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 u64 max_mem_size = U64_MAX;
101 /* add this memory to iomem resource */
102 static struct resource *register_memory_resource(u64 start, u64 size)
104 struct resource *res;
105 unsigned long flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
106 char *resource_name = "System RAM";
108 if (start + size > max_mem_size)
109 return ERR_PTR(-E2BIG);
112 * Request ownership of the new memory range. This might be
113 * a child of an existing resource that was present but
114 * not marked as busy.
116 res = __request_region(&iomem_resource, start, size,
117 resource_name, flags);
120 pr_debug("Unable to reserve System RAM region: %016llx->%016llx\n",
121 start, start + size);
122 return ERR_PTR(-EEXIST);
127 static void release_memory_resource(struct resource *res)
131 release_resource(res);
136 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
137 void get_page_bootmem(unsigned long info, struct page *page,
140 page->freelist = (void *)type;
141 SetPagePrivate(page);
142 set_page_private(page, info);
146 void put_page_bootmem(struct page *page)
150 type = (unsigned long) page->freelist;
151 BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE ||
152 type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE);
154 if (page_ref_dec_return(page) == 1) {
155 page->freelist = NULL;
156 ClearPagePrivate(page);
157 set_page_private(page, 0);
158 INIT_LIST_HEAD(&page->lru);
159 free_reserved_page(page);
163 #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
164 #ifndef CONFIG_SPARSEMEM_VMEMMAP
165 static void register_page_bootmem_info_section(unsigned long start_pfn)
167 unsigned long *usemap, mapsize, section_nr, i;
168 struct mem_section *ms;
169 struct page *page, *memmap;
171 section_nr = pfn_to_section_nr(start_pfn);
172 ms = __nr_to_section(section_nr);
174 /* Get section's memmap address */
175 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
178 * Get page for the memmap's phys address
179 * XXX: need more consideration for sparse_vmemmap...
181 page = virt_to_page(memmap);
182 mapsize = sizeof(struct page) * PAGES_PER_SECTION;
183 mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
185 /* remember memmap's page */
186 for (i = 0; i < mapsize; i++, page++)
187 get_page_bootmem(section_nr, page, SECTION_INFO);
189 usemap = ms->pageblock_flags;
190 page = virt_to_page(usemap);
192 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
194 for (i = 0; i < mapsize; i++, page++)
195 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
198 #else /* CONFIG_SPARSEMEM_VMEMMAP */
199 static void register_page_bootmem_info_section(unsigned long start_pfn)
201 unsigned long *usemap, mapsize, section_nr, i;
202 struct mem_section *ms;
203 struct page *page, *memmap;
205 section_nr = pfn_to_section_nr(start_pfn);
206 ms = __nr_to_section(section_nr);
208 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
210 register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION);
212 usemap = ms->pageblock_flags;
213 page = virt_to_page(usemap);
215 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
217 for (i = 0; i < mapsize; i++, page++)
218 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
220 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
222 void __init register_page_bootmem_info_node(struct pglist_data *pgdat)
224 unsigned long i, pfn, end_pfn, nr_pages;
225 int node = pgdat->node_id;
228 nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
229 page = virt_to_page(pgdat);
231 for (i = 0; i < nr_pages; i++, page++)
232 get_page_bootmem(node, page, NODE_INFO);
234 pfn = pgdat->node_start_pfn;
235 end_pfn = pgdat_end_pfn(pgdat);
237 /* register section info */
238 for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
240 * Some platforms can assign the same pfn to multiple nodes - on
241 * node0 as well as nodeN. To avoid registering a pfn against
242 * multiple nodes we check that this pfn does not already
243 * reside in some other nodes.
245 if (pfn_valid(pfn) && (early_pfn_to_nid(pfn) == node))
246 register_page_bootmem_info_section(pfn);
249 #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
251 static int __meminit __add_section(int nid, unsigned long phys_start_pfn,
252 struct vmem_altmap *altmap, bool want_memblock)
256 if (pfn_valid(phys_start_pfn))
259 ret = sparse_add_one_section(nid, phys_start_pfn, altmap);
266 return hotplug_memory_register(nid, __pfn_to_section(phys_start_pfn));
270 * Reasonably generic function for adding memory. It is
271 * expected that archs that support memory hotplug will
272 * call this function after deciding the zone to which to
275 int __ref __add_pages(int nid, unsigned long phys_start_pfn,
276 unsigned long nr_pages, struct mhp_restrictions *restrictions)
280 int start_sec, end_sec;
281 struct vmem_altmap *altmap = restrictions->altmap;
283 /* during initialize mem_map, align hot-added range to section */
284 start_sec = pfn_to_section_nr(phys_start_pfn);
285 end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
289 * Validate altmap is within bounds of the total request
291 if (altmap->base_pfn != phys_start_pfn
292 || vmem_altmap_offset(altmap) > nr_pages) {
293 pr_warn_once("memory add fail, invalid altmap\n");
300 for (i = start_sec; i <= end_sec; i++) {
301 err = __add_section(nid, section_nr_to_pfn(i), altmap,
302 restrictions->flags & MHP_MEMBLOCK_API);
305 * EEXIST is finally dealt with by ioresource collision
306 * check. see add_memory() => register_memory_resource()
307 * Warning will be printed if there is collision.
