1 // SPDX-License-Identifier: GPL-2.0-only
3 * linux/mm/memory_hotplug.c
8 #include <linux/stddef.h>
10 #include <linux/sched/signal.h>
11 #include <linux/swap.h>
12 #include <linux/interrupt.h>
13 #include <linux/pagemap.h>
14 #include <linux/compiler.h>
15 #include <linux/export.h>
16 #include <linux/pagevec.h>
17 #include <linux/writeback.h>
18 #include <linux/slab.h>
19 #include <linux/sysctl.h>
20 #include <linux/cpu.h>
21 #include <linux/memory.h>
22 #include <linux/memremap.h>
23 #include <linux/memory_hotplug.h>
24 #include <linux/highmem.h>
25 #include <linux/vmalloc.h>
26 #include <linux/ioport.h>
27 #include <linux/delay.h>
28 #include <linux/migrate.h>
29 #include <linux/page-isolation.h>
30 #include <linux/pfn.h>
31 #include <linux/suspend.h>
32 #include <linux/mm_inline.h>
33 #include <linux/firmware-map.h>
34 #include <linux/stop_machine.h>
35 #include <linux/hugetlb.h>
36 #include <linux/memblock.h>
37 #include <linux/compaction.h>
38 #include <linux/rmap.h>
40 #include <asm/tlbflush.h>
46 * online_page_callback contains pointer to current page onlining function.
47 * Initially it is generic_online_page(). If it is required it could be
48 * changed by calling set_online_page_callback() for callback registration
49 * and restore_online_page_callback() for generic callback restore.
52 static void generic_online_page(struct page *page, unsigned int order);
54 static online_page_callback_t online_page_callback = generic_online_page;
55 static DEFINE_MUTEX(online_page_callback_lock);
57 DEFINE_STATIC_PERCPU_RWSEM(mem_hotplug_lock);
59 void get_online_mems(void)
61 percpu_down_read(&mem_hotplug_lock);
64 void put_online_mems(void)
66 percpu_up_read(&mem_hotplug_lock);
69 bool movable_node_enabled = false;
71 #ifndef CONFIG_MEMORY_HOTPLUG_DEFAULT_ONLINE
72 bool memhp_auto_online;
74 bool memhp_auto_online = true;
76 EXPORT_SYMBOL_GPL(memhp_auto_online);
78 static int __init setup_memhp_default_state(char *str)
80 if (!strcmp(str, "online"))
81 memhp_auto_online = true;
82 else if (!strcmp(str, "offline"))
83 memhp_auto_online = false;
87 __setup("memhp_default_state=", setup_memhp_default_state);
89 void mem_hotplug_begin(void)
92 percpu_down_write(&mem_hotplug_lock);
95 void mem_hotplug_done(void)
97 percpu_up_write(&mem_hotplug_lock);
101 u64 max_mem_size = U64_MAX;
103 /* add this memory to iomem resource */
104 static struct resource *register_memory_resource(u64 start, u64 size)
106 struct resource *res;
107 unsigned long flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
108 char *resource_name = "System RAM";
110 if (start + size > max_mem_size)
111 return ERR_PTR(-E2BIG);
114 * Request ownership of the new memory range. This might be
115 * a child of an existing resource that was present but
116 * not marked as busy.
118 res = __request_region(&iomem_resource, start, size,
119 resource_name, flags);
122 pr_debug("Unable to reserve System RAM region: %016llx->%016llx\n",
123 start, start + size);
124 return ERR_PTR(-EEXIST);
129 static void release_memory_resource(struct resource *res)
133 release_resource(res);
138 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
139 void get_page_bootmem(unsigned long info, struct page *page,
142 page->freelist = (void *)type;
143 SetPagePrivate(page);
144 set_page_private(page, info);
148 void put_page_bootmem(struct page *page)
152 type = (unsigned long) page->freelist;
153 BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE ||
154 type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE);
156 if (page_ref_dec_return(page) == 1) {
157 page->freelist = NULL;
158 ClearPagePrivate(page);
159 set_page_private(page, 0);
160 INIT_LIST_HEAD(&page->lru);
161 free_reserved_page(page);
165 #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
166 #ifndef CONFIG_SPARSEMEM_VMEMMAP
167 static void register_page_bootmem_info_section(unsigned long start_pfn)
169 unsigned long *usemap, mapsize, section_nr, i;
170 struct mem_section *ms;
171 struct page *page, *memmap;
173 section_nr = pfn_to_section_nr(start_pfn);
174 ms = __nr_to_section(section_nr);
176 /* Get section's memmap address */
177 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
180 * Get page for the memmap's phys address
181 * XXX: need more consideration for sparse_vmemmap...
183 page = virt_to_page(memmap);
184 mapsize = sizeof(struct page) * PAGES_PER_SECTION;
185 mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
187 /* remember memmap's page */
188 for (i = 0; i < mapsize; i++, page++)
189 get_page_bootmem(section_nr, page, SECTION_INFO);
191 usemap = ms->pageblock_flags;
192 page = virt_to_page(usemap);
194 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
196 for (i = 0; i < mapsize; i++, page++)
197 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
200 #else /* CONFIG_SPARSEMEM_VMEMMAP */
201 static void register_page_bootmem_info_section(unsigned long start_pfn)
203 unsigned long *usemap, mapsize, section_nr, i;
204 struct mem_section *ms;
205 struct page *page, *memmap;
207 section_nr = pfn_to_section_nr(start_pfn);
208 ms = __nr_to_section(section_nr);
210 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
212 register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION);
214 usemap = ms->pageblock_flags;
215 page = virt_to_page(usemap);
217 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
219 for (i = 0; i < mapsize; i++, page++)
220 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
222 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
224 void __init register_page_bootmem_info_node(struct pglist_data *pgdat)
226 unsigned long i, pfn, end_pfn, nr_pages;
227 int node = pgdat->node_id;
230 nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
231 page = virt_to_page(pgdat);
233 for (i = 0; i < nr_pages; i++, page++)
234 get_page_bootmem(node, page, NODE_INFO);
236 pfn = pgdat->node_start_pfn;
237 end_pfn = pgdat_end_pfn(pgdat);
239 /* register section info */
240 for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
242 * Some platforms can assign the same pfn to multiple nodes - on
243 * node0 as well as nodeN. To avoid registering a pfn against
244 * multiple nodes we check that this pfn does not already
245 * reside in some other nodes.
