xref: /linux/mm/memory_hotplug.c (revision 367b8112fe2ea5c39a7bb4d263dcdd9b612fae18)
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 #include <linux/pfn.h>
30 
31 #include <asm/tlbflush.h>
32 
33 #include "internal.h"
34 
35 /* add this memory to iomem resource */
36 static struct resource *register_memory_resource(u64 start, u64 size)
37 {
38 	struct resource *res;
39 	res = kzalloc(sizeof(struct resource), GFP_KERNEL);
40 	BUG_ON(!res);
41 
42 	res->name = "System RAM";
43 	res->start = start;
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);
49 		kfree(res);
50 		res = NULL;
51 	}
52 	return res;
53 }
54 
55 static void release_memory_resource(struct resource *res)
56 {
57 	if (!res)
58 		return;
59 	release_resource(res);
60 	kfree(res);
61 	return;
62 }
63 
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)
67 {
68 	atomic_set(&page->_mapcount, type);
69 	SetPagePrivate(page);
70 	set_page_private(page, info);
71 	atomic_inc(&page->_count);
72 }
73 
74 void put_page_bootmem(struct page *page)
75 {
76 	int type;
77 
78 	type = atomic_read(&page->_mapcount);
79 	BUG_ON(type >= -1);
80 
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);
86 	}
87 
88 }
89 
90 static void register_page_bootmem_info_section(unsigned long start_pfn)
91 {
92 	unsigned long *usemap, mapsize, section_nr, i;
93 	struct mem_section *ms;
94 	struct page *page, *memmap;
95 
96 	if (!pfn_valid(start_pfn))
97 		return;
98 
99 	section_nr = pfn_to_section_nr(start_pfn);
100 	ms = __nr_to_section(section_nr);
101 
102 	/* Get section's memmap address */
103 	memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
104 
105 	/*
106 	 * Get page for the memmap's phys address
107 	 * XXX: need more consideration for sparse_vmemmap...
108 	 */
109 	page = virt_to_page(memmap);
110 	mapsize = sizeof(struct page) * PAGES_PER_SECTION;
111 	mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
112 
113 	/* remember memmap's page */
114 	for (i = 0; i < mapsize; i++, page++)
115 		get_page_bootmem(section_nr, page, SECTION_INFO);
116 
117 	usemap = __nr_to_section(section_nr)->pageblock_flags;
118 	page = virt_to_page(usemap);
119 
120 	mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
121 
122 	for (i = 0; i < mapsize; i++, page++)
123 		get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
124 
125 }
126 
127 void register_page_bootmem_info_node(struct pglist_data *pgdat)
128 {
129 	unsigned long i, pfn, end_pfn, nr_pages;
130 	int node = pgdat->node_id;
131 	struct page *page;
132 	struct zone *zone;
133 
134 	nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
135 	page = virt_to_page(pgdat);
136 
137 	for (i = 0; i < nr_pages; i++, page++)
138 		get_page_bootmem(node, page, NODE_INFO);
139 
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);
147 
148 			for (i = 0; i < nr_pages; i++, page++)
149 				get_page_bootmem(node, page, NODE_INFO);
150 		}
151 	}
152 
153 	pfn = pgdat->node_start_pfn;
154 	end_pfn = pfn + pgdat->node_spanned_pages;
155 
156 	/* register_section info */
157 	for (; pfn < end_pfn; pfn += PAGES_PER_SECTION)
158 		register_page_bootmem_info_section(pfn);
159 
160 }
161 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
162 
163 static void grow_zone_span(struct zone *zone, unsigned long start_pfn,
164 			   unsigned long end_pfn)
165 {
166 	unsigned long old_zone_end_pfn;
167 
168 	zone_span_writelock(zone);
169 
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;
173 
174 	zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
175 				zone->zone_start_pfn;
176 
177 	zone_span_writeunlock(zone);
178 }
179 
180 static void grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn,
