xref: /kernel/linux/linux-5.10/mm/memory_hotplug.c (revision 8c2ecf20)
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#include <linux/zswapd.h>
40
41#include <asm/tlbflush.h>
42
43#include "internal.h"
44#include "shuffle.h"
45
46/*
47 * online_page_callback contains pointer to current page onlining function.
48 * Initially it is generic_online_page(). If it is required it could be
49 * changed by calling set_online_page_callback() for callback registration
50 * and restore_online_page_callback() for generic callback restore.
51 */
52
53static online_page_callback_t online_page_callback = generic_online_page;
54static DEFINE_MUTEX(online_page_callback_lock);
55
56DEFINE_STATIC_PERCPU_RWSEM(mem_hotplug_lock);
57
58void get_online_mems(void)
59{
60	percpu_down_read(&mem_hotplug_lock);
61}
62
63void put_online_mems(void)
64{
65	percpu_up_read(&mem_hotplug_lock);
66}
67
68bool movable_node_enabled = false;
69
70#ifndef CONFIG_MEMORY_HOTPLUG_DEFAULT_ONLINE
71int memhp_default_online_type = MMOP_OFFLINE;
72#else
73int memhp_default_online_type = MMOP_ONLINE;
74#endif
75
76static int __init setup_memhp_default_state(char *str)
77{
78	const int online_type = memhp_online_type_from_str(str);
79
80	if (online_type >= 0)
81		memhp_default_online_type = online_type;
82
83	return 1;
84}
85__setup("memhp_default_state=", setup_memhp_default_state);
86
87void mem_hotplug_begin(void)
88{
89	cpus_read_lock();
90	percpu_down_write(&mem_hotplug_lock);
91}
92
93void mem_hotplug_done(void)
94{
95	percpu_up_write(&mem_hotplug_lock);
96	cpus_read_unlock();
97}
98
99u64 max_mem_size = U64_MAX;
100
101/* add this memory to iomem resource */
102static struct resource *register_memory_resource(u64 start, u64 size,
103						 const char *resource_name)
104{
105	struct resource *res;
106	unsigned long flags =  IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
107
108	if (strcmp(resource_name, "System RAM"))
109		flags |= IORESOURCE_SYSRAM_DRIVER_MANAGED;
110
111	/*
112	 * Make sure value parsed from 'mem=' only restricts memory adding
113	 * while booting, so that memory hotplug won't be impacted. Please
114	 * refer to document of 'mem=' in kernel-parameters.txt for more
115	 * details.
116	 */
117	if (start + size > max_mem_size && system_state < SYSTEM_RUNNING)
118		return ERR_PTR(-E2BIG);
119
120	/*
121	 * Request ownership of the new memory range.  This might be
122	 * a child of an existing resource that was present but
123	 * not marked as busy.
124	 */
125	res = __request_region(&iomem_resource, start, size,
126			       resource_name, flags);
127
128	if (!res) {
129		pr_debug("Unable to reserve System RAM region: %016llx->%016llx\n",
130				start, start + size);
131		return ERR_PTR(-EEXIST);
132	}
133	return res;
134}
135
136static void release_memory_resource(struct resource *res)
137{
138	if (!res)
139		return;
140	release_resource(res);
141	kfree(res);
142}
143
144#ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
145void get_page_bootmem(unsigned long info,  struct page *page,
146		      unsigned long type)
147{
148	page->freelist = (void *)type;
149	SetPagePrivate(page);
150	set_page_private(page, info);
151	page_ref_inc(page);
152}
153
154void put_page_bootmem(struct page *page)
155{
156	unsigned long type;
157
158	type = (unsigned long) page->freelist;
159	BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE ||
160	       type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE);
161
162	if (page_ref_dec_return(page) == 1) {
163		page->freelist = NULL;
164		ClearPagePrivate(page);
165		set_page_private(page, 0);
166		INIT_LIST_HEAD(&page->lru);
167		free_reserved_page(page);
168	}
169}
170
171#ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
172#ifndef CONFIG_SPARSEMEM_VMEMMAP
173static void register_page_bootmem_info_section(unsigned long start_pfn)
174{
175	unsigned long mapsize, section_nr, i;
176	struct mem_section *ms;
177	struct page *page, *memmap;
178	struct mem_section_usage *usage;
179
180	section_nr = pfn_to_section_nr(start_pfn);
181	ms = __nr_to_section(section_nr);
182
183	/* Get section's memmap address */
184	memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
185
186	/*
187	 * Get page for the memmap's phys address
188	 * XXX: need more consideration for sparse_vmemmap...
189	 */
190	page = virt_to_page(memmap);
191	mapsize = sizeof(struct page) * PAGES_PER_SECTION;
192	mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
193
194	/* remember memmap's page */
195	for (i = 0; i < mapsize; i++, page++)
196		get_page_bootmem(section_nr, page, SECTION_INFO);
197
198	usage = ms->usage;
199	page = virt_to_page(usage);
200
201	mapsize = PAGE_ALIGN(mem_section_usage_size()) >> PAGE_SHIFT;
202
203	for (i = 0; i < mapsize; i++, page++)
204		get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
205
206}
207#else /* CONFIG_SPARSEMEM_VMEMMAP */
208static void register_page_bootmem_info_section(unsigned long start_pfn)
209{
210	unsigned long mapsize, section_nr, i;
211	struct mem_section *ms;
212	struct page *page, *memmap;
213	struct mem_section_usage *usage;
214
215	section_nr = pfn_to_section_nr(start_pfn);
216	ms = __nr_to_section(section_nr);
217
218	memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
219
220	register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION);
221
222	usage = ms->usage;
223	page = virt_to_page(usage);
224
225	mapsize = PAGE_ALIGN(mem_section_usage_size()) >> PAGE_SHIFT;
226
227	for (i = 0; i < mapsize; i++, page++)
228		get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
229}
230#endif /* !CONFIG_SPARSEMEM_VMEMMAP */
231
232void __init register_page_bootmem_info_node(struct pglist_data *pgdat)
233{
234	unsigned long i, pfn, end_pfn, nr_pages;
235	int node = pgdat->node_id;
236	struct page *page;
237
238	nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
239	page = virt_to_page(pgdat);
240
241	for (i = 0; i < nr_pages; i++, page++)
242		get_page_bootmem(node, page, NODE_INFO);
243
244	pfn = pgdat->node_start_pfn;
245	end_pfn = pgdat_end_pfn(pgdat);
246
247	/* register section info */
248	for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
249		/*
250		 * Some platforms can assign the same pfn to multiple nodes - on
251		 * node0 as well as nodeN.  To avoid registering a pfn against
252		 * multiple nodes we check that this pfn does not already
253		 * reside in some other nodes.
254		 */
255		if (pfn_valid(pfn) && (early_pfn_to_nid(pfn) == node))
256			register_page_bootmem_info_section(pfn);
257	}
258}
259#endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
260
261static int check_pfn_span(unsigned long pfn, unsigned long nr_pages,
262		const char *reason)
263{
264	/*
265	 * Disallow all operations smaller than a sub-section and only
266	 * allow operations smaller than a section for
267	 * SPARSEMEM_VMEMMAP. Note that check_hotplug_memory_range()
268	 * enforces a larger memory_block_size_bytes() granularity for
269	 * memory that will be marked online, so this check should only
270	 * fire for direct arch_{add,remove}_memory() users outside of
271	 * add_memory_resource().
272	 */
273	unsigned long min_align;
274
275	if (IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP))
276		min_align = PAGES_PER_SUBSECTION;
277	else
278		min_align = PAGES_PER_SECTION;
279	if (!IS_ALIGNED(pfn, min_align)
280			|| !IS_ALIGNED(nr_pages, min_align)) {
281		WARN(1, "Misaligned __%s_pages start: %#lx end: #%lx\n",
282				reason, pfn, pfn + nr_pages - 1);
283		return -EINVAL;
284	}
285	return 0;
286}
287
288static int check_hotplug_memory_addressable(unsigned long pfn,
289					    unsigned long nr_pages)
290{
291	const u64 max_addr = PFN_PHYS(pfn + nr_pages) - 1;
292
293	if (max_addr >> MAX_PHYSMEM_BITS) {
294		const u64 max_allowed = (1ull << (MAX_PHYSMEM_BITS + 1)) - 1;
295		WARN(1,
296		     "Hotplugged memory exceeds maximum addressable address, range=%#llx-%#llx, maximum=%#llx\n",
297		     (u64)PFN_PHYS(pfn), max_addr, max_allowed);
298		return -E2BIG;
299	}
300
301	return 0;
302}
303
304/*
305 * Reasonably generic function for adding memory.  It is
306 * expected that archs that support memory hotplug will
307 * call this function after deciding the zone to which to
308 * add the new pages.