247 if (pfn_valid(pfn) && (early_pfn_to_nid(pfn) == node))
248 register_page_bootmem_info_section(pfn);
251 #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
253 static int __meminit __add_section(int nid, unsigned long phys_start_pfn,
254 struct vmem_altmap *altmap)
258 if (pfn_valid(phys_start_pfn))
261 ret = sparse_add_one_section(nid, phys_start_pfn, altmap);
262 return ret < 0 ? ret : 0;
266 * Reasonably generic function for adding memory. It is
267 * expected that archs that support memory hotplug will
268 * call this function after deciding the zone to which to
271 int __ref __add_pages(int nid, unsigned long phys_start_pfn,
272 unsigned long nr_pages, struct mhp_restrictions *restrictions)
276 int start_sec, end_sec;
277 struct vmem_altmap *altmap = restrictions->altmap;
279 /* during initialize mem_map, align hot-added range to section */
280 start_sec = pfn_to_section_nr(phys_start_pfn);
281 end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
285 * Validate altmap is within bounds of the total request
287 if (altmap->base_pfn != phys_start_pfn
288 || vmem_altmap_offset(altmap) > nr_pages) {
289 pr_warn_once("memory add fail, invalid altmap\n");
296 for (i = start_sec; i <= end_sec; i++) {
297 err = __add_section(nid, section_nr_to_pfn(i), altmap);
300 * EEXIST is finally dealt with by ioresource collision
301 * check. see add_memory() => register_memory_resource()
302 * Warning will be printed if there is collision.
304 if (err && (err != -EEXIST))
309 vmemmap_populate_print_last();
314 /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
315 static unsigned long find_smallest_section_pfn(int nid, struct zone *zone,
316 unsigned long start_pfn,
317 unsigned long end_pfn)
319 struct mem_section *ms;
321 for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) {
322 ms = __pfn_to_section(start_pfn);
324 if (unlikely(!valid_section(ms)))
327 if (unlikely(pfn_to_nid(start_pfn) != nid))
330 if (zone && zone != page_zone(pfn_to_page(start_pfn)))
339 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
340 static unsigned long find_biggest_section_pfn(int nid, struct zone *zone,
341 unsigned long start_pfn,
342 unsigned long end_pfn)
344 struct mem_section *ms;
347 /* pfn is the end pfn of a memory section. */
349 for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) {
350 ms = __pfn_to_section(pfn);
352 if (unlikely(!valid_section(ms)))
355 if (unlikely(pfn_to_nid(pfn) != nid))
358 if (zone && zone != page_zone(pfn_to_page(pfn)))
367 static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
368 unsigned long end_pfn)
370 unsigned long zone_start_pfn = zone->zone_start_pfn;
371 unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */
372 unsigned long zone_end_pfn = z;
374 struct mem_section *ms;
375 int nid = zone_to_nid(zone);
377 zone_span_writelock(zone);
378 if (zone_start_pfn == start_pfn) {
380 * If the section is smallest section in the zone, it need
381 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
382 * In this case, we find second smallest valid mem_section
383 * for shrinking zone.
385 pfn = find_smallest_section_pfn(nid, zone, end_pfn,
388 zone->zone_start_pfn = pfn;
389 zone->spanned_pages = zone_end_pfn - pfn;
391 } else if (zone_end_pfn == end_pfn) {
393 * If the section is biggest section in the zone, it need
394 * shrink zone->spanned_pages.
395 * In this case, we find second biggest valid mem_section for
398 pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn,
401 zone->spanned_pages = pfn - zone_start_pfn + 1;
405 * The section is not biggest or smallest mem_section in the zone, it
406 * only creates a hole in the zone. So in this case, we need not
407 * change the zone. But perhaps, the zone has only hole data. Thus
408 * it check the zone has only hole or not.
410 pfn = zone_start_pfn;
411 for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) {
412 ms = __pfn_to_section(pfn);
414 if (unlikely(!valid_section(ms)))
417 if (page_zone(pfn_to_page(pfn)) != zone)
420 /* If the section is current section, it continues the loop */
421 if (start_pfn == pfn)
424 /* If we find valid section, we have nothing to do */
425 zone_span_writeunlock(zone);
429 /* The zone has no valid section */
430 zone->zone_start_pfn = 0;
431 zone->spanned_pages = 0;
432 zone_span_writeunlock(zone);
435 static void shrink_pgdat_span(struct pglist_data *pgdat,
436 unsigned long start_pfn, unsigned long end_pfn)
438 unsigned long pgdat_start_pfn = pgdat->node_start_pfn;
439 unsigned long p = pgdat_end_pfn(pgdat); /* pgdat_end_pfn namespace clash */
440 unsigned long pgdat_end_pfn = p;
442 struct mem_section *ms;
443 int nid = pgdat->node_id;
445 if (pgdat_start_pfn == start_pfn) {
447 * If the section is smallest section in the pgdat, it need
448 * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages.
449 * In this case, we find second smallest valid mem_section
450 * for shrinking zone.
452 pfn = find_smallest_section_pfn(nid, NULL, end_pfn,
455 pgdat->node_start_pfn = pfn;
456 pgdat->node_spanned_pages = pgdat_end_pfn - pfn;
458 } else if (pgdat_end_pfn == end_pfn) {
460 * If the section is biggest section in the pgdat, it need
461 * shrink pgdat->node_spanned_pages.
462 * In this case, we find second biggest valid mem_section for
465 pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn,
468 pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1;
472 * If the section is not biggest or smallest mem_section in the pgdat,
473 * it only creates a hole in the pgdat. So in this case, we need not
475 * But perhaps, the pgdat has only hole data. Thus it check the pgdat
476 * has only hole or not.
