memory_hotplug.c revision 2f7f24eca31c4fc2fdb134b2ef743ccd67cfb9a9
1/*
2 *  linux/mm/memory_hotplug.c
3 *
4 *  Copyright (C)
5 */
6
7#include <linux/stddef.h>
8#include <linux/mm.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/cpuset.h>
26#include <linux/delay.h>
27#include <linux/migrate.h>
28#include <linux/page-isolation.h>
29
30#include <asm/tlbflush.h>
31
32#include "internal.h"
33
34/* add this memory to iomem resource */
35static struct resource *register_memory_resource(u64 start, u64 size)
36{
37	struct resource *res;
38	res = kzalloc(sizeof(struct resource), GFP_KERNEL);
39	BUG_ON(!res);
40
41	res->name = "System RAM";
42	res->start = start;
43	res->end = start + size - 1;
44	res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
45	if (request_resource(&iomem_resource, res) < 0) {
46		printk("System RAM resource %llx - %llx cannot be added\n",
47		(unsigned long long)res->start, (unsigned long long)res->end);
48		kfree(res);
49		res = NULL;
50	}
51	return res;
52}
53
54static void release_memory_resource(struct resource *res)
55{
56	if (!res)
57		return;
58	release_resource(res);
59	kfree(res);
60	return;
61}
62
63#ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
64#ifndef CONFIG_SPARSEMEM_VMEMMAP
65static void get_page_bootmem(unsigned long info,  struct page *page, int type)
66{
67	atomic_set(&page->_mapcount, type);
68	SetPagePrivate(page);
69	set_page_private(page, info);
70	atomic_inc(&page->_count);
71}
72
73void put_page_bootmem(struct page *page)
74{
75	int type;
76
77	type = atomic_read(&page->_mapcount);
78	BUG_ON(type >= -1);
79
80	if (atomic_dec_return(&page->_count) == 1) {
81		ClearPagePrivate(page);
82		set_page_private(page, 0);
83		reset_page_mapcount(page);
84		__free_pages_bootmem(page, 0);
85	}
86
87}
88
89static void register_page_bootmem_info_section(unsigned long start_pfn)
90{
91	unsigned long *usemap, mapsize, section_nr, i;
92	struct mem_section *ms;
93	struct page *page, *memmap;
94
95	if (!pfn_valid(start_pfn))
96		return;
97
98	section_nr = pfn_to_section_nr(start_pfn);
99	ms = __nr_to_section(section_nr);
100
101	/* Get section's memmap address */
102	memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
103
104	/*
105	 * Get page for the memmap's phys address
106	 * XXX: need more consideration for sparse_vmemmap...
107	 */
108	page = virt_to_page(memmap);
109	mapsize = sizeof(struct page) * PAGES_PER_SECTION;
110	mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
111
112	/* remember memmap's page */
113	for (i = 0; i < mapsize; i++, page++)
114		get_page_bootmem(section_nr, page, SECTION_INFO);
115
116	usemap = __nr_to_section(section_nr)->pageblock_flags;
117	page = virt_to_page(usemap);
118
119	mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
120
121	for (i = 0; i < mapsize; i++, page++)
122		get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
123
124}
125
126void register_page_bootmem_info_node(struct pglist_data *pgdat)
127{
128	unsigned long i, pfn, end_pfn, nr_pages;
129	int node = pgdat->node_id;
130	struct page *page;
131	struct zone *zone;
132
133	nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
134	page = virt_to_page(pgdat);
135
136	for (i = 0; i < nr_pages; i++, page++)
137		get_page_bootmem(node, page, NODE_INFO);
138
139	zone = &pgdat->node_zones[0];
140	for (; zone < pgdat->node_zones + MAX_NR_ZONES - 1; zone++) {
141		if (zone->wait_table) {
142			nr_pages = zone->wait_table_hash_nr_entries
143				* sizeof(wait_queue_head_t);
144			nr_pages = PAGE_ALIGN(nr_pages) >> PAGE_SHIFT;
