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/bootmem.h>
13 #include <linux/compiler.h>
14 #include <linux/export.h>
15 #include <linux/pagevec.h>
16 #include <linux/writeback.h>
17 #include <linux/slab.h>
18 #include <linux/sysctl.h>
19 #include <linux/cpu.h>
20 #include <linux/memory.h>
21 #include <linux/memory_hotplug.h>
22 #include <linux/highmem.h>
23 #include <linux/vmalloc.h>
24 #include <linux/ioport.h>
25 #include <linux/delay.h>
26 #include <linux/migrate.h>
27 #include <linux/page-isolation.h>
28 #include <linux/pfn.h>
29 #include <linux/suspend.h>
30 #include <linux/mm_inline.h>
31 #include <linux/firmware-map.h>
33 #include <asm/tlbflush.h>
38 * online_page_callback contains pointer to current page onlining function.
39 * Initially it is generic_online_page(). If it is required it could be
40 * changed by calling set_online_page_callback() for callback registration
41 * and restore_online_page_callback() for generic callback restore.
44 static void generic_online_page(struct page *page);
46 static online_page_callback_t online_page_callback = generic_online_page;
48 DEFINE_MUTEX(mem_hotplug_mutex);
50 void lock_memory_hotplug(void)
52 mutex_lock(&mem_hotplug_mutex);
54 /* for exclusive hibernation if CONFIG_HIBERNATION=y */
58 void unlock_memory_hotplug(void)
60 unlock_system_sleep();
61 mutex_unlock(&mem_hotplug_mutex);
65 /* add this memory to iomem resource */
66 static struct resource *register_memory_resource(u64 start, u64 size)
69 res = kzalloc(sizeof(struct resource), GFP_KERNEL);
72 res->name = "System RAM";
74 res->end = start + size - 1;
75 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
76 if (request_resource(&iomem_resource, res) < 0) {
77 printk("System RAM resource %pR cannot be added\n", res);
84 static void release_memory_resource(struct resource *res)
88 release_resource(res);
93 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
94 void get_page_bootmem(unsigned long info, struct page *page,
97 page->lru.next = (struct list_head *) type;
99 set_page_private(page, info);
100 atomic_inc(&page->_count);
103 /* reference to __meminit __free_pages_bootmem is valid
104 * so use __ref to tell modpost not to generate a warning */
105 void __ref put_page_bootmem(struct page *page)
108 static DEFINE_MUTEX(ppb_lock);
110 type = (unsigned long) page->lru.next;
111 BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE ||
112 type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE);
114 if (atomic_dec_return(&page->_count) == 1) {
115 ClearPagePrivate(page);
116 set_page_private(page, 0);
117 INIT_LIST_HEAD(&page->lru);
120 * Please refer to comment for __free_pages_bootmem()
121 * for why we serialize here.
123 mutex_lock(&ppb_lock);
124 __free_pages_bootmem(page, 0);
125 mutex_unlock(&ppb_lock);
130 #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
131 #ifndef CONFIG_SPARSEMEM_VMEMMAP
132 static void register_page_bootmem_info_section(unsigned long start_pfn)
134 unsigned long *usemap, mapsize, section_nr, i;
135 struct mem_section *ms;
136 struct page *page, *memmap;
138 section_nr = pfn_to_section_nr(start_pfn);
139 ms = __nr_to_section(section_nr);
141 /* Get section's memmap address */
142 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
145 * Get page for the memmap's phys address
146 * XXX: need more consideration for sparse_vmemmap...
148 page = virt_to_page(memmap);
149 mapsize = sizeof(struct page) * PAGES_PER_SECTION;
150 mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
152 /* remember memmap's page */
153 for (i = 0; i < mapsize; i++, page++)
154 get_page_bootmem(section_nr, page, SECTION_INFO);
156 usemap = __nr_to_section(section_nr)->pageblock_flags;
157 page = virt_to_page(usemap);
159 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
161 for (i = 0; i < mapsize; i++, page++)
162 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
165 #else /* CONFIG_SPARSEMEM_VMEMMAP */
166 static void register_page_bootmem_info_section(unsigned long start_pfn)
168 unsigned long *usemap, mapsize, section_nr, i;
169 struct mem_section *ms;
170 struct page *page, *memmap;
172 if (!pfn_valid(start_pfn))
175 section_nr = pfn_to_section_nr(start_pfn);
176 ms = __nr_to_section(section_nr);
178 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
180 register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION);
182 usemap = __nr_to_section(section_nr)->pageblock_flags;
183 page = virt_to_page(usemap);
185 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
187 for (i = 0; i < mapsize; i++, page++)
188 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
190 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
192 void register_page_bootmem_info_node(struct pglist_data *pgdat)
194 unsigned long i, pfn, end_pfn, nr_pages;
195 int node = pgdat->node_id;
199 nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
200 page = virt_to_page(pgdat);
202 for (i = 0; i < nr_pages; i++, page++)
203 get_page_bootmem(node, page, NODE_INFO);
205 zone = &pgdat->node_zones[0];
206 for (; zone < pgdat->node_zones + MAX_NR_ZONES - 1; zone++) {
207 if (zone->wait_table) {
208 nr_pages = zone->wait_table_hash_nr_entries
209 * sizeof(wait_queue_head_t);
210 nr_pages = PAGE_ALIGN(nr_pages) >> PAGE_SHIFT;
211 page = virt_to_page(zone->wait_table);
213 for (i = 0; i < nr_pages; i++, page++)
214 get_page_bootmem(node, page, NODE_INFO);
218 pfn = pgdat->node_start_pfn;
219 end_pfn = pfn + pgdat->node_spanned_pages;
221 /* register_section info */
222 for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
224 * Some platforms can assign the same pfn to multiple nodes - on
225 * node0 as well as nodeN. To avoid registering a pfn against
226 * multiple nodes we check that this pfn does not already
227 * reside in some other node.
