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>
35 #include <asm/tlbflush.h>
40 * online_page_callback contains pointer to current page onlining function.
41 * Initially it is generic_online_page(). If it is required it could be
42 * changed by calling set_online_page_callback() for callback registration
43 * and restore_online_page_callback() for generic callback restore.
46 static void generic_online_page(struct page *page);
48 static online_page_callback_t online_page_callback = generic_online_page;
49 static DEFINE_MUTEX(online_page_callback_lock);
51 /* The same as the cpu_hotplug lock, but for memory hotplug. */
53 struct task_struct *active_writer;
54 struct mutex lock; /* Synchronizes accesses to refcount, */
56 * Also blocks the new readers during
57 * an ongoing mem hotplug operation.
61 #ifdef CONFIG_DEBUG_LOCK_ALLOC
62 struct lockdep_map dep_map;
65 .active_writer = NULL,
66 .lock = __MUTEX_INITIALIZER(mem_hotplug.lock),
68 #ifdef CONFIG_DEBUG_LOCK_ALLOC
69 .dep_map = {.name = "mem_hotplug.lock" },
73 /* Lockdep annotations for get/put_online_mems() and mem_hotplug_begin/end() */
74 #define memhp_lock_acquire_read() lock_map_acquire_read(&mem_hotplug.dep_map)
75 #define memhp_lock_acquire() lock_map_acquire(&mem_hotplug.dep_map)
76 #define memhp_lock_release() lock_map_release(&mem_hotplug.dep_map)
78 void get_online_mems(void)
81 if (mem_hotplug.active_writer == current)
83 memhp_lock_acquire_read();
84 mutex_lock(&mem_hotplug.lock);
85 mem_hotplug.refcount++;
86 mutex_unlock(&mem_hotplug.lock);
90 void put_online_mems(void)
92 if (mem_hotplug.active_writer == current)
94 mutex_lock(&mem_hotplug.lock);
96 if (WARN_ON(!mem_hotplug.refcount))
97 mem_hotplug.refcount++; /* try to fix things up */
99 if (!--mem_hotplug.refcount && unlikely(mem_hotplug.active_writer))
100 wake_up_process(mem_hotplug.active_writer);
101 mutex_unlock(&mem_hotplug.lock);
102 memhp_lock_release();
106 static void mem_hotplug_begin(void)
108 mem_hotplug.active_writer = current;
110 memhp_lock_acquire();
112 mutex_lock(&mem_hotplug.lock);
113 if (likely(!mem_hotplug.refcount))
115 __set_current_state(TASK_UNINTERRUPTIBLE);
116 mutex_unlock(&mem_hotplug.lock);
121 static void mem_hotplug_done(void)
123 mem_hotplug.active_writer = NULL;
124 mutex_unlock(&mem_hotplug.lock);
125 memhp_lock_release();
128 /* add this memory to iomem resource */
129 static struct resource *register_memory_resource(u64 start, u64 size)
131 struct resource *res;
132 res = kzalloc(sizeof(struct resource), GFP_KERNEL);
135 res->name = "System RAM";
137 res->end = start + size - 1;
138 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
139 if (request_resource(&iomem_resource, res) < 0) {
140 pr_debug("System RAM resource %pR cannot be added\n", res);
147 static void release_memory_resource(struct resource *res)
151 release_resource(res);
156 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
157 void get_page_bootmem(unsigned long info, struct page *page,
160 page->lru.next = (struct list_head *) type;
161 SetPagePrivate(page);
162 set_page_private(page, info);
163 atomic_inc(&page->_count);
166 void put_page_bootmem(struct page *page)
170 type = (unsigned long) page->lru.next;
171 BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE ||
172 type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE);
174 if (atomic_dec_return(&page->_count) == 1) {
175 ClearPagePrivate(page);
176 set_page_private(page, 0);
177 INIT_LIST_HEAD(&page->lru);
178 free_reserved_page(page);
182 #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
183 #ifndef CONFIG_SPARSEMEM_VMEMMAP
184 static void register_page_bootmem_info_section(unsigned long start_pfn)
186 unsigned long *usemap, mapsize, section_nr, i;
187 struct mem_section *ms;
188 struct page *page, *memmap;
190 section_nr = pfn_to_section_nr(start_pfn);
191 ms = __nr_to_section(section_nr);
193 /* Get section's memmap address */
194 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
197 * Get page for the memmap's phys address
198 * XXX: need more consideration for sparse_vmemmap...
200 page = virt_to_page(memmap);
201 mapsize = sizeof(struct page) * PAGES_PER_SECTION;
202 mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
204 /* remember memmap's page */
205 for (i = 0; i < mapsize; i++, page++)
206 get_page_bootmem(section_nr, page, SECTION_INFO);
208 usemap = __nr_to_section(section_nr)->pageblock_flags;
209 page = virt_to_page(usemap);
211 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
213 for (i = 0; i < mapsize; i++, page++)
214 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
217 #else /* CONFIG_SPARSEMEM_VMEMMAP */
218 static void register_page_bootmem_info_section(unsigned long start_pfn)
220 unsigned long *usemap, mapsize, section_nr, i;
221 struct mem_section *ms;
222 struct page *page, *memmap;
224 if (!pfn_valid(start_pfn))
227 section_nr = pfn_to_section_nr(start_pfn);
228 ms = __nr_to_section(section_nr);
230 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
232 register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION);
234 usemap = __nr_to_section(section_nr)->pageblock_flags;
235 page = virt_to_page(usemap);
237 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
239 for (i = 0; i < mapsize; i++, page++)
240 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
242 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
244 void register_page_bootmem_info_node(struct pglist_data *pgdat)
246 unsigned long i, pfn, end_pfn, nr_pages;
247 int node = pgdat->node_id;
251 nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
252 page = virt_to_page(pgdat);
254 for (i = 0; i < nr_pages; i++, page++)
255 get_page_bootmem(node, page, NODE_INFO);
257 zone = &pgdat->node_zones[0];
258 for (; zone < pgdat->node_zones + MAX_NR_ZONES - 1; zone++) {
259 if (zone_is_initialized(zone)) {
260 nr_pages = zone->wait_table_hash_nr_entries
261 * sizeof(wait_queue_head_t);
262 nr_pages = PAGE_ALIGN(nr_pages) >> PAGE_SHIFT;
263 page = virt_to_page(zone->wait_table);
265 for (i = 0; i < nr_pages; i++, page++)
266 get_page_bootmem(node, page, NODE_INFO);
270 pfn = pgdat->node_start_pfn;
271 end_pfn = pgdat_end_pfn(pgdat);
273 /* register section info */
274 for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
276 * Some platforms can assign the same pfn to multiple nodes - on
277 * node0 as well as nodeN. To avoid registering a pfn against
278 * multiple nodes we check that this pfn does not already
279 * reside in some other nodes.
