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, unsigned int order);
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(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, unsigned int order)
661 kernel_map_pages(page, 1 << order, 1);
662 __free_pages_core(page, order);
663 totalram_pages_add(1UL << order);
664 #ifdef CONFIG_HIGHMEM
665 if (PageHighMem(page))
666 totalhigh_pages_add(1UL << order);
670 static int online_pages_blocks(unsigned long start, unsigned long nr_pages)
672 unsigned long end = start + nr_pages;
673 int order, onlined_pages = 0;
675 while (start < end) {
676 order = min(MAX_ORDER - 1,
677 get_order(PFN_PHYS(end) - PFN_PHYS(start)));
678 (*online_page_callback)(pfn_to_page(start), order);
680 onlined_pages += (1UL << order);
681 start += (1UL << order);
683 return onlined_pages;
686 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
689 unsigned long onlined_pages = *(unsigned long *)arg;
691 if (PageReserved(pfn_to_page(start_pfn)))
692 onlined_pages += online_pages_blocks(start_pfn, nr_pages);
694 online_mem_sections(start_pfn, start_pfn + nr_pages);
696 *(unsigned long *)arg = onlined_pages;
700 /* check which state of node_states will be changed when online memory */
701 static void node_states_check_changes_online(unsigned long nr_pages,
702 struct zone *zone, struct memory_notify *arg)
704 int nid = zone_to_nid(zone);
706 arg->status_change_nid = NUMA_NO_NODE;
707 arg->status_change_nid_normal = NUMA_NO_NODE;
708 arg->status_change_nid_high = NUMA_NO_NODE;
710 if (!node_state(nid, N_MEMORY))
711 arg->status_change_nid = nid;
712 if (zone_idx(zone) <= ZONE_NORMAL && !node_state(nid, N_NORMAL_MEMORY))
713 arg->status_change_nid_normal = nid;
714 #ifdef CONFIG_HIGHMEM
715 if (zone_idx(zone) <= N_HIGH_MEMORY && !node_state(nid, N_HIGH_MEMORY))
716 arg->status_change_nid_high = nid;
720 static void node_states_set_node(int node, struct memory_notify *arg)
722 if (arg->status_change_nid_normal >= 0)
723 node_set_state(node, N_NORMAL_MEMORY);
725 if (arg->status_change_nid_high >= 0)
726 node_set_state(node, N_HIGH_MEMORY);
728 if (arg->status_change_nid >= 0)
729 node_set_state(node, N_MEMORY);
732 static void __meminit resize_zone_range(struct zone *zone, unsigned long start_pfn,
733 unsigned long nr_pages)
735 unsigned long old_end_pfn = zone_end_pfn(zone);
737 if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
738 zone->zone_start_pfn = start_pfn;
740 zone->spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - zone->zone_start_pfn;
743 static void __meminit resize_pgdat_range(struct pglist_data *pgdat, unsigned long start_pfn,
744 unsigned long nr_pages)
746 unsigned long old_end_pfn = pgdat_end_pfn(pgdat);
748 if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
749 pgdat->node_start_pfn = start_pfn;
751 pgdat->node_spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - pgdat->node_start_pfn;
754 void __ref move_pfn_range_to_zone(struct zone *zone, unsigned long start_pfn,
755 unsigned long nr_pages, struct vmem_altmap *altmap)
757 struct pglist_data *pgdat = zone->zone_pgdat;
758 int nid = pgdat->node_id;
761 clear_zone_contiguous(zone);
763 /* TODO Huh pgdat is irqsave while zone is not. It used to be like that before */
764 pgdat_resize_lock(pgdat, &flags);
765 zone_span_writelock(zone);
766 if (zone_is_empty(zone))
767 init_currently_empty_zone(zone, start_pfn, nr_pages);
768 resize_zone_range(zone, start_pfn, nr_pages);
769 zone_span_writeunlock(zone);
770 resize_pgdat_range(pgdat, start_pfn, nr_pages);
771 pgdat_resize_unlock(pgdat, &flags);
774 * TODO now we have a visible range of pages which are not associated
775 * with their zone properly. Not nice but set_pfnblock_flags_mask
776 * expects the zone spans the pfn range. All the pages in the range
777 * are reserved so nobody should be touching them so we should be safe
779 memmap_init_zone(nr_pages, nid, zone_idx(zone), start_pfn,
780 MEMMAP_HOTPLUG, altmap);
782 set_zone_contiguous(zone);
786 * Returns a default kernel memory zone for the given pfn range.
