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/sched/signal.h>
  10 #include <linux/swap.h>
  11 #include <linux/interrupt.h>
  12 #include <linux/pagemap.h>
  13 #include <linux/compiler.h>
  14 #include <linux/export.h>
  15 #include <linux/pagevec.h>
  16 #include <linux/writeback.h>
  17 #include <linux/slab.h>
  18 #include <linux/sysctl.h>
  19 #include <linux/cpu.h>
  20 #include <linux/memory.h>
  21 #include <linux/memremap.h>
  22 #include <linux/memory_hotplug.h>
  23 #include <linux/highmem.h>
  24 #include <linux/vmalloc.h>
  25 #include <linux/ioport.h>
  26 #include <linux/delay.h>
  27 #include <linux/migrate.h>
  28 #include <linux/page-isolation.h>
  29 #include <linux/pfn.h>
  30 #include <linux/suspend.h>
  31 #include <linux/mm_inline.h>
  32 #include <linux/firmware-map.h>
  33 #include <linux/stop_machine.h>
  34 #include <linux/hugetlb.h>
  35 #include <linux/memblock.h>
  36 #include <linux/bootmem.h>
  37 #include <linux/compaction.h>
  38
  39 #include <asm/tlbflush.h>
  40
  41 #include "internal.h"
  42
  43 /*
  44  * online_page_callback contains pointer to current page onlining function.
  45  * Initially it is generic_online_page(). If it is required it could be
  46  * changed by calling set_online_page_callback() for callback registration
  47  * and restore_online_page_callback() for generic callback restore.
  48  */
  49
  50 static void generic_online_page(struct page *page);
  51
  52 static online_page_callback_t online_page_callback = generic_online_page;
  53 static DEFINE_MUTEX(online_page_callback_lock);
  54
  55 /* The same as the cpu_hotplug lock, but for memory hotplug. */
  56 static struct {
  57         struct task_struct *active_writer;
  58         struct mutex lock; /* Synchronizes accesses to refcount, */
  59         /*
  60          * Also blocks the new readers during
  61          * an ongoing mem hotplug operation.
  62          */
  63         int refcount;
  64
  65 #ifdef CONFIG_DEBUG_LOCK_ALLOC
  66         struct lockdep_map dep_map;
  67 #endif
  68 } mem_hotplug = {
  69         .active_writer = NULL,
  70         .lock = __MUTEX_INITIALIZER(mem_hotplug.lock),
  71         .refcount = 0,
  72 #ifdef CONFIG_DEBUG_LOCK_ALLOC
  73         .dep_map = {.name = "mem_hotplug.lock" },
  74 #endif
  75 };
  76
  77 /* Lockdep annotations for get/put_online_mems() and mem_hotplug_begin/end() */
  78 #define memhp_lock_acquire_read() lock_map_acquire_read(&mem_hotplug.dep_map)
  79 #define memhp_lock_acquire()      lock_map_acquire(&mem_hotplug.dep_map)
  80 #define memhp_lock_release()      lock_map_release(&mem_hotplug.dep_map)
  81
  82 #ifndef CONFIG_MEMORY_HOTPLUG_DEFAULT_ONLINE
  83 bool memhp_auto_online;
  84 #else
  85 bool memhp_auto_online = true;
  86 #endif
  87 EXPORT_SYMBOL_GPL(memhp_auto_online);
  88
  89 static int __init setup_memhp_default_state(char *str)
  90 {
  91         if (!strcmp(str, "online"))
  92                 memhp_auto_online = true;
  93         else if (!strcmp(str, "offline"))
  94                 memhp_auto_online = false;
  95
  96         return 1;
  97 }
  98 __setup("memhp_default_state=", setup_memhp_default_state);
  99
 100 void get_online_mems(void)
 101 {
 102         might_sleep();
 103         if (mem_hotplug.active_writer == current)
 104                 return;
 105         memhp_lock_acquire_read();
 106         mutex_lock(&mem_hotplug.lock);
 107         mem_hotplug.refcount++;
 108         mutex_unlock(&mem_hotplug.lock);
 109
 110 }
 111
 112 void put_online_mems(void)
 113 {
 114         if (mem_hotplug.active_writer == current)
 115                 return;
 116         mutex_lock(&mem_hotplug.lock);
 117
 118         if (WARN_ON(!mem_hotplug.refcount))
 119                 mem_hotplug.refcount++; /* try to fix things up */
 120
 121         if (!--mem_hotplug.refcount && unlikely(mem_hotplug.active_writer))
 122                 wake_up_process(mem_hotplug.active_writer);
 123         mutex_unlock(&mem_hotplug.lock);
 124         memhp_lock_release();
 125
 126 }
 127
 128 void mem_hotplug_begin(void)
 129 {
 130         assert_held_device_hotplug();
 131
 132         mem_hotplug.active_writer = current;
 133
 134         memhp_lock_acquire();
 135         for (;;) {
 136                 mutex_lock(&mem_hotplug.lock);
 137                 if (likely(!mem_hotplug.refcount))
 138                         break;
 139                 __set_current_state(TASK_UNINTERRUPTIBLE);
 140                 mutex_unlock(&mem_hotplug.lock);
 141                 schedule();
 142         }
 143 }
 144
 145 void mem_hotplug_done(void)
 146 {
 147         mem_hotplug.active_writer = NULL;
 148         mutex_unlock(&mem_hotplug.lock);
 149         memhp_lock_release();
 150 }
 151
 152 /* add this memory to iomem resource */
 153 static struct resource *register_memory_resource(u64 start, u64 size)
 154 {
 155         struct resource *res;
 156         res = kzalloc(sizeof(struct resource), GFP_KERNEL);
 157         if (!res)
 158                 return ERR_PTR(-ENOMEM);
 159
 160         res->name = "System RAM";
 161         res->start = start;
 162         res->end = start + size - 1;
 163         res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
 164         if (request_resource(&iomem_resource, res) < 0) {
 165                 pr_debug("System RAM resource %pR cannot be added\n", res);
 166                 kfree(res);
 167                 return ERR_PTR(-EEXIST);
 168         }
 169         return res;
 170 }
 171
 172 static void release_memory_resource(struct resource *res)
 173 {
 174         if (!res)
 175                 return;
 176         release_resource(res);
 177         kfree(res);
 178         return;
 179 }
 180
 181 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
 182 void get_page_bootmem(unsigned long info,  struct page *page,
 183                       unsigned long type)
 184 {
 185         page->freelist = (void *)type;
 186         SetPagePrivate(page);
 187         set_page_private(page, info);
 188         page_ref_inc(page);
 189 }
 190
 191 void put_page_bootmem(struct page *page)
 192 {
 193         unsigned long type;
 194
 195         type = (unsigned long) page->freelist;
 196         BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE ||
 197                type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE);
 198
 199         if (page_ref_dec_return(page) == 1) {
 200                 page->freelist = NULL;
 201                 ClearPagePrivate(page);
 202                 set_page_private(page, 0);
 203                 INIT_LIST_HEAD(&page->lru);
 204                 free_reserved_page(page);
 205         }
 206 }
 207
 208 #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
 209 #ifndef CONFIG_SPARSEMEM_VMEMMAP
 210 static void register_page_bootmem_info_section(unsigned long start_pfn)
 211 {
 212         unsigned long *usemap, mapsize, section_nr, i;
 213         struct mem_section *ms;
 214         struct page *page, *memmap;
 215
 216         section_nr = pfn_to_section_nr(start_pfn);
 217         ms = __nr_to_section(section_nr);
 218
 219         /* Get section's memmap address */
 220         memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
 221
 222         /*
 223          * Get page for the memmap's phys address
 224          * XXX: need more consideration for sparse_vmemmap...
 225          */
 226         page = virt_to_page(memmap);
 227         mapsize = sizeof(struct page) * PAGES_PER_SECTION;
 228         mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
 229
 230         /* remember memmap's page */
 231         for (i = 0; i < mapsize; i++, page++)
 232                 get_page_bootmem(section_nr, page, SECTION_INFO);
 233
 234         usemap = __nr_to_section(section_nr)->pageblock_flags;
 235         page = virt_to_page(usemap);
 236
 237         mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
 238
 239         for (i = 0; i < mapsize; i++, page++)
 240                 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
 241
 242 }
 243 #else /* CONFIG_SPARSEMEM_VMEMMAP */
 244 static void register_page_bootmem_info_section(unsigned long start_pfn)
 245 {
 246         unsigned long *usemap, mapsize, section_nr, i;
 247         struct mem_section *ms;
 248         struct page *page, *memmap;
 249
 250         if (!pfn_valid(start_pfn))
 251                 return;
 252
 253         section_nr = pfn_to_section_nr(start_pfn);
 254         ms = __nr_to_section(section_nr);
 255
 256         memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
 257
 258         register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION);
 259
 260         usemap = __nr_to_section(section_nr)->pageblock_flags;
 261         page = virt_to_page(usemap);
 262
 263         mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
 264
 265         for (i = 0; i < mapsize; i++, page++)
 266                 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
 267 }
 268 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
 269
 270 void __init register_page_bootmem_info_node(struct pglist_data *pgdat)
 271 {
 272         unsigned long i, pfn, end_pfn, nr_pages;
 273         int node = pgdat->node_id;
 274         struct page *page;
 275
 276         nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
 277         page = virt_to_page(pgdat);
 278
 279         for (i = 0; i < nr_pages; i++, page++)
 280                 get_page_bootmem(node, page, NODE_INFO);
 281
 282         pfn = pgdat->node_start_pfn;
 283         end_pfn = pgdat_end_pfn(pgdat);
 284
 285         /* register section info */
 286         for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
 287                 /*
 288                  * Some platforms can assign the same pfn to multiple nodes - on
 289                  * node0 as well as nodeN.  To avoid registering a pfn against
 290                  * multiple nodes we check that this pfn does not already
 291                  * reside in some other nodes.