309 if (err && (err != -EEXIST))
314 vmemmap_populate_print_last();
319 #ifdef CONFIG_MEMORY_HOTREMOVE
320 /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
321 static unsigned long find_smallest_section_pfn(int nid, struct zone *zone,
322 unsigned long start_pfn,
323 unsigned long end_pfn)
325 struct mem_section *ms;
327 for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) {
328 ms = __pfn_to_section(start_pfn);
330 if (unlikely(!valid_section(ms)))
333 if (unlikely(pfn_to_nid(start_pfn) != nid))
336 if (zone && zone != page_zone(pfn_to_page(start_pfn)))
345 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
346 static unsigned long find_biggest_section_pfn(int nid, struct zone *zone,
347 unsigned long start_pfn,
348 unsigned long end_pfn)
350 struct mem_section *ms;
353 /* pfn is the end pfn of a memory section. */
355 for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) {
356 ms = __pfn_to_section(pfn);
358 if (unlikely(!valid_section(ms)))
361 if (unlikely(pfn_to_nid(pfn) != nid))
364 if (zone && zone != page_zone(pfn_to_page(pfn)))
373 static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
374 unsigned long end_pfn)
376 unsigned long zone_start_pfn = zone->zone_start_pfn;
377 unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */
378 unsigned long zone_end_pfn = z;
380 struct mem_section *ms;
381 int nid = zone_to_nid(zone);
383 zone_span_writelock(zone);
384 if (zone_start_pfn == start_pfn) {
386 * If the section is smallest section in the zone, it need
387 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
388 * In this case, we find second smallest valid mem_section
389 * for shrinking zone.
391 pfn = find_smallest_section_pfn(nid, zone, end_pfn,
394 zone->zone_start_pfn = pfn;
395 zone->spanned_pages = zone_end_pfn - pfn;
397 } else if (zone_end_pfn == end_pfn) {
399 * If the section is biggest section in the zone, it need
400 * shrink zone->spanned_pages.
401 * In this case, we find second biggest valid mem_section for
404 pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn,
407 zone->spanned_pages = pfn - zone_start_pfn + 1;
411 * The section is not biggest or smallest mem_section in the zone, it
412 * only creates a hole in the zone. So in this case, we need not
413 * change the zone. But perhaps, the zone has only hole data. Thus
414 * it check the zone has only hole or not.
416 pfn = zone_start_pfn;
417 for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) {
418 ms = __pfn_to_section(pfn);
420 if (unlikely(!valid_section(ms)))
423 if (page_zone(pfn_to_page(pfn)) != zone)
426 /* If the section is current section, it continues the loop */
427 if (start_pfn == pfn)
430 /* If we find valid section, we have nothing to do */
431 zone_span_writeunlock(zone);
435 /* The zone has no valid section */
436 zone->zone_start_pfn = 0;
437 zone->spanned_pages = 0;
438 zone_span_writeunlock(zone);
441 static void shrink_pgdat_span(struct pglist_data *pgdat,
442 unsigned long start_pfn, unsigned long end_pfn)
444 unsigned long pgdat_start_pfn = pgdat->node_start_pfn;
445 unsigned long p = pgdat_end_pfn(pgdat); /* pgdat_end_pfn namespace clash */
446 unsigned long pgdat_end_pfn = p;
448 struct mem_section *ms;
449 int nid = pgdat->node_id;
451 if (pgdat_start_pfn == start_pfn) {
453 * If the section is smallest section in the pgdat, it need
454 * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages.
455 * In this case, we find second smallest valid mem_section
456 * for shrinking zone.
458 pfn = find_smallest_section_pfn(nid, NULL, end_pfn,
461 pgdat->node_start_pfn = pfn;
462 pgdat->node_spanned_pages = pgdat_end_pfn - pfn;
464 } else if (pgdat_end_pfn == end_pfn) {
466 * If the section is biggest section in the pgdat, it need
467 * shrink pgdat->node_spanned_pages.
468 * In this case, we find second biggest valid mem_section for
471 pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn,
474 pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1;
478 * If the section is not biggest or smallest mem_section in the pgdat,
479 * it only creates a hole in the pgdat. So in this case, we need not
481 * But perhaps, the pgdat has only hole data. Thus it check the pgdat
482 * has only hole or not.
484 pfn = pgdat_start_pfn;
485 for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) {
486 ms = __pfn_to_section(pfn);
488 if (unlikely(!valid_section(ms)))
491 if (pfn_to_nid(pfn) != nid)
494 /* If the section is current section, it continues the loop */
495 if (start_pfn == pfn)
498 /* If we find valid section, we have nothing to do */
502 /* The pgdat has no valid section */
503 pgdat->node_start_pfn = 0;
504 pgdat->node_spanned_pages = 0;
507 static void __remove_zone(struct zone *zone, unsigned long start_pfn)
509 struct pglist_data *pgdat = zone->zone_pgdat;
510 int nr_pages = PAGES_PER_SECTION;
513 pgdat_resize_lock(zone->zone_pgdat, &flags);
514 shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
515 shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages);
516 pgdat_resize_unlock(zone->zone_pgdat, &flags);
519 static void __remove_section(struct zone *zone, struct mem_section *ms,
520 unsigned long map_offset,
521 struct vmem_altmap *altmap)
523 unsigned long start_pfn;
526 if (WARN_ON_ONCE(!valid_section(ms)))
529 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);
539 * __remove_pages() - remove sections of pages from a zone
540 * @zone: zone from which pages need to be removed
541 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
542 * @nr_pages: number of pages to remove (must be multiple of section size)
543 * @altmap: alternative device page map or %NULL if default memmap is used
545 * Generic helper function to remove section mappings and sysfs entries
546 * for the section of the memory we are removing. Caller needs to make
547 * sure that pages are marked reserved and zones are adjust properly by
548 * calling offline_pages().
