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/compiler.h>
13#include <linux/export.h>
14#include <linux/pagevec.h>
15#include <linux/writeback.h>
16#include <linux/slab.h>
17#include <linux/sysctl.h>
18#include <linux/cpu.h>
19#include <linux/memory.h>
20#include <linux/memory_hotplug.h>
21#include <linux/highmem.h>
22#include <linux/vmalloc.h>
23#include <linux/ioport.h>
24#include <linux/delay.h>
25#include <linux/migrate.h>
26#include <linux/page-isolation.h>
27#include <linux/pfn.h>
28#include <linux/suspend.h>
29#include <linux/mm_inline.h>
30#include <linux/firmware-map.h>
31#include <linux/stop_machine.h>
32#include <linux/hugetlb.h>
33#include <linux/memblock.h>
34#include <linux/bootmem.h>
35
36#include <asm/tlbflush.h>
37
38#include "internal.h"
39
40/*
41 * online_page_callback contains pointer to current page onlining function.
42 * Initially it is generic_online_page(). If it is required it could be
43 * changed by calling set_online_page_callback() for callback registration
44 * and restore_online_page_callback() for generic callback restore.
45 */
46
47static void generic_online_page(struct page *page);
48
49static online_page_callback_t online_page_callback = generic_online_page;
50static DEFINE_MUTEX(online_page_callback_lock);
51
52/* The same as the cpu_hotplug lock, but for memory hotplug. */
53static struct {
54	struct task_struct *active_writer;
55	struct mutex lock; /* Synchronizes accesses to refcount, */
56	/*
57	 * Also blocks the new readers during
58	 * an ongoing mem hotplug operation.
59	 */
60	int refcount;
61
62#ifdef CONFIG_DEBUG_LOCK_ALLOC
63	struct lockdep_map dep_map;
64#endif
65} mem_hotplug = {
66	.active_writer = NULL,
67	.lock = __MUTEX_INITIALIZER(mem_hotplug.lock),
68	.refcount = 0,
69#ifdef CONFIG_DEBUG_LOCK_ALLOC
70	.dep_map = {.name = "mem_hotplug.lock" },
71#endif
72};
73
74/* Lockdep annotations for get/put_online_mems() and mem_hotplug_begin/end() */
75#define memhp_lock_acquire_read() lock_map_acquire_read(&mem_hotplug.dep_map)
76#define memhp_lock_acquire()      lock_map_acquire(&mem_hotplug.dep_map)
77#define memhp_lock_release()      lock_map_release(&mem_hotplug.dep_map)
78
79void get_online_mems(void)
80{
81	might_sleep();
82	if (mem_hotplug.active_writer == current)
83		return;
84	memhp_lock_acquire_read();
85	mutex_lock(&mem_hotplug.lock);
86	mem_hotplug.refcount++;
87	mutex_unlock(&mem_hotplug.lock);
88
89}
90
91void put_online_mems(void)
92{
93	if (mem_hotplug.active_writer == current)
94		return;
95	mutex_lock(&mem_hotplug.lock);
96
97	if (WARN_ON(!mem_hotplug.refcount))
98		mem_hotplug.refcount++; /* try to fix things up */
99
100	if (!--mem_hotplug.refcount && unlikely(mem_hotplug.active_writer))
101		wake_up_process(mem_hotplug.active_writer);
102	mutex_unlock(&mem_hotplug.lock);
103	memhp_lock_release();
104
105}
106
107void mem_hotplug_begin(void)
108{
109	mem_hotplug.active_writer = current;
110
111	memhp_lock_acquire();
112	for (;;) {
113		mutex_lock(&mem_hotplug.lock);
114		if (likely(!mem_hotplug.refcount))
115			break;
116		__set_current_state(TASK_UNINTERRUPTIBLE);
117		mutex_unlock(&mem_hotplug.lock);
118		schedule();
119	}
120}
121
122void mem_hotplug_done(void)
123{
124	mem_hotplug.active_writer = NULL;
125	mutex_unlock(&mem_hotplug.lock);
126	memhp_lock_release();
127}
128
129/* add this memory to iomem resource */
130static struct resource *register_memory_resource(u64 start, u64 size)
131{
132	struct resource *res;
133	res = kzalloc(sizeof(struct resource), GFP_KERNEL);
134	BUG_ON(!res);
135
136	res->name = "System RAM";
137	res->start = start;
138	res->end = start + size - 1;
139	res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
140	if (request_resource(&iomem_resource, res) < 0) {
141		pr_debug("System RAM resource %pR cannot be added\n", res);
142		kfree(res);
143		res = NULL;
144	}
145	return res;
146}
147
148static void release_memory_resource(struct resource *res)
149{
150	if (!res)
151		return;
152	release_resource(res);
153	kfree(res);
154	return;
155}
156
157#ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
158void get_page_bootmem(unsigned long info,  struct page *page,
159		      unsigned long type)
160{
161	page->lru.next = (struct list_head *) type;
162	SetPagePrivate(page);
163	set_page_private(page, info);
164	atomic_inc(&page->_count);
165}
166
167void put_page_bootmem(struct page *page)
168{
169	unsigned long type;
170
171	type = (unsigned long) page->lru.next;
172	BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE ||
173	       type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE);
174
175	if (atomic_dec_return(&page->_count) == 1) {
176		ClearPagePrivate(page);
177		set_page_private(page, 0);
178		INIT_LIST_HEAD(&page->lru);
179		free_reserved_page(page);
180	}
181}
182
183#ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
184#ifndef CONFIG_SPARSEMEM_VMEMMAP
185static void register_page_bootmem_info_section(unsigned long start_pfn)
186{
187	unsigned long *usemap, mapsize, section_nr, i;
188	struct mem_section *ms;
189	struct page *page, *memmap;
190
191	section_nr = pfn_to_section_nr(start_pfn);
192	ms = __nr_to_section(section_nr);
193
194	/* Get section's memmap address */
195	memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
196
197	/*
198	 * Get page for the memmap's phys address
199	 * XXX: need more consideration for sparse_vmemmap...
200	 */
201	page = virt_to_page(memmap);
202	mapsize = sizeof(struct page) * PAGES_PER_SECTION;
203	mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
204
205	/* remember memmap's page */
206	for (i = 0; i < mapsize; i++, page++)
207		get_page_bootmem(section_nr, page, SECTION_INFO);
208
209	usemap = __nr_to_section(section_nr)->pageblock_flags;
210	page = virt_to_page(usemap);
211
212	mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
213
214	for (i = 0; i < mapsize; i++, page++)
215		get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
216
217}
218#else /* CONFIG_SPARSEMEM_VMEMMAP */
219static void register_page_bootmem_info_section(unsigned long start_pfn)
220{
221	unsigned long *usemap, mapsize, section_nr, i;
222	struct mem_section *ms;
223	struct page *page, *memmap;
224
225	if (!pfn_valid(start_pfn))
226		return;
227
228	section_nr = pfn_to_section_nr(start_pfn);
229	ms = __nr_to_section(section_nr);
230
231	memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
232
233	register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION);
234
235	usemap = __nr_to_section(section_nr)->pageblock_flags;
236	page = virt_to_page(usemap);
237
238	mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
239
240	for (i = 0; i < mapsize; i++, page++)
241		get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
242}
243#endif /* !CONFIG_SPARSEMEM_VMEMMAP */
244
245void register_page_bootmem_info_node(struct pglist_data *pgdat)
246{
247	unsigned long i, pfn, end_pfn, nr_pages;
248	int node = pgdat->node_id;
249	struct page *page;
250	struct zone *zone;
251
252	nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
253	page = virt_to_page(pgdat);
254
255	for (i = 0; i < nr_pages; i++, page++)
256		get_page_bootmem(node, page, NODE_INFO);
257
258	zone = &pgdat->node_zones[0];
259	for (; zone < pgdat->node_zones + MAX_NR_ZONES - 1; zone++) {
260		if (zone_is_initialized(zone)) {
261			nr_pages = zone->wait_table_hash_nr_entries
262				* sizeof(wait_queue_head_t);
263			nr_pages = PAGE_ALIGN(nr_pages) >> PAGE_SHIFT;
264			page = virt_to_page(zone->wait_table);
265
266			for (i = 0; i < nr_pages; i++, page++)
267				get_page_bootmem(node, page, NODE_INFO);
268		}
269	}
270
271	pfn = pgdat->node_start_pfn;
272	end_pfn = pgdat_end_pfn(pgdat);
273
274	/* register section info */
275	for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
276		/*
277		 * Some platforms can assign the same pfn to multiple nodes - on
278		 * node0 as well as nodeN.  To avoid registering a pfn against
279		 * multiple nodes we check that this pfn does not already
280		 * reside in some other nodes.
