xref: /openbmc/linux/mm/memory_hotplug.c (revision b4f473cf)
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 int memhp_default_online_type = MMOP_OFFLINE;
71 #else
72 int memhp_default_online_type = MMOP_ONLINE;
73 #endif
74 
75 static int __init setup_memhp_default_state(char *str)
76 {
77 	const int online_type = memhp_online_type_from_str(str);
78 
79 	if (online_type >= 0)
80 		memhp_default_online_type = online_type;
81 
82 	return 1;
83 }
84 __setup("memhp_default_state=", setup_memhp_default_state);
85 
86 void mem_hotplug_begin(void)
87 {
88 	cpus_read_lock();
89 	percpu_down_write(&mem_hotplug_lock);
90 }
91 
92 void mem_hotplug_done(void)
93 {
94 	percpu_up_write(&mem_hotplug_lock);
95 	cpus_read_unlock();
96 }
97 
98 u64 max_mem_size = U64_MAX;
99 
100 /* add this memory to iomem resource */
101 static struct resource *register_memory_resource(u64 start, u64 size,
102 						 const char *resource_name)
103 {
104 	struct resource *res;
105 	unsigned long flags =  IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
106 
107 	if (strcmp(resource_name, "System RAM"))
108 		flags |= IORESOURCE_SYSRAM_DRIVER_MANAGED;
109 
110 	/*
111 	 * Make sure value parsed from 'mem=' only restricts memory adding
112 	 * while booting, so that memory hotplug won't be impacted. Please
113 	 * refer to document of 'mem=' in kernel-parameters.txt for more
114 	 * details.
115 	 */
116 	if (start + size > max_mem_size && system_state < SYSTEM_RUNNING)
117 		return ERR_PTR(-E2BIG);
118 
119 	/*
120 	 * Request ownership of the new memory range.  This might be
121 	 * a child of an existing resource that was present but
122 	 * not marked as busy.
123 	 */
124 	res = __request_region(&iomem_resource, start, size,
125 			       resource_name, flags);
126 
127 	if (!res) {
128 		pr_debug("Unable to reserve System RAM region: %016llx->%016llx\n",
129 				start, start + size);
130 		return ERR_PTR(-EEXIST);
131 	}
132 	return res;
133 }
134 
135 static void release_memory_resource(struct resource *res)
136 {
137 	if (!res)
138 		return;
139 	release_resource(res);
140 	kfree(res);
141 }
142 
143 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
144 void get_page_bootmem(unsigned long info,  struct page *page,
145 		      unsigned long type)
146 {
147 	page->freelist = (void *)type;
148 	SetPagePrivate(page);
149 	set_page_private(page, info);
150 	page_ref_inc(page);
151 }
152 
153 void put_page_bootmem(struct page *page)
154 {
155 	unsigned long type;
156 
157 	type = (unsigned long) page->freelist;
158 	BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE ||
159 	       type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE);
160 
161 	if (page_ref_dec_return(page) == 1) {
162 		page->freelist = NULL;
163 		ClearPagePrivate(page);
164 		set_page_private(page, 0);
165 		INIT_LIST_HEAD(&page->lru);
166 		free_reserved_page(page);
167 	}
168 }
169 
170 #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
171 #ifndef CONFIG_SPARSEMEM_VMEMMAP
172 static void register_page_bootmem_info_section(unsigned long start_pfn)
173 {
174 	unsigned long mapsize, section_nr, i;
175 	struct mem_section *ms;
176 	struct page *page, *memmap;
177 	struct mem_section_usage *usage;
178 
179 	section_nr = pfn_to_section_nr(start_pfn);
180 	ms = __nr_to_section(section_nr);
181 
182 	/* Get section's memmap address */
183 	memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
184 
185 	/*
186 	 * Get page for the memmap's phys address
187 	 * XXX: need more consideration for sparse_vmemmap...
188 	 */
189 	page = virt_to_page(memmap);
190 	mapsize = sizeof(struct page) * PAGES_PER_SECTION;
191 	mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
192 
193 	/* remember memmap's page */
194 	for (i = 0; i < mapsize; i++, page++)
195 		get_page_bootmem(section_nr, page, SECTION_INFO);
196 
197 	usage = ms->usage;
198 	page = virt_to_page(usage);
199 
200 	mapsize = PAGE_ALIGN(mem_section_usage_size()) >> PAGE_SHIFT;
201 
202 	for (i = 0; i < mapsize; i++, page++)
203 		get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
204 
205 }
206 #else /* CONFIG_SPARSEMEM_VMEMMAP */
207 static void register_page_bootmem_info_section(unsigned long start_pfn)
208 {
209 	unsigned long mapsize, section_nr, i;
210 	struct mem_section *ms;
211 	struct page *page, *memmap;
212 	struct mem_section_usage *usage;
213 
214 	section_nr = pfn_to_section_nr(start_pfn);
215 	ms = __nr_to_section(section_nr);
216 
217 	memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
218 
219 	register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION);
220 
221 	usage = ms->usage;
222 	page = virt_to_page(usage);
223 
224 	mapsize = PAGE_ALIGN(mem_section_usage_size()) >> PAGE_SHIFT;
225 
226 	for (i = 0; i < mapsize; i++, page++)
227 		get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
228 }
229 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
230 
231 void __init register_page_bootmem_info_node(struct pglist_data *pgdat)
232 {
233 	unsigned long i, pfn, end_pfn, nr_pages;
234 	int node = pgdat->node_id;
235 	struct page *page;
236 
237 	nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
238 	page = virt_to_page(pgdat);
239 
240 	for (i = 0; i < nr_pages; i++, page++)
241 		get_page_bootmem(node, page, NODE_INFO);
242 
243 	pfn = pgdat->node_start_pfn;
244 	end_pfn = pgdat_end_pfn(pgdat);
245 
246 	/* register section info */
247 	for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
248 		/*
249 		 * Some platforms can assign the same pfn to multiple nodes - on
250 		 * node0 as well as nodeN.  To avoid registering a pfn against
251 		 * multiple nodes we check that this pfn does not already
252 		 * reside in some other nodes.
253 		 */
254 		if (pfn_valid(pfn) && (early_pfn_to_nid(pfn) == node))
255 			register_page_bootmem_info_section(pfn);
256 	}
257 }
258 #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
259 
260 static int check_pfn_span(unsigned long pfn, unsigned long nr_pages,
261 		const char *reason)
262 {
263 	/*
264 	 * Disallow all operations smaller than a sub-section and only
265 	 * allow operations smaller than a section for
266 	 * SPARSEMEM_VMEMMAP. Note that check_hotplug_memory_range()
267 	 * enforces a larger memory_block_size_bytes() granularity for
268 	 * memory that will be marked online, so this check should only
269 	 * fire for direct arch_{add,remove}_memory() users outside of
270 	 * add_memory_resource().
271 	 */
272 	unsigned long min_align;
273 
274 	if (IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP))
275 		min_align = PAGES_PER_SUBSECTION;
276 	else
277 		min_align = PAGES_PER_SECTION;
278 	if (!IS_ALIGNED(pfn, min_align)
279 			|| !IS_ALIGNED(nr_pages, min_align)) {
280 		WARN(1, "Misaligned __%s_pages start: %#lx end: #%lx\n",
281 				reason, pfn, pfn + nr_pages - 1);
282 		return -EINVAL;
283 	}
284 	return 0;
285 }
286 
287 static int check_hotplug_memory_addressable(unsigned long pfn,
288 					    unsigned long nr_pages)
289 {
290 	const u64 max_addr = PFN_PHYS(pfn + nr_pages) - 1;
291 
292 	if (max_addr >> MAX_PHYSMEM_BITS) {
293 		const u64 max_allowed = (1ull << (MAX_PHYSMEM_BITS + 1)) - 1;
294 		WARN(1,
295 		     "Hotplugged memory exceeds maximum addressable address, range=%#llx-%#llx, maximum=%#llx\n",
296 		     (u64)PFN_PHYS(pfn), max_addr, max_allowed);
297 		return -E2BIG;
298 	}
299 
300 	return 0;
301 }
302 
303 /*
304  * Reasonably generic function for adding memory.  It is
305  * expected that archs that support memory hotplug will
306  * call this function after deciding the zone to which to
307  * add the new pages.