229 if (pfn_valid(pfn) && (pfn_to_nid(pfn) == node))
230 register_page_bootmem_info_section(pfn);
233 #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
235 static void grow_zone_span(struct zone *zone, unsigned long start_pfn,
236 unsigned long end_pfn)
238 unsigned long old_zone_end_pfn;
240 zone_span_writelock(zone);
242 old_zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages;
243 if (!zone->spanned_pages || start_pfn < zone->zone_start_pfn)
244 zone->zone_start_pfn = start_pfn;
246 zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
247 zone->zone_start_pfn;
249 zone_span_writeunlock(zone);
252 static void resize_zone(struct zone *zone, unsigned long start_pfn,
253 unsigned long end_pfn)
255 zone_span_writelock(zone);
257 if (end_pfn - start_pfn) {
258 zone->zone_start_pfn = start_pfn;
259 zone->spanned_pages = end_pfn - start_pfn;
262 * make it consist as free_area_init_core(),
263 * if spanned_pages = 0, then keep start_pfn = 0
265 zone->zone_start_pfn = 0;
266 zone->spanned_pages = 0;
269 zone_span_writeunlock(zone);
272 static void fix_zone_id(struct zone *zone, unsigned long start_pfn,
273 unsigned long end_pfn)
275 enum zone_type zid = zone_idx(zone);
276 int nid = zone->zone_pgdat->node_id;
279 for (pfn = start_pfn; pfn < end_pfn; pfn++)
280 set_page_links(pfn_to_page(pfn), zid, nid, pfn);
283 static int __meminit move_pfn_range_left(struct zone *z1, struct zone *z2,
284 unsigned long start_pfn, unsigned long end_pfn)
288 unsigned long z1_start_pfn;
290 if (!z1->wait_table) {
291 ret = init_currently_empty_zone(z1, start_pfn,
292 end_pfn - start_pfn, MEMMAP_HOTPLUG);
297 pgdat_resize_lock(z1->zone_pgdat, &flags);
299 /* can't move pfns which are higher than @z2 */
300 if (end_pfn > z2->zone_start_pfn + z2->spanned_pages)
302 /* the move out part mast at the left most of @z2 */
303 if (start_pfn > z2->zone_start_pfn)
305 /* must included/overlap */
306 if (end_pfn <= z2->zone_start_pfn)
309 /* use start_pfn for z1's start_pfn if z1 is empty */
310 if (z1->spanned_pages)
311 z1_start_pfn = z1->zone_start_pfn;
313 z1_start_pfn = start_pfn;
315 resize_zone(z1, z1_start_pfn, end_pfn);
316 resize_zone(z2, end_pfn, z2->zone_start_pfn + z2->spanned_pages);
318 pgdat_resize_unlock(z1->zone_pgdat, &flags);
320 fix_zone_id(z1, start_pfn, end_pfn);
324 pgdat_resize_unlock(z1->zone_pgdat, &flags);
328 static int __meminit move_pfn_range_right(struct zone *z1, struct zone *z2,
329 unsigned long start_pfn, unsigned long end_pfn)
333 unsigned long z2_end_pfn;
335 if (!z2->wait_table) {
336 ret = init_currently_empty_zone(z2, start_pfn,
337 end_pfn - start_pfn, MEMMAP_HOTPLUG);
342 pgdat_resize_lock(z1->zone_pgdat, &flags);
344 /* can't move pfns which are lower than @z1 */
345 if (z1->zone_start_pfn > start_pfn)
347 /* the move out part mast at the right most of @z1 */
348 if (z1->zone_start_pfn + z1->spanned_pages > end_pfn)
350 /* must included/overlap */
351 if (start_pfn >= z1->zone_start_pfn + z1->spanned_pages)
354 /* use end_pfn for z2's end_pfn if z2 is empty */
355 if (z2->spanned_pages)
356 z2_end_pfn = z2->zone_start_pfn + z2->spanned_pages;
358 z2_end_pfn = end_pfn;
360 resize_zone(z1, z1->zone_start_pfn, start_pfn);
361 resize_zone(z2, start_pfn, z2_end_pfn);
