xref: /linux/mm/memory_hotplug.c (revision 722ecdbce68a87de2d9296f91308f44ea900a039)
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/vmalloc.h>
25 #include <linux/ioport.h>
26 #include <linux/delay.h>
27 #include <linux/migrate.h>
28 #include <linux/page-isolation.h>
29 #include <linux/pfn.h>
30 #include <linux/suspend.h>
31 #include <linux/mm_inline.h>
32 #include <linux/firmware-map.h>
33 #include <linux/stop_machine.h>
34 #include <linux/hugetlb.h>
35 #include <linux/memblock.h>
36 #include <linux/compaction.h>
37 #include <linux/rmap.h>
38 #include <linux/module.h>
39 
40 #include <asm/tlbflush.h>
41 
42 #include "internal.h"
43 #include "shuffle.h"
44 
45 #ifdef CONFIG_MHP_MEMMAP_ON_MEMORY
46 static int memmap_on_memory_set(const char *val, const struct kernel_param *kp)
47 {
48 	if (hugetlb_optimize_vmemmap_enabled())
49 		return 0;
50 	return param_set_bool(val, kp);
51 }
52 
53 static const struct kernel_param_ops memmap_on_memory_ops = {
54 	.flags	= KERNEL_PARAM_OPS_FL_NOARG,
55 	.set	= memmap_on_memory_set,
56 	.get	= param_get_bool,
57 };
58 
59 /*
60  * memory_hotplug.memmap_on_memory parameter
61  */
62 static bool memmap_on_memory __ro_after_init;
63 module_param_cb(memmap_on_memory, &memmap_on_memory_ops, &memmap_on_memory, 0444);
64 MODULE_PARM_DESC(memmap_on_memory, "Enable memmap on memory for memory hotplug");
65 
66 bool mhp_memmap_on_memory(void)
67 {
68 	return memmap_on_memory;
69 }
70 #endif
71 
72 enum {
73 	ONLINE_POLICY_CONTIG_ZONES = 0,
74 	ONLINE_POLICY_AUTO_MOVABLE,
75 };
76 
77 static const char * const online_policy_to_str[] = {
78 	[ONLINE_POLICY_CONTIG_ZONES] = "contig-zones",
79 	[ONLINE_POLICY_AUTO_MOVABLE] = "auto-movable",
80 };
81 
82 static int set_online_policy(const char *val, const struct kernel_param *kp)
83 {
84 	int ret = sysfs_match_string(online_policy_to_str, val);
85 
86 	if (ret < 0)
87 		return ret;
88 	*((int *)kp->arg) = ret;
89 	return 0;
90 }
91 
92 static int get_online_policy(char *buffer, const struct kernel_param *kp)
93 {
94 	return sprintf(buffer, "%s\n", online_policy_to_str[*((int *)kp->arg)]);
95 }
96 
97 /*
98  * memory_hotplug.online_policy: configure online behavior when onlining without
99  * specifying a zone (MMOP_ONLINE)
100  *
101  * "contig-zones": keep zone contiguous
102  * "auto-movable": online memory to ZONE_MOVABLE if the configuration
103  *                 (auto_movable_ratio, auto_movable_numa_aware) allows for it
104  */
105 static int online_policy __read_mostly = ONLINE_POLICY_CONTIG_ZONES;
106 static const struct kernel_param_ops online_policy_ops = {
107 	.set = set_online_policy,
108 	.get = get_online_policy,
109 };
110 module_param_cb(online_policy, &online_policy_ops, &online_policy, 0644);
111 MODULE_PARM_DESC(online_policy,
112 		"Set the online policy (\"contig-zones\", \"auto-movable\") "
113 		"Default: \"contig-zones\"");
114 
115 /*
116  * memory_hotplug.auto_movable_ratio: specify maximum MOVABLE:KERNEL ratio
117  *
118  * The ratio represent an upper limit and the kernel might decide to not
119  * online some memory to ZONE_MOVABLE -- e.g., because hotplugged KERNEL memory
120  * doesn't allow for more MOVABLE memory.
121  */
122 static unsigned int auto_movable_ratio __read_mostly = 301;
123 module_param(auto_movable_ratio, uint, 0644);
124 MODULE_PARM_DESC(auto_movable_ratio,
125 		"Set the maximum ratio of MOVABLE:KERNEL memory in the system "
126 		"in percent for \"auto-movable\" online policy. Default: 301");
127 
128 /*
129  * memory_hotplug.auto_movable_numa_aware: consider numa node stats
130  */
131 #ifdef CONFIG_NUMA
132 static bool auto_movable_numa_aware __read_mostly = true;
133 module_param(auto_movable_numa_aware, bool, 0644);
134 MODULE_PARM_DESC(auto_movable_numa_aware,
135 		"Consider numa node stats in addition to global stats in "
136 		"\"auto-movable\" online policy. Default: true");
137 #endif /* CONFIG_NUMA */
138 
139 /*
140  * online_page_callback contains pointer to current page onlining function.
141  * Initially it is generic_online_page(). If it is required it could be
142  * changed by calling set_online_page_callback() for callback registration
143  * and restore_online_page_callback() for generic callback restore.
144  */
145 
146 static online_page_callback_t online_page_callback = generic_online_page;
147 static DEFINE_MUTEX(online_page_callback_lock);
148 
149 DEFINE_STATIC_PERCPU_RWSEM(mem_hotplug_lock);
150 
151 void get_online_mems(void)
152 {
153 	percpu_down_read(&mem_hotplug_lock);
154 }
155 
156 void put_online_mems(void)
157 {
158 	percpu_up_read(&mem_hotplug_lock);
159 }
160 
161 bool movable_node_enabled = false;
162 
163 #ifndef CONFIG_MEMORY_HOTPLUG_DEFAULT_ONLINE
164 int mhp_default_online_type = MMOP_OFFLINE;
165 #else
166 int mhp_default_online_type = MMOP_ONLINE;
167 #endif
168 
169 static int __init setup_memhp_default_state(char *str)
170 {
171 	const int online_type = mhp_online_type_from_str(str);
172 
173 	if (online_type >= 0)
174 		mhp_default_online_type = online_type;
175 
176 	return 1;
177 }
178 __setup("memhp_default_state=", setup_memhp_default_state);
179 
180 void mem_hotplug_begin(void)
181 {
182 	cpus_read_lock();
183 	percpu_down_write(&mem_hotplug_lock);
184 }
185 
186 void mem_hotplug_done(void)
187 {
188 	percpu_up_write(&mem_hotplug_lock);
189 	cpus_read_unlock();
190 }
191 
192 u64 max_mem_size = U64_MAX;
193 
194 /* add this memory to iomem resource */
195 static struct resource *register_memory_resource(u64 start, u64 size,
196 						 const char *resource_name)
197 {
198 	struct resource *res;
199 	unsigned long flags =  IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
200 
201 	if (strcmp(resource_name, "System RAM"))
202 		flags |= IORESOURCE_SYSRAM_DRIVER_MANAGED;
203 
204 	if (!mhp_range_allowed(start, size, true))
205 		return ERR_PTR(-E2BIG);
206 
207 	/*
208 	 * Make sure value parsed from 'mem=' only restricts memory adding
209 	 * while booting, so that memory hotplug won't be impacted. Please
210 	 * refer to document of 'mem=' in kernel-parameters.txt for more
211 	 * details.
212 	 */
213 	if (start + size > max_mem_size && system_state < SYSTEM_RUNNING)
214 		return ERR_PTR(-E2BIG);
215 
216 	/*
217 	 * Request ownership of the new memory range.  This might be
218 	 * a child of an existing resource that was present but
219 	 * not marked as busy.
220 	 */
221 	res = __request_region(&iomem_resource, start, size,
222 			       resource_name, flags);
223 
224 	if (!res) {
225 		pr_debug("Unable to reserve System RAM region: %016llx->%016llx\n",
226 				start, start + size);
227 		return ERR_PTR(-EEXIST);
228 	}
229 	return res;
230 }
231 
232 static void release_memory_resource(struct resource *res)
233 {
234 	if (!res)
235 		return;
236 	release_resource(res);
237 	kfree(res);
238 }
239 
240 static int check_pfn_span(unsigned long pfn, unsigned long nr_pages,
241 		const char *reason)
242 {
243 	/*
244 	 * Disallow all operations smaller than a sub-section and only
245 	 * allow operations smaller than a section for
246 	 * SPARSEMEM_VMEMMAP. Note that check_hotplug_memory_range()
247 	 * enforces a larger memory_block_size_bytes() granularity for
248 	 * memory that will be marked online, so this check should only
249 	 * fire for direct arch_{add,remove}_memory() users outside of
250 	 * add_memory_resource().
251 	 */
252 	unsigned long min_align;
253 
254 	if (IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP))
255 		min_align = PAGES_PER_SUBSECTION;
256 	else
257 		min_align = PAGES_PER_SECTION;
258 	if (!IS_ALIGNED(pfn, min_align)
259 			|| !IS_ALIGNED(nr_pages, min_align)) {
260 		WARN(1, "Misaligned __%s_pages start: %#lx end: #%lx\n",
261 				reason, pfn, pfn + nr_pages - 1);
262 		return -EINVAL;
263 	}
264 	return 0;
265 }
266 
267 /*
268  * Return page for the valid pfn only if the page is online. All pfn
269  * walkers which rely on the fully initialized page->flags and others
270  * should use this rather than pfn_valid && pfn_to_page
271  */
272 struct page *pfn_to_online_page(unsigned long pfn)
273 {
274 	unsigned long nr = pfn_to_section_nr(pfn);
275 	struct dev_pagemap *pgmap;
276 	struct mem_section *ms;
277 
278 	if (nr >= NR_MEM_SECTIONS)
279 		return NULL;
280 
281 	ms = __nr_to_section(nr);
282 	if (!online_section(ms))
283 		return NULL;
284 
285 	/*
286 	 * Save some code text when online_section() +
287 	 * pfn_section_valid() are sufficient.
288 	 */
289 	if (IS_ENABLED(CONFIG_HAVE_ARCH_PFN_VALID) && !pfn_valid(pfn))
290 		return NULL;
291 
292 	if (!pfn_section_valid(ms, pfn))
293 		return NULL;
294 
295 	if (!online_device_section(ms))
296 		return pfn_to_page(pfn);
297 
298 	/*
299 	 * Slowpath: when ZONE_DEVICE collides with
300 	 * ZONE_{NORMAL,MOVABLE} within the same section some pfns in
301 	 * the section may be 'offline' but 'valid'. Only
302 	 * get_dev_pagemap() can determine sub-section online status.