145			page = virt_to_page(zone->wait_table);
146
147			for (i = 0; i < nr_pages; i++, page++)
148				get_page_bootmem(node, page, NODE_INFO);
149		}
150	}
151
152	pfn = pgdat->node_start_pfn;
153	end_pfn = pfn + pgdat->node_spanned_pages;
154
155	/* register_section info */
156	for (; pfn < end_pfn; pfn += PAGES_PER_SECTION)
157		register_page_bootmem_info_section(pfn);
158
159}
160#endif /* !CONFIG_SPARSEMEM_VMEMMAP */
161
162static void grow_zone_span(struct zone *zone, unsigned long start_pfn,
163			   unsigned long end_pfn)
164{
165	unsigned long old_zone_end_pfn;
166
167	zone_span_writelock(zone);
168
169	old_zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages;
170	if (start_pfn < zone->zone_start_pfn)
171		zone->zone_start_pfn = start_pfn;
172
173	zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
174				zone->zone_start_pfn;
175
176	zone_span_writeunlock(zone);
177}
178
179static void grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn,
180			    unsigned long end_pfn)
181{
182	unsigned long old_pgdat_end_pfn =
183		pgdat->node_start_pfn + pgdat->node_spanned_pages;
184
185	if (start_pfn < pgdat->node_start_pfn)
186		pgdat->node_start_pfn = start_pfn;
187
188	pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
189					pgdat->node_start_pfn;
190}
191
192static int __add_zone(struct zone *zone, unsigned long phys_start_pfn)
193{
194	struct pglist_data *pgdat = zone->zone_pgdat;
195	int nr_pages = PAGES_PER_SECTION;
196	int nid = pgdat->node_id;
197	int zone_type;
198	unsigned long flags;
199
200	zone_type = zone - pgdat->node_zones;
201	if (!zone->wait_table) {
202		int ret;
203
204		ret = init_currently_empty_zone(zone, phys_start_pfn,
205						nr_pages, MEMMAP_HOTPLUG);
206		if (ret)
207			return ret;
208	}
209	pgdat_resize_lock(zone->zone_pgdat, &flags);
210	grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages);
211	grow_pgdat_span(zone->zone_pgdat, phys_start_pfn,
212			phys_start_pfn + nr_pages);
213	pgdat_resize_unlock(zone->zone_pgdat, &flags);
214	memmap_init_zone(nr_pages, nid, zone_type,
215			 phys_start_pfn, MEMMAP_HOTPLUG);
216	return 0;
217}
218
219static int __add_section(struct zone *zone, unsigned long phys_start_pfn)
220{
221	int nr_pages = PAGES_PER_SECTION;
222	int ret;
223
224	if (pfn_valid(phys_start_pfn))
225		return -EEXIST;
226
227	ret = sparse_add_one_section(zone, phys_start_pfn, nr_pages);
228
229	if (ret < 0)
230		return ret;
231
232	ret = __add_zone(zone, phys_start_pfn);
233
234	if (ret < 0)
235		return ret;
236
237	return register_new_memory(__pfn_to_section(phys_start_pfn));
238}
239
240#ifdef CONFIG_SPARSEMEM_VMEMMAP
241static int __remove_section(struct zone *zone, struct mem_section *ms)
242{
243	/*
244	 * XXX: Freeing memmap with vmemmap is not implement yet.
245	 *      This should be removed later.
246	 */
247	return -EBUSY;
248}
249#else
250static int __remove_section(struct zone *zone, struct mem_section *ms)
251{
252	unsigned long flags;
253	struct pglist_data *pgdat = zone->zone_pgdat;
254	int ret = -EINVAL;
255
256	if (!valid_section(ms))
257		return ret;
258
259	ret = unregister_memory_section(ms);
260	if (ret)
261		return ret;
262
263	pgdat_resize_lock(pgdat, &flags);
264	sparse_remove_one_section(zone, ms);
265	pgdat_resize_unlock(pgdat, &flags);
266	return 0;
267}
268#endif
269
270/*
271 * Reasonably generic function for adding memory.  It is
272 * expected that archs that support memory hotplug will
273 * call this function after deciding the zone to which to
274 * add the new pages.
