2 * linux/mm/memory_hotplug.c
7 #include <linux/stddef.h>
9 #include <linux/swap.h>
10 #include <linux/interrupt.h>
11 #include <linux/pagemap.h>
12 #include <linux/compiler.h>
13 #include <linux/export.h>
14 #include <linux/pagevec.h>
15 #include <linux/writeback.h>
16 #include <linux/slab.h>
17 #include <linux/sysctl.h>
18 #include <linux/cpu.h>
19 #include <linux/memory.h>
20 #include <linux/memory_hotplug.h>
21 #include <linux/highmem.h>
22 #include <linux/vmalloc.h>
23 #include <linux/ioport.h>
24 #include <linux/delay.h>
25 #include <linux/migrate.h>
26 #include <linux/page-isolation.h>
27 #include <linux/pfn.h>
28 #include <linux/suspend.h>
29 #include <linux/mm_inline.h>
30 #include <linux/firmware-map.h>
31 #include <linux/stop_machine.h>
32 #include <linux/hugetlb.h>
33 #include <linux/memblock.h>
34 #include <linux/bootmem.h>
36 #include <asm/tlbflush.h>
41 * online_page_callback contains pointer to current page onlining function.
42 * Initially it is generic_online_page(). If it is required it could be
43 * changed by calling set_online_page_callback() for callback registration
44 * and restore_online_page_callback() for generic callback restore.
47 static void generic_online_page(struct page *page);
49 static online_page_callback_t online_page_callback = generic_online_page;
50 static DEFINE_MUTEX(online_page_callback_lock);
52 /* The same as the cpu_hotplug lock, but for memory hotplug. */
54 struct task_struct *active_writer;
55 struct mutex lock; /* Synchronizes accesses to refcount, */
57 * Also blocks the new readers during
58 * an ongoing mem hotplug operation.
62 #ifdef CONFIG_DEBUG_LOCK_ALLOC
63 struct lockdep_map dep_map;
66 .active_writer = NULL,
67 .lock = __MUTEX_INITIALIZER(mem_hotplug.lock),
69 #ifdef CONFIG_DEBUG_LOCK_ALLOC
70 .dep_map = {.name = "mem_hotplug.lock" },
74 /* Lockdep annotations for get/put_online_mems() and mem_hotplug_begin/end() */
75 #define memhp_lock_acquire_read() lock_map_acquire_read(&mem_hotplug.dep_map)
76 #define memhp_lock_acquire() lock_map_acquire(&mem_hotplug.dep_map)
77 #define memhp_lock_release() lock_map_release(&mem_hotplug.dep_map)
79 void get_online_mems(void)
82 if (mem_hotplug.active_writer == current)
84 memhp_lock_acquire_read();
85 mutex_lock(&mem_hotplug.lock);
86 mem_hotplug.refcount++;
87 mutex_unlock(&mem_hotplug.lock);
91 void put_online_mems(void)
93 if (mem_hotplug.active_writer == current)
95 mutex_lock(&mem_hotplug.lock);
97 if (WARN_ON(!mem_hotplug.refcount))
98 mem_hotplug.refcount++; /* try to fix things up */
100 if (!--mem_hotplug.refcount && unlikely(mem_hotplug.active_writer))
101 wake_up_process(mem_hotplug.active_writer);
102 mutex_unlock(&mem_hotplug.lock);
103 memhp_lock_release();
107 void mem_hotplug_begin(void)
109 mem_hotplug.active_writer = current;
111 memhp_lock_acquire();
113 mutex_lock(&mem_hotplug.lock);
114 if (likely(!mem_hotplug.refcount))
116 __set_current_state(TASK_UNINTERRUPTIBLE);
117 mutex_unlock(&mem_hotplug.lock);
122 void mem_hotplug_done(void)
124 mem_hotplug.active_writer = NULL;
125 mutex_unlock(&mem_hotplug.lock);
126 memhp_lock_release();
129 /* add this memory to iomem resource */
130 static struct resource *register_memory_resource(u64 start, u64 size)
132 struct resource *res;
133 res = kzalloc(sizeof(struct resource), GFP_KERNEL);
136 res->name = "System RAM";
138 res->end = start + size - 1;
139 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
140 if (request_resource(&iomem_resource, res) < 0) {
141 pr_debug("System RAM resource %pR cannot be added\n", res);
148 static void release_memory_resource(struct resource *res)
152 release_resource(res);
157 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
158 void get_page_bootmem(unsigned long info, struct page *page,
161 page->lru.next = (struct list_head *) type;
162 SetPagePrivate(page);
163 set_page_private(page, info);
164 atomic_inc(&page->_count);
167 void put_page_bootmem(struct page *page)
171 type = (unsigned long) page->lru.next;
172 BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE ||
173 type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE);
175 if (atomic_dec_return(&page->_count) == 1) {
176 ClearPagePrivate(page);
177 set_page_private(page, 0);
178 INIT_LIST_HEAD(&page->lru);
179 free_reserved_page(page);
183 #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
184 #ifndef CONFIG_SPARSEMEM_VMEMMAP
185 static void register_page_bootmem_info_section(unsigned long start_pfn)
187 unsigned long *usemap, mapsize, section_nr, i;
188 struct mem_section *ms;
189 struct page *page, *memmap;
191 section_nr = pfn_to_section_nr(start_pfn);
192 ms = __nr_to_section(section_nr);
194 /* Get section's memmap address */
195 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
198 * Get page for the memmap's phys address
199 * XXX: need more consideration for sparse_vmemmap...
201 page = virt_to_page(memmap);
202 mapsize = sizeof(struct page) * PAGES_PER_SECTION;
203 mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
205 /* remember memmap's page */
206 for (i = 0; i < mapsize; i++, page++)
207 get_page_bootmem(section_nr, page, SECTION_INFO);
209 usemap = __nr_to_section(section_nr)->pageblock_flags;
210 page = virt_to_page(usemap);
212 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
214 for (i = 0; i < mapsize; i++, page++)
215 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
218 #else /* CONFIG_SPARSEMEM_VMEMMAP */
219 static void register_page_bootmem_info_section(unsigned long start_pfn)
221 unsigned long *usemap, mapsize, section_nr, i;
222 struct mem_section *ms;
223 struct page *page, *memmap;
225 if (!pfn_valid(start_pfn))
228 section_nr = pfn_to_section_nr(start_pfn);
229 ms = __nr_to_section(section_nr);
231 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
233 register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION);
235 usemap = __nr_to_section(section_nr)->pageblock_flags;
236 page = virt_to_page(usemap);
238 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
240 for (i = 0; i < mapsize; i++, page++)
241 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
243 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
245 void register_page_bootmem_info_node(struct pglist_data *pgdat)
247 unsigned long i, pfn, end_pfn, nr_pages;
248 int node = pgdat->node_id;
252 nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
253 page = virt_to_page(pgdat);
255 for (i = 0; i < nr_pages; i++, page++)
256 get_page_bootmem(node, page, NODE_INFO);
258 zone = &pgdat->node_zones[0];
259 for (; zone < pgdat->node_zones + MAX_NR_ZONES - 1; zone++) {
260 if (zone_is_initialized(zone)) {
261 nr_pages = zone->wait_table_hash_nr_entries
262 * sizeof(wait_queue_head_t);
263 nr_pages = PAGE_ALIGN(nr_pages) >> PAGE_SHIFT;
264 page = virt_to_page(zone->wait_table);
266 for (i = 0; i < nr_pages; i++, page++)
267 get_page_bootmem(node, page, NODE_INFO);
271 pfn = pgdat->node_start_pfn;
272 end_pfn = pgdat_end_pfn(pgdat);
274 /* register section info */
275 for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
277 * Some platforms can assign the same pfn to multiple nodes - on
278 * node0 as well as nodeN. To avoid registering a pfn against
279 * multiple nodes we check that this pfn does not already
280 * reside in some other nodes.
