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/module.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>
31 #include <asm/tlbflush.h>
35 /* add this memory to iomem resource */
36 static struct resource *register_memory_resource(u64 start, u64 size)
39 res = kzalloc(sizeof(struct resource), GFP_KERNEL);
42 res->name = "System RAM";
44 res->end = start + size - 1;
45 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
46 if (request_resource(&iomem_resource, res) < 0) {
47 printk("System RAM resource %llx - %llx cannot be added\n",
48 (unsigned long long)res->start, (unsigned long long)res->end);
55 static void release_memory_resource(struct resource *res)
59 release_resource(res);
64 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
65 #ifndef CONFIG_SPARSEMEM_VMEMMAP
66 static void get_page_bootmem(unsigned long info, struct page *page, int type)
68 atomic_set(&page->_mapcount, type);
70 set_page_private(page, info);
71 atomic_inc(&page->_count);
74 void put_page_bootmem(struct page *page)
78 type = atomic_read(&page->_mapcount);
81 if (atomic_dec_return(&page->_count) == 1) {
82 ClearPagePrivate(page);
83 set_page_private(page, 0);
84 reset_page_mapcount(page);
85 __free_pages_bootmem(page, 0);
90 static void register_page_bootmem_info_section(unsigned long start_pfn)
92 unsigned long *usemap, mapsize, section_nr, i;
93 struct mem_section *ms;
94 struct page *page, *memmap;
96 if (!pfn_valid(start_pfn))
99 section_nr = pfn_to_section_nr(start_pfn);
100 ms = __nr_to_section(section_nr);
102 /* Get section's memmap address */
103 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
106 * Get page for the memmap's phys address
107 * XXX: need more consideration for sparse_vmemmap...
109 page = virt_to_page(memmap);
110 mapsize = sizeof(struct page) * PAGES_PER_SECTION;
111 mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
113 /* remember memmap's page */
114 for (i = 0; i < mapsize; i++, page++)
115 get_page_bootmem(section_nr, page, SECTION_INFO);
117 usemap = __nr_to_section(section_nr)->pageblock_flags;
118 page = virt_to_page(usemap);
120 mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
122 for (i = 0; i < mapsize; i++, page++)
123 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
127 void register_page_bootmem_info_node(struct pglist_data *pgdat)
129 unsigned long i, pfn, end_pfn, nr_pages;
130 int node = pgdat->node_id;
134 nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
135 page = virt_to_page(pgdat);
137 for (i = 0; i < nr_pages; i++, page++)
138 get_page_bootmem(node, page, NODE_INFO);
140 zone = &pgdat->node_zones[0];
141 for (; zone < pgdat->node_zones + MAX_NR_ZONES - 1; zone++) {
142 if (zone->wait_table) {
143 nr_pages = zone->wait_table_hash_nr_entries
144 * sizeof(wait_queue_head_t);
145 nr_pages = PAGE_ALIGN(nr_pages) >> PAGE_SHIFT;
146 page = virt_to_page(zone->wait_table);
148 for (i = 0; i < nr_pages; i++, page++)
149 get_page_bootmem(node, page, NODE_INFO);
153 pfn = pgdat->node_start_pfn;
154 end_pfn = pfn + pgdat->node_spanned_pages;
156 /* register_section info */
157 for (; pfn < end_pfn; pfn += PAGES_PER_SECTION)
158 register_page_bootmem_info_section(pfn);
161 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
163 static void grow_zone_span(struct zone *zone, unsigned long start_pfn,
164 unsigned long end_pfn)
166 unsigned long old_zone_end_pfn;
168 zone_span_writelock(zone);
170 old_zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages;