309 */
310int __ref __add_pages(int nid, unsigned long pfn, unsigned long nr_pages,
311		struct mhp_params *params)
312{
313	const unsigned long end_pfn = pfn + nr_pages;
314	unsigned long cur_nr_pages;
315	int err;
316	struct vmem_altmap *altmap = params->altmap;
317
318	if (WARN_ON_ONCE(!params->pgprot.pgprot))
319		return -EINVAL;
320
321	err = check_hotplug_memory_addressable(pfn, nr_pages);
322	if (err)
323		return err;
324
325	if (altmap) {
326		/*
327		 * Validate altmap is within bounds of the total request
328		 */
329		if (altmap->base_pfn != pfn
330				|| vmem_altmap_offset(altmap) > nr_pages) {
331			pr_warn_once("memory add fail, invalid altmap\n");
332			return -EINVAL;
333		}
334		altmap->alloc = 0;
335	}
336
337	err = check_pfn_span(pfn, nr_pages, "add");
338	if (err)
339		return err;
340
341	for (; pfn < end_pfn; pfn += cur_nr_pages) {
342		/* Select all remaining pages up to the next section boundary */
343		cur_nr_pages = min(end_pfn - pfn,
344				   SECTION_ALIGN_UP(pfn + 1) - pfn);
345		err = sparse_add_section(nid, pfn, cur_nr_pages, altmap);
346		if (err)
347			break;
348		cond_resched();
349	}
350	vmemmap_populate_print_last();
351	return err;
352}
353
354/* find the smallest valid pfn in the range [start_pfn, end_pfn) */
355static unsigned long find_smallest_section_pfn(int nid, struct zone *zone,
356				     unsigned long start_pfn,
357				     unsigned long end_pfn)
358{
359	for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SUBSECTION) {
360		if (unlikely(!pfn_to_online_page(start_pfn)))
361			continue;
362
363		if (unlikely(pfn_to_nid(start_pfn) != nid))
364			continue;
365
366		if (zone != page_zone(pfn_to_page(start_pfn)))
367			continue;
368
369		return start_pfn;
370	}
371
372	return 0;
373}
374
375/* find the biggest valid pfn in the range [start_pfn, end_pfn). */
376static unsigned long find_biggest_section_pfn(int nid, struct zone *zone,
377				    unsigned long start_pfn,
378				    unsigned long end_pfn)
379{
380	unsigned long pfn;
381
382	/* pfn is the end pfn of a memory section. */
383	pfn = end_pfn - 1;
384	for (; pfn >= start_pfn; pfn -= PAGES_PER_SUBSECTION) {
385		if (unlikely(!pfn_to_online_page(pfn)))
386			continue;
387
388		if (unlikely(pfn_to_nid(pfn) != nid))
389			continue;
390
391		if (zone != page_zone(pfn_to_page(pfn)))
392			continue;
393
394		return pfn;
395	}
396
397	return 0;
398}
399
400static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
401			     unsigned long end_pfn)
402{
403	unsigned long pfn;
404	int nid = zone_to_nid(zone);
405
406	zone_span_writelock(zone);
407	if (zone->zone_start_pfn == start_pfn) {
408		/*
409		 * If the section is smallest section in the zone, it need
410		 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
411		 * In this case, we find second smallest valid mem_section
412		 * for shrinking zone.
413		 */
414		pfn = find_smallest_section_pfn(nid, zone, end_pfn,
415						zone_end_pfn(zone));
416		if (pfn) {
417			zone->spanned_pages = zone_end_pfn(zone) - pfn;
418			zone->zone_start_pfn = pfn;
419		} else {
420			zone->zone_start_pfn = 0;
421			zone->spanned_pages = 0;
422		}
423	} else if (zone_end_pfn(zone) == end_pfn) {
424		/*
425		 * If the section is biggest section in the zone, it need
426		 * shrink zone->spanned_pages.
427		 * In this case, we find second biggest valid mem_section for
428		 * shrinking zone.
429		 */
430		pfn = find_biggest_section_pfn(nid, zone, zone->zone_start_pfn,
431					       start_pfn);
432		if (pfn)
433			zone->spanned_pages = pfn - zone->zone_start_pfn + 1;
434		else {
435			zone->zone_start_pfn = 0;
436			zone->spanned_pages = 0;
437		}
438	}
439	zone_span_writeunlock(zone);
440}
441
442static void update_pgdat_span(struct pglist_data *pgdat)
443{
444	unsigned long node_start_pfn = 0, node_end_pfn = 0;
445	struct zone *zone;
446
447	for (zone = pgdat->node_zones;
448	     zone < pgdat->node_zones + MAX_NR_ZONES; zone++) {
449		unsigned long zone_end_pfn = zone->zone_start_pfn +
450					     zone->spanned_pages;
451
452		/* No need to lock the zones, they can't change. */
453		if (!zone->spanned_pages)
454			continue;
455		if (!node_end_pfn) {
456			node_start_pfn = zone->zone_start_pfn;
457			node_end_pfn = zone_end_pfn;
458			continue;
459		}
460
461		if (zone_end_pfn > node_end_pfn)
462			node_end_pfn = zone_end_pfn;
463		if (zone->zone_start_pfn < node_start_pfn)
464			node_start_pfn = zone->zone_start_pfn;
465	}
466
467	pgdat->node_start_pfn = node_start_pfn;
468	pgdat->node_spanned_pages = node_end_pfn - node_start_pfn;
469}
470
471void __ref remove_pfn_range_from_zone(struct zone *zone,
472				      unsigned long start_pfn,
473				      unsigned long nr_pages)
474{
475	const unsigned long end_pfn = start_pfn + nr_pages;
476	struct pglist_data *pgdat = zone->zone_pgdat;
477	unsigned long pfn, cur_nr_pages, flags;
478
479	/* Poison struct pages because they are now uninitialized again. */
480	for (pfn = start_pfn; pfn < end_pfn; pfn += cur_nr_pages) {
481		cond_resched();
482
483		/* Select all remaining pages up to the next section boundary */
484		cur_nr_pages =
485			min(end_pfn - pfn, SECTION_ALIGN_UP(pfn + 1) - pfn);
486		page_init_poison(pfn_to_page(pfn),
487				 sizeof(struct page) * cur_nr_pages);
488	}
489
490#ifdef CONFIG_ZONE_DEVICE
491	/*
492	 * Zone shrinking code cannot properly deal with ZONE_DEVICE. So
493	 * we will not try to shrink the zones - which is okay as
494	 * set_zone_contiguous() cannot deal with ZONE_DEVICE either way.
495	 */
496	if (zone_idx(zone) == ZONE_DEVICE)
497		return;
498#endif
499
500	clear_zone_contiguous(zone);
501
502	pgdat_resize_lock(zone->zone_pgdat, &flags);
503	shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
504	update_pgdat_span(pgdat);
505	pgdat_resize_unlock(zone->zone_pgdat, &flags);
506
507	set_zone_contiguous(zone);
508}
509
510static void __remove_section(unsigned long pfn, unsigned long nr_pages,
511			     unsigned long map_offset,
512			     struct vmem_altmap *altmap)
513{
514	struct mem_section *ms = __pfn_to_section(pfn);
515
516	if (WARN_ON_ONCE(!valid_section(ms)))
517		return;
518
519	sparse_remove_section(ms, pfn, nr_pages, map_offset, altmap);
520}
521
522/**
523 * __remove_pages() - remove sections of pages
524 * @pfn: starting pageframe (must be aligned to start of a section)
525 * @nr_pages: number of pages to remove (must be multiple of section size)
526 * @altmap: alternative device page map or %NULL if default memmap is used
527 *
528 * Generic helper function to remove section mappings and sysfs entries
529 * for the section of the memory we are removing. Caller needs to make
530 * sure that pages are marked reserved and zones are adjust properly by
531 * calling offline_pages().
