xref: /openbmc/linux/mm/memory_hotplug.c (revision cb8e3c8b4f45e4ed8987a581956dc9c3827a5bcf)
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 #ifdef CONFIG_NUMA
354 int __weak memory_add_physaddr_to_nid(u64 start)
355 {
356 	pr_info_once("Unknown online node for memory at 0x%llx, assuming node 0\n",
357 			start);
358 	return 0;
359 }
360 EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid);
361 
362 int __weak phys_to_target_node(u64 start)
363 {
364 	pr_info_once("Unknown target node for memory at 0x%llx, assuming node 0\n",
365 			start);
366 	return 0;
367 }
368 EXPORT_SYMBOL_GPL(phys_to_target_node);
369 #endif
370 
371 /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
372 static unsigned long find_smallest_section_pfn(int nid, struct zone *zone,
373 				     unsigned long start_pfn,
374 				     unsigned long end_pfn)
375 {
376 	for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SUBSECTION) {
377 		if (unlikely(!pfn_to_online_page(start_pfn)))
378 			continue;
379 
380 		if (unlikely(pfn_to_nid(start_pfn) != nid))
381 			continue;
382 
383 		if (zone != page_zone(pfn_to_page(start_pfn)))
384 			continue;
385 
386 		return start_pfn;
387 	}
388 
389 	return 0;
390 }
391 
392 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
393 static unsigned long find_biggest_section_pfn(int nid, struct zone *zone,
394 				    unsigned long start_pfn,
395 				    unsigned long end_pfn)
396 {
397 	unsigned long pfn;
398 
399 	/* pfn is the end pfn of a memory section. */
400 	pfn = end_pfn - 1;
401 	for (; pfn >= start_pfn; pfn -= PAGES_PER_SUBSECTION) {
402 		if (unlikely(!pfn_to_online_page(pfn)))
403 			continue;
404 
405 		if (unlikely(pfn_to_nid(pfn) != nid))
406 			continue;
407 
408 		if (zone != page_zone(pfn_to_page(pfn)))
409 			continue;
410 
411 		return pfn;
412 	}
413 
414 	return 0;
415 }
416 
417 static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
418 			     unsigned long end_pfn)
419 {
420 	unsigned long pfn;
421 	int nid = zone_to_nid(zone);
422 
423 	zone_span_writelock(zone);
424 	if (zone->zone_start_pfn == start_pfn) {
425 		/*
426 		 * If the section is smallest section in the zone, it need
427 		 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
428 		 * In this case, we find second smallest valid mem_section
429 		 * for shrinking zone.
430 		 */
431 		pfn = find_smallest_section_pfn(nid, zone, end_pfn,
432 						zone_end_pfn(zone));
433 		if (pfn) {
434 			zone->spanned_pages = zone_end_pfn(zone) - pfn;
435 			zone->zone_start_pfn = pfn;
436 		} else {
437 			zone->zone_start_pfn = 0;
438 			zone->spanned_pages = 0;
439 		}
440 	} else if (zone_end_pfn(zone) == end_pfn) {
441 		/*
442 		 * If the section is biggest section in the zone, it need
443 		 * shrink zone->spanned_pages.
444 		 * In this case, we find second biggest valid mem_section for
445 		 * shrinking zone.
446 		 */
447 		pfn = find_biggest_section_pfn(nid, zone, zone->zone_start_pfn,
448 					       start_pfn);
449 		if (pfn)
450 			zone->spanned_pages = pfn - zone->zone_start_pfn + 1;
451 		else {
452 			zone->zone_start_pfn = 0;
453 			zone->spanned_pages = 0;
454 		}
455 	}
456 	zone_span_writeunlock(zone);
457 }
458 
459 static void update_pgdat_span(struct pglist_data *pgdat)
460 {
461 	unsigned long node_start_pfn = 0, node_end_pfn = 0;
462 	struct zone *zone;
463 
464 	for (zone = pgdat->node_zones;
465 	     zone < pgdat->node_zones + MAX_NR_ZONES; zone++) {
466 		unsigned long zone_end_pfn = zone->zone_start_pfn +
467 					     zone->spanned_pages;
468 
469 		/* No need to lock the zones, they can't change. */
470 		if (!zone->spanned_pages)
471 			continue;
472 		if (!node_end_pfn) {
473 			node_start_pfn = zone->zone_start_pfn;
474 			node_end_pfn = zone_end_pfn;
475 			continue;
476 		}
477 
478 		if (zone_end_pfn > node_end_pfn)
479 			node_end_pfn = zone_end_pfn;
480 		if (zone->zone_start_pfn < node_start_pfn)
481 			node_start_pfn = zone->zone_start_pfn;
482 	}
483 
484 	pgdat->node_start_pfn = node_start_pfn;
485 	pgdat->node_spanned_pages = node_end_pfn - node_start_pfn;
486 }
487 
488 void __ref remove_pfn_range_from_zone(struct zone *zone,
489 				      unsigned long start_pfn,
490 				      unsigned long nr_pages)
491 {
492 	const unsigned long end_pfn = start_pfn + nr_pages;
493 	struct pglist_data *pgdat = zone->zone_pgdat;
494 	unsigned long pfn, cur_nr_pages, flags;
495 
496 	/* Poison struct pages because they are now uninitialized again. */
497 	for (pfn = start_pfn; pfn < end_pfn; pfn += cur_nr_pages) {
498 		cond_resched();
499 
500 		/* Select all remaining pages up to the next section boundary */
501 		cur_nr_pages =
502 			min(end_pfn - pfn, SECTION_ALIGN_UP(pfn + 1) - pfn);
503 		page_init_poison(pfn_to_page(pfn),
504 				 sizeof(struct page) * cur_nr_pages);
505 	}
506 
507 #ifdef CONFIG_ZONE_DEVICE
508 	/*
509 	 * Zone shrinking code cannot properly deal with ZONE_DEVICE. So
510 	 * we will not try to shrink the zones - which is okay as
511 	 * set_zone_contiguous() cannot deal with ZONE_DEVICE either way.
512 	 */
513 	if (zone_idx(zone) == ZONE_DEVICE)
514 		return;
515 #endif
516 
517 	clear_zone_contiguous(zone);
518 
519 	pgdat_resize_lock(zone->zone_pgdat, &flags);
520 	shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
521 	update_pgdat_span(pgdat);
522 	pgdat_resize_unlock(zone->zone_pgdat, &flags);
523 
524 	set_zone_contiguous(zone);
525 }
526 
527 static void __remove_section(unsigned long pfn, unsigned long nr_pages,
528 			     unsigned long map_offset,
529 			     struct vmem_altmap *altmap)
530 {
531 	struct mem_section *ms = __pfn_to_section(pfn);
532 
533 	if (WARN_ON_ONCE(!valid_section(ms)))
534 		return;
535 
536 	sparse_remove_section(ms, pfn, nr_pages, map_offset, altmap);
537 }
538 
539 /**
540  * __remove_pages() - remove sections of pages
541  * @pfn: starting pageframe (must be aligned to start of a section)
542  * @nr_pages: number of pages to remove (must be multiple of section size)
543  * @altmap: alternative device page map or %NULL if default memmap is used
544  *
545  * Generic helper function to remove section mappings and sysfs entries
546  * for the section of the memory we are removing. Caller needs to make
547  * sure that pages are marked reserved and zones are adjust properly by
548  * calling offline_pages().