171 if (start_pfn < zone->zone_start_pfn)
172 zone->zone_start_pfn = start_pfn;
174 zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
175 zone->zone_start_pfn;
177 zone_span_writeunlock(zone);
180 static void grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn,
181 unsigned long end_pfn)
183 unsigned long old_pgdat_end_pfn =
184 pgdat->node_start_pfn + pgdat->node_spanned_pages;
186 if (start_pfn < pgdat->node_start_pfn)
187 pgdat->node_start_pfn = start_pfn;
189 pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
190 pgdat->node_start_pfn;
193 static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn)
195 struct pglist_data *pgdat = zone->zone_pgdat;
196 int nr_pages = PAGES_PER_SECTION;
197 int nid = pgdat->node_id;
201 zone_type = zone - pgdat->node_zones;
202 if (!zone->wait_table) {
205 ret = init_currently_empty_zone(zone, phys_start_pfn,
206 nr_pages, MEMMAP_HOTPLUG);
210 pgdat_resize_lock(zone->zone_pgdat, &flags);
211 grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages);
212 grow_pgdat_span(zone->zone_pgdat, phys_start_pfn,
213 phys_start_pfn + nr_pages);
214 pgdat_resize_unlock(zone->zone_pgdat, &flags);
215 memmap_init_zone(nr_pages, nid, zone_type,
216 phys_start_pfn, MEMMAP_HOTPLUG);
220 static int __meminit __add_section(int nid, struct zone *zone,
221 unsigned long phys_start_pfn)
223 int nr_pages = PAGES_PER_SECTION;
226 if (pfn_valid(phys_start_pfn))
229 ret = sparse_add_one_section(zone, phys_start_pfn, nr_pages);
234 ret = __add_zone(zone, phys_start_pfn);
239 return register_new_memory(nid, __pfn_to_section(phys_start_pfn));
242 #ifdef CONFIG_SPARSEMEM_VMEMMAP
243 static int __remove_section(struct zone *zone, struct mem_section *ms)
246 * XXX: Freeing memmap with vmemmap is not implement yet.
247 * This should be removed later.
252 static int __remove_section(struct zone *zone, struct mem_section *ms)
255 struct pglist_data *pgdat = zone->zone_pgdat;
258 if (!valid_section(ms))
261 ret = unregister_memory_section(ms);
265 pgdat_resize_lock(pgdat, &flags);
266 sparse_remove_one_section(zone, ms);
267 pgdat_resize_unlock(pgdat, &flags);
273 * Reasonably generic function for adding memory. It is
274 * expected that archs that support memory hotplug will
275 * call this function after deciding the zone to which to
278 int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn,
279 unsigned long nr_pages)
283 int start_sec, end_sec;
284 /* during initialize mem_map, align hot-added range to section */
285 start_sec = pfn_to_section_nr(phys_start_pfn);
286 end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
288 for (i = start_sec; i <= end_sec; i++) {
289 err = __add_section(nid, zone, i << PFN_SECTION_SHIFT);
292 * EEXIST is finally dealt with by ioresource collision
293 * check. see add_memory() => register_memory_resource()
294 * Warning will be printed if there is collision.
296 if (err && (err != -EEXIST))
303 EXPORT_SYMBOL_GPL(__add_pages);
306 * __remove_pages() - remove sections of pages from a zone
307 * @zone: zone from which pages need to be removed
308 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
309 * @nr_pages: number of pages to remove (must be multiple of section size)
311 * Generic helper function to remove section mappings and sysfs entries
312 * for the section of the memory we are removing. Caller needs to make
313 * sure that pages are marked reserved and zones are adjust properly by
314 * calling offline_pages().