363 pgdat_resize_unlock(z1->zone_pgdat, &flags);
365 fix_zone_id(z2, start_pfn, end_pfn);
369 pgdat_resize_unlock(z1->zone_pgdat, &flags);
373 static void grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn,
374 unsigned long end_pfn)
376 unsigned long old_pgdat_end_pfn =
377 pgdat->node_start_pfn + pgdat->node_spanned_pages;
379 if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
380 pgdat->node_start_pfn = start_pfn;
382 pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
383 pgdat->node_start_pfn;
386 static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn)
388 struct pglist_data *pgdat = zone->zone_pgdat;
389 int nr_pages = PAGES_PER_SECTION;
390 int nid = pgdat->node_id;
394 zone_type = zone - pgdat->node_zones;
395 if (!zone->wait_table) {
398 ret = init_currently_empty_zone(zone, phys_start_pfn,
399 nr_pages, MEMMAP_HOTPLUG);
403 pgdat_resize_lock(zone->zone_pgdat, &flags);
404 grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages);
405 grow_pgdat_span(zone->zone_pgdat, phys_start_pfn,
406 phys_start_pfn + nr_pages);
407 pgdat_resize_unlock(zone->zone_pgdat, &flags);
408 memmap_init_zone(nr_pages, nid, zone_type,
409 phys_start_pfn, MEMMAP_HOTPLUG);
413 static int __meminit __add_section(int nid, struct zone *zone,
414 unsigned long phys_start_pfn)
416 int nr_pages = PAGES_PER_SECTION;
419 if (pfn_valid(phys_start_pfn))
422 ret = sparse_add_one_section(zone, phys_start_pfn, nr_pages);
427 ret = __add_zone(zone, phys_start_pfn);
432 return register_new_memory(nid, __pfn_to_section(phys_start_pfn));
435 #ifdef CONFIG_SPARSEMEM_VMEMMAP
436 static int __remove_section(struct zone *zone, struct mem_section *ms)
439 * XXX: Freeing memmap with vmemmap is not implement yet.
440 * This should be removed later.
445 static int __remove_section(struct zone *zone, struct mem_section *ms)
449 if (!valid_section(ms))
452 ret = unregister_memory_section(ms);
456 sparse_remove_one_section(zone, ms);
462 * Reasonably generic function for adding memory. It is
463 * expected that archs that support memory hotplug will
464 * call this function after deciding the zone to which to
467 int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn,
468 unsigned long nr_pages)
472 int start_sec, end_sec;
473 /* during initialize mem_map, align hot-added range to section */
474 start_sec = pfn_to_section_nr(phys_start_pfn);
475 end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
477 for (i = start_sec; i <= end_sec; i++) {
478 err = __add_section(nid, zone, i << PFN_SECTION_SHIFT);
481 * EEXIST is finally dealt with by ioresource collision
482 * check. see add_memory() => register_memory_resource()
483 * Warning will be printed if there is collision.
485 if (err && (err != -EEXIST))
492 EXPORT_SYMBOL_GPL(__add_pages);
495 * __remove_pages() - remove sections of pages from a zone
496 * @zone: zone from which pages need to be removed
497 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
498 * @nr_pages: number of pages to remove (must be multiple of section size)
500 * Generic helper function to remove section mappings and sysfs entries
501 * for the section of the memory we are removing. Caller needs to make
502 * sure that pages are marked reserved and zones are adjust properly by
503 * calling offline_pages().
