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 /* Serializes write accesses to mem_hotplug.active_writer. */
 129 static DEFINE_MUTEX(memory_add_remove_lock);
 130
 131 void mem_hotplug_begin(void)
 132 {
 133         mutex_lock(&memory_add_remove_lock);
 134
 135         mem_hotplug.active_writer = current;
 136
 137         memhp_lock_acquire();
 138         for (;;) {
 139                 mutex_lock(&mem_hotplug.lock);
 140                 if (likely(!mem_hotplug.refcount))
 141                         break;
 142                 __set_current_state(TASK_UNINTERRUPTIBLE);
 143                 mutex_unlock(&mem_hotplug.lock);
 144                 schedule();
 145         }
 146 }
 147
 148 void mem_hotplug_done(void)
 149 {
 150         mem_hotplug.active_writer = NULL;
 151         mutex_unlock(&mem_hotplug.lock);
 152         memhp_lock_release();
 153         mutex_unlock(&memory_add_remove_lock);
 154 }
 155
 156 /* add this memory to iomem resource */
 157 static struct resource *register_memory_resource(u64 start, u64 size)
 158 {
 159         struct resource *res;
 160         res = kzalloc(sizeof(struct resource), GFP_KERNEL);
 161         if (!res)
 162                 return ERR_PTR(-ENOMEM);
 163
 164         res->name = "System RAM";
 165         res->start = start;
 166         res->end = start + size - 1;
 167         res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
 168         if (request_resource(&iomem_resource, res) < 0) {
 169                 pr_debug("System RAM resource %pR cannot be added\n", res);
 170                 kfree(res);
 171                 return ERR_PTR(-EEXIST);
 172         }
 173         return res;
 174 }
 175
 176 static void release_memory_resource(struct resource *res)
 177 {
 178         if (!res)
 179                 return;
 180         release_resource(res);
 181         kfree(res);
 182         return;
 183 }
 184
 185 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
 186 void get_page_bootmem(unsigned long info,  struct page *page,
 187                       unsigned long type)
 188 {
 189         page->freelist = (void *)type;
 190         SetPagePrivate(page);
 191         set_page_private(page, info);
 192         page_ref_inc(page);
 193 }
 194
 195 void put_page_bootmem(struct page *page)
 196 {
 197         unsigned long type;
 198
 199         type = (unsigned long) page->freelist;
 200         BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE ||
 201                type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE);
 202
 203         if (page_ref_dec_return(page) == 1) {
 204                 page->freelist = NULL;
 205                 ClearPagePrivate(page);
 206                 set_page_private(page, 0);
 207                 INIT_LIST_HEAD(&page->lru);
 208                 free_reserved_page(page);
 209         }
 210 }
 211
 212 #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
 213 #ifndef CONFIG_SPARSEMEM_VMEMMAP
 214 static void register_page_bootmem_info_section(unsigned long start_pfn)
 215 {
 216         unsigned long *usemap, mapsize, section_nr, i;
 217         struct mem_section *ms;
 218         struct page *page, *memmap;
 219
 220         section_nr = pfn_to_section_nr(start_pfn);
 221         ms = __nr_to_section(section_nr);
 222
 223         /* Get section's memmap address */
 224         memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
 225
 226         /*
 227          * Get page for the memmap's phys address
 228          * XXX: need more consideration for sparse_vmemmap...
 229          */
 230         page = virt_to_page(memmap);
 231         mapsize = sizeof(struct page) * PAGES_PER_SECTION;
 232         mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
 233
 234         /* remember memmap's page */
 235         for (i = 0; i < mapsize; i++, page++)
 236                 get_page_bootmem(section_nr, page, SECTION_INFO);
 237
 238         usemap = __nr_to_section(section_nr)->pageblock_flags;
 239         page = virt_to_page(usemap);
 240
 241         mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
 242
 243         for (i = 0; i < mapsize; i++, page++)
 244                 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
 245
 246 }
 247 #else /* CONFIG_SPARSEMEM_VMEMMAP */
 248 static void register_page_bootmem_info_section(unsigned long start_pfn)
 249 {
 250         unsigned long *usemap, mapsize, section_nr, i;
 251         struct mem_section *ms;
 252         struct page *page, *memmap;
 253
 254         if (!pfn_valid(start_pfn))
 255                 return;
 256
 257         section_nr = pfn_to_section_nr(start_pfn);
 258         ms = __nr_to_section(section_nr);
 259
 260         memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
 261
 262         register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION);
 263
 264         usemap = __nr_to_section(section_nr)->pageblock_flags;
 265         page = virt_to_page(usemap);
 266
 267         mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
 268
 269         for (i = 0; i < mapsize; i++, page++)
 270                 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
 271 }
 272 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
 273
 274 void __init register_page_bootmem_info_node(struct pglist_data *pgdat)
 275 {
 276         unsigned long i, pfn, end_pfn, nr_pages;
 277         int node = pgdat->node_id;
 278         struct page *page;
 279
 280         nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
 281         page = virt_to_page(pgdat);
 282
 283         for (i = 0; i < nr_pages; i++, page++)
 284                 get_page_bootmem(node, page, NODE_INFO);
 285
 286         pfn = pgdat->node_start_pfn;
 287         end_pfn = pgdat_end_pfn(pgdat);
 288
 289         /* register section info */
 290         for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
 291                 /*
 292                  * Some platforms can assign the same pfn to multiple nodes - on
 293                  * node0 as well as nodeN.  To avoid registering a pfn against
 294                  * multiple nodes we check that this pfn does not already
 295                  * reside in some other nodes.
 296                  */
 297                 if (pfn_valid(pfn) && (early_pfn_to_nid(pfn) == node))
 298                         register_page_bootmem_info_section(pfn);
 299         }
 300 }
 301 #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
 302
 303 static void __meminit grow_zone_span(struct zone *zone, unsigned long start_pfn,
 304                                      unsigned long end_pfn)
 305 {
 306         unsigned long old_zone_end_pfn;
 307
 308         zone_span_writelock(zone);
 309
 310         old_zone_end_pfn = zone_end_pfn(zone);
 311         if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
 312                 zone->zone_start_pfn = start_pfn;
 313
 314         zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
 315                                 zone->zone_start_pfn;
 316
 317         zone_span_writeunlock(zone);
 318 }
 319
 320 static void resize_zone(struct zone *zone, unsigned long start_pfn,
 321                 unsigned long end_pfn)
 322 {
 323         zone_span_writelock(zone);
 324
 325         if (end_pfn - start_pfn) {
 326                 zone->zone_start_pfn = start_pfn;
 327                 zone->spanned_pages = end_pfn - start_pfn;
 328         } else {
 329                 /*
 330                  * make it consist as free_area_init_core(),
 331                  * if spanned_pages = 0, then keep start_pfn = 0
 332                  */
 333                 zone->zone_start_pfn = 0;
 334                 zone->spanned_pages = 0;
 335         }
 336
 337         zone_span_writeunlock(zone);
 338 }
 339
 340 static void fix_zone_id(struct zone *zone, unsigned long start_pfn,
 341                 unsigned long end_pfn)
 342 {
 343         enum zone_type zid = zone_idx(zone);
 344         int nid = zone->zone_pgdat->node_id;
 345         unsigned long pfn;
 346
 347         for (pfn = start_pfn; pfn < end_pfn; pfn++)
 348                 set_page_links(pfn_to_page(pfn), zid, nid, pfn);
 349 }
 350
 351 /* Can fail with -ENOMEM from allocating a wait table with vmalloc() or
 352  * alloc_bootmem_node_nopanic()/memblock_virt_alloc_node_nopanic() */
 353 static int __ref ensure_zone_is_initialized(struct zone *zone,
 354                         unsigned long start_pfn, unsigned long num_pages)
 355 {
 356         if (!zone_is_initialized(zone))
 357                 return init_currently_empty_zone(zone, start_pfn, num_pages);
 358
 359         return 0;
 360 }
 361
 362 static int __meminit move_pfn_range_left(struct zone *z1, struct zone *z2,
 363                 unsigned long start_pfn, unsigned long end_pfn)
 364 {
 365         int ret;
 366         unsigned long flags;
 367         unsigned long z1_start_pfn;
 368
 369         ret = ensure_zone_is_initialized(z1, start_pfn, end_pfn - start_pfn);
 370         if (ret)
 371                 return ret;
 372
 373         pgdat_resize_lock(z1->zone_pgdat, &flags);
 374
 375         /* can't move pfns which are higher than @z2 */
 376         if (end_pfn > zone_end_pfn(z2))
 377                 goto out_fail;
 378         /* the move out part must be at the left most of @z2 */
 379         if (start_pfn > z2->zone_start_pfn)
 380                 goto out_fail;
 381         /* must included/overlap */
 382         if (end_pfn <= z2->zone_start_pfn)
 383                 goto out_fail;
 384
 385         /* use start_pfn for z1's start_pfn if z1 is empty */
 386         if (!zone_is_empty(z1))
 387                 z1_start_pfn = z1->zone_start_pfn;
 388         else
