xref: /openbmc/linux/mm/memory_hotplug.c (revision 4f205687)
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/memremap.h>
21 #include <linux/memory_hotplug.h>
22 #include <linux/highmem.h>
23 #include <linux/vmalloc.h>
24 #include <linux/ioport.h>
25 #include <linux/delay.h>
26 #include <linux/migrate.h>
27 #include <linux/page-isolation.h>
28 #include <linux/pfn.h>
29 #include <linux/suspend.h>
30 #include <linux/mm_inline.h>
31 #include <linux/firmware-map.h>
32 #include <linux/stop_machine.h>
33 #include <linux/hugetlb.h>
34 #include <linux/memblock.h>
35 #include <linux/bootmem.h>
36 #include <linux/compaction.h>
37 
38 #include <asm/tlbflush.h>
39 
40 #include "internal.h"
41 
42 /*
43  * online_page_callback contains pointer to current page onlining function.
44  * Initially it is generic_online_page(). If it is required it could be
45  * changed by calling set_online_page_callback() for callback registration
46  * and restore_online_page_callback() for generic callback restore.
47  */
48 
49 static void generic_online_page(struct page *page);
50 
51 static online_page_callback_t online_page_callback = generic_online_page;
52 static DEFINE_MUTEX(online_page_callback_lock);
53 
54 /* The same as the cpu_hotplug lock, but for memory hotplug. */
55 static struct {
56 	struct task_struct *active_writer;
57 	struct mutex lock; /* Synchronizes accesses to refcount, */
58 	/*
59 	 * Also blocks the new readers during
60 	 * an ongoing mem hotplug operation.
61 	 */
62 	int refcount;
63 
64 #ifdef CONFIG_DEBUG_LOCK_ALLOC
65 	struct lockdep_map dep_map;
66 #endif
67 } mem_hotplug = {
68 	.active_writer = NULL,
69 	.lock = __MUTEX_INITIALIZER(mem_hotplug.lock),
70 	.refcount = 0,
71 #ifdef CONFIG_DEBUG_LOCK_ALLOC
72 	.dep_map = {.name = "mem_hotplug.lock" },
73 #endif
74 };
75 
76 /* Lockdep annotations for get/put_online_mems() and mem_hotplug_begin/end() */
77 #define memhp_lock_acquire_read() lock_map_acquire_read(&mem_hotplug.dep_map)
78 #define memhp_lock_acquire()      lock_map_acquire(&mem_hotplug.dep_map)
79 #define memhp_lock_release()      lock_map_release(&mem_hotplug.dep_map)
80 
81 #ifndef CONFIG_MEMORY_HOTPLUG_DEFAULT_ONLINE
82 bool memhp_auto_online;
83 #else
84 bool memhp_auto_online = true;
85 #endif
86 EXPORT_SYMBOL_GPL(memhp_auto_online);
87 
88 static int __init setup_memhp_default_state(char *str)
89 {
90 	if (!strcmp(str, "online"))
91 		memhp_auto_online = true;
92 	else if (!strcmp(str, "offline"))
93 		memhp_auto_online = false;
94 
95 	return 1;
96 }
97 __setup("memhp_default_state=", setup_memhp_default_state);
98 
99 void get_online_mems(void)
100 {
101 	might_sleep();
102 	if (mem_hotplug.active_writer == current)
103 		return;
104 	memhp_lock_acquire_read();
105 	mutex_lock(&mem_hotplug.lock);
106 	mem_hotplug.refcount++;
107 	mutex_unlock(&mem_hotplug.lock);
108 
109 }
110 
111 void put_online_mems(void)
112 {
113 	if (mem_hotplug.active_writer == current)
114 		return;
115 	mutex_lock(&mem_hotplug.lock);
116 
117 	if (WARN_ON(!mem_hotplug.refcount))
118 		mem_hotplug.refcount++; /* try to fix things up */
119 
120 	if (!--mem_hotplug.refcount && unlikely(mem_hotplug.active_writer))
121 		wake_up_process(mem_hotplug.active_writer);
122 	mutex_unlock(&mem_hotplug.lock);
123 	memhp_lock_release();
124 
125 }
126 
127 void mem_hotplug_begin(void)
128 {
129 	mem_hotplug.active_writer = current;
130 
131 	memhp_lock_acquire();
132 	for (;;) {
133 		mutex_lock(&mem_hotplug.lock);
134 		if (likely(!mem_hotplug.refcount))
135 			break;
136 		__set_current_state(TASK_UNINTERRUPTIBLE);
137 		mutex_unlock(&mem_hotplug.lock);
138 		schedule();
139 	}
140 }
141 
142 void mem_hotplug_done(void)
143 {
144 	mem_hotplug.active_writer = NULL;
145 	mutex_unlock(&mem_hotplug.lock);
146 	memhp_lock_release();
147 }
148 
149 /* add this memory to iomem resource */
150 static struct resource *register_memory_resource(u64 start, u64 size)
151 {
152 	struct resource *res;
153 	res = kzalloc(sizeof(struct resource), GFP_KERNEL);
154 	if (!res)
155 		return ERR_PTR(-ENOMEM);
156 
157 	res->name = "System RAM";
158 	res->start = start;
159 	res->end = start + size - 1;
160 	res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
161 	if (request_resource(&iomem_resource, res) < 0) {
162 		pr_debug("System RAM resource %pR cannot be added\n", res);
163 		kfree(res);
164 		return ERR_PTR(-EEXIST);
165 	}
166 	return res;
167 }
168 
169 static void release_memory_resource(struct resource *res)
170 {
171 	if (!res)
172 		return;
173 	release_resource(res);
174 	kfree(res);
175 	return;
176 }
177 
178 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
179 void get_page_bootmem(unsigned long info,  struct page *page,
180 		      unsigned long type)
181 {
182 	page->lru.next = (struct list_head *) type;
183 	SetPagePrivate(page);
184 	set_page_private(page, info);
185 	page_ref_inc(page);
186 }
187 
188 void put_page_bootmem(struct page *page)
189 {
190 	unsigned long type;
191 
192 	type = (unsigned long) page->lru.next;
193 	BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE ||
194 	       type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE);
195 
196 	if (page_ref_dec_return(page) == 1) {
197 		ClearPagePrivate(page);
198 		set_page_private(page, 0);
199 		INIT_LIST_HEAD(&page->lru);
200 		free_reserved_page(page);
201 	}
202 }
203 
204 #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
205 #ifndef CONFIG_SPARSEMEM_VMEMMAP
206 static void register_page_bootmem_info_section(unsigned long start_pfn)
207 {
208 	unsigned long *usemap, mapsize, section_nr, i;
209 	struct mem_section *ms;
210 	struct page *page, *memmap;
211 
212 	section_nr = pfn_to_section_nr(start_pfn);
213 	ms = __nr_to_section(section_nr);
214 
215 	/* Get section's memmap address */
216 	memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
217 
218 	/*
219 	 * Get page for the memmap's phys address
220 	 * XXX: need more consideration for sparse_vmemmap...
221 	 */
222 	page = virt_to_page(memmap);
223 	mapsize = sizeof(struct page) * PAGES_PER_SECTION;
224 	mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
225 
226 	/* remember memmap's page */
227 	for (i = 0; i < mapsize; i++, page++)
228 		get_page_bootmem(section_nr, page, SECTION_INFO);
229 
230 	usemap = __nr_to_section(section_nr)->pageblock_flags;
231 	page = virt_to_page(usemap);
232 
233 	mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
234 
235 	for (i = 0; i < mapsize; i++, page++)
236 		get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
237 
238 }
239 #else /* CONFIG_SPARSEMEM_VMEMMAP */
240 static void register_page_bootmem_info_section(unsigned long start_pfn)
241 {
242 	unsigned long *usemap, mapsize, section_nr, i;
243 	struct mem_section *ms;
244 	struct page *page, *memmap;
245 
246 	if (!pfn_valid(start_pfn))
247 		return;
248 
249 	section_nr = pfn_to_section_nr(start_pfn);
250 	ms = __nr_to_section(section_nr);
251 
252 	memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
253 
254 	register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION);
255 
256 	usemap = __nr_to_section(section_nr)->pageblock_flags;
257 	page = virt_to_page(usemap);
258 
259 	mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
260 
261 	for (i = 0; i < mapsize; i++, page++)
262 		get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
263 }
264 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
265 
266 void __init register_page_bootmem_info_node(struct pglist_data *pgdat)
267 {
268 	unsigned long i, pfn, end_pfn, nr_pages;
269 	int node = pgdat->node_id;
270 	struct page *page;
271 	struct zone *zone;
272 
273 	nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
274 	page = virt_to_page(pgdat);
275 
276 	for (i = 0; i < nr_pages; i++, page++)
277 		get_page_bootmem(node, page, NODE_INFO);
278 
279 	zone = &pgdat->node_zones[0];
280 	for (; zone < pgdat->node_zones + MAX_NR_ZONES - 1; zone++) {
281 		if (zone_is_initialized(zone)) {
282 			nr_pages = zone->wait_table_hash_nr_entries
283 				* sizeof(wait_queue_head_t);
284 			nr_pages = PAGE_ALIGN(nr_pages) >> PAGE_SHIFT;
285 			page = virt_to_page(zone->wait_table);
286 
287 			for (i = 0; i < nr_pages; i++, page++)
288 				get_page_bootmem(node, page, NODE_INFO);
289 		}
290 	}
291 
292 	pfn = pgdat->node_start_pfn;
293 	end_pfn = pgdat_end_pfn(pgdat);
294 
295 	/* register section info */
296 	for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
297 		/*
298 		 * Some platforms can assign the same pfn to multiple nodes - on
299 		 * node0 as well as nodeN.  To avoid registering a pfn against
300 		 * multiple nodes we check that this pfn does not already
301 		 * reside in some other nodes.