316 int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
317 unsigned long nr_pages)
319 unsigned long i, ret = 0;
320 int sections_to_remove;
323 * We can only remove entire sections
325 BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
326 BUG_ON(nr_pages % PAGES_PER_SECTION);
328 sections_to_remove = nr_pages / PAGES_PER_SECTION;
329 for (i = 0; i < sections_to_remove; i++) {
330 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
331 release_mem_region(pfn << PAGE_SHIFT,
332 PAGES_PER_SECTION << PAGE_SHIFT);
333 ret = __remove_section(zone, __pfn_to_section(pfn));
339 EXPORT_SYMBOL_GPL(__remove_pages);
341 void online_page(struct page *page)
343 unsigned long pfn = page_to_pfn(page);
346 if (pfn >= num_physpages)
347 num_physpages = pfn + 1;
349 #ifdef CONFIG_HIGHMEM
350 if (PageHighMem(page))
354 #ifdef CONFIG_FLATMEM
355 max_mapnr = max(page_to_pfn(page), max_mapnr);
358 ClearPageReserved(page);
359 init_page_count(page);
363 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
367 unsigned long onlined_pages = *(unsigned long *)arg;
369 if (PageReserved(pfn_to_page(start_pfn)))
370 for (i = 0; i < nr_pages; i++) {
371 page = pfn_to_page(start_pfn + i);
375 *(unsigned long *)arg = onlined_pages;
380 int online_pages(unsigned long pfn, unsigned long nr_pages)
382 unsigned long onlined_pages = 0;
384 int need_zonelists_rebuild = 0;
387 struct memory_notify arg;
390 arg.nr_pages = nr_pages;
391 arg.status_change_nid = -1;
393 nid = page_to_nid(pfn_to_page(pfn));
394 if (node_present_pages(nid) == 0)
395 arg.status_change_nid = nid;
397 ret = memory_notify(MEM_GOING_ONLINE, &arg);
398 ret = notifier_to_errno(ret);
400 memory_notify(MEM_CANCEL_ONLINE, &arg);
404 * This doesn't need a lock to do pfn_to_page().
405 * The section can't be removed here because of the
406 * memory_block->state_mutex.
408 zone = page_zone(pfn_to_page(pfn));
410 * If this zone is not populated, then it is not in zonelist.
411 * This means the page allocator ignores this zone.
412 * So, zonelist must be updated after online.
414 if (!populated_zone(zone))
415 need_zonelists_rebuild = 1;
417 ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
420 printk(KERN_DEBUG "online_pages %lx at %lx failed\n",
422 memory_notify(MEM_CANCEL_ONLINE, &arg);
426 zone->present_pages += onlined_pages;
427 zone->zone_pgdat->node_present_pages += onlined_pages;
429 zone_pcp_update(zone);
430 setup_per_zone_wmarks();
431 calculate_zone_inactive_ratio(zone);
433 kswapd_run(zone_to_nid(zone));
434 node_set_state(zone_to_nid(zone), N_HIGH_MEMORY);
437 if (need_zonelists_rebuild)
438 build_all_zonelists();
440 vm_total_pages = nr_free_pagecache_pages();
442 writeback_set_ratelimit();
445 memory_notify(MEM_ONLINE, &arg);
449 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
451 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
452 static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
454 struct pglist_data *pgdat;
455 unsigned long zones_size[MAX_NR_ZONES] = {0};
456 unsigned long zholes_size[MAX_NR_ZONES] = {0};
457 unsigned long start_pfn = start >> PAGE_SHIFT;
459 pgdat = arch_alloc_nodedata(nid);
463 arch_refresh_nodedata(nid, pgdat);
465 /* we can use NODE_DATA(nid) from here */
467 /* init node's zones as empty zones, we don't have any present pages.*/
468 free_area_init_node(nid, zones_size, start_pfn, zholes_size);
473 static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
475 arch_refresh_nodedata(nid, NULL);
476 arch_free_nodedata(pgdat);
481 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
482 int __ref add_memory(int nid, u64 start, u64 size)
484 pg_data_t *pgdat = NULL;
486 struct resource *res;
491 res = register_memory_resource(start, size);
496 if (!node_online(nid)) {
497 pgdat = hotadd_new_pgdat(nid, start);
504 /* call arch's memory hotadd */
505 ret = arch_add_memory(nid, start, size);
510 /* we online node here. we can't roll back from here. */
511 node_set_online(nid);
514 ret = register_one_node(nid);
516 * If sysfs file of new node can't create, cpu on the node
517 * can't be hot-added. There is no rollback way now.
