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>
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 if (conflict->desc == IORES_DESC_DEVICE_PRIVATE_MEMORY) {
114 pr_debug("Device unaddressable memory block "
115 "memory hotplug at %#010llx !\n",
116 (unsigned long long)start);
118 pr_debug("System RAM resource %pR cannot be added\n", res);
120 return ERR_PTR(-EEXIST);
125 static void release_memory_resource(struct resource *res)
129 release_resource(res);
134 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
135 void get_page_bootmem(unsigned long info, struct page *page,
138 page->freelist = (void *)type;
139 SetPagePrivate(page);
140 set_page_private(page, info);
144 void put_page_bootmem(struct page *page)
148 type = (unsigned long) page->freelist;
149 BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE ||
150 type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE);
152 if (page_ref_dec_return(page) == 1) {
153 page->freelist = NULL;
154 ClearPagePrivate(page);
155 set_page_private(page, 0);
156 INIT_LIST_HEAD(&page->lru);
157 free_reserved_page(page);
161 #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
162 #ifndef CONFIG_SPARSEMEM_VMEMMAP
163 static void register_page_bootmem_info_section(unsigned long start_pfn)
165 unsigned long *usemap, mapsize, section_nr, i;
166 struct mem_section *ms;
167 struct page *page, *memmap;
169 section_nr = pfn_to_section_nr(start_pfn);
170 ms = __nr_to_section(section_nr);
172 /* Get section's memmap address */
173 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
176 * Get page for the memmap's phys address
177 * XXX: need more consideration for sparse_vmemmap...
179 page = virt_to_page(memmap);
180 mapsize = sizeof(struct page) * PAGES_PER_SECTION;
181 mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
183 /* remember memmap's page */
184 for (i = 0; i < mapsize; i++, page++)
185 get_page_bootmem(section_nr, page, SECTION_INFO);
187 usemap = ms->pageblock_flags;
188 page = virt_to_page(usemap);
190 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
192 for (i = 0; i < mapsize; i++, page++)
193 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
196 #else /* CONFIG_SPARSEMEM_VMEMMAP */
197 static void register_page_bootmem_info_section(unsigned long start_pfn)
199 unsigned long *usemap, mapsize, section_nr, i;
200 struct mem_section *ms;
201 struct page *page, *memmap;
203 section_nr = pfn_to_section_nr(start_pfn);
204 ms = __nr_to_section(section_nr);
206 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
208 register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION);
210 usemap = ms->pageblock_flags;
211 page = virt_to_page(usemap);
213 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
215 for (i = 0; i < mapsize; i++, page++)
216 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
218 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
220 void __init register_page_bootmem_info_node(struct pglist_data *pgdat)
222 unsigned long i, pfn, end_pfn, nr_pages;
223 int node = pgdat->node_id;
226 nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
227 page = virt_to_page(pgdat);
229 for (i = 0; i < nr_pages; i++, page++)
230 get_page_bootmem(node, page, NODE_INFO);
232 pfn = pgdat->node_start_pfn;
233 end_pfn = pgdat_end_pfn(pgdat);
235 /* register section info */
236 for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
238 * Some platforms can assign the same pfn to multiple nodes - on
239 * node0 as well as nodeN. To avoid registering a pfn against
240 * multiple nodes we check that this pfn does not already
241 * reside in some other nodes.
243 if (pfn_valid(pfn) && (early_pfn_to_nid(pfn) == node))
244 register_page_bootmem_info_section(pfn);
247 #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
249 static int __meminit __add_section(int nid, unsigned long phys_start_pfn,
250 struct vmem_altmap *altmap, bool want_memblock)
254 if (pfn_valid(phys_start_pfn))
257 ret = sparse_add_one_section(NODE_DATA(nid), phys_start_pfn, altmap);
264 return hotplug_memory_register(nid, __pfn_to_section(phys_start_pfn));
268 * Reasonably generic function for adding memory. It is
269 * expected that archs that support memory hotplug will
270 * call this function after deciding the zone to which to
273 int __ref __add_pages(int nid, unsigned long phys_start_pfn,
274 unsigned long nr_pages, struct vmem_altmap *altmap,
279 int start_sec, end_sec;
281 /* during initialize mem_map, align hot-added range to section */
282 start_sec = pfn_to_section_nr(phys_start_pfn);
283 end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
287 * Validate altmap is within bounds of the total request
289 if (altmap->base_pfn != phys_start_pfn
290 || vmem_altmap_offset(altmap) > nr_pages) {
291 pr_warn_once("memory add fail, invalid altmap\n");
298 for (i = start_sec; i <= end_sec; i++) {
299 err = __add_section(nid, section_nr_to_pfn(i), altmap,
303 * EEXIST is finally dealt with by ioresource collision
304 * check. see add_memory() => register_memory_resource()
305 * Warning will be printed if there is collision.
307 if (err && (err != -EEXIST))
312 vmemmap_populate_print_last();
317 #ifdef CONFIG_MEMORY_HOTREMOVE
318 /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
319 static unsigned long find_smallest_section_pfn(int nid, struct zone *zone,
320 unsigned long start_pfn,
321 unsigned long end_pfn)
323 struct mem_section *ms;
325 for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) {
326 ms = __pfn_to_section(start_pfn);
328 if (unlikely(!valid_section(ms)))
331 if (unlikely(pfn_to_nid(start_pfn) != nid))
334 if (zone && zone != page_zone(pfn_to_page(start_pfn)))
343 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
344 static unsigned long find_biggest_section_pfn(int nid, struct zone *zone,
345 unsigned long start_pfn,
346 unsigned long end_pfn)
348 struct mem_section *ms;
351 /* pfn is the end pfn of a memory section. */
353 for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) {
354 ms = __pfn_to_section(pfn);
356 if (unlikely(!valid_section(ms)))
359 if (unlikely(pfn_to_nid(pfn) != nid))
362 if (zone && zone != page_zone(pfn_to_page(pfn)))
371 static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
372 unsigned long end_pfn)
374 unsigned long zone_start_pfn = zone->zone_start_pfn;
375 unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */
376 unsigned long zone_end_pfn = z;
378 struct mem_section *ms;
379 int nid = zone_to_nid(zone);
381 zone_span_writelock(zone);
382 if (zone_start_pfn == start_pfn) {
384 * If the section is smallest section in the zone, it need
385 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
386 * In this case, we find second smallest valid mem_section
387 * for shrinking zone.
