2 * linux/mm/memory_hotplug.c
7 #include <linux/stddef.h>
9 #include <linux/sched/signal.h>
10 #include <linux/swap.h>
11 #include <linux/interrupt.h>
12 #include <linux/pagemap.h>
13 #include <linux/compiler.h>
14 #include <linux/export.h>
15 #include <linux/pagevec.h>
16 #include <linux/writeback.h>
17 #include <linux/slab.h>
18 #include <linux/sysctl.h>
19 #include <linux/cpu.h>
20 #include <linux/memory.h>
21 #include <linux/memremap.h>
22 #include <linux/memory_hotplug.h>
23 #include <linux/highmem.h>
24 #include <linux/vmalloc.h>
25 #include <linux/ioport.h>
26 #include <linux/delay.h>
27 #include <linux/migrate.h>
28 #include <linux/page-isolation.h>
29 #include <linux/pfn.h>
30 #include <linux/suspend.h>
31 #include <linux/mm_inline.h>
32 #include <linux/firmware-map.h>
33 #include <linux/stop_machine.h>
34 #include <linux/hugetlb.h>
35 #include <linux/memblock.h>
36 #include <linux/bootmem.h>
37 #include <linux/compaction.h>
39 #include <asm/tlbflush.h>
44 * online_page_callback contains pointer to current page onlining function.
45 * Initially it is generic_online_page(). If it is required it could be
46 * changed by calling set_online_page_callback() for callback registration
47 * and restore_online_page_callback() for generic callback restore.
50 static void generic_online_page(struct page *page);
52 static online_page_callback_t online_page_callback = generic_online_page;
53 static DEFINE_MUTEX(online_page_callback_lock);
55 DEFINE_STATIC_PERCPU_RWSEM(mem_hotplug_lock);
57 void get_online_mems(void)
59 percpu_down_read(&mem_hotplug_lock);
62 void put_online_mems(void)
64 percpu_up_read(&mem_hotplug_lock);
67 bool movable_node_enabled = false;
69 #ifndef CONFIG_MEMORY_HOTPLUG_DEFAULT_ONLINE
70 bool memhp_auto_online;
72 bool memhp_auto_online = true;
74 EXPORT_SYMBOL_GPL(memhp_auto_online);
76 static int __init setup_memhp_default_state(char *str)
78 if (!strcmp(str, "online"))
79 memhp_auto_online = true;
80 else if (!strcmp(str, "offline"))
81 memhp_auto_online = false;
85 __setup("memhp_default_state=", setup_memhp_default_state);
87 void mem_hotplug_begin(void)
90 percpu_down_write(&mem_hotplug_lock);
93 void mem_hotplug_done(void)
95 percpu_up_write(&mem_hotplug_lock);
99 /* add this memory to iomem resource */
100 static struct resource *register_memory_resource(u64 start, u64 size)
102 struct resource *res, *conflict;
103 res = kzalloc(sizeof(struct resource), GFP_KERNEL);
105 return ERR_PTR(-ENOMEM);
107 res->name = "System RAM";
109 res->end = start + size - 1;
110 res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
111 conflict = request_resource_conflict(&iomem_resource, res);
113 if (conflict->desc == IORES_DESC_DEVICE_PRIVATE_MEMORY) {
114 pr_debug("Device unaddressable memory block "
115 "memory hotplug at %#010llx !\n",
116 (unsigned long long)start);
118 pr_debug("System RAM resource %pR cannot be added\n", res);
120 return ERR_PTR(-EEXIST);
125 static void release_memory_resource(struct resource *res)
129 release_resource(res);
134 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
135 void get_page_bootmem(unsigned long info, struct page *page,
138 page->freelist = (void *)type;
139 SetPagePrivate(page);
140 set_page_private(page, info);
144 void put_page_bootmem(struct page *page)
148 type = (unsigned long) page->freelist;
149 BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE ||
150 type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE);
152 if (page_ref_dec_return(page) == 1) {
153 page->freelist = NULL;
154 ClearPagePrivate(page);
155 set_page_private(page, 0);
156 INIT_LIST_HEAD(&page->lru);
157 free_reserved_page(page);
161 #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
162 #ifndef CONFIG_SPARSEMEM_VMEMMAP
163 static void register_page_bootmem_info_section(unsigned long start_pfn)
165 unsigned long *usemap, mapsize, section_nr, i;
166 struct mem_section *ms;
167 struct page *page, *memmap;
169 section_nr = pfn_to_section_nr(start_pfn);
170 ms = __nr_to_section(section_nr);
172 /* Get section's memmap address */
173 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
176 * Get page for the memmap's phys address
177 * XXX: need more consideration for sparse_vmemmap...
179 page = virt_to_page(memmap);
180 mapsize = sizeof(struct page) * PAGES_PER_SECTION;
181 mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
183 /* remember memmap's page */
184 for (i = 0; i < mapsize; i++, page++)
185 get_page_bootmem(section_nr, page, SECTION_INFO);
187 usemap = __nr_to_section(section_nr)->pageblock_flags;
188 page = virt_to_page(usemap);
190 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
192 for (i = 0; i < mapsize; i++, page++)
193 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
196 #else /* CONFIG_SPARSEMEM_VMEMMAP */
197 static void register_page_bootmem_info_section(unsigned long start_pfn)
199 unsigned long *usemap, mapsize, section_nr, i;
200 struct mem_section *ms;
201 struct page *page, *memmap;
203 if (!pfn_valid(start_pfn))
206 section_nr = pfn_to_section_nr(start_pfn);
207 ms = __nr_to_section(section_nr);
209 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
211 register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION);
213 usemap = __nr_to_section(section_nr)->pageblock_flags;
214 page = virt_to_page(usemap);
216 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
218 for (i = 0; i < mapsize; i++, page++)
219 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
221 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
223 void __init register_page_bootmem_info_node(struct pglist_data *pgdat)
225 unsigned long i, pfn, end_pfn, nr_pages;
226 int node = pgdat->node_id;
229 nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
230 page = virt_to_page(pgdat);
232 for (i = 0; i < nr_pages; i++, page++)
233 get_page_bootmem(node, page, NODE_INFO);
235 pfn = pgdat->node_start_pfn;
236 end_pfn = pgdat_end_pfn(pgdat);
238 /* register section info */
239 for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
241 * Some platforms can assign the same pfn to multiple nodes - on
242 * node0 as well as nodeN. To avoid registering a pfn against
243 * multiple nodes we check that this pfn does not already
244 * reside in some other nodes.