550 int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
551 unsigned long nr_pages, struct vmem_altmap *altmap)
554 unsigned long map_offset = 0;
555 int sections_to_remove;
557 /* In the ZONE_DEVICE case device driver owns the memory region */
558 if (is_dev_zone(zone)) {
560 map_offset = vmem_altmap_offset(altmap);
563 clear_zone_contiguous(zone);
566 * We can only remove entire sections
568 BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
569 BUG_ON(nr_pages % PAGES_PER_SECTION);
571 sections_to_remove = nr_pages / PAGES_PER_SECTION;
572 for (i = 0; i < sections_to_remove; i++) {
573 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
576 __remove_section(zone, __pfn_to_section(pfn), map_offset,
581 set_zone_contiguous(zone);
584 #endif /* CONFIG_MEMORY_HOTREMOVE */
586 int set_online_page_callback(online_page_callback_t callback)
591 mutex_lock(&online_page_callback_lock);
593 if (online_page_callback == generic_online_page) {
594 online_page_callback = callback;
598 mutex_unlock(&online_page_callback_lock);
603 EXPORT_SYMBOL_GPL(set_online_page_callback);
605 int restore_online_page_callback(online_page_callback_t callback)
610 mutex_lock(&online_page_callback_lock);
612 if (online_page_callback == callback) {
613 online_page_callback = generic_online_page;
617 mutex_unlock(&online_page_callback_lock);
622 EXPORT_SYMBOL_GPL(restore_online_page_callback);
624 void __online_page_set_limits(struct page *page)
627 EXPORT_SYMBOL_GPL(__online_page_set_limits);
629 void __online_page_increment_counters(struct page *page)
631 adjust_managed_page_count(page, 1);
633 EXPORT_SYMBOL_GPL(__online_page_increment_counters);
635 void __online_page_free(struct page *page)
637 __free_reserved_page(page);
639 EXPORT_SYMBOL_GPL(__online_page_free);
641 static void generic_online_page(struct page *page, unsigned int order)
643 kernel_map_pages(page, 1 << order, 1);
644 __free_pages_core(page, order);
645 totalram_pages_add(1UL << order);
646 #ifdef CONFIG_HIGHMEM
647 if (PageHighMem(page))
648 totalhigh_pages_add(1UL << order);
652 static int online_pages_blocks(unsigned long start, unsigned long nr_pages)
654 unsigned long end = start + nr_pages;
655 int order, onlined_pages = 0;
657 while (start < end) {
658 order = min(MAX_ORDER - 1,
659 get_order(PFN_PHYS(end) - PFN_PHYS(start)));
660 (*online_page_callback)(pfn_to_page(start), order);
662 onlined_pages += (1UL << order);
663 start += (1UL << order);
665 return onlined_pages;
668 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
671 unsigned long onlined_pages = *(unsigned long *)arg;
673 if (PageReserved(pfn_to_page(start_pfn)))
674 onlined_pages += online_pages_blocks(start_pfn, nr_pages);
676 online_mem_sections(start_pfn, start_pfn + nr_pages);
678 *(unsigned long *)arg = onlined_pages;
682 /* check which state of node_states will be changed when online memory */
683 static void node_states_check_changes_online(unsigned long nr_pages,
684 struct zone *zone, struct memory_notify *arg)
686 int nid = zone_to_nid(zone);
688 arg->status_change_nid = NUMA_NO_NODE;
689 arg->status_change_nid_normal = NUMA_NO_NODE;
690 arg->status_change_nid_high = NUMA_NO_NODE;
692 if (!node_state(nid, N_MEMORY))
693 arg->status_change_nid = nid;
694 if (zone_idx(zone) <= ZONE_NORMAL && !node_state(nid, N_NORMAL_MEMORY))
695 arg->status_change_nid_normal = nid;
696 #ifdef CONFIG_HIGHMEM
697 if (zone_idx(zone) <= ZONE_HIGHMEM && !node_state(nid, N_HIGH_MEMORY))
698 arg->status_change_nid_high = nid;
702 static void node_states_set_node(int node, struct memory_notify *arg)
704 if (arg->status_change_nid_normal >= 0)
705 node_set_state(node, N_NORMAL_MEMORY);
707 if (arg->status_change_nid_high >= 0)
708 node_set_state(node, N_HIGH_MEMORY);
710 if (arg->status_change_nid >= 0)
711 node_set_state(node, N_MEMORY);
714 static void __meminit resize_zone_range(struct zone *zone, unsigned long start_pfn,
715 unsigned long nr_pages)
717 unsigned long old_end_pfn = zone_end_pfn(zone);
719 if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
720 zone->zone_start_pfn = start_pfn;
722 zone->spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - zone->zone_start_pfn;
725 static void __meminit resize_pgdat_range(struct pglist_data *pgdat, unsigned long start_pfn,
726 unsigned long nr_pages)
728 unsigned long old_end_pfn = pgdat_end_pfn(pgdat);
730 if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
731 pgdat->node_start_pfn = start_pfn;
733 pgdat->node_spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - pgdat->node_start_pfn;
736 void __ref move_pfn_range_to_zone(struct zone *zone, unsigned long start_pfn,
737 unsigned long nr_pages, struct vmem_altmap *altmap)
739 struct pglist_data *pgdat = zone->zone_pgdat;
740 int nid = pgdat->node_id;
743 clear_zone_contiguous(zone);
745 /* TODO Huh pgdat is irqsave while zone is not. It used to be like that before */
746 pgdat_resize_lock(pgdat, &flags);
747 zone_span_writelock(zone);
748 if (zone_is_empty(zone))
749 init_currently_empty_zone(zone, start_pfn, nr_pages);
750 resize_zone_range(zone, start_pfn, nr_pages);
751 zone_span_writeunlock(zone);
752 resize_pgdat_range(pgdat, start_pfn, nr_pages);
753 pgdat_resize_unlock(pgdat, &flags);
756 * TODO now we have a visible range of pages which are not associated
757 * with their zone properly. Not nice but set_pfnblock_flags_mask
758 * expects the zone spans the pfn range. All the pages in the range
759 * are reserved so nobody should be touching them so we should be safe
761 memmap_init_zone(nr_pages, nid, zone_idx(zone), start_pfn,
762 MEMMAP_HOTPLUG, altmap);
764 set_zone_contiguous(zone);
768 * Returns a default kernel memory zone for the given pfn range.
