mm/memory_hotplug.c

   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/compiler.h>
  13 #include <linux/export.h>
  14 #include <linux/pagevec.h>
  15 #include <linux/writeback.h>
  16 #include <linux/slab.h>
  17 #include <linux/sysctl.h>
  18 #include <linux/cpu.h>
  19 #include <linux/memory.h>
  20 #include <linux/memremap.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 #include <linux/stop_machine.h>
  33 #include <linux/hugetlb.h>
  34 #include <linux/memblock.h>
  35 #include <linux/bootmem.h>
  36 #include <linux/compaction.h>
  37
  38 #include <asm/tlbflush.h>
  39
  40 #include "internal.h"
  41
  42 /*
  43  * online_page_callback contains pointer to current page onlining function.
  44  * Initially it is generic_online_page(). If it is required it could be
  45  * changed by calling set_online_page_callback() for callback registration
  46  * and restore_online_page_callback() for generic callback restore.
  47  */
  48
  49 static void generic_online_page(struct page *page);
  50
  51 static online_page_callback_t online_page_callback = generic_online_page;
  52 static DEFINE_MUTEX(online_page_callback_lock);
  53
  54 /* The same as the cpu_hotplug lock, but for memory hotplug. */
  55 static struct {
  56         struct task_struct *active_writer;
  57         struct mutex lock; /* Synchronizes accesses to refcount, */
  58         /*
  59          * Also blocks the new readers during
  60          * an ongoing mem hotplug operation.
  61          */
  62         int refcount;
  63
  64 #ifdef CONFIG_DEBUG_LOCK_ALLOC
  65         struct lockdep_map dep_map;
  66 #endif
  67 } mem_hotplug = {
  68         .active_writer = NULL,
  69         .lock = __MUTEX_INITIALIZER(mem_hotplug.lock),
  70         .refcount = 0,
  71 #ifdef CONFIG_DEBUG_LOCK_ALLOC
  72         .dep_map = {.name = "mem_hotplug.lock" },
  73 #endif
  74 };
  75
  76 /* Lockdep annotations for get/put_online_mems() and mem_hotplug_begin/end() */
  77 #define memhp_lock_acquire_read() lock_map_acquire_read(&mem_hotplug.dep_map)
  78 #define memhp_lock_acquire()      lock_map_acquire(&mem_hotplug.dep_map)
  79 #define memhp_lock_release()      lock_map_release(&mem_hotplug.dep_map)
  80
  81 bool memhp_auto_online;
  82 EXPORT_SYMBOL_GPL(memhp_auto_online);
  83
  84 void get_online_mems(void)
  85 {
  86         might_sleep();
  87         if (mem_hotplug.active_writer == current)
  88                 return;
  89         memhp_lock_acquire_read();
  90         mutex_lock(&mem_hotplug.lock);
  91         mem_hotplug.refcount++;
  92         mutex_unlock(&mem_hotplug.lock);
  93
  94 }
  95
  96 void put_online_mems(void)
  97 {
  98         if (mem_hotplug.active_writer == current)
  99                 return;
 100         mutex_lock(&mem_hotplug.lock);
 101
 102         if (WARN_ON(!mem_hotplug.refcount))
 103                 mem_hotplug.refcount++; /* try to fix things up */
 104
 105         if (!--mem_hotplug.refcount && unlikely(mem_hotplug.active_writer))
 106                 wake_up_process(mem_hotplug.active_writer);
 107         mutex_unlock(&mem_hotplug.lock);
 108         memhp_lock_release();
 109
 110 }
 111
 112 void mem_hotplug_begin(void)
 113 {
 114         mem_hotplug.active_writer = current;
 115
 116         memhp_lock_acquire();
 117         for (;;) {
 118                 mutex_lock(&mem_hotplug.lock);
 119                 if (likely(!mem_hotplug.refcount))
 120                         break;
 121                 __set_current_state(TASK_UNINTERRUPTIBLE);
 122                 mutex_unlock(&mem_hotplug.lock);
 123                 schedule();
 124         }
 125 }
 126
 127 void mem_hotplug_done(void)
 128 {
 129         mem_hotplug.active_writer = NULL;
 130         mutex_unlock(&mem_hotplug.lock);
 131         memhp_lock_release();
 132 }
 133
 134 /* add this memory to iomem resource */
 135 static struct resource *register_memory_resource(u64 start, u64 size)
 136 {
 137         struct resource *res;
 138         res = kzalloc(sizeof(struct resource), GFP_KERNEL);
 139         if (!res)
 140                 return ERR_PTR(-ENOMEM);
 141
 142         res->name = "System RAM";
 143         res->start = start;
 144         res->end = start + size - 1;
 145         res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
 146         if (request_resource(&iomem_resource, res) < 0) {
 147                 pr_debug("System RAM resource %pR cannot be added\n", res);
 148                 kfree(res);
 149                 return ERR_PTR(-EEXIST);
 150         }
 151         return res;
 152 }
 153
 154 static void release_memory_resource(struct resource *res)
 155 {
 156         if (!res)
 157                 return;
 158         release_resource(res);
 159         kfree(res);
 160         return;
 161 }
 162
 163 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
 164 void get_page_bootmem(unsigned long info,  struct page *page,
 165                       unsigned long type)
 166 {
 167         page->lru.next = (struct list_head *) type;
 168         SetPagePrivate(page);
 169         set_page_private(page, info);
 170         page_ref_inc(page);
 171 }
 172
 173 void put_page_bootmem(struct page *page)
 174 {
 175         unsigned long type;
 176
 177         type = (unsigned long) page->lru.next;
 178         BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE ||
 179                type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE);
 180
 181         if (page_ref_dec_return(page) == 1) {
 182                 ClearPagePrivate(page);
 183                 set_page_private(page, 0);
 184                 INIT_LIST_HEAD(&page->lru);
 185                 free_reserved_page(page);
 186         }
 187 }
 188
 189 #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
 190 #ifndef CONFIG_SPARSEMEM_VMEMMAP
 191 static void register_page_bootmem_info_section(unsigned long start_pfn)
 192 {
 193         unsigned long *usemap, mapsize, section_nr, i;
 194         struct mem_section *ms;
 195         struct page *page, *memmap;
 196
 197         section_nr = pfn_to_section_nr(start_pfn);
 198         ms = __nr_to_section(section_nr);
 199
 200         /* Get section's memmap address */
 201         memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
 202
 203         /*
 204          * Get page for the memmap's phys address
 205          * XXX: need more consideration for sparse_vmemmap...
 206          */
 207         page = virt_to_page(memmap);
 208         mapsize = sizeof(struct page) * PAGES_PER_SECTION;
 209         mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
 210
 211         /* remember memmap's page */
 212         for (i = 0; i < mapsize; i++, page++)
 213                 get_page_bootmem(section_nr, page, SECTION_INFO);
 214
 215         usemap = __nr_to_section(section_nr)->pageblock_flags;
 216         page = virt_to_page(usemap);
 217
 218         mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
 219
 220         for (i = 0; i < mapsize; i++, page++)
 221                 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
 222
 223 }
 224 #else /* CONFIG_SPARSEMEM_VMEMMAP */
 225 static void register_page_bootmem_info_section(unsigned long start_pfn)
 226 {
 227         unsigned long *usemap, mapsize, section_nr, i;
 228         struct mem_section *ms;
 229         struct page *page, *memmap;
 230
 231         if (!pfn_valid(start_pfn))
 232                 return;
 233
 234         section_nr = pfn_to_section_nr(start_pfn);
 235         ms = __nr_to_section(section_nr);
 236
 237         memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
 238
 239         register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION);
 240
 241         usemap = __nr_to_section(section_nr)->pageblock_flags;
 242         page = virt_to_page(usemap);
 243
 244         mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
 245
 246         for (i = 0; i < mapsize; i++, page++)
 247                 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
 248 }
 249 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
 250
 251 void register_page_bootmem_info_node(struct pglist_data *pgdat)
 252 {
 253         unsigned long i, pfn, end_pfn, nr_pages;
 254         int node = pgdat->node_id;
 255         struct page *page;
 256         struct zone *zone;
 257
 258         nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
 259         page = virt_to_page(pgdat);
 260
 261         for (i = 0; i < nr_pages; i++, page++)
 262                 get_page_bootmem(node, page, NODE_INFO);
 263
 264         zone = &pgdat->node_zones[0];
 265         for (; zone < pgdat->node_zones + MAX_NR_ZONES - 1; zone++) {
 266                 if (zone_is_initialized(zone)) {
 267                         nr_pages = zone->wait_table_hash_nr_entries
 268                                 * sizeof(wait_queue_head_t);
 269                         nr_pages = PAGE_ALIGN(nr_pages) >> PAGE_SHIFT;
 270                         page = virt_to_page(zone->wait_table);
 271
 272                         for (i = 0; i < nr_pages; i++, page++)
 273                                 get_page_bootmem(node, page, NODE_INFO);
 274                 }
 275         }
 276
 277         pfn = pgdat->node_start_pfn;
 278         end_pfn = pgdat_end_pfn(pgdat);
 279
 280         /* register section info */
 281         for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
 282                 /*
 283                  * Some platforms can assign the same pfn to multiple nodes - on
 284                  * node0 as well as nodeN.  To avoid registering a pfn against
 285                  * multiple nodes we check that this pfn does not already
 286                  * reside in some other nodes.