389 pfn = find_smallest_section_pfn(nid, zone, end_pfn,
392 zone->zone_start_pfn = pfn;
393 zone->spanned_pages = zone_end_pfn - pfn;
395 } else if (zone_end_pfn == end_pfn) {
397 * If the section is biggest section in the zone, it need
398 * shrink zone->spanned_pages.
399 * In this case, we find second biggest valid mem_section for
402 pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn,
405 zone->spanned_pages = pfn - zone_start_pfn + 1;
409 * The section is not biggest or smallest mem_section in the zone, it
410 * only creates a hole in the zone. So in this case, we need not
411 * change the zone. But perhaps, the zone has only hole data. Thus
412 * it check the zone has only hole or not.
414 pfn = zone_start_pfn;
415 for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) {
416 ms = __pfn_to_section(pfn);
418 if (unlikely(!valid_section(ms)))
421 if (page_zone(pfn_to_page(pfn)) != zone)
424 /* If the section is current section, it continues the loop */
425 if (start_pfn == pfn)
428 /* If we find valid section, we have nothing to do */
429 zone_span_writeunlock(zone);
433 /* The zone has no valid section */
434 zone->zone_start_pfn = 0;
435 zone->spanned_pages = 0;
436 zone_span_writeunlock(zone);
439 static void shrink_pgdat_span(struct pglist_data *pgdat,
440 unsigned long start_pfn, unsigned long end_pfn)
442 unsigned long pgdat_start_pfn = pgdat->node_start_pfn;
443 unsigned long p = pgdat_end_pfn(pgdat); /* pgdat_end_pfn namespace clash */
444 unsigned long pgdat_end_pfn = p;
446 struct mem_section *ms;
447 int nid = pgdat->node_id;
449 if (pgdat_start_pfn == start_pfn) {
451 * If the section is smallest section in the pgdat, it need
452 * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages.
453 * In this case, we find second smallest valid mem_section
454 * for shrinking zone.
456 pfn = find_smallest_section_pfn(nid, NULL, end_pfn,
459 pgdat->node_start_pfn = pfn;
460 pgdat->node_spanned_pages = pgdat_end_pfn - pfn;
462 } else if (pgdat_end_pfn == end_pfn) {
464 * If the section is biggest section in the pgdat, it need
465 * shrink pgdat->node_spanned_pages.
466 * In this case, we find second biggest valid mem_section for
469 pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn,
472 pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1;
476 * If the section is not biggest or smallest mem_section in the pgdat,
477 * it only creates a hole in the pgdat. So in this case, we need not
479 * But perhaps, the pgdat has only hole data. Thus it check the pgdat
480 * has only hole or not.
482 pfn = pgdat_start_pfn;
483 for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) {
484 ms = __pfn_to_section(pfn);
486 if (unlikely(!valid_section(ms)))
489 if (pfn_to_nid(pfn) != nid)
492 /* If the section is current section, it continues the loop */
493 if (start_pfn == pfn)
496 /* If we find valid section, we have nothing to do */
500 /* The pgdat has no valid section */
501 pgdat->node_start_pfn = 0;
502 pgdat->node_spanned_pages = 0;
505 static void __remove_zone(struct zone *zone, unsigned long start_pfn)
507 struct pglist_data *pgdat = zone->zone_pgdat;
508 int nr_pages = PAGES_PER_SECTION;
511 pgdat_resize_lock(zone->zone_pgdat, &flags);
512 shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
513 shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages);
514 pgdat_resize_unlock(zone->zone_pgdat, &flags);
517 static int __remove_section(struct zone *zone, struct mem_section *ms,
518 unsigned long map_offset, struct vmem_altmap *altmap)
520 unsigned long start_pfn;
524 if (!valid_section(ms))
527 ret = unregister_memory_section(ms);
531 scn_nr = __section_nr(ms);
532 start_pfn = section_nr_to_pfn((unsigned long)scn_nr);
533 __remove_zone(zone, start_pfn);
535 sparse_remove_one_section(zone, ms, map_offset, altmap);
540 * __remove_pages() - remove sections of pages from a zone
541 * @zone: zone from which pages need to be removed
542 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
543 * @nr_pages: number of pages to remove (must be multiple of section size)
544 * @altmap: alternative device page map or %NULL if default memmap is used
546 * Generic helper function to remove section mappings and sysfs entries
547 * for the section of the memory we are removing. Caller needs to make
548 * sure that pages are marked reserved and zones are adjust properly by
549 * calling offline_pages().