 389                 z1_start_pfn = start_pfn;
 390
 391         resize_zone(z1, z1_start_pfn, end_pfn);
 392         resize_zone(z2, end_pfn, zone_end_pfn(z2));
 393
 394         pgdat_resize_unlock(z1->zone_pgdat, &flags);
 395
 396         fix_zone_id(z1, start_pfn, end_pfn);
 397
 398         return 0;
 399 out_fail:
 400         pgdat_resize_unlock(z1->zone_pgdat, &flags);
 401         return -1;
 402 }
 403
 404 static int __meminit move_pfn_range_right(struct zone *z1, struct zone *z2,
 405                 unsigned long start_pfn, unsigned long end_pfn)
 406 {
 407         int ret;
 408         unsigned long flags;
 409         unsigned long z2_end_pfn;
 410
 411         ret = ensure_zone_is_initialized(z2, start_pfn, end_pfn - start_pfn);
 412         if (ret)
 413                 return ret;
 414
 415         pgdat_resize_lock(z1->zone_pgdat, &flags);
 416
 417         /* can't move pfns which are lower than @z1 */
 418         if (z1->zone_start_pfn > start_pfn)
 419                 goto out_fail;
 420         /* the move out part mast at the right most of @z1 */
 421         if (zone_end_pfn(z1) >  end_pfn)
 422                 goto out_fail;
 423         /* must included/overlap */
 424         if (start_pfn >= zone_end_pfn(z1))
 425                 goto out_fail;
 426
 427         /* use end_pfn for z2's end_pfn if z2 is empty */
 428         if (!zone_is_empty(z2))
 429                 z2_end_pfn = zone_end_pfn(z2);
 430         else
 431                 z2_end_pfn = end_pfn;
 432
 433         resize_zone(z1, z1->zone_start_pfn, start_pfn);
 434         resize_zone(z2, start_pfn, z2_end_pfn);
 435
 436         pgdat_resize_unlock(z1->zone_pgdat, &flags);
 437
 438         fix_zone_id(z2, start_pfn, end_pfn);
 439
 440         return 0;
 441 out_fail:
 442         pgdat_resize_unlock(z1->zone_pgdat, &flags);
 443         return -1;
 444 }
 445
 446 static struct zone * __meminit move_pfn_range(int zone_shift,
 447                 unsigned long start_pfn, unsigned long end_pfn)
 448 {
 449         struct zone *zone = page_zone(pfn_to_page(start_pfn));
 450         int ret = 0;
 451
 452         if (zone_shift < 0)
 453                 ret = move_pfn_range_left(zone + zone_shift, zone,
 454                                           start_pfn, end_pfn);
 455         else if (zone_shift)
 456                 ret = move_pfn_range_right(zone, zone + zone_shift,
 457                                            start_pfn, end_pfn);
 458
 459         if (ret)
 460                 return NULL;
 461
 462         return zone + zone_shift;
 463 }
 464
 465 static void __meminit grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn,
 466                                       unsigned long end_pfn)
 467 {
 468         unsigned long old_pgdat_end_pfn = pgdat_end_pfn(pgdat);
 469
 470         if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
 471                 pgdat->node_start_pfn = start_pfn;
 472
 473         pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
 474                                         pgdat->node_start_pfn;
 475 }
 476
 477 static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn)
 478 {
 479         struct pglist_data *pgdat = zone->zone_pgdat;
 480         int nr_pages = PAGES_PER_SECTION;
 481         int nid = pgdat->node_id;
 482         int zone_type;
 483         unsigned long flags, pfn;
 484         int ret;
 485
 486         zone_type = zone - pgdat->node_zones;
 487         ret = ensure_zone_is_initialized(zone, phys_start_pfn, nr_pages);
 488         if (ret)
 489                 return ret;
 490
 491         pgdat_resize_lock(zone->zone_pgdat, &flags);
 492         grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages);
 493         grow_pgdat_span(zone->zone_pgdat, phys_start_pfn,
 494                         phys_start_pfn + nr_pages);
 495         pgdat_resize_unlock(zone->zone_pgdat, &flags);
 496         memmap_init_zone(nr_pages, nid, zone_type,
 497                          phys_start_pfn, MEMMAP_HOTPLUG);
 498
 499         /* online_page_range is called later and expects pages reserved */
 500         for (pfn = phys_start_pfn; pfn < phys_start_pfn + nr_pages; pfn++) {
 501                 if (!pfn_valid(pfn))
 502                         continue;
 503
 504                 SetPageReserved(pfn_to_page(pfn));
 505         }
 506         return 0;
 507 }
 508
 509 static int __meminit __add_section(int nid, struct zone *zone,
 510                                         unsigned long phys_start_pfn)
 511 {
 512         int ret;
 513
 514         if (pfn_valid(phys_start_pfn))
 515                 return -EEXIST;
 516
 517         ret = sparse_add_one_section(zone, phys_start_pfn);
 518
 519         if (ret < 0)
 520                 return ret;
 521
 522         ret = __add_zone(zone, phys_start_pfn);
 523
 524         if (ret < 0)
 525                 return ret;
 526
 527         return register_new_memory(nid, __pfn_to_section(phys_start_pfn));
 528 }
 529
 530 /*
 531  * Reasonably generic function for adding memory.  It is
 532  * expected that archs that support memory hotplug will
 533  * call this function after deciding the zone to which to
 534  * add the new pages.
 535  */
 536 int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn,
 537                         unsigned long nr_pages)
 538 {
 539         unsigned long i;
 540         int err = 0;
 541         int start_sec, end_sec;
 542         struct vmem_altmap *altmap;
 543
 544         clear_zone_contiguous(zone);
 545
 546         /* during initialize mem_map, align hot-added range to section */
 547         start_sec = pfn_to_section_nr(phys_start_pfn);
 548         end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
 549
 550         altmap = to_vmem_altmap((unsigned long) pfn_to_page(phys_start_pfn));
 551         if (altmap) {
 552                 /*
 553                  * Validate altmap is within bounds of the total request
 554                  */
 555                 if (altmap->base_pfn != phys_start_pfn
 556                                 || vmem_altmap_offset(altmap) > nr_pages) {
 557                         pr_warn_once("memory add fail, invalid altmap\n");
 558                         err = -EINVAL;
 559                         goto out;
 560                 }
 561                 altmap->alloc = 0;
 562         }
 563
 564         for (i = start_sec; i <= end_sec; i++) {
 565                 err = __add_section(nid, zone, section_nr_to_pfn(i));
 566
 567                 /*
 568                  * EEXIST is finally dealt with by ioresource collision
 569                  * check. see add_memory() => register_memory_resource()
 570                  * Warning will be printed if there is collision.
 571                  */
 572                 if (err && (err != -EEXIST))
 573                         break;
 574                 err = 0;
 575         }
 576         vmemmap_populate_print_last();
 577 out:
 578         set_zone_contiguous(zone);
 579         return err;
 580 }
 581 EXPORT_SYMBOL_GPL(__add_pages);
 582
 583 #ifdef CONFIG_MEMORY_HOTREMOVE
 584 /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
 585 static int find_smallest_section_pfn(int nid, struct zone *zone,
 586                                      unsigned long start_pfn,
 587                                      unsigned long end_pfn)
 588 {
 589         struct mem_section *ms;
 590
 591         for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) {
 592                 ms = __pfn_to_section(start_pfn);
 593
 594                 if (unlikely(!valid_section(ms)))
 595                         continue;
 596
 597                 if (unlikely(pfn_to_nid(start_pfn) != nid))
 598                         continue;
 599
 600                 if (zone && zone != page_zone(pfn_to_page(start_pfn)))
 601                         continue;
 602
 603                 return start_pfn;
 604         }
 605
 606         return 0;
 607 }
 608
 609 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
 610 static int find_biggest_section_pfn(int nid, struct zone *zone,
 611                                     unsigned long start_pfn,
 612                                     unsigned long end_pfn)
 613 {
 614         struct mem_section *ms;
 615         unsigned long pfn;
 616
 617         /* pfn is the end pfn of a memory section. */
 618         pfn = end_pfn - 1;
 619         for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) {
 620                 ms = __pfn_to_section(pfn);
 621
 622                 if (unlikely(!valid_section(ms)))
 623                         continue;
 624
 625                 if (unlikely(pfn_to_nid(pfn) != nid))
 626                         continue;
 627
 628                 if (zone && zone != page_zone(pfn_to_page(pfn)))
 629                         continue;
 630
 631                 return pfn;
 632         }
 633
 634         return 0;
 635 }
 636
 637 static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
 638                              unsigned long end_pfn)
 639 {
 640         unsigned long zone_start_pfn = zone->zone_start_pfn;
 641         unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */
 642         unsigned long zone_end_pfn = z;
 643         unsigned long pfn;
 644         struct mem_section *ms;
 645         int nid = zone_to_nid(zone);
 646
 647         zone_span_writelock(zone);
 648         if (zone_start_pfn == start_pfn) {
 649                 /*
 650                  * If the section is smallest section in the zone, it need
 651                  * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
 652                  * In this case, we find second smallest valid mem_section
 653                  * for shrinking zone.
