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 #ifndef CONFIG_SPARSEMEM_VMEMMAP
95 static void get_page_bootmem(unsigned long info, struct page *page,
98 page->lru.next = (struct list_head *) type;
100 set_page_private(page, info);
101 atomic_inc(&page->_count);
104 /* reference to __meminit __free_pages_bootmem is valid
105 * so use __ref to tell modpost not to generate a warning */
106 void __ref put_page_bootmem(struct page *page)
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);
118 __free_pages_bootmem(page, 0);
123 static void register_page_bootmem_info_section(unsigned long start_pfn)
125 unsigned long *usemap, mapsize, section_nr, i;
126 struct mem_section *ms;
127 struct page *page, *memmap;
129 section_nr = pfn_to_section_nr(start_pfn);
130 ms = __nr_to_section(section_nr);
132 /* Get section's memmap address */
133 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
136 * Get page for the memmap's phys address
137 * XXX: need more consideration for sparse_vmemmap...
139 page = virt_to_page(memmap);
140 mapsize = sizeof(struct page) * PAGES_PER_SECTION;
141 mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
143 /* remember memmap's page */
144 for (i = 0; i < mapsize; i++, page++)
145 get_page_bootmem(section_nr, page, SECTION_INFO);
147 usemap = __nr_to_section(section_nr)->pageblock_flags;
148 page = virt_to_page(usemap);
150 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
152 for (i = 0; i < mapsize; i++, page++)
153 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
157 void register_page_bootmem_info_node(struct pglist_data *pgdat)
159 unsigned long i, pfn, end_pfn, nr_pages;
160 int node = pgdat->node_id;
164 nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
165 page = virt_to_page(pgdat);
167 for (i = 0; i < nr_pages; i++, page++)
168 get_page_bootmem(node, page, NODE_INFO);
170 zone = &pgdat->node_zones[0];
171 for (; zone < pgdat->node_zones + MAX_NR_ZONES - 1; zone++) {
172 if (zone->wait_table) {
173 nr_pages = zone->wait_table_hash_nr_entries
174 * sizeof(wait_queue_head_t);
175 nr_pages = PAGE_ALIGN(nr_pages) >> PAGE_SHIFT;
176 page = virt_to_page(zone->wait_table);
178 for (i = 0; i < nr_pages; i++, page++)
179 get_page_bootmem(node, page, NODE_INFO);
183 pfn = pgdat->node_start_pfn;
184 end_pfn = pfn + pgdat->node_spanned_pages;
186 /* register_section info */
187 for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
189 * Some platforms can assign the same pfn to multiple nodes - on
190 * node0 as well as nodeN. To avoid registering a pfn against
191 * multiple nodes we check that this pfn does not already
192 * reside in some other node.
194 if (pfn_valid(pfn) && (pfn_to_nid(pfn) == node))
195 register_page_bootmem_info_section(pfn);
198 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
200 static void grow_zone_span(struct zone *zone, unsigned long start_pfn,
201 unsigned long end_pfn)
203 unsigned long old_zone_end_pfn;
205 zone_span_writelock(zone);
207 old_zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages;
208 if (!zone->spanned_pages || start_pfn < zone->zone_start_pfn)
209 zone->zone_start_pfn = start_pfn;
211 zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
212 zone->zone_start_pfn;
214 zone_span_writeunlock(zone);
217 static void resize_zone(struct zone *zone, unsigned long start_pfn,
218 unsigned long end_pfn)
220 zone_span_writelock(zone);
222 if (end_pfn - start_pfn) {
223 zone->zone_start_pfn = start_pfn;
224 zone->spanned_pages = end_pfn - start_pfn;
227 * make it consist as free_area_init_core(),
228 * if spanned_pages = 0, then keep start_pfn = 0
230 zone->zone_start_pfn = 0;
231 zone->spanned_pages = 0;
234 zone_span_writeunlock(zone);
237 static void fix_zone_id(struct zone *zone, unsigned long start_pfn,
238 unsigned long end_pfn)
