xref: /openbmc/linux/mm/memory_hotplug.c (revision 4cff79e9)
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/bootmem.h>
37 #include <linux/compaction.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);
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(NODE_DATA(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 		ret = __remove_section(zone, __pfn_to_section(pfn), map_offset,
591 				altmap);
592 		map_offset = 0;
593 		if (ret)
594 			break;
595 	}
596 
597 	set_zone_contiguous(zone);
598 
599 	return ret;
600 }
601 #endif /* CONFIG_MEMORY_HOTREMOVE */
602 
603 int set_online_page_callback(online_page_callback_t callback)
604 {
605 	int rc = -EINVAL;
606 
607 	get_online_mems();
608 	mutex_lock(&online_page_callback_lock);
609 
610 	if (online_page_callback == generic_online_page) {
611 		online_page_callback = callback;
612 		rc = 0;
613 	}
614 
615 	mutex_unlock(&online_page_callback_lock);
616 	put_online_mems();
617 
618 	return rc;
619 }
620 EXPORT_SYMBOL_GPL(set_online_page_callback);
621 
622 int restore_online_page_callback(online_page_callback_t callback)
623 {
624 	int rc = -EINVAL;
625 
626 	get_online_mems();
627 	mutex_lock(&online_page_callback_lock);
628 
629 	if (online_page_callback == callback) {
630 		online_page_callback = generic_online_page;
631 		rc = 0;
632 	}
633 
634 	mutex_unlock(&online_page_callback_lock);
635 	put_online_mems();
636 
637 	return rc;
638 }
639 EXPORT_SYMBOL_GPL(restore_online_page_callback);
640 
641 void __online_page_set_limits(struct page *page)
642 {
643 }
644 EXPORT_SYMBOL_GPL(__online_page_set_limits);
645 
646 void __online_page_increment_counters(struct page *page)
647 {
648 	adjust_managed_page_count(page, 1);
649 }
650 EXPORT_SYMBOL_GPL(__online_page_increment_counters);
651 
652 void __online_page_free(struct page *page)
653 {
654 	__free_reserved_page(page);
655 }
656 EXPORT_SYMBOL_GPL(__online_page_free);
657 
658 static void generic_online_page(struct page *page)
659 {
660 	__online_page_set_limits(page);
661 	__online_page_increment_counters(page);
662 	__online_page_free(page);
663 }
664 
665 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
666 			void *arg)
667 {
668 	unsigned long i;
669 	unsigned long onlined_pages = *(unsigned long *)arg;
670 	struct page *page;
671 
672 	if (PageReserved(pfn_to_page(start_pfn)))
673 		for (i = 0; i < nr_pages; i++) {
674 			page = pfn_to_page(start_pfn + i);
675 			(*online_page_callback)(page);
676 			onlined_pages++;
677 		}
678 
679 	online_mem_sections(start_pfn, start_pfn + nr_pages);
680 
681 	*(unsigned long *)arg = onlined_pages;
682 	return 0;
683 }
684 
685 /* check which state of node_states will be changed when online memory */
686 static void node_states_check_changes_online(unsigned long nr_pages,
687 	struct zone *zone, struct memory_notify *arg)
688 {
689 	int nid = zone_to_nid(zone);
690 	enum zone_type zone_last = ZONE_NORMAL;
691 
692 	/*
693 	 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
694 	 * contains nodes which have zones of 0...ZONE_NORMAL,
695 	 * set zone_last to ZONE_NORMAL.
696 	 *
697 	 * If we don't have HIGHMEM nor movable node,
698 	 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
699 	 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
700 	 */
701 	if (N_MEMORY == N_NORMAL_MEMORY)
702 		zone_last = ZONE_MOVABLE;
703 
704 	/*
705 	 * if the memory to be online is in a zone of 0...zone_last, and
706 	 * the zones of 0...zone_last don't have memory before online, we will
707 	 * need to set the node to node_states[N_NORMAL_MEMORY] after
708 	 * the memory is online.
709 	 */
710 	if (zone_idx(zone) <= zone_last && !node_state(nid, N_NORMAL_MEMORY))
711 		arg->status_change_nid_normal = nid;
712 	else
713 		arg->status_change_nid_normal = -1;
714 
715 #ifdef CONFIG_HIGHMEM
716 	/*
717 	 * If we have movable node, node_states[N_HIGH_MEMORY]
718 	 * contains nodes which have zones of 0...ZONE_HIGHMEM,
719 	 * set zone_last to ZONE_HIGHMEM.
720 	 *
721 	 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
722 	 * contains nodes which have zones of 0...ZONE_MOVABLE,
723 	 * set zone_last to ZONE_MOVABLE.
724 	 */
725 	zone_last = ZONE_HIGHMEM;
726 	if (N_MEMORY == N_HIGH_MEMORY)
727 		zone_last = ZONE_MOVABLE;
728 
729 	if (zone_idx(zone) <= zone_last && !node_state(nid, N_HIGH_MEMORY))
730 		arg->status_change_nid_high = nid;
731 	else
732 		arg->status_change_nid_high = -1;
733 #else
734 	arg->status_change_nid_high = arg->status_change_nid_normal;
735 #endif
736 
737 	/*
738 	 * if the node don't have memory befor online, we will need to
739 	 * set the node to node_states[N_MEMORY] after the memory
740 	 * is online.