 654                  */
 655                 pfn = find_smallest_section_pfn(nid, zone, end_pfn,
 656                                                 zone_end_pfn);
 657                 if (pfn) {
 658                         zone->zone_start_pfn = pfn;
 659                         zone->spanned_pages = zone_end_pfn - pfn;
 660                 }
 661         } else if (zone_end_pfn == end_pfn) {
 662                 /*
 663                  * If the section is biggest section in the zone, it need
 664                  * shrink zone->spanned_pages.
 665                  * In this case, we find second biggest valid mem_section for
 666                  * shrinking zone.
 667                  */
 668                 pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn,
 669                                                start_pfn);
 670                 if (pfn)
 671                         zone->spanned_pages = pfn - zone_start_pfn + 1;
 672         }
 673
 674         /*
 675          * The section is not biggest or smallest mem_section in the zone, it
 676          * only creates a hole in the zone. So in this case, we need not
 677          * change the zone. But perhaps, the zone has only hole data. Thus
 678          * it check the zone has only hole or not.
 679          */
 680         pfn = zone_start_pfn;
 681         for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) {
 682                 ms = __pfn_to_section(pfn);
 683
 684                 if (unlikely(!valid_section(ms)))
 685                         continue;
 686
 687                 if (page_zone(pfn_to_page(pfn)) != zone)
 688                         continue;
 689
 690                  /* If the section is current section, it continues the loop */
 691                 if (start_pfn == pfn)
 692                         continue;
 693
 694                 /* If we find valid section, we have nothing to do */
 695                 zone_span_writeunlock(zone);
 696                 return;
 697         }
 698
 699         /* The zone has no valid section */
 700         zone->zone_start_pfn = 0;
 701         zone->spanned_pages = 0;
 702         zone_span_writeunlock(zone);
 703 }
 704
 705 static void shrink_pgdat_span(struct pglist_data *pgdat,
 706                               unsigned long start_pfn, unsigned long end_pfn)
 707 {
 708         unsigned long pgdat_start_pfn = pgdat->node_start_pfn;
 709         unsigned long p = pgdat_end_pfn(pgdat); /* pgdat_end_pfn namespace clash */
 710         unsigned long pgdat_end_pfn = p;
 711         unsigned long pfn;
 712         struct mem_section *ms;
 713         int nid = pgdat->node_id;
 714
 715         if (pgdat_start_pfn == start_pfn) {
 716                 /*
 717                  * If the section is smallest section in the pgdat, it need
 718                  * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages.
 719                  * In this case, we find second smallest valid mem_section
 720                  * for shrinking zone.
 721                  */
 722                 pfn = find_smallest_section_pfn(nid, NULL, end_pfn,
 723                                                 pgdat_end_pfn);
 724                 if (pfn) {
 725                         pgdat->node_start_pfn = pfn;
 726                         pgdat->node_spanned_pages = pgdat_end_pfn - pfn;
 727                 }
 728         } else if (pgdat_end_pfn == end_pfn) {
 729                 /*
 730                  * If the section is biggest section in the pgdat, it need
 731                  * shrink pgdat->node_spanned_pages.
 732                  * In this case, we find second biggest valid mem_section for
 733                  * shrinking zone.
 734                  */
 735                 pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn,
 736                                                start_pfn);
 737                 if (pfn)
 738                         pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1;
 739         }
 740
 741         /*
 742          * If the section is not biggest or smallest mem_section in the pgdat,
 743          * it only creates a hole in the pgdat. So in this case, we need not
 744          * change the pgdat.
 745          * But perhaps, the pgdat has only hole data. Thus it check the pgdat
 746          * has only hole or not.
 747          */
 748         pfn = pgdat_start_pfn;
 749         for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) {
 750                 ms = __pfn_to_section(pfn);
 751
 752                 if (unlikely(!valid_section(ms)))
 753                         continue;
 754
 755                 if (pfn_to_nid(pfn) != nid)
 756                         continue;
 757
 758                  /* If the section is current section, it continues the loop */
 759                 if (start_pfn == pfn)
 760                         continue;
 761
 762                 /* If we find valid section, we have nothing to do */
 763                 return;
 764         }
 765
 766         /* The pgdat has no valid section */
 767         pgdat->node_start_pfn = 0;
 768         pgdat->node_spanned_pages = 0;
 769 }
 770
 771 static void __remove_zone(struct zone *zone, unsigned long start_pfn)
 772 {
 773         struct pglist_data *pgdat = zone->zone_pgdat;
 774         int nr_pages = PAGES_PER_SECTION;
 775         int zone_type;
 776         unsigned long flags;
 777
 778         zone_type = zone - pgdat->node_zones;
 779
 780         pgdat_resize_lock(zone->zone_pgdat, &flags);
 781         shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
 782         shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages);
 783         pgdat_resize_unlock(zone->zone_pgdat, &flags);
 784 }
 785
 786 static int __remove_section(struct zone *zone, struct mem_section *ms,
 787                 unsigned long map_offset)
 788 {
 789         unsigned long start_pfn;
 790         int scn_nr;
 791         int ret = -EINVAL;
 792
 793         if (!valid_section(ms))
 794                 return ret;
 795
 796         ret = unregister_memory_section(ms);
 797         if (ret)
 798                 return ret;
 799
 800         scn_nr = __section_nr(ms);
 801         start_pfn = section_nr_to_pfn(scn_nr);
 802         __remove_zone(zone, start_pfn);
 803
 804         sparse_remove_one_section(zone, ms, map_offset);
 805         return 0;
 806 }
 807
 808 /**
 809  * __remove_pages() - remove sections of pages from a zone
 810  * @zone: zone from which pages need to be removed
 811  * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
 812  * @nr_pages: number of pages to remove (must be multiple of section size)
 813  *
 814  * Generic helper function to remove section mappings and sysfs entries
 815  * for the section of the memory we are removing. Caller needs to make
 816  * sure that pages are marked reserved and zones are adjust properly by
 817  * calling offline_pages().
 818  */
 819 int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
 820                  unsigned long nr_pages)
 821 {
 822         unsigned long i;
 823         unsigned long map_offset = 0;
 824         int sections_to_remove, ret = 0;
 825
 826         /* In the ZONE_DEVICE case device driver owns the memory region */
 827         if (is_dev_zone(zone)) {
 828                 struct page *page = pfn_to_page(phys_start_pfn);
 829                 struct vmem_altmap *altmap;
 830
 831                 altmap = to_vmem_altmap((unsigned long) page);
 832                 if (altmap)
 833                         map_offset = vmem_altmap_offset(altmap);
 834         } else {
 835                 resource_size_t start, size;
 836
 837                 start = phys_start_pfn << PAGE_SHIFT;
 838                 size = nr_pages * PAGE_SIZE;
 839
 840                 ret = release_mem_region_adjustable(&iomem_resource, start,
 841                                         size);
 842                 if (ret) {
 843                         resource_size_t endres = start + size - 1;
 844
 845                         pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
 846                                         &start, &endres, ret);
 847                 }
 848         }
 849
 850         clear_zone_contiguous(zone);
 851
 852         /*
 853          * We can only remove entire sections
 854          */
 855         BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
 856         BUG_ON(nr_pages % PAGES_PER_SECTION);
 857
 858         sections_to_remove = nr_pages / PAGES_PER_SECTION;
 859         for (i = 0; i < sections_to_remove; i++) {
 860                 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
 861
 862                 ret = __remove_section(zone, __pfn_to_section(pfn), map_offset);
 863                 map_offset = 0;
 864                 if (ret)
 865                         break;
 866         }
 867
 868         set_zone_contiguous(zone);
 869
 870         return ret;
 871 }
 872 #endif /* CONFIG_MEMORY_HOTREMOVE */
 873
 874 int set_online_page_callback(online_page_callback_t callback)
 875 {
 876         int rc = -EINVAL;
 877
 878         get_online_mems();
 879         mutex_lock(&online_page_callback_lock);
 880
 881         if (online_page_callback == generic_online_page) {
 882                 online_page_callback = callback;
 883                 rc = 0;
 884         }
 885
 886         mutex_unlock(&online_page_callback_lock);
 887         put_online_mems();
 888
 889         return rc;
 890 }
 891 EXPORT_SYMBOL_GPL(set_online_page_callback);
 892
 893 int restore_online_page_callback(online_page_callback_t callback)
 894 {
 895         int rc = -EINVAL;
 896
 897         get_online_mems();
 898         mutex_lock(&online_page_callback_lock);
 899
 900         if (online_page_callback == callback) {
 901                 online_page_callback = generic_online_page;
 902                 rc = 0;
 903         }
 904
 905         mutex_unlock(&online_page_callback_lock);
 906         put_online_mems();
 907
 908         return rc;
 909 }
 910 EXPORT_SYMBOL_GPL(restore_online_page_callback);
 911
 912 void __online_page_set_limits(struct page *page)
 913 {
 914 }
 915 EXPORT_SYMBOL_GPL(__online_page_set_limits);
 916
 917 void __online_page_increment_counters(struct page *page)
 918 {
 919         adjust_managed_page_count(page, 1);
 920 }
 921 EXPORT_SYMBOL_GPL(__online_page_increment_counters);
 922
 923 void __online_page_free(struct page *page)
 924 {
 925         __free_reserved_page(page);
 926 }
 927 EXPORT_SYMBOL_GPL(__online_page_free);
 928
 929 static void generic_online_page(struct page *page)
 930 {
 931         __online_page_set_limits(page);
 932         __online_page_increment_counters(page);
 933         __online_page_free(page);
 934 }
 935
 936 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
 937                         void *arg)
 938 {
 939         unsigned long i;
 940         unsigned long onlined_pages = *(unsigned long *)arg;
 941         struct page *page;
 942         if (PageReserved(pfn_to_page(start_pfn)))
 943                 for (i = 0; i < nr_pages; i++) {
 944                         page = pfn_to_page(start_pfn + i);
 945                         (*online_page_callback)(page);
 946                         onlined_pages++;
 947                 }
 948         *(unsigned long *)arg = onlined_pages;
 949         return 0;
 950 }
 951
 952 #ifdef CONFIG_MOVABLE_NODE
 953 /*
 954  * When CONFIG_MOVABLE_NODE, we permit onlining of a node which doesn't have
 955  * normal memory.
