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/compaction.h>
37 #include <linux/rmap.h>
39 #include <asm/tlbflush.h>
44 * online_page_callback contains pointer to current page onlining function.
45 * Initially it is generic_online_page(). If it is required it could be
46 * changed by calling set_online_page_callback() for callback registration
47 * and restore_online_page_callback() for generic callback restore.
50 static void generic_online_page(struct page *page);
52 static online_page_callback_t online_page_callback = generic_online_page;
53 static DEFINE_MUTEX(online_page_callback_lock);
55 DEFINE_STATIC_PERCPU_RWSEM(mem_hotplug_lock);
57 void get_online_mems(void)
59 percpu_down_read(&mem_hotplug_lock);
62 void put_online_mems(void)
64 percpu_up_read(&mem_hotplug_lock);
67 bool movable_node_enabled = false;
69 #ifndef CONFIG_MEMORY_HOTPLUG_DEFAULT_ONLINE
70 bool memhp_auto_online;
72 bool memhp_auto_online = true;
74 EXPORT_SYMBOL_GPL(memhp_auto_online);
76 static int __init setup_memhp_default_state(char *str)
78 if (!strcmp(str, "online"))
79 memhp_auto_online = true;
80 else if (!strcmp(str, "offline"))
81 memhp_auto_online = false;
85 __setup("memhp_default_state=", setup_memhp_default_state);
87 void mem_hotplug_begin(void)
90 percpu_down_write(&mem_hotplug_lock);
93 void mem_hotplug_done(void)
95 percpu_up_write(&mem_hotplug_lock);
99 /* add this memory to iomem resource */
100 static struct resource *register_memory_resource(u64 start, u64 size)
102 struct resource *res, *conflict;
103 res = kzalloc(sizeof(struct resource), GFP_KERNEL);
105 return ERR_PTR(-ENOMEM);
107 res->name = "System RAM";
109 res->end = start + size - 1;
110 res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
111 conflict = request_resource_conflict(&iomem_resource, res);
113 pr_debug("System RAM resource %pR cannot be added\n", res);
115 return ERR_PTR(-EEXIST);
120 static void release_memory_resource(struct resource *res)
124 release_resource(res);
129 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
130 void get_page_bootmem(unsigned long info, struct page *page,
133 page->freelist = (void *)type;
134 SetPagePrivate(page);
135 set_page_private(page, info);
139 void put_page_bootmem(struct page *page)
143 type = (unsigned long) page->freelist;
144 BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE ||
145 type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE);
147 if (page_ref_dec_return(page) == 1) {
148 page->freelist = NULL;
149 ClearPagePrivate(page);
150 set_page_private(page, 0);
151 INIT_LIST_HEAD(&page->lru);
152 free_reserved_page(page);
156 #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
157 #ifndef CONFIG_SPARSEMEM_VMEMMAP
158 static void register_page_bootmem_info_section(unsigned long start_pfn)
160 unsigned long *usemap, mapsize, section_nr, i;
161 struct mem_section *ms;
162 struct page *page, *memmap;
164 section_nr = pfn_to_section_nr(start_pfn);
165 ms = __nr_to_section(section_nr);
167 /* Get section's memmap address */
168 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
171 * Get page for the memmap's phys address
172 * XXX: need more consideration for sparse_vmemmap...
174 page = virt_to_page(memmap);
175 mapsize = sizeof(struct page) * PAGES_PER_SECTION;
176 mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
178 /* remember memmap's page */
179 for (i = 0; i < mapsize; i++, page++)
180 get_page_bootmem(section_nr, page, SECTION_INFO);
182 usemap = ms->pageblock_flags;
183 page = virt_to_page(usemap);
185 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
187 for (i = 0; i < mapsize; i++, page++)
188 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
191 #else /* CONFIG_SPARSEMEM_VMEMMAP */
192 static void register_page_bootmem_info_section(unsigned long start_pfn)
194 unsigned long *usemap, mapsize, section_nr, i;
195 struct mem_section *ms;
196 struct page *page, *memmap;
198 section_nr = pfn_to_section_nr(start_pfn);
199 ms = __nr_to_section(section_nr);
201 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
203 register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION);
205 usemap = ms->pageblock_flags;
206 page = virt_to_page(usemap);
208 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
210 for (i = 0; i < mapsize; i++, page++)
211 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
213 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
215 void __init register_page_bootmem_info_node(struct pglist_data *pgdat)
217 unsigned long i, pfn, end_pfn, nr_pages;
218 int node = pgdat->node_id;
221 nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
222 page = virt_to_page(pgdat);
224 for (i = 0; i < nr_pages; i++, page++)
225 get_page_bootmem(node, page, NODE_INFO);
227 pfn = pgdat->node_start_pfn;
228 end_pfn = pgdat_end_pfn(pgdat);
230 /* register section info */
231 for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
233 * Some platforms can assign the same pfn to multiple nodes - on
234 * node0 as well as nodeN. To avoid registering a pfn against
235 * multiple nodes we check that this pfn does not already
236 * reside in some other nodes.
238 if (pfn_valid(pfn) && (early_pfn_to_nid(pfn) == node))
239 register_page_bootmem_info_section(pfn);
242 #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
244 static int __meminit __add_section(int nid, unsigned long phys_start_pfn,
245 struct vmem_altmap *altmap, bool want_memblock)
249 if (pfn_valid(phys_start_pfn))
252 ret = sparse_add_one_section(nid, phys_start_pfn, altmap);
259 return hotplug_memory_register(nid, __pfn_to_section(phys_start_pfn));
263 * Reasonably generic function for adding memory. It is
264 * expected that archs that support memory hotplug will
265 * call this function after deciding the zone to which to
268 int __ref __add_pages(int nid, unsigned long phys_start_pfn,
269 unsigned long nr_pages, struct vmem_altmap *altmap,
274 int start_sec, end_sec;
276 /* during initialize mem_map, align hot-added range to section */
277 start_sec = pfn_to_section_nr(phys_start_pfn);
278 end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
282 * Validate altmap is within bounds of the total request
284 if (altmap->base_pfn != phys_start_pfn
285 || vmem_altmap_offset(altmap) > nr_pages) {
286 pr_warn_once("memory add fail, invalid altmap\n");
293 for (i = start_sec; i <= end_sec; i++) {
294 err = __add_section(nid, section_nr_to_pfn(i), altmap,
298 * EEXIST is finally dealt with by ioresource collision
299 * check. see add_memory() => register_memory_resource()
300 * Warning will be printed if there is collision.