303 	 */
304 	pgmap = get_dev_pagemap(pfn, NULL);
305 	put_dev_pagemap(pgmap);
306 
307 	/* The presence of a pgmap indicates ZONE_DEVICE offline pfn */
308 	if (pgmap)
309 		return NULL;
310 
311 	return pfn_to_page(pfn);
312 }
313 EXPORT_SYMBOL_GPL(pfn_to_online_page);
314 
315 int __ref __add_pages(int nid, unsigned long pfn, unsigned long nr_pages,
316 		struct mhp_params *params)
317 {
318 	const unsigned long end_pfn = pfn + nr_pages;
319 	unsigned long cur_nr_pages;
320 	int err;
321 	struct vmem_altmap *altmap = params->altmap;
322 
323 	if (WARN_ON_ONCE(!pgprot_val(params->pgprot)))
324 		return -EINVAL;
325 
326 	VM_BUG_ON(!mhp_range_allowed(PFN_PHYS(pfn), nr_pages * PAGE_SIZE, false));
327 
328 	if (altmap) {
329 		/*
330 		 * Validate altmap is within bounds of the total request
331 		 */
332 		if (altmap->base_pfn != pfn
333 				|| vmem_altmap_offset(altmap) > nr_pages) {
334 			pr_warn_once("memory add fail, invalid altmap\n");
335 			return -EINVAL;
336 		}
337 		altmap->alloc = 0;
338 	}
339 
340 	err = check_pfn_span(pfn, nr_pages, "add");
341 	if (err)
342 		return err;
343 
344 	for (; pfn < end_pfn; pfn += cur_nr_pages) {
345 		/* Select all remaining pages up to the next section boundary */
346 		cur_nr_pages = min(end_pfn - pfn,
347 				   SECTION_ALIGN_UP(pfn + 1) - pfn);
348 		err = sparse_add_section(nid, pfn, cur_nr_pages, altmap,
349 					 params->pgmap);
350 		if (err)
351 			break;
352 		cond_resched();
353 	}
354 	vmemmap_populate_print_last();
355 	return err;
356 }
357 
358 /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
359 static unsigned long find_smallest_section_pfn(int nid, struct zone *zone,
360 				     unsigned long start_pfn,
361 				     unsigned long end_pfn)
362 {
363 	for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SUBSECTION) {
364 		if (unlikely(!pfn_to_online_page(start_pfn)))
365 			continue;
366 
367 		if (unlikely(pfn_to_nid(start_pfn) != nid))
368 			continue;
369 
370 		if (zone != page_zone(pfn_to_page(start_pfn)))
371 			continue;
372 
373 		return start_pfn;
374 	}
375 
376 	return 0;
377 }
378 
379 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
380 static unsigned long find_biggest_section_pfn(int nid, struct zone *zone,
381 				    unsigned long start_pfn,
382 				    unsigned long end_pfn)
383 {
384 	unsigned long pfn;
385 
386 	/* pfn is the end pfn of a memory section. */
387 	pfn = end_pfn - 1;
388 	for (; pfn >= start_pfn; pfn -= PAGES_PER_SUBSECTION) {
389 		if (unlikely(!pfn_to_online_page(pfn)))
390 			continue;
391 
392 		if (unlikely(pfn_to_nid(pfn) != nid))
393 			continue;
394 
395 		if (zone != page_zone(pfn_to_page(pfn)))
396 			continue;
397 
398 		return pfn;
399 	}
400 
401 	return 0;
402 }
403 
404 static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
405 			     unsigned long end_pfn)
406 {
407 	unsigned long pfn;
408 	int nid = zone_to_nid(zone);
409 
410 	if (zone->zone_start_pfn == start_pfn) {
411 		/*
412 		 * If the section is smallest section in the zone, it need
413 		 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
414 		 * In this case, we find second smallest valid mem_section
415 		 * for shrinking zone.
416 		 */
417 		pfn = find_smallest_section_pfn(nid, zone, end_pfn,
418 						zone_end_pfn(zone));
419 		if (pfn) {
420 			zone->spanned_pages = zone_end_pfn(zone) - pfn;
421 			zone->zone_start_pfn = pfn;
422 		} else {
423 			zone->zone_start_pfn = 0;
424 			zone->spanned_pages = 0;
425 		}
426 	} else if (zone_end_pfn(zone) == end_pfn) {
427 		/*
428 		 * If the section is biggest section in the zone, it need
429 		 * shrink zone->spanned_pages.
430 		 * In this case, we find second biggest valid mem_section for
431 		 * shrinking zone.
432 		 */
433 		pfn = find_biggest_section_pfn(nid, zone, zone->zone_start_pfn,
434 					       start_pfn);
435 		if (pfn)
436 			zone->spanned_pages = pfn - zone->zone_start_pfn + 1;
437 		else {
438 			zone->zone_start_pfn = 0;
439 			zone->spanned_pages = 0;
440 		}
441 	}
442 }
443 
444 static void update_pgdat_span(struct pglist_data *pgdat)
445 {
446 	unsigned long node_start_pfn = 0, node_end_pfn = 0;
447 	struct zone *zone;
448 
449 	for (zone = pgdat->node_zones;
450 	     zone < pgdat->node_zones + MAX_NR_ZONES; zone++) {
451 		unsigned long end_pfn = zone_end_pfn(zone);
452 
453 		/* No need to lock the zones, they can't change. */
454 		if (!zone->spanned_pages)
455 			continue;
456 		if (!node_end_pfn) {
457 			node_start_pfn = zone->zone_start_pfn;
458 			node_end_pfn = end_pfn;
459 			continue;
460 		}
461 
462 		if (end_pfn > node_end_pfn)
463 			node_end_pfn = end_pfn;
464 		if (zone->zone_start_pfn < node_start_pfn)
465 			node_start_pfn = zone->zone_start_pfn;
466 	}
467 
468 	pgdat->node_start_pfn = node_start_pfn;
469 	pgdat->node_spanned_pages = node_end_pfn - node_start_pfn;
470 }
471 
472 void __ref remove_pfn_range_from_zone(struct zone *zone,
473 				      unsigned long start_pfn,
474 				      unsigned long nr_pages)
475 {
476 	const unsigned long end_pfn = start_pfn + nr_pages;
477 	struct pglist_data *pgdat = zone->zone_pgdat;
478 	unsigned long pfn, cur_nr_pages;
479 
480 	/* Poison struct pages because they are now uninitialized again. */
481 	for (pfn = start_pfn; pfn < end_pfn; pfn += cur_nr_pages) {
482 		cond_resched();
483 
484 		/* Select all remaining pages up to the next section boundary */
485 		cur_nr_pages =
486 			min(end_pfn - pfn, SECTION_ALIGN_UP(pfn + 1) - pfn);
487 		page_init_poison(pfn_to_page(pfn),
488 				 sizeof(struct page) * cur_nr_pages);
489 	}
490 
491 	/*
492 	 * Zone shrinking code cannot properly deal with ZONE_DEVICE. So
493 	 * we will not try to shrink the zones - which is okay as
494 	 * set_zone_contiguous() cannot deal with ZONE_DEVICE either way.
495 	 */
496 	if (zone_is_zone_device(zone))
497 		return;
498 
499 	clear_zone_contiguous(zone);
500 
501 	shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
502 	update_pgdat_span(pgdat);
503 
504 	set_zone_contiguous(zone);
505 }
506 
507 static void __remove_section(unsigned long pfn, unsigned long nr_pages,
508 			     unsigned long map_offset,
509 			     struct vmem_altmap *altmap)
510 {
511 	struct mem_section *ms = __pfn_to_section(pfn);
512 
513 	if (WARN_ON_ONCE(!valid_section(ms)))
514 		return;
515 
516 	sparse_remove_section(ms, pfn, nr_pages, map_offset, altmap);
517 }
518 
519 /**
520  * __remove_pages() - remove sections of pages
521  * @pfn: starting pageframe (must be aligned to start of a section)
522  * @nr_pages: number of pages to remove (must be multiple of section size)
523  * @altmap: alternative device page map or %NULL if default memmap is used
524  *
525  * Generic helper function to remove section mappings and sysfs entries
526  * for the section of the memory we are removing. Caller needs to make
527  * sure that pages are marked reserved and zones are adjust properly by
528  * calling offline_pages().
529  */
530 void __remove_pages(unsigned long pfn, unsigned long nr_pages,
531 		    struct vmem_altmap *altmap)
532 {
533 	const unsigned long end_pfn = pfn + nr_pages;
534 	unsigned long cur_nr_pages;
535 	unsigned long map_offset = 0;
536 
537 	map_offset = vmem_altmap_offset(altmap);
538 
539 	if (check_pfn_span(pfn, nr_pages, "remove"))
540 		return;
541 
542 	for (; pfn < end_pfn; pfn += cur_nr_pages) {
543 		cond_resched();
544 		/* Select all remaining pages up to the next section boundary */
545 		cur_nr_pages = min(end_pfn - pfn,
546 				   SECTION_ALIGN_UP(pfn + 1) - pfn);
547 		__remove_section(pfn, cur_nr_pages, map_offset, altmap);
548 		map_offset = 0;
549 	}
550 }
551 
552 int set_online_page_callback(online_page_callback_t callback)
553 {
554 	int rc = -EINVAL;
555 
556 	get_online_mems();
557 	mutex_lock(&online_page_callback_lock);
558 
559 	if (online_page_callback == generic_online_page) {
560 		online_page_callback = callback;
561 		rc = 0;
562 	}
563 
564 	mutex_unlock(&online_page_callback_lock);
565 	put_online_mems();
566 
567 	return rc;
568 }
569 EXPORT_SYMBOL_GPL(set_online_page_callback);
570 
571 int restore_online_page_callback(online_page_callback_t callback)
572 {
573 	int rc = -EINVAL;
574 
575 	get_online_mems();
576 	mutex_lock(&online_page_callback_lock);
577 
578 	if (online_page_callback == callback) {
579 		online_page_callback = generic_online_page;
580 		rc = 0;
581 	}
582 
583 	mutex_unlock(&online_page_callback_lock);
584 	put_online_mems();
585 
586 	return rc;
587 }
588 EXPORT_SYMBOL_GPL(restore_online_page_callback);
589 
590 void generic_online_page(struct page *page, unsigned int order)
591 {
592 	/*
593 	 * Freeing the page with debug_pagealloc enabled will try to unmap it,
594 	 * so we should map it first. This is better than introducing a special
595 	 * case in page freeing fast path.
596 	 */
597 	debug_pagealloc_map_pages(page, 1 << order);
598 	__free_pages_core(page, order);
599 	totalram_pages_add(1UL << order);
600 }
601 EXPORT_SYMBOL_GPL(generic_online_page);
602 
603 static void online_pages_range(unsigned long start_pfn, unsigned long nr_pages)
604 {
605 	const unsigned long end_pfn = start_pfn + nr_pages;
606 	unsigned long pfn;
607 
608 	/*
609 	 * Online the pages in MAX_ORDER - 1 aligned chunks. The callback might
610 	 * decide to not expose all pages to the buddy (e.g., expose them
611 	 * later). We account all pages as being online and belonging to this
612 	 * zone ("present").