281 if (pfn_valid(pfn) && (pfn_to_nid(pfn) == node))
282 register_page_bootmem_info_section(pfn);
285 #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
287 static void __meminit grow_zone_span(struct zone *zone, unsigned long start_pfn,
288 unsigned long end_pfn)
290 unsigned long old_zone_end_pfn;
292 zone_span_writelock(zone);
294 old_zone_end_pfn = zone_end_pfn(zone);
295 if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
296 zone->zone_start_pfn = start_pfn;
298 zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
299 zone->zone_start_pfn;
301 zone_span_writeunlock(zone);
304 static void resize_zone(struct zone *zone, unsigned long start_pfn,
305 unsigned long end_pfn)
307 zone_span_writelock(zone);
309 if (end_pfn - start_pfn) {
310 zone->zone_start_pfn = start_pfn;
311 zone->spanned_pages = end_pfn - start_pfn;
314 * make it consist as free_area_init_core(),
315 * if spanned_pages = 0, then keep start_pfn = 0
317 zone->zone_start_pfn = 0;
318 zone->spanned_pages = 0;
321 zone_span_writeunlock(zone);
324 static void fix_zone_id(struct zone *zone, unsigned long start_pfn,
325 unsigned long end_pfn)
327 enum zone_type zid = zone_idx(zone);
328 int nid = zone->zone_pgdat->node_id;
331 for (pfn = start_pfn; pfn < end_pfn; pfn++)
332 set_page_links(pfn_to_page(pfn), zid, nid, pfn);
335 /* Can fail with -ENOMEM from allocating a wait table with vmalloc() or
336 * alloc_bootmem_node_nopanic()/memblock_virt_alloc_node_nopanic() */
337 static int __ref ensure_zone_is_initialized(struct zone *zone,
338 unsigned long start_pfn, unsigned long num_pages)
340 if (!zone_is_initialized(zone))
341 return init_currently_empty_zone(zone, start_pfn, num_pages,
346 static int __meminit move_pfn_range_left(struct zone *z1, struct zone *z2,
347 unsigned long start_pfn, unsigned long end_pfn)
351 unsigned long z1_start_pfn;
353 ret = ensure_zone_is_initialized(z1, start_pfn, end_pfn - start_pfn);
357 pgdat_resize_lock(z1->zone_pgdat, &flags);
359 /* can't move pfns which are higher than @z2 */
360 if (end_pfn > zone_end_pfn(z2))
362 /* the move out part must be at the left most of @z2 */
363 if (start_pfn > z2->zone_start_pfn)
365 /* must included/overlap */
366 if (end_pfn <= z2->zone_start_pfn)
369 /* use start_pfn for z1's start_pfn if z1 is empty */
370 if (!zone_is_empty(z1))
371 z1_start_pfn = z1->zone_start_pfn;
373 z1_start_pfn = start_pfn;
375 resize_zone(z1, z1_start_pfn, end_pfn);
376 resize_zone(z2, end_pfn, zone_end_pfn(z2));
378 pgdat_resize_unlock(z1->zone_pgdat, &flags);
380 fix_zone_id(z1, start_pfn, end_pfn);
384 pgdat_resize_unlock(z1->zone_pgdat, &flags);
388 static int __meminit move_pfn_range_right(struct zone *z1, struct zone *z2,
389 unsigned long start_pfn, unsigned long end_pfn)
393 unsigned long z2_end_pfn;
395 ret = ensure_zone_is_initialized(z2, start_pfn, end_pfn - start_pfn);
399 pgdat_resize_lock(z1->zone_pgdat, &flags);
401 /* can't move pfns which are lower than @z1 */
402 if (z1->zone_start_pfn > start_pfn)
404 /* the move out part mast at the right most of @z1 */
405 if (zone_end_pfn(z1) > end_pfn)
407 /* must included/overlap */
408 if (start_pfn >= zone_end_pfn(z1))
411 /* use end_pfn for z2's end_pfn if z2 is empty */
412 if (!zone_is_empty(z2))
413 z2_end_pfn = zone_end_pfn(z2);
415 z2_end_pfn = end_pfn;
417 resize_zone(z1, z1->zone_start_pfn, start_pfn);
418 resize_zone(z2, start_pfn, z2_end_pfn);
420 pgdat_resize_unlock(z1->zone_pgdat, &flags);
422 fix_zone_id(z2, start_pfn, end_pfn);
426 pgdat_resize_unlock(z1->zone_pgdat, &flags);
430 static void __meminit grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn,
431 unsigned long end_pfn)
433 unsigned long old_pgdat_end_pfn = pgdat_end_pfn(pgdat);
435 if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
436 pgdat->node_start_pfn = start_pfn;
438 pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
439 pgdat->node_start_pfn;
442 static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn)
444 struct pglist_data *pgdat = zone->zone_pgdat;
445 int nr_pages = PAGES_PER_SECTION;
446 int nid = pgdat->node_id;
451 zone_type = zone - pgdat->node_zones;
452 ret = ensure_zone_is_initialized(zone, phys_start_pfn, nr_pages);
456 pgdat_resize_lock(zone->zone_pgdat, &flags);
457 grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages);
458 grow_pgdat_span(zone->zone_pgdat, phys_start_pfn,
459 phys_start_pfn + nr_pages);
460 pgdat_resize_unlock(zone->zone_pgdat, &flags);
461 memmap_init_zone(nr_pages, nid, zone_type,
462 phys_start_pfn, MEMMAP_HOTPLUG);
466 static int __meminit __add_section(int nid, struct zone *zone,
467 unsigned long phys_start_pfn)
471 if (pfn_valid(phys_start_pfn))
474 ret = sparse_add_one_section(zone, phys_start_pfn);
479 ret = __add_zone(zone, phys_start_pfn);
484 return register_new_memory(nid, __pfn_to_section(phys_start_pfn));
488 * Reasonably generic function for adding memory. It is
489 * expected that archs that support memory hotplug will
490 * call this function after deciding the zone to which to
493 int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn,
494 unsigned long nr_pages)
498 int start_sec, end_sec;
499 /* during initialize mem_map, align hot-added range to section */
500 start_sec = pfn_to_section_nr(phys_start_pfn);
501 end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
503 for (i = start_sec; i <= end_sec; i++) {
504 err = __add_section(nid, zone, i << PFN_SECTION_SHIFT);
507 * EEXIST is finally dealt with by ioresource collision
508 * check. see add_memory() => register_memory_resource()
509 * Warning will be printed if there is collision.