281		 */
282		if (pfn_valid(pfn) && (pfn_to_nid(pfn) == node))
283			register_page_bootmem_info_section(pfn);
284	}
285}
286#endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
287
288static void __meminit grow_zone_span(struct zone *zone, unsigned long start_pfn,
289				     unsigned long end_pfn)
290{
291	unsigned long old_zone_end_pfn;
292
293	zone_span_writelock(zone);
294
295	old_zone_end_pfn = zone_end_pfn(zone);
296	if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
297		zone->zone_start_pfn = start_pfn;
298
299	zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
300				zone->zone_start_pfn;
301
302	zone_span_writeunlock(zone);
303}
304
305static void resize_zone(struct zone *zone, unsigned long start_pfn,
306		unsigned long end_pfn)
307{
308	zone_span_writelock(zone);
309
310	if (end_pfn - start_pfn) {
311		zone->zone_start_pfn = start_pfn;
312		zone->spanned_pages = end_pfn - start_pfn;
313	} else {
314		/*
315		 * make it consist as free_area_init_core(),
316		 * if spanned_pages = 0, then keep start_pfn = 0
317		 */
318		zone->zone_start_pfn = 0;
319		zone->spanned_pages = 0;
320	}
321
322	zone_span_writeunlock(zone);
323}
324
325static void fix_zone_id(struct zone *zone, unsigned long start_pfn,
326		unsigned long end_pfn)
327{
328	enum zone_type zid = zone_idx(zone);
329	int nid = zone->zone_pgdat->node_id;
330	unsigned long pfn;
331
332	for (pfn = start_pfn; pfn < end_pfn; pfn++)
333		set_page_links(pfn_to_page(pfn), zid, nid, pfn);
334}
335
336/* Can fail with -ENOMEM from allocating a wait table with vmalloc() or
337 * alloc_bootmem_node_nopanic()/memblock_virt_alloc_node_nopanic() */
338static int __ref ensure_zone_is_initialized(struct zone *zone,
339			unsigned long start_pfn, unsigned long num_pages)
340{
341	if (!zone_is_initialized(zone))
342		return init_currently_empty_zone(zone, start_pfn, num_pages,
343						 MEMMAP_HOTPLUG);
344	return 0;
345}
346
347static int __meminit move_pfn_range_left(struct zone *z1, struct zone *z2,
348		unsigned long start_pfn, unsigned long end_pfn)
349{
350	int ret;
351	unsigned long flags;
352	unsigned long z1_start_pfn;
353
354	ret = ensure_zone_is_initialized(z1, start_pfn, end_pfn - start_pfn);
355	if (ret)
356		return ret;
357
358	pgdat_resize_lock(z1->zone_pgdat, &flags);
359
360	/* can't move pfns which are higher than @z2 */
361	if (end_pfn > zone_end_pfn(z2))
362		goto out_fail;
363	/* the move out part must be at the left most of @z2 */
364	if (start_pfn > z2->zone_start_pfn)
365		goto out_fail;
366	/* must included/overlap */
367	if (end_pfn <= z2->zone_start_pfn)
368		goto out_fail;
369
370	/* use start_pfn for z1's start_pfn if z1 is empty */
371	if (!zone_is_empty(z1))
372		z1_start_pfn = z1->zone_start_pfn;
373	else
374		z1_start_pfn = start_pfn;
375
376	resize_zone(z1, z1_start_pfn, end_pfn);
377	resize_zone(z2, end_pfn, zone_end_pfn(z2));
378
379	pgdat_resize_unlock(z1->zone_pgdat, &flags);
380
381	fix_zone_id(z1, start_pfn, end_pfn);
382
383	return 0;
384out_fail:
385	pgdat_resize_unlock(z1->zone_pgdat, &flags);
386	return -1;
387}
388
389static int __meminit move_pfn_range_right(struct zone *z1, struct zone *z2,
390		unsigned long start_pfn, unsigned long end_pfn)
391{
392	int ret;
393	unsigned long flags;
394	unsigned long z2_end_pfn;
395
396	ret = ensure_zone_is_initialized(z2, start_pfn, end_pfn - start_pfn);
397	if (ret)
398		return ret;
399
400	pgdat_resize_lock(z1->zone_pgdat, &flags);
401
402	/* can't move pfns which are lower than @z1 */
403	if (z1->zone_start_pfn > start_pfn)
404		goto out_fail;
405	/* the move out part mast at the right most of @z1 */
406	if (zone_end_pfn(z1) >  end_pfn)
407		goto out_fail;
408	/* must included/overlap */
409	if (start_pfn >= zone_end_pfn(z1))
410		goto out_fail;
411
412	/* use end_pfn for z2's end_pfn if z2 is empty */
413	if (!zone_is_empty(z2))
414		z2_end_pfn = zone_end_pfn(z2);
415	else
416		z2_end_pfn = end_pfn;
417
418	resize_zone(z1, z1->zone_start_pfn, start_pfn);
419	resize_zone(z2, start_pfn, z2_end_pfn);
420
421	pgdat_resize_unlock(z1->zone_pgdat, &flags);
422
423	fix_zone_id(z2, start_pfn, end_pfn);
424
425	return 0;
426out_fail:
427	pgdat_resize_unlock(z1->zone_pgdat, &flags);
428	return -1;
429}
430
431static void __meminit grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn,
432				      unsigned long end_pfn)
433{
434	unsigned long old_pgdat_end_pfn = pgdat_end_pfn(pgdat);
435
436	if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
437		pgdat->node_start_pfn = start_pfn;
438
439	pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
440					pgdat->node_start_pfn;
441}
442
443static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn)
444{
445	struct pglist_data *pgdat = zone->zone_pgdat;
446	int nr_pages = PAGES_PER_SECTION;
447	int nid = pgdat->node_id;
448	int zone_type;
449	unsigned long flags;
450	int ret;
451
452	zone_type = zone - pgdat->node_zones;
453	ret = ensure_zone_is_initialized(zone, phys_start_pfn, nr_pages);
454	if (ret)
455		return ret;
456
457	pgdat_resize_lock(zone->zone_pgdat, &flags);
458	grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages);
459	grow_pgdat_span(zone->zone_pgdat, phys_start_pfn,
460			phys_start_pfn + nr_pages);
461	pgdat_resize_unlock(zone->zone_pgdat, &flags);
462	memmap_init_zone(nr_pages, nid, zone_type,
463			 phys_start_pfn, MEMMAP_HOTPLUG);
464	return 0;
465}
466
467static int __meminit __add_section(int nid, struct zone *zone,
468					unsigned long phys_start_pfn)
469{
470	int ret;
471
472	if (pfn_valid(phys_start_pfn))
473		return -EEXIST;
474
475	ret = sparse_add_one_section(zone, phys_start_pfn);
476
477	if (ret < 0)
478		return ret;
479
480	ret = __add_zone(zone, phys_start_pfn);
481
482	if (ret < 0)
483		return ret;
484
485	return register_new_memory(nid, __pfn_to_section(phys_start_pfn));
486}
487
488/*
489 * Reasonably generic function for adding memory.  It is
490 * expected that archs that support memory hotplug will
491 * call this function after deciding the zone to which to
492 * add the new pages.
493 */
494int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn,
495			unsigned long nr_pages)
496{
497	unsigned long i;
498	int err = 0;
499	int start_sec, end_sec;
500	/* during initialize mem_map, align hot-added range to section */
501	start_sec = pfn_to_section_nr(phys_start_pfn);
502	end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
503
504	for (i = start_sec; i <= end_sec; i++) {
505		err = __add_section(nid, zone, section_nr_to_pfn(i));
506
507		/*
508		 * EEXIST is finally dealt with by ioresource collision
509		 * check. see add_memory() => register_memory_resource()
510		 * Warning will be printed if there is collision.