769 * If no kernel zone covers this pfn range it will automatically go
770 * to the ZONE_NORMAL.
772 static struct zone *default_kernel_zone_for_pfn(int nid, unsigned long start_pfn,
773 unsigned long nr_pages)
775 struct pglist_data *pgdat = NODE_DATA(nid);
778 for (zid = 0; zid <= ZONE_NORMAL; zid++) {
779 struct zone *zone = &pgdat->node_zones[zid];
781 if (zone_intersects(zone, start_pfn, nr_pages))
785 return &pgdat->node_zones[ZONE_NORMAL];
788 static inline struct zone *default_zone_for_pfn(int nid, unsigned long start_pfn,
789 unsigned long nr_pages)
791 struct zone *kernel_zone = default_kernel_zone_for_pfn(nid, start_pfn,
793 struct zone *movable_zone = &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
794 bool in_kernel = zone_intersects(kernel_zone, start_pfn, nr_pages);
795 bool in_movable = zone_intersects(movable_zone, start_pfn, nr_pages);
798 * We inherit the existing zone in a simple case where zones do not
799 * overlap in the given range
801 if (in_kernel ^ in_movable)
802 return (in_kernel) ? kernel_zone : movable_zone;
805 * If the range doesn't belong to any zone or two zones overlap in the
806 * given range then we use movable zone only if movable_node is
807 * enabled because we always online to a kernel zone by default.
809 return movable_node_enabled ? movable_zone : kernel_zone;
812 struct zone * zone_for_pfn_range(int online_type, int nid, unsigned start_pfn,
813 unsigned long nr_pages)
815 if (online_type == MMOP_ONLINE_KERNEL)
816 return default_kernel_zone_for_pfn(nid, start_pfn, nr_pages);
818 if (online_type == MMOP_ONLINE_MOVABLE)
819 return &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
821 return default_zone_for_pfn(nid, start_pfn, nr_pages);
825 * Associates the given pfn range with the given node and the zone appropriate
826 * for the given online type.
828 static struct zone * __meminit move_pfn_range(int online_type, int nid,
829 unsigned long start_pfn, unsigned long nr_pages)
833 zone = zone_for_pfn_range(online_type, nid, start_pfn, nr_pages);
834 move_pfn_range_to_zone(zone, start_pfn, nr_pages, NULL);
838 int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type)
841 unsigned long onlined_pages = 0;
843 int need_zonelists_rebuild = 0;
846 struct memory_notify arg;
847 struct memory_block *mem;
852 * We can't use pfn_to_nid() because nid might be stored in struct page
853 * which is not yet initialized. Instead, we find nid from memory block.
855 mem = find_memory_block(__pfn_to_section(pfn));
857 put_device(&mem->dev);
859 /* associate pfn range with the zone */
860 zone = move_pfn_range(online_type, nid, pfn, nr_pages);
863 arg.nr_pages = nr_pages;
864 node_states_check_changes_online(nr_pages, zone, &arg);
866 ret = memory_notify(MEM_GOING_ONLINE, &arg);
867 ret = notifier_to_errno(ret);
869 goto failed_addition;
872 * If this zone is not populated, then it is not in zonelist.
873 * This means the page allocator ignores this zone.
874 * So, zonelist must be updated after online.
876 if (!populated_zone(zone)) {
877 need_zonelists_rebuild = 1;
878 setup_zone_pageset(zone);
881 ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
884 if (need_zonelists_rebuild)
885 zone_pcp_reset(zone);
886 goto failed_addition;
889 zone->present_pages += onlined_pages;
891 pgdat_resize_lock(zone->zone_pgdat, &flags);
892 zone->zone_pgdat->node_present_pages += onlined_pages;
893 pgdat_resize_unlock(zone->zone_pgdat, &flags);
896 node_states_set_node(nid, &arg);
897 if (need_zonelists_rebuild)
898 build_all_zonelists(NULL);
900 zone_pcp_update(zone);
903 init_per_zone_wmark_min();
910 vm_total_pages = nr_free_pagecache_pages();
912 writeback_set_ratelimit();
915 memory_notify(MEM_ONLINE, &arg);
920 pr_debug("online_pages [mem %#010llx-%#010llx] failed\n",
921 (unsigned long long) pfn << PAGE_SHIFT,
922 (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1);
923 memory_notify(MEM_CANCEL_ONLINE, &arg);
927 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
929 static void reset_node_present_pages(pg_data_t *pgdat)
933 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
934 z->present_pages = 0;
936 pgdat->node_present_pages = 0;
939 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
940 static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
942 struct pglist_data *pgdat;
943 unsigned long start_pfn = PFN_DOWN(start);
945 pgdat = NODE_DATA(nid);
947 pgdat = arch_alloc_nodedata(nid);
951 arch_refresh_nodedata(nid, pgdat);
954 * Reset the nr_zones, order and classzone_idx before reuse.
955 * Note that kswapd will init kswapd_classzone_idx properly
956 * when it starts in the near future.
959 pgdat->kswapd_order = 0;
960 pgdat->kswapd_classzone_idx = 0;
963 /* we can use NODE_DATA(nid) from here */
965 pgdat->node_id = nid;
966 pgdat->node_start_pfn = start_pfn;
968 /* init node's zones as empty zones, we don't have any present pages.*/
969 free_area_init_core_hotplug(nid);
970 pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat);
973 * The node we allocated has no zone fallback lists. For avoiding
974 * to access not-initialized zonelist, build here.
976 build_all_zonelists(pgdat);
979 * When memory is hot-added, all the memory is in offline state. So
980 * clear all zones' present_pages because they will be updated in
981 * online_pages() and offline_pages().