511 if (err && (err != -EEXIST))
518 EXPORT_SYMBOL_GPL(__add_pages);
520 #ifdef CONFIG_MEMORY_HOTREMOVE
521 /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
522 static int find_smallest_section_pfn(int nid, struct zone *zone,
523 unsigned long start_pfn,
524 unsigned long end_pfn)
526 struct mem_section *ms;
528 for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) {
529 ms = __pfn_to_section(start_pfn);
531 if (unlikely(!valid_section(ms)))
534 if (unlikely(pfn_to_nid(start_pfn) != nid))
537 if (zone && zone != page_zone(pfn_to_page(start_pfn)))
546 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
547 static int find_biggest_section_pfn(int nid, struct zone *zone,
548 unsigned long start_pfn,
549 unsigned long end_pfn)
551 struct mem_section *ms;
554 /* pfn is the end pfn of a memory section. */
556 for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) {
557 ms = __pfn_to_section(pfn);
559 if (unlikely(!valid_section(ms)))
562 if (unlikely(pfn_to_nid(pfn) != nid))
565 if (zone && zone != page_zone(pfn_to_page(pfn)))
574 static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
575 unsigned long end_pfn)
577 unsigned long zone_start_pfn = zone->zone_start_pfn;
578 unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */
579 unsigned long zone_end_pfn = z;
581 struct mem_section *ms;
582 int nid = zone_to_nid(zone);
584 zone_span_writelock(zone);
585 if (zone_start_pfn == start_pfn) {
587 * If the section is smallest section in the zone, it need
588 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
589 * In this case, we find second smallest valid mem_section
590 * for shrinking zone.
592 pfn = find_smallest_section_pfn(nid, zone, end_pfn,
595 zone->zone_start_pfn = pfn;
596 zone->spanned_pages = zone_end_pfn - pfn;
598 } else if (zone_end_pfn == end_pfn) {
600 * If the section is biggest section in the zone, it need
601 * shrink zone->spanned_pages.
602 * In this case, we find second biggest valid mem_section for
605 pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn,
608 zone->spanned_pages = pfn - zone_start_pfn + 1;
612 * The section is not biggest or smallest mem_section in the zone, it
613 * only creates a hole in the zone. So in this case, we need not
614 * change the zone. But perhaps, the zone has only hole data. Thus
615 * it check the zone has only hole or not.
617 pfn = zone_start_pfn;
618 for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) {
619 ms = __pfn_to_section(pfn);
621 if (unlikely(!valid_section(ms)))
624 if (page_zone(pfn_to_page(pfn)) != zone)
627 /* If the section is current section, it continues the loop */
628 if (start_pfn == pfn)
631 /* If we find valid section, we have nothing to do */
632 zone_span_writeunlock(zone);
636 /* The zone has no valid section */
637 zone->zone_start_pfn = 0;
638 zone->spanned_pages = 0;
639 zone_span_writeunlock(zone);
642 static void shrink_pgdat_span(struct pglist_data *pgdat,
643 unsigned long start_pfn, unsigned long end_pfn)
645 unsigned long pgdat_start_pfn = pgdat->node_start_pfn;
646 unsigned long p = pgdat_end_pfn(pgdat); /* pgdat_end_pfn namespace clash */
647 unsigned long pgdat_end_pfn = p;
649 struct mem_section *ms;
650 int nid = pgdat->node_id;
652 if (pgdat_start_pfn == start_pfn) {
654 * If the section is smallest section in the pgdat, it need
655 * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages.
656 * In this case, we find second smallest valid mem_section
657 * for shrinking zone.
659 pfn = find_smallest_section_pfn(nid, NULL, end_pfn,
662 pgdat->node_start_pfn = pfn;
663 pgdat->node_spanned_pages = pgdat_end_pfn - pfn;
665 } else if (pgdat_end_pfn == end_pfn) {
667 * If the section is biggest section in the pgdat, it need
668 * shrink pgdat->node_spanned_pages.
669 * In this case, we find second biggest valid mem_section for
672 pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn,
675 pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1;
679 * If the section is not biggest or smallest mem_section in the pgdat,
680 * it only creates a hole in the pgdat. So in this case, we need not
682 * But perhaps, the pgdat has only hole data. Thus it check the pgdat
683 * has only hole or not.
685 pfn = pgdat_start_pfn;
686 for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) {
687 ms = __pfn_to_section(pfn);
689 if (unlikely(!valid_section(ms)))
692 if (pfn_to_nid(pfn) != nid)
695 /* If the section is current section, it continues the loop */
696 if (start_pfn == pfn)
699 /* If we find valid section, we have nothing to do */
703 /* The pgdat has no valid section */
704 pgdat->node_start_pfn = 0;
705 pgdat->node_spanned_pages = 0;
708 static void __remove_zone(struct zone *zone, unsigned long start_pfn)
710 struct pglist_data *pgdat = zone->zone_pgdat;
711 int nr_pages = PAGES_PER_SECTION;
715 zone_type = zone - pgdat->node_zones;
717 pgdat_resize_lock(zone->zone_pgdat, &flags);
718 shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
719 shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages);
720 pgdat_resize_unlock(zone->zone_pgdat, &flags);
723 static int __remove_section(struct zone *zone, struct mem_section *ms)
725 unsigned long start_pfn;
729 if (!valid_section(ms))
732 ret = unregister_memory_section(ms);
736 scn_nr = __section_nr(ms);
737 start_pfn = section_nr_to_pfn(scn_nr);
738 __remove_zone(zone, start_pfn);
740 sparse_remove_one_section(zone, ms);
745 * __remove_pages() - remove sections of pages from a zone
746 * @zone: zone from which pages need to be removed
747 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
748 * @nr_pages: number of pages to remove (must be multiple of section size)
750 * Generic helper function to remove section mappings and sysfs entries
751 * for the section of the memory we are removing. Caller needs to make
752 * sure that pages are marked reserved and zones are adjust properly by
753 * calling offline_pages().
