xref: /openbmc/linux/mm/memory_hotplug.c (revision 9b05158f5d805e0cf373f6e5a43efb9306bcb6a2)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *  linux/mm/memory_hotplug.c
4  *
5  *  Copyright (C)
6  */
7 
8 #include <linux/stddef.h>
9 #include <linux/mm.h>
10 #include <linux/sched/signal.h>
11 #include <linux/swap.h>
12 #include <linux/interrupt.h>
13 #include <linux/pagemap.h>
14 #include <linux/compiler.h>
15 #include <linux/export.h>
16 #include <linux/pagevec.h>
17 #include <linux/writeback.h>
18 #include <linux/slab.h>
19 #include <linux/sysctl.h>
20 #include <linux/cpu.h>
21 #include <linux/memory.h>
22 #include <linux/memremap.h>
23 #include <linux/memory_hotplug.h>
24 #include <linux/highmem.h>
25 #include <linux/vmalloc.h>
26 #include <linux/ioport.h>
27 #include <linux/delay.h>
28 #include <linux/migrate.h>
29 #include <linux/page-isolation.h>
30 #include <linux/pfn.h>
31 #include <linux/suspend.h>
32 #include <linux/mm_inline.h>
33 #include <linux/firmware-map.h>
34 #include <linux/stop_machine.h>
35 #include <linux/hugetlb.h>
36 #include <linux/memblock.h>
37 #include <linux/compaction.h>
38 #include <linux/rmap.h>
39 
40 #include <asm/tlbflush.h>
41 
42 #include "internal.h"
43 #include "shuffle.h"
44 
45 /*
46  * online_page_callback contains pointer to current page onlining function.
47  * Initially it is generic_online_page(). If it is required it could be
48  * changed by calling set_online_page_callback() for callback registration
49  * and restore_online_page_callback() for generic callback restore.
50  */
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 u64 max_mem_size = U64_MAX;
100 
101 /* add this memory to iomem resource */
102 static struct resource *register_memory_resource(u64 start, u64 size)
103 {
104 	struct resource *res;
105 	unsigned long flags =  IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
106 	char *resource_name = "System RAM";
107 
108 	if (start + size > max_mem_size)
109 		return ERR_PTR(-E2BIG);
110 
111 	/*
112 	 * Request ownership of the new memory range.  This might be
113 	 * a child of an existing resource that was present but
114 	 * not marked as busy.
115 	 */
116 	res = __request_region(&iomem_resource, start, size,
117 			       resource_name, flags);
118 
119 	if (!res) {
120 		pr_debug("Unable to reserve System RAM region: %016llx->%016llx\n",
121 				start, start + size);
122 		return ERR_PTR(-EEXIST);
123 	}
124 	return res;
125 }
126 
127 static void release_memory_resource(struct resource *res)
128 {
129 	if (!res)
130 		return;
131 	release_resource(res);
132 	kfree(res);
133 }
134 
135 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
136 void get_page_bootmem(unsigned long info,  struct page *page,
137 		      unsigned long type)
138 {
139 	page->freelist = (void *)type;
140 	SetPagePrivate(page);
141 	set_page_private(page, info);
142 	page_ref_inc(page);
143 }
144 
145 void put_page_bootmem(struct page *page)
146 {
147 	unsigned long type;
148 
149 	type = (unsigned long) page->freelist;
150 	BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE ||
151 	       type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE);
152 
153 	if (page_ref_dec_return(page) == 1) {
154 		page->freelist = NULL;
155 		ClearPagePrivate(page);
156 		set_page_private(page, 0);
157 		INIT_LIST_HEAD(&page->lru);
158 		free_reserved_page(page);
159 	}
160 }
161 
162 #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
163 #ifndef CONFIG_SPARSEMEM_VMEMMAP
164 static void register_page_bootmem_info_section(unsigned long start_pfn)
165 {
166 	unsigned long mapsize, section_nr, i;
167 	struct mem_section *ms;
168 	struct page *page, *memmap;
169 	struct mem_section_usage *usage;
170 
171 	section_nr = pfn_to_section_nr(start_pfn);
172 	ms = __nr_to_section(section_nr);
173 
174 	/* Get section's memmap address */
175 	memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
176 
177 	/*
178 	 * Get page for the memmap's phys address
179 	 * XXX: need more consideration for sparse_vmemmap...
180 	 */
181 	page = virt_to_page(memmap);
182 	mapsize = sizeof(struct page) * PAGES_PER_SECTION;
183 	mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
184 
185 	/* remember memmap's page */
186 	for (i = 0; i < mapsize; i++, page++)
187 		get_page_bootmem(section_nr, page, SECTION_INFO);
188 
189 	usage = ms->usage;
190 	page = virt_to_page(usage);
191 
192 	mapsize = PAGE_ALIGN(mem_section_usage_size()) >> PAGE_SHIFT;
193 
194 	for (i = 0; i < mapsize; i++, page++)
195 		get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
196 
197 }
198 #else /* CONFIG_SPARSEMEM_VMEMMAP */
199 static void register_page_bootmem_info_section(unsigned long start_pfn)
200 {
201 	unsigned long mapsize, section_nr, i;
202 	struct mem_section *ms;
203 	struct page *page, *memmap;
204 	struct mem_section_usage *usage;
205 
206 	section_nr = pfn_to_section_nr(start_pfn);
207 	ms = __nr_to_section(section_nr);
208 
209 	memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
210 
211 	register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION);
212 
213 	usage = ms->usage;
214 	page = virt_to_page(usage);
215 
216 	mapsize = PAGE_ALIGN(mem_section_usage_size()) >> PAGE_SHIFT;
217 
218 	for (i = 0; i < mapsize; i++, page++)
219 		get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
220 }
221 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
222 
223 void __init register_page_bootmem_info_node(struct pglist_data *pgdat)
224 {
225 	unsigned long i, pfn, end_pfn, nr_pages;
226 	int node = pgdat->node_id;
227 	struct page *page;
228 
229 	nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
230 	page = virt_to_page(pgdat);
231 
232 	for (i = 0; i < nr_pages; i++, page++)
233 		get_page_bootmem(node, page, NODE_INFO);
234 
235 	pfn = pgdat->node_start_pfn;
236 	end_pfn = pgdat_end_pfn(pgdat);
237 
238 	/* register section info */
239 	for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
240 		/*
241 		 * Some platforms can assign the same pfn to multiple nodes - on
242 		 * node0 as well as nodeN.  To avoid registering a pfn against
243 		 * multiple nodes we check that this pfn does not already
244 		 * reside in some other nodes.
245 		 */
246 		if (pfn_valid(pfn) && (early_pfn_to_nid(pfn) == node))
247 			register_page_bootmem_info_section(pfn);
248 	}
249 }
250 #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
251 
252 static int check_pfn_span(unsigned long pfn, unsigned long nr_pages,
253 		const char *reason)
254 {
255 	/*
256 	 * Disallow all operations smaller than a sub-section and only
257 	 * allow operations smaller than a section for
258 	 * SPARSEMEM_VMEMMAP. Note that check_hotplug_memory_range()
259 	 * enforces a larger memory_block_size_bytes() granularity for
260 	 * memory that will be marked online, so this check should only
261 	 * fire for direct arch_{add,remove}_memory() users outside of
262 	 * add_memory_resource().
263 	 */
264 	unsigned long min_align;
265 
266 	if (IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP))
267 		min_align = PAGES_PER_SUBSECTION;
268 	else
269 		min_align = PAGES_PER_SECTION;
270 	if (!IS_ALIGNED(pfn, min_align)
271 			|| !IS_ALIGNED(nr_pages, min_align)) {
272 		WARN(1, "Misaligned __%s_pages start: %#lx end: #%lx\n",
273 				reason, pfn, pfn + nr_pages - 1);
274 		return -EINVAL;
275 	}
276 	return 0;
277 }
278 
279 static int check_hotplug_memory_addressable(unsigned long pfn,
280 					    unsigned long nr_pages)
281 {
282 	const u64 max_addr = PFN_PHYS(pfn + nr_pages) - 1;
283 
284 	if (max_addr >> MAX_PHYSMEM_BITS) {
285 		const u64 max_allowed = (1ull << (MAX_PHYSMEM_BITS + 1)) - 1;
286 		WARN(1,
287 		     "Hotplugged memory exceeds maximum addressable address, range=%#llx-%#llx, maximum=%#llx\n",
288 		     (u64)PFN_PHYS(pfn), max_addr, max_allowed);
289 		return -E2BIG;
290 	}
291 
292 	return 0;
293 }
294 
295 /*
296  * Reasonably generic function for adding memory.  It is
297  * expected that archs that support memory hotplug will
298  * call this function after deciding the zone to which to
299  * add the new pages.