 287                  */
 288                 if (pfn_valid(pfn) && (pfn_to_nid(pfn) == node))
 289                         register_page_bootmem_info_section(pfn);
 290         }
 291 }
 292 #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
 293
 294 static void __meminit grow_zone_span(struct zone *zone, unsigned long start_pfn,
 295                                      unsigned long end_pfn)
 296 {
 297         unsigned long old_zone_end_pfn;
 298
 299         zone_span_writelock(zone);
 300
 301         old_zone_end_pfn = zone_end_pfn(zone);
 302         if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
 303                 zone->zone_start_pfn = start_pfn;
 304
 305         zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
 306                                 zone->zone_start_pfn;
 307
 308         zone_span_writeunlock(zone);
 309 }
 310
 311 static void resize_zone(struct zone *zone, unsigned long start_pfn,
 312                 unsigned long end_pfn)
 313 {
 314         zone_span_writelock(zone);
 315
 316         if (end_pfn - start_pfn) {
 317                 zone->zone_start_pfn = start_pfn;
 318                 zone->spanned_pages = end_pfn - start_pfn;
 319         } else {
 320                 /*
 321                  * make it consist as free_area_init_core(),
 322                  * if spanned_pages = 0, then keep start_pfn = 0
 323                  */
 324                 zone->zone_start_pfn = 0;
 325                 zone->spanned_pages = 0;
 326         }
 327
 328         zone_span_writeunlock(zone);
 329 }
 330
 331 static void fix_zone_id(struct zone *zone, unsigned long start_pfn,
 332                 unsigned long end_pfn)
 333 {
 334         enum zone_type zid = zone_idx(zone);
 335         int nid = zone->zone_pgdat->node_id;
 336         unsigned long pfn;
 337
 338         for (pfn = start_pfn; pfn < end_pfn; pfn++)
 339                 set_page_links(pfn_to_page(pfn), zid, nid, pfn);
 340 }
 341
 342 /* Can fail with -ENOMEM from allocating a wait table with vmalloc() or
 343  * alloc_bootmem_node_nopanic()/memblock_virt_alloc_node_nopanic() */
 344 static int __ref ensure_zone_is_initialized(struct zone *zone,
 345                         unsigned long start_pfn, unsigned long num_pages)
 346 {
 347         if (!zone_is_initialized(zone))
 348                 return init_currently_empty_zone(zone, start_pfn, num_pages);
 349
 350         return 0;
 351 }
 352
 353 static int __meminit move_pfn_range_left(struct zone *z1, struct zone *z2,
 354                 unsigned long start_pfn, unsigned long end_pfn)
 355 {
 356         int ret;
 357         unsigned long flags;
 358         unsigned long z1_start_pfn;
 359
 360         ret = ensure_zone_is_initialized(z1, start_pfn, end_pfn - start_pfn);
 361         if (ret)
 362                 return ret;
 363
 364         pgdat_resize_lock(z1->zone_pgdat, &flags);
 365
 366         /* can't move pfns which are higher than @z2 */
 367         if (end_pfn > zone_end_pfn(z2))
 368                 goto out_fail;
 369         /* the move out part must be at the left most of @z2 */
 370         if (start_pfn > z2->zone_start_pfn)
 371                 goto out_fail;
 372         /* must included/overlap */
 373         if (end_pfn <= z2->zone_start_pfn)
 374                 goto out_fail;
 375
 376         /* use start_pfn for z1's start_pfn if z1 is empty */
 377         if (!zone_is_empty(z1))
 378                 z1_start_pfn = z1->zone_start_pfn;
 379         else
 380                 z1_start_pfn = start_pfn;
 381
 382         resize_zone(z1, z1_start_pfn, end_pfn);
 383         resize_zone(z2, end_pfn, zone_end_pfn(z2));
 384
 385         pgdat_resize_unlock(z1->zone_pgdat, &flags);
 386
 387         fix_zone_id(z1, start_pfn, end_pfn);
 388
 389         return 0;
 390 out_fail:
 391         pgdat_resize_unlock(z1->zone_pgdat, &flags);
 392         return -1;
 393 }
 394
 395 static int __meminit move_pfn_range_right(struct zone *z1, struct zone *z2,
 396                 unsigned long start_pfn, unsigned long end_pfn)
 397 {
 398         int ret;
 399         unsigned long flags;
 400         unsigned long z2_end_pfn;
 401
 402         ret = ensure_zone_is_initialized(z2, start_pfn, end_pfn - start_pfn);
 403         if (ret)
 404                 return ret;
 405
 406         pgdat_resize_lock(z1->zone_pgdat, &flags);
 407
 408         /* can't move pfns which are lower than @z1 */
 409         if (z1->zone_start_pfn > start_pfn)
 410                 goto out_fail;
 411         /* the move out part mast at the right most of @z1 */
 412         if (zone_end_pfn(z1) >  end_pfn)
 413                 goto out_fail;
 414         /* must included/overlap */
 415         if (start_pfn >= zone_end_pfn(z1))
 416                 goto out_fail;
 417
 418         /* use end_pfn for z2's end_pfn if z2 is empty */
 419         if (!zone_is_empty(z2))
 420                 z2_end_pfn = zone_end_pfn(z2);
 421         else
 422                 z2_end_pfn = end_pfn;
 423
 424         resize_zone(z1, z1->zone_start_pfn, start_pfn);
 425         resize_zone(z2, start_pfn, z2_end_pfn);
 426
 427         pgdat_resize_unlock(z1->zone_pgdat, &flags);
 428
 429         fix_zone_id(z2, start_pfn, end_pfn);
 430
 431         return 0;
 432 out_fail:
 433         pgdat_resize_unlock(z1->zone_pgdat, &flags);
 434         return -1;
 435 }
 436
 437 static void __meminit grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn,
 438                                       unsigned long end_pfn)
 439 {
 440         unsigned long old_pgdat_end_pfn = pgdat_end_pfn(pgdat);
 441
 442         if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
 443                 pgdat->node_start_pfn = start_pfn;
 444
 445         pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
 446                                         pgdat->node_start_pfn;
 447 }
 448
 449 static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn)
 450 {
 451         struct pglist_data *pgdat = zone->zone_pgdat;
 452         int nr_pages = PAGES_PER_SECTION;
 453         int nid = pgdat->node_id;
 454         int zone_type;
 455         unsigned long flags, pfn;
 456         int ret;
 457
 458         zone_type = zone - pgdat->node_zones;
 459         ret = ensure_zone_is_initialized(zone, phys_start_pfn, nr_pages);
 460         if (ret)
 461                 return ret;
 462
 463         pgdat_resize_lock(zone->zone_pgdat, &flags);
 464         grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages);
 465         grow_pgdat_span(zone->zone_pgdat, phys_start_pfn,
 466                         phys_start_pfn + nr_pages);
 467         pgdat_resize_unlock(zone->zone_pgdat, &flags);
 468         memmap_init_zone(nr_pages, nid, zone_type,
 469                          phys_start_pfn, MEMMAP_HOTPLUG);
 470
 471         /* online_page_range is called later and expects pages reserved */
 472         for (pfn = phys_start_pfn; pfn < phys_start_pfn + nr_pages; pfn++) {
 473                 if (!pfn_valid(pfn))
 474                         continue;
 475
 476                 SetPageReserved(pfn_to_page(pfn));
 477         }
 478         return 0;
 479 }
 480
 481 static int __meminit __add_section(int nid, struct zone *zone,
 482                                         unsigned long phys_start_pfn)
 483 {
 484         int ret;
 485
 486         if (pfn_valid(phys_start_pfn))
 487                 return -EEXIST;
 488
 489         ret = sparse_add_one_section(zone, phys_start_pfn);
 490
 491         if (ret < 0)
 492                 return ret;
 493
 494         ret = __add_zone(zone, phys_start_pfn);
 495
 496         if (ret < 0)
 497                 return ret;
 498
 499         return register_new_memory(nid, __pfn_to_section(phys_start_pfn));
 500 }
 501
 502 /*
 503  * Reasonably generic function for adding memory.  It is
 504  * expected that archs that support memory hotplug will
 505  * call this function after deciding the zone to which to
 506  * add the new pages.