308  */
309 int __ref __add_pages(int nid, unsigned long pfn, unsigned long nr_pages,
310 		struct mhp_params *params)
311 {
312 	const unsigned long end_pfn = pfn + nr_pages;
313 	unsigned long cur_nr_pages;
314 	int err;
315 	struct vmem_altmap *altmap = params->altmap;
316 
317 	if (WARN_ON_ONCE(!params->pgprot.pgprot))
318 		return -EINVAL;
319 
320 	err = check_hotplug_memory_addressable(pfn, nr_pages);
321 	if (err)
322 		return err;
323 
324 	if (altmap) {
325 		/*
326 		 * Validate altmap is within bounds of the total request
327 		 */
328 		if (altmap->base_pfn != pfn
329 				|| vmem_altmap_offset(altmap) > nr_pages) {
330 			pr_warn_once("memory add fail, invalid altmap\n");
331 			return -EINVAL;
332 		}
333 		altmap->alloc = 0;
334 	}
335 
336 	err = check_pfn_span(pfn, nr_pages, "add");
337 	if (err)
338 		return err;
339 
340 	for (; pfn < end_pfn; pfn += cur_nr_pages) {
341 		/* Select all remaining pages up to the next section boundary */
342 		cur_nr_pages = min(end_pfn - pfn,
343 				   SECTION_ALIGN_UP(pfn + 1) - pfn);
344 		err = sparse_add_section(nid, pfn, cur_nr_pages, altmap);
345 		if (err)
346 			break;
347 		cond_resched();
348 	}
349 	vmemmap_populate_print_last();
350 	return err;
351 }
352 
353 /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
354 static unsigned long find_smallest_section_pfn(int nid, struct zone *zone,
355 				     unsigned long start_pfn,
356 				     unsigned long end_pfn)
357 {
358 	for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SUBSECTION) {
359 		if (unlikely(!pfn_to_online_page(start_pfn)))
360 			continue;
361 
362 		if (unlikely(pfn_to_nid(start_pfn) != nid))
363 			continue;
364 
365 		if (zone != page_zone(pfn_to_page(start_pfn)))
366 			continue;
367 
368 		return start_pfn;
369 	}
370 
371 	return 0;
372 }
373 
374 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
375 static unsigned long find_biggest_section_pfn(int nid, struct zone *zone,
376 				    unsigned long start_pfn,
377 				    unsigned long end_pfn)
378 {
379 	unsigned long pfn;
380 
381 	/* pfn is the end pfn of a memory section. */
382 	pfn = end_pfn - 1;
383 	for (; pfn >= start_pfn; pfn -= PAGES_PER_SUBSECTION) {
384 		if (unlikely(!pfn_to_online_page(pfn)))
385 			continue;
386 
387 		if (unlikely(pfn_to_nid(pfn) != nid))
388 			continue;
389 
390 		if (zone != page_zone(pfn_to_page(pfn)))
391 			continue;
392 
393 		return pfn;
394 	}
395 
396 	return 0;
397 }
398 
399 static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
400 			     unsigned long end_pfn)
401 {
402 	unsigned long pfn;
403 	int nid = zone_to_nid(zone);
404 
405 	zone_span_writelock(zone);
406 	if (zone->zone_start_pfn == start_pfn) {
407 		/*
408 		 * If the section is smallest section in the zone, it need
409 		 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
410 		 * In this case, we find second smallest valid mem_section
411 		 * for shrinking zone.
412 		 */
413 		pfn = find_smallest_section_pfn(nid, zone, end_pfn,
414 						zone_end_pfn(zone));
415 		if (pfn) {
416 			zone->spanned_pages = zone_end_pfn(zone) - pfn;
417 			zone->zone_start_pfn = pfn;
418 		} else {
419 			zone->zone_start_pfn = 0;
420 			zone->spanned_pages = 0;
421 		}
422 	} else if (zone_end_pfn(zone) == end_pfn) {
423 		/*
424 		 * If the section is biggest section in the zone, it need
425 		 * shrink zone->spanned_pages.
426 		 * In this case, we find second biggest valid mem_section for
427 		 * shrinking zone.
428 		 */
429 		pfn = find_biggest_section_pfn(nid, zone, zone->zone_start_pfn,
430 					       start_pfn);
431 		if (pfn)
432 			zone->spanned_pages = pfn - zone->zone_start_pfn + 1;
433 		else {
434 			zone->zone_start_pfn = 0;
435 			zone->spanned_pages = 0;
436 		}
437 	}
438 	zone_span_writeunlock(zone);
439 }
440 
441 static void update_pgdat_span(struct pglist_data *pgdat)
442 {
443 	unsigned long node_start_pfn = 0, node_end_pfn = 0;
444 	struct zone *zone;
445 
446 	for (zone = pgdat->node_zones;
447 	     zone < pgdat->node_zones + MAX_NR_ZONES; zone++) {
448 		unsigned long zone_end_pfn = zone->zone_start_pfn +
449 					     zone->spanned_pages;
450 
451 		/* No need to lock the zones, they can't change. */
452 		if (!zone->spanned_pages)
453 			continue;
454 		if (!node_end_pfn) {
455 			node_start_pfn = zone->zone_start_pfn;
456 			node_end_pfn = zone_end_pfn;
457 			continue;
458 		}
459 
460 		if (zone_end_pfn > node_end_pfn)
461 			node_end_pfn = zone_end_pfn;
462 		if (zone->zone_start_pfn < node_start_pfn)
463 			node_start_pfn = zone->zone_start_pfn;
464 	}
465 
466 	pgdat->node_start_pfn = node_start_pfn;
467 	pgdat->node_spanned_pages = node_end_pfn - node_start_pfn;
468 }
469 
470 void __ref remove_pfn_range_from_zone(struct zone *zone,
471 				      unsigned long start_pfn,
472 				      unsigned long nr_pages)
473 {
474 	const unsigned long end_pfn = start_pfn + nr_pages;
475 	struct pglist_data *pgdat = zone->zone_pgdat;
476 	unsigned long pfn, cur_nr_pages, flags;
477 
478 	/* Poison struct pages because they are now uninitialized again. */
479 	for (pfn = start_pfn; pfn < end_pfn; pfn += cur_nr_pages) {
480 		cond_resched();
481 
482 		/* Select all remaining pages up to the next section boundary */
483 		cur_nr_pages =
484 			min(end_pfn - pfn, SECTION_ALIGN_UP(pfn + 1) - pfn);
485 		page_init_poison(pfn_to_page(pfn),
486 				 sizeof(struct page) * cur_nr_pages);
487 	}
488 
489 #ifdef CONFIG_ZONE_DEVICE
490 	/*
491 	 * Zone shrinking code cannot properly deal with ZONE_DEVICE. So
492 	 * we will not try to shrink the zones - which is okay as
493 	 * set_zone_contiguous() cannot deal with ZONE_DEVICE either way.
494 	 */
495 	if (zone_idx(zone) == ZONE_DEVICE)
496 		return;
497 #endif
498 
499 	clear_zone_contiguous(zone);
500 
501 	pgdat_resize_lock(zone->zone_pgdat, &flags);
502 	shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
503 	update_pgdat_span(pgdat);
504 	pgdat_resize_unlock(zone->zone_pgdat, &flags);
505 
506 	set_zone_contiguous(zone);
507 }
508 
509 static void __remove_section(unsigned long pfn, unsigned long nr_pages,
510 			     unsigned long map_offset,
511 			     struct vmem_altmap *altmap)
512 {
513 	struct mem_section *ms = __pfn_to_section(pfn);
514 
515 	if (WARN_ON_ONCE(!valid_section(ms)))
516 		return;
517 
518 	sparse_remove_section(ms, pfn, nr_pages, map_offset, altmap);
519 }
520 
521 /**
522  * __remove_pages() - remove sections of pages
523  * @pfn: starting pageframe (must be aligned to start of a section)
524  * @nr_pages: number of pages to remove (must be multiple of section size)
525  * @altmap: alternative device page map or %NULL if default memmap is used
526  *
527  * Generic helper function to remove section mappings and sysfs entries
528  * for the section of the memory we are removing. Caller needs to make
529  * sure that pages are marked reserved and zones are adjust properly by
530  * calling offline_pages().
