xref: /openbmc/linux/mm/memory_hotplug.c (revision 95777591)
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/sched/signal.h>
10 #include <linux/swap.h>
11 #include <linux/interrupt.h>
12 #include <linux/pagemap.h>
13 #include <linux/compiler.h>
14 #include <linux/export.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/memremap.h>
22 #include <linux/memory_hotplug.h>
23 #include <linux/highmem.h>
24 #include <linux/vmalloc.h>
25 #include <linux/ioport.h>
26 #include <linux/delay.h>
27 #include <linux/migrate.h>
28 #include <linux/page-isolation.h>
29 #include <linux/pfn.h>
30 #include <linux/suspend.h>
31 #include <linux/mm_inline.h>
32 #include <linux/firmware-map.h>
33 #include <linux/stop_machine.h>
34 #include <linux/hugetlb.h>
35 #include <linux/memblock.h>
36 #include <linux/compaction.h>
37 #include <linux/rmap.h>
38 
39 #include <asm/tlbflush.h>
40 
41 #include "internal.h"
42 
43 /*
44  * online_page_callback contains pointer to current page onlining function.
45  * Initially it is generic_online_page(). If it is required it could be
46  * changed by calling set_online_page_callback() for callback registration
47  * and restore_online_page_callback() for generic callback restore.
48  */
49 
50 static void generic_online_page(struct page *page, unsigned int order);
51 
52 static online_page_callback_t online_page_callback = generic_online_page;
53 static DEFINE_MUTEX(online_page_callback_lock);
54 
55 DEFINE_STATIC_PERCPU_RWSEM(mem_hotplug_lock);
56 
57 void get_online_mems(void)
58 {
59 	percpu_down_read(&mem_hotplug_lock);
60 }
61 
62 void put_online_mems(void)
63 {
64 	percpu_up_read(&mem_hotplug_lock);
65 }
66 
67 bool movable_node_enabled = false;
68 
69 #ifndef CONFIG_MEMORY_HOTPLUG_DEFAULT_ONLINE
70 bool memhp_auto_online;
71 #else
72 bool memhp_auto_online = true;
73 #endif
74 EXPORT_SYMBOL_GPL(memhp_auto_online);
75 
76 static int __init setup_memhp_default_state(char *str)
77 {
78 	if (!strcmp(str, "online"))
79 		memhp_auto_online = true;
80 	else if (!strcmp(str, "offline"))
81 		memhp_auto_online = false;
82 
83 	return 1;
84 }
85 __setup("memhp_default_state=", setup_memhp_default_state);
86 
87 void mem_hotplug_begin(void)
88 {
89 	cpus_read_lock();
90 	percpu_down_write(&mem_hotplug_lock);
91 }
92 
93 void mem_hotplug_done(void)
94 {
95 	percpu_up_write(&mem_hotplug_lock);
96 	cpus_read_unlock();
97 }
98 
99 /* add this memory to iomem resource */
100 static struct resource *register_memory_resource(u64 start, u64 size)
101 {
102 	struct resource *res, *conflict;
103 	res = kzalloc(sizeof(struct resource), GFP_KERNEL);
104 	if (!res)
105 		return ERR_PTR(-ENOMEM);
106 
107 	res->name = "System RAM";
108 	res->start = start;
109 	res->end = start + size - 1;
110 	res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
111 	conflict =  request_resource_conflict(&iomem_resource, res);
112 	if (conflict) {
113 		if (conflict->desc == IORES_DESC_DEVICE_PRIVATE_MEMORY) {
114 			pr_debug("Device unaddressable memory block "
115 				 "memory hotplug at %#010llx !\n",
116 				 (unsigned long long)start);
117 		}
118 		pr_debug("System RAM resource %pR cannot be added\n", res);
119 		kfree(res);
120 		return ERR_PTR(-EEXIST);
121 	}
122 	return res;
123 }
124 
125 static void release_memory_resource(struct resource *res)
126 {
127 	if (!res)
128 		return;
129 	release_resource(res);
130 	kfree(res);
131 	return;
132 }
133 
134 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
135 void get_page_bootmem(unsigned long info,  struct page *page,
136 		      unsigned long type)
137 {
138 	page->freelist = (void *)type;
139 	SetPagePrivate(page);
140 	set_page_private(page, info);
141 	page_ref_inc(page);
142 }
143 
144 void put_page_bootmem(struct page *page)
145 {
146 	unsigned long type;
147 
148 	type = (unsigned long) page->freelist;
149 	BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE ||
150 	       type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE);
151 
152 	if (page_ref_dec_return(page) == 1) {
153 		page->freelist = NULL;
154 		ClearPagePrivate(page);
155 		set_page_private(page, 0);
156 		INIT_LIST_HEAD(&page->lru);
157 		free_reserved_page(page);
158 	}
159 }
160 
161 #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
162 #ifndef CONFIG_SPARSEMEM_VMEMMAP
163 static void register_page_bootmem_info_section(unsigned long start_pfn)
164 {
165 	unsigned long *usemap, mapsize, section_nr, i;
166 	struct mem_section *ms;
167 	struct page *page, *memmap;
168 
169 	section_nr = pfn_to_section_nr(start_pfn);
170 	ms = __nr_to_section(section_nr);
171 
172 	/* Get section's memmap address */
173 	memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
174 
175 	/*
176 	 * Get page for the memmap's phys address
177 	 * XXX: need more consideration for sparse_vmemmap...
178 	 */
179 	page = virt_to_page(memmap);
180 	mapsize = sizeof(struct page) * PAGES_PER_SECTION;
181 	mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
182 
183 	/* remember memmap's page */
184 	for (i = 0; i < mapsize; i++, page++)
185 		get_page_bootmem(section_nr, page, SECTION_INFO);
186 
187 	usemap = ms->pageblock_flags;
188 	page = virt_to_page(usemap);
189 
190 	mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
191 
192 	for (i = 0; i < mapsize; i++, page++)
193 		get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
194 
195 }
196 #else /* CONFIG_SPARSEMEM_VMEMMAP */
197 static void register_page_bootmem_info_section(unsigned long start_pfn)
198 {
199 	unsigned long *usemap, mapsize, section_nr, i;
200 	struct mem_section *ms;
201 	struct page *page, *memmap;
202 
203 	section_nr = pfn_to_section_nr(start_pfn);
204 	ms = __nr_to_section(section_nr);
205 
206 	memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
207 
208 	register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION);
209 
210 	usemap = ms->pageblock_flags;
211 	page = virt_to_page(usemap);
212 
213 	mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
214 
215 	for (i = 0; i < mapsize; i++, page++)
216 		get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
217 }
218 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
219 
220 void __init register_page_bootmem_info_node(struct pglist_data *pgdat)
221 {
222 	unsigned long i, pfn, end_pfn, nr_pages;
223 	int node = pgdat->node_id;
224 	struct page *page;
225 
226 	nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
227 	page = virt_to_page(pgdat);
228 
229 	for (i = 0; i < nr_pages; i++, page++)
230 		get_page_bootmem(node, page, NODE_INFO);
231 
232 	pfn = pgdat->node_start_pfn;
233 	end_pfn = pgdat_end_pfn(pgdat);
234 
235 	/* register section info */
236 	for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
237 		/*
238 		 * Some platforms can assign the same pfn to multiple nodes - on
239 		 * node0 as well as nodeN.  To avoid registering a pfn against
240 		 * multiple nodes we check that this pfn does not already
241 		 * reside in some other nodes.
242 		 */
243 		if (pfn_valid(pfn) && (early_pfn_to_nid(pfn) == node))
244 			register_page_bootmem_info_section(pfn);
245 	}
246 }
247 #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
248 
249 static int __meminit __add_section(int nid, unsigned long phys_start_pfn,
250 		struct vmem_altmap *altmap, bool want_memblock)
251 {
252 	int ret;
253 
254 	if (pfn_valid(phys_start_pfn))
255 		return -EEXIST;
256 
257 	ret = sparse_add_one_section(nid, phys_start_pfn, altmap);
258 	if (ret < 0)
259 		return ret;
260 
261 	if (!want_memblock)
262 		return 0;
263 
264 	return hotplug_memory_register(nid, __pfn_to_section(phys_start_pfn));
265 }
266 
267 /*
268  * Reasonably generic function for adding memory.  It is
269  * expected that archs that support memory hotplug will
270  * call this function after deciding the zone to which to
271  * add the new pages.
