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>
45 * online_page_callback contains pointer to current page onlining function.
46 * Initially it is generic_online_page(). If it is required it could be
47 * changed by calling set_online_page_callback() for callback registration
48 * and restore_online_page_callback() for generic callback restore.
51 static void generic_online_page(struct page *page, unsigned int order);
53 static online_page_callback_t online_page_callback = generic_online_page;
54 static DEFINE_MUTEX(online_page_callback_lock);
56 DEFINE_STATIC_PERCPU_RWSEM(mem_hotplug_lock);
58 void get_online_mems(void)
60 percpu_down_read(&mem_hotplug_lock);
63 void put_online_mems(void)
65 percpu_up_read(&mem_hotplug_lock);
68 bool movable_node_enabled = false;
70 #ifndef CONFIG_MEMORY_HOTPLUG_DEFAULT_ONLINE
71 bool memhp_auto_online;
73 bool memhp_auto_online = true;
75 EXPORT_SYMBOL_GPL(memhp_auto_online);
77 static int __init setup_memhp_default_state(char *str)
79 if (!strcmp(str, "online"))
80 memhp_auto_online = true;
81 else if (!strcmp(str, "offline"))
82 memhp_auto_online = false;
86 __setup("memhp_default_state=", setup_memhp_default_state);
88 void mem_hotplug_begin(void)
91 percpu_down_write(&mem_hotplug_lock);
94 void mem_hotplug_done(void)
96 percpu_up_write(&mem_hotplug_lock);
100 /* add this memory to iomem resource */
101 static struct resource *register_memory_resource(u64 start, u64 size)
103 struct resource *res, *conflict;
104 res = kzalloc(sizeof(struct resource), GFP_KERNEL);
106 return ERR_PTR(-ENOMEM);
108 res->name = "System RAM";
110 res->end = start + size - 1;
111 res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
112 conflict = request_resource_conflict(&iomem_resource, res);
114 if (conflict->desc == IORES_DESC_DEVICE_PRIVATE_MEMORY) {
115 pr_debug("Device unaddressable memory block "
116 "memory hotplug at %#010llx !\n",
117 (unsigned long long)start);
119 pr_debug("System RAM resource %pR cannot be added\n", res);
121 return ERR_PTR(-EEXIST);
126 static void release_memory_resource(struct resource *res)
130 release_resource(res);
135 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
136 void get_page_bootmem(unsigned long info, struct page *page,
139 page->freelist = (void *)type;
140 SetPagePrivate(page);
141 set_page_private(page, info);
145 void put_page_bootmem(struct page *page)
149 type = (unsigned long) page->freelist;
150 BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE ||
151 type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE);
153 if (page_ref_dec_return(page) == 1) {
154 page->freelist = NULL;
155 ClearPagePrivate(page);
156 set_page_private(page, 0);
157 INIT_LIST_HEAD(&page->lru);
158 free_reserved_page(page);
162 #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
163 #ifndef CONFIG_SPARSEMEM_VMEMMAP
164 static void register_page_bootmem_info_section(unsigned long start_pfn)
166 unsigned long *usemap, mapsize, section_nr, i;
167 struct mem_section *ms;
168 struct page *page, *memmap;
170 section_nr = pfn_to_section_nr(start_pfn);
171 ms = __nr_to_section(section_nr);
173 /* Get section's memmap address */
174 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
177 * Get page for the memmap's phys address
178 * XXX: need more consideration for sparse_vmemmap...
180 page = virt_to_page(memmap);
181 mapsize = sizeof(struct page) * PAGES_PER_SECTION;
182 mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
184 /* remember memmap's page */
185 for (i = 0; i < mapsize; i++, page++)
186 get_page_bootmem(section_nr, page, SECTION_INFO);
188 usemap = ms->pageblock_flags;
189 page = virt_to_page(usemap);
191 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
193 for (i = 0; i < mapsize; i++, page++)
194 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
197 #else /* CONFIG_SPARSEMEM_VMEMMAP */
198 static void register_page_bootmem_info_section(unsigned long start_pfn)
200 unsigned long *usemap, mapsize, section_nr, i;
201 struct mem_section *ms;
202 struct page *page, *memmap;
204 section_nr = pfn_to_section_nr(start_pfn);
205 ms = __nr_to_section(section_nr);
207 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
209 register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION);
211 usemap = ms->pageblock_flags;
212 page = virt_to_page(usemap);
214 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
216 for (i = 0; i < mapsize; i++, page++)
217 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
219 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
221 void __init register_page_bootmem_info_node(struct pglist_data *pgdat)
223 unsigned long i, pfn, end_pfn, nr_pages;
224 int node = pgdat->node_id;
227 nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
228 page = virt_to_page(pgdat);
230 for (i = 0; i < nr_pages; i++, page++)
231 get_page_bootmem(node, page, NODE_INFO);
233 pfn = pgdat->node_start_pfn;
234 end_pfn = pgdat_end_pfn(pgdat);
236 /* register section info */
237 for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
239 * Some platforms can assign the same pfn to multiple nodes - on
240 * node0 as well as nodeN. To avoid registering a pfn against
241 * multiple nodes we check that this pfn does not already
242 * reside in some other nodes.