282 if (pfn_valid(pfn) && (pfn_to_nid(pfn) == node))
283 register_page_bootmem_info_section(pfn);
286 #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
288 static void __meminit grow_zone_span(struct zone *zone, unsigned long start_pfn,
289 unsigned long end_pfn)
291 unsigned long old_zone_end_pfn;
293 zone_span_writelock(zone);
295 old_zone_end_pfn = zone_end_pfn(zone);
296 if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
297 zone->zone_start_pfn = start_pfn;
299 zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
300 zone->zone_start_pfn;
302 zone_span_writeunlock(zone);
305 static void resize_zone(struct zone *zone, unsigned long start_pfn,
306 unsigned long end_pfn)
308 zone_span_writelock(zone);
310 if (end_pfn - start_pfn) {
311 zone->zone_start_pfn = start_pfn;
312 zone->spanned_pages = end_pfn - start_pfn;
315 * make it consist as free_area_init_core(),
316 * if spanned_pages = 0, then keep start_pfn = 0
318 zone->zone_start_pfn = 0;
319 zone->spanned_pages = 0;
322 zone_span_writeunlock(zone);
325 static void fix_zone_id(struct zone *zone, unsigned long start_pfn,
326 unsigned long end_pfn)
328 enum zone_type zid = zone_idx(zone);
329 int nid = zone->zone_pgdat->node_id;
332 for (pfn = start_pfn; pfn < end_pfn; pfn++)
333 set_page_links(pfn_to_page(pfn), zid, nid, pfn);
336 /* Can fail with -ENOMEM from allocating a wait table with vmalloc() or
337 * alloc_bootmem_node_nopanic()/memblock_virt_alloc_node_nopanic() */
338 static int __ref ensure_zone_is_initialized(struct zone *zone,
339 unsigned long start_pfn, unsigned long num_pages)
341 if (!zone_is_initialized(zone))
342 return init_currently_empty_zone(zone, start_pfn, num_pages);
347 static int __meminit move_pfn_range_left(struct zone *z1, struct zone *z2,
348 unsigned long start_pfn, unsigned long end_pfn)
352 unsigned long z1_start_pfn;
354 ret = ensure_zone_is_initialized(z1, start_pfn, end_pfn - start_pfn);
358 pgdat_resize_lock(z1->zone_pgdat, &flags);
360 /* can't move pfns which are higher than @z2 */
361 if (end_pfn > zone_end_pfn(z2))
363 /* the move out part must be at the left most of @z2 */
364 if (start_pfn > z2->zone_start_pfn)
366 /* must included/overlap */
367 if (end_pfn <= z2->zone_start_pfn)
370 /* use start_pfn for z1's start_pfn if z1 is empty */
371 if (!zone_is_empty(z1))
372 z1_start_pfn = z1->zone_start_pfn;
374 z1_start_pfn = start_pfn;
376 resize_zone(z1, z1_start_pfn, end_pfn);
377 resize_zone(z2, end_pfn, zone_end_pfn(z2));
379 pgdat_resize_unlock(z1->zone_pgdat, &flags);
381 fix_zone_id(z1, start_pfn, end_pfn);
385 pgdat_resize_unlock(z1->zone_pgdat, &flags);
389 static int __meminit move_pfn_range_right(struct zone *z1, struct zone *z2,
390 unsigned long start_pfn, unsigned long end_pfn)
394 unsigned long z2_end_pfn;
396 ret = ensure_zone_is_initialized(z2, start_pfn, end_pfn - start_pfn);
400 pgdat_resize_lock(z1->zone_pgdat, &flags);
402 /* can't move pfns which are lower than @z1 */
403 if (z1->zone_start_pfn > start_pfn)
405 /* the move out part mast at the right most of @z1 */
406 if (zone_end_pfn(z1) > end_pfn)
408 /* must included/overlap */
409 if (start_pfn >= zone_end_pfn(z1))
412 /* use end_pfn for z2's end_pfn if z2 is empty */
413 if (!zone_is_empty(z2))
414 z2_end_pfn = zone_end_pfn(z2);
416 z2_end_pfn = end_pfn;
418 resize_zone(z1, z1->zone_start_pfn, start_pfn);
419 resize_zone(z2, start_pfn, z2_end_pfn);
421 pgdat_resize_unlock(z1->zone_pgdat, &flags);
423 fix_zone_id(z2, start_pfn, end_pfn);
427 pgdat_resize_unlock(z1->zone_pgdat, &flags);
431 static void __meminit grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn,
432 unsigned long end_pfn)
434 unsigned long old_pgdat_end_pfn = pgdat_end_pfn(pgdat);
436 if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
437 pgdat->node_start_pfn = start_pfn;
439 pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
440 pgdat->node_start_pfn;
443 static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn)
445 struct pglist_data *pgdat = zone->zone_pgdat;
446 int nr_pages = PAGES_PER_SECTION;
447 int nid = pgdat->node_id;
449 unsigned long flags, pfn;
452 zone_type = zone - pgdat->node_zones;
453 ret = ensure_zone_is_initialized(zone, phys_start_pfn, nr_pages);
457 pgdat_resize_lock(zone->zone_pgdat, &flags);
458 grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages);
459 grow_pgdat_span(zone->zone_pgdat, phys_start_pfn,
460 phys_start_pfn + nr_pages);
461 pgdat_resize_unlock(zone->zone_pgdat, &flags);
462 memmap_init_zone(nr_pages, nid, zone_type,
463 phys_start_pfn, MEMMAP_HOTPLUG);
465 /* online_page_range is called later and expects pages reserved */
466 for (pfn = phys_start_pfn; pfn < phys_start_pfn + nr_pages; pfn++) {
470 SetPageReserved(pfn_to_page(pfn));
475 static int __meminit __add_section(int nid, struct zone *zone,
476 unsigned long phys_start_pfn)
480 if (pfn_valid(phys_start_pfn))
483 ret = sparse_add_one_section(zone, phys_start_pfn);
488 ret = __add_zone(zone, phys_start_pfn);
493 return register_new_memory(nid, __pfn_to_section(phys_start_pfn));
497 * Reasonably generic function for adding memory. It is
498 * expected that archs that support memory hotplug will
499 * call this function after deciding the zone to which to
502 int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn,
503 unsigned long nr_pages)
507 int start_sec, end_sec;
508 /* during initialize mem_map, align hot-added range to section */
509 start_sec = pfn_to_section_nr(phys_start_pfn);
510 end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
512 for (i = start_sec; i <= end_sec; i++) {
513 err = __add_section(nid, zone, section_nr_to_pfn(i));
516 * EEXIST is finally dealt with by ioresource collision
517 * check. see add_memory() => register_memory_resource()
518 * Warning will be printed if there is collision.