302 		 */
303 		if (pfn_valid(pfn) && (early_pfn_to_nid(pfn) == node))
304 			register_page_bootmem_info_section(pfn);
305 	}
306 }
307 #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
308 
309 static void __meminit grow_zone_span(struct zone *zone, unsigned long start_pfn,
310 				     unsigned long end_pfn)
311 {
312 	unsigned long old_zone_end_pfn;
313 
314 	zone_span_writelock(zone);
315 
316 	old_zone_end_pfn = zone_end_pfn(zone);
317 	if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
318 		zone->zone_start_pfn = start_pfn;
319 
320 	zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
321 				zone->zone_start_pfn;
322 
323 	zone_span_writeunlock(zone);
324 }
325 
326 static void resize_zone(struct zone *zone, unsigned long start_pfn,
327 		unsigned long end_pfn)
328 {
329 	zone_span_writelock(zone);
330 
331 	if (end_pfn - start_pfn) {
332 		zone->zone_start_pfn = start_pfn;
333 		zone->spanned_pages = end_pfn - start_pfn;
334 	} else {
335 		/*
336 		 * make it consist as free_area_init_core(),
337 		 * if spanned_pages = 0, then keep start_pfn = 0
338 		 */
339 		zone->zone_start_pfn = 0;
340 		zone->spanned_pages = 0;
341 	}
342 
343 	zone_span_writeunlock(zone);
344 }
345 
346 static void fix_zone_id(struct zone *zone, unsigned long start_pfn,
347 		unsigned long end_pfn)
348 {
349 	enum zone_type zid = zone_idx(zone);
350 	int nid = zone->zone_pgdat->node_id;
351 	unsigned long pfn;
352 
353 	for (pfn = start_pfn; pfn < end_pfn; pfn++)
354 		set_page_links(pfn_to_page(pfn), zid, nid, pfn);
355 }
356 
357 /* Can fail with -ENOMEM from allocating a wait table with vmalloc() or
358  * alloc_bootmem_node_nopanic()/memblock_virt_alloc_node_nopanic() */
359 static int __ref ensure_zone_is_initialized(struct zone *zone,
360 			unsigned long start_pfn, unsigned long num_pages)
361 {
362 	if (!zone_is_initialized(zone))
363 		return init_currently_empty_zone(zone, start_pfn, num_pages);
364 
365 	return 0;
366 }
367 
368 static int __meminit move_pfn_range_left(struct zone *z1, struct zone *z2,
369 		unsigned long start_pfn, unsigned long end_pfn)
370 {
371 	int ret;
372 	unsigned long flags;
373 	unsigned long z1_start_pfn;
374 
375 	ret = ensure_zone_is_initialized(z1, start_pfn, end_pfn - start_pfn);
376 	if (ret)
377 		return ret;
378 
379 	pgdat_resize_lock(z1->zone_pgdat, &flags);
380 
381 	/* can't move pfns which are higher than @z2 */
382 	if (end_pfn > zone_end_pfn(z2))
383 		goto out_fail;
384 	/* the move out part must be at the left most of @z2 */
385 	if (start_pfn > z2->zone_start_pfn)
386 		goto out_fail;
387 	/* must included/overlap */
388 	if (end_pfn <= z2->zone_start_pfn)
389 		goto out_fail;
390 
391 	/* use start_pfn for z1's start_pfn if z1 is empty */
392 	if (!zone_is_empty(z1))
393 		z1_start_pfn = z1->zone_start_pfn;
394 	else
395 		z1_start_pfn = start_pfn;
396 
397 	resize_zone(z1, z1_start_pfn, end_pfn);
398 	resize_zone(z2, end_pfn, zone_end_pfn(z2));
399 
400 	pgdat_resize_unlock(z1->zone_pgdat, &flags);
401 
402 	fix_zone_id(z1, start_pfn, end_pfn);
403 
404 	return 0;
405 out_fail:
406 	pgdat_resize_unlock(z1->zone_pgdat, &flags);
407 	return -1;
408 }
409 
410 static int __meminit move_pfn_range_right(struct zone *z1, struct zone *z2,
411 		unsigned long start_pfn, unsigned long end_pfn)
412 {
413 	int ret;
414 	unsigned long flags;
415 	unsigned long z2_end_pfn;
416 
417 	ret = ensure_zone_is_initialized(z2, start_pfn, end_pfn - start_pfn);
418 	if (ret)
419 		return ret;
420 
421 	pgdat_resize_lock(z1->zone_pgdat, &flags);
422 
423 	/* can't move pfns which are lower than @z1 */
424 	if (z1->zone_start_pfn > start_pfn)
425 		goto out_fail;
426 	/* the move out part mast at the right most of @z1 */
427 	if (zone_end_pfn(z1) >  end_pfn)
428 		goto out_fail;
429 	/* must included/overlap */
430 	if (start_pfn >= zone_end_pfn(z1))
431 		goto out_fail;
432 
433 	/* use end_pfn for z2's end_pfn if z2 is empty */
434 	if (!zone_is_empty(z2))
435 		z2_end_pfn = zone_end_pfn(z2);
436 	else
437 		z2_end_pfn = end_pfn;
438 
439 	resize_zone(z1, z1->zone_start_pfn, start_pfn);
440 	resize_zone(z2, start_pfn, z2_end_pfn);
441 
442 	pgdat_resize_unlock(z1->zone_pgdat, &flags);
443 
444 	fix_zone_id(z2, start_pfn, end_pfn);
445 
446 	return 0;
447 out_fail:
448 	pgdat_resize_unlock(z1->zone_pgdat, &flags);
449 	return -1;
450 }
451 
452 static void __meminit grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn,
453 				      unsigned long end_pfn)
454 {
455 	unsigned long old_pgdat_end_pfn = pgdat_end_pfn(pgdat);
456 
457 	if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
458 		pgdat->node_start_pfn = start_pfn;
459 
460 	pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
461 					pgdat->node_start_pfn;
462 }
463 
464 static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn)
465 {
466 	struct pglist_data *pgdat = zone->zone_pgdat;
467 	int nr_pages = PAGES_PER_SECTION;
468 	int nid = pgdat->node_id;
469 	int zone_type;
470 	unsigned long flags, pfn;
471 	int ret;
472 
473 	zone_type = zone - pgdat->node_zones;
474 	ret = ensure_zone_is_initialized(zone, phys_start_pfn, nr_pages);
475 	if (ret)
476 		return ret;
477 
478 	pgdat_resize_lock(zone->zone_pgdat, &flags);
479 	grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages);
480 	grow_pgdat_span(zone->zone_pgdat, phys_start_pfn,
481 			phys_start_pfn + nr_pages);
482 	pgdat_resize_unlock(zone->zone_pgdat, &flags);
483 	memmap_init_zone(nr_pages, nid, zone_type,
484 			 phys_start_pfn, MEMMAP_HOTPLUG);
485 
486 	/* online_page_range is called later and expects pages reserved */
487 	for (pfn = phys_start_pfn; pfn < phys_start_pfn + nr_pages; pfn++) {
488 		if (!pfn_valid(pfn))
489 			continue;
490 
491 		SetPageReserved(pfn_to_page(pfn));
492 	}
493 	return 0;
494 }
495 
496 static int __meminit __add_section(int nid, struct zone *zone,
497 					unsigned long phys_start_pfn)
498 {
499 	int ret;
500 
501 	if (pfn_valid(phys_start_pfn))
502 		return -EEXIST;
503 
504 	ret = sparse_add_one_section(zone, phys_start_pfn);
505 
506 	if (ret < 0)
507 		return ret;
508 
509 	ret = __add_zone(zone, phys_start_pfn);
510 
511 	if (ret < 0)
512 		return ret;
513 
514 	return register_new_memory(nid, __pfn_to_section(phys_start_pfn));
515 }
516 
517 /*
518  * Reasonably generic function for adding memory.  It is
519  * expected that archs that support memory hotplug will
520  * call this function after deciding the zone to which to
521  * add the new pages.
522  */
523 int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn,
524 			unsigned long nr_pages)
525 {
526 	unsigned long i;
527 	int err = 0;
528 	int start_sec, end_sec;
529 	struct vmem_altmap *altmap;
530 
531 	clear_zone_contiguous(zone);
532 
533 	/* during initialize mem_map, align hot-added range to section */
534 	start_sec = pfn_to_section_nr(phys_start_pfn);
535 	end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
536 
537 	altmap = to_vmem_altmap((unsigned long) pfn_to_page(phys_start_pfn));
538 	if (altmap) {
539 		/*
540 		 * Validate altmap is within bounds of the total request
541 		 */
542 		if (altmap->base_pfn != phys_start_pfn
543 				|| vmem_altmap_offset(altmap) > nr_pages) {
544 			pr_warn_once("memory add fail, invalid altmap\n");
545 			err = -EINVAL;
546 			goto out;
547 		}
548 		altmap->alloc = 0;
549 	}
550 
551 	for (i = start_sec; i <= end_sec; i++) {
552 		err = __add_section(nid, zone, section_nr_to_pfn(i));
553 
554 		/*
555 		 * EEXIST is finally dealt with by ioresource collision
556 		 * check. see add_memory() => register_memory_resource()
557 		 * Warning will be printed if there is collision.