983 reset_node_managed_pages(pgdat);
984 reset_node_present_pages(pgdat);
989 static void rollback_node_hotadd(int nid)
991 pg_data_t *pgdat = NODE_DATA(nid);
993 arch_refresh_nodedata(nid, NULL);
994 free_percpu(pgdat->per_cpu_nodestats);
995 arch_free_nodedata(pgdat);
1001 * try_online_node - online a node if offlined
1003 * @start: start addr of the node
1004 * @set_node_online: Whether we want to online the node
1005 * called by cpu_up() to online a node without onlined memory.
1008 * 1 -> a new node has been allocated
1009 * 0 -> the node is already online
1010 * -ENOMEM -> the node could not be allocated
1012 static int __try_online_node(int nid, u64 start, bool set_node_online)
1017 if (node_online(nid))
1020 pgdat = hotadd_new_pgdat(nid, start);
1022 pr_err("Cannot online node %d due to NULL pgdat\n", nid);
1027 if (set_node_online) {
1028 node_set_online(nid);
1029 ret = register_one_node(nid);
1037 * Users of this function always want to online/register the node
1039 int try_online_node(int nid)
1043 mem_hotplug_begin();
1044 ret = __try_online_node(nid, 0, true);
1049 static int check_hotplug_memory_range(u64 start, u64 size)
1051 unsigned long block_sz = memory_block_size_bytes();
1052 u64 block_nr_pages = block_sz >> PAGE_SHIFT;
1053 u64 nr_pages = size >> PAGE_SHIFT;
1054 u64 start_pfn = PFN_DOWN(start);
1056 /* memory range must be block size aligned */
1057 if (!nr_pages || !IS_ALIGNED(start_pfn, block_nr_pages) ||
1058 !IS_ALIGNED(nr_pages, block_nr_pages)) {
1059 pr_err("Block size [%#lx] unaligned hotplug range: start %#llx, size %#llx",
1060 block_sz, start, size);
1067 static int online_memory_block(struct memory_block *mem, void *arg)
1069 return device_online(&mem->dev);
1073 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1074 * and online/offline operations (triggered e.g. by sysfs).
1076 * we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG
1078 int __ref add_memory_resource(int nid, struct resource *res)
1080 struct mhp_restrictions restrictions = {
1081 .flags = MHP_MEMBLOCK_API,
1084 bool new_node = false;
1088 size = resource_size(res);
1090 ret = check_hotplug_memory_range(start, size);
1094 mem_hotplug_begin();
1097 * Add new range to memblock so that when hotadd_new_pgdat() is called
1098 * to allocate new pgdat, get_pfn_range_for_nid() will be able to find
1099 * this new range and calculate total pages correctly. The range will
1100 * be removed at hot-remove time.
1102 memblock_add_node(start, size, nid);
1104 ret = __try_online_node(nid, start, false);
1109 /* call arch's memory hotadd */
1110 ret = arch_add_memory(nid, start, size, &restrictions);
1115 /* If sysfs file of new node can't be created, cpu on the node
1116 * can't be hot-added. There is no rollback way now.
1117 * So, check by BUG_ON() to catch it reluctantly..
1118 * We online node here. We can't roll back from here.
1120 node_set_online(nid);
1121 ret = __register_one_node(nid);
1125 /* link memory sections under this node.*/
1126 ret = link_mem_sections(nid, PFN_DOWN(start), PFN_UP(start + size - 1));
1129 /* create new memmap entry */
1130 firmware_map_add_hotplug(start, start + size, "System RAM");
1132 /* device_online() will take the lock when calling online_pages() */
1135 /* online pages if requested */
1136 if (memhp_auto_online)
1137 walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1),
1138 NULL, online_memory_block);
1142 /* rollback pgdat allocation and others */
1144 rollback_node_hotadd(nid);
1145 memblock_remove(start, size);
1150 /* requires device_hotplug_lock, see add_memory_resource() */
1151 int __ref __add_memory(int nid, u64 start, u64 size)
1153 struct resource *res;
1156 res = register_memory_resource(start, size);
1158 return PTR_ERR(res);
1160 ret = add_memory_resource(nid, res);
1162 release_memory_resource(res);
1166 int add_memory(int nid, u64 start, u64 size)
1170 lock_device_hotplug();
1171 rc = __add_memory(nid, start, size);
1172 unlock_device_hotplug();
1176 EXPORT_SYMBOL_GPL(add_memory);
1178 #ifdef CONFIG_MEMORY_HOTREMOVE
1180 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1181 * set and the size of the free page is given by page_order(). Using this,
1182 * the function determines if the pageblock contains only free pages.
1183 * Due to buddy contraints, a free page at least the size of a pageblock will
1184 * be located at the start of the pageblock
1186 static inline int pageblock_free(struct page *page)
1188 return PageBuddy(page) && page_order(page) >= pageblock_order;
1191 /* Return the pfn of the start of the next active pageblock after a given pfn */
1192 static unsigned long next_active_pageblock(unsigned long pfn)
1194 struct page *page = pfn_to_page(pfn);
1196 /* Ensure the starting page is pageblock-aligned */
1197 BUG_ON(pfn & (pageblock_nr_pages - 1));
1199 /* If the entire pageblock is free, move to the end of free page */
1200 if (pageblock_free(page)) {
1202 /* be careful. we don't have locks, page_order can be changed.*/
1203 order = page_order(page);
1204 if ((order < MAX_ORDER) && (order >= pageblock_order))
1205 return pfn + (1 << order);
1208 return pfn + pageblock_nr_pages;
1211 static bool is_pageblock_removable_nolock(unsigned long pfn)
1213 struct page *page = pfn_to_page(pfn);
1217 * We have to be careful here because we are iterating over memory
1218 * sections which are not zone aware so we might end up outside of
1219 * the zone but still within the section.