741 	 */
742 	if (!node_state(nid, N_MEMORY))
743 		arg->status_change_nid = nid;
744 	else
745 		arg->status_change_nid = -1;
746 }
747 
748 static void node_states_set_node(int node, struct memory_notify *arg)
749 {
750 	if (arg->status_change_nid_normal >= 0)
751 		node_set_state(node, N_NORMAL_MEMORY);
752 
753 	if (arg->status_change_nid_high >= 0)
754 		node_set_state(node, N_HIGH_MEMORY);
755 
756 	node_set_state(node, N_MEMORY);
757 }
758 
759 static void __meminit resize_zone_range(struct zone *zone, unsigned long start_pfn,
760 		unsigned long nr_pages)
761 {
762 	unsigned long old_end_pfn = zone_end_pfn(zone);
763 
764 	if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
765 		zone->zone_start_pfn = start_pfn;
766 
767 	zone->spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - zone->zone_start_pfn;
768 }
769 
770 static void __meminit resize_pgdat_range(struct pglist_data *pgdat, unsigned long start_pfn,
771                                      unsigned long nr_pages)
772 {
773 	unsigned long old_end_pfn = pgdat_end_pfn(pgdat);
774 
775 	if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
776 		pgdat->node_start_pfn = start_pfn;
777 
778 	pgdat->node_spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - pgdat->node_start_pfn;
779 }
780 
781 void __ref move_pfn_range_to_zone(struct zone *zone, unsigned long start_pfn,
782 		unsigned long nr_pages, struct vmem_altmap *altmap)
783 {
784 	struct pglist_data *pgdat = zone->zone_pgdat;
785 	int nid = pgdat->node_id;
786 	unsigned long flags;
787 
788 	if (zone_is_empty(zone))
789 		init_currently_empty_zone(zone, start_pfn, nr_pages);
790 
791 	clear_zone_contiguous(zone);
792 
793 	/* TODO Huh pgdat is irqsave while zone is not. It used to be like that before */
794 	pgdat_resize_lock(pgdat, &flags);
795 	zone_span_writelock(zone);
796 	resize_zone_range(zone, start_pfn, nr_pages);
797 	zone_span_writeunlock(zone);
798 	resize_pgdat_range(pgdat, start_pfn, nr_pages);
799 	pgdat_resize_unlock(pgdat, &flags);
800 
801 	/*
802 	 * TODO now we have a visible range of pages which are not associated
803 	 * with their zone properly. Not nice but set_pfnblock_flags_mask
804 	 * expects the zone spans the pfn range. All the pages in the range
805 	 * are reserved so nobody should be touching them so we should be safe
806 	 */
807 	memmap_init_zone(nr_pages, nid, zone_idx(zone), start_pfn,
808 			MEMMAP_HOTPLUG, altmap);
809 
810 	set_zone_contiguous(zone);
811 }
812 
813 /*
814  * Returns a default kernel memory zone for the given pfn range.
815  * If no kernel zone covers this pfn range it will automatically go
816  * to the ZONE_NORMAL.
817  */
818 static struct zone *default_kernel_zone_for_pfn(int nid, unsigned long start_pfn,
819 		unsigned long nr_pages)
820 {
821 	struct pglist_data *pgdat = NODE_DATA(nid);
822 	int zid;
823 
824 	for (zid = 0; zid <= ZONE_NORMAL; zid++) {
825 		struct zone *zone = &pgdat->node_zones[zid];
826 
827 		if (zone_intersects(zone, start_pfn, nr_pages))
828 			return zone;
829 	}
830 
831 	return &pgdat->node_zones[ZONE_NORMAL];
832 }
833 
834 static inline struct zone *default_zone_for_pfn(int nid, unsigned long start_pfn,
835 		unsigned long nr_pages)
836 {
837 	struct zone *kernel_zone = default_kernel_zone_for_pfn(nid, start_pfn,
838 			nr_pages);
839 	struct zone *movable_zone = &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
840 	bool in_kernel = zone_intersects(kernel_zone, start_pfn, nr_pages);
841 	bool in_movable = zone_intersects(movable_zone, start_pfn, nr_pages);
842 
843 	/*
844 	 * We inherit the existing zone in a simple case where zones do not
845 	 * overlap in the given range
846 	 */
847 	if (in_kernel ^ in_movable)
848 		return (in_kernel) ? kernel_zone : movable_zone;
849 
850 	/*
851 	 * If the range doesn't belong to any zone or two zones overlap in the
852 	 * given range then we use movable zone only if movable_node is
853 	 * enabled because we always online to a kernel zone by default.
854 	 */
855 	return movable_node_enabled ? movable_zone : kernel_zone;
856 }
857 
858 struct zone * zone_for_pfn_range(int online_type, int nid, unsigned start_pfn,
859 		unsigned long nr_pages)
860 {
861 	if (online_type == MMOP_ONLINE_KERNEL)
862 		return default_kernel_zone_for_pfn(nid, start_pfn, nr_pages);
863 
864 	if (online_type == MMOP_ONLINE_MOVABLE)
865 		return &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
866 
867 	return default_zone_for_pfn(nid, start_pfn, nr_pages);
868 }
869 
870 /*
871  * Associates the given pfn range with the given node and the zone appropriate
872  * for the given online type.
873  */
874 static struct zone * __meminit move_pfn_range(int online_type, int nid,
875 		unsigned long start_pfn, unsigned long nr_pages)
876 {
877 	struct zone *zone;
878 
879 	zone = zone_for_pfn_range(online_type, nid, start_pfn, nr_pages);
880 	move_pfn_range_to_zone(zone, start_pfn, nr_pages, NULL);
881 	return zone;
882 }
883 
884 /* Must be protected by mem_hotplug_begin() or a device_lock */
885 int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type)
886 {
887 	unsigned long flags;
888 	unsigned long onlined_pages = 0;
889 	struct zone *zone;
890 	int need_zonelists_rebuild = 0;
891 	int nid;
892 	int ret;
893 	struct memory_notify arg;
894 	struct memory_block *mem;
895 
896 	/*
897 	 * We can't use pfn_to_nid() because nid might be stored in struct page
898 	 * which is not yet initialized. Instead, we find nid from memory block.