275 */
276int __add_pages(struct zone *zone, unsigned long phys_start_pfn,
277		 unsigned long nr_pages)
278{
279	unsigned long i;
280	int err = 0;
281	int start_sec, end_sec;
282	/* during initialize mem_map, align hot-added range to section */
283	start_sec = pfn_to_section_nr(phys_start_pfn);
284	end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
285
286	for (i = start_sec; i <= end_sec; i++) {
287		err = __add_section(zone, i << PFN_SECTION_SHIFT);
288
289		/*
290		 * EEXIST is finally dealt with by ioresource collision
291		 * check. see add_memory() => register_memory_resource()
292		 * Warning will be printed if there is collision.
293		 */
294		if (err && (err != -EEXIST))
295			break;
296		err = 0;
297	}
298
299	return err;
300}
301EXPORT_SYMBOL_GPL(__add_pages);
302
303/**
304 * __remove_pages() - remove sections of pages from a zone
305 * @zone: zone from which pages need to be removed
306 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
307 * @nr_pages: number of pages to remove (must be multiple of section size)
308 *
309 * Generic helper function to remove section mappings and sysfs entries
310 * for the section of the memory we are removing. Caller needs to make
311 * sure that pages are marked reserved and zones are adjust properly by
312 * calling offline_pages().
313 */
314int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
315		 unsigned long nr_pages)
316{
317	unsigned long i, ret = 0;
318	int sections_to_remove;
319
320	/*
321	 * We can only remove entire sections
322	 */
323	BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
324	BUG_ON(nr_pages % PAGES_PER_SECTION);
325
326	release_mem_region(phys_start_pfn << PAGE_SHIFT, nr_pages * PAGE_SIZE);
327
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		ret = __remove_section(zone, __pfn_to_section(pfn));
332		if (ret)
333			break;
334	}
335	return ret;
336}
337EXPORT_SYMBOL_GPL(__remove_pages);
338
339void online_page(struct page *page)
340{
341	totalram_pages++;
342	num_physpages++;
343
344#ifdef CONFIG_HIGHMEM
345	if (PageHighMem(page))
346		totalhigh_pages++;
347#endif
348
349#ifdef CONFIG_FLATMEM
350	max_mapnr = max(page_to_pfn(page), max_mapnr);
351#endif
352
353	ClearPageReserved(page);
354	init_page_count(page);
355	__free_page(page);
356}
357
358static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
359			void *arg)
360{
361	unsigned long i;
362	unsigned long onlined_pages = *(unsigned long *)arg;
363	struct page *page;
364	if (PageReserved(pfn_to_page(start_pfn)))
365		for (i = 0; i < nr_pages; i++) {
366			page = pfn_to_page(start_pfn + i);
367			online_page(page);
368			onlined_pages++;
369		}
370	*(unsigned long *)arg = onlined_pages;
371	return 0;
372}
373
374
375int online_pages(unsigned long pfn, unsigned long nr_pages)
376{
377	unsigned long onlined_pages = 0;
378	struct zone *zone;
379	int need_zonelists_rebuild = 0;
380	int nid;
381	int ret;
382	struct memory_notify arg;
383
384	arg.start_pfn = pfn;
385	arg.nr_pages = nr_pages;
386	arg.status_change_nid = -1;
387
388	nid = page_to_nid(pfn_to_page(pfn));
389	if (node_present_pages(nid) == 0)
390		arg.status_change_nid = nid;
391
392	ret = memory_notify(MEM_GOING_ONLINE, &arg);
393	ret = notifier_to_errno(ret);
394	if (ret) {
395		memory_notify(MEM_CANCEL_ONLINE, &arg);
396		return ret;
397	}
398	/*
399	 * This doesn't need a lock to do pfn_to_page().
400	 * The section can't be removed here because of the
401	 * memory_block->state_mutex.