272  */
273 int __ref __add_pages(int nid, unsigned long phys_start_pfn,
274 		unsigned long nr_pages, struct vmem_altmap *altmap,
275 		bool want_memblock)
276 {
277 	unsigned long i;
278 	int err = 0;
279 	int start_sec, end_sec;
280 
281 	/* during initialize mem_map, align hot-added range to section */
282 	start_sec = pfn_to_section_nr(phys_start_pfn);
283 	end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
284 
285 	if (altmap) {
286 		/*
287 		 * Validate altmap is within bounds of the total request
288 		 */
289 		if (altmap->base_pfn != phys_start_pfn
290 				|| vmem_altmap_offset(altmap) > nr_pages) {
291 			pr_warn_once("memory add fail, invalid altmap\n");
292 			err = -EINVAL;
293 			goto out;
294 		}
295 		altmap->alloc = 0;
296 	}
297 
298 	for (i = start_sec; i <= end_sec; i++) {
299 		err = __add_section(nid, section_nr_to_pfn(i), altmap,
300 				want_memblock);
301 
302 		/*
303 		 * EEXIST is finally dealt with by ioresource collision
304 		 * check. see add_memory() => register_memory_resource()
305 		 * Warning will be printed if there is collision.
306 		 */
307 		if (err && (err != -EEXIST))
308 			break;
309 		err = 0;
310 		cond_resched();
311 	}
312 	vmemmap_populate_print_last();
313 out:
314 	return err;
315 }
316 
317 #ifdef CONFIG_MEMORY_HOTREMOVE
318 /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
319 static unsigned long find_smallest_section_pfn(int nid, struct zone *zone,
320 				     unsigned long start_pfn,
321 				     unsigned long end_pfn)
322 {
323 	struct mem_section *ms;
324 
325 	for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) {
326 		ms = __pfn_to_section(start_pfn);
327 
328 		if (unlikely(!valid_section(ms)))
329 			continue;
330 
331 		if (unlikely(pfn_to_nid(start_pfn) != nid))
332 			continue;
333 
334 		if (zone && zone != page_zone(pfn_to_page(start_pfn)))
335 			continue;
336 
337 		return start_pfn;
338 	}
339 
340 	return 0;
341 }
342 
343 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
344 static unsigned long find_biggest_section_pfn(int nid, struct zone *zone,
345 				    unsigned long start_pfn,
346 				    unsigned long end_pfn)
347 {
348 	struct mem_section *ms;
349 	unsigned long pfn;
350 
351 	/* pfn is the end pfn of a memory section. */
352 	pfn = end_pfn - 1;
353 	for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) {
354 		ms = __pfn_to_section(pfn);
355 
356 		if (unlikely(!valid_section(ms)))
357 			continue;
358 
359 		if (unlikely(pfn_to_nid(pfn) != nid))
360 			continue;
361 
362 		if (zone && zone != page_zone(pfn_to_page(pfn)))
363 			continue;
364 
365 		return pfn;
366 	}
367 
368 	return 0;
369 }
370 
371 static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
372 			     unsigned long end_pfn)
373 {
374 	unsigned long zone_start_pfn = zone->zone_start_pfn;
375 	unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */
376 	unsigned long zone_end_pfn = z;
377 	unsigned long pfn;
378 	struct mem_section *ms;
379 	int nid = zone_to_nid(zone);
380 
381 	zone_span_writelock(zone);
382 	if (zone_start_pfn == start_pfn) {
383 		/*
384 		 * If the section is smallest section in the zone, it need
385 		 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
386 		 * In this case, we find second smallest valid mem_section
387 		 * for shrinking zone.
388 		 */
389 		pfn = find_smallest_section_pfn(nid, zone, end_pfn,
390 						zone_end_pfn);
391 		if (pfn) {
392 			zone->zone_start_pfn = pfn;
393 			zone->spanned_pages = zone_end_pfn - pfn;
394 		}
395 	} else if (zone_end_pfn == end_pfn) {
396 		/*
397 		 * If the section is biggest section in the zone, it need
398 		 * shrink zone->spanned_pages.
399 		 * In this case, we find second biggest valid mem_section for
400 		 * shrinking zone.
401 		 */
402 		pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn,
403 					       start_pfn);
404 		if (pfn)
405 			zone->spanned_pages = pfn - zone_start_pfn + 1;
406 	}
407 
408 	/*
409 	 * The section is not biggest or smallest mem_section in the zone, it
410 	 * only creates a hole in the zone. So in this case, we need not
411 	 * change the zone. But perhaps, the zone has only hole data. Thus
412 	 * it check the zone has only hole or not.
413 	 */
414 	pfn = zone_start_pfn;
415 	for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) {
416 		ms = __pfn_to_section(pfn);
417 
418 		if (unlikely(!valid_section(ms)))
419 			continue;
420 
421 		if (page_zone(pfn_to_page(pfn)) != zone)
422 			continue;
423 
424 		 /* If the section is current section, it continues the loop */
425 		if (start_pfn == pfn)
426 			continue;
427 
428 		/* If we find valid section, we have nothing to do */
429 		zone_span_writeunlock(zone);
430 		return;
431 	}
432 
433 	/* The zone has no valid section */
434 	zone->zone_start_pfn = 0;
435 	zone->spanned_pages = 0;
436 	zone_span_writeunlock(zone);
437 }
438 
439 static void shrink_pgdat_span(struct pglist_data *pgdat,
440 			      unsigned long start_pfn, unsigned long end_pfn)
441 {
442 	unsigned long pgdat_start_pfn = pgdat->node_start_pfn;
443 	unsigned long p = pgdat_end_pfn(pgdat); /* pgdat_end_pfn namespace clash */
444 	unsigned long pgdat_end_pfn = p;
445 	unsigned long pfn;
446 	struct mem_section *ms;
447 	int nid = pgdat->node_id;
448 
449 	if (pgdat_start_pfn == start_pfn) {
450 		/*
451 		 * If the section is smallest section in the pgdat, it need
452 		 * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages.
453 		 * In this case, we find second smallest valid mem_section
454 		 * for shrinking zone.
455 		 */
456 		pfn = find_smallest_section_pfn(nid, NULL, end_pfn,
457 						pgdat_end_pfn);
458 		if (pfn) {
459 			pgdat->node_start_pfn = pfn;
460 			pgdat->node_spanned_pages = pgdat_end_pfn - pfn;
461 		}
462 	} else if (pgdat_end_pfn == end_pfn) {
463 		/*
464 		 * If the section is biggest section in the pgdat, it need
465 		 * shrink pgdat->node_spanned_pages.
466 		 * In this case, we find second biggest valid mem_section for
467 		 * shrinking zone.
468 		 */
469 		pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn,
470 					       start_pfn);
471 		if (pfn)
472 			pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1;
473 	}
474 
475 	/*
476 	 * If the section is not biggest or smallest mem_section in the pgdat,
477 	 * it only creates a hole in the pgdat. So in this case, we need not
478 	 * change the pgdat.
479 	 * But perhaps, the pgdat has only hole data. Thus it check the pgdat
480 	 * has only hole or not.