 956  */
 957 static bool can_online_high_movable(struct zone *zone)
 958 {
 959         return true;
 960 }
 961 #else /* CONFIG_MOVABLE_NODE */
 962 /* ensure every online node has NORMAL memory */
 963 static bool can_online_high_movable(struct zone *zone)
 964 {
 965         return node_state(zone_to_nid(zone), N_NORMAL_MEMORY);
 966 }
 967 #endif /* CONFIG_MOVABLE_NODE */
 968
 969 /* check which state of node_states will be changed when online memory */
 970 static void node_states_check_changes_online(unsigned long nr_pages,
 971         struct zone *zone, struct memory_notify *arg)
 972 {
 973         int nid = zone_to_nid(zone);
 974         enum zone_type zone_last = ZONE_NORMAL;
 975
 976         /*
 977          * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
 978          * contains nodes which have zones of 0...ZONE_NORMAL,
 979          * set zone_last to ZONE_NORMAL.
 980          *
 981          * If we don't have HIGHMEM nor movable node,
 982          * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
 983          * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
 984          */
 985         if (N_MEMORY == N_NORMAL_MEMORY)
 986                 zone_last = ZONE_MOVABLE;
 987
 988         /*
 989          * if the memory to be online is in a zone of 0...zone_last, and
 990          * the zones of 0...zone_last don't have memory before online, we will
 991          * need to set the node to node_states[N_NORMAL_MEMORY] after
 992          * the memory is online.
 993          */
 994         if (zone_idx(zone) <= zone_last && !node_state(nid, N_NORMAL_MEMORY))
 995                 arg->status_change_nid_normal = nid;
 996         else
 997                 arg->status_change_nid_normal = -1;
 998
 999 #ifdef CONFIG_HIGHMEM
1000         /*
1001          * If we have movable node, node_states[N_HIGH_MEMORY]
1002          * contains nodes which have zones of 0...ZONE_HIGHMEM,
1003          * set zone_last to ZONE_HIGHMEM.
1004          *
1005          * If we don't have movable node, node_states[N_NORMAL_MEMORY]
1006          * contains nodes which have zones of 0...ZONE_MOVABLE,
1007          * set zone_last to ZONE_MOVABLE.
1008          */
1009         zone_last = ZONE_HIGHMEM;
1010         if (N_MEMORY == N_HIGH_MEMORY)
1011                 zone_last = ZONE_MOVABLE;
1012
1013         if (zone_idx(zone) <= zone_last && !node_state(nid, N_HIGH_MEMORY))
1014                 arg->status_change_nid_high = nid;
1015         else
1016                 arg->status_change_nid_high = -1;
1017 #else
1018         arg->status_change_nid_high = arg->status_change_nid_normal;
1019 #endif
1020
1021         /*
1022          * if the node don't have memory befor online, we will need to
1023          * set the node to node_states[N_MEMORY] after the memory
1024          * is online.
1025          */
1026         if (!node_state(nid, N_MEMORY))
1027                 arg->status_change_nid = nid;
1028         else
1029                 arg->status_change_nid = -1;
1030 }
1031
1032 static void node_states_set_node(int node, struct memory_notify *arg)
1033 {
1034         if (arg->status_change_nid_normal >= 0)
1035                 node_set_state(node, N_NORMAL_MEMORY);
1036
1037         if (arg->status_change_nid_high >= 0)
1038                 node_set_state(node, N_HIGH_MEMORY);
1039
1040         node_set_state(node, N_MEMORY);
1041 }
1042
1043 bool zone_can_shift(unsigned long pfn, unsigned long nr_pages,
1044                    enum zone_type target, int *zone_shift)
1045 {
1046         struct zone *zone = page_zone(pfn_to_page(pfn));
1047         enum zone_type idx = zone_idx(zone);
1048         int i;
1049
1050         *zone_shift = 0;
1051
1052         if (idx < target) {
1053                 /* pages must be at end of current zone */
1054                 if (pfn + nr_pages != zone_end_pfn(zone))
1055                         return false;
1056
1057                 /* no zones in use between current zone and target */
1058                 for (i = idx + 1; i < target; i++)
1059                         if (zone_is_initialized(zone - idx + i))
1060                                 return false;
1061         }
1062
1063         if (target < idx) {
1064                 /* pages must be at beginning of current zone */
1065                 if (pfn != zone->zone_start_pfn)
1066                         return false;
1067
1068                 /* no zones in use between current zone and target */
1069                 for (i = target + 1; i < idx; i++)
1070                         if (zone_is_initialized(zone - idx + i))
1071                                 return false;
1072         }
1073
1074         *zone_shift = target - idx;
1075         return true;
1076 }
1077
1078 /* Must be protected by mem_hotplug_begin() */
1079 int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type)
1080 {
1081         unsigned long flags;
1082         unsigned long onlined_pages = 0;
1083         struct zone *zone;
1084         int need_zonelists_rebuild = 0;
1085         int nid;
1086         int ret;
1087         struct memory_notify arg;
1088         int zone_shift = 0;
1089
1090         /*
1091          * This doesn't need a lock to do pfn_to_page().
1092          * The section can't be removed here because of the
1093          * memory_block->state_mutex.
1094          */
1095         zone = page_zone(pfn_to_page(pfn));
1096
1097         if ((zone_idx(zone) > ZONE_NORMAL ||
1098             online_type == MMOP_ONLINE_MOVABLE) &&
1099             !can_online_high_movable(zone))
1100                 return -EINVAL;
1101
1102         if (online_type == MMOP_ONLINE_KERNEL) {
1103                 if (!zone_can_shift(pfn, nr_pages, ZONE_NORMAL, &zone_shift))
1104                         return -EINVAL;
1105         } else if (online_type == MMOP_ONLINE_MOVABLE) {
1106                 if (!zone_can_shift(pfn, nr_pages, ZONE_MOVABLE, &zone_shift))
1107                         return -EINVAL;
1108         }
1109
1110         zone = move_pfn_range(zone_shift, pfn, pfn + nr_pages);
1111         if (!zone)
1112                 return -EINVAL;
1113
1114         arg.start_pfn = pfn;
1115         arg.nr_pages = nr_pages;
1116         node_states_check_changes_online(nr_pages, zone, &arg);
1117
1118         nid = zone_to_nid(zone);
1119
1120         ret = memory_notify(MEM_GOING_ONLINE, &arg);
1121         ret = notifier_to_errno(ret);
1122         if (ret)
1123                 goto failed_addition;
1124
1125         /*
1126          * If this zone is not populated, then it is not in zonelist.
1127          * This means the page allocator ignores this zone.
1128          * So, zonelist must be updated after online.