613 	 * When using memmap_on_memory, the range might not be aligned to
614 	 * MAX_ORDER_NR_PAGES - 1, but pageblock aligned. __ffs() will detect
615 	 * this and the first chunk to online will be pageblock_nr_pages.
616 	 */
617 	for (pfn = start_pfn; pfn < end_pfn;) {
618 		int order = min(MAX_ORDER - 1UL, __ffs(pfn));
619 
620 		(*online_page_callback)(pfn_to_page(pfn), order);
621 		pfn += (1UL << order);
622 	}
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 
637 	if (!node_state(nid, N_MEMORY))
638 		arg->status_change_nid = nid;
639 	if (zone_idx(zone) <= ZONE_NORMAL && !node_state(nid, N_NORMAL_MEMORY))
640 		arg->status_change_nid_normal = nid;
641 }
642 
643 static void node_states_set_node(int node, struct memory_notify *arg)
644 {
645 	if (arg->status_change_nid_normal >= 0)
646 		node_set_state(node, N_NORMAL_MEMORY);
647 
648 	if (arg->status_change_nid >= 0)
649 		node_set_state(node, N_MEMORY);
650 }
651 
652 static void __meminit resize_zone_range(struct zone *zone, unsigned long start_pfn,
653 		unsigned long nr_pages)
654 {
655 	unsigned long old_end_pfn = zone_end_pfn(zone);
656 
657 	if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
658 		zone->zone_start_pfn = start_pfn;
659 
660 	zone->spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - zone->zone_start_pfn;
661 }
662 
663 static void __meminit resize_pgdat_range(struct pglist_data *pgdat, unsigned long start_pfn,
664                                      unsigned long nr_pages)
665 {
666 	unsigned long old_end_pfn = pgdat_end_pfn(pgdat);
667 
668 	if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
669 		pgdat->node_start_pfn = start_pfn;
670 
671 	pgdat->node_spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - pgdat->node_start_pfn;
672 
673 }
674 
675 static void section_taint_zone_device(unsigned long pfn)
676 {
677 	struct mem_section *ms = __pfn_to_section(pfn);
678 
679 	ms->section_mem_map |= SECTION_TAINT_ZONE_DEVICE;
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 
698 	clear_zone_contiguous(zone);
699 
700 	if (zone_is_empty(zone))
701 		init_currently_empty_zone(zone, start_pfn, nr_pages);
702 	resize_zone_range(zone, start_pfn, nr_pages);
703 	resize_pgdat_range(pgdat, start_pfn, nr_pages);
704 
705 	/*
706 	 * Subsection population requires care in pfn_to_online_page().
707 	 * Set the taint to enable the slow path detection of
708 	 * ZONE_DEVICE pages in an otherwise  ZONE_{NORMAL,MOVABLE}
709 	 * section.
710 	 */
711 	if (zone_is_zone_device(zone)) {
712 		if (!IS_ALIGNED(start_pfn, PAGES_PER_SECTION))
713 			section_taint_zone_device(start_pfn);
714 		if (!IS_ALIGNED(start_pfn + nr_pages, PAGES_PER_SECTION))
715 			section_taint_zone_device(start_pfn + nr_pages);
716 	}
717 
718 	/*
719 	 * TODO now we have a visible range of pages which are not associated
720 	 * with their zone properly. Not nice but set_pfnblock_flags_mask
721 	 * expects the zone spans the pfn range. All the pages in the range
722 	 * are reserved so nobody should be touching them so we should be safe
723 	 */
724 	memmap_init_range(nr_pages, nid, zone_idx(zone), start_pfn, 0,
725 			 MEMINIT_HOTPLUG, altmap, migratetype);
726 
727 	set_zone_contiguous(zone);
728 }
729 
730 struct auto_movable_stats {
731 	unsigned long kernel_early_pages;
732 	unsigned long movable_pages;
733 };
734 
735 static void auto_movable_stats_account_zone(struct auto_movable_stats *stats,
736 					    struct zone *zone)
737 {
738 	if (zone_idx(zone) == ZONE_MOVABLE) {
739 		stats->movable_pages += zone->present_pages;
740 	} else {
741 		stats->kernel_early_pages += zone->present_early_pages;
742 #ifdef CONFIG_CMA
743 		/*
744 		 * CMA pages (never on hotplugged memory) behave like
745 		 * ZONE_MOVABLE.
746 		 */
747 		stats->movable_pages += zone->cma_pages;
748 		stats->kernel_early_pages -= zone->cma_pages;
749 #endif /* CONFIG_CMA */
750 	}
751 }
752 struct auto_movable_group_stats {
753 	unsigned long movable_pages;
754 	unsigned long req_kernel_early_pages;
755 };
756 
757 static int auto_movable_stats_account_group(struct memory_group *group,
758 					   void *arg)
759 {
760 	const int ratio = READ_ONCE(auto_movable_ratio);
761 	struct auto_movable_group_stats *stats = arg;
762 	long pages;
763 
764 	/*
765 	 * We don't support modifying the config while the auto-movable online
766 	 * policy is already enabled. Just avoid the division by zero below.
767 	 */
768 	if (!ratio)
769 		return 0;
770 
771 	/*
772 	 * Calculate how many early kernel pages this group requires to
773 	 * satisfy the configured zone ratio.
774 	 */
775 	pages = group->present_movable_pages * 100 / ratio;
776 	pages -= group->present_kernel_pages;
777 
778 	if (pages > 0)
779 		stats->req_kernel_early_pages += pages;
780 	stats->movable_pages += group->present_movable_pages;
781 	return 0;
782 }
783 
784 static bool auto_movable_can_online_movable(int nid, struct memory_group *group,
785 					    unsigned long nr_pages)
786 {
787 	unsigned long kernel_early_pages, movable_pages;
788 	struct auto_movable_group_stats group_stats = {};
789 	struct auto_movable_stats stats = {};
790 	pg_data_t *pgdat = NODE_DATA(nid);
791 	struct zone *zone;
792 	int i;
793 
794 	/* Walk all relevant zones and collect MOVABLE vs. KERNEL stats. */
795 	if (nid == NUMA_NO_NODE) {
796 		/* TODO: cache values */
797 		for_each_populated_zone(zone)
798 			auto_movable_stats_account_zone(&stats, zone);
799 	} else {
800 		for (i = 0; i < MAX_NR_ZONES; i++) {
801 			zone = pgdat->node_zones + i;
802 			if (populated_zone(zone))
803 				auto_movable_stats_account_zone(&stats, zone);
804 		}
805 	}
806 
807 	kernel_early_pages = stats.kernel_early_pages;
808 	movable_pages = stats.movable_pages;
809 
810 	/*
811 	 * Kernel memory inside dynamic memory group allows for more MOVABLE
812 	 * memory within the same group. Remove the effect of all but the
813 	 * current group from the stats.
814 	 */
815 	walk_dynamic_memory_groups(nid, auto_movable_stats_account_group,
816 				   group, &group_stats);
817 	if (kernel_early_pages <= group_stats.req_kernel_early_pages)
818 		return false;
819 	kernel_early_pages -= group_stats.req_kernel_early_pages;
820 	movable_pages -= group_stats.movable_pages;
821 
822 	if (group && group->is_dynamic)
823 		kernel_early_pages += group->present_kernel_pages;
824 
825 	/*
826 	 * Test if we could online the given number of pages to ZONE_MOVABLE
827 	 * and still stay in the configured ratio.
828 	 */
829 	movable_pages += nr_pages;
830 	return movable_pages <= (auto_movable_ratio * kernel_early_pages) / 100;
831 }
832 
833 /*
834  * Returns a default kernel memory zone for the given pfn range.
835  * If no kernel zone covers this pfn range it will automatically go
836  * to the ZONE_NORMAL.
837  */
838 static struct zone *default_kernel_zone_for_pfn(int nid, unsigned long start_pfn,
839 		unsigned long nr_pages)
840 {
841 	struct pglist_data *pgdat = NODE_DATA(nid);
842 	int zid;
843 
844 	for (zid = 0; zid < ZONE_NORMAL; zid++) {
845 		struct zone *zone = &pgdat->node_zones[zid];
846 
847 		if (zone_intersects(zone, start_pfn, nr_pages))
848 			return zone;
849 	}
850 
851 	return &pgdat->node_zones[ZONE_NORMAL];
852 }
853 
854 /*
855  * Determine to which zone to online memory dynamically based on user
856  * configuration and system stats. We care about the following ratio:
857  *
858  *   MOVABLE : KERNEL
859  *
860  * Whereby MOVABLE is memory in ZONE_MOVABLE and KERNEL is memory in
861  * one of the kernel zones. CMA pages inside one of the kernel zones really
862  * behaves like ZONE_MOVABLE, so we treat them accordingly.
863  *
864  * We don't allow for hotplugged memory in a KERNEL zone to increase the
865  * amount of MOVABLE memory we can have, so we end up with:
866  *
867  *   MOVABLE : KERNEL_EARLY
868  *
869  * Whereby KERNEL_EARLY is memory in one of the kernel zones, available sinze
870  * boot. We base our calculation on KERNEL_EARLY internally, because:
871  *
872  * a) Hotplugged memory in one of the kernel zones can sometimes still get
873  *    hotunplugged, especially when hot(un)plugging individual memory blocks.
874  *    There is no coordination across memory devices, therefore "automatic"
875  *    hotunplugging, as implemented in hypervisors, could result in zone
876  *    imbalances.
877  * b) Early/boot memory in one of the kernel zones can usually not get
878  *    hotunplugged again (e.g., no firmware interface to unplug, fragmented
879  *    with unmovable allocations). While there are corner cases where it might
880  *    still work, it is barely relevant in practice.
881  *
882  * Exceptions are dynamic memory groups, which allow for more MOVABLE
883  * memory within the same memory group -- because in that case, there is
884  * coordination within the single memory device managed by a single driver.
885  *
886  * We rely on "present pages" instead of "managed pages", as the latter is
887  * highly unreliable and dynamic in virtualized environments, and does not
888  * consider boot time allocations. For example, memory ballooning adjusts the
889  * managed pages when inflating/deflating the balloon, and balloon compaction
890  * can even migrate inflated pages between zones.
891  *
892  * Using "present pages" is better but some things to keep in mind are:
893  *
894  * a) Some memblock allocations, such as for the crashkernel area, are
895  *    effectively unused by the kernel, yet they account to "present pages".
896  *    Fortunately, these allocations are comparatively small in relevant setups
897  *    (e.g., fraction of system memory).
898  * b) Some hotplugged memory blocks in virtualized environments, esecially
899  *    hotplugged by virtio-mem, look like they are completely present, however,
900  *    only parts of the memory block are actually currently usable.
901  *    "present pages" is an upper limit that can get reached at runtime. As
902  *    we base our calculations on KERNEL_EARLY, this is not an issue.