481 	 */
482 	pfn = pgdat_start_pfn;
483 	for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) {
484 		ms = __pfn_to_section(pfn);
485 
486 		if (unlikely(!valid_section(ms)))
487 			continue;
488 
489 		if (pfn_to_nid(pfn) != nid)
490 			continue;
491 
492 		 /* If the section is current section, it continues the loop */
493 		if (start_pfn == pfn)
494 			continue;
495 
496 		/* If we find valid section, we have nothing to do */
497 		return;
498 	}
499 
500 	/* The pgdat has no valid section */
501 	pgdat->node_start_pfn = 0;
502 	pgdat->node_spanned_pages = 0;
503 }
504 
505 static void __remove_zone(struct zone *zone, unsigned long start_pfn)
506 {
507 	struct pglist_data *pgdat = zone->zone_pgdat;
508 	int nr_pages = PAGES_PER_SECTION;
509 	unsigned long flags;
510 
511 	pgdat_resize_lock(zone->zone_pgdat, &flags);
512 	shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
513 	shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages);
514 	pgdat_resize_unlock(zone->zone_pgdat, &flags);
515 }
516 
517 static int __remove_section(struct zone *zone, struct mem_section *ms,
518 		unsigned long map_offset, struct vmem_altmap *altmap)
519 {
520 	unsigned long start_pfn;
521 	int scn_nr;
522 	int ret = -EINVAL;
523 
524 	if (!valid_section(ms))
525 		return ret;
526 
527 	ret = unregister_memory_section(ms);
528 	if (ret)
529 		return ret;
530 
531 	scn_nr = __section_nr(ms);
532 	start_pfn = section_nr_to_pfn((unsigned long)scn_nr);
533 	__remove_zone(zone, start_pfn);
534 
535 	sparse_remove_one_section(zone, ms, map_offset, altmap);
536 	return 0;
537 }
538 
539 /**
540  * __remove_pages() - remove sections of pages from a zone
541  * @zone: zone from which pages need to be removed
542  * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
543  * @nr_pages: number of pages to remove (must be multiple of section size)
544  * @altmap: alternative device page map or %NULL if default memmap is used
545  *
546  * Generic helper function to remove section mappings and sysfs entries
547  * for the section of the memory we are removing. Caller needs to make
548  * sure that pages are marked reserved and zones are adjust properly by
549  * calling offline_pages().
550  */
551 int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
552 		 unsigned long nr_pages, struct vmem_altmap *altmap)
553 {
554 	unsigned long i;
555 	unsigned long map_offset = 0;
556 	int sections_to_remove, ret = 0;
557 
558 	/* In the ZONE_DEVICE case device driver owns the memory region */
559 	if (is_dev_zone(zone)) {
560 		if (altmap)
561 			map_offset = vmem_altmap_offset(altmap);
562 	} else {
563 		resource_size_t start, size;
564 
565 		start = phys_start_pfn << PAGE_SHIFT;
566 		size = nr_pages * PAGE_SIZE;
567 
568 		ret = release_mem_region_adjustable(&iomem_resource, start,
569 					size);
570 		if (ret) {
571 			resource_size_t endres = start + size - 1;
572 
573 			pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
574 					&start, &endres, ret);
575 		}
576 	}
577 
578 	clear_zone_contiguous(zone);
579 
580 	/*
581 	 * We can only remove entire sections
582 	 */
583 	BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
584 	BUG_ON(nr_pages % PAGES_PER_SECTION);
585 
586 	sections_to_remove = nr_pages / PAGES_PER_SECTION;
587 	for (i = 0; i < sections_to_remove; i++) {
588 		unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
589 
590 		cond_resched();
591 		ret = __remove_section(zone, __pfn_to_section(pfn), map_offset,
592 				altmap);
593 		map_offset = 0;
594 		if (ret)
595 			break;
596 	}
597 
598 	set_zone_contiguous(zone);
599 
600 	return ret;
601 }
602 #endif /* CONFIG_MEMORY_HOTREMOVE */
603 
604 int set_online_page_callback(online_page_callback_t callback)
605 {
606 	int rc = -EINVAL;
607 
608 	get_online_mems();
609 	mutex_lock(&online_page_callback_lock);
610 
611 	if (online_page_callback == generic_online_page) {
612 		online_page_callback = callback;
613 		rc = 0;
614 	}
615 
616 	mutex_unlock(&online_page_callback_lock);
617 	put_online_mems();
618 
619 	return rc;
620 }
621 EXPORT_SYMBOL_GPL(set_online_page_callback);
622 
623 int restore_online_page_callback(online_page_callback_t callback)
624 {
625 	int rc = -EINVAL;
626 
627 	get_online_mems();
628 	mutex_lock(&online_page_callback_lock);
629 
630 	if (online_page_callback == callback) {
631 		online_page_callback = generic_online_page;
632 		rc = 0;
633 	}
634 
635 	mutex_unlock(&online_page_callback_lock);
636 	put_online_mems();
637 
638 	return rc;
639 }
640 EXPORT_SYMBOL_GPL(restore_online_page_callback);
641 
642 void __online_page_set_limits(struct page *page)
643 {
644 }
645 EXPORT_SYMBOL_GPL(__online_page_set_limits);
646 
647 void __online_page_increment_counters(struct page *page)
648 {
649 	adjust_managed_page_count(page, 1);
650 }
651 EXPORT_SYMBOL_GPL(__online_page_increment_counters);
652 
653 void __online_page_free(struct page *page)
654 {
655 	__free_reserved_page(page);
656 }
657 EXPORT_SYMBOL_GPL(__online_page_free);
658 
659 static void generic_online_page(struct page *page, unsigned int order)
660 {
661 	kernel_map_pages(page, 1 << order, 1);
662 	__free_pages_core(page, order);
663 	totalram_pages_add(1UL << order);
664 #ifdef CONFIG_HIGHMEM
665 	if (PageHighMem(page))
666 		totalhigh_pages_add(1UL << order);
667 #endif
668 }
669 
670 static int online_pages_blocks(unsigned long start, unsigned long nr_pages)
671 {
672 	unsigned long end = start + nr_pages;
673 	int order, onlined_pages = 0;
674 
675 	while (start < end) {
676 		order = min(MAX_ORDER - 1,
677 			get_order(PFN_PHYS(end) - PFN_PHYS(start)));
678 		(*online_page_callback)(pfn_to_page(start), order);
679 
680 		onlined_pages += (1UL << order);
681 		start += (1UL << order);
682 	}
683 	return onlined_pages;
684 }
685 
686 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
687 			void *arg)
688 {
689 	unsigned long onlined_pages = *(unsigned long *)arg;
690 
691 	if (PageReserved(pfn_to_page(start_pfn)))
692 		onlined_pages += online_pages_blocks(start_pfn, nr_pages);
693 
694 	online_mem_sections(start_pfn, start_pfn + nr_pages);
695 
696 	*(unsigned long *)arg = onlined_pages;
697 	return 0;
698 }
699 
700 /* check which state of node_states will be changed when online memory */
701 static void node_states_check_changes_online(unsigned long nr_pages,
702 	struct zone *zone, struct memory_notify *arg)
703 {
704 	int nid = zone_to_nid(zone);
705 
706 	arg->status_change_nid = NUMA_NO_NODE;
707 	arg->status_change_nid_normal = NUMA_NO_NODE;
708 	arg->status_change_nid_high = NUMA_NO_NODE;
709 
710 	if (!node_state(nid, N_MEMORY))
711 		arg->status_change_nid = nid;
712 	if (zone_idx(zone) <= ZONE_NORMAL && !node_state(nid, N_NORMAL_MEMORY))
713 		arg->status_change_nid_normal = nid;
714 #ifdef CONFIG_HIGHMEM
715 	if (zone_idx(zone) <= N_HIGH_MEMORY && !node_state(nid, N_HIGH_MEMORY))
716 		arg->status_change_nid_high = nid;
717 #endif
718 }
719 
720 static void node_states_set_node(int node, struct memory_notify *arg)
721 {
722 	if (arg->status_change_nid_normal >= 0)
723 		node_set_state(node, N_NORMAL_MEMORY);
724 
725 	if (arg->status_change_nid_high >= 0)
726 		node_set_state(node, N_HIGH_MEMORY);
727 
728 	if (arg->status_change_nid >= 0)
729 		node_set_state(node, N_MEMORY);
730 }
731 
732 static void __meminit resize_zone_range(struct zone *zone, unsigned long start_pfn,
733 		unsigned long nr_pages)
734 {
735 	unsigned long old_end_pfn = zone_end_pfn(zone);
736 
737 	if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
738 		zone->zone_start_pfn = start_pfn;
739 
740 	zone->spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - zone->zone_start_pfn;
741 }
742 
743 static void __meminit resize_pgdat_range(struct pglist_data *pgdat, unsigned long start_pfn,
744                                      unsigned long nr_pages)
745 {
746 	unsigned long old_end_pfn = pgdat_end_pfn(pgdat);
747 
748 	if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
749 		pgdat->node_start_pfn = start_pfn;
750 
751 	pgdat->node_spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - pgdat->node_start_pfn;
752 }
753 
754 void __ref move_pfn_range_to_zone(struct zone *zone, unsigned long start_pfn,
755 		unsigned long nr_pages, struct vmem_altmap *altmap)
756 {
757 	struct pglist_data *pgdat = zone->zone_pgdat;
758 	int nid = pgdat->node_id;
759 	unsigned long flags;
760 
761 	clear_zone_contiguous(zone);
762 
763 	/* TODO Huh pgdat is irqsave while zone is not. It used to be like that before */
764 	pgdat_resize_lock(pgdat, &flags);
765 	zone_span_writelock(zone);
766 	if (zone_is_empty(zone))
767 		init_currently_empty_zone(zone, start_pfn, nr_pages);
768 	resize_zone_range(zone, start_pfn, nr_pages);
769 	zone_span_writeunlock(zone);
770 	resize_pgdat_range(pgdat, start_pfn, nr_pages);
771 	pgdat_resize_unlock(pgdat, &flags);
772 
773 	/*
774 	 * TODO now we have a visible range of pages which are not associated
775 	 * with their zone properly. Not nice but set_pfnblock_flags_mask
776 	 * expects the zone spans the pfn range. All the pages in the range
777 	 * are reserved so nobody should be touching them so we should be safe
778 	 */
779 	memmap_init_zone(nr_pages, nid, zone_idx(zone), start_pfn,
780 			MEMMAP_HOTPLUG, altmap);
781 
782 	set_zone_contiguous(zone);
783 }
784 
785 /*
786  * Returns a default kernel memory zone for the given pfn range.