899 	 */
900 	mem = find_memory_block(__pfn_to_section(pfn));
901 	nid = mem->nid;
902 
903 	/* associate pfn range with the zone */
904 	zone = move_pfn_range(online_type, nid, pfn, nr_pages);
905 
906 	arg.start_pfn = pfn;
907 	arg.nr_pages = nr_pages;
908 	node_states_check_changes_online(nr_pages, zone, &arg);
909 
910 	ret = memory_notify(MEM_GOING_ONLINE, &arg);
911 	ret = notifier_to_errno(ret);
912 	if (ret)
913 		goto failed_addition;
914 
915 	/*
916 	 * If this zone is not populated, then it is not in zonelist.
917 	 * This means the page allocator ignores this zone.
918 	 * So, zonelist must be updated after online.
919 	 */
920 	if (!populated_zone(zone)) {
921 		need_zonelists_rebuild = 1;
922 		setup_zone_pageset(zone);
923 	}
924 
925 	ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
926 		online_pages_range);
927 	if (ret) {
928 		if (need_zonelists_rebuild)
929 			zone_pcp_reset(zone);
930 		goto failed_addition;
931 	}
932 
933 	zone->present_pages += onlined_pages;
934 
935 	pgdat_resize_lock(zone->zone_pgdat, &flags);
936 	zone->zone_pgdat->node_present_pages += onlined_pages;
937 	pgdat_resize_unlock(zone->zone_pgdat, &flags);
938 
939 	if (onlined_pages) {
940 		node_states_set_node(nid, &arg);
941 		if (need_zonelists_rebuild)
942 			build_all_zonelists(NULL);
943 		else
944 			zone_pcp_update(zone);
945 	}
946 
947 	init_per_zone_wmark_min();
948 
949 	if (onlined_pages) {
950 		kswapd_run(nid);
951 		kcompactd_run(nid);
952 	}
953 
954 	vm_total_pages = nr_free_pagecache_pages();
955 
956 	writeback_set_ratelimit();
957 
958 	if (onlined_pages)
959 		memory_notify(MEM_ONLINE, &arg);
960 	return 0;
961 
962 failed_addition:
963 	pr_debug("online_pages [mem %#010llx-%#010llx] failed\n",
964 		 (unsigned long long) pfn << PAGE_SHIFT,
965 		 (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1);
966 	memory_notify(MEM_CANCEL_ONLINE, &arg);
967 	return ret;
968 }
969 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
970 
971 static void reset_node_present_pages(pg_data_t *pgdat)
972 {
973 	struct zone *z;
974 
975 	for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
976 		z->present_pages = 0;
977 
978 	pgdat->node_present_pages = 0;
979 }
980 
981 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
982 static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
983 {
984 	struct pglist_data *pgdat;
985 	unsigned long zones_size[MAX_NR_ZONES] = {0};
986 	unsigned long zholes_size[MAX_NR_ZONES] = {0};
987 	unsigned long start_pfn = PFN_DOWN(start);
988 
989 	pgdat = NODE_DATA(nid);
990 	if (!pgdat) {
991 		pgdat = arch_alloc_nodedata(nid);
992 		if (!pgdat)
993 			return NULL;
994 
995 		arch_refresh_nodedata(nid, pgdat);
996 	} else {
997 		/*
998 		 * Reset the nr_zones, order and classzone_idx before reuse.
999 		 * Note that kswapd will init kswapd_classzone_idx properly
1000 		 * when it starts in the near future.
1001 		 */
1002 		pgdat->nr_zones = 0;
1003 		pgdat->kswapd_order = 0;
1004 		pgdat->kswapd_classzone_idx = 0;
1005 	}
1006 
1007 	/* we can use NODE_DATA(nid) from here */
1008 
1009 	/* init node's zones as empty zones, we don't have any present pages.*/
1010 	free_area_init_node(nid, zones_size, start_pfn, zholes_size);
1011 	pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat);
1012 
1013 	/*
1014 	 * The node we allocated has no zone fallback lists. For avoiding
1015 	 * to access not-initialized zonelist, build here.
1016 	 */
1017 	build_all_zonelists(pgdat);
1018 
1019 	/*
1020 	 * zone->managed_pages is set to an approximate value in
1021 	 * free_area_init_core(), which will cause
1022 	 * /sys/device/system/node/nodeX/meminfo has wrong data.
1023 	 * So reset it to 0 before any memory is onlined.
1024 	 */
1025 	reset_node_managed_pages(pgdat);
1026 
1027 	/*
1028 	 * When memory is hot-added, all the memory is in offline state. So
1029 	 * clear all zones' present_pages because they will be updated in
1030 	 * online_pages() and offline_pages().
1031 	 */
1032 	reset_node_present_pages(pgdat);
1033 
1034 	return pgdat;
1035 }
1036 
1037 static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
1038 {
1039 	arch_refresh_nodedata(nid, NULL);
1040 	free_percpu(pgdat->per_cpu_nodestats);
1041 	arch_free_nodedata(pgdat);
1042 	return;
1043 }
1044 
1045 
1046 /**
1047  * try_online_node - online a node if offlined
1048  * @nid: the node ID
1049  *
1050  * called by cpu_up() to online a node without onlined memory.