505 int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
506 unsigned long nr_pages)
508 unsigned long i, ret = 0;
509 int sections_to_remove;
512 * We can only remove entire sections
514 BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
515 BUG_ON(nr_pages % PAGES_PER_SECTION);
517 release_mem_region(phys_start_pfn << PAGE_SHIFT, nr_pages * PAGE_SIZE);
519 sections_to_remove = nr_pages / PAGES_PER_SECTION;
520 for (i = 0; i < sections_to_remove; i++) {
521 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
522 ret = __remove_section(zone, __pfn_to_section(pfn));
528 EXPORT_SYMBOL_GPL(__remove_pages);
530 int set_online_page_callback(online_page_callback_t callback)
534 lock_memory_hotplug();
536 if (online_page_callback == generic_online_page) {
537 online_page_callback = callback;
541 unlock_memory_hotplug();
545 EXPORT_SYMBOL_GPL(set_online_page_callback);
547 int restore_online_page_callback(online_page_callback_t callback)
551 lock_memory_hotplug();
553 if (online_page_callback == callback) {
554 online_page_callback = generic_online_page;
558 unlock_memory_hotplug();
562 EXPORT_SYMBOL_GPL(restore_online_page_callback);
564 void __online_page_set_limits(struct page *page)
566 unsigned long pfn = page_to_pfn(page);
568 if (pfn >= num_physpages)
569 num_physpages = pfn + 1;
571 EXPORT_SYMBOL_GPL(__online_page_set_limits);
573 void __online_page_increment_counters(struct page *page)
577 #ifdef CONFIG_HIGHMEM
578 if (PageHighMem(page))
582 EXPORT_SYMBOL_GPL(__online_page_increment_counters);
584 void __online_page_free(struct page *page)
586 ClearPageReserved(page);
587 init_page_count(page);
590 EXPORT_SYMBOL_GPL(__online_page_free);
592 static void generic_online_page(struct page *page)
594 __online_page_set_limits(page);
595 __online_page_increment_counters(page);
596 __online_page_free(page);
599 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
603 unsigned long onlined_pages = *(unsigned long *)arg;
605 if (PageReserved(pfn_to_page(start_pfn)))
606 for (i = 0; i < nr_pages; i++) {
607 page = pfn_to_page(start_pfn + i);
608 (*online_page_callback)(page);
611 *(unsigned long *)arg = onlined_pages;
615 #ifdef CONFIG_MOVABLE_NODE
617 * When CONFIG_MOVABLE_NODE, we permit onlining of a node which doesn't have
620 static bool can_online_high_movable(struct zone *zone)
624 #else /* CONFIG_MOVABLE_NODE */
625 /* ensure every online node has NORMAL memory */
626 static bool can_online_high_movable(struct zone *zone)
628 return node_state(zone_to_nid(zone), N_NORMAL_MEMORY);
630 #endif /* CONFIG_MOVABLE_NODE */
632 /* check which state of node_states will be changed when online memory */
633 static void node_states_check_changes_online(unsigned long nr_pages,
634 struct zone *zone, struct memory_notify *arg)
636 int nid = zone_to_nid(zone);
637 enum zone_type zone_last = ZONE_NORMAL;
640 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
641 * contains nodes which have zones of 0...ZONE_NORMAL,
642 * set zone_last to ZONE_NORMAL.
644 * If we don't have HIGHMEM nor movable node,
645 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
646 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
648 if (N_MEMORY == N_NORMAL_MEMORY)
649 zone_last = ZONE_MOVABLE;
652 * if the memory to be online is in a zone of 0...zone_last, and
653 * the zones of 0...zone_last don't have memory before online, we will
654 * need to set the node to node_states[N_NORMAL_MEMORY] after
655 * the memory is online.
657 if (zone_idx(zone) <= zone_last && !node_state(nid, N_NORMAL_MEMORY))
658 arg->status_change_nid_normal = nid;
660 arg->status_change_nid_normal = -1;
662 #ifdef CONFIG_HIGHMEM
664 * If we have movable node, node_states[N_HIGH_MEMORY]
665 * contains nodes which have zones of 0...ZONE_HIGHMEM,
666 * set zone_last to ZONE_HIGHMEM.
668 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
669 * contains nodes which have zones of 0...ZONE_MOVABLE,
670 * set zone_last to ZONE_MOVABLE.
672 zone_last = ZONE_HIGHMEM;
673 if (N_MEMORY == N_HIGH_MEMORY)
674 zone_last = ZONE_MOVABLE;
676 if (zone_idx(zone) <= zone_last && !node_state(nid, N_HIGH_MEMORY))
677 arg->status_change_nid_high = nid;
679 arg->status_change_nid_high = -1;
681 arg->status_change_nid_high = arg->status_change_nid_normal;
685 * if the node don't have memory befor online, we will need to
686 * set the node to node_states[N_MEMORY] after the memory
689 if (!node_state(nid, N_MEMORY))
690 arg->status_change_nid = nid;
692 arg->status_change_nid = -1;
695 static void node_states_set_node(int node, struct memory_notify *arg)
697 if (arg->status_change_nid_normal >= 0)
698 node_set_state(node, N_NORMAL_MEMORY);
700 if (arg->status_change_nid_high >= 0)
701 node_set_state(node, N_HIGH_MEMORY);
703 node_set_state(node, N_MEMORY);
707 int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type)
709 unsigned long onlined_pages = 0;
711 int need_zonelists_rebuild = 0;
714 struct memory_notify arg;
716 lock_memory_hotplug();
718 * This doesn't need a lock to do pfn_to_page().
719 * The section can't be removed here because of the
720 * memory_block->state_mutex.
722 zone = page_zone(pfn_to_page(pfn));
724 if ((zone_idx(zone) > ZONE_NORMAL || online_type == ONLINE_MOVABLE) &&
725 !can_online_high_movable(zone)) {
726 unlock_memory_hotplug();
730 if (online_type == ONLINE_KERNEL && zone_idx(zone) == ZONE_MOVABLE) {
731 if (move_pfn_range_left(zone - 1, zone, pfn, pfn + nr_pages)) {
732 unlock_memory_hotplug();
736 if (online_type == ONLINE_MOVABLE && zone_idx(zone) == ZONE_MOVABLE - 1) {
737 if (move_pfn_range_right(zone, zone + 1, pfn, pfn + nr_pages)) {
738 unlock_memory_hotplug();
743 /* Previous code may changed the zone of the pfn range */
744 zone = page_zone(pfn_to_page(pfn));
747 arg.nr_pages = nr_pages;
748 node_states_check_changes_online(nr_pages, zone, &arg);
750 nid = page_to_nid(pfn_to_page(pfn));
752 ret = memory_notify(MEM_GOING_ONLINE, &arg);
753 ret = notifier_to_errno(ret);
755 memory_notify(MEM_CANCEL_ONLINE, &arg);
756 unlock_memory_hotplug();
760 * If this zone is not populated, then it is not in zonelist.