531  */
532 void __remove_pages(unsigned long pfn, unsigned long nr_pages,
533 		    struct vmem_altmap *altmap)
534 {
535 	const unsigned long end_pfn = pfn + nr_pages;
536 	unsigned long cur_nr_pages;
537 	unsigned long map_offset = 0;
538 
539 	map_offset = vmem_altmap_offset(altmap);
540 
541 	if (check_pfn_span(pfn, nr_pages, "remove"))
542 		return;
543 
544 	for (; pfn < end_pfn; pfn += cur_nr_pages) {
545 		cond_resched();
546 		/* Select all remaining pages up to the next section boundary */
547 		cur_nr_pages = min(end_pfn - pfn,
548 				   SECTION_ALIGN_UP(pfn + 1) - pfn);
549 		__remove_section(pfn, cur_nr_pages, map_offset, altmap);
550 		map_offset = 0;
551 	}
552 }
553 
554 int set_online_page_callback(online_page_callback_t callback)
555 {
556 	int rc = -EINVAL;
557 
558 	get_online_mems();
559 	mutex_lock(&online_page_callback_lock);
560 
561 	if (online_page_callback == generic_online_page) {
562 		online_page_callback = callback;
563 		rc = 0;
564 	}
565 
566 	mutex_unlock(&online_page_callback_lock);
567 	put_online_mems();
568 
569 	return rc;
570 }
571 EXPORT_SYMBOL_GPL(set_online_page_callback);
572 
573 int restore_online_page_callback(online_page_callback_t callback)
574 {
575 	int rc = -EINVAL;
576 
577 	get_online_mems();
578 	mutex_lock(&online_page_callback_lock);
579 
580 	if (online_page_callback == callback) {
581 		online_page_callback = generic_online_page;
582 		rc = 0;
583 	}
584 
585 	mutex_unlock(&online_page_callback_lock);
586 	put_online_mems();
587 
588 	return rc;
589 }
590 EXPORT_SYMBOL_GPL(restore_online_page_callback);
591 
592 void generic_online_page(struct page *page, unsigned int order)
593 {
594 	/*
595 	 * Freeing the page with debug_pagealloc enabled will try to unmap it,
596 	 * so we should map it first. This is better than introducing a special
597 	 * case in page freeing fast path.
598 	 */
599 	if (debug_pagealloc_enabled_static())
600 		kernel_map_pages(page, 1 << order, 1);
601 	__free_pages_core(page, order);
602 	totalram_pages_add(1UL << order);
603 #ifdef CONFIG_HIGHMEM
604 	if (PageHighMem(page))
605 		totalhigh_pages_add(1UL << order);
606 #endif
607 }
608 EXPORT_SYMBOL_GPL(generic_online_page);
609 
610 static void online_pages_range(unsigned long start_pfn, unsigned long nr_pages)
611 {
612 	const unsigned long end_pfn = start_pfn + nr_pages;
613 	unsigned long pfn;
614 
615 	/*
616 	 * Online the pages in MAX_ORDER - 1 aligned chunks. The callback might
617 	 * decide to not expose all pages to the buddy (e.g., expose them
618 	 * later). We account all pages as being online and belonging to this
619 	 * zone ("present").
620 	 */
621 	for (pfn = start_pfn; pfn < end_pfn; pfn += MAX_ORDER_NR_PAGES)
622 		(*online_page_callback)(pfn_to_page(pfn), MAX_ORDER - 1);
623 
624 	/* mark all involved sections as online */
625 	online_mem_sections(start_pfn, end_pfn);
626 }
627 
628 /* check which state of node_states will be changed when online memory */
629 static void node_states_check_changes_online(unsigned long nr_pages,
630 	struct zone *zone, struct memory_notify *arg)
631 {
632 	int nid = zone_to_nid(zone);
633 
634 	arg->status_change_nid = NUMA_NO_NODE;
635 	arg->status_change_nid_normal = NUMA_NO_NODE;
636 	arg->status_change_nid_high = NUMA_NO_NODE;
637 
638 	if (!node_state(nid, N_MEMORY))
639 		arg->status_change_nid = nid;
640 	if (zone_idx(zone) <= ZONE_NORMAL && !node_state(nid, N_NORMAL_MEMORY))
641 		arg->status_change_nid_normal = nid;
642 #ifdef CONFIG_HIGHMEM
643 	if (zone_idx(zone) <= ZONE_HIGHMEM && !node_state(nid, N_HIGH_MEMORY))
644 		arg->status_change_nid_high = nid;
645 #endif
646 }
647 
648 static void node_states_set_node(int node, struct memory_notify *arg)
649 {
650 	if (arg->status_change_nid_normal >= 0)
651 		node_set_state(node, N_NORMAL_MEMORY);
652 
653 	if (arg->status_change_nid_high >= 0)
654 		node_set_state(node, N_HIGH_MEMORY);
655 
656 	if (arg->status_change_nid >= 0)
657 		node_set_state(node, N_MEMORY);
658 }
659 
660 static void __meminit resize_zone_range(struct zone *zone, unsigned long start_pfn,
661 		unsigned long nr_pages)
662 {
663 	unsigned long old_end_pfn = zone_end_pfn(zone);
664 
665 	if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
666 		zone->zone_start_pfn = start_pfn;
667 
668 	zone->spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - zone->zone_start_pfn;
669 }
670 
671 static void __meminit resize_pgdat_range(struct pglist_data *pgdat, unsigned long start_pfn,
672                                      unsigned long nr_pages)
673 {
674 	unsigned long old_end_pfn = pgdat_end_pfn(pgdat);
675 
676 	if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
677 		pgdat->node_start_pfn = start_pfn;
678 
679 	pgdat->node_spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - pgdat->node_start_pfn;
680 
681 }
682 /*
683  * Associate the pfn range with the given zone, initializing the memmaps
684  * and resizing the pgdat/zone data to span the added pages. After this
685  * call, all affected pages are PG_reserved.
686  *
687  * All aligned pageblocks are initialized to the specified migratetype
688  * (usually MIGRATE_MOVABLE). Besides setting the migratetype, no related
689  * zone stats (e.g., nr_isolate_pageblock) are touched.
690  */
691 void __ref move_pfn_range_to_zone(struct zone *zone, unsigned long start_pfn,
692 				  unsigned long nr_pages,
693 				  struct vmem_altmap *altmap, int migratetype)
694 {
695 	struct pglist_data *pgdat = zone->zone_pgdat;
696 	int nid = pgdat->node_id;
697 	unsigned long flags;
698 
699 	clear_zone_contiguous(zone);
700 
701 	/* TODO Huh pgdat is irqsave while zone is not. It used to be like that before */
702 	pgdat_resize_lock(pgdat, &flags);
703 	zone_span_writelock(zone);
704 	if (zone_is_empty(zone))
705 		init_currently_empty_zone(zone, start_pfn, nr_pages);
706 	resize_zone_range(zone, start_pfn, nr_pages);
707 	zone_span_writeunlock(zone);
708 	resize_pgdat_range(pgdat, start_pfn, nr_pages);
709 	pgdat_resize_unlock(pgdat, &flags);
710 
711 	/*
712 	 * TODO now we have a visible range of pages which are not associated
713 	 * with their zone properly. Not nice but set_pfnblock_flags_mask
714 	 * expects the zone spans the pfn range. All the pages in the range
715 	 * are reserved so nobody should be touching them so we should be safe
716 	 */
717 	memmap_init_zone(nr_pages, nid, zone_idx(zone), start_pfn,
718 			 MEMINIT_HOTPLUG, altmap, migratetype);
719 
720 	set_zone_contiguous(zone);
721 }
722 
723 /*
724  * Returns a default kernel memory zone for the given pfn range.
725  * If no kernel zone covers this pfn range it will automatically go
726  * to the ZONE_NORMAL.