1051  */
1052 int try_online_node(int nid)
1053 {
1054 	pg_data_t	*pgdat;
1055 	int	ret;
1056 
1057 	if (node_online(nid))
1058 		return 0;
1059 
1060 	mem_hotplug_begin();
1061 	pgdat = hotadd_new_pgdat(nid, 0);
1062 	if (!pgdat) {
1063 		pr_err("Cannot online node %d due to NULL pgdat\n", nid);
1064 		ret = -ENOMEM;
1065 		goto out;
1066 	}
1067 	node_set_online(nid);
1068 	ret = register_one_node(nid);
1069 	BUG_ON(ret);
1070 out:
1071 	mem_hotplug_done();
1072 	return ret;
1073 }
1074 
1075 static int check_hotplug_memory_range(u64 start, u64 size)
1076 {
1077 	unsigned long block_sz = memory_block_size_bytes();
1078 	u64 block_nr_pages = block_sz >> PAGE_SHIFT;
1079 	u64 nr_pages = size >> PAGE_SHIFT;
1080 	u64 start_pfn = PFN_DOWN(start);
1081 
1082 	/* memory range must be block size aligned */
1083 	if (!nr_pages || !IS_ALIGNED(start_pfn, block_nr_pages) ||
1084 	    !IS_ALIGNED(nr_pages, block_nr_pages)) {
1085 		pr_err("Block size [%#lx] unaligned hotplug range: start %#llx, size %#llx",
1086 		       block_sz, start, size);
1087 		return -EINVAL;
1088 	}
1089 
1090 	return 0;
1091 }
1092 
1093 static int online_memory_block(struct memory_block *mem, void *arg)
1094 {
1095 	return device_online(&mem->dev);
1096 }
1097 
1098 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1099 int __ref add_memory_resource(int nid, struct resource *res, bool online)
1100 {
1101 	u64 start, size;
1102 	pg_data_t *pgdat = NULL;
1103 	bool new_pgdat;
1104 	bool new_node;
1105 	int ret;
1106 
1107 	start = res->start;
1108 	size = resource_size(res);
1109 
1110 	ret = check_hotplug_memory_range(start, size);
1111 	if (ret)
1112 		return ret;
1113 
1114 	{	/* Stupid hack to suppress address-never-null warning */
1115 		void *p = NODE_DATA(nid);
1116 		new_pgdat = !p;
1117 	}
1118 
1119 	mem_hotplug_begin();
1120 
1121 	/*
1122 	 * Add new range to memblock so that when hotadd_new_pgdat() is called
1123 	 * to allocate new pgdat, get_pfn_range_for_nid() will be able to find
1124 	 * this new range and calculate total pages correctly.  The range will
1125 	 * be removed at hot-remove time.
1126 	 */
1127 	memblock_add_node(start, size, nid);
1128 
1129 	new_node = !node_online(nid);
1130 	if (new_node) {
1131 		pgdat = hotadd_new_pgdat(nid, start);
1132 		ret = -ENOMEM;
1133 		if (!pgdat)
1134 			goto error;
1135 	}
1136 
1137 	/* call arch's memory hotadd */
1138 	ret = arch_add_memory(nid, start, size, NULL, true);
1139 
1140 	if (ret < 0)
1141 		goto error;
1142 
1143 	/* we online node here. we can't roll back from here. */
1144 	node_set_online(nid);
1145 
1146 	if (new_node) {
1147 		unsigned long start_pfn = start >> PAGE_SHIFT;
1148 		unsigned long nr_pages = size >> PAGE_SHIFT;
1149 
1150 		ret = __register_one_node(nid);
1151 		if (ret)
1152 			goto register_fail;
1153 
1154 		/*
1155 		 * link memory sections under this node. This is already
1156 		 * done when creatig memory section in register_new_memory
1157 		 * but that depends to have the node registered so offline
1158 		 * nodes have to go through register_node.
1159 		 * TODO clean up this mess.
1160 		 */
1161 		ret = link_mem_sections(nid, start_pfn, nr_pages);
1162 register_fail:
1163 		/*
1164 		 * If sysfs file of new node can't create, cpu on the node
1165 		 * can't be hot-added. There is no rollback way now.
1166 		 * So, check by BUG_ON() to catch it reluctantly..
1167 		 */
1168 		BUG_ON(ret);
1169 	}
1170 
1171 	/* create new memmap entry */
1172 	firmware_map_add_hotplug(start, start + size, "System RAM");
1173 
1174 	/* online pages if requested */
1175 	if (online)
1176 		walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1),
1177 				  NULL, online_memory_block);
1178 
1179 	goto out;
1180 
1181 error:
1182 	/* rollback pgdat allocation and others */
1183 	if (new_pgdat && pgdat)
1184 		rollback_node_hotadd(nid, pgdat);
1185 	memblock_remove(start, size);
1186 
1187 out:
1188 	mem_hotplug_done();
1189 	return ret;
1190 }
1191 EXPORT_SYMBOL_GPL(add_memory_resource);
1192 
1193 int __ref add_memory(int nid, u64 start, u64 size)
1194 {
1195 	struct resource *res;
1196 	int ret;
1197 
1198 	res = register_memory_resource(start, size);
1199 	if (IS_ERR(res))
1200 		return PTR_ERR(res);
1201 
1202 	ret = add_memory_resource(nid, res, memhp_auto_online);
1203 	if (ret < 0)
1204 		release_memory_resource(res);
1205 	return ret;
1206 }
1207 EXPORT_SYMBOL_GPL(add_memory);
1208 
1209 #ifdef CONFIG_MEMORY_HOTREMOVE
1210 /*
1211  * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1212  * set and the size of the free page is given by page_order(). Using this,
1213  * the function determines if the pageblock contains only free pages.