761 * This means the page allocator ignores this zone.
762 * So, zonelist must be updated after online.
764 mutex_lock(&zonelists_mutex);
765 if (!populated_zone(zone)) {
766 need_zonelists_rebuild = 1;
767 build_all_zonelists(NULL, zone);
770 ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
773 if (need_zonelists_rebuild)
774 zone_pcp_reset(zone);
775 mutex_unlock(&zonelists_mutex);
776 printk(KERN_DEBUG "online_pages [mem %#010llx-%#010llx] failed\n",
777 (unsigned long long) pfn << PAGE_SHIFT,
778 (((unsigned long long) pfn + nr_pages)
780 memory_notify(MEM_CANCEL_ONLINE, &arg);
781 unlock_memory_hotplug();
785 zone->managed_pages += onlined_pages;
786 zone->present_pages += onlined_pages;
787 zone->zone_pgdat->node_present_pages += onlined_pages;
789 node_states_set_node(zone_to_nid(zone), &arg);
790 if (need_zonelists_rebuild)
791 build_all_zonelists(NULL, NULL);
793 zone_pcp_update(zone);
796 mutex_unlock(&zonelists_mutex);
798 init_per_zone_wmark_min();
801 kswapd_run(zone_to_nid(zone));
803 vm_total_pages = nr_free_pagecache_pages();
805 writeback_set_ratelimit();
808 memory_notify(MEM_ONLINE, &arg);
809 unlock_memory_hotplug();
813 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
815 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
816 static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
818 struct pglist_data *pgdat;
819 unsigned long zones_size[MAX_NR_ZONES] = {0};
820 unsigned long zholes_size[MAX_NR_ZONES] = {0};
821 unsigned long start_pfn = start >> PAGE_SHIFT;
823 pgdat = arch_alloc_nodedata(nid);
827 arch_refresh_nodedata(nid, pgdat);
829 /* we can use NODE_DATA(nid) from here */
831 /* init node's zones as empty zones, we don't have any present pages.*/
832 free_area_init_node(nid, zones_size, start_pfn, zholes_size);
835 * The node we allocated has no zone fallback lists. For avoiding
836 * to access not-initialized zonelist, build here.
838 mutex_lock(&zonelists_mutex);
839 build_all_zonelists(pgdat, NULL);
840 mutex_unlock(&zonelists_mutex);
845 static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
847 arch_refresh_nodedata(nid, NULL);
848 arch_free_nodedata(pgdat);
854 * called by cpu_up() to online a node without onlined memory.
856 int mem_online_node(int nid)
861 lock_memory_hotplug();
862 pgdat = hotadd_new_pgdat(nid, 0);
867 node_set_online(nid);
868 ret = register_one_node(nid);
872 unlock_memory_hotplug();
876 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
877 int __ref add_memory(int nid, u64 start, u64 size)
879 pg_data_t *pgdat = NULL;
881 struct resource *res;
884 lock_memory_hotplug();
886 res = register_memory_resource(start, size);
891 if (!node_online(nid)) {
892 pgdat = hotadd_new_pgdat(nid, start);
899 /* call arch's memory hotadd */
900 ret = arch_add_memory(nid, start, size);
905 /* we online node here. we can't roll back from here. */
906 node_set_online(nid);
909 ret = register_one_node(nid);
911 * If sysfs file of new node can't create, cpu on the node
912 * can't be hot-added. There is no rollback way now.
913 * So, check by BUG_ON() to catch it reluctantly..
918 /* create new memmap entry */
919 firmware_map_add_hotplug(start, start + size, "System RAM");
924 /* rollback pgdat allocation and others */
926 rollback_node_hotadd(nid, pgdat);
927 release_memory_resource(res);
930 unlock_memory_hotplug();
933 EXPORT_SYMBOL_GPL(add_memory);
935 #ifdef CONFIG_MEMORY_HOTREMOVE
937 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
938 * set and the size of the free page is given by page_order(). Using this,
939 * the function determines if the pageblock contains only free pages.