532 */
533void __remove_pages(unsigned long pfn, unsigned long nr_pages,
534		    struct vmem_altmap *altmap)
535{
536	const unsigned long end_pfn = pfn + nr_pages;
537	unsigned long cur_nr_pages;
538	unsigned long map_offset = 0;
539
540	map_offset = vmem_altmap_offset(altmap);
541
542	if (check_pfn_span(pfn, nr_pages, "remove"))
543		return;
544
545	for (; pfn < end_pfn; pfn += cur_nr_pages) {
546		cond_resched();
547		/* Select all remaining pages up to the next section boundary */
548		cur_nr_pages = min(end_pfn - pfn,
549				   SECTION_ALIGN_UP(pfn + 1) - pfn);
550		__remove_section(pfn, cur_nr_pages, map_offset, altmap);
551		map_offset = 0;
552	}
553}
554
555int set_online_page_callback(online_page_callback_t callback)
556{
557	int rc = -EINVAL;
558
559	get_online_mems();
560	mutex_lock(&online_page_callback_lock);
561
562	if (online_page_callback == generic_online_page) {
563		online_page_callback = callback;
564		rc = 0;
565	}
566
567	mutex_unlock(&online_page_callback_lock);
568	put_online_mems();
569
570	return rc;
571}
572EXPORT_SYMBOL_GPL(set_online_page_callback);
573
574int restore_online_page_callback(online_page_callback_t callback)
575{
576	int rc = -EINVAL;
577
578	get_online_mems();
579	mutex_lock(&online_page_callback_lock);
580
581	if (online_page_callback == callback) {
582		online_page_callback = generic_online_page;
583		rc = 0;
584	}
585
586	mutex_unlock(&online_page_callback_lock);
587	put_online_mems();
588
589	return rc;
590}
591EXPORT_SYMBOL_GPL(restore_online_page_callback);
592
593void generic_online_page(struct page *page, unsigned int order)
594{
595	/*
596	 * Freeing the page with debug_pagealloc enabled will try to unmap it,
597	 * so we should map it first. This is better than introducing a special
598	 * case in page freeing fast path.
599	 */
600	if (debug_pagealloc_enabled_static())
601		kernel_map_pages(page, 1 << order, 1);
602	__free_pages_core(page, order);
603	totalram_pages_add(1UL << order);
604#ifdef CONFIG_HIGHMEM
605	if (PageHighMem(page))
606		totalhigh_pages_add(1UL << order);
607#endif
608}
609EXPORT_SYMBOL_GPL(generic_online_page);
610
611static void online_pages_range(unsigned long start_pfn, unsigned long nr_pages)
612{
613	const unsigned long end_pfn = start_pfn + nr_pages;
614	unsigned long pfn;
615
616	/*
617	 * Online the pages in MAX_ORDER - 1 aligned chunks. The callback might
618	 * decide to not expose all pages to the buddy (e.g., expose them
619	 * later). We account all pages as being online and belonging to this
620	 * zone ("present").
621	 */
622	for (pfn = start_pfn; pfn < end_pfn; pfn += MAX_ORDER_NR_PAGES)
623		(*online_page_callback)(pfn_to_page(pfn), MAX_ORDER - 1);
624
625	/* mark all involved sections as online */
626	online_mem_sections(start_pfn, end_pfn);
627}
628
629/* check which state of node_states will be changed when online memory */
630static void node_states_check_changes_online(unsigned long nr_pages,
631	struct zone *zone, struct memory_notify *arg)
632{
633	int nid = zone_to_nid(zone);
634
635	arg->status_change_nid = NUMA_NO_NODE;
636	arg->status_change_nid_normal = NUMA_NO_NODE;
637	arg->status_change_nid_high = NUMA_NO_NODE;
638
639	if (!node_state(nid, N_MEMORY))
640		arg->status_change_nid = nid;
641	if (zone_idx(zone) <= ZONE_NORMAL && !node_state(nid, N_NORMAL_MEMORY))
642		arg->status_change_nid_normal = nid;
643#ifdef CONFIG_HIGHMEM
644	if (zone_idx(zone) <= ZONE_HIGHMEM && !node_state(nid, N_HIGH_MEMORY))
645		arg->status_change_nid_high = nid;
646#endif
647}
648
649static void node_states_set_node(int node, struct memory_notify *arg)
650{
651	if (arg->status_change_nid_normal >= 0)
652		node_set_state(node, N_NORMAL_MEMORY);
653
654	if (arg->status_change_nid_high >= 0)
655		node_set_state(node, N_HIGH_MEMORY);
656
657	if (arg->status_change_nid >= 0)
658		node_set_state(node, N_MEMORY);
659}
660
661static void __meminit resize_zone_range(struct zone *zone, unsigned long start_pfn,
662		unsigned long nr_pages)
663{
664	unsigned long old_end_pfn = zone_end_pfn(zone);
665
666	if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
667		zone->zone_start_pfn = start_pfn;
668
669	zone->spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - zone->zone_start_pfn;
670}
671
672static void __meminit resize_pgdat_range(struct pglist_data *pgdat, unsigned long start_pfn,
673                                     unsigned long nr_pages)
674{
675	unsigned long old_end_pfn = pgdat_end_pfn(pgdat);
676
677	if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
678		pgdat->node_start_pfn = start_pfn;
679
680	pgdat->node_spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - pgdat->node_start_pfn;
681
682}
683/*
684 * Associate the pfn range with the given zone, initializing the memmaps
685 * and resizing the pgdat/zone data to span the added pages. After this
686 * call, all affected pages are PG_reserved.
687 *
688 * All aligned pageblocks are initialized to the specified migratetype
689 * (usually MIGRATE_MOVABLE). Besides setting the migratetype, no related
690 * zone stats (e.g., nr_isolate_pageblock) are touched.
691 */
692void __ref move_pfn_range_to_zone(struct zone *zone, unsigned long start_pfn,
693				  unsigned long nr_pages,
694				  struct vmem_altmap *altmap, int migratetype)
695{
696	struct pglist_data *pgdat = zone->zone_pgdat;
697	int nid = pgdat->node_id;
698	unsigned long flags;
699
700	clear_zone_contiguous(zone);
701
702	/* TODO Huh pgdat is irqsave while zone is not. It used to be like that before */
703	pgdat_resize_lock(pgdat, &flags);
704	zone_span_writelock(zone);
705	if (zone_is_empty(zone))
706		init_currently_empty_zone(zone, start_pfn, nr_pages);
707	resize_zone_range(zone, start_pfn, nr_pages);
708	zone_span_writeunlock(zone);
709	resize_pgdat_range(pgdat, start_pfn, nr_pages);
710	pgdat_resize_unlock(pgdat, &flags);
711
712	/*
713	 * TODO now we have a visible range of pages which are not associated
714	 * with their zone properly. Not nice but set_pfnblock_flags_mask
715	 * expects the zone spans the pfn range. All the pages in the range
716	 * are reserved so nobody should be touching them so we should be safe
717	 */
718	memmap_init_zone(nr_pages, nid, zone_idx(zone), start_pfn, 0,
719			 MEMINIT_HOTPLUG, altmap, migratetype);
720
721	set_zone_contiguous(zone);
722}
723
724/*
725 * Returns a default kernel memory zone for the given pfn range.
726 * If no kernel zone covers this pfn range it will automatically go
727 * to the ZONE_NORMAL.
728 */
729static struct zone *default_kernel_zone_for_pfn(int nid, unsigned long start_pfn,
730		unsigned long nr_pages)
731{
732	struct pglist_data *pgdat = NODE_DATA(nid);
733	int zid;
734
735	for (zid = 0; zid <= ZONE_NORMAL; zid++) {
736		struct zone *zone = &pgdat->node_zones[zid];
737
738		if (zone_intersects(zone, start_pfn, nr_pages))
739			return zone;
740	}
741
742	return &pgdat->node_zones[ZONE_NORMAL];
743}
744
745static inline struct zone *default_zone_for_pfn(int nid, unsigned long start_pfn,
746		unsigned long nr_pages)
747{
748	struct zone *kernel_zone = default_kernel_zone_for_pfn(nid, start_pfn,
749			nr_pages);
750	struct zone *movable_zone = &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
751	bool in_kernel = zone_intersects(kernel_zone, start_pfn, nr_pages);
752	bool in_movable = zone_intersects(movable_zone, start_pfn, nr_pages);
753
754	/*
755	 * We inherit the existing zone in a simple case where zones do not
756	 * overlap in the given range
757	 */
758	if (in_kernel ^ in_movable)
759		return (in_kernel) ? kernel_zone : movable_zone;
760
761	/*
762	 * If the range doesn't belong to any zone or two zones overlap in the
763	 * given range then we use movable zone only if movable_node is
764	 * enabled because we always online to a kernel zone by default.