244 if (pfn_valid(pfn) && (early_pfn_to_nid(pfn) == node))
245 register_page_bootmem_info_section(pfn);
248 #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
250 static int __meminit __add_section(int nid, unsigned long phys_start_pfn,
251 struct vmem_altmap *altmap, bool want_memblock)
255 if (pfn_valid(phys_start_pfn))
258 ret = sparse_add_one_section(nid, phys_start_pfn, altmap);
265 return hotplug_memory_register(nid, __pfn_to_section(phys_start_pfn));
269 * Reasonably generic function for adding memory. It is
270 * expected that archs that support memory hotplug will
271 * call this function after deciding the zone to which to
274 int __ref __add_pages(int nid, unsigned long phys_start_pfn,
275 unsigned long nr_pages, struct vmem_altmap *altmap,
280 int start_sec, end_sec;
282 /* during initialize mem_map, align hot-added range to section */
283 start_sec = pfn_to_section_nr(phys_start_pfn);
284 end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
288 * Validate altmap is within bounds of the total request
290 if (altmap->base_pfn != phys_start_pfn
291 || vmem_altmap_offset(altmap) > nr_pages) {
292 pr_warn_once("memory add fail, invalid altmap\n");
299 for (i = start_sec; i <= end_sec; i++) {
300 err = __add_section(nid, section_nr_to_pfn(i), altmap,
304 * EEXIST is finally dealt with by ioresource collision
305 * check. see add_memory() => register_memory_resource()
306 * Warning will be printed if there is collision.
308 if (err && (err != -EEXIST))
313 vmemmap_populate_print_last();
318 #ifdef CONFIG_MEMORY_HOTREMOVE
319 /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
320 static unsigned long find_smallest_section_pfn(int nid, struct zone *zone,
321 unsigned long start_pfn,
322 unsigned long end_pfn)
324 struct mem_section *ms;
326 for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) {
327 ms = __pfn_to_section(start_pfn);
329 if (unlikely(!valid_section(ms)))
332 if (unlikely(pfn_to_nid(start_pfn) != nid))
335 if (zone && zone != page_zone(pfn_to_page(start_pfn)))
344 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
345 static unsigned long find_biggest_section_pfn(int nid, struct zone *zone,
346 unsigned long start_pfn,
347 unsigned long end_pfn)
349 struct mem_section *ms;
352 /* pfn is the end pfn of a memory section. */
354 for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) {
355 ms = __pfn_to_section(pfn);
357 if (unlikely(!valid_section(ms)))
360 if (unlikely(pfn_to_nid(pfn) != nid))
363 if (zone && zone != page_zone(pfn_to_page(pfn)))
372 static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
373 unsigned long end_pfn)
375 unsigned long zone_start_pfn = zone->zone_start_pfn;
376 unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */
377 unsigned long zone_end_pfn = z;
379 struct mem_section *ms;
380 int nid = zone_to_nid(zone);
382 zone_span_writelock(zone);
383 if (zone_start_pfn == start_pfn) {
385 * If the section is smallest section in the zone, it need
386 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
387 * In this case, we find second smallest valid mem_section
388 * for shrinking zone.
390 pfn = find_smallest_section_pfn(nid, zone, end_pfn,
393 zone->zone_start_pfn = pfn;
394 zone->spanned_pages = zone_end_pfn - pfn;
396 } else if (zone_end_pfn == end_pfn) {
398 * If the section is biggest section in the zone, it need
399 * shrink zone->spanned_pages.
400 * In this case, we find second biggest valid mem_section for
403 pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn,
406 zone->spanned_pages = pfn - zone_start_pfn + 1;
410 * The section is not biggest or smallest mem_section in the zone, it
411 * only creates a hole in the zone. So in this case, we need not
412 * change the zone. But perhaps, the zone has only hole data. Thus
413 * it check the zone has only hole or not.
415 pfn = zone_start_pfn;
416 for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) {
417 ms = __pfn_to_section(pfn);
419 if (unlikely(!valid_section(ms)))
422 if (page_zone(pfn_to_page(pfn)) != zone)
425 /* If the section is current section, it continues the loop */
426 if (start_pfn == pfn)
429 /* If we find valid section, we have nothing to do */
430 zone_span_writeunlock(zone);
434 /* The zone has no valid section */
435 zone->zone_start_pfn = 0;
436 zone->spanned_pages = 0;
437 zone_span_writeunlock(zone);
440 static void shrink_pgdat_span(struct pglist_data *pgdat,
441 unsigned long start_pfn, unsigned long end_pfn)
443 unsigned long pgdat_start_pfn = pgdat->node_start_pfn;
444 unsigned long p = pgdat_end_pfn(pgdat); /* pgdat_end_pfn namespace clash */
445 unsigned long pgdat_end_pfn = p;
447 struct mem_section *ms;
448 int nid = pgdat->node_id;
450 if (pgdat_start_pfn == start_pfn) {
452 * If the section is smallest section in the pgdat, it need
453 * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages.
454 * In this case, we find second smallest valid mem_section
455 * for shrinking zone.
457 pfn = find_smallest_section_pfn(nid, NULL, end_pfn,
460 pgdat->node_start_pfn = pfn;
461 pgdat->node_spanned_pages = pgdat_end_pfn - pfn;
463 } else if (pgdat_end_pfn == end_pfn) {
465 * If the section is biggest section in the pgdat, it need
466 * shrink pgdat->node_spanned_pages.
467 * In this case, we find second biggest valid mem_section for
470 pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn,
473 pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1;
477 * If the section is not biggest or smallest mem_section in the pgdat,
478 * it only creates a hole in the pgdat. So in this case, we need not
480 * But perhaps, the pgdat has only hole data. Thus it check the pgdat
481 * has only hole or not.
