Linux Audio

Check our new training course

Loading...
v3.1
 
   1/*
   2 *  linux/mm/memory_hotplug.c
   3 *
   4 *  Copyright (C)
   5 */
   6
   7#include <linux/stddef.h>
   8#include <linux/mm.h>
 
   9#include <linux/swap.h>
  10#include <linux/interrupt.h>
  11#include <linux/pagemap.h>
  12#include <linux/bootmem.h>
  13#include <linux/compiler.h>
  14#include <linux/module.h>
  15#include <linux/pagevec.h>
  16#include <linux/writeback.h>
  17#include <linux/slab.h>
  18#include <linux/sysctl.h>
  19#include <linux/cpu.h>
  20#include <linux/memory.h>
 
  21#include <linux/memory_hotplug.h>
  22#include <linux/highmem.h>
  23#include <linux/vmalloc.h>
  24#include <linux/ioport.h>
  25#include <linux/delay.h>
  26#include <linux/migrate.h>
  27#include <linux/page-isolation.h>
  28#include <linux/pfn.h>
  29#include <linux/suspend.h>
  30#include <linux/mm_inline.h>
  31#include <linux/firmware-map.h>
 
 
 
 
 
 
  32
  33#include <asm/tlbflush.h>
  34
  35#include "internal.h"
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  36
  37/*
  38 * online_page_callback contains pointer to current page onlining function.
  39 * Initially it is generic_online_page(). If it is required it could be
  40 * changed by calling set_online_page_callback() for callback registration
  41 * and restore_online_page_callback() for generic callback restore.
  42 */
  43
  44static void generic_online_page(struct page *page);
  45
  46static online_page_callback_t online_page_callback = generic_online_page;
 
  47
  48DEFINE_MUTEX(mem_hotplug_mutex);
  49
  50void lock_memory_hotplug(void)
  51{
  52	mutex_lock(&mem_hotplug_mutex);
 
  53
  54	/* for exclusive hibernation if CONFIG_HIBERNATION=y */
  55	lock_system_sleep();
 
  56}
  57
  58void unlock_memory_hotplug(void)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  59{
  60	unlock_system_sleep();
  61	mutex_unlock(&mem_hotplug_mutex);
  62}
  63
 
  64
  65/* add this memory to iomem resource */
  66static struct resource *register_memory_resource(u64 start, u64 size)
 
  67{
  68	struct resource *res;
  69	res = kzalloc(sizeof(struct resource), GFP_KERNEL);
  70	BUG_ON(!res);
 
 
 
 
 
  71
  72	res->name = "System RAM";
  73	res->start = start;
  74	res->end = start + size - 1;
  75	res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
  76	if (request_resource(&iomem_resource, res) < 0) {
  77		printk("System RAM resource %llx - %llx cannot be added\n",
  78		(unsigned long long)res->start, (unsigned long long)res->end);
  79		kfree(res);
  80		res = NULL;
 
 
 
 
 
 
 
 
 
 
 
 
  81	}
  82	return res;
  83}
  84
  85static void release_memory_resource(struct resource *res)
  86{
  87	if (!res)
  88		return;
  89	release_resource(res);
  90	kfree(res);
  91	return;
  92}
  93
  94#ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
  95#ifndef CONFIG_SPARSEMEM_VMEMMAP
  96static void get_page_bootmem(unsigned long info,  struct page *page,
  97			     unsigned long type)
  98{
  99	page->lru.next = (struct list_head *) type;
 100	SetPagePrivate(page);
 101	set_page_private(page, info);
 102	atomic_inc(&page->_count);
 103}
 104
 105/* reference to __meminit __free_pages_bootmem is valid
 106 * so use __ref to tell modpost not to generate a warning */
 107void __ref put_page_bootmem(struct page *page)
 108{
 109	unsigned long type;
 110
 111	type = (unsigned long) page->lru.next;
 112	BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE ||
 113	       type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE);
 114
 115	if (atomic_dec_return(&page->_count) == 1) {
 116		ClearPagePrivate(page);
 117		set_page_private(page, 0);
 118		INIT_LIST_HEAD(&page->lru);
 119		__free_pages_bootmem(page, 0);
 120	}
 121
 
 
 
 
 
 
 
 122}
 123
 124static void register_page_bootmem_info_section(unsigned long start_pfn)
 
 
 
 
 
 125{
 126	unsigned long *usemap, mapsize, section_nr, i;
 
 127	struct mem_section *ms;
 128	struct page *page, *memmap;
 129
 130	if (!pfn_valid(start_pfn))
 131		return;
 132
 133	section_nr = pfn_to_section_nr(start_pfn);
 134	ms = __nr_to_section(section_nr);
 135
 136	/* Get section's memmap address */
 137	memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
 
 138
 139	/*
 140	 * Get page for the memmap's phys address
 141	 * XXX: need more consideration for sparse_vmemmap...
 142	 */
 143	page = virt_to_page(memmap);
 144	mapsize = sizeof(struct page) * PAGES_PER_SECTION;
 145	mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
 146
 147	/* remember memmap's page */
 148	for (i = 0; i < mapsize; i++, page++)
 149		get_page_bootmem(section_nr, page, SECTION_INFO);
 150
 151	usemap = __nr_to_section(section_nr)->pageblock_flags;
 152	page = virt_to_page(usemap);
 153
 154	mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
 
 
 
 
 
 
 
 155
 156	for (i = 0; i < mapsize; i++, page++)
 157		get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
 
 158
 
 159}
 
 160
 161void register_page_bootmem_info_node(struct pglist_data *pgdat)
 
 162{
 163	unsigned long i, pfn, end_pfn, nr_pages;
 164	int node = pgdat->node_id;
 165	struct page *page;
 166	struct zone *zone;
 167
 168	nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
 169	page = virt_to_page(pgdat);
 170
 171	for (i = 0; i < nr_pages; i++, page++)
 172		get_page_bootmem(node, page, NODE_INFO);
 173
 174	zone = &pgdat->node_zones[0];
 175	for (; zone < pgdat->node_zones + MAX_NR_ZONES - 1; zone++) {
 176		if (zone->wait_table) {
 177			nr_pages = zone->wait_table_hash_nr_entries
 178				* sizeof(wait_queue_head_t);
 179			nr_pages = PAGE_ALIGN(nr_pages) >> PAGE_SHIFT;
 180			page = virt_to_page(zone->wait_table);
 181
 182			for (i = 0; i < nr_pages; i++, page++)
 183				get_page_bootmem(node, page, NODE_INFO);
 
 
 
 
 
 
 184		}
 
 185	}
 186
 187	pfn = pgdat->node_start_pfn;
 188	end_pfn = pfn + pgdat->node_spanned_pages;
 189
 190	/* register_section info */
 191	for (; pfn < end_pfn; pfn += PAGES_PER_SECTION)
 192		register_page_bootmem_info_section(pfn);
 193
 
 
 
 
 
 
 
 
 
 
 
 
 194}
 195#endif /* !CONFIG_SPARSEMEM_VMEMMAP */
 196
 197static void grow_zone_span(struct zone *zone, unsigned long start_pfn,
 198			   unsigned long end_pfn)
 
 
 199{
 200	unsigned long old_zone_end_pfn;
 
 
 201
 202	zone_span_writelock(zone);
 
 203
 204	old_zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages;
 205	if (start_pfn < zone->zone_start_pfn)
 206		zone->zone_start_pfn = start_pfn;
 207
 208	zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
 209				zone->zone_start_pfn;
 210
 211	zone_span_writeunlock(zone);
 212}
 213
 214static void grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn,
 215			    unsigned long end_pfn)
 
 
 216{
 217	unsigned long old_pgdat_end_pfn =
 218		pgdat->node_start_pfn + pgdat->node_spanned_pages;
 219
 220	if (start_pfn < pgdat->node_start_pfn)
 221		pgdat->node_start_pfn = start_pfn;
 222
 223	pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
 224					pgdat->node_start_pfn;
 225}
 
 
 226
 227static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn)
 228{
 229	struct pglist_data *pgdat = zone->zone_pgdat;
 230	int nr_pages = PAGES_PER_SECTION;
 231	int nid = pgdat->node_id;
 232	int zone_type;
 233	unsigned long flags;
 234
 235	zone_type = zone - pgdat->node_zones;
 236	if (!zone->wait_table) {
 237		int ret;
 238
 239		ret = init_currently_empty_zone(zone, phys_start_pfn,
 240						nr_pages, MEMMAP_HOTPLUG);
 241		if (ret)
 242			return ret;
 243	}
 244	pgdat_resize_lock(zone->zone_pgdat, &flags);
 245	grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages);
 246	grow_pgdat_span(zone->zone_pgdat, phys_start_pfn,
 247			phys_start_pfn + nr_pages);
 248	pgdat_resize_unlock(zone->zone_pgdat, &flags);
 249	memmap_init_zone(nr_pages, nid, zone_type,
 250			 phys_start_pfn, MEMMAP_HOTPLUG);
 251	return 0;
 252}
 253
 254static int __meminit __add_section(int nid, struct zone *zone,
 255					unsigned long phys_start_pfn)
 256{
 257	int nr_pages = PAGES_PER_SECTION;
 258	int ret;
 259
 260	if (pfn_valid(phys_start_pfn))
 261		return -EEXIST;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 262
 263	ret = sparse_add_one_section(zone, phys_start_pfn, nr_pages);
 
 
 
 264
 265	if (ret < 0)
 266		return ret;
 
 267
 268	ret = __add_zone(zone, phys_start_pfn);
 
 
 
 
 
 
 
 269
 270	if (ret < 0)
 271		return ret;
 
 
 
 272
 273	return register_new_memory(nid, __pfn_to_section(phys_start_pfn));
 
 274}
 275
 276#ifdef CONFIG_SPARSEMEM_VMEMMAP
 277static int __remove_section(struct zone *zone, struct mem_section *ms)
 
 278{
 279	/*
 280	 * XXX: Freeing memmap with vmemmap is not implement yet.
 281	 *      This should be removed later.
 282	 */
 283	return -EBUSY;
 284}
 285#else
 286static int __remove_section(struct zone *zone, struct mem_section *ms)
 287{
 288	unsigned long flags;
 289	struct pglist_data *pgdat = zone->zone_pgdat;
 290	int ret = -EINVAL;
 291
 292	if (!valid_section(ms))
 293		return ret;
 
 294
 295	ret = unregister_memory_section(ms);
 296	if (ret)
 297		return ret;
 
 
 
 298
 299	pgdat_resize_lock(pgdat, &flags);
 300	sparse_remove_one_section(zone, ms);
 301	pgdat_resize_unlock(pgdat, &flags);
 302	return 0;
 303}
 304#endif
 
 305
 306/*
 307 * Reasonably generic function for adding memory.  It is
 308 * expected that archs that support memory hotplug will
 309 * call this function after deciding the zone to which to
 310 * add the new pages.
 311 */
 312int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn,
 313			unsigned long nr_pages)
 314{
 315	unsigned long i;
 316	int err = 0;
 317	int start_sec, end_sec;
 318	/* during initialize mem_map, align hot-added range to section */
 319	start_sec = pfn_to_section_nr(phys_start_pfn);
 320	end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
 321
 322	for (i = start_sec; i <= end_sec; i++) {
 323		err = __add_section(nid, zone, i << PFN_SECTION_SHIFT);
 324
 325		/*
 326		 * EEXIST is finally dealt with by ioresource collision
 327		 * check. see add_memory() => register_memory_resource()
 328		 * Warning will be printed if there is collision.
 329		 */
 330		if (err && (err != -EEXIST))
 331			break;
 332		err = 0;
 333	}
 334
 335	return err;
 
 
 
 336}
 337EXPORT_SYMBOL_GPL(__add_pages);
 338
 339/**
 340 * __remove_pages() - remove sections of pages from a zone
 341 * @zone: zone from which pages need to be removed
 342 * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
 343 * @nr_pages: number of pages to remove (must be multiple of section size)
 
 344 *
 345 * Generic helper function to remove section mappings and sysfs entries
 346 * for the section of the memory we are removing. Caller needs to make
 347 * sure that pages are marked reserved and zones are adjust properly by
 348 * calling offline_pages().
 349 */
 350int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
 351		 unsigned long nr_pages)
 352{
 353	unsigned long i, ret = 0;
 354	int sections_to_remove;
 355
 356	/*
 357	 * We can only remove entire sections
 358	 */
 359	BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
 360	BUG_ON(nr_pages % PAGES_PER_SECTION);
 361
 362	sections_to_remove = nr_pages / PAGES_PER_SECTION;
 363	for (i = 0; i < sections_to_remove; i++) {
 364		unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
 365		release_mem_region(pfn << PAGE_SHIFT,
 366				   PAGES_PER_SECTION << PAGE_SHIFT);
 367		ret = __remove_section(zone, __pfn_to_section(pfn));
 368		if (ret)
 369			break;
 370	}
 371	return ret;
 372}
 373EXPORT_SYMBOL_GPL(__remove_pages);
 374
 375int set_online_page_callback(online_page_callback_t callback)
 376{
 377	int rc = -EINVAL;
 378
 379	lock_memory_hotplug();
 
 380
 381	if (online_page_callback == generic_online_page) {
 382		online_page_callback = callback;
 383		rc = 0;
 384	}
 385
 386	unlock_memory_hotplug();
 
 387
 388	return rc;
 389}
 390EXPORT_SYMBOL_GPL(set_online_page_callback);
 391
 392int restore_online_page_callback(online_page_callback_t callback)
 393{
 394	int rc = -EINVAL;
 395
 396	lock_memory_hotplug();
 
 397
 398	if (online_page_callback == callback) {
 399		online_page_callback = generic_online_page;
 400		rc = 0;
 401	}
 402
 403	unlock_memory_hotplug();
 
 404
 405	return rc;
 406}
 407EXPORT_SYMBOL_GPL(restore_online_page_callback);
 408
 409void __online_page_set_limits(struct page *page)
 
 
 
 
 
 
 
 
 
 
 
 
 
 410{
 411	unsigned long pfn = page_to_pfn(page);
 
 412
 413	if (pfn >= num_physpages)
 414		num_physpages = pfn + 1;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 415}
 416EXPORT_SYMBOL_GPL(__online_page_set_limits);
 417
 418void __online_page_increment_counters(struct page *page)
 
 
 419{
 420	totalram_pages++;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 421
 422#ifdef CONFIG_HIGHMEM
 423	if (PageHighMem(page))
 424		totalhigh_pages++;
 
 
 
 
 
 
 
 
 
 
 425#endif
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 426}
 427EXPORT_SYMBOL_GPL(__online_page_increment_counters);
 428
 429void __online_page_free(struct page *page)
 
 
 
 
 
 
 430{
 431	ClearPageReserved(page);
 432	init_page_count(page);
 433	__free_page(page);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 434}
 435EXPORT_SYMBOL_GPL(__online_page_free);
 436
 437static void generic_online_page(struct page *page)
 
 438{
 439	__online_page_set_limits(page);
 440	__online_page_increment_counters(page);
 441	__online_page_free(page);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 442}
 443
 444static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
 445			void *arg)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 446{
 447	unsigned long i;
 448	unsigned long onlined_pages = *(unsigned long *)arg;
 449	struct page *page;
 450	if (PageReserved(pfn_to_page(start_pfn)))
 451		for (i = 0; i < nr_pages; i++) {
 452			page = pfn_to_page(start_pfn + i);
 453			(*online_page_callback)(page);
 454			onlined_pages++;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 455		}
 456	*(unsigned long *)arg = onlined_pages;
 457	return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 458}
 459
 
 
 
 
 