246 if (pfn_valid(pfn) && (early_pfn_to_nid(pfn) == node))
247 register_page_bootmem_info_section(pfn);
250 #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
252 static int __meminit __add_section(int nid, unsigned long phys_start_pfn,
258 if (pfn_valid(phys_start_pfn))
261 ret = sparse_add_one_section(NODE_DATA(nid), phys_start_pfn);
266 * Make all the pages reserved so that nobody will stumble over half
268 * FIXME: We also have to associate it with a node because pfn_to_node
269 * relies on having page with the proper node.
271 for (i = 0; i < PAGES_PER_SECTION; i++) {
272 unsigned long pfn = phys_start_pfn + i;
277 page = pfn_to_page(pfn);
278 set_page_node(page, nid);
279 SetPageReserved(page);
285 return register_new_memory(nid, __pfn_to_section(phys_start_pfn));
289 * Reasonably generic function for adding memory. It is
290 * expected that archs that support memory hotplug will
291 * call this function after deciding the zone to which to
294 int __ref __add_pages(int nid, unsigned long phys_start_pfn,
295 unsigned long nr_pages, bool want_memblock)
299 int start_sec, end_sec;
300 struct vmem_altmap *altmap;
302 /* during initialize mem_map, align hot-added range to section */
303 start_sec = pfn_to_section_nr(phys_start_pfn);
304 end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
306 altmap = to_vmem_altmap((unsigned long) pfn_to_page(phys_start_pfn));
309 * Validate altmap is within bounds of the total request
311 if (altmap->base_pfn != phys_start_pfn
312 || vmem_altmap_offset(altmap) > nr_pages) {
313 pr_warn_once("memory add fail, invalid altmap\n");
320 for (i = start_sec; i <= end_sec; i++) {
321 err = __add_section(nid, section_nr_to_pfn(i), want_memblock);
324 * EEXIST is finally dealt with by ioresource collision
325 * check. see add_memory() => register_memory_resource()
326 * Warning will be printed if there is collision.
328 if (err && (err != -EEXIST))
333 vmemmap_populate_print_last();
337 EXPORT_SYMBOL_GPL(__add_pages);
339 #ifdef CONFIG_MEMORY_HOTREMOVE
340 /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
341 static unsigned long find_smallest_section_pfn(int nid, struct zone *zone,
342 unsigned long start_pfn,
343 unsigned long end_pfn)
345 struct mem_section *ms;
347 for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) {
348 ms = __pfn_to_section(start_pfn);
350 if (unlikely(!valid_section(ms)))
353 if (unlikely(pfn_to_nid(start_pfn) != nid))
356 if (zone && zone != page_zone(pfn_to_page(start_pfn)))
365 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
366 static unsigned long find_biggest_section_pfn(int nid, struct zone *zone,
367 unsigned long start_pfn,
368 unsigned long end_pfn)
370 struct mem_section *ms;
373 /* pfn is the end pfn of a memory section. */
375 for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) {
376 ms = __pfn_to_section(pfn);
378 if (unlikely(!valid_section(ms)))
381 if (unlikely(pfn_to_nid(pfn) != nid))
384 if (zone && zone != page_zone(pfn_to_page(pfn)))
393 static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
394 unsigned long end_pfn)
396 unsigned long zone_start_pfn = zone->zone_start_pfn;
397 unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */
398 unsigned long zone_end_pfn = z;
400 struct mem_section *ms;
401 int nid = zone_to_nid(zone);
403 zone_span_writelock(zone);
404 if (zone_start_pfn == start_pfn) {
406 * If the section is smallest section in the zone, it need
407 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
408 * In this case, we find second smallest valid mem_section
409 * for shrinking zone.
411 pfn = find_smallest_section_pfn(nid, zone, end_pfn,
414 zone->zone_start_pfn = pfn;
415 zone->spanned_pages = zone_end_pfn - pfn;
417 } else if (zone_end_pfn == end_pfn) {
419 * If the section is biggest section in the zone, it need
420 * shrink zone->spanned_pages.
421 * In this case, we find second biggest valid mem_section for
424 pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn,
427 zone->spanned_pages = pfn - zone_start_pfn + 1;
431 * The section is not biggest or smallest mem_section in the zone, it
432 * only creates a hole in the zone. So in this case, we need not
433 * change the zone. But perhaps, the zone has only hole data. Thus
434 * it check the zone has only hole or not.
436 pfn = zone_start_pfn;
437 for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) {
438 ms = __pfn_to_section(pfn);
440 if (unlikely(!valid_section(ms)))
443 if (page_zone(pfn_to_page(pfn)) != zone)
446 /* If the section is current section, it continues the loop */
447 if (start_pfn == pfn)
450 /* If we find valid section, we have nothing to do */
451 zone_span_writeunlock(zone);
455 /* The zone has no valid section */
456 zone->zone_start_pfn = 0;
457 zone->spanned_pages = 0;
458 zone_span_writeunlock(zone);
461 static void shrink_pgdat_span(struct pglist_data *pgdat,
462 unsigned long start_pfn, unsigned long end_pfn)
464 unsigned long pgdat_start_pfn = pgdat->node_start_pfn;
465 unsigned long p = pgdat_end_pfn(pgdat); /* pgdat_end_pfn namespace clash */
466 unsigned long pgdat_end_pfn = p;
468 struct mem_section *ms;
469 int nid = pgdat->node_id;
471 if (pgdat_start_pfn == start_pfn) {
473 * If the section is smallest section in the pgdat, it need
474 * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages.
475 * In this case, we find second smallest valid mem_section
476 * for shrinking zone.
478 pfn = find_smallest_section_pfn(nid, NULL, end_pfn,
481 pgdat->node_start_pfn = pfn;
482 pgdat->node_spanned_pages = pgdat_end_pfn - pfn;
484 } else if (pgdat_end_pfn == end_pfn) {
486 * If the section is biggest section in the pgdat, it need
487 * shrink pgdat->node_spanned_pages.