302 if (err && (err != -EEXIST))
307 vmemmap_populate_print_last();
312 #ifdef CONFIG_MEMORY_HOTREMOVE
313 /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
314 static unsigned long find_smallest_section_pfn(int nid, struct zone *zone,
315 unsigned long start_pfn,
316 unsigned long end_pfn)
318 struct mem_section *ms;
320 for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) {
321 ms = __pfn_to_section(start_pfn);
323 if (unlikely(!valid_section(ms)))
326 if (unlikely(pfn_to_nid(start_pfn) != nid))
329 if (zone && zone != page_zone(pfn_to_page(start_pfn)))
338 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
339 static unsigned long find_biggest_section_pfn(int nid, struct zone *zone,
340 unsigned long start_pfn,
341 unsigned long end_pfn)
343 struct mem_section *ms;
346 /* pfn is the end pfn of a memory section. */
348 for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) {
349 ms = __pfn_to_section(pfn);
351 if (unlikely(!valid_section(ms)))
354 if (unlikely(pfn_to_nid(pfn) != nid))
357 if (zone && zone != page_zone(pfn_to_page(pfn)))
366 static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
367 unsigned long end_pfn)
369 unsigned long zone_start_pfn = zone->zone_start_pfn;
370 unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */
371 unsigned long zone_end_pfn = z;
373 struct mem_section *ms;
374 int nid = zone_to_nid(zone);
376 zone_span_writelock(zone);
377 if (zone_start_pfn == start_pfn) {
379 * If the section is smallest section in the zone, it need
380 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
381 * In this case, we find second smallest valid mem_section
382 * for shrinking zone.
384 pfn = find_smallest_section_pfn(nid, zone, end_pfn,
387 zone->zone_start_pfn = pfn;
388 zone->spanned_pages = zone_end_pfn - pfn;
390 } else if (zone_end_pfn == end_pfn) {
392 * If the section is biggest section in the zone, it need
393 * shrink zone->spanned_pages.
394 * In this case, we find second biggest valid mem_section for
397 pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn,
400 zone->spanned_pages = pfn - zone_start_pfn + 1;
404 * The section is not biggest or smallest mem_section in the zone, it
405 * only creates a hole in the zone. So in this case, we need not
406 * change the zone. But perhaps, the zone has only hole data. Thus
407 * it check the zone has only hole or not.
409 pfn = zone_start_pfn;
410 for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) {
411 ms = __pfn_to_section(pfn);
413 if (unlikely(!valid_section(ms)))
416 if (page_zone(pfn_to_page(pfn)) != zone)
419 /* If the section is current section, it continues the loop */
420 if (start_pfn == pfn)
423 /* If we find valid section, we have nothing to do */
424 zone_span_writeunlock(zone);
428 /* The zone has no valid section */
429 zone->zone_start_pfn = 0;
430 zone->spanned_pages = 0;
431 zone_span_writeunlock(zone);
434 static void shrink_pgdat_span(struct pglist_data *pgdat,
435 unsigned long start_pfn, unsigned long end_pfn)
437 unsigned long pgdat_start_pfn = pgdat->node_start_pfn;
438 unsigned long p = pgdat_end_pfn(pgdat); /* pgdat_end_pfn namespace clash */
439 unsigned long pgdat_end_pfn = p;
441 struct mem_section *ms;
442 int nid = pgdat->node_id;
444 if (pgdat_start_pfn == start_pfn) {
446 * If the section is smallest section in the pgdat, it need
447 * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages.
448 * In this case, we find second smallest valid mem_section
449 * for shrinking zone.
451 pfn = find_smallest_section_pfn(nid, NULL, end_pfn,
454 pgdat->node_start_pfn = pfn;
455 pgdat->node_spanned_pages = pgdat_end_pfn - pfn;
457 } else if (pgdat_end_pfn == end_pfn) {
459 * If the section is biggest section in the pgdat, it need
460 * shrink pgdat->node_spanned_pages.
461 * In this case, we find second biggest valid mem_section for
464 pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn,
467 pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1;
471 * If the section is not biggest or smallest mem_section in the pgdat,
472 * it only creates a hole in the pgdat. So in this case, we need not
474 * But perhaps, the pgdat has only hole data. Thus it check the pgdat
475 * has only hole or not.