 292                  */
 293                 if (pfn_valid(pfn) && (early_pfn_to_nid(pfn) == node))
 294                         register_page_bootmem_info_section(pfn);
 295         }
 296 }
 297 #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
 298
 299 static void __meminit grow_zone_span(struct zone *zone, unsigned long start_pfn,
 300                                      unsigned long end_pfn)
 301 {
 302         unsigned long old_zone_end_pfn;
 303
 304         zone_span_writelock(zone);
 305
 306         old_zone_end_pfn = zone_end_pfn(zone);
 307         if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
 308                 zone->zone_start_pfn = start_pfn;
 309
 310         zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
 311                                 zone->zone_start_pfn;
 312
 313         zone_span_writeunlock(zone);
 314 }
 315
 316 static void resize_zone(struct zone *zone, unsigned long start_pfn,
 317                 unsigned long end_pfn)
 318 {
 319         zone_span_writelock(zone);
 320
 321         if (end_pfn - start_pfn) {
 322                 zone->zone_start_pfn = start_pfn;
 323                 zone->spanned_pages = end_pfn - start_pfn;
 324         } else {
 325                 /*
 326                  * make it consist as free_area_init_core(),
 327                  * if spanned_pages = 0, then keep start_pfn = 0
 328                  */
 329                 zone->zone_start_pfn = 0;
 330                 zone->spanned_pages = 0;
 331         }
 332
 333         zone_span_writeunlock(zone);
 334 }
 335
 336 static void fix_zone_id(struct zone *zone, unsigned long start_pfn,
 337                 unsigned long end_pfn)
 338 {
 339         enum zone_type zid = zone_idx(zone);
 340         int nid = zone->zone_pgdat->node_id;
 341         unsigned long pfn;
 342
 343         for (pfn = start_pfn; pfn < end_pfn; pfn++)
 344                 set_page_links(pfn_to_page(pfn), zid, nid, pfn);
 345 }
 346
 347 /* Can fail with -ENOMEM from allocating a wait table with vmalloc() or
 348  * alloc_bootmem_node_nopanic()/memblock_virt_alloc_node_nopanic() */
 349 static int __ref ensure_zone_is_initialized(struct zone *zone,
 350                         unsigned long start_pfn, unsigned long num_pages)
 351 {
 352         if (!zone_is_initialized(zone))
 353                 return init_currently_empty_zone(zone, start_pfn, num_pages);
 354
 355         return 0;
 356 }
 357
 358 static int __meminit move_pfn_range_left(struct zone *z1, struct zone *z2,
 359                 unsigned long start_pfn, unsigned long end_pfn)
 360 {
 361         int ret;
 362         unsigned long flags;
 363         unsigned long z1_start_pfn;
 364
 365         ret = ensure_zone_is_initialized(z1, start_pfn, end_pfn - start_pfn);
 366         if (ret)
 367                 return ret;
 368
 369         pgdat_resize_lock(z1->zone_pgdat, &flags);
 370
 371         /* can't move pfns which are higher than @z2 */
 372         if (end_pfn > zone_end_pfn(z2))
 373                 goto out_fail;
 374         /* the move out part must be at the left most of @z2 */
 375         if (start_pfn > z2->zone_start_pfn)
 376                 goto out_fail;
 377         /* must included/overlap */
 378         if (end_pfn <= z2->zone_start_pfn)
 379                 goto out_fail;
 380
 381         /* use start_pfn for z1's start_pfn if z1 is empty */
 382         if (!zone_is_empty(z1))
 383                 z1_start_pfn = z1->zone_start_pfn;
 384         else
 385                 z1_start_pfn = start_pfn;
 386
 387         resize_zone(z1, z1_start_pfn, end_pfn);
 388         resize_zone(z2, end_pfn, zone_end_pfn(z2));
 389
 390         pgdat_resize_unlock(z1->zone_pgdat, &flags);
 391
 392         fix_zone_id(z1, start_pfn, end_pfn);
 393
 394         return 0;
 395 out_fail:
 396         pgdat_resize_unlock(z1->zone_pgdat, &flags);
 397         return -1;
 398 }
 399
 400 static int __meminit move_pfn_range_right(struct zone *z1, struct zone *z2,
 401                 unsigned long start_pfn, unsigned long end_pfn)
 402 {
 403         int ret;
 404         unsigned long flags;
 405         unsigned long z2_end_pfn;
 406
 407         ret = ensure_zone_is_initialized(z2, start_pfn, end_pfn - start_pfn);
 408         if (ret)
 409                 return ret;
 410
 411         pgdat_resize_lock(z1->zone_pgdat, &flags);
 412
 413         /* can't move pfns which are lower than @z1 */
 414         if (z1->zone_start_pfn > start_pfn)
 415                 goto out_fail;
 416         /* the move out part mast at the right most of @z1 */
 417         if (zone_end_pfn(z1) >  end_pfn)
 418                 goto out_fail;
 419         /* must included/overlap */
 420         if (start_pfn >= zone_end_pfn(z1))
 421                 goto out_fail;
 422
 423         /* use end_pfn for z2's end_pfn if z2 is empty */
 424         if (!zone_is_empty(z2))
 425                 z2_end_pfn = zone_end_pfn(z2);
 426         else
 427                 z2_end_pfn = end_pfn;
 428
 429         resize_zone(z1, z1->zone_start_pfn, start_pfn);
 430         resize_zone(z2, start_pfn, z2_end_pfn);
 431
 432         pgdat_resize_unlock(z1->zone_pgdat, &flags);
 433
 434         fix_zone_id(z2, start_pfn, end_pfn);
 435
 436         return 0;
 437 out_fail:
 438         pgdat_resize_unlock(z1->zone_pgdat, &flags);
 439         return -1;
 440 }
 441
 442 static struct zone * __meminit move_pfn_range(int zone_shift,
 443                 unsigned long start_pfn, unsigned long end_pfn)
 444 {
 445         struct zone *zone = page_zone(pfn_to_page(start_pfn));
 446         int ret = 0;
 447
 448         if (zone_shift < 0)
 449                 ret = move_pfn_range_left(zone + zone_shift, zone,
 450                                           start_pfn, end_pfn);
 451         else if (zone_shift)
 452                 ret = move_pfn_range_right(zone, zone + zone_shift,
 453                                            start_pfn, end_pfn);
 454
 455         if (ret)
 456                 return NULL;
 457
 458         return zone + zone_shift;
 459 }
 460
 461 static void __meminit grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn,
 462                                       unsigned long end_pfn)
 463 {
 464         unsigned long old_pgdat_end_pfn = pgdat_end_pfn(pgdat);
 465
 466         if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
 467                 pgdat->node_start_pfn = start_pfn;
 468
 469         pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
 470                                         pgdat->node_start_pfn;
 471 }
 472
 473 static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn)
 474 {
 475         struct pglist_data *pgdat = zone->zone_pgdat;
 476         int nr_pages = PAGES_PER_SECTION;
 477         int nid = pgdat->node_id;
 478         int zone_type;
 479         unsigned long flags, pfn;
 480         int ret;
 481
 482         zone_type = zone - pgdat->node_zones;
 483         ret = ensure_zone_is_initialized(zone, phys_start_pfn, nr_pages);
 484         if (ret)
 485                 return ret;
 486
 487         pgdat_resize_lock(zone->zone_pgdat, &flags);
 488         grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages);
 489         grow_pgdat_span(zone->zone_pgdat, phys_start_pfn,
 490                         phys_start_pfn + nr_pages);
 491         pgdat_resize_unlock(zone->zone_pgdat, &flags);
 492         memmap_init_zone(nr_pages, nid, zone_type,
 493                          phys_start_pfn, MEMMAP_HOTPLUG);
 494
 495         /* online_page_range is called later and expects pages reserved */
 496         for (pfn = phys_start_pfn; pfn < phys_start_pfn + nr_pages; pfn++) {
 497                 if (!pfn_valid(pfn))
 498                         continue;
 499
 500                 SetPageReserved(pfn_to_page(pfn));
 501         }
 502         return 0;
 503 }
 504
 505 static int __meminit __add_section(int nid, struct zone *zone,
 506                                         unsigned long phys_start_pfn)
 507 {
 508         int ret;
 509
 510         if (pfn_valid(phys_start_pfn))
 511                 return -EEXIST;
 512
 513         ret = sparse_add_one_section(zone, phys_start_pfn);
 514
 515         if (ret < 0)
 516                 return ret;
 517
 518         ret = __add_zone(zone, phys_start_pfn);
 519
 520         if (ret < 0)
 521                 return ret;
 522
 523         return register_new_memory(nid, __pfn_to_section(phys_start_pfn));
 524 }
 525
 526 /*
 527  * Reasonably generic function for adding memory.  It is
 528  * expected that archs that support memory hotplug will
 529  * call this function after deciding the zone to which to
 530  * add the new pages.