755 int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
756 unsigned long nr_pages)
759 int sections_to_remove;
760 resource_size_t start, size;
764 * We can only remove entire sections
766 BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
767 BUG_ON(nr_pages % PAGES_PER_SECTION);
769 start = phys_start_pfn << PAGE_SHIFT;
770 size = nr_pages * PAGE_SIZE;
771 ret = release_mem_region_adjustable(&iomem_resource, start, size);
773 resource_size_t endres = start + size - 1;
775 pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
776 &start, &endres, ret);
779 sections_to_remove = nr_pages / PAGES_PER_SECTION;
780 for (i = 0; i < sections_to_remove; i++) {
781 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
782 ret = __remove_section(zone, __pfn_to_section(pfn));
788 EXPORT_SYMBOL_GPL(__remove_pages);
789 #endif /* CONFIG_MEMORY_HOTREMOVE */
791 int set_online_page_callback(online_page_callback_t callback)
796 mutex_lock(&online_page_callback_lock);
798 if (online_page_callback == generic_online_page) {
799 online_page_callback = callback;
803 mutex_unlock(&online_page_callback_lock);
808 EXPORT_SYMBOL_GPL(set_online_page_callback);
810 int restore_online_page_callback(online_page_callback_t callback)
815 mutex_lock(&online_page_callback_lock);
817 if (online_page_callback == callback) {
818 online_page_callback = generic_online_page;
822 mutex_unlock(&online_page_callback_lock);
827 EXPORT_SYMBOL_GPL(restore_online_page_callback);
829 void __online_page_set_limits(struct page *page)
832 EXPORT_SYMBOL_GPL(__online_page_set_limits);
834 void __online_page_increment_counters(struct page *page)
836 adjust_managed_page_count(page, 1);
838 EXPORT_SYMBOL_GPL(__online_page_increment_counters);
840 void __online_page_free(struct page *page)
842 __free_reserved_page(page);
844 EXPORT_SYMBOL_GPL(__online_page_free);
846 static void generic_online_page(struct page *page)
848 __online_page_set_limits(page);
849 __online_page_increment_counters(page);
850 __online_page_free(page);
853 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
857 unsigned long onlined_pages = *(unsigned long *)arg;
859 if (PageReserved(pfn_to_page(start_pfn)))
860 for (i = 0; i < nr_pages; i++) {
861 page = pfn_to_page(start_pfn + i);
862 (*online_page_callback)(page);
865 *(unsigned long *)arg = onlined_pages;
869 #ifdef CONFIG_MOVABLE_NODE
871 * When CONFIG_MOVABLE_NODE, we permit onlining of a node which doesn't have
874 static bool can_online_high_movable(struct zone *zone)
878 #else /* CONFIG_MOVABLE_NODE */
879 /* ensure every online node has NORMAL memory */
880 static bool can_online_high_movable(struct zone *zone)
882 return node_state(zone_to_nid(zone), N_NORMAL_MEMORY);
884 #endif /* CONFIG_MOVABLE_NODE */
886 /* check which state of node_states will be changed when online memory */
887 static void node_states_check_changes_online(unsigned long nr_pages,
888 struct zone *zone, struct memory_notify *arg)
890 int nid = zone_to_nid(zone);
891 enum zone_type zone_last = ZONE_NORMAL;
894 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
895 * contains nodes which have zones of 0...ZONE_NORMAL,
896 * set zone_last to ZONE_NORMAL.
898 * If we don't have HIGHMEM nor movable node,
899 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
900 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
902 if (N_MEMORY == N_NORMAL_MEMORY)
903 zone_last = ZONE_MOVABLE;
906 * if the memory to be online is in a zone of 0...zone_last, and
907 * the zones of 0...zone_last don't have memory before online, we will
908 * need to set the node to node_states[N_NORMAL_MEMORY] after
909 * the memory is online.
911 if (zone_idx(zone) <= zone_last && !node_state(nid, N_NORMAL_MEMORY))
912 arg->status_change_nid_normal = nid;
914 arg->status_change_nid_normal = -1;
916 #ifdef CONFIG_HIGHMEM
918 * If we have movable node, node_states[N_HIGH_MEMORY]
919 * contains nodes which have zones of 0...ZONE_HIGHMEM,
920 * set zone_last to ZONE_HIGHMEM.
922 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
923 * contains nodes which have zones of 0...ZONE_MOVABLE,
924 * set zone_last to ZONE_MOVABLE.
926 zone_last = ZONE_HIGHMEM;
927 if (N_MEMORY == N_HIGH_MEMORY)
928 zone_last = ZONE_MOVABLE;
930 if (zone_idx(zone) <= zone_last && !node_state(nid, N_HIGH_MEMORY))
931 arg->status_change_nid_high = nid;
933 arg->status_change_nid_high = -1;
935 arg->status_change_nid_high = arg->status_change_nid_normal;
939 * if the node don't have memory befor online, we will need to
940 * set the node to node_states[N_MEMORY] after the memory
943 if (!node_state(nid, N_MEMORY))
944 arg->status_change_nid = nid;
946 arg->status_change_nid = -1;
949 static void node_states_set_node(int node, struct memory_notify *arg)
951 if (arg->status_change_nid_normal >= 0)
952 node_set_state(node, N_NORMAL_MEMORY);
954 if (arg->status_change_nid_high >= 0)
955 node_set_state(node, N_HIGH_MEMORY);
957 node_set_state(node, N_MEMORY);
961 int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type)
964 unsigned long onlined_pages = 0;
966 int need_zonelists_rebuild = 0;
969 struct memory_notify arg;
973 * This doesn't need a lock to do pfn_to_page().
974 * The section can't be removed here because of the
975 * memory_block->state_mutex.