 
 460
 461int __ref online_pages(unsigned long pfn, unsigned long nr_pages)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 462{
 463	unsigned long onlined_pages = 0;
 464	struct zone *zone;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 465	int need_zonelists_rebuild = 0;
 466	int nid;
 467	int ret;
 468	struct memory_notify arg;
 469
 470	lock_memory_hotplug();
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 471	arg.start_pfn = pfn;
 472	arg.nr_pages = nr_pages;
 473	arg.status_change_nid = -1;
 474
 475	nid = page_to_nid(pfn_to_page(pfn));
 476	if (node_present_pages(nid) == 0)
 477		arg.status_change_nid = nid;
 478
 479	ret = memory_notify(MEM_GOING_ONLINE, &arg);
 480	ret = notifier_to_errno(ret);
 481	if (ret) {
 482		memory_notify(MEM_CANCEL_ONLINE, &arg);
 483		unlock_memory_hotplug();
 484		return ret;
 485	}
 486	/*
 487	 * This doesn't need a lock to do pfn_to_page().
 488	 * The section can't be removed here because of the
 489	 * memory_block->state_mutex.
 490	 */
 491	zone = page_zone(pfn_to_page(pfn));
 
 
 
 492	/*
 493	 * If this zone is not populated, then it is not in zonelist.
 494	 * This means the page allocator ignores this zone.
 495	 * So, zonelist must be updated after online.
 496	 */
 497	mutex_lock(&zonelists_mutex);
 498	if (!populated_zone(zone))
 499		need_zonelists_rebuild = 1;
 500
 501	ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
 502		online_pages_range);
 503	if (ret) {
 504		mutex_unlock(&zonelists_mutex);
 505		printk(KERN_DEBUG "online_pages %lx at %lx failed\n",
 506			nr_pages, pfn);
 507		memory_notify(MEM_CANCEL_ONLINE, &arg);
 508		unlock_memory_hotplug();
 509		return ret;
 510	}
 511
 512	zone->present_pages += onlined_pages;
 513	zone->zone_pgdat->node_present_pages += onlined_pages;
 
 
 514	if (need_zonelists_rebuild)
 515		build_all_zonelists(zone);
 516	else
 517		zone_pcp_update(zone);
 518
 519	mutex_unlock(&zonelists_mutex);
 
 520
 521	init_per_zone_wmark_min();
 
 
 
 
 
 
 522
 523	if (onlined_pages) {
 524		kswapd_run(zone_to_nid(zone));
 525		node_set_state(zone_to_nid(zone), N_HIGH_MEMORY);
 526	}
 527
 528	vm_total_pages = nr_free_pagecache_pages();
 
 529
 530	writeback_set_ratelimit();
 531
 532	if (onlined_pages)
 533		memory_notify(MEM_ONLINE, &arg);
 534	unlock_memory_hotplug();
 535
 536	return 0;
 
 
 
 
 
 
 
 
 537}
 538#endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
 539
 540/* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
 541static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
 542{
 543	struct pglist_data *pgdat;
 544	unsigned long zones_size[MAX_NR_ZONES] = {0};
 545	unsigned long zholes_size[MAX_NR_ZONES] = {0};
 546	unsigned long start_pfn = start >> PAGE_SHIFT;
 547
 548	pgdat = arch_alloc_nodedata(nid);
 549	if (!pgdat)
 550		return NULL;
 551
 552	arch_refresh_nodedata(nid, pgdat);
 553
 554	/* we can use NODE_DATA(nid) from here */
 
 
 
 
 
 
 555
 556	/* init node's zones as empty zones, we don't have any present pages.*/
 557	free_area_init_node(nid, zones_size, start_pfn, zholes_size);
 558
 559	/*
 560	 * The node we allocated has no zone fallback lists. For avoiding
 561	 * to access not-initialized zonelist, build here.
 562	 */
 563	mutex_lock(&zonelists_mutex);
 564	build_all_zonelists(NULL);
 565	mutex_unlock(&zonelists_mutex);
 566
 567	return pgdat;
 568}
 569
 570static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
 571{
 572	arch_refresh_nodedata(nid, NULL);
 573	arch_free_nodedata(pgdat);
 574	return;
 575}
 576
 577
 578/*
 
 
 
 579 * called by cpu_up() to online a node without onlined memory.
 
 
 
 
 
 580 */
 581int mem_online_node(int nid)
 582{
 583	pg_data_t	*pgdat;
 584	int	ret;
 
 
 
 585
 586	lock_memory_hotplug();
 587	pgdat = hotadd_new_pgdat(nid, 0);
 588	if (!pgdat) {
 
 589		ret = -ENOMEM;
 590		goto out;
 591	}
 592	node_set_online(nid);
 593	ret = register_one_node(nid);
 594	BUG_ON(ret);
 595
 
 
 
 
 
 596out:
 597	unlock_memory_hotplug();
 598	return ret;
 599}
 600
 601/* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
 602int __ref add_memory(int nid, u64 start, u64 size)
 
 
 603{
 604	pg_data_t *pgdat = NULL;
 605	int new_pgdat = 0;
 606	struct resource *res;
 607	int ret;
 608
 609	lock_memory_hotplug();
 
 
 
 
 610
 611	res = register_memory_resource(start, size);
 612	ret = -EEXIST;
 613	if (!res)
 614		goto out;
 
 
 
 
 
 615
 616	if (!node_online(nid)) {
 617		pgdat = hotadd_new_pgdat(nid, start);
 618		ret = -ENOMEM;
 619		if (!pgdat)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 620			goto out;
 621		new_pgdat = 1;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 622	}
 623
 624	/* call arch's memory hotadd */
 625	ret = arch_add_memory(nid, start, size);
 626
 
 
 
 
 
 
 
 
 
 627	if (ret < 0)
 628		goto error;
 
 629
 630	/* we online node here. we can't roll back from here. */
 631	node_set_online(nid);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 632
 633	if (new_pgdat) {
 634		ret = register_one_node(nid);
 635		/*
 636		 * If sysfs file of new node can't create, cpu on the node
 637		 * can't be hot-added. There is no rollback way now.
 638		 * So, check by BUG_ON() to catch it reluctantly..
 
 639		 */
 
 
 640		BUG_ON(ret);
 641	}
 642
 
 
 
 
 643	/* create new memmap entry */
 644	firmware_map_add_hotplug(start, start + size, "System RAM");
 
 645
 646	goto out;
 
 647
 648error:
 649	/* rollback pgdat allocation and others */
 650	if (new_pgdat)
 651		rollback_node_hotadd(nid, pgdat);
 652	if (res)
 653		release_memory_resource(res);
 654
 655out:
 656	unlock_memory_hotplug();
 
 
 
 
 
 
 
 
 657	return ret;
 658}
 659EXPORT_SYMBOL_GPL(add_memory);
 660
 661#ifdef CONFIG_MEMORY_HOTREMOVE
 662/*
 663 * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
 664 * set and the size of the free page is given by page_order(). Using this,
 665 * the function determines if the pageblock contains only free pages.
 666 * Due to buddy contraints, a free page at least the size of a pageblock will
 667 * be located at the start of the pageblock
 668 */
 669static inline int pageblock_free(struct page *page)
 670{
 671	return PageBuddy(page) && page_order(page) >= pageblock_order;
 
 
 
 
 
 
 
 
 
 
 672}
 673
 674/* Return the start of the next active pageblock after a given page */
 675static struct page *next_active_pageblock(struct page *page)
 676{
 677	/* Ensure the starting page is pageblock-aligned */
 678	BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
 679
 680	/* If the entire pageblock is free, move to the end of free page */
 681	if (pageblock_free(page)) {
 682		int order;
 683		/* be careful. we don't have locks, page_order can be changed.*/
 684		order = page_order(page);
 685		if ((order < MAX_ORDER) && (order >= pageblock_order))
 686			return page + (1 << order);
 687	}
 688
 689	return page + pageblock_nr_pages;
 690}
 
 691
 692/* Checks if this range of memory is likely to be hot-removable. */
 693int is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 694{
 695	struct page *page = pfn_to_page(start_pfn);
 696	struct page *end_page = page + nr_pages;
 697
 698	/* Check the starting page of each pageblock within the range */
 699	for (; page < end_page; page = next_active_pageblock(page)) {
 700		if (!is_pageblock_removable_nolock(page))
 701			return 0;
 702		cond_resched();
 
 
 
 
 
 
 703	}
 704
 705	/* All pageblocks in the memory block are likely to be hot-removable */
 706	return 1;
 
 
 
 
 
 707}
 
 708
 709/*
 710 * Confirm all pages in a range [start, end) is belongs to the same zone.
 
 
 
 
 
 
 
 
 
 
 711 */
 712static int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn)
 713{
 714	unsigned long pfn;
 715	struct zone *zone = NULL;
 716	struct page *page;
 717	int i;
 718	for (pfn = start_pfn;
 719	     pfn < end_pfn;
 720	     pfn += MAX_ORDER_NR_PAGES) {
 721		i = 0;
 722		/* This is just a CONFIG_HOLES_IN_ZONE check.*/
 723		while ((i < MAX_ORDER_NR_PAGES) && !pfn_valid_within(pfn + i))
 724			i++;
 725		if (i == MAX_ORDER_NR_PAGES)
 726			continue;
 727		page = pfn_to_page(pfn + i);
 728		if (zone && page_zone(page) != zone)
 729			return 0;
 730		zone = page_zone(page);
 
 
 
 
 
 731	}
 732	return 1;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 733}
 734
 
 735/*
 736 * Scanning pfn is much easier than scanning lru list.
 737 * Scan pfn from start to end and Find LRU page.
 
 
 
 
 
 
 
 738 */
 739static unsigned long scan_lru_pages(unsigned long start, unsigned long end)
 
 740{
 741	unsigned long pfn;
 742	struct page *page;
 743	for (pfn = start; pfn < end; pfn++) {
 744		if (pfn_valid(pfn)) {
 745			page = pfn_to_page(pfn);
 746			if (PageLRU(page))
 747				return pfn;
 748		}
 749	}
 750	return 0;
 751}
 752
 753static struct page *
 754hotremove_migrate_alloc(struct page *page, unsigned long private, int **x)
 755{
 756	/* This should be improooooved!! */
 757	return alloc_page(GFP_HIGHUSER_MOVABLE);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 758}
 759
 760#define NR_OFFLINE_AT_ONCE_PAGES	(256)
 761static int
 762do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
 763{
 764	unsigned long pfn;
 765	struct page *page;
 766	int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
 767	int not_managed = 0;
 768	int ret = 0;
 769	LIST_HEAD(source);
 
 
 
 
 
 
 770
 771	for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
 772		if (!pfn_valid(pfn))
 773			continue;
 774		page = pfn_to_page(pfn);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 775		if (!get_page_unless_zero(page))
 776			continue;
 777		/*
 778		 * We can skip free pages. And we can only deal with pages on
 779		 * LRU.
 780		 */
 781		ret = isolate_lru_page(page);
 782		if (!ret) { /* Success */
 783			put_page(page);
 
 
 784			list_add_tail(&page->lru, &source);
 785			move_pages--;
 786			inc_zone_page_state(page, NR_ISOLATED_ANON +
 787					    page_is_file_cache(page));
 788
 789		} else {
 790#ifdef CONFIG_DEBUG_VM
 791			printk(KERN_ALERT "removing pfn %lx from LRU failed\n",
 792			       pfn);
 793			dump_page(page);
 794#endif
 795			put_page(page);
 796			/* Because we don't have big zone->lock. we should
 797			   check this again here. */
 798			if (page_count(page)) {
 799				not_managed++;
 800				ret = -EBUSY;
 801				break;
 802			}
 803		}
 
 804	}
 805	if (!list_empty(&source)) {
 806		if (not_managed) {
 807			putback_lru_pages(&source);
 808			goto out;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 809		}
 810		/* this function returns # of failed pages */
 811		ret = migrate_pages(&source, hotremove_migrate_alloc, 0,
 812								true, true);
 813		if (ret)
 814			putback_lru_pages(&source);
 815	}
 816out:
 817	return ret;
 818}
 819
 820/*
 821 * remove from free_area[] and mark all as Reserved.
 822 */
 823static int
 824offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
 825			void *data)
 826{
 827	__offline_isolated_pages(start, start + nr_pages);
 828	return 0;
 829}
 
 830
 831static void
 832offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
 
 833{
 834	walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
 835				offline_isolated_pages_cb);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 836}
 837
 838/*
 839 * Check all pages in range, recoreded as memory resource, are isolated.
 840 */
 841static int
 842check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
 843			void *data)
 844{
 845	int ret;
 846	long offlined = *(long *)data;
 847	ret = test_pages_isolated(start_pfn, start_pfn + nr_pages);
 848	offlined = nr_pages;
 849	if (!ret)
 850		*(long *)data += offlined;
 851	return ret;
 852}
 853
 854static long
 855check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
 856{
 857	long offlined = 0;
 858	int ret;
 859
 860	ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
 861			check_pages_isolated_cb);
 862	if (ret < 0)
 863		offlined = (long)ret;
 864	return offlined;
 865}
 866
 867static int __ref offline_pages(unsigned long start_pfn,
 868		  unsigned long end_pfn, unsigned long timeout)
 
 
 
 869{
 870	unsigned long pfn, nr_pages, expire;
 871	long offlined_pages;
 872	int ret, drain, retry_max, node;
 873	struct zone *zone;
 874	struct memory_notify arg;
 
 
 875
 876	BUG_ON(start_pfn >= end_pfn);
 877	/* at least, alignment against pageblock is necessary */
 878	if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
 879		return -EINVAL;
 880	if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
 881		return -EINVAL;
 882	/* This makes hotplug much easier...and readable.
 883	   we assume this for now. .*/
 884	if (!test_pages_in_a_zone(start_pfn, end_pfn))
 885		return -EINVAL;
 886
 887	lock_memory_hotplug();
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 888
 889	zone = page_zone(pfn_to_page(start_pfn));
 890	node = zone_to_nid(zone);
 891	nr_pages = end_pfn - start_pfn;
 
 
 
 892
 893	/* set above range as isolated */
 894	ret = start_isolate_page_range(start_pfn, end_pfn);
 895	if (ret)
 896		goto out;
 
 
 
 
 
 897
 898	arg.start_pfn = start_pfn;
 899	arg.nr_pages = nr_pages;
 900	arg.status_change_nid = -1;
 901	if (nr_pages >= node_present_pages(node))
 902		arg.status_change_nid = node;
 903
 904	ret = memory_notify(MEM_GOING_OFFLINE, &arg);
 905	ret = notifier_to_errno(ret);
 906	if (ret)
 907		goto failed_removal;
 908
 909	pfn = start_pfn;
 910	expire = jiffies + timeout;
 911	drain = 0;
 912	retry_max = 5;
 913repeat:
 914	/* start memory hot removal */
 915	ret = -EAGAIN;
 916	if (time_after(jiffies, expire))
 917		goto failed_removal;
 918	ret = -EINTR;
 919	if (signal_pending(current))
 920		goto failed_removal;
 921	ret = 0;
 922	if (drain) {
 923		lru_add_drain_all();
 924		cond_resched();
 925		drain_all_pages();
 926	}
 927
 928	pfn = scan_lru_pages(start_pfn, end_pfn);
 929	if (pfn) { /* We have page on LRU */
 930		ret = do_migrate_range(pfn, end_pfn);
 931		if (!ret) {
 932			drain = 1;
 933			goto repeat;
 934		} else {
 935			if (ret < 0)
 936				if (--retry_max == 0)
 937					goto failed_removal;
 938			yield();
 939			drain = 1;
 940			goto repeat;
 941		}
 942	}
 943	/* drain all zone's lru pagevec, this is asyncronous... */
 944	lru_add_drain_all();
 945	yield();
 946	/* drain pcp pages , this is synchrouns. */
 947	drain_all_pages();
 948	/* check again */
 949	offlined_pages = check_pages_isolated(start_pfn, end_pfn);
 950	if (offlined_pages < 0) {
 951		ret = -EBUSY;
 952		goto failed_removal;
 953	}
 954	printk(KERN_INFO "Offlined Pages %ld\n", offlined_pages);
 955	/* Ok, all of our target is islaoted.
 956	   We cannot do rollback at this point. */
 957	offline_isolated_pages(start_pfn, end_pfn);
 958	/* reset pagetype flags and makes migrate type to be MOVABLE */
 959	undo_isolate_page_range(start_pfn, end_pfn);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 960	/* removal success */
 961	zone->present_pages -= offlined_pages;
 962	zone->zone_pgdat->node_present_pages -= offlined_pages;
 963	totalram_pages -= offlined_pages;
 964
 