 531  */
 532 int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn,
 533                         unsigned long nr_pages)
 534 {
 535         unsigned long i;
 536         int err = 0;
 537         int start_sec, end_sec;
 538         struct vmem_altmap *altmap;
 539
 540         clear_zone_contiguous(zone);
 541
 542         /* during initialize mem_map, align hot-added range to section */
 543         start_sec = pfn_to_section_nr(phys_start_pfn);
 544         end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
 545
 546         altmap = to_vmem_altmap((unsigned long) pfn_to_page(phys_start_pfn));
 547         if (altmap) {
 548                 /*
 549                  * Validate altmap is within bounds of the total request
 550                  */
 551                 if (altmap->base_pfn != phys_start_pfn
 552                                 || vmem_altmap_offset(altmap) > nr_pages) {
 553                         pr_warn_once("memory add fail, invalid altmap\n");
 554                         err = -EINVAL;
 555                         goto out;
 556                 }
 557                 altmap->alloc = 0;
 558         }
 559
 560         for (i = start_sec; i <= end_sec; i++) {
 561                 err = __add_section(nid, zone, section_nr_to_pfn(i));
 562
 563                 /*
 564                  * EEXIST is finally dealt with by ioresource collision
 565                  * check. see add_memory() => register_memory_resource()
 566                  * Warning will be printed if there is collision.
 567                  */
 568                 if (err && (err != -EEXIST))
 569                         break;
 570                 err = 0;
 571         }
 572         vmemmap_populate_print_last();
 573 out:
 574         set_zone_contiguous(zone);
 575         return err;
 576 }
 577 EXPORT_SYMBOL_GPL(__add_pages);
 578
 579 #ifdef CONFIG_MEMORY_HOTREMOVE
 580 /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
 581 static int find_smallest_section_pfn(int nid, struct zone *zone,
 582                                      unsigned long start_pfn,
 583                                      unsigned long end_pfn)
 584 {
 585         struct mem_section *ms;
 586
 587         for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) {
 588                 ms = __pfn_to_section(start_pfn);
 589
 590                 if (unlikely(!valid_section(ms)))
 591                         continue;
 592
 593                 if (unlikely(pfn_to_nid(start_pfn) != nid))
 594                         continue;
 595
 596                 if (zone && zone != page_zone(pfn_to_page(start_pfn)))
 597                         continue;
 598
 599                 return start_pfn;
 600         }
 601
 602         return 0;
 603 }
 604
 605 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
 606 static int find_biggest_section_pfn(int nid, struct zone *zone,
 607                                     unsigned long start_pfn,
 608                                     unsigned long end_pfn)
 609 {
 610         struct mem_section *ms;
 611         unsigned long pfn;
 612
 613         /* pfn is the end pfn of a memory section. */
 614         pfn = end_pfn - 1;
 615         for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) {
 616                 ms = __pfn_to_section(pfn);
 617
 618                 if (unlikely(!valid_section(ms)))
 619                         continue;
 620
 621                 if (unlikely(pfn_to_nid(pfn) != nid))
 622                         continue;
 623
 624                 if (zone && zone != page_zone(pfn_to_page(pfn)))
 625                         continue;
 626
 627                 return pfn;
 628         }
 629
 630         return 0;
 631 }
 632
 633 static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
 634                              unsigned long end_pfn)
 635 {
 636         unsigned long zone_start_pfn = zone->zone_start_pfn;
 637         unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */
 638         unsigned long zone_end_pfn = z;
 639         unsigned long pfn;
 640         struct mem_section *ms;
 641         int nid = zone_to_nid(zone);
 642
 643         zone_span_writelock(zone);
 644         if (zone_start_pfn == start_pfn) {
 645                 /*
 646                  * If the section is smallest section in the zone, it need
 647                  * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
 648                  * In this case, we find second smallest valid mem_section
 649                  * for shrinking zone.
 650                  */
 651                 pfn = find_smallest_section_pfn(nid, zone, end_pfn,
 652                                                 zone_end_pfn);
 653                 if (pfn) {
 654                         zone->zone_start_pfn = pfn;
 655                         zone->spanned_pages = zone_end_pfn - pfn;
 656                 }
 657         } else if (zone_end_pfn == end_pfn) {
 658                 /*
 659                  * If the section is biggest section in the zone, it need
 660                  * shrink zone->spanned_pages.
 661                  * In this case, we find second biggest valid mem_section for
 662                  * shrinking zone.
 663                  */
 664                 pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn,
 665                                                start_pfn);
 666                 if (pfn)
 667                         zone->spanned_pages = pfn - zone_start_pfn + 1;
 668         }
 669
 670         /*
 671          * The section is not biggest or smallest mem_section in the zone, it
 672          * only creates a hole in the zone. So in this case, we need not
 673          * change the zone. But perhaps, the zone has only hole data. Thus
 674          * it check the zone has only hole or not.
 675          */
 676         pfn = zone_start_pfn;
 677         for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) {
 678                 ms = __pfn_to_section(pfn);
 679
 680                 if (unlikely(!valid_section(ms)))
 681                         continue;
 682
 683                 if (page_zone(pfn_to_page(pfn)) != zone)
 684                         continue;
 685
 686                  /* If the section is current section, it continues the loop */
 687                 if (start_pfn == pfn)
 688                         continue;
 689
 690                 /* If we find valid section, we have nothing to do */
 691                 zone_span_writeunlock(zone);
 692                 return;
 693         }
 694
 695         /* The zone has no valid section */
 696         zone->zone_start_pfn = 0;
 697         zone->spanned_pages = 0;
 698         zone_span_writeunlock(zone);
 699 }
 700
 701 static void shrink_pgdat_span(struct pglist_data *pgdat,
 702                               unsigned long start_pfn, unsigned long end_pfn)
 703 {
 704         unsigned long pgdat_start_pfn = pgdat->node_start_pfn;
 705         unsigned long p = pgdat_end_pfn(pgdat); /* pgdat_end_pfn namespace clash */
 706         unsigned long pgdat_end_pfn = p;
 707         unsigned long pfn;
 708         struct mem_section *ms;
 709         int nid = pgdat->node_id;
 710
 711         if (pgdat_start_pfn == start_pfn) {
 712                 /*
 713                  * If the section is smallest section in the pgdat, it need
 714                  * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages.
 715                  * In this case, we find second smallest valid mem_section
 716                  * for shrinking zone.
 717                  */
 718                 pfn = find_smallest_section_pfn(nid, NULL, end_pfn,
 719                                                 pgdat_end_pfn);
 720                 if (pfn) {
 721                         pgdat->node_start_pfn = pfn;
 722                         pgdat->node_spanned_pages = pgdat_end_pfn - pfn;
 723                 }
 724         } else if (pgdat_end_pfn == end_pfn) {
 725                 /*
 726                  * If the section is biggest section in the pgdat, it need
 727                  * shrink pgdat->node_spanned_pages.
 728                  * In this case, we find second biggest valid mem_section for
 729                  * shrinking zone.
 730                  */
 731                 pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn,
 732                                                start_pfn);
 733                 if (pfn)
 734                         pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1;
 735         }
 736
 737         /*
 738          * If the section is not biggest or smallest mem_section in the pgdat,
 739          * it only creates a hole in the pgdat. So in this case, we need not
 740          * change the pgdat.
 741          * But perhaps, the pgdat has only hole data. Thus it check the pgdat
 742          * has only hole or not.
 743          */
 744         pfn = pgdat_start_pfn;
 745         for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) {
 746                 ms = __pfn_to_section(pfn);
 747
 748                 if (unlikely(!valid_section(ms)))
 749                         continue;
 750
 751                 if (pfn_to_nid(pfn) != nid)
 752                         continue;
 753
 754                  /* If the section is current section, it continues the loop */
 755                 if (start_pfn == pfn)
 756                         continue;
 757
 758                 /* If we find valid section, we have nothing to do */
 759                 return;
 760         }
 761
 762         /* The pgdat has no valid section */
 763         pgdat->node_start_pfn = 0;
 764         pgdat->node_spanned_pages = 0;
 765 }
 766
 767 static void __remove_zone(struct zone *zone, unsigned long start_pfn)
 768 {
 769         struct pglist_data *pgdat = zone->zone_pgdat;
 770         int nr_pages = PAGES_PER_SECTION;
 771         int zone_type;
 772         unsigned long flags;
 773
 774         zone_type = zone - pgdat->node_zones;
 775
 776         pgdat_resize_lock(zone->zone_pgdat, &flags);
 777         shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
 778         shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages);
 779         pgdat_resize_unlock(zone->zone_pgdat, &flags);
 780 }
 781
 782 static int __remove_section(struct zone *zone, struct mem_section *ms,
 783                 unsigned long map_offset)
 784 {
 785         unsigned long start_pfn;
 786         int scn_nr;
 787         int ret = -EINVAL;
 788
 789         if (!valid_section(ms))
 790                 return ret;
 791
 792         ret = unregister_memory_section(ms);
 793         if (ret)
 794                 return ret;
 795
 796         scn_nr = __section_nr(ms);
 797         start_pfn = section_nr_to_pfn(scn_nr);
 798         __remove_zone(zone, start_pfn);
 799
 800         sparse_remove_one_section(zone, ms, map_offset);
 801         return 0;
 802 }
 803
 804 /**
 805  * __remove_pages() - remove sections of pages from a zone
 806  * @zone: zone from which pages need to be removed
 807  * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
 808  * @nr_pages: number of pages to remove (must be multiple of section size)
 809  *
 810  * Generic helper function to remove section mappings and sysfs entries
 811  * for the section of the memory we are removing. Caller needs to make
 812  * sure that pages are marked reserved and zones are adjust properly by
 813  * calling offline_pages().