511		 */
512		if (err && (err != -EEXIST))
513			break;
514		err = 0;
515	}
516
517	return err;
518}
519EXPORT_SYMBOL_GPL(__add_pages);
520
521#ifdef CONFIG_MEMORY_HOTREMOVE
522/* find the smallest valid pfn in the range [start_pfn, end_pfn) */
523static int find_smallest_section_pfn(int nid, struct zone *zone,
524				     unsigned long start_pfn,
525				     unsigned long end_pfn)
526{
527	struct mem_section *ms;
528
529	for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) {
530		ms = __pfn_to_section(start_pfn);
531
532		if (unlikely(!valid_section(ms)))
533			continue;
534
535		if (unlikely(pfn_to_nid(start_pfn) != nid))
536			continue;
537
538		if (zone && zone != page_zone(pfn_to_page(start_pfn)))
539			continue;
540
541		return start_pfn;
542	}
543
544	return 0;
545}
546
547/* find the biggest valid pfn in the range [start_pfn, end_pfn). */
548static int find_biggest_section_pfn(int nid, struct zone *zone,
549				    unsigned long start_pfn,
550				    unsigned long end_pfn)
551{
552	struct mem_section *ms;
553	unsigned long pfn;
554
555	/* pfn is the end pfn of a memory section. */
556	pfn = end_pfn - 1;
557	for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) {
558		ms = __pfn_to_section(pfn);
559
560		if (unlikely(!valid_section(ms)))
561			continue;
562
563		if (unlikely(pfn_to_nid(pfn) != nid))
564			continue;
565
566		if (zone && zone != page_zone(pfn_to_page(pfn)))
567			continue;
568
569		return pfn;
570	}
571
572	return 0;
573}
574
575static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
576			     unsigned long end_pfn)
577{
578	unsigned long zone_start_pfn = zone->zone_start_pfn;
579	unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */
580	unsigned long zone_end_pfn = z;
581	unsigned long pfn;
582	struct mem_section *ms;
583	int nid = zone_to_nid(zone);
584
585	zone_span_writelock(zone);
586	if (zone_start_pfn == start_pfn) {
587		/*
588		 * If the section is smallest section in the zone, it need
589		 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
590		 * In this case, we find second smallest valid mem_section
591		 * for shrinking zone.
592		 */
593		pfn = find_smallest_section_pfn(nid, zone, end_pfn,
594						zone_end_pfn);
595		if (pfn) {
596			zone->zone_start_pfn = pfn;
597			zone->spanned_pages = zone_end_pfn - pfn;
598		}
599	} else if (zone_end_pfn == end_pfn) {
600		/*
601		 * If the section is biggest section in the zone, it need
602		 * shrink zone->spanned_pages.
603		 * In this case, we find second biggest valid mem_section for
604		 * shrinking zone.
605		 */
606		pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn,
607					       start_pfn);
608		if (pfn)
609			zone->spanned_pages = pfn - zone_start_pfn + 1;
610	}
611
612	/*
613	 * The section is not biggest or smallest mem_section in the zone, it
614	 * only creates a hole in the zone. So in this case, we need not
615	 * change the zone. But perhaps, the zone has only hole data. Thus
616	 * it check the zone has only hole or not.
617	 */
618	pfn = zone_start_pfn;
619	for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) {
620		ms = __pfn_to_section(pfn);
621
622		if (unlikely(!valid_section(ms)))
623			continue;
624
625		if (page_zone(pfn_to_page(pfn)) != zone)
626			continue;
627
628		 /* If the section is current section, it continues the loop */
629		if (start_pfn == pfn)
630			continue;
631
632		/* If we find valid section, we have nothing to do */
633		zone_span_writeunlock(zone);
634		return;
635	}
636
637	/* The zone has no valid section */
638	zone->zone_start_pfn = 0;
639	zone->spanned_pages = 0;
640	zone_span_writeunlock(zone);
641}
642
643static void shrink_pgdat_span(struct pglist_data *pgdat,
644			      unsigned long start_pfn, unsigned long end_pfn)
645{
646	unsigned long pgdat_start_pfn = pgdat->node_start_pfn;
647	unsigned long p = pgdat_end_pfn(pgdat); /* pgdat_end_pfn namespace clash */
648	unsigned long pgdat_end_pfn = p;
649	unsigned long pfn;
650	struct mem_section *ms;
651	int nid = pgdat->node_id;
652
653	if (pgdat_start_pfn == start_pfn) {
654		/*
655		 * If the section is smallest section in the pgdat, it need
656		 * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages.
657		 * In this case, we find second smallest valid mem_section
658		 * for shrinking zone.
659		 */
660		pfn = find_smallest_section_pfn(nid, NULL, end_pfn,
661						pgdat_end_pfn);
662		if (pfn) {
663			pgdat->node_start_pfn = pfn;
664			pgdat->node_spanned_pages = pgdat_end_pfn - pfn;
665		}
666	} else if (pgdat_end_pfn == end_pfn) {
667		/*
668		 * If the section is biggest section in the pgdat, it need
669		 * shrink pgdat->node_spanned_pages.
670		 * In this case, we find second biggest valid mem_section for
671		 * shrinking zone.
672		 */
673		pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn,
674					       start_pfn);
675		if (pfn)
676			pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1;
677	}
678
679	/*
680	 * If the section is not biggest or smallest mem_section in the pgdat,
681	 * it only creates a hole in the pgdat. So in this case, we need not
682	 * change the pgdat.
683	 * But perhaps, the pgdat has only hole data. Thus it check the pgdat
684	 * has only hole or not.
685	 */
686	pfn = pgdat_start_pfn;
687	for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) {
688		ms = __pfn_to_section(pfn);
689
690		if (unlikely(!valid_section(ms)))
691			continue;
692
693		if (pfn_to_nid(pfn) != nid)
694			continue;
695
696		 /* If the section is current section, it continues the loop */
697		if (start_pfn == pfn)
698			continue;
699
700		/* If we find valid section, we have nothing to do */
701		return;
702	}
703
704	/* The pgdat has no valid section */
705	pgdat->node_start_pfn = 0;
706	pgdat->node_spanned_pages = 0;
707}
708
709static void __remove_zone(struct zone *zone, unsigned long start_pfn)
710{
711	struct pglist_data *pgdat = zone->zone_pgdat;
712	int nr_pages = PAGES_PER_SECTION;
713	int zone_type;
714	unsigned long flags;
715
716	zone_type = zone - pgdat->node_zones;
717
718	pgdat_resize_lock(zone->zone_pgdat, &flags);
719	shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
720	shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages);
721	pgdat_resize_unlock(zone->zone_pgdat, &flags);
722}
723
724static int __remove_section(struct zone *zone, struct mem_section *ms)
725{
726	unsigned long start_pfn;
727	int scn_nr;
728	int ret = -EINVAL;
729
730	if (!valid_section(ms))
731		return ret;
732
733	ret = unregister_memory_section(ms);
734	if (ret)
735		return ret;
736
737	scn_nr = __section_nr(ms);
738	start_pfn = section_nr_to_pfn(scn_nr);
739	__remove_zone(zone, start_pfn);
740
741	sparse_remove_one_section(zone, ms);
742	return 0;
743}
744
745/**
746 * __remove_pages() - remove sections of pages from a zone
747 * @zone: zone from which pages need to be removed
748 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
749 * @nr_pages: number of pages to remove (must be multiple of section size)
750 *
751 * Generic helper function to remove section mappings and sysfs entries
752 * for the section of the memory we are removing. Caller needs to make
753 * sure that pages are marked reserved and zones are adjust properly by
754 * calling offline_pages().