 965	init_per_zone_wmark_min();
 966
 967	if (!node_present_pages(node)) {
 968		node_clear_state(node, N_HIGH_MEMORY);
 
 
 
 
 
 
 
 
 
 
 969		kswapd_stop(node);
 970	}
 971
 972	vm_total_pages = nr_free_pagecache_pages();
 973	writeback_set_ratelimit();
 974
 975	memory_notify(MEM_OFFLINE, &arg);
 976	unlock_memory_hotplug();
 977	return 0;
 978
 979failed_removal:
 980	printk(KERN_INFO "memory offlining %lx to %lx failed\n",
 981		start_pfn, end_pfn);
 982	memory_notify(MEM_CANCEL_OFFLINE, &arg);
 983	/* pushback to free area */
 984	undo_isolate_page_range(start_pfn, end_pfn);
 985
 986out:
 987	unlock_memory_hotplug();
 
 
 
 
 
 
 988	return ret;
 989}
 990
 991int remove_memory(u64 start, u64 size)
 992{
 993	unsigned long start_pfn, end_pfn;
 
 
 
 
 994
 995	start_pfn = PFN_DOWN(start);
 996	end_pfn = start_pfn + PFN_DOWN(size);
 997	return offline_pages(start_pfn, end_pfn, 120 * HZ);
 
 
 
 
 
 998}
 999#else
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1000int remove_memory(u64 start, u64 size)
1001{
1002	return -EINVAL;
 
 
 
 
 
 
1003}
1004#endif /* CONFIG_MEMORY_HOTREMOVE */
1005EXPORT_SYMBOL_GPL(remove_memory);
v6.8
   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/writeback.h>
  17#include <linux/slab.h>
  18#include <linux/sysctl.h>
  19#include <linux/cpu.h>
  20#include <linux/memory.h>
  21#include <linux/memremap.h>
  22#include <linux/memory_hotplug.h>
 
  23#include <linux/vmalloc.h>
  24#include <linux/ioport.h>
  25#include <linux/delay.h>
  26#include <linux/migrate.h>
  27#include <linux/page-isolation.h>
  28#include <linux/pfn.h>
  29#include <linux/suspend.h>
  30#include <linux/mm_inline.h>
  31#include <linux/firmware-map.h>
  32#include <linux/stop_machine.h>
  33#include <linux/hugetlb.h>
  34#include <linux/memblock.h>
  35#include <linux/compaction.h>
  36#include <linux/rmap.h>
  37#include <linux/module.h>
  38
  39#include <asm/tlbflush.h>
  40
  41#include "internal.h"
  42#include "shuffle.h"
  43
  44enum {
  45	MEMMAP_ON_MEMORY_DISABLE = 0,
  46	MEMMAP_ON_MEMORY_ENABLE,
  47	MEMMAP_ON_MEMORY_FORCE,
  48};
  49
  50static int memmap_mode __read_mostly = MEMMAP_ON_MEMORY_DISABLE;
  51
  52static inline unsigned long memory_block_memmap_size(void)
  53{
  54	return PHYS_PFN(memory_block_size_bytes()) * sizeof(struct page);
  55}
  56
  57static inline unsigned long memory_block_memmap_on_memory_pages(void)
  58{
  59	unsigned long nr_pages = PFN_UP(memory_block_memmap_size());
  60
  61	/*
  62	 * In "forced" memmap_on_memory mode, we add extra pages to align the
  63	 * vmemmap size to cover full pageblocks. That way, we can add memory
  64	 * even if the vmemmap size is not properly aligned, however, we might waste
  65	 * memory.
  66	 */
  67	if (memmap_mode == MEMMAP_ON_MEMORY_FORCE)
  68		return pageblock_align(nr_pages);
  69	return nr_pages;
  70}
  71
  72#ifdef CONFIG_MHP_MEMMAP_ON_MEMORY
  73/*
  74 * memory_hotplug.memmap_on_memory parameter
  75 */
  76static int set_memmap_mode(const char *val, const struct kernel_param *kp)
  77{
  78	int ret, mode;
  79	bool enabled;
  80
  81	if (sysfs_streq(val, "force") ||  sysfs_streq(val, "FORCE")) {
  82		mode = MEMMAP_ON_MEMORY_FORCE;
  83	} else {
  84		ret = kstrtobool(val, &enabled);
  85		if (ret < 0)
  86			return ret;
  87		if (enabled)
  88			mode = MEMMAP_ON_MEMORY_ENABLE;
  89		else
  90			mode = MEMMAP_ON_MEMORY_DISABLE;
  91	}
  92	*((int *)kp->arg) = mode;
  93	if (mode == MEMMAP_ON_MEMORY_FORCE) {
  94		unsigned long memmap_pages = memory_block_memmap_on_memory_pages();
  95
  96		pr_info_once("Memory hotplug will waste %ld pages in each memory block\n",
  97			     memmap_pages - PFN_UP(memory_block_memmap_size()));
  98	}
  99	return 0;
 100}
 101
 102static int get_memmap_mode(char *buffer, const struct kernel_param *kp)
 103{
 104	int mode = *((int *)kp->arg);
 105
 106	if (mode == MEMMAP_ON_MEMORY_FORCE)
 107		return sprintf(buffer, "force\n");
 108	return sprintf(buffer, "%c\n", mode ? 'Y' : 'N');
 109}
 110
 111static const struct kernel_param_ops memmap_mode_ops = {
 112	.set = set_memmap_mode,
 113	.get = get_memmap_mode,
 114};
 115module_param_cb(memmap_on_memory, &memmap_mode_ops, &memmap_mode, 0444);
 116MODULE_PARM_DESC(memmap_on_memory, "Enable memmap on memory for memory hotplug\n"
 117		 "With value \"force\" it could result in memory wastage due "
 118		 "to memmap size limitations (Y/N/force)");
 119
 120static inline bool mhp_memmap_on_memory(void)
 121{
 122	return memmap_mode != MEMMAP_ON_MEMORY_DISABLE;
 123}
 124#else
 125static inline bool mhp_memmap_on_memory(void)
 126{
 127	return false;
 128}
 129#endif
 130
 131enum {
 132	ONLINE_POLICY_CONTIG_ZONES = 0,
 133	ONLINE_POLICY_AUTO_MOVABLE,
 134};
 135
 136static const char * const online_policy_to_str[] = {
 137	[ONLINE_POLICY_CONTIG_ZONES] = "contig-zones",
 138	[ONLINE_POLICY_AUTO_MOVABLE] = "auto-movable",
 139};
 140
 141static int set_online_policy(const char *val, const struct kernel_param *kp)
 142{
 143	int ret = sysfs_match_string(online_policy_to_str, val);
 144
 145	if (ret < 0)
 146		return ret;
 147	*((int *)kp->arg) = ret;
 148	return 0;
 149}
 150
 151static int get_online_policy(char *buffer, const struct kernel_param *kp)
 152{
 153	return sprintf(buffer, "%s\n", online_policy_to_str[*((int *)kp->arg)]);
 154}
 155
 156/*
 157 * memory_hotplug.online_policy: configure online behavior when onlining without
 158 * specifying a zone (MMOP_ONLINE)
 159 *
 160 * "contig-zones": keep zone contiguous
 161 * "auto-movable": online memory to ZONE_MOVABLE if the configuration
 162 *                 (auto_movable_ratio, auto_movable_numa_aware) allows for it
 163 */
 164static int online_policy __read_mostly = ONLINE_POLICY_CONTIG_ZONES;
 165static const struct kernel_param_ops online_policy_ops = {
 166	.set = set_online_policy,
 167	.get = get_online_policy,
 168};
 169module_param_cb(online_policy, &online_policy_ops, &online_policy, 0644);
 170MODULE_PARM_DESC(online_policy,
 171		"Set the online policy (\"contig-zones\", \"auto-movable\") "
 172		"Default: \"contig-zones\"");
 173
 174/*
 175 * memory_hotplug.auto_movable_ratio: specify maximum MOVABLE:KERNEL ratio
 176 *
 177 * The ratio represent an upper limit and the kernel might decide to not
 178 * online some memory to ZONE_MOVABLE -- e.g., because hotplugged KERNEL memory
 179 * doesn't allow for more MOVABLE memory.
 180 */
 181static unsigned int auto_movable_ratio __read_mostly = 301;
 182module_param(auto_movable_ratio, uint, 0644);
 183MODULE_PARM_DESC(auto_movable_ratio,
 184		"Set the maximum ratio of MOVABLE:KERNEL memory in the system "
 185		"in percent for \"auto-movable\" online policy. Default: 301");
 186
 187/*
 188 * memory_hotplug.auto_movable_numa_aware: consider numa node stats
 189 */
 190#ifdef CONFIG_NUMA
 191static bool auto_movable_numa_aware __read_mostly = true;
 192module_param(auto_movable_numa_aware, bool, 0644);
 193MODULE_PARM_DESC(auto_movable_numa_aware,
 194		"Consider numa node stats in addition to global stats in "
 195		"\"auto-movable\" online policy. Default: true");
 196#endif /* CONFIG_NUMA */
 197
 198/*
 199 * online_page_callback contains pointer to current page onlining function.
 200 * Initially it is generic_online_page(). If it is required it could be
 201 * changed by calling set_online_page_callback() for callback registration
 202 * and restore_online_page_callback() for generic callback restore.
 203 */
 204
 
 
 205static online_page_callback_t online_page_callback = generic_online_page;
 206static DEFINE_MUTEX(online_page_callback_lock);
 207
 208DEFINE_STATIC_PERCPU_RWSEM(mem_hotplug_lock);
 209
 210void get_online_mems(void)
 211{
 212	percpu_down_read(&mem_hotplug_lock);
 213}
 214
 215void put_online_mems(void)
 216{
 217	percpu_up_read(&mem_hotplug_lock);
 218}
 219
 220bool movable_node_enabled = false;
 221
 222#ifndef CONFIG_MEMORY_HOTPLUG_DEFAULT_ONLINE
 223int mhp_default_online_type = MMOP_OFFLINE;
 224#else
 225int mhp_default_online_type = MMOP_ONLINE;
 226#endif
 227
 228static int __init setup_memhp_default_state(char *str)
 229{
 230	const int online_type = mhp_online_type_from_str(str);
 231
 232	if (online_type >= 0)
 233		mhp_default_online_type = online_type;
 234
 235	return 1;
 236}
 237__setup("memhp_default_state=", setup_memhp_default_state);
 238
 239void mem_hotplug_begin(void)
 240{
 241	cpus_read_lock();
 242	percpu_down_write(&mem_hotplug_lock);
 243}
 244
 245void mem_hotplug_done(void)
 246{
 247	percpu_up_write(&mem_hotplug_lock);
 248	cpus_read_unlock();
 249}
 250
 251u64 max_mem_size = U64_MAX;
 252
 253/* add this memory to iomem resource */
 254static struct resource *register_memory_resource(u64 start, u64 size,
 255						 const char *resource_name)
 256{
 257	struct resource *res;
 258	unsigned long flags =  IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
 259
 260	if (strcmp(resource_name, "System RAM"))
 261		flags |= IORESOURCE_SYSRAM_DRIVER_MANAGED;
 262
 263	if (!mhp_range_allowed(start, size, true))
 264		return ERR_PTR(-E2BIG);
 265
 266	/*
 267	 * Make sure value parsed from 'mem=' only restricts memory adding
 268	 * while booting, so that memory hotplug won't be impacted. Please
 269	 * refer to document of 'mem=' in kernel-parameters.txt for more
 270	 * details.
 271	 */
 272	if (start + size > max_mem_size && system_state < SYSTEM_RUNNING)
 273		return ERR_PTR(-E2BIG);
 274
 275	/*
 276	 * Request ownership of the new memory range.  This might be
 277	 * a child of an existing resource that was present but
 278	 * not marked as busy.
 279	 */
 280	res = __request_region(&iomem_resource, start, size,
 281			       resource_name, flags);
 282
 283	if (!res) {
 284		pr_debug("Unable to reserve System RAM region: %016llx->%016llx\n",
 285				start, start + size);
 286		return ERR_PTR(-EEXIST);
 287	}
 288	return res;
 289}
 290
 291static void release_memory_resource(struct resource *res)
 292{
 293	if (!res)
 294		return;
 295	release_resource(res);
 296	kfree(res);
 
 297}
 298
 299static int check_pfn_span(unsigned long pfn, unsigned long nr_pages)
 
 
 
 300{
 301	/*
 302	 * Disallow all operations smaller than a sub-section and only
 303	 * allow operations smaller than a section for
 304	 * SPARSEMEM_VMEMMAP. Note that check_hotplug_memory_range()
 305	 * enforces a larger memory_block_size_bytes() granularity for
 306	 * memory that will be marked online, so this check should only
 307	 * fire for direct arch_{add,remove}_memory() users outside of
 308	 * add_memory_resource().
 309	 */
 310	unsigned long min_align;
 
 
 
 
 
 
 
 
 
 
 
 
 311
 312	if (IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP))
 313		min_align = PAGES_PER_SUBSECTION;
 314	else
 315		min_align = PAGES_PER_SECTION;
 316	if (!IS_ALIGNED(pfn | nr_pages, min_align))
 317		return -EINVAL;
 318	return 0;
 319}
 320
 321/*
 322 * Return page for the valid pfn only if the page is online. All pfn
 323 * walkers which rely on the fully initialized page->flags and others
 324 * should use this rather than pfn_valid && pfn_to_page
 325 */
 326struct page *pfn_to_online_page(unsigned long pfn)
 327{
 328	unsigned long nr = pfn_to_section_nr(pfn);
 329	struct dev_pagemap *pgmap;
 330	struct mem_section *ms;
 
 
 
 
 331
 332	if (nr >= NR_MEM_SECTIONS)
 333		return NULL;
 334
 335	ms = __nr_to_section(nr);
 336	if (!online_section(ms))
 337		return NULL;
 338
 339	/*
 340	 * Save some code text when online_section() +
 341	 * pfn_section_valid() are sufficient.
 342	 */
 343	if (IS_ENABLED(CONFIG_HAVE_ARCH_PFN_VALID) && !pfn_valid(pfn))
 344		return NULL;
 
 345
 346	if (!pfn_section_valid(ms, pfn))
 347		return NULL;
 
 348
 349	if (!online_device_section(ms))
 350		return pfn_to_page(pfn);
 351
 352	/*
 353	 * Slowpath: when ZONE_DEVICE collides with
 354	 * ZONE_{NORMAL,MOVABLE} within the same section some pfns in
 355	 * the section may be 'offline' but 'valid'. Only
 356	 * get_dev_pagemap() can determine sub-section online status.
 357	 */
 358	pgmap = get_dev_pagemap(pfn, NULL);
 359	put_dev_pagemap(pgmap);
 360
 361	/* The presence of a pgmap indicates ZONE_DEVICE offline pfn */
 362	if (pgmap)
 363		return NULL;
 364
 365	return pfn_to_page(pfn);
 366}
 367EXPORT_SYMBOL_GPL(pfn_to_online_page);
 368
 369int __ref __add_pages(int nid, unsigned long pfn, unsigned long nr_pages,
 370		struct mhp_params *params)
 371{
 372	const unsigned long end_pfn = pfn + nr_pages;
 373	unsigned long cur_nr_pages;
 374	int err;
 375	struct vmem_altmap *altmap = params->altmap;
 