1220 * We have to take care about the node as well. If the node is offline
1221 * its NODE_DATA will be NULL - see page_zone.
1223 if (!node_online(page_to_nid(page)))
1226 zone = page_zone(page);
1227 pfn = page_to_pfn(page);
1228 if (!zone_spans_pfn(zone, pfn))
1231 return !has_unmovable_pages(zone, page, 0, MIGRATE_MOVABLE, SKIP_HWPOISON);
1234 /* Checks if this range of memory is likely to be hot-removable. */
1235 bool is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
1237 unsigned long end_pfn, pfn;
1239 end_pfn = min(start_pfn + nr_pages,
1240 zone_end_pfn(page_zone(pfn_to_page(start_pfn))));
1242 /* Check the starting page of each pageblock within the range */
1243 for (pfn = start_pfn; pfn < end_pfn; pfn = next_active_pageblock(pfn)) {
1244 if (!is_pageblock_removable_nolock(pfn))
1249 /* All pageblocks in the memory block are likely to be hot-removable */
1254 * Confirm all pages in a range [start, end) belong to the same zone.
1255 * When true, return its valid [start, end).
1257 int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn,
1258 unsigned long *valid_start, unsigned long *valid_end)
1260 unsigned long pfn, sec_end_pfn;
1261 unsigned long start, end;
1262 struct zone *zone = NULL;
1265 for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1);
1267 pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) {
1268 /* Make sure the memory section is present first */
1269 if (!present_section_nr(pfn_to_section_nr(pfn)))
1271 for (; pfn < sec_end_pfn && pfn < end_pfn;
1272 pfn += MAX_ORDER_NR_PAGES) {
1274 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1275 while ((i < MAX_ORDER_NR_PAGES) &&
1276 !pfn_valid_within(pfn + i))
1278 if (i == MAX_ORDER_NR_PAGES || pfn + i >= end_pfn)
1280 /* Check if we got outside of the zone */
1281 if (zone && !zone_spans_pfn(zone, pfn + i))
1283 page = pfn_to_page(pfn + i);
1284 if (zone && page_zone(page) != zone)
1288 zone = page_zone(page);
1289 end = pfn + MAX_ORDER_NR_PAGES;
1294 *valid_start = start;
1295 *valid_end = min(end, end_pfn);
1303 * Scan pfn range [start,end) to find movable/migratable pages (LRU pages,
1304 * non-lru movable pages and hugepages). We scan pfn because it's much
1305 * easier than scanning over linked list. This function returns the pfn
1306 * of the first found movable page if it's found, otherwise 0.
1308 static unsigned long scan_movable_pages(unsigned long start, unsigned long end)
1312 for (pfn = start; pfn < end; pfn++) {
1313 struct page *page, *head;
1316 if (!pfn_valid(pfn))
1318 page = pfn_to_page(pfn);
1321 if (__PageMovable(page))
1324 if (!PageHuge(page))
1326 head = compound_head(page);
1327 if (page_huge_active(head))
1329 skip = (1 << compound_order(head)) - (page - head);
1335 static struct page *new_node_page(struct page *page, unsigned long private)
1337 int nid = page_to_nid(page);
1338 nodemask_t nmask = node_states[N_MEMORY];
1341 * try to allocate from a different node but reuse this node if there
1342 * are no other online nodes to be used (e.g. we are offlining a part
1343 * of the only existing node)
1345 node_clear(nid, nmask);
1346 if (nodes_empty(nmask))
1347 node_set(nid, nmask);
1349 return new_page_nodemask(page, nid, &nmask);
1353 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1360 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1361 if (!pfn_valid(pfn))
1363 page = pfn_to_page(pfn);
1365 if (PageHuge(page)) {
1366 struct page *head = compound_head(page);
1367 pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1;
1368 isolate_huge_page(head, &source);
1370 } else if (PageTransHuge(page))
1371 pfn = page_to_pfn(compound_head(page))
1372 + hpage_nr_pages(page) - 1;
1375 * HWPoison pages have elevated reference counts so the migration would
1376 * fail on them. It also doesn't make any sense to migrate them in the
1377 * first place. Still try to unmap such a page in case it is still mapped
1378 * (e.g. current hwpoison implementation doesn't unmap KSM pages but keep
1379 * the unmap as the catch all safety net).
1381 if (PageHWPoison(page)) {
1382 if (WARN_ON(PageLRU(page)))
1383 isolate_lru_page(page);
1384 if (page_mapped(page))
1385 try_to_unmap(page, TTU_IGNORE_MLOCK | TTU_IGNORE_ACCESS);
1389 if (!get_page_unless_zero(page))
1392 * We can skip free pages. And we can deal with pages on
1393 * LRU and non-lru movable pages.
1396 ret = isolate_lru_page(page);
1398 ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE);
1399 if (!ret) { /* Success */
1400 list_add_tail(&page->lru, &source);
1401 if (!__PageMovable(page))
1402 inc_node_page_state(page, NR_ISOLATED_ANON +
1403 page_is_file_cache(page));
1406 pr_warn("failed to isolate pfn %lx\n", pfn);
1407 dump_page(page, "isolation failed");
1411 if (!list_empty(&source)) {
1412 /* Allocate a new page from the nearest neighbor node */
1413 ret = migrate_pages(&source, new_node_page, NULL, 0,
1414 MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
1416 list_for_each_entry(page, &source, lru) {
1417 pr_warn("migrating pfn %lx failed ret:%d ",
1418 page_to_pfn(page), ret);
1419 dump_page(page, "migration failure");
1421 putback_movable_pages(&source);
1429 * remove from free_area[] and mark all as Reserved.
1432 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1435 unsigned long *offlined_pages = (unsigned long *)data;
1437 *offlined_pages += __offline_isolated_pages(start, start + nr_pages);
1442 * Check all pages in range, recoreded as memory resource, are isolated.
