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