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 grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn,
218 unsigned long end_pfn)
220 unsigned long old_pgdat_end_pfn =
221 pgdat->node_start_pfn + pgdat->node_spanned_pages;
223 if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
224 pgdat->node_start_pfn = start_pfn;
226 pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
227 pgdat->node_start_pfn;
230 static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn)
232 struct pglist_data *pgdat = zone->zone_pgdat;
233 int nr_pages = PAGES_PER_SECTION;
234 int nid = pgdat->node_id;
238 zone_type = zone - pgdat->node_zones;
239 if (!zone->wait_table) {
242 ret = init_currently_empty_zone(zone, phys_start_pfn,
243 nr_pages, MEMMAP_HOTPLUG);
247 pgdat_resize_lock(zone->zone_pgdat, &flags);
248 grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages);
249 grow_pgdat_span(zone->zone_pgdat, phys_start_pfn,
250 phys_start_pfn + nr_pages);
251 pgdat_resize_unlock(zone->zone_pgdat, &flags);
252 memmap_init_zone(nr_pages, nid, zone_type,
253 phys_start_pfn, MEMMAP_HOTPLUG);
257 static int __meminit __add_section(int nid, struct zone *zone,
258 unsigned long phys_start_pfn)
260 int nr_pages = PAGES_PER_SECTION;
263 if (pfn_valid(phys_start_pfn))
266 ret = sparse_add_one_section(zone, phys_start_pfn, nr_pages);
271 ret = __add_zone(zone, phys_start_pfn);
276 return register_new_memory(nid, __pfn_to_section(phys_start_pfn));
279 #ifdef CONFIG_SPARSEMEM_VMEMMAP
280 static int __remove_section(struct zone *zone, struct mem_section *ms)
283 * XXX: Freeing memmap with vmemmap is not implement yet.
284 * This should be removed later.
289 static int __remove_section(struct zone *zone, struct mem_section *ms)
292 struct pglist_data *pgdat = zone->zone_pgdat;
295 if (!valid_section(ms))
298 ret = unregister_memory_section(ms);
302 pgdat_resize_lock(pgdat, &flags);
303 sparse_remove_one_section(zone, ms);
304 pgdat_resize_unlock(pgdat, &flags);
310 * Reasonably generic function for adding memory. It is
311 * expected that archs that support memory hotplug will
312 * call this function after deciding the zone to which to
315 int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn,
316 unsigned long nr_pages)
320 int start_sec, end_sec;
321 /* during initialize mem_map, align hot-added range to section */
322 start_sec = pfn_to_section_nr(phys_start_pfn);
323 end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
325 for (i = start_sec; i <= end_sec; i++) {
326 err = __add_section(nid, zone, i << PFN_SECTION_SHIFT);
329 * EEXIST is finally dealt with by ioresource collision
330 * check. see add_memory() => register_memory_resource()
331 * Warning will be printed if there is collision.
333 if (err && (err != -EEXIST))
340 EXPORT_SYMBOL_GPL(__add_pages);
343 * __remove_pages() - remove sections of pages from a zone
344 * @zone: zone from which pages need to be removed
345 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
346 * @nr_pages: number of pages to remove (must be multiple of section size)
348 * Generic helper function to remove section mappings and sysfs entries
349 * for the section of the memory we are removing. Caller needs to make
350 * sure that pages are marked reserved and zones are adjust properly by
351 * calling offline_pages().