478 pfn = pgdat_start_pfn;
479 for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) {
480 ms = __pfn_to_section(pfn);
482 if (unlikely(!valid_section(ms)))
485 if (pfn_to_nid(pfn) != nid)
488 /* If the section is current section, it continues the loop */
489 if (start_pfn == pfn)
492 /* If we find valid section, we have nothing to do */
496 /* The pgdat has no valid section */
497 pgdat->node_start_pfn = 0;
498 pgdat->node_spanned_pages = 0;
501 static void __remove_zone(struct zone *zone, unsigned long start_pfn)
503 struct pglist_data *pgdat = zone->zone_pgdat;
504 int nr_pages = PAGES_PER_SECTION;
507 pgdat_resize_lock(zone->zone_pgdat, &flags);
508 shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
509 shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages);
510 pgdat_resize_unlock(zone->zone_pgdat, &flags);
513 static void __remove_section(struct zone *zone, struct mem_section *ms,
514 unsigned long map_offset,
515 struct vmem_altmap *altmap)
517 unsigned long start_pfn;
520 if (WARN_ON_ONCE(!valid_section(ms)))
523 scn_nr = __section_nr(ms);
524 start_pfn = section_nr_to_pfn((unsigned long)scn_nr);
525 __remove_zone(zone, start_pfn);
527 sparse_remove_one_section(zone, ms, map_offset, altmap);
531 * __remove_pages() - remove sections of pages from a zone
532 * @zone: zone from which pages need to be removed
533 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
534 * @nr_pages: number of pages to remove (must be multiple of section size)
535 * @altmap: alternative device page map or %NULL if default memmap is used
537 * Generic helper function to remove section mappings and sysfs entries
538 * for the section of the memory we are removing. Caller needs to make
539 * sure that pages are marked reserved and zones are adjust properly by
540 * calling offline_pages().
542 void __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
543 unsigned long nr_pages, struct vmem_altmap *altmap)
546 unsigned long map_offset = 0;
547 int sections_to_remove;
549 /* In the ZONE_DEVICE case device driver owns the memory region */
550 if (is_dev_zone(zone))
551 map_offset = vmem_altmap_offset(altmap);
553 clear_zone_contiguous(zone);
556 * We can only remove entire sections
558 BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
559 BUG_ON(nr_pages % PAGES_PER_SECTION);
561 sections_to_remove = nr_pages / PAGES_PER_SECTION;
562 for (i = 0; i < sections_to_remove; i++) {
563 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
566 __remove_section(zone, __pfn_to_section(pfn), map_offset,
571 set_zone_contiguous(zone);
574 int set_online_page_callback(online_page_callback_t callback)
579 mutex_lock(&online_page_callback_lock);
581 if (online_page_callback == generic_online_page) {
582 online_page_callback = callback;
586 mutex_unlock(&online_page_callback_lock);
591 EXPORT_SYMBOL_GPL(set_online_page_callback);
593 int restore_online_page_callback(online_page_callback_t callback)
598 mutex_lock(&online_page_callback_lock);
600 if (online_page_callback == callback) {
601 online_page_callback = generic_online_page;
605 mutex_unlock(&online_page_callback_lock);
610 EXPORT_SYMBOL_GPL(restore_online_page_callback);
612 void __online_page_set_limits(struct page *page)
615 EXPORT_SYMBOL_GPL(__online_page_set_limits);
617 void __online_page_increment_counters(struct page *page)
619 adjust_managed_page_count(page, 1);
621 EXPORT_SYMBOL_GPL(__online_page_increment_counters);
623 void __online_page_free(struct page *page)
625 __free_reserved_page(page);
627 EXPORT_SYMBOL_GPL(__online_page_free);
629 static void generic_online_page(struct page *page, unsigned int order)
631 kernel_map_pages(page, 1 << order, 1);
632 __free_pages_core(page, order);
633 totalram_pages_add(1UL << order);
634 #ifdef CONFIG_HIGHMEM
635 if (PageHighMem(page))
636 totalhigh_pages_add(1UL << order);
640 static int online_pages_blocks(unsigned long start, unsigned long nr_pages)
642 unsigned long end = start + nr_pages;
643 int order, onlined_pages = 0;
645 while (start < end) {
646 order = min(MAX_ORDER - 1,
647 get_order(PFN_PHYS(end) - PFN_PHYS(start)));
648 (*online_page_callback)(pfn_to_page(start), order);
650 onlined_pages += (1UL << order);
651 start += (1UL << order);
653 return onlined_pages;
656 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
659 unsigned long onlined_pages = *(unsigned long *)arg;
661 if (PageReserved(pfn_to_page(start_pfn)))
662 onlined_pages += online_pages_blocks(start_pfn, nr_pages);
664 online_mem_sections(start_pfn, start_pfn + nr_pages);
666 *(unsigned long *)arg = onlined_pages;
670 /* check which state of node_states will be changed when online memory */
671 static void node_states_check_changes_online(unsigned long nr_pages,
672 struct zone *zone, struct memory_notify *arg)
674 int nid = zone_to_nid(zone);
676 arg->status_change_nid = NUMA_NO_NODE;
677 arg->status_change_nid_normal = NUMA_NO_NODE;
678 arg->status_change_nid_high = NUMA_NO_NODE;
680 if (!node_state(nid, N_MEMORY))
681 arg->status_change_nid = nid;
682 if (zone_idx(zone) <= ZONE_NORMAL && !node_state(nid, N_NORMAL_MEMORY))
683 arg->status_change_nid_normal = nid;
684 #ifdef CONFIG_HIGHMEM
685 if (zone_idx(zone) <= ZONE_HIGHMEM && !node_state(nid, N_HIGH_MEMORY))
686 arg->status_change_nid_high = nid;
690 static void node_states_set_node(int node, struct memory_notify *arg)
692 if (arg->status_change_nid_normal >= 0)
693 node_set_state(node, N_NORMAL_MEMORY);
695 if (arg->status_change_nid_high >= 0)
696 node_set_state(node, N_HIGH_MEMORY);
698 if (arg->status_change_nid >= 0)
699 node_set_state(node, N_MEMORY);
702 static void __meminit resize_zone_range(struct zone *zone, unsigned long start_pfn,
703 unsigned long nr_pages)
705 unsigned long old_end_pfn = zone_end_pfn(zone);
707 if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
708 zone->zone_start_pfn = start_pfn;
710 zone->spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - zone->zone_start_pfn;
713 static void __meminit resize_pgdat_range(struct pglist_data *pgdat, unsigned long start_pfn,
714 unsigned long nr_pages)
716 unsigned long old_end_pfn = pgdat_end_pfn(pgdat);
718 if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
719 pgdat->node_start_pfn = start_pfn;
721 pgdat->node_spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - pgdat->node_start_pfn;
724 void __ref move_pfn_range_to_zone(struct zone *zone, unsigned long start_pfn,
725 unsigned long nr_pages, struct vmem_altmap *altmap)
727 struct pglist_data *pgdat = zone->zone_pgdat;
728 int nid = pgdat->node_id;
731 clear_zone_contiguous(zone);
733 /* TODO Huh pgdat is irqsave while zone is not. It used to be like that before */
734 pgdat_resize_lock(pgdat, &flags);
735 zone_span_writelock(zone);
736 if (zone_is_empty(zone))
737 init_currently_empty_zone(zone, start_pfn, nr_pages);
738 resize_zone_range(zone, start_pfn, nr_pages);
739 zone_span_writeunlock(zone);
740 resize_pgdat_range(pgdat, start_pfn, nr_pages);
741 pgdat_resize_unlock(pgdat, &flags);
744 * TODO now we have a visible range of pages which are not associated
745 * with their zone properly. Not nice but set_pfnblock_flags_mask
746 * expects the zone spans the pfn range. All the pages in the range
747 * are reserved so nobody should be touching them so we should be safe
749 memmap_init_zone(nr_pages, nid, zone_idx(zone), start_pfn,
750 MEMMAP_HOTPLUG, altmap);
752 set_zone_contiguous(zone);
756 * Returns a default kernel memory zone for the given pfn range.