787 * If no kernel zone covers this pfn range it will automatically go
788 * to the ZONE_NORMAL.
790 static struct zone *default_kernel_zone_for_pfn(int nid, unsigned long start_pfn,
791 unsigned long nr_pages)
793 struct pglist_data *pgdat = NODE_DATA(nid);
796 for (zid = 0; zid <= ZONE_NORMAL; zid++) {
797 struct zone *zone = &pgdat->node_zones[zid];
799 if (zone_intersects(zone, start_pfn, nr_pages))
803 return &pgdat->node_zones[ZONE_NORMAL];
806 static inline struct zone *default_zone_for_pfn(int nid, unsigned long start_pfn,
807 unsigned long nr_pages)
809 struct zone *kernel_zone = default_kernel_zone_for_pfn(nid, start_pfn,
811 struct zone *movable_zone = &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
812 bool in_kernel = zone_intersects(kernel_zone, start_pfn, nr_pages);
813 bool in_movable = zone_intersects(movable_zone, start_pfn, nr_pages);
816 * We inherit the existing zone in a simple case where zones do not
817 * overlap in the given range
819 if (in_kernel ^ in_movable)
820 return (in_kernel) ? kernel_zone : movable_zone;
823 * If the range doesn't belong to any zone or two zones overlap in the
824 * given range then we use movable zone only if movable_node is
825 * enabled because we always online to a kernel zone by default.
827 return movable_node_enabled ? movable_zone : kernel_zone;
830 struct zone * zone_for_pfn_range(int online_type, int nid, unsigned start_pfn,
831 unsigned long nr_pages)
833 if (online_type == MMOP_ONLINE_KERNEL)
834 return default_kernel_zone_for_pfn(nid, start_pfn, nr_pages);
836 if (online_type == MMOP_ONLINE_MOVABLE)
837 return &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
839 return default_zone_for_pfn(nid, start_pfn, nr_pages);
843 * Associates the given pfn range with the given node and the zone appropriate
844 * for the given online type.
846 static struct zone * __meminit move_pfn_range(int online_type, int nid,
847 unsigned long start_pfn, unsigned long nr_pages)
851 zone = zone_for_pfn_range(online_type, nid, start_pfn, nr_pages);
852 move_pfn_range_to_zone(zone, start_pfn, nr_pages, NULL);
856 int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type)
859 unsigned long onlined_pages = 0;
861 int need_zonelists_rebuild = 0;
864 struct memory_notify arg;
865 struct memory_block *mem;
870 * We can't use pfn_to_nid() because nid might be stored in struct page
871 * which is not yet initialized. Instead, we find nid from memory block.
873 mem = find_memory_block(__pfn_to_section(pfn));
876 /* associate pfn range with the zone */
877 zone = move_pfn_range(online_type, nid, pfn, nr_pages);
880 arg.nr_pages = nr_pages;
881 node_states_check_changes_online(nr_pages, zone, &arg);
883 ret = memory_notify(MEM_GOING_ONLINE, &arg);
884 ret = notifier_to_errno(ret);
886 goto failed_addition;
889 * If this zone is not populated, then it is not in zonelist.
890 * This means the page allocator ignores this zone.
891 * So, zonelist must be updated after online.