787  * If no kernel zone covers this pfn range it will automatically go
788  * to the ZONE_NORMAL.
789  */
790 static struct zone *default_kernel_zone_for_pfn(int nid, unsigned long start_pfn,
791 		unsigned long nr_pages)
792 {
793 	struct pglist_data *pgdat = NODE_DATA(nid);
794 	int zid;
795 
796 	for (zid = 0; zid <= ZONE_NORMAL; zid++) {
797 		struct zone *zone = &pgdat->node_zones[zid];
798 
799 		if (zone_intersects(zone, start_pfn, nr_pages))
800 			return zone;
801 	}
802 
803 	return &pgdat->node_zones[ZONE_NORMAL];
804 }
805 
806 static inline struct zone *default_zone_for_pfn(int nid, unsigned long start_pfn,
807 		unsigned long nr_pages)
808 {
809 	struct zone *kernel_zone = default_kernel_zone_for_pfn(nid, start_pfn,
810 			nr_pages);
811 	struct zone *movable_zone = &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
812 	bool in_kernel = zone_intersects(kernel_zone, start_pfn, nr_pages);
813 	bool in_movable = zone_intersects(movable_zone, start_pfn, nr_pages);
814 
815 	/*
816 	 * We inherit the existing zone in a simple case where zones do not
817 	 * overlap in the given range
818 	 */
819 	if (in_kernel ^ in_movable)
820 		return (in_kernel) ? kernel_zone : movable_zone;
821 
822 	/*
823 	 * If the range doesn't belong to any zone or two zones overlap in the
824 	 * given range then we use movable zone only if movable_node is
825 	 * enabled because we always online to a kernel zone by default.
826 	 */
827 	return movable_node_enabled ? movable_zone : kernel_zone;
828 }
829 
830 struct zone * zone_for_pfn_range(int online_type, int nid, unsigned start_pfn,
831 		unsigned long nr_pages)
832 {
833 	if (online_type == MMOP_ONLINE_KERNEL)
834 		return default_kernel_zone_for_pfn(nid, start_pfn, nr_pages);
835 
836 	if (online_type == MMOP_ONLINE_MOVABLE)
837 		return &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
838 
839 	return default_zone_for_pfn(nid, start_pfn, nr_pages);
840 }
841 
842 /*
843  * Associates the given pfn range with the given node and the zone appropriate
844  * for the given online type.
845  */
846 static struct zone * __meminit move_pfn_range(int online_type, int nid,
847 		unsigned long start_pfn, unsigned long nr_pages)
848 {
849 	struct zone *zone;
850 
851 	zone = zone_for_pfn_range(online_type, nid, start_pfn, nr_pages);
852 	move_pfn_range_to_zone(zone, start_pfn, nr_pages, NULL);
853 	return zone;
854 }
855 
856 int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type)
857 {
858 	unsigned long flags;
859 	unsigned long onlined_pages = 0;
860 	struct zone *zone;
861 	int need_zonelists_rebuild = 0;
862 	int nid;
863 	int ret;
864 	struct memory_notify arg;
865 	struct memory_block *mem;
866 
867 	mem_hotplug_begin();
868 
869 	/*
870 	 * We can't use pfn_to_nid() because nid might be stored in struct page
871 	 * which is not yet initialized. Instead, we find nid from memory block.
872 	 */
873 	mem = find_memory_block(__pfn_to_section(pfn));
874 	nid = mem->nid;
875 
876 	/* associate pfn range with the zone */
877 	zone = move_pfn_range(online_type, nid, pfn, nr_pages);
878 
879 	arg.start_pfn = pfn;
880 	arg.nr_pages = nr_pages;
881 	node_states_check_changes_online(nr_pages, zone, &arg);
882 
883 	ret = memory_notify(MEM_GOING_ONLINE, &arg);
884 	ret = notifier_to_errno(ret);
885 	if (ret)
886 		goto failed_addition;
887 
888 	/*
889 	 * If this zone is not populated, then it is not in zonelist.
890 	 * This means the page allocator ignores this zone.
891 	 * So, zonelist must be updated after online.
892 	 */
893 	if (!populated_zone(zone)) {
894 		need_zonelists_rebuild = 1;
895 		setup_zone_pageset(zone);
896 	}
897 
898 	ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
899 		online_pages_range);
900 	if (ret) {
901 		if (need_zonelists_rebuild)
902 			zone_pcp_reset(zone);
903 		goto failed_addition;
904 	}
905 
906 	zone->present_pages += onlined_pages;
907 
908 	pgdat_resize_lock(zone->zone_pgdat, &flags);
909 	zone->zone_pgdat->node_present_pages += onlined_pages;
910 	pgdat_resize_unlock(zone->zone_pgdat, &flags);
911 
912 	if (onlined_pages) {
913 		node_states_set_node(nid, &arg);
914 		if (need_zonelists_rebuild)
915 			build_all_zonelists(NULL);
916 		else
917 			zone_pcp_update(zone);
918 	}
919 
920 	init_per_zone_wmark_min();
921 
922 	if (onlined_pages) {
923 		kswapd_run(nid);
924 		kcompactd_run(nid);
925 	}
926 
927 	vm_total_pages = nr_free_pagecache_pages();
928 
929 	writeback_set_ratelimit();
930 
931 	if (onlined_pages)
932 		memory_notify(MEM_ONLINE, &arg);
933 	mem_hotplug_done();
934 	return 0;
935 
936 failed_addition:
937 	pr_debug("online_pages [mem %#010llx-%#010llx] failed\n",
938 		 (unsigned long long) pfn << PAGE_SHIFT,
939 		 (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1);
940 	memory_notify(MEM_CANCEL_ONLINE, &arg);
941 	mem_hotplug_done();
942 	return ret;
943 }
944 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
945 
946 static void reset_node_present_pages(pg_data_t *pgdat)
947 {
948 	struct zone *z;
949 
950 	for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
951 		z->present_pages = 0;
952 
953 	pgdat->node_present_pages = 0;
954 }
955 
956 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
957 static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
958 {
959 	struct pglist_data *pgdat;
960 	unsigned long start_pfn = PFN_DOWN(start);
961 
962 	pgdat = NODE_DATA(nid);
963 	if (!pgdat) {
964 		pgdat = arch_alloc_nodedata(nid);
965 		if (!pgdat)
966 			return NULL;
967 
968 		arch_refresh_nodedata(nid, pgdat);
969 	} else {
970 		/*
971 		 * Reset the nr_zones, order and classzone_idx before reuse.