483 pfn = pgdat_start_pfn;
484 for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) {
485 ms = __pfn_to_section(pfn);
487 if (unlikely(!valid_section(ms)))
490 if (pfn_to_nid(pfn) != nid)
493 /* If the section is current section, it continues the loop */
494 if (start_pfn == pfn)
497 /* If we find valid section, we have nothing to do */
501 /* The pgdat has no valid section */
502 pgdat->node_start_pfn = 0;
503 pgdat->node_spanned_pages = 0;
506 static void __remove_zone(struct zone *zone, unsigned long start_pfn)
508 struct pglist_data *pgdat = zone->zone_pgdat;
509 int nr_pages = PAGES_PER_SECTION;
512 pgdat_resize_lock(zone->zone_pgdat, &flags);
513 shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
514 shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages);
515 pgdat_resize_unlock(zone->zone_pgdat, &flags);
518 static int __remove_section(struct zone *zone, struct mem_section *ms,
519 unsigned long map_offset, struct vmem_altmap *altmap)
521 unsigned long start_pfn;
525 if (!valid_section(ms))
528 ret = unregister_memory_section(ms);
532 scn_nr = __section_nr(ms);
533 start_pfn = section_nr_to_pfn((unsigned long)scn_nr);
534 __remove_zone(zone, start_pfn);
536 sparse_remove_one_section(zone, ms, map_offset, altmap);
541 * __remove_pages() - remove sections of pages from a zone
542 * @zone: zone from which pages need to be removed
543 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
544 * @nr_pages: number of pages to remove (must be multiple of section size)
545 * @altmap: alternative device page map or %NULL if default memmap is used
547 * Generic helper function to remove section mappings and sysfs entries
548 * for the section of the memory we are removing. Caller needs to make
549 * sure that pages are marked reserved and zones are adjust properly by
550 * calling offline_pages().
552 int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
553 unsigned long nr_pages, struct vmem_altmap *altmap)
556 unsigned long map_offset = 0;
557 int sections_to_remove, ret = 0;
559 /* In the ZONE_DEVICE case device driver owns the memory region */
560 if (is_dev_zone(zone)) {
562 map_offset = vmem_altmap_offset(altmap);
564 resource_size_t start, size;
566 start = phys_start_pfn << PAGE_SHIFT;
567 size = nr_pages * PAGE_SIZE;
569 ret = release_mem_region_adjustable(&iomem_resource, start,
572 resource_size_t endres = start + size - 1;
574 pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
575 &start, &endres, ret);
579 clear_zone_contiguous(zone);
582 * We can only remove entire sections
584 BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
585 BUG_ON(nr_pages % PAGES_PER_SECTION);
587 sections_to_remove = nr_pages / PAGES_PER_SECTION;
588 for (i = 0; i < sections_to_remove; i++) {
589 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
592 ret = __remove_section(zone, __pfn_to_section(pfn), map_offset,
599 set_zone_contiguous(zone);
603 #endif /* CONFIG_MEMORY_HOTREMOVE */
605 int set_online_page_callback(online_page_callback_t callback)
610 mutex_lock(&online_page_callback_lock);
612 if (online_page_callback == generic_online_page) {
613 online_page_callback = callback;
617 mutex_unlock(&online_page_callback_lock);
622 EXPORT_SYMBOL_GPL(set_online_page_callback);
624 int restore_online_page_callback(online_page_callback_t callback)
629 mutex_lock(&online_page_callback_lock);
631 if (online_page_callback == callback) {
632 online_page_callback = generic_online_page;
636 mutex_unlock(&online_page_callback_lock);
641 EXPORT_SYMBOL_GPL(restore_online_page_callback);
643 void __online_page_set_limits(struct page *page)
646 EXPORT_SYMBOL_GPL(__online_page_set_limits);
648 void __online_page_increment_counters(struct page *page)
650 adjust_managed_page_count(page, 1);
652 EXPORT_SYMBOL_GPL(__online_page_increment_counters);
654 void __online_page_free(struct page *page)
656 __free_reserved_page(page);
658 EXPORT_SYMBOL_GPL(__online_page_free);
660 static void generic_online_page(struct page *page, unsigned int order)
662 __free_pages_core(page, order);
663 totalram_pages_add(1UL << order);
664 #ifdef CONFIG_HIGHMEM
665 if (PageHighMem(page))
666 totalhigh_pages_add(1UL << order);
670 static int online_pages_blocks(unsigned long start, unsigned long nr_pages)
672 unsigned long end = start + nr_pages;
673 int order, onlined_pages = 0;
675 while (start < end) {
676 order = min(MAX_ORDER - 1,
677 get_order(PFN_PHYS(end) - PFN_PHYS(start)));
678 (*online_page_callback)(pfn_to_page(start), order);
680 onlined_pages += (1UL << order);
681 start += (1UL << order);
683 return onlined_pages;
686 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
689 unsigned long onlined_pages = *(unsigned long *)arg;
691 if (PageReserved(pfn_to_page(start_pfn)))
692 onlined_pages += online_pages_blocks(start_pfn, nr_pages);
694 online_mem_sections(start_pfn, start_pfn + nr_pages);
696 *(unsigned long *)arg = onlined_pages;
700 /* check which state of node_states will be changed when online memory */
701 static void node_states_check_changes_online(unsigned long nr_pages,
702 struct zone *zone, struct memory_notify *arg)
704 int nid = zone_to_nid(zone);
706 arg->status_change_nid = NUMA_NO_NODE;
707 arg->status_change_nid_normal = NUMA_NO_NODE;
708 arg->status_change_nid_high = NUMA_NO_NODE;
710 if (!node_state(nid, N_MEMORY))
711 arg->status_change_nid = nid;
712 if (zone_idx(zone) <= ZONE_NORMAL && !node_state(nid, N_NORMAL_MEMORY))
713 arg->status_change_nid_normal = nid;
714 #ifdef CONFIG_HIGHMEM
715 if (zone_idx(zone) <= N_HIGH_MEMORY && !node_state(nid, N_HIGH_MEMORY))
716 arg->status_change_nid_high = nid;
720 static void node_states_set_node(int node, struct memory_notify *arg)
722 if (arg->status_change_nid_normal >= 0)
723 node_set_state(node, N_NORMAL_MEMORY);
725 if (arg->status_change_nid_high >= 0)
726 node_set_state(node, N_HIGH_MEMORY);
728 if (arg->status_change_nid >= 0)
729 node_set_state(node, N_MEMORY);
732 static void __meminit resize_zone_range(struct zone *zone, unsigned long start_pfn,
733 unsigned long nr_pages)
735 unsigned long old_end_pfn = zone_end_pfn(zone);
737 if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
738 zone->zone_start_pfn = start_pfn;
740 zone->spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - zone->zone_start_pfn;
743 static void __meminit resize_pgdat_range(struct pglist_data *pgdat, unsigned long start_pfn,
744 unsigned long nr_pages)
746 unsigned long old_end_pfn = pgdat_end_pfn(pgdat);
748 if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
749 pgdat->node_start_pfn = start_pfn;
751 pgdat->node_spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - pgdat->node_start_pfn;
754 void __ref move_pfn_range_to_zone(struct zone *zone, unsigned long start_pfn,
755 unsigned long nr_pages, struct vmem_altmap *altmap)
757 struct pglist_data *pgdat = zone->zone_pgdat;
758 int nid = pgdat->node_id;
761 clear_zone_contiguous(zone);
763 /* TODO Huh pgdat is irqsave while zone is not. It used to be like that before */
764 pgdat_resize_lock(pgdat, &flags);
765 zone_span_writelock(zone);
766 if (zone_is_empty(zone))
767 init_currently_empty_zone(zone, start_pfn, nr_pages);
768 resize_zone_range(zone, start_pfn, nr_pages);
769 zone_span_writeunlock(zone);
770 resize_pgdat_range(pgdat, start_pfn, nr_pages);
771 pgdat_resize_unlock(pgdat, &flags);
774 * TODO now we have a visible range of pages which are not associated
775 * with their zone properly. Not nice but set_pfnblock_flags_mask
776 * expects the zone spans the pfn range. All the pages in the range
777 * are reserved so nobody should be touching them so we should be safe
779 memmap_init_zone(nr_pages, nid, zone_idx(zone), start_pfn,
780 MEMMAP_HOTPLUG, altmap);
782 set_zone_contiguous(zone);
786 * Returns a default kernel memory zone for the given pfn range.