353 int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
354 unsigned long nr_pages)
356 unsigned long i, ret = 0;
357 int sections_to_remove;
360 * We can only remove entire sections
362 BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
363 BUG_ON(nr_pages % PAGES_PER_SECTION);
365 release_mem_region(phys_start_pfn << PAGE_SHIFT, nr_pages * PAGE_SIZE);
367 sections_to_remove = nr_pages / PAGES_PER_SECTION;
368 for (i = 0; i < sections_to_remove; i++) {
369 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
370 ret = __remove_section(zone, __pfn_to_section(pfn));
376 EXPORT_SYMBOL_GPL(__remove_pages);
378 int set_online_page_callback(online_page_callback_t callback)
382 lock_memory_hotplug();
384 if (online_page_callback == generic_online_page) {
385 online_page_callback = callback;
389 unlock_memory_hotplug();
393 EXPORT_SYMBOL_GPL(set_online_page_callback);
395 int restore_online_page_callback(online_page_callback_t callback)
399 lock_memory_hotplug();
401 if (online_page_callback == callback) {
402 online_page_callback = generic_online_page;
406 unlock_memory_hotplug();
410 EXPORT_SYMBOL_GPL(restore_online_page_callback);
412 void __online_page_set_limits(struct page *page)
414 unsigned long pfn = page_to_pfn(page);
416 if (pfn >= num_physpages)
417 num_physpages = pfn + 1;
419 EXPORT_SYMBOL_GPL(__online_page_set_limits);
421 void __online_page_increment_counters(struct page *page)
425 #ifdef CONFIG_HIGHMEM
426 if (PageHighMem(page))
430 EXPORT_SYMBOL_GPL(__online_page_increment_counters);
432 void __online_page_free(struct page *page)
434 ClearPageReserved(page);
435 init_page_count(page);
438 EXPORT_SYMBOL_GPL(__online_page_free);
440 static void generic_online_page(struct page *page)
442 __online_page_set_limits(page);
443 __online_page_increment_counters(page);
444 __online_page_free(page);
447 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
451 unsigned long onlined_pages = *(unsigned long *)arg;
453 if (PageReserved(pfn_to_page(start_pfn)))
454 for (i = 0; i < nr_pages; i++) {
455 page = pfn_to_page(start_pfn + i);
456 (*online_page_callback)(page);
459 *(unsigned long *)arg = onlined_pages;
463 /* check which state of node_states will be changed when online memory */
464 static void node_states_check_changes_online(unsigned long nr_pages,
465 struct zone *zone, struct memory_notify *arg)
467 int nid = zone_to_nid(zone);
468 enum zone_type zone_last = ZONE_NORMAL;
471 * If we have HIGHMEM, node_states[N_NORMAL_MEMORY] contains nodes
472 * which have 0...ZONE_NORMAL, set zone_last to ZONE_NORMAL.
474 * If we don't have HIGHMEM, node_states[N_NORMAL_MEMORY] contains nodes
475 * which have 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
477 if (N_HIGH_MEMORY == N_NORMAL_MEMORY)
478 zone_last = ZONE_MOVABLE;
481 * if the memory to be online is in a zone of 0...zone_last, and
482 * the zones of 0...zone_last don't have memory before online, we will
483 * need to set the node to node_states[N_NORMAL_MEMORY] after
484 * the memory is online.
486 if (zone_idx(zone) <= zone_last && !node_state(nid, N_NORMAL_MEMORY))
487 arg->status_change_nid_normal = nid;
489 arg->status_change_nid_normal = -1;
492 * if the node don't have memory befor online, we will need to
493 * set the node to node_states[N_HIGH_MEMORY] after the memory
496 if (!node_state(nid, N_HIGH_MEMORY))
497 arg->status_change_nid = nid;
499 arg->status_change_nid = -1;
502 static void node_states_set_node(int node, struct memory_notify *arg)
504 if (arg->status_change_nid_normal >= 0)
505 node_set_state(node, N_NORMAL_MEMORY);
507 node_set_state(node, N_HIGH_MEMORY);
511 int __ref online_pages(unsigned long pfn, unsigned long nr_pages)
513 unsigned long onlined_pages = 0;
515 int need_zonelists_rebuild = 0;
518 struct memory_notify arg;
520 lock_memory_hotplug();
522 * This doesn't need a lock to do pfn_to_page().
523 * The section can't be removed here because of the
524 * memory_block->state_mutex.
526 zone = page_zone(pfn_to_page(pfn));
529 arg.nr_pages = nr_pages;
530 node_states_check_changes_online(nr_pages, zone, &arg);
532 nid = page_to_nid(pfn_to_page(pfn));
534 ret = memory_notify(MEM_GOING_ONLINE, &arg);
535 ret = notifier_to_errno(ret);
537 memory_notify(MEM_CANCEL_ONLINE, &arg);
538 unlock_memory_hotplug();
542 * If this zone is not populated, then it is not in zonelist.
543 * This means the page allocator ignores this zone.
544 * So, zonelist must be updated after online.