549  */
550 void __remove_pages(unsigned long pfn, unsigned long nr_pages,
551 		    struct vmem_altmap *altmap)
552 {
553 	const unsigned long end_pfn = pfn + nr_pages;
554 	unsigned long cur_nr_pages;
555 	unsigned long map_offset = 0;
556 
557 	map_offset = vmem_altmap_offset(altmap);
558 
559 	if (check_pfn_span(pfn, nr_pages, "remove"))
560 		return;
561 
562 	for (; pfn < end_pfn; pfn += cur_nr_pages) {
563 		cond_resched();
564 		/* Select all remaining pages up to the next section boundary */
565 		cur_nr_pages = min(end_pfn - pfn,
566 				   SECTION_ALIGN_UP(pfn + 1) - pfn);
567 		__remove_section(pfn, cur_nr_pages, map_offset, altmap);
568 		map_offset = 0;
569 	}
570 }
571 
572 int set_online_page_callback(online_page_callback_t callback)
573 {
574 	int rc = -EINVAL;
575 
576 	get_online_mems();
577 	mutex_lock(&online_page_callback_lock);
578 
579 	if (online_page_callback == generic_online_page) {
580 		online_page_callback = callback;
581 		rc = 0;
582 	}
583 
584 	mutex_unlock(&online_page_callback_lock);
585 	put_online_mems();
586 
587 	return rc;
588 }
589 EXPORT_SYMBOL_GPL(set_online_page_callback);
590 
591 int restore_online_page_callback(online_page_callback_t callback)
592 {
593 	int rc = -EINVAL;
594 
595 	get_online_mems();
596 	mutex_lock(&online_page_callback_lock);
597 
598 	if (online_page_callback == callback) {
599 		online_page_callback = generic_online_page;
600 		rc = 0;
601 	}
602 
603 	mutex_unlock(&online_page_callback_lock);
604 	put_online_mems();
605 
606 	return rc;
607 }
608 EXPORT_SYMBOL_GPL(restore_online_page_callback);
609 
610 void generic_online_page(struct page *page, unsigned int order)
611 {
612 	/*
613 	 * Freeing the page with debug_pagealloc enabled will try to unmap it,
614 	 * so we should map it first. This is better than introducing a special
615 	 * case in page freeing fast path.
616 	 */
617 	if (debug_pagealloc_enabled_static())
618 		kernel_map_pages(page, 1 << order, 1);
619 	__free_pages_core(page, order);
620 	totalram_pages_add(1UL << order);
621 #ifdef CONFIG_HIGHMEM
622 	if (PageHighMem(page))
623 		totalhigh_pages_add(1UL << order);
624 #endif
625 }
626 EXPORT_SYMBOL_GPL(generic_online_page);
627 
628 static void online_pages_range(unsigned long start_pfn, unsigned long nr_pages)
629 {
630 	const unsigned long end_pfn = start_pfn + nr_pages;
631 	unsigned long pfn;
632 
633 	/*
634 	 * Online the pages in MAX_ORDER - 1 aligned chunks. The callback might
635 	 * decide to not expose all pages to the buddy (e.g., expose them
636 	 * later). We account all pages as being online and belonging to this
637 	 * zone ("present").
638 	 */
639 	for (pfn = start_pfn; pfn < end_pfn; pfn += MAX_ORDER_NR_PAGES)
640 		(*online_page_callback)(pfn_to_page(pfn), MAX_ORDER - 1);
641 
642 	/* mark all involved sections as online */
643 	online_mem_sections(start_pfn, end_pfn);
644 }
645 
646 /* check which state of node_states will be changed when online memory */
647 static void node_states_check_changes_online(unsigned long nr_pages,
648 	struct zone *zone, struct memory_notify *arg)
649 {
650 	int nid = zone_to_nid(zone);
651 
652 	arg->status_change_nid = NUMA_NO_NODE;
653 	arg->status_change_nid_normal = NUMA_NO_NODE;
654 	arg->status_change_nid_high = NUMA_NO_NODE;
655 
656 	if (!node_state(nid, N_MEMORY))
657 		arg->status_change_nid = nid;
658 	if (zone_idx(zone) <= ZONE_NORMAL && !node_state(nid, N_NORMAL_MEMORY))
659 		arg->status_change_nid_normal = nid;
660 #ifdef CONFIG_HIGHMEM
661 	if (zone_idx(zone) <= ZONE_HIGHMEM && !node_state(nid, N_HIGH_MEMORY))
662 		arg->status_change_nid_high = nid;
663 #endif
664 }
665 
666 static void node_states_set_node(int node, struct memory_notify *arg)
667 {
668 	if (arg->status_change_nid_normal >= 0)
669 		node_set_state(node, N_NORMAL_MEMORY);
670 
671 	if (arg->status_change_nid_high >= 0)
672 		node_set_state(node, N_HIGH_MEMORY);
673 
674 	if (arg->status_change_nid >= 0)
675 		node_set_state(node, N_MEMORY);
676 }
677 
678 static void __meminit resize_zone_range(struct zone *zone, unsigned long start_pfn,
679 		unsigned long nr_pages)
680 {
681 	unsigned long old_end_pfn = zone_end_pfn(zone);
682 
683 	if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
684 		zone->zone_start_pfn = start_pfn;
685 
686 	zone->spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - zone->zone_start_pfn;
687 }
688 
689 static void __meminit resize_pgdat_range(struct pglist_data *pgdat, unsigned long start_pfn,
690                                      unsigned long nr_pages)
691 {
692 	unsigned long old_end_pfn = pgdat_end_pfn(pgdat);
693 
694 	if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
695 		pgdat->node_start_pfn = start_pfn;
696 
697 	pgdat->node_spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - pgdat->node_start_pfn;
698 
699 }
700 /*
701  * Associate the pfn range with the given zone, initializing the memmaps
702  * and resizing the pgdat/zone data to span the added pages. After this
703  * call, all affected pages are PG_reserved.
704  *
705  * All aligned pageblocks are initialized to the specified migratetype
706  * (usually MIGRATE_MOVABLE). Besides setting the migratetype, no related
707  * zone stats (e.g., nr_isolate_pageblock) are touched.