765	 */
766	return movable_node_enabled ? movable_zone : kernel_zone;
767}
768
769struct zone *zone_for_pfn_range(int online_type, int nid,
770		unsigned long start_pfn, unsigned long nr_pages)
771{
772	if (online_type == MMOP_ONLINE_KERNEL)
773		return default_kernel_zone_for_pfn(nid, start_pfn, nr_pages);
774
775	if (online_type == MMOP_ONLINE_MOVABLE)
776		return &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
777
778	return default_zone_for_pfn(nid, start_pfn, nr_pages);
779}
780
781int __ref online_pages(unsigned long pfn, unsigned long nr_pages,
782		       int online_type, int nid)
783{
784	unsigned long flags;
785	struct zone *zone;
786	int need_zonelists_rebuild = 0;
787	int ret;
788	struct memory_notify arg;
789
790	/* We can only online full sections (e.g., SECTION_IS_ONLINE) */
791	if (WARN_ON_ONCE(!nr_pages ||
792			 !IS_ALIGNED(pfn | nr_pages, PAGES_PER_SECTION)))
793		return -EINVAL;
794
795	mem_hotplug_begin();
796
797	/* associate pfn range with the zone */
798	zone = zone_for_pfn_range(online_type, nid, pfn, nr_pages);
799	move_pfn_range_to_zone(zone, pfn, nr_pages, NULL, MIGRATE_ISOLATE);
800
801	arg.start_pfn = pfn;
802	arg.nr_pages = nr_pages;
803	node_states_check_changes_online(nr_pages, zone, &arg);
804
805	ret = memory_notify(MEM_GOING_ONLINE, &arg);
806	ret = notifier_to_errno(ret);
807	if (ret)
808		goto failed_addition;
809
810	/*
811	 * Fixup the number of isolated pageblocks before marking the sections
812	 * onlining, such that undo_isolate_page_range() works correctly.
813	 */
814	spin_lock_irqsave(&zone->lock, flags);
815	zone->nr_isolate_pageblock += nr_pages / pageblock_nr_pages;
816	spin_unlock_irqrestore(&zone->lock, flags);
817
818	/*
819	 * If this zone is not populated, then it is not in zonelist.
820	 * This means the page allocator ignores this zone.
821	 * So, zonelist must be updated after online.
822	 */
823	if (!populated_zone(zone)) {
824		need_zonelists_rebuild = 1;
825		setup_zone_pageset(zone);
826	}
827
828	online_pages_range(pfn, nr_pages);
829	zone->present_pages += nr_pages;
830
831	pgdat_resize_lock(zone->zone_pgdat, &flags);
832	zone->zone_pgdat->node_present_pages += nr_pages;
833	pgdat_resize_unlock(zone->zone_pgdat, &flags);
834
835	node_states_set_node(nid, &arg);
836	if (need_zonelists_rebuild)
837		build_all_zonelists(NULL);
838	zone_pcp_update(zone);
839
840	/* Basic onlining is complete, allow allocation of onlined pages. */
841	undo_isolate_page_range(pfn, pfn + nr_pages, MIGRATE_MOVABLE);
842
843	/*
844	 * Freshly onlined pages aren't shuffled (e.g., all pages are placed to
845	 * the tail of the freelist when undoing isolation). Shuffle the whole
846	 * zone to make sure the just onlined pages are properly distributed
847	 * across the whole freelist - to create an initial shuffle.
848	 */
849	shuffle_zone(zone);
850
851	init_per_zone_wmark_min();
852
853	kswapd_run(nid);
854	kcompactd_run(nid);
855#ifdef CONFIG_HYPERHOLD_ZSWAPD
856	zswapd_run(nid);
857#endif
858
859	writeback_set_ratelimit();
860
861	memory_notify(MEM_ONLINE, &arg);
862	mem_hotplug_done();
863	return 0;
864
865failed_addition:
866	pr_debug("online_pages [mem %#010llx-%#010llx] failed\n",
867		 (unsigned long long) pfn << PAGE_SHIFT,
868		 (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1);
869	memory_notify(MEM_CANCEL_ONLINE, &arg);
870	remove_pfn_range_from_zone(zone, pfn, nr_pages);
871	mem_hotplug_done();
872	return ret;
873}
874#endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
875
876static void reset_node_present_pages(pg_data_t *pgdat)
877{
878	struct zone *z;
879
880	for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
881		z->present_pages = 0;
882
883	pgdat->node_present_pages = 0;
884}
885
886/* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
887static pg_data_t __ref *hotadd_new_pgdat(int nid)
888{
889	struct pglist_data *pgdat;
890
891	pgdat = NODE_DATA(nid);
892	if (!pgdat) {
893		pgdat = arch_alloc_nodedata(nid);
894		if (!pgdat)
895			return NULL;
896
897		pgdat->per_cpu_nodestats =
898			alloc_percpu(struct per_cpu_nodestat);
899		arch_refresh_nodedata(nid, pgdat);
900	} else {
901		int cpu;
902		/*
903		 * Reset the nr_zones, order and highest_zoneidx before reuse.
904		 * Note that kswapd will init kswapd_highest_zoneidx properly
905		 * when it starts in the near future.
906		 */
907		pgdat->nr_zones = 0;
908		pgdat->kswapd_order = 0;
909		pgdat->kswapd_highest_zoneidx = 0;
910		for_each_online_cpu(cpu) {
911			struct per_cpu_nodestat *p;
912
913			p = per_cpu_ptr(pgdat->per_cpu_nodestats, cpu);
914			memset(p, 0, sizeof(*p));
915		}
916	}
917
918	/* we can use NODE_DATA(nid) from here */
919	pgdat->node_id = nid;
920	pgdat->node_start_pfn = 0;
921
922	/* init node's zones as empty zones, we don't have any present pages.*/
923	free_area_init_core_hotplug(nid);
924
925	/*
926	 * The node we allocated has no zone fallback lists. For avoiding
927	 * to access not-initialized zonelist, build here.
928	 */
929	build_all_zonelists(pgdat);
930
931	/*
932	 * When memory is hot-added, all the memory is in offline state. So
933	 * clear all zones' present_pages because they will be updated in
934	 * online_pages() and offline_pages().
935	 */
936	reset_node_managed_pages(pgdat);
937	reset_node_present_pages(pgdat);
938
939	return pgdat;
940}
941
942static void rollback_node_hotadd(int nid)
943{
944	pg_data_t *pgdat = NODE_DATA(nid);
945
946	arch_refresh_nodedata(nid, NULL);
947	free_percpu(pgdat->per_cpu_nodestats);
948	arch_free_nodedata(pgdat);
949}
950
951
952/**
953 * try_online_node - online a node if offlined
954 * @nid: the node ID
955 * @set_node_online: Whether we want to online the node
956 * called by cpu_up() to online a node without onlined memory.
957 *
958 * Returns:
959 * 1 -> a new node has been allocated
960 * 0 -> the node is already online
961 * -ENOMEM -> the node could not be allocated
962 */
963static int __try_online_node(int nid, bool set_node_online)
964{
965	pg_data_t *pgdat;
966	int ret = 1;
967
968	if (node_online(nid))
969		return 0;
970
971	pgdat = hotadd_new_pgdat(nid);
972	if (!pgdat) {
973		pr_err("Cannot online node %d due to NULL pgdat\n", nid);
974		ret = -ENOMEM;
975		goto out;
976	}
977
978	if (set_node_online) {
979		node_set_online(nid);
980		ret = register_one_node(nid);
981		BUG_ON(ret);
982	}
983out:
984	return ret;
985}
986
987/*
988 * Users of this function always want to online/register the node
989 */
990int try_online_node(int nid)
991{
992	int ret;
993
994	mem_hotplug_begin();
995	ret =  __try_online_node(nid, true);
996	mem_hotplug_done();
997	return ret;
998}
999
1000static int check_hotplug_memory_range(u64 start, u64 size)
1001{
1002	/* memory range must be block size aligned */
1003	if (!size || !IS_ALIGNED(start, memory_block_size_bytes()) ||
1004	    !IS_ALIGNED(size, memory_block_size_bytes())) {
1005		pr_err("Block size [%#lx] unaligned hotplug range: start %#llx, size %#llx",
1006		       memory_block_size_bytes(), start, size);
1007		return -EINVAL;
1008	}
1009
1010	return 0;
1011}
1012
1013static int online_memory_block(struct memory_block *mem, void *arg)
1014{
1015	mem->online_type = memhp_default_online_type;
1016	return device_online(&mem->dev);
1017}
1018
1019/*
1020 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1021 * and online/offline operations (triggered e.g. by sysfs).