1129          */
1130         mutex_lock(&zonelists_mutex);
1131         if (!populated_zone(zone)) {
1132                 need_zonelists_rebuild = 1;
1133                 build_all_zonelists(NULL, zone);
1134         }
1135
1136         ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
1137                 online_pages_range);
1138         if (ret) {
1139                 if (need_zonelists_rebuild)
1140                         zone_pcp_reset(zone);
1141                 mutex_unlock(&zonelists_mutex);
1142                 goto failed_addition;
1143         }
1144
1145         zone->present_pages += onlined_pages;
1146
1147         pgdat_resize_lock(zone->zone_pgdat, &flags);
1148         zone->zone_pgdat->node_present_pages += onlined_pages;
1149         pgdat_resize_unlock(zone->zone_pgdat, &flags);
1150
1151         if (onlined_pages) {
1152                 node_states_set_node(nid, &arg);
1153                 if (need_zonelists_rebuild)
1154                         build_all_zonelists(NULL, NULL);
1155                 else
1156                         zone_pcp_update(zone);
1157         }
1158
1159         mutex_unlock(&zonelists_mutex);
1160
1161         init_per_zone_wmark_min();
1162
1163         if (onlined_pages) {
1164                 kswapd_run(nid);
1165                 kcompactd_run(nid);
1166         }
1167
1168         vm_total_pages = nr_free_pagecache_pages();
1169
1170         writeback_set_ratelimit();
1171
1172         if (onlined_pages)
1173                 memory_notify(MEM_ONLINE, &arg);
1174         return 0;
1175
1176 failed_addition:
1177         pr_debug("online_pages [mem %#010llx-%#010llx] failed\n",
1178                  (unsigned long long) pfn << PAGE_SHIFT,
1179                  (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1);
1180         memory_notify(MEM_CANCEL_ONLINE, &arg);
1181         return ret;
1182 }
1183 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
1184
1185 static void reset_node_present_pages(pg_data_t *pgdat)
1186 {
1187         struct zone *z;
1188
1189         for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
1190                 z->present_pages = 0;
1191
1192         pgdat->node_present_pages = 0;
1193 }
1194
1195 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1196 static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
1197 {
1198         struct pglist_data *pgdat;
1199         unsigned long zones_size[MAX_NR_ZONES] = {0};
1200         unsigned long zholes_size[MAX_NR_ZONES] = {0};
1201         unsigned long start_pfn = PFN_DOWN(start);
1202
1203         pgdat = NODE_DATA(nid);
1204         if (!pgdat) {
1205                 pgdat = arch_alloc_nodedata(nid);
1206                 if (!pgdat)
1207                         return NULL;
1208
1209                 arch_refresh_nodedata(nid, pgdat);
1210         } else {
1211                 /*
1212                  * Reset the nr_zones, order and classzone_idx before reuse.
1213                  * Note that kswapd will init kswapd_classzone_idx properly
1214                  * when it starts in the near future.
1215                  */
1216                 pgdat->nr_zones = 0;
1217                 pgdat->kswapd_order = 0;
1218                 pgdat->kswapd_classzone_idx = 0;
1219         }
1220
1221         /* we can use NODE_DATA(nid) from here */
1222
1223         /* init node's zones as empty zones, we don't have any present pages.*/
1224         free_area_init_node(nid, zones_size, start_pfn, zholes_size);
1225         pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat);
1226
1227         /*
1228          * The node we allocated has no zone fallback lists. For avoiding
1229          * to access not-initialized zonelist, build here.
1230          */
1231         mutex_lock(&zonelists_mutex);
1232         build_all_zonelists(pgdat, NULL);
1233         mutex_unlock(&zonelists_mutex);
1234
1235         /*
1236          * zone->managed_pages is set to an approximate value in
1237          * free_area_init_core(), which will cause
1238          * /sys/device/system/node/nodeX/meminfo has wrong data.
1239          * So reset it to 0 before any memory is onlined.
1240          */
1241         reset_node_managed_pages(pgdat);
1242
1243         /*
1244          * When memory is hot-added, all the memory is in offline state. So
1245          * clear all zones' present_pages because they will be updated in
1246          * online_pages() and offline_pages().
1247          */
1248         reset_node_present_pages(pgdat);
1249
1250         return pgdat;
1251 }
1252
1253 static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
1254 {
1255         arch_refresh_nodedata(nid, NULL);
1256         free_percpu(pgdat->per_cpu_nodestats);
1257         arch_free_nodedata(pgdat);
1258         return;
1259 }
1260
1261
1262 /**
1263  * try_online_node - online a node if offlined
1264  *
1265  * called by cpu_up() to online a node without onlined memory.
1266  */
1267 int try_online_node(int nid)
1268 {
1269         pg_data_t       *pgdat;
1270         int     ret;
1271
1272         if (node_online(nid))
1273                 return 0;
1274
1275         mem_hotplug_begin();
1276         pgdat = hotadd_new_pgdat(nid, 0);
1277         if (!pgdat) {
1278                 pr_err("Cannot online node %d due to NULL pgdat\n", nid);
1279                 ret = -ENOMEM;
1280                 goto out;
1281         }
1282         node_set_online(nid);
1283         ret = register_one_node(nid);
1284         BUG_ON(ret);
1285
1286         if (pgdat->node_zonelists->_zonerefs->zone == NULL) {
1287                 mutex_lock(&zonelists_mutex);
1288                 build_all_zonelists(NULL, NULL);
1289                 mutex_unlock(&zonelists_mutex);
1290         }
1291
1292 out:
1293         mem_hotplug_done();
1294         return ret;
1295 }
1296
1297 static int check_hotplug_memory_range(u64 start, u64 size)
1298 {
1299         u64 start_pfn = PFN_DOWN(start);
1300         u64 nr_pages = size >> PAGE_SHIFT;
1301
1302         /* Memory range must be aligned with section */
1303         if ((start_pfn & ~PAGE_SECTION_MASK) ||
1304             (nr_pages % PAGES_PER_SECTION) || (!nr_pages)) {
1305                 pr_err("Section-unaligned hotplug range: start 0x%llx, size 0x%llx\n",
1306                                 (unsigned long long)start,
1307                                 (unsigned long long)size);
1308                 return -EINVAL;
1309         }
1310
1311         return 0;
1312 }
1313
1314 /*
1315  * If movable zone has already been setup, newly added memory should be check.
1316  * If its address is higher than movable zone, it should be added as movable.
1317  * Without this check, movable zone may overlap with other zone.
1318  */
1319 static int should_add_memory_movable(int nid, u64 start, u64 size)
1320 {
1321         unsigned long start_pfn = start >> PAGE_SHIFT;
1322         pg_data_t *pgdat = NODE_DATA(nid);
1323         struct zone *movable_zone = pgdat->node_zones + ZONE_MOVABLE;
1324
1325         if (zone_is_empty(movable_zone))
1326                 return 0;
1327
1328         if (movable_zone->zone_start_pfn <= start_pfn)
1329                 return 1;
1330
1331         return 0;
1332 }
1333
1334 int zone_for_memory(int nid, u64 start, u64 size, int zone_default,
1335                 bool for_device)
1336 {
1337 #ifdef CONFIG_ZONE_DEVICE
1338         if (for_device)
1339                 return ZONE_DEVICE;
1340 #endif
1341         if (should_add_memory_movable(nid, start, size))
1342                 return ZONE_MOVABLE;
1343
1344         return zone_default;
1345 }
1346
1347 static int online_memory_block(struct memory_block *mem, void *arg)
1348 {
1349         return device_online(&mem->dev);
1350 }
1351
1352 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1353 int __ref add_memory_resource(int nid, struct resource *res, bool online)
1354 {
1355         u64 start, size;
1356         pg_data_t *pgdat = NULL;
1357         bool new_pgdat;
1358         bool new_node;
1359         int ret;
1360
1361         start = res->start;
1362         size = resource_size(res);
1363
1364         ret = check_hotplug_memory_range(start, size);
1365         if (ret)
1366                 return ret;
1367
1368         {       /* Stupid hack to suppress address-never-null warning */
1369                 void *p = NODE_DATA(nid);
1370                 new_pgdat = !p;
1371         }
1372
1373         mem_hotplug_begin();
1374
1375         /*
1376          * Add new range to memblock so that when hotadd_new_pgdat() is called
1377          * to allocate new pgdat, get_pfn_range_for_nid() will be able to find
1378          * this new range and calculate total pages correctly.  The range will
1379          * be removed at hot-remove time.
1380          */
1381         memblock_add_node(start, size, nid);
1382
1383         new_node = !node_online(nid);
1384         if (new_node) {
1385                 pgdat = hotadd_new_pgdat(nid, start);
1386                 ret = -ENOMEM;
1387                 if (!pgdat)
1388                         goto error;
1389         }
1390
1391         /* call arch's memory hotadd */
1392         ret = arch_add_memory(nid, start, size, false);
1393
1394         if (ret < 0)
1395                 goto error;
1396
1397         /* we online node here. we can't roll back from here. */
1398         node_set_online(nid);
1399
1400         if (new_node) {
1401                 ret = register_one_node(nid);
1402                 /*
1403                  * If sysfs file of new node can't create, cpu on the node
1404                  * can't be hot-added. There is no rollback way now.
1405                  * So, check by BUG_ON() to catch it reluctantly..