1214  * Due to buddy contraints, a free page at least the size of a pageblock will
1215  * be located at the start of the pageblock
1216  */
1217 static inline int pageblock_free(struct page *page)
1218 {
1219 	return PageBuddy(page) && page_order(page) >= pageblock_order;
1220 }
1221 
1222 /* Return the start of the next active pageblock after a given page */
1223 static struct page *next_active_pageblock(struct page *page)
1224 {
1225 	/* Ensure the starting page is pageblock-aligned */
1226 	BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
1227 
1228 	/* If the entire pageblock is free, move to the end of free page */
1229 	if (pageblock_free(page)) {
1230 		int order;
1231 		/* be careful. we don't have locks, page_order can be changed.*/
1232 		order = page_order(page);
1233 		if ((order < MAX_ORDER) && (order >= pageblock_order))
1234 			return page + (1 << order);
1235 	}
1236 
1237 	return page + pageblock_nr_pages;
1238 }
1239 
1240 /* Checks if this range of memory is likely to be hot-removable. */
1241 bool is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
1242 {
1243 	struct page *page = pfn_to_page(start_pfn);
1244 	struct page *end_page = page + nr_pages;
1245 
1246 	/* Check the starting page of each pageblock within the range */
1247 	for (; page < end_page; page = next_active_pageblock(page)) {
1248 		if (!is_pageblock_removable_nolock(page))
1249 			return false;
1250 		cond_resched();
1251 	}
1252 
1253 	/* All pageblocks in the memory block are likely to be hot-removable */
1254 	return true;
1255 }
1256 
1257 /*
1258  * Confirm all pages in a range [start, end) belong to the same zone.
1259  * When true, return its valid [start, end).
1260  */
1261 int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn,
1262 			 unsigned long *valid_start, unsigned long *valid_end)
1263 {
1264 	unsigned long pfn, sec_end_pfn;
1265 	unsigned long start, end;
1266 	struct zone *zone = NULL;
1267 	struct page *page;
1268 	int i;
1269 	for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1);
1270 	     pfn < end_pfn;
1271 	     pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) {
1272 		/* Make sure the memory section is present first */
1273 		if (!present_section_nr(pfn_to_section_nr(pfn)))
1274 			continue;
1275 		for (; pfn < sec_end_pfn && pfn < end_pfn;
1276 		     pfn += MAX_ORDER_NR_PAGES) {
1277 			i = 0;
1278 			/* This is just a CONFIG_HOLES_IN_ZONE check.*/
1279 			while ((i < MAX_ORDER_NR_PAGES) &&
1280 				!pfn_valid_within(pfn + i))
1281 				i++;
1282 			if (i == MAX_ORDER_NR_PAGES || pfn + i >= end_pfn)
1283 				continue;
1284 			page = pfn_to_page(pfn + i);
1285 			if (zone && page_zone(page) != zone)
1286 				return 0;
1287 			if (!zone)
1288 				start = pfn + i;
1289 			zone = page_zone(page);
1290 			end = pfn + MAX_ORDER_NR_PAGES;
1291 		}
1292 	}
1293 
1294 	if (zone) {
1295 		*valid_start = start;
1296 		*valid_end = min(end, end_pfn);
1297 		return 1;
1298 	} else {
1299 		return 0;
1300 	}
1301 }
1302 
1303 /*
1304  * Scan pfn range [start,end) to find movable/migratable pages (LRU pages,
1305  * non-lru movable pages and hugepages). We scan pfn because it's much
1306  * easier than scanning over linked list. This function returns the pfn
1307  * of the first found movable page if it's found, otherwise 0.
1308  */
1309 static unsigned long scan_movable_pages(unsigned long start, unsigned long end)
1310 {
1311 	unsigned long pfn;
1312 	struct page *page;
1313 	for (pfn = start; pfn < end; pfn++) {
1314 		if (pfn_valid(pfn)) {
1315 			page = pfn_to_page(pfn);
1316 			if (PageLRU(page))
1317 				return pfn;
1318 			if (__PageMovable(page))
1319 				return pfn;
1320 			if (PageHuge(page)) {
1321 				if (page_huge_active(page))
1322 					return pfn;
1323 				else
1324 					pfn = round_up(pfn + 1,
1325 						1 << compound_order(page)) - 1;
1326 			}
1327 		}
1328 	}
1329 	return 0;
1330 }
1331 
1332 static struct page *new_node_page(struct page *page, unsigned long private)
1333 {
1334 	int nid = page_to_nid(page);
1335 	nodemask_t nmask = node_states[N_MEMORY];
1336 
1337 	/*
1338 	 * try to allocate from a different node but reuse this node if there
1339 	 * are no other online nodes to be used (e.g. we are offlining a part
1340 	 * of the only existing node)
1341 	 */
1342 	node_clear(nid, nmask);
1343 	if (nodes_empty(nmask))
1344 		node_set(nid, nmask);
1345 
1346 	return new_page_nodemask(page, nid, &nmask);
1347 }
1348 
1349 #define NR_OFFLINE_AT_ONCE_PAGES	(256)
1350 static int
1351 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1352 {
1353 	unsigned long pfn;
1354 	struct page *page;
1355 	int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
1356 	int not_managed = 0;
1357 	int ret = 0;
1358 	LIST_HEAD(source);
1359 
1360 	for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
1361 		if (!pfn_valid(pfn))
1362 			continue;
1363 		page = pfn_to_page(pfn);
1364 
1365 		if (PageHuge(page)) {
1366 			struct page *head = compound_head(page);
1367 			pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1;
1368 			if (compound_order(head) > PFN_SECTION_SHIFT) {
1369 				ret = -EBUSY;
1370 				break;
1371 			}
1372 			if (isolate_huge_page(page, &source))
1373 				move_pages -= 1 << compound_order(head);
1374 			continue;
1375 		} else if (PageTransHuge(page))
1376 			pfn = page_to_pfn(compound_head(page))
1377 				+ hpage_nr_pages(page) - 1;
1378 
1379 		if (!get_page_unless_zero(page))
1380 			continue;
1381 		/*
1382 		 * We can skip free pages. And we can deal with pages on
1383 		 * LRU and non-lru movable pages.