477 pfn = pgdat_start_pfn;
478 for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) {
479 ms = __pfn_to_section(pfn);
481 if (unlikely(!valid_section(ms)))
484 if (pfn_to_nid(pfn) != nid)
487 /* If the section is current section, it continues the loop */
488 if (start_pfn == pfn)
491 /* If we find valid section, we have nothing to do */
495 /* The pgdat has no valid section */
496 pgdat->node_start_pfn = 0;
497 pgdat->node_spanned_pages = 0;
500 static void __remove_zone(struct zone *zone, unsigned long start_pfn)
502 struct pglist_data *pgdat = zone->zone_pgdat;
503 int nr_pages = PAGES_PER_SECTION;
506 pgdat_resize_lock(zone->zone_pgdat, &flags);
507 shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
508 shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages);
509 pgdat_resize_unlock(zone->zone_pgdat, &flags);
512 static int __remove_section(struct zone *zone, struct mem_section *ms,
513 unsigned long map_offset, struct vmem_altmap *altmap)
515 unsigned long start_pfn;
519 if (!valid_section(ms))
522 ret = unregister_memory_section(ms);
526 scn_nr = __section_nr(ms);
527 start_pfn = section_nr_to_pfn((unsigned long)scn_nr);
528 __remove_zone(zone, start_pfn);
530 sparse_remove_one_section(zone, ms, map_offset, altmap);
535 * __remove_pages() - remove sections of pages from a zone
536 * @zone: zone from which pages need to be removed
537 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
538 * @nr_pages: number of pages to remove (must be multiple of section size)
539 * @altmap: alternative device page map or %NULL if default memmap is used
541 * Generic helper function to remove section mappings and sysfs entries
542 * for the section of the memory we are removing. Caller needs to make
543 * sure that pages are marked reserved and zones are adjust properly by
544 * calling offline_pages().
546 int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
547 unsigned long nr_pages, struct vmem_altmap *altmap)
550 unsigned long map_offset = 0;
551 int sections_to_remove, ret = 0;
553 /* In the ZONE_DEVICE case device driver owns the memory region */
554 if (is_dev_zone(zone)) {
556 map_offset = vmem_altmap_offset(altmap);
558 resource_size_t start, size;
560 start = phys_start_pfn << PAGE_SHIFT;
561 size = nr_pages * PAGE_SIZE;
563 ret = release_mem_region_adjustable(&iomem_resource, start,
566 resource_size_t endres = start + size - 1;
568 pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
569 &start, &endres, ret);
573 clear_zone_contiguous(zone);
576 * We can only remove entire sections
578 BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
579 BUG_ON(nr_pages % PAGES_PER_SECTION);
581 sections_to_remove = nr_pages / PAGES_PER_SECTION;
582 for (i = 0; i < sections_to_remove; i++) {
583 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
586 ret = __remove_section(zone, __pfn_to_section(pfn), map_offset,
593 set_zone_contiguous(zone);
597 #endif /* CONFIG_MEMORY_HOTREMOVE */
599 int set_online_page_callback(online_page_callback_t callback)
604 mutex_lock(&online_page_callback_lock);
606 if (online_page_callback == generic_online_page) {
607 online_page_callback = callback;
611 mutex_unlock(&online_page_callback_lock);
616 EXPORT_SYMBOL_GPL(set_online_page_callback);
618 int restore_online_page_callback(online_page_callback_t callback)
623 mutex_lock(&online_page_callback_lock);
625 if (online_page_callback == callback) {
626 online_page_callback = generic_online_page;
630 mutex_unlock(&online_page_callback_lock);
635 EXPORT_SYMBOL_GPL(restore_online_page_callback);
637 void __online_page_set_limits(struct page *page)
640 EXPORT_SYMBOL_GPL(__online_page_set_limits);
642 void __online_page_increment_counters(struct page *page)
644 adjust_managed_page_count(page, 1);
646 EXPORT_SYMBOL_GPL(__online_page_increment_counters);
648 void __online_page_free(struct page *page)
650 __free_reserved_page(page);
652 EXPORT_SYMBOL_GPL(__online_page_free);
654 static void generic_online_page(struct page *page)
656 __online_page_set_limits(page);
657 __online_page_increment_counters(page);
658 __online_page_free(page);
661 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
665 unsigned long onlined_pages = *(unsigned long *)arg;
668 if (PageReserved(pfn_to_page(start_pfn)))
669 for (i = 0; i < nr_pages; i++) {
670 page = pfn_to_page(start_pfn + i);
671 (*online_page_callback)(page);
675 online_mem_sections(start_pfn, start_pfn + nr_pages);
677 *(unsigned long *)arg = onlined_pages;
681 /* check which state of node_states will be changed when online memory */
682 static void node_states_check_changes_online(unsigned long nr_pages,
683 struct zone *zone, struct memory_notify *arg)
685 int nid = zone_to_nid(zone);
687 arg->status_change_nid = -1;
688 arg->status_change_nid_normal = -1;
689 arg->status_change_nid_high = -1;
691 if (!node_state(nid, N_MEMORY))
692 arg->status_change_nid = nid;
693 if (zone_idx(zone) <= ZONE_NORMAL && !node_state(nid, N_NORMAL_MEMORY))
694 arg->status_change_nid_normal = nid;
695 #ifdef CONFIG_HIGHMEM
696 if (zone_idx(zone) <= N_HIGH_MEMORY && !node_state(nid, N_HIGH_MEMORY))
697 arg->status_change_nid_high = nid;
701 static void node_states_set_node(int node, struct memory_notify *arg)
703 if (arg->status_change_nid_normal >= 0)
704 node_set_state(node, N_NORMAL_MEMORY);
706 if (arg->status_change_nid_high >= 0)
707 node_set_state(node, N_HIGH_MEMORY);
709 if (arg->status_change_nid >= 0)
710 node_set_state(node, N_MEMORY);
713 static void __meminit resize_zone_range(struct zone *zone, unsigned long start_pfn,
714 unsigned long nr_pages)
716 unsigned long old_end_pfn = zone_end_pfn(zone);
718 if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
719 zone->zone_start_pfn = start_pfn;
721 zone->spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - zone->zone_start_pfn;
724 static void __meminit resize_pgdat_range(struct pglist_data *pgdat, unsigned long start_pfn,
725 unsigned long nr_pages)
727 unsigned long old_end_pfn = pgdat_end_pfn(pgdat);
729 if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
730 pgdat->node_start_pfn = start_pfn;
732 pgdat->node_spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - pgdat->node_start_pfn;
735 void __ref move_pfn_range_to_zone(struct zone *zone, unsigned long start_pfn,
736 unsigned long nr_pages, struct vmem_altmap *altmap)
738 struct pglist_data *pgdat = zone->zone_pgdat;
739 int nid = pgdat->node_id;
742 clear_zone_contiguous(zone);
744 /* TODO Huh pgdat is irqsave while zone is not. It used to be like that before */
745 pgdat_resize_lock(pgdat, &flags);
746 zone_span_writelock(zone);
747 if (zone_is_empty(zone))
748 init_currently_empty_zone(zone, start_pfn, nr_pages);
749 resize_zone_range(zone, start_pfn, nr_pages);
750 zone_span_writeunlock(zone);
751 resize_pgdat_range(pgdat, start_pfn, nr_pages);
752 pgdat_resize_unlock(pgdat, &flags);
755 * TODO now we have a visible range of pages which are not associated
756 * with their zone properly. Not nice but set_pfnblock_flags_mask
757 * expects the zone spans the pfn range. All the pages in the range
758 * are reserved so nobody should be touching them so we should be safe
760 memmap_init_zone(nr_pages, nid, zone_idx(zone), start_pfn,
761 MEMMAP_HOTPLUG, altmap);
763 set_zone_contiguous(zone);
767 * Returns a default kernel memory zone for the given pfn range.