893 if (!populated_zone(zone)) {
894 need_zonelists_rebuild = 1;
895 setup_zone_pageset(zone);
898 ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
901 if (need_zonelists_rebuild)
902 zone_pcp_reset(zone);
903 goto failed_addition;
906 zone->present_pages += onlined_pages;
908 pgdat_resize_lock(zone->zone_pgdat, &flags);
909 zone->zone_pgdat->node_present_pages += onlined_pages;
910 pgdat_resize_unlock(zone->zone_pgdat, &flags);
913 node_states_set_node(nid, &arg);
914 if (need_zonelists_rebuild)
915 build_all_zonelists(NULL);
917 zone_pcp_update(zone);
920 init_per_zone_wmark_min();
927 vm_total_pages = nr_free_pagecache_pages();
929 writeback_set_ratelimit();
932 memory_notify(MEM_ONLINE, &arg);
937 pr_debug("online_pages [mem %#010llx-%#010llx] failed\n",
938 (unsigned long long) pfn << PAGE_SHIFT,
939 (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1);
940 memory_notify(MEM_CANCEL_ONLINE, &arg);
944 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
946 static void reset_node_present_pages(pg_data_t *pgdat)
950 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
951 z->present_pages = 0;
953 pgdat->node_present_pages = 0;
956 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
957 static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
959 struct pglist_data *pgdat;
960 unsigned long start_pfn = PFN_DOWN(start);
962 pgdat = NODE_DATA(nid);
964 pgdat = arch_alloc_nodedata(nid);
968 arch_refresh_nodedata(nid, pgdat);
971 * Reset the nr_zones, order and classzone_idx before reuse.
972 * Note that kswapd will init kswapd_classzone_idx properly
973 * when it starts in the near future.
976 pgdat->kswapd_order = 0;
977 pgdat->kswapd_classzone_idx = 0;
980 /* we can use NODE_DATA(nid) from here */
982 pgdat->node_id = nid;
983 pgdat->node_start_pfn = start_pfn;
985 /* init node's zones as empty zones, we don't have any present pages.*/
986 free_area_init_core_hotplug(nid);
987 pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat);
990 * The node we allocated has no zone fallback lists. For avoiding
991 * to access not-initialized zonelist, build here.
993 build_all_zonelists(pgdat);
996 * When memory is hot-added, all the memory is in offline state. So
997 * clear all zones' present_pages because they will be updated in
998 * online_pages() and offline_pages().
1000 reset_node_managed_pages(pgdat);
1001 reset_node_present_pages(pgdat);
1006 static void rollback_node_hotadd(int nid)
1008 pg_data_t *pgdat = NODE_DATA(nid);
1010 arch_refresh_nodedata(nid, NULL);
1011 free_percpu(pgdat->per_cpu_nodestats);
1012 arch_free_nodedata(pgdat);
1018 * try_online_node - online a node if offlined
1020 * @start: start addr of the node
1021 * @set_node_online: Whether we want to online the node
1022 * called by cpu_up() to online a node without onlined memory.
1025 * 1 -> a new node has been allocated
1026 * 0 -> the node is already online
1027 * -ENOMEM -> the node could not be allocated
1029 static int __try_online_node(int nid, u64 start, bool set_node_online)
1034 if (node_online(nid))
1037 pgdat = hotadd_new_pgdat(nid, start);
1039 pr_err("Cannot online node %d due to NULL pgdat\n", nid);
1044 if (set_node_online) {
1045 node_set_online(nid);
1046 ret = register_one_node(nid);
1054 * Users of this function always want to online/register the node
1056 int try_online_node(int nid)
1060 mem_hotplug_begin();
1061 ret = __try_online_node(nid, 0, true);
1066 static int check_hotplug_memory_range(u64 start, u64 size)
1068 unsigned long block_sz = memory_block_size_bytes();
1069 u64 block_nr_pages = block_sz >> PAGE_SHIFT;
1070 u64 nr_pages = size >> PAGE_SHIFT;
1071 u64 start_pfn = PFN_DOWN(start);
1073 /* memory range must be block size aligned */
1074 if (!nr_pages || !IS_ALIGNED(start_pfn, block_nr_pages) ||
1075 !IS_ALIGNED(nr_pages, block_nr_pages)) {
1076 pr_err("Block size [%#lx] unaligned hotplug range: start %#llx, size %#llx",
1077 block_sz, start, size);
1084 static int online_memory_block(struct memory_block *mem, void *arg)
1086 return device_online(&mem->dev);
1090 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1091 * and online/offline operations (triggered e.g. by sysfs).