488 * In this case, we find second biggest valid mem_section for
491 pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn,
494 pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1;
498 * If the section is not biggest or smallest mem_section in the pgdat,
499 * it only creates a hole in the pgdat. So in this case, we need not
501 * But perhaps, the pgdat has only hole data. Thus it check the pgdat
502 * has only hole or not.
504 pfn = pgdat_start_pfn;
505 for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) {
506 ms = __pfn_to_section(pfn);
508 if (unlikely(!valid_section(ms)))
511 if (pfn_to_nid(pfn) != nid)
514 /* If the section is current section, it continues the loop */
515 if (start_pfn == pfn)
518 /* If we find valid section, we have nothing to do */
522 /* The pgdat has no valid section */
523 pgdat->node_start_pfn = 0;
524 pgdat->node_spanned_pages = 0;
527 static void __remove_zone(struct zone *zone, unsigned long start_pfn)
529 struct pglist_data *pgdat = zone->zone_pgdat;
530 int nr_pages = PAGES_PER_SECTION;
533 pgdat_resize_lock(zone->zone_pgdat, &flags);
534 shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
535 shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages);
536 pgdat_resize_unlock(zone->zone_pgdat, &flags);
539 static int __remove_section(struct zone *zone, struct mem_section *ms,
540 unsigned long map_offset)
542 unsigned long start_pfn;
546 if (!valid_section(ms))
549 ret = unregister_memory_section(ms);
553 scn_nr = __section_nr(ms);
554 start_pfn = section_nr_to_pfn((unsigned long)scn_nr);
555 __remove_zone(zone, start_pfn);
557 sparse_remove_one_section(zone, ms, map_offset);
562 * __remove_pages() - remove sections of pages from a zone
563 * @zone: zone from which pages need to be removed
564 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
565 * @nr_pages: number of pages to remove (must be multiple of section size)
567 * Generic helper function to remove section mappings and sysfs entries
568 * for the section of the memory we are removing. Caller needs to make
569 * sure that pages are marked reserved and zones are adjust properly by
570 * calling offline_pages().
572 int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
573 unsigned long nr_pages)
576 unsigned long map_offset = 0;
577 int sections_to_remove, ret = 0;
579 /* In the ZONE_DEVICE case device driver owns the memory region */
580 if (is_dev_zone(zone)) {
581 struct page *page = pfn_to_page(phys_start_pfn);
582 struct vmem_altmap *altmap;
584 altmap = to_vmem_altmap((unsigned long) page);
586 map_offset = vmem_altmap_offset(altmap);
588 resource_size_t start, size;
590 start = phys_start_pfn << PAGE_SHIFT;
591 size = nr_pages * PAGE_SIZE;
593 ret = release_mem_region_adjustable(&iomem_resource, start,
596 resource_size_t endres = start + size - 1;
598 pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
599 &start, &endres, ret);
603 clear_zone_contiguous(zone);
606 * We can only remove entire sections
608 BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
609 BUG_ON(nr_pages % PAGES_PER_SECTION);
611 sections_to_remove = nr_pages / PAGES_PER_SECTION;
612 for (i = 0; i < sections_to_remove; i++) {
613 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
615 ret = __remove_section(zone, __pfn_to_section(pfn), map_offset);
621 set_zone_contiguous(zone);
625 #endif /* CONFIG_MEMORY_HOTREMOVE */
627 int set_online_page_callback(online_page_callback_t callback)
632 mutex_lock(&online_page_callback_lock);
634 if (online_page_callback == generic_online_page) {
635 online_page_callback = callback;
639 mutex_unlock(&online_page_callback_lock);
644 EXPORT_SYMBOL_GPL(set_online_page_callback);
646 int restore_online_page_callback(online_page_callback_t callback)
651 mutex_lock(&online_page_callback_lock);
653 if (online_page_callback == callback) {
654 online_page_callback = generic_online_page;
658 mutex_unlock(&online_page_callback_lock);
663 EXPORT_SYMBOL_GPL(restore_online_page_callback);
665 void __online_page_set_limits(struct page *page)
668 EXPORT_SYMBOL_GPL(__online_page_set_limits);
670 void __online_page_increment_counters(struct page *page)
672 adjust_managed_page_count(page, 1);
674 EXPORT_SYMBOL_GPL(__online_page_increment_counters);
676 void __online_page_free(struct page *page)
678 __free_reserved_page(page);
680 EXPORT_SYMBOL_GPL(__online_page_free);
682 static void generic_online_page(struct page *page)
684 __online_page_set_limits(page);
685 __online_page_increment_counters(page);
686 __online_page_free(page);
689 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
693 unsigned long onlined_pages = *(unsigned long *)arg;
696 if (PageReserved(pfn_to_page(start_pfn)))
697 for (i = 0; i < nr_pages; i++) {
698 page = pfn_to_page(start_pfn + i);
699 (*online_page_callback)(page);
703 online_mem_sections(start_pfn, start_pfn + nr_pages);
705 *(unsigned long *)arg = onlined_pages;
709 /* check which state of node_states will be changed when online memory */
710 static void node_states_check_changes_online(unsigned long nr_pages,
711 struct zone *zone, struct memory_notify *arg)
713 int nid = zone_to_nid(zone);
714 enum zone_type zone_last = ZONE_NORMAL;
717 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
718 * contains nodes which have zones of 0...ZONE_NORMAL,
719 * set zone_last to ZONE_NORMAL.
721 * If we don't have HIGHMEM nor movable node,
722 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
723 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
725 if (N_MEMORY == N_NORMAL_MEMORY)
726 zone_last = ZONE_MOVABLE;
729 * if the memory to be online is in a zone of 0...zone_last, and
730 * the zones of 0...zone_last don't have memory before online, we will
731 * need to set the node to node_states[N_NORMAL_MEMORY] after
732 * the memory is online.