 507  */
 508 int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn,
 509                         unsigned long nr_pages)
 510 {
 511         unsigned long i;
 512         int err = 0;
 513         int start_sec, end_sec;
 514         struct vmem_altmap *altmap;
 515
 516         clear_zone_contiguous(zone);
 517
 518         /* during initialize mem_map, align hot-added range to section */
 519         start_sec = pfn_to_section_nr(phys_start_pfn);
 520         end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
 521
 522         altmap = to_vmem_altmap((unsigned long) pfn_to_page(phys_start_pfn));
 523         if (altmap) {
 524                 /*
 525                  * Validate altmap is within bounds of the total request
 526                  */
 527                 if (altmap->base_pfn != phys_start_pfn
 528                                 || vmem_altmap_offset(altmap) > nr_pages) {
 529                         pr_warn_once("memory add fail, invalid altmap\n");
 530                         err = -EINVAL;
 531                         goto out;
 532                 }
 533                 altmap->alloc = 0;
 534         }
 535
 536         for (i = start_sec; i <= end_sec; i++) {
 537                 err = __add_section(nid, zone, section_nr_to_pfn(i));
 538
 539                 /*
 540                  * EEXIST is finally dealt with by ioresource collision
 541                  * check. see add_memory() => register_memory_resource()
 542                  * Warning will be printed if there is collision.
 543                  */
 544                 if (err && (err != -EEXIST))
 545                         break;
 546                 err = 0;
 547         }
 548         vmemmap_populate_print_last();
 549 out:
 550         set_zone_contiguous(zone);
 551         return err;
 552 }
 553 EXPORT_SYMBOL_GPL(__add_pages);
 554
 555 #ifdef CONFIG_MEMORY_HOTREMOVE
 556 /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
 557 static int find_smallest_section_pfn(int nid, struct zone *zone,
 558                                      unsigned long start_pfn,
 559                                      unsigned long end_pfn)
 560 {
 561         struct mem_section *ms;
 562
 563         for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) {
 564                 ms = __pfn_to_section(start_pfn);
 565
 566                 if (unlikely(!valid_section(ms)))
 567                         continue;
 568
 569                 if (unlikely(pfn_to_nid(start_pfn) != nid))
 570                         continue;
 571
 572                 if (zone && zone != page_zone(pfn_to_page(start_pfn)))
 573                         continue;
 574
 575                 return start_pfn;
 576         }
 577
 578         return 0;
 579 }
 580
 581 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
 582 static int find_biggest_section_pfn(int nid, struct zone *zone,
 583                                     unsigned long start_pfn,
 584                                     unsigned long end_pfn)
 585 {
 586         struct mem_section *ms;
 587         unsigned long pfn;
 588
 589         /* pfn is the end pfn of a memory section. */
 590         pfn = end_pfn - 1;
 591         for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) {
 592                 ms = __pfn_to_section(pfn);
 593
 594                 if (unlikely(!valid_section(ms)))
 595                         continue;
 596
 597                 if (unlikely(pfn_to_nid(pfn) != nid))
 598                         continue;
 599
 600                 if (zone && zone != page_zone(pfn_to_page(pfn)))
 601                         continue;
 602
 603                 return pfn;
 604         }
 605
 606         return 0;
 607 }
 608
 609 static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
 610                              unsigned long end_pfn)
 611 {
 612         unsigned long zone_start_pfn = zone->zone_start_pfn;
 613         unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */
 614         unsigned long zone_end_pfn = z;
 615         unsigned long pfn;
 616         struct mem_section *ms;
 617         int nid = zone_to_nid(zone);
 618
 619         zone_span_writelock(zone);
 620         if (zone_start_pfn == start_pfn) {
 621                 /*
 622                  * If the section is smallest section in the zone, it need
 623                  * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
 624                  * In this case, we find second smallest valid mem_section
 625                  * for shrinking zone.
 626                  */
 627                 pfn = find_smallest_section_pfn(nid, zone, end_pfn,
 628                                                 zone_end_pfn);
 629                 if (pfn) {
 630                         zone->zone_start_pfn = pfn;
 631                         zone->spanned_pages = zone_end_pfn - pfn;
 632                 }
 633         } else if (zone_end_pfn == end_pfn) {
 634                 /*
 635                  * If the section is biggest section in the zone, it need
 636                  * shrink zone->spanned_pages.
 637                  * In this case, we find second biggest valid mem_section for
 638                  * shrinking zone.
 639                  */
 640                 pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn,
 641                                                start_pfn);
 642                 if (pfn)
 643                         zone->spanned_pages = pfn - zone_start_pfn + 1;
 644         }
 645
 646         /*
 647          * The section is not biggest or smallest mem_section in the zone, it
 648          * only creates a hole in the zone. So in this case, we need not
 649          * change the zone. But perhaps, the zone has only hole data. Thus
 650          * it check the zone has only hole or not.
 651          */
 652         pfn = zone_start_pfn;
 653         for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) {
 654                 ms = __pfn_to_section(pfn);
 655
 656                 if (unlikely(!valid_section(ms)))
 657                         continue;
 658
 659                 if (page_zone(pfn_to_page(pfn)) != zone)
 660                         continue;
 661
 662                  /* If the section is current section, it continues the loop */
 663                 if (start_pfn == pfn)
 664                         continue;
 665
 666                 /* If we find valid section, we have nothing to do */
 667                 zone_span_writeunlock(zone);
 668                 return;
 669         }
 670
 671         /* The zone has no valid section */
 672         zone->zone_start_pfn = 0;
 673         zone->spanned_pages = 0;
 674         zone_span_writeunlock(zone);
 675 }
 676
 677 static void shrink_pgdat_span(struct pglist_data *pgdat,
 678                               unsigned long start_pfn, unsigned long end_pfn)
 679 {
 680         unsigned long pgdat_start_pfn = pgdat->node_start_pfn;
 681         unsigned long p = pgdat_end_pfn(pgdat); /* pgdat_end_pfn namespace clash */
 682         unsigned long pgdat_end_pfn = p;
 683         unsigned long pfn;
 684         struct mem_section *ms;
 685         int nid = pgdat->node_id;
 686
 687         if (pgdat_start_pfn == start_pfn) {
 688                 /*
 689                  * If the section is smallest section in the pgdat, it need
 690                  * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages.
 691                  * In this case, we find second smallest valid mem_section
 692                  * for shrinking zone.
 693                  */
 694                 pfn = find_smallest_section_pfn(nid, NULL, end_pfn,
 695                                                 pgdat_end_pfn);
 696                 if (pfn) {
 697                         pgdat->node_start_pfn = pfn;
 698                         pgdat->node_spanned_pages = pgdat_end_pfn - pfn;
 699                 }
 700         } else if (pgdat_end_pfn == end_pfn) {
 701                 /*
 702                  * If the section is biggest section in the pgdat, it need
 703                  * shrink pgdat->node_spanned_pages.
 704                  * In this case, we find second biggest valid mem_section for
 705                  * shrinking zone.
 706                  */
 707                 pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn,
 708                                                start_pfn);
 709                 if (pfn)
 710                         pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1;
 711         }
 712
 713         /*
 714          * If the section is not biggest or smallest mem_section in the pgdat,
 715          * it only creates a hole in the pgdat. So in this case, we need not
 716          * change the pgdat.
 717          * But perhaps, the pgdat has only hole data. Thus it check the pgdat
 718          * has only hole or not.
 719          */
 720         pfn = pgdat_start_pfn;
 721         for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) {
 722                 ms = __pfn_to_section(pfn);
 723
 724                 if (unlikely(!valid_section(ms)))
 725                         continue;
 726
 727                 if (pfn_to_nid(pfn) != nid)
 728                         continue;
 729
 730                  /* If the section is current section, it continues the loop */
 731                 if (start_pfn == pfn)
 732                         continue;
 733
 734                 /* If we find valid section, we have nothing to do */
 735                 return;
 736         }
 737
 738         /* The pgdat has no valid section */
 739         pgdat->node_start_pfn = 0;
 740         pgdat->node_spanned_pages = 0;
 741 }
 742
 743 static void __remove_zone(struct zone *zone, unsigned long start_pfn)
 744 {
 745         struct pglist_data *pgdat = zone->zone_pgdat;
 746         int nr_pages = PAGES_PER_SECTION;
 747         int zone_type;
 748         unsigned long flags;
 749
 750         zone_type = zone - pgdat->node_zones;
 751
 752         pgdat_resize_lock(zone->zone_pgdat, &flags);
 753         shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
 754         shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages);
 755         pgdat_resize_unlock(zone->zone_pgdat, &flags);
 756 }
 757
 758 static int __remove_section(struct zone *zone, struct mem_section *ms,
 759                 unsigned long map_offset)
 760 {
 761         unsigned long start_pfn;
 762         int scn_nr;
 763         int ret = -EINVAL;
 764
 765         if (!valid_section(ms))
 766                 return ret;
 767
 768         ret = unregister_memory_section(ms);
 769         if (ret)
 770                 return ret;
 771
 772         scn_nr = __section_nr(ms);
 773         start_pfn = section_nr_to_pfn(scn_nr);
 774         __remove_zone(zone, start_pfn);
 775
 776         sparse_remove_one_section(zone, ms, map_offset);
 777         return 0;
 778 }
 779
 780 /**
 781  * __remove_pages() - remove sections of pages from a zone
 782  * @zone: zone from which pages need to be removed
 783  * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
 784  * @nr_pages: number of pages to remove (must be multiple of section size)
 785  *
 786  * Generic helper function to remove section mappings and sysfs entries
 787  * for the section of the memory we are removing. Caller needs to make
 788  * sure that pages are marked reserved and zones are adjust properly by
 789  * calling offline_pages().