1093 * we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG
1095 int __ref add_memory_resource(int nid, struct resource *res)
1098 bool new_node = false;
1102 size = resource_size(res);
1104 ret = check_hotplug_memory_range(start, size);
1108 mem_hotplug_begin();
1111 * Add new range to memblock so that when hotadd_new_pgdat() is called
1112 * to allocate new pgdat, get_pfn_range_for_nid() will be able to find
1113 * this new range and calculate total pages correctly. The range will
1114 * be removed at hot-remove time.
1116 memblock_add_node(start, size, nid);
1118 ret = __try_online_node(nid, start, false);
1123 /* call arch's memory hotadd */
1124 ret = arch_add_memory(nid, start, size, NULL, true);
1129 /* If sysfs file of new node can't be created, cpu on the node
1130 * can't be hot-added. There is no rollback way now.
1131 * So, check by BUG_ON() to catch it reluctantly..
1132 * We online node here. We can't roll back from here.
1134 node_set_online(nid);
1135 ret = __register_one_node(nid);
1139 /* link memory sections under this node.*/
1140 ret = link_mem_sections(nid, PFN_DOWN(start), PFN_UP(start + size - 1));
1143 /* create new memmap entry */
1144 firmware_map_add_hotplug(start, start + size, "System RAM");
1146 /* device_online() will take the lock when calling online_pages() */
1149 /* online pages if requested */
1150 if (memhp_auto_online)
1151 walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1),
1152 NULL, online_memory_block);
1156 /* rollback pgdat allocation and others */
1158 rollback_node_hotadd(nid);
1159 memblock_remove(start, size);
1164 /* requires device_hotplug_lock, see add_memory_resource() */
1165 int __ref __add_memory(int nid, u64 start, u64 size)
1167 struct resource *res;
1170 res = register_memory_resource(start, size);
1172 return PTR_ERR(res);
1174 ret = add_memory_resource(nid, res);
1176 release_memory_resource(res);
1180 int add_memory(int nid, u64 start, u64 size)
1184 lock_device_hotplug();
1185 rc = __add_memory(nid, start, size);
1186 unlock_device_hotplug();
1190 EXPORT_SYMBOL_GPL(add_memory);
1192 #ifdef CONFIG_MEMORY_HOTREMOVE
1194 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1195 * set and the size of the free page is given by page_order(). Using this,
1196 * the function determines if the pageblock contains only free pages.
1197 * Due to buddy contraints, a free page at least the size of a pageblock will
1198 * be located at the start of the pageblock
1200 static inline int pageblock_free(struct page *page)
1202 return PageBuddy(page) && page_order(page) >= pageblock_order;
1205 /* Return the pfn of the start of the next active pageblock after a given pfn */
1206 static unsigned long next_active_pageblock(unsigned long pfn)
1208 struct page *page = pfn_to_page(pfn);
1210 /* Ensure the starting page is pageblock-aligned */
1211 BUG_ON(pfn & (pageblock_nr_pages - 1));
1213 /* If the entire pageblock is free, move to the end of free page */
1214 if (pageblock_free(page)) {
1216 /* be careful. we don't have locks, page_order can be changed.*/
1217 order = page_order(page);
1218 if ((order < MAX_ORDER) && (order >= pageblock_order))
1219 return pfn + (1 << order);
1222 return pfn + pageblock_nr_pages;
1225 static bool is_pageblock_removable_nolock(unsigned long pfn)
1227 struct page *page = pfn_to_page(pfn);
1231 * We have to be careful here because we are iterating over memory
1232 * sections which are not zone aware so we might end up outside of
1233 * the zone but still within the section.
1234 * We have to take care about the node as well. If the node is offline
1235 * its NODE_DATA will be NULL - see page_zone.