755 */
756int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
757		 unsigned long nr_pages)
758{
759	unsigned long i;
760	int sections_to_remove;
761	resource_size_t start, size;
762	int ret = 0;
763
764	/*
765	 * We can only remove entire sections
766	 */
767	BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
768	BUG_ON(nr_pages % PAGES_PER_SECTION);
769
770	start = phys_start_pfn << PAGE_SHIFT;
771	size = nr_pages * PAGE_SIZE;
772	ret = release_mem_region_adjustable(&iomem_resource, start, size);
773	if (ret) {
774		resource_size_t endres = start + size - 1;
775
776		pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
777				&start, &endres, ret);
778	}
779
780	sections_to_remove = nr_pages / PAGES_PER_SECTION;
781	for (i = 0; i < sections_to_remove; i++) {
782		unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
783		ret = __remove_section(zone, __pfn_to_section(pfn));
784		if (ret)
785			break;
786	}
787	return ret;
788}
789EXPORT_SYMBOL_GPL(__remove_pages);
790#endif /* CONFIG_MEMORY_HOTREMOVE */
791
792int set_online_page_callback(online_page_callback_t callback)
793{
794	int rc = -EINVAL;
795
796	get_online_mems();
797	mutex_lock(&online_page_callback_lock);
798
799	if (online_page_callback == generic_online_page) {
800		online_page_callback = callback;
801		rc = 0;
802	}
803
804	mutex_unlock(&online_page_callback_lock);
805	put_online_mems();
806
807	return rc;
808}
809EXPORT_SYMBOL_GPL(set_online_page_callback);
810
811int restore_online_page_callback(online_page_callback_t callback)
812{
813	int rc = -EINVAL;
814
815	get_online_mems();
816	mutex_lock(&online_page_callback_lock);
817
818	if (online_page_callback == callback) {
819		online_page_callback = generic_online_page;
820		rc = 0;
821	}
822
823	mutex_unlock(&online_page_callback_lock);
824	put_online_mems();
825
826	return rc;
827}
828EXPORT_SYMBOL_GPL(restore_online_page_callback);
829
830void __online_page_set_limits(struct page *page)
831{
832}
833EXPORT_SYMBOL_GPL(__online_page_set_limits);
834
835void __online_page_increment_counters(struct page *page)
836{
837	adjust_managed_page_count(page, 1);
838}
839EXPORT_SYMBOL_GPL(__online_page_increment_counters);
840
841void __online_page_free(struct page *page)
842{
843	__free_reserved_page(page);
844}
845EXPORT_SYMBOL_GPL(__online_page_free);
846
847static void generic_online_page(struct page *page)
848{
849	__online_page_set_limits(page);
850	__online_page_increment_counters(page);
851	__online_page_free(page);
852}
853
854static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
855			void *arg)
856{
857	unsigned long i;
858	unsigned long onlined_pages = *(unsigned long *)arg;
859	struct page *page;
860	if (PageReserved(pfn_to_page(start_pfn)))
861		for (i = 0; i < nr_pages; i++) {
862			page = pfn_to_page(start_pfn + i);
863			(*online_page_callback)(page);
864			onlined_pages++;
865		}
866	*(unsigned long *)arg = onlined_pages;
867	return 0;
868}
869
870#ifdef CONFIG_MOVABLE_NODE
871/*
872 * When CONFIG_MOVABLE_NODE, we permit onlining of a node which doesn't have
873 * normal memory.
874 */
875static bool can_online_high_movable(struct zone *zone)
876{
877	return true;
878}
879#else /* CONFIG_MOVABLE_NODE */
880/* ensure every online node has NORMAL memory */
881static bool can_online_high_movable(struct zone *zone)
882{
883	return node_state(zone_to_nid(zone), N_NORMAL_MEMORY);
884}
885#endif /* CONFIG_MOVABLE_NODE */
886
887/* check which state of node_states will be changed when online memory */
888static void node_states_check_changes_online(unsigned long nr_pages,
889	struct zone *zone, struct memory_notify *arg)
890{
891	int nid = zone_to_nid(zone);
892	enum zone_type zone_last = ZONE_NORMAL;
893
894	/*
895	 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
896	 * contains nodes which have zones of 0...ZONE_NORMAL,
897	 * set zone_last to ZONE_NORMAL.
898	 *
899	 * If we don't have HIGHMEM nor movable node,
900	 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
901	 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
902	 */
903	if (N_MEMORY == N_NORMAL_MEMORY)
904		zone_last = ZONE_MOVABLE;
905
906	/*
907	 * if the memory to be online is in a zone of 0...zone_last, and
908	 * the zones of 0...zone_last don't have memory before online, we will
909	 * need to set the node to node_states[N_NORMAL_MEMORY] after
910	 * the memory is online.
911	 */
912	if (zone_idx(zone) <= zone_last && !node_state(nid, N_NORMAL_MEMORY))
913		arg->status_change_nid_normal = nid;
914	else
915		arg->status_change_nid_normal = -1;
916
917#ifdef CONFIG_HIGHMEM
918	/*
919	 * If we have movable node, node_states[N_HIGH_MEMORY]
920	 * contains nodes which have zones of 0...ZONE_HIGHMEM,
921	 * set zone_last to ZONE_HIGHMEM.
922	 *
923	 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
924	 * contains nodes which have zones of 0...ZONE_MOVABLE,
925	 * set zone_last to ZONE_MOVABLE.
926	 */
927	zone_last = ZONE_HIGHMEM;
928	if (N_MEMORY == N_HIGH_MEMORY)
929		zone_last = ZONE_MOVABLE;
930
931	if (zone_idx(zone) <= zone_last && !node_state(nid, N_HIGH_MEMORY))
932		arg->status_change_nid_high = nid;
933	else
934		arg->status_change_nid_high = -1;
935#else
936	arg->status_change_nid_high = arg->status_change_nid_normal;
937#endif
938
939	/*
940	 * if the node don't have memory befor online, we will need to
941	 * set the node to node_states[N_MEMORY] after the memory
942	 * is online.
943	 */
944	if (!node_state(nid, N_MEMORY))
945		arg->status_change_nid = nid;
946	else
947		arg->status_change_nid = -1;
948}
949
950static void node_states_set_node(int node, struct memory_notify *arg)
951{
952	if (arg->status_change_nid_normal >= 0)
953		node_set_state(node, N_NORMAL_MEMORY);
954
955	if (arg->status_change_nid_high >= 0)
956		node_set_state(node, N_HIGH_MEMORY);
957
958	node_set_state(node, N_MEMORY);
959}
960
961
962/* Must be protected by mem_hotplug_begin() */
963int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type)
964{
965	unsigned long flags;
966	unsigned long onlined_pages = 0;
967	struct zone *zone;
968	int need_zonelists_rebuild = 0;
969	int nid;
970	int ret;
971	struct memory_notify arg;
972
973	/*
974	 * This doesn't need a lock to do pfn_to_page().
975	 * The section can't be removed here because of the
976	 * memory_block->state_mutex.
977	 */
978	zone = page_zone(pfn_to_page(pfn));
979
980	if ((zone_idx(zone) > ZONE_NORMAL ||
981	    online_type == MMOP_ONLINE_MOVABLE) &&
982	    !can_online_high_movable(zone))
983		return -EINVAL;
984
985	if (online_type == MMOP_ONLINE_KERNEL &&
986	    zone_idx(zone) == ZONE_MOVABLE) {
987		if (move_pfn_range_left(zone - 1, zone, pfn, pfn + nr_pages))
988			return -EINVAL;
989	}
990	if (online_type == MMOP_ONLINE_MOVABLE &&
991	    zone_idx(zone) == ZONE_MOVABLE - 1) {
992		if (move_pfn_range_right(zone, zone + 1, pfn, pfn + nr_pages))
993			return -EINVAL;
994	}
995
996	/* Previous code may changed the zone of the pfn range */
997	zone = page_zone(pfn_to_page(pfn));
998
999	arg.start_pfn = pfn;
1000	arg.nr_pages = nr_pages;
1001	node_states_check_changes_online(nr_pages, zone, &arg);
1002
1003	nid = pfn_to_nid(pfn);
1004
1005	ret = memory_notify(MEM_GOING_ONLINE, &arg);
1006	ret = notifier_to_errno(ret);
1007	if (ret) {
1008		memory_notify(MEM_CANCEL_ONLINE, &arg);
1009		return ret;
1010	}
1011	/*
1012	 * If this zone is not populated, then it is not in zonelist.
1013	 * This means the page allocator ignores this zone.
1014	 * So, zonelist must be updated after online.