903  */
904 static struct zone *auto_movable_zone_for_pfn(int nid,
905 					      struct memory_group *group,
906 					      unsigned long pfn,
907 					      unsigned long nr_pages)
908 {
909 	unsigned long online_pages = 0, max_pages, end_pfn;
910 	struct page *page;
911 
912 	if (!auto_movable_ratio)
913 		goto kernel_zone;
914 
915 	if (group && !group->is_dynamic) {
916 		max_pages = group->s.max_pages;
917 		online_pages = group->present_movable_pages;
918 
919 		/* If anything is !MOVABLE online the rest !MOVABLE. */
920 		if (group->present_kernel_pages)
921 			goto kernel_zone;
922 	} else if (!group || group->d.unit_pages == nr_pages) {
923 		max_pages = nr_pages;
924 	} else {
925 		max_pages = group->d.unit_pages;
926 		/*
927 		 * Take a look at all online sections in the current unit.
928 		 * We can safely assume that all pages within a section belong
929 		 * to the same zone, because dynamic memory groups only deal
930 		 * with hotplugged memory.
931 		 */
932 		pfn = ALIGN_DOWN(pfn, group->d.unit_pages);
933 		end_pfn = pfn + group->d.unit_pages;
934 		for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
935 			page = pfn_to_online_page(pfn);
936 			if (!page)
937 				continue;
938 			/* If anything is !MOVABLE online the rest !MOVABLE. */
939 			if (page_zonenum(page) != ZONE_MOVABLE)
940 				goto kernel_zone;
941 			online_pages += PAGES_PER_SECTION;
942 		}
943 	}
944 
945 	/*
946 	 * Online MOVABLE if we could *currently* online all remaining parts
947 	 * MOVABLE. We expect to (add+) online them immediately next, so if
948 	 * nobody interferes, all will be MOVABLE if possible.
949 	 */
950 	nr_pages = max_pages - online_pages;
951 	if (!auto_movable_can_online_movable(NUMA_NO_NODE, group, nr_pages))
952 		goto kernel_zone;
953 
954 #ifdef CONFIG_NUMA
955 	if (auto_movable_numa_aware &&
956 	    !auto_movable_can_online_movable(nid, group, nr_pages))
957 		goto kernel_zone;
958 #endif /* CONFIG_NUMA */
959 
960 	return &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
961 kernel_zone:
962 	return default_kernel_zone_for_pfn(nid, pfn, nr_pages);
963 }
964 
965 static inline struct zone *default_zone_for_pfn(int nid, unsigned long start_pfn,
966 		unsigned long nr_pages)
967 {
968 	struct zone *kernel_zone = default_kernel_zone_for_pfn(nid, start_pfn,
969 			nr_pages);
970 	struct zone *movable_zone = &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
971 	bool in_kernel = zone_intersects(kernel_zone, start_pfn, nr_pages);
972 	bool in_movable = zone_intersects(movable_zone, start_pfn, nr_pages);
973 
974 	/*
975 	 * We inherit the existing zone in a simple case where zones do not
976 	 * overlap in the given range
977 	 */
978 	if (in_kernel ^ in_movable)
979 		return (in_kernel) ? kernel_zone : movable_zone;
980 
981 	/*
982 	 * If the range doesn't belong to any zone or two zones overlap in the
983 	 * given range then we use movable zone only if movable_node is
984 	 * enabled because we always online to a kernel zone by default.
985 	 */
986 	return movable_node_enabled ? movable_zone : kernel_zone;
987 }
988 
989 struct zone *zone_for_pfn_range(int online_type, int nid,
990 		struct memory_group *group, unsigned long start_pfn,
991 		unsigned long nr_pages)
992 {
993 	if (online_type == MMOP_ONLINE_KERNEL)
994 		return default_kernel_zone_for_pfn(nid, start_pfn, nr_pages);
995 
996 	if (online_type == MMOP_ONLINE_MOVABLE)
997 		return &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
998 
999 	if (online_policy == ONLINE_POLICY_AUTO_MOVABLE)
1000 		return auto_movable_zone_for_pfn(nid, group, start_pfn, nr_pages);
1001 
1002 	return default_zone_for_pfn(nid, start_pfn, nr_pages);
1003 }
1004 
1005 /*
1006  * This function should only be called by memory_block_{online,offline},
1007  * and {online,offline}_pages.
1008  */
1009 void adjust_present_page_count(struct page *page, struct memory_group *group,
1010 			       long nr_pages)
1011 {
1012 	struct zone *zone = page_zone(page);
1013 	const bool movable = zone_idx(zone) == ZONE_MOVABLE;
1014 
1015 	/*
1016 	 * We only support onlining/offlining/adding/removing of complete
1017 	 * memory blocks; therefore, either all is either early or hotplugged.
1018 	 */
1019 	if (early_section(__pfn_to_section(page_to_pfn(page))))
1020 		zone->present_early_pages += nr_pages;
1021 	zone->present_pages += nr_pages;
1022 	zone->zone_pgdat->node_present_pages += nr_pages;
1023 
1024 	if (group && movable)
1025 		group->present_movable_pages += nr_pages;
1026 	else if (group && !movable)
1027 		group->present_kernel_pages += nr_pages;
1028 }
1029 
1030 int mhp_init_memmap_on_memory(unsigned long pfn, unsigned long nr_pages,
1031 			      struct zone *zone)
1032 {
1033 	unsigned long end_pfn = pfn + nr_pages;
1034 	int ret;
1035 
1036 	ret = kasan_add_zero_shadow(__va(PFN_PHYS(pfn)), PFN_PHYS(nr_pages));
1037 	if (ret)
1038 		return ret;
1039 
1040 	move_pfn_range_to_zone(zone, pfn, nr_pages, NULL, MIGRATE_UNMOVABLE);
1041 
1042 	/*
1043 	 * It might be that the vmemmap_pages fully span sections. If that is
1044 	 * the case, mark those sections online here as otherwise they will be
1045 	 * left offline.
1046 	 */
1047 	if (nr_pages >= PAGES_PER_SECTION)
1048 	        online_mem_sections(pfn, ALIGN_DOWN(end_pfn, PAGES_PER_SECTION));
1049 
1050 	return ret;
1051 }
1052 
1053 void mhp_deinit_memmap_on_memory(unsigned long pfn, unsigned long nr_pages)
1054 {
1055 	unsigned long end_pfn = pfn + nr_pages;
1056 
1057 	/*
1058 	 * It might be that the vmemmap_pages fully span sections. If that is
1059 	 * the case, mark those sections offline here as otherwise they will be
1060 	 * left online.
1061 	 */
1062 	if (nr_pages >= PAGES_PER_SECTION)
1063 		offline_mem_sections(pfn, ALIGN_DOWN(end_pfn, PAGES_PER_SECTION));
1064 
1065         /*
1066 	 * The pages associated with this vmemmap have been offlined, so
1067 	 * we can reset its state here.
1068 	 */
1069 	remove_pfn_range_from_zone(page_zone(pfn_to_page(pfn)), pfn, nr_pages);
1070 	kasan_remove_zero_shadow(__va(PFN_PHYS(pfn)), PFN_PHYS(nr_pages));
1071 }
1072 
1073 int __ref online_pages(unsigned long pfn, unsigned long nr_pages,
1074 		       struct zone *zone, struct memory_group *group)
1075 {
1076 	unsigned long flags;
1077 	int need_zonelists_rebuild = 0;
1078 	const int nid = zone_to_nid(zone);
1079 	int ret;
1080 	struct memory_notify arg;
1081 
1082 	/*
1083 	 * {on,off}lining is constrained to full memory sections (or more
1084 	 * precisely to memory blocks from the user space POV).
1085 	 * memmap_on_memory is an exception because it reserves initial part
1086 	 * of the physical memory space for vmemmaps. That space is pageblock
1087 	 * aligned.
1088 	 */
1089 	if (WARN_ON_ONCE(!nr_pages ||
1090 			 !IS_ALIGNED(pfn, pageblock_nr_pages) ||
1091 			 !IS_ALIGNED(pfn + nr_pages, PAGES_PER_SECTION)))
1092 		return -EINVAL;
1093 
1094 	mem_hotplug_begin();
1095 
1096 	/* associate pfn range with the zone */
1097 	move_pfn_range_to_zone(zone, pfn, nr_pages, NULL, MIGRATE_ISOLATE);
1098 
1099 	arg.start_pfn = pfn;
1100 	arg.nr_pages = nr_pages;
1101 	node_states_check_changes_online(nr_pages, zone, &arg);
1102 
1103 	ret = memory_notify(MEM_GOING_ONLINE, &arg);
1104 	ret = notifier_to_errno(ret);
1105 	if (ret)
1106 		goto failed_addition;
1107 
1108 	/*
1109 	 * Fixup the number of isolated pageblocks before marking the sections
1110 	 * onlining, such that undo_isolate_page_range() works correctly.
1111 	 */
1112 	spin_lock_irqsave(&zone->lock, flags);
1113 	zone->nr_isolate_pageblock += nr_pages / pageblock_nr_pages;
1114 	spin_unlock_irqrestore(&zone->lock, flags);
1115 
1116 	/*
1117 	 * If this zone is not populated, then it is not in zonelist.
1118 	 * This means the page allocator ignores this zone.
1119 	 * So, zonelist must be updated after online.
1120 	 */
1121 	if (!populated_zone(zone)) {
1122 		need_zonelists_rebuild = 1;
1123 		setup_zone_pageset(zone);
1124 	}
1125 
1126 	online_pages_range(pfn, nr_pages);
1127 	adjust_present_page_count(pfn_to_page(pfn), group, nr_pages);
1128 
1129 	node_states_set_node(nid, &arg);
1130 	if (need_zonelists_rebuild)
1131 		build_all_zonelists(NULL);
1132 
1133 	/* Basic onlining is complete, allow allocation of onlined pages. */
1134 	undo_isolate_page_range(pfn, pfn + nr_pages, MIGRATE_MOVABLE);
1135 
1136 	/*
1137 	 * Freshly onlined pages aren't shuffled (e.g., all pages are placed to
1138 	 * the tail of the freelist when undoing isolation). Shuffle the whole
1139 	 * zone to make sure the just onlined pages are properly distributed
1140 	 * across the whole freelist - to create an initial shuffle.