757 * If no kernel zone covers this pfn range it will automatically go
758 * to the ZONE_NORMAL.
760 static struct zone *default_kernel_zone_for_pfn(int nid, unsigned long start_pfn,
761 unsigned long nr_pages)
763 struct pglist_data *pgdat = NODE_DATA(nid);
766 for (zid = 0; zid <= ZONE_NORMAL; zid++) {
767 struct zone *zone = &pgdat->node_zones[zid];
769 if (zone_intersects(zone, start_pfn, nr_pages))
773 return &pgdat->node_zones[ZONE_NORMAL];
776 static inline struct zone *default_zone_for_pfn(int nid, unsigned long start_pfn,
777 unsigned long nr_pages)
779 struct zone *kernel_zone = default_kernel_zone_for_pfn(nid, start_pfn,
781 struct zone *movable_zone = &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
782 bool in_kernel = zone_intersects(kernel_zone, start_pfn, nr_pages);
783 bool in_movable = zone_intersects(movable_zone, start_pfn, nr_pages);
786 * We inherit the existing zone in a simple case where zones do not
787 * overlap in the given range
789 if (in_kernel ^ in_movable)
790 return (in_kernel) ? kernel_zone : movable_zone;
793 * If the range doesn't belong to any zone or two zones overlap in the
794 * given range then we use movable zone only if movable_node is
795 * enabled because we always online to a kernel zone by default.
797 return movable_node_enabled ? movable_zone : kernel_zone;
800 struct zone * zone_for_pfn_range(int online_type, int nid, unsigned start_pfn,
801 unsigned long nr_pages)
803 if (online_type == MMOP_ONLINE_KERNEL)
804 return default_kernel_zone_for_pfn(nid, start_pfn, nr_pages);
806 if (online_type == MMOP_ONLINE_MOVABLE)
807 return &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
809 return default_zone_for_pfn(nid, start_pfn, nr_pages);
813 * Associates the given pfn range with the given node and the zone appropriate
814 * for the given online type.
816 static struct zone * __meminit move_pfn_range(int online_type, int nid,
817 unsigned long start_pfn, unsigned long nr_pages)
821 zone = zone_for_pfn_range(online_type, nid, start_pfn, nr_pages);
822 move_pfn_range_to_zone(zone, start_pfn, nr_pages, NULL);
826 int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type)
829 unsigned long onlined_pages = 0;
831 int need_zonelists_rebuild = 0;
834 struct memory_notify arg;
835 struct memory_block *mem;
840 * We can't use pfn_to_nid() because nid might be stored in struct page
841 * which is not yet initialized. Instead, we find nid from memory block.
843 mem = find_memory_block(__pfn_to_section(pfn));
845 put_device(&mem->dev);
847 /* associate pfn range with the zone */
848 zone = move_pfn_range(online_type, nid, pfn, nr_pages);
851 arg.nr_pages = nr_pages;
852 node_states_check_changes_online(nr_pages, zone, &arg);
854 ret = memory_notify(MEM_GOING_ONLINE, &arg);
855 ret = notifier_to_errno(ret);
857 goto failed_addition;
860 * If this zone is not populated, then it is not in zonelist.
861 * This means the page allocator ignores this zone.
862 * So, zonelist must be updated after online.
864 if (!populated_zone(zone)) {
865 need_zonelists_rebuild = 1;
866 setup_zone_pageset(zone);
869 ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
872 if (need_zonelists_rebuild)
873 zone_pcp_reset(zone);
874 goto failed_addition;
877 zone->present_pages += onlined_pages;
879 pgdat_resize_lock(zone->zone_pgdat, &flags);
880 zone->zone_pgdat->node_present_pages += onlined_pages;
881 pgdat_resize_unlock(zone->zone_pgdat, &flags);
886 node_states_set_node(nid, &arg);
887 if (need_zonelists_rebuild)
888 build_all_zonelists(NULL);
890 zone_pcp_update(zone);
893 init_per_zone_wmark_min();
900 vm_total_pages = nr_free_pagecache_pages();
902 writeback_set_ratelimit();
905 memory_notify(MEM_ONLINE, &arg);
910 pr_debug("online_pages [mem %#010llx-%#010llx] failed\n",
911 (unsigned long long) pfn << PAGE_SHIFT,
912 (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1);
913 memory_notify(MEM_CANCEL_ONLINE, &arg);
917 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
919 static void reset_node_present_pages(pg_data_t *pgdat)
923 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
924 z->present_pages = 0;
926 pgdat->node_present_pages = 0;
929 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
930 static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
932 struct pglist_data *pgdat;
933 unsigned long start_pfn = PFN_DOWN(start);
935 pgdat = NODE_DATA(nid);
937 pgdat = arch_alloc_nodedata(nid);
941 arch_refresh_nodedata(nid, pgdat);
944 * Reset the nr_zones, order and classzone_idx before reuse.