518 * So, check by BUG_ON() to catch it reluctantly..
526 /* rollback pgdat allocation and others */
528 rollback_node_hotadd(nid, pgdat);
530 release_memory_resource(res);
533 unlock_system_sleep();
536 EXPORT_SYMBOL_GPL(add_memory);
538 #ifdef CONFIG_MEMORY_HOTREMOVE
540 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
541 * set and the size of the free page is given by page_order(). Using this,
542 * the function determines if the pageblock contains only free pages.
543 * Due to buddy contraints, a free page at least the size of a pageblock will
544 * be located at the start of the pageblock
546 static inline int pageblock_free(struct page *page)
548 return PageBuddy(page) && page_order(page) >= pageblock_order;
551 /* Return the start of the next active pageblock after a given page */
552 static struct page *next_active_pageblock(struct page *page)
554 int pageblocks_stride;
556 /* Ensure the starting page is pageblock-aligned */
557 BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
559 /* Move forward by at least 1 * pageblock_nr_pages */
560 pageblocks_stride = 1;
562 /* If the entire pageblock is free, move to the end of free page */
563 if (pageblock_free(page))
564 pageblocks_stride += page_order(page) - pageblock_order;
566 return page + (pageblocks_stride * pageblock_nr_pages);
569 /* Checks if this range of memory is likely to be hot-removable. */
570 int is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
573 struct page *page = pfn_to_page(start_pfn);
574 struct page *end_page = page + nr_pages;
576 /* Check the starting page of each pageblock within the range */
577 for (; page < end_page; page = next_active_pageblock(page)) {
578 type = get_pageblock_migratetype(page);
581 * A pageblock containing MOVABLE or free pages is considered
584 if (type != MIGRATE_MOVABLE && !pageblock_free(page))
588 * A pageblock starting with a PageReserved page is not
589 * considered removable.
591 if (PageReserved(page))
595 /* All pageblocks in the memory block are likely to be hot-removable */
600 * Confirm all pages in a range [start, end) is belongs to the same zone.
602 static int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn)
605 struct zone *zone = NULL;
608 for (pfn = start_pfn;
610 pfn += MAX_ORDER_NR_PAGES) {
612 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
613 while ((i < MAX_ORDER_NR_PAGES) && !pfn_valid_within(pfn + i))
615 if (i == MAX_ORDER_NR_PAGES)
617 page = pfn_to_page(pfn + i);
618 if (zone && page_zone(page) != zone)
620 zone = page_zone(page);
626 * Scanning pfn is much easier than scanning lru list.
627 * Scan pfn from start to end and Find LRU page.
629 int scan_lru_pages(unsigned long start, unsigned long end)
633 for (pfn = start; pfn < end; pfn++) {
634 if (pfn_valid(pfn)) {
635 page = pfn_to_page(pfn);
644 hotremove_migrate_alloc(struct page *page, unsigned long private, int **x)
646 /* This should be improooooved!! */
647 return alloc_page(GFP_HIGHUSER_MOVABLE);
650 #define NR_OFFLINE_AT_ONCE_PAGES (256)
652 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
656 int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
661 for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
664 page = pfn_to_page(pfn);
665 if (!page_count(page))
668 * We can skip free pages. And we can only deal with pages on
671 ret = isolate_lru_page(page);
672 if (!ret) { /* Success */
673 list_add_tail(&page->lru, &source);
676 /* Becasue we don't have big zone->lock. we should
677 check this again here. */
678 if (page_count(page))
680 #ifdef CONFIG_DEBUG_VM
681 printk(KERN_INFO "removing from LRU failed"
683 pfn, page_count(page), page->flags);
689 if (!list_empty(&source))
690 putback_lru_pages(&source);
694 if (list_empty(&source))
696 /* this function returns # of failed pages */
697 ret = migrate_pages(&source, hotremove_migrate_alloc, 0);
704 * remove from free_area[] and mark all as Reserved.