1022 *
1023 * we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG
1024 */
1025int __ref add_memory_resource(int nid, struct resource *res, mhp_t mhp_flags)
1026{
1027	struct mhp_params params = { .pgprot = pgprot_mhp(PAGE_KERNEL) };
1028	u64 start, size;
1029	bool new_node = false;
1030	int ret;
1031
1032	start = res->start;
1033	size = resource_size(res);
1034
1035	ret = check_hotplug_memory_range(start, size);
1036	if (ret)
1037		return ret;
1038
1039	if (!node_possible(nid)) {
1040		WARN(1, "node %d was absent from the node_possible_map\n", nid);
1041		return -EINVAL;
1042	}
1043
1044	mem_hotplug_begin();
1045
1046	if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK))
1047		memblock_add_node(start, size, nid);
1048
1049	ret = __try_online_node(nid, false);
1050	if (ret < 0)
1051		goto error;
1052	new_node = ret;
1053
1054	/* call arch's memory hotadd */
1055	ret = arch_add_memory(nid, start, size, &params);
1056	if (ret < 0)
1057		goto error;
1058
1059	/* create memory block devices after memory was added */
1060	ret = create_memory_block_devices(start, size);
1061	if (ret) {
1062		arch_remove_memory(nid, start, size, NULL);
1063		goto error;
1064	}
1065
1066	if (new_node) {
1067		/* If sysfs file of new node can't be created, cpu on the node
1068		 * can't be hot-added. There is no rollback way now.
1069		 * So, check by BUG_ON() to catch it reluctantly..
1070		 * We online node here. We can't roll back from here.
1071		 */
1072		node_set_online(nid);
1073		ret = __register_one_node(nid);
1074		BUG_ON(ret);
1075	}
1076
1077	/* link memory sections under this node.*/
1078	link_mem_sections(nid, PFN_DOWN(start), PFN_UP(start + size - 1),
1079			  MEMINIT_HOTPLUG);
1080
1081	/* create new memmap entry */
1082	if (!strcmp(res->name, "System RAM"))
1083		firmware_map_add_hotplug(start, start + size, "System RAM");
1084
1085	/* device_online() will take the lock when calling online_pages() */
1086	mem_hotplug_done();
1087
1088	/*
1089	 * In case we're allowed to merge the resource, flag it and trigger
1090	 * merging now that adding succeeded.
1091	 */
1092	if (mhp_flags & MEMHP_MERGE_RESOURCE)
1093		merge_system_ram_resource(res);
1094
1095	/* online pages if requested */
1096	if (memhp_default_online_type != MMOP_OFFLINE)
1097		walk_memory_blocks(start, size, NULL, online_memory_block);
1098
1099	return ret;
1100error:
1101	/* rollback pgdat allocation and others */
1102	if (new_node)
1103		rollback_node_hotadd(nid);
1104	if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK))
1105		memblock_remove(start, size);
1106	mem_hotplug_done();
1107	return ret;
1108}
1109
1110/* requires device_hotplug_lock, see add_memory_resource() */
1111int __ref __add_memory(int nid, u64 start, u64 size, mhp_t mhp_flags)
1112{
1113	struct resource *res;
1114	int ret;
1115
1116	res = register_memory_resource(start, size, "System RAM");
1117	if (IS_ERR(res))
1118		return PTR_ERR(res);
1119
1120	ret = add_memory_resource(nid, res, mhp_flags);
1121	if (ret < 0)
1122		release_memory_resource(res);
1123	return ret;
1124}
1125
1126int add_memory(int nid, u64 start, u64 size, mhp_t mhp_flags)
1127{
1128	int rc;
1129
1130	lock_device_hotplug();
1131	rc = __add_memory(nid, start, size, mhp_flags);
1132	unlock_device_hotplug();
1133
1134	return rc;
1135}
1136EXPORT_SYMBOL_GPL(add_memory);
1137
1138/*
1139 * Add special, driver-managed memory to the system as system RAM. Such
1140 * memory is not exposed via the raw firmware-provided memmap as system
1141 * RAM, instead, it is detected and added by a driver - during cold boot,
1142 * after a reboot, and after kexec.
1143 *
1144 * Reasons why this memory should not be used for the initial memmap of a
1145 * kexec kernel or for placing kexec images:
1146 * - The booting kernel is in charge of determining how this memory will be
1147 *   used (e.g., use persistent memory as system RAM)
1148 * - Coordination with a hypervisor is required before this memory
1149 *   can be used (e.g., inaccessible parts).
1150 *
1151 * For this memory, no entries in /sys/firmware/memmap ("raw firmware-provided
1152 * memory map") are created. Also, the created memory resource is flagged
1153 * with IORESOURCE_SYSRAM_DRIVER_MANAGED, so in-kernel users can special-case
1154 * this memory as well (esp., not place kexec images onto it).
1155 *
1156 * The resource_name (visible via /proc/iomem) has to have the format
1157 * "System RAM ($DRIVER)".
1158 */
1159int add_memory_driver_managed(int nid, u64 start, u64 size,
1160			      const char *resource_name, mhp_t mhp_flags)
1161{
1162	struct resource *res;
1163	int rc;
1164
1165	if (!resource_name ||
1166	    strstr(resource_name, "System RAM (") != resource_name ||
1167	    resource_name[strlen(resource_name) - 1] != ')')
1168		return -EINVAL;
1169
1170	lock_device_hotplug();
1171
1172	res = register_memory_resource(start, size, resource_name);
1173	if (IS_ERR(res)) {
1174		rc = PTR_ERR(res);
1175		goto out_unlock;
1176	}
1177
1178	rc = add_memory_resource(nid, res, mhp_flags);
1179	if (rc < 0)
1180		release_memory_resource(res);
1181
1182out_unlock:
1183	unlock_device_hotplug();
1184	return rc;
1185}
1186EXPORT_SYMBOL_GPL(add_memory_driver_managed);
1187
1188#ifdef CONFIG_MEMORY_HOTREMOVE
1189/*
1190 * Confirm all pages in a range [start, end) belong to the same zone (skipping
1191 * memory holes). When true, return the zone.
1192 */
1193struct zone *test_pages_in_a_zone(unsigned long start_pfn,
1194				  unsigned long end_pfn)
1195{
1196	unsigned long pfn, sec_end_pfn;
1197	struct zone *zone = NULL;
1198	struct page *page;
1199	int i;
1200	for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1);
1201	     pfn < end_pfn;
1202	     pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) {
1203		/* Make sure the memory section is present first */
1204		if (!present_section_nr(pfn_to_section_nr(pfn)))
1205			continue;
1206		for (; pfn < sec_end_pfn && pfn < end_pfn;
1207		     pfn += MAX_ORDER_NR_PAGES) {
1208			i = 0;
1209			/* This is just a CONFIG_HOLES_IN_ZONE check.*/
1210			while ((i < MAX_ORDER_NR_PAGES) &&
1211				!pfn_valid_within(pfn + i))
1212				i++;
1213			if (i == MAX_ORDER_NR_PAGES || pfn + i >= end_pfn)
1214				continue;
1215			/* Check if we got outside of the zone */
1216			if (zone && !zone_spans_pfn(zone, pfn + i))
1217				return NULL;
1218			page = pfn_to_page(pfn + i);
1219			if (zone && page_zone(page) != zone)
1220				return NULL;
1221			zone = page_zone(page);
1222		}
1223	}
1224
1225	return zone;
1226}
1227
1228/*
1229 * Scan pfn range [start,end) to find movable/migratable pages (LRU pages,
1230 * non-lru movable pages and hugepages). Will skip over most unmovable
1231 * pages (esp., pages that can be skipped when offlining), but bail out on
1232 * definitely unmovable pages.