240 enum zone_type zid = zone_idx(zone);
241 int nid = zone->zone_pgdat->node_id;
244 for (pfn = start_pfn; pfn < end_pfn; pfn++)
245 set_page_links(pfn_to_page(pfn), zid, nid, pfn);
248 static int __meminit move_pfn_range_left(struct zone *z1, struct zone *z2,
249 unsigned long start_pfn, unsigned long end_pfn)
253 unsigned long z1_start_pfn;
255 if (!z1->wait_table) {
256 ret = init_currently_empty_zone(z1, start_pfn,
257 end_pfn - start_pfn, MEMMAP_HOTPLUG);
262 pgdat_resize_lock(z1->zone_pgdat, &flags);
264 /* can't move pfns which are higher than @z2 */
265 if (end_pfn > z2->zone_start_pfn + z2->spanned_pages)
267 /* the move out part mast at the left most of @z2 */
268 if (start_pfn > z2->zone_start_pfn)
270 /* must included/overlap */
271 if (end_pfn <= z2->zone_start_pfn)
274 /* use start_pfn for z1's start_pfn if z1 is empty */
275 if (z1->spanned_pages)
276 z1_start_pfn = z1->zone_start_pfn;
278 z1_start_pfn = start_pfn;
280 resize_zone(z1, z1_start_pfn, end_pfn);
281 resize_zone(z2, end_pfn, z2->zone_start_pfn + z2->spanned_pages);
283 pgdat_resize_unlock(z1->zone_pgdat, &flags);
285 fix_zone_id(z1, start_pfn, end_pfn);
289 pgdat_resize_unlock(z1->zone_pgdat, &flags);
293 static int __meminit move_pfn_range_right(struct zone *z1, struct zone *z2,
294 unsigned long start_pfn, unsigned long end_pfn)
298 unsigned long z2_end_pfn;
300 if (!z2->wait_table) {
301 ret = init_currently_empty_zone(z2, start_pfn,
302 end_pfn - start_pfn, MEMMAP_HOTPLUG);
307 pgdat_resize_lock(z1->zone_pgdat, &flags);
309 /* can't move pfns which are lower than @z1 */
310 if (z1->zone_start_pfn > start_pfn)
312 /* the move out part mast at the right most of @z1 */
313 if (z1->zone_start_pfn + z1->spanned_pages > end_pfn)
315 /* must included/overlap */
316 if (start_pfn >= z1->zone_start_pfn + z1->spanned_pages)
319 /* use end_pfn for z2's end_pfn if z2 is empty */
320 if (z2->spanned_pages)
321 z2_end_pfn = z2->zone_start_pfn + z2->spanned_pages;
323 z2_end_pfn = end_pfn;
325 resize_zone(z1, z1->zone_start_pfn, start_pfn);
326 resize_zone(z2, start_pfn, z2_end_pfn);
328 pgdat_resize_unlock(z1->zone_pgdat, &flags);
330 fix_zone_id(z2, start_pfn, end_pfn);
334 pgdat_resize_unlock(z1->zone_pgdat, &flags);
338 static void grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn,
339 unsigned long end_pfn)
341 unsigned long old_pgdat_end_pfn =
342 pgdat->node_start_pfn + pgdat->node_spanned_pages;
344 if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
345 pgdat->node_start_pfn = start_pfn;
347 pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
348 pgdat->node_start_pfn;
351 static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn)
353 struct pglist_data *pgdat = zone->zone_pgdat;
354 int nr_pages = PAGES_PER_SECTION;
355 int nid = pgdat->node_id;
359 zone_type = zone - pgdat->node_zones;
360 if (!zone->wait_table) {
363 ret = init_currently_empty_zone(zone, phys_start_pfn,
364 nr_pages, MEMMAP_HOTPLUG);
368 pgdat_resize_lock(zone->zone_pgdat, &flags);
369 grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages);
370 grow_pgdat_span(zone->zone_pgdat, phys_start_pfn,
371 phys_start_pfn + nr_pages);
372 pgdat_resize_unlock(zone->zone_pgdat, &flags);
373 memmap_init_zone(nr_pages, nid, zone_type,
374 phys_start_pfn, MEMMAP_HOTPLUG);
378 static int __meminit __add_section(int nid, struct zone *zone,
379 unsigned long phys_start_pfn)
381 int nr_pages = PAGES_PER_SECTION;
384 if (pfn_valid(phys_start_pfn))
387 ret = sparse_add_one_section(zone, phys_start_pfn, nr_pages);
392 ret = __add_zone(zone, phys_start_pfn);
397 return register_new_memory(nid, __pfn_to_section(phys_start_pfn));
400 #ifdef CONFIG_SPARSEMEM_VMEMMAP
401 static int __remove_section(struct zone *zone, struct mem_section *ms)
404 * XXX: Freeing memmap with vmemmap is not implement yet.