1384 		 */
1385 		if (PageLRU(page))
1386 			ret = isolate_lru_page(page);
1387 		else
1388 			ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE);
1389 		if (!ret) { /* Success */
1390 			put_page(page);
1391 			list_add_tail(&page->lru, &source);
1392 			move_pages--;
1393 			if (!__PageMovable(page))
1394 				inc_node_page_state(page, NR_ISOLATED_ANON +
1395 						    page_is_file_cache(page));
1396 
1397 		} else {
1398 #ifdef CONFIG_DEBUG_VM
1399 			pr_alert("failed to isolate pfn %lx\n", pfn);
1400 			dump_page(page, "isolation failed");
1401 #endif
1402 			put_page(page);
1403 			/* Because we don't have big zone->lock. we should
1404 			   check this again here. */
1405 			if (page_count(page)) {
1406 				not_managed++;
1407 				ret = -EBUSY;
1408 				break;
1409 			}
1410 		}
1411 	}
1412 	if (!list_empty(&source)) {
1413 		if (not_managed) {
1414 			putback_movable_pages(&source);
1415 			goto out;
1416 		}
1417 
1418 		/* Allocate a new page from the nearest neighbor node */
1419 		ret = migrate_pages(&source, new_node_page, NULL, 0,
1420 					MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
1421 		if (ret)
1422 			putback_movable_pages(&source);
1423 	}
1424 out:
1425 	return ret;
1426 }
1427 
1428 /*
1429  * remove from free_area[] and mark all as Reserved.
1430  */
1431 static int
1432 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1433 			void *data)
1434 {
1435 	__offline_isolated_pages(start, start + nr_pages);
1436 	return 0;
1437 }
1438 
1439 static void
1440 offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
1441 {
1442 	walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
1443 				offline_isolated_pages_cb);
1444 }
1445 
1446 /*
1447  * Check all pages in range, recoreded as memory resource, are isolated.
1448  */
1449 static int
1450 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1451 			void *data)
1452 {
1453 	int ret;
1454 	long offlined = *(long *)data;
1455 	ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true);
1456 	offlined = nr_pages;
1457 	if (!ret)
1458 		*(long *)data += offlined;
1459 	return ret;
1460 }
1461 
1462 static long
1463 check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
1464 {
1465 	long offlined = 0;
1466 	int ret;
1467 
1468 	ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
1469 			check_pages_isolated_cb);
1470 	if (ret < 0)
1471 		offlined = (long)ret;
1472 	return offlined;
1473 }
1474 
1475 static int __init cmdline_parse_movable_node(char *p)
1476 {
1477 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1478 	movable_node_enabled = true;
1479 #else
1480 	pr_warn("movable_node parameter depends on CONFIG_HAVE_MEMBLOCK_NODE_MAP to work properly\n");
1481 #endif
1482 	return 0;
1483 }
1484 early_param("movable_node", cmdline_parse_movable_node);
1485 
1486 /* check which state of node_states will be changed when offline memory */
1487 static void node_states_check_changes_offline(unsigned long nr_pages,
1488 		struct zone *zone, struct memory_notify *arg)
1489 {
1490 	struct pglist_data *pgdat = zone->zone_pgdat;
1491 	unsigned long present_pages = 0;
1492 	enum zone_type zt, zone_last = ZONE_NORMAL;
1493 
1494 	/*
1495 	 * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
1496 	 * contains nodes which have zones of 0...ZONE_NORMAL,
1497 	 * set zone_last to ZONE_NORMAL.
1498 	 *
1499 	 * If we don't have HIGHMEM nor movable node,
1500 	 * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
1501 	 * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
1502 	 */
1503 	if (N_MEMORY == N_NORMAL_MEMORY)
1504 		zone_last = ZONE_MOVABLE;
1505 
1506 	/*
1507 	 * check whether node_states[N_NORMAL_MEMORY] will be changed.
1508 	 * If the memory to be offline is in a zone of 0...zone_last,
1509 	 * and it is the last present memory, 0...zone_last will
1510 	 * become empty after offline , thus we can determind we will
1511 	 * need to clear the node from node_states[N_NORMAL_MEMORY].
1512 	 */
1513 	for (zt = 0; zt <= zone_last; zt++)
1514 		present_pages += pgdat->node_zones[zt].present_pages;
1515 	if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1516 		arg->status_change_nid_normal = zone_to_nid(zone);
1517 	else
1518 		arg->status_change_nid_normal = -1;
1519 
1520 #ifdef CONFIG_HIGHMEM
1521 	/*
1522 	 * If we have movable node, node_states[N_HIGH_MEMORY]
1523 	 * contains nodes which have zones of 0...ZONE_HIGHMEM,
1524 	 * set zone_last to ZONE_HIGHMEM.
1525 	 *
1526 	 * If we don't have movable node, node_states[N_NORMAL_MEMORY]
1527 	 * contains nodes which have zones of 0...ZONE_MOVABLE,
1528 	 * set zone_last to ZONE_MOVABLE.
1529 	 */
1530 	zone_last = ZONE_HIGHMEM;
1531 	if (N_MEMORY == N_HIGH_MEMORY)
1532 		zone_last = ZONE_MOVABLE;
1533 
1534 	for (; zt <= zone_last; zt++)
1535 		present_pages += pgdat->node_zones[zt].present_pages;
1536 	if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1537 		arg->status_change_nid_high = zone_to_nid(zone);
1538 	else
1539 		arg->status_change_nid_high = -1;
1540 #else
1541 	arg->status_change_nid_high = arg->status_change_nid_normal;
1542 #endif
1543 
1544 	/*
1545 	 * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE
1546 	 */
1547 	zone_last = ZONE_MOVABLE;
1548 
1549 	/*
1550 	 * check whether node_states[N_HIGH_MEMORY] will be changed
1551 	 * If we try to offline the last present @nr_pages from the node,
1552 	 * we can determind we will need to clear the node from
1553 	 * node_states[N_HIGH_MEMORY].