551 int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
552 unsigned long nr_pages, struct vmem_altmap *altmap)
555 unsigned long map_offset = 0;
556 int sections_to_remove, ret = 0;
558 /* In the ZONE_DEVICE case device driver owns the memory region */
559 if (is_dev_zone(zone)) {
561 map_offset = vmem_altmap_offset(altmap);
563 resource_size_t start, size;
565 start = phys_start_pfn << PAGE_SHIFT;
566 size = nr_pages * PAGE_SIZE;
568 ret = release_mem_region_adjustable(&iomem_resource, start,
571 resource_size_t endres = start + size - 1;
573 pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
574 &start, &endres, ret);
578 clear_zone_contiguous(zone);
581 * We can only remove entire sections
583 BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
584 BUG_ON(nr_pages % PAGES_PER_SECTION);
586 sections_to_remove = nr_pages / PAGES_PER_SECTION;
587 for (i = 0; i < sections_to_remove; i++) {
588 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
591 ret = __remove_section(zone, __pfn_to_section(pfn), map_offset,
598 set_zone_contiguous(zone);
602 #endif /* CONFIG_MEMORY_HOTREMOVE */
604 int set_online_page_callback(online_page_callback_t callback)
609 mutex_lock(&online_page_callback_lock);
611 if (online_page_callback == generic_online_page) {
612 online_page_callback = callback;
616 mutex_unlock(&online_page_callback_lock);
621 EXPORT_SYMBOL_GPL(set_online_page_callback);
623 int restore_online_page_callback(online_page_callback_t callback)
628 mutex_lock(&online_page_callback_lock);
630 if (online_page_callback == callback) {
631 online_page_callback = generic_online_page;
635 mutex_unlock(&online_page_callback_lock);
640 EXPORT_SYMBOL_GPL(restore_online_page_callback);
642 void __online_page_set_limits(struct page *page)
645 EXPORT_SYMBOL_GPL(__online_page_set_limits);
647 void __online_page_increment_counters(struct page *page)
649 adjust_managed_page_count(page, 1);
651 EXPORT_SYMBOL_GPL(__online_page_increment_counters);
653 void __online_page_free(struct page *page)
655 __free_reserved_page(page);
657 EXPORT_SYMBOL_GPL(__online_page_free);
659 static void generic_online_page(struct page *page)
661 __online_page_set_limits(page);
662 __online_page_increment_counters(page);
663 __online_page_free(page);
666 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
670 unsigned long onlined_pages = *(unsigned long *)arg;
673 if (PageReserved(pfn_to_page(start_pfn)))
674 for (i = 0; i < nr_pages; i++) {
675 page = pfn_to_page(start_pfn + i);
676 (*online_page_callback)(page);
680 online_mem_sections(start_pfn, start_pfn + nr_pages);
682 *(unsigned long *)arg = onlined_pages;
686 /* check which state of node_states will be changed when online memory */
687 static void node_states_check_changes_online(unsigned long nr_pages,
688 struct zone *zone, struct memory_notify *arg)
690 int nid = zone_to_nid(zone);
691 enum zone_type zone_last = ZONE_NORMAL;
694 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
695 * contains nodes which have zones of 0...ZONE_NORMAL,
696 * set zone_last to ZONE_NORMAL.
698 * If we don't have HIGHMEM nor movable node,
699 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
700 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
702 if (N_MEMORY == N_NORMAL_MEMORY)
703 zone_last = ZONE_MOVABLE;
706 * if the memory to be online is in a zone of 0...zone_last, and
707 * the zones of 0...zone_last don't have memory before online, we will
708 * need to set the node to node_states[N_NORMAL_MEMORY] after
709 * the memory is online.
711 if (zone_idx(zone) <= zone_last && !node_state(nid, N_NORMAL_MEMORY))
712 arg->status_change_nid_normal = nid;
714 arg->status_change_nid_normal = -1;
716 #ifdef CONFIG_HIGHMEM
718 * If we have movable node, node_states[N_HIGH_MEMORY]
719 * contains nodes which have zones of 0...ZONE_HIGHMEM,
720 * set zone_last to ZONE_HIGHMEM.
722 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
723 * contains nodes which have zones of 0...ZONE_MOVABLE,
724 * set zone_last to ZONE_MOVABLE.
726 zone_last = ZONE_HIGHMEM;
727 if (N_MEMORY == N_HIGH_MEMORY)
728 zone_last = ZONE_MOVABLE;
730 if (zone_idx(zone) <= zone_last && !node_state(nid, N_HIGH_MEMORY))
731 arg->status_change_nid_high = nid;
733 arg->status_change_nid_high = -1;
735 arg->status_change_nid_high = arg->status_change_nid_normal;
739 * if the node don't have memory befor online, we will need to
740 * set the node to node_states[N_MEMORY] after the memory
743 if (!node_state(nid, N_MEMORY))
744 arg->status_change_nid = nid;
746 arg->status_change_nid = -1;
749 static void node_states_set_node(int node, struct memory_notify *arg)
751 if (arg->status_change_nid_normal >= 0)
752 node_set_state(node, N_NORMAL_MEMORY);
754 if (arg->status_change_nid_high >= 0)
755 node_set_state(node, N_HIGH_MEMORY);
757 node_set_state(node, N_MEMORY);
760 static void __meminit resize_zone_range(struct zone *zone, unsigned long start_pfn,
761 unsigned long nr_pages)
763 unsigned long old_end_pfn = zone_end_pfn(zone);
765 if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
766 zone->zone_start_pfn = start_pfn;
768 zone->spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - zone->zone_start_pfn;
771 static void __meminit resize_pgdat_range(struct pglist_data *pgdat, unsigned long start_pfn,
772 unsigned long nr_pages)
774 unsigned long old_end_pfn = pgdat_end_pfn(pgdat);
776 if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
777 pgdat->node_start_pfn = start_pfn;
779 pgdat->node_spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - pgdat->node_start_pfn;
782 void __ref move_pfn_range_to_zone(struct zone *zone, unsigned long start_pfn,
783 unsigned long nr_pages, struct vmem_altmap *altmap)
785 struct pglist_data *pgdat = zone->zone_pgdat;
786 int nid = pgdat->node_id;
789 if (zone_is_empty(zone))
790 init_currently_empty_zone(zone, start_pfn, nr_pages);
792 clear_zone_contiguous(zone);
794 /* TODO Huh pgdat is irqsave while zone is not. It used to be like that before */
795 pgdat_resize_lock(pgdat, &flags);
796 zone_span_writelock(zone);
797 resize_zone_range(zone, start_pfn, nr_pages);
798 zone_span_writeunlock(zone);
799 resize_pgdat_range(pgdat, start_pfn, nr_pages);
800 pgdat_resize_unlock(pgdat, &flags);
803 * TODO now we have a visible range of pages which are not associated
804 * with their zone properly. Not nice but set_pfnblock_flags_mask
805 * expects the zone spans the pfn range. All the pages in the range
806 * are reserved so nobody should be touching them so we should be safe
808 memmap_init_zone(nr_pages, nid, zone_idx(zone), start_pfn,
809 MEMMAP_HOTPLUG, altmap);
811 set_zone_contiguous(zone);
815 * Returns a default kernel memory zone for the given pfn range.