520 if (err && (err != -EEXIST))
524 vmemmap_populate_print_last();
528 EXPORT_SYMBOL_GPL(__add_pages);
530 #ifdef CONFIG_MEMORY_HOTREMOVE
531 /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
532 static int find_smallest_section_pfn(int nid, struct zone *zone,
533 unsigned long start_pfn,
534 unsigned long end_pfn)
536 struct mem_section *ms;
538 for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) {
539 ms = __pfn_to_section(start_pfn);
541 if (unlikely(!valid_section(ms)))
544 if (unlikely(pfn_to_nid(start_pfn) != nid))
547 if (zone && zone != page_zone(pfn_to_page(start_pfn)))
556 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
557 static int find_biggest_section_pfn(int nid, struct zone *zone,
558 unsigned long start_pfn,
559 unsigned long end_pfn)
561 struct mem_section *ms;
564 /* pfn is the end pfn of a memory section. */
566 for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) {
567 ms = __pfn_to_section(pfn);
569 if (unlikely(!valid_section(ms)))
572 if (unlikely(pfn_to_nid(pfn) != nid))
575 if (zone && zone != page_zone(pfn_to_page(pfn)))
584 static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
585 unsigned long end_pfn)
587 unsigned long zone_start_pfn = zone->zone_start_pfn;
588 unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */
589 unsigned long zone_end_pfn = z;
591 struct mem_section *ms;
592 int nid = zone_to_nid(zone);
594 zone_span_writelock(zone);
595 if (zone_start_pfn == start_pfn) {
597 * If the section is smallest section in the zone, it need
598 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
599 * In this case, we find second smallest valid mem_section
600 * for shrinking zone.
602 pfn = find_smallest_section_pfn(nid, zone, end_pfn,
605 zone->zone_start_pfn = pfn;
606 zone->spanned_pages = zone_end_pfn - pfn;
608 } else if (zone_end_pfn == end_pfn) {
610 * If the section is biggest section in the zone, it need
611 * shrink zone->spanned_pages.
612 * In this case, we find second biggest valid mem_section for
615 pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn,
618 zone->spanned_pages = pfn - zone_start_pfn + 1;
622 * The section is not biggest or smallest mem_section in the zone, it
623 * only creates a hole in the zone. So in this case, we need not
624 * change the zone. But perhaps, the zone has only hole data. Thus
625 * it check the zone has only hole or not.
627 pfn = zone_start_pfn;
628 for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) {
629 ms = __pfn_to_section(pfn);
631 if (unlikely(!valid_section(ms)))
634 if (page_zone(pfn_to_page(pfn)) != zone)
637 /* If the section is current section, it continues the loop */
638 if (start_pfn == pfn)
641 /* If we find valid section, we have nothing to do */
642 zone_span_writeunlock(zone);
646 /* The zone has no valid section */
647 zone->zone_start_pfn = 0;
648 zone->spanned_pages = 0;
649 zone_span_writeunlock(zone);
652 static void shrink_pgdat_span(struct pglist_data *pgdat,
653 unsigned long start_pfn, unsigned long end_pfn)
655 unsigned long pgdat_start_pfn = pgdat->node_start_pfn;
656 unsigned long p = pgdat_end_pfn(pgdat); /* pgdat_end_pfn namespace clash */
657 unsigned long pgdat_end_pfn = p;
659 struct mem_section *ms;
660 int nid = pgdat->node_id;
662 if (pgdat_start_pfn == start_pfn) {
664 * If the section is smallest section in the pgdat, it need
665 * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages.
666 * In this case, we find second smallest valid mem_section
667 * for shrinking zone.
669 pfn = find_smallest_section_pfn(nid, NULL, end_pfn,
672 pgdat->node_start_pfn = pfn;
673 pgdat->node_spanned_pages = pgdat_end_pfn - pfn;
675 } else if (pgdat_end_pfn == end_pfn) {
677 * If the section is biggest section in the pgdat, it need
678 * shrink pgdat->node_spanned_pages.
679 * In this case, we find second biggest valid mem_section for
682 pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn,
685 pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1;
689 * If the section is not biggest or smallest mem_section in the pgdat,
690 * it only creates a hole in the pgdat. So in this case, we need not
692 * But perhaps, the pgdat has only hole data. Thus it check the pgdat
693 * has only hole or not.
695 pfn = pgdat_start_pfn;
696 for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) {
697 ms = __pfn_to_section(pfn);
699 if (unlikely(!valid_section(ms)))
702 if (pfn_to_nid(pfn) != nid)
705 /* If the section is current section, it continues the loop */
706 if (start_pfn == pfn)
709 /* If we find valid section, we have nothing to do */
713 /* The pgdat has no valid section */
714 pgdat->node_start_pfn = 0;
715 pgdat->node_spanned_pages = 0;
718 static void __remove_zone(struct zone *zone, unsigned long start_pfn)
720 struct pglist_data *pgdat = zone->zone_pgdat;
721 int nr_pages = PAGES_PER_SECTION;
725 zone_type = zone - pgdat->node_zones;
727 pgdat_resize_lock(zone->zone_pgdat, &flags);
728 shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
729 shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages);
730 pgdat_resize_unlock(zone->zone_pgdat, &flags);
733 static int __remove_section(struct zone *zone, struct mem_section *ms)
735 unsigned long start_pfn;
739 if (!valid_section(ms))
742 ret = unregister_memory_section(ms);
746 scn_nr = __section_nr(ms);
747 start_pfn = section_nr_to_pfn(scn_nr);
748 __remove_zone(zone, start_pfn);
750 sparse_remove_one_section(zone, ms);
755 * __remove_pages() - remove sections of pages from a zone
756 * @zone: zone from which pages need to be removed
757 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
758 * @nr_pages: number of pages to remove (must be multiple of section size)
760 * Generic helper function to remove section mappings and sysfs entries
761 * for the section of the memory we are removing. Caller needs to make
762 * sure that pages are marked reserved and zones are adjust properly by
763 * calling offline_pages().