 
 
 376
 377	if (WARN_ON_ONCE(!pgprot_val(params->pgprot)))
 378		return -EINVAL;
 379
 380	VM_BUG_ON(!mhp_range_allowed(PFN_PHYS(pfn), nr_pages * PAGE_SIZE, false));
 
 
 
 
 
 
 381
 382	if (altmap) {
 383		/*
 384		 * Validate altmap is within bounds of the total request
 385		 */
 386		if (altmap->base_pfn != pfn
 387				|| vmem_altmap_offset(altmap) > nr_pages) {
 388			pr_warn_once("memory add fail, invalid altmap\n");
 389			return -EINVAL;
 390		}
 391		altmap->alloc = 0;
 392	}
 393
 394	if (check_pfn_span(pfn, nr_pages)) {
 395		WARN(1, "Misaligned %s start: %#lx end: %#lx\n", __func__, pfn, pfn + nr_pages - 1);
 396		return -EINVAL;
 397	}
 
 
 398
 399	for (; pfn < end_pfn; pfn += cur_nr_pages) {
 400		/* Select all remaining pages up to the next section boundary */
 401		cur_nr_pages = min(end_pfn - pfn,
 402				   SECTION_ALIGN_UP(pfn + 1) - pfn);
 403		err = sparse_add_section(nid, pfn, cur_nr_pages, altmap,
 404					 params->pgmap);
 405		if (err)
 406			break;
 407		cond_resched();
 408	}
 409	vmemmap_populate_print_last();
 410	return err;
 411}
 
 412
 413/* find the smallest valid pfn in the range [start_pfn, end_pfn) */
 414static unsigned long find_smallest_section_pfn(int nid, struct zone *zone,
 415				     unsigned long start_pfn,
 416				     unsigned long end_pfn)
 417{
 418	for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SUBSECTION) {
 419		if (unlikely(!pfn_to_online_page(start_pfn)))
 420			continue;
 421
 422		if (unlikely(pfn_to_nid(start_pfn) != nid))
 423			continue;
 424
 425		if (zone != page_zone(pfn_to_page(start_pfn)))
 426			continue;
 
 427
 428		return start_pfn;
 429	}
 430
 431	return 0;
 432}
 433
 434/* find the biggest valid pfn in the range [start_pfn, end_pfn). */
 435static unsigned long find_biggest_section_pfn(int nid, struct zone *zone,
 436				    unsigned long start_pfn,
 437				    unsigned long end_pfn)
 438{
 439	unsigned long pfn;
 
 
 
 
 440
 441	/* pfn is the end pfn of a memory section. */
 442	pfn = end_pfn - 1;
 443	for (; pfn >= start_pfn; pfn -= PAGES_PER_SUBSECTION) {
 444		if (unlikely(!pfn_to_online_page(pfn)))
 445			continue;
 446
 447		if (unlikely(pfn_to_nid(pfn) != nid))
 448			continue;
 
 
 
 
 
 449
 450		if (zone != page_zone(pfn_to_page(pfn)))
 451			continue;
 
 452
 453		return pfn;
 
 
 
 454	}
 455
 
 
 
 
 
 
 456	return 0;
 457}
 458
 459static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
 460			     unsigned long end_pfn)
 461{
 462	unsigned long pfn;
 463	int nid = zone_to_nid(zone);
 464
 465	if (zone->zone_start_pfn == start_pfn) {
 466		/*
 467		 * If the section is smallest section in the zone, it need
 468		 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
 469		 * In this case, we find second smallest valid mem_section
 470		 * for shrinking zone.
 471		 */
 472		pfn = find_smallest_section_pfn(nid, zone, end_pfn,
 473						zone_end_pfn(zone));
 474		if (pfn) {
 475			zone->spanned_pages = zone_end_pfn(zone) - pfn;
 476			zone->zone_start_pfn = pfn;
 477		} else {
 478			zone->zone_start_pfn = 0;
 479			zone->spanned_pages = 0;
 480		}
 481	} else if (zone_end_pfn(zone) == end_pfn) {
 482		/*
 483		 * If the section is biggest section in the zone, it need
 484		 * shrink zone->spanned_pages.
 485		 * In this case, we find second biggest valid mem_section for
 486		 * shrinking zone.
 487		 */
 488		pfn = find_biggest_section_pfn(nid, zone, zone->zone_start_pfn,
 489					       start_pfn);
 490		if (pfn)
 491			zone->spanned_pages = pfn - zone->zone_start_pfn + 1;
 492		else {
 493			zone->zone_start_pfn = 0;
 494			zone->spanned_pages = 0;
 495		}
 496	}
 497}
 498
 499static void update_pgdat_span(struct pglist_data *pgdat)
 500{
 501	unsigned long node_start_pfn = 0, node_end_pfn = 0;
 502	struct zone *zone;
 503
 504	for (zone = pgdat->node_zones;
 505	     zone < pgdat->node_zones + MAX_NR_ZONES; zone++) {
 506		unsigned long end_pfn = zone_end_pfn(zone);
 507
 508		/* No need to lock the zones, they can't change. */
 509		if (!zone->spanned_pages)
 510			continue;
 511		if (!node_end_pfn) {
 512			node_start_pfn = zone->zone_start_pfn;
 513			node_end_pfn = end_pfn;
 514			continue;
 515		}
 516
 517		if (end_pfn > node_end_pfn)
 518			node_end_pfn = end_pfn;
 519		if (zone->zone_start_pfn < node_start_pfn)
 520			node_start_pfn = zone->zone_start_pfn;
 521	}
 522
 523	pgdat->node_start_pfn = node_start_pfn;
 524	pgdat->node_spanned_pages = node_end_pfn - node_start_pfn;
 525}
 526
 527void __ref remove_pfn_range_from_zone(struct zone *zone,
 528				      unsigned long start_pfn,
 529				      unsigned long nr_pages)
 530{
 531	const unsigned long end_pfn = start_pfn + nr_pages;
 
 
 
 
 
 
 
 
 
 532	struct pglist_data *pgdat = zone->zone_pgdat;
 533	unsigned long pfn, cur_nr_pages;
 534
 535	/* Poison struct pages because they are now uninitialized again. */
 536	for (pfn = start_pfn; pfn < end_pfn; pfn += cur_nr_pages) {
 537		cond_resched();
 538
 539		/* Select all remaining pages up to the next section boundary */
 540		cur_nr_pages =
 541			min(end_pfn - pfn, SECTION_ALIGN_UP(pfn + 1) - pfn);
 542		page_init_poison(pfn_to_page(pfn),
 543				 sizeof(struct page) * cur_nr_pages);
 544	}
 545
 546	/*
 547	 * Zone shrinking code cannot properly deal with ZONE_DEVICE. So
 548	 * we will not try to shrink the zones - which is okay as
 549	 * set_zone_contiguous() cannot deal with ZONE_DEVICE either way.
 550	 */
 551	if (zone_is_zone_device(zone))
 552		return;
 553
 554	clear_zone_contiguous(zone);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 555
 556	shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
 557	update_pgdat_span(pgdat);
 558
 559	set_zone_contiguous(zone);
 560}
 
 561
 562/**
 563 * __remove_pages() - remove sections of pages
 564 * @pfn: starting pageframe (must be aligned to start of a section)
 
 565 * @nr_pages: number of pages to remove (must be multiple of section size)
 566 * @altmap: alternative device page map or %NULL if default memmap is used
 567 *
 568 * Generic helper function to remove section mappings and sysfs entries
 569 * for the section of the memory we are removing. Caller needs to make
 570 * sure that pages are marked reserved and zones are adjust properly by
 571 * calling offline_pages().
 572 */
 573void __remove_pages(unsigned long pfn, unsigned long nr_pages,
 574		    struct vmem_altmap *altmap)
 575{
 576	const unsigned long end_pfn = pfn + nr_pages;
 577	unsigned long cur_nr_pages;
 578
 579	if (check_pfn_span(pfn, nr_pages)) {
 580		WARN(1, "Misaligned %s start: %#lx end: %#lx\n", __func__, pfn, pfn + nr_pages - 1);
 581		return;
 582	}
 
 583
 584	for (; pfn < end_pfn; pfn += cur_nr_pages) {
 585		cond_resched();
 586		/* Select all remaining pages up to the next section boundary */
 587		cur_nr_pages = min(end_pfn - pfn,
 588				   SECTION_ALIGN_UP(pfn + 1) - pfn);
 589		sparse_remove_section(pfn, cur_nr_pages, altmap);
 
 
 590	}
 
 591}
 
 592
 593int set_online_page_callback(online_page_callback_t callback)
 594{
 595	int rc = -EINVAL;
 596
 597	get_online_mems();
 598	mutex_lock(&online_page_callback_lock);
 599
 600	if (online_page_callback == generic_online_page) {
 601		online_page_callback = callback;
 602		rc = 0;
 603	}
 604
 605	mutex_unlock(&online_page_callback_lock);
 606	put_online_mems();
 607
 608	return rc;
 609}
 610EXPORT_SYMBOL_GPL(set_online_page_callback);
 611
 612int restore_online_page_callback(online_page_callback_t callback)
 613{
 614	int rc = -EINVAL;
 615
 616	get_online_mems();
 617	mutex_lock(&online_page_callback_lock);
 618
 619	if (online_page_callback == callback) {
 620		online_page_callback = generic_online_page;
 621		rc = 0;
 622	}
 623
 624	mutex_unlock(&online_page_callback_lock);
 625	put_online_mems();
 626
 627	return rc;
 628}
 629EXPORT_SYMBOL_GPL(restore_online_page_callback);
 630
 631void generic_online_page(struct page *page, unsigned int order)
 632{
 633	/*
 634	 * Freeing the page with debug_pagealloc enabled will try to unmap it,
 635	 * so we should map it first. This is better than introducing a special
 636	 * case in page freeing fast path.
 637	 */
 638	debug_pagealloc_map_pages(page, 1 << order);
 639	__free_pages_core(page, order);
 640	totalram_pages_add(1UL << order);
 641}
 642EXPORT_SYMBOL_GPL(generic_online_page);
 643
 644static void online_pages_range(unsigned long start_pfn, unsigned long nr_pages)
 645{
 646	const unsigned long end_pfn = start_pfn + nr_pages;
 647	unsigned long pfn;
 648
 649	/*
 650	 * Online the pages in MAX_PAGE_ORDER aligned chunks. The callback might
 651	 * decide to not expose all pages to the buddy (e.g., expose them
 652	 * later). We account all pages as being online and belonging to this
 653	 * zone ("present").
 654	 * When using memmap_on_memory, the range might not be aligned to
 655	 * MAX_ORDER_NR_PAGES - 1, but pageblock aligned. __ffs() will detect
 656	 * this and the first chunk to online will be pageblock_nr_pages.
 657	 */
 658	for (pfn = start_pfn; pfn < end_pfn;) {
 659		int order;
 660
 661		/*
 662		 * Free to online pages in the largest chunks alignment allows.
 663		 *
 664		 * __ffs() behaviour is undefined for 0. start == 0 is
 665		 * MAX_PAGE_ORDER-aligned, Set order to MAX_PAGE_ORDER for
 666		 * the case.
 667		 */
 668		if (pfn)
 669			order = min_t(int, MAX_PAGE_ORDER, __ffs(pfn));
 670		else
 671			order = MAX_PAGE_ORDER;
 672
 673		(*online_page_callback)(pfn_to_page(pfn), order);
 674		pfn += (1UL << order);
 675	}
 676
 677	/* mark all involved sections as online */
 678	online_mem_sections(start_pfn, end_pfn);
 679}
 
 680
 681/* check which state of node_states will be changed when online memory */
 682static void node_states_check_changes_online(unsigned long nr_pages,
 683	struct zone *zone, struct memory_notify *arg)
 684{
 685	int nid = zone_to_nid(zone);
 686
 687	arg->status_change_nid = NUMA_NO_NODE;
 688	arg->status_change_nid_normal = NUMA_NO_NODE;
 689
 690	if (!node_state(nid, N_MEMORY))
 691		arg->status_change_nid = nid;
 692	if (zone_idx(zone) <= ZONE_NORMAL && !node_state(nid, N_NORMAL_MEMORY))
 693		arg->status_change_nid_normal = nid;
 694}
 695
 696static void node_states_set_node(int node, struct memory_notify *arg)
 697{
 698	if (arg->status_change_nid_normal >= 0)
 699		node_set_state(node, N_NORMAL_MEMORY);
 700
 701	if (arg->status_change_nid >= 0)
 702		node_set_state(node, N_MEMORY);
 703}
 704
 705static void __meminit resize_zone_range(struct zone *zone, unsigned long start_pfn,
 706		unsigned long nr_pages)
 707{
 708	unsigned long old_end_pfn = zone_end_pfn(zone);
 709
 710	if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
 711		zone->zone_start_pfn = start_pfn;
 712
 713	zone->spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - zone->zone_start_pfn;
 714}
 715
 716static void __meminit resize_pgdat_range(struct pglist_data *pgdat, unsigned long start_pfn,
 717                                     unsigned long nr_pages)
 718{
 719	unsigned long old_end_pfn = pgdat_end_pfn(pgdat);
 720
 721	if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
 722		pgdat->node_start_pfn = start_pfn;
 723
 724	pgdat->node_spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - pgdat->node_start_pfn;
 725
 726}
 727
 728#ifdef CONFIG_ZONE_DEVICE
 729static void section_taint_zone_device(unsigned long pfn)
 730{
 731	struct mem_section *ms = __pfn_to_section(pfn);
 732
 733	ms->section_mem_map |= SECTION_TAINT_ZONE_DEVICE;
 734}
 735#else
 736static inline void section_taint_zone_device(unsigned long pfn)
 737{
 738}
 739#endif
 740
 741/*
 742 * Associate the pfn range with the given zone, initializing the memmaps
 743 * and resizing the pgdat/zone data to span the added pages. After this
 744 * call, all affected pages are PG_reserved.
 745 *
 746 * All aligned pageblocks are initialized to the specified migratetype
 747 * (usually MIGRATE_MOVABLE). Besides setting the migratetype, no related
 748 * zone stats (e.g., nr_isolate_pageblock) are touched.
 749 */
 750void __ref move_pfn_range_to_zone(struct zone *zone, unsigned long start_pfn,
 751				  unsigned long nr_pages,
 752				  struct vmem_altmap *altmap, int migratetype)
 753{
 754	struct pglist_data *pgdat = zone->zone_pgdat;
 755	int nid = pgdat->node_id;
 756
 757	clear_zone_contiguous(zone);
 758
 759	if (zone_is_empty(zone))
 760		init_currently_empty_zone(zone, start_pfn, nr_pages);
 761	resize_zone_range(zone, start_pfn, nr_pages);
 762	resize_pgdat_range(pgdat, start_pfn, nr_pages);
 763
 764	/*
 765	 * Subsection population requires care in pfn_to_online_page().
 766	 * Set the taint to enable the slow path detection of
 767	 * ZONE_DEVICE pages in an otherwise  ZONE_{NORMAL,MOVABLE}
 768	 * section.
 769	 */
 770	if (zone_is_zone_device(zone)) {
 771		if (!IS_ALIGNED(start_pfn, PAGES_PER_SECTION))
 772			section_taint_zone_device(start_pfn);
 773		if (!IS_ALIGNED(start_pfn + nr_pages, PAGES_PER_SECTION))
 774			section_taint_zone_device(start_pfn + nr_pages);
 775	}
 776
 777	/*
 778	 * TODO now we have a visible range of pages which are not associated
 779	 * with their zone properly. Not nice but set_pfnblock_flags_mask
 780	 * expects the zone spans the pfn range. All the pages in the range
 781	 * are reserved so nobody should be touching them so we should be safe
 782	 */
 783	memmap_init_range(nr_pages, nid, zone_idx(zone), start_pfn, 0,
 784			 MEMINIT_HOTPLUG, altmap, migratetype);
 785
 786	set_zone_contiguous(zone);
 787}
 