181 			    unsigned long end_pfn)
182 {
183 	unsigned long old_pgdat_end_pfn =
184 		pgdat->node_start_pfn + pgdat->node_spanned_pages;
185 
186 	if (start_pfn < pgdat->node_start_pfn)
187 		pgdat->node_start_pfn = start_pfn;
188 
189 	pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
190 					pgdat->node_start_pfn;
191 }
192 
193 static int __add_zone(struct zone *zone, unsigned long phys_start_pfn)
194 {
195 	struct pglist_data *pgdat = zone->zone_pgdat;
196 	int nr_pages = PAGES_PER_SECTION;
197 	int nid = pgdat->node_id;
198 	int zone_type;
199 	unsigned long flags;
200 
201 	zone_type = zone - pgdat->node_zones;
202 	if (!zone->wait_table) {
203 		int ret;
204 
205 		ret = init_currently_empty_zone(zone, phys_start_pfn,
206 						nr_pages, MEMMAP_HOTPLUG);
207 		if (ret)
208 			return ret;
209 	}
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);
217 	return 0;
218 }
219 
220 static int __add_section(struct zone *zone, unsigned long phys_start_pfn)
221 {
222 	int nr_pages = PAGES_PER_SECTION;
223 	int ret;
224 
225 	if (pfn_valid(phys_start_pfn))
226 		return -EEXIST;
227 
228 	ret = sparse_add_one_section(zone, phys_start_pfn, nr_pages);
229 
230 	if (ret < 0)
231 		return ret;
232 
233 	ret = __add_zone(zone, phys_start_pfn);
234 
235 	if (ret < 0)
236 		return ret;
237 
238 	return register_new_memory(__pfn_to_section(phys_start_pfn));
239 }
240 
241 #ifdef CONFIG_SPARSEMEM_VMEMMAP
242 static int __remove_section(struct zone *zone, struct mem_section *ms)
243 {
244 	/*
245 	 * XXX: Freeing memmap with vmemmap is not implement yet.
246 	 *      This should be removed later.
247 	 */
248 	return -EBUSY;
249 }
250 #else
251 static int __remove_section(struct zone *zone, struct mem_section *ms)
252 {
253 	unsigned long flags;
254 	struct pglist_data *pgdat = zone->zone_pgdat;
255 	int ret = -EINVAL;
256 
257 	if (!valid_section(ms))
258 		return ret;
259 
260 	ret = unregister_memory_section(ms);
261 	if (ret)
262 		return ret;
263 
264 	pgdat_resize_lock(pgdat, &flags);
265 	sparse_remove_one_section(zone, ms);
266 	pgdat_resize_unlock(pgdat, &flags);
267 	return 0;
268 }
269 #endif
270 
271 /*
272  * Reasonably generic function for adding memory.  It is
273  * expected that archs that support memory hotplug will
274  * call this function after deciding the zone to which to
275  * add the new pages.
276  */
277 int __add_pages(struct zone *zone, unsigned long phys_start_pfn,
278 		 unsigned long nr_pages)
279 {
280 	unsigned long i;
281 	int err = 0;
282 	int start_sec, end_sec;
283 	/* during initialize mem_map, align hot-added range to section */
284 	start_sec = pfn_to_section_nr(phys_start_pfn);
285 	end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
286 
287 	for (i = start_sec; i <= end_sec; i++) {
288 		err = __add_section(zone, i << PFN_SECTION_SHIFT);
289 
290 		/*
291 		 * EEXIST is finally dealt with by ioresource collision
292 		 * check. see add_memory() => register_memory_resource()
293 		 * Warning will be printed if there is collision.
294 		 */
295 		if (err && (err != -EEXIST))
296 			break;
297 		err = 0;
298 	}
299 
300 	return err;
301 }
302 EXPORT_SYMBOL_GPL(__add_pages);
303 
304 /**
305  * __remove_pages() - remove sections of pages from a zone
306  * @zone: zone from which pages need to be removed
307  * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
308  * @nr_pages: number of pages to remove (must be multiple of section size)
309  *
310  * Generic helper function to remove section mappings and sysfs entries
311  * for the section of the memory we are removing. Caller needs to make
312  * sure that pages are marked reserved and zones are adjust properly by
313  * calling offline_pages().