300  */
301 int __ref __add_pages(int nid, unsigned long pfn, unsigned long nr_pages,
302 		struct mhp_restrictions *restrictions)
303 {
304 	int err;
305 	unsigned long nr, start_sec, end_sec;
306 	struct vmem_altmap *altmap = restrictions->altmap;
307 
308 	err = check_hotplug_memory_addressable(pfn, nr_pages);
309 	if (err)
310 		return err;
311 
312 	if (altmap) {
313 		/*
314 		 * Validate altmap is within bounds of the total request
315 		 */
316 		if (altmap->base_pfn != pfn
317 				|| vmem_altmap_offset(altmap) > nr_pages) {
318 			pr_warn_once("memory add fail, invalid altmap\n");
319 			return -EINVAL;
320 		}
321 		altmap->alloc = 0;
322 	}
323 
324 	err = check_pfn_span(pfn, nr_pages, "add");
325 	if (err)
326 		return err;
327 
328 	start_sec = pfn_to_section_nr(pfn);
329 	end_sec = pfn_to_section_nr(pfn + nr_pages - 1);
330 	for (nr = start_sec; nr <= end_sec; nr++) {
331 		unsigned long pfns;
332 
333 		pfns = min(nr_pages, PAGES_PER_SECTION
334 				- (pfn & ~PAGE_SECTION_MASK));
335 		err = sparse_add_section(nid, pfn, pfns, altmap);
336 		if (err)
337 			break;
338 		pfn += pfns;
339 		nr_pages -= pfns;
340 		cond_resched();
341 	}
342 	vmemmap_populate_print_last();
343 	return err;
344 }
345 
346 /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
347 static unsigned long find_smallest_section_pfn(int nid, struct zone *zone,
348 				     unsigned long start_pfn,
349 				     unsigned long end_pfn)
350 {
351 	for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SUBSECTION) {
352 		if (unlikely(!pfn_to_online_page(start_pfn)))
353 			continue;
354 
355 		if (unlikely(pfn_to_nid(start_pfn) != nid))
356 			continue;
357 
358 		if (zone != page_zone(pfn_to_page(start_pfn)))
359 			continue;
360 
361 		return start_pfn;
362 	}
363 
364 	return 0;
365 }
366 
367 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
368 static unsigned long find_biggest_section_pfn(int nid, struct zone *zone,
369 				    unsigned long start_pfn,
370 				    unsigned long end_pfn)
371 {
372 	unsigned long pfn;
373 
374 	/* pfn is the end pfn of a memory section. */
375 	pfn = end_pfn - 1;
376 	for (; pfn >= start_pfn; pfn -= PAGES_PER_SUBSECTION) {
377 		if (unlikely(!pfn_to_online_page(pfn)))
378 			continue;
379 
380 		if (unlikely(pfn_to_nid(pfn) != nid))
381 			continue;
382 
383 		if (zone != page_zone(pfn_to_page(pfn)))
384 			continue;
385 
386 		return pfn;
387 	}
388 
389 	return 0;
390 }
391 
392 static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
393 			     unsigned long end_pfn)
394 {
395 	unsigned long zone_start_pfn = zone->zone_start_pfn;
396 	unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */
397 	unsigned long zone_end_pfn = z;
398 	unsigned long pfn;
399 	int nid = zone_to_nid(zone);
400 
401 	zone_span_writelock(zone);
402 	if (zone_start_pfn == start_pfn) {
403 		/*
404 		 * If the section is smallest section in the zone, it need
405 		 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
406 		 * In this case, we find second smallest valid mem_section
407 		 * for shrinking zone.
408 		 */
409 		pfn = find_smallest_section_pfn(nid, zone, end_pfn,
410 						zone_end_pfn);
411 		if (pfn) {
412 			zone->zone_start_pfn = pfn;
413 			zone->spanned_pages = zone_end_pfn - pfn;
414 		}
415 	} else if (zone_end_pfn == end_pfn) {
416 		/*
417 		 * If the section is biggest section in the zone, it need
418 		 * shrink zone->spanned_pages.
419 		 * In this case, we find second biggest valid mem_section for
420 		 * shrinking zone.
421 		 */
422 		pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn,
423 					       start_pfn);
424 		if (pfn)
425 			zone->spanned_pages = pfn - zone_start_pfn + 1;
426 	}
427 
428 	/*
429 	 * The section is not biggest or smallest mem_section in the zone, it
430 	 * only creates a hole in the zone. So in this case, we need not
431 	 * change the zone. But perhaps, the zone has only hole data. Thus
432 	 * it check the zone has only hole or not.
433 	 */
434 	pfn = zone_start_pfn;
435 	for (; pfn < zone_end_pfn; pfn += PAGES_PER_SUBSECTION) {
436 		if (unlikely(!pfn_to_online_page(pfn)))
437 			continue;
438 
439 		if (page_zone(pfn_to_page(pfn)) != zone)
440 			continue;
441 
442 		/* Skip range to be removed */
443 		if (pfn >= start_pfn && pfn < end_pfn)
444 			continue;
445 
446 		/* If we find valid section, we have nothing to do */
447 		zone_span_writeunlock(zone);
448 		return;
449 	}
450 
451 	/* The zone has no valid section */
452 	zone->zone_start_pfn = 0;
453 	zone->spanned_pages = 0;
454 	zone_span_writeunlock(zone);
455 }
456 
457 static void update_pgdat_span(struct pglist_data *pgdat)
458 {
459 	unsigned long node_start_pfn = 0, node_end_pfn = 0;
460 	struct zone *zone;
461 
462 	for (zone = pgdat->node_zones;
463 	     zone < pgdat->node_zones + MAX_NR_ZONES; zone++) {
464 		unsigned long zone_end_pfn = zone->zone_start_pfn +
465 					     zone->spanned_pages;
466 
467 		/* No need to lock the zones, they can't change. */
468 		if (!zone->spanned_pages)
469 			continue;
470 		if (!node_end_pfn) {
471 			node_start_pfn = zone->zone_start_pfn;
472 			node_end_pfn = zone_end_pfn;
473 			continue;
474 		}
475 
476 		if (zone_end_pfn > node_end_pfn)
477 			node_end_pfn = zone_end_pfn;
478 		if (zone->zone_start_pfn < node_start_pfn)
479 			node_start_pfn = zone->zone_start_pfn;
480 	}
481 
482 	pgdat->node_start_pfn = node_start_pfn;
483 	pgdat->node_spanned_pages = node_end_pfn - node_start_pfn;
484 }
485 
486 void __ref remove_pfn_range_from_zone(struct zone *zone,
487 				      unsigned long start_pfn,
488 				      unsigned long nr_pages)
489 {
490 	struct pglist_data *pgdat = zone->zone_pgdat;
491 	unsigned long flags;
492 
493 	/* Poison struct pages because they are now uninitialized again. */
494 	page_init_poison(pfn_to_page(start_pfn), sizeof(struct page) * nr_pages);
495 
496 #ifdef CONFIG_ZONE_DEVICE
497 	/*
498 	 * Zone shrinking code cannot properly deal with ZONE_DEVICE. So
499 	 * we will not try to shrink the zones - which is okay as
500 	 * set_zone_contiguous() cannot deal with ZONE_DEVICE either way.