768 * If no kernel zone covers this pfn range it will automatically go
769 * to the ZONE_NORMAL.
771 static struct zone *default_kernel_zone_for_pfn(int nid, unsigned long start_pfn,
772 unsigned long nr_pages)
774 struct pglist_data *pgdat = NODE_DATA(nid);
777 for (zid = 0; zid <= ZONE_NORMAL; zid++) {
778 struct zone *zone = &pgdat->node_zones[zid];
780 if (zone_intersects(zone, start_pfn, nr_pages))
784 return &pgdat->node_zones[ZONE_NORMAL];
787 static inline struct zone *default_zone_for_pfn(int nid, unsigned long start_pfn,
788 unsigned long nr_pages)
790 struct zone *kernel_zone = default_kernel_zone_for_pfn(nid, start_pfn,
792 struct zone *movable_zone = &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
793 bool in_kernel = zone_intersects(kernel_zone, start_pfn, nr_pages);
794 bool in_movable = zone_intersects(movable_zone, start_pfn, nr_pages);
797 * We inherit the existing zone in a simple case where zones do not
798 * overlap in the given range
800 if (in_kernel ^ in_movable)
801 return (in_kernel) ? kernel_zone : movable_zone;
804 * If the range doesn't belong to any zone or two zones overlap in the
805 * given range then we use movable zone only if movable_node is
806 * enabled because we always online to a kernel zone by default.
808 return movable_node_enabled ? movable_zone : kernel_zone;
811 struct zone * zone_for_pfn_range(int online_type, int nid, unsigned start_pfn,
812 unsigned long nr_pages)
814 if (online_type == MMOP_ONLINE_KERNEL)
815 return default_kernel_zone_for_pfn(nid, start_pfn, nr_pages);
817 if (online_type == MMOP_ONLINE_MOVABLE)
818 return &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
820 return default_zone_for_pfn(nid, start_pfn, nr_pages);
824 * Associates the given pfn range with the given node and the zone appropriate
825 * for the given online type.
827 static struct zone * __meminit move_pfn_range(int online_type, int nid,
828 unsigned long start_pfn, unsigned long nr_pages)
832 zone = zone_for_pfn_range(online_type, nid, start_pfn, nr_pages);
833 move_pfn_range_to_zone(zone, start_pfn, nr_pages, NULL);
837 int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type)
840 unsigned long onlined_pages = 0;
842 int need_zonelists_rebuild = 0;
845 struct memory_notify arg;
846 struct memory_block *mem;
851 * We can't use pfn_to_nid() because nid might be stored in struct page
852 * which is not yet initialized. Instead, we find nid from memory block.
854 mem = find_memory_block(__pfn_to_section(pfn));
857 /* associate pfn range with the zone */
858 zone = move_pfn_range(online_type, nid, pfn, nr_pages);
861 arg.nr_pages = nr_pages;
862 node_states_check_changes_online(nr_pages, zone, &arg);
864 ret = memory_notify(MEM_GOING_ONLINE, &arg);
865 ret = notifier_to_errno(ret);
867 goto failed_addition;
870 * If this zone is not populated, then it is not in zonelist.
871 * This means the page allocator ignores this zone.
872 * So, zonelist must be updated after online.
874 if (!populated_zone(zone)) {
875 need_zonelists_rebuild = 1;
876 setup_zone_pageset(zone);
879 ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
882 if (need_zonelists_rebuild)
883 zone_pcp_reset(zone);
884 goto failed_addition;
887 zone->present_pages += onlined_pages;
889 pgdat_resize_lock(zone->zone_pgdat, &flags);
890 zone->zone_pgdat->node_present_pages += onlined_pages;
891 pgdat_resize_unlock(zone->zone_pgdat, &flags);
894 node_states_set_node(nid, &arg);
895 if (need_zonelists_rebuild)
896 build_all_zonelists(NULL);
898 zone_pcp_update(zone);
901 init_per_zone_wmark_min();
908 vm_total_pages = nr_free_pagecache_pages();
910 writeback_set_ratelimit();
913 memory_notify(MEM_ONLINE, &arg);
918 pr_debug("online_pages [mem %#010llx-%#010llx] failed\n",
919 (unsigned long long) pfn << PAGE_SHIFT,
920 (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1);
921 memory_notify(MEM_CANCEL_ONLINE, &arg);
925 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
927 static void reset_node_present_pages(pg_data_t *pgdat)
931 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
932 z->present_pages = 0;
934 pgdat->node_present_pages = 0;
937 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
938 static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
940 struct pglist_data *pgdat;
941 unsigned long start_pfn = PFN_DOWN(start);
943 pgdat = NODE_DATA(nid);
945 pgdat = arch_alloc_nodedata(nid);
949 arch_refresh_nodedata(nid, pgdat);
952 * Reset the nr_zones, order and classzone_idx before reuse.