314  */
315 int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
316 		 unsigned long nr_pages)
317 {
318 	unsigned long i, ret = 0;
319 	int sections_to_remove;
320 
321 	/*
322 	 * We can only remove entire sections
323 	 */
324 	BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
325 	BUG_ON(nr_pages % PAGES_PER_SECTION);
326 
327 	sections_to_remove = nr_pages / PAGES_PER_SECTION;
328 	for (i = 0; i < sections_to_remove; i++) {
329 		unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
330 		release_mem_region(pfn << PAGE_SHIFT,
331 				   PAGES_PER_SECTION << PAGE_SHIFT);
332 		ret = __remove_section(zone, __pfn_to_section(pfn));
333 		if (ret)
334 			break;
335 	}
336 	return ret;
337 }
338 EXPORT_SYMBOL_GPL(__remove_pages);
339 
340 void online_page(struct page *page)
341 {
342 	totalram_pages++;
343 	num_physpages++;
344 
345 #ifdef CONFIG_HIGHMEM
346 	if (PageHighMem(page))
347 		totalhigh_pages++;
348 #endif
349 
350 #ifdef CONFIG_FLATMEM
351 	max_mapnr = max(page_to_pfn(page), max_mapnr);
352 #endif
353 
354 	ClearPageReserved(page);
355 	init_page_count(page);
356 	__free_page(page);
357 }
358 
359 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
360 			void *arg)
361 {
362 	unsigned long i;
363 	unsigned long onlined_pages = *(unsigned long *)arg;
364 	struct page *page;
365 	if (PageReserved(pfn_to_page(start_pfn)))
366 		for (i = 0; i < nr_pages; i++) {
367 			page = pfn_to_page(start_pfn + i);
368 			online_page(page);
369 			onlined_pages++;
370 		}
371 	*(unsigned long *)arg = onlined_pages;
372 	return 0;
373 }
374 
375 
376 int online_pages(unsigned long pfn, unsigned long nr_pages)
377 {
378 	unsigned long onlined_pages = 0;
379 	struct zone *zone;
380 	int need_zonelists_rebuild = 0;
381 	int nid;
382 	int ret;
383 	struct memory_notify arg;
384 
385 	arg.start_pfn = pfn;
386 	arg.nr_pages = nr_pages;
387 	arg.status_change_nid = -1;
388 
389 	nid = page_to_nid(pfn_to_page(pfn));
390 	if (node_present_pages(nid) == 0)
391 		arg.status_change_nid = nid;
392 
393 	ret = memory_notify(MEM_GOING_ONLINE, &arg);
394 	ret = notifier_to_errno(ret);
395 	if (ret) {
396 		memory_notify(MEM_CANCEL_ONLINE, &arg);
397 		return ret;
398 	}
399 	/*
400 	 * This doesn't need a lock to do pfn_to_page().
401 	 * The section can't be removed here because of the
402 	 * memory_block->state_mutex.
403 	 */
404 	zone = page_zone(pfn_to_page(pfn));
405 	/*
406 	 * If this zone is not populated, then it is not in zonelist.
407 	 * This means the page allocator ignores this zone.
408 	 * So, zonelist must be updated after online.
409 	 */
410 	if (!populated_zone(zone))
411 		need_zonelists_rebuild = 1;
412 
413 	ret = walk_memory_resource(pfn, nr_pages, &onlined_pages,
414 		online_pages_range);
415 	if (ret) {
416 		printk(KERN_DEBUG "online_pages %lx at %lx failed\n",
417 			nr_pages, pfn);
418 		memory_notify(MEM_CANCEL_ONLINE, &arg);
419 		return ret;
420 	}
421 
422 	zone->present_pages += onlined_pages;
423 	zone->zone_pgdat->node_present_pages += onlined_pages;
424 
425 	setup_per_zone_pages_min();
426 	if (onlined_pages) {
427 		kswapd_run(zone_to_nid(zone));
428 		node_set_state(zone_to_nid(zone), N_HIGH_MEMORY);
429 	}
430 
431 	if (need_zonelists_rebuild)
432 		build_all_zonelists();
433 	else
434 		vm_total_pages = nr_free_pagecache_pages();
435 
436 	writeback_set_ratelimit();
437 
438 	if (onlined_pages)
439 		memory_notify(MEM_ONLINE, &arg);
440 
441 	return 0;
442 }
443 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
444 
445 static pg_data_t *hotadd_new_pgdat(int nid, u64 start)
446 {
447 	struct pglist_data *pgdat;
448 	unsigned long zones_size[MAX_NR_ZONES] = {0};
449 	unsigned long zholes_size[MAX_NR_ZONES] = {0};
450 	unsigned long start_pfn = start >> PAGE_SHIFT;
451 
452 	pgdat = arch_alloc_nodedata(nid);
453 	if (!pgdat)
454 		return NULL;
455 
456 	arch_refresh_nodedata(nid, pgdat);
457 
458 	/* we can use NODE_DATA(nid) from here */
459 
460 	/* init node's zones as empty zones, we don't have any present pages.*/
461 	free_area_init_node(nid, zones_size, start_pfn, zholes_size);
462 
463 	return pgdat;
464 }
465 
466 static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