940 * Due to buddy contraints, a free page at least the size of a pageblock will
941 * be located at the start of the pageblock
943 static inline int pageblock_free(struct page *page)
945 return PageBuddy(page) && page_order(page) >= pageblock_order;
948 /* Return the start of the next active pageblock after a given page */
949 static struct page *next_active_pageblock(struct page *page)
951 /* Ensure the starting page is pageblock-aligned */
952 BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
954 /* If the entire pageblock is free, move to the end of free page */
955 if (pageblock_free(page)) {
957 /* be careful. we don't have locks, page_order can be changed.*/
958 order = page_order(page);
959 if ((order < MAX_ORDER) && (order >= pageblock_order))
960 return page + (1 << order);
963 return page + pageblock_nr_pages;
966 /* Checks if this range of memory is likely to be hot-removable. */
967 int is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
969 struct page *page = pfn_to_page(start_pfn);
970 struct page *end_page = page + nr_pages;
972 /* Check the starting page of each pageblock within the range */
973 for (; page < end_page; page = next_active_pageblock(page)) {
974 if (!is_pageblock_removable_nolock(page))
979 /* All pageblocks in the memory block are likely to be hot-removable */
984 * Confirm all pages in a range [start, end) is belongs to the same zone.
986 static int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn)
989 struct zone *zone = NULL;
992 for (pfn = start_pfn;
994 pfn += MAX_ORDER_NR_PAGES) {
996 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
997 while ((i < MAX_ORDER_NR_PAGES) && !pfn_valid_within(pfn + i))
999 if (i == MAX_ORDER_NR_PAGES)
1001 page = pfn_to_page(pfn + i);
1002 if (zone && page_zone(page) != zone)
1004 zone = page_zone(page);
1010 * Scanning pfn is much easier than scanning lru list.
1011 * Scan pfn from start to end and Find LRU page.
1013 static unsigned long scan_lru_pages(unsigned long start, unsigned long end)
1017 for (pfn = start; pfn < end; pfn++) {
1018 if (pfn_valid(pfn)) {
1019 page = pfn_to_page(pfn);
1027 #define NR_OFFLINE_AT_ONCE_PAGES (256)
1029 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1033 int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
1034 int not_managed = 0;
1038 for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
1039 if (!pfn_valid(pfn))
1041 page = pfn_to_page(pfn);
1042 if (!get_page_unless_zero(page))
1045 * We can skip free pages. And we can only deal with pages on
1048 ret = isolate_lru_page(page);
1049 if (!ret) { /* Success */
1051 list_add_tail(&page->lru, &source);
1053 inc_zone_page_state(page, NR_ISOLATED_ANON +
1054 page_is_file_cache(page));
1057 #ifdef CONFIG_DEBUG_VM
1058 printk(KERN_ALERT "removing pfn %lx from LRU failed\n",
1063 /* Because we don't have big zone->lock. we should
1064 check this again here. */
1065 if (page_count(page)) {
1072 if (!list_empty(&source)) {
1074 putback_lru_pages(&source);
1079 * alloc_migrate_target should be improooooved!!
1080 * migrate_pages returns # of failed pages.
1082 ret = migrate_pages(&source, alloc_migrate_target, 0,
1086 putback_lru_pages(&source);
1093 * remove from free_area[] and mark all as Reserved.
1096 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1099 __offline_isolated_pages(start, start + nr_pages);
1104 offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
1106 walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
1107 offline_isolated_pages_cb);
1111 * Check all pages in range, recoreded as memory resource, are isolated.
1114 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1118 long offlined = *(long *)data;
1119 ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true);
1120 offlined = nr_pages;
1122 *(long *)data += offlined;
1127 check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
1132 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
1133 check_pages_isolated_cb);
1135 offlined = (long)ret;
1139 #ifdef CONFIG_MOVABLE_NODE
1141 * When CONFIG_MOVABLE_NODE, we permit offlining of a node which doesn't have
1144 static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1148 #else /* CONFIG_MOVABLE_NODE */
1149 /* ensure the node has NORMAL memory if it is still online */
1150 static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1152 struct pglist_data *pgdat = zone->zone_pgdat;
1153 unsigned long present_pages = 0;
1156 for (zt = 0; zt <= ZONE_NORMAL; zt++)
1157 present_pages += pgdat->node_zones[zt].present_pages;
1159 if (present_pages > nr_pages)
1163 for (; zt <= ZONE_MOVABLE; zt++)
1164 present_pages += pgdat->node_zones[zt].present_pages;
1167 * we can't offline the last normal memory until all
1168 * higher memory is offlined.
1170 return present_pages == 0;
1172 #endif /* CONFIG_MOVABLE_NODE */
1174 /* check which state of node_states will be changed when offline memory */
1175 static void node_states_check_changes_offline(unsigned long nr_pages,
1176 struct zone *zone, struct memory_notify *arg)
1178 struct pglist_data *pgdat = zone->zone_pgdat;
1179 unsigned long present_pages = 0;
1180 enum zone_type zt, zone_last = ZONE_NORMAL;
1183 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
1184 * contains nodes which have zones of 0...ZONE_NORMAL,
1185 * set zone_last to ZONE_NORMAL.
1187 * If we don't have HIGHMEM nor movable node,
1188 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
1189 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
1191 if (N_MEMORY == N_NORMAL_MEMORY)
1192 zone_last = ZONE_MOVABLE;
1195 * check whether node_states[N_NORMAL_MEMORY] will be changed.