816 * If no kernel zone covers this pfn range it will automatically go
817 * to the ZONE_NORMAL.
819 static struct zone *default_kernel_zone_for_pfn(int nid, unsigned long start_pfn,
820 unsigned long nr_pages)
822 struct pglist_data *pgdat = NODE_DATA(nid);
825 for (zid = 0; zid <= ZONE_NORMAL; zid++) {
826 struct zone *zone = &pgdat->node_zones[zid];
828 if (zone_intersects(zone, start_pfn, nr_pages))
832 return &pgdat->node_zones[ZONE_NORMAL];
835 static inline struct zone *default_zone_for_pfn(int nid, unsigned long start_pfn,
836 unsigned long nr_pages)
838 struct zone *kernel_zone = default_kernel_zone_for_pfn(nid, start_pfn,
840 struct zone *movable_zone = &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
841 bool in_kernel = zone_intersects(kernel_zone, start_pfn, nr_pages);
842 bool in_movable = zone_intersects(movable_zone, start_pfn, nr_pages);
845 * We inherit the existing zone in a simple case where zones do not
846 * overlap in the given range
848 if (in_kernel ^ in_movable)
849 return (in_kernel) ? kernel_zone : movable_zone;
852 * If the range doesn't belong to any zone or two zones overlap in the
853 * given range then we use movable zone only if movable_node is
854 * enabled because we always online to a kernel zone by default.
856 return movable_node_enabled ? movable_zone : kernel_zone;
859 struct zone * zone_for_pfn_range(int online_type, int nid, unsigned start_pfn,
860 unsigned long nr_pages)
862 if (online_type == MMOP_ONLINE_KERNEL)
863 return default_kernel_zone_for_pfn(nid, start_pfn, nr_pages);
865 if (online_type == MMOP_ONLINE_MOVABLE)
866 return &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
868 return default_zone_for_pfn(nid, start_pfn, nr_pages);
872 * Associates the given pfn range with the given node and the zone appropriate
873 * for the given online type.
875 static struct zone * __meminit move_pfn_range(int online_type, int nid,
876 unsigned long start_pfn, unsigned long nr_pages)
880 zone = zone_for_pfn_range(online_type, nid, start_pfn, nr_pages);
881 move_pfn_range_to_zone(zone, start_pfn, nr_pages, NULL);
885 /* Must be protected by mem_hotplug_begin() or a device_lock */
886 int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type)
889 unsigned long onlined_pages = 0;
891 int need_zonelists_rebuild = 0;
894 struct memory_notify arg;
895 struct memory_block *mem;
898 * We can't use pfn_to_nid() because nid might be stored in struct page
899 * which is not yet initialized. Instead, we find nid from memory block.
901 mem = find_memory_block(__pfn_to_section(pfn));
904 /* associate pfn range with the zone */
905 zone = move_pfn_range(online_type, nid, pfn, nr_pages);
908 arg.nr_pages = nr_pages;
909 node_states_check_changes_online(nr_pages, zone, &arg);
911 ret = memory_notify(MEM_GOING_ONLINE, &arg);
912 ret = notifier_to_errno(ret);
914 goto failed_addition;
917 * If this zone is not populated, then it is not in zonelist.
918 * This means the page allocator ignores this zone.
919 * So, zonelist must be updated after online.
921 if (!populated_zone(zone)) {
922 need_zonelists_rebuild = 1;
923 setup_zone_pageset(zone);
926 ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
929 if (need_zonelists_rebuild)
930 zone_pcp_reset(zone);
931 goto failed_addition;
934 zone->present_pages += onlined_pages;
936 pgdat_resize_lock(zone->zone_pgdat, &flags);
937 zone->zone_pgdat->node_present_pages += onlined_pages;
938 pgdat_resize_unlock(zone->zone_pgdat, &flags);
941 node_states_set_node(nid, &arg);
942 if (need_zonelists_rebuild)
943 build_all_zonelists(NULL);
945 zone_pcp_update(zone);
948 init_per_zone_wmark_min();
955 vm_total_pages = nr_free_pagecache_pages();
957 writeback_set_ratelimit();
960 memory_notify(MEM_ONLINE, &arg);
964 pr_debug("online_pages [mem %#010llx-%#010llx] failed\n",
965 (unsigned long long) pfn << PAGE_SHIFT,
966 (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1);
967 memory_notify(MEM_CANCEL_ONLINE, &arg);
970 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
972 static void reset_node_present_pages(pg_data_t *pgdat)
976 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
977 z->present_pages = 0;
979 pgdat->node_present_pages = 0;
982 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
983 static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
985 struct pglist_data *pgdat;
986 unsigned long start_pfn = PFN_DOWN(start);
988 pgdat = NODE_DATA(nid);
990 pgdat = arch_alloc_nodedata(nid);
994 arch_refresh_nodedata(nid, pgdat);
997 * Reset the nr_zones, order and classzone_idx before reuse.