405 * This should be removed later.
410 static int __remove_section(struct zone *zone, struct mem_section *ms)
413 struct pglist_data *pgdat = zone->zone_pgdat;
416 if (!valid_section(ms))
419 ret = unregister_memory_section(ms);
423 pgdat_resize_lock(pgdat, &flags);
424 sparse_remove_one_section(zone, ms);
425 pgdat_resize_unlock(pgdat, &flags);
431 * Reasonably generic function for adding memory. It is
432 * expected that archs that support memory hotplug will
433 * call this function after deciding the zone to which to
436 int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn,
437 unsigned long nr_pages)
441 int start_sec, end_sec;
442 /* during initialize mem_map, align hot-added range to section */
443 start_sec = pfn_to_section_nr(phys_start_pfn);
444 end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
446 for (i = start_sec; i <= end_sec; i++) {
447 err = __add_section(nid, zone, i << PFN_SECTION_SHIFT);
450 * EEXIST is finally dealt with by ioresource collision
451 * check. see add_memory() => register_memory_resource()
452 * Warning will be printed if there is collision.
454 if (err && (err != -EEXIST))
461 EXPORT_SYMBOL_GPL(__add_pages);
464 * __remove_pages() - remove sections of pages from a zone
465 * @zone: zone from which pages need to be removed
466 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
467 * @nr_pages: number of pages to remove (must be multiple of section size)
469 * Generic helper function to remove section mappings and sysfs entries
470 * for the section of the memory we are removing. Caller needs to make
471 * sure that pages are marked reserved and zones are adjust properly by
472 * calling offline_pages().
474 int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
475 unsigned long nr_pages)
477 unsigned long i, ret = 0;
478 int sections_to_remove;
481 * We can only remove entire sections
483 BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
484 BUG_ON(nr_pages % PAGES_PER_SECTION);
486 release_mem_region(phys_start_pfn << PAGE_SHIFT, nr_pages * PAGE_SIZE);
488 sections_to_remove = nr_pages / PAGES_PER_SECTION;
489 for (i = 0; i < sections_to_remove; i++) {
490 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
491 ret = __remove_section(zone, __pfn_to_section(pfn));
497 EXPORT_SYMBOL_GPL(__remove_pages);
499 int set_online_page_callback(online_page_callback_t callback)
503 lock_memory_hotplug();
505 if (online_page_callback == generic_online_page) {
506 online_page_callback = callback;
510 unlock_memory_hotplug();
514 EXPORT_SYMBOL_GPL(set_online_page_callback);
516 int restore_online_page_callback(online_page_callback_t callback)
520 lock_memory_hotplug();
522 if (online_page_callback == callback) {
523 online_page_callback = generic_online_page;
527 unlock_memory_hotplug();
531 EXPORT_SYMBOL_GPL(restore_online_page_callback);
533 void __online_page_set_limits(struct page *page)
535 unsigned long pfn = page_to_pfn(page);
537 if (pfn >= num_physpages)
538 num_physpages = pfn + 1;
540 EXPORT_SYMBOL_GPL(__online_page_set_limits);
542 void __online_page_increment_counters(struct page *page)
546 #ifdef CONFIG_HIGHMEM
547 if (PageHighMem(page))
551 EXPORT_SYMBOL_GPL(__online_page_increment_counters);
553 void __online_page_free(struct page *page)
555 ClearPageReserved(page);
556 init_page_count(page);
559 EXPORT_SYMBOL_GPL(__online_page_free);
561 static void generic_online_page(struct page *page)