 788
 789struct auto_movable_stats {
 790	unsigned long kernel_early_pages;
 791	unsigned long movable_pages;
 792};
 793
 794static void auto_movable_stats_account_zone(struct auto_movable_stats *stats,
 795					    struct zone *zone)
 796{
 797	if (zone_idx(zone) == ZONE_MOVABLE) {
 798		stats->movable_pages += zone->present_pages;
 799	} else {
 800		stats->kernel_early_pages += zone->present_early_pages;
 801#ifdef CONFIG_CMA
 802		/*
 803		 * CMA pages (never on hotplugged memory) behave like
 804		 * ZONE_MOVABLE.
 805		 */
 806		stats->movable_pages += zone->cma_pages;
 807		stats->kernel_early_pages -= zone->cma_pages;
 808#endif /* CONFIG_CMA */
 809	}
 810}
 811struct auto_movable_group_stats {
 812	unsigned long movable_pages;
 813	unsigned long req_kernel_early_pages;
 814};
 815
 816static int auto_movable_stats_account_group(struct memory_group *group,
 817					   void *arg)
 818{
 819	const int ratio = READ_ONCE(auto_movable_ratio);
 820	struct auto_movable_group_stats *stats = arg;
 821	long pages;
 822
 823	/*
 824	 * We don't support modifying the config while the auto-movable online
 825	 * policy is already enabled. Just avoid the division by zero below.
 826	 */
 827	if (!ratio)
 828		return 0;
 829
 830	/*
 831	 * Calculate how many early kernel pages this group requires to
 832	 * satisfy the configured zone ratio.
 833	 */
 834	pages = group->present_movable_pages * 100 / ratio;
 835	pages -= group->present_kernel_pages;
 836
 837	if (pages > 0)
 838		stats->req_kernel_early_pages += pages;
 839	stats->movable_pages += group->present_movable_pages;
 840	return 0;
 841}
 
 842
 843static bool auto_movable_can_online_movable(int nid, struct memory_group *group,
 844					    unsigned long nr_pages)
 845{
 846	unsigned long kernel_early_pages, movable_pages;
 847	struct auto_movable_group_stats group_stats = {};
 848	struct auto_movable_stats stats = {};
 849	pg_data_t *pgdat = NODE_DATA(nid);
 850	struct zone *zone;
 851	int i;
 852
 853	/* Walk all relevant zones and collect MOVABLE vs. KERNEL stats. */
 854	if (nid == NUMA_NO_NODE) {
 855		/* TODO: cache values */
 856		for_each_populated_zone(zone)
 857			auto_movable_stats_account_zone(&stats, zone);
 858	} else {
 859		for (i = 0; i < MAX_NR_ZONES; i++) {
 860			zone = pgdat->node_zones + i;
 861			if (populated_zone(zone))
 862				auto_movable_stats_account_zone(&stats, zone);
 863		}
 864	}
 865
 866	kernel_early_pages = stats.kernel_early_pages;
 867	movable_pages = stats.movable_pages;
 868
 869	/*
 870	 * Kernel memory inside dynamic memory group allows for more MOVABLE
 871	 * memory within the same group. Remove the effect of all but the
 872	 * current group from the stats.
 873	 */
 874	walk_dynamic_memory_groups(nid, auto_movable_stats_account_group,
 875				   group, &group_stats);
 876	if (kernel_early_pages <= group_stats.req_kernel_early_pages)
 877		return false;
 878	kernel_early_pages -= group_stats.req_kernel_early_pages;
 879	movable_pages -= group_stats.movable_pages;
 880
 881	if (group && group->is_dynamic)
 882		kernel_early_pages += group->present_kernel_pages;
 883
 884	/*
 885	 * Test if we could online the given number of pages to ZONE_MOVABLE
 886	 * and still stay in the configured ratio.
 887	 */
 888	movable_pages += nr_pages;
 889	return movable_pages <= (auto_movable_ratio * kernel_early_pages) / 100;
 890}
 891
 892/*
 893 * Returns a default kernel memory zone for the given pfn range.
 894 * If no kernel zone covers this pfn range it will automatically go
 895 * to the ZONE_NORMAL.
 896 */
 897static struct zone *default_kernel_zone_for_pfn(int nid, unsigned long start_pfn,
 898		unsigned long nr_pages)
 899{
 900	struct pglist_data *pgdat = NODE_DATA(nid);
 901	int zid;
 902
 903	for (zid = 0; zid < ZONE_NORMAL; zid++) {
 904		struct zone *zone = &pgdat->node_zones[zid];
 905
 906		if (zone_intersects(zone, start_pfn, nr_pages))
 907			return zone;
 908	}
 909
 910	return &pgdat->node_zones[ZONE_NORMAL];
 911}
 912
 913/*
 914 * Determine to which zone to online memory dynamically based on user
 915 * configuration and system stats. We care about the following ratio:
 916 *
 917 *   MOVABLE : KERNEL
 918 *
 919 * Whereby MOVABLE is memory in ZONE_MOVABLE and KERNEL is memory in
 920 * one of the kernel zones. CMA pages inside one of the kernel zones really
 921 * behaves like ZONE_MOVABLE, so we treat them accordingly.
 922 *
 923 * We don't allow for hotplugged memory in a KERNEL zone to increase the
 924 * amount of MOVABLE memory we can have, so we end up with:
 925 *
 926 *   MOVABLE : KERNEL_EARLY
 927 *
 928 * Whereby KERNEL_EARLY is memory in one of the kernel zones, available sinze
 929 * boot. We base our calculation on KERNEL_EARLY internally, because:
 930 *
 931 * a) Hotplugged memory in one of the kernel zones can sometimes still get
 932 *    hotunplugged, especially when hot(un)plugging individual memory blocks.
 933 *    There is no coordination across memory devices, therefore "automatic"
 934 *    hotunplugging, as implemented in hypervisors, could result in zone
 935 *    imbalances.
 936 * b) Early/boot memory in one of the kernel zones can usually not get
 937 *    hotunplugged again (e.g., no firmware interface to unplug, fragmented
 938 *    with unmovable allocations). While there are corner cases where it might
 939 *    still work, it is barely relevant in practice.
 940 *
 941 * Exceptions are dynamic memory groups, which allow for more MOVABLE
 942 * memory within the same memory group -- because in that case, there is
 943 * coordination within the single memory device managed by a single driver.
 944 *
 945 * We rely on "present pages" instead of "managed pages", as the latter is
 946 * highly unreliable and dynamic in virtualized environments, and does not
 947 * consider boot time allocations. For example, memory ballooning adjusts the
 948 * managed pages when inflating/deflating the balloon, and balloon compaction
 949 * can even migrate inflated pages between zones.
 950 *
 951 * Using "present pages" is better but some things to keep in mind are:
 952 *
 953 * a) Some memblock allocations, such as for the crashkernel area, are
 954 *    effectively unused by the kernel, yet they account to "present pages".
 955 *    Fortunately, these allocations are comparatively small in relevant setups
 956 *    (e.g., fraction of system memory).
 957 * b) Some hotplugged memory blocks in virtualized environments, esecially
 958 *    hotplugged by virtio-mem, look like they are completely present, however,
 959 *    only parts of the memory block are actually currently usable.
 960 *    "present pages" is an upper limit that can get reached at runtime. As
 961 *    we base our calculations on KERNEL_EARLY, this is not an issue.
 962 */
 963static struct zone *auto_movable_zone_for_pfn(int nid,
 964					      struct memory_group *group,
 965					      unsigned long pfn,
 966					      unsigned long nr_pages)
 967{
 968	unsigned long online_pages = 0, max_pages, end_pfn;
 
 969	struct page *page;
 970
 971	if (!auto_movable_ratio)
 972		goto kernel_zone;
 973
 974	if (group && !group->is_dynamic) {
 975		max_pages = group->s.max_pages;
 976		online_pages = group->present_movable_pages;
 977
 978		/* If anything is !MOVABLE online the rest !MOVABLE. */
 979		if (group->present_kernel_pages)
 980			goto kernel_zone;
 981	} else if (!group || group->d.unit_pages == nr_pages) {
 982		max_pages = nr_pages;
 983	} else {
 984		max_pages = group->d.unit_pages;
 985		/*
 986		 * Take a look at all online sections in the current unit.
 987		 * We can safely assume that all pages within a section belong
 988		 * to the same zone, because dynamic memory groups only deal
 989		 * with hotplugged memory.
 990		 */
 991		pfn = ALIGN_DOWN(pfn, group->d.unit_pages);
 992		end_pfn = pfn + group->d.unit_pages;
 993		for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
 994			page = pfn_to_online_page(pfn);
 995			if (!page)
 996				continue;
 997			/* If anything is !MOVABLE online the rest !MOVABLE. */
 998			if (!is_zone_movable_page(page))
 999				goto kernel_zone;
1000			online_pages += PAGES_PER_SECTION;
1001		}
1002	}
1003
1004	/*
1005	 * Online MOVABLE if we could *currently* online all remaining parts
1006	 * MOVABLE. We expect to (add+) online them immediately next, so if
1007	 * nobody interferes, all will be MOVABLE if possible.
1008	 */
1009	nr_pages = max_pages - online_pages;
1010	if (!auto_movable_can_online_movable(NUMA_NO_NODE, group, nr_pages))
1011		goto kernel_zone;
1012
1013#ifdef CONFIG_NUMA
1014	if (auto_movable_numa_aware &&
1015	    !auto_movable_can_online_movable(nid, group, nr_pages))
1016		goto kernel_zone;
1017#endif /* CONFIG_NUMA */
1018
1019	return &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
1020kernel_zone:
1021	return default_kernel_zone_for_pfn(nid, pfn, nr_pages);
1022}
1023
1024static inline struct zone *default_zone_for_pfn(int nid, unsigned long start_pfn,
1025		unsigned long nr_pages)
1026{
1027	struct zone *kernel_zone = default_kernel_zone_for_pfn(nid, start_pfn,
1028			nr_pages);
1029	struct zone *movable_zone = &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
1030	bool in_kernel = zone_intersects(kernel_zone, start_pfn, nr_pages);
1031	bool in_movable = zone_intersects(movable_zone, start_pfn, nr_pages);
1032
1033	/*
1034	 * We inherit the existing zone in a simple case where zones do not
1035	 * overlap in the given range
1036	 */
1037	if (in_kernel ^ in_movable)
1038		return (in_kernel) ? kernel_zone : movable_zone;
1039
1040	/*
1041	 * If the range doesn't belong to any zone or two zones overlap in the
1042	 * given range then we use movable zone only if movable_node is
1043	 * enabled because we always online to a kernel zone by default.
1044	 */
1045	return movable_node_enabled ? movable_zone : kernel_zone;
1046}
1047
1048struct zone *zone_for_pfn_range(int online_type, int nid,
1049		struct memory_group *group, unsigned long start_pfn,
1050		unsigned long nr_pages)
1051{
1052	if (online_type == MMOP_ONLINE_KERNEL)
1053		return default_kernel_zone_for_pfn(nid, start_pfn, nr_pages);
1054
1055	if (online_type == MMOP_ONLINE_MOVABLE)
1056		return &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
1057
1058	if (online_policy == ONLINE_POLICY_AUTO_MOVABLE)
1059		return auto_movable_zone_for_pfn(nid, group, start_pfn, nr_pages);
1060
1061	return default_zone_for_pfn(nid, start_pfn, nr_pages);
1062}
1063
1064/*
1065 * This function should only be called by memory_block_{online,offline},
1066 * and {online,offline}_pages.
1067 */
1068void adjust_present_page_count(struct page *page, struct memory_group *group,
1069			       long nr_pages)
1070{
1071	struct zone *zone = page_zone(page);
1072	const bool movable = zone_idx(zone) == ZONE_MOVABLE;
1073
1074	/*
1075	 * We only support onlining/offlining/adding/removing of complete
1076	 * memory blocks; therefore, either all is either early or hotplugged.
1077	 */
1078	if (early_section(__pfn_to_section(page_to_pfn(page))))
1079		zone->present_early_pages += nr_pages;
1080	zone->present_pages += nr_pages;
1081	zone->zone_pgdat->node_present_pages += nr_pages;
1082
1083	if (group && movable)
1084		group->present_movable_pages += nr_pages;
1085	else if (group && !movable)
1086		group->present_kernel_pages += nr_pages;
1087}
1088
1089int mhp_init_memmap_on_memory(unsigned long pfn, unsigned long nr_pages,
1090			      struct zone *zone)
1091{
1092	unsigned long end_pfn = pfn + nr_pages;
1093	int ret, i;
1094
1095	ret = kasan_add_zero_shadow(__va(PFN_PHYS(pfn)), PFN_PHYS(nr_pages));
1096	if (ret)
1097		return ret;
1098
1099	move_pfn_range_to_zone(zone, pfn, nr_pages, NULL, MIGRATE_UNMOVABLE);
1100
1101	for (i = 0; i < nr_pages; i++)
1102		SetPageVmemmapSelfHosted(pfn_to_page(pfn + i));
1103
1104	/*
1105	 * It might be that the vmemmap_pages fully span sections. If that is
1106	 * the case, mark those sections online here as otherwise they will be
1107	 * left offline.
1108	 */
1109	if (nr_pages >= PAGES_PER_SECTION)
1110	        online_mem_sections(pfn, ALIGN_DOWN(end_pfn, PAGES_PER_SECTION));
1111
1112	return ret;
1113}
1114
1115void mhp_deinit_memmap_on_memory(unsigned long pfn, unsigned long nr_pages)
1116{
1117	unsigned long end_pfn = pfn + nr_pages;
1118
1119	/*
1120	 * It might be that the vmemmap_pages fully span sections. If that is
1121	 * the case, mark those sections offline here as otherwise they will be
1122	 * left online.
1123	 */
1124	if (nr_pages >= PAGES_PER_SECTION)
1125		offline_mem_sections(pfn, ALIGN_DOWN(end_pfn, PAGES_PER_SECTION));
1126
1127        /*
1128	 * The pages associated with this vmemmap have been offlined, so
1129	 * we can reset its state here.
1130	 */
1131	remove_pfn_range_from_zone(page_zone(pfn_to_page(pfn)), pfn, nr_pages);
1132	kasan_remove_zero_shadow(__va(PFN_PHYS(pfn)), PFN_PHYS(nr_pages));
1133}
1134
1135/*
1136 * Must be called with mem_hotplug_lock in write mode.
1137 */
1138int __ref online_pages(unsigned long pfn, unsigned long nr_pages,
1139		       struct zone *zone, struct memory_group *group)
1140{
1141	unsigned long flags;
1142	int need_zonelists_rebuild = 0;
1143	const int nid = zone_to_nid(zone);
1144	int ret;
1145	struct memory_notify arg;
1146
1147	/*
1148	 * {on,off}lining is constrained to full memory sections (or more
1149	 * precisely to memory blocks from the user space POV).
1150	 * memmap_on_memory is an exception because it reserves initial part
1151	 * of the physical memory space for vmemmaps. That space is pageblock
1152	 * aligned.
1153	 */
1154	if (WARN_ON_ONCE(!nr_pages || !pageblock_aligned(pfn) ||
1155			 !IS_ALIGNED(pfn + nr_pages, PAGES_PER_SECTION)))
1156		return -EINVAL;
1157
1158
1159	/* associate pfn range with the zone */
1160	move_pfn_range_to_zone(zone, pfn, nr_pages, NULL, MIGRATE_ISOLATE);
1161
1162	arg.start_pfn = pfn;
1163	arg.nr_pages = nr_pages;
1164	node_states_check_changes_online(nr_pages, zone, &arg);
 
 
 
 
1165
1166	ret = memory_notify(MEM_GOING_ONLINE, &arg);
1167	ret = notifier_to_errno(ret);
1168	if (ret)
1169		goto failed_addition;
1170
 
 
1171	/*
1172	 * Fixup the number of isolated pageblocks before marking the sections
1173	 * onlining, such that undo_isolate_page_range() works correctly.
 