402	 */
403	zone = page_zone(pfn_to_page(pfn));
404	/*
405	 * If this zone is not populated, then it is not in zonelist.
406	 * This means the page allocator ignores this zone.
407	 * So, zonelist must be updated after online.
408	 */
409	if (!populated_zone(zone))
410		need_zonelists_rebuild = 1;
411
412	ret = walk_memory_resource(pfn, nr_pages, &onlined_pages,
413		online_pages_range);
414	if (ret) {
415		printk(KERN_DEBUG "online_pages %lx at %lx failed\n",
416			nr_pages, pfn);
417		memory_notify(MEM_CANCEL_ONLINE, &arg);
418		return ret;
419	}
420
421	zone->present_pages += onlined_pages;
422	zone->zone_pgdat->node_present_pages += onlined_pages;
423
424	setup_per_zone_pages_min();
425	if (onlined_pages) {
426		kswapd_run(zone_to_nid(zone));
427		node_set_state(zone_to_nid(zone), N_HIGH_MEMORY);
428	}
429
430	if (need_zonelists_rebuild)
431		build_all_zonelists();
432	else
433		vm_total_pages = nr_free_pagecache_pages();
434
435	writeback_set_ratelimit();
436
437	if (onlined_pages)
438		memory_notify(MEM_ONLINE, &arg);
439
440	return 0;
441}
442#endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
443
444static pg_data_t *hotadd_new_pgdat(int nid, u64 start)
445{
446	struct pglist_data *pgdat;
447	unsigned long zones_size[MAX_NR_ZONES] = {0};
448	unsigned long zholes_size[MAX_NR_ZONES] = {0};
449	unsigned long start_pfn = start >> PAGE_SHIFT;
450
451	pgdat = arch_alloc_nodedata(nid);
452	if (!pgdat)
453		return NULL;
454
455	arch_refresh_nodedata(nid, pgdat);
456
457	/* we can use NODE_DATA(nid) from here */
458
459	/* init node's zones as empty zones, we don't have any present pages.*/
460	free_area_init_node(nid, zones_size, start_pfn, zholes_size);
461
462	return pgdat;
463}
464
465static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
466{
467	arch_refresh_nodedata(nid, NULL);
468	arch_free_nodedata(pgdat);
469	return;
470}
471
472
473int add_memory(int nid, u64 start, u64 size)
474{
475	pg_data_t *pgdat = NULL;
476	int new_pgdat = 0;
477	struct resource *res;
478	int ret;
479
480	res = register_memory_resource(start, size);
481	if (!res)
482		return -EEXIST;
483
484	if (!node_online(nid)) {
485		pgdat = hotadd_new_pgdat(nid, start);
486		if (!pgdat)
487			return -ENOMEM;
488		new_pgdat = 1;
489	}
490
491	/* call arch's memory hotadd */
492	ret = arch_add_memory(nid, start, size);
493
494	if (ret < 0)
495		goto error;
496
497	/* we online node here. we can't roll back from here. */
498	node_set_online(nid);
499
500	cpuset_track_online_nodes();
501
502	if (new_pgdat) {
503		ret = register_one_node(nid);
504		/*
505		 * If sysfs file of new node can't create, cpu on the node
506		 * can't be hot-added. There is no rollback way now.
507		 * So, check by BUG_ON() to catch it reluctantly..
508		 */
509		BUG_ON(ret);
510	}
511
512	return ret;
513error:
514	/* rollback pgdat allocation and others */
515	if (new_pgdat)
516		rollback_node_hotadd(nid, pgdat);
517	if (res)
518		release_memory_resource(res);
519
520	return ret;
521}
522EXPORT_SYMBOL_GPL(add_memory);
523
524#ifdef CONFIG_MEMORY_HOTREMOVE
525/*
526 * Confirm all pages in a range [start, end) is belongs to the same zone.