1141 	 */
1142 	shuffle_zone(zone);
1143 
1144 	/* reinitialise watermarks and update pcp limits */
1145 	init_per_zone_wmark_min();
1146 
1147 	kswapd_run(nid);
1148 	kcompactd_run(nid);
1149 
1150 	writeback_set_ratelimit();
1151 
1152 	memory_notify(MEM_ONLINE, &arg);
1153 	mem_hotplug_done();
1154 	return 0;
1155 
1156 failed_addition:
1157 	pr_debug("online_pages [mem %#010llx-%#010llx] failed\n",
1158 		 (unsigned long long) pfn << PAGE_SHIFT,
1159 		 (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1);
1160 	memory_notify(MEM_CANCEL_ONLINE, &arg);
1161 	remove_pfn_range_from_zone(zone, pfn, nr_pages);
1162 	mem_hotplug_done();
1163 	return ret;
1164 }
1165 
1166 static void reset_node_present_pages(pg_data_t *pgdat)
1167 {
1168 	struct zone *z;
1169 
1170 	for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
1171 		z->present_pages = 0;
1172 
1173 	pgdat->node_present_pages = 0;
1174 }
1175 
1176 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1177 static pg_data_t __ref *hotadd_init_pgdat(int nid)
1178 {
1179 	struct pglist_data *pgdat;
1180 
1181 	/*
1182 	 * NODE_DATA is preallocated (free_area_init) but its internal
1183 	 * state is not allocated completely. Add missing pieces.
1184 	 * Completely offline nodes stay around and they just need
1185 	 * reintialization.
1186 	 */
1187 	pgdat = NODE_DATA(nid);
1188 
1189 	/* init node's zones as empty zones, we don't have any present pages.*/
1190 	free_area_init_core_hotplug(pgdat);
1191 
1192 	/*
1193 	 * The node we allocated has no zone fallback lists. For avoiding
1194 	 * to access not-initialized zonelist, build here.
1195 	 */
1196 	build_all_zonelists(pgdat);
1197 
1198 	/*
1199 	 * When memory is hot-added, all the memory is in offline state. So
1200 	 * clear all zones' present_pages because they will be updated in
1201 	 * online_pages() and offline_pages().
1202 	 * TODO: should be in free_area_init_core_hotplug?
1203 	 */
1204 	reset_node_managed_pages(pgdat);
1205 	reset_node_present_pages(pgdat);
1206 
1207 	return pgdat;
1208 }
1209 
1210 /*
1211  * __try_online_node - online a node if offlined
1212  * @nid: the node ID
1213  * @set_node_online: Whether we want to online the node
1214  * called by cpu_up() to online a node without onlined memory.
1215  *
1216  * Returns:
1217  * 1 -> a new node has been allocated
1218  * 0 -> the node is already online
1219  * -ENOMEM -> the node could not be allocated
1220  */
1221 static int __try_online_node(int nid, bool set_node_online)
1222 {
1223 	pg_data_t *pgdat;
1224 	int ret = 1;
1225 
1226 	if (node_online(nid))
1227 		return 0;
1228 
1229 	pgdat = hotadd_init_pgdat(nid);
1230 	if (!pgdat) {
1231 		pr_err("Cannot online node %d due to NULL pgdat\n", nid);
1232 		ret = -ENOMEM;
1233 		goto out;
1234 	}
1235 
1236 	if (set_node_online) {
1237 		node_set_online(nid);
1238 		ret = register_one_node(nid);
1239 		BUG_ON(ret);
1240 	}
1241 out:
1242 	return ret;
1243 }
1244 
1245 /*
1246  * Users of this function always want to online/register the node
1247  */
1248 int try_online_node(int nid)
1249 {
1250 	int ret;
1251 
1252 	mem_hotplug_begin();
1253 	ret =  __try_online_node(nid, true);
1254 	mem_hotplug_done();
1255 	return ret;
1256 }
1257 
1258 static int check_hotplug_memory_range(u64 start, u64 size)
1259 {
1260 	/* memory range must be block size aligned */
1261 	if (!size || !IS_ALIGNED(start, memory_block_size_bytes()) ||
1262 	    !IS_ALIGNED(size, memory_block_size_bytes())) {
1263 		pr_err("Block size [%#lx] unaligned hotplug range: start %#llx, size %#llx",
1264 		       memory_block_size_bytes(), start, size);
1265 		return -EINVAL;
1266 	}
1267 
1268 	return 0;
1269 }
1270 
1271 static int online_memory_block(struct memory_block *mem, void *arg)
1272 {
1273 	mem->online_type = mhp_default_online_type;
1274 	return device_online(&mem->dev);
1275 }
1276 
1277 bool mhp_supports_memmap_on_memory(unsigned long size)
1278 {
1279 	unsigned long nr_vmemmap_pages = size / PAGE_SIZE;
1280 	unsigned long vmemmap_size = nr_vmemmap_pages * sizeof(struct page);
1281 	unsigned long remaining_size = size - vmemmap_size;
1282 
1283 	/*
1284 	 * Besides having arch support and the feature enabled at runtime, we
1285 	 * need a few more assumptions to hold true:
1286 	 *
1287 	 * a) We span a single memory block: memory onlining/offlinin;g happens
1288 	 *    in memory block granularity. We don't want the vmemmap of online
1289 	 *    memory blocks to reside on offline memory blocks. In the future,
1290 	 *    we might want to support variable-sized memory blocks to make the
1291 	 *    feature more versatile.
1292 	 *
1293 	 * b) The vmemmap pages span complete PMDs: We don't want vmemmap code
1294 	 *    to populate memory from the altmap for unrelated parts (i.e.,
1295 	 *    other memory blocks)
1296 	 *
1297 	 * c) The vmemmap pages (and thereby the pages that will be exposed to
1298 	 *    the buddy) have to cover full pageblocks: memory onlining/offlining
1299 	 *    code requires applicable ranges to be page-aligned, for example, to
1300 	 *    set the migratetypes properly.
1301 	 *
1302 	 * TODO: Although we have a check here to make sure that vmemmap pages
1303 	 *       fully populate a PMD, it is not the right place to check for
1304 	 *       this. A much better solution involves improving vmemmap code
1305 	 *       to fallback to base pages when trying to populate vmemmap using
1306 	 *       altmap as an alternative source of memory, and we do not exactly
1307 	 *       populate a single PMD.
1308 	 */
1309 	return mhp_memmap_on_memory() &&
1310 	       size == memory_block_size_bytes() &&
1311 	       IS_ALIGNED(vmemmap_size, PMD_SIZE) &&
1312 	       IS_ALIGNED(remaining_size, (pageblock_nr_pages << PAGE_SHIFT));
1313 }
1314 
1315 /*
1316  * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1317  * and online/offline operations (triggered e.g. by sysfs).
1318  *
1319  * we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG
1320  */
1321 int __ref add_memory_resource(int nid, struct resource *res, mhp_t mhp_flags)
1322 {
1323 	struct mhp_params params = { .pgprot = pgprot_mhp(PAGE_KERNEL) };
1324 	enum memblock_flags memblock_flags = MEMBLOCK_NONE;
1325 	struct vmem_altmap mhp_altmap = {};
1326 	struct memory_group *group = NULL;
1327 	u64 start, size;
1328 	bool new_node = false;
1329 	int ret;
1330 
1331 	start = res->start;
1332 	size = resource_size(res);
1333 
1334 	ret = check_hotplug_memory_range(start, size);
1335 	if (ret)
1336 		return ret;
1337 
1338 	if (mhp_flags & MHP_NID_IS_MGID) {
1339 		group = memory_group_find_by_id(nid);
1340 		if (!group)
1341 			return -EINVAL;
1342 		nid = group->nid;
1343 	}
1344 
1345 	if (!node_possible(nid)) {
1346 		WARN(1, "node %d was absent from the node_possible_map\n", nid);
1347 		return -EINVAL;
1348 	}
1349 
1350 	mem_hotplug_begin();
1351 
1352 	if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK)) {
1353 		if (res->flags & IORESOURCE_SYSRAM_DRIVER_MANAGED)
1354 			memblock_flags = MEMBLOCK_DRIVER_MANAGED;
1355 		ret = memblock_add_node(start, size, nid, memblock_flags);
1356 		if (ret)
1357 			goto error_mem_hotplug_end;
1358 	}
1359 
1360 	ret = __try_online_node(nid, false);
1361 	if (ret < 0)
1362 		goto error;
1363 	new_node = ret;
1364 
1365 	/*
1366 	 * Self hosted memmap array
1367 	 */
1368 	if (mhp_flags & MHP_MEMMAP_ON_MEMORY) {
1369 		if (!mhp_supports_memmap_on_memory(size)) {
1370 			ret = -EINVAL;
1371 			goto error;
1372 		}
1373 		mhp_altmap.free = PHYS_PFN(size);
1374 		mhp_altmap.base_pfn = PHYS_PFN(start);
1375 		params.altmap = &mhp_altmap;
1376 	}
1377 
1378 	/* call arch's memory hotadd */
1379 	ret = arch_add_memory(nid, start, size, &params);
1380 	if (ret < 0)
1381 		goto error;
1382 
1383 	/* create memory block devices after memory was added */
1384 	ret = create_memory_block_devices(start, size, mhp_altmap.alloc,
1385 					  group);
1386 	if (ret) {
1387 		arch_remove_memory(start, size, NULL);
1388 		goto error;
1389 	}
1390 
1391 	if (new_node) {
1392 		/* If sysfs file of new node can't be created, cpu on the node
1393 		 * can't be hot-added. There is no rollback way now.
1394 		 * So, check by BUG_ON() to catch it reluctantly..
1395 		 * We online node here. We can't roll back from here.
1396 		 */
1397 		node_set_online(nid);
1398 		ret = __register_one_node(nid);
1399 		BUG_ON(ret);
1400 	}
1401 
1402 	register_memory_blocks_under_node(nid, PFN_DOWN(start),
1403 					  PFN_UP(start + size - 1),
1404 					  MEMINIT_HOTPLUG);
1405 
1406 	/* create new memmap entry */
1407 	if (!strcmp(res->name, "System RAM"))
1408 		firmware_map_add_hotplug(start, start + size, "System RAM");
1409 
1410 	/* device_online() will take the lock when calling online_pages() */
1411 	mem_hotplug_done();
1412 
1413 	/*
1414 	 * In case we're allowed to merge the resource, flag it and trigger
1415 	 * merging now that adding succeeded.
1416 	 */
1417 	if (mhp_flags & MHP_MERGE_RESOURCE)
1418 		merge_system_ram_resource(res);
1419 
1420 	/* online pages if requested */
1421 	if (mhp_default_online_type != MMOP_OFFLINE)
1422 		walk_memory_blocks(start, size, NULL, online_memory_block);
1423 
1424 	return ret;
1425 error:
1426 	if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK))
1427 		memblock_remove(start, size);
1428 error_mem_hotplug_end:
1429 	mem_hotplug_done();
1430 	return ret;
1431 }
1432 
1433 /* requires device_hotplug_lock, see add_memory_resource() */
1434 int __ref __add_memory(int nid, u64 start, u64 size, mhp_t mhp_flags)
1435 {
1436 	struct resource *res;
1437 	int ret;
1438 
1439 	res = register_memory_resource(start, size, "System RAM");
1440 	if (IS_ERR(res))
1441 		return PTR_ERR(res);
1442 
1443 	ret = add_memory_resource(nid, res, mhp_flags);
1444 	if (ret < 0)
1445 		release_memory_resource(res);
1446 	return ret;
1447 }
1448 
1449 int add_memory(int nid, u64 start, u64 size, mhp_t mhp_flags)
1450 {
1451 	int rc;
1452 
1453 	lock_device_hotplug();
1454 	rc = __add_memory(nid, start, size, mhp_flags);
1455 	unlock_device_hotplug();
1456 
1457 	return rc;
1458 }
1459 EXPORT_SYMBOL_GPL(add_memory);
1460 
1461 /*
1462  * Add special, driver-managed memory to the system as system RAM. Such
1463  * memory is not exposed via the raw firmware-provided memmap as system
1464  * RAM, instead, it is detected and added by a driver - during cold boot,
1465  * after a reboot, and after kexec.