 790  */
 791 int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
 792                  unsigned long nr_pages)
 793 {
 794         unsigned long i;
 795         unsigned long map_offset = 0;
 796         int sections_to_remove, ret = 0;
 797
 798         /* In the ZONE_DEVICE case device driver owns the memory region */
 799         if (is_dev_zone(zone)) {
 800                 struct page *page = pfn_to_page(phys_start_pfn);
 801                 struct vmem_altmap *altmap;
 802
 803                 altmap = to_vmem_altmap((unsigned long) page);
 804                 if (altmap)
 805                         map_offset = vmem_altmap_offset(altmap);
 806         } else {
 807                 resource_size_t start, size;
 808
 809                 start = phys_start_pfn << PAGE_SHIFT;
 810                 size = nr_pages * PAGE_SIZE;
 811
 812                 ret = release_mem_region_adjustable(&iomem_resource, start,
 813                                         size);
 814                 if (ret) {
 815                         resource_size_t endres = start + size - 1;
 816
 817                         pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
 818                                         &start, &endres, ret);
 819                 }
 820         }
 821
 822         clear_zone_contiguous(zone);
 823
 824         /*
 825          * We can only remove entire sections
 826          */
 827         BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
 828         BUG_ON(nr_pages % PAGES_PER_SECTION);
 829
 830         sections_to_remove = nr_pages / PAGES_PER_SECTION;
 831         for (i = 0; i < sections_to_remove; i++) {
 832                 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
 833
 834                 ret = __remove_section(zone, __pfn_to_section(pfn), map_offset);
 835                 map_offset = 0;
 836                 if (ret)
 837                         break;
 838         }
 839
 840         set_zone_contiguous(zone);
 841
 842         return ret;
 843 }
 844 EXPORT_SYMBOL_GPL(__remove_pages);
 845 #endif /* CONFIG_MEMORY_HOTREMOVE */
 846
 847 int set_online_page_callback(online_page_callback_t callback)
 848 {
 849         int rc = -EINVAL;
 850
 851         get_online_mems();
 852         mutex_lock(&online_page_callback_lock);
 853
 854         if (online_page_callback == generic_online_page) {
 855                 online_page_callback = callback;
 856                 rc = 0;
 857         }
 858
 859         mutex_unlock(&online_page_callback_lock);
 860         put_online_mems();
 861
 862         return rc;
 863 }
 864 EXPORT_SYMBOL_GPL(set_online_page_callback);
 865
 866 int restore_online_page_callback(online_page_callback_t callback)
 867 {
 868         int rc = -EINVAL;
 869
 870         get_online_mems();
 871         mutex_lock(&online_page_callback_lock);
 872
 873         if (online_page_callback == callback) {
 874                 online_page_callback = generic_online_page;
 875                 rc = 0;
 876         }
 877
 878         mutex_unlock(&online_page_callback_lock);
 879         put_online_mems();
 880
 881         return rc;
 882 }
 883 EXPORT_SYMBOL_GPL(restore_online_page_callback);
 884
 885 void __online_page_set_limits(struct page *page)
 886 {
 887 }
 888 EXPORT_SYMBOL_GPL(__online_page_set_limits);
 889
 890 void __online_page_increment_counters(struct page *page)
 891 {
 892         adjust_managed_page_count(page, 1);
 893 }
 894 EXPORT_SYMBOL_GPL(__online_page_increment_counters);
 895
 896 void __online_page_free(struct page *page)
 897 {
 898         __free_reserved_page(page);
 899 }
 900 EXPORT_SYMBOL_GPL(__online_page_free);
 901
 902 static void generic_online_page(struct page *page)
 903 {
 904         __online_page_set_limits(page);
 905         __online_page_increment_counters(page);
 906         __online_page_free(page);
 907 }
 908
 909 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
 910                         void *arg)
 911 {
 912         unsigned long i;
 913         unsigned long onlined_pages = *(unsigned long *)arg;
 914         struct page *page;
 915         if (PageReserved(pfn_to_page(start_pfn)))
 916                 for (i = 0; i < nr_pages; i++) {
 917                         page = pfn_to_page(start_pfn + i);
 918                         (*online_page_callback)(page);
 919                         onlined_pages++;
 920                 }
 921         *(unsigned long *)arg = onlined_pages;
 922         return 0;
 923 }
 924
 925 #ifdef CONFIG_MOVABLE_NODE
 926 /*
 927  * When CONFIG_MOVABLE_NODE, we permit onlining of a node which doesn't have
 928  * normal memory.
 929  */
 930 static bool can_online_high_movable(struct zone *zone)
 931 {
 932         return true;
 933 }
 934 #else /* CONFIG_MOVABLE_NODE */
 935 /* ensure every online node has NORMAL memory */
 936 static bool can_online_high_movable(struct zone *zone)
 937 {
 938         return node_state(zone_to_nid(zone), N_NORMAL_MEMORY);
 939 }
 940 #endif /* CONFIG_MOVABLE_NODE */
 941
 942 /* check which state of node_states will be changed when online memory */
 943 static void node_states_check_changes_online(unsigned long nr_pages,
 944         struct zone *zone, struct memory_notify *arg)
 945 {
 946         int nid = zone_to_nid(zone);
 947         enum zone_type zone_last = ZONE_NORMAL;
 948
 949         /*
 950          * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
 951          * contains nodes which have zones of 0...ZONE_NORMAL,
 952          * set zone_last to ZONE_NORMAL.
 953          *
 954          * If we don't have HIGHMEM nor movable node,
 955          * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
 956          * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
 957          */
 958         if (N_MEMORY == N_NORMAL_MEMORY)
 959                 zone_last = ZONE_MOVABLE;
 960
 961         /*
 962          * if the memory to be online is in a zone of 0...zone_last, and
 963          * the zones of 0...zone_last don't have memory before online, we will
 964          * need to set the node to node_states[N_NORMAL_MEMORY] after
 965          * the memory is online.
 966          */
 967         if (zone_idx(zone) <= zone_last && !node_state(nid, N_NORMAL_MEMORY))
 968                 arg->status_change_nid_normal = nid;
 969         else
 970                 arg->status_change_nid_normal = -1;
 971
 972 #ifdef CONFIG_HIGHMEM
 973         /*
 974          * If we have movable node, node_states[N_HIGH_MEMORY]
 975          * contains nodes which have zones of 0...ZONE_HIGHMEM,
 976          * set zone_last to ZONE_HIGHMEM.
 977          *
 978          * If we don't have movable node, node_states[N_NORMAL_MEMORY]
 979          * contains nodes which have zones of 0...ZONE_MOVABLE,
 980          * set zone_last to ZONE_MOVABLE.
 981          */
 982         zone_last = ZONE_HIGHMEM;
 983         if (N_MEMORY == N_HIGH_MEMORY)
 984                 zone_last = ZONE_MOVABLE;
 985
 986         if (zone_idx(zone) <= zone_last && !node_state(nid, N_HIGH_MEMORY))
 987                 arg->status_change_nid_high = nid;
 988         else
 989                 arg->status_change_nid_high = -1;
 990 #else
 991         arg->status_change_nid_high = arg->status_change_nid_normal;
 992 #endif
 993
 994         /*
 995          * if the node don't have memory befor online, we will need to
 996          * set the node to node_states[N_MEMORY] after the memory
 997          * is online.
 998          */
 999         if (!node_state(nid, N_MEMORY))
1000                 arg->status_change_nid = nid;
1001         else
1002                 arg->status_change_nid = -1;
1003 }
1004
1005 static void node_states_set_node(int node, struct memory_notify *arg)
1006 {
1007         if (arg->status_change_nid_normal >= 0)
1008                 node_set_state(node, N_NORMAL_MEMORY);
1009
1010         if (arg->status_change_nid_high >= 0)
1011                 node_set_state(node, N_HIGH_MEMORY);
1012
1013         node_set_state(node, N_MEMORY);
1014 }
1015
1016
1017 /* Must be protected by mem_hotplug_begin() */
1018 int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type)
1019 {
1020         unsigned long flags;
1021         unsigned long onlined_pages = 0;
1022         struct zone *zone;
1023         int need_zonelists_rebuild = 0;
1024         int nid;
1025         int ret;
1026         struct memory_notify arg;
1027
1028         /*
1029          * This doesn't need a lock to do pfn_to_page().
1030          * The section can't be removed here because of the
1031          * memory_block->state_mutex.