727  */
728 static struct zone *default_kernel_zone_for_pfn(int nid, unsigned long start_pfn,
729 		unsigned long nr_pages)
730 {
731 	struct pglist_data *pgdat = NODE_DATA(nid);
732 	int zid;
733 
734 	for (zid = 0; zid <= ZONE_NORMAL; zid++) {
735 		struct zone *zone = &pgdat->node_zones[zid];
736 
737 		if (zone_intersects(zone, start_pfn, nr_pages))
738 			return zone;
739 	}
740 
741 	return &pgdat->node_zones[ZONE_NORMAL];
742 }
743 
744 static inline struct zone *default_zone_for_pfn(int nid, unsigned long start_pfn,
745 		unsigned long nr_pages)
746 {
747 	struct zone *kernel_zone = default_kernel_zone_for_pfn(nid, start_pfn,
748 			nr_pages);
749 	struct zone *movable_zone = &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
750 	bool in_kernel = zone_intersects(kernel_zone, start_pfn, nr_pages);
751 	bool in_movable = zone_intersects(movable_zone, start_pfn, nr_pages);
752 
753 	/*
754 	 * We inherit the existing zone in a simple case where zones do not
755 	 * overlap in the given range
756 	 */
757 	if (in_kernel ^ in_movable)
758 		return (in_kernel) ? kernel_zone : movable_zone;
759 
760 	/*
761 	 * If the range doesn't belong to any zone or two zones overlap in the
762 	 * given range then we use movable zone only if movable_node is
763 	 * enabled because we always online to a kernel zone by default.
764 	 */
765 	return movable_node_enabled ? movable_zone : kernel_zone;
766 }
767 
768 struct zone * zone_for_pfn_range(int online_type, int nid, unsigned start_pfn,
769 		unsigned long nr_pages)
770 {
771 	if (online_type == MMOP_ONLINE_KERNEL)
772 		return default_kernel_zone_for_pfn(nid, start_pfn, nr_pages);
773 
774 	if (online_type == MMOP_ONLINE_MOVABLE)
775 		return &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
776 
777 	return default_zone_for_pfn(nid, start_pfn, nr_pages);
778 }
779 
780 int __ref online_pages(unsigned long pfn, unsigned long nr_pages,
781 		       int online_type, int nid)
782 {
783 	unsigned long flags;
784 	struct zone *zone;
785 	int need_zonelists_rebuild = 0;
786 	int ret;
787 	struct memory_notify arg;
788 
789 	/* We can only online full sections (e.g., SECTION_IS_ONLINE) */
790 	if (WARN_ON_ONCE(!nr_pages ||
791 			 !IS_ALIGNED(pfn | nr_pages, PAGES_PER_SECTION)))
792 		return -EINVAL;
793 
794 	mem_hotplug_begin();
795 
796 	/* associate pfn range with the zone */
797 	zone = zone_for_pfn_range(online_type, nid, pfn, nr_pages);
798 	move_pfn_range_to_zone(zone, pfn, nr_pages, NULL, MIGRATE_ISOLATE);
799 
800 	arg.start_pfn = pfn;
801 	arg.nr_pages = nr_pages;
802 	node_states_check_changes_online(nr_pages, zone, &arg);
803 
804 	ret = memory_notify(MEM_GOING_ONLINE, &arg);
805 	ret = notifier_to_errno(ret);
806 	if (ret)
807 		goto failed_addition;
808 
809 	/*
810 	 * Fixup the number of isolated pageblocks before marking the sections
811 	 * onlining, such that undo_isolate_page_range() works correctly.
812 	 */
813 	spin_lock_irqsave(&zone->lock, flags);
814 	zone->nr_isolate_pageblock += nr_pages / pageblock_nr_pages;
815 	spin_unlock_irqrestore(&zone->lock, flags);
816 
817 	/*
818 	 * If this zone is not populated, then it is not in zonelist.
819 	 * This means the page allocator ignores this zone.
820 	 * So, zonelist must be updated after online.
821 	 */
822 	if (!populated_zone(zone)) {
823 		need_zonelists_rebuild = 1;
824 		setup_zone_pageset(zone);
825 	}
826 
827 	online_pages_range(pfn, nr_pages);
828 	zone->present_pages += nr_pages;
829 
830 	pgdat_resize_lock(zone->zone_pgdat, &flags);
831 	zone->zone_pgdat->node_present_pages += nr_pages;
832 	pgdat_resize_unlock(zone->zone_pgdat, &flags);
833 
834 	node_states_set_node(nid, &arg);
835 	if (need_zonelists_rebuild)
836 		build_all_zonelists(NULL);
837 	zone_pcp_update(zone);
838 
839 	/* Basic onlining is complete, allow allocation of onlined pages. */
840 	undo_isolate_page_range(pfn, pfn + nr_pages, MIGRATE_MOVABLE);
841 
842 	/*
843 	 * Freshly onlined pages aren't shuffled (e.g., all pages are placed to
844 	 * the tail of the freelist when undoing isolation). Shuffle the whole
845 	 * zone to make sure the just onlined pages are properly distributed
846 	 * across the whole freelist - to create an initial shuffle.
847 	 */
848 	shuffle_zone(zone);
849 
850 	init_per_zone_wmark_min();
851 
852 	kswapd_run(nid);
853 	kcompactd_run(nid);
854 
855 	writeback_set_ratelimit();
856 
857 	memory_notify(MEM_ONLINE, &arg);
858 	mem_hotplug_done();
859 	return 0;
860 
861 failed_addition:
862 	pr_debug("online_pages [mem %#010llx-%#010llx] failed\n",
863 		 (unsigned long long) pfn << PAGE_SHIFT,
864 		 (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1);
865 	memory_notify(MEM_CANCEL_ONLINE, &arg);
866 	remove_pfn_range_from_zone(zone, pfn, nr_pages);
867 	mem_hotplug_done();
868 	return ret;
869 }
870 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
871 
872 static void reset_node_present_pages(pg_data_t *pgdat)
873 {
874 	struct zone *z;
875 
876 	for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
877 		z->present_pages = 0;
878 
879 	pgdat->node_present_pages = 0;
880 }
881 
882 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
883 static pg_data_t __ref *hotadd_new_pgdat(int nid)
884 {
885 	struct pglist_data *pgdat;
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 highest_zoneidx before reuse.
900 		 * Note that kswapd will init kswapd_highest_zoneidx properly
901 		 * when it starts in the near future.
902 		 */
903 		pgdat->nr_zones = 0;
904 		pgdat->kswapd_order = 0;
905 		pgdat->kswapd_highest_zoneidx = 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 	pgdat->node_id = nid;
916 	pgdat->node_start_pfn = 0;
917 
918 	/* init node's zones as empty zones, we don't have any present pages.*/
919 	free_area_init_core_hotplug(nid);
920 
921 	/*
922 	 * The node we allocated has no zone fallback lists. For avoiding
923 	 * to access not-initialized zonelist, build here.
924 	 */
925 	build_all_zonelists(pgdat);
926 
927 	/*
928 	 * When memory is hot-added, all the memory is in offline state. So
929 	 * clear all zones' present_pages because they will be updated in
930 	 * online_pages() and offline_pages().
931 	 */
932 	reset_node_managed_pages(pgdat);
933 	reset_node_present_pages(pgdat);
934 
935 	return pgdat;
936 }
937 
938 static void rollback_node_hotadd(int nid)
939 {
940 	pg_data_t *pgdat = NODE_DATA(nid);
941 
942 	arch_refresh_nodedata(nid, NULL);
943 	free_percpu(pgdat->per_cpu_nodestats);
944 	arch_free_nodedata(pgdat);
945 }
946 
947 
948 /**
949  * try_online_node - online a node if offlined
950  * @nid: the node ID
951  * @set_node_online: Whether we want to online the node
952  * called by cpu_up() to online a node without onlined memory.