1466  *
1467  * Reasons why this memory should not be used for the initial memmap of a
1468  * kexec kernel or for placing kexec images:
1469  * - The booting kernel is in charge of determining how this memory will be
1470  *   used (e.g., use persistent memory as system RAM)
1471  * - Coordination with a hypervisor is required before this memory
1472  *   can be used (e.g., inaccessible parts).
1473  *
1474  * For this memory, no entries in /sys/firmware/memmap ("raw firmware-provided
1475  * memory map") are created. Also, the created memory resource is flagged
1476  * with IORESOURCE_SYSRAM_DRIVER_MANAGED, so in-kernel users can special-case
1477  * this memory as well (esp., not place kexec images onto it).
1478  *
1479  * The resource_name (visible via /proc/iomem) has to have the format
1480  * "System RAM ($DRIVER)".
1481  */
1482 int add_memory_driver_managed(int nid, u64 start, u64 size,
1483 			      const char *resource_name, mhp_t mhp_flags)
1484 {
1485 	struct resource *res;
1486 	int rc;
1487 
1488 	if (!resource_name ||
1489 	    strstr(resource_name, "System RAM (") != resource_name ||
1490 	    resource_name[strlen(resource_name) - 1] != ')')
1491 		return -EINVAL;
1492 
1493 	lock_device_hotplug();
1494 
1495 	res = register_memory_resource(start, size, resource_name);
1496 	if (IS_ERR(res)) {
1497 		rc = PTR_ERR(res);
1498 		goto out_unlock;
1499 	}
1500 
1501 	rc = add_memory_resource(nid, res, mhp_flags);
1502 	if (rc < 0)
1503 		release_memory_resource(res);
1504 
1505 out_unlock:
1506 	unlock_device_hotplug();
1507 	return rc;
1508 }
1509 EXPORT_SYMBOL_GPL(add_memory_driver_managed);
1510 
1511 /*
1512  * Platforms should define arch_get_mappable_range() that provides
1513  * maximum possible addressable physical memory range for which the
1514  * linear mapping could be created. The platform returned address
1515  * range must adhere to these following semantics.
1516  *
1517  * - range.start <= range.end
1518  * - Range includes both end points [range.start..range.end]
1519  *
1520  * There is also a fallback definition provided here, allowing the
1521  * entire possible physical address range in case any platform does
1522  * not define arch_get_mappable_range().
1523  */
1524 struct range __weak arch_get_mappable_range(void)
1525 {
1526 	struct range mhp_range = {
1527 		.start = 0UL,
1528 		.end = -1ULL,
1529 	};
1530 	return mhp_range;
1531 }
1532 
1533 struct range mhp_get_pluggable_range(bool need_mapping)
1534 {
1535 	const u64 max_phys = (1ULL << MAX_PHYSMEM_BITS) - 1;
1536 	struct range mhp_range;
1537 
1538 	if (need_mapping) {
1539 		mhp_range = arch_get_mappable_range();
1540 		if (mhp_range.start > max_phys) {
1541 			mhp_range.start = 0;
1542 			mhp_range.end = 0;
1543 		}
1544 		mhp_range.end = min_t(u64, mhp_range.end, max_phys);
1545 	} else {
1546 		mhp_range.start = 0;
1547 		mhp_range.end = max_phys;
1548 	}
1549 	return mhp_range;
1550 }
1551 EXPORT_SYMBOL_GPL(mhp_get_pluggable_range);
1552 
1553 bool mhp_range_allowed(u64 start, u64 size, bool need_mapping)
1554 {
1555 	struct range mhp_range = mhp_get_pluggable_range(need_mapping);
1556 	u64 end = start + size;
1557 
1558 	if (start < end && start >= mhp_range.start && (end - 1) <= mhp_range.end)
1559 		return true;
1560 
1561 	pr_warn("Hotplug memory [%#llx-%#llx] exceeds maximum addressable range [%#llx-%#llx]\n",
1562 		start, end, mhp_range.start, mhp_range.end);
1563 	return false;
1564 }
1565 
1566 #ifdef CONFIG_MEMORY_HOTREMOVE
1567 /*
1568  * Scan pfn range [start,end) to find movable/migratable pages (LRU pages,
1569  * non-lru movable pages and hugepages). Will skip over most unmovable
1570  * pages (esp., pages that can be skipped when offlining), but bail out on
1571  * definitely unmovable pages.
1572  *
1573  * Returns:
1574  *	0 in case a movable page is found and movable_pfn was updated.
1575  *	-ENOENT in case no movable page was found.
1576  *	-EBUSY in case a definitely unmovable page was found.
1577  */
1578 static int scan_movable_pages(unsigned long start, unsigned long end,
1579 			      unsigned long *movable_pfn)
1580 {
1581 	unsigned long pfn;
1582 
1583 	for (pfn = start; pfn < end; pfn++) {
1584 		struct page *page, *head;
1585 		unsigned long skip;
1586 
1587 		if (!pfn_valid(pfn))
1588 			continue;
1589 		page = pfn_to_page(pfn);
1590 		if (PageLRU(page))
1591 			goto found;
1592 		if (__PageMovable(page))
1593 			goto found;
1594 
1595 		/*
1596 		 * PageOffline() pages that are not marked __PageMovable() and
1597 		 * have a reference count > 0 (after MEM_GOING_OFFLINE) are
1598 		 * definitely unmovable. If their reference count would be 0,
1599 		 * they could at least be skipped when offlining memory.
1600 		 */
1601 		if (PageOffline(page) && page_count(page))
1602 			return -EBUSY;
1603 
1604 		if (!PageHuge(page))
1605 			continue;
1606 		head = compound_head(page);
1607 		/*
1608 		 * This test is racy as we hold no reference or lock.  The
1609 		 * hugetlb page could have been free'ed and head is no longer
1610 		 * a hugetlb page before the following check.  In such unlikely
1611 		 * cases false positives and negatives are possible.  Calling
1612 		 * code must deal with these scenarios.
1613 		 */
1614 		if (HPageMigratable(head))
1615 			goto found;
1616 		skip = compound_nr(head) - (page - head);
1617 		pfn += skip - 1;
1618 	}
1619 	return -ENOENT;
1620 found:
1621 	*movable_pfn = pfn;
1622 	return 0;
1623 }
1624 
1625 static int
1626 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1627 {
1628 	unsigned long pfn;
1629 	struct page *page, *head;
1630 	int ret = 0;
1631 	LIST_HEAD(source);
1632 	static DEFINE_RATELIMIT_STATE(migrate_rs, DEFAULT_RATELIMIT_INTERVAL,
1633 				      DEFAULT_RATELIMIT_BURST);
1634 
1635 	for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1636 		struct folio *folio;
1637 
1638 		if (!pfn_valid(pfn))
1639 			continue;
1640 		page = pfn_to_page(pfn);
1641 		folio = page_folio(page);
1642 		head = &folio->page;
1643 
1644 		if (PageHuge(page)) {
1645 			pfn = page_to_pfn(head) + compound_nr(head) - 1;
1646 			isolate_huge_page(head, &source);
1647 			continue;
1648 		} else if (PageTransHuge(page))
1649 			pfn = page_to_pfn(head) + thp_nr_pages(page) - 1;
1650 
1651 		/*
1652 		 * HWPoison pages have elevated reference counts so the migration would
1653 		 * fail on them. It also doesn't make any sense to migrate them in the
1654 		 * first place. Still try to unmap such a page in case it is still mapped
1655 		 * (e.g. current hwpoison implementation doesn't unmap KSM pages but keep
1656 		 * the unmap as the catch all safety net).
1657 		 */
1658 		if (PageHWPoison(page)) {
1659 			if (WARN_ON(folio_test_lru(folio)))
1660 				folio_isolate_lru(folio);
1661 			if (folio_mapped(folio))
1662 				try_to_unmap(folio, TTU_IGNORE_MLOCK);
1663 			continue;
1664 		}
1665 
1666 		if (!get_page_unless_zero(page))
1667 			continue;
1668 		/*
1669 		 * We can skip free pages. And we can deal with pages on
1670 		 * LRU and non-lru movable pages.
1671 		 */
1672 		if (PageLRU(page))
1673 			ret = isolate_lru_page(page);
1674 		else
1675 			ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE);
1676 		if (!ret) { /* Success */
1677 			list_add_tail(&page->lru, &source);
1678 			if (!__PageMovable(page))
1679 				inc_node_page_state(page, NR_ISOLATED_ANON +
1680 						    page_is_file_lru(page));
1681 
1682 		} else {
1683 			if (__ratelimit(&migrate_rs)) {
1684 				pr_warn("failed to isolate pfn %lx\n", pfn);
1685 				dump_page(page, "isolation failed");
1686 			}
1687 		}
1688 		put_page(page);
1689 	}
1690 	if (!list_empty(&source)) {
1691 		nodemask_t nmask = node_states[N_MEMORY];
1692 		struct migration_target_control mtc = {
1693 			.nmask = &nmask,
1694 			.gfp_mask = GFP_USER | __GFP_MOVABLE | __GFP_RETRY_MAYFAIL,
1695 		};
1696 
1697 		/*
1698 		 * We have checked that migration range is on a single zone so
1699 		 * we can use the nid of the first page to all the others.