563 __online_page_set_limits(page);
564 __online_page_increment_counters(page);
565 __online_page_free(page);
568 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
572 unsigned long onlined_pages = *(unsigned long *)arg;
574 if (PageReserved(pfn_to_page(start_pfn)))
575 for (i = 0; i < nr_pages; i++) {
576 page = pfn_to_page(start_pfn + i);
577 (*online_page_callback)(page);
580 *(unsigned long *)arg = onlined_pages;
584 /* ensure every online node has NORMAL memory */
585 static bool can_online_high_movable(struct zone *zone)
587 return node_state(zone_to_nid(zone), N_NORMAL_MEMORY);
590 /* check which state of node_states will be changed when online memory */
591 static void node_states_check_changes_online(unsigned long nr_pages,
592 struct zone *zone, struct memory_notify *arg)
594 int nid = zone_to_nid(zone);
595 enum zone_type zone_last = ZONE_NORMAL;
598 * If we have HIGHMEM, node_states[N_NORMAL_MEMORY] contains nodes
599 * which have 0...ZONE_NORMAL, set zone_last to ZONE_NORMAL.
601 * If we don't have HIGHMEM, node_states[N_NORMAL_MEMORY] contains nodes
602 * which have 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
604 if (N_HIGH_MEMORY == N_NORMAL_MEMORY)
605 zone_last = ZONE_MOVABLE;
608 * if the memory to be online is in a zone of 0...zone_last, and
609 * the zones of 0...zone_last don't have memory before online, we will
610 * need to set the node to node_states[N_NORMAL_MEMORY] after
611 * the memory is online.
613 if (zone_idx(zone) <= zone_last && !node_state(nid, N_NORMAL_MEMORY))
614 arg->status_change_nid_normal = nid;
616 arg->status_change_nid_normal = -1;
619 * if the node don't have memory befor online, we will need to
620 * set the node to node_states[N_HIGH_MEMORY] after the memory
623 if (!node_state(nid, N_HIGH_MEMORY))
624 arg->status_change_nid = nid;
626 arg->status_change_nid = -1;
629 static void node_states_set_node(int node, struct memory_notify *arg)
631 if (arg->status_change_nid_normal >= 0)
632 node_set_state(node, N_NORMAL_MEMORY);
634 node_set_state(node, N_HIGH_MEMORY);
638 int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type)
640 unsigned long onlined_pages = 0;
642 int need_zonelists_rebuild = 0;
645 struct memory_notify arg;
647 lock_memory_hotplug();
649 * This doesn't need a lock to do pfn_to_page().
650 * The section can't be removed here because of the
651 * memory_block->state_mutex.
653 zone = page_zone(pfn_to_page(pfn));
655 if ((zone_idx(zone) > ZONE_NORMAL || online_type == ONLINE_MOVABLE) &&
656 !can_online_high_movable(zone)) {
657 unlock_memory_hotplug();
661 if (online_type == ONLINE_KERNEL && zone_idx(zone) == ZONE_MOVABLE) {
662 if (move_pfn_range_left(zone - 1, zone, pfn, pfn + nr_pages)) {
663 unlock_memory_hotplug();
667 if (online_type == ONLINE_MOVABLE && zone_idx(zone) == ZONE_MOVABLE - 1) {
668 if (move_pfn_range_right(zone, zone + 1, pfn, pfn + nr_pages)) {
669 unlock_memory_hotplug();
674 /* Previous code may changed the zone of the pfn range */
675 zone = page_zone(pfn_to_page(pfn));
678 arg.nr_pages = nr_pages;
679 node_states_check_changes_online(nr_pages, zone, &arg);
681 nid = page_to_nid(pfn_to_page(pfn));
683 ret = memory_notify(MEM_GOING_ONLINE, &arg);
684 ret = notifier_to_errno(ret);
686 memory_notify(MEM_CANCEL_ONLINE, &arg);
687 unlock_memory_hotplug();
691 * If this zone is not populated, then it is not in zonelist.