708  */
709 void __ref move_pfn_range_to_zone(struct zone *zone, unsigned long start_pfn,
710 				  unsigned long nr_pages,
711 				  struct vmem_altmap *altmap, int migratetype)
712 {
713 	struct pglist_data *pgdat = zone->zone_pgdat;
714 	int nid = pgdat->node_id;
715 	unsigned long flags;
716 
717 	clear_zone_contiguous(zone);
718 
719 	/* TODO Huh pgdat is irqsave while zone is not. It used to be like that before */
720 	pgdat_resize_lock(pgdat, &flags);
721 	zone_span_writelock(zone);
722 	if (zone_is_empty(zone))
723 		init_currently_empty_zone(zone, start_pfn, nr_pages);
724 	resize_zone_range(zone, start_pfn, nr_pages);
725 	zone_span_writeunlock(zone);
726 	resize_pgdat_range(pgdat, start_pfn, nr_pages);
727 	pgdat_resize_unlock(pgdat, &flags);
728 
729 	/*
730 	 * TODO now we have a visible range of pages which are not associated
731 	 * with their zone properly. Not nice but set_pfnblock_flags_mask
732 	 * expects the zone spans the pfn range. All the pages in the range
733 	 * are reserved so nobody should be touching them so we should be safe
734 	 */
735 	memmap_init_zone(nr_pages, nid, zone_idx(zone), start_pfn,
736 			 MEMINIT_HOTPLUG, altmap, migratetype);
737 
738 	set_zone_contiguous(zone);
739 }
740 
741 /*
742  * Returns a default kernel memory zone for the given pfn range.
743  * If no kernel zone covers this pfn range it will automatically go
744  * to the ZONE_NORMAL.
745  */
746 static struct zone *default_kernel_zone_for_pfn(int nid, unsigned long start_pfn,
747 		unsigned long nr_pages)
748 {
749 	struct pglist_data *pgdat = NODE_DATA(nid);
750 	int zid;
751 
752 	for (zid = 0; zid <= ZONE_NORMAL; zid++) {
753 		struct zone *zone = &pgdat->node_zones[zid];
754 
755 		if (zone_intersects(zone, start_pfn, nr_pages))
756 			return zone;
757 	}
758 
759 	return &pgdat->node_zones[ZONE_NORMAL];
760 }
761 
762 static inline struct zone *default_zone_for_pfn(int nid, unsigned long start_pfn,
763 		unsigned long nr_pages)
764 {
765 	struct zone *kernel_zone = default_kernel_zone_for_pfn(nid, start_pfn,
766 			nr_pages);
767 	struct zone *movable_zone = &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
768 	bool in_kernel = zone_intersects(kernel_zone, start_pfn, nr_pages);
769 	bool in_movable = zone_intersects(movable_zone, start_pfn, nr_pages);
770 
771 	/*
772 	 * We inherit the existing zone in a simple case where zones do not
773 	 * overlap in the given range
774 	 */
775 	if (in_kernel ^ in_movable)
776 		return (in_kernel) ? kernel_zone : movable_zone;
777 
778 	/*
779 	 * If the range doesn't belong to any zone or two zones overlap in the
780 	 * given range then we use movable zone only if movable_node is
781 	 * enabled because we always online to a kernel zone by default.
782 	 */
783 	return movable_node_enabled ? movable_zone : kernel_zone;
784 }
785 
786 struct zone * zone_for_pfn_range(int online_type, int nid, unsigned start_pfn,
787 		unsigned long nr_pages)
788 {
789 	if (online_type == MMOP_ONLINE_KERNEL)
790 		return default_kernel_zone_for_pfn(nid, start_pfn, nr_pages);
791 
792 	if (online_type == MMOP_ONLINE_MOVABLE)
793 		return &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
794 
795 	return default_zone_for_pfn(nid, start_pfn, nr_pages);
796 }
797 
798 int __ref online_pages(unsigned long pfn, unsigned long nr_pages,
799 		       int online_type, int nid)
800 {
801 	unsigned long flags;
802 	struct zone *zone;
803 	int need_zonelists_rebuild = 0;
804 	int ret;
805 	struct memory_notify arg;
806 
807 	/* We can only online full sections (e.g., SECTION_IS_ONLINE) */
808 	if (WARN_ON_ONCE(!nr_pages ||
809 			 !IS_ALIGNED(pfn | nr_pages, PAGES_PER_SECTION)))
810 		return -EINVAL;
811 
812 	mem_hotplug_begin();
813 
814 	/* associate pfn range with the zone */
815 	zone = zone_for_pfn_range(online_type, nid, pfn, nr_pages);
816 	move_pfn_range_to_zone(zone, pfn, nr_pages, NULL, MIGRATE_ISOLATE);
817 
818 	arg.start_pfn = pfn;
819 	arg.nr_pages = nr_pages;
820 	node_states_check_changes_online(nr_pages, zone, &arg);
821 
822 	ret = memory_notify(MEM_GOING_ONLINE, &arg);
823 	ret = notifier_to_errno(ret);
824 	if (ret)
825 		goto failed_addition;
826 
827 	/*
828 	 * Fixup the number of isolated pageblocks before marking the sections
829 	 * onlining, such that undo_isolate_page_range() works correctly.
830 	 */
831 	spin_lock_irqsave(&zone->lock, flags);
832 	zone->nr_isolate_pageblock += nr_pages / pageblock_nr_pages;
833 	spin_unlock_irqrestore(&zone->lock, flags);
834 
835 	/*
836 	 * If this zone is not populated, then it is not in zonelist.
837 	 * This means the page allocator ignores this zone.
838 	 * So, zonelist must be updated after online.
839 	 */
840 	if (!populated_zone(zone)) {
841 		need_zonelists_rebuild = 1;
842 		setup_zone_pageset(zone);
843 	}
844 
845 	online_pages_range(pfn, nr_pages);
846 	zone->present_pages += nr_pages;
847 
848 	pgdat_resize_lock(zone->zone_pgdat, &flags);
849 	zone->zone_pgdat->node_present_pages += nr_pages;
850 	pgdat_resize_unlock(zone->zone_pgdat, &flags);
851 
852 	node_states_set_node(nid, &arg);
853 	if (need_zonelists_rebuild)
854 		build_all_zonelists(NULL);
855 	zone_pcp_update(zone);
856 
857 	/* Basic onlining is complete, allow allocation of onlined pages. */
858 	undo_isolate_page_range(pfn, pfn + nr_pages, MIGRATE_MOVABLE);
859 
860 	/*
861 	 * When exposing larger, physically contiguous memory areas to the
862 	 * buddy, shuffling in the buddy (when freeing onlined pages, putting
863 	 * them either to the head or the tail of the freelist) is only helpful
864 	 * for maintaining the shuffle, but not for creating the initial
865 	 * shuffle. Shuffle the whole zone to make sure the just onlined pages
866 	 * are properly distributed across the whole freelist. Make sure to
867 	 * shuffle once pageblocks are no longer isolated.
868 	 */
869 	shuffle_zone(zone);
870 
871 	init_per_zone_wmark_min();
872 
873 	kswapd_run(nid);
874 	kcompactd_run(nid);
875 
876 	writeback_set_ratelimit();
877 
878 	memory_notify(MEM_ONLINE, &arg);
879 	mem_hotplug_done();
880 	return 0;
881 
882 failed_addition:
883 	pr_debug("online_pages [mem %#010llx-%#010llx] failed\n",
884 		 (unsigned long long) pfn << PAGE_SHIFT,
885 		 (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1);
886 	memory_notify(MEM_CANCEL_ONLINE, &arg);
887 	remove_pfn_range_from_zone(zone, pfn, nr_pages);
888 	mem_hotplug_done();
889 	return ret;
890 }
891 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
892 
893 static void reset_node_present_pages(pg_data_t *pgdat)
894 {
895 	struct zone *z;
896 
897 	for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
898 		z->present_pages = 0;
899 
900 	pgdat->node_present_pages = 0;
901 }
902 
903 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
904 static pg_data_t __ref *hotadd_new_pgdat(int nid)
905 {
906 	struct pglist_data *pgdat;
907 
908 	pgdat = NODE_DATA(nid);
909 	if (!pgdat) {
910 		pgdat = arch_alloc_nodedata(nid);
911 		if (!pgdat)
912 			return NULL;
913 
914 		pgdat->per_cpu_nodestats =
915 			alloc_percpu(struct per_cpu_nodestat);
916 		arch_refresh_nodedata(nid, pgdat);
917 	} else {
918 		int cpu;
919 		/*
920 		 * Reset the nr_zones, order and highest_zoneidx before reuse.