945 * Note that kswapd will init kswapd_classzone_idx properly
946 * when it starts in the near future.
949 pgdat->kswapd_order = 0;
950 pgdat->kswapd_classzone_idx = 0;
953 /* we can use NODE_DATA(nid) from here */
955 pgdat->node_id = nid;
956 pgdat->node_start_pfn = start_pfn;
958 /* init node's zones as empty zones, we don't have any present pages.*/
959 free_area_init_core_hotplug(nid);
960 pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat);
963 * The node we allocated has no zone fallback lists. For avoiding
964 * to access not-initialized zonelist, build here.
966 build_all_zonelists(pgdat);
969 * When memory is hot-added, all the memory is in offline state. So
970 * clear all zones' present_pages because they will be updated in
971 * online_pages() and offline_pages().
973 reset_node_managed_pages(pgdat);
974 reset_node_present_pages(pgdat);
979 static void rollback_node_hotadd(int nid)
981 pg_data_t *pgdat = NODE_DATA(nid);
983 arch_refresh_nodedata(nid, NULL);
984 free_percpu(pgdat->per_cpu_nodestats);
985 arch_free_nodedata(pgdat);
991 * try_online_node - online a node if offlined
993 * @start: start addr of the node
994 * @set_node_online: Whether we want to online the node
995 * called by cpu_up() to online a node without onlined memory.
998 * 1 -> a new node has been allocated
999 * 0 -> the node is already online
1000 * -ENOMEM -> the node could not be allocated
1002 static int __try_online_node(int nid, u64 start, bool set_node_online)
1007 if (node_online(nid))
1010 pgdat = hotadd_new_pgdat(nid, start);
1012 pr_err("Cannot online node %d due to NULL pgdat\n", nid);
1017 if (set_node_online) {
1018 node_set_online(nid);
1019 ret = register_one_node(nid);
1027 * Users of this function always want to online/register the node
1029 int try_online_node(int nid)
1033 mem_hotplug_begin();
1034 ret = __try_online_node(nid, 0, true);
1039 static int check_hotplug_memory_range(u64 start, u64 size)
1041 /* memory range must be block size aligned */
1042 if (!size || !IS_ALIGNED(start, memory_block_size_bytes()) ||
1043 !IS_ALIGNED(size, memory_block_size_bytes())) {
1044 pr_err("Block size [%#lx] unaligned hotplug range: start %#llx, size %#llx",
1045 memory_block_size_bytes(), start, size);
1052 static int online_memory_block(struct memory_block *mem, void *arg)
1054 return device_online(&mem->dev);
1058 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1059 * and online/offline operations (triggered e.g. by sysfs).
1061 * we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG
1063 int __ref add_memory_resource(int nid, struct resource *res)
1065 struct mhp_restrictions restrictions = {};
1067 bool new_node = false;
1071 size = resource_size(res);
1073 ret = check_hotplug_memory_range(start, size);
1077 mem_hotplug_begin();
1080 * Add new range to memblock so that when hotadd_new_pgdat() is called
1081 * to allocate new pgdat, get_pfn_range_for_nid() will be able to find
1082 * this new range and calculate total pages correctly. The range will
1083 * be removed at hot-remove time.
1085 memblock_add_node(start, size, nid);
1087 ret = __try_online_node(nid, start, false);
1092 /* call arch's memory hotadd */
1093 ret = arch_add_memory(nid, start, size, &restrictions);
1097 /* create memory block devices after memory was added */
1098 ret = create_memory_block_devices(start, size);
1100 arch_remove_memory(nid, start, size, NULL);
1105 /* If sysfs file of new node can't be created, cpu on the node
1106 * can't be hot-added. There is no rollback way now.
1107 * So, check by BUG_ON() to catch it reluctantly..
1108 * We online node here. We can't roll back from here.
1110 node_set_online(nid);
1111 ret = __register_one_node(nid);
1115 /* link memory sections under this node.*/
1116 ret = link_mem_sections(nid, PFN_DOWN(start), PFN_UP(start + size - 1));
1119 /* create new memmap entry */
1120 firmware_map_add_hotplug(start, start + size, "System RAM");
1122 /* device_online() will take the lock when calling online_pages() */
1125 /* online pages if requested */
1126 if (memhp_auto_online)
1127 walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1),
1128 NULL, online_memory_block);
1132 /* rollback pgdat allocation and others */
1134 rollback_node_hotadd(nid);
1135 memblock_remove(start, size);
1140 /* requires device_hotplug_lock, see add_memory_resource() */
1141 int __ref __add_memory(int nid, u64 start, u64 size)
1143 struct resource *res;
1146 res = register_memory_resource(start, size);
1148 return PTR_ERR(res);
1150 ret = add_memory_resource(nid, res);
1152 release_memory_resource(res);
1156 int add_memory(int nid, u64 start, u64 size)
1160 lock_device_hotplug();
1161 rc = __add_memory(nid, start, size);
1162 unlock_device_hotplug();
1166 EXPORT_SYMBOL_GPL(add_memory);
1168 #ifdef CONFIG_MEMORY_HOTREMOVE
1170 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1171 * set and the size of the free page is given by page_order(). Using this,
1172 * the function determines if the pageblock contains only free pages.