998 * Note that kswapd will init kswapd_classzone_idx properly
999 * when it starts in the near future.
1001 pgdat->nr_zones = 0;
1002 pgdat->kswapd_order = 0;
1003 pgdat->kswapd_classzone_idx = 0;
1006 /* we can use NODE_DATA(nid) from here */
1008 pgdat->node_id = nid;
1009 pgdat->node_start_pfn = start_pfn;
1011 /* init node's zones as empty zones, we don't have any present pages.*/
1012 free_area_init_core_hotplug(nid);
1013 pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat);
1016 * The node we allocated has no zone fallback lists. For avoiding
1017 * to access not-initialized zonelist, build here.
1019 build_all_zonelists(pgdat);
1022 * When memory is hot-added, all the memory is in offline state. So
1023 * clear all zones' present_pages because they will be updated in
1024 * online_pages() and offline_pages().
1026 reset_node_managed_pages(pgdat);
1027 reset_node_present_pages(pgdat);
1032 static void rollback_node_hotadd(int nid)
1034 pg_data_t *pgdat = NODE_DATA(nid);
1036 arch_refresh_nodedata(nid, NULL);
1037 free_percpu(pgdat->per_cpu_nodestats);
1038 arch_free_nodedata(pgdat);
1044 * try_online_node - online a node if offlined
1046 * @start: start addr of the node
1047 * @set_node_online: Whether we want to online the node
1048 * called by cpu_up() to online a node without onlined memory.
1051 * 1 -> a new node has been allocated
1052 * 0 -> the node is already online
1053 * -ENOMEM -> the node could not be allocated
1055 static int __try_online_node(int nid, u64 start, bool set_node_online)
1060 if (node_online(nid))
1063 pgdat = hotadd_new_pgdat(nid, start);
1065 pr_err("Cannot online node %d due to NULL pgdat\n", nid);
1070 if (set_node_online) {
1071 node_set_online(nid);
1072 ret = register_one_node(nid);
1080 * Users of this function always want to online/register the node
1082 int try_online_node(int nid)
1086 mem_hotplug_begin();
1087 ret = __try_online_node(nid, 0, true);
1092 static int check_hotplug_memory_range(u64 start, u64 size)
1094 unsigned long block_sz = memory_block_size_bytes();
1095 u64 block_nr_pages = block_sz >> PAGE_SHIFT;
1096 u64 nr_pages = size >> PAGE_SHIFT;
1097 u64 start_pfn = PFN_DOWN(start);
1099 /* memory range must be block size aligned */
1100 if (!nr_pages || !IS_ALIGNED(start_pfn, block_nr_pages) ||
1101 !IS_ALIGNED(nr_pages, block_nr_pages)) {
1102 pr_err("Block size [%#lx] unaligned hotplug range: start %#llx, size %#llx",
1103 block_sz, start, size);
1110 static int online_memory_block(struct memory_block *mem, void *arg)
1112 return device_online(&mem->dev);
1115 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1116 int __ref add_memory_resource(int nid, struct resource *res, bool online)
1119 bool new_node = false;
1123 size = resource_size(res);
1125 ret = check_hotplug_memory_range(start, size);
1129 mem_hotplug_begin();
1132 * Add new range to memblock so that when hotadd_new_pgdat() is called
1133 * to allocate new pgdat, get_pfn_range_for_nid() will be able to find
1134 * this new range and calculate total pages correctly. The range will
1135 * be removed at hot-remove time.
1137 memblock_add_node(start, size, nid);
1139 ret = __try_online_node(nid, start, false);
1144 /* call arch's memory hotadd */
1145 ret = arch_add_memory(nid, start, size, NULL, true);
1150 /* If sysfs file of new node can't be created, cpu on the node
1151 * can't be hot-added. There is no rollback way now.
1152 * So, check by BUG_ON() to catch it reluctantly..
1153 * We online node here. We can't roll back from here.
1155 node_set_online(nid);
1156 ret = __register_one_node(nid);
1160 /* link memory sections under this node.*/
1161 ret = link_mem_sections(nid, PFN_DOWN(start), PFN_UP(start + size - 1));
1164 /* create new memmap entry */
1165 firmware_map_add_hotplug(start, start + size, "System RAM");
1167 /* online pages if requested */
1169 walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1),
1170 NULL, online_memory_block);
1175 /* rollback pgdat allocation and others */
1177 rollback_node_hotadd(nid);
1178 memblock_remove(start, size);
1184 EXPORT_SYMBOL_GPL(add_memory_resource);
1186 int __ref add_memory(int nid, u64 start, u64 size)
1188 struct resource *res;
1191 res = register_memory_resource(start, size);
1193 return PTR_ERR(res);
1195 ret = add_memory_resource(nid, res, memhp_auto_online);
1197 release_memory_resource(res);
1200 EXPORT_SYMBOL_GPL(add_memory);
1202 #ifdef CONFIG_MEMORY_HOTREMOVE
1204 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1205 * set and the size of the free page is given by page_order(). Using this,
1206 * the function determines if the pageblock contains only free pages.
1207 * Due to buddy contraints, a free page at least the size of a pageblock will
1208 * be located at the start of the pageblock
1210 static inline int pageblock_free(struct page *page)
1212 return PageBuddy(page) && page_order(page) >= pageblock_order;
1215 /* Return the start of the next active pageblock after a given page */
1216 static struct page *next_active_pageblock(struct page *page)
1218 /* Ensure the starting page is pageblock-aligned */
1219 BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
1221 /* If the entire pageblock is free, move to the end of free page */
1222 if (pageblock_free(page)) {
1224 /* be careful. we don't have locks, page_order can be changed.*/
1225 order = page_order(page);
1226 if ((order < MAX_ORDER) && (order >= pageblock_order))
1227 return page + (1 << order);
1230 return page + pageblock_nr_pages;
1233 static bool is_pageblock_removable_nolock(struct page *page)
1239 * We have to be careful here because we are iterating over memory
1240 * sections which are not zone aware so we might end up outside of
1241 * the zone but still within the section.