501 	 */
502 	if (zone_idx(zone) == ZONE_DEVICE)
503 		return;
504 #endif
505 
506 	clear_zone_contiguous(zone);
507 
508 	pgdat_resize_lock(zone->zone_pgdat, &flags);
509 	shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
510 	update_pgdat_span(pgdat);
511 	pgdat_resize_unlock(zone->zone_pgdat, &flags);
512 
513 	set_zone_contiguous(zone);
514 }
515 
516 static void __remove_section(unsigned long pfn, unsigned long nr_pages,
517 			     unsigned long map_offset,
518 			     struct vmem_altmap *altmap)
519 {
520 	struct mem_section *ms = __nr_to_section(pfn_to_section_nr(pfn));
521 
522 	if (WARN_ON_ONCE(!valid_section(ms)))
523 		return;
524 
525 	sparse_remove_section(ms, pfn, nr_pages, map_offset, altmap);
526 }
527 
528 /**
529  * __remove_pages() - remove sections of pages
530  * @pfn: starting pageframe (must be aligned to start of a section)
531  * @nr_pages: number of pages to remove (must be multiple of section size)
532  * @altmap: alternative device page map or %NULL if default memmap is used
533  *
534  * Generic helper function to remove section mappings and sysfs entries
535  * for the section of the memory we are removing. Caller needs to make
536  * sure that pages are marked reserved and zones are adjust properly by
537  * calling offline_pages().
538  */
539 void __remove_pages(unsigned long pfn, unsigned long nr_pages,
540 		    struct vmem_altmap *altmap)
541 {
542 	unsigned long map_offset = 0;
543 	unsigned long nr, start_sec, end_sec;
544 
545 	map_offset = vmem_altmap_offset(altmap);
546 
547 	if (check_pfn_span(pfn, nr_pages, "remove"))
548 		return;
549 
550 	start_sec = pfn_to_section_nr(pfn);
551 	end_sec = pfn_to_section_nr(pfn + nr_pages - 1);
552 	for (nr = start_sec; nr <= end_sec; nr++) {
553 		unsigned long pfns;
554 
555 		cond_resched();
556 		pfns = min(nr_pages, PAGES_PER_SECTION
557 				- (pfn & ~PAGE_SECTION_MASK));
558 		__remove_section(pfn, pfns, map_offset, altmap);
559 		pfn += pfns;
560 		nr_pages -= pfns;
561 		map_offset = 0;
562 	}
563 }
564 
565 int set_online_page_callback(online_page_callback_t callback)
566 {
567 	int rc = -EINVAL;
568 
569 	get_online_mems();
570 	mutex_lock(&online_page_callback_lock);
571 
572 	if (online_page_callback == generic_online_page) {
573 		online_page_callback = callback;
574 		rc = 0;
575 	}
576 
577 	mutex_unlock(&online_page_callback_lock);
578 	put_online_mems();
579 
580 	return rc;
581 }
582 EXPORT_SYMBOL_GPL(set_online_page_callback);
583 
584 int restore_online_page_callback(online_page_callback_t callback)
585 {
586 	int rc = -EINVAL;
587 
588 	get_online_mems();
589 	mutex_lock(&online_page_callback_lock);
590 
591 	if (online_page_callback == callback) {
592 		online_page_callback = generic_online_page;
593 		rc = 0;
594 	}
595 
596 	mutex_unlock(&online_page_callback_lock);
597 	put_online_mems();
598 
599 	return rc;
600 }
601 EXPORT_SYMBOL_GPL(restore_online_page_callback);
602 
603 void generic_online_page(struct page *page, unsigned int order)
604 {
605 	kernel_map_pages(page, 1 << order, 1);
606 	__free_pages_core(page, order);
607 	totalram_pages_add(1UL << order);
608 #ifdef CONFIG_HIGHMEM
609 	if (PageHighMem(page))
610 		totalhigh_pages_add(1UL << order);
611 #endif
612 }
613 EXPORT_SYMBOL_GPL(generic_online_page);
614 
615 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
616 			void *arg)
617 {
618 	const unsigned long end_pfn = start_pfn + nr_pages;
619 	unsigned long pfn;
620 	int order;
621 
622 	/*
623 	 * Online the pages. The callback might decide to keep some pages
624 	 * PG_reserved (to add them to the buddy later), but we still account
625 	 * them as being online/belonging to this zone ("present").
626 	 */
627 	for (pfn = start_pfn; pfn < end_pfn; pfn += 1ul << order) {
628 		order = min(MAX_ORDER - 1, get_order(PFN_PHYS(end_pfn - pfn)));
629 		/* __free_pages_core() wants pfns to be aligned to the order */
630 		if (WARN_ON_ONCE(!IS_ALIGNED(pfn, 1ul << order)))
631 			order = 0;
632 		(*online_page_callback)(pfn_to_page(pfn), order);
633 	}
634 
635 	/* mark all involved sections as online */
636 	online_mem_sections(start_pfn, end_pfn);
637 
638 	*(unsigned long *)arg += nr_pages;
639 	return 0;
640 }
641 
642 /* check which state of node_states will be changed when online memory */
643 static void node_states_check_changes_online(unsigned long nr_pages,
644 	struct zone *zone, struct memory_notify *arg)
645 {
646 	int nid = zone_to_nid(zone);
647 
648 	arg->status_change_nid = NUMA_NO_NODE;
649 	arg->status_change_nid_normal = NUMA_NO_NODE;
650 	arg->status_change_nid_high = NUMA_NO_NODE;
651 
652 	if (!node_state(nid, N_MEMORY))
653 		arg->status_change_nid = nid;
654 	if (zone_idx(zone) <= ZONE_NORMAL && !node_state(nid, N_NORMAL_MEMORY))
655 		arg->status_change_nid_normal = nid;
656 #ifdef CONFIG_HIGHMEM
657 	if (zone_idx(zone) <= ZONE_HIGHMEM && !node_state(nid, N_HIGH_MEMORY))
658 		arg->status_change_nid_high = nid;
659 #endif
660 }
661 
662 static void node_states_set_node(int node, struct memory_notify *arg)
663 {
664 	if (arg->status_change_nid_normal >= 0)
665 		node_set_state(node, N_NORMAL_MEMORY);
666 
667 	if (arg->status_change_nid_high >= 0)
668 		node_set_state(node, N_HIGH_MEMORY);
669 
670 	if (arg->status_change_nid >= 0)
671 		node_set_state(node, N_MEMORY);
672 }
673 
674 static void __meminit resize_zone_range(struct zone *zone, unsigned long start_pfn,
675 		unsigned long nr_pages)
676 {
677 	unsigned long old_end_pfn = zone_end_pfn(zone);
678 
679 	if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
680 		zone->zone_start_pfn = start_pfn;
681 
682 	zone->spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - zone->zone_start_pfn;
683 }
684 
685 static void __meminit resize_pgdat_range(struct pglist_data *pgdat, unsigned long start_pfn,
686                                      unsigned long nr_pages)
687 {
688 	unsigned long old_end_pfn = pgdat_end_pfn(pgdat);
689 
690 	if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
691 		pgdat->node_start_pfn = start_pfn;
692 
693 	pgdat->node_spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - pgdat->node_start_pfn;
694 
695 }
696 /*
697  * Associate the pfn range with the given zone, initializing the memmaps
698  * and resizing the pgdat/zone data to span the added pages. After this
699  * call, all affected pages are PG_reserved.
700  */
701 void __ref move_pfn_range_to_zone(struct zone *zone, unsigned long start_pfn,
702 		unsigned long nr_pages, struct vmem_altmap *altmap)
703 {
704 	struct pglist_data *pgdat = zone->zone_pgdat;
705 	int nid = pgdat->node_id;
706 	unsigned long flags;
707 
708 	clear_zone_contiguous(zone);
709 
710 	/* TODO Huh pgdat is irqsave while zone is not. It used to be like that before */
711 	pgdat_resize_lock(pgdat, &flags);
712 	zone_span_writelock(zone);
713 	if (zone_is_empty(zone))
714 		init_currently_empty_zone(zone, start_pfn, nr_pages);
715 	resize_zone_range(zone, start_pfn, nr_pages);
716 	zone_span_writeunlock(zone);
717 	resize_pgdat_range(pgdat, start_pfn, nr_pages);
718 	pgdat_resize_unlock(pgdat, &flags);
719 
720 	/*
721 	 * TODO now we have a visible range of pages which are not associated
722 	 * with their zone properly. Not nice but set_pfnblock_flags_mask
723 	 * expects the zone spans the pfn range. All the pages in the range
724 	 * are reserved so nobody should be touching them so we should be safe
725 	 */
726 	memmap_init_zone(nr_pages, nid, zone_idx(zone), start_pfn,
727 			MEMMAP_HOTPLUG, altmap);
728 
729 	set_zone_contiguous(zone);
730 }
731 
732 /*
733  * Returns a default kernel memory zone for the given pfn range.