1196 * If the memory to be offline is in a zone of 0...zone_last,
1197 * and it is the last present memory, 0...zone_last will
1198 * become empty after offline , thus we can determind we will
1199 * need to clear the node from node_states[N_NORMAL_MEMORY].
1201 for (zt = 0; zt <= zone_last; zt++)
1202 present_pages += pgdat->node_zones[zt].present_pages;
1203 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1204 arg->status_change_nid_normal = zone_to_nid(zone);
1206 arg->status_change_nid_normal = -1;
1208 #ifdef CONFIG_HIGHMEM
1210 * If we have movable node, node_states[N_HIGH_MEMORY]
1211 * contains nodes which have zones of 0...ZONE_HIGHMEM,
1212 * set zone_last to ZONE_HIGHMEM.
1214 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
1215 * contains nodes which have zones of 0...ZONE_MOVABLE,
1216 * set zone_last to ZONE_MOVABLE.
1218 zone_last = ZONE_HIGHMEM;
1219 if (N_MEMORY == N_HIGH_MEMORY)
1220 zone_last = ZONE_MOVABLE;
1222 for (; zt <= zone_last; zt++)
1223 present_pages += pgdat->node_zones[zt].present_pages;
1224 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1225 arg->status_change_nid_high = zone_to_nid(zone);
1227 arg->status_change_nid_high = -1;
1229 arg->status_change_nid_high = arg->status_change_nid_normal;
1233 * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE
1235 zone_last = ZONE_MOVABLE;
1238 * check whether node_states[N_HIGH_MEMORY] will be changed
1239 * If we try to offline the last present @nr_pages from the node,
1240 * we can determind we will need to clear the node from
1241 * node_states[N_HIGH_MEMORY].
1243 for (; zt <= zone_last; zt++)
1244 present_pages += pgdat->node_zones[zt].present_pages;
1245 if (nr_pages >= present_pages)
1246 arg->status_change_nid = zone_to_nid(zone);
1248 arg->status_change_nid = -1;
1251 static void node_states_clear_node(int node, struct memory_notify *arg)
1253 if (arg->status_change_nid_normal >= 0)
1254 node_clear_state(node, N_NORMAL_MEMORY);
1256 if ((N_MEMORY != N_NORMAL_MEMORY) &&
1257 (arg->status_change_nid_high >= 0))
1258 node_clear_state(node, N_HIGH_MEMORY);
1260 if ((N_MEMORY != N_HIGH_MEMORY) &&
1261 (arg->status_change_nid >= 0))
1262 node_clear_state(node, N_MEMORY);
1265 static int __ref __offline_pages(unsigned long start_pfn,
1266 unsigned long end_pfn, unsigned long timeout)
1268 unsigned long pfn, nr_pages, expire;
1269 long offlined_pages;
1270 int ret, drain, retry_max, node;
1272 struct memory_notify arg;
1274 BUG_ON(start_pfn >= end_pfn);
1275 /* at least, alignment against pageblock is necessary */
1276 if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
1278 if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
1280 /* This makes hotplug much easier...and readable.
1281 we assume this for now. .*/
1282 if (!test_pages_in_a_zone(start_pfn, end_pfn))
1285 lock_memory_hotplug();
1287 zone = page_zone(pfn_to_page(start_pfn));
1288 node = zone_to_nid(zone);
1289 nr_pages = end_pfn - start_pfn;
1292 if (zone_idx(zone) <= ZONE_NORMAL && !can_offline_normal(zone, nr_pages))
1295 /* set above range as isolated */
1296 ret = start_isolate_page_range(start_pfn, end_pfn,
1297 MIGRATE_MOVABLE, true);
1301 arg.start_pfn = start_pfn;
1302 arg.nr_pages = nr_pages;
1303 node_states_check_changes_offline(nr_pages, zone, &arg);
1305 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1306 ret = notifier_to_errno(ret);
1308 goto failed_removal;
1311 expire = jiffies + timeout;
1315 /* start memory hot removal */
1317 if (time_after(jiffies, expire))
1318 goto failed_removal;
1320 if (signal_pending(current))
1321 goto failed_removal;
1324 lru_add_drain_all();
1329 pfn = scan_lru_pages(start_pfn, end_pfn);
1330 if (pfn) { /* We have page on LRU */
1331 ret = do_migrate_range(pfn, end_pfn);
1337 if (--retry_max == 0)
1338 goto failed_removal;
1344 /* drain all zone's lru pagevec, this is asynchronous... */
1345 lru_add_drain_all();
1347 /* drain pcp pages, this is synchronous. */
1350 offlined_pages = check_pages_isolated(start_pfn, end_pfn);
1351 if (offlined_pages < 0) {
1353 goto failed_removal;
1355 printk(KERN_INFO "Offlined Pages %ld\n", offlined_pages);
1356 /* Ok, all of our target is isolated.