1242 * We have to take care about the node as well. If the node is offline
1243 * its NODE_DATA will be NULL - see page_zone.
1245 if (!node_online(page_to_nid(page)))
1248 zone = page_zone(page);
1249 pfn = page_to_pfn(page);
1250 if (!zone_spans_pfn(zone, pfn))
1253 return !has_unmovable_pages(zone, page, 0, MIGRATE_MOVABLE, true);
1256 /* Checks if this range of memory is likely to be hot-removable. */
1257 bool is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
1259 struct page *page = pfn_to_page(start_pfn);
1260 struct page *end_page = page + nr_pages;
1262 /* Check the starting page of each pageblock within the range */
1263 for (; page < end_page; page = next_active_pageblock(page)) {
1264 if (!is_pageblock_removable_nolock(page))
1269 /* All pageblocks in the memory block are likely to be hot-removable */
1274 * Confirm all pages in a range [start, end) belong to the same zone.
1275 * When true, return its valid [start, end).
1277 int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn,
1278 unsigned long *valid_start, unsigned long *valid_end)
1280 unsigned long pfn, sec_end_pfn;
1281 unsigned long start, end;
1282 struct zone *zone = NULL;
1285 for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1);
1287 pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) {
1288 /* Make sure the memory section is present first */
1289 if (!present_section_nr(pfn_to_section_nr(pfn)))
1291 for (; pfn < sec_end_pfn && pfn < end_pfn;
1292 pfn += MAX_ORDER_NR_PAGES) {
1294 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1295 while ((i < MAX_ORDER_NR_PAGES) &&
1296 !pfn_valid_within(pfn + i))
1298 if (i == MAX_ORDER_NR_PAGES || pfn + i >= end_pfn)
1300 page = pfn_to_page(pfn + i);
1301 if (zone && page_zone(page) != zone)
1305 zone = page_zone(page);
1306 end = pfn + MAX_ORDER_NR_PAGES;
1311 *valid_start = start;
1312 *valid_end = min(end, end_pfn);
1320 * Scan pfn range [start,end) to find movable/migratable pages (LRU pages,
1321 * non-lru movable pages and hugepages). We scan pfn because it's much
1322 * easier than scanning over linked list. This function returns the pfn
1323 * of the first found movable page if it's found, otherwise 0.
1325 static unsigned long scan_movable_pages(unsigned long start, unsigned long end)
1329 for (pfn = start; pfn < end; pfn++) {
1330 if (pfn_valid(pfn)) {
1331 page = pfn_to_page(pfn);
1334 if (__PageMovable(page))
1336 if (PageHuge(page)) {
1337 if (hugepage_migration_supported(page_hstate(page)) &&
1338 page_huge_active(page))
1341 pfn = round_up(pfn + 1,
1342 1 << compound_order(page)) - 1;
1349 static struct page *new_node_page(struct page *page, unsigned long private)
1351 int nid = page_to_nid(page);
1352 nodemask_t nmask = node_states[N_MEMORY];
1355 * try to allocate from a different node but reuse this node if there
1356 * are no other online nodes to be used (e.g. we are offlining a part
1357 * of the only existing node)
1359 node_clear(nid, nmask);
1360 if (nodes_empty(nmask))
1361 node_set(nid, nmask);
1363 return new_page_nodemask(page, nid, &nmask);
1366 #define NR_OFFLINE_AT_ONCE_PAGES (256)
1368 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1372 int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
1373 int not_managed = 0;
1377 for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
1378 if (!pfn_valid(pfn))
1380 page = pfn_to_page(pfn);
1382 if (PageHuge(page)) {
1383 struct page *head = compound_head(page);
1384 pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1;
1385 if (compound_order(head) > PFN_SECTION_SHIFT) {
1389 if (isolate_huge_page(page, &source))
1390 move_pages -= 1 << compound_order(head);
1392 } else if (PageTransHuge(page))
1393 pfn = page_to_pfn(compound_head(page))
1394 + hpage_nr_pages(page) - 1;
1396 if (!get_page_unless_zero(page))
1399 * We can skip free pages. And we can deal with pages on
1400 * LRU and non-lru movable pages.
1403 ret = isolate_lru_page(page);
1405 ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE);
1406 if (!ret) { /* Success */
1408 list_add_tail(&page->lru, &source);
1410 if (!__PageMovable(page))
1411 inc_node_page_state(page, NR_ISOLATED_ANON +
1412 page_is_file_cache(page));
1415 #ifdef CONFIG_DEBUG_VM
1416 pr_alert("failed to isolate pfn %lx\n", pfn);
1417 dump_page(page, "isolation failed");
1420 /* Because we don't have big zone->lock. we should
1421 check this again here. */
1422 if (page_count(page)) {
1429 if (!list_empty(&source)) {
1431 putback_movable_pages(&source);
1435 /* Allocate a new page from the nearest neighbor node */
1436 ret = migrate_pages(&source, new_node_page, NULL, 0,
1437 MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
1439 putback_movable_pages(&source);
1446 * remove from free_area[] and mark all as Reserved.
1449 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1452 __offline_isolated_pages(start, start + nr_pages);
1457 offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
1459 walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
1460 offline_isolated_pages_cb);
1464 * Check all pages in range, recoreded as memory resource, are isolated.