707 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
710 __offline_isolated_pages(start, start + nr_pages);
715 offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
717 walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
718 offline_isolated_pages_cb);
722 * Check all pages in range, recoreded as memory resource, are isolated.
725 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
729 long offlined = *(long *)data;
730 ret = test_pages_isolated(start_pfn, start_pfn + nr_pages);
733 *(long *)data += offlined;
738 check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
743 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
744 check_pages_isolated_cb);
746 offlined = (long)ret;
750 int offline_pages(unsigned long start_pfn,
751 unsigned long end_pfn, unsigned long timeout)
753 unsigned long pfn, nr_pages, expire;
755 int ret, drain, retry_max, node;
757 struct memory_notify arg;
759 BUG_ON(start_pfn >= end_pfn);
760 /* at least, alignment against pageblock is necessary */
761 if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
763 if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
765 /* This makes hotplug much easier...and readable.
766 we assume this for now. .*/
767 if (!test_pages_in_a_zone(start_pfn, end_pfn))
772 zone = page_zone(pfn_to_page(start_pfn));
773 node = zone_to_nid(zone);
774 nr_pages = end_pfn - start_pfn;
776 /* set above range as isolated */
777 ret = start_isolate_page_range(start_pfn, end_pfn);
781 arg.start_pfn = start_pfn;
782 arg.nr_pages = nr_pages;
783 arg.status_change_nid = -1;
784 if (nr_pages >= node_present_pages(node))
785 arg.status_change_nid = node;
787 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
788 ret = notifier_to_errno(ret);
793 expire = jiffies + timeout;
797 /* start memory hot removal */
799 if (time_after(jiffies, expire))
802 if (signal_pending(current))
807 flush_scheduled_work();
812 pfn = scan_lru_pages(start_pfn, end_pfn);
813 if (pfn) { /* We have page on LRU */
814 ret = do_migrate_range(pfn, end_pfn);
820 if (--retry_max == 0)
827 /* drain all zone's lru pagevec, this is asyncronous... */
829 flush_scheduled_work();
831 /* drain pcp pages , this is synchrouns. */
834 offlined_pages = check_pages_isolated(start_pfn, end_pfn);
835 if (offlined_pages < 0) {
839 printk(KERN_INFO "Offlined Pages %ld\n", offlined_pages);
840 /* Ok, all of our target is islaoted.
841 We cannot do rollback at this point. */
842 offline_isolated_pages(start_pfn, end_pfn);
843 /* reset pagetype flags and makes migrate type to be MOVABLE */
844 undo_isolate_page_range(start_pfn, end_pfn);
845 /* removal success */
846 zone->present_pages -= offlined_pages;
847 zone->zone_pgdat->node_present_pages -= offlined_pages;
848 totalram_pages -= offlined_pages;
850 setup_per_zone_wmarks();
851 calculate_zone_inactive_ratio(zone);
853 vm_total_pages = nr_free_pagecache_pages();
854 writeback_set_ratelimit();
856 memory_notify(MEM_OFFLINE, &arg);
857 unlock_system_sleep();
861 printk(KERN_INFO "memory offlining %lx to %lx failed\n",
863 memory_notify(MEM_CANCEL_OFFLINE, &arg);
864 /* pushback to free area */
865 undo_isolate_page_range(start_pfn, end_pfn);
868 unlock_system_sleep();
872 int remove_memory(u64 start, u64 size)
874 unsigned long start_pfn, end_pfn;
876 start_pfn = PFN_DOWN(start);
877 end_pfn = start_pfn + PFN_DOWN(size);
878 return offline_pages(start_pfn, end_pfn, 120 * HZ);
881 int remove_memory(u64 start, u64 size)
885 #endif /* CONFIG_MEMORY_HOTREMOVE */
886 EXPORT_SYMBOL_GPL(remove_memory);