972 		 * Note that kswapd will init kswapd_classzone_idx properly
973 		 * when it starts in the near future.
974 		 */
975 		pgdat->nr_zones = 0;
976 		pgdat->kswapd_order = 0;
977 		pgdat->kswapd_classzone_idx = 0;
978 	}
979 
980 	/* we can use NODE_DATA(nid) from here */
981 
982 	pgdat->node_id = nid;
983 	pgdat->node_start_pfn = start_pfn;
984 
985 	/* init node's zones as empty zones, we don't have any present pages.*/
986 	free_area_init_core_hotplug(nid);
987 	pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat);
988 
989 	/*
990 	 * The node we allocated has no zone fallback lists. For avoiding
991 	 * to access not-initialized zonelist, build here.
992 	 */
993 	build_all_zonelists(pgdat);
994 
995 	/*
996 	 * When memory is hot-added, all the memory is in offline state. So
997 	 * clear all zones' present_pages because they will be updated in
998 	 * online_pages() and offline_pages().
999 	 */
1000 	reset_node_managed_pages(pgdat);
1001 	reset_node_present_pages(pgdat);
1002 
1003 	return pgdat;
1004 }
1005 
1006 static void rollback_node_hotadd(int nid)
1007 {
1008 	pg_data_t *pgdat = NODE_DATA(nid);
1009 
1010 	arch_refresh_nodedata(nid, NULL);
1011 	free_percpu(pgdat->per_cpu_nodestats);
1012 	arch_free_nodedata(pgdat);
1013 	return;
1014 }
1015 
1016 
1017 /**
1018  * try_online_node - online a node if offlined
1019  * @nid: the node ID
1020  * @start: start addr of the node
1021  * @set_node_online: Whether we want to online the node
1022  * called by cpu_up() to online a node without onlined memory.
1023  *
1024  * Returns:
1025  * 1 -> a new node has been allocated
1026  * 0 -> the node is already online
1027  * -ENOMEM -> the node could not be allocated
1028  */
1029 static int __try_online_node(int nid, u64 start, bool set_node_online)
1030 {
1031 	pg_data_t *pgdat;
1032 	int ret = 1;
1033 
1034 	if (node_online(nid))
1035 		return 0;
1036 
1037 	pgdat = hotadd_new_pgdat(nid, start);
1038 	if (!pgdat) {
1039 		pr_err("Cannot online node %d due to NULL pgdat\n", nid);
1040 		ret = -ENOMEM;
1041 		goto out;
1042 	}
1043 
1044 	if (set_node_online) {
1045 		node_set_online(nid);
1046 		ret = register_one_node(nid);
1047 		BUG_ON(ret);
1048 	}
1049 out:
1050 	return ret;
1051 }
1052 
1053 /*
1054  * Users of this function always want to online/register the node
1055  */
1056 int try_online_node(int nid)
1057 {
1058 	int ret;
1059 
1060 	mem_hotplug_begin();
1061 	ret =  __try_online_node(nid, 0, true);
1062 	mem_hotplug_done();
1063 	return ret;
1064 }
1065 
1066 static int check_hotplug_memory_range(u64 start, u64 size)
1067 {
1068 	unsigned long block_sz = memory_block_size_bytes();
1069 	u64 block_nr_pages = block_sz >> PAGE_SHIFT;
1070 	u64 nr_pages = size >> PAGE_SHIFT;
1071 	u64 start_pfn = PFN_DOWN(start);
1072 
1073 	/* memory range must be block size aligned */
1074 	if (!nr_pages || !IS_ALIGNED(start_pfn, block_nr_pages) ||
1075 	    !IS_ALIGNED(nr_pages, block_nr_pages)) {
1076 		pr_err("Block size [%#lx] unaligned hotplug range: start %#llx, size %#llx",
1077 		       block_sz, start, size);
1078 		return -EINVAL;
1079 	}
1080 
1081 	return 0;
1082 }
1083 
1084 static int online_memory_block(struct memory_block *mem, void *arg)
1085 {
1086 	return device_online(&mem->dev);
1087 }
1088 
1089 /*
1090  * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1091  * and online/offline operations (triggered e.g. by sysfs).
1092  *
1093  * we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG
1094  */
1095 int __ref add_memory_resource(int nid, struct resource *res)
1096 {
1097 	u64 start, size;
1098 	bool new_node = false;
1099 	int ret;
1100 
1101 	start = res->start;
1102 	size = resource_size(res);
1103 
1104 	ret = check_hotplug_memory_range(start, size);
1105 	if (ret)
1106 		return ret;
1107 
1108 	mem_hotplug_begin();
1109 
1110 	/*
1111 	 * Add new range to memblock so that when hotadd_new_pgdat() is called
1112 	 * to allocate new pgdat, get_pfn_range_for_nid() will be able to find
1113 	 * this new range and calculate total pages correctly.  The range will
1114 	 * be removed at hot-remove time.
1115 	 */
1116 	memblock_add_node(start, size, nid);
1117 
1118 	ret = __try_online_node(nid, start, false);
1119 	if (ret < 0)
1120 		goto error;
1121 	new_node = ret;
1122 
1123 	/* call arch's memory hotadd */
1124 	ret = arch_add_memory(nid, start, size, NULL, true);
1125 	if (ret < 0)
1126 		goto error;
1127 
1128 	if (new_node) {
1129 		/* If sysfs file of new node can't be created, cpu on the node
1130 		 * can't be hot-added. There is no rollback way now.
1131 		 * So, check by BUG_ON() to catch it reluctantly..
1132 		 * We online node here. We can't roll back from here.
1133 		 */
1134 		node_set_online(nid);
1135 		ret = __register_one_node(nid);
1136 		BUG_ON(ret);
1137 	}
1138 
1139 	/* link memory sections under this node.*/
1140 	ret = link_mem_sections(nid, PFN_DOWN(start), PFN_UP(start + size - 1));
1141 	BUG_ON(ret);
1142 
1143 	/* create new memmap entry */
1144 	firmware_map_add_hotplug(start, start + size, "System RAM");
1145 
1146 	/* device_online() will take the lock when calling online_pages() */
1147 	mem_hotplug_done();
1148 
1149 	/* online pages if requested */
1150 	if (memhp_auto_online)
1151 		walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1),
1152 				  NULL, online_memory_block);
1153 
1154 	return ret;
1155 error:
1156 	/* rollback pgdat allocation and others */
1157 	if (new_node)
1158 		rollback_node_hotadd(nid);
1159 	memblock_remove(start, size);
1160 	mem_hotplug_done();
1161 	return ret;
1162 }
1163 
1164 /* requires device_hotplug_lock, see add_memory_resource() */
1165 int __ref __add_memory(int nid, u64 start, u64 size)
1166 {
1167 	struct resource *res;
1168 	int ret;
1169 
1170 	res = register_memory_resource(start, size);
1171 	if (IS_ERR(res))
1172 		return PTR_ERR(res);
1173 
1174 	ret = add_memory_resource(nid, res);
1175 	if (ret < 0)
1176 		release_memory_resource(res);
1177 	return ret;
1178 }
1179 
1180 int add_memory(int nid, u64 start, u64 size)
1181 {
1182 	int rc;
1183 
1184 	lock_device_hotplug();
1185 	rc = __add_memory(nid, start, size);
1186 	unlock_device_hotplug();
1187 
1188 	return rc;
1189 }
1190 EXPORT_SYMBOL_GPL(add_memory);
1191 
1192 #ifdef CONFIG_MEMORY_HOTREMOVE
1193 /*
1194  * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1195  * set and the size of the free page is given by page_order(). Using this,
1196  * the function determines if the pageblock contains only free pages.