558 		 */
559 		if (err && (err != -EEXIST))
560 			break;
561 		err = 0;
562 	}
563 	vmemmap_populate_print_last();
564 out:
565 	set_zone_contiguous(zone);
566 	return err;
567 }
568 EXPORT_SYMBOL_GPL(__add_pages);
569 
570 #ifdef CONFIG_MEMORY_HOTREMOVE
571 /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
572 static int find_smallest_section_pfn(int nid, struct zone *zone,
573 				     unsigned long start_pfn,
574 				     unsigned long end_pfn)
575 {
576 	struct mem_section *ms;
577 
578 	for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) {
579 		ms = __pfn_to_section(start_pfn);
580 
581 		if (unlikely(!valid_section(ms)))
582 			continue;
583 
584 		if (unlikely(pfn_to_nid(start_pfn) != nid))
585 			continue;
586 
587 		if (zone && zone != page_zone(pfn_to_page(start_pfn)))
588 			continue;
589 
590 		return start_pfn;
591 	}
592 
593 	return 0;
594 }
595 
596 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
597 static int find_biggest_section_pfn(int nid, struct zone *zone,
598 				    unsigned long start_pfn,
599 				    unsigned long end_pfn)
600 {
601 	struct mem_section *ms;
602 	unsigned long pfn;
603 
604 	/* pfn is the end pfn of a memory section. */
605 	pfn = end_pfn - 1;
606 	for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) {
607 		ms = __pfn_to_section(pfn);
608 
609 		if (unlikely(!valid_section(ms)))
610 			continue;
611 
612 		if (unlikely(pfn_to_nid(pfn) != nid))
613 			continue;
614 
615 		if (zone && zone != page_zone(pfn_to_page(pfn)))
616 			continue;
617 
618 		return pfn;
619 	}
620 
621 	return 0;
622 }
623 
624 static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
625 			     unsigned long end_pfn)
626 {
627 	unsigned long zone_start_pfn = zone->zone_start_pfn;
628 	unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */
629 	unsigned long zone_end_pfn = z;
630 	unsigned long pfn;
631 	struct mem_section *ms;
632 	int nid = zone_to_nid(zone);
633 
634 	zone_span_writelock(zone);
635 	if (zone_start_pfn == start_pfn) {
636 		/*
637 		 * If the section is smallest section in the zone, it need
638 		 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
639 		 * In this case, we find second smallest valid mem_section
640 		 * for shrinking zone.
641 		 */
642 		pfn = find_smallest_section_pfn(nid, zone, end_pfn,
643 						zone_end_pfn);
644 		if (pfn) {
645 			zone->zone_start_pfn = pfn;
646 			zone->spanned_pages = zone_end_pfn - pfn;
647 		}
648 	} else if (zone_end_pfn == end_pfn) {
649 		/*
650 		 * If the section is biggest section in the zone, it need
651 		 * shrink zone->spanned_pages.
652 		 * In this case, we find second biggest valid mem_section for
653 		 * shrinking zone.
654 		 */
655 		pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn,
656 					       start_pfn);
657 		if (pfn)
658 			zone->spanned_pages = pfn - zone_start_pfn + 1;
659 	}
660 
661 	/*
662 	 * The section is not biggest or smallest mem_section in the zone, it
663 	 * only creates a hole in the zone. So in this case, we need not
664 	 * change the zone. But perhaps, the zone has only hole data. Thus
665 	 * it check the zone has only hole or not.
666 	 */
667 	pfn = zone_start_pfn;
668 	for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) {
669 		ms = __pfn_to_section(pfn);
670 
671 		if (unlikely(!valid_section(ms)))
672 			continue;
673 
674 		if (page_zone(pfn_to_page(pfn)) != zone)
675 			continue;
676 
677 		 /* If the section is current section, it continues the loop */
678 		if (start_pfn == pfn)
679 			continue;
680 
681 		/* If we find valid section, we have nothing to do */
682 		zone_span_writeunlock(zone);
683 		return;
684 	}
685 
686 	/* The zone has no valid section */
687 	zone->zone_start_pfn = 0;
688 	zone->spanned_pages = 0;
689 	zone_span_writeunlock(zone);
690 }
691 
692 static void shrink_pgdat_span(struct pglist_data *pgdat,
693 			      unsigned long start_pfn, unsigned long end_pfn)
694 {
695 	unsigned long pgdat_start_pfn = pgdat->node_start_pfn;
696 	unsigned long p = pgdat_end_pfn(pgdat); /* pgdat_end_pfn namespace clash */
697 	unsigned long pgdat_end_pfn = p;
698 	unsigned long pfn;
699 	struct mem_section *ms;
700 	int nid = pgdat->node_id;
701 
702 	if (pgdat_start_pfn == start_pfn) {
703 		/*
704 		 * If the section is smallest section in the pgdat, it need
705 		 * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages.
706 		 * In this case, we find second smallest valid mem_section
707 		 * for shrinking zone.
708 		 */
709 		pfn = find_smallest_section_pfn(nid, NULL, end_pfn,
710 						pgdat_end_pfn);
711 		if (pfn) {
712 			pgdat->node_start_pfn = pfn;
713 			pgdat->node_spanned_pages = pgdat_end_pfn - pfn;
714 		}
715 	} else if (pgdat_end_pfn == end_pfn) {
716 		/*
717 		 * If the section is biggest section in the pgdat, it need
718 		 * shrink pgdat->node_spanned_pages.
719 		 * In this case, we find second biggest valid mem_section for
720 		 * shrinking zone.
721 		 */
722 		pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn,
723 					       start_pfn);
724 		if (pfn)
725 			pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1;
726 	}
727 
728 	/*
729 	 * If the section is not biggest or smallest mem_section in the pgdat,
730 	 * it only creates a hole in the pgdat. So in this case, we need not
731 	 * change the pgdat.
732 	 * But perhaps, the pgdat has only hole data. Thus it check the pgdat
733 	 * has only hole or not.
734 	 */
735 	pfn = pgdat_start_pfn;
736 	for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) {
737 		ms = __pfn_to_section(pfn);
738 
739 		if (unlikely(!valid_section(ms)))
740 			continue;
741 
742 		if (pfn_to_nid(pfn) != nid)
743 			continue;
744 
745 		 /* If the section is current section, it continues the loop */
746 		if (start_pfn == pfn)
747 			continue;
748 
749 		/* If we find valid section, we have nothing to do */
750 		return;
751 	}
752 
753 	/* The pgdat has no valid section */
754 	pgdat->node_start_pfn = 0;
755 	pgdat->node_spanned_pages = 0;
756 }
757 
758 static void __remove_zone(struct zone *zone, unsigned long start_pfn)
759 {
760 	struct pglist_data *pgdat = zone->zone_pgdat;
761 	int nr_pages = PAGES_PER_SECTION;
762 	int zone_type;
763 	unsigned long flags;
764 
765 	zone_type = zone - pgdat->node_zones;
766 
767 	pgdat_resize_lock(zone->zone_pgdat, &flags);
768 	shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
769 	shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages);
770 	pgdat_resize_unlock(zone->zone_pgdat, &flags);
771 }
772 
773 static int __remove_section(struct zone *zone, struct mem_section *ms,
774 		unsigned long map_offset)
775 {
776 	unsigned long start_pfn;
777 	int scn_nr;
778 	int ret = -EINVAL;
779 
780 	if (!valid_section(ms))
781 		return ret;
782 
783 	ret = unregister_memory_section(ms);
784 	if (ret)
785 		return ret;
786 
787 	scn_nr = __section_nr(ms);
788 	start_pfn = section_nr_to_pfn(scn_nr);
789 	__remove_zone(zone, start_pfn);
790 
791 	sparse_remove_one_section(zone, ms, map_offset);
792 	return 0;
793 }
794 
795 /**
796  * __remove_pages() - remove sections of pages from a zone
797  * @zone: zone from which pages need to be removed
798  * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
799  * @nr_pages: number of pages to remove (must be multiple of section size)
800  *
801  * Generic helper function to remove section mappings and sysfs entries
802  * for the section of the memory we are removing. Caller needs to make
803  * sure that pages are marked reserved and zones are adjust properly by
804  * calling offline_pages().