1015	 */
1016	mutex_lock(&zonelists_mutex);
1017	if (!populated_zone(zone)) {
1018		need_zonelists_rebuild = 1;
1019		build_all_zonelists(NULL, zone);
1020	}
1021
1022	ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
1023		online_pages_range);
1024	if (ret) {
1025		if (need_zonelists_rebuild)
1026			zone_pcp_reset(zone);
1027		mutex_unlock(&zonelists_mutex);
1028		printk(KERN_DEBUG "online_pages [mem %#010llx-%#010llx] failed\n",
1029		       (unsigned long long) pfn << PAGE_SHIFT,
1030		       (((unsigned long long) pfn + nr_pages)
1031			    << PAGE_SHIFT) - 1);
1032		memory_notify(MEM_CANCEL_ONLINE, &arg);
1033		return ret;
1034	}
1035
1036	zone->present_pages += onlined_pages;
1037
1038	pgdat_resize_lock(zone->zone_pgdat, &flags);
1039	zone->zone_pgdat->node_present_pages += onlined_pages;
1040	pgdat_resize_unlock(zone->zone_pgdat, &flags);
1041
1042	if (onlined_pages) {
1043		node_states_set_node(zone_to_nid(zone), &arg);
1044		if (need_zonelists_rebuild)
1045			build_all_zonelists(NULL, NULL);
1046		else
1047			zone_pcp_update(zone);
1048	}
1049
1050	mutex_unlock(&zonelists_mutex);
1051
1052	init_per_zone_wmark_min();
1053
1054	if (onlined_pages)
1055		kswapd_run(zone_to_nid(zone));
1056
1057	vm_total_pages = nr_free_pagecache_pages();
1058
1059	writeback_set_ratelimit();
1060
1061	if (onlined_pages)
1062		memory_notify(MEM_ONLINE, &arg);
1063	return 0;
1064}
1065#endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
1066
1067static void reset_node_present_pages(pg_data_t *pgdat)
1068{
1069	struct zone *z;
1070
1071	for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
1072		z->present_pages = 0;
1073
1074	pgdat->node_present_pages = 0;
1075}
1076
1077/* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1078static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
1079{
1080	struct pglist_data *pgdat;
1081	unsigned long zones_size[MAX_NR_ZONES] = {0};
1082	unsigned long zholes_size[MAX_NR_ZONES] = {0};
1083	unsigned long start_pfn = PFN_DOWN(start);
1084
1085	pgdat = NODE_DATA(nid);
1086	if (!pgdat) {
1087		pgdat = arch_alloc_nodedata(nid);
1088		if (!pgdat)
1089			return NULL;
1090
1091		arch_refresh_nodedata(nid, pgdat);
1092	} else {
1093		/* Reset the nr_zones and classzone_idx to 0 before reuse */
1094		pgdat->nr_zones = 0;
1095		pgdat->classzone_idx = 0;
1096	}
1097
1098	/* we can use NODE_DATA(nid) from here */
1099
1100	/* init node's zones as empty zones, we don't have any present pages.*/
1101	free_area_init_node(nid, zones_size, start_pfn, zholes_size);
1102
1103	/*
1104	 * The node we allocated has no zone fallback lists. For avoiding
1105	 * to access not-initialized zonelist, build here.
1106	 */
1107	mutex_lock(&zonelists_mutex);
1108	build_all_zonelists(pgdat, NULL);
1109	mutex_unlock(&zonelists_mutex);
1110
1111	/*
1112	 * zone->managed_pages is set to an approximate value in
1113	 * free_area_init_core(), which will cause
1114	 * /sys/device/system/node/nodeX/meminfo has wrong data.
1115	 * So reset it to 0 before any memory is onlined.
1116	 */
1117	reset_node_managed_pages(pgdat);
1118
1119	/*
1120	 * When memory is hot-added, all the memory is in offline state. So
1121	 * clear all zones' present_pages because they will be updated in
1122	 * online_pages() and offline_pages().
1123	 */
1124	reset_node_present_pages(pgdat);
1125
1126	return pgdat;
1127}
1128
1129static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
1130{
1131	arch_refresh_nodedata(nid, NULL);
1132	arch_free_nodedata(pgdat);
1133	return;
1134}
1135
1136
1137/**
1138 * try_online_node - online a node if offlined
1139 *
1140 * called by cpu_up() to online a node without onlined memory.
1141 */
1142int try_online_node(int nid)
1143{
1144	pg_data_t	*pgdat;
1145	int	ret;
1146
1147	if (node_online(nid))
1148		return 0;
1149
1150	mem_hotplug_begin();
1151	pgdat = hotadd_new_pgdat(nid, 0);
1152	if (!pgdat) {
1153		pr_err("Cannot online node %d due to NULL pgdat\n", nid);
1154		ret = -ENOMEM;
1155		goto out;
1156	}
1157	node_set_online(nid);
1158	ret = register_one_node(nid);
1159	BUG_ON(ret);
1160
1161	if (pgdat->node_zonelists->_zonerefs->zone == NULL) {
1162		mutex_lock(&zonelists_mutex);
1163		build_all_zonelists(NULL, NULL);
1164		mutex_unlock(&zonelists_mutex);
1165	}
1166
1167out:
1168	mem_hotplug_done();
1169	return ret;
1170}
1171
1172static int check_hotplug_memory_range(u64 start, u64 size)
1173{
1174	u64 start_pfn = PFN_DOWN(start);
1175	u64 nr_pages = size >> PAGE_SHIFT;
1176
1177	/* Memory range must be aligned with section */
1178	if ((start_pfn & ~PAGE_SECTION_MASK) ||
1179	    (nr_pages % PAGES_PER_SECTION) || (!nr_pages)) {
1180		pr_err("Section-unaligned hotplug range: start 0x%llx, size 0x%llx\n",
1181				(unsigned long long)start,
1182				(unsigned long long)size);
1183		return -EINVAL;
1184	}
1185
1186	return 0;
1187}
1188
1189/*
1190 * If movable zone has already been setup, newly added memory should be check.
1191 * If its address is higher than movable zone, it should be added as movable.
1192 * Without this check, movable zone may overlap with other zone.
1193 */
1194static int should_add_memory_movable(int nid, u64 start, u64 size)
1195{
1196	unsigned long start_pfn = start >> PAGE_SHIFT;
1197	pg_data_t *pgdat = NODE_DATA(nid);
1198	struct zone *movable_zone = pgdat->node_zones + ZONE_MOVABLE;
1199
1200	if (zone_is_empty(movable_zone))
1201		return 0;
1202
1203	if (movable_zone->zone_start_pfn <= start_pfn)
1204		return 1;
1205
1206	return 0;
1207}
1208
1209int zone_for_memory(int nid, u64 start, u64 size, int zone_default)
1210{
1211	if (should_add_memory_movable(nid, start, size))
1212		return ZONE_MOVABLE;
1213
1214	return zone_default;
1215}
1216
1217/* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1218int __ref add_memory(int nid, u64 start, u64 size)
1219{
1220	pg_data_t *pgdat = NULL;
1221	bool new_pgdat;
1222	bool new_node;
1223	struct resource *res;
1224	int ret;
1225
1226	ret = check_hotplug_memory_range(start, size);
1227	if (ret)
1228		return ret;
1229
1230	res = register_memory_resource(start, size);
1231	ret = -EEXIST;
1232	if (!res)
1233		return ret;
1234
1235	{	/* Stupid hack to suppress address-never-null warning */
1236		void *p = NODE_DATA(nid);
1237		new_pgdat = !p;
1238	}
1239
1240	mem_hotplug_begin();
1241
1242	new_node = !node_online(nid);
1243	if (new_node) {
1244		pgdat = hotadd_new_pgdat(nid, start);
1245		ret = -ENOMEM;
1246		if (!pgdat)
1247			goto error;
1248	}
1249
1250	/* call arch's memory hotadd */
1251	ret = arch_add_memory(nid, start, size);
1252
1253	if (ret < 0)
1254		goto error;
1255
1256	/* we online node here. we can't roll back from here. */
1257	node_set_online(nid);
1258
1259	if (new_node) {
1260		ret = register_one_node(nid);
1261		/*
1262		 * If sysfs file of new node can't create, cpu on the node
1263		 * can't be hot-added. There is no rollback way now.