467 {
468 	arch_refresh_nodedata(nid, NULL);
469 	arch_free_nodedata(pgdat);
470 	return;
471 }
472 
473 
474 int add_memory(int nid, u64 start, u64 size)
475 {
476 	pg_data_t *pgdat = NULL;
477 	int new_pgdat = 0;
478 	struct resource *res;
479 	int ret;
480 
481 	res = register_memory_resource(start, size);
482 	if (!res)
483 		return -EEXIST;
484 
485 	if (!node_online(nid)) {
486 		pgdat = hotadd_new_pgdat(nid, start);
487 		if (!pgdat)
488 			return -ENOMEM;
489 		new_pgdat = 1;
490 	}
491 
492 	/* call arch's memory hotadd */
493 	ret = arch_add_memory(nid, start, size);
494 
495 	if (ret < 0)
496 		goto error;
497 
498 	/* we online node here. we can't roll back from here. */
499 	node_set_online(nid);
500 
501 	cpuset_track_online_nodes();
502 
503 	if (new_pgdat) {
504 		ret = register_one_node(nid);
505 		/*
506 		 * If sysfs file of new node can't create, cpu on the node
507 		 * can't be hot-added. There is no rollback way now.
508 		 * So, check by BUG_ON() to catch it reluctantly..
509 		 */
510 		BUG_ON(ret);
511 	}
512 
513 	return ret;
514 error:
515 	/* rollback pgdat allocation and others */
516 	if (new_pgdat)
517 		rollback_node_hotadd(nid, pgdat);
518 	if (res)
519 		release_memory_resource(res);
520 
521 	return ret;
522 }
523 EXPORT_SYMBOL_GPL(add_memory);
524 
525 #ifdef CONFIG_MEMORY_HOTREMOVE
526 /*
527  * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
528  * set and the size of the free page is given by page_order(). Using this,
529  * the function determines if the pageblock contains only free pages.
530  * Due to buddy contraints, a free page at least the size of a pageblock will
531  * be located at the start of the pageblock
532  */
533 static inline int pageblock_free(struct page *page)
534 {
535 	return PageBuddy(page) && page_order(page) >= pageblock_order;
536 }
537 
538 /* Return the start of the next active pageblock after a given page */
539 static struct page *next_active_pageblock(struct page *page)
540 {
541 	int pageblocks_stride;
542 
543 	/* Ensure the starting page is pageblock-aligned */
544 	BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
545 
546 	/* Move forward by at least 1 * pageblock_nr_pages */
547 	pageblocks_stride = 1;
548 
549 	/* If the entire pageblock is free, move to the end of free page */
550 	if (pageblock_free(page))
551 		pageblocks_stride += page_order(page) - pageblock_order;
552 
553 	return page + (pageblocks_stride * pageblock_nr_pages);
554 }
555 
556 /* Checks if this range of memory is likely to be hot-removable. */
557 int is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
558 {
559 	int type;
560 	struct page *page = pfn_to_page(start_pfn);
561 	struct page *end_page = page + nr_pages;
562 
563 	/* Check the starting page of each pageblock within the range */
564 	for (; page < end_page; page = next_active_pageblock(page)) {
565 		type = get_pageblock_migratetype(page);
566 
567 		/*
568 		 * A pageblock containing MOVABLE or free pages is considered
569 		 * removable
570 		 */
571 		if (type != MIGRATE_MOVABLE && !pageblock_free(page))
572 			return 0;
573 
574 		/*
575 		 * A pageblock starting with a PageReserved page is not
576 		 * considered removable.
577 		 */
578 		if (PageReserved(page))
579 			return 0;
580 	}
581 
582 	/* All pageblocks in the memory block are likely to be hot-removable */
583 	return 1;
584 }
585 
586 /*
587  * Confirm all pages in a range [start, end) is belongs to the same zone.
588  */
589 static int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn)
590 {
591 	unsigned long pfn;
592 	struct zone *zone = NULL;
593 	struct page *page;
594 	int i;
595 	for (pfn = start_pfn;
596 	     pfn < end_pfn;
597 	     pfn += MAX_ORDER_NR_PAGES) {
598 		i = 0;
599 		/* This is just a CONFIG_HOLES_IN_ZONE check.*/
600 		while ((i < MAX_ORDER_NR_PAGES) && !pfn_valid_within(pfn + i))
601 			i++;
602 		if (i == MAX_ORDER_NR_PAGES)
603 			continue;
604 		page = pfn_to_page(pfn + i);
605 		if (zone && page_zone(page) != zone)
606 			return 0;
607 		zone = page_zone(page);
608 	}
609 	return 1;
610 }
611 
612 /*
613  * Scanning pfn is much easier than scanning lru list.