787 * If no kernel zone covers this pfn range it will automatically go
788 * to the ZONE_NORMAL.
790 static struct zone *default_kernel_zone_for_pfn(int nid, unsigned long start_pfn,
791 unsigned long nr_pages)
793 struct pglist_data *pgdat = NODE_DATA(nid);
796 for (zid = 0; zid <= ZONE_NORMAL; zid++) {
797 struct zone *zone = &pgdat->node_zones[zid];
799 if (zone_intersects(zone, start_pfn, nr_pages))
803 return &pgdat->node_zones[ZONE_NORMAL];
806 static inline struct zone *default_zone_for_pfn(int nid, unsigned long start_pfn,
807 unsigned long nr_pages)
809 struct zone *kernel_zone = default_kernel_zone_for_pfn(nid, start_pfn,
811 struct zone *movable_zone = &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
812 bool in_kernel = zone_intersects(kernel_zone, start_pfn, nr_pages);
813 bool in_movable = zone_intersects(movable_zone, start_pfn, nr_pages);
816 * We inherit the existing zone in a simple case where zones do not
817 * overlap in the given range
819 if (in_kernel ^ in_movable)
820 return (in_kernel) ? kernel_zone : movable_zone;
823 * If the range doesn't belong to any zone or two zones overlap in the
824 * given range then we use movable zone only if movable_node is
825 * enabled because we always online to a kernel zone by default.
827 return movable_node_enabled ? movable_zone : kernel_zone;
830 struct zone * zone_for_pfn_range(int online_type, int nid, unsigned start_pfn,
831 unsigned long nr_pages)
833 if (online_type == MMOP_ONLINE_KERNEL)
834 return default_kernel_zone_for_pfn(nid, start_pfn, nr_pages);
836 if (online_type == MMOP_ONLINE_MOVABLE)
837 return &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
839 return default_zone_for_pfn(nid, start_pfn, nr_pages);
843 * Associates the given pfn range with the given node and the zone appropriate
844 * for the given online type.
846 static struct zone * __meminit move_pfn_range(int online_type, int nid,
847 unsigned long start_pfn, unsigned long nr_pages)
851 zone = zone_for_pfn_range(online_type, nid, start_pfn, nr_pages);
852 move_pfn_range_to_zone(zone, start_pfn, nr_pages, NULL);
856 int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type)
859 unsigned long onlined_pages = 0;
861 int need_zonelists_rebuild = 0;
864 struct memory_notify arg;
865 struct memory_block *mem;
870 * We can't use pfn_to_nid() because nid might be stored in struct page
871 * which is not yet initialized. Instead, we find nid from memory block.
873 mem = find_memory_block(__pfn_to_section(pfn));
876 /* associate pfn range with the zone */
877 zone = move_pfn_range(online_type, nid, pfn, nr_pages);
880 arg.nr_pages = nr_pages;
881 node_states_check_changes_online(nr_pages, zone, &arg);
883 ret = memory_notify(MEM_GOING_ONLINE, &arg);
884 ret = notifier_to_errno(ret);
886 goto failed_addition;
889 * If this zone is not populated, then it is not in zonelist.
890 * This means the page allocator ignores this zone.
891 * So, zonelist must be updated after online.
893 if (!populated_zone(zone)) {
894 need_zonelists_rebuild = 1;
895 setup_zone_pageset(zone);
898 ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
901 if (need_zonelists_rebuild)
902 zone_pcp_reset(zone);
903 goto failed_addition;
906 zone->present_pages += onlined_pages;
908 pgdat_resize_lock(zone->zone_pgdat, &flags);
909 zone->zone_pgdat->node_present_pages += onlined_pages;
910 pgdat_resize_unlock(zone->zone_pgdat, &flags);
915 node_states_set_node(nid, &arg);
916 if (need_zonelists_rebuild)
917 build_all_zonelists(NULL);
919 zone_pcp_update(zone);
922 init_per_zone_wmark_min();
929 vm_total_pages = nr_free_pagecache_pages();
931 writeback_set_ratelimit();
934 memory_notify(MEM_ONLINE, &arg);
939 pr_debug("online_pages [mem %#010llx-%#010llx] failed\n",
940 (unsigned long long) pfn << PAGE_SHIFT,
941 (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1);
942 memory_notify(MEM_CANCEL_ONLINE, &arg);
946 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
948 static void reset_node_present_pages(pg_data_t *pgdat)
952 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
953 z->present_pages = 0;
955 pgdat->node_present_pages = 0;
958 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
959 static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
961 struct pglist_data *pgdat;
962 unsigned long start_pfn = PFN_DOWN(start);
964 pgdat = NODE_DATA(nid);
966 pgdat = arch_alloc_nodedata(nid);
970 arch_refresh_nodedata(nid, pgdat);
973 * Reset the nr_zones, order and classzone_idx before reuse.