 814  */
 815 int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
 816                  unsigned long nr_pages)
 817 {
 818         unsigned long i;
 819         unsigned long map_offset = 0;
 820         int sections_to_remove, ret = 0;
 821
 822         /* In the ZONE_DEVICE case device driver owns the memory region */
 823         if (is_dev_zone(zone)) {
 824                 struct page *page = pfn_to_page(phys_start_pfn);
 825                 struct vmem_altmap *altmap;
 826
 827                 altmap = to_vmem_altmap((unsigned long) page);
 828                 if (altmap)
 829                         map_offset = vmem_altmap_offset(altmap);
 830         } else {
 831                 resource_size_t start, size;
 832
 833                 start = phys_start_pfn << PAGE_SHIFT;
 834                 size = nr_pages * PAGE_SIZE;
 835
 836                 ret = release_mem_region_adjustable(&iomem_resource, start,
 837                                         size);
 838                 if (ret) {
 839                         resource_size_t endres = start + size - 1;
 840
 841                         pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
 842                                         &start, &endres, ret);
 843                 }
 844         }
 845
 846         clear_zone_contiguous(zone);
 847
 848         /*
 849          * We can only remove entire sections
 850          */
 851         BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
 852         BUG_ON(nr_pages % PAGES_PER_SECTION);
 853
 854         sections_to_remove = nr_pages / PAGES_PER_SECTION;
 855         for (i = 0; i < sections_to_remove; i++) {
 856                 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
 857
 858                 ret = __remove_section(zone, __pfn_to_section(pfn), map_offset);
 859                 map_offset = 0;
 860                 if (ret)
 861                         break;
 862         }
 863
 864         set_zone_contiguous(zone);
 865
 866         return ret;
 867 }
 868 #endif /* CONFIG_MEMORY_HOTREMOVE */
 869
 870 int set_online_page_callback(online_page_callback_t callback)
 871 {
 872         int rc = -EINVAL;
 873
 874         get_online_mems();
 875         mutex_lock(&online_page_callback_lock);
 876
 877         if (online_page_callback == generic_online_page) {
 878                 online_page_callback = callback;
 879                 rc = 0;
 880         }
 881
 882         mutex_unlock(&online_page_callback_lock);
 883         put_online_mems();
 884
 885         return rc;
 886 }
 887 EXPORT_SYMBOL_GPL(set_online_page_callback);
 888
 889 int restore_online_page_callback(online_page_callback_t callback)
 890 {
 891         int rc = -EINVAL;
 892
 893         get_online_mems();
 894         mutex_lock(&online_page_callback_lock);
 895
 896         if (online_page_callback == callback) {
 897                 online_page_callback = generic_online_page;
 898                 rc = 0;
 899         }
 900
 901         mutex_unlock(&online_page_callback_lock);
 902         put_online_mems();
 903
 904         return rc;
 905 }
 906 EXPORT_SYMBOL_GPL(restore_online_page_callback);
 907
 908 void __online_page_set_limits(struct page *page)
 909 {
 910 }
 911 EXPORT_SYMBOL_GPL(__online_page_set_limits);
 912
 913 void __online_page_increment_counters(struct page *page)
 914 {
 915         adjust_managed_page_count(page, 1);
 916 }
 917 EXPORT_SYMBOL_GPL(__online_page_increment_counters);
 918
 919 void __online_page_free(struct page *page)
 920 {
 921         __free_reserved_page(page);
 922 }
 923 EXPORT_SYMBOL_GPL(__online_page_free);
 924
 925 static void generic_online_page(struct page *page)
 926 {
 927         __online_page_set_limits(page);
 928         __online_page_increment_counters(page);
 929         __online_page_free(page);
 930 }
 931
 932 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
 933                         void *arg)
 934 {
 935         unsigned long i;
 936         unsigned long onlined_pages = *(unsigned long *)arg;
 937         struct page *page;
 938         if (PageReserved(pfn_to_page(start_pfn)))
 939                 for (i = 0; i < nr_pages; i++) {
 940                         page = pfn_to_page(start_pfn + i);
 941                         (*online_page_callback)(page);
 942                         onlined_pages++;
 943                 }
 944         *(unsigned long *)arg = onlined_pages;
 945         return 0;
 946 }
 947
 948 #ifdef CONFIG_MOVABLE_NODE
 949 /*
 950  * When CONFIG_MOVABLE_NODE, we permit onlining of a node which doesn't have
 951  * normal memory.
 952  */
 953 static bool can_online_high_movable(struct zone *zone)
 954 {
 955         return true;
 956 }
 957 #else /* CONFIG_MOVABLE_NODE */
 958 /* ensure every online node has NORMAL memory */
 959 static bool can_online_high_movable(struct zone *zone)
 960 {
 961         return node_state(zone_to_nid(zone), N_NORMAL_MEMORY);
 962 }
 963 #endif /* CONFIG_MOVABLE_NODE */
 964
 965 /* check which state of node_states will be changed when online memory */
 966 static void node_states_check_changes_online(unsigned long nr_pages,
 967         struct zone *zone, struct memory_notify *arg)
 968 {
 969         int nid = zone_to_nid(zone);
 970         enum zone_type zone_last = ZONE_NORMAL;
 971
 972         /*
 973          * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
 974          * contains nodes which have zones of 0...ZONE_NORMAL,
 975          * set zone_last to ZONE_NORMAL.
 976          *
 977          * If we don't have HIGHMEM nor movable node,
 978          * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
 979          * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
 980          */
 981         if (N_MEMORY == N_NORMAL_MEMORY)
 982                 zone_last = ZONE_MOVABLE;
 983
 984         /*
 985          * if the memory to be online is in a zone of 0...zone_last, and
 986          * the zones of 0...zone_last don't have memory before online, we will
 987          * need to set the node to node_states[N_NORMAL_MEMORY] after
 988          * the memory is online.
 989          */
 990         if (zone_idx(zone) <= zone_last && !node_state(nid, N_NORMAL_MEMORY))
 991                 arg->status_change_nid_normal = nid;
 992         else
 993                 arg->status_change_nid_normal = -1;
 994
 995 #ifdef CONFIG_HIGHMEM
 996         /*
 997          * If we have movable node, node_states[N_HIGH_MEMORY]
 998          * contains nodes which have zones of 0...ZONE_HIGHMEM,
 999          * set zone_last to ZONE_HIGHMEM.
1000          *
1001          * If we don't have movable node, node_states[N_NORMAL_MEMORY]
1002          * contains nodes which have zones of 0...ZONE_MOVABLE,
1003          * set zone_last to ZONE_MOVABLE.
1004          */
1005         zone_last = ZONE_HIGHMEM;
1006         if (N_MEMORY == N_HIGH_MEMORY)
1007                 zone_last = ZONE_MOVABLE;
1008
1009         if (zone_idx(zone) <= zone_last && !node_state(nid, N_HIGH_MEMORY))
1010                 arg->status_change_nid_high = nid;
1011         else
1012                 arg->status_change_nid_high = -1;
1013 #else
1014         arg->status_change_nid_high = arg->status_change_nid_normal;
1015 #endif
1016
1017         /*
1018          * if the node don't have memory befor online, we will need to
1019          * set the node to node_states[N_MEMORY] after the memory
1020          * is online.
1021          */
1022         if (!node_state(nid, N_MEMORY))
1023                 arg->status_change_nid = nid;
1024         else
1025                 arg->status_change_nid = -1;
1026 }
1027
1028 static void node_states_set_node(int node, struct memory_notify *arg)
1029 {
1030         if (arg->status_change_nid_normal >= 0)
1031                 node_set_state(node, N_NORMAL_MEMORY);
1032
1033         if (arg->status_change_nid_high >= 0)
1034                 node_set_state(node, N_HIGH_MEMORY);
1035
1036         node_set_state(node, N_MEMORY);
1037 }
1038
1039 bool zone_can_shift(unsigned long pfn, unsigned long nr_pages,
1040                    enum zone_type target, int *zone_shift)
1041 {
1042         struct zone *zone = page_zone(pfn_to_page(pfn));
1043         enum zone_type idx = zone_idx(zone);
1044         int i;
1045
1046         *zone_shift = 0;
1047
1048         if (idx < target) {
1049                 /* pages must be at end of current zone */
1050                 if (pfn + nr_pages != zone_end_pfn(zone))
1051                         return false;
1052
1053                 /* no zones in use between current zone and target */
1054                 for (i = idx + 1; i < target; i++)
1055                         if (zone_is_initialized(zone - idx + i))
1056                                 return false;
1057         }
1058
1059         if (target < idx) {
1060                 /* pages must be at beginning of current zone */
1061                 if (pfn != zone->zone_start_pfn)
1062                         return false;
1063
1064                 /* no zones in use between current zone and target */
1065                 for (i = target + 1; i < idx; i++)
1066                         if (zone_is_initialized(zone - idx + i))
1067                                 return false;
1068         }
1069
1070         *zone_shift = target - idx;
1071         return true;
1072 }
1073
1074 /* Must be protected by mem_hotplug_begin() */
1075 int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type)
1076 {
1077         unsigned long flags;
1078         unsigned long onlined_pages = 0;
1079         struct zone *zone;
1080         int need_zonelists_rebuild = 0;
1081         int nid;
1082         int ret;
1083         struct memory_notify arg;
1084         int zone_shift = 0;
1085
1086         /*
1087          * This doesn't need a lock to do pfn_to_page().