1554 	 */
1555 	for (; zt <= zone_last; zt++)
1556 		present_pages += pgdat->node_zones[zt].present_pages;
1557 	if (nr_pages >= present_pages)
1558 		arg->status_change_nid = zone_to_nid(zone);
1559 	else
1560 		arg->status_change_nid = -1;
1561 }
1562 
1563 static void node_states_clear_node(int node, struct memory_notify *arg)
1564 {
1565 	if (arg->status_change_nid_normal >= 0)
1566 		node_clear_state(node, N_NORMAL_MEMORY);
1567 
1568 	if ((N_MEMORY != N_NORMAL_MEMORY) &&
1569 	    (arg->status_change_nid_high >= 0))
1570 		node_clear_state(node, N_HIGH_MEMORY);
1571 
1572 	if ((N_MEMORY != N_HIGH_MEMORY) &&
1573 	    (arg->status_change_nid >= 0))
1574 		node_clear_state(node, N_MEMORY);
1575 }
1576 
1577 static int __ref __offline_pages(unsigned long start_pfn,
1578 		  unsigned long end_pfn)
1579 {
1580 	unsigned long pfn, nr_pages;
1581 	long offlined_pages;
1582 	int ret, node;
1583 	unsigned long flags;
1584 	unsigned long valid_start, valid_end;
1585 	struct zone *zone;
1586 	struct memory_notify arg;
1587 
1588 	/* at least, alignment against pageblock is necessary */
1589 	if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
1590 		return -EINVAL;
1591 	if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
1592 		return -EINVAL;
1593 	/* This makes hotplug much easier...and readable.
1594 	   we assume this for now. .*/
1595 	if (!test_pages_in_a_zone(start_pfn, end_pfn, &valid_start, &valid_end))
1596 		return -EINVAL;
1597 
1598 	zone = page_zone(pfn_to_page(valid_start));
1599 	node = zone_to_nid(zone);
1600 	nr_pages = end_pfn - start_pfn;
1601 
1602 	/* set above range as isolated */
1603 	ret = start_isolate_page_range(start_pfn, end_pfn,
1604 				       MIGRATE_MOVABLE, true);
1605 	if (ret)
1606 		return ret;
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 		goto failed_removal;
1616 
1617 	pfn = start_pfn;
1618 repeat:
1619 	/* start memory hot removal */
1620 	ret = -EINTR;
1621 	if (signal_pending(current))
1622 		goto failed_removal;
1623 
1624 	cond_resched();
1625 	lru_add_drain_all();
1626 	drain_all_pages(zone);
1627 
1628 	pfn = scan_movable_pages(start_pfn, end_pfn);
1629 	if (pfn) { /* We have movable pages */
1630 		ret = do_migrate_range(pfn, end_pfn);
1631 		goto repeat;
1632 	}
1633 
1634 	/*
1635 	 * dissolve free hugepages in the memory block before doing offlining
1636 	 * actually in order to make hugetlbfs's object counting consistent.
1637 	 */
1638 	ret = dissolve_free_huge_pages(start_pfn, end_pfn);
1639 	if (ret)
1640 		goto failed_removal;
1641 	/* check again */
1642 	offlined_pages = check_pages_isolated(start_pfn, end_pfn);
1643 	if (offlined_pages < 0)
1644 		goto repeat;
1645 	pr_info("Offlined Pages %ld\n", offlined_pages);
1646 	/* Ok, all of our target is isolated.
1647 	   We cannot do rollback at this point. */
1648 	offline_isolated_pages(start_pfn, end_pfn);
1649 	/* reset pagetype flags and makes migrate type to be MOVABLE */
1650 	undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1651 	/* removal success */
1652 	adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages);
1653 	zone->present_pages -= offlined_pages;
1654 
1655 	pgdat_resize_lock(zone->zone_pgdat, &flags);
1656 	zone->zone_pgdat->node_present_pages -= offlined_pages;
1657 	pgdat_resize_unlock(zone->zone_pgdat, &flags);
1658 
1659 	init_per_zone_wmark_min();
1660 
1661 	if (!populated_zone(zone)) {
1662 		zone_pcp_reset(zone);
1663 		build_all_zonelists(NULL);
1664 	} else
1665 		zone_pcp_update(zone);
1666 
1667 	node_states_clear_node(node, &arg);
1668 	if (arg.status_change_nid >= 0) {
1669 		kswapd_stop(node);
1670 		kcompactd_stop(node);
1671 	}
1672 
1673 	vm_total_pages = nr_free_pagecache_pages();
1674 	writeback_set_ratelimit();
1675 
1676 	memory_notify(MEM_OFFLINE, &arg);
1677 	return 0;
1678 
1679 failed_removal:
1680 	pr_debug("memory offlining [mem %#010llx-%#010llx] failed\n",
1681 		 (unsigned long long) start_pfn << PAGE_SHIFT,
1682 		 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
1683 	memory_notify(MEM_CANCEL_OFFLINE, &arg);
1684 	/* pushback to free area */
1685 	undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1686 	return ret;
1687 }
1688 
1689 /* Must be protected by mem_hotplug_begin() or a device_lock */
1690 int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1691 {
1692 	return __offline_pages(start_pfn, start_pfn + nr_pages);
1693 }
1694 #endif /* CONFIG_MEMORY_HOTREMOVE */
1695 
1696 /**
1697  * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
1698  * @start_pfn: start pfn of the memory range
1699  * @end_pfn: end pfn of the memory range
1700  * @arg: argument passed to func
1701  * @func: callback for each memory section walked
1702  *
1703  * This function walks through all present mem sections in range
1704  * [start_pfn, end_pfn) and call func on each mem section.