1032          */
1033         zone = page_zone(pfn_to_page(pfn));
1034
1035         if ((zone_idx(zone) > ZONE_NORMAL ||
1036             online_type == MMOP_ONLINE_MOVABLE) &&
1037             !can_online_high_movable(zone))
1038                 return -EINVAL;
1039
1040         if (online_type == MMOP_ONLINE_KERNEL &&
1041             zone_idx(zone) == ZONE_MOVABLE) {
1042                 if (move_pfn_range_left(zone - 1, zone, pfn, pfn + nr_pages))
1043                         return -EINVAL;
1044         }
1045         if (online_type == MMOP_ONLINE_MOVABLE &&
1046             zone_idx(zone) == ZONE_MOVABLE - 1) {
1047                 if (move_pfn_range_right(zone, zone + 1, pfn, pfn + nr_pages))
1048                         return -EINVAL;
1049         }
1050
1051         /* Previous code may changed the zone of the pfn range */
1052         zone = page_zone(pfn_to_page(pfn));
1053
1054         arg.start_pfn = pfn;
1055         arg.nr_pages = nr_pages;
1056         node_states_check_changes_online(nr_pages, zone, &arg);
1057
1058         nid = zone_to_nid(zone);
1059
1060         ret = memory_notify(MEM_GOING_ONLINE, &arg);
1061         ret = notifier_to_errno(ret);
1062         if (ret)
1063                 goto failed_addition;
1064
1065         /*
1066          * If this zone is not populated, then it is not in zonelist.
1067          * This means the page allocator ignores this zone.
1068          * So, zonelist must be updated after online.
1069          */
1070         mutex_lock(&zonelists_mutex);
1071         if (!populated_zone(zone)) {
1072                 need_zonelists_rebuild = 1;
1073                 build_all_zonelists(NULL, zone);
1074         }
1075
1076         ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
1077                 online_pages_range);
1078         if (ret) {
1079                 if (need_zonelists_rebuild)
1080                         zone_pcp_reset(zone);
1081                 mutex_unlock(&zonelists_mutex);
1082                 goto failed_addition;
1083         }
1084
1085         zone->present_pages += onlined_pages;
1086
1087         pgdat_resize_lock(zone->zone_pgdat, &flags);
1088         zone->zone_pgdat->node_present_pages += onlined_pages;
1089         pgdat_resize_unlock(zone->zone_pgdat, &flags);
1090
1091         if (onlined_pages) {
1092                 node_states_set_node(nid, &arg);
1093                 if (need_zonelists_rebuild)
1094                         build_all_zonelists(NULL, NULL);
1095                 else
1096                         zone_pcp_update(zone);
1097         }
1098
1099         mutex_unlock(&zonelists_mutex);
1100
1101         init_per_zone_wmark_min();
1102
1103         if (onlined_pages) {
1104                 kswapd_run(nid);
1105                 kcompactd_run(nid);
1106         }
1107
1108         vm_total_pages = nr_free_pagecache_pages();
1109
1110         writeback_set_ratelimit();
1111
1112         if (onlined_pages)
1113                 memory_notify(MEM_ONLINE, &arg);
1114         return 0;
1115
1116 failed_addition:
1117         pr_debug("online_pages [mem %#010llx-%#010llx] failed\n",
1118                  (unsigned long long) pfn << PAGE_SHIFT,
1119                  (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1);
1120         memory_notify(MEM_CANCEL_ONLINE, &arg);
1121         return ret;
1122 }
1123 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
1124
1125 static void reset_node_present_pages(pg_data_t *pgdat)
1126 {
1127         struct zone *z;
1128
1129         for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
1130                 z->present_pages = 0;
1131
1132         pgdat->node_present_pages = 0;
1133 }
1134
1135 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1136 static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
1137 {
1138         struct pglist_data *pgdat;
1139         unsigned long zones_size[MAX_NR_ZONES] = {0};
1140         unsigned long zholes_size[MAX_NR_ZONES] = {0};
1141         unsigned long start_pfn = PFN_DOWN(start);
1142
1143         pgdat = NODE_DATA(nid);
1144         if (!pgdat) {
1145                 pgdat = arch_alloc_nodedata(nid);
1146                 if (!pgdat)
1147                         return NULL;
1148
1149                 arch_refresh_nodedata(nid, pgdat);
1150         } else {
1151                 /* Reset the nr_zones and classzone_idx to 0 before reuse */
1152                 pgdat->nr_zones = 0;
1153                 pgdat->classzone_idx = 0;
1154         }
1155
1156         /* we can use NODE_DATA(nid) from here */
1157
1158         /* init node's zones as empty zones, we don't have any present pages.*/
1159         free_area_init_node(nid, zones_size, start_pfn, zholes_size);
1160
1161         /*
1162          * The node we allocated has no zone fallback lists. For avoiding
1163          * to access not-initialized zonelist, build here.
1164          */
1165         mutex_lock(&zonelists_mutex);
1166         build_all_zonelists(pgdat, NULL);
1167         mutex_unlock(&zonelists_mutex);
1168
1169         /*
1170          * zone->managed_pages is set to an approximate value in
1171          * free_area_init_core(), which will cause
1172          * /sys/device/system/node/nodeX/meminfo has wrong data.
1173          * So reset it to 0 before any memory is onlined.
1174          */
1175         reset_node_managed_pages(pgdat);
1176
1177         /*
1178          * When memory is hot-added, all the memory is in offline state. So
1179          * clear all zones' present_pages because they will be updated in
1180          * online_pages() and offline_pages().
1181          */
1182         reset_node_present_pages(pgdat);
1183
1184         return pgdat;
1185 }
1186
1187 static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
1188 {
1189         arch_refresh_nodedata(nid, NULL);
1190         arch_free_nodedata(pgdat);
1191         return;
1192 }
1193
1194
1195 /**
1196  * try_online_node - online a node if offlined
1197  *
1198  * called by cpu_up() to online a node without onlined memory.
1199  */
1200 int try_online_node(int nid)
1201 {
1202         pg_data_t       *pgdat;
1203         int     ret;
1204
1205         if (node_online(nid))
1206                 return 0;
1207
1208         mem_hotplug_begin();
1209         pgdat = hotadd_new_pgdat(nid, 0);
1210         if (!pgdat) {
1211                 pr_err("Cannot online node %d due to NULL pgdat\n", nid);
1212                 ret = -ENOMEM;
1213                 goto out;
1214         }
1215         node_set_online(nid);
1216         ret = register_one_node(nid);
1217         BUG_ON(ret);
1218
1219         if (pgdat->node_zonelists->_zonerefs->zone == NULL) {
1220                 mutex_lock(&zonelists_mutex);
1221                 build_all_zonelists(NULL, NULL);
1222                 mutex_unlock(&zonelists_mutex);
1223         }
1224
1225 out:
1226         mem_hotplug_done();
1227         return ret;
1228 }
1229
1230 static int check_hotplug_memory_range(u64 start, u64 size)
1231 {
1232         u64 start_pfn = PFN_DOWN(start);
1233         u64 nr_pages = size >> PAGE_SHIFT;
1234
1235         /* Memory range must be aligned with section */
1236         if ((start_pfn & ~PAGE_SECTION_MASK) ||
1237             (nr_pages % PAGES_PER_SECTION) || (!nr_pages)) {
1238                 pr_err("Section-unaligned hotplug range: start 0x%llx, size 0x%llx\n",
1239                                 (unsigned long long)start,
1240                                 (unsigned long long)size);
1241                 return -EINVAL;
1242         }
1243
1244         return 0;
1245 }
1246
1247 /*
1248  * If movable zone has already been setup, newly added memory should be check.
1249  * If its address is higher than movable zone, it should be added as movable.
1250  * Without this check, movable zone may overlap with other zone.
1251  */
1252 static int should_add_memory_movable(int nid, u64 start, u64 size)
1253 {
1254         unsigned long start_pfn = start >> PAGE_SHIFT;
1255         pg_data_t *pgdat = NODE_DATA(nid);
1256         struct zone *movable_zone = pgdat->node_zones + ZONE_MOVABLE;
1257
1258         if (zone_is_empty(movable_zone))
1259                 return 0;
1260
1261         if (movable_zone->zone_start_pfn <= start_pfn)
1262                 return 1;
1263
1264         return 0;
1265 }
1266
1267 int zone_for_memory(int nid, u64 start, u64 size, int zone_default,
1268                 bool for_device)
1269 {
1270 #ifdef CONFIG_ZONE_DEVICE
1271         if (for_device)
1272                 return ZONE_DEVICE;
1273 #endif
1274         if (should_add_memory_movable(nid, start, size))
1275                 return ZONE_MOVABLE;
1276
1277         return zone_default;
1278 }
1279
1280 static int online_memory_block(struct memory_block *mem, void *arg)
1281 {
1282         return memory_block_change_state(mem, MEM_ONLINE, MEM_OFFLINE);
1283 }
1284
1285 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1286 int __ref add_memory_resource(int nid, struct resource *res, bool online)
1287 {
1288         u64 start, size;
1289         pg_data_t *pgdat = NULL;
1290         bool new_pgdat;
1291         bool new_node;
1292         int ret;
1293
1294         start = res->start;
1295         size = resource_size(res);
1296
1297         ret = check_hotplug_memory_range(start, size);
1298         if (ret)
1299                 return ret;
1300
1301         {       /* Stupid hack to suppress address-never-null warning */
1302                 void *p = NODE_DATA(nid);
1303                 new_pgdat = !p;
1304         }
1305
1306         mem_hotplug_begin();
1307
1308         /*
1309          * Add new range to memblock so that when hotadd_new_pgdat() is called
1310          * to allocate new pgdat, get_pfn_range_for_nid() will be able to find
1311          * this new range and calculate total pages correctly.  The range will
1312          * be removed at hot-remove time.