977 zone = page_zone(pfn_to_page(pfn));
980 if ((zone_idx(zone) > ZONE_NORMAL ||
981 online_type == MMOP_ONLINE_MOVABLE) &&
982 !can_online_high_movable(zone))
985 if (online_type == MMOP_ONLINE_KERNEL &&
986 zone_idx(zone) == ZONE_MOVABLE) {
987 if (move_pfn_range_left(zone - 1, zone, pfn, pfn + nr_pages))
990 if (online_type == MMOP_ONLINE_MOVABLE &&
991 zone_idx(zone) == ZONE_MOVABLE - 1) {
992 if (move_pfn_range_right(zone, zone + 1, pfn, pfn + nr_pages))
996 /* Previous code may changed the zone of the pfn range */
997 zone = page_zone(pfn_to_page(pfn));
1000 arg.nr_pages = nr_pages;
1001 node_states_check_changes_online(nr_pages, zone, &arg);
1003 nid = pfn_to_nid(pfn);
1005 ret = memory_notify(MEM_GOING_ONLINE, &arg);
1006 ret = notifier_to_errno(ret);
1008 memory_notify(MEM_CANCEL_ONLINE, &arg);
1012 * If this zone is not populated, then it is not in zonelist.
1013 * This means the page allocator ignores this zone.
1014 * So, zonelist must be updated after online.
1016 mutex_lock(&zonelists_mutex);
1017 if (!populated_zone(zone)) {
1018 need_zonelists_rebuild = 1;
1019 build_all_zonelists(NULL, zone);
1022 ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
1023 online_pages_range);
1025 if (need_zonelists_rebuild)
1026 zone_pcp_reset(zone);
1027 mutex_unlock(&zonelists_mutex);
1028 printk(KERN_DEBUG "online_pages [mem %#010llx-%#010llx] failed\n",
1029 (unsigned long long) pfn << PAGE_SHIFT,
1030 (((unsigned long long) pfn + nr_pages)
1031 << PAGE_SHIFT) - 1);
1032 memory_notify(MEM_CANCEL_ONLINE, &arg);
1036 zone->present_pages += onlined_pages;
1038 pgdat_resize_lock(zone->zone_pgdat, &flags);
1039 zone->zone_pgdat->node_present_pages += onlined_pages;
1040 pgdat_resize_unlock(zone->zone_pgdat, &flags);
1042 if (onlined_pages) {
1043 node_states_set_node(zone_to_nid(zone), &arg);
1044 if (need_zonelists_rebuild)
1045 build_all_zonelists(NULL, NULL);
1047 zone_pcp_update(zone);
1050 mutex_unlock(&zonelists_mutex);
1052 init_per_zone_wmark_min();
1055 kswapd_run(zone_to_nid(zone));
1057 vm_total_pages = nr_free_pagecache_pages();
1059 writeback_set_ratelimit();
1062 memory_notify(MEM_ONLINE, &arg);
1067 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
1069 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1070 static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
1072 struct pglist_data *pgdat;
1073 unsigned long zones_size[MAX_NR_ZONES] = {0};
1074 unsigned long zholes_size[MAX_NR_ZONES] = {0};
1075 unsigned long start_pfn = PFN_DOWN(start);
1077 pgdat = NODE_DATA(nid);
1079 pgdat = arch_alloc_nodedata(nid);
1083 arch_refresh_nodedata(nid, pgdat);
1086 /* we can use NODE_DATA(nid) from here */
1088 /* init node's zones as empty zones, we don't have any present pages.*/
1089 free_area_init_node(nid, zones_size, start_pfn, zholes_size);
1092 * The node we allocated has no zone fallback lists. For avoiding
1093 * to access not-initialized zonelist, build here.
1095 mutex_lock(&zonelists_mutex);
1096 build_all_zonelists(pgdat, NULL);
1097 mutex_unlock(&zonelists_mutex);
1102 static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
1104 arch_refresh_nodedata(nid, NULL);
1105 arch_free_nodedata(pgdat);
1111 * try_online_node - online a node if offlined
1113 * called by cpu_up() to online a node without onlined memory.
1115 int try_online_node(int nid)
1120 if (node_online(nid))
1123 mem_hotplug_begin();
1124 pgdat = hotadd_new_pgdat(nid, 0);
1126 pr_err("Cannot online node %d due to NULL pgdat\n", nid);
1130 node_set_online(nid);
1131 ret = register_one_node(nid);
1134 if (pgdat->node_zonelists->_zonerefs->zone == NULL) {
1135 mutex_lock(&zonelists_mutex);
1136 build_all_zonelists(NULL, NULL);
1137 mutex_unlock(&zonelists_mutex);
1145 static int check_hotplug_memory_range(u64 start, u64 size)
1147 u64 start_pfn = PFN_DOWN(start);
1148 u64 nr_pages = size >> PAGE_SHIFT;
1150 /* Memory range must be aligned with section */
1151 if ((start_pfn & ~PAGE_SECTION_MASK) ||
1152 (nr_pages % PAGES_PER_SECTION) || (!nr_pages)) {
1153 pr_err("Section-unaligned hotplug range: start 0x%llx, size 0x%llx\n",
1154 (unsigned long long)start,
1155 (unsigned long long)size);
1163 * If movable zone has already been setup, newly added memory should be check.
1164 * If its address is higher than movable zone, it should be added as movable.
1165 * Without this check, movable zone may overlap with other zone.
1167 static int should_add_memory_movable(int nid, u64 start, u64 size)
1169 unsigned long start_pfn = start >> PAGE_SHIFT;
1170 pg_data_t *pgdat = NODE_DATA(nid);
1171 struct zone *movable_zone = pgdat->node_zones + ZONE_MOVABLE;
1173 if (zone_is_empty(movable_zone))
1176 if (movable_zone->zone_start_pfn <= start_pfn)
1182 int zone_for_memory(int nid, u64 start, u64 size, int zone_default)
1184 if (should_add_memory_movable(nid, start, size))
1185 return ZONE_MOVABLE;
1187 return zone_default;
1190 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1191 int __ref add_memory(int nid, u64 start, u64 size)
1193 pg_data_t *pgdat = NULL;
1196 struct resource *res;
1199 ret = check_hotplug_memory_range(start, size);
1203 res = register_memory_resource(start, size);
1208 { /* Stupid hack to suppress address-never-null warning */
1209 void *p = NODE_DATA(nid);
1213 mem_hotplug_begin();
1215 new_node = !node_online(nid);
1217 pgdat = hotadd_new_pgdat(nid, start);
1223 /* call arch's memory hotadd */
1224 ret = arch_add_memory(nid, start, size);
1229 /* we online node here. we can't roll back from here. */
1230 node_set_online(nid);
1233 ret = register_one_node(nid);
1235 * If sysfs file of new node can't create, cpu on the node
1236 * can't be hot-added. There is no rollback way now.
1237 * So, check by BUG_ON() to catch it reluctantly..