692 * This means the page allocator ignores this zone.
693 * So, zonelist must be updated after online.
695 mutex_lock(&zonelists_mutex);
696 if (!populated_zone(zone)) {
697 need_zonelists_rebuild = 1;
698 build_all_zonelists(NULL, zone);
701 ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
704 if (need_zonelists_rebuild)
705 zone_pcp_reset(zone);
706 mutex_unlock(&zonelists_mutex);
707 printk(KERN_DEBUG "online_pages [mem %#010llx-%#010llx] failed\n",
708 (unsigned long long) pfn << PAGE_SHIFT,
709 (((unsigned long long) pfn + nr_pages)
711 memory_notify(MEM_CANCEL_ONLINE, &arg);
712 unlock_memory_hotplug();
716 zone->present_pages += onlined_pages;
717 zone->zone_pgdat->node_present_pages += onlined_pages;
719 node_states_set_node(zone_to_nid(zone), &arg);
720 if (need_zonelists_rebuild)
721 build_all_zonelists(NULL, NULL);
723 zone_pcp_update(zone);
726 mutex_unlock(&zonelists_mutex);
728 init_per_zone_wmark_min();
731 kswapd_run(zone_to_nid(zone));
733 vm_total_pages = nr_free_pagecache_pages();
735 writeback_set_ratelimit();
738 memory_notify(MEM_ONLINE, &arg);
739 unlock_memory_hotplug();
743 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
745 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
746 static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
748 struct pglist_data *pgdat;
749 unsigned long zones_size[MAX_NR_ZONES] = {0};
750 unsigned long zholes_size[MAX_NR_ZONES] = {0};
751 unsigned long start_pfn = start >> PAGE_SHIFT;
753 pgdat = arch_alloc_nodedata(nid);
757 arch_refresh_nodedata(nid, pgdat);
759 /* we can use NODE_DATA(nid) from here */
761 /* init node's zones as empty zones, we don't have any present pages.*/
762 free_area_init_node(nid, zones_size, start_pfn, zholes_size);
765 * The node we allocated has no zone fallback lists. For avoiding
766 * to access not-initialized zonelist, build here.
768 mutex_lock(&zonelists_mutex);
769 build_all_zonelists(pgdat, NULL);
770 mutex_unlock(&zonelists_mutex);
775 static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
777 arch_refresh_nodedata(nid, NULL);
778 arch_free_nodedata(pgdat);
784 * called by cpu_up() to online a node without onlined memory.
786 int mem_online_node(int nid)
791 lock_memory_hotplug();
792 pgdat = hotadd_new_pgdat(nid, 0);
797 node_set_online(nid);
798 ret = register_one_node(nid);
802 unlock_memory_hotplug();
806 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
807 int __ref add_memory(int nid, u64 start, u64 size)
809 pg_data_t *pgdat = NULL;
811 struct resource *res;
814 lock_memory_hotplug();
816 res = register_memory_resource(start, size);
821 if (!node_online(nid)) {
822 pgdat = hotadd_new_pgdat(nid, start);
829 /* call arch's memory hotadd */
830 ret = arch_add_memory(nid, start, size);
835 /* we online node here. we can't roll back from here. */
836 node_set_online(nid);
839 ret = register_one_node(nid);
841 * If sysfs file of new node can't create, cpu on the node
842 * can't be hot-added. There is no rollback way now.
843 * So, check by BUG_ON() to catch it reluctantly..