805  */
806 int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
807 		 unsigned long nr_pages)
808 {
809 	unsigned long i;
810 	unsigned long map_offset = 0;
811 	int sections_to_remove, ret = 0;
812 
813 	/* In the ZONE_DEVICE case device driver owns the memory region */
814 	if (is_dev_zone(zone)) {
815 		struct page *page = pfn_to_page(phys_start_pfn);
816 		struct vmem_altmap *altmap;
817 
818 		altmap = to_vmem_altmap((unsigned long) page);
819 		if (altmap)
820 			map_offset = vmem_altmap_offset(altmap);
821 	} else {
822 		resource_size_t start, size;
823 
824 		start = phys_start_pfn << PAGE_SHIFT;
825 		size = nr_pages * PAGE_SIZE;
826 
827 		ret = release_mem_region_adjustable(&iomem_resource, start,
828 					size);
829 		if (ret) {
830 			resource_size_t endres = start + size - 1;
831 
832 			pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
833 					&start, &endres, ret);
834 		}
835 	}
836 
837 	clear_zone_contiguous(zone);
838 
839 	/*
840 	 * We can only remove entire sections
841 	 */
842 	BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
843 	BUG_ON(nr_pages % PAGES_PER_SECTION);
844 
845 	sections_to_remove = nr_pages / PAGES_PER_SECTION;
846 	for (i = 0; i < sections_to_remove; i++) {
847 		unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
848 
849 		ret = __remove_section(zone, __pfn_to_section(pfn), map_offset);
850 		map_offset = 0;
851 		if (ret)
852 			break;
853 	}
854 
855 	set_zone_contiguous(zone);
856 
857 	return ret;
858 }
859 EXPORT_SYMBOL_GPL(__remove_pages);
860 #endif /* CONFIG_MEMORY_HOTREMOVE */
861 
862 int set_online_page_callback(online_page_callback_t callback)
863 {
864 	int rc = -EINVAL;
865 
866 	get_online_mems();
867 	mutex_lock(&online_page_callback_lock);
868 
869 	if (online_page_callback == generic_online_page) {
870 		online_page_callback = callback;
871 		rc = 0;
872 	}
873 
874 	mutex_unlock(&online_page_callback_lock);
875 	put_online_mems();
876 
877 	return rc;
878 }
879 EXPORT_SYMBOL_GPL(set_online_page_callback);
880 
881 int restore_online_page_callback(online_page_callback_t callback)
882 {
883 	int rc = -EINVAL;
884 
885 	get_online_mems();
886 	mutex_lock(&online_page_callback_lock);
887 
888 	if (online_page_callback == callback) {
889 		online_page_callback = generic_online_page;
890 		rc = 0;
891 	}
892 
893 	mutex_unlock(&online_page_callback_lock);
894 	put_online_mems();
895 
896 	return rc;
897 }
898 EXPORT_SYMBOL_GPL(restore_online_page_callback);
899 
900 void __online_page_set_limits(struct page *page)
901 {
902 }
903 EXPORT_SYMBOL_GPL(__online_page_set_limits);
904 
905 void __online_page_increment_counters(struct page *page)
906 {
907 	adjust_managed_page_count(page, 1);
908 }
909 EXPORT_SYMBOL_GPL(__online_page_increment_counters);
910 
911 void __online_page_free(struct page *page)
912 {
913 	__free_reserved_page(page);
914 }
915 EXPORT_SYMBOL_GPL(__online_page_free);
916 
917 static void generic_online_page(struct page *page)
918 {
919 	__online_page_set_limits(page);
920 	__online_page_increment_counters(page);
921 	__online_page_free(page);
922 }
923 
924 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
925 			void *arg)
926 {
927 	unsigned long i;
928 	unsigned long onlined_pages = *(unsigned long *)arg;
929 	struct page *page;
930 	if (PageReserved(pfn_to_page(start_pfn)))
931 		for (i = 0; i < nr_pages; i++) {
932 			page = pfn_to_page(start_pfn + i);
933 			(*online_page_callback)(page);
934 			onlined_pages++;
935 		}
936 	*(unsigned long *)arg = onlined_pages;
937 	return 0;
938 }
939 
940 #ifdef CONFIG_MOVABLE_NODE
941 /*
942  * When CONFIG_MOVABLE_NODE, we permit onlining of a node which doesn't have
943  * normal memory.
944  */
945 static bool can_online_high_movable(struct zone *zone)
946 {
947 	return true;
948 }
949 #else /* CONFIG_MOVABLE_NODE */
950 /* ensure every online node has NORMAL memory */
951 static bool can_online_high_movable(struct zone *zone)
952 {
953 	return node_state(zone_to_nid(zone), N_NORMAL_MEMORY);
954 }
955 #endif /* CONFIG_MOVABLE_NODE */
956 
957 /* check which state of node_states will be changed when online memory */
958 static void node_states_check_changes_online(unsigned long nr_pages,
959 	struct zone *zone, struct memory_notify *arg)
960 {
961 	int nid = zone_to_nid(zone);
962 	enum zone_type zone_last = ZONE_NORMAL;
963 
964 	/*
965 	 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
966 	 * contains nodes which have zones of 0...ZONE_NORMAL,
967 	 * set zone_last to ZONE_NORMAL.
968 	 *
969 	 * If we don't have HIGHMEM nor movable node,
970 	 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
971 	 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
972 	 */
973 	if (N_MEMORY == N_NORMAL_MEMORY)
974 		zone_last = ZONE_MOVABLE;
975 
976 	/*
977 	 * if the memory to be online is in a zone of 0...zone_last, and
978 	 * the zones of 0...zone_last don't have memory before online, we will
979 	 * need to set the node to node_states[N_NORMAL_MEMORY] after
980 	 * the memory is online.
981 	 */
982 	if (zone_idx(zone) <= zone_last && !node_state(nid, N_NORMAL_MEMORY))
983 		arg->status_change_nid_normal = nid;
984 	else
985 		arg->status_change_nid_normal = -1;
986 
987 #ifdef CONFIG_HIGHMEM
988 	/*
989 	 * If we have movable node, node_states[N_HIGH_MEMORY]
990 	 * contains nodes which have zones of 0...ZONE_HIGHMEM,
991 	 * set zone_last to ZONE_HIGHMEM.
992 	 *
993 	 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
994 	 * contains nodes which have zones of 0...ZONE_MOVABLE,
995 	 * set zone_last to ZONE_MOVABLE.
996 	 */
997 	zone_last = ZONE_HIGHMEM;
998 	if (N_MEMORY == N_HIGH_MEMORY)
999 		zone_last = ZONE_MOVABLE;
1000 
1001 	if (zone_idx(zone) <= zone_last && !node_state(nid, N_HIGH_MEMORY))
1002 		arg->status_change_nid_high = nid;
1003 	else
1004 		arg->status_change_nid_high = -1;
1005 #else
1006 	arg->status_change_nid_high = arg->status_change_nid_normal;
1007 #endif
1008 
1009 	/*
1010 	 * if the node don't have memory befor online, we will need to
1011 	 * set the node to node_states[N_MEMORY] after the memory
1012 	 * is online.
1013 	 */
1014 	if (!node_state(nid, N_MEMORY))
1015 		arg->status_change_nid = nid;
1016 	else
1017 		arg->status_change_nid = -1;
1018 }
1019 
1020 static void node_states_set_node(int node, struct memory_notify *arg)
1021 {
1022 	if (arg->status_change_nid_normal >= 0)
1023 		node_set_state(node, N_NORMAL_MEMORY);
1024 
1025 	if (arg->status_change_nid_high >= 0)
1026 		node_set_state(node, N_HIGH_MEMORY);
1027 
1028 	node_set_state(node, N_MEMORY);
1029 }
1030 
1031 
1032 /* Must be protected by mem_hotplug_begin() */
1033 int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type)
1034 {
1035 	unsigned long flags;
1036 	unsigned long onlined_pages = 0;
1037 	struct zone *zone;
1038 	int need_zonelists_rebuild = 0;
1039 	int nid;
1040 	int ret;
1041 	struct memory_notify arg;
1042 
1043 	/*
1044 	 * This doesn't need a lock to do pfn_to_page().
1045 	 * The section can't be removed here because of the
1046 	 * memory_block->state_mutex.
1047 	 */
1048 	zone = page_zone(pfn_to_page(pfn));
1049 
1050 	if ((zone_idx(zone) > ZONE_NORMAL ||
1051 	    online_type == MMOP_ONLINE_MOVABLE) &&
1052 	    !can_online_high_movable(zone))
1053 		return -EINVAL;
1054 
1055 	if (online_type == MMOP_ONLINE_KERNEL &&
1056 	    zone_idx(zone) == ZONE_MOVABLE) {
1057 		if (move_pfn_range_left(zone - 1, zone, pfn, pfn + nr_pages))
1058 			return -EINVAL;
1059 	}
1060 	if (online_type == MMOP_ONLINE_MOVABLE &&
1061 	    zone_idx(zone) == ZONE_MOVABLE - 1) {
1062 		if (move_pfn_range_right(zone, zone + 1, pfn, pfn + nr_pages))
1063 			return -EINVAL;
1064 	}
1065 
1066 	/* Previous code may changed the zone of the pfn range */
1067 	zone = page_zone(pfn_to_page(pfn));
1068 
1069 	arg.start_pfn = pfn;
1070 	arg.nr_pages = nr_pages;
1071 	node_states_check_changes_online(nr_pages, zone, &arg);
1072 
1073 	nid = zone_to_nid(zone);
1074 
1075 	ret = memory_notify(MEM_GOING_ONLINE, &arg);
1076 	ret = notifier_to_errno(ret);
1077 	if (ret)
1078 		goto failed_addition;
1079 
1080 	/*
1081 	 * If this zone is not populated, then it is not in zonelist.