1173 * Due to buddy contraints, a free page at least the size of a pageblock will
1174 * be located at the start of the pageblock
1176 static inline int pageblock_free(struct page *page)
1178 return PageBuddy(page) && page_order(page) >= pageblock_order;
1181 /* Return the pfn of the start of the next active pageblock after a given pfn */
1182 static unsigned long next_active_pageblock(unsigned long pfn)
1184 struct page *page = pfn_to_page(pfn);
1186 /* Ensure the starting page is pageblock-aligned */
1187 BUG_ON(pfn & (pageblock_nr_pages - 1));
1189 /* If the entire pageblock is free, move to the end of free page */
1190 if (pageblock_free(page)) {
1192 /* be careful. we don't have locks, page_order can be changed.*/
1193 order = page_order(page);
1194 if ((order < MAX_ORDER) && (order >= pageblock_order))
1195 return pfn + (1 << order);
1198 return pfn + pageblock_nr_pages;
1201 static bool is_pageblock_removable_nolock(unsigned long pfn)
1203 struct page *page = pfn_to_page(pfn);
1207 * We have to be careful here because we are iterating over memory
1208 * sections which are not zone aware so we might end up outside of
1209 * the zone but still within the section.
1210 * We have to take care about the node as well. If the node is offline
1211 * its NODE_DATA will be NULL - see page_zone.
1213 if (!node_online(page_to_nid(page)))
1216 zone = page_zone(page);
1217 pfn = page_to_pfn(page);
1218 if (!zone_spans_pfn(zone, pfn))
1221 return !has_unmovable_pages(zone, page, 0, MIGRATE_MOVABLE, SKIP_HWPOISON);
1224 /* Checks if this range of memory is likely to be hot-removable. */
1225 bool is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
1227 unsigned long end_pfn, pfn;
1229 end_pfn = min(start_pfn + nr_pages,
1230 zone_end_pfn(page_zone(pfn_to_page(start_pfn))));
1232 /* Check the starting page of each pageblock within the range */
1233 for (pfn = start_pfn; pfn < end_pfn; pfn = next_active_pageblock(pfn)) {
1234 if (!is_pageblock_removable_nolock(pfn))
1239 /* All pageblocks in the memory block are likely to be hot-removable */
1244 * Confirm all pages in a range [start, end) belong to the same zone.
1245 * When true, return its valid [start, end).
1247 int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn,
1248 unsigned long *valid_start, unsigned long *valid_end)
1250 unsigned long pfn, sec_end_pfn;
1251 unsigned long start, end;
1252 struct zone *zone = NULL;
1255 for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1);
1257 pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) {
1258 /* Make sure the memory section is present first */
1259 if (!present_section_nr(pfn_to_section_nr(pfn)))
1261 for (; pfn < sec_end_pfn && pfn < end_pfn;
1262 pfn += MAX_ORDER_NR_PAGES) {
1264 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1265 while ((i < MAX_ORDER_NR_PAGES) &&
1266 !pfn_valid_within(pfn + i))
1268 if (i == MAX_ORDER_NR_PAGES || pfn + i >= end_pfn)
1270 /* Check if we got outside of the zone */
1271 if (zone && !zone_spans_pfn(zone, pfn + i))
1273 page = pfn_to_page(pfn + i);
1274 if (zone && page_zone(page) != zone)
1278 zone = page_zone(page);
1279 end = pfn + MAX_ORDER_NR_PAGES;
1284 *valid_start = start;
1285 *valid_end = min(end, end_pfn);
1293 * Scan pfn range [start,end) to find movable/migratable pages (LRU pages,
1294 * non-lru movable pages and hugepages). We scan pfn because it's much
1295 * easier than scanning over linked list. This function returns the pfn
1296 * of the first found movable page if it's found, otherwise 0.
1298 static unsigned long scan_movable_pages(unsigned long start, unsigned long end)
1302 for (pfn = start; pfn < end; pfn++) {
1303 struct page *page, *head;
1306 if (!pfn_valid(pfn))
1308 page = pfn_to_page(pfn);
1311 if (__PageMovable(page))
1314 if (!PageHuge(page))
1316 head = compound_head(page);
1317 if (page_huge_active(head))
1319 skip = (1 << compound_order(head)) - (page - head);
1325 static struct page *new_node_page(struct page *page, unsigned long private)
1327 int nid = page_to_nid(page);
1328 nodemask_t nmask = node_states[N_MEMORY];
1331 * try to allocate from a different node but reuse this node if there
1332 * are no other online nodes to be used (e.g. we are offlining a part
1333 * of the only existing node)
1335 node_clear(nid, nmask);
1336 if (nodes_empty(nmask))
1337 node_set(nid, nmask);
1339 return new_page_nodemask(page, nid, &nmask);
1343 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1350 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1351 if (!pfn_valid(pfn))
1353 page = pfn_to_page(pfn);
1355 if (PageHuge(page)) {
1356 struct page *head = compound_head(page);
1357 pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1;
1358 isolate_huge_page(head, &source);
1360 } else if (PageTransHuge(page))
1361 pfn = page_to_pfn(compound_head(page))
1362 + hpage_nr_pages(page) - 1;
1365 * HWPoison pages have elevated reference counts so the migration would
1366 * fail on them. It also doesn't make any sense to migrate them in the
1367 * first place. Still try to unmap such a page in case it is still mapped
1368 * (e.g. current hwpoison implementation doesn't unmap KSM pages but keep
1369 * the unmap as the catch all safety net).
1371 if (PageHWPoison(page)) {
1372 if (WARN_ON(PageLRU(page)))
1373 isolate_lru_page(page);
1374 if (page_mapped(page))
1375 try_to_unmap(page, TTU_IGNORE_MLOCK | TTU_IGNORE_ACCESS);
1379 if (!get_page_unless_zero(page))
1382 * We can skip free pages. And we can deal with pages on
1383 * LRU and non-lru movable pages.
1386 ret = isolate_lru_page(page);
1388 ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE);
1389 if (!ret) { /* Success */
1390 list_add_tail(&page->lru, &source);
1391 if (!__PageMovable(page))
1392 inc_node_page_state(page, NR_ISOLATED_ANON +
1393 page_is_file_cache(page));
1396 pr_warn("failed to isolate pfn %lx\n", pfn);
1397 dump_page(page, "isolation failed");
1401 if (!list_empty(&source)) {
1402 /* Allocate a new page from the nearest neighbor node */
1403 ret = migrate_pages(&source, new_node_page, NULL, 0,
1404 MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
1406 list_for_each_entry(page, &source, lru) {
1407 pr_warn("migrating pfn %lx failed ret:%d ",
1408 page_to_pfn(page), ret);
1409 dump_page(page, "migration failure");
1411 putback_movable_pages(&source);
1419 * remove from free_area[] and mark all as Reserved.