1088          * The section can't be removed here because of the
1089          * memory_block->state_mutex.
1090          */
1091         zone = page_zone(pfn_to_page(pfn));
1092
1093         if ((zone_idx(zone) > ZONE_NORMAL ||
1094             online_type == MMOP_ONLINE_MOVABLE) &&
1095             !can_online_high_movable(zone))
1096                 return -EINVAL;
1097
1098         if (online_type == MMOP_ONLINE_KERNEL) {
1099                 if (!zone_can_shift(pfn, nr_pages, ZONE_NORMAL, &zone_shift))
1100                         return -EINVAL;
1101         } else if (online_type == MMOP_ONLINE_MOVABLE) {
1102                 if (!zone_can_shift(pfn, nr_pages, ZONE_MOVABLE, &zone_shift))
1103                         return -EINVAL;
1104         }
1105
1106         zone = move_pfn_range(zone_shift, pfn, pfn + nr_pages);
1107         if (!zone)
1108                 return -EINVAL;
1109
1110         arg.start_pfn = pfn;
1111         arg.nr_pages = nr_pages;
1112         node_states_check_changes_online(nr_pages, zone, &arg);
1113
1114         nid = zone_to_nid(zone);
1115
1116         ret = memory_notify(MEM_GOING_ONLINE, &arg);
1117         ret = notifier_to_errno(ret);
1118         if (ret)
1119                 goto failed_addition;
1120
1121         /*
1122          * If this zone is not populated, then it is not in zonelist.
1123          * This means the page allocator ignores this zone.
1124          * So, zonelist must be updated after online.
1125          */
1126         mutex_lock(&zonelists_mutex);
1127         if (!populated_zone(zone)) {
1128                 need_zonelists_rebuild = 1;
1129                 build_all_zonelists(NULL, zone);
1130         }
1131
1132         ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
1133                 online_pages_range);
1134         if (ret) {
1135                 if (need_zonelists_rebuild)
1136                         zone_pcp_reset(zone);
1137                 mutex_unlock(&zonelists_mutex);
1138                 goto failed_addition;
1139         }
1140
1141         zone->present_pages += onlined_pages;
1142
1143         pgdat_resize_lock(zone->zone_pgdat, &flags);
1144         zone->zone_pgdat->node_present_pages += onlined_pages;
1145         pgdat_resize_unlock(zone->zone_pgdat, &flags);
1146
1147         if (onlined_pages) {
1148                 node_states_set_node(nid, &arg);
1149                 if (need_zonelists_rebuild)
1150                         build_all_zonelists(NULL, NULL);
1151                 else
1152                         zone_pcp_update(zone);
1153         }
1154
1155         mutex_unlock(&zonelists_mutex);
1156
1157         init_per_zone_wmark_min();
1158
1159         if (onlined_pages) {
1160                 kswapd_run(nid);
1161                 kcompactd_run(nid);
1162         }
1163
1164         vm_total_pages = nr_free_pagecache_pages();
1165
1166         writeback_set_ratelimit();
1167
1168         if (onlined_pages)
1169                 memory_notify(MEM_ONLINE, &arg);
1170         return 0;
1171
1172 failed_addition:
1173         pr_debug("online_pages [mem %#010llx-%#010llx] failed\n",
1174                  (unsigned long long) pfn << PAGE_SHIFT,
1175                  (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1);
1176         memory_notify(MEM_CANCEL_ONLINE, &arg);
1177         return ret;
1178 }
1179 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
1180
1181 static void reset_node_present_pages(pg_data_t *pgdat)
1182 {
1183         struct zone *z;
1184
1185         for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
1186                 z->present_pages = 0;
1187
1188         pgdat->node_present_pages = 0;
1189 }
1190
1191 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1192 static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
1193 {
1194         struct pglist_data *pgdat;
1195         unsigned long zones_size[MAX_NR_ZONES] = {0};
1196         unsigned long zholes_size[MAX_NR_ZONES] = {0};
1197         unsigned long start_pfn = PFN_DOWN(start);
1198
1199         pgdat = NODE_DATA(nid);
1200         if (!pgdat) {
1201                 pgdat = arch_alloc_nodedata(nid);
1202                 if (!pgdat)
1203                         return NULL;
1204
1205                 arch_refresh_nodedata(nid, pgdat);
1206         } else {
1207                 /* Reset the nr_zones, order and classzone_idx before reuse */
1208                 pgdat->nr_zones = 0;
1209                 pgdat->kswapd_order = 0;
1210                 pgdat->kswapd_classzone_idx = 0;
1211         }
1212
1213         /* we can use NODE_DATA(nid) from here */
1214
1215         /* init node's zones as empty zones, we don't have any present pages.*/
1216         free_area_init_node(nid, zones_size, start_pfn, zholes_size);
1217         pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat);
1218
1219         /*
1220          * The node we allocated has no zone fallback lists. For avoiding
1221          * to access not-initialized zonelist, build here.
1222          */
1223         mutex_lock(&zonelists_mutex);
1224         build_all_zonelists(pgdat, NULL);
1225         mutex_unlock(&zonelists_mutex);
1226
1227         /*
1228          * zone->managed_pages is set to an approximate value in
1229          * free_area_init_core(), which will cause
1230          * /sys/device/system/node/nodeX/meminfo has wrong data.
1231          * So reset it to 0 before any memory is onlined.
1232          */
1233         reset_node_managed_pages(pgdat);
1234
1235         /*
1236          * When memory is hot-added, all the memory is in offline state. So
1237          * clear all zones' present_pages because they will be updated in
1238          * online_pages() and offline_pages().
1239          */
1240         reset_node_present_pages(pgdat);
1241
1242         return pgdat;
1243 }
1244
1245 static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
1246 {
1247         arch_refresh_nodedata(nid, NULL);
1248         free_percpu(pgdat->per_cpu_nodestats);
1249         arch_free_nodedata(pgdat);
1250         return;
1251 }
1252
1253
1254 /**
1255  * try_online_node - online a node if offlined
1256  *
1257  * called by cpu_up() to online a node without onlined memory.
1258  */
1259 int try_online_node(int nid)
1260 {
1261         pg_data_t       *pgdat;
1262         int     ret;
1263
1264         if (node_online(nid))
1265                 return 0;
1266
1267         mem_hotplug_begin();
1268         pgdat = hotadd_new_pgdat(nid, 0);
1269         if (!pgdat) {
1270                 pr_err("Cannot online node %d due to NULL pgdat\n", nid);
1271                 ret = -ENOMEM;
1272                 goto out;
1273         }
1274         node_set_online(nid);
1275         ret = register_one_node(nid);
1276         BUG_ON(ret);
1277
1278         if (pgdat->node_zonelists->_zonerefs->zone == NULL) {
1279                 mutex_lock(&zonelists_mutex);
1280                 build_all_zonelists(NULL, NULL);
1281                 mutex_unlock(&zonelists_mutex);
1282         }
1283
1284 out:
1285         mem_hotplug_done();
1286         return ret;
1287 }
1288
1289 static int check_hotplug_memory_range(u64 start, u64 size)
1290 {
1291         u64 start_pfn = PFN_DOWN(start);
1292         u64 nr_pages = size >> PAGE_SHIFT;
1293
1294         /* Memory range must be aligned with section */
1295         if ((start_pfn & ~PAGE_SECTION_MASK) ||
1296             (nr_pages % PAGES_PER_SECTION) || (!nr_pages)) {
1297                 pr_err("Section-unaligned hotplug range: start 0x%llx, size 0x%llx\n",
1298                                 (unsigned long long)start,
1299                                 (unsigned long long)size);
1300                 return -EINVAL;
1301         }
1302
1303         return 0;
1304 }
1305
1306 /*
1307  * If movable zone has already been setup, newly added memory should be check.
1308  * If its address is higher than movable zone, it should be added as movable.
1309  * Without this check, movable zone may overlap with other zone.
1310  */
1311 static int should_add_memory_movable(int nid, u64 start, u64 size)
1312 {
1313         unsigned long start_pfn = start >> PAGE_SHIFT;
1314         pg_data_t *pgdat = NODE_DATA(nid);
1315         struct zone *movable_zone = pgdat->node_zones + ZONE_MOVABLE;
1316
1317         if (zone_is_empty(movable_zone))
1318                 return 0;
1319
1320         if (movable_zone->zone_start_pfn <= start_pfn)
1321                 return 1;
1322
1323         return 0;
1324 }
1325
1326 int zone_for_memory(int nid, u64 start, u64 size, int zone_default,
1327                 bool for_device)
1328 {
1329 #ifdef CONFIG_ZONE_DEVICE
1330         if (for_device)
1331                 return ZONE_DEVICE;
1332 #endif
1333         if (should_add_memory_movable(nid, start, size))
1334                 return ZONE_MOVABLE;
1335
1336         return zone_default;
1337 }
1338
1339 static int online_memory_block(struct memory_block *mem, void *arg)
1340 {
1341         return device_online(&mem->dev);
1342 }
1343
1344 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1345 int __ref add_memory_resource(int nid, struct resource *res, bool online)
1346 {
1347         u64 start, size;
1348         pg_data_t *pgdat = NULL;
1349         bool new_pgdat;
1350         bool new_node;
1351         int ret;
1352
1353         start = res->start;
1354         size = resource_size(res);
1355
1356         ret = check_hotplug_memory_range(start, size);
1357         if (ret)
1358                 return ret;
1359
1360         {       /* Stupid hack to suppress address-never-null warning */
1361                 void *p = NODE_DATA(nid);
1362                 new_pgdat = !p;
1363         }
1364
1365         mem_hotplug_begin();
1366
1367         /*
1368          * Add new range to memblock so that when hotadd_new_pgdat() is called
1369          * to allocate new pgdat, get_pfn_range_for_nid() will be able to find
1370          * this new range and calculate total pages correctly.  The range will
1371          * be removed at hot-remove time.