921 		 * Note that kswapd will init kswapd_highest_zoneidx properly
922 		 * when it starts in the near future.
923 		 */
924 		pgdat->nr_zones = 0;
925 		pgdat->kswapd_order = 0;
926 		pgdat->kswapd_highest_zoneidx = 0;
927 		for_each_online_cpu(cpu) {
928 			struct per_cpu_nodestat *p;
929 
930 			p = per_cpu_ptr(pgdat->per_cpu_nodestats, cpu);
931 			memset(p, 0, sizeof(*p));
932 		}
933 	}
934 
935 	/* we can use NODE_DATA(nid) from here */
936 	pgdat->node_id = nid;
937 	pgdat->node_start_pfn = 0;
938 
939 	/* init node's zones as empty zones, we don't have any present pages.*/
940 	free_area_init_core_hotplug(nid);
941 
942 	/*
943 	 * The node we allocated has no zone fallback lists. For avoiding
944 	 * to access not-initialized zonelist, build here.
945 	 */
946 	build_all_zonelists(pgdat);
947 
948 	/*
949 	 * When memory is hot-added, all the memory is in offline state. So
950 	 * clear all zones' present_pages because they will be updated in
951 	 * online_pages() and offline_pages().
952 	 */
953 	reset_node_managed_pages(pgdat);
954 	reset_node_present_pages(pgdat);
955 
956 	return pgdat;
957 }
958 
959 static void rollback_node_hotadd(int nid)
960 {
961 	pg_data_t *pgdat = NODE_DATA(nid);
962 
963 	arch_refresh_nodedata(nid, NULL);
964 	free_percpu(pgdat->per_cpu_nodestats);
965 	arch_free_nodedata(pgdat);
966 }
967 
968 
969 /**
970  * try_online_node - online a node if offlined
971  * @nid: the node ID
972  * @set_node_online: Whether we want to online the node
973  * called by cpu_up() to online a node without onlined memory.
974  *
975  * Returns:
976  * 1 -> a new node has been allocated
977  * 0 -> the node is already online
978  * -ENOMEM -> the node could not be allocated
979  */
980 static int __try_online_node(int nid, bool set_node_online)
981 {
982 	pg_data_t *pgdat;
983 	int ret = 1;
984 
985 	if (node_online(nid))
986 		return 0;
987 
988 	pgdat = hotadd_new_pgdat(nid);
989 	if (!pgdat) {
990 		pr_err("Cannot online node %d due to NULL pgdat\n", nid);
991 		ret = -ENOMEM;
992 		goto out;
993 	}
994 
995 	if (set_node_online) {
996 		node_set_online(nid);
997 		ret = register_one_node(nid);
998 		BUG_ON(ret);
999 	}
1000 out:
1001 	return ret;
1002 }
1003 
1004 /*
1005  * Users of this function always want to online/register the node
1006  */
1007 int try_online_node(int nid)
1008 {
1009 	int ret;
1010 
1011 	mem_hotplug_begin();
1012 	ret =  __try_online_node(nid, true);
1013 	mem_hotplug_done();
1014 	return ret;
1015 }
1016 
1017 static int check_hotplug_memory_range(u64 start, u64 size)
1018 {
1019 	/* memory range must be block size aligned */
1020 	if (!size || !IS_ALIGNED(start, memory_block_size_bytes()) ||
1021 	    !IS_ALIGNED(size, memory_block_size_bytes())) {
1022 		pr_err("Block size [%#lx] unaligned hotplug range: start %#llx, size %#llx",
1023 		       memory_block_size_bytes(), start, size);
1024 		return -EINVAL;
1025 	}
1026 
1027 	return 0;
1028 }
1029 
1030 static int online_memory_block(struct memory_block *mem, void *arg)
1031 {
1032 	mem->online_type = memhp_default_online_type;
1033 	return device_online(&mem->dev);
1034 }
1035 
1036 /*
1037  * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1038  * and online/offline operations (triggered e.g. by sysfs).
1039  *
1040  * we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG
1041  */
1042 int __ref add_memory_resource(int nid, struct resource *res, mhp_t mhp_flags)
1043 {
1044 	struct mhp_params params = { .pgprot = PAGE_KERNEL };
1045 	u64 start, size;
1046 	bool new_node = false;
1047 	int ret;
1048 
1049 	start = res->start;
1050 	size = resource_size(res);
1051 
1052 	ret = check_hotplug_memory_range(start, size);
1053 	if (ret)
1054 		return ret;
1055 
1056 	if (!node_possible(nid)) {
1057 		WARN(1, "node %d was absent from the node_possible_map\n", nid);
1058 		return -EINVAL;
1059 	}
1060 
1061 	mem_hotplug_begin();
1062 
1063 	if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK))
1064 		memblock_add_node(start, size, nid);
1065 
1066 	ret = __try_online_node(nid, false);
1067 	if (ret < 0)
1068 		goto error;
1069 	new_node = ret;
1070 
1071 	/* call arch's memory hotadd */
1072 	ret = arch_add_memory(nid, start, size, &params);
1073 	if (ret < 0)
1074 		goto error;
1075 
1076 	/* create memory block devices after memory was added */
1077 	ret = create_memory_block_devices(start, size);
1078 	if (ret) {
1079 		arch_remove_memory(nid, start, size, NULL);
1080 		goto error;
1081 	}
1082 
1083 	if (new_node) {
1084 		/* If sysfs file of new node can't be created, cpu on the node
1085 		 * can't be hot-added. There is no rollback way now.
1086 		 * So, check by BUG_ON() to catch it reluctantly..
1087 		 * We online node here. We can't roll back from here.
1088 		 */
1089 		node_set_online(nid);
1090 		ret = __register_one_node(nid);
1091 		BUG_ON(ret);
1092 	}
1093 
1094 	/* link memory sections under this node.*/
1095 	ret = link_mem_sections(nid, PFN_DOWN(start), PFN_UP(start + size - 1),
1096 				MEMINIT_HOTPLUG);
1097 	BUG_ON(ret);
1098 
1099 	/* create new memmap entry */
1100 	if (!strcmp(res->name, "System RAM"))
1101 		firmware_map_add_hotplug(start, start + size, "System RAM");
1102 
1103 	/* device_online() will take the lock when calling online_pages() */
1104 	mem_hotplug_done();
1105 
1106 	/*
1107 	 * In case we're allowed to merge the resource, flag it and trigger
1108 	 * merging now that adding succeeded.