1237 if (!node_online(page_to_nid(page)))
1240 zone = page_zone(page);
1241 pfn = page_to_pfn(page);
1242 if (!zone_spans_pfn(zone, pfn))
1245 return !has_unmovable_pages(zone, page, 0, MIGRATE_MOVABLE, SKIP_HWPOISON);
1248 /* Checks if this range of memory is likely to be hot-removable. */
1249 bool is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
1251 unsigned long end_pfn, pfn;
1253 end_pfn = min(start_pfn + nr_pages,
1254 zone_end_pfn(page_zone(pfn_to_page(start_pfn))));
1256 /* Check the starting page of each pageblock within the range */
1257 for (pfn = start_pfn; pfn < end_pfn; pfn = next_active_pageblock(pfn)) {
1258 if (!is_pageblock_removable_nolock(pfn))
1263 /* All pageblocks in the memory block are likely to be hot-removable */
1268 * Confirm all pages in a range [start, end) belong to the same zone.
1269 * When true, return its valid [start, end).
1271 int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn,
1272 unsigned long *valid_start, unsigned long *valid_end)
1274 unsigned long pfn, sec_end_pfn;
1275 unsigned long start, end;
1276 struct zone *zone = NULL;
1279 for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1);
1281 pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) {
1282 /* Make sure the memory section is present first */
1283 if (!present_section_nr(pfn_to_section_nr(pfn)))
1285 for (; pfn < sec_end_pfn && pfn < end_pfn;
1286 pfn += MAX_ORDER_NR_PAGES) {
1288 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1289 while ((i < MAX_ORDER_NR_PAGES) &&
1290 !pfn_valid_within(pfn + i))
1292 if (i == MAX_ORDER_NR_PAGES || pfn + i >= end_pfn)
1294 /* Check if we got outside of the zone */
1295 if (zone && !zone_spans_pfn(zone, pfn + i))
1297 page = pfn_to_page(pfn + i);
1298 if (zone && page_zone(page) != zone)
1302 zone = page_zone(page);
1303 end = pfn + MAX_ORDER_NR_PAGES;
1308 *valid_start = start;
1309 *valid_end = min(end, end_pfn);
1317 * Scan pfn range [start,end) to find movable/migratable pages (LRU pages,
1318 * non-lru movable pages and hugepages). We scan pfn because it's much
1319 * easier than scanning over linked list. This function returns the pfn
1320 * of the first found movable page if it's found, otherwise 0.
1322 static unsigned long scan_movable_pages(unsigned long start, unsigned long end)
1326 for (pfn = start; pfn < end; pfn++) {
1327 struct page *page, *head;
1330 if (!pfn_valid(pfn))
1332 page = pfn_to_page(pfn);
1335 if (__PageMovable(page))
1338 if (!PageHuge(page))
1340 head = compound_head(page);
1341 if (hugepage_migration_supported(page_hstate(head)) &&
1342 page_huge_active(head))
1344 skip = (1 << compound_order(head)) - (page - head);
1350 static struct page *new_node_page(struct page *page, unsigned long private)
1352 int nid = page_to_nid(page);
1353 nodemask_t nmask = node_states[N_MEMORY];
1356 * try to allocate from a different node but reuse this node if there
1357 * are no other online nodes to be used (e.g. we are offlining a part
1358 * of the only existing node)
1360 node_clear(nid, nmask);
1361 if (nodes_empty(nmask))
1362 node_set(nid, nmask);
1364 return new_page_nodemask(page, nid, &nmask);
1368 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1375 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1376 if (!pfn_valid(pfn))
1378 page = pfn_to_page(pfn);
1380 if (PageHuge(page)) {
1381 struct page *head = compound_head(page);
1382 if (compound_order(head) > PFN_SECTION_SHIFT) {
1386 pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1;
1387 isolate_huge_page(head, &source);
1389 } else if (PageTransHuge(page))
1390 pfn = page_to_pfn(compound_head(page))
1391 + hpage_nr_pages(page) - 1;
1394 * HWPoison pages have elevated reference counts so the migration would
1395 * fail on them. It also doesn't make any sense to migrate them in the
1396 * first place. Still try to unmap such a page in case it is still mapped
1397 * (e.g. current hwpoison implementation doesn't unmap KSM pages but keep
1398 * the unmap as the catch all safety net).