1174	 */
1175	spin_lock_irqsave(&zone->lock, flags);
1176	zone->nr_isolate_pageblock += nr_pages / pageblock_nr_pages;
1177	spin_unlock_irqrestore(&zone->lock, flags);
1178
1179	/*
1180	 * If this zone is not populated, then it is not in zonelist.
1181	 * This means the page allocator ignores this zone.
1182	 * So, zonelist must be updated after online.
1183	 */
1184	if (!populated_zone(zone)) {
 
1185		need_zonelists_rebuild = 1;
1186		setup_zone_pageset(zone);
 
 
 
 
 
 
 
 
 
1187	}
1188
1189	online_pages_range(pfn, nr_pages);
1190	adjust_present_page_count(pfn_to_page(pfn), group, nr_pages);
1191
1192	node_states_set_node(nid, &arg);
1193	if (need_zonelists_rebuild)
1194		build_all_zonelists(NULL);
 
 
1195
1196	/* Basic onlining is complete, allow allocation of onlined pages. */
1197	undo_isolate_page_range(pfn, pfn + nr_pages, MIGRATE_MOVABLE);
1198
1199	/*
1200	 * Freshly onlined pages aren't shuffled (e.g., all pages are placed to
1201	 * the tail of the freelist when undoing isolation). Shuffle the whole
1202	 * zone to make sure the just onlined pages are properly distributed
1203	 * across the whole freelist - to create an initial shuffle.
1204	 */
1205	shuffle_zone(zone);
1206
1207	/* reinitialise watermarks and update pcp limits */
1208	init_per_zone_wmark_min();
 
 
1209
1210	kswapd_run(nid);
1211	kcompactd_run(nid);
1212
1213	writeback_set_ratelimit();
1214
1215	memory_notify(MEM_ONLINE, &arg);
 
 
 
1216	return 0;
1217
1218failed_addition:
1219	pr_debug("online_pages [mem %#010llx-%#010llx] failed\n",
1220		 (unsigned long long) pfn << PAGE_SHIFT,
1221		 (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1);
1222	memory_notify(MEM_CANCEL_ONLINE, &arg);
1223	remove_pfn_range_from_zone(zone, pfn, nr_pages);
1224	return ret;
1225}
 
1226
1227/* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1228static pg_data_t __ref *hotadd_init_pgdat(int nid)
1229{
1230	struct pglist_data *pgdat;
 
 
 
 
 
 
 
 
 
1231
1232	/*
1233	 * NODE_DATA is preallocated (free_area_init) but its internal
1234	 * state is not allocated completely. Add missing pieces.
1235	 * Completely offline nodes stay around and they just need
1236	 * reintialization.
1237	 */
1238	pgdat = NODE_DATA(nid);
1239
1240	/* init node's zones as empty zones, we don't have any present pages.*/
1241	free_area_init_core_hotplug(pgdat);
1242
1243	/*
1244	 * The node we allocated has no zone fallback lists. For avoiding
1245	 * to access not-initialized zonelist, build here.
1246	 */
1247	build_all_zonelists(pgdat);
 
 
1248
1249	return pgdat;
1250}
1251
 
 
 
 
 
 
 
 
1252/*
1253 * __try_online_node - online a node if offlined
1254 * @nid: the node ID
1255 * @set_node_online: Whether we want to online the node
1256 * called by cpu_up() to online a node without onlined memory.
1257 *
1258 * Returns:
1259 * 1 -> a new node has been allocated
1260 * 0 -> the node is already online
1261 * -ENOMEM -> the node could not be allocated
1262 */
1263static int __try_online_node(int nid, bool set_node_online)
1264{
1265	pg_data_t *pgdat;
1266	int ret = 1;
1267
1268	if (node_online(nid))
1269		return 0;
1270
1271	pgdat = hotadd_init_pgdat(nid);
 
1272	if (!pgdat) {
1273		pr_err("Cannot online node %d due to NULL pgdat\n", nid);
1274		ret = -ENOMEM;
1275		goto out;
1276	}
 
 
 
1277
1278	if (set_node_online) {
1279		node_set_online(nid);
1280		ret = register_one_node(nid);
1281		BUG_ON(ret);
1282	}
1283out:
 
1284	return ret;
1285}
1286
1287/*
1288 * Users of this function always want to online/register the node
1289 */
1290int try_online_node(int nid)
1291{
 
 
 
1292	int ret;
1293
1294	mem_hotplug_begin();
1295	ret =  __try_online_node(nid, true);
1296	mem_hotplug_done();
1297	return ret;
1298}
1299
1300static int check_hotplug_memory_range(u64 start, u64 size)
1301{
1302	/* memory range must be block size aligned */
1303	if (!size || !IS_ALIGNED(start, memory_block_size_bytes()) ||
1304	    !IS_ALIGNED(size, memory_block_size_bytes())) {
1305		pr_err("Block size [%#lx] unaligned hotplug range: start %#llx, size %#llx",
1306		       memory_block_size_bytes(), start, size);
1307		return -EINVAL;
1308	}
1309
1310	return 0;
1311}
1312
1313static int online_memory_block(struct memory_block *mem, void *arg)
1314{
1315	mem->online_type = mhp_default_online_type;
1316	return device_online(&mem->dev);
1317}
1318
1319#ifndef arch_supports_memmap_on_memory
1320static inline bool arch_supports_memmap_on_memory(unsigned long vmemmap_size)
1321{
1322	/*
1323	 * As default, we want the vmemmap to span a complete PMD such that we
1324	 * can map the vmemmap using a single PMD if supported by the
1325	 * architecture.
1326	 */
1327	return IS_ALIGNED(vmemmap_size, PMD_SIZE);
1328}
1329#endif
1330
1331static bool mhp_supports_memmap_on_memory(unsigned long size)
1332{
1333	unsigned long vmemmap_size = memory_block_memmap_size();
1334	unsigned long memmap_pages = memory_block_memmap_on_memory_pages();
1335
1336	/*
1337	 * Besides having arch support and the feature enabled at runtime, we
1338	 * need a few more assumptions to hold true:
1339	 *
1340	 * a) We span a single memory block: memory onlining/offlinin;g happens
1341	 *    in memory block granularity. We don't want the vmemmap of online
1342	 *    memory blocks to reside on offline memory blocks. In the future,
1343	 *    we might want to support variable-sized memory blocks to make the
1344	 *    feature more versatile.
1345	 *
1346	 * b) The vmemmap pages span complete PMDs: We don't want vmemmap code
1347	 *    to populate memory from the altmap for unrelated parts (i.e.,
1348	 *    other memory blocks)
1349	 *
1350	 * c) The vmemmap pages (and thereby the pages that will be exposed to
1351	 *    the buddy) have to cover full pageblocks: memory onlining/offlining
1352	 *    code requires applicable ranges to be page-aligned, for example, to
1353	 *    set the migratetypes properly.
1354	 *
1355	 * TODO: Although we have a check here to make sure that vmemmap pages
1356	 *       fully populate a PMD, it is not the right place to check for
1357	 *       this. A much better solution involves improving vmemmap code
1358	 *       to fallback to base pages when trying to populate vmemmap using
1359	 *       altmap as an alternative source of memory, and we do not exactly
1360	 *       populate a single PMD.
1361	 */
1362	if (!mhp_memmap_on_memory() || size != memory_block_size_bytes())
1363		return false;
1364
1365	/*
1366	 * Make sure the vmemmap allocation is fully contained
1367	 * so that we always allocate vmemmap memory from altmap area.
1368	 */
1369	if (!IS_ALIGNED(vmemmap_size, PAGE_SIZE))
1370		return false;
1371
1372	/*
1373	 * start pfn should be pageblock_nr_pages aligned for correctly
1374	 * setting migrate types
1375	 */
1376	if (!pageblock_aligned(memmap_pages))
1377		return false;
1378
1379	if (memmap_pages == PHYS_PFN(memory_block_size_bytes()))
1380		/* No effective hotplugged memory doesn't make sense. */
1381		return false;
1382
1383	return arch_supports_memmap_on_memory(vmemmap_size);
1384}
1385
1386static void __ref remove_memory_blocks_and_altmaps(u64 start, u64 size)
1387{
1388	unsigned long memblock_size = memory_block_size_bytes();
1389	u64 cur_start;
1390
1391	/*
1392	 * For memmap_on_memory, the altmaps were added on a per-memblock
1393	 * basis; we have to process each individual memory block.
1394	 */
1395	for (cur_start = start; cur_start < start + size;
1396	     cur_start += memblock_size) {
1397		struct vmem_altmap *altmap = NULL;
1398		struct memory_block *mem;
1399
1400		mem = find_memory_block(pfn_to_section_nr(PFN_DOWN(cur_start)));
1401		if (WARN_ON_ONCE(!mem))
1402			continue;
1403
1404		altmap = mem->altmap;
1405		mem->altmap = NULL;
1406
1407		remove_memory_block_devices(cur_start, memblock_size);
1408
1409		arch_remove_memory(cur_start, memblock_size, altmap);
1410
1411		/* Verify that all vmemmap pages have actually been freed. */
1412		WARN(altmap->alloc, "Altmap not fully unmapped");
1413		kfree(altmap);
1414	}
1415}
1416
1417static int create_altmaps_and_memory_blocks(int nid, struct memory_group *group,
1418					    u64 start, u64 size)
1419{
1420	unsigned long memblock_size = memory_block_size_bytes();
1421	u64 cur_start;
1422	int ret;
1423
1424	for (cur_start = start; cur_start < start + size;
1425	     cur_start += memblock_size) {
1426		struct mhp_params params = { .pgprot =
1427						     pgprot_mhp(PAGE_KERNEL) };
1428		struct vmem_altmap mhp_altmap = {
1429			.base_pfn = PHYS_PFN(cur_start),
1430			.end_pfn = PHYS_PFN(cur_start + memblock_size - 1),
1431		};
1432
1433		mhp_altmap.free = memory_block_memmap_on_memory_pages();
1434		params.altmap = kmemdup(&mhp_altmap, sizeof(struct vmem_altmap),
1435					GFP_KERNEL);
1436		if (!params.altmap) {
1437			ret = -ENOMEM;
1438			goto out;
1439		}
1440
1441		/* call arch's memory hotadd */
1442		ret = arch_add_memory(nid, cur_start, memblock_size, &params);
1443		if (ret < 0) {
1444			kfree(params.altmap);
1445			goto out;
1446		}
1447
1448		/* create memory block devices after memory was added */
1449		ret = create_memory_block_devices(cur_start, memblock_size,
1450						  params.altmap, group);
1451		if (ret) {
1452			arch_remove_memory(cur_start, memblock_size, NULL);
1453			kfree(params.altmap);
1454			goto out;
1455		}
1456	}
1457
1458	return 0;
1459out:
1460	if (ret && cur_start != start)
1461		remove_memory_blocks_and_altmaps(start, cur_start - start);
1462	return ret;
1463}
1464
1465/*
1466 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1467 * and online/offline operations (triggered e.g. by sysfs).
1468 *
1469 * we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG
1470 */
1471int __ref add_memory_resource(int nid, struct resource *res, mhp_t mhp_flags)
1472{
1473	struct mhp_params params = { .pgprot = pgprot_mhp(PAGE_KERNEL) };
1474	enum memblock_flags memblock_flags = MEMBLOCK_NONE;
1475	struct memory_group *group = NULL;
1476	u64 start, size;
1477	bool new_node = false;
1478	int ret;
1479
1480	start = res->start;
1481	size = resource_size(res);
1482
1483	ret = check_hotplug_memory_range(start, size);
1484	if (ret)
1485		return ret;
1486
1487	if (mhp_flags & MHP_NID_IS_MGID) {
1488		group = memory_group_find_by_id(nid);
1489		if (!group)
1490			return -EINVAL;
1491		nid = group->nid;
1492	}
1493
1494	if (!node_possible(nid)) {
1495		WARN(1, "node %d was absent from the node_possible_map\n", nid);
1496		return -EINVAL;
1497	}
1498
1499	mem_hotplug_begin();
 
1500
1501	if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK)) {
1502		if (res->flags & IORESOURCE_SYSRAM_DRIVER_MANAGED)
1503			memblock_flags = MEMBLOCK_DRIVER_MANAGED;
1504		ret = memblock_add_node(start, size, nid, memblock_flags);
1505		if (ret)
1506			goto error_mem_hotplug_end;
1507	}
1508
1509	ret = __try_online_node(nid, false);
1510	if (ret < 0)
1511		goto error;
1512	new_node = ret;
1513
1514	/*
1515	 * Self hosted memmap array
1516	 */
1517	if ((mhp_flags & MHP_MEMMAP_ON_MEMORY) &&
1518	    mhp_supports_memmap_on_memory(memory_block_size_bytes())) {
1519		ret = create_altmaps_and_memory_blocks(nid, group, start, size);
1520		if (ret)
1521			goto error;
1522	} else {
1523		ret = arch_add_memory(nid, start, size, &params);
1524		if (ret < 0)
1525			goto error;
1526
1527		/* create memory block devices after memory was added */
1528		ret = create_memory_block_devices(start, size, NULL, group);
1529		if (ret) {
1530			arch_remove_memory(start, size, params.altmap);
1531			goto error;
1532		}
1533	}
1534
1535	if (new_node) {
1536		/* If sysfs file of new node can't be created, cpu on the node
 
 
1537		 * can't be hot-added. There is no rollback way now.
1538		 * So, check by BUG_ON() to catch it reluctantly..
1539		 * We online node here. We can't roll back from here.
1540		 */
1541		node_set_online(nid);
1542		ret = __register_one_node(nid);
1543		BUG_ON(ret);
1544	}
1545
1546	register_memory_blocks_under_node(nid, PFN_DOWN(start),
1547					  PFN_UP(start + size - 1),
1548					  MEMINIT_HOTPLUG);
1549
1550	/* create new memmap entry */
1551	if (!strcmp(res->name, "System RAM"))
1552		firmware_map_add_hotplug(start, start + size, "System RAM");
1553
1554	/* device_online() will take the lock when calling online_pages() */
1555	mem_hotplug_done();
1556
1557	/*
1558	 * In case we're allowed to merge the resource, flag it and trigger
1559	 * merging now that adding succeeded.
1560	 */
1561	if (mhp_flags & MHP_MERGE_RESOURCE)
1562		merge_system_ram_resource(res);
1563
1564	/* online pages if requested */
1565	if (mhp_default_online_type != MMOP_OFFLINE)
1566		walk_memory_blocks(start, size, NULL, online_memory_block);
1567
1568	return ret;
1569error:
1570	if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK))
1571		memblock_remove(start, size);
1572error_mem_hotplug_end:
1573	mem_hotplug_done();
1574	return ret;
1575}
 
