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