1313          */
1314         memblock_add_node(start, size, nid);
1315
1316         new_node = !node_online(nid);
1317         if (new_node) {
1318                 pgdat = hotadd_new_pgdat(nid, start);
1319                 ret = -ENOMEM;
1320                 if (!pgdat)
1321                         goto error;
1322         }
1323
1324         /* call arch's memory hotadd */
1325         ret = arch_add_memory(nid, start, size, false);
1326
1327         if (ret < 0)
1328                 goto error;
1329
1330         /* we online node here. we can't roll back from here. */
1331         node_set_online(nid);
1332
1333         if (new_node) {
1334                 ret = register_one_node(nid);
1335                 /*
1336                  * If sysfs file of new node can't create, cpu on the node
1337                  * can't be hot-added. There is no rollback way now.
1338                  * So, check by BUG_ON() to catch it reluctantly..
1339                  */
1340                 BUG_ON(ret);
1341         }
1342
1343         /* create new memmap entry */
1344         firmware_map_add_hotplug(start, start + size, "System RAM");
1345
1346         /* online pages if requested */
1347         if (online)
1348                 walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1),
1349                                   NULL, online_memory_block);
1350
1351         goto out;
1352
1353 error:
1354         /* rollback pgdat allocation and others */
1355         if (new_pgdat)
1356                 rollback_node_hotadd(nid, pgdat);
1357         memblock_remove(start, size);
1358
1359 out:
1360         mem_hotplug_done();
1361         return ret;
1362 }
1363 EXPORT_SYMBOL_GPL(add_memory_resource);
1364
1365 int __ref add_memory(int nid, u64 start, u64 size)
1366 {
1367         struct resource *res;
1368         int ret;
1369
1370         res = register_memory_resource(start, size);
1371         if (IS_ERR(res))
1372                 return PTR_ERR(res);
1373
1374         ret = add_memory_resource(nid, res, memhp_auto_online);
1375         if (ret < 0)
1376                 release_memory_resource(res);
1377         return ret;
1378 }
1379 EXPORT_SYMBOL_GPL(add_memory);
1380
1381 #ifdef CONFIG_MEMORY_HOTREMOVE
1382 /*
1383  * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1384  * set and the size of the free page is given by page_order(). Using this,
1385  * the function determines if the pageblock contains only free pages.
1386  * Due to buddy contraints, a free page at least the size of a pageblock will
1387  * be located at the start of the pageblock
1388  */
1389 static inline int pageblock_free(struct page *page)
1390 {
1391         return PageBuddy(page) && page_order(page) >= pageblock_order;
1392 }
1393
1394 /* Return the start of the next active pageblock after a given page */
1395 static struct page *next_active_pageblock(struct page *page)
1396 {
1397         /* Ensure the starting page is pageblock-aligned */
1398         BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
1399
1400         /* If the entire pageblock is free, move to the end of free page */
1401         if (pageblock_free(page)) {
1402                 int order;
1403                 /* be careful. we don't have locks, page_order can be changed.*/
1404                 order = page_order(page);
1405                 if ((order < MAX_ORDER) && (order >= pageblock_order))
1406                         return page + (1 << order);
1407         }
1408
1409         return page + pageblock_nr_pages;
1410 }
1411
1412 /* Checks if this range of memory is likely to be hot-removable. */
1413 int is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
1414 {
1415         struct page *page = pfn_to_page(start_pfn);
1416         struct page *end_page = page + nr_pages;
1417
1418         /* Check the starting page of each pageblock within the range */
1419         for (; page < end_page; page = next_active_pageblock(page)) {
1420                 if (!is_pageblock_removable_nolock(page))
1421                         return 0;
1422                 cond_resched();
1423         }
1424
1425         /* All pageblocks in the memory block are likely to be hot-removable */
1426         return 1;
1427 }
1428
1429 /*
1430  * Confirm all pages in a range [start, end) is belongs to the same zone.
1431  */
1432 int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn)
1433 {
1434         unsigned long pfn, sec_end_pfn;
1435         struct zone *zone = NULL;
1436         struct page *page;
1437         int i;
1438         for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn);
1439              pfn < end_pfn;
1440              pfn = sec_end_pfn + 1, sec_end_pfn += PAGES_PER_SECTION) {
1441                 /* Make sure the memory section is present first */
1442                 if (!present_section_nr(pfn_to_section_nr(pfn)))
1443                         continue;
1444                 for (; pfn < sec_end_pfn && pfn < end_pfn;
1445                      pfn += MAX_ORDER_NR_PAGES) {
1446                         i = 0;
1447                         /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1448                         while ((i < MAX_ORDER_NR_PAGES) &&
1449                                 !pfn_valid_within(pfn + i))
1450                                 i++;
1451                         if (i == MAX_ORDER_NR_PAGES)
1452                                 continue;
1453                         page = pfn_to_page(pfn + i);
1454                         if (zone && page_zone(page) != zone)
1455                                 return 0;
1456                         zone = page_zone(page);
1457                 }
1458         }
1459         return 1;
1460 }
1461
1462 /*
1463  * Scan pfn range [start,end) to find movable/migratable pages (LRU pages
1464  * and hugepages). We scan pfn because it's much easier than scanning over
1465  * linked list. This function returns the pfn of the first found movable
1466  * page if it's found, otherwise 0.
1467  */
1468 static unsigned long scan_movable_pages(unsigned long start, unsigned long end)
1469 {
1470         unsigned long pfn;
1471         struct page *page;
1472         for (pfn = start; pfn < end; pfn++) {
1473                 if (pfn_valid(pfn)) {
1474                         page = pfn_to_page(pfn);
1475                         if (PageLRU(page))
1476                                 return pfn;
1477                         if (PageHuge(page)) {
1478                                 if (page_huge_active(page))
1479                                         return pfn;
1480                                 else
1481                                         pfn = round_up(pfn + 1,
1482                                                 1 << compound_order(page)) - 1;
1483                         }
1484                 }
1485         }
1486         return 0;
1487 }
1488
1489 #define NR_OFFLINE_AT_ONCE_PAGES        (256)
1490 static int
1491 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1492 {
1493         unsigned long pfn;
1494         struct page *page;
1495         int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
1496         int not_managed = 0;
1497         int ret = 0;
1498         LIST_HEAD(source);
1499
1500         for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
1501                 if (!pfn_valid(pfn))
1502                         continue;
1503                 page = pfn_to_page(pfn);
1504
1505                 if (PageHuge(page)) {
1506                         struct page *head = compound_head(page);
1507                         pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1;
1508                         if (compound_order(head) > PFN_SECTION_SHIFT) {
1509                                 ret = -EBUSY;
1510                                 break;
1511                         }
1512                         if (isolate_huge_page(page, &source))
1513                                 move_pages -= 1 << compound_order(head);
1514                         continue;
1515                 }
1516
1517                 if (!get_page_unless_zero(page))
1518                         continue;
1519                 /*
1520                  * We can skip free pages. And we can only deal with pages on
1521                  * LRU.
1522                  */
1523                 ret = isolate_lru_page(page);
1524                 if (!ret) { /* Success */
1525                         put_page(page);
1526                         list_add_tail(&page->lru, &source);
1527                         move_pages--;
1528                         inc_zone_page_state(page, NR_ISOLATED_ANON +
1529                                             page_is_file_cache(page));
1530
1531                 } else {
1532 #ifdef CONFIG_DEBUG_VM
1533                         pr_alert("removing pfn %lx from LRU failed\n", pfn);
1534                         dump_page(page, "failed to remove from LRU");
1535 #endif
1536                         put_page(page);
1537                         /* Because we don't have big zone->lock. we should
1538                            check this again here. */
1539                         if (page_count(page)) {
1540                                 not_managed++;
1541                                 ret = -EBUSY;
1542                                 break;
1543                         }
1544                 }
1545         }
1546         if (!list_empty(&source)) {
1547                 if (not_managed) {
1548                         putback_movable_pages(&source);
1549                         goto out;
1550                 }
1551
1552                 /*
1553                  * alloc_migrate_target should be improooooved!!
1554                  * migrate_pages returns # of failed pages.
1555                  */
1556                 ret = migrate_pages(&source, alloc_migrate_target, NULL, 0,
1557                                         MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
1558                 if (ret)
1559                         putback_movable_pages(&source);
1560         }
1561 out:
1562         return ret;
1563 }
1564
1565 /*
1566  * remove from free_area[] and mark all as Reserved.
1567  */
1568 static int
1569 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1570                         void *data)
1571 {
1572         __offline_isolated_pages(start, start + nr_pages);
1573         return 0;
1574 }
1575
1576 static void
1577 offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
1578 {
1579         walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
1580                                 offline_isolated_pages_cb);
1581 }
1582
1583 /*
1584  * Check all pages in range, recoreded as memory resource, are isolated.