734  * If no kernel zone covers this pfn range it will automatically go
735  * to the ZONE_NORMAL.
736  */
737 static struct zone *default_kernel_zone_for_pfn(int nid, unsigned long start_pfn,
738 		unsigned long nr_pages)
739 {
740 	struct pglist_data *pgdat = NODE_DATA(nid);
741 	int zid;
742 
743 	for (zid = 0; zid <= ZONE_NORMAL; zid++) {
744 		struct zone *zone = &pgdat->node_zones[zid];
745 
746 		if (zone_intersects(zone, start_pfn, nr_pages))
747 			return zone;
748 	}
749 
750 	return &pgdat->node_zones[ZONE_NORMAL];
751 }
752 
753 static inline struct zone *default_zone_for_pfn(int nid, unsigned long start_pfn,
754 		unsigned long nr_pages)
755 {
756 	struct zone *kernel_zone = default_kernel_zone_for_pfn(nid, start_pfn,
757 			nr_pages);
758 	struct zone *movable_zone = &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
759 	bool in_kernel = zone_intersects(kernel_zone, start_pfn, nr_pages);
760 	bool in_movable = zone_intersects(movable_zone, start_pfn, nr_pages);
761 
762 	/*
763 	 * We inherit the existing zone in a simple case where zones do not
764 	 * overlap in the given range
765 	 */
766 	if (in_kernel ^ in_movable)
767 		return (in_kernel) ? kernel_zone : movable_zone;
768 
769 	/*
770 	 * If the range doesn't belong to any zone or two zones overlap in the
771 	 * given range then we use movable zone only if movable_node is
772 	 * enabled because we always online to a kernel zone by default.
773 	 */
774 	return movable_node_enabled ? movable_zone : kernel_zone;
775 }
776 
777 struct zone * zone_for_pfn_range(int online_type, int nid, unsigned start_pfn,
778 		unsigned long nr_pages)
779 {
780 	if (online_type == MMOP_ONLINE_KERNEL)
781 		return default_kernel_zone_for_pfn(nid, start_pfn, nr_pages);
782 
783 	if (online_type == MMOP_ONLINE_MOVABLE)
784 		return &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
785 
786 	return default_zone_for_pfn(nid, start_pfn, nr_pages);
787 }
788 
789 int __ref online_pages(unsigned long pfn, unsigned long nr_pages,
790 		       int online_type, int nid)
791 {
792 	unsigned long flags;
793 	unsigned long onlined_pages = 0;
794 	struct zone *zone;
795 	int need_zonelists_rebuild = 0;
796 	int ret;
797 	struct memory_notify arg;
798 
799 	mem_hotplug_begin();
800 
801 	/* associate pfn range with the zone */
802 	zone = zone_for_pfn_range(online_type, nid, pfn, nr_pages);
803 	move_pfn_range_to_zone(zone, pfn, nr_pages, NULL);
804 
805 	arg.start_pfn = pfn;
806 	arg.nr_pages = nr_pages;
807 	node_states_check_changes_online(nr_pages, zone, &arg);
808 
809 	ret = memory_notify(MEM_GOING_ONLINE, &arg);
810 	ret = notifier_to_errno(ret);
811 	if (ret)
812 		goto failed_addition;
813 
814 	/*
815 	 * If this zone is not populated, then it is not in zonelist.
816 	 * This means the page allocator ignores this zone.
817 	 * So, zonelist must be updated after online.
818 	 */
819 	if (!populated_zone(zone)) {
820 		need_zonelists_rebuild = 1;
821 		setup_zone_pageset(zone);
822 	}
823 
824 	ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
825 		online_pages_range);
826 	if (ret) {
827 		/* not a single memory resource was applicable */
828 		if (need_zonelists_rebuild)
829 			zone_pcp_reset(zone);
830 		goto failed_addition;
831 	}
832 
833 	zone->present_pages += onlined_pages;
834 
835 	pgdat_resize_lock(zone->zone_pgdat, &flags);
836 	zone->zone_pgdat->node_present_pages += onlined_pages;
837 	pgdat_resize_unlock(zone->zone_pgdat, &flags);
838 
839 	shuffle_zone(zone);
840 
841 	node_states_set_node(nid, &arg);
842 	if (need_zonelists_rebuild)
843 		build_all_zonelists(NULL);
844 	else
845 		zone_pcp_update(zone);
846 
847 	init_per_zone_wmark_min();
848 
849 	kswapd_run(nid);
850 	kcompactd_run(nid);
851 
852 	vm_total_pages = nr_free_pagecache_pages();
853 
854 	writeback_set_ratelimit();
855 
856 	memory_notify(MEM_ONLINE, &arg);
857 	mem_hotplug_done();
858 	return 0;
859 
860 failed_addition:
861 	pr_debug("online_pages [mem %#010llx-%#010llx] failed\n",
862 		 (unsigned long long) pfn << PAGE_SHIFT,
863 		 (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1);
864 	memory_notify(MEM_CANCEL_ONLINE, &arg);
865 	remove_pfn_range_from_zone(zone, pfn, nr_pages);
866 	mem_hotplug_done();
867 	return ret;
868 }
869 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
870 
871 static void reset_node_present_pages(pg_data_t *pgdat)
872 {
873 	struct zone *z;
874 
875 	for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
876 		z->present_pages = 0;
877 
878 	pgdat->node_present_pages = 0;
879 }
880 
881 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
882 static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
883 {
884 	struct pglist_data *pgdat;
885 	unsigned long start_pfn = PFN_DOWN(start);
886 
887 	pgdat = NODE_DATA(nid);
888 	if (!pgdat) {
889 		pgdat = arch_alloc_nodedata(nid);
890 		if (!pgdat)
891 			return NULL;
892 
893 		pgdat->per_cpu_nodestats =
894 			alloc_percpu(struct per_cpu_nodestat);
895 		arch_refresh_nodedata(nid, pgdat);
896 	} else {
897 		int cpu;
898 		/*
899 		 * Reset the nr_zones, order and classzone_idx before reuse.
900 		 * Note that kswapd will init kswapd_classzone_idx properly
901 		 * when it starts in the near future.
902 		 */
903 		pgdat->nr_zones = 0;
904 		pgdat->kswapd_order = 0;
905 		pgdat->kswapd_classzone_idx = 0;
906 		for_each_online_cpu(cpu) {
907 			struct per_cpu_nodestat *p;
908 
909 			p = per_cpu_ptr(pgdat->per_cpu_nodestats, cpu);
910 			memset(p, 0, sizeof(*p));
911 		}
912 	}
913 
914 	/* we can use NODE_DATA(nid) from here */
915 
916 	pgdat->node_id = nid;
917 	pgdat->node_start_pfn = start_pfn;
918 
919 	/* init node's zones as empty zones, we don't have any present pages.*/
920 	free_area_init_core_hotplug(nid);
921 
922 	/*
923 	 * The node we allocated has no zone fallback lists. For avoiding
924 	 * to access not-initialized zonelist, build here.
925 	 */
926 	build_all_zonelists(pgdat);
927 
928 	/*
929 	 * When memory is hot-added, all the memory is in offline state. So
930 	 * clear all zones' present_pages because they will be updated in
931 	 * online_pages() and offline_pages().