1422 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1425 unsigned long *offlined_pages = (unsigned long *)data;
1427 *offlined_pages += __offline_isolated_pages(start, start + nr_pages);
1432 * Check all pages in range, recoreded as memory resource, are isolated.
1435 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1438 return test_pages_isolated(start_pfn, start_pfn + nr_pages, true);
1441 static int __init cmdline_parse_movable_node(char *p)
1443 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1444 movable_node_enabled = true;
1446 pr_warn("movable_node parameter depends on CONFIG_HAVE_MEMBLOCK_NODE_MAP to work properly\n");
1450 early_param("movable_node", cmdline_parse_movable_node);
1452 /* check which state of node_states will be changed when offline memory */
1453 static void node_states_check_changes_offline(unsigned long nr_pages,
1454 struct zone *zone, struct memory_notify *arg)
1456 struct pglist_data *pgdat = zone->zone_pgdat;
1457 unsigned long present_pages = 0;
1460 arg->status_change_nid = NUMA_NO_NODE;
1461 arg->status_change_nid_normal = NUMA_NO_NODE;
1462 arg->status_change_nid_high = NUMA_NO_NODE;
1465 * Check whether node_states[N_NORMAL_MEMORY] will be changed.
1466 * If the memory to be offline is within the range
1467 * [0..ZONE_NORMAL], and it is the last present memory there,
1468 * the zones in that range will become empty after the offlining,
1469 * thus we can determine that we need to clear the node from
1470 * node_states[N_NORMAL_MEMORY].
1472 for (zt = 0; zt <= ZONE_NORMAL; zt++)
1473 present_pages += pgdat->node_zones[zt].present_pages;
1474 if (zone_idx(zone) <= ZONE_NORMAL && nr_pages >= present_pages)
1475 arg->status_change_nid_normal = zone_to_nid(zone);
1477 #ifdef CONFIG_HIGHMEM
1479 * node_states[N_HIGH_MEMORY] contains nodes which
1480 * have normal memory or high memory.
1481 * Here we add the present_pages belonging to ZONE_HIGHMEM.
1482 * If the zone is within the range of [0..ZONE_HIGHMEM), and
1483 * we determine that the zones in that range become empty,
1484 * we need to clear the node for N_HIGH_MEMORY.
1486 present_pages += pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1487 if (zone_idx(zone) <= ZONE_HIGHMEM && nr_pages >= present_pages)
1488 arg->status_change_nid_high = zone_to_nid(zone);
1492 * We have accounted the pages from [0..ZONE_NORMAL), and
1493 * in case of CONFIG_HIGHMEM the pages from ZONE_HIGHMEM
1495 * Here we count the possible pages from ZONE_MOVABLE.
1496 * If after having accounted all the pages, we see that the nr_pages
1497 * to be offlined is over or equal to the accounted pages,
1498 * we know that the node will become empty, and so, we can clear
1499 * it for N_MEMORY as well.
1501 present_pages += pgdat->node_zones[ZONE_MOVABLE].present_pages;
1503 if (nr_pages >= present_pages)
1504 arg->status_change_nid = zone_to_nid(zone);
1507 static void node_states_clear_node(int node, struct memory_notify *arg)
1509 if (arg->status_change_nid_normal >= 0)
1510 node_clear_state(node, N_NORMAL_MEMORY);
1512 if (arg->status_change_nid_high >= 0)
1513 node_clear_state(node, N_HIGH_MEMORY);
1515 if (arg->status_change_nid >= 0)
1516 node_clear_state(node, N_MEMORY);
1519 static int __ref __offline_pages(unsigned long start_pfn,
1520 unsigned long end_pfn)
1522 unsigned long pfn, nr_pages;
1523 unsigned long offlined_pages = 0;
1524 int ret, node, nr_isolate_pageblock;
1525 unsigned long flags;
1526 unsigned long valid_start, valid_end;
1528 struct memory_notify arg;
1531 mem_hotplug_begin();
1533 /* This makes hotplug much easier...and readable.
1534 we assume this for now. .*/
1535 if (!test_pages_in_a_zone(start_pfn, end_pfn, &valid_start,
1538 reason = "multizone range";
1539 goto failed_removal;
1542 zone = page_zone(pfn_to_page(valid_start));
1543 node = zone_to_nid(zone);
1544 nr_pages = end_pfn - start_pfn;
1546 /* set above range as isolated */
1547 ret = start_isolate_page_range(start_pfn, end_pfn,
1549 SKIP_HWPOISON | REPORT_FAILURE);
1551 reason = "failure to isolate range";
1552 goto failed_removal;
1554 nr_isolate_pageblock = ret;
1556 arg.start_pfn = start_pfn;
1557 arg.nr_pages = nr_pages;
1558 node_states_check_changes_offline(nr_pages, zone, &arg);
1560 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1561 ret = notifier_to_errno(ret);
1563 reason = "notifier failure";
1564 goto failed_removal_isolated;
1568 for (pfn = start_pfn; pfn;) {
1569 if (signal_pending(current)) {
1571 reason = "signal backoff";
1572 goto failed_removal_isolated;
1576 lru_add_drain_all();
1578 pfn = scan_movable_pages(pfn, end_pfn);
1581 * TODO: fatal migration failures should bail
1584 do_migrate_range(pfn, end_pfn);
1589 * Dissolve free hugepages in the memory block before doing
1590 * offlining actually in order to make hugetlbfs's object
1591 * counting consistent.
1593 ret = dissolve_free_huge_pages(start_pfn, end_pfn);
1595 reason = "failure to dissolve huge pages";
1596 goto failed_removal_isolated;
1599 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn,
1600 NULL, check_pages_isolated_cb);
1603 /* Ok, all of our target is isolated.
1604 We cannot do rollback at this point. */
1605 walk_system_ram_range(start_pfn, end_pfn - start_pfn,
1606 &offlined_pages, offline_isolated_pages_cb);
1607 pr_info("Offlined Pages %ld\n", offlined_pages);
1609 * Onlining will reset pagetype flags and makes migrate type
1610 * MOVABLE, so just need to decrease the number of isolated
1611 * pageblocks zone counter here.