953 * Note that kswapd will init kswapd_classzone_idx properly
954 * when it starts in the near future.
957 pgdat->kswapd_order = 0;
958 pgdat->kswapd_classzone_idx = 0;
961 /* we can use NODE_DATA(nid) from here */
963 pgdat->node_id = nid;
964 pgdat->node_start_pfn = start_pfn;
966 /* init node's zones as empty zones, we don't have any present pages.*/
967 free_area_init_core_hotplug(nid);
968 pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat);
971 * The node we allocated has no zone fallback lists. For avoiding
972 * to access not-initialized zonelist, build here.
974 build_all_zonelists(pgdat);
977 * When memory is hot-added, all the memory is in offline state. So
978 * clear all zones' present_pages because they will be updated in
979 * online_pages() and offline_pages().
981 reset_node_managed_pages(pgdat);
982 reset_node_present_pages(pgdat);
987 static void rollback_node_hotadd(int nid)
989 pg_data_t *pgdat = NODE_DATA(nid);
991 arch_refresh_nodedata(nid, NULL);
992 free_percpu(pgdat->per_cpu_nodestats);
993 arch_free_nodedata(pgdat);
999 * try_online_node - online a node if offlined
1001 * @start: start addr of the node
1002 * @set_node_online: Whether we want to online the node
1003 * called by cpu_up() to online a node without onlined memory.
1006 * 1 -> a new node has been allocated
1007 * 0 -> the node is already online
1008 * -ENOMEM -> the node could not be allocated
1010 static int __try_online_node(int nid, u64 start, bool set_node_online)
1015 if (node_online(nid))
1018 pgdat = hotadd_new_pgdat(nid, start);
1020 pr_err("Cannot online node %d due to NULL pgdat\n", nid);
1025 if (set_node_online) {
1026 node_set_online(nid);
1027 ret = register_one_node(nid);
1035 * Users of this function always want to online/register the node
1037 int try_online_node(int nid)
1041 mem_hotplug_begin();
1042 ret = __try_online_node(nid, 0, true);
1047 static int check_hotplug_memory_range(u64 start, u64 size)
1049 unsigned long block_sz = memory_block_size_bytes();
1050 u64 block_nr_pages = block_sz >> PAGE_SHIFT;
1051 u64 nr_pages = size >> PAGE_SHIFT;
1052 u64 start_pfn = PFN_DOWN(start);
1054 /* memory range must be block size aligned */
1055 if (!nr_pages || !IS_ALIGNED(start_pfn, block_nr_pages) ||
1056 !IS_ALIGNED(nr_pages, block_nr_pages)) {
1057 pr_err("Block size [%#lx] unaligned hotplug range: start %#llx, size %#llx",
1058 block_sz, start, size);
1065 static int online_memory_block(struct memory_block *mem, void *arg)
1067 return device_online(&mem->dev);
1071 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1072 * and online/offline operations (triggered e.g. by sysfs).
1074 * we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG
1076 int __ref add_memory_resource(int nid, struct resource *res)
1079 bool new_node = false;
1083 size = resource_size(res);
1085 ret = check_hotplug_memory_range(start, size);
1089 mem_hotplug_begin();
1092 * Add new range to memblock so that when hotadd_new_pgdat() is called
1093 * to allocate new pgdat, get_pfn_range_for_nid() will be able to find
1094 * this new range and calculate total pages correctly. The range will
1095 * be removed at hot-remove time.
1097 memblock_add_node(start, size, nid);
1099 ret = __try_online_node(nid, start, false);
1104 /* call arch's memory hotadd */
1105 ret = arch_add_memory(nid, start, size, NULL, true);
1110 /* If sysfs file of new node can't be created, cpu on the node
1111 * can't be hot-added. There is no rollback way now.
1112 * So, check by BUG_ON() to catch it reluctantly..
1113 * We online node here. We can't roll back from here.
1115 node_set_online(nid);
1116 ret = __register_one_node(nid);
1120 /* link memory sections under this node.*/
1121 ret = link_mem_sections(nid, PFN_DOWN(start), PFN_UP(start + size - 1));
1124 /* create new memmap entry */
1125 firmware_map_add_hotplug(start, start + size, "System RAM");
1127 /* device_online() will take the lock when calling online_pages() */
1130 /* online pages if requested */
1131 if (memhp_auto_online)
1132 walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1),
1133 NULL, online_memory_block);
1137 /* rollback pgdat allocation and others */
1139 rollback_node_hotadd(nid);
1140 memblock_remove(start, size);
1145 /* requires device_hotplug_lock, see add_memory_resource() */
1146 int __ref __add_memory(int nid, u64 start, u64 size)
1148 struct resource *res;
1151 res = register_memory_resource(start, size);
1153 return PTR_ERR(res);
1155 ret = add_memory_resource(nid, res);
1157 release_memory_resource(res);
1161 int add_memory(int nid, u64 start, u64 size)
1165 lock_device_hotplug();
1166 rc = __add_memory(nid, start, size);
1167 unlock_device_hotplug();
1171 EXPORT_SYMBOL_GPL(add_memory);
1173 #ifdef CONFIG_MEMORY_HOTREMOVE
1175 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1176 * set and the size of the free page is given by page_order(). Using this,
1177 * the function determines if the pageblock contains only free pages.