1445 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1448 return test_pages_isolated(start_pfn, start_pfn + nr_pages, true);
1451 static int __init cmdline_parse_movable_node(char *p)
1453 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1454 movable_node_enabled = true;
1456 pr_warn("movable_node parameter depends on CONFIG_HAVE_MEMBLOCK_NODE_MAP to work properly\n");
1460 early_param("movable_node", cmdline_parse_movable_node);
1462 /* check which state of node_states will be changed when offline memory */
1463 static void node_states_check_changes_offline(unsigned long nr_pages,
1464 struct zone *zone, struct memory_notify *arg)
1466 struct pglist_data *pgdat = zone->zone_pgdat;
1467 unsigned long present_pages = 0;
1470 arg->status_change_nid = NUMA_NO_NODE;
1471 arg->status_change_nid_normal = NUMA_NO_NODE;
1472 arg->status_change_nid_high = NUMA_NO_NODE;
1475 * Check whether node_states[N_NORMAL_MEMORY] will be changed.
1476 * If the memory to be offline is within the range
1477 * [0..ZONE_NORMAL], and it is the last present memory there,
1478 * the zones in that range will become empty after the offlining,
1479 * thus we can determine that we need to clear the node from
1480 * node_states[N_NORMAL_MEMORY].
1482 for (zt = 0; zt <= ZONE_NORMAL; zt++)
1483 present_pages += pgdat->node_zones[zt].present_pages;
1484 if (zone_idx(zone) <= ZONE_NORMAL && nr_pages >= present_pages)
1485 arg->status_change_nid_normal = zone_to_nid(zone);
1487 #ifdef CONFIG_HIGHMEM
1489 * node_states[N_HIGH_MEMORY] contains nodes which
1490 * have normal memory or high memory.
1491 * Here we add the present_pages belonging to ZONE_HIGHMEM.
1492 * If the zone is within the range of [0..ZONE_HIGHMEM), and
1493 * we determine that the zones in that range become empty,
1494 * we need to clear the node for N_HIGH_MEMORY.
1496 present_pages += pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1497 if (zone_idx(zone) <= ZONE_HIGHMEM && nr_pages >= present_pages)
1498 arg->status_change_nid_high = zone_to_nid(zone);
1502 * We have accounted the pages from [0..ZONE_NORMAL), and
1503 * in case of CONFIG_HIGHMEM the pages from ZONE_HIGHMEM
1505 * Here we count the possible pages from ZONE_MOVABLE.
1506 * If after having accounted all the pages, we see that the nr_pages
1507 * to be offlined is over or equal to the accounted pages,
1508 * we know that the node will become empty, and so, we can clear
1509 * it for N_MEMORY as well.
1511 present_pages += pgdat->node_zones[ZONE_MOVABLE].present_pages;
1513 if (nr_pages >= present_pages)
1514 arg->status_change_nid = zone_to_nid(zone);
1517 static void node_states_clear_node(int node, struct memory_notify *arg)
1519 if (arg->status_change_nid_normal >= 0)
1520 node_clear_state(node, N_NORMAL_MEMORY);
1522 if (arg->status_change_nid_high >= 0)
1523 node_clear_state(node, N_HIGH_MEMORY);
1525 if (arg->status_change_nid >= 0)
1526 node_clear_state(node, N_MEMORY);
1529 static int __ref __offline_pages(unsigned long start_pfn,
1530 unsigned long end_pfn)
1532 unsigned long pfn, nr_pages;
1533 unsigned long offlined_pages = 0;
1534 int ret, node, nr_isolate_pageblock;
1535 unsigned long flags;
1536 unsigned long valid_start, valid_end;
1538 struct memory_notify arg;
1541 mem_hotplug_begin();
1543 /* This makes hotplug much easier...and readable.
1544 we assume this for now. .*/
1545 if (!test_pages_in_a_zone(start_pfn, end_pfn, &valid_start,
1548 reason = "multizone range";
1549 goto failed_removal;
1552 zone = page_zone(pfn_to_page(valid_start));
1553 node = zone_to_nid(zone);
1554 nr_pages = end_pfn - start_pfn;
1556 /* set above range as isolated */
1557 ret = start_isolate_page_range(start_pfn, end_pfn,
1559 SKIP_HWPOISON | REPORT_FAILURE);
1561 reason = "failure to isolate range";
1562 goto failed_removal;
1564 nr_isolate_pageblock = ret;
1566 arg.start_pfn = start_pfn;
1567 arg.nr_pages = nr_pages;
1568 node_states_check_changes_offline(nr_pages, zone, &arg);
1570 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1571 ret = notifier_to_errno(ret);
1573 reason = "notifier failure";
1574 goto failed_removal_isolated;
1578 for (pfn = start_pfn; pfn;) {
1579 if (signal_pending(current)) {
1581 reason = "signal backoff";
1582 goto failed_removal_isolated;
1586 lru_add_drain_all();
1588 pfn = scan_movable_pages(pfn, end_pfn);
1591 * TODO: fatal migration failures should bail
1594 do_migrate_range(pfn, end_pfn);
1599 * Dissolve free hugepages in the memory block before doing
1600 * offlining actually in order to make hugetlbfs's object
1601 * counting consistent.
1603 ret = dissolve_free_huge_pages(start_pfn, end_pfn);
1605 reason = "failure to dissolve huge pages";
1606 goto failed_removal_isolated;
1609 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn,
1610 NULL, check_pages_isolated_cb);
1613 /* Ok, all of our target is isolated.
1614 We cannot do rollback at this point. */
1615 walk_system_ram_range(start_pfn, end_pfn - start_pfn,
1616 &offlined_pages, offline_isolated_pages_cb);
1617 pr_info("Offlined Pages %ld\n", offlined_pages);
1619 * Onlining will reset pagetype flags and makes migrate type
1620 * MOVABLE, so just need to decrease the number of isolated
1621 * pageblocks zone counter here.