1197  * Due to buddy contraints, a free page at least the size of a pageblock will
1198  * be located at the start of the pageblock
1199  */
1200 static inline int pageblock_free(struct page *page)
1201 {
1202 	return PageBuddy(page) && page_order(page) >= pageblock_order;
1203 }
1204 
1205 /* Return the pfn of the start of the next active pageblock after a given pfn */
1206 static unsigned long next_active_pageblock(unsigned long pfn)
1207 {
1208 	struct page *page = pfn_to_page(pfn);
1209 
1210 	/* Ensure the starting page is pageblock-aligned */
1211 	BUG_ON(pfn & (pageblock_nr_pages - 1));
1212 
1213 	/* If the entire pageblock is free, move to the end of free page */
1214 	if (pageblock_free(page)) {
1215 		int order;
1216 		/* be careful. we don't have locks, page_order can be changed.*/
1217 		order = page_order(page);
1218 		if ((order < MAX_ORDER) && (order >= pageblock_order))
1219 			return pfn + (1 << order);
1220 	}
1221 
1222 	return pfn + pageblock_nr_pages;
1223 }
1224 
1225 static bool is_pageblock_removable_nolock(unsigned long pfn)
1226 {
1227 	struct page *page = pfn_to_page(pfn);
1228 	struct zone *zone;
1229 
1230 	/*
1231 	 * We have to be careful here because we are iterating over memory
1232 	 * sections which are not zone aware so we might end up outside of
1233 	 * the zone but still within the section.
1234 	 * We have to take care about the node as well. If the node is offline
1235 	 * its NODE_DATA will be NULL - see page_zone.
1236 	 */
1237 	if (!node_online(page_to_nid(page)))
1238 		return false;
1239 
1240 	zone = page_zone(page);
1241 	pfn = page_to_pfn(page);
1242 	if (!zone_spans_pfn(zone, pfn))
1243 		return false;
1244 
1245 	return !has_unmovable_pages(zone, page, 0, MIGRATE_MOVABLE, SKIP_HWPOISON);
1246 }
1247 
1248 /* Checks if this range of memory is likely to be hot-removable. */
1249 bool is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
1250 {
1251 	unsigned long end_pfn, pfn;
1252 
1253 	end_pfn = min(start_pfn + nr_pages,
1254 			zone_end_pfn(page_zone(pfn_to_page(start_pfn))));
1255 
1256 	/* Check the starting page of each pageblock within the range */
1257 	for (pfn = start_pfn; pfn < end_pfn; pfn = next_active_pageblock(pfn)) {
1258 		if (!is_pageblock_removable_nolock(pfn))
1259 			return false;
1260 		cond_resched();
1261 	}
1262 
1263 	/* All pageblocks in the memory block are likely to be hot-removable */
1264 	return true;
1265 }
1266 
1267 /*
1268  * Confirm all pages in a range [start, end) belong to the same zone.
1269  * When true, return its valid [start, end).
1270  */
1271 int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn,
1272 			 unsigned long *valid_start, unsigned long *valid_end)
1273 {
1274 	unsigned long pfn, sec_end_pfn;
1275 	unsigned long start, end;
1276 	struct zone *zone = NULL;
1277 	struct page *page;
1278 	int i;
1279 	for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1);
1280 	     pfn < end_pfn;
1281 	     pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) {
1282 		/* Make sure the memory section is present first */
1283 		if (!present_section_nr(pfn_to_section_nr(pfn)))
1284 			continue;
1285 		for (; pfn < sec_end_pfn && pfn < end_pfn;
1286 		     pfn += MAX_ORDER_NR_PAGES) {
1287 			i = 0;
1288 			/* This is just a CONFIG_HOLES_IN_ZONE check.*/
1289 			while ((i < MAX_ORDER_NR_PAGES) &&
1290 				!pfn_valid_within(pfn + i))
1291 				i++;
1292 			if (i == MAX_ORDER_NR_PAGES || pfn + i >= end_pfn)
1293 				continue;
1294 			/* Check if we got outside of the zone */
1295 			if (zone && !zone_spans_pfn(zone, pfn + i))
1296 				return 0;
1297 			page = pfn_to_page(pfn + i);
1298 			if (zone && page_zone(page) != zone)
1299 				return 0;
1300 			if (!zone)
1301 				start = pfn + i;
1302 			zone = page_zone(page);
1303 			end = pfn + MAX_ORDER_NR_PAGES;
1304 		}
1305 	}
1306 
1307 	if (zone) {
1308 		*valid_start = start;
1309 		*valid_end = min(end, end_pfn);
1310 		return 1;
1311 	} else {
1312 		return 0;
1313 	}
1314 }
1315 
1316 /*
1317  * Scan pfn range [start,end) to find movable/migratable pages (LRU pages,
1318  * non-lru movable pages and hugepages). We scan pfn because it's much
1319  * easier than scanning over linked list. This function returns the pfn
1320  * of the first found movable page if it's found, otherwise 0.
1321  */
1322 static unsigned long scan_movable_pages(unsigned long start, unsigned long end)
1323 {
1324 	unsigned long pfn;
1325 
1326 	for (pfn = start; pfn < end; pfn++) {
1327 		struct page *page, *head;
1328 		unsigned long skip;
1329 
1330 		if (!pfn_valid(pfn))
1331 			continue;
1332 		page = pfn_to_page(pfn);
1333 		if (PageLRU(page))
1334 			return pfn;
1335 		if (__PageMovable(page))
1336 			return pfn;
1337 
1338 		if (!PageHuge(page))
1339 			continue;
1340 		head = compound_head(page);
1341 		if (hugepage_migration_supported(page_hstate(head)) &&
1342 		    page_huge_active(head))
1343 			return pfn;
1344 		skip = (1 << compound_order(head)) - (page - head);
1345 		pfn += skip - 1;
1346 	}
1347 	return 0;
1348 }
1349 
1350 static struct page *new_node_page(struct page *page, unsigned long private)
1351 {
1352 	int nid = page_to_nid(page);
1353 	nodemask_t nmask = node_states[N_MEMORY];
1354 
1355 	/*
1356 	 * try to allocate from a different node but reuse this node if there
1357 	 * are no other online nodes to be used (e.g. we are offlining a part
1358 	 * of the only existing node)
1359 	 */
1360 	node_clear(nid, nmask);
1361 	if (nodes_empty(nmask))
1362 		node_set(nid, nmask);
1363 
1364 	return new_page_nodemask(page, nid, &nmask);
1365 }
1366 
1367 static int
1368 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1369 {
1370 	unsigned long pfn;
1371 	struct page *page;
1372 	int ret = 0;
1373 	LIST_HEAD(source);
1374 
1375 	for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1376 		if (!pfn_valid(pfn))
1377 			continue;
1378 		page = pfn_to_page(pfn);
1379 
1380 		if (PageHuge(page)) {
1381 			struct page *head = compound_head(page);
1382 			if (compound_order(head) > PFN_SECTION_SHIFT) {
1383 				ret = -EBUSY;
1384 				break;
1385 			}
1386 			pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1;
1387 			isolate_huge_page(head, &source);
1388 			continue;
1389 		} else if (PageTransHuge(page))
1390 			pfn = page_to_pfn(compound_head(page))
1391 				+ hpage_nr_pages(page) - 1;
1392 
1393 		/*
1394 		 * HWPoison pages have elevated reference counts so the migration would
1395 		 * fail on them. It also doesn't make any sense to migrate them in the
1396 		 * first place. Still try to unmap such a page in case it is still mapped
1397 		 * (e.g. current hwpoison implementation doesn't unmap KSM pages but keep
1398 		 * the unmap as the catch all safety net).
1399 		 */
1400 		if (PageHWPoison(page)) {
1401 			if (WARN_ON(PageLRU(page)))
1402 				isolate_lru_page(page);
1403 			if (page_mapped(page))
1404 				try_to_unmap(page, TTU_IGNORE_MLOCK | TTU_IGNORE_ACCESS);
1405 			continue;
1406 		}
1407 
1408 		if (!get_page_unless_zero(page))
1409 			continue;
1410 		/*
1411 		 * We can skip free pages. And we can deal with pages on
1412 		 * LRU and non-lru movable pages.