1178 * Due to buddy contraints, a free page at least the size of a pageblock will
1179 * be located at the start of the pageblock
1181 static inline int pageblock_free(struct page *page)
1183 return PageBuddy(page) && page_order(page) >= pageblock_order;
1186 /* Return the start of the next active pageblock after a given page */
1187 static struct page *next_active_pageblock(struct page *page)
1189 /* Ensure the starting page is pageblock-aligned */
1190 BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
1192 /* If the entire pageblock is free, move to the end of free page */
1193 if (pageblock_free(page)) {
1195 /* be careful. we don't have locks, page_order can be changed.*/
1196 order = page_order(page);
1197 if ((order < MAX_ORDER) && (order >= pageblock_order))
1198 return page + (1 << order);
1201 return page + pageblock_nr_pages;
1204 static bool is_pageblock_removable_nolock(struct page *page)
1210 * We have to be careful here because we are iterating over memory
1211 * sections which are not zone aware so we might end up outside of
1212 * the zone but still within the section.
1213 * We have to take care about the node as well. If the node is offline
1214 * its NODE_DATA will be NULL - see page_zone.
1216 if (!node_online(page_to_nid(page)))
1219 zone = page_zone(page);
1220 pfn = page_to_pfn(page);
1221 if (!zone_spans_pfn(zone, pfn))
1224 return !has_unmovable_pages(zone, page, 0, MIGRATE_MOVABLE, SKIP_HWPOISON);
1227 /* Checks if this range of memory is likely to be hot-removable. */
1228 bool is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
1230 struct page *page = pfn_to_page(start_pfn);
1231 struct page *end_page = page + nr_pages;
1233 /* Check the starting page of each pageblock within the range */
1234 for (; page < end_page; page = next_active_pageblock(page)) {
1235 if (!is_pageblock_removable_nolock(page))
1240 /* All pageblocks in the memory block are likely to be hot-removable */
1245 * Confirm all pages in a range [start, end) belong to the same zone.
1246 * When true, return its valid [start, end).
1248 int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn,
1249 unsigned long *valid_start, unsigned long *valid_end)
1251 unsigned long pfn, sec_end_pfn;
1252 unsigned long start, end;
1253 struct zone *zone = NULL;
1256 for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1);
1258 pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) {
1259 /* Make sure the memory section is present first */
1260 if (!present_section_nr(pfn_to_section_nr(pfn)))
1262 for (; pfn < sec_end_pfn && pfn < end_pfn;
1263 pfn += MAX_ORDER_NR_PAGES) {
1265 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1266 while ((i < MAX_ORDER_NR_PAGES) &&
1267 !pfn_valid_within(pfn + i))
1269 if (i == MAX_ORDER_NR_PAGES || pfn + i >= end_pfn)
1271 page = pfn_to_page(pfn + i);
1272 if (zone && page_zone(page) != zone)
1276 zone = page_zone(page);
1277 end = pfn + MAX_ORDER_NR_PAGES;
1282 *valid_start = start;
1283 *valid_end = min(end, end_pfn);
1291 * Scan pfn range [start,end) to find movable/migratable pages (LRU pages,
1292 * non-lru movable pages and hugepages). We scan pfn because it's much
1293 * easier than scanning over linked list. This function returns the pfn
1294 * of the first found movable page if it's found, otherwise 0.
1296 static unsigned long scan_movable_pages(unsigned long start, unsigned long end)
1300 for (pfn = start; pfn < end; pfn++) {
1301 if (pfn_valid(pfn)) {
1302 page = pfn_to_page(pfn);
1305 if (__PageMovable(page))
1307 if (PageHuge(page)) {
1308 if (hugepage_migration_supported(page_hstate(page)) &&
1309 page_huge_active(page))
1312 pfn = round_up(pfn + 1,
1313 1 << compound_order(page)) - 1;
1320 static struct page *new_node_page(struct page *page, unsigned long private)
1322 int nid = page_to_nid(page);
1323 nodemask_t nmask = node_states[N_MEMORY];
1326 * try to allocate from a different node but reuse this node if there
1327 * are no other online nodes to be used (e.g. we are offlining a part
1328 * of the only existing node)
1330 node_clear(nid, nmask);
1331 if (nodes_empty(nmask))
1332 node_set(nid, nmask);
1334 return new_page_nodemask(page, nid, &nmask);
1338 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1342 int not_managed = 0;
1346 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1347 if (!pfn_valid(pfn))
1349 page = pfn_to_page(pfn);
1351 if (PageHuge(page)) {
1352 struct page *head = compound_head(page);
1353 pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1;
1354 if (compound_order(head) > PFN_SECTION_SHIFT) {
1358 isolate_huge_page(page, &source);
1360 } else if (PageTransHuge(page))
1361 pfn = page_to_pfn(compound_head(page))
1362 + hpage_nr_pages(page) - 1;
1365 * HWPoison pages have elevated reference counts so the migration would
1366 * fail on them. It also doesn't make any sense to migrate them in the
1367 * first place. Still try to unmap such a page in case it is still mapped
1368 * (e.g. current hwpoison implementation doesn't unmap KSM pages but keep
1369 * the unmap as the catch all safety net).
1371 if (PageHWPoison(page)) {
1372 if (WARN_ON(PageLRU(page)))
1373 isolate_lru_page(page);
1374 if (page_mapped(page))
1375 try_to_unmap(page, TTU_IGNORE_MLOCK | TTU_IGNORE_ACCESS);
1379 if (!get_page_unless_zero(page))
1382 * We can skip free pages. And we can deal with pages on
1383 * LRU and non-lru movable pages.