848 /* create new memmap entry */
849 firmware_map_add_hotplug(start, start + size, "System RAM");
854 /* rollback pgdat allocation and others */
856 rollback_node_hotadd(nid, pgdat);
858 release_memory_resource(res);
861 unlock_memory_hotplug();
864 EXPORT_SYMBOL_GPL(add_memory);
866 #ifdef CONFIG_MEMORY_HOTREMOVE
868 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
869 * set and the size of the free page is given by page_order(). Using this,
870 * the function determines if the pageblock contains only free pages.
871 * Due to buddy contraints, a free page at least the size of a pageblock will
872 * be located at the start of the pageblock
874 static inline int pageblock_free(struct page *page)
876 return PageBuddy(page) && page_order(page) >= pageblock_order;
879 /* Return the start of the next active pageblock after a given page */
880 static struct page *next_active_pageblock(struct page *page)
882 /* Ensure the starting page is pageblock-aligned */
883 BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
885 /* If the entire pageblock is free, move to the end of free page */
886 if (pageblock_free(page)) {
888 /* be careful. we don't have locks, page_order can be changed.*/
889 order = page_order(page);
890 if ((order < MAX_ORDER) && (order >= pageblock_order))
891 return page + (1 << order);
894 return page + pageblock_nr_pages;
897 /* Checks if this range of memory is likely to be hot-removable. */
898 int is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
900 struct page *page = pfn_to_page(start_pfn);
901 struct page *end_page = page + nr_pages;
903 /* Check the starting page of each pageblock within the range */
904 for (; page < end_page; page = next_active_pageblock(page)) {
905 if (!is_pageblock_removable_nolock(page))
910 /* All pageblocks in the memory block are likely to be hot-removable */
915 * Confirm all pages in a range [start, end) is belongs to the same zone.
917 static int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn)
920 struct zone *zone = NULL;
923 for (pfn = start_pfn;
925 pfn += MAX_ORDER_NR_PAGES) {
927 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
928 while ((i < MAX_ORDER_NR_PAGES) && !pfn_valid_within(pfn + i))
930 if (i == MAX_ORDER_NR_PAGES)
932 page = pfn_to_page(pfn + i);
933 if (zone && page_zone(page) != zone)
935 zone = page_zone(page);
941 * Scanning pfn is much easier than scanning lru list.
942 * Scan pfn from start to end and Find LRU page.
944 static unsigned long scan_lru_pages(unsigned long start, unsigned long end)
948 for (pfn = start; pfn < end; pfn++) {
949 if (pfn_valid(pfn)) {
950 page = pfn_to_page(pfn);
958 #define NR_OFFLINE_AT_ONCE_PAGES (256)
960 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
964 int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
969 for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
972 page = pfn_to_page(pfn);
973 if (!get_page_unless_zero(page))
976 * We can skip free pages. And we can only deal with pages on
979 ret = isolate_lru_page(page);
980 if (!ret) { /* Success */
982 list_add_tail(&page->lru, &source);
984 inc_zone_page_state(page, NR_ISOLATED_ANON +
985 page_is_file_cache(page));
988 #ifdef CONFIG_DEBUG_VM
989 printk(KERN_ALERT "removing pfn %lx from LRU failed\n",
994 /* Because we don't have big zone->lock. we should
995 check this again here. */
996 if (page_count(page)) {
1003 if (!list_empty(&source)) {
1005 putback_lru_pages(&source);
1010 * alloc_migrate_target should be improooooved!!
1011 * migrate_pages returns # of failed pages.
1013 ret = migrate_pages(&source, alloc_migrate_target, 0,
1014 true, MIGRATE_SYNC);
1016 putback_lru_pages(&source);
1023 * remove from free_area[] and mark all as Reserved.
1026 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1029 __offline_isolated_pages(start, start + nr_pages);
1034 offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
1036 walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
1037 offline_isolated_pages_cb);
1041 * Check all pages in range, recoreded as memory resource, are isolated.