546 mutex_lock(&zonelists_mutex);
547 if (!populated_zone(zone)) {
548 need_zonelists_rebuild = 1;
549 build_all_zonelists(NULL, zone);
552 ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
555 if (need_zonelists_rebuild)
556 zone_pcp_reset(zone);
557 mutex_unlock(&zonelists_mutex);
558 printk(KERN_DEBUG "online_pages [mem %#010llx-%#010llx] failed\n",
559 (unsigned long long) pfn << PAGE_SHIFT,
560 (((unsigned long long) pfn + nr_pages)
562 memory_notify(MEM_CANCEL_ONLINE, &arg);
563 unlock_memory_hotplug();
567 zone->present_pages += onlined_pages;
568 zone->zone_pgdat->node_present_pages += onlined_pages;
570 node_states_set_node(zone_to_nid(zone), &arg);
571 if (need_zonelists_rebuild)
572 build_all_zonelists(NULL, NULL);
574 zone_pcp_update(zone);
577 mutex_unlock(&zonelists_mutex);
579 init_per_zone_wmark_min();
582 kswapd_run(zone_to_nid(zone));
584 vm_total_pages = nr_free_pagecache_pages();
586 writeback_set_ratelimit();
589 memory_notify(MEM_ONLINE, &arg);
590 unlock_memory_hotplug();
594 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
596 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
597 static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
599 struct pglist_data *pgdat;
600 unsigned long zones_size[MAX_NR_ZONES] = {0};
601 unsigned long zholes_size[MAX_NR_ZONES] = {0};
602 unsigned long start_pfn = start >> PAGE_SHIFT;
604 pgdat = arch_alloc_nodedata(nid);
608 arch_refresh_nodedata(nid, pgdat);
610 /* we can use NODE_DATA(nid) from here */
612 /* init node's zones as empty zones, we don't have any present pages.*/
613 free_area_init_node(nid, zones_size, start_pfn, zholes_size);
616 * The node we allocated has no zone fallback lists. For avoiding
617 * to access not-initialized zonelist, build here.
619 mutex_lock(&zonelists_mutex);
620 build_all_zonelists(pgdat, NULL);
621 mutex_unlock(&zonelists_mutex);
626 static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
628 arch_refresh_nodedata(nid, NULL);
629 arch_free_nodedata(pgdat);
635 * called by cpu_up() to online a node without onlined memory.
637 int mem_online_node(int nid)
642 lock_memory_hotplug();
643 pgdat = hotadd_new_pgdat(nid, 0);
648 node_set_online(nid);
649 ret = register_one_node(nid);
653 unlock_memory_hotplug();
657 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
658 int __ref add_memory(int nid, u64 start, u64 size)
660 pg_data_t *pgdat = NULL;
662 struct resource *res;
665 lock_memory_hotplug();
667 res = register_memory_resource(start, size);
672 if (!node_online(nid)) {
673 pgdat = hotadd_new_pgdat(nid, start);
680 /* call arch's memory hotadd */
681 ret = arch_add_memory(nid, start, size);
686 /* we online node here. we can't roll back from here. */
687 node_set_online(nid);
690 ret = register_one_node(nid);
692 * If sysfs file of new node can't create, cpu on the node
693 * can't be hot-added. There is no rollback way now.
694 * So, check by BUG_ON() to catch it reluctantly..
699 /* create new memmap entry */
700 firmware_map_add_hotplug(start, start + size, "System RAM");
705 /* rollback pgdat allocation and others */
707 rollback_node_hotadd(nid, pgdat);
709 release_memory_resource(res);
712 unlock_memory_hotplug();
715 EXPORT_SYMBOL_GPL(add_memory);
717 #ifdef CONFIG_MEMORY_HOTREMOVE
719 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
720 * set and the size of the free page is given by page_order(). Using this,
721 * the function determines if the pageblock contains only free pages.
722 * Due to buddy contraints, a free page at least the size of a pageblock will
723 * be located at the start of the pageblock
725 static inline int pageblock_free(struct page *page)
727 return PageBuddy(page) && page_order(page) >= pageblock_order;
730 /* Return the start of the next active pageblock after a given page */
731 static struct page *next_active_pageblock(struct page *page)
733 /* Ensure the starting page is pageblock-aligned */
734 BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
736 /* If the entire pageblock is free, move to the end of free page */
737 if (pageblock_free(page)) {
739 /* be careful. we don't have locks, page_order can be changed.*/
740 order = page_order(page);
741 if ((order < MAX_ORDER) && (order >= pageblock_order))
742 return page + (1 << order);
745 return page + pageblock_nr_pages;
748 /* Checks if this range of memory is likely to be hot-removable. */
749 int is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
751 struct page *page = pfn_to_page(start_pfn);
752 struct page *end_page = page + nr_pages;
754 /* Check the starting page of each pageblock within the range */
755 for (; page < end_page; page = next_active_pageblock(page)) {
756 if (!is_pageblock_removable_nolock(page))
761 /* All pageblocks in the memory block are likely to be hot-removable */
766 * Confirm all pages in a range [start, end) is belongs to the same zone.