1386 ret = isolate_lru_page(page);
1388 ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE);
1389 if (!ret) { /* Success */
1391 list_add_tail(&page->lru, &source);
1392 if (!__PageMovable(page))
1393 inc_node_page_state(page, NR_ISOLATED_ANON +
1394 page_is_file_cache(page));
1397 pr_warn("failed to isolate pfn %lx\n", pfn);
1398 dump_page(page, "isolation failed");
1400 /* Because we don't have big zone->lock. we should
1401 check this again here. */
1402 if (page_count(page)) {
1409 if (!list_empty(&source)) {
1411 putback_movable_pages(&source);
1415 /* Allocate a new page from the nearest neighbor node */
1416 ret = migrate_pages(&source, new_node_page, NULL, 0,
1417 MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
1419 list_for_each_entry(page, &source, lru) {
1420 pr_warn("migrating pfn %lx failed ret:%d ",
1421 page_to_pfn(page), ret);
1422 dump_page(page, "migration failure");
1424 putback_movable_pages(&source);
1432 * remove from free_area[] and mark all as Reserved.
1435 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1438 __offline_isolated_pages(start, start + nr_pages);
1443 offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
1445 walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
1446 offline_isolated_pages_cb);
1450 * Check all pages in range, recoreded as memory resource, are isolated.
1453 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1457 long offlined = *(long *)data;
1458 ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true);
1459 offlined = nr_pages;
1461 *(long *)data += offlined;
1466 check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
1471 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
1472 check_pages_isolated_cb);
1474 offlined = (long)ret;
1478 static int __init cmdline_parse_movable_node(char *p)
1480 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1481 movable_node_enabled = true;
1483 pr_warn("movable_node parameter depends on CONFIG_HAVE_MEMBLOCK_NODE_MAP to work properly\n");
1487 early_param("movable_node", cmdline_parse_movable_node);
1489 /* check which state of node_states will be changed when offline memory */
1490 static void node_states_check_changes_offline(unsigned long nr_pages,
1491 struct zone *zone, struct memory_notify *arg)
1493 struct pglist_data *pgdat = zone->zone_pgdat;
1494 unsigned long present_pages = 0;
1497 arg->status_change_nid = -1;
1498 arg->status_change_nid_normal = -1;
1499 arg->status_change_nid_high = -1;
1502 * Check whether node_states[N_NORMAL_MEMORY] will be changed.
1503 * If the memory to be offline is within the range
1504 * [0..ZONE_NORMAL], and it is the last present memory there,
1505 * the zones in that range will become empty after the offlining,
1506 * thus we can determine that we need to clear the node from
1507 * node_states[N_NORMAL_MEMORY].
1509 for (zt = 0; zt <= ZONE_NORMAL; zt++)
1510 present_pages += pgdat->node_zones[zt].present_pages;
1511 if (zone_idx(zone) <= ZONE_NORMAL && nr_pages >= present_pages)
1512 arg->status_change_nid_normal = zone_to_nid(zone);
1514 #ifdef CONFIG_HIGHMEM
1516 * node_states[N_HIGH_MEMORY] contains nodes which
1517 * have normal memory or high memory.
1518 * Here we add the present_pages belonging to ZONE_HIGHMEM.
1519 * If the zone is within the range of [0..ZONE_HIGHMEM), and
1520 * we determine that the zones in that range become empty,
1521 * we need to clear the node for N_HIGH_MEMORY.
1523 present_pages += pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1524 if (zone_idx(zone) <= ZONE_HIGHMEM && nr_pages >= present_pages)
1525 arg->status_change_nid_high = zone_to_nid(zone);
1529 * We have accounted the pages from [0..ZONE_NORMAL), and
1530 * in case of CONFIG_HIGHMEM the pages from ZONE_HIGHMEM
1532 * Here we count the possible pages from ZONE_MOVABLE.
1533 * If after having accounted all the pages, we see that the nr_pages
1534 * to be offlined is over or equal to the accounted pages,
1535 * we know that the node will become empty, and so, we can clear
1536 * it for N_MEMORY as well.
1538 present_pages += pgdat->node_zones[ZONE_MOVABLE].present_pages;
1540 if (nr_pages >= present_pages)
1541 arg->status_change_nid = zone_to_nid(zone);
1544 static void node_states_clear_node(int node, struct memory_notify *arg)
1546 if (arg->status_change_nid_normal >= 0)
1547 node_clear_state(node, N_NORMAL_MEMORY);
1549 if (arg->status_change_nid_high >= 0)
1550 node_clear_state(node, N_HIGH_MEMORY);
1552 if (arg->status_change_nid >= 0)
1553 node_clear_state(node, N_MEMORY);
1556 static int __ref __offline_pages(unsigned long start_pfn,
1557 unsigned long end_pfn)
1559 unsigned long pfn, nr_pages;
1560 long offlined_pages;
1562 unsigned long flags;
1563 unsigned long valid_start, valid_end;
1565 struct memory_notify arg;
1568 mem_hotplug_begin();
1570 /* This makes hotplug much easier...and readable.
1571 we assume this for now. .*/
1572 if (!test_pages_in_a_zone(start_pfn, end_pfn, &valid_start,
1576 reason = "multizone range";
1577 goto failed_removal;
1580 zone = page_zone(pfn_to_page(valid_start));
1581 node = zone_to_nid(zone);
1582 nr_pages = end_pfn - start_pfn;
1584 /* set above range as isolated */
1585 ret = start_isolate_page_range(start_pfn, end_pfn,
1587 SKIP_HWPOISON | REPORT_FAILURE);
1590 reason = "failure to isolate range";
1591 goto failed_removal;
1594 arg.start_pfn = start_pfn;
1595 arg.nr_pages = nr_pages;
1596 node_states_check_changes_offline(nr_pages, zone, &arg);
1598 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1599 ret = notifier_to_errno(ret);
1601 reason = "notifier failure";
1602 goto failed_removal_isolated;
1606 for (pfn = start_pfn; pfn;) {
1607 if (signal_pending(current)) {
1609 reason = "signal backoff";
1610 goto failed_removal_isolated;
1614 lru_add_drain_all();
1615 drain_all_pages(zone);
1617 pfn = scan_movable_pages(pfn, end_pfn);
1620 * TODO: fatal migration failures should bail
1623 do_migrate_range(pfn, end_pfn);
1628 * Dissolve free hugepages in the memory block before doing
1629 * offlining actually in order to make hugetlbfs's object
1630 * counting consistent.