527 */
528static int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn)
529{
530	unsigned long pfn;
531	struct zone *zone = NULL;
532	struct page *page;
533	int i;
534	for (pfn = start_pfn;
535	     pfn < end_pfn;
536	     pfn += MAX_ORDER_NR_PAGES) {
537		i = 0;
538		/* This is just a CONFIG_HOLES_IN_ZONE check.*/
539		while ((i < MAX_ORDER_NR_PAGES) && !pfn_valid_within(pfn + i))
540			i++;
541		if (i == MAX_ORDER_NR_PAGES)
542			continue;
543		page = pfn_to_page(pfn + i);
544		if (zone && page_zone(page) != zone)
545			return 0;
546		zone = page_zone(page);
547	}
548	return 1;
549}
550
551/*
552 * Scanning pfn is much easier than scanning lru list.
553 * Scan pfn from start to end and Find LRU page.
554 */
555int scan_lru_pages(unsigned long start, unsigned long end)
556{
557	unsigned long pfn;
558	struct page *page;
559	for (pfn = start; pfn < end; pfn++) {
560		if (pfn_valid(pfn)) {
561			page = pfn_to_page(pfn);
562			if (PageLRU(page))
563				return pfn;
564		}
565	}
566	return 0;
567}
568
569static struct page *
570hotremove_migrate_alloc(struct page *page,
571			unsigned long private,
572			int **x)
573{
574	/* This should be improoooooved!! */
575	return alloc_page(GFP_HIGHUSER_PAGECACHE);
576}
577
578
579#define NR_OFFLINE_AT_ONCE_PAGES	(256)
580static int
581do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
582{
583	unsigned long pfn;
584	struct page *page;
585	int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
586	int not_managed = 0;
587	int ret = 0;
588	LIST_HEAD(source);
589
590	for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
591		if (!pfn_valid(pfn))
592			continue;
593		page = pfn_to_page(pfn);
594		if (!page_count(page))
595			continue;
596		/*
597		 * We can skip free pages. And we can only deal with pages on
598		 * LRU.
599		 */
600		ret = isolate_lru_page(page, &source);
601		if (!ret) { /* Success */
602			move_pages--;
603		} else {
604			/* Becasue we don't have big zone->lock. we should
605			   check this again here. */
606			if (page_count(page))
607				not_managed++;
608#ifdef CONFIG_DEBUG_VM
609			printk(KERN_INFO "removing from LRU failed"
610					 " %lx/%d/%lx\n",
611				pfn, page_count(page), page->flags);
612#endif
613		}
614	}
615	ret = -EBUSY;
616	if (not_managed) {
617		if (!list_empty(&source))
618			putback_lru_pages(&source);
619		goto out;
620	}
621	ret = 0;
622	if (list_empty(&source))
623		goto out;
624	/* this function returns # of failed pages */
625	ret = migrate_pages(&source, hotremove_migrate_alloc, 0);
626
627out:
628	return ret;
629}
630
631/*
632 * remove from free_area[] and mark all as Reserved.
633 */
634static int
635offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
636			void *data)
637{
638	__offline_isolated_pages(start, start + nr_pages);
639	return 0;
640}
641
642static void
643offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
644{
645	walk_memory_resource(start_pfn, end_pfn - start_pfn, NULL,
646				offline_isolated_pages_cb);
647}
648
649/*
650 * Check all pages in range, recoreded as memory resource, are isolated.
651 */
652static int
653check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
654			void *data)
655{
656	int ret;
657	long offlined = *(long *)data;
658	ret = test_pages_isolated(start_pfn, start_pfn + nr_pages);
659	offlined = nr_pages;
660	if (!ret)
661		*(long *)data += offlined;
662	return ret;
663}
664
665static long
666check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
667{
668	long offlined = 0;
669	int ret;
670
671	ret = walk_memory_resource(start_pfn, end_pfn - start_pfn, &offlined,
672			check_pages_isolated_cb);
673	if (ret < 0)
674		offlined = (long)ret;
675	return offlined;
676}
677
678int offline_pages(unsigned long start_pfn,
679		  unsigned long end_pfn, unsigned long timeout)
680{
681	unsigned long pfn, nr_pages, expire;
682	long offlined_pages;
683	int ret, drain, retry_max, node;
684	struct zone *zone;
685	struct memory_notify arg;
686
687	BUG_ON(start_pfn >= end_pfn);
688	/* at least, alignment against pageblock is necessary */
689	if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
690		return -EINVAL;
691	if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
692		return -EINVAL;
693	/* This makes hotplug much easier...and readable.