1357 We cannot do rollback at this point. */
1358 offline_isolated_pages(start_pfn, end_pfn);
1359 /* reset pagetype flags and makes migrate type to be MOVABLE */
1360 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1361 /* removal success */
1362 zone->managed_pages -= offlined_pages;
1363 zone->present_pages -= offlined_pages;
1364 zone->zone_pgdat->node_present_pages -= offlined_pages;
1365 totalram_pages -= offlined_pages;
1367 init_per_zone_wmark_min();
1369 if (!populated_zone(zone)) {
1370 zone_pcp_reset(zone);
1371 mutex_lock(&zonelists_mutex);
1372 build_all_zonelists(NULL, NULL);
1373 mutex_unlock(&zonelists_mutex);
1375 zone_pcp_update(zone);
1377 node_states_clear_node(node, &arg);
1378 if (arg.status_change_nid >= 0)
1381 vm_total_pages = nr_free_pagecache_pages();
1382 writeback_set_ratelimit();
1384 memory_notify(MEM_OFFLINE, &arg);
1385 unlock_memory_hotplug();
1389 printk(KERN_INFO "memory offlining [mem %#010llx-%#010llx] failed\n",
1390 (unsigned long long) start_pfn << PAGE_SHIFT,
1391 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
1392 memory_notify(MEM_CANCEL_OFFLINE, &arg);
1393 /* pushback to free area */
1394 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1397 unlock_memory_hotplug();
1401 int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1403 return __offline_pages(start_pfn, start_pfn + nr_pages, 120 * HZ);
1407 * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
1408 * @start_pfn: start pfn of the memory range
1409 * @end_pfn: end pft of the memory range
1410 * @arg: argument passed to func
1411 * @func: callback for each memory section walked
1413 * This function walks through all present mem sections in range
1414 * [start_pfn, end_pfn) and call func on each mem section.
1416 * Returns the return value of func.
1418 static int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn,
1419 void *arg, int (*func)(struct memory_block *, void *))
1421 struct memory_block *mem = NULL;
1422 struct mem_section *section;
1423 unsigned long pfn, section_nr;
1426 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1427 section_nr = pfn_to_section_nr(pfn);
1428 if (!present_section_nr(section_nr))
1431 section = __nr_to_section(section_nr);
1432 /* same memblock? */
1434 if ((section_nr >= mem->start_section_nr) &&
1435 (section_nr <= mem->end_section_nr))
1438 mem = find_memory_block_hinted(section, mem);
1442 ret = func(mem, arg);
1444 kobject_put(&mem->dev.kobj);
1450 kobject_put(&mem->dev.kobj);
1456 * offline_memory_block_cb - callback function for offlining memory block
1457 * @mem: the memory block to be offlined
1458 * @arg: buffer to hold error msg
1460 * Always return 0, and put the error msg in arg if any.
1462 static int offline_memory_block_cb(struct memory_block *mem, void *arg)
1465 int error = offline_memory_block(mem);
1467 if (error != 0 && *ret == 0)
1473 static int is_memblock_offlined_cb(struct memory_block *mem, void *arg)
1475 int ret = !is_memblock_offlined(mem);
1478 pr_warn("removing memory fails, because memory "
1479 "[%#010llx-%#010llx] is onlined\n",
1480 PFN_PHYS(section_nr_to_pfn(mem->start_section_nr)),
1481 PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1);
1486 int __ref remove_memory(u64 start, u64 size)
1488 unsigned long start_pfn, end_pfn;
1492 start_pfn = PFN_DOWN(start);
1493 end_pfn = start_pfn + PFN_DOWN(size);
1496 * When CONFIG_MEMCG is on, one memory block may be used by other
1497 * blocks to store page cgroup when onlining pages. But we don't know
1498 * in what order pages are onlined. So we iterate twice to offline
1500 * 1st iterate: offline every non primary memory block.
1501 * 2nd iterate: offline primary (i.e. first added) memory block.
1504 walk_memory_range(start_pfn, end_pfn, &ret,
1505 offline_memory_block_cb);
1515 lock_memory_hotplug();
1518 * we have offlined all memory blocks like this:
1519 * 1. lock memory hotplug
1520 * 2. offline a memory block
1521 * 3. unlock memory hotplug
1523 * repeat step1-3 to offline the memory block. All memory blocks
1524 * must be offlined before removing memory. But we don't hold the
1525 * lock in the whole operation. So we should check whether all
1526 * memory blocks are offlined.
1529 ret = walk_memory_range(start_pfn, end_pfn, NULL,
1530 is_memblock_offlined_cb);
1532 unlock_memory_hotplug();
1536 /* remove memmap entry */
1537 firmware_map_remove(start, start + size, "System RAM");
1539 arch_remove_memory(start, size);
1541 unlock_memory_hotplug();
1546 int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1550 int remove_memory(u64 start, u64 size)
1554 #endif /* CONFIG_MEMORY_HOTREMOVE */
1555 EXPORT_SYMBOL_GPL(remove_memory);