1576
1577/* requires device_hotplug_lock, see add_memory_resource() */
1578int __ref __add_memory(int nid, u64 start, u64 size, mhp_t mhp_flags)
 
 
 
 
 
 
 
1579{
1580	struct resource *res;
1581	int ret;
1582
1583	res = register_memory_resource(start, size, "System RAM");
1584	if (IS_ERR(res))
1585		return PTR_ERR(res);
1586
1587	ret = add_memory_resource(nid, res, mhp_flags);
1588	if (ret < 0)
1589		release_memory_resource(res);
1590	return ret;
1591}
1592
1593int add_memory(int nid, u64 start, u64 size, mhp_t mhp_flags)
 
1594{
1595	int rc;
 
1596
1597	lock_device_hotplug();
1598	rc = __add_memory(nid, start, size, mhp_flags);
1599	unlock_device_hotplug();
 
 
 
 
 
1600
1601	return rc;
1602}
1603EXPORT_SYMBOL_GPL(add_memory);
1604
1605/*
1606 * Add special, driver-managed memory to the system as system RAM. Such
1607 * memory is not exposed via the raw firmware-provided memmap as system
1608 * RAM, instead, it is detected and added by a driver - during cold boot,
1609 * after a reboot, and after kexec.
1610 *
1611 * Reasons why this memory should not be used for the initial memmap of a
1612 * kexec kernel or for placing kexec images:
1613 * - The booting kernel is in charge of determining how this memory will be
1614 *   used (e.g., use persistent memory as system RAM)
1615 * - Coordination with a hypervisor is required before this memory
1616 *   can be used (e.g., inaccessible parts).
1617 *
1618 * For this memory, no entries in /sys/firmware/memmap ("raw firmware-provided
1619 * memory map") are created. Also, the created memory resource is flagged
1620 * with IORESOURCE_SYSRAM_DRIVER_MANAGED, so in-kernel users can special-case
1621 * this memory as well (esp., not place kexec images onto it).
1622 *
1623 * The resource_name (visible via /proc/iomem) has to have the format
1624 * "System RAM ($DRIVER)".
1625 */
1626int add_memory_driver_managed(int nid, u64 start, u64 size,
1627			      const char *resource_name, mhp_t mhp_flags)
1628{
1629	struct resource *res;
1630	int rc;
1631
1632	if (!resource_name ||
1633	    strstr(resource_name, "System RAM (") != resource_name ||
1634	    resource_name[strlen(resource_name) - 1] != ')')
1635		return -EINVAL;
1636
1637	lock_device_hotplug();
1638
1639	res = register_memory_resource(start, size, resource_name);
1640	if (IS_ERR(res)) {
1641		rc = PTR_ERR(res);
1642		goto out_unlock;
1643	}
1644
1645	rc = add_memory_resource(nid, res, mhp_flags);
1646	if (rc < 0)
1647		release_memory_resource(res);
1648
1649out_unlock:
1650	unlock_device_hotplug();
1651	return rc;
1652}
1653EXPORT_SYMBOL_GPL(add_memory_driver_managed);
1654
1655/*
1656 * Platforms should define arch_get_mappable_range() that provides
1657 * maximum possible addressable physical memory range for which the
1658 * linear mapping could be created. The platform returned address
1659 * range must adhere to these following semantics.
1660 *
1661 * - range.start <= range.end
1662 * - Range includes both end points [range.start..range.end]
1663 *
1664 * There is also a fallback definition provided here, allowing the
1665 * entire possible physical address range in case any platform does
1666 * not define arch_get_mappable_range().
1667 */
1668struct range __weak arch_get_mappable_range(void)
1669{
1670	struct range mhp_range = {
1671		.start = 0UL,
1672		.end = -1ULL,
1673	};
1674	return mhp_range;
1675}
1676
1677struct range mhp_get_pluggable_range(bool need_mapping)
1678{
1679	const u64 max_phys = (1ULL << MAX_PHYSMEM_BITS) - 1;
1680	struct range mhp_range;
1681
1682	if (need_mapping) {
1683		mhp_range = arch_get_mappable_range();
1684		if (mhp_range.start > max_phys) {
1685			mhp_range.start = 0;
1686			mhp_range.end = 0;
1687		}
1688		mhp_range.end = min_t(u64, mhp_range.end, max_phys);
1689	} else {
1690		mhp_range.start = 0;
1691		mhp_range.end = max_phys;
1692	}
1693	return mhp_range;
1694}
1695EXPORT_SYMBOL_GPL(mhp_get_pluggable_range);
1696
1697bool mhp_range_allowed(u64 start, u64 size, bool need_mapping)
1698{
1699	struct range mhp_range = mhp_get_pluggable_range(need_mapping);
1700	u64 end = start + size;
1701
1702	if (start < end && start >= mhp_range.start && (end - 1) <= mhp_range.end)
1703		return true;
1704
1705	pr_warn("Hotplug memory [%#llx-%#llx] exceeds maximum addressable range [%#llx-%#llx]\n",
1706		start, end, mhp_range.start, mhp_range.end);
1707	return false;
1708}
1709
1710#ifdef CONFIG_MEMORY_HOTREMOVE
1711/*
1712 * Scan pfn range [start,end) to find movable/migratable pages (LRU pages,
1713 * non-lru movable pages and hugepages). Will skip over most unmovable
1714 * pages (esp., pages that can be skipped when offlining), but bail out on
1715 * definitely unmovable pages.
1716 *
1717 * Returns:
1718 *	0 in case a movable page is found and movable_pfn was updated.
1719 *	-ENOENT in case no movable page was found.
1720 *	-EBUSY in case a definitely unmovable page was found.
1721 */
1722static int scan_movable_pages(unsigned long start, unsigned long end,
1723			      unsigned long *movable_pfn)
1724{
1725	unsigned long pfn;
1726
1727	for (pfn = start; pfn < end; pfn++) {
1728		struct page *page, *head;
1729		unsigned long skip;
 
 
 
 
 
 
1730
1731		if (!pfn_valid(pfn))
1732			continue;
1733		page = pfn_to_page(pfn);
1734		if (PageLRU(page))
1735			goto found;
1736		if (__PageMovable(page))
1737			goto found;
1738
1739		/*
1740		 * PageOffline() pages that are not marked __PageMovable() and
1741		 * have a reference count > 0 (after MEM_GOING_OFFLINE) are
1742		 * definitely unmovable. If their reference count would be 0,
1743		 * they could at least be skipped when offlining memory.
1744		 */
1745		if (PageOffline(page) && page_count(page))
1746			return -EBUSY;
1747
1748		if (!PageHuge(page))
1749			continue;
1750		head = compound_head(page);
1751		/*
1752		 * This test is racy as we hold no reference or lock.  The
1753		 * hugetlb page could have been free'ed and head is no longer
1754		 * a hugetlb page before the following check.  In such unlikely
1755		 * cases false positives and negatives are possible.  Calling
1756		 * code must deal with these scenarios.
1757		 */
1758		if (HPageMigratable(head))
1759			goto found;
1760		skip = compound_nr(head) - (pfn - page_to_pfn(head));
1761		pfn += skip - 1;
1762	}
1763	return -ENOENT;
1764found:
1765	*movable_pfn = pfn;
1766	return 0;
1767}
1768
1769static void do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
 
 
1770{
1771	unsigned long pfn;
1772	struct page *page, *head;
 
 
 
1773	LIST_HEAD(source);
1774	static DEFINE_RATELIMIT_STATE(migrate_rs, DEFAULT_RATELIMIT_INTERVAL,
1775				      DEFAULT_RATELIMIT_BURST);
1776
1777	for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1778		struct folio *folio;
1779		bool isolated;
1780
 
1781		if (!pfn_valid(pfn))
1782			continue;
1783		page = pfn_to_page(pfn);
1784		folio = page_folio(page);
1785		head = &folio->page;
1786
1787		if (PageHuge(page)) {
1788			pfn = page_to_pfn(head) + compound_nr(head) - 1;
1789			isolate_hugetlb(folio, &source);
1790			continue;
1791		} else if (PageTransHuge(page))
1792			pfn = page_to_pfn(head) + thp_nr_pages(page) - 1;
1793
1794		/*
1795		 * HWPoison pages have elevated reference counts so the migration would
1796		 * fail on them. It also doesn't make any sense to migrate them in the
1797		 * first place. Still try to unmap such a page in case it is still mapped
1798		 * (e.g. current hwpoison implementation doesn't unmap KSM pages but keep
1799		 * the unmap as the catch all safety net).
1800		 */
1801		if (PageHWPoison(page)) {
1802			if (WARN_ON(folio_test_lru(folio)))
1803				folio_isolate_lru(folio);
1804			if (folio_mapped(folio))
1805				try_to_unmap(folio, TTU_IGNORE_MLOCK);
1806			continue;
1807		}
1808
1809		if (!get_page_unless_zero(page))
1810			continue;
1811		/*
1812		 * We can skip free pages. And we can deal with pages on
1813		 * LRU and non-lru movable pages.
1814		 */
1815		if (PageLRU(page))
1816			isolated = isolate_lru_page(page);
1817		else
1818			isolated = isolate_movable_page(page, ISOLATE_UNEVICTABLE);
1819		if (isolated) {
1820			list_add_tail(&page->lru, &source);
1821			if (!__PageMovable(page))
1822				inc_node_page_state(page, NR_ISOLATED_ANON +
1823						    page_is_file_lru(page));
1824
1825		} else {
1826			if (__ratelimit(&migrate_rs)) {
1827				pr_warn("failed to isolate pfn %lx\n", pfn);
1828				dump_page(page, "isolation failed");
 
 
 
 
 
 
 
 
 
1829			}
1830		}
1831		put_page(page);
1832	}
1833	if (!list_empty(&source)) {
1834		nodemask_t nmask = node_states[N_MEMORY];
1835		struct migration_target_control mtc = {
1836			.nmask = &nmask,
1837			.gfp_mask = GFP_USER | __GFP_MOVABLE | __GFP_RETRY_MAYFAIL,
1838		};
1839		int ret;
1840
1841		/*
1842		 * We have checked that migration range is on a single zone so
1843		 * we can use the nid of the first page to all the others.
1844		 */
1845		mtc.nid = page_to_nid(list_first_entry(&source, struct page, lru));
1846
1847		/*
1848		 * try to allocate from a different node but reuse this node
1849		 * if there are no other online nodes to be used (e.g. we are
1850		 * offlining a part of the only existing node)
1851		 */
1852		node_clear(mtc.nid, nmask);
1853		if (nodes_empty(nmask))
1854			node_set(mtc.nid, nmask);
1855		ret = migrate_pages(&source, alloc_migration_target, NULL,
1856			(unsigned long)&mtc, MIGRATE_SYNC, MR_MEMORY_HOTPLUG, NULL);
1857		if (ret) {
1858			list_for_each_entry(page, &source, lru) {
1859				if (__ratelimit(&migrate_rs)) {
1860					pr_warn("migrating pfn %lx failed ret:%d\n",
1861						page_to_pfn(page), ret);
1862					dump_page(page, "migration failure");
1863				}
1864			}
1865			putback_movable_pages(&source);
1866		}
 
 
 
 
 
1867	}
 
 
1868}
1869
1870static int __init cmdline_parse_movable_node(char *p)
 
 
 
 
 
1871{
1872	movable_node_enabled = true;
1873	return 0;
1874}
1875early_param("movable_node", cmdline_parse_movable_node);
1876
1877/* check which state of node_states will be changed when offline memory */
1878static void node_states_check_changes_offline(unsigned long nr_pages,
1879		struct zone *zone, struct memory_notify *arg)
1880{
1881	struct pglist_data *pgdat = zone->zone_pgdat;
1882	unsigned long present_pages = 0;
1883	enum zone_type zt;
1884
1885	arg->status_change_nid = NUMA_NO_NODE;
1886	arg->status_change_nid_normal = NUMA_NO_NODE;
1887
1888	/*
1889	 * Check whether node_states[N_NORMAL_MEMORY] will be changed.
1890	 * If the memory to be offline is within the range
1891	 * [0..ZONE_NORMAL], and it is the last present memory there,
1892	 * the zones in that range will become empty after the offlining,
1893	 * thus we can determine that we need to clear the node from
1894	 * node_states[N_NORMAL_MEMORY].
1895	 */
1896	for (zt = 0; zt <= ZONE_NORMAL; zt++)
1897		present_pages += pgdat->node_zones[zt].present_pages;
1898	if (zone_idx(zone) <= ZONE_NORMAL && nr_pages >= present_pages)
1899		arg->status_change_nid_normal = zone_to_nid(zone);
1900
1901	/*
1902	 * We have accounted the pages from [0..ZONE_NORMAL); ZONE_HIGHMEM
1903	 * does not apply as we don't support 32bit.
1904	 * Here we count the possible pages from ZONE_MOVABLE.
1905	 * If after having accounted all the pages, we see that the nr_pages
1906	 * to be offlined is over or equal to the accounted pages,
1907	 * we know that the node will become empty, and so, we can clear
1908	 * it for N_MEMORY as well.
1909	 */
1910	present_pages += pgdat->node_zones[ZONE_MOVABLE].present_pages;
1911
1912	if (nr_pages >= present_pages)
1913		arg->status_change_nid = zone_to_nid(zone);
1914}
1915
1916static void node_states_clear_node(int node, struct memory_notify *arg)
 
 
 
 
 
1917{
1918	if (arg->status_change_nid_normal >= 0)
1919		node_clear_state(node, N_NORMAL_MEMORY);
1920
1921	if (arg->status_change_nid >= 0)
1922		node_clear_state(node, N_MEMORY);
 
 
1923}
1924
1925static int count_system_ram_pages_cb(unsigned long start_pfn,
1926				     unsigned long nr_pages, void *data)
1927{
1928	unsigned long *nr_system_ram_pages = data;
 
1929
1930	*nr_system_ram_pages += nr_pages;
1931	return 0;
 
 
 
1932}
1933
1934/*
1935 * Must be called with mem_hotplug_lock in write mode.
1936 */
1937int __ref offline_pages(unsigned long start_pfn, unsigned long nr_pages,
1938			struct zone *zone, struct memory_group *group)
1939{
1940	const unsigned long end_pfn = start_pfn + nr_pages;
1941	unsigned long pfn, system_ram_pages = 0;
1942	const int node = zone_to_nid(zone);
1943	unsigned long flags;
1944	struct memory_notify arg;
1945	char *reason;
1946	int ret;
1947
1948	/*
1949	 * {on,off}lining is constrained to full memory sections (or more
1950	 * precisely to memory blocks from the user space POV).
1951	 * memmap_on_memory is an exception because it reserves initial part
1952	 * of the physical memory space for vmemmaps. That space is pageblock
1953	 * aligned.
1954	 */
1955	if (WARN_ON_ONCE(!nr_pages || !pageblock_aligned(start_pfn) ||
1956			 !IS_ALIGNED(start_pfn + nr_pages, PAGES_PER_SECTION)))
1957		return -EINVAL;
1958
1959	/*
1960	 * Don't allow to offline memory blocks that contain holes.
1961	 * Consequently, memory blocks with holes can never get onlined
1962	 * via the hotplug path - online_pages() - as hotplugged memory has
1963	 * no holes. This way, we e.g., don't have to worry about marking
1964	 * memory holes PG_reserved, don't need pfn_valid() checks, and can
1965	 * avoid using walk_system_ram_range() later.
1966	 */
1967	walk_system_ram_range(start_pfn, nr_pages, &system_ram_pages,
1968			      count_system_ram_pages_cb);
1969	if (system_ram_pages != nr_pages) {
1970		ret = -EINVAL;
1971		reason = "memory holes";
1972		goto failed_removal;
1973	}
1974
1975	/*
1976	 * We only support offlining of memory blocks managed by a single zone,
1977	 * checked by calling code. This is just a sanity check that we might
1978	 * want to remove in the future.
1979	 */
1980	if (WARN_ON_ONCE(page_zone(pfn_to_page(start_pfn)) != zone ||
1981			 page_zone(pfn_to_page(end_pfn - 1)) != zone)) {
1982		ret = -EINVAL;
1983		reason = "multizone range";
1984		goto failed_removal;
1985	}
1986
1987	/*
1988	 * Disable pcplists so that page isolation cannot race with freeing
1989	 * in a way that pages from isolated pageblock are left on pcplists.
1990	 */
1991	zone_pcp_disable(zone);
1992	lru_cache_disable();
1993
1994	/* set above range as isolated */
1995	ret = start_isolate_page_range(start_pfn, end_pfn,
1996				       MIGRATE_MOVABLE,
1997				       MEMORY_OFFLINE | REPORT_FAILURE,
1998				       GFP_USER | __GFP_MOVABLE | __GFP_RETRY_MAYFAIL);
1999	if (ret) {
2000		reason = "failure to isolate range";
2001		goto failed_removal_pcplists_disabled;
2002	}
2003
2004	arg.start_pfn = start_pfn;
2005	arg.nr_pages = nr_pages;
2006	node_states_check_changes_offline(nr_pages, zone, &arg);
 