1623 spin_lock_irqsave(&zone->lock, flags);
1624 zone->nr_isolate_pageblock -= nr_isolate_pageblock;
1625 spin_unlock_irqrestore(&zone->lock, flags);
1627 /* removal success */
1628 adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages);
1629 zone->present_pages -= offlined_pages;
1631 pgdat_resize_lock(zone->zone_pgdat, &flags);
1632 zone->zone_pgdat->node_present_pages -= offlined_pages;
1633 pgdat_resize_unlock(zone->zone_pgdat, &flags);
1635 init_per_zone_wmark_min();
1637 if (!populated_zone(zone)) {
1638 zone_pcp_reset(zone);
1639 build_all_zonelists(NULL);
1641 zone_pcp_update(zone);
1643 node_states_clear_node(node, &arg);
1644 if (arg.status_change_nid >= 0) {
1646 kcompactd_stop(node);
1649 vm_total_pages = nr_free_pagecache_pages();
1650 writeback_set_ratelimit();
1652 memory_notify(MEM_OFFLINE, &arg);
1656 failed_removal_isolated:
1657 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1658 memory_notify(MEM_CANCEL_OFFLINE, &arg);
1660 pr_debug("memory offlining [mem %#010llx-%#010llx] failed due to %s\n",
1661 (unsigned long long) start_pfn << PAGE_SHIFT,
1662 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1,
1664 /* pushback to free area */
1669 int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1671 return __offline_pages(start_pfn, start_pfn + nr_pages);
1673 #endif /* CONFIG_MEMORY_HOTREMOVE */
1676 * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
1677 * @start_pfn: start pfn of the memory range
1678 * @end_pfn: end pfn of the memory range
1679 * @arg: argument passed to func
1680 * @func: callback for each memory section walked
1682 * This function walks through all present mem sections in range
1683 * [start_pfn, end_pfn) and call func on each mem section.
1685 * Returns the return value of func.
1687 int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn,
1688 void *arg, int (*func)(struct memory_block *, void *))
1690 struct memory_block *mem = NULL;
1691 struct mem_section *section;
1692 unsigned long pfn, section_nr;
1695 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1696 section_nr = pfn_to_section_nr(pfn);
1697 if (!present_section_nr(section_nr))
1700 section = __nr_to_section(section_nr);
1701 /* same memblock? */
1703 if ((section_nr >= mem->start_section_nr) &&
1704 (section_nr <= mem->end_section_nr))
1707 mem = find_memory_block_hinted(section, mem);
1711 ret = func(mem, arg);
1713 kobject_put(&mem->dev.kobj);
1719 kobject_put(&mem->dev.kobj);
1724 #ifdef CONFIG_MEMORY_HOTREMOVE
1725 static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
1727 int ret = !is_memblock_offlined(mem);
1729 if (unlikely(ret)) {
1730 phys_addr_t beginpa, endpa;
1732 beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
1733 endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1;
1734 pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n",
1741 static int check_cpu_on_node(pg_data_t *pgdat)
1745 for_each_present_cpu(cpu) {
1746 if (cpu_to_node(cpu) == pgdat->node_id)
1748 * the cpu on this node isn't removed, and we can't
1749 * offline this node.
1761 * Offline a node if all memory sections and cpus of the node are removed.
1763 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1764 * and online/offline operations before this call.
1766 void try_offline_node(int nid)
1768 pg_data_t *pgdat = NODE_DATA(nid);
1769 unsigned long start_pfn = pgdat->node_start_pfn;
1770 unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
1773 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1774 unsigned long section_nr = pfn_to_section_nr(pfn);
1776 if (!present_section_nr(section_nr))
1779 if (pfn_to_nid(pfn) != nid)
1783 * some memory sections of this node are not removed, and we
1784 * can't offline node now.
1789 if (check_cpu_on_node(pgdat))
1793 * all memory/cpu of this node are removed, we can offline this
1796 node_set_offline(nid);
1797 unregister_one_node(nid);
1799 EXPORT_SYMBOL(try_offline_node);
1801 static void __release_memory_resource(resource_size_t start,
1802 resource_size_t size)
1807 * When removing memory in the same granularity as it was added,
1808 * this function never fails. It might only fail if resources
1809 * have to be adjusted or split. We'll ignore the error, as
1810 * removing of memory cannot fail.
1812 ret = release_mem_region_adjustable(&iomem_resource, start, size);
1814 resource_size_t endres = start + size - 1;
1816 pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
1817 &start, &endres, ret);
1824 * @start: physical address of the region to remove
1825 * @size: size of the region to remove
1827 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1828 * and online/offline operations before this call, as required by
1829 * try_offline_node().
1831 void __ref __remove_memory(int nid, u64 start, u64 size)
1835 BUG_ON(check_hotplug_memory_range(start, size));
1837 mem_hotplug_begin();
1840 * All memory blocks must be offlined before removing memory. Check
1841 * whether all memory blocks in question are offline and trigger a BUG()
1842 * if this is not the case.
1844 ret = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL,
1845 check_memblock_offlined_cb);
1849 /* remove memmap entry */
1850 firmware_map_remove(start, start + size, "System RAM");
1851 memblock_free(start, size);
1852 memblock_remove(start, size);
1854 arch_remove_memory(nid, start, size, NULL);
1855 __release_memory_resource(start, size);
1857 try_offline_node(nid);
1862 void remove_memory(int nid, u64 start, u64 size)
1864 lock_device_hotplug();
1865 __remove_memory(nid, start, size);
1866 unlock_device_hotplug();
1868 EXPORT_SYMBOL_GPL(remove_memory);
1869 #endif /* CONFIG_MEMORY_HOTREMOVE */