1233 *
1234 * Returns:
1235 *	0 in case a movable page is found and movable_pfn was updated.
1236 *	-ENOENT in case no movable page was found.
1237 *	-EBUSY in case a definitely unmovable page was found.
1238 */
1239static int scan_movable_pages(unsigned long start, unsigned long end,
1240			      unsigned long *movable_pfn)
1241{
1242	unsigned long pfn;
1243
1244	for (pfn = start; pfn < end; pfn++) {
1245		struct page *page, *head;
1246		unsigned long skip;
1247
1248		if (!pfn_valid(pfn))
1249			continue;
1250		page = pfn_to_page(pfn);
1251		if (PageLRU(page))
1252			goto found;
1253		if (__PageMovable(page))
1254			goto found;
1255
1256		/*
1257		 * PageOffline() pages that are not marked __PageMovable() and
1258		 * have a reference count > 0 (after MEM_GOING_OFFLINE) are
1259		 * definitely unmovable. If their reference count would be 0,
1260		 * they could at least be skipped when offlining memory.
1261		 */
1262		if (PageOffline(page) && page_count(page))
1263			return -EBUSY;
1264
1265		if (!PageHuge(page))
1266			continue;
1267		head = compound_head(page);
1268		if (page_huge_active(head))
1269			goto found;
1270		skip = compound_nr(head) - (pfn - page_to_pfn(head));
1271		pfn += skip - 1;
1272	}
1273	return -ENOENT;
1274found:
1275	*movable_pfn = pfn;
1276	return 0;
1277}
1278
1279static int
1280do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1281{
1282	unsigned long pfn;
1283	struct page *page, *head;
1284	int ret = 0;
1285	LIST_HEAD(source);
1286	static DEFINE_RATELIMIT_STATE(migrate_rs, DEFAULT_RATELIMIT_INTERVAL,
1287				      DEFAULT_RATELIMIT_BURST);
1288
1289	for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1290		if (!pfn_valid(pfn))
1291			continue;
1292		page = pfn_to_page(pfn);
1293		head = compound_head(page);
1294
1295		if (PageHuge(page)) {
1296			pfn = page_to_pfn(head) + compound_nr(head) - 1;
1297			isolate_hugetlb(head, &source);
1298			continue;
1299		} else if (PageTransHuge(page))
1300			pfn = page_to_pfn(head) + thp_nr_pages(page) - 1;
1301
1302		/*
1303		 * HWPoison pages have elevated reference counts so the migration would
1304		 * fail on them. It also doesn't make any sense to migrate them in the
1305		 * first place. Still try to unmap such a page in case it is still mapped
1306		 * (e.g. current hwpoison implementation doesn't unmap KSM pages but keep
1307		 * the unmap as the catch all safety net).
1308		 */
1309		if (PageHWPoison(page)) {
1310			if (WARN_ON(PageLRU(page)))
1311				isolate_lru_page(page);
1312			if (page_mapped(page))
1313				try_to_unmap(page, TTU_IGNORE_MLOCK);
1314			continue;
1315		}
1316
1317		if (!get_page_unless_zero(page))
1318			continue;
1319		/*
1320		 * We can skip free pages. And we can deal with pages on
1321		 * LRU and non-lru movable pages.
1322		 */
1323		if (PageLRU(page))
1324			ret = isolate_lru_page(page);
1325		else
1326			ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE);
1327		if (!ret) { /* Success */
1328			list_add_tail(&page->lru, &source);
1329			if (!__PageMovable(page))
1330				inc_node_page_state(page, NR_ISOLATED_ANON +
1331						    page_is_file_lru(page));
1332
1333		} else {
1334			if (__ratelimit(&migrate_rs)) {
1335				pr_warn("failed to isolate pfn %lx\n", pfn);
1336				dump_page(page, "isolation failed");
1337			}
1338		}
1339		put_page(page);
1340	}
1341	if (!list_empty(&source)) {
1342		nodemask_t nmask = node_states[N_MEMORY];
1343		struct migration_target_control mtc = {
1344			.nmask = &nmask,
1345			.gfp_mask = GFP_USER | __GFP_MOVABLE | __GFP_RETRY_MAYFAIL,
1346		};
1347
1348		/*
1349		 * We have checked that migration range is on a single zone so
1350		 * we can use the nid of the first page to all the others.
1351		 */
1352		mtc.nid = page_to_nid(list_first_entry(&source, struct page, lru));
1353
1354		/*
1355		 * try to allocate from a different node but reuse this node
1356		 * if there are no other online nodes to be used (e.g. we are
1357		 * offlining a part of the only existing node)
1358		 */
1359		node_clear(mtc.nid, nmask);
1360		if (nodes_empty(nmask))
1361			node_set(mtc.nid, nmask);
1362		ret = migrate_pages(&source, alloc_migration_target, NULL,
1363			(unsigned long)&mtc, MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
1364		if (ret) {
1365			list_for_each_entry(page, &source, lru) {
1366				if (__ratelimit(&migrate_rs)) {
1367					pr_warn("migrating pfn %lx failed ret:%d\n",
1368						page_to_pfn(page), ret);
1369					dump_page(page, "migration failure");
1370				}
1371			}
1372			putback_movable_pages(&source);
1373		}
1374	}
1375
1376	return ret;
1377}
1378
1379static int __init cmdline_parse_movable_node(char *p)
1380{
1381	movable_node_enabled = true;
1382	return 0;
1383}
1384early_param("movable_node", cmdline_parse_movable_node);
1385
1386/* check which state of node_states will be changed when offline memory */
1387static void node_states_check_changes_offline(unsigned long nr_pages,
1388		struct zone *zone, struct memory_notify *arg)
1389{
1390	struct pglist_data *pgdat = zone->zone_pgdat;
1391	unsigned long present_pages = 0;
1392	enum zone_type zt;
1393
1394	arg->status_change_nid = NUMA_NO_NODE;
1395	arg->status_change_nid_normal = NUMA_NO_NODE;
1396	arg->status_change_nid_high = NUMA_NO_NODE;
1397
1398	/*
1399	 * Check whether node_states[N_NORMAL_MEMORY] will be changed.
1400	 * If the memory to be offline is within the range
1401	 * [0..ZONE_NORMAL], and it is the last present memory there,
1402	 * the zones in that range will become empty after the offlining,
1403	 * thus we can determine that we need to clear the node from
1404	 * node_states[N_NORMAL_MEMORY].
1405	 */
1406	for (zt = 0; zt <= ZONE_NORMAL; zt++)
1407		present_pages += pgdat->node_zones[zt].present_pages;
1408	if (zone_idx(zone) <= ZONE_NORMAL && nr_pages >= present_pages)
1409		arg->status_change_nid_normal = zone_to_nid(zone);
1410
1411#ifdef CONFIG_HIGHMEM
1412	/*
1413	 * node_states[N_HIGH_MEMORY] contains nodes which
1414	 * have normal memory or high memory.
1415	 * Here we add the present_pages belonging to ZONE_HIGHMEM.
1416	 * If the zone is within the range of [0..ZONE_HIGHMEM), and
1417	 * we determine that the zones in that range become empty,
1418	 * we need to clear the node for N_HIGH_MEMORY.
1419	 */
1420	present_pages += pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1421	if (zone_idx(zone) <= ZONE_HIGHMEM && nr_pages >= present_pages)
1422		arg->status_change_nid_high = zone_to_nid(zone);
1423#endif
1424
1425	/*
1426	 * We have accounted the pages from [0..ZONE_NORMAL), and
1427	 * in case of CONFIG_HIGHMEM the pages from ZONE_HIGHMEM
1428	 * as well.
1429	 * Here we count the possible pages from ZONE_MOVABLE.
1430	 * If after having accounted all the pages, we see that the nr_pages
1431	 * to be offlined is over or equal to the accounted pages,
1432	 * we know that the node will become empty, and so, we can clear
1433	 * it for N_MEMORY as well.