1406                  */
1407                 BUG_ON(ret);
1408         }
1409
1410         /* create new memmap entry */
1411         firmware_map_add_hotplug(start, start + size, "System RAM");
1412
1413         /* online pages if requested */
1414         if (online)
1415                 walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1),
1416                                   NULL, online_memory_block);
1417
1418         goto out;
1419
1420 error:
1421         /* rollback pgdat allocation and others */
1422         if (new_pgdat)
1423                 rollback_node_hotadd(nid, pgdat);
1424         memblock_remove(start, size);
1425
1426 out:
1427         mem_hotplug_done();
1428         return ret;
1429 }
1430 EXPORT_SYMBOL_GPL(add_memory_resource);
1431
1432 int __ref add_memory(int nid, u64 start, u64 size)
1433 {
1434         struct resource *res;
1435         int ret;
1436
1437         res = register_memory_resource(start, size);
1438         if (IS_ERR(res))
1439                 return PTR_ERR(res);
1440
1441         ret = add_memory_resource(nid, res, memhp_auto_online);
1442         if (ret < 0)
1443                 release_memory_resource(res);
1444         return ret;
1445 }
1446 EXPORT_SYMBOL_GPL(add_memory);
1447
1448 #ifdef CONFIG_MEMORY_HOTREMOVE
1449 /*
1450  * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1451  * set and the size of the free page is given by page_order(). Using this,
1452  * the function determines if the pageblock contains only free pages.
1453  * Due to buddy contraints, a free page at least the size of a pageblock will
1454  * be located at the start of the pageblock
1455  */
1456 static inline int pageblock_free(struct page *page)
1457 {
1458         return PageBuddy(page) && page_order(page) >= pageblock_order;
1459 }
1460
1461 /* Return the start of the next active pageblock after a given page */
1462 static struct page *next_active_pageblock(struct page *page)
1463 {
1464         /* Ensure the starting page is pageblock-aligned */
1465         BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
1466
1467         /* If the entire pageblock is free, move to the end of free page */
1468         if (pageblock_free(page)) {
1469                 int order;
1470                 /* be careful. we don't have locks, page_order can be changed.*/
1471                 order = page_order(page);
1472                 if ((order < MAX_ORDER) && (order >= pageblock_order))
1473                         return page + (1 << order);
1474         }
1475
1476         return page + pageblock_nr_pages;
1477 }
1478
1479 /* Checks if this range of memory is likely to be hot-removable. */
1480 bool is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
1481 {
1482         struct page *page = pfn_to_page(start_pfn);
1483         struct page *end_page = page + nr_pages;
1484
1485         /* Check the starting page of each pageblock within the range */
1486         for (; page < end_page; page = next_active_pageblock(page)) {
1487                 if (!is_pageblock_removable_nolock(page))
1488                         return false;
1489                 cond_resched();
1490         }
1491
1492         /* All pageblocks in the memory block are likely to be hot-removable */
1493         return true;
1494 }
1495
1496 /*
1497  * Confirm all pages in a range [start, end) belong to the same zone.
1498  * When true, return its valid [start, end).
1499  */
1500 int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn,
1501                          unsigned long *valid_start, unsigned long *valid_end)
1502 {
1503         unsigned long pfn, sec_end_pfn;
1504         unsigned long start, end;
1505         struct zone *zone = NULL;
1506         struct page *page;
1507         int i;
1508         for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1);
1509              pfn < end_pfn;
1510              pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) {
1511                 /* Make sure the memory section is present first */
1512                 if (!present_section_nr(pfn_to_section_nr(pfn)))
1513                         continue;
1514                 for (; pfn < sec_end_pfn && pfn < end_pfn;
1515                      pfn += MAX_ORDER_NR_PAGES) {
1516                         i = 0;
1517                         /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1518                         while ((i < MAX_ORDER_NR_PAGES) &&
1519                                 !pfn_valid_within(pfn + i))
1520                                 i++;
1521                         if (i == MAX_ORDER_NR_PAGES || pfn + i >= end_pfn)
1522                                 continue;
1523                         page = pfn_to_page(pfn + i);
1524                         if (zone && page_zone(page) != zone)
1525                                 return 0;
1526                         if (!zone)
1527                                 start = pfn + i;
1528                         zone = page_zone(page);
1529                         end = pfn + MAX_ORDER_NR_PAGES;
1530                 }
1531         }
1532
1533         if (zone) {
1534                 *valid_start = start;
1535                 *valid_end = min(end, end_pfn);
1536                 return 1;
1537         } else {
1538                 return 0;
1539         }
1540 }
1541
1542 /*
1543  * Scan pfn range [start,end) to find movable/migratable pages (LRU pages,
1544  * non-lru movable pages and hugepages). We scan pfn because it's much
1545  * easier than scanning over linked list. This function returns the pfn
1546  * of the first found movable page if it's found, otherwise 0.
1547  */
1548 static unsigned long scan_movable_pages(unsigned long start, unsigned long end)
1549 {
1550         unsigned long pfn;
1551         struct page *page;
1552         for (pfn = start; pfn < end; pfn++) {
1553                 if (pfn_valid(pfn)) {
1554                         page = pfn_to_page(pfn);
1555                         if (PageLRU(page))
1556                                 return pfn;
1557                         if (__PageMovable(page))
1558                                 return pfn;
1559                         if (PageHuge(page)) {
1560                                 if (page_huge_active(page))
1561                                         return pfn;
1562                                 else
1563                                         pfn = round_up(pfn + 1,
1564                                                 1 << compound_order(page)) - 1;
1565                         }
1566                 }
1567         }
1568         return 0;
1569 }
1570
1571 static struct page *new_node_page(struct page *page, unsigned long private,
1572                 int **result)
1573 {
1574         gfp_t gfp_mask = GFP_USER | __GFP_MOVABLE;
1575         int nid = page_to_nid(page);
1576         nodemask_t nmask = node_states[N_MEMORY];
1577         struct page *new_page = NULL;
1578
1579         /*
1580          * TODO: allocate a destination hugepage from a nearest neighbor node,
1581          * accordance with memory policy of the user process if possible. For
1582          * now as a simple work-around, we use the next node for destination.
1583          */
1584         if (PageHuge(page))
1585                 return alloc_huge_page_node(page_hstate(compound_head(page)),
1586                                         next_node_in(nid, nmask));
1587
1588         node_clear(nid, nmask);
1589
1590         if (PageHighMem(page)
1591             || (zone_idx(page_zone(page)) == ZONE_MOVABLE))
1592                 gfp_mask |= __GFP_HIGHMEM;
1593
1594         if (!nodes_empty(nmask))
1595                 new_page = __alloc_pages_nodemask(gfp_mask, 0,
1596                                         node_zonelist(nid, gfp_mask), &nmask);
1597         if (!new_page)
1598                 new_page = __alloc_pages(gfp_mask, 0,
1599                                         node_zonelist(nid, gfp_mask));
1600
1601         return new_page;
1602 }
1603
1604 #define NR_OFFLINE_AT_ONCE_PAGES        (256)
1605 static int
1606 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1607 {
1608         unsigned long pfn;
1609         struct page *page;
1610         int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
1611         int not_managed = 0;
1612         int ret = 0;
1613         LIST_HEAD(source);
1614
1615         for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
1616                 if (!pfn_valid(pfn))
1617                         continue;
1618                 page = pfn_to_page(pfn);
1619
1620                 if (PageHuge(page)) {
1621                         struct page *head = compound_head(page);
1622                         pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1;
1623                         if (compound_order(head) > PFN_SECTION_SHIFT) {
1624                                 ret = -EBUSY;
1625                                 break;
1626                         }
1627                         if (isolate_huge_page(page, &source))
1628                                 move_pages -= 1 << compound_order(head);
1629                         continue;
1630                 }
1631
1632                 if (!get_page_unless_zero(page))
1633                         continue;
1634                 /*
1635                  * We can skip free pages. And we can deal with pages on
1636                  * LRU and non-lru movable pages.
1637                  */
1638                 if (PageLRU(page))
1639                         ret = isolate_lru_page(page);
1640                 else
1641                         ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE);
1642                 if (!ret) { /* Success */
1643                         put_page(page);
1644                         list_add_tail(&page->lru, &source);
1645                         move_pages--;
1646                         if (!__PageMovable(page))
1647                                 inc_node_page_state(page, NR_ISOLATED_ANON +
1648                                                     page_is_file_cache(page));
1649
1650                 } else {
1651 #ifdef CONFIG_DEBUG_VM
1652                         pr_alert("failed to isolate pfn %lx\n", pfn);
1653                         dump_page(page, "isolation failed");
1654 #endif
1655                         put_page(page);
1656                         /* Because we don't have big zone->lock. we should
1657                            check this again here. */
1658                         if (page_count(page)) {
1659                                 not_managed++;
1660                                 ret = -EBUSY;
1661                                 break;
1662                         }
1663                 }
1664         }
1665         if (!list_empty(&source)) {
1666                 if (not_managed) {
1667                         putback_movable_pages(&source);
1668                         goto out;
1669                 }
1670
1671                 /* Allocate a new page from the nearest neighbor node */
1672                 ret = migrate_pages(&source, new_node_page, NULL, 0,
1673                                         MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
1674                 if (ret)
1675                         putback_movable_pages(&source);
1676         }
1677 out:
1678         return ret;
1679 }
1680
1681 /*
1682  * remove from free_area[] and mark all as Reserved.