1400 if (PageHWPoison(page)) {
1401 if (WARN_ON(PageLRU(page)))
1402 isolate_lru_page(page);
1403 if (page_mapped(page))
1404 try_to_unmap(page, TTU_IGNORE_MLOCK | TTU_IGNORE_ACCESS);
1408 if (!get_page_unless_zero(page))
1411 * We can skip free pages. And we can deal with pages on
1412 * LRU and non-lru movable pages.
1415 ret = isolate_lru_page(page);
1417 ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE);
1418 if (!ret) { /* Success */
1419 list_add_tail(&page->lru, &source);
1420 if (!__PageMovable(page))
1421 inc_node_page_state(page, NR_ISOLATED_ANON +
1422 page_is_file_cache(page));
1425 pr_warn("failed to isolate pfn %lx\n", pfn);
1426 dump_page(page, "isolation failed");
1430 if (!list_empty(&source)) {
1431 /* Allocate a new page from the nearest neighbor node */
1432 ret = migrate_pages(&source, new_node_page, NULL, 0,
1433 MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
1435 list_for_each_entry(page, &source, lru) {
1436 pr_warn("migrating pfn %lx failed ret:%d ",
1437 page_to_pfn(page), ret);
1438 dump_page(page, "migration failure");
1440 putback_movable_pages(&source);
1448 * remove from free_area[] and mark all as Reserved.
1451 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1454 __offline_isolated_pages(start, start + nr_pages);
1459 offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
1461 walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
1462 offline_isolated_pages_cb);
1466 * Check all pages in range, recoreded as memory resource, are isolated.
1469 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1473 long offlined = *(long *)data;
1474 ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true);
1475 offlined = nr_pages;
1477 *(long *)data += offlined;
1482 check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
1487 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
1488 check_pages_isolated_cb);
1490 offlined = (long)ret;
1494 static int __init cmdline_parse_movable_node(char *p)
1496 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1497 movable_node_enabled = true;
1499 pr_warn("movable_node parameter depends on CONFIG_HAVE_MEMBLOCK_NODE_MAP to work properly\n");
1503 early_param("movable_node", cmdline_parse_movable_node);
1505 /* check which state of node_states will be changed when offline memory */
1506 static void node_states_check_changes_offline(unsigned long nr_pages,
1507 struct zone *zone, struct memory_notify *arg)
1509 struct pglist_data *pgdat = zone->zone_pgdat;
1510 unsigned long present_pages = 0;
1513 arg->status_change_nid = NUMA_NO_NODE;
1514 arg->status_change_nid_normal = NUMA_NO_NODE;
1515 arg->status_change_nid_high = NUMA_NO_NODE;
1518 * Check whether node_states[N_NORMAL_MEMORY] will be changed.
1519 * If the memory to be offline is within the range
1520 * [0..ZONE_NORMAL], and it is the last present memory there,
1521 * the zones in that range will become empty after the offlining,
1522 * thus we can determine that we need to clear the node from
1523 * node_states[N_NORMAL_MEMORY].
1525 for (zt = 0; zt <= ZONE_NORMAL; zt++)
1526 present_pages += pgdat->node_zones[zt].present_pages;
1527 if (zone_idx(zone) <= ZONE_NORMAL && nr_pages >= present_pages)
1528 arg->status_change_nid_normal = zone_to_nid(zone);
1530 #ifdef CONFIG_HIGHMEM
1532 * node_states[N_HIGH_MEMORY] contains nodes which
1533 * have normal memory or high memory.
1534 * Here we add the present_pages belonging to ZONE_HIGHMEM.
1535 * If the zone is within the range of [0..ZONE_HIGHMEM), and
1536 * we determine that the zones in that range become empty,
1537 * we need to clear the node for N_HIGH_MEMORY.
1539 present_pages += pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1540 if (zone_idx(zone) <= ZONE_HIGHMEM && nr_pages >= present_pages)
1541 arg->status_change_nid_high = zone_to_nid(zone);
1545 * We have accounted the pages from [0..ZONE_NORMAL), and
1546 * in case of CONFIG_HIGHMEM the pages from ZONE_HIGHMEM
1548 * Here we count the possible pages from ZONE_MOVABLE.