932 	 */
933 	reset_node_managed_pages(pgdat);
934 	reset_node_present_pages(pgdat);
935 
936 	return pgdat;
937 }
938 
939 static void rollback_node_hotadd(int nid)
940 {
941 	pg_data_t *pgdat = NODE_DATA(nid);
942 
943 	arch_refresh_nodedata(nid, NULL);
944 	free_percpu(pgdat->per_cpu_nodestats);
945 	arch_free_nodedata(pgdat);
946 }
947 
948 
949 /**
950  * try_online_node - online a node if offlined
951  * @nid: the node ID
952  * @start: start addr of the node
953  * @set_node_online: Whether we want to online the node
954  * called by cpu_up() to online a node without onlined memory.
955  *
956  * Returns:
957  * 1 -> a new node has been allocated
958  * 0 -> the node is already online
959  * -ENOMEM -> the node could not be allocated
960  */
961 static int __try_online_node(int nid, u64 start, bool set_node_online)
962 {
963 	pg_data_t *pgdat;
964 	int ret = 1;
965 
966 	if (node_online(nid))
967 		return 0;
968 
969 	pgdat = hotadd_new_pgdat(nid, start);
970 	if (!pgdat) {
971 		pr_err("Cannot online node %d due to NULL pgdat\n", nid);
972 		ret = -ENOMEM;
973 		goto out;
974 	}
975 
976 	if (set_node_online) {
977 		node_set_online(nid);
978 		ret = register_one_node(nid);
979 		BUG_ON(ret);
980 	}
981 out:
982 	return ret;
983 }
984 
985 /*
986  * Users of this function always want to online/register the node
987  */
988 int try_online_node(int nid)
989 {
990 	int ret;
991 
992 	mem_hotplug_begin();
993 	ret =  __try_online_node(nid, 0, true);
994 	mem_hotplug_done();
995 	return ret;
996 }
997 
998 static int check_hotplug_memory_range(u64 start, u64 size)
999 {
1000 	/* memory range must be block size aligned */
1001 	if (!size || !IS_ALIGNED(start, memory_block_size_bytes()) ||
1002 	    !IS_ALIGNED(size, memory_block_size_bytes())) {
1003 		pr_err("Block size [%#lx] unaligned hotplug range: start %#llx, size %#llx",
1004 		       memory_block_size_bytes(), start, size);
1005 		return -EINVAL;
1006 	}
1007 
1008 	return 0;
1009 }
1010 
1011 static int online_memory_block(struct memory_block *mem, void *arg)
1012 {
1013 	return device_online(&mem->dev);
1014 }
1015 
1016 /*
1017  * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1018  * and online/offline operations (triggered e.g. by sysfs).
1019  *
1020  * we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG
1021  */
1022 int __ref add_memory_resource(int nid, struct resource *res)
1023 {
1024 	struct mhp_restrictions restrictions = {};
1025 	u64 start, size;
1026 	bool new_node = false;
1027 	int ret;
1028 
1029 	start = res->start;
1030 	size = resource_size(res);
1031 
1032 	ret = check_hotplug_memory_range(start, size);
1033 	if (ret)
1034 		return ret;
1035 
1036 	mem_hotplug_begin();
1037 
1038 	/*
1039 	 * Add new range to memblock so that when hotadd_new_pgdat() is called
1040 	 * to allocate new pgdat, get_pfn_range_for_nid() will be able to find
1041 	 * this new range and calculate total pages correctly.  The range will
1042 	 * be removed at hot-remove time.
1043 	 */
1044 	memblock_add_node(start, size, nid);
1045 
1046 	ret = __try_online_node(nid, start, false);
1047 	if (ret < 0)
1048 		goto error;
1049 	new_node = ret;
1050 
1051 	/* call arch's memory hotadd */
1052 	ret = arch_add_memory(nid, start, size, &restrictions);
1053 	if (ret < 0)
1054 		goto error;
1055 
1056 	/* create memory block devices after memory was added */
1057 	ret = create_memory_block_devices(start, size);
1058 	if (ret) {
1059 		arch_remove_memory(nid, start, size, NULL);
1060 		goto error;
1061 	}
1062 
1063 	if (new_node) {
1064 		/* If sysfs file of new node can't be created, cpu on the node
1065 		 * can't be hot-added. There is no rollback way now.
1066 		 * So, check by BUG_ON() to catch it reluctantly..
1067 		 * We online node here. We can't roll back from here.
1068 		 */
1069 		node_set_online(nid);
1070 		ret = __register_one_node(nid);
1071 		BUG_ON(ret);
1072 	}
1073 
1074 	/* link memory sections under this node.*/
1075 	ret = link_mem_sections(nid, PFN_DOWN(start), PFN_UP(start + size - 1));
1076 	BUG_ON(ret);
1077 
1078 	/* create new memmap entry */
1079 	firmware_map_add_hotplug(start, start + size, "System RAM");
1080 
1081 	/* device_online() will take the lock when calling online_pages() */
1082 	mem_hotplug_done();
1083 
1084 	/* online pages if requested */
1085 	if (memhp_auto_online)
1086 		walk_memory_blocks(start, size, NULL, online_memory_block);
1087 
1088 	return ret;
1089 error:
1090 	/* rollback pgdat allocation and others */
1091 	if (new_node)
1092 		rollback_node_hotadd(nid);
1093 	memblock_remove(start, size);
1094 	mem_hotplug_done();
1095 	return ret;
1096 }
1097 
1098 /* requires device_hotplug_lock, see add_memory_resource() */
1099 int __ref __add_memory(int nid, u64 start, u64 size)
1100 {
1101 	struct resource *res;
1102 	int ret;
1103 
1104 	res = register_memory_resource(start, size);
1105 	if (IS_ERR(res))
1106 		return PTR_ERR(res);
1107 
1108 	ret = add_memory_resource(nid, res);
1109 	if (ret < 0)
1110 		release_memory_resource(res);
1111 	return ret;
1112 }
1113 
1114 int add_memory(int nid, u64 start, u64 size)
1115 {
1116 	int rc;
1117 
1118 	lock_device_hotplug();
1119 	rc = __add_memory(nid, start, size);
1120 	unlock_device_hotplug();
1121 
1122 	return rc;
1123 }
1124 EXPORT_SYMBOL_GPL(add_memory);
1125 
1126 #ifdef CONFIG_MEMORY_HOTREMOVE
1127 /*
1128  * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1129  * set and the size of the free page is given by page_order(). Using this,
1130  * the function determines if the pageblock contains only free pages.
1131  * Due to buddy contraints, a free page at least the size of a pageblock will
1132  * be located at the start of the pageblock
1133  */
1134 static inline int pageblock_free(struct page *page)
1135 {
1136 	return PageBuddy(page) && page_order(page) >= pageblock_order;
1137 }
1138 
1139 /* Return the pfn of the start of the next active pageblock after a given pfn */
1140 static unsigned long next_active_pageblock(unsigned long pfn)
1141 {
1142 	struct page *page = pfn_to_page(pfn);
1143 
1144 	/* Ensure the starting page is pageblock-aligned */
1145 	BUG_ON(pfn & (pageblock_nr_pages - 1));
1146 
1147 	/* If the entire pageblock is free, move to the end of free page */
1148 	if (pageblock_free(page)) {
1149 		int order;
1150 		/* be careful. we don't have locks, page_order can be changed.*/
1151 		order = page_order(page);
1152 		if ((order < MAX_ORDER) && (order >= pageblock_order))
1153 			return pfn + (1 << order);
1154 	}
1155 
1156 	return pfn + pageblock_nr_pages;
1157 }
1158 
1159 static bool is_pageblock_removable_nolock(unsigned long pfn)
1160 {
1161 	struct page *page = pfn_to_page(pfn);
1162 	struct zone *zone;
1163 
1164 	/*
1165 	 * We have to be careful here because we are iterating over memory
1166 	 * sections which are not zone aware so we might end up outside of
1167 	 * the zone but still within the section.
1168 	 * We have to take care about the node as well. If the node is offline
1169 	 * its NODE_DATA will be NULL - see page_zone.