1683  */
1684 static int
1685 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1686                         void *data)
1687 {
1688         __offline_isolated_pages(start, start + nr_pages);
1689         return 0;
1690 }
1691
1692 static void
1693 offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
1694 {
1695         walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
1696                                 offline_isolated_pages_cb);
1697 }
1698
1699 /*
1700  * Check all pages in range, recoreded as memory resource, are isolated.
1701  */
1702 static int
1703 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1704                         void *data)
1705 {
1706         int ret;
1707         long offlined = *(long *)data;
1708         ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true);
1709         offlined = nr_pages;
1710         if (!ret)
1711                 *(long *)data += offlined;
1712         return ret;
1713 }
1714
1715 static long
1716 check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
1717 {
1718         long offlined = 0;
1719         int ret;
1720
1721         ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
1722                         check_pages_isolated_cb);
1723         if (ret < 0)
1724                 offlined = (long)ret;
1725         return offlined;
1726 }
1727
1728 #ifdef CONFIG_MOVABLE_NODE
1729 /*
1730  * When CONFIG_MOVABLE_NODE, we permit offlining of a node which doesn't have
1731  * normal memory.
1732  */
1733 static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1734 {
1735         return true;
1736 }
1737 #else /* CONFIG_MOVABLE_NODE */
1738 /* ensure the node has NORMAL memory if it is still online */
1739 static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1740 {
1741         struct pglist_data *pgdat = zone->zone_pgdat;
1742         unsigned long present_pages = 0;
1743         enum zone_type zt;
1744
1745         for (zt = 0; zt <= ZONE_NORMAL; zt++)
1746                 present_pages += pgdat->node_zones[zt].present_pages;
1747
1748         if (present_pages > nr_pages)
1749                 return true;
1750
1751         present_pages = 0;
1752         for (; zt <= ZONE_MOVABLE; zt++)
1753                 present_pages += pgdat->node_zones[zt].present_pages;
1754
1755         /*
1756          * we can't offline the last normal memory until all
1757          * higher memory is offlined.
1758          */
1759         return present_pages == 0;
1760 }
1761 #endif /* CONFIG_MOVABLE_NODE */
1762
1763 static int __init cmdline_parse_movable_node(char *p)
1764 {
1765 #ifdef CONFIG_MOVABLE_NODE
1766         movable_node_enabled = true;
1767 #else
1768         pr_warn("movable_node option not supported\n");
1769 #endif
1770         return 0;
1771 }
1772 early_param("movable_node", cmdline_parse_movable_node);
1773
1774 /* check which state of node_states will be changed when offline memory */
1775 static void node_states_check_changes_offline(unsigned long nr_pages,
1776                 struct zone *zone, struct memory_notify *arg)
1777 {
1778         struct pglist_data *pgdat = zone->zone_pgdat;
1779         unsigned long present_pages = 0;
1780         enum zone_type zt, zone_last = ZONE_NORMAL;
1781
1782         /*
1783          * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
1784          * contains nodes which have zones of 0...ZONE_NORMAL,
1785          * set zone_last to ZONE_NORMAL.
1786          *
1787          * If we don't have HIGHMEM nor movable node,
1788          * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
1789          * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
1790          */
1791         if (N_MEMORY == N_NORMAL_MEMORY)
1792                 zone_last = ZONE_MOVABLE;
1793
1794         /*
1795          * check whether node_states[N_NORMAL_MEMORY] will be changed.
1796          * If the memory to be offline is in a zone of 0...zone_last,
1797          * and it is the last present memory, 0...zone_last will
1798          * become empty after offline , thus we can determind we will
1799          * need to clear the node from node_states[N_NORMAL_MEMORY].
1800          */
1801         for (zt = 0; zt <= zone_last; zt++)
1802                 present_pages += pgdat->node_zones[zt].present_pages;
1803         if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1804                 arg->status_change_nid_normal = zone_to_nid(zone);
1805         else
1806                 arg->status_change_nid_normal = -1;
1807
1808 #ifdef CONFIG_HIGHMEM
1809         /*
1810          * If we have movable node, node_states[N_HIGH_MEMORY]
1811          * contains nodes which have zones of 0...ZONE_HIGHMEM,
1812          * set zone_last to ZONE_HIGHMEM.
1813          *
1814          * If we don't have movable node, node_states[N_NORMAL_MEMORY]
1815          * contains nodes which have zones of 0...ZONE_MOVABLE,
1816          * set zone_last to ZONE_MOVABLE.
1817          */
1818         zone_last = ZONE_HIGHMEM;
1819         if (N_MEMORY == N_HIGH_MEMORY)
1820                 zone_last = ZONE_MOVABLE;
1821
1822         for (; zt <= zone_last; zt++)
1823                 present_pages += pgdat->node_zones[zt].present_pages;
1824         if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1825                 arg->status_change_nid_high = zone_to_nid(zone);
1826         else
1827                 arg->status_change_nid_high = -1;
1828 #else
1829         arg->status_change_nid_high = arg->status_change_nid_normal;
1830 #endif
1831
1832         /*
1833          * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE
1834          */
1835         zone_last = ZONE_MOVABLE;
1836
1837         /*
1838          * check whether node_states[N_HIGH_MEMORY] will be changed
1839          * If we try to offline the last present @nr_pages from the node,
1840          * we can determind we will need to clear the node from
1841          * node_states[N_HIGH_MEMORY].
1842          */
1843         for (; zt <= zone_last; zt++)
1844                 present_pages += pgdat->node_zones[zt].present_pages;
1845         if (nr_pages >= present_pages)
1846                 arg->status_change_nid = zone_to_nid(zone);
1847         else
1848                 arg->status_change_nid = -1;
1849 }
1850
1851 static void node_states_clear_node(int node, struct memory_notify *arg)
1852 {
1853         if (arg->status_change_nid_normal >= 0)
1854                 node_clear_state(node, N_NORMAL_MEMORY);
1855
1856         if ((N_MEMORY != N_NORMAL_MEMORY) &&
1857             (arg->status_change_nid_high >= 0))
1858                 node_clear_state(node, N_HIGH_MEMORY);
1859
1860         if ((N_MEMORY != N_HIGH_MEMORY) &&
1861             (arg->status_change_nid >= 0))
1862                 node_clear_state(node, N_MEMORY);
1863 }
1864
1865 static int __ref __offline_pages(unsigned long start_pfn,
1866                   unsigned long end_pfn, unsigned long timeout)
1867 {
1868         unsigned long pfn, nr_pages, expire;
1869         long offlined_pages;
1870         int ret, drain, retry_max, node;
1871         unsigned long flags;
1872         unsigned long valid_start, valid_end;
1873         struct zone *zone;
1874         struct memory_notify arg;
1875
1876         /* at least, alignment against pageblock is necessary */
1877         if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
1878                 return -EINVAL;
1879         if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
1880                 return -EINVAL;
1881         /* This makes hotplug much easier...and readable.
1882            we assume this for now. .*/
1883         if (!test_pages_in_a_zone(start_pfn, end_pfn, &valid_start, &valid_end))
1884                 return -EINVAL;
1885
1886         zone = page_zone(pfn_to_page(valid_start));
1887         node = zone_to_nid(zone);
1888         nr_pages = end_pfn - start_pfn;
1889
1890         if (zone_idx(zone) <= ZONE_NORMAL && !can_offline_normal(zone, nr_pages))
1891                 return -EINVAL;
1892
1893         /* set above range as isolated */
1894         ret = start_isolate_page_range(start_pfn, end_pfn,
1895                                        MIGRATE_MOVABLE, true);
1896         if (ret)
1897                 return ret;
1898
1899         arg.start_pfn = start_pfn;
1900         arg.nr_pages = nr_pages;
1901         node_states_check_changes_offline(nr_pages, zone, &arg);
1902
1903         ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1904         ret = notifier_to_errno(ret);
1905         if (ret)
1906                 goto failed_removal;
1907
1908         pfn = start_pfn;
1909         expire = jiffies + timeout;
1910         drain = 0;
1911         retry_max = 5;
1912 repeat:
1913         /* start memory hot removal */
1914         ret = -EAGAIN;
1915         if (time_after(jiffies, expire))
1916                 goto failed_removal;
1917         ret = -EINTR;
1918         if (signal_pending(current))
1919                 goto failed_removal;
1920         ret = 0;
1921         if (drain) {
1922                 lru_add_drain_all();
1923                 cond_resched();
1924                 drain_all_pages(zone);
1925         }
1926
1927         pfn = scan_movable_pages(start_pfn, end_pfn);
1928         if (pfn) { /* We have movable pages */
1929                 ret = do_migrate_range(pfn, end_pfn);
1930                 if (!ret) {
1931                         drain = 1;
1932                         goto repeat;
1933                 } else {
1934                         if (ret < 0)
1935                                 if (--retry_max == 0)
1936                                         goto failed_removal;
1937                         yield();
1938                         drain = 1;
1939                         goto repeat;
1940                 }
1941         }
1942         /* drain all zone's lru pagevec, this is asynchronous... */
1943         lru_add_drain_all();
1944         yield();
1945         /* drain pcp pages, this is synchronous. */
1946         drain_all_pages(zone);
1947         /*
1948          * dissolve free hugepages in the memory block before doing offlining
1949          * actually in order to make hugetlbfs's object counting consistent.