734 if (zone_idx(zone) <= zone_last && !node_state(nid, N_NORMAL_MEMORY))
735 arg->status_change_nid_normal = nid;
737 arg->status_change_nid_normal = -1;
739 #ifdef CONFIG_HIGHMEM
741 * If we have movable node, node_states[N_HIGH_MEMORY]
742 * contains nodes which have zones of 0...ZONE_HIGHMEM,
743 * set zone_last to ZONE_HIGHMEM.
745 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
746 * contains nodes which have zones of 0...ZONE_MOVABLE,
747 * set zone_last to ZONE_MOVABLE.
749 zone_last = ZONE_HIGHMEM;
750 if (N_MEMORY == N_HIGH_MEMORY)
751 zone_last = ZONE_MOVABLE;
753 if (zone_idx(zone) <= zone_last && !node_state(nid, N_HIGH_MEMORY))
754 arg->status_change_nid_high = nid;
756 arg->status_change_nid_high = -1;
758 arg->status_change_nid_high = arg->status_change_nid_normal;
762 * if the node don't have memory befor online, we will need to
763 * set the node to node_states[N_MEMORY] after the memory
766 if (!node_state(nid, N_MEMORY))
767 arg->status_change_nid = nid;
769 arg->status_change_nid = -1;
772 static void node_states_set_node(int node, struct memory_notify *arg)
774 if (arg->status_change_nid_normal >= 0)
775 node_set_state(node, N_NORMAL_MEMORY);
777 if (arg->status_change_nid_high >= 0)
778 node_set_state(node, N_HIGH_MEMORY);
780 node_set_state(node, N_MEMORY);
783 static void __meminit resize_zone_range(struct zone *zone, unsigned long start_pfn,
784 unsigned long nr_pages)
786 unsigned long old_end_pfn = zone_end_pfn(zone);
788 if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
789 zone->zone_start_pfn = start_pfn;
791 zone->spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - zone->zone_start_pfn;
794 static void __meminit resize_pgdat_range(struct pglist_data *pgdat, unsigned long start_pfn,
795 unsigned long nr_pages)
797 unsigned long old_end_pfn = pgdat_end_pfn(pgdat);
799 if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
800 pgdat->node_start_pfn = start_pfn;
802 pgdat->node_spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - pgdat->node_start_pfn;
805 void __ref move_pfn_range_to_zone(struct zone *zone,
806 unsigned long start_pfn, unsigned long nr_pages)
808 struct pglist_data *pgdat = zone->zone_pgdat;
809 int nid = pgdat->node_id;
812 if (zone_is_empty(zone))
813 init_currently_empty_zone(zone, start_pfn, nr_pages);
815 clear_zone_contiguous(zone);
817 /* TODO Huh pgdat is irqsave while zone is not. It used to be like that before */
818 pgdat_resize_lock(pgdat, &flags);
819 zone_span_writelock(zone);
820 resize_zone_range(zone, start_pfn, nr_pages);
821 zone_span_writeunlock(zone);
822 resize_pgdat_range(pgdat, start_pfn, nr_pages);
823 pgdat_resize_unlock(pgdat, &flags);
826 * TODO now we have a visible range of pages which are not associated
827 * with their zone properly. Not nice but set_pfnblock_flags_mask
828 * expects the zone spans the pfn range. All the pages in the range
829 * are reserved so nobody should be touching them so we should be safe
831 memmap_init_zone(nr_pages, nid, zone_idx(zone), start_pfn, MEMMAP_HOTPLUG);
833 set_zone_contiguous(zone);
837 * Returns a default kernel memory zone for the given pfn range.
838 * If no kernel zone covers this pfn range it will automatically go
839 * to the ZONE_NORMAL.
841 static struct zone *default_kernel_zone_for_pfn(int nid, unsigned long start_pfn,
842 unsigned long nr_pages)
844 struct pglist_data *pgdat = NODE_DATA(nid);
847 for (zid = 0; zid <= ZONE_NORMAL; zid++) {
848 struct zone *zone = &pgdat->node_zones[zid];
850 if (zone_intersects(zone, start_pfn, nr_pages))
854 return &pgdat->node_zones[ZONE_NORMAL];
857 static inline struct zone *default_zone_for_pfn(int nid, unsigned long start_pfn,
858 unsigned long nr_pages)
860 struct zone *kernel_zone = default_kernel_zone_for_pfn(nid, start_pfn,
862 struct zone *movable_zone = &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
863 bool in_kernel = zone_intersects(kernel_zone, start_pfn, nr_pages);
864 bool in_movable = zone_intersects(movable_zone, start_pfn, nr_pages);
867 * We inherit the existing zone in a simple case where zones do not
868 * overlap in the given range
870 if (in_kernel ^ in_movable)
871 return (in_kernel) ? kernel_zone : movable_zone;
874 * If the range doesn't belong to any zone or two zones overlap in the
875 * given range then we use movable zone only if movable_node is
876 * enabled because we always online to a kernel zone by default.
878 return movable_node_enabled ? movable_zone : kernel_zone;
881 struct zone * zone_for_pfn_range(int online_type, int nid, unsigned start_pfn,
882 unsigned long nr_pages)
884 if (online_type == MMOP_ONLINE_KERNEL)
885 return default_kernel_zone_for_pfn(nid, start_pfn, nr_pages);
887 if (online_type == MMOP_ONLINE_MOVABLE)
888 return &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
890 return default_zone_for_pfn(nid, start_pfn, nr_pages);
894 * Associates the given pfn range with the given node and the zone appropriate
895 * for the given online type.
897 static struct zone * __meminit move_pfn_range(int online_type, int nid,
898 unsigned long start_pfn, unsigned long nr_pages)
902 zone = zone_for_pfn_range(online_type, nid, start_pfn, nr_pages);
903 move_pfn_range_to_zone(zone, start_pfn, nr_pages);
907 /* Must be protected by mem_hotplug_begin() or a device_lock */
908 int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type)
911 unsigned long onlined_pages = 0;
913 int need_zonelists_rebuild = 0;
916 struct memory_notify arg;
918 nid = pfn_to_nid(pfn);
919 /* associate pfn range with the zone */
920 zone = move_pfn_range(online_type, nid, pfn, nr_pages);
923 arg.nr_pages = nr_pages;
924 node_states_check_changes_online(nr_pages, zone, &arg);
926 ret = memory_notify(MEM_GOING_ONLINE, &arg);
927 ret = notifier_to_errno(ret);
929 goto failed_addition;
932 * If this zone is not populated, then it is not in zonelist.