1264		 * So, check by BUG_ON() to catch it reluctantly..
1265		 */
1266		BUG_ON(ret);
1267	}
1268
1269	/* create new memmap entry */
1270	firmware_map_add_hotplug(start, start + size, "System RAM");
1271
1272	goto out;
1273
1274error:
1275	/* rollback pgdat allocation and others */
1276	if (new_pgdat)
1277		rollback_node_hotadd(nid, pgdat);
1278	release_memory_resource(res);
1279
1280out:
1281	mem_hotplug_done();
1282	return ret;
1283}
1284EXPORT_SYMBOL_GPL(add_memory);
1285
1286#ifdef CONFIG_MEMORY_HOTREMOVE
1287/*
1288 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1289 * set and the size of the free page is given by page_order(). Using this,
1290 * the function determines if the pageblock contains only free pages.
1291 * Due to buddy contraints, a free page at least the size of a pageblock will
1292 * be located at the start of the pageblock
1293 */
1294static inline int pageblock_free(struct page *page)
1295{
1296	return PageBuddy(page) && page_order(page) >= pageblock_order;
1297}
1298
1299/* Return the start of the next active pageblock after a given page */
1300static struct page *next_active_pageblock(struct page *page)
1301{
1302	/* Ensure the starting page is pageblock-aligned */
1303	BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
1304
1305	/* If the entire pageblock is free, move to the end of free page */
1306	if (pageblock_free(page)) {
1307		int order;
1308		/* be careful. we don't have locks, page_order can be changed.*/
1309		order = page_order(page);
1310		if ((order < MAX_ORDER) && (order >= pageblock_order))
1311			return page + (1 << order);
1312	}
1313
1314	return page + pageblock_nr_pages;
1315}
1316
1317/* Checks if this range of memory is likely to be hot-removable. */
1318int is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
1319{
1320	struct page *page = pfn_to_page(start_pfn);
1321	struct page *end_page = page + nr_pages;
1322
1323	/* Check the starting page of each pageblock within the range */
1324	for (; page < end_page; page = next_active_pageblock(page)) {
1325		if (!is_pageblock_removable_nolock(page))
1326			return 0;
1327		cond_resched();
1328	}
1329
1330	/* All pageblocks in the memory block are likely to be hot-removable */
1331	return 1;
1332}
1333
1334/*
1335 * Confirm all pages in a range [start, end) is belongs to the same zone.
1336 */
1337int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn)
1338{
1339	unsigned long pfn;
1340	struct zone *zone = NULL;
1341	struct page *page;
1342	int i;
1343	for (pfn = start_pfn;
1344	     pfn < end_pfn;
1345	     pfn += MAX_ORDER_NR_PAGES) {
1346		i = 0;
1347		/* This is just a CONFIG_HOLES_IN_ZONE check.*/
1348		while ((i < MAX_ORDER_NR_PAGES) && !pfn_valid_within(pfn + i))
1349			i++;
1350		if (i == MAX_ORDER_NR_PAGES)
1351			continue;
1352		page = pfn_to_page(pfn + i);
1353		if (zone && page_zone(page) != zone)
1354			return 0;
1355		zone = page_zone(page);
1356	}
1357	return 1;
1358}
1359
1360/*
1361 * Scan pfn range [start,end) to find movable/migratable pages (LRU pages
1362 * and hugepages). We scan pfn because it's much easier than scanning over
1363 * linked list. This function returns the pfn of the first found movable
1364 * page if it's found, otherwise 0.
1365 */
1366static unsigned long scan_movable_pages(unsigned long start, unsigned long end)
1367{
1368	unsigned long pfn;
1369	struct page *page;
1370	for (pfn = start; pfn < end; pfn++) {
1371		if (pfn_valid(pfn)) {
1372			page = pfn_to_page(pfn);
1373			if (PageLRU(page))
1374				return pfn;
1375			if (PageHuge(page)) {
1376				if (page_huge_active(page))
1377					return pfn;
1378				else
1379					pfn = round_up(pfn + 1,
1380						1 << compound_order(page)) - 1;
1381			}
1382		}
1383	}
1384	return 0;
1385}
1386
1387#define NR_OFFLINE_AT_ONCE_PAGES	(256)
1388static int
1389do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1390{
1391	unsigned long pfn;
1392	struct page *page;
1393	int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
1394	int not_managed = 0;
1395	int ret = 0;
1396	LIST_HEAD(source);
1397
1398	for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
1399		if (!pfn_valid(pfn))
1400			continue;
1401		page = pfn_to_page(pfn);
1402
1403		if (PageHuge(page)) {
1404			struct page *head = compound_head(page);
1405			pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1;
1406			if (compound_order(head) > PFN_SECTION_SHIFT) {
1407				ret = -EBUSY;
1408				break;
1409			}
1410			if (isolate_huge_page(page, &source))
1411				move_pages -= 1 << compound_order(head);
1412			continue;
1413		}
1414
1415		if (!get_page_unless_zero(page))
1416			continue;
1417		/*
1418		 * We can skip free pages. And we can only deal with pages on
1419		 * LRU.
1420		 */
1421		ret = isolate_lru_page(page);
1422		if (!ret) { /* Success */
1423			put_page(page);
1424			list_add_tail(&page->lru, &source);
1425			move_pages--;
1426			inc_zone_page_state(page, NR_ISOLATED_ANON +
1427					    page_is_file_cache(page));
1428
1429		} else {
1430#ifdef CONFIG_DEBUG_VM
1431			printk(KERN_ALERT "removing pfn %lx from LRU failed\n",
1432			       pfn);
1433			dump_page(page, "failed to remove from LRU");
1434#endif
1435			put_page(page);
1436			/* Because we don't have big zone->lock. we should
1437			   check this again here. */
1438			if (page_count(page)) {
1439				not_managed++;
1440				ret = -EBUSY;
1441				break;
1442			}
1443		}
1444	}
1445	if (!list_empty(&source)) {
1446		if (not_managed) {
1447			putback_movable_pages(&source);
1448			goto out;
1449		}
1450
1451		/*
1452		 * alloc_migrate_target should be improooooved!!
1453		 * migrate_pages returns # of failed pages.
1454		 */
1455		ret = migrate_pages(&source, alloc_migrate_target, NULL, 0,
1456					MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
1457		if (ret)
1458			putback_movable_pages(&source);
1459	}
1460out:
1461	return ret;
1462}
1463
1464/*
1465 * remove from free_area[] and mark all as Reserved.
1466 */
1467static int
1468offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1469			void *data)
1470{
1471	__offline_isolated_pages(start, start + nr_pages);
1472	return 0;
1473}
1474
1475static void
1476offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
1477{
1478	walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
1479				offline_isolated_pages_cb);
1480}
1481
1482/*
1483 * Check all pages in range, recoreded as memory resource, are isolated.
1484 */
1485static int
1486check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1487			void *data)
1488{
1489	int ret;
1490	long offlined = *(long *)data;
1491	ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true);
1492	offlined = nr_pages;
1493	if (!ret)
1494		*(long *)data += offlined;
1495	return ret;
1496}
1497
1498static long
1499check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
1500{
1501	long offlined = 0;
1502	int ret;
1503
1504	ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
1505			check_pages_isolated_cb);
1506	if (ret < 0)
1507		offlined = (long)ret;
1508	return offlined;
1509}
1510
1511#ifdef CONFIG_MOVABLE_NODE
1512/*
1513 * When CONFIG_MOVABLE_NODE, we permit offlining of a node which doesn't have
1514 * normal memory.
1515 */
1516static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1517{
1518	return true;
1519}
1520#else /* CONFIG_MOVABLE_NODE */
1521/* ensure the node has NORMAL memory if it is still online */
1522static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1523{
1524	struct pglist_data *pgdat = zone->zone_pgdat;
1525	unsigned long present_pages = 0;
1526	enum zone_type zt;
1527
1528	for (zt = 0; zt <= ZONE_NORMAL; zt++)
1529		present_pages += pgdat->node_zones[zt].present_pages;
1530
1531	if (present_pages > nr_pages)
1532		return true;
1533
1534	present_pages = 0;
1535	for (; zt <= ZONE_MOVABLE; zt++)
1536		present_pages += pgdat->node_zones[zt].present_pages;
1537
1538	/*
1539	 * we can't offline the last normal memory until all
1540	 * higher memory is offlined.