1242 /* create new memmap entry */
1243 firmware_map_add_hotplug(start, start + size, "System RAM");
1248 /* rollback pgdat allocation and others */
1250 rollback_node_hotadd(nid, pgdat);
1251 release_memory_resource(res);
1257 EXPORT_SYMBOL_GPL(add_memory);
1259 #ifdef CONFIG_MEMORY_HOTREMOVE
1261 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1262 * set and the size of the free page is given by page_order(). Using this,
1263 * the function determines if the pageblock contains only free pages.
1264 * Due to buddy contraints, a free page at least the size of a pageblock will
1265 * be located at the start of the pageblock
1267 static inline int pageblock_free(struct page *page)
1269 return PageBuddy(page) && page_order(page) >= pageblock_order;
1272 /* Return the start of the next active pageblock after a given page */
1273 static struct page *next_active_pageblock(struct page *page)
1275 /* Ensure the starting page is pageblock-aligned */
1276 BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
1278 /* If the entire pageblock is free, move to the end of free page */
1279 if (pageblock_free(page)) {
1281 /* be careful. we don't have locks, page_order can be changed.*/
1282 order = page_order(page);
1283 if ((order < MAX_ORDER) && (order >= pageblock_order))
1284 return page + (1 << order);
1287 return page + pageblock_nr_pages;
1290 /* Checks if this range of memory is likely to be hot-removable. */
1291 int is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
1293 struct page *page = pfn_to_page(start_pfn);
1294 struct page *end_page = page + nr_pages;
1296 /* Check the starting page of each pageblock within the range */
1297 for (; page < end_page; page = next_active_pageblock(page)) {
1298 if (!is_pageblock_removable_nolock(page))
1303 /* All pageblocks in the memory block are likely to be hot-removable */
1308 * Confirm all pages in a range [start, end) is belongs to the same zone.
1310 int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn)
1313 struct zone *zone = NULL;
1316 for (pfn = start_pfn;
1318 pfn += MAX_ORDER_NR_PAGES) {
1320 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1321 while ((i < MAX_ORDER_NR_PAGES) && !pfn_valid_within(pfn + i))
1323 if (i == MAX_ORDER_NR_PAGES)
1325 page = pfn_to_page(pfn + i);
1326 if (zone && page_zone(page) != zone)
1328 zone = page_zone(page);
1334 * Scan pfn range [start,end) to find movable/migratable pages (LRU pages
1335 * and hugepages). We scan pfn because it's much easier than scanning over
1336 * linked list. This function returns the pfn of the first found movable
1337 * page if it's found, otherwise 0.
1339 static unsigned long scan_movable_pages(unsigned long start, unsigned long end)
1343 for (pfn = start; pfn < end; pfn++) {
1344 if (pfn_valid(pfn)) {
1345 page = pfn_to_page(pfn);
1348 if (PageHuge(page)) {
1349 if (is_hugepage_active(page))
1352 pfn = round_up(pfn + 1,
1353 1 << compound_order(page)) - 1;
1360 #define NR_OFFLINE_AT_ONCE_PAGES (256)
1362 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1366 int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
1367 int not_managed = 0;
1371 for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
1372 if (!pfn_valid(pfn))
1374 page = pfn_to_page(pfn);
1376 if (PageHuge(page)) {
1377 struct page *head = compound_head(page);
1378 pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1;
1379 if (compound_order(head) > PFN_SECTION_SHIFT) {
1383 if (isolate_huge_page(page, &source))
1384 move_pages -= 1 << compound_order(head);
1388 if (!get_page_unless_zero(page))
1391 * We can skip free pages. And we can only deal with pages on
1394 ret = isolate_lru_page(page);
1395 if (!ret) { /* Success */
1397 list_add_tail(&page->lru, &source);
1399 inc_zone_page_state(page, NR_ISOLATED_ANON +
1400 page_is_file_cache(page));
1403 #ifdef CONFIG_DEBUG_VM
1404 printk(KERN_ALERT "removing pfn %lx from LRU failed\n",
1406 dump_page(page, "failed to remove from LRU");
1409 /* Because we don't have big zone->lock. we should
1410 check this again here. */
1411 if (page_count(page)) {
1418 if (!list_empty(&source)) {
1420 putback_movable_pages(&source);
1425 * alloc_migrate_target should be improooooved!!
1426 * migrate_pages returns # of failed pages.
1428 ret = migrate_pages(&source, alloc_migrate_target, NULL, 0,
1429 MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
1431 putback_movable_pages(&source);
1438 * remove from free_area[] and mark all as Reserved.
1441 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1444 __offline_isolated_pages(start, start + nr_pages);
1449 offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
1451 walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
1452 offline_isolated_pages_cb);
1456 * Check all pages in range, recoreded as memory resource, are isolated.
1459 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1463 long offlined = *(long *)data;
1464 ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true);
1465 offlined = nr_pages;
1467 *(long *)data += offlined;
1472 check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
1477 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
1478 check_pages_isolated_cb);
1480 offlined = (long)ret;
1484 #ifdef CONFIG_MOVABLE_NODE
1486 * When CONFIG_MOVABLE_NODE, we permit offlining of a node which doesn't have
1489 static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1493 #else /* CONFIG_MOVABLE_NODE */
1494 /* ensure the node has NORMAL memory if it is still online */
1495 static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1497 struct pglist_data *pgdat = zone->zone_pgdat;
1498 unsigned long present_pages = 0;
1501 for (zt = 0; zt <= ZONE_NORMAL; zt++)
1502 present_pages += pgdat->node_zones[zt].present_pages;
1504 if (present_pages > nr_pages)
1508 for (; zt <= ZONE_MOVABLE; zt++)
1509 present_pages += pgdat->node_zones[zt].present_pages;
1512 * we can't offline the last normal memory until all
1513 * higher memory is offlined.
1515 return present_pages == 0;
1517 #endif /* CONFIG_MOVABLE_NODE */
1519 static int __init cmdline_parse_movable_node(char *p)
1521 #ifdef CONFIG_MOVABLE_NODE
1523 * Memory used by the kernel cannot be hot-removed because Linux
1524 * cannot migrate the kernel pages. When memory hotplug is
1525 * enabled, we should prevent memblock from allocating memory
1528 * ACPI SRAT records all hotpluggable memory ranges. But before
1529 * SRAT is parsed, we don't know about it.
1531 * The kernel image is loaded into memory at very early time. We
1532 * cannot prevent this anyway. So on NUMA system, we set any
1533 * node the kernel resides in as un-hotpluggable.