933 * This means the page allocator ignores this zone.
934 * So, zonelist must be updated after online.
936 if (!populated_zone(zone)) {
937 need_zonelists_rebuild = 1;
938 setup_zone_pageset(zone);
941 ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
944 if (need_zonelists_rebuild)
945 zone_pcp_reset(zone);
946 goto failed_addition;
949 zone->present_pages += onlined_pages;
951 pgdat_resize_lock(zone->zone_pgdat, &flags);
952 zone->zone_pgdat->node_present_pages += onlined_pages;
953 pgdat_resize_unlock(zone->zone_pgdat, &flags);
956 node_states_set_node(nid, &arg);
957 if (need_zonelists_rebuild)
958 build_all_zonelists(NULL);
960 zone_pcp_update(zone);
963 init_per_zone_wmark_min();
970 vm_total_pages = nr_free_pagecache_pages();
972 writeback_set_ratelimit();
975 memory_notify(MEM_ONLINE, &arg);
979 pr_debug("online_pages [mem %#010llx-%#010llx] failed\n",
980 (unsigned long long) pfn << PAGE_SHIFT,
981 (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1);
982 memory_notify(MEM_CANCEL_ONLINE, &arg);
985 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
987 static void reset_node_present_pages(pg_data_t *pgdat)
991 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
992 z->present_pages = 0;
994 pgdat->node_present_pages = 0;
997 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
998 static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
1000 struct pglist_data *pgdat;
1001 unsigned long zones_size[MAX_NR_ZONES] = {0};
1002 unsigned long zholes_size[MAX_NR_ZONES] = {0};
1003 unsigned long start_pfn = PFN_DOWN(start);
1005 pgdat = NODE_DATA(nid);
1007 pgdat = arch_alloc_nodedata(nid);
1011 arch_refresh_nodedata(nid, pgdat);
1014 * Reset the nr_zones, order and classzone_idx before reuse.
1015 * Note that kswapd will init kswapd_classzone_idx properly
1016 * when it starts in the near future.
1018 pgdat->nr_zones = 0;
1019 pgdat->kswapd_order = 0;
1020 pgdat->kswapd_classzone_idx = 0;
1023 /* we can use NODE_DATA(nid) from here */
1025 /* init node's zones as empty zones, we don't have any present pages.*/
1026 free_area_init_node(nid, zones_size, start_pfn, zholes_size);
1027 pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat);
1030 * The node we allocated has no zone fallback lists. For avoiding
1031 * to access not-initialized zonelist, build here.
1033 build_all_zonelists(pgdat);
1036 * zone->managed_pages is set to an approximate value in
1037 * free_area_init_core(), which will cause
1038 * /sys/device/system/node/nodeX/meminfo has wrong data.
1039 * So reset it to 0 before any memory is onlined.
1041 reset_node_managed_pages(pgdat);
1044 * When memory is hot-added, all the memory is in offline state. So
1045 * clear all zones' present_pages because they will be updated in
1046 * online_pages() and offline_pages().
1048 reset_node_present_pages(pgdat);
1053 static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
1055 arch_refresh_nodedata(nid, NULL);
1056 free_percpu(pgdat->per_cpu_nodestats);
1057 arch_free_nodedata(pgdat);
1063 * try_online_node - online a node if offlined
1065 * called by cpu_up() to online a node without onlined memory.
1067 int try_online_node(int nid)
1072 if (node_online(nid))
1075 mem_hotplug_begin();
1076 pgdat = hotadd_new_pgdat(nid, 0);
1078 pr_err("Cannot online node %d due to NULL pgdat\n", nid);
1082 node_set_online(nid);
1083 ret = register_one_node(nid);
1090 static int check_hotplug_memory_range(u64 start, u64 size)
1092 u64 start_pfn = PFN_DOWN(start);
1093 u64 nr_pages = size >> PAGE_SHIFT;
1095 /* Memory range must be aligned with section */
1096 if ((start_pfn & ~PAGE_SECTION_MASK) ||
1097 (nr_pages % PAGES_PER_SECTION) || (!nr_pages)) {
1098 pr_err("Section-unaligned hotplug range: start 0x%llx, size 0x%llx\n",
1099 (unsigned long long)start,
1100 (unsigned long long)size);
1107 static int online_memory_block(struct memory_block *mem, void *arg)
1109 return device_online(&mem->dev);
1112 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1113 int __ref add_memory_resource(int nid, struct resource *res, bool online)
1116 pg_data_t *pgdat = NULL;
1122 size = resource_size(res);
1124 ret = check_hotplug_memory_range(start, size);
1128 { /* Stupid hack to suppress address-never-null warning */
1129 void *p = NODE_DATA(nid);
1133 mem_hotplug_begin();
1136 * Add new range to memblock so that when hotadd_new_pgdat() is called
1137 * to allocate new pgdat, get_pfn_range_for_nid() will be able to find
1138 * this new range and calculate total pages correctly. The range will
1139 * be removed at hot-remove time.
1141 memblock_add_node(start, size, nid);
1143 new_node = !node_online(nid);
1145 pgdat = hotadd_new_pgdat(nid, start);
1151 /* call arch's memory hotadd */
1152 ret = arch_add_memory(nid, start, size, true);
1157 /* we online node here. we can't roll back from here. */
1158 node_set_online(nid);
1161 unsigned long start_pfn = start >> PAGE_SHIFT;
1162 unsigned long nr_pages = size >> PAGE_SHIFT;
1164 ret = __register_one_node(nid);
1169 * link memory sections under this node. This is already
1170 * done when creatig memory section in register_new_memory
1171 * but that depends to have the node registered so offline
1172 * nodes have to go through register_node.
1173 * TODO clean up this mess.
1175 ret = link_mem_sections(nid, start_pfn, nr_pages);
1178 * If sysfs file of new node can't create, cpu on the node
1179 * can't be hot-added. There is no rollback way now.
1180 * So, check by BUG_ON() to catch it reluctantly..