1372          */
1373         memblock_add_node(start, size, nid);
1374
1375         new_node = !node_online(nid);
1376         if (new_node) {
1377                 pgdat = hotadd_new_pgdat(nid, start);
1378                 ret = -ENOMEM;
1379                 if (!pgdat)
1380                         goto error;
1381         }
1382
1383         /* call arch's memory hotadd */
1384         ret = arch_add_memory(nid, start, size, false);
1385
1386         if (ret < 0)
1387                 goto error;
1388
1389         /* we online node here. we can't roll back from here. */
1390         node_set_online(nid);
1391
1392         if (new_node) {
1393                 ret = register_one_node(nid);
1394                 /*
1395                  * If sysfs file of new node can't create, cpu on the node
1396                  * can't be hot-added. There is no rollback way now.
1397                  * So, check by BUG_ON() to catch it reluctantly..
1398                  */
1399                 BUG_ON(ret);
1400         }
1401
1402         /* create new memmap entry */
1403         firmware_map_add_hotplug(start, start + size, "System RAM");
1404
1405         /* online pages if requested */
1406         if (online)
1407                 walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1),
1408                                   NULL, online_memory_block);
1409
1410         goto out;
1411
1412 error:
1413         /* rollback pgdat allocation and others */
1414         if (new_pgdat)
1415                 rollback_node_hotadd(nid, pgdat);
1416         memblock_remove(start, size);
1417
1418 out:
1419         mem_hotplug_done();
1420         return ret;
1421 }
1422 EXPORT_SYMBOL_GPL(add_memory_resource);
1423
1424 int __ref add_memory(int nid, u64 start, u64 size)
1425 {
1426         struct resource *res;
1427         int ret;
1428
1429         res = register_memory_resource(start, size);
1430         if (IS_ERR(res))
1431                 return PTR_ERR(res);
1432
1433         ret = add_memory_resource(nid, res, memhp_auto_online);
1434         if (ret < 0)
1435                 release_memory_resource(res);
1436         return ret;
1437 }
1438 EXPORT_SYMBOL_GPL(add_memory);
1439
1440 #ifdef CONFIG_MEMORY_HOTREMOVE
1441 /*
1442  * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1443  * set and the size of the free page is given by page_order(). Using this,
1444  * the function determines if the pageblock contains only free pages.
1445  * Due to buddy contraints, a free page at least the size of a pageblock will
1446  * be located at the start of the pageblock
1447  */
1448 static inline int pageblock_free(struct page *page)
1449 {
1450         return PageBuddy(page) && page_order(page) >= pageblock_order;
1451 }
1452
1453 /* Return the start of the next active pageblock after a given page */
1454 static struct page *next_active_pageblock(struct page *page)
1455 {
1456         /* Ensure the starting page is pageblock-aligned */
1457         BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
1458
1459         /* If the entire pageblock is free, move to the end of free page */
1460         if (pageblock_free(page)) {
1461                 int order;
1462                 /* be careful. we don't have locks, page_order can be changed.*/
1463                 order = page_order(page);
1464                 if ((order < MAX_ORDER) && (order >= pageblock_order))
1465                         return page + (1 << order);
1466         }
1467
1468         return page + pageblock_nr_pages;
1469 }
1470
1471 /* Checks if this range of memory is likely to be hot-removable. */
1472 bool is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
1473 {
1474         struct page *page = pfn_to_page(start_pfn);
1475         struct page *end_page = page + nr_pages;
1476
1477         /* Check the starting page of each pageblock within the range */
1478         for (; page < end_page; page = next_active_pageblock(page)) {
1479                 if (!is_pageblock_removable_nolock(page))
1480                         return false;
1481                 cond_resched();
1482         }
1483
1484         /* All pageblocks in the memory block are likely to be hot-removable */
1485         return true;
1486 }
1487
1488 /*
1489  * Confirm all pages in a range [start, end) belong to the same zone.
1490  * When true, return its valid [start, end).
1491  */
1492 int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn,
1493                          unsigned long *valid_start, unsigned long *valid_end)
1494 {
1495         unsigned long pfn, sec_end_pfn;
1496         unsigned long start, end;
1497         struct zone *zone = NULL;
1498         struct page *page;
1499         int i;
1500         for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1);
1501              pfn < end_pfn;
1502              pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) {
1503                 /* Make sure the memory section is present first */
1504                 if (!present_section_nr(pfn_to_section_nr(pfn)))
1505                         continue;
1506                 for (; pfn < sec_end_pfn && pfn < end_pfn;
1507                      pfn += MAX_ORDER_NR_PAGES) {
1508                         i = 0;
1509                         /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1510                         while ((i < MAX_ORDER_NR_PAGES) &&
1511                                 !pfn_valid_within(pfn + i))
1512                                 i++;
1513                         if (i == MAX_ORDER_NR_PAGES || pfn + i >= end_pfn)
1514                                 continue;
1515                         page = pfn_to_page(pfn + i);
1516                         if (zone && page_zone(page) != zone)
1517                                 return 0;
1518                         if (!zone)
1519                                 start = pfn + i;
1520                         zone = page_zone(page);
1521                         end = pfn + MAX_ORDER_NR_PAGES;
1522                 }
1523         }
1524
1525         if (zone) {
1526                 *valid_start = start;
1527                 *valid_end = min(end, end_pfn);
1528                 return 1;
1529         } else {
1530                 return 0;
1531         }
1532 }
1533
1534 /*
1535  * Scan pfn range [start,end) to find movable/migratable pages (LRU pages,
1536  * non-lru movable pages and hugepages). We scan pfn because it's much
1537  * easier than scanning over linked list. This function returns the pfn
1538  * of the first found movable page if it's found, otherwise 0.
1539  */
1540 static unsigned long scan_movable_pages(unsigned long start, unsigned long end)
1541 {
1542         unsigned long pfn;
1543         struct page *page;
1544         for (pfn = start; pfn < end; pfn++) {
1545                 if (pfn_valid(pfn)) {
1546                         page = pfn_to_page(pfn);
1547                         if (PageLRU(page))
1548                                 return pfn;
1549                         if (__PageMovable(page))
1550                                 return pfn;
1551                         if (PageHuge(page)) {
1552                                 if (page_huge_active(page))
1553                                         return pfn;
1554                                 else
1555                                         pfn = round_up(pfn + 1,
1556                                                 1 << compound_order(page)) - 1;
1557                         }
1558                 }
1559         }
1560         return 0;
1561 }
1562
1563 static struct page *new_node_page(struct page *page, unsigned long private,
1564                 int **result)
1565 {
1566         gfp_t gfp_mask = GFP_USER | __GFP_MOVABLE;
1567         int nid = page_to_nid(page);
1568         nodemask_t nmask = node_states[N_MEMORY];
1569         struct page *new_page = NULL;
1570
1571         /*
1572          * TODO: allocate a destination hugepage from a nearest neighbor node,
1573          * accordance with memory policy of the user process if possible. For
1574          * now as a simple work-around, we use the next node for destination.
1575          */
1576         if (PageHuge(page))
1577                 return alloc_huge_page_node(page_hstate(compound_head(page)),
1578                                         next_node_in(nid, nmask));
1579
1580         node_clear(nid, nmask);
1581
1582         if (PageHighMem(page)
1583             || (zone_idx(page_zone(page)) == ZONE_MOVABLE))
1584                 gfp_mask |= __GFP_HIGHMEM;
1585
1586         if (!nodes_empty(nmask))
1587                 new_page = __alloc_pages_nodemask(gfp_mask, 0,
1588                                         node_zonelist(nid, gfp_mask), &nmask);
1589         if (!new_page)
1590                 new_page = __alloc_pages(gfp_mask, 0,
1591                                         node_zonelist(nid, gfp_mask));
1592
1593         return new_page;
1594 }
1595
1596 #define NR_OFFLINE_AT_ONCE_PAGES        (256)
1597 static int
1598 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1599 {
1600         unsigned long pfn;
1601         struct page *page;
1602         int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
1603         int not_managed = 0;
1604         int ret = 0;
1605         LIST_HEAD(source);
1606
1607         for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
1608                 if (!pfn_valid(pfn))
1609                         continue;
1610                 page = pfn_to_page(pfn);
1611
1612                 if (PageHuge(page)) {
1613                         struct page *head = compound_head(page);
1614                         pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1;
1615                         if (compound_order(head) > PFN_SECTION_SHIFT) {
1616                                 ret = -EBUSY;
1617                                 break;
1618                         }
1619                         if (isolate_huge_page(page, &source))
1620                                 move_pages -= 1 << compound_order(head);
1621                         continue;
1622                 }
1623
1624                 if (!get_page_unless_zero(page))
1625                         continue;
1626                 /*
1627                  * We can skip free pages. And we can deal with pages on
1628                  * LRU and non-lru movable pages.