1549 * If after having accounted all the pages, we see that the nr_pages
1550 * to be offlined is over or equal to the accounted pages,
1551 * we know that the node will become empty, and so, we can clear
1552 * it for N_MEMORY as well.
1554 present_pages += pgdat->node_zones[ZONE_MOVABLE].present_pages;
1556 if (nr_pages >= present_pages)
1557 arg->status_change_nid = zone_to_nid(zone);
1560 static void node_states_clear_node(int node, struct memory_notify *arg)
1562 if (arg->status_change_nid_normal >= 0)
1563 node_clear_state(node, N_NORMAL_MEMORY);
1565 if (arg->status_change_nid_high >= 0)
1566 node_clear_state(node, N_HIGH_MEMORY);
1568 if (arg->status_change_nid >= 0)
1569 node_clear_state(node, N_MEMORY);
1572 static int __ref __offline_pages(unsigned long start_pfn,
1573 unsigned long end_pfn)
1575 unsigned long pfn, nr_pages;
1576 long offlined_pages;
1578 unsigned long flags;
1579 unsigned long valid_start, valid_end;
1581 struct memory_notify arg;
1584 mem_hotplug_begin();
1586 /* This makes hotplug much easier...and readable.
1587 we assume this for now. .*/
1588 if (!test_pages_in_a_zone(start_pfn, end_pfn, &valid_start,
1591 reason = "multizone range";
1592 goto failed_removal;
1595 zone = page_zone(pfn_to_page(valid_start));
1596 node = zone_to_nid(zone);
1597 nr_pages = end_pfn - start_pfn;
1599 /* set above range as isolated */
1600 ret = start_isolate_page_range(start_pfn, end_pfn,
1602 SKIP_HWPOISON | REPORT_FAILURE);
1604 reason = "failure to isolate range";
1605 goto failed_removal;
1608 arg.start_pfn = start_pfn;
1609 arg.nr_pages = nr_pages;
1610 node_states_check_changes_offline(nr_pages, zone, &arg);
1612 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1613 ret = notifier_to_errno(ret);
1615 reason = "notifier failure";
1616 goto failed_removal_isolated;
1620 for (pfn = start_pfn; pfn;) {
1621 if (signal_pending(current)) {
1623 reason = "signal backoff";
1624 goto failed_removal_isolated;
1628 lru_add_drain_all();
1630 pfn = scan_movable_pages(pfn, end_pfn);
1633 * TODO: fatal migration failures should bail
1636 do_migrate_range(pfn, end_pfn);
1641 * Dissolve free hugepages in the memory block before doing
1642 * offlining actually in order to make hugetlbfs's object
1643 * counting consistent.
1645 ret = dissolve_free_huge_pages(start_pfn, end_pfn);
1647 reason = "failure to dissolve huge pages";
1648 goto failed_removal_isolated;
1651 offlined_pages = check_pages_isolated(start_pfn, end_pfn);
1652 } while (offlined_pages < 0);
1654 pr_info("Offlined Pages %ld\n", offlined_pages);
1655 /* Ok, all of our target is isolated.
1656 We cannot do rollback at this point. */
1657 offline_isolated_pages(start_pfn, end_pfn);
1658 /* reset pagetype flags and makes migrate type to be MOVABLE */
1659 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1660 /* removal success */
1661 adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages);
1662 zone->present_pages -= offlined_pages;
1664 pgdat_resize_lock(zone->zone_pgdat, &flags);
1665 zone->zone_pgdat->node_present_pages -= offlined_pages;
1666 pgdat_resize_unlock(zone->zone_pgdat, &flags);
1668 init_per_zone_wmark_min();
1670 if (!populated_zone(zone)) {
1671 zone_pcp_reset(zone);
1672 build_all_zonelists(NULL);
1674 zone_pcp_update(zone);
1676 node_states_clear_node(node, &arg);
1677 if (arg.status_change_nid >= 0) {
1679 kcompactd_stop(node);
1682 vm_total_pages = nr_free_pagecache_pages();
1683 writeback_set_ratelimit();
1685 memory_notify(MEM_OFFLINE, &arg);
1689 failed_removal_isolated:
1690 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1692 pr_debug("memory offlining [mem %#010llx-%#010llx] failed due to %s\n",
1693 (unsigned long long) start_pfn << PAGE_SHIFT,
1694 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1,
1696 memory_notify(MEM_CANCEL_OFFLINE, &arg);
1697 /* pushback to free area */
1702 int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1704 return __offline_pages(start_pfn, start_pfn + nr_pages);
1706 #endif /* CONFIG_MEMORY_HOTREMOVE */
1709 * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
1710 * @start_pfn: start pfn of the memory range
1711 * @end_pfn: end pfn of the memory range
1712 * @arg: argument passed to func
1713 * @func: callback for each memory section walked
1715 * This function walks through all present mem sections in range
1716 * [start_pfn, end_pfn) and call func on each mem section.