1109 	 */
1110 	if (mhp_flags & MEMHP_MERGE_RESOURCE)
1111 		merge_system_ram_resource(res);
1112 
1113 	/* online pages if requested */
1114 	if (memhp_default_online_type != MMOP_OFFLINE)
1115 		walk_memory_blocks(start, size, NULL, online_memory_block);
1116 
1117 	return ret;
1118 error:
1119 	/* rollback pgdat allocation and others */
1120 	if (new_node)
1121 		rollback_node_hotadd(nid);
1122 	if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK))
1123 		memblock_remove(start, size);
1124 	mem_hotplug_done();
1125 	return ret;
1126 }
1127 
1128 /* requires device_hotplug_lock, see add_memory_resource() */
1129 int __ref __add_memory(int nid, u64 start, u64 size, mhp_t mhp_flags)
1130 {
1131 	struct resource *res;
1132 	int ret;
1133 
1134 	res = register_memory_resource(start, size, "System RAM");
1135 	if (IS_ERR(res))
1136 		return PTR_ERR(res);
1137 
1138 	ret = add_memory_resource(nid, res, mhp_flags);
1139 	if (ret < 0)
1140 		release_memory_resource(res);
1141 	return ret;
1142 }
1143 
1144 int add_memory(int nid, u64 start, u64 size, mhp_t mhp_flags)
1145 {
1146 	int rc;
1147 
1148 	lock_device_hotplug();
1149 	rc = __add_memory(nid, start, size, mhp_flags);
1150 	unlock_device_hotplug();
1151 
1152 	return rc;
1153 }
1154 EXPORT_SYMBOL_GPL(add_memory);
1155 
1156 /*
1157  * Add special, driver-managed memory to the system as system RAM. Such
1158  * memory is not exposed via the raw firmware-provided memmap as system
1159  * RAM, instead, it is detected and added by a driver - during cold boot,
1160  * after a reboot, and after kexec.
1161  *
1162  * Reasons why this memory should not be used for the initial memmap of a
1163  * kexec kernel or for placing kexec images:
1164  * - The booting kernel is in charge of determining how this memory will be
1165  *   used (e.g., use persistent memory as system RAM)
1166  * - Coordination with a hypervisor is required before this memory
1167  *   can be used (e.g., inaccessible parts).
1168  *
1169  * For this memory, no entries in /sys/firmware/memmap ("raw firmware-provided
1170  * memory map") are created. Also, the created memory resource is flagged
1171  * with IORESOURCE_SYSRAM_DRIVER_MANAGED, so in-kernel users can special-case
1172  * this memory as well (esp., not place kexec images onto it).
1173  *
1174  * The resource_name (visible via /proc/iomem) has to have the format
1175  * "System RAM ($DRIVER)".
1176  */
1177 int add_memory_driver_managed(int nid, u64 start, u64 size,
1178 			      const char *resource_name, mhp_t mhp_flags)
1179 {
1180 	struct resource *res;
1181 	int rc;
1182 
1183 	if (!resource_name ||
1184 	    strstr(resource_name, "System RAM (") != resource_name ||
1185 	    resource_name[strlen(resource_name) - 1] != ')')
1186 		return -EINVAL;
1187 
1188 	lock_device_hotplug();
1189 
1190 	res = register_memory_resource(start, size, resource_name);
1191 	if (IS_ERR(res)) {
1192 		rc = PTR_ERR(res);
1193 		goto out_unlock;
1194 	}
1195 
1196 	rc = add_memory_resource(nid, res, mhp_flags);
1197 	if (rc < 0)
1198 		release_memory_resource(res);
1199 
1200 out_unlock:
1201 	unlock_device_hotplug();
1202 	return rc;
1203 }
1204 EXPORT_SYMBOL_GPL(add_memory_driver_managed);
1205 
1206 #ifdef CONFIG_MEMORY_HOTREMOVE
1207 /*
1208  * Confirm all pages in a range [start, end) belong to the same zone (skipping
1209  * memory holes). When true, return the zone.
1210  */
1211 struct zone *test_pages_in_a_zone(unsigned long start_pfn,
1212 				  unsigned long end_pfn)
1213 {
1214 	unsigned long pfn, sec_end_pfn;
1215 	struct zone *zone = NULL;
1216 	struct page *page;
1217 	int i;
1218 	for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1);
1219 	     pfn < end_pfn;
1220 	     pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) {
1221 		/* Make sure the memory section is present first */
1222 		if (!present_section_nr(pfn_to_section_nr(pfn)))
1223 			continue;
1224 		for (; pfn < sec_end_pfn && pfn < end_pfn;
1225 		     pfn += MAX_ORDER_NR_PAGES) {
1226 			i = 0;
1227 			/* This is just a CONFIG_HOLES_IN_ZONE check.*/
1228 			while ((i < MAX_ORDER_NR_PAGES) &&
1229 				!pfn_valid_within(pfn + i))
1230 				i++;
1231 			if (i == MAX_ORDER_NR_PAGES || pfn + i >= end_pfn)
1232 				continue;
1233 			/* Check if we got outside of the zone */
1234 			if (zone && !zone_spans_pfn(zone, pfn + i))
1235 				return NULL;
1236 			page = pfn_to_page(pfn + i);
1237 			if (zone && page_zone(page) != zone)
1238 				return NULL;
1239 			zone = page_zone(page);
1240 		}
1241 	}
1242 
1243 	return zone;
1244 }
1245 
1246 /*
1247  * Scan pfn range [start,end) to find movable/migratable pages (LRU pages,
1248  * non-lru movable pages and hugepages). Will skip over most unmovable
1249  * pages (esp., pages that can be skipped when offlining), but bail out on
1250  * definitely unmovable pages.
1251  *
1252  * Returns:
1253  *	0 in case a movable page is found and movable_pfn was updated.
1254  *	-ENOENT in case no movable page was found.
1255  *	-EBUSY in case a definitely unmovable page was found.
1256  */
1257 static int scan_movable_pages(unsigned long start, unsigned long end,
1258 			      unsigned long *movable_pfn)
1259 {
1260 	unsigned long pfn;
1261 
1262 	for (pfn = start; pfn < end; pfn++) {
1263 		struct page *page, *head;
1264 		unsigned long skip;
1265 
1266 		if (!pfn_valid(pfn))
1267 			continue;
1268 		page = pfn_to_page(pfn);
1269 		if (PageLRU(page))
1270 			goto found;
1271 		if (__PageMovable(page))
1272 			goto found;
1273 
1274 		/*
1275 		 * PageOffline() pages that are not marked __PageMovable() and
1276 		 * have a reference count > 0 (after MEM_GOING_OFFLINE) are
1277 		 * definitely unmovable. If their reference count would be 0,
1278 		 * they could at least be skipped when offlining memory.