1082 	 * This means the page allocator ignores this zone.
1083 	 * So, zonelist must be updated after online.
1084 	 */
1085 	mutex_lock(&zonelists_mutex);
1086 	if (!populated_zone(zone)) {
1087 		need_zonelists_rebuild = 1;
1088 		build_all_zonelists(NULL, zone);
1089 	}
1090 
1091 	ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
1092 		online_pages_range);
1093 	if (ret) {
1094 		if (need_zonelists_rebuild)
1095 			zone_pcp_reset(zone);
1096 		mutex_unlock(&zonelists_mutex);
1097 		goto failed_addition;
1098 	}
1099 
1100 	zone->present_pages += onlined_pages;
1101 
1102 	pgdat_resize_lock(zone->zone_pgdat, &flags);
1103 	zone->zone_pgdat->node_present_pages += onlined_pages;
1104 	pgdat_resize_unlock(zone->zone_pgdat, &flags);
1105 
1106 	if (onlined_pages) {
1107 		node_states_set_node(nid, &arg);
1108 		if (need_zonelists_rebuild)
1109 			build_all_zonelists(NULL, NULL);
1110 		else
1111 			zone_pcp_update(zone);
1112 	}
1113 
1114 	mutex_unlock(&zonelists_mutex);
1115 
1116 	init_per_zone_wmark_min();
1117 
1118 	if (onlined_pages) {
1119 		kswapd_run(nid);
1120 		kcompactd_run(nid);
1121 	}
1122 
1123 	vm_total_pages = nr_free_pagecache_pages();
1124 
1125 	writeback_set_ratelimit();
1126 
1127 	if (onlined_pages)
1128 		memory_notify(MEM_ONLINE, &arg);
1129 	return 0;
1130 
1131 failed_addition:
1132 	pr_debug("online_pages [mem %#010llx-%#010llx] failed\n",
1133 		 (unsigned long long) pfn << PAGE_SHIFT,
1134 		 (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1);
1135 	memory_notify(MEM_CANCEL_ONLINE, &arg);
1136 	return ret;
1137 }
1138 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
1139 
1140 static void reset_node_present_pages(pg_data_t *pgdat)
1141 {
1142 	struct zone *z;
1143 
1144 	for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
1145 		z->present_pages = 0;
1146 
1147 	pgdat->node_present_pages = 0;
1148 }
1149 
1150 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1151 static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
1152 {
1153 	struct pglist_data *pgdat;
1154 	unsigned long zones_size[MAX_NR_ZONES] = {0};
1155 	unsigned long zholes_size[MAX_NR_ZONES] = {0};
1156 	unsigned long start_pfn = PFN_DOWN(start);
1157 
1158 	pgdat = NODE_DATA(nid);
1159 	if (!pgdat) {
1160 		pgdat = arch_alloc_nodedata(nid);
1161 		if (!pgdat)
1162 			return NULL;
1163 
1164 		arch_refresh_nodedata(nid, pgdat);
1165 	} else {
1166 		/* Reset the nr_zones and classzone_idx to 0 before reuse */
1167 		pgdat->nr_zones = 0;
1168 		pgdat->classzone_idx = 0;
1169 	}
1170 
1171 	/* we can use NODE_DATA(nid) from here */
1172 
1173 	/* init node's zones as empty zones, we don't have any present pages.*/
1174 	free_area_init_node(nid, zones_size, start_pfn, zholes_size);
1175 
1176 	/*
1177 	 * The node we allocated has no zone fallback lists. For avoiding
1178 	 * to access not-initialized zonelist, build here.
1179 	 */
1180 	mutex_lock(&zonelists_mutex);
1181 	build_all_zonelists(pgdat, NULL);
1182 	mutex_unlock(&zonelists_mutex);
1183 
1184 	/*
1185 	 * zone->managed_pages is set to an approximate value in
1186 	 * free_area_init_core(), which will cause
1187 	 * /sys/device/system/node/nodeX/meminfo has wrong data.
1188 	 * So reset it to 0 before any memory is onlined.
1189 	 */
1190 	reset_node_managed_pages(pgdat);
1191 
1192 	/*
1193 	 * When memory is hot-added, all the memory is in offline state. So
1194 	 * clear all zones' present_pages because they will be updated in
1195 	 * online_pages() and offline_pages().
1196 	 */
1197 	reset_node_present_pages(pgdat);
1198 
1199 	return pgdat;
1200 }
1201 
1202 static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
1203 {
1204 	arch_refresh_nodedata(nid, NULL);
1205 	arch_free_nodedata(pgdat);
1206 	return;
1207 }
1208 
1209 
1210 /**
1211  * try_online_node - online a node if offlined
1212  *
1213  * called by cpu_up() to online a node without onlined memory.
1214  */
1215 int try_online_node(int nid)
1216 {
1217 	pg_data_t	*pgdat;
1218 	int	ret;
1219 
1220 	if (node_online(nid))
1221 		return 0;
1222 
1223 	mem_hotplug_begin();
1224 	pgdat = hotadd_new_pgdat(nid, 0);
1225 	if (!pgdat) {
1226 		pr_err("Cannot online node %d due to NULL pgdat\n", nid);
1227 		ret = -ENOMEM;
1228 		goto out;
1229 	}
1230 	node_set_online(nid);
1231 	ret = register_one_node(nid);
1232 	BUG_ON(ret);
1233 
1234 	if (pgdat->node_zonelists->_zonerefs->zone == NULL) {
1235 		mutex_lock(&zonelists_mutex);
1236 		build_all_zonelists(NULL, NULL);
1237 		mutex_unlock(&zonelists_mutex);
1238 	}
1239 
1240 out:
1241 	mem_hotplug_done();
1242 	return ret;
1243 }
1244 
1245 static int check_hotplug_memory_range(u64 start, u64 size)
1246 {
1247 	u64 start_pfn = PFN_DOWN(start);
1248 	u64 nr_pages = size >> PAGE_SHIFT;
1249 
1250 	/* Memory range must be aligned with section */
1251 	if ((start_pfn & ~PAGE_SECTION_MASK) ||
1252 	    (nr_pages % PAGES_PER_SECTION) || (!nr_pages)) {
1253 		pr_err("Section-unaligned hotplug range: start 0x%llx, size 0x%llx\n",
1254 				(unsigned long long)start,
1255 				(unsigned long long)size);
1256 		return -EINVAL;
1257 	}
1258 
1259 	return 0;
1260 }
1261 
1262 /*
1263  * If movable zone has already been setup, newly added memory should be check.
1264  * If its address is higher than movable zone, it should be added as movable.
1265  * Without this check, movable zone may overlap with other zone.
1266  */
1267 static int should_add_memory_movable(int nid, u64 start, u64 size)
1268 {
1269 	unsigned long start_pfn = start >> PAGE_SHIFT;
1270 	pg_data_t *pgdat = NODE_DATA(nid);
1271 	struct zone *movable_zone = pgdat->node_zones + ZONE_MOVABLE;
1272 
1273 	if (zone_is_empty(movable_zone))
1274 		return 0;
1275 
1276 	if (movable_zone->zone_start_pfn <= start_pfn)
1277 		return 1;
1278 
1279 	return 0;
1280 }
1281 
1282 int zone_for_memory(int nid, u64 start, u64 size, int zone_default,
1283 		bool for_device)
1284 {
1285 #ifdef CONFIG_ZONE_DEVICE
1286 	if (for_device)
1287 		return ZONE_DEVICE;
1288 #endif
1289 	if (should_add_memory_movable(nid, start, size))
1290 		return ZONE_MOVABLE;
1291 
1292 	return zone_default;
1293 }
1294 
1295 static int online_memory_block(struct memory_block *mem, void *arg)
1296 {
1297 	return memory_block_change_state(mem, MEM_ONLINE, MEM_OFFLINE);
1298 }
1299 
1300 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1301 int __ref add_memory_resource(int nid, struct resource *res, bool online)
1302 {
1303 	u64 start, size;
1304 	pg_data_t *pgdat = NULL;
1305 	bool new_pgdat;
1306 	bool new_node;
1307 	int ret;
1308 
1309 	start = res->start;
1310 	size = resource_size(res);
1311 
1312 	ret = check_hotplug_memory_range(start, size);
1313 	if (ret)
1314 		return ret;
1315 
1316 	{	/* Stupid hack to suppress address-never-null warning */
1317 		void *p = NODE_DATA(nid);
1318 		new_pgdat = !p;
1319 	}
1320 
1321 	mem_hotplug_begin();
1322 
1323 	/*
1324 	 * Add new range to memblock so that when hotadd_new_pgdat() is called
1325 	 * to allocate new pgdat, get_pfn_range_for_nid() will be able to find
1326 	 * this new range and calculate total pages correctly.  The range will
1327 	 * be removed at hot-remove time.