768 static int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn)
771 struct zone *zone = NULL;
774 for (pfn = start_pfn;
776 pfn += MAX_ORDER_NR_PAGES) {
778 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
779 while ((i < MAX_ORDER_NR_PAGES) && !pfn_valid_within(pfn + i))
781 if (i == MAX_ORDER_NR_PAGES)
783 page = pfn_to_page(pfn + i);
784 if (zone && page_zone(page) != zone)
786 zone = page_zone(page);
792 * Scanning pfn is much easier than scanning lru list.
793 * Scan pfn from start to end and Find LRU page.
795 static unsigned long scan_lru_pages(unsigned long start, unsigned long end)
799 for (pfn = start; pfn < end; pfn++) {
800 if (pfn_valid(pfn)) {
801 page = pfn_to_page(pfn);
809 #define NR_OFFLINE_AT_ONCE_PAGES (256)
811 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
815 int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
820 for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
823 page = pfn_to_page(pfn);
824 if (!get_page_unless_zero(page))
827 * We can skip free pages. And we can only deal with pages on
830 ret = isolate_lru_page(page);
831 if (!ret) { /* Success */
833 list_add_tail(&page->lru, &source);
835 inc_zone_page_state(page, NR_ISOLATED_ANON +
836 page_is_file_cache(page));
839 #ifdef CONFIG_DEBUG_VM
840 printk(KERN_ALERT "removing pfn %lx from LRU failed\n",
845 /* Because we don't have big zone->lock. we should
846 check this again here. */
847 if (page_count(page)) {
854 if (!list_empty(&source)) {
856 putback_lru_pages(&source);
861 * alloc_migrate_target should be improooooved!!
862 * migrate_pages returns # of failed pages.
864 ret = migrate_pages(&source, alloc_migrate_target, 0,
867 putback_lru_pages(&source);
874 * remove from free_area[] and mark all as Reserved.
877 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
880 __offline_isolated_pages(start, start + nr_pages);
885 offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
887 walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
888 offline_isolated_pages_cb);
892 * Check all pages in range, recoreded as memory resource, are isolated.
895 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
899 long offlined = *(long *)data;
900 ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true);
903 *(long *)data += offlined;
908 check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
913 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
914 check_pages_isolated_cb);
916 offlined = (long)ret;
920 /* check which state of node_states will be changed when offline memory */
921 static void node_states_check_changes_offline(unsigned long nr_pages,
922 struct zone *zone, struct memory_notify *arg)
924 struct pglist_data *pgdat = zone->zone_pgdat;
925 unsigned long present_pages = 0;
926 enum zone_type zt, zone_last = ZONE_NORMAL;
929 * If we have HIGHMEM, node_states[N_NORMAL_MEMORY] contains nodes
930 * which have 0...ZONE_NORMAL, set zone_last to ZONE_NORMAL.
932 * If we don't have HIGHMEM, node_states[N_NORMAL_MEMORY] contains nodes
933 * which have 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
935 if (N_HIGH_MEMORY == N_NORMAL_MEMORY)
936 zone_last = ZONE_MOVABLE;
939 * check whether node_states[N_NORMAL_MEMORY] will be changed.
940 * If the memory to be offline is in a zone of 0...zone_last,
941 * and it is the last present memory, 0...zone_last will
942 * become empty after offline , thus we can determind we will
943 * need to clear the node from node_states[N_NORMAL_MEMORY].
945 for (zt = 0; zt <= zone_last; zt++)
946 present_pages += pgdat->node_zones[zt].present_pages;
947 if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
948 arg->status_change_nid_normal = zone_to_nid(zone);
950 arg->status_change_nid_normal = -1;
953 * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE
955 zone_last = ZONE_MOVABLE;
958 * check whether node_states[N_HIGH_MEMORY] will be changed
959 * If we try to offline the last present @nr_pages from the node,
960 * we can determind we will need to clear the node from
961 * node_states[N_HIGH_MEMORY].
963 for (; zt <= zone_last; zt++)
964 present_pages += pgdat->node_zones[zt].present_pages;
965 if (nr_pages >= present_pages)
966 arg->status_change_nid = zone_to_nid(zone);
968 arg->status_change_nid = -1;
971 static void node_states_clear_node(int node, struct memory_notify *arg)
973 if (arg->status_change_nid_normal >= 0)
974 node_clear_state(node, N_NORMAL_MEMORY);
976 if ((N_HIGH_MEMORY != N_NORMAL_MEMORY) &&
977 (arg->status_change_nid >= 0))
978 node_clear_state(node, N_HIGH_MEMORY);
981 static int __ref __offline_pages(unsigned long start_pfn,
982 unsigned long end_pfn, unsigned long timeout)
984 unsigned long pfn, nr_pages, expire;
986 int ret, drain, retry_max, node;
988 struct memory_notify arg;
990 BUG_ON(start_pfn >= end_pfn);
991 /* at least, alignment against pageblock is necessary */
992 if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
994 if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
996 /* This makes hotplug much easier...and readable.