1279 		 */
1280 		if (PageOffline(page) && page_count(page))
1281 			return -EBUSY;
1282 
1283 		if (!PageHuge(page))
1284 			continue;
1285 		head = compound_head(page);
1286 		if (page_huge_active(head))
1287 			goto found;
1288 		skip = compound_nr(head) - (page - head);
1289 		pfn += skip - 1;
1290 	}
1291 	return -ENOENT;
1292 found:
1293 	*movable_pfn = pfn;
1294 	return 0;
1295 }
1296 
1297 static struct page *new_node_page(struct page *page, unsigned long private)
1298 {
1299 	nodemask_t nmask = node_states[N_MEMORY];
1300 	struct migration_target_control mtc = {
1301 		.nid = page_to_nid(page),
1302 		.nmask = &nmask,
1303 		.gfp_mask = GFP_USER | __GFP_MOVABLE | __GFP_RETRY_MAYFAIL,
1304 	};
1305 
1306 	/*
1307 	 * try to allocate from a different node but reuse this node if there
1308 	 * are no other online nodes to be used (e.g. we are offlining a part
1309 	 * of the only existing node)
1310 	 */
1311 	node_clear(mtc.nid, nmask);
1312 	if (nodes_empty(nmask))
1313 		node_set(mtc.nid, nmask);
1314 
1315 	return alloc_migration_target(page, (unsigned long)&mtc);
1316 }
1317 
1318 static int
1319 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1320 {
1321 	unsigned long pfn;
1322 	struct page *page, *head;
1323 	int ret = 0;
1324 	LIST_HEAD(source);
1325 
1326 	for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1327 		if (!pfn_valid(pfn))
1328 			continue;
1329 		page = pfn_to_page(pfn);
1330 		head = compound_head(page);
1331 
1332 		if (PageHuge(page)) {
1333 			pfn = page_to_pfn(head) + compound_nr(head) - 1;
1334 			isolate_huge_page(head, &source);
1335 			continue;
1336 		} else if (PageTransHuge(page))
1337 			pfn = page_to_pfn(head) + thp_nr_pages(page) - 1;
1338 
1339 		/*
1340 		 * HWPoison pages have elevated reference counts so the migration would
1341 		 * fail on them. It also doesn't make any sense to migrate them in the
1342 		 * first place. Still try to unmap such a page in case it is still mapped
1343 		 * (e.g. current hwpoison implementation doesn't unmap KSM pages but keep
1344 		 * the unmap as the catch all safety net).
1345 		 */
1346 		if (PageHWPoison(page)) {
1347 			if (WARN_ON(PageLRU(page)))
1348 				isolate_lru_page(page);
1349 			if (page_mapped(page))
1350 				try_to_unmap(page, TTU_IGNORE_MLOCK | TTU_IGNORE_ACCESS);
1351 			continue;
1352 		}
1353 
1354 		if (!get_page_unless_zero(page))
1355 			continue;
1356 		/*
1357 		 * We can skip free pages. And we can deal with pages on
1358 		 * LRU and non-lru movable pages.
1359 		 */
1360 		if (PageLRU(page))
1361 			ret = isolate_lru_page(page);
1362 		else
1363 			ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE);
1364 		if (!ret) { /* Success */
1365 			list_add_tail(&page->lru, &source);
1366 			if (!__PageMovable(page))
1367 				inc_node_page_state(page, NR_ISOLATED_ANON +
1368 						    page_is_file_lru(page));
1369 
1370 		} else {
1371 			pr_warn("failed to isolate pfn %lx\n", pfn);
1372 			dump_page(page, "isolation failed");
1373 		}
1374 		put_page(page);
1375 	}
1376 	if (!list_empty(&source)) {
1377 		/* Allocate a new page from the nearest neighbor node */
1378 		ret = migrate_pages(&source, new_node_page, NULL, 0,
1379 					MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
1380 		if (ret) {
1381 			list_for_each_entry(page, &source, lru) {
1382 				pr_warn("migrating pfn %lx failed ret:%d ",
1383 				       page_to_pfn(page), ret);
1384 				dump_page(page, "migration failure");
1385 			}
1386 			putback_movable_pages(&source);
1387 		}
1388 	}
1389 
1390 	return ret;
1391 }
1392 
1393 static int __init cmdline_parse_movable_node(char *p)
1394 {
1395 	movable_node_enabled = true;
1396 	return 0;
1397 }
1398 early_param("movable_node", cmdline_parse_movable_node);
1399 
1400 /* check which state of node_states will be changed when offline memory */
1401 static void node_states_check_changes_offline(unsigned long nr_pages,
1402 		struct zone *zone, struct memory_notify *arg)
1403 {
1404 	struct pglist_data *pgdat = zone->zone_pgdat;
1405 	unsigned long present_pages = 0;
1406 	enum zone_type zt;
1407 
1408 	arg->status_change_nid = NUMA_NO_NODE;
1409 	arg->status_change_nid_normal = NUMA_NO_NODE;
1410 	arg->status_change_nid_high = NUMA_NO_NODE;
1411 
1412 	/*
1413 	 * Check whether node_states[N_NORMAL_MEMORY] will be changed.
1414 	 * If the memory to be offline is within the range
1415 	 * [0..ZONE_NORMAL], and it is the last present memory there,
1416 	 * the zones in that range will become empty after the offlining,
1417 	 * thus we can determine that we need to clear the node from
1418 	 * node_states[N_NORMAL_MEMORY].
1419 	 */
1420 	for (zt = 0; zt <= ZONE_NORMAL; zt++)
1421 		present_pages += pgdat->node_zones[zt].present_pages;
1422 	if (zone_idx(zone) <= ZONE_NORMAL && nr_pages >= present_pages)
1423 		arg->status_change_nid_normal = zone_to_nid(zone);
1424 
1425 #ifdef CONFIG_HIGHMEM
1426 	/*
1427 	 * node_states[N_HIGH_MEMORY] contains nodes which
1428 	 * have normal memory or high memory.
1429 	 * Here we add the present_pages belonging to ZONE_HIGHMEM.
1430 	 * If the zone is within the range of [0..ZONE_HIGHMEM), and
1431 	 * we determine that the zones in that range become empty,
1432 	 * we need to clear the node for N_HIGH_MEMORY.
1433 	 */
1434 	present_pages += pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1435 	if (zone_idx(zone) <= ZONE_HIGHMEM && nr_pages >= present_pages)
1436 		arg->status_change_nid_high = zone_to_nid(zone);
1437 #endif
1438 
1439 	/*
1440 	 * We have accounted the pages from [0..ZONE_NORMAL), and
1441 	 * in case of CONFIG_HIGHMEM the pages from ZONE_HIGHMEM
1442 	 * as well.
1443 	 * Here we count the possible pages from ZONE_MOVABLE.
1444 	 * If after having accounted all the pages, we see that the nr_pages
1445 	 * to be offlined is over or equal to the accounted pages,
1446 	 * we know that the node will become empty, and so, we can clear
1447 	 * it for N_MEMORY as well.