614  * Scan pfn from start to end and Find LRU page.
615  */
616 int scan_lru_pages(unsigned long start, unsigned long end)
617 {
618 	unsigned long pfn;
619 	struct page *page;
620 	for (pfn = start; pfn < end; pfn++) {
621 		if (pfn_valid(pfn)) {
622 			page = pfn_to_page(pfn);
623 			if (PageLRU(page))
624 				return pfn;
625 		}
626 	}
627 	return 0;
628 }
629 
630 static struct page *
631 hotremove_migrate_alloc(struct page *page,
632 			unsigned long private,
633 			int **x)
634 {
635 	/* This should be improoooooved!! */
636 	return alloc_page(GFP_HIGHUSER_PAGECACHE);
637 }
638 
639 
640 #define NR_OFFLINE_AT_ONCE_PAGES	(256)
641 static int
642 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
643 {
644 	unsigned long pfn;
645 	struct page *page;
646 	int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
647 	int not_managed = 0;
648 	int ret = 0;
649 	LIST_HEAD(source);
650 
651 	for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
652 		if (!pfn_valid(pfn))
653 			continue;
654 		page = pfn_to_page(pfn);
655 		if (!page_count(page))
656 			continue;
657 		/*
658 		 * We can skip free pages. And we can only deal with pages on
659 		 * LRU.
660 		 */
661 		ret = isolate_lru_page(page);
662 		if (!ret) { /* Success */
663 			list_add_tail(&page->lru, &source);
664 			move_pages--;
665 		} else {
666 			/* Becasue we don't have big zone->lock. we should
667 			   check this again here. */
668 			if (page_count(page))
669 				not_managed++;
670 #ifdef CONFIG_DEBUG_VM
671 			printk(KERN_INFO "removing from LRU failed"
672 					 " %lx/%d/%lx\n",
673 				pfn, page_count(page), page->flags);
674 #endif
675 		}
676 	}
677 	ret = -EBUSY;
678 	if (not_managed) {
679 		if (!list_empty(&source))
680 			putback_lru_pages(&source);
681 		goto out;
682 	}
683 	ret = 0;
684 	if (list_empty(&source))
685 		goto out;
686 	/* this function returns # of failed pages */
687 	ret = migrate_pages(&source, hotremove_migrate_alloc, 0);
688 
689 out:
690 	return ret;
691 }
692 
693 /*
694  * remove from free_area[] and mark all as Reserved.
695  */
696 static int
697 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
698 			void *data)
699 {
700 	__offline_isolated_pages(start, start + nr_pages);
701 	return 0;
702 }
703 
704 static void
705 offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
706 {
707 	walk_memory_resource(start_pfn, end_pfn - start_pfn, NULL,
708 				offline_isolated_pages_cb);
709 }
710 
711 /*
712  * Check all pages in range, recoreded as memory resource, are isolated.
713  */
714 static int
715 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
716 			void *data)
717 {
718 	int ret;
719 	long offlined = *(long *)data;
720 	ret = test_pages_isolated(start_pfn, start_pfn + nr_pages);
721 	offlined = nr_pages;
722 	if (!ret)
723 		*(long *)data += offlined;
724 	return ret;
725 }
726 
727 static long
728 check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
729 {
730 	long offlined = 0;
731 	int ret;
732 
733 	ret = walk_memory_resource(start_pfn, end_pfn - start_pfn, &offlined,
734 			check_pages_isolated_cb);
735 	if (ret < 0)
736 		offlined = (long)ret;
737 	return offlined;
738 }
739 
740 int offline_pages(unsigned long start_pfn,
741 		  unsigned long end_pfn, unsigned long timeout)
742 {
743 	unsigned long pfn, nr_pages, expire;
744 	long offlined_pages;
745 	int ret, drain, retry_max, node;
746 	struct zone *zone;
747 	struct memory_notify arg;
748 
749 	BUG_ON(start_pfn >= end_pfn);
750 	/* at least, alignment against pageblock is necessary */
751 	if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
752 		return -EINVAL;
753 	if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
754 		return -EINVAL;
755 	/* This makes hotplug much easier...and readable.