1328 	 */
1329 	memblock_add_node(start, size, nid);
1330 
1331 	new_node = !node_online(nid);
1332 	if (new_node) {
1333 		pgdat = hotadd_new_pgdat(nid, start);
1334 		ret = -ENOMEM;
1335 		if (!pgdat)
1336 			goto error;
1337 	}
1338 
1339 	/* call arch's memory hotadd */
1340 	ret = arch_add_memory(nid, start, size, false);
1341 
1342 	if (ret < 0)
1343 		goto error;
1344 
1345 	/* we online node here. we can't roll back from here. */
1346 	node_set_online(nid);
1347 
1348 	if (new_node) {
1349 		ret = register_one_node(nid);
1350 		/*
1351 		 * If sysfs file of new node can't create, cpu on the node
1352 		 * can't be hot-added. There is no rollback way now.
1353 		 * So, check by BUG_ON() to catch it reluctantly..
1354 		 */
1355 		BUG_ON(ret);
1356 	}
1357 
1358 	/* create new memmap entry */
1359 	firmware_map_add_hotplug(start, start + size, "System RAM");
1360 
1361 	/* online pages if requested */
1362 	if (online)
1363 		walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1),
1364 				  NULL, online_memory_block);
1365 
1366 	goto out;
1367 
1368 error:
1369 	/* rollback pgdat allocation and others */
1370 	if (new_pgdat)
1371 		rollback_node_hotadd(nid, pgdat);
1372 	memblock_remove(start, size);
1373 
1374 out:
1375 	mem_hotplug_done();
1376 	return ret;
1377 }
1378 EXPORT_SYMBOL_GPL(add_memory_resource);
1379 
1380 int __ref add_memory(int nid, u64 start, u64 size)
1381 {
1382 	struct resource *res;
1383 	int ret;
1384 
1385 	res = register_memory_resource(start, size);
1386 	if (IS_ERR(res))
1387 		return PTR_ERR(res);
1388 
1389 	ret = add_memory_resource(nid, res, memhp_auto_online);
1390 	if (ret < 0)
1391 		release_memory_resource(res);
1392 	return ret;
1393 }
1394 EXPORT_SYMBOL_GPL(add_memory);
1395 
1396 #ifdef CONFIG_MEMORY_HOTREMOVE
1397 /*
1398  * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1399  * set and the size of the free page is given by page_order(). Using this,
1400  * the function determines if the pageblock contains only free pages.
1401  * Due to buddy contraints, a free page at least the size of a pageblock will
1402  * be located at the start of the pageblock
1403  */
1404 static inline int pageblock_free(struct page *page)
1405 {
1406 	return PageBuddy(page) && page_order(page) >= pageblock_order;
1407 }
1408 
1409 /* Return the start of the next active pageblock after a given page */
1410 static struct page *next_active_pageblock(struct page *page)
1411 {
1412 	/* Ensure the starting page is pageblock-aligned */
1413 	BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
1414 
1415 	/* If the entire pageblock is free, move to the end of free page */
1416 	if (pageblock_free(page)) {
1417 		int order;
1418 		/* be careful. we don't have locks, page_order can be changed.*/
1419 		order = page_order(page);
1420 		if ((order < MAX_ORDER) && (order >= pageblock_order))
1421 			return page + (1 << order);
1422 	}
1423 
1424 	return page + pageblock_nr_pages;
1425 }
1426 
1427 /* Checks if this range of memory is likely to be hot-removable. */
1428 bool is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
1429 {
1430 	struct page *page = pfn_to_page(start_pfn);
1431 	struct page *end_page = page + nr_pages;
1432 
1433 	/* Check the starting page of each pageblock within the range */
1434 	for (; page < end_page; page = next_active_pageblock(page)) {
1435 		if (!is_pageblock_removable_nolock(page))
1436 			return false;
1437 		cond_resched();
1438 	}
1439 
1440 	/* All pageblocks in the memory block are likely to be hot-removable */
1441 	return true;
1442 }
1443 
1444 /*
1445  * Confirm all pages in a range [start, end) is belongs to the same zone.
1446  */
1447 int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn)
1448 {
1449 	unsigned long pfn, sec_end_pfn;
1450 	struct zone *zone = NULL;
1451 	struct page *page;
1452 	int i;
1453 	for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn);
1454 	     pfn < end_pfn;
1455 	     pfn = sec_end_pfn + 1, sec_end_pfn += PAGES_PER_SECTION) {
1456 		/* Make sure the memory section is present first */
1457 		if (!present_section_nr(pfn_to_section_nr(pfn)))
1458 			continue;
1459 		for (; pfn < sec_end_pfn && pfn < end_pfn;
1460 		     pfn += MAX_ORDER_NR_PAGES) {
1461 			i = 0;
1462 			/* This is just a CONFIG_HOLES_IN_ZONE check.*/
1463 			while ((i < MAX_ORDER_NR_PAGES) &&
1464 				!pfn_valid_within(pfn + i))
1465 				i++;
1466 			if (i == MAX_ORDER_NR_PAGES)
1467 				continue;
1468 			page = pfn_to_page(pfn + i);
1469 			if (zone && page_zone(page) != zone)
1470 				return 0;
1471 			zone = page_zone(page);
1472 		}
1473 	}
1474 	return 1;
1475 }
1476 
1477 /*
1478  * Scan pfn range [start,end) to find movable/migratable pages (LRU pages
1479  * and hugepages). We scan pfn because it's much easier than scanning over
1480  * linked list. This function returns the pfn of the first found movable
1481  * page if it's found, otherwise 0.
1482  */
1483 static unsigned long scan_movable_pages(unsigned long start, unsigned long end)
1484 {
1485 	unsigned long pfn;
1486 	struct page *page;
1487 	for (pfn = start; pfn < end; pfn++) {
1488 		if (pfn_valid(pfn)) {
1489 			page = pfn_to_page(pfn);
1490 			if (PageLRU(page))
1491 				return pfn;
1492 			if (PageHuge(page)) {
1493 				if (page_huge_active(page))
1494 					return pfn;
1495 				else
1496 					pfn = round_up(pfn + 1,
1497 						1 << compound_order(page)) - 1;
1498 			}
1499 		}
1500 	}
1501 	return 0;
1502 }
1503 
1504 #define NR_OFFLINE_AT_ONCE_PAGES	(256)
1505 static int
1506 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1507 {
1508 	unsigned long pfn;
1509 	struct page *page;
1510 	int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
1511 	int not_managed = 0;
1512 	int ret = 0;
1513 	LIST_HEAD(source);
1514 
1515 	for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
1516 		if (!pfn_valid(pfn))
1517 			continue;
1518 		page = pfn_to_page(pfn);
1519 
1520 		if (PageHuge(page)) {
1521 			struct page *head = compound_head(page);
1522 			pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1;
1523 			if (compound_order(head) > PFN_SECTION_SHIFT) {
1524 				ret = -EBUSY;
1525 				break;
1526 			}
1527 			if (isolate_huge_page(page, &source))
1528 				move_pages -= 1 << compound_order(head);
1529 			continue;
1530 		}
1531 
1532 		if (!get_page_unless_zero(page))
1533 			continue;
1534 		/*
1535 		 * We can skip free pages. And we can only deal with pages on
1536 		 * LRU.
1537 		 */
1538 		ret = isolate_lru_page(page);
1539 		if (!ret) { /* Success */
1540 			put_page(page);
1541 			list_add_tail(&page->lru, &source);
1542 			move_pages--;
1543 			inc_zone_page_state(page, NR_ISOLATED_ANON +
1544 					    page_is_file_cache(page));
1545 
1546 		} else {
1547 #ifdef CONFIG_DEBUG_VM
1548 			pr_alert("removing pfn %lx from LRU failed\n", pfn);
1549 			dump_page(page, "failed to remove from LRU");
1550 #endif
1551 			put_page(page);
1552 			/* Because we don't have big zone->lock. we should
1553 			   check this again here. */
1554 			if (page_count(page)) {
1555 				not_managed++;
1556 				ret = -EBUSY;
1557 				break;
1558 			}
1559 		}
1560 	}
1561 	if (!list_empty(&source)) {
1562 		if (not_managed) {
1563 			putback_movable_pages(&source);
1564 			goto out;
1565 		}
1566 
1567 		/*
1568 		 * alloc_migrate_target should be improooooved!!
1569 		 * migrate_pages returns # of failed pages.
1570 		 */
1571 		ret = migrate_pages(&source, alloc_migrate_target, NULL, 0,
1572 					MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
1573 		if (ret)
1574 			putback_movable_pages(&source);
1575 	}
1576 out:
1577 	return ret;
1578 }
1579 
1580 /*
1581  * remove from free_area[] and mark all as Reserved.
1582  */
1583 static int
1584 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1585 			void *data)
1586 {
1587 	__offline_isolated_pages(start, start + nr_pages);
1588 	return 0;
1589 }
1590 
1591 static void
1592 offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
1593 {
1594 	walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
1595 				offline_isolated_pages_cb);
1596 }
1597 
1598 /*
1599  * Check all pages in range, recoreded as memory resource, are isolated.
1600  */
1601 static int
1602 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1603 			void *data)
1604 {
1605 	int ret;
1606 	long offlined = *(long *)data;
1607 	ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true);
1608 	offlined = nr_pages;
1609 	if (!ret)
1610 		*(long *)data += offlined;
1611 	return ret;
1612 }
1613 
1614 static long
1615 check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
1616 {
1617 	long offlined = 0;
1618 	int ret;
1619 
1620 	ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
1621 			check_pages_isolated_cb);
1622 	if (ret < 0)
1623 		offlined = (long)ret;
1624 	return offlined;
1625 }
1626 
1627 #ifdef CONFIG_MOVABLE_NODE
1628 /*
1629  * When CONFIG_MOVABLE_NODE, we permit offlining of a node which doesn't have
1630  * normal memory.