1448 	 */
1449 	present_pages += pgdat->node_zones[ZONE_MOVABLE].present_pages;
1450 
1451 	if (nr_pages >= present_pages)
1452 		arg->status_change_nid = zone_to_nid(zone);
1453 }
1454 
1455 static void node_states_clear_node(int node, struct memory_notify *arg)
1456 {
1457 	if (arg->status_change_nid_normal >= 0)
1458 		node_clear_state(node, N_NORMAL_MEMORY);
1459 
1460 	if (arg->status_change_nid_high >= 0)
1461 		node_clear_state(node, N_HIGH_MEMORY);
1462 
1463 	if (arg->status_change_nid >= 0)
1464 		node_clear_state(node, N_MEMORY);
1465 }
1466 
1467 static int count_system_ram_pages_cb(unsigned long start_pfn,
1468 				     unsigned long nr_pages, void *data)
1469 {
1470 	unsigned long *nr_system_ram_pages = data;
1471 
1472 	*nr_system_ram_pages += nr_pages;
1473 	return 0;
1474 }
1475 
1476 int __ref offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1477 {
1478 	const unsigned long end_pfn = start_pfn + nr_pages;
1479 	unsigned long pfn, system_ram_pages = 0;
1480 	unsigned long flags;
1481 	struct zone *zone;
1482 	struct memory_notify arg;
1483 	int ret, node;
1484 	char *reason;
1485 
1486 	/* We can only offline full sections (e.g., SECTION_IS_ONLINE) */
1487 	if (WARN_ON_ONCE(!nr_pages ||
1488 			 !IS_ALIGNED(start_pfn | nr_pages, PAGES_PER_SECTION)))
1489 		return -EINVAL;
1490 
1491 	mem_hotplug_begin();
1492 
1493 	/*
1494 	 * Don't allow to offline memory blocks that contain holes.
1495 	 * Consequently, memory blocks with holes can never get onlined
1496 	 * via the hotplug path - online_pages() - as hotplugged memory has
1497 	 * no holes. This way, we e.g., don't have to worry about marking
1498 	 * memory holes PG_reserved, don't need pfn_valid() checks, and can
1499 	 * avoid using walk_system_ram_range() later.
1500 	 */
1501 	walk_system_ram_range(start_pfn, nr_pages, &system_ram_pages,
1502 			      count_system_ram_pages_cb);
1503 	if (system_ram_pages != nr_pages) {
1504 		ret = -EINVAL;
1505 		reason = "memory holes";
1506 		goto failed_removal;
1507 	}
1508 
1509 	/* This makes hotplug much easier...and readable.
1510 	   we assume this for now. .*/
1511 	zone = test_pages_in_a_zone(start_pfn, end_pfn);
1512 	if (!zone) {
1513 		ret = -EINVAL;
1514 		reason = "multizone range";
1515 		goto failed_removal;
1516 	}
1517 	node = zone_to_nid(zone);
1518 
1519 	/* set above range as isolated */
1520 	ret = start_isolate_page_range(start_pfn, end_pfn,
1521 				       MIGRATE_MOVABLE,
1522 				       MEMORY_OFFLINE | REPORT_FAILURE);
1523 	if (ret) {
1524 		reason = "failure to isolate range";
1525 		goto failed_removal;
1526 	}
1527 
1528 	arg.start_pfn = start_pfn;
1529 	arg.nr_pages = nr_pages;
1530 	node_states_check_changes_offline(nr_pages, zone, &arg);
1531 
1532 	ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1533 	ret = notifier_to_errno(ret);
1534 	if (ret) {
1535 		reason = "notifier failure";
1536 		goto failed_removal_isolated;
1537 	}
1538 
1539 	do {
1540 		pfn = start_pfn;
1541 		do {
1542 			if (signal_pending(current)) {
1543 				ret = -EINTR;
1544 				reason = "signal backoff";
1545 				goto failed_removal_isolated;
1546 			}
1547 
1548 			cond_resched();
1549 			lru_add_drain_all();
1550 
1551 			ret = scan_movable_pages(pfn, end_pfn, &pfn);
1552 			if (!ret) {
1553 				/*
1554 				 * TODO: fatal migration failures should bail
1555 				 * out
1556 				 */
1557 				do_migrate_range(pfn, end_pfn);
1558 			}
1559 		} while (!ret);
1560 
1561 		if (ret != -ENOENT) {
1562 			reason = "unmovable page";
1563 			goto failed_removal_isolated;
1564 		}
1565 
1566 		/*
1567 		 * Dissolve free hugepages in the memory block before doing
1568 		 * offlining actually in order to make hugetlbfs's object
1569 		 * counting consistent.
1570 		 */
1571 		ret = dissolve_free_huge_pages(start_pfn, end_pfn);
1572 		if (ret) {
1573 			reason = "failure to dissolve huge pages";
1574 			goto failed_removal_isolated;
1575 		}
1576 
1577 		/*
1578 		 * per-cpu pages are drained in start_isolate_page_range, but if
1579 		 * there are still pages that are not free, make sure that we
1580 		 * drain again, because when we isolated range we might
1581 		 * have raced with another thread that was adding pages to pcp
1582 		 * list.
1583 		 *
1584 		 * Forward progress should be still guaranteed because
1585 		 * pages on the pcp list can only belong to MOVABLE_ZONE
1586 		 * because has_unmovable_pages explicitly checks for
1587 		 * PageBuddy on freed pages on other zones.
1588 		 */
1589 		ret = test_pages_isolated(start_pfn, end_pfn, MEMORY_OFFLINE);
1590 		if (ret)
1591 			drain_all_pages(zone);
1592 	} while (ret);
1593 
1594 	/* Mark all sections offline and remove free pages from the buddy. */
1595 	__offline_isolated_pages(start_pfn, end_pfn);
1596 	pr_info("Offlined Pages %ld\n", nr_pages);
1597 
1598 	/*
1599 	 * The memory sections are marked offline, and the pageblock flags
1600 	 * effectively stale; nobody should be touching them. Fixup the number
1601 	 * of isolated pageblocks, memory onlining will properly revert this.
1602 	 */
1603 	spin_lock_irqsave(&zone->lock, flags);
1604 	zone->nr_isolate_pageblock -= nr_pages / pageblock_nr_pages;
1605 	spin_unlock_irqrestore(&zone->lock, flags);
1606 
1607 	/* removal success */
1608 	adjust_managed_page_count(pfn_to_page(start_pfn), -nr_pages);
1609 	zone->present_pages -= nr_pages;
1610 
1611 	pgdat_resize_lock(zone->zone_pgdat, &flags);
1612 	zone->zone_pgdat->node_present_pages -= nr_pages;
1613 	pgdat_resize_unlock(zone->zone_pgdat, &flags);
1614 
1615 	init_per_zone_wmark_min();
1616 
1617 	if (!populated_zone(zone)) {
1618 		zone_pcp_reset(zone);
1619 		build_all_zonelists(NULL);
1620 	} else
1621 		zone_pcp_update(zone);
1622 
1623 	node_states_clear_node(node, &arg);
1624 	if (arg.status_change_nid >= 0) {
1625 		kswapd_stop(node);
1626 		kcompactd_stop(node);
1627 	}
1628 
1629 	writeback_set_ratelimit();
1630 
1631 	memory_notify(MEM_OFFLINE, &arg);
1632 	remove_pfn_range_from_zone(zone, start_pfn, nr_pages);
1633 	mem_hotplug_done();
1634 	return 0;
1635 
1636 failed_removal_isolated:
1637 	undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1638 	memory_notify(MEM_CANCEL_OFFLINE, &arg);
1639 failed_removal:
1640 	pr_debug("memory offlining [mem %#010llx-%#010llx] failed due to %s\n",
1641 		 (unsigned long long) start_pfn << PAGE_SHIFT,
1642 		 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1,
1643 		 reason);
1644 	/* pushback to free area */
1645 	mem_hotplug_done();
1646 	return ret;
1647 }
1648 
1649 static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
1650 {
1651 	int ret = !is_memblock_offlined(mem);
1652 
1653 	if (unlikely(ret)) {
1654 		phys_addr_t beginpa, endpa;
1655 
1656 		beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
1657 		endpa = beginpa + memory_block_size_bytes() - 1;
1658 		pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n",
1659 			&beginpa, &endpa);
1660 
1661 		return -EBUSY;
1662 	}
1663 	return 0;
1664 }
1665 
1666 static int check_cpu_on_node(pg_data_t *pgdat)
1667 {
1668 	int cpu;
1669 
1670 	for_each_present_cpu(cpu) {
1671 		if (cpu_to_node(cpu) == pgdat->node_id)
1672 			/*
1673 			 * the cpu on this node isn't removed, and we can't
1674 			 * offline this node.