1434	 */
1435	present_pages += pgdat->node_zones[ZONE_MOVABLE].present_pages;
1436
1437	if (nr_pages >= present_pages)
1438		arg->status_change_nid = zone_to_nid(zone);
1439}
1440
1441static void node_states_clear_node(int node, struct memory_notify *arg)
1442{
1443	if (arg->status_change_nid_normal >= 0)
1444		node_clear_state(node, N_NORMAL_MEMORY);
1445
1446	if (arg->status_change_nid_high >= 0)
1447		node_clear_state(node, N_HIGH_MEMORY);
1448
1449	if (arg->status_change_nid >= 0)
1450		node_clear_state(node, N_MEMORY);
1451}
1452
1453static int count_system_ram_pages_cb(unsigned long start_pfn,
1454				     unsigned long nr_pages, void *data)
1455{
1456	unsigned long *nr_system_ram_pages = data;
1457
1458	*nr_system_ram_pages += nr_pages;
1459	return 0;
1460}
1461
1462int __ref offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1463{
1464	const unsigned long end_pfn = start_pfn + nr_pages;
1465	unsigned long pfn, system_ram_pages = 0;
1466	unsigned long flags;
1467	struct zone *zone;
1468	struct memory_notify arg;
1469	int ret, node;
1470	char *reason;
1471
1472	/* We can only offline full sections (e.g., SECTION_IS_ONLINE) */
1473	if (WARN_ON_ONCE(!nr_pages ||
1474			 !IS_ALIGNED(start_pfn | nr_pages, PAGES_PER_SECTION)))
1475		return -EINVAL;
1476
1477	mem_hotplug_begin();
1478
1479	/*
1480	 * Don't allow to offline memory blocks that contain holes.
1481	 * Consequently, memory blocks with holes can never get onlined
1482	 * via the hotplug path - online_pages() - as hotplugged memory has
1483	 * no holes. This way, we e.g., don't have to worry about marking
1484	 * memory holes PG_reserved, don't need pfn_valid() checks, and can
1485	 * avoid using walk_system_ram_range() later.
1486	 */
1487	walk_system_ram_range(start_pfn, nr_pages, &system_ram_pages,
1488			      count_system_ram_pages_cb);
1489	if (system_ram_pages != nr_pages) {
1490		ret = -EINVAL;
1491		reason = "memory holes";
1492		goto failed_removal;
1493	}
1494
1495	/* This makes hotplug much easier...and readable.
1496	   we assume this for now. .*/
1497	zone = test_pages_in_a_zone(start_pfn, end_pfn);
1498	if (!zone) {
1499		ret = -EINVAL;
1500		reason = "multizone range";
1501		goto failed_removal;
1502	}
1503	node = zone_to_nid(zone);
1504
1505	/* set above range as isolated */
1506	ret = start_isolate_page_range(start_pfn, end_pfn,
1507				       MIGRATE_MOVABLE,
1508				       MEMORY_OFFLINE | REPORT_FAILURE);
1509	if (ret) {
1510		reason = "failure to isolate range";
1511		goto failed_removal;
1512	}
1513
1514	arg.start_pfn = start_pfn;
1515	arg.nr_pages = nr_pages;
1516	node_states_check_changes_offline(nr_pages, zone, &arg);
1517
1518	ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1519	ret = notifier_to_errno(ret);
1520	if (ret) {
1521		reason = "notifier failure";
1522		goto failed_removal_isolated;
1523	}
1524
1525	do {
1526		pfn = start_pfn;
1527		do {
1528			if (signal_pending(current)) {
1529				ret = -EINTR;
1530				reason = "signal backoff";
1531				goto failed_removal_isolated;
1532			}
1533
1534			cond_resched();
1535			lru_add_drain_all();
1536
1537			ret = scan_movable_pages(pfn, end_pfn, &pfn);
1538			if (!ret) {
1539				/*
1540				 * TODO: fatal migration failures should bail
1541				 * out
1542				 */
1543				do_migrate_range(pfn, end_pfn);
1544			}
1545		} while (!ret);
1546
1547		if (ret != -ENOENT) {
1548			reason = "unmovable page";
1549			goto failed_removal_isolated;
1550		}
1551
1552		/*
1553		 * Dissolve free hugepages in the memory block before doing
1554		 * offlining actually in order to make hugetlbfs's object
1555		 * counting consistent.
1556		 */
1557		ret = dissolve_free_huge_pages(start_pfn, end_pfn);
1558		if (ret) {
1559			reason = "failure to dissolve huge pages";
1560			goto failed_removal_isolated;
1561		}
1562
1563		/*
1564		 * per-cpu pages are drained in start_isolate_page_range, but if
1565		 * there are still pages that are not free, make sure that we
1566		 * drain again, because when we isolated range we might
1567		 * have raced with another thread that was adding pages to pcp
1568		 * list.
1569		 *
1570		 * Forward progress should be still guaranteed because
1571		 * pages on the pcp list can only belong to MOVABLE_ZONE
1572		 * because has_unmovable_pages explicitly checks for
1573		 * PageBuddy on freed pages on other zones.
1574		 */
1575		ret = test_pages_isolated(start_pfn, end_pfn, MEMORY_OFFLINE);
1576		if (ret)
1577			drain_all_pages(zone);
1578	} while (ret);
1579
1580	/* Mark all sections offline and remove free pages from the buddy. */
1581	__offline_isolated_pages(start_pfn, end_pfn);
1582	pr_info("Offlined Pages %ld\n", nr_pages);
1583
1584	/*
1585	 * The memory sections are marked offline, and the pageblock flags
1586	 * effectively stale; nobody should be touching them. Fixup the number
1587	 * of isolated pageblocks, memory onlining will properly revert this.
1588	 */
1589	spin_lock_irqsave(&zone->lock, flags);
1590	zone->nr_isolate_pageblock -= nr_pages / pageblock_nr_pages;
1591	spin_unlock_irqrestore(&zone->lock, flags);
1592
1593	/* removal success */
1594	adjust_managed_page_count(pfn_to_page(start_pfn), -nr_pages);
1595	zone->present_pages -= nr_pages;
1596
1597	pgdat_resize_lock(zone->zone_pgdat, &flags);
1598	zone->zone_pgdat->node_present_pages -= nr_pages;
1599	pgdat_resize_unlock(zone->zone_pgdat, &flags);
1600
1601	init_per_zone_wmark_min();
1602
1603	if (!populated_zone(zone)) {
1604		zone_pcp_reset(zone);
1605		build_all_zonelists(NULL);
1606	} else
1607		zone_pcp_update(zone);
1608
1609	node_states_clear_node(node, &arg);
1610	if (arg.status_change_nid >= 0) {
1611		kswapd_stop(node);
1612		kcompactd_stop(node);
1613#ifdef CONFIG_HYPERHOLD_ZSWAPD
1614		zswapd_stop(node);
1615#endif
1616	}
1617
1618	writeback_set_ratelimit();
1619
1620	memory_notify(MEM_OFFLINE, &arg);
1621	remove_pfn_range_from_zone(zone, start_pfn, nr_pages);
1622	mem_hotplug_done();
1623	return 0;
1624
1625failed_removal_isolated:
1626	undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1627	memory_notify(MEM_CANCEL_OFFLINE, &arg);
1628failed_removal:
1629	pr_debug("memory offlining [mem %#010llx-%#010llx] failed due to %s\n",
1630		 (unsigned long long) start_pfn << PAGE_SHIFT,
1631		 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1,
1632		 reason);
1633	/* pushback to free area */
1634	mem_hotplug_done();
1635	return ret;
1636}
1637
1638static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
1639{
1640	int ret = !is_memblock_offlined(mem);
1641
1642	if (unlikely(ret)) {
1643		phys_addr_t beginpa, endpa;
1644
1645		beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
1646		endpa = beginpa + memory_block_size_bytes() - 1;
1647		pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n",
1648			&beginpa, &endpa);
1649
1650		return -EBUSY;
1651	}
1652	return 0;
1653}
1654
1655static int check_cpu_on_node(pg_data_t *pgdat)
1656{
1657	int cpu;
1658
1659	for_each_present_cpu(cpu) {
1660		if (cpu_to_node(cpu) == pgdat->node_id)
1661			/*
1662			 * the cpu on this node isn't removed, and we can't
1663			 * offline this node.