765 int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
766 unsigned long nr_pages)
769 int sections_to_remove;
770 resource_size_t start, size;
774 * We can only remove entire sections
776 BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
777 BUG_ON(nr_pages % PAGES_PER_SECTION);
779 start = phys_start_pfn << PAGE_SHIFT;
780 size = nr_pages * PAGE_SIZE;
782 /* in the ZONE_DEVICE case device driver owns the memory region */
783 if (!is_dev_zone(zone))
784 ret = release_mem_region_adjustable(&iomem_resource, start, size);
786 resource_size_t endres = start + size - 1;
788 pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
789 &start, &endres, ret);
792 sections_to_remove = nr_pages / PAGES_PER_SECTION;
793 for (i = 0; i < sections_to_remove; i++) {
794 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
795 ret = __remove_section(zone, __pfn_to_section(pfn));
801 EXPORT_SYMBOL_GPL(__remove_pages);
802 #endif /* CONFIG_MEMORY_HOTREMOVE */
804 int set_online_page_callback(online_page_callback_t callback)
809 mutex_lock(&online_page_callback_lock);
811 if (online_page_callback == generic_online_page) {
812 online_page_callback = callback;
816 mutex_unlock(&online_page_callback_lock);
821 EXPORT_SYMBOL_GPL(set_online_page_callback);
823 int restore_online_page_callback(online_page_callback_t callback)
828 mutex_lock(&online_page_callback_lock);
830 if (online_page_callback == callback) {
831 online_page_callback = generic_online_page;
835 mutex_unlock(&online_page_callback_lock);
840 EXPORT_SYMBOL_GPL(restore_online_page_callback);
842 void __online_page_set_limits(struct page *page)
845 EXPORT_SYMBOL_GPL(__online_page_set_limits);
847 void __online_page_increment_counters(struct page *page)
849 adjust_managed_page_count(page, 1);
851 EXPORT_SYMBOL_GPL(__online_page_increment_counters);
853 void __online_page_free(struct page *page)
855 __free_reserved_page(page);
857 EXPORT_SYMBOL_GPL(__online_page_free);
859 static void generic_online_page(struct page *page)
861 __online_page_set_limits(page);
862 __online_page_increment_counters(page);
863 __online_page_free(page);
866 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
870 unsigned long onlined_pages = *(unsigned long *)arg;
872 if (PageReserved(pfn_to_page(start_pfn)))
873 for (i = 0; i < nr_pages; i++) {
874 page = pfn_to_page(start_pfn + i);
875 (*online_page_callback)(page);
878 *(unsigned long *)arg = onlined_pages;
882 #ifdef CONFIG_MOVABLE_NODE
884 * When CONFIG_MOVABLE_NODE, we permit onlining of a node which doesn't have
887 static bool can_online_high_movable(struct zone *zone)
891 #else /* CONFIG_MOVABLE_NODE */
892 /* ensure every online node has NORMAL memory */
893 static bool can_online_high_movable(struct zone *zone)
895 return node_state(zone_to_nid(zone), N_NORMAL_MEMORY);
897 #endif /* CONFIG_MOVABLE_NODE */
899 /* check which state of node_states will be changed when online memory */
900 static void node_states_check_changes_online(unsigned long nr_pages,
901 struct zone *zone, struct memory_notify *arg)
903 int nid = zone_to_nid(zone);
904 enum zone_type zone_last = ZONE_NORMAL;
907 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
908 * contains nodes which have zones of 0...ZONE_NORMAL,
909 * set zone_last to ZONE_NORMAL.
911 * If we don't have HIGHMEM nor movable node,
912 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
913 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
915 if (N_MEMORY == N_NORMAL_MEMORY)
916 zone_last = ZONE_MOVABLE;
919 * if the memory to be online is in a zone of 0...zone_last, and
920 * the zones of 0...zone_last don't have memory before online, we will
921 * need to set the node to node_states[N_NORMAL_MEMORY] after
922 * the memory is online.
924 if (zone_idx(zone) <= zone_last && !node_state(nid, N_NORMAL_MEMORY))
925 arg->status_change_nid_normal = nid;
927 arg->status_change_nid_normal = -1;
929 #ifdef CONFIG_HIGHMEM
931 * If we have movable node, node_states[N_HIGH_MEMORY]
932 * contains nodes which have zones of 0...ZONE_HIGHMEM,
933 * set zone_last to ZONE_HIGHMEM.
935 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
936 * contains nodes which have zones of 0...ZONE_MOVABLE,
937 * set zone_last to ZONE_MOVABLE.
939 zone_last = ZONE_HIGHMEM;
940 if (N_MEMORY == N_HIGH_MEMORY)
941 zone_last = ZONE_MOVABLE;
943 if (zone_idx(zone) <= zone_last && !node_state(nid, N_HIGH_MEMORY))
944 arg->status_change_nid_high = nid;
946 arg->status_change_nid_high = -1;
948 arg->status_change_nid_high = arg->status_change_nid_normal;
952 * if the node don't have memory befor online, we will need to
953 * set the node to node_states[N_MEMORY] after the memory
956 if (!node_state(nid, N_MEMORY))
957 arg->status_change_nid = nid;
959 arg->status_change_nid = -1;
962 static void node_states_set_node(int node, struct memory_notify *arg)
964 if (arg->status_change_nid_normal >= 0)
965 node_set_state(node, N_NORMAL_MEMORY);
967 if (arg->status_change_nid_high >= 0)
968 node_set_state(node, N_HIGH_MEMORY);
970 node_set_state(node, N_MEMORY);
974 /* Must be protected by mem_hotplug_begin() */
975 int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type)
978 unsigned long onlined_pages = 0;
980 int need_zonelists_rebuild = 0;
983 struct memory_notify arg;
986 * This doesn't need a lock to do pfn_to_page().
987 * The section can't be removed here because of the
988 * memory_block->state_mutex.
