)
1621 unsigned long pfn, nr_pages;
1622 long offlined_pages;
1624 unsigned long flags;
1625 unsigned long valid_start, valid_end;
1627 struct memory_notify arg;
1629 /* at least, alignment against pageblock is necessary */
1630 if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
1632 if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
1634 /* This makes hotplug much easier...and readable.
1635 we assume this for now. .*/
1636 if (!test_pages_in_a_zone(start_pfn, end_pfn, &valid_start, &valid_end))
1639 zone = page_zone(pfn_to_page(valid_start));
1640 node = zone_to_nid(zone);
1641 nr_pages = end_pfn - start_pfn;
1643 /* set above range as isolated */
1644 ret = start_isolate_page_range(start_pfn, end_pfn,
1645 MIGRATE_MOVABLE, true);
1649 arg.start_pfn = start_pfn;
1650 arg.nr_pages = nr_pages;
1651 node_states_check_changes_offline(nr_pages, zone, &arg);
1653 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1654 ret = notifier_to_errno(ret);
1656 goto failed_removal;
1660 /* start memory hot removal */
1662 if (signal_pending(current))
1663 goto failed_removal;
1666 lru_add_drain_all();
1667 drain_all_pages(zone);
1669 pfn = scan_movable_pages(start_pfn, end_pfn);
1670 if (pfn) { /* We have movable pages */
1671 ret = do_migrate_range(pfn, end_pfn);
1676 * dissolve free hugepages in the memory block before doing offlining
1677 * actually in order to make hugetlbfs's object counting consistent.
1679 ret = dissolve_free_huge_pages(start_pfn, end_pfn);
1681 goto failed_removal;
1683 offlined_pages = check_pages_isolated(start_pfn, end_pfn);
1684 if (offlined_pages < 0)
1686 pr_info("Offlined Pages %ld\n", offlined_pages);
1687 /* Ok, all of our target is isolated.
1688 We cannot do rollback at this point. */
1689 offline_isolated_pages(start_pfn, end_pfn);
1690 /* reset pagetype flags and makes migrate type to be MOVABLE */
1691 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1692 /* removal success */
1693 adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages);
1694 zone->present_pages -= offlined_pages;
1696 pgdat_resize_lock(zone->zone_pgdat, &flags);
1697 zone->zone_pgdat->node_present_pages -= offlined_pages;
1698 pgdat_resize_unlock(zone->zone_pgdat, &flags);
1700 init_per_zone_wmark_min();
1702 if (!populated_zone(zone)) {
1703 zone_pcp_reset(zone);
1704 build_all_zonelists(NULL);
1706 zone_pcp_update(zone);
1708 node_states_clear_node(node, &arg);
1709 if (arg.status_change_nid >= 0) {
1711 kcompactd_stop(node);
1714 vm_total_pages = nr_free_pagecache_pages();
1715 writeback_set_ratelimit();
1717 memory_notify(MEM_OFFLINE, &arg);
1721 pr_debug("memory offlining [mem %#010llx-%#010llx] failed\n",
1722 (unsigned long long) start_pfn << PAGE_SHIFT,
1723 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
1724 memory_notify(MEM_CANCEL_OFFLINE, &arg);
1725 /* pushback to free area */
1726 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1730 /* Must be protected by mem_hotplug_begin() or a device_lock */
1731 int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1733 return __offline_pages(start_pfn, start_pfn + nr_pages);
1735 #endif /* CONFIG_MEMORY_HOTREMOVE */
1738 * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
1739 * @start_pfn: start pfn of the memory range
1740 * @end_pfn: end pfn of the memory range
1741 * @arg: argument passed to func
1742 * @func: callback for each memory section walked
1744 * This function walks through all present mem sections in range
1745 * [start_pfn, end_pfn) and call func on each mem section.
1747 * Returns the return value of func.
1749 int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn,
1750 void *arg, int (*func)(struct memory_block *, void *))
1752 struct memory_block *mem = NULL;
1753 struct mem_section *section;
1754 unsigned long pfn, section_nr;
1757 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1758 section_nr = pfn_to_section_nr(pfn);
1759 if (!present_section_nr(section_nr))
1762 section = __nr_to_section(section_nr);
1763 /* same memblock? */
1765 if ((section_nr >= mem->start_section_nr) &&
1766 (section_nr <= mem->end_section_nr))
1769 mem = find_memory_block_hinted(section, mem);
1773 ret = func(mem, arg);
1775 kobject_put(&mem->dev.kobj);
1781 kobject_put(&mem->dev.kobj);
1786 #ifdef CONFIG_MEMORY_HOTREMOVE
1787 static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
1789 int ret = !is_memblock_offlined(mem);
1791 if (unlikely(ret)) {
1792 phys_addr_t beginpa, endpa;
1794 beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
1795 endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1;
1796 pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n",
1803 static int check_cpu_on_node(pg_data_t *pgdat)
1807 for_each_present_cpu(cpu) {
1808 if (cpu_to_node(cpu) == pgdat->node_id)
1810 * the cpu on this node isn't removed, and we can't
1811 * offline this node.
1819 static void unmap_cpu_on_node(pg_data_t *pgdat)
1821 #ifdef CONFIG_ACPI_NUMA
1824 for_each_possible_cpu(cpu)
1825 if (cpu_to_node(cpu) == pgdat->node_id)
1826 numa_clear_node(cpu);
1830 static int check_and_unmap_cpu_on_node(pg_data_t *pgdat)
1834 ret = check_cpu_on_node(pgdat);
1839 * the node will be offlined when we come here, so we can clear
1840 * the cpu_to_node() now.
1843 unmap_cpu_on_node(pgdat);
1851 * Offline a node if all memory sections and cpus of the node are removed.
1853 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1854 * and online/offline operations before this call.
1856 void try_offline_node(int nid)
1858 pg_data_t *pgdat = NODE_DATA(nid);
1859 unsigned long start_pfn = pgdat->node_start_pfn;
1860 unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
1863 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1864 unsigned long section_nr = pfn_to_section_nr(pfn);
1866 if (!present_section_nr(section_nr))
1869 if (pfn_to_nid(pfn) != nid)
1873 * some memory sections of this node are not removed, and we
1874 * can't offline node now.
1879 if (check_and_unmap_cpu_on_node(pgdat))
1883 * all memory/cpu of this node are removed, we can offline this
1886 node_set_offline(nid);
1887 unregister_one_node(nid);
1889 EXPORT_SYMBOL(try_offline_node);
1894 * @start: physical address of the region to remove
1895 * @size: size of the region to remove
1897 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1898 * and online/offline operations before this call, as required by
1899 * try_offline_node().
1901 void __ref remove_memory(int nid, u64 start, u64 size)
1905 BUG_ON(check_hotplug_memory_range(start, size));
1907 mem_hotplug_begin();
1910 * All memory blocks must be offlined before removing memory. Check
1911 * whether all memory blocks in question are offline and trigger a BUG()
1912 * if this is not the case.
1914 ret = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL,
1915 check_memblock_offlined_cb);
1919 /* remove memmap entry */
1920 firmware_map_remove(start, start + size, "System RAM");
1921 memblock_free(start, size);
1922 memblock_remove(start, size);
1924 arch_remove_memory(start, size, NULL);
1926 try_offline_node(nid);
1930 EXPORT_SYMBOL_GPL(remove_memory);
1931 #endif /* CONFIG_MEMORY_HOTREMOVE */