1675 			 */
1676 			return -EBUSY;
1677 	}
1678 
1679 	return 0;
1680 }
1681 
1682 static int check_no_memblock_for_node_cb(struct memory_block *mem, void *arg)
1683 {
1684 	int nid = *(int *)arg;
1685 
1686 	/*
1687 	 * If a memory block belongs to multiple nodes, the stored nid is not
1688 	 * reliable. However, such blocks are always online (e.g., cannot get
1689 	 * offlined) and, therefore, are still spanned by the node.
1690 	 */
1691 	return mem->nid == nid ? -EEXIST : 0;
1692 }
1693 
1694 /**
1695  * try_offline_node
1696  * @nid: the node ID
1697  *
1698  * Offline a node if all memory sections and cpus of the node are removed.
1699  *
1700  * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1701  * and online/offline operations before this call.
1702  */
1703 void try_offline_node(int nid)
1704 {
1705 	pg_data_t *pgdat = NODE_DATA(nid);
1706 	int rc;
1707 
1708 	/*
1709 	 * If the node still spans pages (especially ZONE_DEVICE), don't
1710 	 * offline it. A node spans memory after move_pfn_range_to_zone(),
1711 	 * e.g., after the memory block was onlined.
1712 	 */
1713 	if (pgdat->node_spanned_pages)
1714 		return;
1715 
1716 	/*
1717 	 * Especially offline memory blocks might not be spanned by the
1718 	 * node. They will get spanned by the node once they get onlined.
1719 	 * However, they link to the node in sysfs and can get onlined later.
1720 	 */
1721 	rc = for_each_memory_block(&nid, check_no_memblock_for_node_cb);
1722 	if (rc)
1723 		return;
1724 
1725 	if (check_cpu_on_node(pgdat))
1726 		return;
1727 
1728 	/*
1729 	 * all memory/cpu of this node are removed, we can offline this
1730 	 * node now.
1731 	 */
1732 	node_set_offline(nid);
1733 	unregister_one_node(nid);
1734 }
1735 EXPORT_SYMBOL(try_offline_node);
1736 
1737 static int __ref try_remove_memory(int nid, u64 start, u64 size)
1738 {
1739 	int rc = 0;
1740 
1741 	BUG_ON(check_hotplug_memory_range(start, size));
1742 
1743 	/*
1744 	 * All memory blocks must be offlined before removing memory.  Check
1745 	 * whether all memory blocks in question are offline and return error
1746 	 * if this is not the case.
1747 	 */
1748 	rc = walk_memory_blocks(start, size, NULL, check_memblock_offlined_cb);
1749 	if (rc)
1750 		return rc;
1751 
1752 	/* remove memmap entry */
1753 	firmware_map_remove(start, start + size, "System RAM");
1754 
1755 	/*
1756 	 * Memory block device removal under the device_hotplug_lock is
1757 	 * a barrier against racing online attempts.
1758 	 */
1759 	remove_memory_block_devices(start, size);
1760 
1761 	mem_hotplug_begin();
1762 
1763 	arch_remove_memory(nid, start, size, NULL);
1764 
1765 	if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK)) {
1766 		memblock_free(start, size);
1767 		memblock_remove(start, size);
1768 	}
1769 
1770 	release_mem_region_adjustable(start, size);
1771 
1772 	try_offline_node(nid);
1773 
1774 	mem_hotplug_done();
1775 	return 0;
1776 }
1777 
1778 /**
1779  * remove_memory
1780  * @nid: the node ID
1781  * @start: physical address of the region to remove
1782  * @size: size of the region to remove
1783  *
1784  * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1785  * and online/offline operations before this call, as required by
1786  * try_offline_node().
1787  */
1788 void __remove_memory(int nid, u64 start, u64 size)
1789 {
1790 
1791 	/*
1792 	 * trigger BUG() if some memory is not offlined prior to calling this
1793 	 * function
1794 	 */
1795 	if (try_remove_memory(nid, start, size))
1796 		BUG();
1797 }
1798 
1799 /*
1800  * Remove memory if every memory block is offline, otherwise return -EBUSY is
1801  * some memory is not offline
1802  */
1803 int remove_memory(int nid, u64 start, u64 size)
1804 {
1805 	int rc;
1806 
1807 	lock_device_hotplug();
1808 	rc  = try_remove_memory(nid, start, size);
1809 	unlock_device_hotplug();
1810 
1811 	return rc;
1812 }
1813 EXPORT_SYMBOL_GPL(remove_memory);
1814 
1815 /*
1816  * Try to offline and remove a memory block. Might take a long time to
1817  * finish in case memory is still in use. Primarily useful for memory devices
1818  * that logically unplugged all memory (so it's no longer in use) and want to
1819  * offline + remove the memory block.
1820  */
1821 int offline_and_remove_memory(int nid, u64 start, u64 size)
1822 {
1823 	struct memory_block *mem;
1824 	int rc = -EINVAL;
1825 
1826 	if (!IS_ALIGNED(start, memory_block_size_bytes()) ||
1827 	    size != memory_block_size_bytes())
1828 		return rc;
1829 
1830 	lock_device_hotplug();
1831 	mem = find_memory_block(__pfn_to_section(PFN_DOWN(start)));
1832 	if (mem)
1833 		rc = device_offline(&mem->dev);
1834 	/* Ignore if the device is already offline. */
1835 	if (rc > 0)
1836 		rc = 0;
1837 
1838 	/*
1839 	 * In case we succeeded to offline the memory block, remove it.
1840 	 * This cannot fail as it cannot get onlined in the meantime.
1841 	 */
1842 	if (!rc) {
1843 		rc = try_remove_memory(nid, start, size);
1844 		WARN_ON_ONCE(rc);
1845 	}
1846 	unlock_device_hotplug();
1847 
1848 	return rc;
1849 }
1850 EXPORT_SYMBOL_GPL(offline_and_remove_memory);
1851 #endif /* CONFIG_MEMORY_HOTREMOVE */
1852