990 zone = page_zone(pfn_to_page(pfn));
992 if ((zone_idx(zone) > ZONE_NORMAL ||
993 online_type == MMOP_ONLINE_MOVABLE) &&
994 !can_online_high_movable(zone))
997 if (online_type == MMOP_ONLINE_KERNEL &&
998 zone_idx(zone) == ZONE_MOVABLE) {
999 if (move_pfn_range_left(zone - 1, zone, pfn, pfn + nr_pages))
1002 if (online_type == MMOP_ONLINE_MOVABLE &&
1003 zone_idx(zone) == ZONE_MOVABLE - 1) {
1004 if (move_pfn_range_right(zone, zone + 1, pfn, pfn + nr_pages))
1008 /* Previous code may changed the zone of the pfn range */
1009 zone = page_zone(pfn_to_page(pfn));
1011 arg.start_pfn = pfn;
1012 arg.nr_pages = nr_pages;
1013 node_states_check_changes_online(nr_pages, zone, &arg);
1015 nid = pfn_to_nid(pfn);
1017 ret = memory_notify(MEM_GOING_ONLINE, &arg);
1018 ret = notifier_to_errno(ret);
1020 memory_notify(MEM_CANCEL_ONLINE, &arg);
1024 * If this zone is not populated, then it is not in zonelist.
1025 * This means the page allocator ignores this zone.
1026 * So, zonelist must be updated after online.
1028 mutex_lock(&zonelists_mutex);
1029 if (!populated_zone(zone)) {
1030 need_zonelists_rebuild = 1;
1031 build_all_zonelists(NULL, zone);
1034 ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
1035 online_pages_range);
1037 if (need_zonelists_rebuild)
1038 zone_pcp_reset(zone);
1039 mutex_unlock(&zonelists_mutex);
1040 printk(KERN_DEBUG "online_pages [mem %#010llx-%#010llx] failed\n",
1041 (unsigned long long) pfn << PAGE_SHIFT,
1042 (((unsigned long long) pfn + nr_pages)
1043 << PAGE_SHIFT) - 1);
1044 memory_notify(MEM_CANCEL_ONLINE, &arg);
1048 zone->present_pages += onlined_pages;
1050 pgdat_resize_lock(zone->zone_pgdat, &flags);
1051 zone->zone_pgdat->node_present_pages += onlined_pages;
1052 pgdat_resize_unlock(zone->zone_pgdat, &flags);
1054 if (onlined_pages) {
1055 node_states_set_node(zone_to_nid(zone), &arg);
1056 if (need_zonelists_rebuild)
1057 build_all_zonelists(NULL, NULL);
1059 zone_pcp_update(zone);
1062 mutex_unlock(&zonelists_mutex);
1064 init_per_zone_wmark_min();
1067 kswapd_run(zone_to_nid(zone));
1069 vm_total_pages = nr_free_pagecache_pages();
1071 writeback_set_ratelimit();
1074 memory_notify(MEM_ONLINE, &arg);
1077 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
1079 static void reset_node_present_pages(pg_data_t *pgdat)
1083 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
1084 z->present_pages = 0;
1086 pgdat->node_present_pages = 0;
1089 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1090 static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
1092 struct pglist_data *pgdat;
1093 unsigned long zones_size[MAX_NR_ZONES] = {0};
1094 unsigned long zholes_size[MAX_NR_ZONES] = {0};
1095 unsigned long start_pfn = PFN_DOWN(start);
1097 pgdat = NODE_DATA(nid);
1099 pgdat = arch_alloc_nodedata(nid);
1103 arch_refresh_nodedata(nid, pgdat);
1105 /* Reset the nr_zones and classzone_idx to 0 before reuse */
1106 pgdat->nr_zones = 0;
1107 pgdat->classzone_idx = 0;
1110 /* we can use NODE_DATA(nid) from here */
1112 /* init node's zones as empty zones, we don't have any present pages.*/
1113 free_area_init_node(nid, zones_size, start_pfn, zholes_size);
1116 * The node we allocated has no zone fallback lists. For avoiding
1117 * to access not-initialized zonelist, build here.
1119 mutex_lock(&zonelists_mutex);
1120 build_all_zonelists(pgdat, NULL);
1121 mutex_unlock(&zonelists_mutex);
1124 * zone->managed_pages is set to an approximate value in
1125 * free_area_init_core(), which will cause
1126 * /sys/device/system/node/nodeX/meminfo has wrong data.
1127 * So reset it to 0 before any memory is onlined.
1129 reset_node_managed_pages(pgdat);
1132 * When memory is hot-added, all the memory is in offline state. So
1133 * clear all zones' present_pages because they will be updated in
1134 * online_pages() and offline_pages().
1136 reset_node_present_pages(pgdat);
1141 static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
1143 arch_refresh_nodedata(nid, NULL);
1144 arch_free_nodedata(pgdat);
1150 * try_online_node - online a node if offlined
1152 * called by cpu_up() to online a node without onlined memory.
1154 int try_online_node(int nid)
1159 if (node_online(nid))
1162 mem_hotplug_begin();
1163 pgdat = hotadd_new_pgdat(nid, 0);
1165 pr_err("Cannot online node %d due to NULL pgdat\n", nid);
1169 node_set_online(nid);
1170 ret = register_one_node(nid);
1173 if (pgdat->node_zonelists->_zonerefs->zone == NULL) {
1174 mutex_lock(&zonelists_mutex);
1175 build_all_zonelists(NULL, NULL);
1176 mutex_unlock(&zonelists_mutex);
1184 static int check_hotplug_memory_range(u64 start, u64 size)
1186 u64 start_pfn = PFN_DOWN(start);
1187 u64 nr_pages = size >> PAGE_SHIFT;
1189 /* Memory range must be aligned with section */
1190 if ((start_pfn & ~PAGE_SECTION_MASK) ||
1191 (nr_pages % PAGES_PER_SECTION) || (!nr_pages)) {
1192 pr_err("Section-unaligned hotplug range: start 0x%llx, size 0x%llx\n",
1193 (unsigned long long)start,
1194 (unsigned long long)size);
1202 * If movable zone has already been setup, newly added memory should be check.
1203 * If its address is higher than movable zone, it should be added as movable.
1204 * Without this check, movable zone may overlap with other zone.
1206 static int should_add_memory_movable(int nid, u64 start, u64 size)
1208 unsigned long start_pfn = start >> PAGE_SHIFT;
1209 pg_data_t *pgdat = NODE_DATA(nid);
1210 struct zone *movable_zone = pgdat->node_zones + ZONE_MOVABLE;
1212 if (zone_is_empty(movable_zone))
1215 if (movable_zone->zone_start_pfn <= start_pfn)
1221 int zone_for_memory(int nid, u64 start, u64 size, int zone_default,
1224 #ifdef CONFIG_ZONE_DEVICE
1228 if (should_add_memory_movable(nid, start, size))
1229 return ZONE_MOVABLE;
1231 return zone_default;
1234 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1235 int __ref add_memory_resource(int nid, struct resource *res)
1238 pg_data_t *pgdat = NULL;
1244 size = resource_size(res);
1246 ret = check_hotplug_memory_range(start, size);
1250 { /* Stupid hack to suppress address-never-null warning */
1251 void *p = NODE_DATA(nid);
1255 mem_hotplug_begin();
1258 * Add new range to memblock so that when hotadd_new_pgdat() is called
1259 * to allocate new pgdat, get_pfn_range_for_nid() will be able to find
1260 * this new range and calculate total pages correctly. The range will
1261 * be removed at hot-remove time.