1700 		 */
1701 		mtc.nid = page_to_nid(list_first_entry(&source, struct page, lru));
1702 
1703 		/*
1704 		 * try to allocate from a different node but reuse this node
1705 		 * if there are no other online nodes to be used (e.g. we are
1706 		 * offlining a part of the only existing node)
1707 		 */
1708 		node_clear(mtc.nid, nmask);
1709 		if (nodes_empty(nmask))
1710 			node_set(mtc.nid, nmask);
1711 		ret = migrate_pages(&source, alloc_migration_target, NULL,
1712 			(unsigned long)&mtc, MIGRATE_SYNC, MR_MEMORY_HOTPLUG, NULL);
1713 		if (ret) {
1714 			list_for_each_entry(page, &source, lru) {
1715 				if (__ratelimit(&migrate_rs)) {
1716 					pr_warn("migrating pfn %lx failed ret:%d\n",
1717 						page_to_pfn(page), ret);
1718 					dump_page(page, "migration failure");
1719 				}
1720 			}
1721 			putback_movable_pages(&source);
1722 		}
1723 	}
1724 
1725 	return ret;
1726 }
1727 
1728 static int __init cmdline_parse_movable_node(char *p)
1729 {
1730 	movable_node_enabled = true;
1731 	return 0;
1732 }
1733 early_param("movable_node", cmdline_parse_movable_node);
1734 
1735 /* check which state of node_states will be changed when offline memory */
1736 static void node_states_check_changes_offline(unsigned long nr_pages,
1737 		struct zone *zone, struct memory_notify *arg)
1738 {
1739 	struct pglist_data *pgdat = zone->zone_pgdat;
1740 	unsigned long present_pages = 0;
1741 	enum zone_type zt;
1742 
1743 	arg->status_change_nid = NUMA_NO_NODE;
1744 	arg->status_change_nid_normal = NUMA_NO_NODE;
1745 
1746 	/*
1747 	 * Check whether node_states[N_NORMAL_MEMORY] will be changed.
1748 	 * If the memory to be offline is within the range
1749 	 * [0..ZONE_NORMAL], and it is the last present memory there,
1750 	 * the zones in that range will become empty after the offlining,
1751 	 * thus we can determine that we need to clear the node from
1752 	 * node_states[N_NORMAL_MEMORY].
1753 	 */
1754 	for (zt = 0; zt <= ZONE_NORMAL; zt++)
1755 		present_pages += pgdat->node_zones[zt].present_pages;
1756 	if (zone_idx(zone) <= ZONE_NORMAL && nr_pages >= present_pages)
1757 		arg->status_change_nid_normal = zone_to_nid(zone);
1758 
1759 	/*
1760 	 * We have accounted the pages from [0..ZONE_NORMAL); ZONE_HIGHMEM
1761 	 * does not apply as we don't support 32bit.
1762 	 * Here we count the possible pages from ZONE_MOVABLE.
1763 	 * If after having accounted all the pages, we see that the nr_pages
1764 	 * to be offlined is over or equal to the accounted pages,
1765 	 * we know that the node will become empty, and so, we can clear
1766 	 * it for N_MEMORY as well.
1767 	 */
1768 	present_pages += pgdat->node_zones[ZONE_MOVABLE].present_pages;
1769 
1770 	if (nr_pages >= present_pages)
1771 		arg->status_change_nid = zone_to_nid(zone);
1772 }
1773 
1774 static void node_states_clear_node(int node, struct memory_notify *arg)
1775 {
1776 	if (arg->status_change_nid_normal >= 0)
1777 		node_clear_state(node, N_NORMAL_MEMORY);
1778 
1779 	if (arg->status_change_nid >= 0)
1780 		node_clear_state(node, N_MEMORY);
1781 }
1782 
1783 static int count_system_ram_pages_cb(unsigned long start_pfn,
1784 				     unsigned long nr_pages, void *data)
1785 {
1786 	unsigned long *nr_system_ram_pages = data;
1787 
1788 	*nr_system_ram_pages += nr_pages;
1789 	return 0;
1790 }
1791 
1792 int __ref offline_pages(unsigned long start_pfn, unsigned long nr_pages,
1793 			struct zone *zone, struct memory_group *group)
1794 {
1795 	const unsigned long end_pfn = start_pfn + nr_pages;
1796 	unsigned long pfn, system_ram_pages = 0;
1797 	const int node = zone_to_nid(zone);
1798 	unsigned long flags;
1799 	struct memory_notify arg;
1800 	char *reason;
1801 	int ret;
1802 
1803 	/*
1804 	 * {on,off}lining is constrained to full memory sections (or more
1805 	 * precisely to memory blocks from the user space POV).
1806 	 * memmap_on_memory is an exception because it reserves initial part
1807 	 * of the physical memory space for vmemmaps. That space is pageblock
1808 	 * aligned.
1809 	 */
1810 	if (WARN_ON_ONCE(!nr_pages ||
1811 			 !IS_ALIGNED(start_pfn, pageblock_nr_pages) ||
1812 			 !IS_ALIGNED(start_pfn + nr_pages, PAGES_PER_SECTION)))
1813 		return -EINVAL;
1814 
1815 	mem_hotplug_begin();
1816 
1817 	/*
1818 	 * Don't allow to offline memory blocks that contain holes.
1819 	 * Consequently, memory blocks with holes can never get onlined
1820 	 * via the hotplug path - online_pages() - as hotplugged memory has
1821 	 * no holes. This way, we e.g., don't have to worry about marking
1822 	 * memory holes PG_reserved, don't need pfn_valid() checks, and can
1823 	 * avoid using walk_system_ram_range() later.
1824 	 */
1825 	walk_system_ram_range(start_pfn, nr_pages, &system_ram_pages,
1826 			      count_system_ram_pages_cb);
1827 	if (system_ram_pages != nr_pages) {
1828 		ret = -EINVAL;
1829 		reason = "memory holes";
1830 		goto failed_removal;
1831 	}
1832 
1833 	/*
1834 	 * We only support offlining of memory blocks managed by a single zone,
1835 	 * checked by calling code. This is just a sanity check that we might
1836 	 * want to remove in the future.
1837 	 */
1838 	if (WARN_ON_ONCE(page_zone(pfn_to_page(start_pfn)) != zone ||
1839 			 page_zone(pfn_to_page(end_pfn - 1)) != zone)) {
1840 		ret = -EINVAL;
1841 		reason = "multizone range";
1842 		goto failed_removal;
1843 	}
1844 
1845 	/*
1846 	 * Disable pcplists so that page isolation cannot race with freeing
1847 	 * in a way that pages from isolated pageblock are left on pcplists.
1848 	 */
1849 	zone_pcp_disable(zone);
1850 	lru_cache_disable();
1851 
1852 	/* set above range as isolated */
1853 	ret = start_isolate_page_range(start_pfn, end_pfn,
1854 				       MIGRATE_MOVABLE,
1855 				       MEMORY_OFFLINE | REPORT_FAILURE,
1856 				       GFP_USER | __GFP_MOVABLE | __GFP_RETRY_MAYFAIL);
1857 	if (ret) {
1858 		reason = "failure to isolate range";
1859 		goto failed_removal_pcplists_disabled;
1860 	}
1861 
1862 	arg.start_pfn = start_pfn;
1863 	arg.nr_pages = nr_pages;
1864 	node_states_check_changes_offline(nr_pages, zone, &arg);
1865 
1866 	ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1867 	ret = notifier_to_errno(ret);
1868 	if (ret) {
1869 		reason = "notifier failure";
1870 		goto failed_removal_isolated;
1871 	}
1872 
1873 	do {
1874 		pfn = start_pfn;
1875 		do {
1876 			if (signal_pending(current)) {
1877 				ret = -EINTR;
1878 				reason = "signal backoff";
1879 				goto failed_removal_isolated;
1880 			}
1881 
1882 			cond_resched();
1883 
1884 			ret = scan_movable_pages(pfn, end_pfn, &pfn);
1885 			if (!ret) {
1886 				/*
1887 				 * TODO: fatal migration failures should bail
1888 				 * out
1889 				 */
1890 				do_migrate_range(pfn, end_pfn);
1891 			}
1892 		} while (!ret);
1893 
1894 		if (ret != -ENOENT) {
1895 			reason = "unmovable page";
1896 			goto failed_removal_isolated;
1897 		}
1898 
1899 		/*
1900 		 * Dissolve free hugepages in the memory block before doing
1901 		 * offlining actually in order to make hugetlbfs's object
1902 		 * counting consistent.
1903 		 */
1904 		ret = dissolve_free_huge_pages(start_pfn, end_pfn);
1905 		if (ret) {
1906 			reason = "failure to dissolve huge pages";
1907 			goto failed_removal_isolated;
1908 		}
1909 
1910 		ret = test_pages_isolated(start_pfn, end_pfn, MEMORY_OFFLINE);
1911 
1912 	} while (ret);
1913 
1914 	/* Mark all sections offline and remove free pages from the buddy. */
1915 	__offline_isolated_pages(start_pfn, end_pfn);
1916 	pr_debug("Offlined Pages %ld\n", nr_pages);
1917 
1918 	/*
1919 	 * The memory sections are marked offline, and the pageblock flags
1920 	 * effectively stale; nobody should be touching them. Fixup the number
1921 	 * of isolated pageblocks, memory onlining will properly revert this.
1922 	 */
1923 	spin_lock_irqsave(&zone->lock, flags);
1924 	zone->nr_isolate_pageblock -= nr_pages / pageblock_nr_pages;
1925 	spin_unlock_irqrestore(&zone->lock, flags);
1926 
1927 	lru_cache_enable();
1928 	zone_pcp_enable(zone);
1929 
1930 	/* removal success */
1931 	adjust_managed_page_count(pfn_to_page(start_pfn), -nr_pages);
1932 	adjust_present_page_count(pfn_to_page(start_pfn), group, -nr_pages);
1933 
1934 	/* reinitialise watermarks and update pcp limits */
1935 	init_per_zone_wmark_min();
1936 
1937 	if (!populated_zone(zone)) {
1938 		zone_pcp_reset(zone);
1939 		build_all_zonelists(NULL);
1940 	}
1941 
1942 	node_states_clear_node(node, &arg);
1943 	if (arg.status_change_nid >= 0) {
1944 		kswapd_stop(node);
1945 		kcompactd_stop(node);
1946 	}
1947 
1948 	writeback_set_ratelimit();
1949 
1950 	memory_notify(MEM_OFFLINE, &arg);
1951 	remove_pfn_range_from_zone(zone, start_pfn, nr_pages);
1952 	mem_hotplug_done();
1953 	return 0;
1954 
1955 failed_removal_isolated:
1956 	/* pushback to free area */
1957 	undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1958 	memory_notify(MEM_CANCEL_OFFLINE, &arg);
1959 failed_removal_pcplists_disabled:
1960 	lru_cache_enable();
1961 	zone_pcp_enable(zone);
1962 failed_removal:
1963 	pr_debug("memory offlining [mem %#010llx-%#010llx] failed due to %s\n",
1964 		 (unsigned long long) start_pfn << PAGE_SHIFT,
1965 		 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1,
1966 		 reason);
1967 	mem_hotplug_done();
1968 	return ret;
1969 }
1970 
1971 static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
1972 {
1973 	int ret = !is_memblock_offlined(mem);
1974 	int *nid = arg;
1975 
1976 	*nid = mem->nid;
1977 	if (unlikely(ret)) {
1978 		phys_addr_t beginpa, endpa;
1979 
1980 		beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
1981 		endpa = beginpa + memory_block_size_bytes() - 1;
1982 		pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n",
1983 			&beginpa, &endpa);
1984 
1985 		return -EBUSY;
1986 	}
1987 	return 0;
1988 }
1989 
1990 static int get_nr_vmemmap_pages_cb(struct memory_block *mem, void *arg)
1991 {
1992 	/*
1993 	 * If not set, continue with the next block.