1170 	 */
1171 	if (!node_online(page_to_nid(page)))
1172 		return false;
1173 
1174 	zone = page_zone(page);
1175 	pfn = page_to_pfn(page);
1176 	if (!zone_spans_pfn(zone, pfn))
1177 		return false;
1178 
1179 	return !has_unmovable_pages(zone, page, MIGRATE_MOVABLE,
1180 				    MEMORY_OFFLINE);
1181 }
1182 
1183 /* Checks if this range of memory is likely to be hot-removable. */
1184 bool is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
1185 {
1186 	unsigned long end_pfn, pfn;
1187 
1188 	end_pfn = min(start_pfn + nr_pages,
1189 			zone_end_pfn(page_zone(pfn_to_page(start_pfn))));
1190 
1191 	/* Check the starting page of each pageblock within the range */
1192 	for (pfn = start_pfn; pfn < end_pfn; pfn = next_active_pageblock(pfn)) {
1193 		if (!is_pageblock_removable_nolock(pfn))
1194 			return false;
1195 		cond_resched();
1196 	}
1197 
1198 	/* All pageblocks in the memory block are likely to be hot-removable */
1199 	return true;
1200 }
1201 
1202 /*
1203  * Confirm all pages in a range [start, end) belong to the same zone.
1204  * When true, return its valid [start, end).
1205  */
1206 int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn,
1207 			 unsigned long *valid_start, unsigned long *valid_end)
1208 {
1209 	unsigned long pfn, sec_end_pfn;
1210 	unsigned long start, end;
1211 	struct zone *zone = NULL;
1212 	struct page *page;
1213 	int i;
1214 	for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1);
1215 	     pfn < end_pfn;
1216 	     pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) {
1217 		/* Make sure the memory section is present first */
1218 		if (!present_section_nr(pfn_to_section_nr(pfn)))
1219 			continue;
1220 		for (; pfn < sec_end_pfn && pfn < end_pfn;
1221 		     pfn += MAX_ORDER_NR_PAGES) {
1222 			i = 0;
1223 			/* This is just a CONFIG_HOLES_IN_ZONE check.*/
1224 			while ((i < MAX_ORDER_NR_PAGES) &&
1225 				!pfn_valid_within(pfn + i))
1226 				i++;
1227 			if (i == MAX_ORDER_NR_PAGES || pfn + i >= end_pfn)
1228 				continue;
1229 			/* Check if we got outside of the zone */
1230 			if (zone && !zone_spans_pfn(zone, pfn + i))
1231 				return 0;
1232 			page = pfn_to_page(pfn + i);
1233 			if (zone && page_zone(page) != zone)
1234 				return 0;
1235 			if (!zone)
1236 				start = pfn + i;
1237 			zone = page_zone(page);
1238 			end = pfn + MAX_ORDER_NR_PAGES;
1239 		}
1240 	}
1241 
1242 	if (zone) {
1243 		*valid_start = start;
1244 		*valid_end = min(end, end_pfn);
1245 		return 1;
1246 	} else {
1247 		return 0;
1248 	}
1249 }
1250 
1251 /*
1252  * Scan pfn range [start,end) to find movable/migratable pages (LRU pages,
1253  * non-lru movable pages and hugepages). We scan pfn because it's much
1254  * easier than scanning over linked list. This function returns the pfn
1255  * of the first found movable page if it's found, otherwise 0.
1256  */
1257 static unsigned long scan_movable_pages(unsigned long start, unsigned long end)
1258 {
1259 	unsigned long pfn;
1260 
1261 	for (pfn = start; pfn < end; pfn++) {
1262 		struct page *page, *head;
1263 		unsigned long skip;
1264 
1265 		if (!pfn_valid(pfn))
1266 			continue;
1267 		page = pfn_to_page(pfn);
1268 		if (PageLRU(page))
1269 			return pfn;
1270 		if (__PageMovable(page))
1271 			return pfn;
1272 
1273 		if (!PageHuge(page))
1274 			continue;
1275 		head = compound_head(page);
1276 		if (page_huge_active(head))
1277 			return pfn;
1278 		skip = compound_nr(head) - (page - head);
1279 		pfn += skip - 1;
1280 	}
1281 	return 0;
1282 }
1283 
1284 static struct page *new_node_page(struct page *page, unsigned long private)
1285 {
1286 	int nid = page_to_nid(page);
1287 	nodemask_t nmask = node_states[N_MEMORY];
1288 
1289 	/*
1290 	 * try to allocate from a different node but reuse this node if there
1291 	 * are no other online nodes to be used (e.g. we are offlining a part
1292 	 * of the only existing node)
1293 	 */
1294 	node_clear(nid, nmask);
1295 	if (nodes_empty(nmask))
1296 		node_set(nid, nmask);
1297 
1298 	return new_page_nodemask(page, nid, &nmask);
1299 }
1300 
1301 static int
1302 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1303 {
1304 	unsigned long pfn;
1305 	struct page *page;
1306 	int ret = 0;
1307 	LIST_HEAD(source);
1308 
1309 	for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1310 		if (!pfn_valid(pfn))
1311 			continue;
1312 		page = pfn_to_page(pfn);
1313 
1314 		if (PageHuge(page)) {
1315 			struct page *head = compound_head(page);
1316 			pfn = page_to_pfn(head) + compound_nr(head) - 1;
1317 			isolate_huge_page(head, &source);
1318 			continue;
1319 		} else if (PageTransHuge(page))
1320 			pfn = page_to_pfn(compound_head(page))
1321 				+ hpage_nr_pages(page) - 1;
1322 
1323 		/*
1324 		 * HWPoison pages have elevated reference counts so the migration would
1325 		 * fail on them. It also doesn't make any sense to migrate them in the
1326 		 * first place. Still try to unmap such a page in case it is still mapped
1327 		 * (e.g. current hwpoison implementation doesn't unmap KSM pages but keep
1328 		 * the unmap as the catch all safety net).
1329 		 */
1330 		if (PageHWPoison(page)) {
1331 			if (WARN_ON(PageLRU(page)))
1332 				isolate_lru_page(page);
1333 			if (page_mapped(page))
1334 				try_to_unmap(page, TTU_IGNORE_MLOCK | TTU_IGNORE_ACCESS);
1335 			continue;
1336 		}
1337 
1338 		if (!get_page_unless_zero(page))
1339 			continue;
1340 		/*
1341 		 * We can skip free pages. And we can deal with pages on
1342 		 * LRU and non-lru movable pages.
1343 		 */
1344 		if (PageLRU(page))
1345 			ret = isolate_lru_page(page);
1346 		else
1347 			ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE);
1348 		if (!ret) { /* Success */
1349 			list_add_tail(&page->lru, &source);
1350 			if (!__PageMovable(page))
1351 				inc_node_page_state(page, NR_ISOLATED_ANON +
1352 						    page_is_file_cache(page));
1353 
1354 		} else {
1355 			pr_warn("failed to isolate pfn %lx\n", pfn);
1356 			dump_page(page, "isolation failed");
1357 		}
1358 		put_page(page);
1359 	}
1360 	if (!list_empty(&source)) {
1361 		/* Allocate a new page from the nearest neighbor node */
1362 		ret = migrate_pages(&source, new_node_page, NULL, 0,
1363 					MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
1364 		if (ret) {
1365 			list_for_each_entry(page, &source, lru) {
1366 				pr_warn("migrating pfn %lx failed ret:%d ",
1367 				       page_to_pfn(page), ret);
1368 				dump_page(page, "migration failure");
1369 			}
1370 			putback_movable_pages(&source);
1371 		}
1372 	}
1373 
1374 	return ret;
1375 }
1376 
1377 /* Mark all sections offline and remove all free pages from the buddy. */
1378 static int
1379 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1380 			void *data)
1381 {
1382 	unsigned long *offlined_pages = (unsigned long *)data;
1383 
1384 	*offlined_pages += __offline_isolated_pages(start, start + nr_pages);
1385 	return 0;
1386 }
1387 
1388 /*
1389  * Check all pages in range, recoreded as memory resource, are isolated.