1631  */
1632 static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1633 {
1634 	return true;
1635 }
1636 #else /* CONFIG_MOVABLE_NODE */
1637 /* ensure the node has NORMAL memory if it is still online */
1638 static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1639 {
1640 	struct pglist_data *pgdat = zone->zone_pgdat;
1641 	unsigned long present_pages = 0;
1642 	enum zone_type zt;
1643 
1644 	for (zt = 0; zt <= ZONE_NORMAL; zt++)
1645 		present_pages += pgdat->node_zones[zt].present_pages;
1646 
1647 	if (present_pages > nr_pages)
1648 		return true;
1649 
1650 	present_pages = 0;
1651 	for (; zt <= ZONE_MOVABLE; zt++)
1652 		present_pages += pgdat->node_zones[zt].present_pages;
1653 
1654 	/*
1655 	 * we can't offline the last normal memory until all
1656 	 * higher memory is offlined.
1657 	 */
1658 	return present_pages == 0;
1659 }
1660 #endif /* CONFIG_MOVABLE_NODE */
1661 
1662 static int __init cmdline_parse_movable_node(char *p)
1663 {
1664 #ifdef CONFIG_MOVABLE_NODE
1665 	/*
1666 	 * Memory used by the kernel cannot be hot-removed because Linux
1667 	 * cannot migrate the kernel pages. When memory hotplug is
1668 	 * enabled, we should prevent memblock from allocating memory
1669 	 * for the kernel.
1670 	 *
1671 	 * ACPI SRAT records all hotpluggable memory ranges. But before
1672 	 * SRAT is parsed, we don't know about it.
1673 	 *
1674 	 * The kernel image is loaded into memory at very early time. We
1675 	 * cannot prevent this anyway. So on NUMA system, we set any
1676 	 * node the kernel resides in as un-hotpluggable.
1677 	 *
1678 	 * Since on modern servers, one node could have double-digit
1679 	 * gigabytes memory, we can assume the memory around the kernel
1680 	 * image is also un-hotpluggable. So before SRAT is parsed, just
1681 	 * allocate memory near the kernel image to try the best to keep
1682 	 * the kernel away from hotpluggable memory.
1683 	 */
1684 	memblock_set_bottom_up(true);
1685 	movable_node_enabled = true;
1686 #else
1687 	pr_warn("movable_node option not supported\n");
1688 #endif
1689 	return 0;
1690 }
1691 early_param("movable_node", cmdline_parse_movable_node);
1692 
1693 /* check which state of node_states will be changed when offline memory */
1694 static void node_states_check_changes_offline(unsigned long nr_pages,
1695 		struct zone *zone, struct memory_notify *arg)
1696 {
1697 	struct pglist_data *pgdat = zone->zone_pgdat;
1698 	unsigned long present_pages = 0;
1699 	enum zone_type zt, zone_last = ZONE_NORMAL;
1700 
1701 	/*
1702 	 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
1703 	 * contains nodes which have zones of 0...ZONE_NORMAL,
1704 	 * set zone_last to ZONE_NORMAL.
1705 	 *
1706 	 * If we don't have HIGHMEM nor movable node,
1707 	 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
1708 	 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
1709 	 */
1710 	if (N_MEMORY == N_NORMAL_MEMORY)
1711 		zone_last = ZONE_MOVABLE;
1712 
1713 	/*
1714 	 * check whether node_states[N_NORMAL_MEMORY] will be changed.
1715 	 * If the memory to be offline is in a zone of 0...zone_last,
1716 	 * and it is the last present memory, 0...zone_last will
1717 	 * become empty after offline , thus we can determind we will
1718 	 * need to clear the node from node_states[N_NORMAL_MEMORY].
1719 	 */
1720 	for (zt = 0; zt <= zone_last; zt++)
1721 		present_pages += pgdat->node_zones[zt].present_pages;
1722 	if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1723 		arg->status_change_nid_normal = zone_to_nid(zone);
1724 	else
1725 		arg->status_change_nid_normal = -1;
1726 
1727 #ifdef CONFIG_HIGHMEM
1728 	/*
1729 	 * If we have movable node, node_states[N_HIGH_MEMORY]
1730 	 * contains nodes which have zones of 0...ZONE_HIGHMEM,
1731 	 * set zone_last to ZONE_HIGHMEM.
1732 	 *
1733 	 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
1734 	 * contains nodes which have zones of 0...ZONE_MOVABLE,
1735 	 * set zone_last to ZONE_MOVABLE.
1736 	 */
1737 	zone_last = ZONE_HIGHMEM;
1738 	if (N_MEMORY == N_HIGH_MEMORY)
1739 		zone_last = ZONE_MOVABLE;
1740 
1741 	for (; zt <= zone_last; zt++)
1742 		present_pages += pgdat->node_zones[zt].present_pages;
1743 	if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1744 		arg->status_change_nid_high = zone_to_nid(zone);
1745 	else
1746 		arg->status_change_nid_high = -1;
1747 #else
1748 	arg->status_change_nid_high = arg->status_change_nid_normal;
1749 #endif
1750 
1751 	/*
1752 	 * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE
1753 	 */
1754 	zone_last = ZONE_MOVABLE;
1755 
1756 	/*
1757 	 * check whether node_states[N_HIGH_MEMORY] will be changed
1758 	 * If we try to offline the last present @nr_pages from the node,
1759 	 * we can determind we will need to clear the node from
1760 	 * node_states[N_HIGH_MEMORY].
1761 	 */
1762 	for (; zt <= zone_last; zt++)
1763 		present_pages += pgdat->node_zones[zt].present_pages;
1764 	if (nr_pages >= present_pages)
1765 		arg->status_change_nid = zone_to_nid(zone);
1766 	else
1767 		arg->status_change_nid = -1;
1768 }
1769 
1770 static void node_states_clear_node(int node, struct memory_notify *arg)
1771 {
1772 	if (arg->status_change_nid_normal >= 0)
1773 		node_clear_state(node, N_NORMAL_MEMORY);
1774 
1775 	if ((N_MEMORY != N_NORMAL_MEMORY) &&
1776 	    (arg->status_change_nid_high >= 0))
1777 		node_clear_state(node, N_HIGH_MEMORY);
1778 
1779 	if ((N_MEMORY != N_HIGH_MEMORY) &&
1780 	    (arg->status_change_nid >= 0))
1781 		node_clear_state(node, N_MEMORY);
1782 }
1783 
1784 static int __ref __offline_pages(unsigned long start_pfn,
1785 		  unsigned long end_pfn, unsigned long timeout)
1786 {
1787 	unsigned long pfn, nr_pages, expire;
1788 	long offlined_pages;
1789 	int ret, drain, retry_max, node;
1790 	unsigned long flags;
1791 	struct zone *zone;
1792 	struct memory_notify arg;
1793 
1794 	/* at least, alignment against pageblock is necessary */
1795 	if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
1796 		return -EINVAL;
1797 	if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
1798 		return -EINVAL;
1799 	/* This makes hotplug much easier...and readable.
1800 	   we assume this for now. .*/
1801 	if (!test_pages_in_a_zone(start_pfn, end_pfn))
1802 		return -EINVAL;
1803 
1804 	zone = page_zone(pfn_to_page(start_pfn));
1805 	node = zone_to_nid(zone);
1806 	nr_pages = end_pfn - start_pfn;
1807 
1808 	if (zone_idx(zone) <= ZONE_NORMAL && !can_offline_normal(zone, nr_pages))
1809 		return -EINVAL;
1810 
1811 	/* set above range as isolated */
1812 	ret = start_isolate_page_range(start_pfn, end_pfn,
1813 				       MIGRATE_MOVABLE, true);
1814 	if (ret)
1815 		return ret;
1816 
1817 	arg.start_pfn = start_pfn;
1818 	arg.nr_pages = nr_pages;
1819 	node_states_check_changes_offline(nr_pages, zone, &arg);
1820 
1821 	ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1822 	ret = notifier_to_errno(ret);
1823 	if (ret)
1824 		goto failed_removal;
1825 
1826 	pfn = start_pfn;
1827 	expire = jiffies + timeout;
1828 	drain = 0;
1829 	retry_max = 5;
1830 repeat:
1831 	/* start memory hot removal */
1832 	ret = -EAGAIN;
1833 	if (time_after(jiffies, expire))
1834 		goto failed_removal;
1835 	ret = -EINTR;
1836 	if (signal_pending(current))
1837 		goto failed_removal;
1838 	ret = 0;
1839 	if (drain) {
1840 		lru_add_drain_all();
1841 		cond_resched();
1842 		drain_all_pages(zone);
1843 	}
1844 
1845 	pfn = scan_movable_pages(start_pfn, end_pfn);
1846 	if (pfn) { /* We have movable pages */
1847 		ret = do_migrate_range(pfn, end_pfn);
1848 		if (!ret) {
1849 			drain = 1;
1850 			goto repeat;
1851 		} else {
1852 			if (ret < 0)
1853 				if (--retry_max == 0)
1854 					goto failed_removal;
1855 			yield();
1856 			drain = 1;
1857 			goto repeat;
1858 		}
1859 	}
1860 	/* drain all zone's lru pagevec, this is asynchronous... */
1861 	lru_add_drain_all();
1862 	yield();
1863 	/* drain pcp pages, this is synchronous. */
1864 	drain_all_pages(zone);
1865 	/*
1866 	 * dissolve free hugepages in the memory block before doing offlining
1867 	 * actually in order to make hugetlbfs's object counting consistent.