1705  *
1706  * Returns the return value of func.
1707  */
1708 int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn,
1709 		void *arg, int (*func)(struct memory_block *, void *))
1710 {
1711 	struct memory_block *mem = NULL;
1712 	struct mem_section *section;
1713 	unsigned long pfn, section_nr;
1714 	int ret;
1715 
1716 	for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1717 		section_nr = pfn_to_section_nr(pfn);
1718 		if (!present_section_nr(section_nr))
1719 			continue;
1720 
1721 		section = __nr_to_section(section_nr);
1722 		/* same memblock? */
1723 		if (mem)
1724 			if ((section_nr >= mem->start_section_nr) &&
1725 			    (section_nr <= mem->end_section_nr))
1726 				continue;
1727 
1728 		mem = find_memory_block_hinted(section, mem);
1729 		if (!mem)
1730 			continue;
1731 
1732 		ret = func(mem, arg);
1733 		if (ret) {
1734 			kobject_put(&mem->dev.kobj);
1735 			return ret;
1736 		}
1737 	}
1738 
1739 	if (mem)
1740 		kobject_put(&mem->dev.kobj);
1741 
1742 	return 0;
1743 }
1744 
1745 #ifdef CONFIG_MEMORY_HOTREMOVE
1746 static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
1747 {
1748 	int ret = !is_memblock_offlined(mem);
1749 
1750 	if (unlikely(ret)) {
1751 		phys_addr_t beginpa, endpa;
1752 
1753 		beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
1754 		endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1;
1755 		pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n",
1756 			&beginpa, &endpa);
1757 	}
1758 
1759 	return ret;
1760 }
1761 
1762 static int check_cpu_on_node(pg_data_t *pgdat)
1763 {
1764 	int cpu;
1765 
1766 	for_each_present_cpu(cpu) {
1767 		if (cpu_to_node(cpu) == pgdat->node_id)
1768 			/*
1769 			 * the cpu on this node isn't removed, and we can't
1770 			 * offline this node.
1771 			 */
1772 			return -EBUSY;
1773 	}
1774 
1775 	return 0;
1776 }
1777 
1778 static void unmap_cpu_on_node(pg_data_t *pgdat)
1779 {
1780 #ifdef CONFIG_ACPI_NUMA
1781 	int cpu;
1782 
1783 	for_each_possible_cpu(cpu)
1784 		if (cpu_to_node(cpu) == pgdat->node_id)
1785 			numa_clear_node(cpu);
1786 #endif
1787 }
1788 
1789 static int check_and_unmap_cpu_on_node(pg_data_t *pgdat)
1790 {
1791 	int ret;
1792 
1793 	ret = check_cpu_on_node(pgdat);
1794 	if (ret)
1795 		return ret;
1796 
1797 	/*
1798 	 * the node will be offlined when we come here, so we can clear
1799 	 * the cpu_to_node() now.
1800 	 */
1801 
1802 	unmap_cpu_on_node(pgdat);
1803 	return 0;
1804 }
1805 
1806 /**
1807  * try_offline_node
1808  * @nid: the node ID
1809  *
1810  * Offline a node if all memory sections and cpus of the node are removed.
1811  *
1812  * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1813  * and online/offline operations before this call.
1814  */
1815 void try_offline_node(int nid)
1816 {
1817 	pg_data_t *pgdat = NODE_DATA(nid);
1818 	unsigned long start_pfn = pgdat->node_start_pfn;
1819 	unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
1820 	unsigned long pfn;
1821 
1822 	for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1823 		unsigned long section_nr = pfn_to_section_nr(pfn);
1824 
1825 		if (!present_section_nr(section_nr))
1826 			continue;
1827 
1828 		if (pfn_to_nid(pfn) != nid)
1829 			continue;
1830 
1831 		/*
1832 		 * some memory sections of this node are not removed, and we
1833 		 * can't offline node now.
1834 		 */
1835 		return;
1836 	}
1837 
1838 	if (check_and_unmap_cpu_on_node(pgdat))
1839 		return;
1840 
1841 	/*
1842 	 * all memory/cpu of this node are removed, we can offline this
1843 	 * node now.
1844 	 */
1845 	node_set_offline(nid);
1846 	unregister_one_node(nid);
1847 }
1848 EXPORT_SYMBOL(try_offline_node);
1849 
1850 /**
1851  * remove_memory
1852  * @nid: the node ID
1853  * @start: physical address of the region to remove
1854  * @size: size of the region to remove
1855  *
1856  * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1857  * and online/offline operations before this call, as required by
1858  * try_offline_node().
1859  */
1860 void __ref remove_memory(int nid, u64 start, u64 size)
1861 {
1862 	int ret;
1863 
1864 	BUG_ON(check_hotplug_memory_range(start, size));
1865 
1866 	mem_hotplug_begin();
1867 
1868 	/*
1869 	 * All memory blocks must be offlined before removing memory.  Check
1870 	 * whether all memory blocks in question are offline and trigger a BUG()
1871 	 * if this is not the case.
1872 	 */
1873 	ret = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL,
1874 				check_memblock_offlined_cb);
1875 	if (ret)
1876 		BUG();
1877 
1878 	/* remove memmap entry */
1879 	firmware_map_remove(start, start + size, "System RAM");
1880 	memblock_free(start, size);
1881 	memblock_remove(start, size);
1882 
1883 	arch_remove_memory(start, size, NULL);
1884 
1885 	try_offline_node(nid);
1886 
1887 	mem_hotplug_done();
1888 }
1889 EXPORT_SYMBOL_GPL(remove_memory);
1890 #endif /* CONFIG_MEMORY_HOTREMOVE */
1891