 
2007
2008	ret = memory_notify(MEM_GOING_OFFLINE, &arg);
2009	ret = notifier_to_errno(ret);
2010	if (ret) {
2011		reason = "notifier failure";
2012		goto failed_removal_isolated;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2013	}
2014
2015	do {
2016		pfn = start_pfn;
2017		do {
2018			/*
2019			 * Historically we always checked for any signal and
2020			 * can't limit it to fatal signals without eventually
2021			 * breaking user space.
2022			 */
2023			if (signal_pending(current)) {
2024				ret = -EINTR;
2025				reason = "signal backoff";
2026				goto failed_removal_isolated;
2027			}
2028
2029			cond_resched();
2030
2031			ret = scan_movable_pages(pfn, end_pfn, &pfn);
2032			if (!ret) {
2033				/*
2034				 * TODO: fatal migration failures should bail
2035				 * out
2036				 */
2037				do_migrate_range(pfn, end_pfn);
2038			}
2039		} while (!ret);
2040
2041		if (ret != -ENOENT) {
2042			reason = "unmovable page";
2043			goto failed_removal_isolated;
2044		}
2045
2046		/*
2047		 * Dissolve free hugepages in the memory block before doing
2048		 * offlining actually in order to make hugetlbfs's object
2049		 * counting consistent.
2050		 */
2051		ret = dissolve_free_huge_pages(start_pfn, end_pfn);
2052		if (ret) {
2053			reason = "failure to dissolve huge pages";
2054			goto failed_removal_isolated;
2055		}
2056
2057		ret = test_pages_isolated(start_pfn, end_pfn, MEMORY_OFFLINE);
2058
2059	} while (ret);
2060
2061	/* Mark all sections offline and remove free pages from the buddy. */
2062	__offline_isolated_pages(start_pfn, end_pfn);
2063	pr_debug("Offlined Pages %ld\n", nr_pages);
2064
2065	/*
2066	 * The memory sections are marked offline, and the pageblock flags
2067	 * effectively stale; nobody should be touching them. Fixup the number
2068	 * of isolated pageblocks, memory onlining will properly revert this.
2069	 */
2070	spin_lock_irqsave(&zone->lock, flags);
2071	zone->nr_isolate_pageblock -= nr_pages / pageblock_nr_pages;
2072	spin_unlock_irqrestore(&zone->lock, flags);
2073
2074	lru_cache_enable();
2075	zone_pcp_enable(zone);
2076
2077	/* removal success */
2078	adjust_managed_page_count(pfn_to_page(start_pfn), -nr_pages);
2079	adjust_present_page_count(pfn_to_page(start_pfn), group, -nr_pages);
 
2080
2081	/* reinitialise watermarks and update pcp limits */
2082	init_per_zone_wmark_min();
2083
2084	/*
2085	 * Make sure to mark the node as memory-less before rebuilding the zone
2086	 * list. Otherwise this node would still appear in the fallback lists.
2087	 */
2088	node_states_clear_node(node, &arg);
2089	if (!populated_zone(zone)) {
2090		zone_pcp_reset(zone);
2091		build_all_zonelists(NULL);
2092	}
2093
2094	if (arg.status_change_nid >= 0) {
2095		kcompactd_stop(node);
2096		kswapd_stop(node);
2097	}
2098
 
2099	writeback_set_ratelimit();
2100
2101	memory_notify(MEM_OFFLINE, &arg);
2102	remove_pfn_range_from_zone(zone, start_pfn, nr_pages);
2103	return 0;
2104
2105failed_removal_isolated:
 
 
 
2106	/* pushback to free area */
2107	undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
2108	memory_notify(MEM_CANCEL_OFFLINE, &arg);
2109failed_removal_pcplists_disabled:
2110	lru_cache_enable();
2111	zone_pcp_enable(zone);
2112failed_removal:
2113	pr_debug("memory offlining [mem %#010llx-%#010llx] failed due to %s\n",
2114		 (unsigned long long) start_pfn << PAGE_SHIFT,
2115		 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1,
2116		 reason);
2117	return ret;
2118}
2119
2120static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
2121{
2122	int *nid = arg;
2123
2124	*nid = mem->nid;
2125	if (unlikely(mem->state != MEM_OFFLINE)) {
2126		phys_addr_t beginpa, endpa;
2127
2128		beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
2129		endpa = beginpa + memory_block_size_bytes() - 1;
2130		pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n",
2131			&beginpa, &endpa);
2132
2133		return -EBUSY;
2134	}
2135	return 0;
2136}
2137
2138static int count_memory_range_altmaps_cb(struct memory_block *mem, void *arg)
2139{
2140	u64 *num_altmaps = (u64 *)arg;
2141
2142	if (mem->altmap)
2143		*num_altmaps += 1;
2144
2145	return 0;
2146}
2147
2148static int check_cpu_on_node(int nid)
2149{
2150	int cpu;
2151
2152	for_each_present_cpu(cpu) {
2153		if (cpu_to_node(cpu) == nid)
2154			/*
2155			 * the cpu on this node isn't removed, and we can't
2156			 * offline this node.
2157			 */
2158			return -EBUSY;
2159	}
2160
2161	return 0;
2162}
2163
2164static int check_no_memblock_for_node_cb(struct memory_block *mem, void *arg)
2165{
2166	int nid = *(int *)arg;
2167
2168	/*
2169	 * If a memory block belongs to multiple nodes, the stored nid is not
2170	 * reliable. However, such blocks are always online (e.g., cannot get
2171	 * offlined) and, therefore, are still spanned by the node.
2172	 */
2173	return mem->nid == nid ? -EEXIST : 0;
2174}
2175
2176/**
2177 * try_offline_node
2178 * @nid: the node ID
2179 *
2180 * Offline a node if all memory sections and cpus of the node are removed.
2181 *
2182 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
2183 * and online/offline operations before this call.
2184 */
2185void try_offline_node(int nid)
2186{
2187	int rc;
2188
2189	/*
2190	 * If the node still spans pages (especially ZONE_DEVICE), don't
2191	 * offline it. A node spans memory after move_pfn_range_to_zone(),
2192	 * e.g., after the memory block was onlined.
2193	 */
2194	if (node_spanned_pages(nid))
2195		return;
2196
2197	/*
2198	 * Especially offline memory blocks might not be spanned by the
2199	 * node. They will get spanned by the node once they get onlined.
2200	 * However, they link to the node in sysfs and can get onlined later.
2201	 */
2202	rc = for_each_memory_block(&nid, check_no_memblock_for_node_cb);
2203	if (rc)
2204		return;
2205
2206	if (check_cpu_on_node(nid))
2207		return;
2208
2209	/*
2210	 * all memory/cpu of this node are removed, we can offline this
2211	 * node now.
2212	 */
2213	node_set_offline(nid);
2214	unregister_one_node(nid);
2215}
2216EXPORT_SYMBOL(try_offline_node);
2217
2218static int memory_blocks_have_altmaps(u64 start, u64 size)
2219{
2220	u64 num_memblocks = size / memory_block_size_bytes();
2221	u64 num_altmaps = 0;
2222
2223	if (!mhp_memmap_on_memory())
2224		return 0;
2225
2226	walk_memory_blocks(start, size, &num_altmaps,
2227			   count_memory_range_altmaps_cb);
2228
2229	if (num_altmaps == 0)
2230		return 0;
2231
2232	if (WARN_ON_ONCE(num_memblocks != num_altmaps))
2233		return -EINVAL;
2234
2235	return 1;
2236}
2237
2238static int __ref try_remove_memory(u64 start, u64 size)
2239{
2240	int rc, nid = NUMA_NO_NODE;
2241
2242	BUG_ON(check_hotplug_memory_range(start, size));
2243
2244	/*
2245	 * All memory blocks must be offlined before removing memory.  Check
2246	 * whether all memory blocks in question are offline and return error
2247	 * if this is not the case.
2248	 *
2249	 * While at it, determine the nid. Note that if we'd have mixed nodes,
2250	 * we'd only try to offline the last determined one -- which is good
2251	 * enough for the cases we care about.
2252	 */
2253	rc = walk_memory_blocks(start, size, &nid, check_memblock_offlined_cb);
2254	if (rc)
2255		return rc;
2256
2257	/* remove memmap entry */
2258	firmware_map_remove(start, start + size, "System RAM");
2259
2260	mem_hotplug_begin();
2261
2262	rc = memory_blocks_have_altmaps(start, size);
2263	if (rc < 0) {
2264		mem_hotplug_done();
2265		return rc;
2266	} else if (!rc) {
2267		/*
2268		 * Memory block device removal under the device_hotplug_lock is
2269		 * a barrier against racing online attempts.
2270		 * No altmaps present, do the removal directly
2271		 */
2272		remove_memory_block_devices(start, size);
2273		arch_remove_memory(start, size, NULL);
2274	} else {
2275		/* all memblocks in the range have altmaps */
2276		remove_memory_blocks_and_altmaps(start, size);
2277	}
2278
2279	if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK)) {
2280		memblock_phys_free(start, size);
2281		memblock_remove(start, size);
2282	}
2283
2284	release_mem_region_adjustable(start, size);
2285
2286	if (nid != NUMA_NO_NODE)
2287		try_offline_node(nid);
2288
2289	mem_hotplug_done();
2290	return 0;
2291}
2292
2293/**
2294 * __remove_memory - Remove memory if every memory block is offline
2295 * @start: physical address of the region to remove
2296 * @size: size of the region to remove
2297 *
2298 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
2299 * and online/offline operations before this call, as required by
2300 * try_offline_node().
2301 */
2302void __remove_memory(u64 start, u64 size)
2303{
2304
2305	/*
2306	 * trigger BUG() if some memory is not offlined prior to calling this
2307	 * function
2308	 */
2309	if (try_remove_memory(start, size))
2310		BUG();
2311}
2312
2313/*
2314 * Remove memory if every memory block is offline, otherwise return -EBUSY is
2315 * some memory is not offline
2316 */
2317int remove_memory(u64 start, u64 size)
2318{
2319	int rc;
2320
2321	lock_device_hotplug();
2322	rc = try_remove_memory(start, size);
2323	unlock_device_hotplug();
2324
2325	return rc;
2326}
 
2327EXPORT_SYMBOL_GPL(remove_memory);
2328
2329static int try_offline_memory_block(struct memory_block *mem, void *arg)
2330{
2331	uint8_t online_type = MMOP_ONLINE_KERNEL;
2332	uint8_t **online_types = arg;
2333	struct page *page;
2334	int rc;
2335
2336	/*
2337	 * Sense the online_type via the zone of the memory block. Offlining
2338	 * with multiple zones within one memory block will be rejected
2339	 * by offlining code ... so we don't care about that.
2340	 */
2341	page = pfn_to_online_page(section_nr_to_pfn(mem->start_section_nr));
2342	if (page && zone_idx(page_zone(page)) == ZONE_MOVABLE)
2343		online_type = MMOP_ONLINE_MOVABLE;
2344
2345	rc = device_offline(&mem->dev);
2346	/*
2347	 * Default is MMOP_OFFLINE - change it only if offlining succeeded,
2348	 * so try_reonline_memory_block() can do the right thing.
2349	 */
2350	if (!rc)
2351		**online_types = online_type;
2352
2353	(*online_types)++;
2354	/* Ignore if already offline. */
2355	return rc < 0 ? rc : 0;
2356}
2357
2358static int try_reonline_memory_block(struct memory_block *mem, void *arg)
2359{
2360	uint8_t **online_types = arg;
2361	int rc;
2362
2363	if (**online_types != MMOP_OFFLINE) {
2364		mem->online_type = **online_types;
2365		rc = device_online(&mem->dev);
2366		if (rc < 0)
2367			pr_warn("%s: Failed to re-online memory: %d",
2368				__func__, rc);
2369	}
2370
2371	/* Continue processing all remaining memory blocks. */
2372	(*online_types)++;
2373	return 0;
2374}
2375
2376/*
2377 * Try to offline and remove memory. Might take a long time to finish in case
2378 * memory is still in use. Primarily useful for memory devices that logically
2379 * unplugged all memory (so it's no longer in use) and want to offline + remove
2380 * that memory.
2381 */
2382int offline_and_remove_memory(u64 start, u64 size)
2383{
2384	const unsigned long mb_count = size / memory_block_size_bytes();
2385	uint8_t *online_types, *tmp;
2386	int rc;
2387
2388	if (!IS_ALIGNED(start, memory_block_size_bytes()) ||
2389	    !IS_ALIGNED(size, memory_block_size_bytes()) || !size)
2390		return -EINVAL;
2391
2392	/*
2393	 * We'll remember the old online type of each memory block, so we can
2394	 * try to revert whatever we did when offlining one memory block fails
2395	 * after offlining some others succeeded.
2396	 */
2397	online_types = kmalloc_array(mb_count, sizeof(*online_types),
2398				     GFP_KERNEL);
2399	if (!online_types)
2400		return -ENOMEM;
2401	/*
2402	 * Initialize all states to MMOP_OFFLINE, so when we abort processing in
2403	 * try_offline_memory_block(), we'll skip all unprocessed blocks in
2404	 * try_reonline_memory_block().
2405	 */
2406	memset(online_types, MMOP_OFFLINE, mb_count);
2407
2408	lock_device_hotplug();
2409
2410	tmp = online_types;
2411	rc = walk_memory_blocks(start, size, &tmp, try_offline_memory_block);
2412
2413	/*
2414	 * In case we succeeded to offline all memory, remove it.
2415	 * This cannot fail as it cannot get onlined in the meantime.
2416	 */
2417	if (!rc) {
2418		rc = try_remove_memory(start, size);
2419		if (rc)
2420			pr_err("%s: Failed to remove memory: %d", __func__, rc);
2421	}
2422
2423	/*
2424	 * Rollback what we did. While memory onlining might theoretically fail
2425	 * (nacked by a notifier), it barely ever happens.
2426	 */
2427	if (rc) {
2428		tmp = online_types;
2429		walk_memory_blocks(start, size, &tmp,
2430				   try_reonline_memory_block);
2431	}
2432	unlock_device_hotplug();
2433
2434	kfree(online_types);
2435	return rc;
2436}
2437EXPORT_SYMBOL_GPL(offline_and_remove_memory);
2438#endif /* CONFIG_MEMORY_HOTREMOVE */