1263 memblock_add_node(start, size, nid);
1265 new_node = !node_online(nid);
1267 pgdat = hotadd_new_pgdat(nid, start);
1273 /* call arch's memory hotadd */
1274 ret = arch_add_memory(nid, start, size, false);
1279 /* we online node here. we can't roll back from here. */
1280 node_set_online(nid);
1283 ret = register_one_node(nid);
1285 * If sysfs file of new node can't create, cpu on the node
1286 * can't be hot-added. There is no rollback way now.
1287 * So, check by BUG_ON() to catch it reluctantly..
1292 /* create new memmap entry */
1293 firmware_map_add_hotplug(start, start + size, "System RAM");
1298 /* rollback pgdat allocation and others */
1300 rollback_node_hotadd(nid, pgdat);
1301 memblock_remove(start, size);
1307 EXPORT_SYMBOL_GPL(add_memory_resource);
1309 int __ref add_memory(int nid, u64 start, u64 size)
1311 struct resource *res;
1314 res = register_memory_resource(start, size);
1318 ret = add_memory_resource(nid, res);
1320 release_memory_resource(res);
1323 EXPORT_SYMBOL_GPL(add_memory);
1325 #ifdef CONFIG_MEMORY_HOTREMOVE
1327 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1328 * set and the size of the free page is given by page_order(). Using this,
1329 * the function determines if the pageblock contains only free pages.
1330 * Due to buddy contraints, a free page at least the size of a pageblock will
1331 * be located at the start of the pageblock
1333 static inline int pageblock_free(struct page *page)
1335 return PageBuddy(page) && page_order(page) >= pageblock_order;
1338 /* Return the start of the next active pageblock after a given page */
1339 static struct page *next_active_pageblock(struct page *page)
1341 /* Ensure the starting page is pageblock-aligned */
1342 BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
1344 /* If the entire pageblock is free, move to the end of free page */
1345 if (pageblock_free(page)) {
1347 /* be careful. we don't have locks, page_order can be changed.*/
1348 order = page_order(page);
1349 if ((order < MAX_ORDER) && (order >= pageblock_order))
1350 return page + (1 << order);
1353 return page + pageblock_nr_pages;
1356 /* Checks if this range of memory is likely to be hot-removable. */
1357 int is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
1359 struct page *page = pfn_to_page(start_pfn);
1360 struct page *end_page = page + nr_pages;
1362 /* Check the starting page of each pageblock within the range */
1363 for (; page < end_page; page = next_active_pageblock(page)) {
1364 if (!is_pageblock_removable_nolock(page))
1369 /* All pageblocks in the memory block are likely to be hot-removable */
1374 * Confirm all pages in a range [start, end) is belongs to the same zone.
1376 int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn)
1379 struct zone *zone = NULL;
1382 for (pfn = start_pfn;
1384 pfn += MAX_ORDER_NR_PAGES) {
1386 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1387 while ((i < MAX_ORDER_NR_PAGES) && !pfn_valid_within(pfn + i))
1389 if (i == MAX_ORDER_NR_PAGES)
1391 page = pfn_to_page(pfn + i);
1392 if (zone && page_zone(page) != zone)
1394 zone = page_zone(page);
1400 * Scan pfn range [start,end) to find movable/migratable pages (LRU pages
1401 * and hugepages). We scan pfn because it's much easier than scanning over
1402 * linked list. This function returns the pfn of the first found movable
1403 * page if it's found, otherwise 0.
1405 static unsigned long scan_movable_pages(unsigned long start, unsigned long end)
1409 for (pfn = start; pfn < end; pfn++) {
1410 if (pfn_valid(pfn)) {
1411 page = pfn_to_page(pfn);
1414 if (PageHuge(page)) {
1415 if (page_huge_active(page))
1418 pfn = round_up(pfn + 1,
1419 1 << compound_order(page)) - 1;
1426 #define NR_OFFLINE_AT_ONCE_PAGES (256)
1428 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1432 int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
1433 int not_managed = 0;
1437 for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
1438 if (!pfn_valid(pfn))
1440 page = pfn_to_page(pfn);
1442 if (PageHuge(page)) {
1443 struct page *head = compound_head(page);
1444 pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1;
1445 if (compound_order(head) > PFN_SECTION_SHIFT) {
1449 if (isolate_huge_page(page, &source))
1450 move_pages -= 1 << compound_order(head);
1454 if (!get_page_unless_zero(page))
1457 * We can skip free pages. And we can only deal with pages on
1460 ret = isolate_lru_page(page);
1461 if (!ret) { /* Success */
1463 list_add_tail(&page->lru, &source);
1465 inc_zone_page_state(page, NR_ISOLATED_ANON +
1466 page_is_file_cache(page));
1469 #ifdef CONFIG_DEBUG_VM
1470 printk(KERN_ALERT "removing pfn %lx from LRU failed\n",
1472 dump_page(page, "failed to remove from LRU");
1475 /* Because we don't have big zone->lock. we should
1476 check this again here. */
1477 if (page_count(page)) {
1484 if (!list_empty(&source)) {
1486 putback_movable_pages(&source);
1491 * alloc_migrate_target should be improooooved!!
1492 * migrate_pages returns # of failed pages.
1494 ret = migrate_pages(&source, alloc_migrate_target, NULL, 0,
1495 MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
1497 putback_movable_pages(&source);
1504 * remove from free_area[] and mark all as Reserved.
1507 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1510 __offline_isolated_pages(start, start + nr_pages);
1515 offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
1517 walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
1518 offline_isolated_pages_cb);
1522 * Check all pages in range, recoreded as memory resource, are isolated.
1525 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1529 long offlined = *(long *)data;
1530 ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true);
1531 offlined = nr_pages;
1533 *(long *)data += offlined;
1538 check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
1543 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
1544 check_pages_isolated_cb);
1546 offlined = (long)ret;
1550 #ifdef CONFIG_MOVABLE_NODE
1552 * When CONFIG_MOVABLE_NODE, we permit offlining of a node which doesn't have
1555 static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1559 #else /* CONFIG_MOVABLE_NODE */
1560 /* ensure the node has NORMAL memory if it is still online */
1561 static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1563 struct pglist_data *pgdat = zone->zone_pgdat;
1564 unsigned long present_pages = 0;
1567 for (zt = 0; zt <= ZONE_NORMAL; zt++)
1568 present_pages += pgdat->node_zones[zt].present_pages;
1570 if (present_pages > nr_pages)
1574 for (; zt <= ZONE_MOVABLE; zt++)
1575 present_pages += pgdat->node_zones[zt].present_pages;
1578 * we can't offline the last normal memory until all
1579 * higher memory is offlined.