997 we assume this for now. .*/
998 if (!test_pages_in_a_zone(start_pfn, end_pfn))
1001 lock_memory_hotplug();
1003 zone = page_zone(pfn_to_page(start_pfn));
1004 node = zone_to_nid(zone);
1005 nr_pages = end_pfn - start_pfn;
1007 /* set above range as isolated */
1008 ret = start_isolate_page_range(start_pfn, end_pfn,
1009 MIGRATE_MOVABLE, true);
1013 arg.start_pfn = start_pfn;
1014 arg.nr_pages = nr_pages;
1015 node_states_check_changes_offline(nr_pages, zone, &arg);
1017 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1018 ret = notifier_to_errno(ret);
1020 goto failed_removal;
1023 expire = jiffies + timeout;
1027 /* start memory hot removal */
1029 if (time_after(jiffies, expire))
1030 goto failed_removal;
1032 if (signal_pending(current))
1033 goto failed_removal;
1036 lru_add_drain_all();
1041 pfn = scan_lru_pages(start_pfn, end_pfn);
1042 if (pfn) { /* We have page on LRU */
1043 ret = do_migrate_range(pfn, end_pfn);
1049 if (--retry_max == 0)
1050 goto failed_removal;
1056 /* drain all zone's lru pagevec, this is asyncronous... */
1057 lru_add_drain_all();
1059 /* drain pcp pages , this is synchrouns. */
1062 offlined_pages = check_pages_isolated(start_pfn, end_pfn);
1063 if (offlined_pages < 0) {
1065 goto failed_removal;
1067 printk(KERN_INFO "Offlined Pages %ld\n", offlined_pages);
1068 /* Ok, all of our target is islaoted.
1069 We cannot do rollback at this point. */
1070 offline_isolated_pages(start_pfn, end_pfn);
1071 /* reset pagetype flags and makes migrate type to be MOVABLE */
1072 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1073 /* removal success */
1074 zone->present_pages -= offlined_pages;
1075 zone->zone_pgdat->node_present_pages -= offlined_pages;
1076 totalram_pages -= offlined_pages;
1078 init_per_zone_wmark_min();
1080 if (!populated_zone(zone)) {
1081 zone_pcp_reset(zone);
1082 mutex_lock(&zonelists_mutex);
1083 build_all_zonelists(NULL, NULL);
1084 mutex_unlock(&zonelists_mutex);
1086 zone_pcp_update(zone);
1088 node_states_clear_node(node, &arg);
1089 if (arg.status_change_nid >= 0)
1092 vm_total_pages = nr_free_pagecache_pages();
1093 writeback_set_ratelimit();
1095 memory_notify(MEM_OFFLINE, &arg);
1096 unlock_memory_hotplug();
1100 printk(KERN_INFO "memory offlining [mem %#010llx-%#010llx] failed\n",
1101 (unsigned long long) start_pfn << PAGE_SHIFT,
1102 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
1103 memory_notify(MEM_CANCEL_OFFLINE, &arg);
1104 /* pushback to free area */
1105 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1108 unlock_memory_hotplug();
1112 int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1114 return __offline_pages(start_pfn, start_pfn + nr_pages, 120 * HZ);
1117 int remove_memory(u64 start, u64 size)
1119 struct memory_block *mem = NULL;
1120 struct mem_section *section;
1121 unsigned long start_pfn, end_pfn;
1122 unsigned long pfn, section_nr;
1125 start_pfn = PFN_DOWN(start);
1126 end_pfn = start_pfn + PFN_DOWN(size);
1128 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1129 section_nr = pfn_to_section_nr(pfn);
1130 if (!present_section_nr(section_nr))
1133 section = __nr_to_section(section_nr);
1134 /* same memblock? */
1136 if ((section_nr >= mem->start_section_nr) &&
1137 (section_nr <= mem->end_section_nr))
1140 mem = find_memory_block_hinted(section, mem);
1144 ret = offline_memory_block(mem);
1146 kobject_put(&mem->dev.kobj);
1152 kobject_put(&mem->dev.kobj);
1157 int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1161 int remove_memory(u64 start, u64 size)
1165 #endif /* CONFIG_MEMORY_HOTREMOVE */
1166 EXPORT_SYMBOL_GPL(remove_memory);