1185 /* create new memmap entry */
1186 firmware_map_add_hotplug(start, start + size, "System RAM");
1188 /* online pages if requested */
1190 walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1),
1191 NULL, online_memory_block);
1196 /* rollback pgdat allocation and others */
1197 if (new_pgdat && pgdat)
1198 rollback_node_hotadd(nid, pgdat);
1199 memblock_remove(start, size);
1205 EXPORT_SYMBOL_GPL(add_memory_resource);
1207 int __ref add_memory(int nid, u64 start, u64 size)
1209 struct resource *res;
1212 res = register_memory_resource(start, size);
1214 return PTR_ERR(res);
1216 ret = add_memory_resource(nid, res, memhp_auto_online);
1218 release_memory_resource(res);
1221 EXPORT_SYMBOL_GPL(add_memory);
1223 #ifdef CONFIG_MEMORY_HOTREMOVE
1225 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1226 * set and the size of the free page is given by page_order(). Using this,
1227 * the function determines if the pageblock contains only free pages.
1228 * Due to buddy contraints, a free page at least the size of a pageblock will
1229 * be located at the start of the pageblock
1231 static inline int pageblock_free(struct page *page)
1233 return PageBuddy(page) && page_order(page) >= pageblock_order;
1236 /* Return the start of the next active pageblock after a given page */
1237 static struct page *next_active_pageblock(struct page *page)
1239 /* Ensure the starting page is pageblock-aligned */
1240 BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
1242 /* If the entire pageblock is free, move to the end of free page */
1243 if (pageblock_free(page)) {
1245 /* be careful. we don't have locks, page_order can be changed.*/
1246 order = page_order(page);
1247 if ((order < MAX_ORDER) && (order >= pageblock_order))
1248 return page + (1 << order);
1251 return page + pageblock_nr_pages;
1254 /* Checks if this range of memory is likely to be hot-removable. */
1255 bool is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
1257 struct page *page = pfn_to_page(start_pfn);
1258 struct page *end_page = page + nr_pages;
1260 /* Check the starting page of each pageblock within the range */
1261 for (; page < end_page; page = next_active_pageblock(page)) {
1262 if (!is_pageblock_removable_nolock(page))
1267 /* All pageblocks in the memory block are likely to be hot-removable */
1272 * Confirm all pages in a range [start, end) belong to the same zone.
1273 * When true, return its valid [start, end).
1275 int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn,
1276 unsigned long *valid_start, unsigned long *valid_end)
1278 unsigned long pfn, sec_end_pfn;
1279 unsigned long start, end;
1280 struct zone *zone = NULL;
1283 for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1);
1285 pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) {
1286 /* Make sure the memory section is present first */
1287 if (!present_section_nr(pfn_to_section_nr(pfn)))
1289 for (; pfn < sec_end_pfn && pfn < end_pfn;
1290 pfn += MAX_ORDER_NR_PAGES) {
1292 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1293 while ((i < MAX_ORDER_NR_PAGES) &&
1294 !pfn_valid_within(pfn + i))
1296 if (i == MAX_ORDER_NR_PAGES || pfn + i >= end_pfn)
1298 page = pfn_to_page(pfn + i);
1299 if (zone && page_zone(page) != zone)
1303 zone = page_zone(page);
1304 end = pfn + MAX_ORDER_NR_PAGES;
1309 *valid_start = start;
1310 *valid_end = min(end, end_pfn);
1318 * Scan pfn range [start,end) to find movable/migratable pages (LRU pages,
1319 * non-lru movable pages and hugepages). We scan pfn because it's much
1320 * easier than scanning over linked list. This function returns the pfn
1321 * of the first found movable page if it's found, otherwise 0.
1323 static unsigned long scan_movable_pages(unsigned long start, unsigned long end)
1327 for (pfn = start; pfn < end; pfn++) {
1328 if (pfn_valid(pfn)) {
1329 page = pfn_to_page(pfn);
1332 if (__PageMovable(page))
1334 if (PageHuge(page)) {
1335 if (page_huge_active(page))
1338 pfn = round_up(pfn + 1,
1339 1 << compound_order(page)) - 1;
1346 static struct page *new_node_page(struct page *page, unsigned long private,
1349 int nid = page_to_nid(page);
1350 nodemask_t nmask = node_states[N_MEMORY];
1353 * try to allocate from a different node but reuse this node if there
1354 * are no other online nodes to be used (e.g. we are offlining a part
1355 * of the only existing node)
1357 node_clear(nid, nmask);
1358 if (nodes_empty(nmask))
1359 node_set(nid, nmask);
1361 return new_page_nodemask(page, nid, &nmask);
1364 #define NR_OFFLINE_AT_ONCE_PAGES (256)
1366 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1370 int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
1371 int not_managed = 0;
1375 for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
1376 if (!pfn_valid(pfn))
1378 page = pfn_to_page(pfn);
1380 if (PageHuge(page)) {
1381 struct page *head = compound_head(page);
1382 pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1;
1383 if (compound_order(head) > PFN_SECTION_SHIFT) {
1387 if (isolate_huge_page(page, &source))
1388 move_pages -= 1 << compound_order(head);
1390 } else if (thp_migration_supported() && PageTransHuge(page))
1391 pfn = page_to_pfn(compound_head(page))
1392 + hpage_nr_pages(page) - 1;
1394 if (!get_page_unless_zero(page))
1397 * We can skip free pages. And we can deal with pages on
1398 * LRU and non-lru movable pages.
1401 ret = isolate_lru_page(page);
1403 ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE);
1404 if (!ret) { /* Success */
1406 list_add_tail(&page->lru, &source);
1408 if (!__PageMovable(page))
1409 inc_node_page_state(page, NR_ISOLATED_ANON +
1410 page_is_file_cache(page));
1413 #ifdef CONFIG_DEBUG_VM
1414 pr_alert("failed to isolate pfn %lx\n", pfn);
1415 dump_page(page, "isolation failed");
1418 /* Because we don't have big zone->lock. we should
1419 check this again here. */
1420 if (page_count(page)) {
1427 if (!list_empty(&source)) {
1429 putback_movable_pages(&source);
1433 /* Allocate a new page from the nearest neighbor node */
1434 ret = migrate_pages(&source, new_node_page, NULL, 0,
1435 MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
1437 putback_movable_pages(&source);
1444 * remove from free_area[] and mark all as Reserved.