1467 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1471 long offlined = *(long *)data;
1472 ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true);
1473 offlined = nr_pages;
1475 *(long *)data += offlined;
1480 check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
1485 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
1486 check_pages_isolated_cb);
1488 offlined = (long)ret;
1492 static int __init cmdline_parse_movable_node(char *p)
1494 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1495 movable_node_enabled = true;
1497 pr_warn("movable_node parameter depends on CONFIG_HAVE_MEMBLOCK_NODE_MAP to work properly\n");
1501 early_param("movable_node", cmdline_parse_movable_node);
1503 /* check which state of node_states will be changed when offline memory */
1504 static void node_states_check_changes_offline(unsigned long nr_pages,
1505 struct zone *zone, struct memory_notify *arg)
1507 struct pglist_data *pgdat = zone->zone_pgdat;
1508 unsigned long present_pages = 0;
1509 enum zone_type zt, zone_last = ZONE_NORMAL;
1512 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
1513 * contains nodes which have zones of 0...ZONE_NORMAL,
1514 * set zone_last to ZONE_NORMAL.
1516 * If we don't have HIGHMEM nor movable node,
1517 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
1518 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
1520 if (N_MEMORY == N_NORMAL_MEMORY)
1521 zone_last = ZONE_MOVABLE;
1524 * check whether node_states[N_NORMAL_MEMORY] will be changed.
1525 * If the memory to be offline is in a zone of 0...zone_last,
1526 * and it is the last present memory, 0...zone_last will
1527 * become empty after offline , thus we can determind we will
1528 * need to clear the node from node_states[N_NORMAL_MEMORY].
1530 for (zt = 0; zt <= zone_last; zt++)
1531 present_pages += pgdat->node_zones[zt].present_pages;
1532 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1533 arg->status_change_nid_normal = zone_to_nid(zone);
1535 arg->status_change_nid_normal = -1;
1537 #ifdef CONFIG_HIGHMEM
1539 * If we have movable node, node_states[N_HIGH_MEMORY]
1540 * contains nodes which have zones of 0...ZONE_HIGHMEM,
1541 * set zone_last to ZONE_HIGHMEM.
1543 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
1544 * contains nodes which have zones of 0...ZONE_MOVABLE,
1545 * set zone_last to ZONE_MOVABLE.
1547 zone_last = ZONE_HIGHMEM;
1548 if (N_MEMORY == N_HIGH_MEMORY)
1549 zone_last = ZONE_MOVABLE;
1551 for (; zt <= zone_last; zt++)
1552 present_pages += pgdat->node_zones[zt].present_pages;
1553 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1554 arg->status_change_nid_high = zone_to_nid(zone);
1556 arg->status_change_nid_high = -1;
1558 arg->status_change_nid_high = arg->status_change_nid_normal;
1562 * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE
1564 zone_last = ZONE_MOVABLE;
1567 * check whether node_states[N_HIGH_MEMORY] will be changed
1568 * If we try to offline the last present @nr_pages from the node,
1569 * we can determind we will need to clear the node from
1570 * node_states[N_HIGH_MEMORY].
1572 for (; zt <= zone_last; zt++)
1573 present_pages += pgdat->node_zones[zt].present_pages;
1574 if (nr_pages >= present_pages)
1575 arg->status_change_nid = zone_to_nid(zone);
1577 arg->status_change_nid = -1;
1580 static void node_states_clear_node(int node, struct memory_notify *arg)
1582 if (arg->status_change_nid_normal >= 0)
1583 node_clear_state(node, N_NORMAL_MEMORY);
1585 if ((N_MEMORY != N_NORMAL_MEMORY) &&
1586 (arg->status_change_nid_high >= 0))
1587 node_clear_state(node, N_HIGH_MEMORY);
1589 if ((N_MEMORY != N_HIGH_MEMORY) &&
1590 (arg->status_change_nid >= 0))
1591 node_clear_state(node, N_MEMORY);
1594 static int __ref __offline_pages(unsigned long start_pfn,
1595 unsigned long end_pfn)
1597 unsigned long pfn, nr_pages;
1598 long offlined_pages;
1600 unsigned long flags;
1601 unsigned long valid_start, valid_end;
1603 struct memory_notify arg;
1605 /* at least, alignment against pageblock is necessary */
1606 if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
1608 if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
1610 /* This makes hotplug much easier...and readable.
1611 we assume this for now. .*/
1612 if (!test_pages_in_a_zone(start_pfn, end_pfn, &valid_start, &valid_end))
1615 zone = page_zone(pfn_to_page(valid_start));
1616 node = zone_to_nid(zone);
1617 nr_pages = end_pfn - start_pfn;
1619 /* set above range as isolated */
1620 ret = start_isolate_page_range(start_pfn, end_pfn,
1621 MIGRATE_MOVABLE, true);
1625 arg.start_pfn = start_pfn;
1626 arg.nr_pages = nr_pages;
1627 node_states_check_changes_offline(nr_pages, zone, &arg);
1629 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1630 ret = notifier_to_errno(ret);
1632 goto failed_removal;
1636 /* start memory hot removal */
1638 if (signal_pending(current))
1639 goto failed_removal;
1642 lru_add_drain_all();
1643 drain_all_pages(zone);
1645 pfn = scan_movable_pages(start_pfn, end_pfn);
1646 if (pfn) { /* We have movable pages */
1647 ret = do_migrate_range(pfn, end_pfn);
1652 * dissolve free hugepages in the memory block before doing offlining
1653 * actually in order to make hugetlbfs's object counting consistent.
1655 ret = dissolve_free_huge_pages(start_pfn, end_pfn);
1657 goto failed_removal;
1659 offlined_pages = check_pages_isolated(start_pfn, end_pfn);
1660 if (offlined_pages < 0)
1662 pr_info("Offlined Pages %ld\n", offlined_pages);
1663 /* Ok, all of our target is isolated.