694	   we assume this for now. .*/
695	if (!test_pages_in_a_zone(start_pfn, end_pfn))
696		return -EINVAL;
697
698	zone = page_zone(pfn_to_page(start_pfn));
699	node = zone_to_nid(zone);
700	nr_pages = end_pfn - start_pfn;
701
702	/* set above range as isolated */
703	ret = start_isolate_page_range(start_pfn, end_pfn);
704	if (ret)
705		return ret;
706
707	arg.start_pfn = start_pfn;
708	arg.nr_pages = nr_pages;
709	arg.status_change_nid = -1;
710	if (nr_pages >= node_present_pages(node))
711		arg.status_change_nid = node;
712
713	ret = memory_notify(MEM_GOING_OFFLINE, &arg);
714	ret = notifier_to_errno(ret);
715	if (ret)
716		goto failed_removal;
717
718	pfn = start_pfn;
719	expire = jiffies + timeout;
720	drain = 0;
721	retry_max = 5;
722repeat:
723	/* start memory hot removal */
724	ret = -EAGAIN;
725	if (time_after(jiffies, expire))
726		goto failed_removal;
727	ret = -EINTR;
728	if (signal_pending(current))
729		goto failed_removal;
730	ret = 0;
731	if (drain) {
732		lru_add_drain_all();
733		flush_scheduled_work();
734		cond_resched();
735		drain_all_pages();
736	}
737
738	pfn = scan_lru_pages(start_pfn, end_pfn);
739	if (pfn) { /* We have page on LRU */
740		ret = do_migrate_range(pfn, end_pfn);
741		if (!ret) {
742			drain = 1;
743			goto repeat;
744		} else {
745			if (ret < 0)
746				if (--retry_max == 0)
747					goto failed_removal;
748			yield();
749			drain = 1;
750			goto repeat;
751		}
752	}
753	/* drain all zone's lru pagevec, this is asyncronous... */
754	lru_add_drain_all();
755	flush_scheduled_work();
756	yield();
757	/* drain pcp pages , this is synchrouns. */
758	drain_all_pages();
759	/* check again */
760	offlined_pages = check_pages_isolated(start_pfn, end_pfn);
761	if (offlined_pages < 0) {
762		ret = -EBUSY;
763		goto failed_removal;
764	}
765	printk(KERN_INFO "Offlined Pages %ld\n", offlined_pages);
766	/* Ok, all of our target is islaoted.
767	   We cannot do rollback at this point. */
768	offline_isolated_pages(start_pfn, end_pfn);
769	/* reset pagetype flags and makes migrate type to be MOVABLE */
770	undo_isolate_page_range(start_pfn, end_pfn);
771	/* removal success */
772	zone->present_pages -= offlined_pages;
773	zone->zone_pgdat->node_present_pages -= offlined_pages;
774	totalram_pages -= offlined_pages;
775	num_physpages -= offlined_pages;
776
777	vm_total_pages = nr_free_pagecache_pages();
778	writeback_set_ratelimit();
779
780	memory_notify(MEM_OFFLINE, &arg);
781	return 0;
782
783failed_removal:
784	printk(KERN_INFO "memory offlining %lx to %lx failed\n",
785		start_pfn, end_pfn);
786	memory_notify(MEM_CANCEL_OFFLINE, &arg);
787	/* pushback to free area */
788	undo_isolate_page_range(start_pfn, end_pfn);
789
790	return ret;
791}
792#else
793int remove_memory(u64 start, u64 size)
794{
795	return -EINVAL;
796}
797EXPORT_SYMBOL_GPL(remove_memory);
798#endif /* CONFIG_MEMORY_HOTREMOVE */
799