1585  */
1586 static int
1587 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1588                         void *data)
1589 {
1590         int ret;
1591         long offlined = *(long *)data;
1592         ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true);
1593         offlined = nr_pages;
1594         if (!ret)
1595                 *(long *)data += offlined;
1596         return ret;
1597 }
1598
1599 static long
1600 check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
1601 {
1602         long offlined = 0;
1603         int ret;
1604
1605         ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
1606                         check_pages_isolated_cb);
1607         if (ret < 0)
1608                 offlined = (long)ret;
1609         return offlined;
1610 }
1611
1612 #ifdef CONFIG_MOVABLE_NODE
1613 /*
1614  * When CONFIG_MOVABLE_NODE, we permit offlining of a node which doesn't have
1615  * normal memory.
1616  */
1617 static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1618 {
1619         return true;
1620 }
1621 #else /* CONFIG_MOVABLE_NODE */
1622 /* ensure the node has NORMAL memory if it is still online */
1623 static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1624 {
1625         struct pglist_data *pgdat = zone->zone_pgdat;
1626         unsigned long present_pages = 0;
1627         enum zone_type zt;
1628
1629         for (zt = 0; zt <= ZONE_NORMAL; zt++)
1630                 present_pages += pgdat->node_zones[zt].present_pages;
1631
1632         if (present_pages > nr_pages)
1633                 return true;
1634
1635         present_pages = 0;
1636         for (; zt <= ZONE_MOVABLE; zt++)
1637                 present_pages += pgdat->node_zones[zt].present_pages;
1638
1639         /*
1640          * we can't offline the last normal memory until all
1641          * higher memory is offlined.
1642          */
1643         return present_pages == 0;
1644 }
1645 #endif /* CONFIG_MOVABLE_NODE */
1646
1647 static int __init cmdline_parse_movable_node(char *p)
1648 {
1649 #ifdef CONFIG_MOVABLE_NODE
1650         /*
1651          * Memory used by the kernel cannot be hot-removed because Linux
1652          * cannot migrate the kernel pages. When memory hotplug is
1653          * enabled, we should prevent memblock from allocating memory
1654          * for the kernel.
1655          *
1656          * ACPI SRAT records all hotpluggable memory ranges. But before
1657          * SRAT is parsed, we don't know about it.
1658          *
1659          * The kernel image is loaded into memory at very early time. We
1660          * cannot prevent this anyway. So on NUMA system, we set any
1661          * node the kernel resides in as un-hotpluggable.
1662          *
1663          * Since on modern servers, one node could have double-digit
1664          * gigabytes memory, we can assume the memory around the kernel
1665          * image is also un-hotpluggable. So before SRAT is parsed, just
1666          * allocate memory near the kernel image to try the best to keep
1667          * the kernel away from hotpluggable memory.
1668          */
1669         memblock_set_bottom_up(true);
1670         movable_node_enabled = true;
1671 #else
1672         pr_warn("movable_node option not supported\n");
1673 #endif
1674         return 0;
1675 }
1676 early_param("movable_node", cmdline_parse_movable_node);
1677
1678 /* check which state of node_states will be changed when offline memory */
1679 static void node_states_check_changes_offline(unsigned long nr_pages,
1680                 struct zone *zone, struct memory_notify *arg)
1681 {
1682         struct pglist_data *pgdat = zone->zone_pgdat;
1683         unsigned long present_pages = 0;
1684         enum zone_type zt, zone_last = ZONE_NORMAL;
1685
1686         /*
1687          * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
1688          * contains nodes which have zones of 0...ZONE_NORMAL,
1689          * set zone_last to ZONE_NORMAL.
1690          *
1691          * If we don't have HIGHMEM nor movable node,
1692          * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
1693          * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
1694          */
1695         if (N_MEMORY == N_NORMAL_MEMORY)
1696                 zone_last = ZONE_MOVABLE;
1697
1698         /*
1699          * check whether node_states[N_NORMAL_MEMORY] will be changed.
1700          * If the memory to be offline is in a zone of 0...zone_last,
1701          * and it is the last present memory, 0...zone_last will
1702          * become empty after offline , thus we can determind we will
1703          * need to clear the node from node_states[N_NORMAL_MEMORY].
1704          */
1705         for (zt = 0; zt <= zone_last; zt++)
1706                 present_pages += pgdat->node_zones[zt].present_pages;
1707         if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1708                 arg->status_change_nid_normal = zone_to_nid(zone);
1709         else
1710                 arg->status_change_nid_normal = -1;
1711
1712 #ifdef CONFIG_HIGHMEM
1713         /*
1714          * If we have movable node, node_states[N_HIGH_MEMORY]
1715          * contains nodes which have zones of 0...ZONE_HIGHMEM,
1716          * set zone_last to ZONE_HIGHMEM.
1717          *
1718          * If we don't have movable node, node_states[N_NORMAL_MEMORY]
1719          * contains nodes which have zones of 0...ZONE_MOVABLE,
1720          * set zone_last to ZONE_MOVABLE.
1721          */
1722         zone_last = ZONE_HIGHMEM;
1723         if (N_MEMORY == N_HIGH_MEMORY)
1724                 zone_last = ZONE_MOVABLE;
1725
1726         for (; zt <= zone_last; zt++)
1727                 present_pages += pgdat->node_zones[zt].present_pages;
1728         if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1729                 arg->status_change_nid_high = zone_to_nid(zone);
1730         else
1731                 arg->status_change_nid_high = -1;
1732 #else
1733         arg->status_change_nid_high = arg->status_change_nid_normal;
1734 #endif
1735
1736         /*
1737          * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE
1738          */
1739         zone_last = ZONE_MOVABLE;
1740
1741         /*
1742          * check whether node_states[N_HIGH_MEMORY] will be changed
1743          * If we try to offline the last present @nr_pages from the node,
1744          * we can determind we will need to clear the node from
1745          * node_states[N_HIGH_MEMORY].
1746          */
1747         for (; zt <= zone_last; zt++)
1748                 present_pages += pgdat->node_zones[zt].present_pages;
1749         if (nr_pages >= present_pages)
1750                 arg->status_change_nid = zone_to_nid(zone);
1751         else
1752                 arg->status_change_nid = -1;
1753 }
1754
1755 static void node_states_clear_node(int node, struct memory_notify *arg)
1756 {
1757         if (arg->status_change_nid_normal >= 0)
1758                 node_clear_state(node, N_NORMAL_MEMORY);
1759
1760         if ((N_MEMORY != N_NORMAL_MEMORY) &&
1761             (arg->status_change_nid_high >= 0))
1762                 node_clear_state(node, N_HIGH_MEMORY);
1763
1764         if ((N_MEMORY != N_HIGH_MEMORY) &&
1765             (arg->status_change_nid >= 0))
1766                 node_clear_state(node, N_MEMORY);
1767 }
1768
1769 static int __ref __offline_pages(unsigned long start_pfn,
1770                   unsigned long end_pfn, unsigned long timeout)
1771 {
1772         unsigned long pfn, nr_pages, expire;
1773         long offlined_pages;
1774         int ret, drain, retry_max, node;
1775         unsigned long flags;
1776         struct zone *zone;
1777         struct memory_notify arg;
1778
1779         /* at least, alignment against pageblock is necessary */
1780         if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
1781                 return -EINVAL;
1782         if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
1783                 return -EINVAL;
1784         /* This makes hotplug much easier...and readable.
1785            we assume this for now. .*/
1786         if (!test_pages_in_a_zone(start_pfn, end_pfn))
1787                 return -EINVAL;
1788
1789         zone = page_zone(pfn_to_page(start_pfn));
1790         node = zone_to_nid(zone);
1791         nr_pages = end_pfn - start_pfn;
1792
1793         if (zone_idx(zone) <= ZONE_NORMAL && !can_offline_normal(zone, nr_pages))
1794                 return -EINVAL;
1795
1796         /* set above range as isolated */
1797         ret = start_isolate_page_range(start_pfn, end_pfn,
1798                                        MIGRATE_MOVABLE, true);
1799         if (ret)
1800                 return ret;
1801
1802         arg.start_pfn = start_pfn;
1803         arg.nr_pages = nr_pages;
1804         node_states_check_changes_offline(nr_pages, zone, &arg);
1805
1806         ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1807         ret = notifier_to_errno(ret);
1808         if (ret)
1809                 goto failed_removal;
1810
1811         pfn = start_pfn;
1812         expire = jiffies + timeout;
1813         drain = 0;
1814         retry_max = 5;
1815 repeat:
1816         /* start memory hot removal */
1817         ret = -EAGAIN;
1818         if (time_after(jiffies, expire))
1819                 goto failed_removal;
1820         ret = -EINTR;
1821         if (signal_pending(current))
1822                 goto failed_removal;
1823         ret = 0;
1824         if (drain) {
1825                 lru_add_drain_all();
1826                 cond_resched();
1827                 drain_all_pages(zone);
1828         }
1829
1830         pfn = scan_movable_pages(start_pfn, end_pfn);
1831         if (pfn) { /* We have movable pages */
1832                 ret = do_migrate_range(pfn, end_pfn);
1833                 if (!ret) {
1834                         drain = 1;
1835                         goto repeat;
1836                 } else {
1837                         if (ret < 0)
1838                                 if (--retry_max == 0)
1839                                         goto failed_removal;
1840                         yield();
1841                         drain = 1;
1842                         goto repeat;
1843                 }
1844         }
1845         /* drain all zone's lru pagevec, this is asynchronous... */
1846         lru_add_drain_all();
1847         yield();
1848         /* drain pcp pages, this is synchronous. */
1849         drain_all_pages(zone);
1850         /*
1851          * dissolve free hugepages in the memory block before doing offlining
1852          * actually in order to make hugetlbfs's object counting consistent.