1413 		 */
1414 		if (PageLRU(page))
1415 			ret = isolate_lru_page(page);
1416 		else
1417 			ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE);
1418 		if (!ret) { /* Success */
1419 			list_add_tail(&page->lru, &source);
1420 			if (!__PageMovable(page))
1421 				inc_node_page_state(page, NR_ISOLATED_ANON +
1422 						    page_is_file_cache(page));
1423 
1424 		} else {
1425 			pr_warn("failed to isolate pfn %lx\n", pfn);
1426 			dump_page(page, "isolation failed");
1427 		}
1428 		put_page(page);
1429 	}
1430 	if (!list_empty(&source)) {
1431 		/* Allocate a new page from the nearest neighbor node */
1432 		ret = migrate_pages(&source, new_node_page, NULL, 0,
1433 					MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
1434 		if (ret) {
1435 			list_for_each_entry(page, &source, lru) {
1436 				pr_warn("migrating pfn %lx failed ret:%d ",
1437 				       page_to_pfn(page), ret);
1438 				dump_page(page, "migration failure");
1439 			}
1440 			putback_movable_pages(&source);
1441 		}
1442 	}
1443 
1444 	return ret;
1445 }
1446 
1447 /*
1448  * remove from free_area[] and mark all as Reserved.
1449  */
1450 static int
1451 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1452 			void *data)
1453 {
1454 	__offline_isolated_pages(start, start + nr_pages);
1455 	return 0;
1456 }
1457 
1458 static void
1459 offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
1460 {
1461 	walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
1462 				offline_isolated_pages_cb);
1463 }
1464 
1465 /*
1466  * Check all pages in range, recoreded as memory resource, are isolated.
1467  */
1468 static int
1469 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1470 			void *data)
1471 {
1472 	int ret;
1473 	long offlined = *(long *)data;
1474 	ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true);
1475 	offlined = nr_pages;
1476 	if (!ret)
1477 		*(long *)data += offlined;
1478 	return ret;
1479 }
1480 
1481 static long
1482 check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
1483 {
1484 	long offlined = 0;
1485 	int ret;
1486 
1487 	ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
1488 			check_pages_isolated_cb);
1489 	if (ret < 0)
1490 		offlined = (long)ret;
1491 	return offlined;
1492 }
1493 
1494 static int __init cmdline_parse_movable_node(char *p)
1495 {
1496 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1497 	movable_node_enabled = true;
1498 #else
1499 	pr_warn("movable_node parameter depends on CONFIG_HAVE_MEMBLOCK_NODE_MAP to work properly\n");
1500 #endif
1501 	return 0;
1502 }
1503 early_param("movable_node", cmdline_parse_movable_node);
1504 
1505 /* check which state of node_states will be changed when offline memory */
1506 static void node_states_check_changes_offline(unsigned long nr_pages,
1507 		struct zone *zone, struct memory_notify *arg)
1508 {
1509 	struct pglist_data *pgdat = zone->zone_pgdat;
1510 	unsigned long present_pages = 0;
1511 	enum zone_type zt;
1512 
1513 	arg->status_change_nid = NUMA_NO_NODE;
1514 	arg->status_change_nid_normal = NUMA_NO_NODE;
1515 	arg->status_change_nid_high = NUMA_NO_NODE;
1516 
1517 	/*
1518 	 * Check whether node_states[N_NORMAL_MEMORY] will be changed.
1519 	 * If the memory to be offline is within the range
1520 	 * [0..ZONE_NORMAL], and it is the last present memory there,
1521 	 * the zones in that range will become empty after the offlining,
1522 	 * thus we can determine that we need to clear the node from
1523 	 * node_states[N_NORMAL_MEMORY].
1524 	 */
1525 	for (zt = 0; zt <= ZONE_NORMAL; zt++)
1526 		present_pages += pgdat->node_zones[zt].present_pages;
1527 	if (zone_idx(zone) <= ZONE_NORMAL && nr_pages >= present_pages)
1528 		arg->status_change_nid_normal = zone_to_nid(zone);
1529 
1530 #ifdef CONFIG_HIGHMEM
1531 	/*
1532 	 * node_states[N_HIGH_MEMORY] contains nodes which
1533 	 * have normal memory or high memory.
1534 	 * Here we add the present_pages belonging to ZONE_HIGHMEM.
1535 	 * If the zone is within the range of [0..ZONE_HIGHMEM), and
1536 	 * we determine that the zones in that range become empty,
1537 	 * we need to clear the node for N_HIGH_MEMORY.
1538 	 */
1539 	present_pages += pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1540 	if (zone_idx(zone) <= ZONE_HIGHMEM && nr_pages >= present_pages)
1541 		arg->status_change_nid_high = zone_to_nid(zone);
1542 #endif
1543 
1544 	/*
1545 	 * We have accounted the pages from [0..ZONE_NORMAL), and
1546 	 * in case of CONFIG_HIGHMEM the pages from ZONE_HIGHMEM
1547 	 * as well.
1548 	 * Here we count the possible pages from ZONE_MOVABLE.
1549 	 * If after having accounted all the pages, we see that the nr_pages
1550 	 * to be offlined is over or equal to the accounted pages,
1551 	 * we know that the node will become empty, and so, we can clear
1552 	 * it for N_MEMORY as well.
1553 	 */
1554 	present_pages += pgdat->node_zones[ZONE_MOVABLE].present_pages;
1555 
1556 	if (nr_pages >= present_pages)
1557 		arg->status_change_nid = zone_to_nid(zone);
1558 }
1559 
1560 static void node_states_clear_node(int node, struct memory_notify *arg)
1561 {
1562 	if (arg->status_change_nid_normal >= 0)
1563 		node_clear_state(node, N_NORMAL_MEMORY);
1564 
1565 	if (arg->status_change_nid_high >= 0)
1566 		node_clear_state(node, N_HIGH_MEMORY);
1567 
1568 	if (arg->status_change_nid >= 0)
1569 		node_clear_state(node, N_MEMORY);
1570 }
1571 
1572 static int __ref __offline_pages(unsigned long start_pfn,
1573 		  unsigned long end_pfn)
1574 {
1575 	unsigned long pfn, nr_pages;
1576 	long offlined_pages;
1577 	int ret, node;
1578 	unsigned long flags;
1579 	unsigned long valid_start, valid_end;
1580 	struct zone *zone;
1581 	struct memory_notify arg;
1582 	char *reason;
1583 
1584 	mem_hotplug_begin();
1585 
1586 	/* This makes hotplug much easier...and readable.
1587 	   we assume this for now. .*/
1588 	if (!test_pages_in_a_zone(start_pfn, end_pfn, &valid_start,
1589 				  &valid_end)) {
1590 		ret = -EINVAL;
1591 		reason = "multizone range";
1592 		goto failed_removal;
1593 	}
1594 
1595 	zone = page_zone(pfn_to_page(valid_start));
1596 	node = zone_to_nid(zone);
1597 	nr_pages = end_pfn - start_pfn;
1598 
1599 	/* set above range as isolated */
1600 	ret = start_isolate_page_range(start_pfn, end_pfn,
1601 				       MIGRATE_MOVABLE,
1602 				       SKIP_HWPOISON | REPORT_FAILURE);
1603 	if (ret) {
1604 		reason = "failure to isolate range";
1605 		goto failed_removal;
1606 	}
1607 
1608 	arg.start_pfn = start_pfn;
1609 	arg.nr_pages = nr_pages;
1610 	node_states_check_changes_offline(nr_pages, zone, &arg);
1611 
1612 	ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1613 	ret = notifier_to_errno(ret);
1614 	if (ret) {
1615 		reason = "notifier failure";
1616 		goto failed_removal_isolated;
1617 	}
1618 
1619 	do {
1620 		for (pfn = start_pfn; pfn;) {
1621 			if (signal_pending(current)) {
1622 				ret = -EINTR;
1623 				reason = "signal backoff";
1624 				goto failed_removal_isolated;
1625 			}
1626 
1627 			cond_resched();
1628 			lru_add_drain_all();
1629 
1630 			pfn = scan_movable_pages(pfn, end_pfn);
1631 			if (pfn) {
1632 				/*
1633 				 * TODO: fatal migration failures should bail
1634 				 * out
1635 				 */
1636 				do_migrate_range(pfn, end_pfn);
1637 			}
1638 		}
1639 
1640 		/*
1641 		 * Dissolve free hugepages in the memory block before doing
1642 		 * offlining actually in order to make hugetlbfs's object
1643 		 * counting consistent.