1868 	 */
1869 	dissolve_free_huge_pages(start_pfn, end_pfn);
1870 	/* check again */
1871 	offlined_pages = check_pages_isolated(start_pfn, end_pfn);
1872 	if (offlined_pages < 0) {
1873 		ret = -EBUSY;
1874 		goto failed_removal;
1875 	}
1876 	pr_info("Offlined Pages %ld\n", offlined_pages);
1877 	/* Ok, all of our target is isolated.
1878 	   We cannot do rollback at this point. */
1879 	offline_isolated_pages(start_pfn, end_pfn);
1880 	/* reset pagetype flags and makes migrate type to be MOVABLE */
1881 	undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1882 	/* removal success */
1883 	adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages);
1884 	zone->present_pages -= offlined_pages;
1885 
1886 	pgdat_resize_lock(zone->zone_pgdat, &flags);
1887 	zone->zone_pgdat->node_present_pages -= offlined_pages;
1888 	pgdat_resize_unlock(zone->zone_pgdat, &flags);
1889 
1890 	init_per_zone_wmark_min();
1891 
1892 	if (!populated_zone(zone)) {
1893 		zone_pcp_reset(zone);
1894 		mutex_lock(&zonelists_mutex);
1895 		build_all_zonelists(NULL, NULL);
1896 		mutex_unlock(&zonelists_mutex);
1897 	} else
1898 		zone_pcp_update(zone);
1899 
1900 	node_states_clear_node(node, &arg);
1901 	if (arg.status_change_nid >= 0) {
1902 		kswapd_stop(node);
1903 		kcompactd_stop(node);
1904 	}
1905 
1906 	vm_total_pages = nr_free_pagecache_pages();
1907 	writeback_set_ratelimit();
1908 
1909 	memory_notify(MEM_OFFLINE, &arg);
1910 	return 0;
1911 
1912 failed_removal:
1913 	pr_debug("memory offlining [mem %#010llx-%#010llx] failed\n",
1914 		 (unsigned long long) start_pfn << PAGE_SHIFT,
1915 		 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
1916 	memory_notify(MEM_CANCEL_OFFLINE, &arg);
1917 	/* pushback to free area */
1918 	undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1919 	return ret;
1920 }
1921 
1922 /* Must be protected by mem_hotplug_begin() */
1923 int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1924 {
1925 	return __offline_pages(start_pfn, start_pfn + nr_pages, 120 * HZ);
1926 }
1927 #endif /* CONFIG_MEMORY_HOTREMOVE */
1928 
1929 /**
1930  * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
1931  * @start_pfn: start pfn of the memory range
1932  * @end_pfn: end pfn of the memory range
1933  * @arg: argument passed to func
1934  * @func: callback for each memory section walked
1935  *
1936  * This function walks through all present mem sections in range
1937  * [start_pfn, end_pfn) and call func on each mem section.
1938  *
1939  * Returns the return value of func.
1940  */
1941 int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn,
1942 		void *arg, int (*func)(struct memory_block *, void *))
1943 {
1944 	struct memory_block *mem = NULL;
1945 	struct mem_section *section;
1946 	unsigned long pfn, section_nr;
1947 	int ret;
1948 
1949 	for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1950 		section_nr = pfn_to_section_nr(pfn);
1951 		if (!present_section_nr(section_nr))
1952 			continue;
1953 
1954 		section = __nr_to_section(section_nr);
1955 		/* same memblock? */
1956 		if (mem)
1957 			if ((section_nr >= mem->start_section_nr) &&
1958 			    (section_nr <= mem->end_section_nr))
1959 				continue;
1960 
1961 		mem = find_memory_block_hinted(section, mem);
1962 		if (!mem)
1963 			continue;
1964 
1965 		ret = func(mem, arg);
1966 		if (ret) {
1967 			kobject_put(&mem->dev.kobj);
1968 			return ret;
1969 		}
1970 	}
1971 
1972 	if (mem)
1973 		kobject_put(&mem->dev.kobj);
1974 
1975 	return 0;
1976 }
1977 
1978 #ifdef CONFIG_MEMORY_HOTREMOVE
1979 static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
1980 {
1981 	int ret = !is_memblock_offlined(mem);
1982 
1983 	if (unlikely(ret)) {
1984 		phys_addr_t beginpa, endpa;
1985 
1986 		beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
1987 		endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1;
1988 		pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n",
1989 			&beginpa, &endpa);
1990 	}
1991 
1992 	return ret;
1993 }
1994 
1995 static int check_cpu_on_node(pg_data_t *pgdat)
1996 {
1997 	int cpu;
1998 
1999 	for_each_present_cpu(cpu) {
2000 		if (cpu_to_node(cpu) == pgdat->node_id)
2001 			/*
2002 			 * the cpu on this node isn't removed, and we can't
2003 			 * offline this node.
2004 			 */
2005 			return -EBUSY;
2006 	}
2007 
2008 	return 0;
2009 }
2010 
2011 static void unmap_cpu_on_node(pg_data_t *pgdat)
2012 {
2013 #ifdef CONFIG_ACPI_NUMA
2014 	int cpu;
2015 
2016 	for_each_possible_cpu(cpu)
2017 		if (cpu_to_node(cpu) == pgdat->node_id)
2018 			numa_clear_node(cpu);
2019 #endif
2020 }
2021 
2022 static int check_and_unmap_cpu_on_node(pg_data_t *pgdat)
2023 {
2024 	int ret;
2025 
2026 	ret = check_cpu_on_node(pgdat);
2027 	if (ret)
2028 		return ret;
2029 
2030 	/*
2031 	 * the node will be offlined when we come here, so we can clear
2032 	 * the cpu_to_node() now.
2033 	 */
2034 
2035 	unmap_cpu_on_node(pgdat);
2036 	return 0;
2037 }
2038 
2039 /**
2040  * try_offline_node
2041  *
2042  * Offline a node if all memory sections and cpus of the node are removed.
2043  *
2044  * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
2045  * and online/offline operations before this call.
2046  */
2047 void try_offline_node(int nid)
2048 {
2049 	pg_data_t *pgdat = NODE_DATA(nid);
2050 	unsigned long start_pfn = pgdat->node_start_pfn;
2051 	unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
2052 	unsigned long pfn;
2053 	int i;
2054 
2055 	for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
2056 		unsigned long section_nr = pfn_to_section_nr(pfn);
2057 
2058 		if (!present_section_nr(section_nr))
2059 			continue;
2060 
2061 		if (pfn_to_nid(pfn) != nid)
2062 			continue;
2063 
2064 		/*
2065 		 * some memory sections of this node are not removed, and we
2066 		 * can't offline node now.
2067 		 */
2068 		return;
2069 	}
2070 
2071 	if (check_and_unmap_cpu_on_node(pgdat))
2072 		return;
2073 
2074 	/*
2075 	 * all memory/cpu of this node are removed, we can offline this
2076 	 * node now.
2077 	 */
2078 	node_set_offline(nid);
2079 	unregister_one_node(nid);
2080 
2081 	/* free waittable in each zone */
2082 	for (i = 0; i < MAX_NR_ZONES; i++) {
2083 		struct zone *zone = pgdat->node_zones + i;
2084 
2085 		/*
2086 		 * wait_table may be allocated from boot memory,
2087 		 * here only free if it's allocated by vmalloc.
2088 		 */
2089 		if (is_vmalloc_addr(zone->wait_table)) {
2090 			vfree(zone->wait_table);
2091 			zone->wait_table = NULL;
2092 		}
2093 	}
2094 }
2095 EXPORT_SYMBOL(try_offline_node);
2096 
2097 /**
2098  * remove_memory
2099  *
2100  * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
2101  * and online/offline operations before this call, as required by
2102  * try_offline_node().
2103  */
2104 void __ref remove_memory(int nid, u64 start, u64 size)
2105 {
2106 	int ret;
2107 
2108 	BUG_ON(check_hotplug_memory_range(start, size));
2109 
2110 	mem_hotplug_begin();
2111 
2112 	/*
2113 	 * All memory blocks must be offlined before removing memory.  Check
2114 	 * whether all memory blocks in question are offline and trigger a BUG()
2115 	 * if this is not the case.
2116 	 */
2117 	ret = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL,
2118 				check_memblock_offlined_cb);
2119 	if (ret)
2120 		BUG();
2121 
2122 	/* remove memmap entry */
2123 	firmware_map_remove(start, start + size, "System RAM");
2124 	memblock_free(start, size);
2125 	memblock_remove(start, size);
2126 
2127 	arch_remove_memory(start, size);
2128 
2129 	try_offline_node(nid);
2130 
2131 	mem_hotplug_done();
2132 }
2133 EXPORT_SYMBOL_GPL(remove_memory);
2134 #endif /* CONFIG_MEMORY_HOTREMOVE */
2135