1950          */
1951         ret = dissolve_free_huge_pages(start_pfn, end_pfn);
1952         if (ret)
1953                 goto failed_removal;
1954         /* check again */
1955         offlined_pages = check_pages_isolated(start_pfn, end_pfn);
1956         if (offlined_pages < 0) {
1957                 ret = -EBUSY;
1958                 goto failed_removal;
1959         }
1960         pr_info("Offlined Pages %ld\n", offlined_pages);
1961         /* Ok, all of our target is isolated.
1962            We cannot do rollback at this point. */
1963         offline_isolated_pages(start_pfn, end_pfn);
1964         /* reset pagetype flags and makes migrate type to be MOVABLE */
1965         undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1966         /* removal success */
1967         adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages);
1968         zone->present_pages -= offlined_pages;
1969
1970         pgdat_resize_lock(zone->zone_pgdat, &flags);
1971         zone->zone_pgdat->node_present_pages -= offlined_pages;
1972         pgdat_resize_unlock(zone->zone_pgdat, &flags);
1973
1974         init_per_zone_wmark_min();
1975
1976         if (!populated_zone(zone)) {
1977                 zone_pcp_reset(zone);
1978                 mutex_lock(&zonelists_mutex);
1979                 build_all_zonelists(NULL, NULL);
1980                 mutex_unlock(&zonelists_mutex);
1981         } else
1982                 zone_pcp_update(zone);
1983
1984         node_states_clear_node(node, &arg);
1985         if (arg.status_change_nid >= 0) {
1986                 kswapd_stop(node);
1987                 kcompactd_stop(node);
1988         }
1989
1990         vm_total_pages = nr_free_pagecache_pages();
1991         writeback_set_ratelimit();
1992
1993         memory_notify(MEM_OFFLINE, &arg);
1994         return 0;
1995
1996 failed_removal:
1997         pr_debug("memory offlining [mem %#010llx-%#010llx] failed\n",
1998                  (unsigned long long) start_pfn << PAGE_SHIFT,
1999                  ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
2000         memory_notify(MEM_CANCEL_OFFLINE, &arg);
2001         /* pushback to free area */
2002         undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
2003         return ret;
2004 }
2005
2006 /* Must be protected by mem_hotplug_begin() */
2007 int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
2008 {
2009         return __offline_pages(start_pfn, start_pfn + nr_pages, 120 * HZ);
2010 }
2011 #endif /* CONFIG_MEMORY_HOTREMOVE */
2012
2013 /**
2014  * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
2015  * @start_pfn: start pfn of the memory range
2016  * @end_pfn: end pfn of the memory range
2017  * @arg: argument passed to func
2018  * @func: callback for each memory section walked
2019  *
2020  * This function walks through all present mem sections in range
2021  * [start_pfn, end_pfn) and call func on each mem section.
2022  *
2023  * Returns the return value of func.
2024  */
2025 int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn,
2026                 void *arg, int (*func)(struct memory_block *, void *))
2027 {
2028         struct memory_block *mem = NULL;
2029         struct mem_section *section;
2030         unsigned long pfn, section_nr;
2031         int ret;
2032
2033         for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
2034                 section_nr = pfn_to_section_nr(pfn);
2035                 if (!present_section_nr(section_nr))
2036                         continue;
2037
2038                 section = __nr_to_section(section_nr);
2039                 /* same memblock? */
2040                 if (mem)
2041                         if ((section_nr >= mem->start_section_nr) &&
2042                             (section_nr <= mem->end_section_nr))
2043                                 continue;
2044
2045                 mem = find_memory_block_hinted(section, mem);
2046                 if (!mem)
2047                         continue;
2048
2049                 ret = func(mem, arg);
2050                 if (ret) {
2051                         kobject_put(&mem->dev.kobj);
2052                         return ret;
2053                 }
2054         }
2055
2056         if (mem)
2057                 kobject_put(&mem->dev.kobj);
2058
2059         return 0;
2060 }
2061
2062 #ifdef CONFIG_MEMORY_HOTREMOVE
2063 static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
2064 {
2065         int ret = !is_memblock_offlined(mem);
2066
2067         if (unlikely(ret)) {
2068                 phys_addr_t beginpa, endpa;
2069
2070                 beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
2071                 endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1;
2072                 pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n",
2073                         &beginpa, &endpa);
2074         }
2075
2076         return ret;
2077 }
2078
2079 static int check_cpu_on_node(pg_data_t *pgdat)
2080 {
2081         int cpu;
2082
2083         for_each_present_cpu(cpu) {
2084                 if (cpu_to_node(cpu) == pgdat->node_id)
2085                         /*
2086                          * the cpu on this node isn't removed, and we can't
2087                          * offline this node.
2088                          */
2089                         return -EBUSY;
2090         }
2091
2092         return 0;
2093 }
2094
2095 static void unmap_cpu_on_node(pg_data_t *pgdat)
2096 {
2097 #ifdef CONFIG_ACPI_NUMA
2098         int cpu;
2099
2100         for_each_possible_cpu(cpu)
2101                 if (cpu_to_node(cpu) == pgdat->node_id)
2102                         numa_clear_node(cpu);
2103 #endif
2104 }
2105
2106 static int check_and_unmap_cpu_on_node(pg_data_t *pgdat)
2107 {
2108         int ret;
2109
2110         ret = check_cpu_on_node(pgdat);
2111         if (ret)
2112                 return ret;
2113
2114         /*
2115          * the node will be offlined when we come here, so we can clear
2116          * the cpu_to_node() now.
2117          */
2118
2119         unmap_cpu_on_node(pgdat);
2120         return 0;
2121 }
2122
2123 /**
2124  * try_offline_node
2125  *
2126  * Offline a node if all memory sections and cpus of the node are removed.
2127  *
2128  * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
2129  * and online/offline operations before this call.
2130  */
2131 void try_offline_node(int nid)
2132 {
2133         pg_data_t *pgdat = NODE_DATA(nid);
2134         unsigned long start_pfn = pgdat->node_start_pfn;
2135         unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
2136         unsigned long pfn;
2137
2138         for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
2139                 unsigned long section_nr = pfn_to_section_nr(pfn);
2140
2141                 if (!present_section_nr(section_nr))
2142                         continue;
2143
2144                 if (pfn_to_nid(pfn) != nid)
2145                         continue;
2146
2147                 /*
2148                  * some memory sections of this node are not removed, and we
2149                  * can't offline node now.
2150                  */
2151                 return;
2152         }
2153
2154         if (check_and_unmap_cpu_on_node(pgdat))
2155                 return;
2156
2157         /*
2158          * all memory/cpu of this node are removed, we can offline this
2159          * node now.
2160          */
2161         node_set_offline(nid);
2162         unregister_one_node(nid);
2163 }
2164 EXPORT_SYMBOL(try_offline_node);
2165
2166 /**
2167  * remove_memory
2168  *
2169  * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
2170  * and online/offline operations before this call, as required by
2171  * try_offline_node().
2172  */
2173 void __ref remove_memory(int nid, u64 start, u64 size)
2174 {
2175         int ret;
2176
2177         BUG_ON(check_hotplug_memory_range(start, size));
2178
2179         mem_hotplug_begin();
2180
2181         /*
2182          * All memory blocks must be offlined before removing memory.  Check
2183          * whether all memory blocks in question are offline and trigger a BUG()
2184          * if this is not the case.
2185          */
2186         ret = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL,
2187                                 check_memblock_offlined_cb);
2188         if (ret)
2189                 BUG();
2190
2191         /* remove memmap entry */
2192         firmware_map_remove(start, start + size, "System RAM");
2193         memblock_free(start, size);
2194         memblock_remove(start, size);
2195
2196         arch_remove_memory(start, size);
2197
2198         try_offline_node(nid);
2199
2200         mem_hotplug_done();
2201 }
2202 EXPORT_SYMBOL_GPL(remove_memory);
2203 #endif /* CONFIG_MEMORY_HOTREMOVE */