1644 		 */
1645 		ret = dissolve_free_huge_pages(start_pfn, end_pfn);
1646 		if (ret) {
1647 			reason = "failure to dissolve huge pages";
1648 			goto failed_removal_isolated;
1649 		}
1650 		/* check again */
1651 		offlined_pages = check_pages_isolated(start_pfn, end_pfn);
1652 	} while (offlined_pages < 0);
1653 
1654 	pr_info("Offlined Pages %ld\n", offlined_pages);
1655 	/* Ok, all of our target is isolated.
1656 	   We cannot do rollback at this point. */
1657 	offline_isolated_pages(start_pfn, end_pfn);
1658 	/* reset pagetype flags and makes migrate type to be MOVABLE */
1659 	undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1660 	/* removal success */
1661 	adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages);
1662 	zone->present_pages -= offlined_pages;
1663 
1664 	pgdat_resize_lock(zone->zone_pgdat, &flags);
1665 	zone->zone_pgdat->node_present_pages -= offlined_pages;
1666 	pgdat_resize_unlock(zone->zone_pgdat, &flags);
1667 
1668 	init_per_zone_wmark_min();
1669 
1670 	if (!populated_zone(zone)) {
1671 		zone_pcp_reset(zone);
1672 		build_all_zonelists(NULL);
1673 	} else
1674 		zone_pcp_update(zone);
1675 
1676 	node_states_clear_node(node, &arg);
1677 	if (arg.status_change_nid >= 0) {
1678 		kswapd_stop(node);
1679 		kcompactd_stop(node);
1680 	}
1681 
1682 	vm_total_pages = nr_free_pagecache_pages();
1683 	writeback_set_ratelimit();
1684 
1685 	memory_notify(MEM_OFFLINE, &arg);
1686 	mem_hotplug_done();
1687 	return 0;
1688 
1689 failed_removal_isolated:
1690 	undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1691 failed_removal:
1692 	pr_debug("memory offlining [mem %#010llx-%#010llx] failed due to %s\n",
1693 		 (unsigned long long) start_pfn << PAGE_SHIFT,
1694 		 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1,
1695 		 reason);
1696 	memory_notify(MEM_CANCEL_OFFLINE, &arg);
1697 	/* pushback to free area */
1698 	mem_hotplug_done();
1699 	return ret;
1700 }
1701 
1702 int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1703 {
1704 	return __offline_pages(start_pfn, start_pfn + nr_pages);
1705 }
1706 #endif /* CONFIG_MEMORY_HOTREMOVE */
1707 
1708 /**
1709  * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
1710  * @start_pfn: start pfn of the memory range
1711  * @end_pfn: end pfn of the memory range
1712  * @arg: argument passed to func
1713  * @func: callback for each memory section walked
1714  *
1715  * This function walks through all present mem sections in range
1716  * [start_pfn, end_pfn) and call func on each mem section.
1717  *
1718  * Returns the return value of func.
1719  */
1720 int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn,
1721 		void *arg, int (*func)(struct memory_block *, void *))
1722 {
1723 	struct memory_block *mem = NULL;
1724 	struct mem_section *section;
1725 	unsigned long pfn, section_nr;
1726 	int ret;
1727 
1728 	for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1729 		section_nr = pfn_to_section_nr(pfn);
1730 		if (!present_section_nr(section_nr))
1731 			continue;
1732 
1733 		section = __nr_to_section(section_nr);
1734 		/* same memblock? */
1735 		if (mem)
1736 			if ((section_nr >= mem->start_section_nr) &&
1737 			    (section_nr <= mem->end_section_nr))
1738 				continue;
1739 
1740 		mem = find_memory_block_hinted(section, mem);
1741 		if (!mem)
1742 			continue;
1743 
1744 		ret = func(mem, arg);
1745 		if (ret) {
1746 			kobject_put(&mem->dev.kobj);
1747 			return ret;
1748 		}
1749 	}
1750 
1751 	if (mem)
1752 		kobject_put(&mem->dev.kobj);
1753 
1754 	return 0;
1755 }
1756 
1757 #ifdef CONFIG_MEMORY_HOTREMOVE
1758 static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
1759 {
1760 	int ret = !is_memblock_offlined(mem);
1761 
1762 	if (unlikely(ret)) {
1763 		phys_addr_t beginpa, endpa;
1764 
1765 		beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
1766 		endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1;
1767 		pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n",
1768 			&beginpa, &endpa);
1769 	}
1770 
1771 	return ret;
1772 }
1773 
1774 static int check_cpu_on_node(pg_data_t *pgdat)
1775 {
1776 	int cpu;
1777 
1778 	for_each_present_cpu(cpu) {
1779 		if (cpu_to_node(cpu) == pgdat->node_id)
1780 			/*
1781 			 * the cpu on this node isn't removed, and we can't
1782 			 * offline this node.
1783 			 */
1784 			return -EBUSY;
1785 	}
1786 
1787 	return 0;
1788 }
1789 
1790 /**
1791  * try_offline_node
1792  * @nid: the node ID
1793  *
1794  * Offline a node if all memory sections and cpus of the node are removed.
1795  *
1796  * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1797  * and online/offline operations before this call.
1798  */
1799 void try_offline_node(int nid)
1800 {
1801 	pg_data_t *pgdat = NODE_DATA(nid);
1802 	unsigned long start_pfn = pgdat->node_start_pfn;
1803 	unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
1804 	unsigned long pfn;
1805 
1806 	for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1807 		unsigned long section_nr = pfn_to_section_nr(pfn);
1808 
1809 		if (!present_section_nr(section_nr))
1810 			continue;
1811 
1812 		if (pfn_to_nid(pfn) != nid)
1813 			continue;
1814 
1815 		/*
1816 		 * some memory sections of this node are not removed, and we
1817 		 * can't offline node now.
1818 		 */
1819 		return;
1820 	}
1821 
1822 	if (check_cpu_on_node(pgdat))
1823 		return;
1824 
1825 	/*
1826 	 * all memory/cpu of this node are removed, we can offline this
1827 	 * node now.
1828 	 */
1829 	node_set_offline(nid);
1830 	unregister_one_node(nid);
1831 }
1832 EXPORT_SYMBOL(try_offline_node);
1833 
1834 /**
1835  * remove_memory
1836  * @nid: the node ID
1837  * @start: physical address of the region to remove
1838  * @size: size of the region to remove
1839  *
1840  * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1841  * and online/offline operations before this call, as required by
1842  * try_offline_node().
1843  */
1844 void __ref __remove_memory(int nid, u64 start, u64 size)
1845 {
1846 	int ret;
1847 
1848 	BUG_ON(check_hotplug_memory_range(start, size));
1849 
1850 	mem_hotplug_begin();
1851 
1852 	/*
1853 	 * All memory blocks must be offlined before removing memory.  Check
1854 	 * whether all memory blocks in question are offline and trigger a BUG()
1855 	 * if this is not the case.
1856 	 */
1857 	ret = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL,
1858 				check_memblock_offlined_cb);
1859 	if (ret)
1860 		BUG();
1861 
1862 	/* remove memmap entry */
1863 	firmware_map_remove(start, start + size, "System RAM");
1864 	memblock_free(start, size);
1865 	memblock_remove(start, size);
1866 
1867 	arch_remove_memory(nid, start, size, NULL);
1868 
1869 	try_offline_node(nid);
1870 
1871 	mem_hotplug_done();
1872 }
1873 
1874 void remove_memory(int nid, u64 start, u64 size)
1875 {
1876 	lock_device_hotplug();
1877 	__remove_memory(nid, start, size);
1878 	unlock_device_hotplug();
1879 }
1880 EXPORT_SYMBOL_GPL(remove_memory);
1881 #endif /* CONFIG_MEMORY_HOTREMOVE */
1882