1390  */
1391 static int
1392 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1393 			void *data)
1394 {
1395 	return test_pages_isolated(start_pfn, start_pfn + nr_pages,
1396 				   MEMORY_OFFLINE);
1397 }
1398 
1399 static int __init cmdline_parse_movable_node(char *p)
1400 {
1401 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1402 	movable_node_enabled = true;
1403 #else
1404 	pr_warn("movable_node parameter depends on CONFIG_HAVE_MEMBLOCK_NODE_MAP to work properly\n");
1405 #endif
1406 	return 0;
1407 }
1408 early_param("movable_node", cmdline_parse_movable_node);
1409 
1410 /* check which state of node_states will be changed when offline memory */
1411 static void node_states_check_changes_offline(unsigned long nr_pages,
1412 		struct zone *zone, struct memory_notify *arg)
1413 {
1414 	struct pglist_data *pgdat = zone->zone_pgdat;
1415 	unsigned long present_pages = 0;
1416 	enum zone_type zt;
1417 
1418 	arg->status_change_nid = NUMA_NO_NODE;
1419 	arg->status_change_nid_normal = NUMA_NO_NODE;
1420 	arg->status_change_nid_high = NUMA_NO_NODE;
1421 
1422 	/*
1423 	 * Check whether node_states[N_NORMAL_MEMORY] will be changed.
1424 	 * If the memory to be offline is within the range
1425 	 * [0..ZONE_NORMAL], and it is the last present memory there,
1426 	 * the zones in that range will become empty after the offlining,
1427 	 * thus we can determine that we need to clear the node from
1428 	 * node_states[N_NORMAL_MEMORY].
1429 	 */
1430 	for (zt = 0; zt <= ZONE_NORMAL; zt++)
1431 		present_pages += pgdat->node_zones[zt].present_pages;
1432 	if (zone_idx(zone) <= ZONE_NORMAL && nr_pages >= present_pages)
1433 		arg->status_change_nid_normal = zone_to_nid(zone);
1434 
1435 #ifdef CONFIG_HIGHMEM
1436 	/*
1437 	 * node_states[N_HIGH_MEMORY] contains nodes which
1438 	 * have normal memory or high memory.
1439 	 * Here we add the present_pages belonging to ZONE_HIGHMEM.
1440 	 * If the zone is within the range of [0..ZONE_HIGHMEM), and
1441 	 * we determine that the zones in that range become empty,
1442 	 * we need to clear the node for N_HIGH_MEMORY.
1443 	 */
1444 	present_pages += pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1445 	if (zone_idx(zone) <= ZONE_HIGHMEM && nr_pages >= present_pages)
1446 		arg->status_change_nid_high = zone_to_nid(zone);
1447 #endif
1448 
1449 	/*
1450 	 * We have accounted the pages from [0..ZONE_NORMAL), and
1451 	 * in case of CONFIG_HIGHMEM the pages from ZONE_HIGHMEM
1452 	 * as well.
1453 	 * Here we count the possible pages from ZONE_MOVABLE.
1454 	 * If after having accounted all the pages, we see that the nr_pages
1455 	 * to be offlined is over or equal to the accounted pages,
1456 	 * we know that the node will become empty, and so, we can clear
1457 	 * it for N_MEMORY as well.
1458 	 */
1459 	present_pages += pgdat->node_zones[ZONE_MOVABLE].present_pages;
1460 
1461 	if (nr_pages >= present_pages)
1462 		arg->status_change_nid = zone_to_nid(zone);
1463 }
1464 
1465 static void node_states_clear_node(int node, struct memory_notify *arg)
1466 {
1467 	if (arg->status_change_nid_normal >= 0)
1468 		node_clear_state(node, N_NORMAL_MEMORY);
1469 
1470 	if (arg->status_change_nid_high >= 0)
1471 		node_clear_state(node, N_HIGH_MEMORY);
1472 
1473 	if (arg->status_change_nid >= 0)
1474 		node_clear_state(node, N_MEMORY);
1475 }
1476 
1477 static int count_system_ram_pages_cb(unsigned long start_pfn,
1478 				     unsigned long nr_pages, void *data)
1479 {
1480 	unsigned long *nr_system_ram_pages = data;
1481 
1482 	*nr_system_ram_pages += nr_pages;
1483 	return 0;
1484 }
1485 
1486 static int __ref __offline_pages(unsigned long start_pfn,
1487 		  unsigned long end_pfn)
1488 {
1489 	unsigned long pfn, nr_pages = 0;
1490 	unsigned long offlined_pages = 0;
1491 	int ret, node, nr_isolate_pageblock;
1492 	unsigned long flags;
1493 	unsigned long valid_start, valid_end;
1494 	struct zone *zone;
1495 	struct memory_notify arg;
1496 	char *reason;
1497 
1498 	mem_hotplug_begin();
1499 
1500 	/*
1501 	 * Don't allow to offline memory blocks that contain holes.
1502 	 * Consequently, memory blocks with holes can never get onlined
1503 	 * via the hotplug path - online_pages() - as hotplugged memory has
1504 	 * no holes. This way, we e.g., don't have to worry about marking
1505 	 * memory holes PG_reserved, don't need pfn_valid() checks, and can
1506 	 * avoid using walk_system_ram_range() later.
1507 	 */
1508 	walk_system_ram_range(start_pfn, end_pfn - start_pfn, &nr_pages,
1509 			      count_system_ram_pages_cb);
1510 	if (nr_pages != end_pfn - start_pfn) {
1511 		ret = -EINVAL;
1512 		reason = "memory holes";
1513 		goto failed_removal;
1514 	}
1515 
1516 	/* This makes hotplug much easier...and readable.
1517 	   we assume this for now. .*/
1518 	if (!test_pages_in_a_zone(start_pfn, end_pfn, &valid_start,
1519 				  &valid_end)) {
1520 		ret = -EINVAL;
1521 		reason = "multizone range";
1522 		goto failed_removal;
1523 	}
1524 
1525 	zone = page_zone(pfn_to_page(valid_start));
1526 	node = zone_to_nid(zone);
1527 
1528 	/* set above range as isolated */
1529 	ret = start_isolate_page_range(start_pfn, end_pfn,
1530 				       MIGRATE_MOVABLE,
1531 				       MEMORY_OFFLINE | REPORT_FAILURE);
1532 	if (ret < 0) {
1533 		reason = "failure to isolate range";
1534 		goto failed_removal;
1535 	}
1536 	nr_isolate_pageblock = ret;
1537 
1538 	arg.start_pfn = start_pfn;
1539 	arg.nr_pages = nr_pages;
1540 	node_states_check_changes_offline(nr_pages, zone, &arg);
1541 
1542 	ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1543 	ret = notifier_to_errno(ret);
1544 	if (ret) {
1545 		reason = "notifier failure";
1546 		goto failed_removal_isolated;
1547 	}
1548 
1549 	do {
1550 		for (pfn = start_pfn; pfn;) {
1551 			if (signal_pending(current)) {
1552 				ret = -EINTR;
1553 				reason = "signal backoff";
1554 				goto failed_removal_isolated;
1555 			}
1556 
1557 			cond_resched();
1558 			lru_add_drain_all();
1559 
1560 			pfn = scan_movable_pages(pfn, end_pfn);
1561 			if (pfn) {
1562 				/*
1563 				 * TODO: fatal migration failures should bail
1564 				 * out
1565 				 */
1566 				do_migrate_range(pfn, end_pfn);
1567 			}
1568 		}
1569 
1570 		/*
1571 		 * Dissolve free hugepages in the memory block before doing
1572 		 * offlining actually in order to make hugetlbfs's object
1573 		 * counting consistent.
1574 		 */
1575 		ret = dissolve_free_huge_pages(start_pfn, end_pfn);
1576 		if (ret) {
1577 			reason = "failure to dissolve huge pages";
1578 			goto failed_removal_isolated;
1579 		}
1580 		/* check again */
1581 		ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn,
1582 					    NULL, check_pages_isolated_cb);
1583 	} while (ret);
1584 
1585 	/* Ok, all of our target is isolated.
1586 	   We cannot do rollback at this point. */
1587 	walk_system_ram_range(start_pfn, end_pfn - start_pfn,
1588 			      &offlined_pages, offline_isolated_pages_cb);
1589 	pr_info("Offlined Pages %ld\n", offlined_pages);
1590 	/*
1591 	 * Onlining will reset pagetype flags and makes migrate type
1592 	 * MOVABLE, so just need to decrease the number of isolated
1593 	 * pageblocks zone counter here.