974 * Note that kswapd will init kswapd_classzone_idx properly
975 * when it starts in the near future.
978 pgdat->kswapd_order = 0;
979 pgdat->kswapd_classzone_idx = 0;
982 /* we can use NODE_DATA(nid) from here */
984 pgdat->node_id = nid;
985 pgdat->node_start_pfn = start_pfn;
987 /* init node's zones as empty zones, we don't have any present pages.*/
988 free_area_init_core_hotplug(nid);
989 pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat);
992 * The node we allocated has no zone fallback lists. For avoiding
993 * to access not-initialized zonelist, build here.
995 build_all_zonelists(pgdat);
998 * When memory is hot-added, all the memory is in offline state. So
999 * clear all zones' present_pages because they will be updated in
1000 * online_pages() and offline_pages().
1002 reset_node_managed_pages(pgdat);
1003 reset_node_present_pages(pgdat);
1008 static void rollback_node_hotadd(int nid)
1010 pg_data_t *pgdat = NODE_DATA(nid);
1012 arch_refresh_nodedata(nid, NULL);
1013 free_percpu(pgdat->per_cpu_nodestats);
1014 arch_free_nodedata(pgdat);
1020 * try_online_node - online a node if offlined
1022 * @start: start addr of the node
1023 * @set_node_online: Whether we want to online the node
1024 * called by cpu_up() to online a node without onlined memory.
1027 * 1 -> a new node has been allocated
1028 * 0 -> the node is already online
1029 * -ENOMEM -> the node could not be allocated
1031 static int __try_online_node(int nid, u64 start, bool set_node_online)
1036 if (node_online(nid))
1039 pgdat = hotadd_new_pgdat(nid, start);
1041 pr_err("Cannot online node %d due to NULL pgdat\n", nid);
1046 if (set_node_online) {
1047 node_set_online(nid);
1048 ret = register_one_node(nid);
1056 * Users of this function always want to online/register the node
1058 int try_online_node(int nid)
1062 mem_hotplug_begin();
1063 ret = __try_online_node(nid, 0, true);
1068 static int check_hotplug_memory_range(u64 start, u64 size)
1070 unsigned long block_sz = memory_block_size_bytes();
1071 u64 block_nr_pages = block_sz >> PAGE_SHIFT;
1072 u64 nr_pages = size >> PAGE_SHIFT;
1073 u64 start_pfn = PFN_DOWN(start);
1075 /* memory range must be block size aligned */
1076 if (!nr_pages || !IS_ALIGNED(start_pfn, block_nr_pages) ||
1077 !IS_ALIGNED(nr_pages, block_nr_pages)) {
1078 pr_err("Block size [%#lx] unaligned hotplug range: start %#llx, size %#llx",
1079 block_sz, start, size);
1086 static int online_memory_block(struct memory_block *mem, void *arg)
1088 return device_online(&mem->dev);
1092 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1093 * and online/offline operations (triggered e.g. by sysfs).
1095 * we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG
1097 int __ref add_memory_resource(int nid, struct resource *res)
1100 bool new_node = false;
1104 size = resource_size(res);
1106 ret = check_hotplug_memory_range(start, size);
1110 mem_hotplug_begin();
1113 * Add new range to memblock so that when hotadd_new_pgdat() is called
1114 * to allocate new pgdat, get_pfn_range_for_nid() will be able to find
1115 * this new range and calculate total pages correctly. The range will
1116 * be removed at hot-remove time.
1118 memblock_add_node(start, size, nid);
1120 ret = __try_online_node(nid, start, false);
1125 /* call arch's memory hotadd */
1126 ret = arch_add_memory(nid, start, size, NULL, true);
1131 /* If sysfs file of new node can't be created, cpu on the node
1132 * can't be hot-added. There is no rollback way now.
1133 * So, check by BUG_ON() to catch it reluctantly..
1134 * We online node here. We can't roll back from here.
1136 node_set_online(nid);
1137 ret = __register_one_node(nid);
1141 /* link memory sections under this node.*/
1142 ret = link_mem_sections(nid, PFN_DOWN(start), PFN_UP(start + size - 1));
1145 /* create new memmap entry */
1146 firmware_map_add_hotplug(start, start + size, "System RAM");
1148 /* device_online() will take the lock when calling online_pages() */
1151 /* online pages if requested */
1152 if (memhp_auto_online)
1153 walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1),
1154 NULL, online_memory_block);
1158 /* rollback pgdat allocation and others */
1160 rollback_node_hotadd(nid);
1161 memblock_remove(start, size);
1166 /* requires device_hotplug_lock, see add_memory_resource() */
1167 int __ref __add_memory(int nid, u64 start, u64 size)
1169 struct resource *res;
1172 res = register_memory_resource(start, size);
1174 return PTR_ERR(res);
1176 ret = add_memory_resource(nid, res);
1178 release_memory_resource(res);
1182 int add_memory(int nid, u64 start, u64 size)
1186 lock_device_hotplug();
1187 rc = __add_memory(nid, start, size);
1188 unlock_device_hotplug();
1192 EXPORT_SYMBOL_GPL(add_memory);
1194 #ifdef CONFIG_MEMORY_HOTREMOVE
1196 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1197 * set and the size of the free page is given by page_order(). Using this,
1198 * the function determines if the pageblock contains only free pages.