1581 return present_pages == 0;
1583 #endif /* CONFIG_MOVABLE_NODE */
1585 static int __init cmdline_parse_movable_node(char *p)
1587 #ifdef CONFIG_MOVABLE_NODE
1589 * Memory used by the kernel cannot be hot-removed because Linux
1590 * cannot migrate the kernel pages. When memory hotplug is
1591 * enabled, we should prevent memblock from allocating memory
1594 * ACPI SRAT records all hotpluggable memory ranges. But before
1595 * SRAT is parsed, we don't know about it.
1597 * The kernel image is loaded into memory at very early time. We
1598 * cannot prevent this anyway. So on NUMA system, we set any
1599 * node the kernel resides in as un-hotpluggable.
1601 * Since on modern servers, one node could have double-digit
1602 * gigabytes memory, we can assume the memory around the kernel
1603 * image is also un-hotpluggable. So before SRAT is parsed, just
1604 * allocate memory near the kernel image to try the best to keep
1605 * the kernel away from hotpluggable memory.
1607 memblock_set_bottom_up(true);
1608 movable_node_enabled = true;
1610 pr_warn("movable_node option not supported\n");
1614 early_param("movable_node", cmdline_parse_movable_node);
1616 /* check which state of node_states will be changed when offline memory */
1617 static void node_states_check_changes_offline(unsigned long nr_pages,
1618 struct zone *zone, struct memory_notify *arg)
1620 struct pglist_data *pgdat = zone->zone_pgdat;
1621 unsigned long present_pages = 0;
1622 enum zone_type zt, zone_last = ZONE_NORMAL;
1625 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
1626 * contains nodes which have zones of 0...ZONE_NORMAL,
1627 * set zone_last to ZONE_NORMAL.
1629 * If we don't have HIGHMEM nor movable node,
1630 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
1631 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
1633 if (N_MEMORY == N_NORMAL_MEMORY)
1634 zone_last = ZONE_MOVABLE;
1637 * check whether node_states[N_NORMAL_MEMORY] will be changed.
1638 * If the memory to be offline is in a zone of 0...zone_last,
1639 * and it is the last present memory, 0...zone_last will
1640 * become empty after offline , thus we can determind we will
1641 * need to clear the node from node_states[N_NORMAL_MEMORY].
1643 for (zt = 0; zt <= zone_last; zt++)
1644 present_pages += pgdat->node_zones[zt].present_pages;
1645 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1646 arg->status_change_nid_normal = zone_to_nid(zone);
1648 arg->status_change_nid_normal = -1;
1650 #ifdef CONFIG_HIGHMEM
1652 * If we have movable node, node_states[N_HIGH_MEMORY]
1653 * contains nodes which have zones of 0...ZONE_HIGHMEM,
1654 * set zone_last to ZONE_HIGHMEM.
1656 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
1657 * contains nodes which have zones of 0...ZONE_MOVABLE,
1658 * set zone_last to ZONE_MOVABLE.
1660 zone_last = ZONE_HIGHMEM;
1661 if (N_MEMORY == N_HIGH_MEMORY)
1662 zone_last = ZONE_MOVABLE;
1664 for (; zt <= zone_last; zt++)
1665 present_pages += pgdat->node_zones[zt].present_pages;
1666 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1667 arg->status_change_nid_high = zone_to_nid(zone);
1669 arg->status_change_nid_high = -1;
1671 arg->status_change_nid_high = arg->status_change_nid_normal;
1675 * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE
1677 zone_last = ZONE_MOVABLE;
1680 * check whether node_states[N_HIGH_MEMORY] will be changed
1681 * If we try to offline the last present @nr_pages from the node,
1682 * we can determind we will need to clear the node from
1683 * node_states[N_HIGH_MEMORY].
1685 for (; zt <= zone_last; zt++)
1686 present_pages += pgdat->node_zones[zt].present_pages;
1687 if (nr_pages >= present_pages)
1688 arg->status_change_nid = zone_to_nid(zone);
1690 arg->status_change_nid = -1;
1693 static void node_states_clear_node(int node, struct memory_notify *arg)
1695 if (arg->status_change_nid_normal >= 0)
1696 node_clear_state(node, N_NORMAL_MEMORY);
1698 if ((N_MEMORY != N_NORMAL_MEMORY) &&
1699 (arg->status_change_nid_high >= 0))
1700 node_clear_state(node, N_HIGH_MEMORY);
1702 if ((N_MEMORY != N_HIGH_MEMORY) &&
1703 (arg->status_change_nid >= 0))
1704 node_clear_state(node, N_MEMORY);
1707 static int __ref __offline_pages(unsigned long start_pfn,
1708 unsigned long end_pfn, unsigned long timeout)
1710 unsigned long pfn, nr_pages, expire;
1711 long offlined_pages;
1712 int ret, drain, retry_max, node;
1713 unsigned long flags;
1715 struct memory_notify arg;
1717 /* at least, alignment against pageblock is necessary */
1718 if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
1720 if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
1722 /* This makes hotplug much easier...and readable.
1723 we assume this for now. .*/
1724 if (!test_pages_in_a_zone(start_pfn, end_pfn))
1727 zone = page_zone(pfn_to_page(start_pfn));
1728 node = zone_to_nid(zone);
1729 nr_pages = end_pfn - start_pfn;
1731 if (zone_idx(zone) <= ZONE_NORMAL && !can_offline_normal(zone, nr_pages))
1734 /* set above range as isolated */
1735 ret = start_isolate_page_range(start_pfn, end_pfn,
1736 MIGRATE_MOVABLE, true);
1740 arg.start_pfn = start_pfn;
1741 arg.nr_pages = nr_pages;
1742 node_states_check_changes_offline(nr_pages, zone, &arg);
1744 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1745 ret = notifier_to_errno(ret);
1747 goto failed_removal;
1750 expire = jiffies + timeout;
1754 /* start memory hot removal */
1756 if (time_after(jiffies, expire))
1757 goto failed_removal;
1759 if (signal_pending(current))
1760 goto failed_removal;
1763 lru_add_drain_all();
1765 drain_all_pages(zone);
1768 pfn = scan_movable_pages(start_pfn, end_pfn);
1769 if (pfn) { /* We have movable pages */
1770 ret = do_migrate_range(pfn, end_pfn);
1776 if (--retry_max == 0)
1777 goto failed_removal;
1783 /* drain all zone's lru pagevec, this is asynchronous... */
1784 lru_add_drain_all();
1786 /* drain pcp pages, this is synchronous. */
1787 drain_all_pages(zone);
1789 * dissolve free hugepages in the memory block before doing offlining
1790 * actually in order to make hugetlbfs's object counting consistent.