1613 spin_lock_irqsave(&zone->lock, flags);
1614 zone->nr_isolate_pageblock -= nr_isolate_pageblock;
1615 spin_unlock_irqrestore(&zone->lock, flags);
1617 /* removal success */
1618 adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages);
1619 zone->present_pages -= offlined_pages;
1621 pgdat_resize_lock(zone->zone_pgdat, &flags);
1622 zone->zone_pgdat->node_present_pages -= offlined_pages;
1623 pgdat_resize_unlock(zone->zone_pgdat, &flags);
1625 init_per_zone_wmark_min();
1627 if (!populated_zone(zone)) {
1628 zone_pcp_reset(zone);
1629 build_all_zonelists(NULL);
1631 zone_pcp_update(zone);
1633 node_states_clear_node(node, &arg);
1634 if (arg.status_change_nid >= 0) {
1636 kcompactd_stop(node);
1639 vm_total_pages = nr_free_pagecache_pages();
1640 writeback_set_ratelimit();
1642 memory_notify(MEM_OFFLINE, &arg);
1646 failed_removal_isolated:
1647 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1648 memory_notify(MEM_CANCEL_OFFLINE, &arg);
1650 pr_debug("memory offlining [mem %#010llx-%#010llx] failed due to %s\n",
1651 (unsigned long long) start_pfn << PAGE_SHIFT,
1652 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1,
1654 /* pushback to free area */
1659 int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1661 return __offline_pages(start_pfn, start_pfn + nr_pages);
1663 #endif /* CONFIG_MEMORY_HOTREMOVE */
1666 * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
1667 * @start_pfn: start pfn of the memory range
1668 * @end_pfn: end pfn of the memory range
1669 * @arg: argument passed to func
1670 * @func: callback for each memory section walked
1672 * This function walks through all present mem sections in range
1673 * [start_pfn, end_pfn) and call func on each mem section.
1675 * Returns the return value of func.
1677 int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn,
1678 void *arg, int (*func)(struct memory_block *, void *))
1680 struct memory_block *mem = NULL;
1681 struct mem_section *section;
1682 unsigned long pfn, section_nr;
1685 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1686 section_nr = pfn_to_section_nr(pfn);
1687 if (!present_section_nr(section_nr))
1690 section = __nr_to_section(section_nr);
1691 /* same memblock? */
1693 if ((section_nr >= mem->start_section_nr) &&
1694 (section_nr <= mem->end_section_nr))
1697 mem = find_memory_block_hinted(section, mem);
1701 ret = func(mem, arg);
1703 kobject_put(&mem->dev.kobj);
1709 kobject_put(&mem->dev.kobj);
1714 #ifdef CONFIG_MEMORY_HOTREMOVE
1715 static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
1717 int ret = !is_memblock_offlined(mem);
1719 if (unlikely(ret)) {
1720 phys_addr_t beginpa, endpa;
1722 beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
1723 endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1;
1724 pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n",
1732 static int check_cpu_on_node(pg_data_t *pgdat)
1736 for_each_present_cpu(cpu) {
1737 if (cpu_to_node(cpu) == pgdat->node_id)
1739 * the cpu on this node isn't removed, and we can't
1740 * offline this node.
1752 * Offline a node if all memory sections and cpus of the node are removed.
1754 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1755 * and online/offline operations before this call.
1757 void try_offline_node(int nid)
1759 pg_data_t *pgdat = NODE_DATA(nid);
1760 unsigned long start_pfn = pgdat->node_start_pfn;
1761 unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
1764 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1765 unsigned long section_nr = pfn_to_section_nr(pfn);
1767 if (!present_section_nr(section_nr))
1770 if (pfn_to_nid(pfn) != nid)
1774 * some memory sections of this node are not removed, and we
1775 * can't offline node now.
1780 if (check_cpu_on_node(pgdat))
1784 * all memory/cpu of this node are removed, we can offline this
1787 node_set_offline(nid);
1788 unregister_one_node(nid);
1790 EXPORT_SYMBOL(try_offline_node);
1792 static void __release_memory_resource(resource_size_t start,
1793 resource_size_t size)
1798 * When removing memory in the same granularity as it was added,
1799 * this function never fails. It might only fail if resources
1800 * have to be adjusted or split. We'll ignore the error, as
1801 * removing of memory cannot fail.
1803 ret = release_mem_region_adjustable(&iomem_resource, start, size);
1805 resource_size_t endres = start + size - 1;
1807 pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
1808 &start, &endres, ret);
1812 static int __ref try_remove_memory(int nid, u64 start, u64 size)
1816 BUG_ON(check_hotplug_memory_range(start, size));
1818 mem_hotplug_begin();
1821 * All memory blocks must be offlined before removing memory. Check
1822 * whether all memory blocks in question are offline and return error
1823 * if this is not the case.
1825 rc = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL,
1826 check_memblock_offlined_cb);
1830 /* remove memmap entry */
1831 firmware_map_remove(start, start + size, "System RAM");
1832 memblock_free(start, size);
1833 memblock_remove(start, size);
1835 /* remove memory block devices before removing memory */
1836 remove_memory_block_devices(start, size);
1838 arch_remove_memory(nid, start, size, NULL);
1839 __release_memory_resource(start, size);
1841 try_offline_node(nid);
1851 * @start: physical address of the region to remove
1852 * @size: size of the region to remove
1854 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1855 * and online/offline operations before this call, as required by
1856 * try_offline_node().
1858 void __remove_memory(int nid, u64 start, u64 size)
1862 * trigger BUG() is some memory is not offlined prior to calling this
1865 if (try_remove_memory(nid, start, size))
1870 * Remove memory if every memory block is offline, otherwise return -EBUSY is
1871 * some memory is not offline
1873 int remove_memory(int nid, u64 start, u64 size)
1877 lock_device_hotplug();
1878 rc = try_remove_memory(nid, start, size);
1879 unlock_device_hotplug();
1883 EXPORT_SYMBOL_GPL(remove_memory);
1884 #endif /* CONFIG_MEMORY_HOTREMOVE */