1535 * Since on modern servers, one node could have double-digit
1536 * gigabytes memory, we can assume the memory around the kernel
1537 * image is also un-hotpluggable. So before SRAT is parsed, just
1538 * allocate memory near the kernel image to try the best to keep
1539 * the kernel away from hotpluggable memory.
1541 memblock_set_bottom_up(true);
1542 movable_node_enabled = true;
1544 pr_warn("movable_node option not supported\n");
1548 early_param("movable_node", cmdline_parse_movable_node);
1550 /* check which state of node_states will be changed when offline memory */
1551 static void node_states_check_changes_offline(unsigned long nr_pages,
1552 struct zone *zone, struct memory_notify *arg)
1554 struct pglist_data *pgdat = zone->zone_pgdat;
1555 unsigned long present_pages = 0;
1556 enum zone_type zt, zone_last = ZONE_NORMAL;
1559 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
1560 * contains nodes which have zones of 0...ZONE_NORMAL,
1561 * set zone_last to ZONE_NORMAL.
1563 * If we don't have HIGHMEM nor movable node,
1564 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
1565 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
1567 if (N_MEMORY == N_NORMAL_MEMORY)
1568 zone_last = ZONE_MOVABLE;
1571 * check whether node_states[N_NORMAL_MEMORY] will be changed.
1572 * If the memory to be offline is in a zone of 0...zone_last,
1573 * and it is the last present memory, 0...zone_last will
1574 * become empty after offline , thus we can determind we will
1575 * need to clear the node from node_states[N_NORMAL_MEMORY].
1577 for (zt = 0; zt <= zone_last; zt++)
1578 present_pages += pgdat->node_zones[zt].present_pages;
1579 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1580 arg->status_change_nid_normal = zone_to_nid(zone);
1582 arg->status_change_nid_normal = -1;
1584 #ifdef CONFIG_HIGHMEM
1586 * If we have movable node, node_states[N_HIGH_MEMORY]
1587 * contains nodes which have zones of 0...ZONE_HIGHMEM,
1588 * set zone_last to ZONE_HIGHMEM.
1590 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
1591 * contains nodes which have zones of 0...ZONE_MOVABLE,
1592 * set zone_last to ZONE_MOVABLE.
1594 zone_last = ZONE_HIGHMEM;
1595 if (N_MEMORY == N_HIGH_MEMORY)
1596 zone_last = ZONE_MOVABLE;
1598 for (; zt <= zone_last; zt++)
1599 present_pages += pgdat->node_zones[zt].present_pages;
1600 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1601 arg->status_change_nid_high = zone_to_nid(zone);
1603 arg->status_change_nid_high = -1;
1605 arg->status_change_nid_high = arg->status_change_nid_normal;
1609 * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE
1611 zone_last = ZONE_MOVABLE;
1614 * check whether node_states[N_HIGH_MEMORY] will be changed
1615 * If we try to offline the last present @nr_pages from the node,
1616 * we can determind we will need to clear the node from
1617 * node_states[N_HIGH_MEMORY].
1619 for (; zt <= zone_last; zt++)
1620 present_pages += pgdat->node_zones[zt].present_pages;
1621 if (nr_pages >= present_pages)
1622 arg->status_change_nid = zone_to_nid(zone);
1624 arg->status_change_nid = -1;
1627 static void node_states_clear_node(int node, struct memory_notify *arg)
1629 if (arg->status_change_nid_normal >= 0)
1630 node_clear_state(node, N_NORMAL_MEMORY);
1632 if ((N_MEMORY != N_NORMAL_MEMORY) &&
1633 (arg->status_change_nid_high >= 0))
1634 node_clear_state(node, N_HIGH_MEMORY);
1636 if ((N_MEMORY != N_HIGH_MEMORY) &&
1637 (arg->status_change_nid >= 0))
1638 node_clear_state(node, N_MEMORY);
1641 static int __ref __offline_pages(unsigned long start_pfn,
1642 unsigned long end_pfn, unsigned long timeout)
1644 unsigned long pfn, nr_pages, expire;
1645 long offlined_pages;
1646 int ret, drain, retry_max, node;
1647 unsigned long flags;
1649 struct memory_notify arg;
1651 /* at least, alignment against pageblock is necessary */
1652 if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
1654 if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
1656 /* This makes hotplug much easier...and readable.
1657 we assume this for now. .*/
1658 if (!test_pages_in_a_zone(start_pfn, end_pfn))
1661 mem_hotplug_begin();
1663 zone = page_zone(pfn_to_page(start_pfn));
1664 node = zone_to_nid(zone);
1665 nr_pages = end_pfn - start_pfn;
1668 if (zone_idx(zone) <= ZONE_NORMAL && !can_offline_normal(zone, nr_pages))
1671 /* set above range as isolated */
1672 ret = start_isolate_page_range(start_pfn, end_pfn,
1673 MIGRATE_MOVABLE, true);
1677 arg.start_pfn = start_pfn;
1678 arg.nr_pages = nr_pages;
1679 node_states_check_changes_offline(nr_pages, zone, &arg);
1681 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1682 ret = notifier_to_errno(ret);
1684 goto failed_removal;
1687 expire = jiffies + timeout;
1691 /* start memory hot removal */
1693 if (time_after(jiffies, expire))
1694 goto failed_removal;
1696 if (signal_pending(current))
1697 goto failed_removal;
1700 lru_add_drain_all();
1705 pfn = scan_movable_pages(start_pfn, end_pfn);
1706 if (pfn) { /* We have movable pages */
1707 ret = do_migrate_range(pfn, end_pfn);
1713 if (--retry_max == 0)
1714 goto failed_removal;
1720 /* drain all zone's lru pagevec, this is asynchronous... */
1721 lru_add_drain_all();
1723 /* drain pcp pages, this is synchronous. */
1726 * dissolve free hugepages in the memory block before doing offlining
1727 * actually in order to make hugetlbfs's object counting consistent.
1729 dissolve_free_huge_pages(start_pfn, end_pfn);
1731 offlined_pages = check_pages_isolated(start_pfn, end_pfn);
1732 if (offlined_pages < 0) {
1734 goto failed_removal;
1736 printk(KERN_INFO "Offlined Pages %ld\n", offlined_pages);
1737 /* Ok, all of our target is isolated.