1447 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1450 __offline_isolated_pages(start, start + nr_pages);
1455 offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
1457 walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
1458 offline_isolated_pages_cb);
1462 * Check all pages in range, recoreded as memory resource, are isolated.
1465 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1469 long offlined = *(long *)data;
1470 ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true);
1471 offlined = nr_pages;
1473 *(long *)data += offlined;
1478 check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
1483 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
1484 check_pages_isolated_cb);
1486 offlined = (long)ret;
1490 static int __init cmdline_parse_movable_node(char *p)
1492 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1493 movable_node_enabled = true;
1495 pr_warn("movable_node parameter depends on CONFIG_HAVE_MEMBLOCK_NODE_MAP to work properly\n");
1499 early_param("movable_node", cmdline_parse_movable_node);
1501 /* check which state of node_states will be changed when offline memory */
1502 static void node_states_check_changes_offline(unsigned long nr_pages,
1503 struct zone *zone, struct memory_notify *arg)
1505 struct pglist_data *pgdat = zone->zone_pgdat;
1506 unsigned long present_pages = 0;
1507 enum zone_type zt, zone_last = ZONE_NORMAL;
1510 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
1511 * contains nodes which have zones of 0...ZONE_NORMAL,
1512 * set zone_last to ZONE_NORMAL.
1514 * If we don't have HIGHMEM nor movable node,
1515 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
1516 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
1518 if (N_MEMORY == N_NORMAL_MEMORY)
1519 zone_last = ZONE_MOVABLE;
1522 * check whether node_states[N_NORMAL_MEMORY] will be changed.
1523 * If the memory to be offline is in a zone of 0...zone_last,
1524 * and it is the last present memory, 0...zone_last will
1525 * become empty after offline , thus we can determind we will
1526 * need to clear the node from node_states[N_NORMAL_MEMORY].
1528 for (zt = 0; zt <= zone_last; zt++)
1529 present_pages += pgdat->node_zones[zt].present_pages;
1530 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1531 arg->status_change_nid_normal = zone_to_nid(zone);
1533 arg->status_change_nid_normal = -1;
1535 #ifdef CONFIG_HIGHMEM
1537 * If we have movable node, node_states[N_HIGH_MEMORY]
1538 * contains nodes which have zones of 0...ZONE_HIGHMEM,
1539 * set zone_last to ZONE_HIGHMEM.
1541 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
1542 * contains nodes which have zones of 0...ZONE_MOVABLE,
1543 * set zone_last to ZONE_MOVABLE.
1545 zone_last = ZONE_HIGHMEM;
1546 if (N_MEMORY == N_HIGH_MEMORY)
1547 zone_last = ZONE_MOVABLE;
1549 for (; zt <= zone_last; zt++)
1550 present_pages += pgdat->node_zones[zt].present_pages;
1551 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1552 arg->status_change_nid_high = zone_to_nid(zone);
1554 arg->status_change_nid_high = -1;
1556 arg->status_change_nid_high = arg->status_change_nid_normal;
1560 * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE
1562 zone_last = ZONE_MOVABLE;
1565 * check whether node_states[N_HIGH_MEMORY] will be changed
1566 * If we try to offline the last present @nr_pages from the node,
1567 * we can determind we will need to clear the node from
1568 * node_states[N_HIGH_MEMORY].
1570 for (; zt <= zone_last; zt++)
1571 present_pages += pgdat->node_zones[zt].present_pages;
1572 if (nr_pages >= present_pages)
1573 arg->status_change_nid = zone_to_nid(zone);
1575 arg->status_change_nid = -1;
1578 static void node_states_clear_node(int node, struct memory_notify *arg)
1580 if (arg->status_change_nid_normal >= 0)
1581 node_clear_state(node, N_NORMAL_MEMORY);
1583 if ((N_MEMORY != N_NORMAL_MEMORY) &&
1584 (arg->status_change_nid_high >= 0))
1585 node_clear_state(node, N_HIGH_MEMORY);
1587 if ((N_MEMORY != N_HIGH_MEMORY) &&
1588 (arg->status_change_nid >= 0))
1589 node_clear_state(node, N_MEMORY);
1592 static int __ref __offline_pages(unsigned long start_pfn,
1593 unsigned long end_pfn, unsigned long timeout)
1595 unsigned long pfn, nr_pages, expire;
1596 long offlined_pages;
1597 int ret, drain, retry_max, node;
1598 unsigned long flags;
1599 unsigned long valid_start, valid_end;
1601 struct memory_notify arg;
1603 /* at least, alignment against pageblock is necessary */
1604 if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
1606 if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
1608 /* This makes hotplug much easier...and readable.
1609 we assume this for now. .*/
1610 if (!test_pages_in_a_zone(start_pfn, end_pfn, &valid_start, &valid_end))
1613 zone = page_zone(pfn_to_page(valid_start));
1614 node = zone_to_nid(zone);
1615 nr_pages = end_pfn - start_pfn;
1617 /* set above range as isolated */
1618 ret = start_isolate_page_range(start_pfn, end_pfn,
1619 MIGRATE_MOVABLE, true);
1623 arg.start_pfn = start_pfn;
1624 arg.nr_pages = nr_pages;
1625 node_states_check_changes_offline(nr_pages, zone, &arg);
1627 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1628 ret = notifier_to_errno(ret);
1630 goto failed_removal;
1633 expire = jiffies + timeout;
1637 /* start memory hot removal */
1639 if (time_after(jiffies, expire))
1640 goto failed_removal;
1642 if (signal_pending(current))
1643 goto failed_removal;
1646 lru_add_drain_all_cpuslocked();
1648 drain_all_pages(zone);
1651 pfn = scan_movable_pages(start_pfn, end_pfn);
1652 if (pfn) { /* We have movable pages */
1653 ret = do_migrate_range(pfn, end_pfn);
1659 if (--retry_max == 0)
1660 goto failed_removal;
1666 /* drain all zone's lru pagevec, this is asynchronous... */
1667 lru_add_drain_all_cpuslocked();
1669 /* drain pcp pages, this is synchronous. */
1670 drain_all_pages(zone);
1672 * dissolve free hugepages in the memory block before doing offlining
1673 * actually in order to make hugetlbfs's object counting consistent.