1629                  */
1630                 if (PageLRU(page))
1631                         ret = isolate_lru_page(page);
1632                 else
1633                         ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE);
1634                 if (!ret) { /* Success */
1635                         put_page(page);
1636                         list_add_tail(&page->lru, &source);
1637                         move_pages--;
1638                         if (!__PageMovable(page))
1639                                 inc_node_page_state(page, NR_ISOLATED_ANON +
1640                                                     page_is_file_cache(page));
1641
1642                 } else {
1643 #ifdef CONFIG_DEBUG_VM
1644                         pr_alert("failed to isolate pfn %lx\n", pfn);
1645                         dump_page(page, "isolation failed");
1646 #endif
1647                         put_page(page);
1648                         /* Because we don't have big zone->lock. we should
1649                            check this again here. */
1650                         if (page_count(page)) {
1651                                 not_managed++;
1652                                 ret = -EBUSY;
1653                                 break;
1654                         }
1655                 }
1656         }
1657         if (!list_empty(&source)) {
1658                 if (not_managed) {
1659                         putback_movable_pages(&source);
1660                         goto out;
1661                 }
1662
1663                 /* Allocate a new page from the nearest neighbor node */
1664                 ret = migrate_pages(&source, new_node_page, NULL, 0,
1665                                         MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
1666                 if (ret)
1667                         putback_movable_pages(&source);
1668         }
1669 out:
1670         return ret;
1671 }
1672
1673 /*
1674  * remove from free_area[] and mark all as Reserved.
1675  */
1676 static int
1677 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1678                         void *data)
1679 {
1680         __offline_isolated_pages(start, start + nr_pages);
1681         return 0;
1682 }
1683
1684 static void
1685 offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
1686 {
1687         walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
1688                                 offline_isolated_pages_cb);
1689 }
1690
1691 /*
1692  * Check all pages in range, recoreded as memory resource, are isolated.
1693  */
1694 static int
1695 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1696                         void *data)
1697 {
1698         int ret;
1699         long offlined = *(long *)data;
1700         ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true);
1701         offlined = nr_pages;
1702         if (!ret)
1703                 *(long *)data += offlined;
1704         return ret;
1705 }
1706
1707 static long
1708 check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
1709 {
1710         long offlined = 0;
1711         int ret;
1712
1713         ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
1714                         check_pages_isolated_cb);
1715         if (ret < 0)
1716                 offlined = (long)ret;
1717         return offlined;
1718 }
1719
1720 #ifdef CONFIG_MOVABLE_NODE
1721 /*
1722  * When CONFIG_MOVABLE_NODE, we permit offlining of a node which doesn't have
1723  * normal memory.
1724  */
1725 static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1726 {
1727         return true;
1728 }
1729 #else /* CONFIG_MOVABLE_NODE */
1730 /* ensure the node has NORMAL memory if it is still online */
1731 static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1732 {
1733         struct pglist_data *pgdat = zone->zone_pgdat;
1734         unsigned long present_pages = 0;
1735         enum zone_type zt;
1736
1737         for (zt = 0; zt <= ZONE_NORMAL; zt++)
1738                 present_pages += pgdat->node_zones[zt].present_pages;
1739
1740         if (present_pages > nr_pages)
1741                 return true;
1742
1743         present_pages = 0;
1744         for (; zt <= ZONE_MOVABLE; zt++)
1745                 present_pages += pgdat->node_zones[zt].present_pages;
1746
1747         /*
1748          * we can't offline the last normal memory until all
1749          * higher memory is offlined.
1750          */
1751         return present_pages == 0;
1752 }
1753 #endif /* CONFIG_MOVABLE_NODE */
1754
1755 static int __init cmdline_parse_movable_node(char *p)
1756 {
1757 #ifdef CONFIG_MOVABLE_NODE
1758         movable_node_enabled = true;
1759 #else
1760         pr_warn("movable_node option not supported\n");
1761 #endif
1762         return 0;
1763 }
1764 early_param("movable_node", cmdline_parse_movable_node);
1765
1766 /* check which state of node_states will be changed when offline memory */
1767 static void node_states_check_changes_offline(unsigned long nr_pages,
1768                 struct zone *zone, struct memory_notify *arg)
1769 {
1770         struct pglist_data *pgdat = zone->zone_pgdat;
1771         unsigned long present_pages = 0;
1772         enum zone_type zt, zone_last = ZONE_NORMAL;
1773
1774         /*
1775          * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
1776          * contains nodes which have zones of 0...ZONE_NORMAL,
1777          * set zone_last to ZONE_NORMAL.
1778          *
1779          * If we don't have HIGHMEM nor movable node,
1780          * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
1781          * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
1782          */
1783         if (N_MEMORY == N_NORMAL_MEMORY)
1784                 zone_last = ZONE_MOVABLE;
1785
1786         /*
1787          * check whether node_states[N_NORMAL_MEMORY] will be changed.
1788          * If the memory to be offline is in a zone of 0...zone_last,
1789          * and it is the last present memory, 0...zone_last will
1790          * become empty after offline , thus we can determind we will
1791          * need to clear the node from node_states[N_NORMAL_MEMORY].
1792          */
1793         for (zt = 0; zt <= zone_last; zt++)
1794                 present_pages += pgdat->node_zones[zt].present_pages;
1795         if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1796                 arg->status_change_nid_normal = zone_to_nid(zone);
1797         else
1798                 arg->status_change_nid_normal = -1;
1799
1800 #ifdef CONFIG_HIGHMEM
1801         /*
1802          * If we have movable node, node_states[N_HIGH_MEMORY]
1803          * contains nodes which have zones of 0...ZONE_HIGHMEM,
1804          * set zone_last to ZONE_HIGHMEM.
1805          *
1806          * If we don't have movable node, node_states[N_NORMAL_MEMORY]
1807          * contains nodes which have zones of 0...ZONE_MOVABLE,
1808          * set zone_last to ZONE_MOVABLE.
1809          */
1810         zone_last = ZONE_HIGHMEM;
1811         if (N_MEMORY == N_HIGH_MEMORY)
1812                 zone_last = ZONE_MOVABLE;
1813
1814         for (; zt <= zone_last; zt++)
1815                 present_pages += pgdat->node_zones[zt].present_pages;
1816         if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1817                 arg->status_change_nid_high = zone_to_nid(zone);
1818         else
1819                 arg->status_change_nid_high = -1;
1820 #else
1821         arg->status_change_nid_high = arg->status_change_nid_normal;
1822 #endif
1823
1824         /*
1825          * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE
1826          */
1827         zone_last = ZONE_MOVABLE;
1828
1829         /*
1830          * check whether node_states[N_HIGH_MEMORY] will be changed
1831          * If we try to offline the last present @nr_pages from the node,
1832          * we can determind we will need to clear the node from
1833          * node_states[N_HIGH_MEMORY].
1834          */
1835         for (; zt <= zone_last; zt++)
1836                 present_pages += pgdat->node_zones[zt].present_pages;
1837         if (nr_pages >= present_pages)
1838                 arg->status_change_nid = zone_to_nid(zone);
1839         else
1840                 arg->status_change_nid = -1;
1841 }
1842
1843 static void node_states_clear_node(int node, struct memory_notify *arg)
1844 {
1845         if (arg->status_change_nid_normal >= 0)
1846                 node_clear_state(node, N_NORMAL_MEMORY);
1847
1848         if ((N_MEMORY != N_NORMAL_MEMORY) &&
1849             (arg->status_change_nid_high >= 0))
1850                 node_clear_state(node, N_HIGH_MEMORY);
1851
1852         if ((N_MEMORY != N_HIGH_MEMORY) &&
1853             (arg->status_change_nid >= 0))
1854                 node_clear_state(node, N_MEMORY);
1855 }
1856
1857 static int __ref __offline_pages(unsigned long start_pfn,
1858                   unsigned long end_pfn, unsigned long timeout)
1859 {
1860         unsigned long pfn, nr_pages, expire;
1861         long offlined_pages;
1862         int ret, drain, retry_max, node;
1863         unsigned long flags;
1864         unsigned long valid_start, valid_end;
1865         struct zone *zone;
1866         struct memory_notify arg;
1867
1868         /* at least, alignment against pageblock is necessary */
1869         if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
1870                 return -EINVAL;
1871         if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
1872                 return -EINVAL;
1873         /* This makes hotplug much easier...and readable.