1044 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1048 long offlined = *(long *)data;
1049 ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true);
1050 offlined = nr_pages;
1052 *(long *)data += offlined;
1057 check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
1062 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
1063 check_pages_isolated_cb);
1065 offlined = (long)ret;
1069 /* ensure the node has NORMAL memory if it is still online */
1070 static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1072 struct pglist_data *pgdat = zone->zone_pgdat;
1073 unsigned long present_pages = 0;
1076 for (zt = 0; zt <= ZONE_NORMAL; zt++)
1077 present_pages += pgdat->node_zones[zt].present_pages;
1079 if (present_pages > nr_pages)
1083 for (; zt <= ZONE_MOVABLE; zt++)
1084 present_pages += pgdat->node_zones[zt].present_pages;
1087 * we can't offline the last normal memory until all
1088 * higher memory is offlined.
1090 return present_pages == 0;
1093 /* check which state of node_states will be changed when offline memory */
1094 static void node_states_check_changes_offline(unsigned long nr_pages,
1095 struct zone *zone, struct memory_notify *arg)
1097 struct pglist_data *pgdat = zone->zone_pgdat;
1098 unsigned long present_pages = 0;
1099 enum zone_type zt, zone_last = ZONE_NORMAL;
1102 * If we have HIGHMEM, node_states[N_NORMAL_MEMORY] contains nodes
1103 * which have 0...ZONE_NORMAL, set zone_last to ZONE_NORMAL.
1105 * If we don't have HIGHMEM, node_states[N_NORMAL_MEMORY] contains nodes
1106 * which have 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
1108 if (N_HIGH_MEMORY == N_NORMAL_MEMORY)
1109 zone_last = ZONE_MOVABLE;
1112 * check whether node_states[N_NORMAL_MEMORY] will be changed.
1113 * If the memory to be offline is in a zone of 0...zone_last,
1114 * and it is the last present memory, 0...zone_last will
1115 * become empty after offline , thus we can determind we will
1116 * need to clear the node from node_states[N_NORMAL_MEMORY].
1118 for (zt = 0; zt <= zone_last; zt++)
1119 present_pages += pgdat->node_zones[zt].present_pages;
1120 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1121 arg->status_change_nid_normal = zone_to_nid(zone);
1123 arg->status_change_nid_normal = -1;
1126 * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE
1128 zone_last = ZONE_MOVABLE;
1131 * check whether node_states[N_HIGH_MEMORY] will be changed
1132 * If we try to offline the last present @nr_pages from the node,
1133 * we can determind we will need to clear the node from
1134 * node_states[N_HIGH_MEMORY].
1136 for (; zt <= zone_last; zt++)
1137 present_pages += pgdat->node_zones[zt].present_pages;
1138 if (nr_pages >= present_pages)
1139 arg->status_change_nid = zone_to_nid(zone);
1141 arg->status_change_nid = -1;
1144 static void node_states_clear_node(int node, struct memory_notify *arg)
1146 if (arg->status_change_nid_normal >= 0)
1147 node_clear_state(node, N_NORMAL_MEMORY);
1149 if ((N_HIGH_MEMORY != N_NORMAL_MEMORY) &&
1150 (arg->status_change_nid >= 0))
1151 node_clear_state(node, N_HIGH_MEMORY);
1154 static int __ref __offline_pages(unsigned long start_pfn,
1155 unsigned long end_pfn, unsigned long timeout)
1157 unsigned long pfn, nr_pages, expire;
1158 long offlined_pages;
1159 int ret, drain, retry_max, node;
1161 struct memory_notify arg;
1163 BUG_ON(start_pfn >= end_pfn);
1164 /* at least, alignment against pageblock is necessary */
1165 if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
1167 if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
1169 /* This makes hotplug much easier...and readable.