1994 	 */
1995 	return mem->nr_vmemmap_pages;
1996 }
1997 
1998 static int check_cpu_on_node(int nid)
1999 {
2000 	int cpu;
2001 
2002 	for_each_present_cpu(cpu) {
2003 		if (cpu_to_node(cpu) == nid)
2004 			/*
2005 			 * the cpu on this node isn't removed, and we can't
2006 			 * offline this node.
2007 			 */
2008 			return -EBUSY;
2009 	}
2010 
2011 	return 0;
2012 }
2013 
2014 static int check_no_memblock_for_node_cb(struct memory_block *mem, void *arg)
2015 {
2016 	int nid = *(int *)arg;
2017 
2018 	/*
2019 	 * If a memory block belongs to multiple nodes, the stored nid is not
2020 	 * reliable. However, such blocks are always online (e.g., cannot get
2021 	 * offlined) and, therefore, are still spanned by the node.
2022 	 */
2023 	return mem->nid == nid ? -EEXIST : 0;
2024 }
2025 
2026 /**
2027  * try_offline_node
2028  * @nid: the node ID
2029  *
2030  * Offline a node if all memory sections and cpus of the node are removed.
2031  *
2032  * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
2033  * and online/offline operations before this call.
2034  */
2035 void try_offline_node(int nid)
2036 {
2037 	int rc;
2038 
2039 	/*
2040 	 * If the node still spans pages (especially ZONE_DEVICE), don't
2041 	 * offline it. A node spans memory after move_pfn_range_to_zone(),
2042 	 * e.g., after the memory block was onlined.
2043 	 */
2044 	if (node_spanned_pages(nid))
2045 		return;
2046 
2047 	/*
2048 	 * Especially offline memory blocks might not be spanned by the
2049 	 * node. They will get spanned by the node once they get onlined.
2050 	 * However, they link to the node in sysfs and can get onlined later.
2051 	 */
2052 	rc = for_each_memory_block(&nid, check_no_memblock_for_node_cb);
2053 	if (rc)
2054 		return;
2055 
2056 	if (check_cpu_on_node(nid))
2057 		return;
2058 
2059 	/*
2060 	 * all memory/cpu of this node are removed, we can offline this
2061 	 * node now.
2062 	 */
2063 	node_set_offline(nid);
2064 	unregister_one_node(nid);
2065 }
2066 EXPORT_SYMBOL(try_offline_node);
2067 
2068 static int __ref try_remove_memory(u64 start, u64 size)
2069 {
2070 	struct vmem_altmap mhp_altmap = {};
2071 	struct vmem_altmap *altmap = NULL;
2072 	unsigned long nr_vmemmap_pages;
2073 	int rc = 0, nid = NUMA_NO_NODE;
2074 
2075 	BUG_ON(check_hotplug_memory_range(start, size));
2076 
2077 	/*
2078 	 * All memory blocks must be offlined before removing memory.  Check
2079 	 * whether all memory blocks in question are offline and return error
2080 	 * if this is not the case.
2081 	 *
2082 	 * While at it, determine the nid. Note that if we'd have mixed nodes,
2083 	 * we'd only try to offline the last determined one -- which is good
2084 	 * enough for the cases we care about.
2085 	 */
2086 	rc = walk_memory_blocks(start, size, &nid, check_memblock_offlined_cb);
2087 	if (rc)
2088 		return rc;
2089 
2090 	/*
2091 	 * We only support removing memory added with MHP_MEMMAP_ON_MEMORY in
2092 	 * the same granularity it was added - a single memory block.
2093 	 */
2094 	if (mhp_memmap_on_memory()) {
2095 		nr_vmemmap_pages = walk_memory_blocks(start, size, NULL,
2096 						      get_nr_vmemmap_pages_cb);
2097 		if (nr_vmemmap_pages) {
2098 			if (size != memory_block_size_bytes()) {
2099 				pr_warn("Refuse to remove %#llx - %#llx,"
2100 					"wrong granularity\n",
2101 					start, start + size);
2102 				return -EINVAL;
2103 			}
2104 
2105 			/*
2106 			 * Let remove_pmd_table->free_hugepage_table do the
2107 			 * right thing if we used vmem_altmap when hot-adding
2108 			 * the range.
2109 			 */
2110 			mhp_altmap.alloc = nr_vmemmap_pages;
2111 			altmap = &mhp_altmap;
2112 		}
2113 	}
2114 
2115 	/* remove memmap entry */
2116 	firmware_map_remove(start, start + size, "System RAM");
2117 
2118 	/*
2119 	 * Memory block device removal under the device_hotplug_lock is
2120 	 * a barrier against racing online attempts.
2121 	 */
2122 	remove_memory_block_devices(start, size);
2123 
2124 	mem_hotplug_begin();
2125 
2126 	arch_remove_memory(start, size, altmap);
2127 
2128 	if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK)) {
2129 		memblock_phys_free(start, size);
2130 		memblock_remove(start, size);
2131 	}
2132 
2133 	release_mem_region_adjustable(start, size);
2134 
2135 	if (nid != NUMA_NO_NODE)
2136 		try_offline_node(nid);
2137 
2138 	mem_hotplug_done();
2139 	return 0;
2140 }
2141 
2142 /**
2143  * __remove_memory - Remove memory if every memory block is offline
2144  * @start: physical address of the region to remove
2145  * @size: size of the region to remove
2146  *
2147  * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
2148  * and online/offline operations before this call, as required by
2149  * try_offline_node().
2150  */
2151 void __remove_memory(u64 start, u64 size)
2152 {
2153 
2154 	/*
2155 	 * trigger BUG() if some memory is not offlined prior to calling this
2156 	 * function
2157 	 */
2158 	if (try_remove_memory(start, size))
2159 		BUG();
2160 }
2161 
2162 /*
2163  * Remove memory if every memory block is offline, otherwise return -EBUSY is
2164  * some memory is not offline
2165  */
2166 int remove_memory(u64 start, u64 size)
2167 {
2168 	int rc;
2169 
2170 	lock_device_hotplug();
2171 	rc = try_remove_memory(start, size);
2172 	unlock_device_hotplug();
2173 
2174 	return rc;
2175 }
2176 EXPORT_SYMBOL_GPL(remove_memory);
2177 
2178 static int try_offline_memory_block(struct memory_block *mem, void *arg)
2179 {
2180 	uint8_t online_type = MMOP_ONLINE_KERNEL;
2181 	uint8_t **online_types = arg;
2182 	struct page *page;
2183 	int rc;
2184 
2185 	/*
2186 	 * Sense the online_type via the zone of the memory block. Offlining
2187 	 * with multiple zones within one memory block will be rejected
2188 	 * by offlining code ... so we don't care about that.
2189 	 */
2190 	page = pfn_to_online_page(section_nr_to_pfn(mem->start_section_nr));
2191 	if (page && zone_idx(page_zone(page)) == ZONE_MOVABLE)
2192 		online_type = MMOP_ONLINE_MOVABLE;
2193 
2194 	rc = device_offline(&mem->dev);
2195 	/*
2196 	 * Default is MMOP_OFFLINE - change it only if offlining succeeded,
2197 	 * so try_reonline_memory_block() can do the right thing.
2198 	 */
2199 	if (!rc)
2200 		**online_types = online_type;
2201 
2202 	(*online_types)++;
2203 	/* Ignore if already offline. */
2204 	return rc < 0 ? rc : 0;
2205 }
2206 
2207 static int try_reonline_memory_block(struct memory_block *mem, void *arg)
2208 {
2209 	uint8_t **online_types = arg;
2210 	int rc;
2211 
2212 	if (**online_types != MMOP_OFFLINE) {
2213 		mem->online_type = **online_types;
2214 		rc = device_online(&mem->dev);
2215 		if (rc < 0)
2216 			pr_warn("%s: Failed to re-online memory: %d",
2217 				__func__, rc);
2218 	}
2219 
2220 	/* Continue processing all remaining memory blocks. */
2221 	(*online_types)++;
2222 	return 0;
2223 }
2224 
2225 /*
2226  * Try to offline and remove memory. Might take a long time to finish in case
2227  * memory is still in use. Primarily useful for memory devices that logically
2228  * unplugged all memory (so it's no longer in use) and want to offline + remove
2229  * that memory.
2230  */
2231 int offline_and_remove_memory(u64 start, u64 size)
2232 {
2233 	const unsigned long mb_count = size / memory_block_size_bytes();
2234 	uint8_t *online_types, *tmp;
2235 	int rc;
2236 
2237 	if (!IS_ALIGNED(start, memory_block_size_bytes()) ||
2238 	    !IS_ALIGNED(size, memory_block_size_bytes()) || !size)
2239 		return -EINVAL;
2240 
2241 	/*
2242 	 * We'll remember the old online type of each memory block, so we can
2243 	 * try to revert whatever we did when offlining one memory block fails
2244 	 * after offlining some others succeeded.
2245 	 */
2246 	online_types = kmalloc_array(mb_count, sizeof(*online_types),
2247 				     GFP_KERNEL);
2248 	if (!online_types)
2249 		return -ENOMEM;
2250 	/*
2251 	 * Initialize all states to MMOP_OFFLINE, so when we abort processing in
2252 	 * try_offline_memory_block(), we'll skip all unprocessed blocks in
2253 	 * try_reonline_memory_block().
2254 	 */
2255 	memset(online_types, MMOP_OFFLINE, mb_count);
2256 
2257 	lock_device_hotplug();
2258 
2259 	tmp = online_types;
2260 	rc = walk_memory_blocks(start, size, &tmp, try_offline_memory_block);
2261 
2262 	/*
2263 	 * In case we succeeded to offline all memory, remove it.
2264 	 * This cannot fail as it cannot get onlined in the meantime.
2265 	 */
2266 	if (!rc) {
2267 		rc = try_remove_memory(start, size);
2268 		if (rc)
2269 			pr_err("%s: Failed to remove memory: %d", __func__, rc);
2270 	}
2271 
2272 	/*
2273 	 * Rollback what we did. While memory onlining might theoretically fail
2274 	 * (nacked by a notifier), it barely ever happens.
2275 	 */
2276 	if (rc) {
2277 		tmp = online_types;
2278 		walk_memory_blocks(start, size, &tmp,
2279 				   try_reonline_memory_block);
2280 	}
2281 	unlock_device_hotplug();
2282 
2283 	kfree(online_types);
2284 	return rc;
2285 }
2286 EXPORT_SYMBOL_GPL(offline_and_remove_memory);
2287 #endif /* CONFIG_MEMORY_HOTREMOVE */
2288