1199 * Due to buddy contraints, a free page at least the size of a pageblock will
1200 * be located at the start of the pageblock
1202 static inline int pageblock_free(struct page *page)
1204 return PageBuddy(page) && page_order(page) >= pageblock_order;
1207 /* Return the start of the next active pageblock after a given page */
1208 static struct page *next_active_pageblock(struct page *page)
1210 /* Ensure the starting page is pageblock-aligned */
1211 BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
1213 /* If the entire pageblock is free, move to the end of free page */
1214 if (pageblock_free(page)) {
1216 /* be careful. we don't have locks, page_order can be changed.*/
1217 order = page_order(page);
1218 if ((order < MAX_ORDER) && (order >= pageblock_order))
1219 return page + (1 << order);
1222 return page + pageblock_nr_pages;
1225 static bool is_pageblock_removable_nolock(struct page *page)
1231 * We have to be careful here because we are iterating over memory
1232 * sections which are not zone aware so we might end up outside of
1233 * the zone but still within the section.
1234 * We have to take care about the node as well. If the node is offline
1235 * its NODE_DATA will be NULL - see page_zone.
1237 if (!node_online(page_to_nid(page)))
1240 zone = page_zone(page);
1241 pfn = page_to_pfn(page);
1242 if (!zone_spans_pfn(zone, pfn))
1245 return !has_unmovable_pages(zone, page, 0, MIGRATE_MOVABLE, SKIP_HWPOISON);
1248 /* Checks if this range of memory is likely to be hot-removable. */
1249 bool is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
1251 struct page *page = pfn_to_page(start_pfn);
1252 unsigned long end_pfn = min(start_pfn + nr_pages, zone_end_pfn(page_zone(page)));
1253 struct page *end_page = pfn_to_page(end_pfn);
1255 /* Check the starting page of each pageblock within the range */
1256 for (; page < end_page; page = next_active_pageblock(page)) {
1257 if (!is_pageblock_removable_nolock(page))
1262 /* All pageblocks in the memory block are likely to be hot-removable */
1267 * Confirm all pages in a range [start, end) belong to the same zone.
1268 * When true, return its valid [start, end).
1270 int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn,
1271 unsigned long *valid_start, unsigned long *valid_end)
1273 unsigned long pfn, sec_end_pfn;
1274 unsigned long start, end;
1275 struct zone *zone = NULL;
1278 for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1);
1280 pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) {
1281 /* Make sure the memory section is present first */
1282 if (!present_section_nr(pfn_to_section_nr(pfn)))
1284 for (; pfn < sec_end_pfn && pfn < end_pfn;
1285 pfn += MAX_ORDER_NR_PAGES) {
1287 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1288 while ((i < MAX_ORDER_NR_PAGES) &&
1289 !pfn_valid_within(pfn + i))
1291 if (i == MAX_ORDER_NR_PAGES || pfn + i >= end_pfn)
1293 /* Check if we got outside of the zone */
1294 if (zone && !zone_spans_pfn(zone, pfn + i))
1296 page = pfn_to_page(pfn + i);
1297 if (zone && page_zone(page) != zone)
1301 zone = page_zone(page);
1302 end = pfn + MAX_ORDER_NR_PAGES;
1307 *valid_start = start;
1308 *valid_end = min(end, end_pfn);
1316 * Scan pfn range [start,end) to find movable/migratable pages (LRU pages,
1317 * non-lru movable pages and hugepages). We scan pfn because it's much
1318 * easier than scanning over linked list. This function returns the pfn
1319 * of the first found movable page if it's found, otherwise 0.
1321 static unsigned long scan_movable_pages(unsigned long start, unsigned long end)
1325 for (pfn = start; pfn < end; pfn++) {
1326 struct page *page, *head;
1329 if (!pfn_valid(pfn))
1331 page = pfn_to_page(pfn);
1334 if (__PageMovable(page))
1337 if (!PageHuge(page))
1339 head = compound_head(page);
1340 if (hugepage_migration_supported(page_hstate(head)) &&
1341 page_huge_active(head))
1343 skip = (1 << compound_order(head)) - (page - head);
1349 static struct page *new_node_page(struct page *page, unsigned long private)
1351 int nid = page_to_nid(page);
1352 nodemask_t nmask = node_states[N_MEMORY];
1355 * try to allocate from a different node but reuse this node if there
1356 * are no other online nodes to be used (e.g. we are offlining a part
1357 * of the only existing node)
1359 node_clear(nid, nmask);
1360 if (nodes_empty(nmask))
1361 node_set(nid, nmask);
1363 return new_page_nodemask(page, nid, &nmask);
1367 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1374 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1375 if (!pfn_valid(pfn))
1377 page = pfn_to_page(pfn);
1379 if (PageHuge(page)) {
1380 struct page *head = compound_head(page);
1381 pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1;
1382 if (compound_order(head) > PFN_SECTION_SHIFT) {
1386 isolate_huge_page(page, &source);
1388 } else if (PageTransHuge(page))
1389 pfn = page_to_pfn(compound_head(page))
1390 + hpage_nr_pages(page) - 1;
1393 * HWPoison pages have elevated reference counts so the migration would
1394 * fail on them. It also doesn't make any sense to migrate them in the
1395 * first place. Still try to unmap such a page in case it is still mapped
1396 * (e.g. current hwpoison implementation doesn't unmap KSM pages but keep
1397 * the unmap as the catch all safety net).
1399 if (PageHWPoison(page)) {
1400 if (WARN_ON(PageLRU(page)))
1401 isolate_lru_page(page);
1402 if (page_mapped(page))
1403 try_to_unmap(page, TTU_IGNORE_MLOCK | TTU_IGNORE_ACCESS);
1407 if (!get_page_unless_zero(page))
1410 * We can skip free pages. And we can deal with pages on
1411 * LRU and non-lru movable pages.