1664			 */
1665			return -EBUSY;
1666	}
1667
1668	return 0;
1669}
1670
1671static int check_no_memblock_for_node_cb(struct memory_block *mem, void *arg)
1672{
1673	int nid = *(int *)arg;
1674
1675	/*
1676	 * If a memory block belongs to multiple nodes, the stored nid is not
1677	 * reliable. However, such blocks are always online (e.g., cannot get
1678	 * offlined) and, therefore, are still spanned by the node.
1679	 */
1680	return mem->nid == nid ? -EEXIST : 0;
1681}
1682
1683/**
1684 * try_offline_node
1685 * @nid: the node ID
1686 *
1687 * Offline a node if all memory sections and cpus of the node are removed.
1688 *
1689 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1690 * and online/offline operations before this call.
1691 */
1692void try_offline_node(int nid)
1693{
1694	pg_data_t *pgdat = NODE_DATA(nid);
1695	int rc;
1696
1697	/*
1698	 * If the node still spans pages (especially ZONE_DEVICE), don't
1699	 * offline it. A node spans memory after move_pfn_range_to_zone(),
1700	 * e.g., after the memory block was onlined.
1701	 */
1702	if (pgdat->node_spanned_pages)
1703		return;
1704
1705	/*
1706	 * Especially offline memory blocks might not be spanned by the
1707	 * node. They will get spanned by the node once they get onlined.
1708	 * However, they link to the node in sysfs and can get onlined later.
1709	 */
1710	rc = for_each_memory_block(&nid, check_no_memblock_for_node_cb);
1711	if (rc)
1712		return;
1713
1714	if (check_cpu_on_node(pgdat))
1715		return;
1716
1717	/*
1718	 * all memory/cpu of this node are removed, we can offline this
1719	 * node now.
1720	 */
1721	node_set_offline(nid);
1722	unregister_one_node(nid);
1723}
1724EXPORT_SYMBOL(try_offline_node);
1725
1726static int __ref try_remove_memory(int nid, u64 start, u64 size)
1727{
1728	int rc = 0;
1729
1730	BUG_ON(check_hotplug_memory_range(start, size));
1731
1732	/*
1733	 * All memory blocks must be offlined before removing memory.  Check
1734	 * whether all memory blocks in question are offline and return error
1735	 * if this is not the case.
1736	 */
1737	rc = walk_memory_blocks(start, size, NULL, check_memblock_offlined_cb);
1738	if (rc)
1739		return rc;
1740
1741	/* remove memmap entry */
1742	firmware_map_remove(start, start + size, "System RAM");
1743
1744	/*
1745	 * Memory block device removal under the device_hotplug_lock is
1746	 * a barrier against racing online attempts.
1747	 */
1748	remove_memory_block_devices(start, size);
1749
1750	mem_hotplug_begin();
1751
1752	arch_remove_memory(nid, start, size, NULL);
1753
1754	if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK)) {
1755		memblock_free(start, size);
1756		memblock_remove(start, size);
1757	}
1758
1759	release_mem_region_adjustable(start, size);
1760
1761	try_offline_node(nid);
1762
1763	mem_hotplug_done();
1764	return 0;
1765}
1766
1767/**
1768 * remove_memory
1769 * @nid: the node ID
1770 * @start: physical address of the region to remove
1771 * @size: size of the region to remove
1772 *
1773 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1774 * and online/offline operations before this call, as required by
1775 * try_offline_node().
1776 */
1777void __remove_memory(int nid, u64 start, u64 size)
1778{
1779
1780	/*
1781	 * trigger BUG() if some memory is not offlined prior to calling this
1782	 * function
1783	 */
1784	if (try_remove_memory(nid, start, size))
1785		BUG();
1786}
1787
1788/*
1789 * Remove memory if every memory block is offline, otherwise return -EBUSY is
1790 * some memory is not offline
1791 */
1792int remove_memory(int nid, u64 start, u64 size)
1793{
1794	int rc;
1795
1796	lock_device_hotplug();
1797	rc  = try_remove_memory(nid, start, size);
1798	unlock_device_hotplug();
1799
1800	return rc;
1801}
1802EXPORT_SYMBOL_GPL(remove_memory);
1803
1804static int try_offline_memory_block(struct memory_block *mem, void *arg)
1805{
1806	uint8_t online_type = MMOP_ONLINE_KERNEL;
1807	uint8_t **online_types = arg;
1808	struct page *page;
1809	int rc;
1810
1811	/*
1812	 * Sense the online_type via the zone of the memory block. Offlining
1813	 * with multiple zones within one memory block will be rejected
1814	 * by offlining code ... so we don't care about that.
1815	 */
1816	page = pfn_to_online_page(section_nr_to_pfn(mem->start_section_nr));
1817	if (page && zone_idx(page_zone(page)) == ZONE_MOVABLE)
1818		online_type = MMOP_ONLINE_MOVABLE;
1819
1820	rc = device_offline(&mem->dev);
1821	/*
1822	 * Default is MMOP_OFFLINE - change it only if offlining succeeded,
1823	 * so try_reonline_memory_block() can do the right thing.
1824	 */
1825	if (!rc)
1826		**online_types = online_type;
1827
1828	(*online_types)++;
1829	/* Ignore if already offline. */
1830	return rc < 0 ? rc : 0;
1831}
1832
1833static int try_reonline_memory_block(struct memory_block *mem, void *arg)
1834{
1835	uint8_t **online_types = arg;
1836	int rc;
1837
1838	if (**online_types != MMOP_OFFLINE) {
1839		mem->online_type = **online_types;
1840		rc = device_online(&mem->dev);
1841		if (rc < 0)
1842			pr_warn("%s: Failed to re-online memory: %d",
1843				__func__, rc);
1844	}
1845
1846	/* Continue processing all remaining memory blocks. */
1847	(*online_types)++;
1848	return 0;
1849}
1850
1851/*
1852 * Try to offline and remove memory. Might take a long time to finish in case
1853 * memory is still in use. Primarily useful for memory devices that logically
1854 * unplugged all memory (so it's no longer in use) and want to offline + remove
1855 * that memory.
1856 */
1857int offline_and_remove_memory(int nid, u64 start, u64 size)
1858{
1859	const unsigned long mb_count = size / memory_block_size_bytes();
1860	uint8_t *online_types, *tmp;
1861	int rc;
1862
1863	if (!IS_ALIGNED(start, memory_block_size_bytes()) ||
1864	    !IS_ALIGNED(size, memory_block_size_bytes()) || !size)
1865		return -EINVAL;
1866
1867	/*
1868	 * We'll remember the old online type of each memory block, so we can
1869	 * try to revert whatever we did when offlining one memory block fails
1870	 * after offlining some others succeeded.
1871	 */
1872	online_types = kmalloc_array(mb_count, sizeof(*online_types),
1873				     GFP_KERNEL);
1874	if (!online_types)
1875		return -ENOMEM;
1876	/*
1877	 * Initialize all states to MMOP_OFFLINE, so when we abort processing in
1878	 * try_offline_memory_block(), we'll skip all unprocessed blocks in
1879	 * try_reonline_memory_block().
1880	 */
1881	memset(online_types, MMOP_OFFLINE, mb_count);
1882
1883	lock_device_hotplug();
1884
1885	tmp = online_types;
1886	rc = walk_memory_blocks(start, size, &tmp, try_offline_memory_block);
1887
1888	/*
1889	 * In case we succeeded to offline all memory, remove it.
1890	 * This cannot fail as it cannot get onlined in the meantime.
1891	 */
1892	if (!rc) {
1893		rc = try_remove_memory(nid, start, size);
1894		if (rc)
1895			pr_err("%s: Failed to remove memory: %d", __func__, rc);
1896	}
1897
1898	/*
1899	 * Rollback what we did. While memory onlining might theoretically fail
1900	 * (nacked by a notifier), it barely ever happens.
1901	 */
1902	if (rc) {
1903		tmp = online_types;
1904		walk_memory_blocks(start, size, &tmp,
1905				   try_reonline_memory_block);
1906	}
1907	unlock_device_hotplug();
1908
1909	kfree(online_types);
1910	return rc;
1911}
1912EXPORT_SYMBOL_GPL(offline_and_remove_memory);
1913#endif /* CONFIG_MEMORY_HOTREMOVE */
1914