1675 ret = dissolve_free_huge_pages(start_pfn, end_pfn);
1677 goto failed_removal;
1679 offlined_pages = check_pages_isolated(start_pfn, end_pfn);
1680 if (offlined_pages < 0) {
1682 goto failed_removal;
1684 pr_info("Offlined Pages %ld\n", offlined_pages);
1685 /* Ok, all of our target is isolated.
1686 We cannot do rollback at this point. */
1687 offline_isolated_pages(start_pfn, end_pfn);
1688 /* reset pagetype flags and makes migrate type to be MOVABLE */
1689 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1690 /* removal success */
1691 adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages);
1692 zone->present_pages -= offlined_pages;
1694 pgdat_resize_lock(zone->zone_pgdat, &flags);
1695 zone->zone_pgdat->node_present_pages -= offlined_pages;
1696 pgdat_resize_unlock(zone->zone_pgdat, &flags);
1698 init_per_zone_wmark_min();
1700 if (!populated_zone(zone)) {
1701 zone_pcp_reset(zone);
1702 build_all_zonelists(NULL);
1704 zone_pcp_update(zone);
1706 node_states_clear_node(node, &arg);
1707 if (arg.status_change_nid >= 0) {
1709 kcompactd_stop(node);
1712 vm_total_pages = nr_free_pagecache_pages();
1713 writeback_set_ratelimit();
1715 memory_notify(MEM_OFFLINE, &arg);
1719 pr_debug("memory offlining [mem %#010llx-%#010llx] failed\n",
1720 (unsigned long long) start_pfn << PAGE_SHIFT,
1721 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
1722 memory_notify(MEM_CANCEL_OFFLINE, &arg);
1723 /* pushback to free area */
1724 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1728 /* Must be protected by mem_hotplug_begin() or a device_lock */
1729 int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1731 return __offline_pages(start_pfn, start_pfn + nr_pages, 120 * HZ);
1733 #endif /* CONFIG_MEMORY_HOTREMOVE */
1736 * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
1737 * @start_pfn: start pfn of the memory range
1738 * @end_pfn: end pfn of the memory range
1739 * @arg: argument passed to func
1740 * @func: callback for each memory section walked
1742 * This function walks through all present mem sections in range
1743 * [start_pfn, end_pfn) and call func on each mem section.
1745 * Returns the return value of func.
1747 int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn,
1748 void *arg, int (*func)(struct memory_block *, void *))
1750 struct memory_block *mem = NULL;
1751 struct mem_section *section;
1752 unsigned long pfn, section_nr;
1755 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1756 section_nr = pfn_to_section_nr(pfn);
1757 if (!present_section_nr(section_nr))
1760 section = __nr_to_section(section_nr);
1761 /* same memblock? */
1763 if ((section_nr >= mem->start_section_nr) &&
1764 (section_nr <= mem->end_section_nr))
1767 mem = find_memory_block_hinted(section, mem);
1771 ret = func(mem, arg);
1773 kobject_put(&mem->dev.kobj);
1779 kobject_put(&mem->dev.kobj);
1784 #ifdef CONFIG_MEMORY_HOTREMOVE
1785 static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
1787 int ret = !is_memblock_offlined(mem);
1789 if (unlikely(ret)) {
1790 phys_addr_t beginpa, endpa;
1792 beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
1793 endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1;
1794 pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n",
1801 static int check_cpu_on_node(pg_data_t *pgdat)
1805 for_each_present_cpu(cpu) {
1806 if (cpu_to_node(cpu) == pgdat->node_id)
1808 * the cpu on this node isn't removed, and we can't
1809 * offline this node.
1817 static void unmap_cpu_on_node(pg_data_t *pgdat)
1819 #ifdef CONFIG_ACPI_NUMA
1822 for_each_possible_cpu(cpu)
1823 if (cpu_to_node(cpu) == pgdat->node_id)
1824 numa_clear_node(cpu);
1828 static int check_and_unmap_cpu_on_node(pg_data_t *pgdat)
1832 ret = check_cpu_on_node(pgdat);
1837 * the node will be offlined when we come here, so we can clear
1838 * the cpu_to_node() now.
1841 unmap_cpu_on_node(pgdat);
1848 * Offline a node if all memory sections and cpus of the node are removed.
1850 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1851 * and online/offline operations before this call.
1853 void try_offline_node(int nid)
1855 pg_data_t *pgdat = NODE_DATA(nid);
1856 unsigned long start_pfn = pgdat->node_start_pfn;
1857 unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
1860 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1861 unsigned long section_nr = pfn_to_section_nr(pfn);
1863 if (!present_section_nr(section_nr))
1866 if (pfn_to_nid(pfn) != nid)
1870 * some memory sections of this node are not removed, and we
1871 * can't offline node now.
1876 if (check_and_unmap_cpu_on_node(pgdat))
1880 * all memory/cpu of this node are removed, we can offline this
1883 node_set_offline(nid);
1884 unregister_one_node(nid);
1886 EXPORT_SYMBOL(try_offline_node);
1891 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1892 * and online/offline operations before this call, as required by
1893 * try_offline_node().
1895 void __ref remove_memory(int nid, u64 start, u64 size)
1899 BUG_ON(check_hotplug_memory_range(start, size));
1901 mem_hotplug_begin();
1904 * All memory blocks must be offlined before removing memory. Check
1905 * whether all memory blocks in question are offline and trigger a BUG()
1906 * if this is not the case.
1908 ret = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL,
1909 check_memblock_offlined_cb);
1913 /* remove memmap entry */
1914 firmware_map_remove(start, start + size, "System RAM");
1915 memblock_free(start, size);
1916 memblock_remove(start, size);
1918 arch_remove_memory(start, size);
1920 try_offline_node(nid);
1924 EXPORT_SYMBOL_GPL(remove_memory);
1925 #endif /* CONFIG_MEMORY_HOTREMOVE */