953  *
954  * Returns:
955  * 1 -> a new node has been allocated
956  * 0 -> the node is already online
957  * -ENOMEM -> the node could not be allocated
958  */
959 static int __try_online_node(int nid, bool set_node_online)
960 {
961 	pg_data_t *pgdat;
962 	int ret = 1;
963 
964 	if (node_online(nid))
965 		return 0;
966 
967 	pgdat = hotadd_new_pgdat(nid);
968 	if (!pgdat) {
969 		pr_err("Cannot online node %d due to NULL pgdat\n", nid);
970 		ret = -ENOMEM;
971 		goto out;
972 	}
973 
974 	if (set_node_online) {
975 		node_set_online(nid);
976 		ret = register_one_node(nid);
977 		BUG_ON(ret);
978 	}
979 out:
980 	return ret;
981 }
982 
983 /*
984  * Users of this function always want to online/register the node
985  */
986 int try_online_node(int nid)
987 {
988 	int ret;
989 
990 	mem_hotplug_begin();
991 	ret =  __try_online_node(nid, true);
992 	mem_hotplug_done();
993 	return ret;
994 }
995 
996 static int check_hotplug_memory_range(u64 start, u64 size)
997 {
998 	/* memory range must be block size aligned */
999 	if (!size || !IS_ALIGNED(start, memory_block_size_bytes()) ||
1000 	    !IS_ALIGNED(size, memory_block_size_bytes())) {
1001 		pr_err("Block size [%#lx] unaligned hotplug range: start %#llx, size %#llx",
1002 		       memory_block_size_bytes(), start, size);
1003 		return -EINVAL;
1004 	}
1005 
1006 	return 0;
1007 }
1008 
1009 static int online_memory_block(struct memory_block *mem, void *arg)
1010 {
1011 	mem->online_type = memhp_default_online_type;
1012 	return device_online(&mem->dev);
1013 }
1014 
1015 /*
1016  * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1017  * and online/offline operations (triggered e.g. by sysfs).
1018  *
1019  * we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG
1020  */
1021 int __ref add_memory_resource(int nid, struct resource *res, mhp_t mhp_flags)
1022 {
1023 	struct mhp_params params = { .pgprot = PAGE_KERNEL };
1024 	u64 start, size;
1025 	bool new_node = false;
1026 	int ret;
1027 
1028 	start = res->start;
1029 	size = resource_size(res);
1030 
1031 	ret = check_hotplug_memory_range(start, size);
1032 	if (ret)
1033 		return ret;
1034 
1035 	if (!node_possible(nid)) {
1036 		WARN(1, "node %d was absent from the node_possible_map\n", nid);
1037 		return -EINVAL;
1038 	}
1039 
1040 	mem_hotplug_begin();
1041 
1042 	if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK))
1043 		memblock_add_node(start, size, nid);
1044 
1045 	ret = __try_online_node(nid, false);
1046 	if (ret < 0)
1047 		goto error;
1048 	new_node = ret;
1049 
1050 	/* call arch's memory hotadd */
1051 	ret = arch_add_memory(nid, start, size, &params);
1052 	if (ret < 0)
1053 		goto error;
1054 
1055 	/* create memory block devices after memory was added */
1056 	ret = create_memory_block_devices(start, size);
1057 	if (ret) {
1058 		arch_remove_memory(nid, start, size, NULL);
1059 		goto error;
1060 	}
1061 
1062 	if (new_node) {
1063 		/* If sysfs file of new node can't be created, cpu on the node
1064 		 * can't be hot-added. There is no rollback way now.
1065 		 * So, check by BUG_ON() to catch it reluctantly..
1066 		 * We online node here. We can't roll back from here.
1067 		 */
1068 		node_set_online(nid);
1069 		ret = __register_one_node(nid);
1070 		BUG_ON(ret);
1071 	}
1072 
1073 	/* link memory sections under this node.*/
1074 	link_mem_sections(nid, PFN_DOWN(start), PFN_UP(start + size - 1),
1075 			  MEMINIT_HOTPLUG);
1076 
1077 	/* create new memmap entry */
1078 	if (!strcmp(res->name, "System RAM"))
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 	/*
1085 	 * In case we're allowed to merge the resource, flag it and trigger
1086 	 * merging now that adding succeeded.
1087 	 */
1088 	if (mhp_flags & MEMHP_MERGE_RESOURCE)
1089 		merge_system_ram_resource(res);
1090 
1091 	/* online pages if requested */
1092 	if (memhp_default_online_type != MMOP_OFFLINE)
1093 		walk_memory_blocks(start, size, NULL, online_memory_block);
1094 
1095 	return ret;
1096 error:
1097 	/* rollback pgdat allocation and others */
1098 	if (new_node)
1099 		rollback_node_hotadd(nid);
1100 	if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK))
1101 		memblock_remove(start, size);
1102 	mem_hotplug_done();
1103 	return ret;
1104 }
1105 
1106 /* requires device_hotplug_lock, see add_memory_resource() */
1107 int __ref __add_memory(int nid, u64 start, u64 size, mhp_t mhp_flags)
1108 {
1109 	struct resource *res;
1110 	int ret;
1111 
1112 	res = register_memory_resource(start, size, "System RAM");
1113 	if (IS_ERR(res))
1114 		return PTR_ERR(res);
1115 
1116 	ret = add_memory_resource(nid, res, mhp_flags);
1117 	if (ret < 0)
1118 		release_memory_resource(res);
1119 	return ret;
1120 }
1121 
1122 int add_memory(int nid, u64 start, u64 size, mhp_t mhp_flags)
1123 {
1124 	int rc;
1125 
1126 	lock_device_hotplug();
1127 	rc = __add_memory(nid, start, size, mhp_flags);
1128 	unlock_device_hotplug();
1129 
1130 	return rc;
1131 }
1132 EXPORT_SYMBOL_GPL(add_memory);
1133 
1134 /*
1135  * Add special, driver-managed memory to the system as system RAM. Such
1136  * memory is not exposed via the raw firmware-provided memmap as system
1137  * RAM, instead, it is detected and added by a driver - during cold boot,
1138  * after a reboot, and after kexec.
1139  *
1140  * Reasons why this memory should not be used for the initial memmap of a
1141  * kexec kernel or for placing kexec images:
1142  * - The booting kernel is in charge of determining how this memory will be
1143  *   used (e.g., use persistent memory as system RAM)
1144  * - Coordination with a hypervisor is required before this memory
1145  *   can be used (e.g., inaccessible parts).
1146  *
1147  * For this memory, no entries in /sys/firmware/memmap ("raw firmware-provided
1148  * memory map") are created. Also, the created memory resource is flagged
1149  * with IORESOURCE_SYSRAM_DRIVER_MANAGED, so in-kernel users can special-case
1150  * this memory as well (esp., not place kexec images onto it).
1151  *
1152  * The resource_name (visible via /proc/iomem) has to have the format
1153  * "System RAM ($DRIVER)".
1154  */
1155 int add_memory_driver_managed(int nid, u64 start, u64 size,
1156 			      const char *resource_name, mhp_t mhp_flags)
1157 {
1158 	struct resource *res;
1159 	int rc;
1160 
1161 	if (!resource_name ||
1162 	    strstr(resource_name, "System RAM (") != resource_name ||
1163 	    resource_name[strlen(resource_name) - 1] != ')')
1164 		return -EINVAL;
1165 
1166 	lock_device_hotplug();
1167 
1168 	res = register_memory_resource(start, size, resource_name);
1169 	if (IS_ERR(res)) {
1170 		rc = PTR_ERR(res);
1171 		goto out_unlock;
1172 	}
1173 
1174 	rc = add_memory_resource(nid, res, mhp_flags);
1175 	if (rc < 0)
1176 		release_memory_resource(res);
1177 
1178 out_unlock:
1179 	unlock_device_hotplug();
1180 	return rc;
1181 }
1182 EXPORT_SYMBOL_GPL(add_memory_driver_managed);
1183 
1184 #ifdef CONFIG_MEMORY_HOTREMOVE
1185 /*
1186  * Confirm all pages in a range [start, end) belong to the same zone (skipping
1187  * memory holes). When true, return the zone.