1874            we assume this for now. .*/
1875         if (!test_pages_in_a_zone(start_pfn, end_pfn, &valid_start, &valid_end))
1876                 return -EINVAL;
1877
1878         zone = page_zone(pfn_to_page(valid_start));
1879         node = zone_to_nid(zone);
1880         nr_pages = end_pfn - start_pfn;
1881
1882         if (zone_idx(zone) <= ZONE_NORMAL && !can_offline_normal(zone, nr_pages))
1883                 return -EINVAL;
1884
1885         /* set above range as isolated */
1886         ret = start_isolate_page_range(start_pfn, end_pfn,
1887                                        MIGRATE_MOVABLE, true);
1888         if (ret)
1889                 return ret;
1890
1891         arg.start_pfn = start_pfn;
1892         arg.nr_pages = nr_pages;
1893         node_states_check_changes_offline(nr_pages, zone, &arg);
1894
1895         ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1896         ret = notifier_to_errno(ret);
1897         if (ret)
1898                 goto failed_removal;
1899
1900         pfn = start_pfn;
1901         expire = jiffies + timeout;
1902         drain = 0;
1903         retry_max = 5;
1904 repeat:
1905         /* start memory hot removal */
1906         ret = -EAGAIN;
1907         if (time_after(jiffies, expire))
1908                 goto failed_removal;
1909         ret = -EINTR;
1910         if (signal_pending(current))
1911                 goto failed_removal;
1912         ret = 0;
1913         if (drain) {
1914                 lru_add_drain_all();
1915                 cond_resched();
1916                 drain_all_pages(zone);
1917         }
1918
1919         pfn = scan_movable_pages(start_pfn, end_pfn);
1920         if (pfn) { /* We have movable pages */
1921                 ret = do_migrate_range(pfn, end_pfn);
1922                 if (!ret) {
1923                         drain = 1;
1924                         goto repeat;
1925                 } else {
1926                         if (ret < 0)
1927                                 if (--retry_max == 0)
1928                                         goto failed_removal;
1929                         yield();
1930                         drain = 1;
1931                         goto repeat;
1932                 }
1933         }
1934         /* drain all zone's lru pagevec, this is asynchronous... */
1935         lru_add_drain_all();
1936         yield();
1937         /* drain pcp pages, this is synchronous. */
1938         drain_all_pages(zone);
1939         /*
1940          * dissolve free hugepages in the memory block before doing offlining
1941          * actually in order to make hugetlbfs's object counting consistent.
1942          */
1943         ret = dissolve_free_huge_pages(start_pfn, end_pfn);
1944         if (ret)
1945                 goto failed_removal;
1946         /* check again */
1947         offlined_pages = check_pages_isolated(start_pfn, end_pfn);
1948         if (offlined_pages < 0) {
1949                 ret = -EBUSY;
1950                 goto failed_removal;
1951         }
1952         pr_info("Offlined Pages %ld\n", offlined_pages);
1953         /* Ok, all of our target is isolated.
1954            We cannot do rollback at this point. */
1955         offline_isolated_pages(start_pfn, end_pfn);
1956         /* reset pagetype flags and makes migrate type to be MOVABLE */
1957         undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1958         /* removal success */
1959         adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages);
1960         zone->present_pages -= offlined_pages;
1961
1962         pgdat_resize_lock(zone->zone_pgdat, &flags);
1963         zone->zone_pgdat->node_present_pages -= offlined_pages;
1964         pgdat_resize_unlock(zone->zone_pgdat, &flags);
1965
1966         init_per_zone_wmark_min();
1967
1968         if (!populated_zone(zone)) {
1969                 zone_pcp_reset(zone);
1970                 mutex_lock(&zonelists_mutex);
1971                 build_all_zonelists(NULL, NULL);
1972                 mutex_unlock(&zonelists_mutex);
1973         } else
1974                 zone_pcp_update(zone);
1975
1976         node_states_clear_node(node, &arg);
1977         if (arg.status_change_nid >= 0) {
1978                 kswapd_stop(node);
1979                 kcompactd_stop(node);
1980         }
1981
1982         vm_total_pages = nr_free_pagecache_pages();
1983         writeback_set_ratelimit();
1984
1985         memory_notify(MEM_OFFLINE, &arg);
1986         return 0;
1987
1988 failed_removal:
1989         pr_debug("memory offlining [mem %#010llx-%#010llx] failed\n",
1990                  (unsigned long long) start_pfn << PAGE_SHIFT,
1991                  ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
1992         memory_notify(MEM_CANCEL_OFFLINE, &arg);
1993         /* pushback to free area */
1994         undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1995         return ret;
1996 }
1997
1998 /* Must be protected by mem_hotplug_begin() */
1999 int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
2000 {
2001         return __offline_pages(start_pfn, start_pfn + nr_pages, 120 * HZ);
2002 }
2003 #endif /* CONFIG_MEMORY_HOTREMOVE */
2004
2005 /**
2006  * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
2007  * @start_pfn: start pfn of the memory range
2008  * @end_pfn: end pfn of the memory range
2009  * @arg: argument passed to func
2010  * @func: callback for each memory section walked
2011  *
2012  * This function walks through all present mem sections in range
2013  * [start_pfn, end_pfn) and call func on each mem section.
2014  *
2015  * Returns the return value of func.
2016  */
2017 int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn,
2018                 void *arg, int (*func)(struct memory_block *, void *))
2019 {
2020         struct memory_block *mem = NULL;
2021         struct mem_section *section;
2022         unsigned long pfn, section_nr;
2023         int ret;
2024
2025         for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
2026                 section_nr = pfn_to_section_nr(pfn);
2027                 if (!present_section_nr(section_nr))
2028                         continue;
2029
2030                 section = __nr_to_section(section_nr);
2031                 /* same memblock? */
2032                 if (mem)
2033                         if ((section_nr >= mem->start_section_nr) &&
2034                             (section_nr <= mem->end_section_nr))
2035                                 continue;
2036
2037                 mem = find_memory_block_hinted(section, mem);
2038                 if (!mem)
2039                         continue;
2040
2041                 ret = func(mem, arg);
2042                 if (ret) {
2043                         kobject_put(&mem->dev.kobj);
2044                         return ret;
2045                 }
2046         }
2047
2048         if (mem)
2049                 kobject_put(&mem->dev.kobj);
2050
2051         return 0;
2052 }
2053
2054 #ifdef CONFIG_MEMORY_HOTREMOVE
2055 static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
2056 {
2057         int ret = !is_memblock_offlined(mem);
2058
2059         if (unlikely(ret)) {
2060                 phys_addr_t beginpa, endpa;
2061
2062                 beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
2063                 endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1;
2064                 pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n",
2065                         &beginpa, &endpa);
2066         }
2067
2068         return ret;
2069 }
2070
2071 static int check_cpu_on_node(pg_data_t *pgdat)
2072 {
2073         int cpu;
2074
2075         for_each_present_cpu(cpu) {
2076                 if (cpu_to_node(cpu) == pgdat->node_id)
2077                         /*
2078                          * the cpu on this node isn't removed, and we can't
2079                          * offline this node.
2080                          */
2081                         return -EBUSY;
2082         }
2083
2084         return 0;
2085 }
2086
2087 static void unmap_cpu_on_node(pg_data_t *pgdat)
2088 {
2089 #ifdef CONFIG_ACPI_NUMA
2090         int cpu;
2091
2092         for_each_possible_cpu(cpu)
2093                 if (cpu_to_node(cpu) == pgdat->node_id)
2094                         numa_clear_node(cpu);
2095 #endif
2096 }
2097
2098 static int check_and_unmap_cpu_on_node(pg_data_t *pgdat)
2099 {
2100         int ret;
2101
2102         ret = check_cpu_on_node(pgdat);
2103         if (ret)
2104                 return ret;
2105
2106         /*
2107          * the node will be offlined when we come here, so we can clear
2108          * the cpu_to_node() now.
2109          */
2110
2111         unmap_cpu_on_node(pgdat);
2112         return 0;
2113 }
2114
2115 /**
2116  * try_offline_node
2117  *
2118  * Offline a node if all memory sections and cpus of the node are removed.
2119  *
2120  * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
2121  * and online/offline operations before this call.
2122  */
2123 void try_offline_node(int nid)
2124 {
2125         pg_data_t *pgdat = NODE_DATA(nid);
2126         unsigned long start_pfn = pgdat->node_start_pfn;
2127         unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
2128         unsigned long pfn;
2129
2130         for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
2131                 unsigned long section_nr = pfn_to_section_nr(pfn);
2132
2133                 if (!present_section_nr(section_nr))
2134                         continue;
2135
2136                 if (pfn_to_nid(pfn) != nid)
2137                         continue;
2138
2139                 /*
2140                  * some memory sections of this node are not removed, and we
2141                  * can't offline node now.
2142                  */
2143                 return;
2144         }
2145
2146         if (check_and_unmap_cpu_on_node(pgdat))
2147                 return;
2148
2149         /*
2150          * all memory/cpu of this node are removed, we can offline this
2151          * node now.
2152          */
2153         node_set_offline(nid);
2154         unregister_one_node(nid);
2155 }
2156 EXPORT_SYMBOL(try_offline_node);
2157
2158 /**
2159  * remove_memory
2160  *
2161  * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
2162  * and online/offline operations before this call, as required by
2163  * try_offline_node().
2164  */
2165 void __ref remove_memory(int nid, u64 start, u64 size)
2166 {
2167         int ret;
2168
2169         BUG_ON(check_hotplug_memory_range(start, size));
2170
2171         mem_hotplug_begin();
2172
2173         /*
2174          * All memory blocks must be offlined before removing memory.  Check
2175          * whether all memory blocks in question are offline and trigger a BUG()
2176          * if this is not the case.
2177          */
2178         ret = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL,
2179                                 check_memblock_offlined_cb);
2180         if (ret)
2181                 BUG();
2182
2183         /* remove memmap entry */
2184         firmware_map_remove(start, start + size, "System RAM");
2185         memblock_free(start, size);
2186         memblock_remove(start, size);
2187
2188         arch_remove_memory(start, size);
2189
2190         try_offline_node(nid);
2191
2192         mem_hotplug_done();
2193 }
2194 EXPORT_SYMBOL_GPL(remove_memory);
2195 #endif /* CONFIG_MEMORY_HOTREMOVE */