1853          */
1854         dissolve_free_huge_pages(start_pfn, end_pfn);
1855         /* check again */
1856         offlined_pages = check_pages_isolated(start_pfn, end_pfn);
1857         if (offlined_pages < 0) {
1858                 ret = -EBUSY;
1859                 goto failed_removal;
1860         }
1861         pr_info("Offlined Pages %ld\n", offlined_pages);
1862         /* Ok, all of our target is isolated.
1863            We cannot do rollback at this point. */
1864         offline_isolated_pages(start_pfn, end_pfn);
1865         /* reset pagetype flags and makes migrate type to be MOVABLE */
1866         undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1867         /* removal success */
1868         adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages);
1869         zone->present_pages -= offlined_pages;
1870
1871         pgdat_resize_lock(zone->zone_pgdat, &flags);
1872         zone->zone_pgdat->node_present_pages -= offlined_pages;
1873         pgdat_resize_unlock(zone->zone_pgdat, &flags);
1874
1875         init_per_zone_wmark_min();
1876
1877         if (!populated_zone(zone)) {
1878                 zone_pcp_reset(zone);
1879                 mutex_lock(&zonelists_mutex);
1880                 build_all_zonelists(NULL, NULL);
1881                 mutex_unlock(&zonelists_mutex);
1882         } else
1883                 zone_pcp_update(zone);
1884
1885         node_states_clear_node(node, &arg);
1886         if (arg.status_change_nid >= 0) {
1887                 kswapd_stop(node);
1888                 kcompactd_stop(node);
1889         }
1890
1891         vm_total_pages = nr_free_pagecache_pages();
1892         writeback_set_ratelimit();
1893
1894         memory_notify(MEM_OFFLINE, &arg);
1895         return 0;
1896
1897 failed_removal:
1898         pr_debug("memory offlining [mem %#010llx-%#010llx] failed\n",
1899                  (unsigned long long) start_pfn << PAGE_SHIFT,
1900                  ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
1901         memory_notify(MEM_CANCEL_OFFLINE, &arg);
1902         /* pushback to free area */
1903         undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1904         return ret;
1905 }
1906
1907 /* Must be protected by mem_hotplug_begin() */
1908 int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1909 {
1910         return __offline_pages(start_pfn, start_pfn + nr_pages, 120 * HZ);
1911 }
1912 #endif /* CONFIG_MEMORY_HOTREMOVE */
1913
1914 /**
1915  * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
1916  * @start_pfn: start pfn of the memory range
1917  * @end_pfn: end pfn of the memory range
1918  * @arg: argument passed to func
1919  * @func: callback for each memory section walked
1920  *
1921  * This function walks through all present mem sections in range
1922  * [start_pfn, end_pfn) and call func on each mem section.
1923  *
1924  * Returns the return value of func.
1925  */
1926 int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn,
1927                 void *arg, int (*func)(struct memory_block *, void *))
1928 {
1929         struct memory_block *mem = NULL;
1930         struct mem_section *section;
1931         unsigned long pfn, section_nr;
1932         int ret;
1933
1934         for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1935                 section_nr = pfn_to_section_nr(pfn);
1936                 if (!present_section_nr(section_nr))
1937                         continue;
1938
1939                 section = __nr_to_section(section_nr);
1940                 /* same memblock? */
1941                 if (mem)
1942                         if ((section_nr >= mem->start_section_nr) &&
1943                             (section_nr <= mem->end_section_nr))
1944                                 continue;
1945
1946                 mem = find_memory_block_hinted(section, mem);
1947                 if (!mem)
1948                         continue;
1949
1950                 ret = func(mem, arg);
1951                 if (ret) {
1952                         kobject_put(&mem->dev.kobj);
1953                         return ret;
1954                 }
1955         }
1956
1957         if (mem)
1958                 kobject_put(&mem->dev.kobj);
1959
1960         return 0;
1961 }
1962
1963 #ifdef CONFIG_MEMORY_HOTREMOVE
1964 static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
1965 {
1966         int ret = !is_memblock_offlined(mem);
1967
1968         if (unlikely(ret)) {
1969                 phys_addr_t beginpa, endpa;
1970
1971                 beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
1972                 endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1;
1973                 pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n",
1974                         &beginpa, &endpa);
1975         }
1976
1977         return ret;
1978 }
1979
1980 static int check_cpu_on_node(pg_data_t *pgdat)
1981 {
1982         int cpu;
1983
1984         for_each_present_cpu(cpu) {
1985                 if (cpu_to_node(cpu) == pgdat->node_id)
1986                         /*
1987                          * the cpu on this node isn't removed, and we can't
1988                          * offline this node.
1989                          */
1990                         return -EBUSY;
1991         }
1992
1993         return 0;
1994 }
1995
1996 static void unmap_cpu_on_node(pg_data_t *pgdat)
1997 {
1998 #ifdef CONFIG_ACPI_NUMA
1999         int cpu;
2000
2001         for_each_possible_cpu(cpu)
2002                 if (cpu_to_node(cpu) == pgdat->node_id)
2003                         numa_clear_node(cpu);
2004 #endif
2005 }
2006
2007 static int check_and_unmap_cpu_on_node(pg_data_t *pgdat)
2008 {
2009         int ret;
2010
2011         ret = check_cpu_on_node(pgdat);
2012         if (ret)
2013                 return ret;
2014
2015         /*
2016          * the node will be offlined when we come here, so we can clear
2017          * the cpu_to_node() now.
2018          */
2019
2020         unmap_cpu_on_node(pgdat);
2021         return 0;
2022 }
2023
2024 /**
2025  * try_offline_node
2026  *
2027  * Offline a node if all memory sections and cpus of the node are removed.
2028  *
2029  * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
2030  * and online/offline operations before this call.
2031  */
2032 void try_offline_node(int nid)
2033 {
2034         pg_data_t *pgdat = NODE_DATA(nid);
2035         unsigned long start_pfn = pgdat->node_start_pfn;
2036         unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
2037         unsigned long pfn;
2038         int i;
2039
2040         for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
2041                 unsigned long section_nr = pfn_to_section_nr(pfn);
2042
2043                 if (!present_section_nr(section_nr))
2044                         continue;
2045
2046                 if (pfn_to_nid(pfn) != nid)
2047                         continue;
2048
2049                 /*
2050                  * some memory sections of this node are not removed, and we
2051                  * can't offline node now.
2052                  */
2053                 return;
2054         }
2055
2056         if (check_and_unmap_cpu_on_node(pgdat))
2057                 return;
2058
2059         /*
2060          * all memory/cpu of this node are removed, we can offline this
2061          * node now.
2062          */
2063         node_set_offline(nid);
2064         unregister_one_node(nid);
2065
2066         /* free waittable in each zone */
2067         for (i = 0; i < MAX_NR_ZONES; i++) {
2068                 struct zone *zone = pgdat->node_zones + i;
2069
2070                 /*
2071                  * wait_table may be allocated from boot memory,
2072                  * here only free if it's allocated by vmalloc.
2073                  */
2074                 if (is_vmalloc_addr(zone->wait_table)) {
2075                         vfree(zone->wait_table);
2076                         zone->wait_table = NULL;
2077                 }
2078         }
2079 }
2080 EXPORT_SYMBOL(try_offline_node);
2081
2082 /**
2083  * remove_memory
2084  *
2085  * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
2086  * and online/offline operations before this call, as required by
2087  * try_offline_node().
2088  */
2089 void __ref remove_memory(int nid, u64 start, u64 size)
2090 {
2091         int ret;
2092
2093         BUG_ON(check_hotplug_memory_range(start, size));
2094
2095         mem_hotplug_begin();
2096
2097         /*
2098          * All memory blocks must be offlined before removing memory.  Check
2099          * whether all memory blocks in question are offline and trigger a BUG()
2100          * if this is not the case.
2101          */
2102         ret = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL,
2103                                 check_memblock_offlined_cb);
2104         if (ret)
2105                 BUG();
2106
2107         /* remove memmap entry */
2108         firmware_map_remove(start, start + size, "System RAM");
2109         memblock_free(start, size);
2110         memblock_remove(start, size);
2111
2112         arch_remove_memory(start, size);
2113
2114         try_offline_node(nid);
2115
2116         mem_hotplug_done();
2117 }
2118 EXPORT_SYMBOL_GPL(remove_memory);
2119 #endif /* CONFIG_MEMORY_HOTREMOVE */