1541	 */
1542	return present_pages == 0;
1543}
1544#endif /* CONFIG_MOVABLE_NODE */
1545
1546static int __init cmdline_parse_movable_node(char *p)
1547{
1548#ifdef CONFIG_MOVABLE_NODE
1549	/*
1550	 * Memory used by the kernel cannot be hot-removed because Linux
1551	 * cannot migrate the kernel pages. When memory hotplug is
1552	 * enabled, we should prevent memblock from allocating memory
1553	 * for the kernel.
1554	 *
1555	 * ACPI SRAT records all hotpluggable memory ranges. But before
1556	 * SRAT is parsed, we don't know about it.
1557	 *
1558	 * The kernel image is loaded into memory at very early time. We
1559	 * cannot prevent this anyway. So on NUMA system, we set any
1560	 * node the kernel resides in as un-hotpluggable.
1561	 *
1562	 * Since on modern servers, one node could have double-digit
1563	 * gigabytes memory, we can assume the memory around the kernel
1564	 * image is also un-hotpluggable. So before SRAT is parsed, just
1565	 * allocate memory near the kernel image to try the best to keep
1566	 * the kernel away from hotpluggable memory.
1567	 */
1568	memblock_set_bottom_up(true);
1569	movable_node_enabled = true;
1570#else
1571	pr_warn("movable_node option not supported\n");
1572#endif
1573	return 0;
1574}
1575early_param("movable_node", cmdline_parse_movable_node);
1576
1577/* check which state of node_states will be changed when offline memory */
1578static void node_states_check_changes_offline(unsigned long nr_pages,
1579		struct zone *zone, struct memory_notify *arg)
1580{
1581	struct pglist_data *pgdat = zone->zone_pgdat;
1582	unsigned long present_pages = 0;
1583	enum zone_type zt, zone_last = ZONE_NORMAL;
1584
1585	/*
1586	 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
1587	 * contains nodes which have zones of 0...ZONE_NORMAL,
1588	 * set zone_last to ZONE_NORMAL.
1589	 *
1590	 * If we don't have HIGHMEM nor movable node,
1591	 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
1592	 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
1593	 */
1594	if (N_MEMORY == N_NORMAL_MEMORY)
1595		zone_last = ZONE_MOVABLE;
1596
1597	/*
1598	 * check whether node_states[N_NORMAL_MEMORY] will be changed.
1599	 * If the memory to be offline is in a zone of 0...zone_last,
1600	 * and it is the last present memory, 0...zone_last will
1601	 * become empty after offline , thus we can determind we will
1602	 * need to clear the node from node_states[N_NORMAL_MEMORY].
1603	 */
1604	for (zt = 0; zt <= zone_last; zt++)
1605		present_pages += pgdat->node_zones[zt].present_pages;
1606	if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1607		arg->status_change_nid_normal = zone_to_nid(zone);
1608	else
1609		arg->status_change_nid_normal = -1;
1610
1611#ifdef CONFIG_HIGHMEM
1612	/*
1613	 * If we have movable node, node_states[N_HIGH_MEMORY]
1614	 * contains nodes which have zones of 0...ZONE_HIGHMEM,
1615	 * set zone_last to ZONE_HIGHMEM.
1616	 *
1617	 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
1618	 * contains nodes which have zones of 0...ZONE_MOVABLE,
1619	 * set zone_last to ZONE_MOVABLE.
1620	 */
1621	zone_last = ZONE_HIGHMEM;
1622	if (N_MEMORY == N_HIGH_MEMORY)
1623		zone_last = ZONE_MOVABLE;
1624
1625	for (; zt <= zone_last; zt++)
1626		present_pages += pgdat->node_zones[zt].present_pages;
1627	if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1628		arg->status_change_nid_high = zone_to_nid(zone);
1629	else
1630		arg->status_change_nid_high = -1;
1631#else
1632	arg->status_change_nid_high = arg->status_change_nid_normal;
1633#endif
1634
1635	/*
1636	 * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE
1637	 */
1638	zone_last = ZONE_MOVABLE;
1639
1640	/*
1641	 * check whether node_states[N_HIGH_MEMORY] will be changed
1642	 * If we try to offline the last present @nr_pages from the node,
1643	 * we can determind we will need to clear the node from
1644	 * node_states[N_HIGH_MEMORY].
1645	 */
1646	for (; zt <= zone_last; zt++)
1647		present_pages += pgdat->node_zones[zt].present_pages;
1648	if (nr_pages >= present_pages)
1649		arg->status_change_nid = zone_to_nid(zone);
1650	else
1651		arg->status_change_nid = -1;
1652}
1653
1654static void node_states_clear_node(int node, struct memory_notify *arg)
1655{
1656	if (arg->status_change_nid_normal >= 0)
1657		node_clear_state(node, N_NORMAL_MEMORY);
1658
1659	if ((N_MEMORY != N_NORMAL_MEMORY) &&
1660	    (arg->status_change_nid_high >= 0))
1661		node_clear_state(node, N_HIGH_MEMORY);
1662
1663	if ((N_MEMORY != N_HIGH_MEMORY) &&
1664	    (arg->status_change_nid >= 0))
1665		node_clear_state(node, N_MEMORY);
1666}
1667
1668static int __ref __offline_pages(unsigned long start_pfn,
1669		  unsigned long end_pfn, unsigned long timeout)
1670{
1671	unsigned long pfn, nr_pages, expire;
1672	long offlined_pages;
1673	int ret, drain, retry_max, node;
1674	unsigned long flags;
1675	struct zone *zone;
1676	struct memory_notify arg;
1677
1678	/* at least, alignment against pageblock is necessary */
1679	if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
1680		return -EINVAL;
1681	if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
1682		return -EINVAL;
1683	/* This makes hotplug much easier...and readable.
1684	   we assume this for now. .*/
1685	if (!test_pages_in_a_zone(start_pfn, end_pfn))
1686		return -EINVAL;
1687
1688	zone = page_zone(pfn_to_page(start_pfn));
1689	node = zone_to_nid(zone);
1690	nr_pages = end_pfn - start_pfn;
1691
1692	if (zone_idx(zone) <= ZONE_NORMAL && !can_offline_normal(zone, nr_pages))
1693		return -EINVAL;
1694
1695	/* set above range as isolated */
1696	ret = start_isolate_page_range(start_pfn, end_pfn,
1697				       MIGRATE_MOVABLE, true);
1698	if (ret)
1699		return ret;
1700
1701	arg.start_pfn = start_pfn;
1702	arg.nr_pages = nr_pages;
1703	node_states_check_changes_offline(nr_pages, zone, &arg);
1704
1705	ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1706	ret = notifier_to_errno(ret);
1707	if (ret)
1708		goto failed_removal;
1709
1710	pfn = start_pfn;
1711	expire = jiffies + timeout;
1712	drain = 0;
1713	retry_max = 5;
1714repeat:
1715	/* start memory hot removal */
1716	ret = -EAGAIN;
1717	if (time_after(jiffies, expire))
1718		goto failed_removal;
1719	ret = -EINTR;
1720	if (signal_pending(current))
1721		goto failed_removal;
1722	ret = 0;
1723	if (drain) {
1724		lru_add_drain_all();
1725		cond_resched();
1726		drain_all_pages(zone);
1727	}
1728
1729	pfn = scan_movable_pages(start_pfn, end_pfn);
1730	if (pfn) { /* We have movable pages */
1731		ret = do_migrate_range(pfn, end_pfn);
1732		if (!ret) {
1733			drain = 1;
1734			goto repeat;
1735		} else {
1736			if (ret < 0)
1737				if (--retry_max == 0)
1738					goto failed_removal;
1739			yield();
1740			drain = 1;
1741			goto repeat;
1742		}
1743	}
1744	/* drain all zone's lru pagevec, this is asynchronous... */
1745	lru_add_drain_all();
1746	yield();
1747	/* drain pcp pages, this is synchronous. */
1748	drain_all_pages(zone);
1749	/*
1750	 * dissolve free hugepages in the memory block before doing offlining
1751	 * actually in order to make hugetlbfs's object counting consistent.