1414 ret = isolate_lru_page(page);
1416 ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE);
1417 if (!ret) { /* Success */
1418 list_add_tail(&page->lru, &source);
1419 if (!__PageMovable(page))
1420 inc_node_page_state(page, NR_ISOLATED_ANON +
1421 page_is_file_cache(page));
1424 pr_warn("failed to isolate pfn %lx\n", pfn);
1425 dump_page(page, "isolation failed");
1429 if (!list_empty(&source)) {
1430 /* Allocate a new page from the nearest neighbor node */
1431 ret = migrate_pages(&source, new_node_page, NULL, 0,
1432 MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
1434 list_for_each_entry(page, &source, lru) {
1435 pr_warn("migrating pfn %lx failed ret:%d ",
1436 page_to_pfn(page), ret);
1437 dump_page(page, "migration failure");
1439 putback_movable_pages(&source);
1447 * remove from free_area[] and mark all as Reserved.
1450 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1453 __offline_isolated_pages(start, start + nr_pages);
1458 offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
1460 walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
1461 offline_isolated_pages_cb);
1465 * Check all pages in range, recoreded as memory resource, are isolated.
1468 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1472 long offlined = *(long *)data;
1473 ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true);
1474 offlined = nr_pages;
1476 *(long *)data += offlined;
1481 check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
1486 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
1487 check_pages_isolated_cb);
1489 offlined = (long)ret;
1493 static int __init cmdline_parse_movable_node(char *p)
1495 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1496 movable_node_enabled = true;
1498 pr_warn("movable_node parameter depends on CONFIG_HAVE_MEMBLOCK_NODE_MAP to work properly\n");
1502 early_param("movable_node", cmdline_parse_movable_node);
1504 /* check which state of node_states will be changed when offline memory */
1505 static void node_states_check_changes_offline(unsigned long nr_pages,
1506 struct zone *zone, struct memory_notify *arg)
1508 struct pglist_data *pgdat = zone->zone_pgdat;
1509 unsigned long present_pages = 0;
1512 arg->status_change_nid = NUMA_NO_NODE;
1513 arg->status_change_nid_normal = NUMA_NO_NODE;
1514 arg->status_change_nid_high = NUMA_NO_NODE;
1517 * Check whether node_states[N_NORMAL_MEMORY] will be changed.
1518 * If the memory to be offline is within the range
1519 * [0..ZONE_NORMAL], and it is the last present memory there,
1520 * the zones in that range will become empty after the offlining,
1521 * thus we can determine that we need to clear the node from
1522 * node_states[N_NORMAL_MEMORY].
1524 for (zt = 0; zt <= ZONE_NORMAL; zt++)
1525 present_pages += pgdat->node_zones[zt].present_pages;
1526 if (zone_idx(zone) <= ZONE_NORMAL && nr_pages >= present_pages)
1527 arg->status_change_nid_normal = zone_to_nid(zone);
1529 #ifdef CONFIG_HIGHMEM
1531 * node_states[N_HIGH_MEMORY] contains nodes which
1532 * have normal memory or high memory.
1533 * Here we add the present_pages belonging to ZONE_HIGHMEM.
1534 * If the zone is within the range of [0..ZONE_HIGHMEM), and
1535 * we determine that the zones in that range become empty,
1536 * we need to clear the node for N_HIGH_MEMORY.
1538 present_pages += pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1539 if (zone_idx(zone) <= ZONE_HIGHMEM && nr_pages >= present_pages)
1540 arg->status_change_nid_high = zone_to_nid(zone);
1544 * We have accounted the pages from [0..ZONE_NORMAL), and
1545 * in case of CONFIG_HIGHMEM the pages from ZONE_HIGHMEM
1547 * Here we count the possible pages from ZONE_MOVABLE.
1548 * If after having accounted all the pages, we see that the nr_pages
1549 * to be offlined is over or equal to the accounted pages,
1550 * we know that the node will become empty, and so, we can clear
1551 * it for N_MEMORY as well.
1553 present_pages += pgdat->node_zones[ZONE_MOVABLE].present_pages;
1555 if (nr_pages >= present_pages)
1556 arg->status_change_nid = zone_to_nid(zone);
1559 static void node_states_clear_node(int node, struct memory_notify *arg)
1561 if (arg->status_change_nid_normal >= 0)
1562 node_clear_state(node, N_NORMAL_MEMORY);
1564 if (arg->status_change_nid_high >= 0)
1565 node_clear_state(node, N_HIGH_MEMORY);
1567 if (arg->status_change_nid >= 0)
1568 node_clear_state(node, N_MEMORY);
1571 static int __ref __offline_pages(unsigned long start_pfn,
1572 unsigned long end_pfn)
1574 unsigned long pfn, nr_pages;
1575 long offlined_pages;
1577 unsigned long flags;
1578 unsigned long valid_start, valid_end;
1580 struct memory_notify arg;
1583 mem_hotplug_begin();
1585 /* This makes hotplug much easier...and readable.
1586 we assume this for now. .*/
1587 if (!test_pages_in_a_zone(start_pfn, end_pfn, &valid_start,
1590 reason = "multizone range";
1591 goto failed_removal;
1594 zone = page_zone(pfn_to_page(valid_start));
1595 node = zone_to_nid(zone);
1596 nr_pages = end_pfn - start_pfn;
1598 /* set above range as isolated */
1599 ret = start_isolate_page_range(start_pfn, end_pfn,
1601 SKIP_HWPOISON | REPORT_FAILURE);
1603 reason = "failure to isolate range";
1604 goto failed_removal;
1607 arg.start_pfn = start_pfn;
1608 arg.nr_pages = nr_pages;
1609 node_states_check_changes_offline(nr_pages, zone, &arg);
1611 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1612 ret = notifier_to_errno(ret);
1614 reason = "notifier failure";
1615 goto failed_removal_isolated;
1619 for (pfn = start_pfn; pfn;) {
1620 if (signal_pending(current)) {
1622 reason = "signal backoff";
1623 goto failed_removal_isolated;
1627 lru_add_drain_all();
1629 pfn = scan_movable_pages(pfn, end_pfn);
1632 * TODO: fatal migration failures should bail
1635 do_migrate_range(pfn, end_pfn);
1640 * Dissolve free hugepages in the memory block before doing
1641 * offlining actually in order to make hugetlbfs's object
1642 * counting consistent.