1188  */
1189 struct zone *test_pages_in_a_zone(unsigned long start_pfn,
1190 				  unsigned long end_pfn)
1191 {
1192 	unsigned long pfn, sec_end_pfn;
1193 	struct zone *zone = NULL;
1194 	struct page *page;
1195 	int i;
1196 	for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1);
1197 	     pfn < end_pfn;
1198 	     pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) {
1199 		/* Make sure the memory section is present first */
1200 		if (!present_section_nr(pfn_to_section_nr(pfn)))
1201 			continue;
1202 		for (; pfn < sec_end_pfn && pfn < end_pfn;
1203 		     pfn += MAX_ORDER_NR_PAGES) {
1204 			i = 0;
1205 			/* This is just a CONFIG_HOLES_IN_ZONE check.*/
1206 			while ((i < MAX_ORDER_NR_PAGES) &&
1207 				!pfn_valid_within(pfn + i))
1208 				i++;
1209 			if (i == MAX_ORDER_NR_PAGES || pfn + i >= end_pfn)
1210 				continue;
1211 			/* Check if we got outside of the zone */
1212 			if (zone && !zone_spans_pfn(zone, pfn + i))
1213 				return NULL;
1214 			page = pfn_to_page(pfn + i);
1215 			if (zone && page_zone(page) != zone)
1216 				return NULL;
1217 			zone = page_zone(page);
1218 		}
1219 	}
1220 
1221 	return zone;
1222 }
1223 
1224 /*
1225  * Scan pfn range [start,end) to find movable/migratable pages (LRU pages,
1226  * non-lru movable pages and hugepages). Will skip over most unmovable
1227  * pages (esp., pages that can be skipped when offlining), but bail out on
1228  * definitely unmovable pages.
1229  *
1230  * Returns:
1231  *	0 in case a movable page is found and movable_pfn was updated.
1232  *	-ENOENT in case no movable page was found.
1233  *	-EBUSY in case a definitely unmovable page was found.
1234  */
1235 static int scan_movable_pages(unsigned long start, unsigned long end,
1236 			      unsigned long *movable_pfn)
1237 {
1238 	unsigned long pfn;
1239 
1240 	for (pfn = start; pfn < end; pfn++) {
1241 		struct page *page, *head;
1242 		unsigned long skip;
1243 
1244 		if (!pfn_valid(pfn))
1245 			continue;
1246 		page = pfn_to_page(pfn);
1247 		if (PageLRU(page))
1248 			goto found;
1249 		if (__PageMovable(page))
1250 			goto found;
1251 
1252 		/*
1253 		 * PageOffline() pages that are not marked __PageMovable() and
1254 		 * have a reference count > 0 (after MEM_GOING_OFFLINE) are
1255 		 * definitely unmovable. If their reference count would be 0,
1256 		 * they could at least be skipped when offlining memory.
1257 		 */
1258 		if (PageOffline(page) && page_count(page))
1259 			return -EBUSY;
1260 
1261 		if (!PageHuge(page))
1262 			continue;
1263 		head = compound_head(page);
1264 		if (page_huge_active(head))
1265 			goto found;
1266 		skip = compound_nr(head) - (page - head);
1267 		pfn += skip - 1;
1268 	}
1269 	return -ENOENT;
1270 found:
1271 	*movable_pfn = pfn;
1272 	return 0;
1273 }
1274 
1275 static int
1276 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1277 {
1278 	unsigned long pfn;
1279 	struct page *page, *head;
1280 	int ret = 0;
1281 	LIST_HEAD(source);
1282 
1283 	for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1284 		if (!pfn_valid(pfn))
1285 			continue;
1286 		page = pfn_to_page(pfn);
1287 		head = compound_head(page);
1288 
1289 		if (PageHuge(page)) {
1290 			pfn = page_to_pfn(head) + compound_nr(head) - 1;
1291 			isolate_huge_page(head, &source);
1292 			continue;
1293 		} else if (PageTransHuge(page))
1294 			pfn = page_to_pfn(head) + thp_nr_pages(page) - 1;
1295 
1296 		/*
1297 		 * HWPoison pages have elevated reference counts so the migration would
1298 		 * fail on them. It also doesn't make any sense to migrate them in the
1299 		 * first place. Still try to unmap such a page in case it is still mapped
1300 		 * (e.g. current hwpoison implementation doesn't unmap KSM pages but keep
1301 		 * the unmap as the catch all safety net).
1302 		 */
1303 		if (PageHWPoison(page)) {
1304 			if (WARN_ON(PageLRU(page)))
1305 				isolate_lru_page(page);
1306 			if (page_mapped(page))
1307 				try_to_unmap(page, TTU_IGNORE_MLOCK | TTU_IGNORE_ACCESS);
1308 			continue;
1309 		}
1310 
1311 		if (!get_page_unless_zero(page))
1312 			continue;
1313 		/*
1314 		 * We can skip free pages. And we can deal with pages on
1315 		 * LRU and non-lru movable pages.
1316 		 */
1317 		if (PageLRU(page))
1318 			ret = isolate_lru_page(page);
1319 		else
1320 			ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE);
1321 		if (!ret) { /* Success */
1322 			list_add_tail(&page->lru, &source);
1323 			if (!__PageMovable(page))
1324 				inc_node_page_state(page, NR_ISOLATED_ANON +
1325 						    page_is_file_lru(page));
1326 
1327 		} else {
1328 			pr_warn("failed to isolate pfn %lx\n", pfn);
1329 			dump_page(page, "isolation failed");
1330 		}
1331 		put_page(page);
1332 	}
1333 	if (!list_empty(&source)) {
1334 		nodemask_t nmask = node_states[N_MEMORY];
1335 		struct migration_target_control mtc = {
1336 			.nmask = &nmask,
1337 			.gfp_mask = GFP_USER | __GFP_MOVABLE | __GFP_RETRY_MAYFAIL,
1338 		};
1339 
1340 		/*
1341 		 * We have checked that migration range is on a single zone so
1342 		 * we can use the nid of the first page to all the others.
1343 		 */
1344 		mtc.nid = page_to_nid(list_first_entry(&source, struct page, lru));
1345 
1346 		/*
1347 		 * try to allocate from a different node but reuse this node
1348 		 * if there are no other online nodes to be used (e.g. we are
1349 		 * offlining a part of the only existing node)
1350 		 */
1351 		node_clear(mtc.nid, nmask);
1352 		if (nodes_empty(nmask))
1353 			node_set(mtc.nid, nmask);
1354 		ret = migrate_pages(&source, alloc_migration_target, NULL,
1355 			(unsigned long)&mtc, MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
1356 		if (ret) {
1357 			list_for_each_entry(page, &source, lru) {
1358 				pr_warn("migrating pfn %lx failed ret:%d ",
1359 				       page_to_pfn(page), ret);
1360 				dump_page(page, "migration failure");
1361 			}
1362 			putback_movable_pages(&source);
1363 		}
1364 	}
1365 
1366 	return ret;
1367 }
1368 
1369 static int __init cmdline_parse_movable_node(char *p)
1370 {
1371 	movable_node_enabled = true;
1372 	return 0;
1373 }
1374 early_param("movable_node", cmdline_parse_movable_node);
1375 
1376 /* check which state of node_states will be changed when offline memory */
1377 static void node_states_check_changes_offline(unsigned long nr_pages,
1378 		struct zone *zone, struct memory_notify *arg)
1379 {
1380 	struct pglist_data *pgdat = zone->zone_pgdat;
1381 	unsigned long present_pages = 0;
1382 	enum zone_type zt;
1383 
1384 	arg->status_change_nid = NUMA_NO_NODE;
1385 	arg->status_change_nid_normal = NUMA_NO_NODE;
1386 	arg->status_change_nid_high = NUMA_NO_NODE;
1387 
1388 	/*
1389 	 * Check whether node_states[N_NORMAL_MEMORY] will be changed.
1390 	 * If the memory to be offline is within the range
1391 	 * [0..ZONE_NORMAL], and it is the last present memory there,
1392 	 * the zones in that range will become empty after the offlining,
1393 	 * thus we can determine that we need to clear the node from
1394 	 * node_states[N_NORMAL_MEMORY].