1752	 */
1753	dissolve_free_huge_pages(start_pfn, end_pfn);
1754	/* check again */
1755	offlined_pages = check_pages_isolated(start_pfn, end_pfn);
1756	if (offlined_pages < 0) {
1757		ret = -EBUSY;
1758		goto failed_removal;
1759	}
1760	printk(KERN_INFO "Offlined Pages %ld\n", offlined_pages);
1761	/* Ok, all of our target is isolated.
1762	   We cannot do rollback at this point. */
1763	offline_isolated_pages(start_pfn, end_pfn);
1764	/* reset pagetype flags and makes migrate type to be MOVABLE */
1765	undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1766	/* removal success */
1767	adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages);
1768	zone->present_pages -= offlined_pages;
1769
1770	pgdat_resize_lock(zone->zone_pgdat, &flags);
1771	zone->zone_pgdat->node_present_pages -= offlined_pages;
1772	pgdat_resize_unlock(zone->zone_pgdat, &flags);
1773
1774	init_per_zone_wmark_min();
1775
1776	if (!populated_zone(zone)) {
1777		zone_pcp_reset(zone);
1778		mutex_lock(&zonelists_mutex);
1779		build_all_zonelists(NULL, NULL);
1780		mutex_unlock(&zonelists_mutex);
1781	} else
1782		zone_pcp_update(zone);
1783
1784	node_states_clear_node(node, &arg);
1785	if (arg.status_change_nid >= 0)
1786		kswapd_stop(node);
1787
1788	vm_total_pages = nr_free_pagecache_pages();
1789	writeback_set_ratelimit();
1790
1791	memory_notify(MEM_OFFLINE, &arg);
1792	return 0;
1793
1794failed_removal:
1795	printk(KERN_INFO "memory offlining [mem %#010llx-%#010llx] failed\n",
1796	       (unsigned long long) start_pfn << PAGE_SHIFT,
1797	       ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
1798	memory_notify(MEM_CANCEL_OFFLINE, &arg);
1799	/* pushback to free area */
1800	undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1801	return ret;
1802}
1803
1804/* Must be protected by mem_hotplug_begin() */
1805int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1806{
1807	return __offline_pages(start_pfn, start_pfn + nr_pages, 120 * HZ);
1808}
1809#endif /* CONFIG_MEMORY_HOTREMOVE */
1810
1811/**
1812 * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
1813 * @start_pfn: start pfn of the memory range
1814 * @end_pfn: end pfn of the memory range
1815 * @arg: argument passed to func
1816 * @func: callback for each memory section walked
1817 *
1818 * This function walks through all present mem sections in range
1819 * [start_pfn, end_pfn) and call func on each mem section.
1820 *
1821 * Returns the return value of func.
1822 */
1823int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn,
1824		void *arg, int (*func)(struct memory_block *, void *))
1825{
1826	struct memory_block *mem = NULL;
1827	struct mem_section *section;
1828	unsigned long pfn, section_nr;
1829	int ret;
1830
1831	for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1832		section_nr = pfn_to_section_nr(pfn);
1833		if (!present_section_nr(section_nr))
1834			continue;
1835
1836		section = __nr_to_section(section_nr);
1837		/* same memblock? */
1838		if (mem)
1839			if ((section_nr >= mem->start_section_nr) &&
1840			    (section_nr <= mem->end_section_nr))
1841				continue;
1842
1843		mem = find_memory_block_hinted(section, mem);
1844		if (!mem)
1845			continue;
1846
1847		ret = func(mem, arg);
1848		if (ret) {
1849			kobject_put(&mem->dev.kobj);
1850			return ret;
1851		}
1852	}
1853
1854	if (mem)
1855		kobject_put(&mem->dev.kobj);
1856
1857	return 0;
1858}
1859
1860#ifdef CONFIG_MEMORY_HOTREMOVE
1861static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
1862{
1863	int ret = !is_memblock_offlined(mem);
1864
1865	if (unlikely(ret)) {
1866		phys_addr_t beginpa, endpa;
1867
1868		beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
1869		endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1;
1870		pr_warn("removing memory fails, because memory "
1871			"[%pa-%pa] is onlined\n",
1872			&beginpa, &endpa);
1873	}
1874
1875	return ret;
1876}
1877
1878static int check_cpu_on_node(pg_data_t *pgdat)
1879{
1880	int cpu;
1881
1882	for_each_present_cpu(cpu) {
1883		if (cpu_to_node(cpu) == pgdat->node_id)
1884			/*
1885			 * the cpu on this node isn't removed, and we can't
1886			 * offline this node.
1887			 */
1888			return -EBUSY;
1889	}
1890
1891	return 0;
1892}
1893
1894static void unmap_cpu_on_node(pg_data_t *pgdat)
1895{
1896#ifdef CONFIG_ACPI_NUMA
1897	int cpu;
1898
1899	for_each_possible_cpu(cpu)
1900		if (cpu_to_node(cpu) == pgdat->node_id)
1901			numa_clear_node(cpu);
1902#endif
1903}
1904
1905static int check_and_unmap_cpu_on_node(pg_data_t *pgdat)
1906{
1907	int ret;
1908
1909	ret = check_cpu_on_node(pgdat);
1910	if (ret)
1911		return ret;
1912
1913	/*
1914	 * the node will be offlined when we come here, so we can clear
1915	 * the cpu_to_node() now.
1916	 */
1917
1918	unmap_cpu_on_node(pgdat);
1919	return 0;
1920}
1921
1922/**
1923 * try_offline_node
1924 *
1925 * Offline a node if all memory sections and cpus of the node are removed.
1926 *
1927 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1928 * and online/offline operations before this call.
1929 */
1930void try_offline_node(int nid)
1931{
1932	pg_data_t *pgdat = NODE_DATA(nid);
1933	unsigned long start_pfn = pgdat->node_start_pfn;
1934	unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
1935	unsigned long pfn;
1936	int i;
1937
1938	for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1939		unsigned long section_nr = pfn_to_section_nr(pfn);
1940
1941		if (!present_section_nr(section_nr))
1942			continue;
1943
1944		if (pfn_to_nid(pfn) != nid)
1945			continue;
1946
1947		/*
1948		 * some memory sections of this node are not removed, and we
1949		 * can't offline node now.
1950		 */
1951		return;
1952	}
1953
1954	if (check_and_unmap_cpu_on_node(pgdat))
1955		return;
1956
1957	/*
1958	 * all memory/cpu of this node are removed, we can offline this
1959	 * node now.
1960	 */
1961	node_set_offline(nid);
1962	unregister_one_node(nid);
1963
1964	/* free waittable in each zone */
1965	for (i = 0; i < MAX_NR_ZONES; i++) {
1966		struct zone *zone = pgdat->node_zones + i;
1967
1968		/*
1969		 * wait_table may be allocated from boot memory,
1970		 * here only free if it's allocated by vmalloc.
1971		 */
1972		if (is_vmalloc_addr(zone->wait_table)) {
1973			vfree(zone->wait_table);
1974			zone->wait_table = NULL;
1975		}
1976	}
1977}
1978EXPORT_SYMBOL(try_offline_node);
1979
1980/**
1981 * remove_memory
1982 *
1983 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1984 * and online/offline operations before this call, as required by
1985 * try_offline_node().
1986 */
1987void __ref remove_memory(int nid, u64 start, u64 size)
1988{
1989	int ret;
1990
1991	BUG_ON(check_hotplug_memory_range(start, size));
1992
1993	mem_hotplug_begin();
1994
1995	/*
1996	 * All memory blocks must be offlined before removing memory.  Check
1997	 * whether all memory blocks in question are offline and trigger a BUG()
1998	 * if this is not the case.
1999	 */
2000	ret = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL,
2001				check_memblock_offlined_cb);
2002	if (ret)
2003		BUG();
2004
2005	/* remove memmap entry */
2006	firmware_map_remove(start, start + size, "System RAM");
2007
2008	arch_remove_memory(start, size);
2009
2010	try_offline_node(nid);
2011
2012	mem_hotplug_done();
2013}
2014EXPORT_SYMBOL_GPL(remove_memory);
2015#endif /* CONFIG_MEMORY_HOTREMOVE */
2016