1738 We cannot do rollback at this point. */
1739 offline_isolated_pages(start_pfn, end_pfn);
1740 /* reset pagetype flags and makes migrate type to be MOVABLE */
1741 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1742 /* removal success */
1743 adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages);
1744 zone->present_pages -= offlined_pages;
1746 pgdat_resize_lock(zone->zone_pgdat, &flags);
1747 zone->zone_pgdat->node_present_pages -= offlined_pages;
1748 pgdat_resize_unlock(zone->zone_pgdat, &flags);
1750 init_per_zone_wmark_min();
1752 if (!populated_zone(zone)) {
1753 zone_pcp_reset(zone);
1754 mutex_lock(&zonelists_mutex);
1755 build_all_zonelists(NULL, NULL);
1756 mutex_unlock(&zonelists_mutex);
1758 zone_pcp_update(zone);
1760 node_states_clear_node(node, &arg);
1761 if (arg.status_change_nid >= 0)
1764 vm_total_pages = nr_free_pagecache_pages();
1765 writeback_set_ratelimit();
1767 memory_notify(MEM_OFFLINE, &arg);
1772 printk(KERN_INFO "memory offlining [mem %#010llx-%#010llx] failed\n",
1773 (unsigned long long) start_pfn << PAGE_SHIFT,
1774 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
1775 memory_notify(MEM_CANCEL_OFFLINE, &arg);
1776 /* pushback to free area */
1777 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1784 int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1786 return __offline_pages(start_pfn, start_pfn + nr_pages, 120 * HZ);
1788 #endif /* CONFIG_MEMORY_HOTREMOVE */
1791 * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
1792 * @start_pfn: start pfn of the memory range
1793 * @end_pfn: end pfn of the memory range
1794 * @arg: argument passed to func
1795 * @func: callback for each memory section walked
1797 * This function walks through all present mem sections in range
1798 * [start_pfn, end_pfn) and call func on each mem section.
1800 * Returns the return value of func.
1802 int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn,
1803 void *arg, int (*func)(struct memory_block *, void *))
1805 struct memory_block *mem = NULL;
1806 struct mem_section *section;
1807 unsigned long pfn, section_nr;
1810 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1811 section_nr = pfn_to_section_nr(pfn);
1812 if (!present_section_nr(section_nr))
1815 section = __nr_to_section(section_nr);
1816 /* same memblock? */
1818 if ((section_nr >= mem->start_section_nr) &&
1819 (section_nr <= mem->end_section_nr))
1822 mem = find_memory_block_hinted(section, mem);
1826 ret = func(mem, arg);
1828 kobject_put(&mem->dev.kobj);
1834 kobject_put(&mem->dev.kobj);
1839 #ifdef CONFIG_MEMORY_HOTREMOVE
1840 static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
1842 int ret = !is_memblock_offlined(mem);
1844 if (unlikely(ret)) {
1845 phys_addr_t beginpa, endpa;
1847 beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
1848 endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1;
1849 pr_warn("removing memory fails, because memory "
1850 "[%pa-%pa] is onlined\n",
1857 static int check_cpu_on_node(pg_data_t *pgdat)
1861 for_each_present_cpu(cpu) {
1862 if (cpu_to_node(cpu) == pgdat->node_id)
1864 * the cpu on this node isn't removed, and we can't
1865 * offline this node.
1873 static void unmap_cpu_on_node(pg_data_t *pgdat)
1875 #ifdef CONFIG_ACPI_NUMA
1878 for_each_possible_cpu(cpu)
1879 if (cpu_to_node(cpu) == pgdat->node_id)
1880 numa_clear_node(cpu);
1884 static int check_and_unmap_cpu_on_node(pg_data_t *pgdat)
1888 ret = check_cpu_on_node(pgdat);
1893 * the node will be offlined when we come here, so we can clear
1894 * the cpu_to_node() now.
1897 unmap_cpu_on_node(pgdat);
1904 * Offline a node if all memory sections and cpus of the node are removed.
1906 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1907 * and online/offline operations before this call.
1909 void try_offline_node(int nid)
1911 pg_data_t *pgdat = NODE_DATA(nid);
1912 unsigned long start_pfn = pgdat->node_start_pfn;
1913 unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
1915 struct page *pgdat_page = virt_to_page(pgdat);
1918 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1919 unsigned long section_nr = pfn_to_section_nr(pfn);
1921 if (!present_section_nr(section_nr))
1924 if (pfn_to_nid(pfn) != nid)
1928 * some memory sections of this node are not removed, and we
1929 * can't offline node now.
1934 if (check_and_unmap_cpu_on_node(pgdat))
1938 * all memory/cpu of this node are removed, we can offline this
1941 node_set_offline(nid);
1942 unregister_one_node(nid);
1944 if (!PageSlab(pgdat_page) && !PageCompound(pgdat_page))
1945 /* node data is allocated from boot memory */
1948 /* free waittable in each zone */
1949 for (i = 0; i < MAX_NR_ZONES; i++) {
1950 struct zone *zone = pgdat->node_zones + i;
1953 * wait_table may be allocated from boot memory,
1954 * here only free if it's allocated by vmalloc.
1956 if (is_vmalloc_addr(zone->wait_table))
1957 vfree(zone->wait_table);
1961 * Since there is no way to guarentee the address of pgdat/zone is not
1962 * on stack of any kernel threads or used by other kernel objects
1963 * without reference counting or other symchronizing method, do not
1964 * reset node_data and free pgdat here. Just reset it to 0 and reuse
1965 * the memory when the node is online again.
1967 memset(pgdat, 0, sizeof(*pgdat));
1969 EXPORT_SYMBOL(try_offline_node);
1974 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1975 * and online/offline operations before this call, as required by
1976 * try_offline_node().
1978 void __ref remove_memory(int nid, u64 start, u64 size)
1982 BUG_ON(check_hotplug_memory_range(start, size));
1984 mem_hotplug_begin();
1987 * All memory blocks must be offlined before removing memory. Check
1988 * whether all memory blocks in question are offline and trigger a BUG()
1989 * if this is not the case.
1991 ret = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL,
1992 check_memblock_offlined_cb);
1996 /* remove memmap entry */
1997 firmware_map_remove(start, start + size, "System RAM");
1999 arch_remove_memory(start, size);
2001 try_offline_node(nid);
2005 EXPORT_SYMBOL_GPL(remove_memory);
2006 #endif /* CONFIG_MEMORY_HOTREMOVE */