1664 We cannot do rollback at this point. */
1665 offline_isolated_pages(start_pfn, end_pfn);
1666 /* reset pagetype flags and makes migrate type to be MOVABLE */
1667 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1668 /* removal success */
1669 adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages);
1670 zone->present_pages -= offlined_pages;
1672 pgdat_resize_lock(zone->zone_pgdat, &flags);
1673 zone->zone_pgdat->node_present_pages -= offlined_pages;
1674 pgdat_resize_unlock(zone->zone_pgdat, &flags);
1676 init_per_zone_wmark_min();
1678 if (!populated_zone(zone)) {
1679 zone_pcp_reset(zone);
1680 build_all_zonelists(NULL);
1682 zone_pcp_update(zone);
1684 node_states_clear_node(node, &arg);
1685 if (arg.status_change_nid >= 0) {
1687 kcompactd_stop(node);
1690 vm_total_pages = nr_free_pagecache_pages();
1691 writeback_set_ratelimit();
1693 memory_notify(MEM_OFFLINE, &arg);
1697 pr_debug("memory offlining [mem %#010llx-%#010llx] failed\n",
1698 (unsigned long long) start_pfn << PAGE_SHIFT,
1699 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
1700 memory_notify(MEM_CANCEL_OFFLINE, &arg);
1701 /* pushback to free area */
1702 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1706 /* Must be protected by mem_hotplug_begin() or a device_lock */
1707 int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1709 return __offline_pages(start_pfn, start_pfn + nr_pages);
1711 #endif /* CONFIG_MEMORY_HOTREMOVE */
1714 * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
1715 * @start_pfn: start pfn of the memory range
1716 * @end_pfn: end pfn of the memory range
1717 * @arg: argument passed to func
1718 * @func: callback for each memory section walked
1720 * This function walks through all present mem sections in range
1721 * [start_pfn, end_pfn) and call func on each mem section.
1723 * Returns the return value of func.
1725 int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn,
1726 void *arg, int (*func)(struct memory_block *, void *))
1728 struct memory_block *mem = NULL;
1729 struct mem_section *section;
1730 unsigned long pfn, section_nr;
1733 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1734 section_nr = pfn_to_section_nr(pfn);
1735 if (!present_section_nr(section_nr))
1738 section = __nr_to_section(section_nr);
1739 /* same memblock? */
1741 if ((section_nr >= mem->start_section_nr) &&
1742 (section_nr <= mem->end_section_nr))
1745 mem = find_memory_block_hinted(section, mem);
1749 ret = func(mem, arg);
1751 kobject_put(&mem->dev.kobj);
1757 kobject_put(&mem->dev.kobj);
1762 #ifdef CONFIG_MEMORY_HOTREMOVE
1763 static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
1765 int ret = !is_memblock_offlined(mem);
1767 if (unlikely(ret)) {
1768 phys_addr_t beginpa, endpa;
1770 beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
1771 endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1;
1772 pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n",
1779 static int check_cpu_on_node(pg_data_t *pgdat)
1783 for_each_present_cpu(cpu) {
1784 if (cpu_to_node(cpu) == pgdat->node_id)
1786 * the cpu on this node isn't removed, and we can't
1787 * offline this node.
1795 static void unmap_cpu_on_node(pg_data_t *pgdat)
1797 #ifdef CONFIG_ACPI_NUMA
1800 for_each_possible_cpu(cpu)
1801 if (cpu_to_node(cpu) == pgdat->node_id)
1802 numa_clear_node(cpu);
1806 static int check_and_unmap_cpu_on_node(pg_data_t *pgdat)
1810 ret = check_cpu_on_node(pgdat);
1815 * the node will be offlined when we come here, so we can clear
1816 * the cpu_to_node() now.
1819 unmap_cpu_on_node(pgdat);
1827 * Offline a node if all memory sections and cpus of the node are removed.
1829 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1830 * and online/offline operations before this call.
1832 void try_offline_node(int nid)
1834 pg_data_t *pgdat = NODE_DATA(nid);
1835 unsigned long start_pfn = pgdat->node_start_pfn;
1836 unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
1839 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1840 unsigned long section_nr = pfn_to_section_nr(pfn);
1842 if (!present_section_nr(section_nr))
1845 if (pfn_to_nid(pfn) != nid)
1849 * some memory sections of this node are not removed, and we
1850 * can't offline node now.
1855 if (check_and_unmap_cpu_on_node(pgdat))
1859 * all memory/cpu of this node are removed, we can offline this
1862 node_set_offline(nid);
1863 unregister_one_node(nid);
1865 EXPORT_SYMBOL(try_offline_node);
1870 * @start: physical address of the region to remove
1871 * @size: size of the region to remove
1873 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1874 * and online/offline operations before this call, as required by
1875 * try_offline_node().
1877 void __ref remove_memory(int nid, u64 start, u64 size)
1881 BUG_ON(check_hotplug_memory_range(start, size));
1883 mem_hotplug_begin();
1886 * All memory blocks must be offlined before removing memory. Check
1887 * whether all memory blocks in question are offline and trigger a BUG()
1888 * if this is not the case.
1890 ret = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL,
1891 check_memblock_offlined_cb);
1895 /* remove memmap entry */
1896 firmware_map_remove(start, start + size, "System RAM");
1897 memblock_free(start, size);
1898 memblock_remove(start, size);
1900 arch_remove_memory(start, size, NULL);
1902 try_offline_node(nid);
1906 EXPORT_SYMBOL_GPL(remove_memory);
1907 #endif /* CONFIG_MEMORY_HOTREMOVE */