1170 we assume this for now. .*/
1171 if (!test_pages_in_a_zone(start_pfn, end_pfn))
1174 lock_memory_hotplug();
1176 zone = page_zone(pfn_to_page(start_pfn));
1177 node = zone_to_nid(zone);
1178 nr_pages = end_pfn - start_pfn;
1181 if (zone_idx(zone) <= ZONE_NORMAL && !can_offline_normal(zone, nr_pages))
1184 /* set above range as isolated */
1185 ret = start_isolate_page_range(start_pfn, end_pfn,
1186 MIGRATE_MOVABLE, true);
1190 arg.start_pfn = start_pfn;
1191 arg.nr_pages = nr_pages;
1192 node_states_check_changes_offline(nr_pages, zone, &arg);
1194 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1195 ret = notifier_to_errno(ret);
1197 goto failed_removal;
1200 expire = jiffies + timeout;
1204 /* start memory hot removal */
1206 if (time_after(jiffies, expire))
1207 goto failed_removal;
1209 if (signal_pending(current))
1210 goto failed_removal;
1213 lru_add_drain_all();
1218 pfn = scan_lru_pages(start_pfn, end_pfn);
1219 if (pfn) { /* We have page on LRU */
1220 ret = do_migrate_range(pfn, end_pfn);
1226 if (--retry_max == 0)
1227 goto failed_removal;
1233 /* drain all zone's lru pagevec, this is asyncronous... */
1234 lru_add_drain_all();
1236 /* drain pcp pages , this is synchrouns. */
1239 offlined_pages = check_pages_isolated(start_pfn, end_pfn);
1240 if (offlined_pages < 0) {
1242 goto failed_removal;
1244 printk(KERN_INFO "Offlined Pages %ld\n", offlined_pages);
1245 /* Ok, all of our target is islaoted.
1246 We cannot do rollback at this point. */
1247 offline_isolated_pages(start_pfn, end_pfn);
1248 /* reset pagetype flags and makes migrate type to be MOVABLE */
1249 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1250 /* removal success */
1251 zone->present_pages -= offlined_pages;
1252 zone->zone_pgdat->node_present_pages -= offlined_pages;
1253 totalram_pages -= offlined_pages;
1255 init_per_zone_wmark_min();
1257 if (!populated_zone(zone)) {
1258 zone_pcp_reset(zone);
1259 mutex_lock(&zonelists_mutex);
1260 build_all_zonelists(NULL, NULL);
1261 mutex_unlock(&zonelists_mutex);
1263 zone_pcp_update(zone);
1265 node_states_clear_node(node, &arg);
1266 if (arg.status_change_nid >= 0)
1269 vm_total_pages = nr_free_pagecache_pages();
1270 writeback_set_ratelimit();
1272 memory_notify(MEM_OFFLINE, &arg);
1273 unlock_memory_hotplug();
1277 printk(KERN_INFO "memory offlining [mem %#010llx-%#010llx] failed\n",
1278 (unsigned long long) start_pfn << PAGE_SHIFT,
1279 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
1280 memory_notify(MEM_CANCEL_OFFLINE, &arg);
1281 /* pushback to free area */
1282 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1285 unlock_memory_hotplug();
1289 int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1291 return __offline_pages(start_pfn, start_pfn + nr_pages, 120 * HZ);
1294 int remove_memory(u64 start, u64 size)
1296 struct memory_block *mem = NULL;
1297 struct mem_section *section;
1298 unsigned long start_pfn, end_pfn;
1299 unsigned long pfn, section_nr;
1302 start_pfn = PFN_DOWN(start);
1303 end_pfn = start_pfn + PFN_DOWN(size);
1305 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1306 section_nr = pfn_to_section_nr(pfn);
1307 if (!present_section_nr(section_nr))
1310 section = __nr_to_section(section_nr);
1311 /* same memblock? */
1313 if ((section_nr >= mem->start_section_nr) &&
1314 (section_nr <= mem->end_section_nr))
1317 mem = find_memory_block_hinted(section, mem);
1321 ret = offline_memory_block(mem);
1323 kobject_put(&mem->dev.kobj);
1329 kobject_put(&mem->dev.kobj);
1334 int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1338 int remove_memory(u64 start, u64 size)
1342 #endif /* CONFIG_MEMORY_HOTREMOVE */
1343 EXPORT_SYMBOL_GPL(remove_memory);