1395 	 */
1396 	for (zt = 0; zt <= ZONE_NORMAL; zt++)
1397 		present_pages += pgdat->node_zones[zt].present_pages;
1398 	if (zone_idx(zone) <= ZONE_NORMAL && nr_pages >= present_pages)
1399 		arg->status_change_nid_normal = zone_to_nid(zone);
1400 
1401 #ifdef CONFIG_HIGHMEM
1402 	/*
1403 	 * node_states[N_HIGH_MEMORY] contains nodes which
1404 	 * have normal memory or high memory.
1405 	 * Here we add the present_pages belonging to ZONE_HIGHMEM.
1406 	 * If the zone is within the range of [0..ZONE_HIGHMEM), and
1407 	 * we determine that the zones in that range become empty,
1408 	 * we need to clear the node for N_HIGH_MEMORY.
1409 	 */
1410 	present_pages += pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1411 	if (zone_idx(zone) <= ZONE_HIGHMEM && nr_pages >= present_pages)
1412 		arg->status_change_nid_high = zone_to_nid(zone);
1413 #endif
1414 
1415 	/*
1416 	 * We have accounted the pages from [0..ZONE_NORMAL), and
1417 	 * in case of CONFIG_HIGHMEM the pages from ZONE_HIGHMEM
1418 	 * as well.
1419 	 * Here we count the possible pages from ZONE_MOVABLE.
1420 	 * If after having accounted all the pages, we see that the nr_pages
1421 	 * to be offlined is over or equal to the accounted pages,
1422 	 * we know that the node will become empty, and so, we can clear
1423 	 * it for N_MEMORY as well.
1424 	 */
1425 	present_pages += pgdat->node_zones[ZONE_MOVABLE].present_pages;
1426 
1427 	if (nr_pages >= present_pages)
1428 		arg->status_change_nid = zone_to_nid(zone);
1429 }
1430 
1431 static void node_states_clear_node(int node, struct memory_notify *arg)
1432 {
1433 	if (arg->status_change_nid_normal >= 0)
1434 		node_clear_state(node, N_NORMAL_MEMORY);
1435 
1436 	if (arg->status_change_nid_high >= 0)
1437 		node_clear_state(node, N_HIGH_MEMORY);
1438 
1439 	if (arg->status_change_nid >= 0)
1440 		node_clear_state(node, N_MEMORY);
1441 }
1442 
1443 static int count_system_ram_pages_cb(unsigned long start_pfn,
1444 				     unsigned long nr_pages, void *data)
1445 {
1446 	unsigned long *nr_system_ram_pages = data;
1447 
1448 	*nr_system_ram_pages += nr_pages;
1449 	return 0;
1450 }
1451 
1452 int __ref offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1453 {
1454 	const unsigned long end_pfn = start_pfn + nr_pages;
1455 	unsigned long pfn, system_ram_pages = 0;
1456 	unsigned long flags;
1457 	struct zone *zone;
1458 	struct memory_notify arg;
1459 	int ret, node;
1460 	char *reason;
1461 
1462 	/* We can only offline full sections (e.g., SECTION_IS_ONLINE) */
1463 	if (WARN_ON_ONCE(!nr_pages ||
1464 			 !IS_ALIGNED(start_pfn | nr_pages, PAGES_PER_SECTION)))
1465 		return -EINVAL;
1466 
1467 	mem_hotplug_begin();
1468 
1469 	/*
1470 	 * Don't allow to offline memory blocks that contain holes.
1471 	 * Consequently, memory blocks with holes can never get onlined
1472 	 * via the hotplug path - online_pages() - as hotplugged memory has
1473 	 * no holes. This way, we e.g., don't have to worry about marking
1474 	 * memory holes PG_reserved, don't need pfn_valid() checks, and can
1475 	 * avoid using walk_system_ram_range() later.
1476 	 */
1477 	walk_system_ram_range(start_pfn, nr_pages, &system_ram_pages,
1478 			      count_system_ram_pages_cb);
1479 	if (system_ram_pages != nr_pages) {
1480 		ret = -EINVAL;
1481 		reason = "memory holes";
1482 		goto failed_removal;
1483 	}
1484 
1485 	/* This makes hotplug much easier...and readable.
1486 	   we assume this for now. .*/
1487 	zone = test_pages_in_a_zone(start_pfn, end_pfn);
1488 	if (!zone) {
1489 		ret = -EINVAL;
1490 		reason = "multizone range";
1491 		goto failed_removal;
1492 	}
1493 	node = zone_to_nid(zone);
1494 
1495 	/* set above range as isolated */
1496 	ret = start_isolate_page_range(start_pfn, end_pfn,
1497 				       MIGRATE_MOVABLE,
1498 				       MEMORY_OFFLINE | REPORT_FAILURE);
1499 	if (ret) {
1500 		reason = "failure to isolate range";
1501 		goto failed_removal;
1502 	}
1503 
1504 	arg.start_pfn = start_pfn;
1505 	arg.nr_pages = nr_pages;
1506 	node_states_check_changes_offline(nr_pages, zone, &arg);
1507 
1508 	ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1509 	ret = notifier_to_errno(ret);
1510 	if (ret) {
1511 		reason = "notifier failure";
1512 		goto failed_removal_isolated;
1513 	}
1514 
1515 	do {
1516 		pfn = start_pfn;
1517 		do {
1518 			if (signal_pending(current)) {
1519 				ret = -EINTR;
1520 				reason = "signal backoff";
1521 				goto failed_removal_isolated;
1522 			}
1523 
1524 			cond_resched();
1525 			lru_add_drain_all();
1526 
1527 			ret = scan_movable_pages(pfn, end_pfn, &pfn);
1528 			if (!ret) {
1529 				/*
1530 				 * TODO: fatal migration failures should bail
1531 				 * out
1532 				 */
1533 				do_migrate_range(pfn, end_pfn);
1534 			}
1535 		} while (!ret);
1536 
1537 		if (ret != -ENOENT) {
1538 			reason = "unmovable page";
1539 			goto failed_removal_isolated;
1540 		}
1541 
1542 		/*
1543 		 * Dissolve free hugepages in the memory block before doing
1544 		 * offlining actually in order to make hugetlbfs's object
1545 		 * counting consistent.
1546 		 */
1547 		ret = dissolve_free_huge_pages(start_pfn, end_pfn);
1548 		if (ret) {
1549 			reason = "failure to dissolve huge pages";
1550 			goto failed_removal_isolated;
1551 		}
1552 
1553 		/*
1554 		 * per-cpu pages are drained in start_isolate_page_range, but if
1555 		 * there are still pages that are not free, make sure that we
1556 		 * drain again, because when we isolated range we might
1557 		 * have raced with another thread that was adding pages to pcp
1558 		 * list.
1559 		 *
1560 		 * Forward progress should be still guaranteed because
1561 		 * pages on the pcp list can only belong to MOVABLE_ZONE
1562 		 * because has_unmovable_pages explicitly checks for
1563 		 * PageBuddy on freed pages on other zones.
1564 		 */
1565 		ret = test_pages_isolated(start_pfn, end_pfn, MEMORY_OFFLINE);
1566 		if (ret)
1567 			drain_all_pages(zone);
1568 	} while (ret);
1569 
1570 	/* Mark all sections offline and remove free pages from the buddy. */
1571 	__offline_isolated_pages(start_pfn, end_pfn);
1572 	pr_info("Offlined Pages %ld\n", nr_pages);
1573 
1574 	/*
1575 	 * The memory sections are marked offline, and the pageblock flags
1576 	 * effectively stale; nobody should be touching them. Fixup the number
1577 	 * of isolated pageblocks, memory onlining will properly revert this.