1644 ret = dissolve_free_huge_pages(start_pfn, end_pfn);
1646 reason = "failure to dissolve huge pages";
1647 goto failed_removal_isolated;
1650 offlined_pages = check_pages_isolated(start_pfn, end_pfn);
1651 } while (offlined_pages < 0);
1653 pr_info("Offlined Pages %ld\n", offlined_pages);
1654 /* Ok, all of our target is isolated.
1655 We cannot do rollback at this point. */
1656 offline_isolated_pages(start_pfn, end_pfn);
1657 /* reset pagetype flags and makes migrate type to be MOVABLE */
1658 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1659 /* removal success */
1660 adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages);
1661 zone->present_pages -= offlined_pages;
1663 pgdat_resize_lock(zone->zone_pgdat, &flags);
1664 zone->zone_pgdat->node_present_pages -= offlined_pages;
1665 pgdat_resize_unlock(zone->zone_pgdat, &flags);
1667 init_per_zone_wmark_min();
1669 if (!populated_zone(zone)) {
1670 zone_pcp_reset(zone);
1671 build_all_zonelists(NULL);
1673 zone_pcp_update(zone);
1675 node_states_clear_node(node, &arg);
1676 if (arg.status_change_nid >= 0) {
1678 kcompactd_stop(node);
1681 vm_total_pages = nr_free_pagecache_pages();
1682 writeback_set_ratelimit();
1684 memory_notify(MEM_OFFLINE, &arg);
1688 failed_removal_isolated:
1689 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1691 pr_debug("memory offlining [mem %#010llx-%#010llx] failed due to %s\n",
1692 (unsigned long long) start_pfn << PAGE_SHIFT,
1693 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1,
1695 memory_notify(MEM_CANCEL_OFFLINE, &arg);
1696 /* pushback to free area */
1701 int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1703 return __offline_pages(start_pfn, start_pfn + nr_pages);
1705 #endif /* CONFIG_MEMORY_HOTREMOVE */
1708 * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
1709 * @start_pfn: start pfn of the memory range
1710 * @end_pfn: end pfn of the memory range
1711 * @arg: argument passed to func
1712 * @func: callback for each memory section walked
1714 * This function walks through all present mem sections in range
1715 * [start_pfn, end_pfn) and call func on each mem section.
1717 * Returns the return value of func.
1719 int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn,
1720 void *arg, int (*func)(struct memory_block *, void *))
1722 struct memory_block *mem = NULL;
1723 struct mem_section *section;
1724 unsigned long pfn, section_nr;
1727 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1728 section_nr = pfn_to_section_nr(pfn);
1729 if (!present_section_nr(section_nr))
1732 section = __nr_to_section(section_nr);
1733 /* same memblock? */
1735 if ((section_nr >= mem->start_section_nr) &&
1736 (section_nr <= mem->end_section_nr))
1739 mem = find_memory_block_hinted(section, mem);
1743 ret = func(mem, arg);
1745 kobject_put(&mem->dev.kobj);
1751 kobject_put(&mem->dev.kobj);
1756 #ifdef CONFIG_MEMORY_HOTREMOVE
1757 static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
1759 int ret = !is_memblock_offlined(mem);
1761 if (unlikely(ret)) {
1762 phys_addr_t beginpa, endpa;
1764 beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
1765 endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1;
1766 pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n",
1773 static int check_cpu_on_node(pg_data_t *pgdat)
1777 for_each_present_cpu(cpu) {
1778 if (cpu_to_node(cpu) == pgdat->node_id)
1780 * the cpu on this node isn't removed, and we can't
1781 * offline this node.
1793 * Offline a node if all memory sections and cpus of the node are removed.
1795 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1796 * and online/offline operations before this call.
1798 void try_offline_node(int nid)
1800 pg_data_t *pgdat = NODE_DATA(nid);
1801 unsigned long start_pfn = pgdat->node_start_pfn;
1802 unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
1805 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1806 unsigned long section_nr = pfn_to_section_nr(pfn);
1808 if (!present_section_nr(section_nr))
1811 if (pfn_to_nid(pfn) != nid)
1815 * some memory sections of this node are not removed, and we
1816 * can't offline node now.
1821 if (check_cpu_on_node(pgdat))
1825 * all memory/cpu of this node are removed, we can offline this
1828 node_set_offline(nid);
1829 unregister_one_node(nid);
1831 EXPORT_SYMBOL(try_offline_node);
1836 * @start: physical address of the region to remove
1837 * @size: size of the region to remove
1839 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1840 * and online/offline operations before this call, as required by
1841 * try_offline_node().
1843 void __ref __remove_memory(int nid, u64 start, u64 size)
1847 BUG_ON(check_hotplug_memory_range(start, size));
1849 mem_hotplug_begin();
1852 * All memory blocks must be offlined before removing memory. Check
1853 * whether all memory blocks in question are offline and trigger a BUG()
1854 * if this is not the case.
1856 ret = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL,
1857 check_memblock_offlined_cb);
1861 /* remove memmap entry */
1862 firmware_map_remove(start, start + size, "System RAM");
1863 memblock_free(start, size);
1864 memblock_remove(start, size);
1866 arch_remove_memory(nid, start, size, NULL);
1868 try_offline_node(nid);
1873 void remove_memory(int nid, u64 start, u64 size)
1875 lock_device_hotplug();
1876 __remove_memory(nid, start, size);
1877 unlock_device_hotplug();
1879 EXPORT_SYMBOL_GPL(remove_memory);
1880 #endif /* CONFIG_MEMORY_HOTREMOVE */