1632 ret = dissolve_free_huge_pages(start_pfn, end_pfn);
1634 reason = "failure to dissolve huge pages";
1635 goto failed_removal_isolated;
1638 offlined_pages = check_pages_isolated(start_pfn, end_pfn);
1639 } while (offlined_pages < 0);
1641 pr_info("Offlined Pages %ld\n", offlined_pages);
1642 /* Ok, all of our target is isolated.
1643 We cannot do rollback at this point. */
1644 offline_isolated_pages(start_pfn, end_pfn);
1645 /* reset pagetype flags and makes migrate type to be MOVABLE */
1646 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1647 /* removal success */
1648 adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages);
1649 zone->present_pages -= offlined_pages;
1651 pgdat_resize_lock(zone->zone_pgdat, &flags);
1652 zone->zone_pgdat->node_present_pages -= offlined_pages;
1653 pgdat_resize_unlock(zone->zone_pgdat, &flags);
1655 init_per_zone_wmark_min();
1657 if (!populated_zone(zone)) {
1658 zone_pcp_reset(zone);
1659 build_all_zonelists(NULL);
1661 zone_pcp_update(zone);
1663 node_states_clear_node(node, &arg);
1664 if (arg.status_change_nid >= 0) {
1666 kcompactd_stop(node);
1669 vm_total_pages = nr_free_pagecache_pages();
1670 writeback_set_ratelimit();
1672 memory_notify(MEM_OFFLINE, &arg);
1676 failed_removal_isolated:
1677 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1679 pr_debug("memory offlining [mem %#010llx-%#010llx] failed due to %s\n",
1680 (unsigned long long) start_pfn << PAGE_SHIFT,
1681 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1,
1683 memory_notify(MEM_CANCEL_OFFLINE, &arg);
1684 /* pushback to free area */
1689 int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1691 return __offline_pages(start_pfn, start_pfn + nr_pages);
1693 #endif /* CONFIG_MEMORY_HOTREMOVE */
1696 * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
1697 * @start_pfn: start pfn of the memory range
1698 * @end_pfn: end pfn of the memory range
1699 * @arg: argument passed to func
1700 * @func: callback for each memory section walked
1702 * This function walks through all present mem sections in range
1703 * [start_pfn, end_pfn) and call func on each mem section.
1705 * Returns the return value of func.
1707 int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn,
1708 void *arg, int (*func)(struct memory_block *, void *))
1710 struct memory_block *mem = NULL;
1711 struct mem_section *section;
1712 unsigned long pfn, section_nr;
1715 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1716 section_nr = pfn_to_section_nr(pfn);
1717 if (!present_section_nr(section_nr))
1720 section = __nr_to_section(section_nr);
1721 /* same memblock? */
1723 if ((section_nr >= mem->start_section_nr) &&
1724 (section_nr <= mem->end_section_nr))
1727 mem = find_memory_block_hinted(section, mem);
1731 ret = func(mem, arg);
1733 kobject_put(&mem->dev.kobj);
1739 kobject_put(&mem->dev.kobj);
1744 #ifdef CONFIG_MEMORY_HOTREMOVE
1745 static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
1747 int ret = !is_memblock_offlined(mem);
1749 if (unlikely(ret)) {
1750 phys_addr_t beginpa, endpa;
1752 beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
1753 endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1;
1754 pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n",
1761 static int check_cpu_on_node(pg_data_t *pgdat)
1765 for_each_present_cpu(cpu) {
1766 if (cpu_to_node(cpu) == pgdat->node_id)
1768 * the cpu on this node isn't removed, and we can't
1769 * offline this node.
1781 * Offline a node if all memory sections and cpus of the node are removed.
1783 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1784 * and online/offline operations before this call.
1786 void try_offline_node(int nid)
1788 pg_data_t *pgdat = NODE_DATA(nid);
1789 unsigned long start_pfn = pgdat->node_start_pfn;
1790 unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
1793 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1794 unsigned long section_nr = pfn_to_section_nr(pfn);
1796 if (!present_section_nr(section_nr))
1799 if (pfn_to_nid(pfn) != nid)
1803 * some memory sections of this node are not removed, and we
1804 * can't offline node now.
1809 if (check_cpu_on_node(pgdat))
1813 * all memory/cpu of this node are removed, we can offline this
1816 node_set_offline(nid);
1817 unregister_one_node(nid);
1819 EXPORT_SYMBOL(try_offline_node);
1824 * @start: physical address of the region to remove
1825 * @size: size of the region to remove
1827 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1828 * and online/offline operations before this call, as required by
1829 * try_offline_node().
1831 void __ref __remove_memory(int nid, u64 start, u64 size)
1835 BUG_ON(check_hotplug_memory_range(start, size));
1837 mem_hotplug_begin();
1840 * All memory blocks must be offlined before removing memory. Check
1841 * whether all memory blocks in question are offline and trigger a BUG()
1842 * if this is not the case.
1844 ret = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL,
1845 check_memblock_offlined_cb);
1849 /* remove memmap entry */
1850 firmware_map_remove(start, start + size, "System RAM");
1851 memblock_free(start, size);
1852 memblock_remove(start, size);
1854 arch_remove_memory(nid, start, size, NULL);
1856 try_offline_node(nid);
1861 void remove_memory(int nid, u64 start, u64 size)
1863 lock_device_hotplug();
1864 __remove_memory(nid, start, size);
1865 unlock_device_hotplug();
1867 EXPORT_SYMBOL_GPL(remove_memory);
1868 #endif /* CONFIG_MEMORY_HOTREMOVE */