1594 	 */
1595 	spin_lock_irqsave(&zone->lock, flags);
1596 	zone->nr_isolate_pageblock -= nr_isolate_pageblock;
1597 	spin_unlock_irqrestore(&zone->lock, flags);
1598 
1599 	/* removal success */
1600 	adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages);
1601 	zone->present_pages -= offlined_pages;
1602 
1603 	pgdat_resize_lock(zone->zone_pgdat, &flags);
1604 	zone->zone_pgdat->node_present_pages -= offlined_pages;
1605 	pgdat_resize_unlock(zone->zone_pgdat, &flags);
1606 
1607 	init_per_zone_wmark_min();
1608 
1609 	if (!populated_zone(zone)) {
1610 		zone_pcp_reset(zone);
1611 		build_all_zonelists(NULL);
1612 	} else
1613 		zone_pcp_update(zone);
1614 
1615 	node_states_clear_node(node, &arg);
1616 	if (arg.status_change_nid >= 0) {
1617 		kswapd_stop(node);
1618 		kcompactd_stop(node);
1619 	}
1620 
1621 	vm_total_pages = nr_free_pagecache_pages();
1622 	writeback_set_ratelimit();
1623 
1624 	memory_notify(MEM_OFFLINE, &arg);
1625 	remove_pfn_range_from_zone(zone, start_pfn, nr_pages);
1626 	mem_hotplug_done();
1627 	return 0;
1628 
1629 failed_removal_isolated:
1630 	undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1631 	memory_notify(MEM_CANCEL_OFFLINE, &arg);
1632 failed_removal:
1633 	pr_debug("memory offlining [mem %#010llx-%#010llx] failed due to %s\n",
1634 		 (unsigned long long) start_pfn << PAGE_SHIFT,
1635 		 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1,
1636 		 reason);
1637 	/* pushback to free area */
1638 	mem_hotplug_done();
1639 	return ret;
1640 }
1641 
1642 int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1643 {
1644 	return __offline_pages(start_pfn, start_pfn + nr_pages);
1645 }
1646 
1647 static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
1648 {
1649 	int ret = !is_memblock_offlined(mem);
1650 
1651 	if (unlikely(ret)) {
1652 		phys_addr_t beginpa, endpa;
1653 
1654 		beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
1655 		endpa = beginpa + memory_block_size_bytes() - 1;
1656 		pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n",
1657 			&beginpa, &endpa);
1658 
1659 		return -EBUSY;
1660 	}
1661 	return 0;
1662 }
1663 
1664 static int check_cpu_on_node(pg_data_t *pgdat)
1665 {
1666 	int cpu;
1667 
1668 	for_each_present_cpu(cpu) {
1669 		if (cpu_to_node(cpu) == pgdat->node_id)
1670 			/*
1671 			 * the cpu on this node isn't removed, and we can't
1672 			 * offline this node.
1673 			 */
1674 			return -EBUSY;
1675 	}
1676 
1677 	return 0;
1678 }
1679 
1680 static int check_no_memblock_for_node_cb(struct memory_block *mem, void *arg)
1681 {
1682 	int nid = *(int *)arg;
1683 
1684 	/*
1685 	 * If a memory block belongs to multiple nodes, the stored nid is not
1686 	 * reliable. However, such blocks are always online (e.g., cannot get
1687 	 * offlined) and, therefore, are still spanned by the node.
1688 	 */
1689 	return mem->nid == nid ? -EEXIST : 0;
1690 }
1691 
1692 /**
1693  * try_offline_node
1694  * @nid: the node ID
1695  *
1696  * Offline a node if all memory sections and cpus of the node are removed.
1697  *
1698  * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1699  * and online/offline operations before this call.
1700  */
1701 void try_offline_node(int nid)
1702 {
1703 	pg_data_t *pgdat = NODE_DATA(nid);
1704 	int rc;
1705 
1706 	/*
1707 	 * If the node still spans pages (especially ZONE_DEVICE), don't
1708 	 * offline it. A node spans memory after move_pfn_range_to_zone(),
1709 	 * e.g., after the memory block was onlined.
1710 	 */
1711 	if (pgdat->node_spanned_pages)
1712 		return;
1713 
1714 	/*
1715 	 * Especially offline memory blocks might not be spanned by the
1716 	 * node. They will get spanned by the node once they get onlined.
1717 	 * However, they link to the node in sysfs and can get onlined later.
1718 	 */
1719 	rc = for_each_memory_block(&nid, check_no_memblock_for_node_cb);
1720 	if (rc)
1721 		return;
1722 
1723 	if (check_cpu_on_node(pgdat))
1724 		return;
1725 
1726 	/*
1727 	 * all memory/cpu of this node are removed, we can offline this
1728 	 * node now.
1729 	 */
1730 	node_set_offline(nid);
1731 	unregister_one_node(nid);
1732 }
1733 EXPORT_SYMBOL(try_offline_node);
1734 
1735 static void __release_memory_resource(resource_size_t start,
1736 				      resource_size_t size)
1737 {
1738 	int ret;
1739 
1740 	/*
1741 	 * When removing memory in the same granularity as it was added,
1742 	 * this function never fails. It might only fail if resources
1743 	 * have to be adjusted or split. We'll ignore the error, as
1744 	 * removing of memory cannot fail.
1745 	 */
1746 	ret = release_mem_region_adjustable(&iomem_resource, start, size);
1747 	if (ret) {
1748 		resource_size_t endres = start + size - 1;
1749 
1750 		pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
1751 			&start, &endres, ret);
1752 	}
1753 }
1754 
1755 static int __ref try_remove_memory(int nid, u64 start, u64 size)
1756 {
1757 	int rc = 0;
1758 
1759 	BUG_ON(check_hotplug_memory_range(start, size));
1760 
1761 	/*
1762 	 * All memory blocks must be offlined before removing memory.  Check
1763 	 * whether all memory blocks in question are offline and return error
1764 	 * if this is not the case.
1765 	 */
1766 	rc = walk_memory_blocks(start, size, NULL, check_memblock_offlined_cb);
1767 	if (rc)
1768 		goto done;
1769 
1770 	/* remove memmap entry */
1771 	firmware_map_remove(start, start + size, "System RAM");
1772 
1773 	/*
1774 	 * Memory block device removal under the device_hotplug_lock is
1775 	 * a barrier against racing online attempts.
1776 	 */
1777 	remove_memory_block_devices(start, size);
1778 
1779 	mem_hotplug_begin();
1780 
1781 	arch_remove_memory(nid, start, size, NULL);
1782 	memblock_free(start, size);
1783 	memblock_remove(start, size);
1784 	__release_memory_resource(start, size);
1785 
1786 	try_offline_node(nid);
1787 
1788 done:
1789 	mem_hotplug_done();
1790 	return rc;
1791 }
1792 
1793 /**
1794  * remove_memory
1795  * @nid: the node ID
1796  * @start: physical address of the region to remove
1797  * @size: size of the region to remove
1798  *
1799  * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1800  * and online/offline operations before this call, as required by
1801  * try_offline_node().
1802  */
1803 void __remove_memory(int nid, u64 start, u64 size)
1804 {
1805 
1806 	/*
1807 	 * trigger BUG() if some memory is not offlined prior to calling this
1808 	 * function
1809 	 */
1810 	if (try_remove_memory(nid, start, size))
1811 		BUG();
1812 }
1813 
1814 /*
1815  * Remove memory if every memory block is offline, otherwise return -EBUSY is
1816  * some memory is not offline
1817  */
1818 int remove_memory(int nid, u64 start, u64 size)
1819 {
1820 	int rc;
1821 
1822 	lock_device_hotplug();
1823 	rc  = try_remove_memory(nid, start, size);
1824 	unlock_device_hotplug();
1825 
1826 	return rc;
1827 }
1828 EXPORT_SYMBOL_GPL(remove_memory);
1829 #endif /* CONFIG_MEMORY_HOTREMOVE */
1830