1792 dissolve_free_huge_pages(start_pfn, end_pfn);
1794 offlined_pages = check_pages_isolated(start_pfn, end_pfn);
1795 if (offlined_pages < 0) {
1797 goto failed_removal;
1799 printk(KERN_INFO "Offlined Pages %ld\n", offlined_pages);
1800 /* Ok, all of our target is isolated.
1801 We cannot do rollback at this point. */
1802 offline_isolated_pages(start_pfn, end_pfn);
1803 /* reset pagetype flags and makes migrate type to be MOVABLE */
1804 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1805 /* removal success */
1806 adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages);
1807 zone->present_pages -= offlined_pages;
1809 pgdat_resize_lock(zone->zone_pgdat, &flags);
1810 zone->zone_pgdat->node_present_pages -= offlined_pages;
1811 pgdat_resize_unlock(zone->zone_pgdat, &flags);
1813 init_per_zone_wmark_min();
1815 if (!populated_zone(zone)) {
1816 zone_pcp_reset(zone);
1817 mutex_lock(&zonelists_mutex);
1818 build_all_zonelists(NULL, NULL);
1819 mutex_unlock(&zonelists_mutex);
1821 zone_pcp_update(zone);
1823 node_states_clear_node(node, &arg);
1824 if (arg.status_change_nid >= 0)
1827 vm_total_pages = nr_free_pagecache_pages();
1828 writeback_set_ratelimit();
1830 memory_notify(MEM_OFFLINE, &arg);
1834 printk(KERN_INFO "memory offlining [mem %#010llx-%#010llx] failed\n",
1835 (unsigned long long) start_pfn << PAGE_SHIFT,
1836 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
1837 memory_notify(MEM_CANCEL_OFFLINE, &arg);
1838 /* pushback to free area */
1839 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1843 /* Must be protected by mem_hotplug_begin() */
1844 int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1846 return __offline_pages(start_pfn, start_pfn + nr_pages, 120 * HZ);
1848 #endif /* CONFIG_MEMORY_HOTREMOVE */
1851 * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
1852 * @start_pfn: start pfn of the memory range
1853 * @end_pfn: end pfn of the memory range
1854 * @arg: argument passed to func
1855 * @func: callback for each memory section walked
1857 * This function walks through all present mem sections in range
1858 * [start_pfn, end_pfn) and call func on each mem section.
1860 * Returns the return value of func.
1862 int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn,
1863 void *arg, int (*func)(struct memory_block *, void *))
1865 struct memory_block *mem = NULL;
1866 struct mem_section *section;
1867 unsigned long pfn, section_nr;
1870 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1871 section_nr = pfn_to_section_nr(pfn);
1872 if (!present_section_nr(section_nr))
1875 section = __nr_to_section(section_nr);
1876 /* same memblock? */
1878 if ((section_nr >= mem->start_section_nr) &&
1879 (section_nr <= mem->end_section_nr))
1882 mem = find_memory_block_hinted(section, mem);
1886 ret = func(mem, arg);
1888 kobject_put(&mem->dev.kobj);
1894 kobject_put(&mem->dev.kobj);
1899 #ifdef CONFIG_MEMORY_HOTREMOVE
1900 static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
1902 int ret = !is_memblock_offlined(mem);
1904 if (unlikely(ret)) {
1905 phys_addr_t beginpa, endpa;
1907 beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
1908 endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1;
1909 pr_warn("removing memory fails, because memory "
1910 "[%pa-%pa] is onlined\n",
1917 static int check_cpu_on_node(pg_data_t *pgdat)
1921 for_each_present_cpu(cpu) {
1922 if (cpu_to_node(cpu) == pgdat->node_id)
1924 * the cpu on this node isn't removed, and we can't
1925 * offline this node.
1933 static void unmap_cpu_on_node(pg_data_t *pgdat)
1935 #ifdef CONFIG_ACPI_NUMA
1938 for_each_possible_cpu(cpu)
1939 if (cpu_to_node(cpu) == pgdat->node_id)
1940 numa_clear_node(cpu);
1944 static int check_and_unmap_cpu_on_node(pg_data_t *pgdat)
1948 ret = check_cpu_on_node(pgdat);
1953 * the node will be offlined when we come here, so we can clear
1954 * the cpu_to_node() now.
1957 unmap_cpu_on_node(pgdat);
1964 * Offline a node if all memory sections and cpus of the node are removed.
1966 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1967 * and online/offline operations before this call.
1969 void try_offline_node(int nid)
1971 pg_data_t *pgdat = NODE_DATA(nid);
1972 unsigned long start_pfn = pgdat->node_start_pfn;
1973 unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
1977 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1978 unsigned long section_nr = pfn_to_section_nr(pfn);
1980 if (!present_section_nr(section_nr))
1983 if (pfn_to_nid(pfn) != nid)
1987 * some memory sections of this node are not removed, and we
1988 * can't offline node now.
1993 if (check_and_unmap_cpu_on_node(pgdat))
1997 * all memory/cpu of this node are removed, we can offline this
2000 node_set_offline(nid);
2001 unregister_one_node(nid);
2003 /* free waittable in each zone */
2004 for (i = 0; i < MAX_NR_ZONES; i++) {
2005 struct zone *zone = pgdat->node_zones + i;
2008 * wait_table may be allocated from boot memory,
2009 * here only free if it's allocated by vmalloc.
2011 if (is_vmalloc_addr(zone->wait_table)) {
2012 vfree(zone->wait_table);
2013 zone->wait_table = NULL;
2017 EXPORT_SYMBOL(try_offline_node);
2022 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
2023 * and online/offline operations before this call, as required by
2024 * try_offline_node().
2026 void __ref remove_memory(int nid, u64 start, u64 size)
2030 BUG_ON(check_hotplug_memory_range(start, size));
2032 mem_hotplug_begin();
2035 * All memory blocks must be offlined before removing memory. Check
2036 * whether all memory blocks in question are offline and trigger a BUG()
2037 * if this is not the case.
2039 ret = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL,
2040 check_memblock_offlined_cb);
2044 /* remove memmap entry */
2045 firmware_map_remove(start, start + size, "System RAM");
2046 memblock_free(start, size);
2047 memblock_remove(start, size);
2049 arch_remove_memory(start, size);
2051 try_offline_node(nid);
2055 EXPORT_SYMBOL_GPL(remove_memory);
2056 #endif /* CONFIG_MEMORY_HOTREMOVE */