1718 * Returns the return value of func.
1720 int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn,
1721 void *arg, int (*func)(struct memory_block *, void *))
1723 struct memory_block *mem = NULL;
1724 struct mem_section *section;
1725 unsigned long pfn, section_nr;
1728 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1729 section_nr = pfn_to_section_nr(pfn);
1730 if (!present_section_nr(section_nr))
1733 section = __nr_to_section(section_nr);
1734 /* same memblock? */
1736 if ((section_nr >= mem->start_section_nr) &&
1737 (section_nr <= mem->end_section_nr))
1740 mem = find_memory_block_hinted(section, mem);
1744 ret = func(mem, arg);
1746 kobject_put(&mem->dev.kobj);
1752 kobject_put(&mem->dev.kobj);
1757 #ifdef CONFIG_MEMORY_HOTREMOVE
1758 static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
1760 int ret = !is_memblock_offlined(mem);
1762 if (unlikely(ret)) {
1763 phys_addr_t beginpa, endpa;
1765 beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
1766 endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1;
1767 pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n",
1774 static int check_cpu_on_node(pg_data_t *pgdat)
1778 for_each_present_cpu(cpu) {
1779 if (cpu_to_node(cpu) == pgdat->node_id)
1781 * the cpu on this node isn't removed, and we can't
1782 * offline this node.
1794 * Offline a node if all memory sections and cpus of the node are removed.
1796 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1797 * and online/offline operations before this call.
1799 void try_offline_node(int nid)
1801 pg_data_t *pgdat = NODE_DATA(nid);
1802 unsigned long start_pfn = pgdat->node_start_pfn;
1803 unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
1806 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1807 unsigned long section_nr = pfn_to_section_nr(pfn);
1809 if (!present_section_nr(section_nr))
1812 if (pfn_to_nid(pfn) != nid)
1816 * some memory sections of this node are not removed, and we
1817 * can't offline node now.
1822 if (check_cpu_on_node(pgdat))
1826 * all memory/cpu of this node are removed, we can offline this
1829 node_set_offline(nid);
1830 unregister_one_node(nid);
1832 EXPORT_SYMBOL(try_offline_node);
1837 * @start: physical address of the region to remove
1838 * @size: size of the region to remove
1840 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1841 * and online/offline operations before this call, as required by
1842 * try_offline_node().
1844 void __ref __remove_memory(int nid, u64 start, u64 size)
1848 BUG_ON(check_hotplug_memory_range(start, size));
1850 mem_hotplug_begin();
1853 * All memory blocks must be offlined before removing memory. Check
1854 * whether all memory blocks in question are offline and trigger a BUG()
1855 * if this is not the case.
1857 ret = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL,
1858 check_memblock_offlined_cb);
1862 /* remove memmap entry */
1863 firmware_map_remove(start, start + size, "System RAM");
1864 memblock_free(start, size);
1865 memblock_remove(start, size);
1867 arch_remove_memory(nid, start, size, NULL);
1869 try_offline_node(nid);
1874 void remove_memory(int nid, u64 start, u64 size)
1876 lock_device_hotplug();
1877 __remove_memory(nid, start, size);
1878 unlock_device_hotplug();
1880 EXPORT_SYMBOL_GPL(remove_memory);
1881 #endif /* CONFIG_MEMORY_HOTREMOVE */