756 	   we assume this for now. .*/
757 	if (!test_pages_in_a_zone(start_pfn, end_pfn))
758 		return -EINVAL;
759 
760 	zone = page_zone(pfn_to_page(start_pfn));
761 	node = zone_to_nid(zone);
762 	nr_pages = end_pfn - start_pfn;
763 
764 	/* set above range as isolated */
765 	ret = start_isolate_page_range(start_pfn, end_pfn);
766 	if (ret)
767 		return ret;
768 
769 	arg.start_pfn = start_pfn;
770 	arg.nr_pages = nr_pages;
771 	arg.status_change_nid = -1;
772 	if (nr_pages >= node_present_pages(node))
773 		arg.status_change_nid = node;
774 
775 	ret = memory_notify(MEM_GOING_OFFLINE, &arg);
776 	ret = notifier_to_errno(ret);
777 	if (ret)
778 		goto failed_removal;
779 
780 	pfn = start_pfn;
781 	expire = jiffies + timeout;
782 	drain = 0;
783 	retry_max = 5;
784 repeat:
785 	/* start memory hot removal */
786 	ret = -EAGAIN;
787 	if (time_after(jiffies, expire))
788 		goto failed_removal;
789 	ret = -EINTR;
790 	if (signal_pending(current))
791 		goto failed_removal;
792 	ret = 0;
793 	if (drain) {
794 		lru_add_drain_all();
795 		flush_scheduled_work();
796 		cond_resched();
797 		drain_all_pages();
798 	}
799 
800 	pfn = scan_lru_pages(start_pfn, end_pfn);
801 	if (pfn) { /* We have page on LRU */
802 		ret = do_migrate_range(pfn, end_pfn);
803 		if (!ret) {
804 			drain = 1;
805 			goto repeat;
806 		} else {
807 			if (ret < 0)
808 				if (--retry_max == 0)
809 					goto failed_removal;
810 			yield();
811 			drain = 1;
812 			goto repeat;
813 		}
814 	}
815 	/* drain all zone's lru pagevec, this is asyncronous... */
816 	lru_add_drain_all();
817 	flush_scheduled_work();
818 	yield();
819 	/* drain pcp pages , this is synchrouns. */
820 	drain_all_pages();
821 	/* check again */
822 	offlined_pages = check_pages_isolated(start_pfn, end_pfn);
823 	if (offlined_pages < 0) {
824 		ret = -EBUSY;
825 		goto failed_removal;
826 	}
827 	printk(KERN_INFO "Offlined Pages %ld\n", offlined_pages);
828 	/* Ok, all of our target is islaoted.
829 	   We cannot do rollback at this point. */
830 	offline_isolated_pages(start_pfn, end_pfn);
831 	/* reset pagetype flags and makes migrate type to be MOVABLE */
832 	undo_isolate_page_range(start_pfn, end_pfn);
833 	/* removal success */
834 	zone->present_pages -= offlined_pages;
835 	zone->zone_pgdat->node_present_pages -= offlined_pages;
836 	totalram_pages -= offlined_pages;
837 	num_physpages -= offlined_pages;
838 
839 	vm_total_pages = nr_free_pagecache_pages();
840 	writeback_set_ratelimit();
841 
842 	memory_notify(MEM_OFFLINE, &arg);
843 	return 0;
844 
845 failed_removal:
846 	printk(KERN_INFO "memory offlining %lx to %lx failed\n",
847 		start_pfn, end_pfn);
848 	memory_notify(MEM_CANCEL_OFFLINE, &arg);
849 	/* pushback to free area */
850 	undo_isolate_page_range(start_pfn, end_pfn);
851 
852 	return ret;
853 }
854 
855 int remove_memory(u64 start, u64 size)
856 {
857 	unsigned long start_pfn, end_pfn;
858 
859 	start_pfn = PFN_DOWN(start);
860 	end_pfn = start_pfn + PFN_DOWN(size);
861 	return offline_pages(start_pfn, end_pfn, 120 * HZ);
862 }
863 #else
864 int remove_memory(u64 start, u64 size)
865 {
866 	return -EINVAL;
867 }
868 #endif /* CONFIG_MEMORY_HOTREMOVE */
869 EXPORT_SYMBOL_GPL(remove_memory);
870