1578 	 */
1579 	spin_lock_irqsave(&zone->lock, flags);
1580 	zone->nr_isolate_pageblock -= nr_pages / pageblock_nr_pages;
1581 	spin_unlock_irqrestore(&zone->lock, flags);
1582 
1583 	/* removal success */
1584 	adjust_managed_page_count(pfn_to_page(start_pfn), -nr_pages);
1585 	zone->present_pages -= nr_pages;
1586 
1587 	pgdat_resize_lock(zone->zone_pgdat, &flags);
1588 	zone->zone_pgdat->node_present_pages -= nr_pages;
1589 	pgdat_resize_unlock(zone->zone_pgdat, &flags);
1590 
1591 	init_per_zone_wmark_min();
1592 
1593 	if (!populated_zone(zone)) {
1594 		zone_pcp_reset(zone);
1595 		build_all_zonelists(NULL);
1596 	} else
1597 		zone_pcp_update(zone);
1598 
1599 	node_states_clear_node(node, &arg);
1600 	if (arg.status_change_nid >= 0) {
1601 		kswapd_stop(node);
1602 		kcompactd_stop(node);
1603 	}
1604 
1605 	writeback_set_ratelimit();
1606 
1607 	memory_notify(MEM_OFFLINE, &arg);
1608 	remove_pfn_range_from_zone(zone, start_pfn, nr_pages);
1609 	mem_hotplug_done();
1610 	return 0;
1611 
1612 failed_removal_isolated:
1613 	undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1614 	memory_notify(MEM_CANCEL_OFFLINE, &arg);
1615 failed_removal:
1616 	pr_debug("memory offlining [mem %#010llx-%#010llx] failed due to %s\n",
1617 		 (unsigned long long) start_pfn << PAGE_SHIFT,
1618 		 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1,
1619 		 reason);
1620 	/* pushback to free area */
1621 	mem_hotplug_done();
1622 	return ret;
1623 }
1624 
1625 static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
1626 {
1627 	int ret = !is_memblock_offlined(mem);
1628 
1629 	if (unlikely(ret)) {
1630 		phys_addr_t beginpa, endpa;
1631 
1632 		beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
1633 		endpa = beginpa + memory_block_size_bytes() - 1;
1634 		pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n",
1635 			&beginpa, &endpa);
1636 
1637 		return -EBUSY;
1638 	}
1639 	return 0;
1640 }
1641 
1642 static int check_cpu_on_node(pg_data_t *pgdat)
1643 {
1644 	int cpu;
1645 
1646 	for_each_present_cpu(cpu) {
1647 		if (cpu_to_node(cpu) == pgdat->node_id)
1648 			/*
1649 			 * the cpu on this node isn't removed, and we can't
1650 			 * offline this node.
1651 			 */
1652 			return -EBUSY;
1653 	}
1654 
1655 	return 0;
1656 }
1657 
1658 static int check_no_memblock_for_node_cb(struct memory_block *mem, void *arg)
1659 {
1660 	int nid = *(int *)arg;
1661 
1662 	/*
1663 	 * If a memory block belongs to multiple nodes, the stored nid is not
1664 	 * reliable. However, such blocks are always online (e.g., cannot get
1665 	 * offlined) and, therefore, are still spanned by the node.
1666 	 */
1667 	return mem->nid == nid ? -EEXIST : 0;
1668 }
1669 
1670 /**
1671  * try_offline_node
1672  * @nid: the node ID
1673  *
1674  * Offline a node if all memory sections and cpus of the node are removed.
1675  *
1676  * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1677  * and online/offline operations before this call.
1678  */
1679 void try_offline_node(int nid)
1680 {
1681 	pg_data_t *pgdat = NODE_DATA(nid);
1682 	int rc;
1683 
1684 	/*
1685 	 * If the node still spans pages (especially ZONE_DEVICE), don't
1686 	 * offline it. A node spans memory after move_pfn_range_to_zone(),
1687 	 * e.g., after the memory block was onlined.
1688 	 */
1689 	if (pgdat->node_spanned_pages)
1690 		return;
1691 
1692 	/*
1693 	 * Especially offline memory blocks might not be spanned by the
1694 	 * node. They will get spanned by the node once they get onlined.
1695 	 * However, they link to the node in sysfs and can get onlined later.
1696 	 */
1697 	rc = for_each_memory_block(&nid, check_no_memblock_for_node_cb);
1698 	if (rc)
1699 		return;
1700 
1701 	if (check_cpu_on_node(pgdat))
1702 		return;
1703 
1704 	/*
1705 	 * all memory/cpu of this node are removed, we can offline this
1706 	 * node now.
1707 	 */
1708 	node_set_offline(nid);
1709 	unregister_one_node(nid);
1710 }
1711 EXPORT_SYMBOL(try_offline_node);
1712 
1713 static int __ref try_remove_memory(int nid, u64 start, u64 size)
1714 {
1715 	int rc = 0;
1716 
1717 	BUG_ON(check_hotplug_memory_range(start, size));
1718 
1719 	/*
1720 	 * All memory blocks must be offlined before removing memory.  Check
1721 	 * whether all memory blocks in question are offline and return error
1722 	 * if this is not the case.
1723 	 */
1724 	rc = walk_memory_blocks(start, size, NULL, check_memblock_offlined_cb);
1725 	if (rc)
1726 		return rc;
1727 
1728 	/* remove memmap entry */
1729 	firmware_map_remove(start, start + size, "System RAM");
1730 
1731 	/*
1732 	 * Memory block device removal under the device_hotplug_lock is
1733 	 * a barrier against racing online attempts.
1734 	 */
1735 	remove_memory_block_devices(start, size);
1736 
1737 	mem_hotplug_begin();
1738 
1739 	arch_remove_memory(nid, start, size, NULL);
1740 
1741 	if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK)) {
1742 		memblock_free(start, size);
1743 		memblock_remove(start, size);
1744 	}
1745 
1746 	release_mem_region_adjustable(start, size);
1747 
1748 	try_offline_node(nid);
1749 
1750 	mem_hotplug_done();
1751 	return 0;
1752 }
1753 
1754 /**
1755  * remove_memory
1756  * @nid: the node ID
1757  * @start: physical address of the region to remove
1758  * @size: size of the region to remove
1759  *
1760  * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1761  * and online/offline operations before this call, as required by
1762  * try_offline_node().
1763  */
1764 void __remove_memory(int nid, u64 start, u64 size)
1765 {
1766 
1767 	/*
1768 	 * trigger BUG() if some memory is not offlined prior to calling this
1769 	 * function
1770 	 */
1771 	if (try_remove_memory(nid, start, size))
1772 		BUG();
1773 }
1774 
1775 /*
1776  * Remove memory if every memory block is offline, otherwise return -EBUSY is
1777  * some memory is not offline
1778  */
1779 int remove_memory(int nid, u64 start, u64 size)
1780 {
1781 	int rc;
1782 
1783 	lock_device_hotplug();
1784 	rc  = try_remove_memory(nid, start, size);
1785 	unlock_device_hotplug();
1786 
1787 	return rc;
1788 }
1789 EXPORT_SYMBOL_GPL(remove_memory);
1790 
1791 /*
1792  * Try to offline and remove a memory block. Might take a long time to
1793  * finish in case memory is still in use. Primarily useful for memory devices
1794  * that logically unplugged all memory (so it's no longer in use) and want to
1795  * offline + remove the memory block.
1796  */
1797 int offline_and_remove_memory(int nid, u64 start, u64 size)
1798 {
1799 	struct memory_block *mem;
1800 	int rc = -EINVAL;
1801 
1802 	if (!IS_ALIGNED(start, memory_block_size_bytes()) ||
1803 	    size != memory_block_size_bytes())
1804 		return rc;
1805 
1806 	lock_device_hotplug();
1807 	mem = find_memory_block(__pfn_to_section(PFN_DOWN(start)));
1808 	if (mem)
1809 		rc = device_offline(&mem->dev);
1810 	/* Ignore if the device is already offline. */
1811 	if (rc > 0)
1812 		rc = 0;
1813 
1814 	/*
1815 	 * In case we succeeded to offline the memory block, remove it.
1816 	 * This cannot fail as it cannot get onlined in the meantime.
1817 	 */
1818 	if (!rc) {
1819 		rc = try_remove_memory(nid, start, size);
1820 		WARN_ON_ONCE(rc);
1821 	}
1822 	unlock_device_hotplug();
1823 
1824 	return rc;
1825 }
1826 EXPORT_SYMBOL_GPL(offline_and_remove_memory);
1827 #endif /* CONFIG_MEMORY_HOTREMOVE */
1828