mm/memory_hotplug.c

   1 /*
   2  *  linux/mm/memory_hotplug.c
   3  *
   4  *  Copyright (C)
   5  */
   6
   7 #include <linux/stddef.h>
   8 #include <linux/mm.h>
   9 #include <linux/swap.h>
  10 #include <linux/interrupt.h>
  11 #include <linux/pagemap.h>
  12 #include <linux/bootmem.h>
  13 #include <linux/compiler.h>
  14 #include <linux/module.h>
  15 #include <linux/pagevec.h>
  16 #include <linux/writeback.h>
  17 #include <linux/slab.h>
  18 #include <linux/sysctl.h>
  19 #include <linux/cpu.h>
  20 #include <linux/memory.h>
  21 #include <linux/memory_hotplug.h>
  22 #include <linux/highmem.h>
  23 #include <linux/vmalloc.h>
  24 #include <linux/ioport.h>
  25 #include <linux/delay.h>
  26 #include <linux/migrate.h>
  27 #include <linux/page-isolation.h>
  28 #include <linux/pfn.h>
  29 #include <linux/suspend.h>
  30 #include <linux/mm_inline.h>
  31 #include <linux/firmware-map.h>
  32
  33 #include <asm/tlbflush.h>
  34
  35 #include "internal.h"
  36
  37 DEFINE_MUTEX(mem_hotplug_mutex);
  38
  39 void lock_memory_hotplug(void)
  40 {
  41         mutex_lock(&mem_hotplug_mutex);
  42
  43         /* for exclusive hibernation if CONFIG_HIBERNATION=y */
  44         lock_system_sleep();
  45 }
  46
  47 void unlock_memory_hotplug(void)
  48 {
  49         unlock_system_sleep();
  50         mutex_unlock(&mem_hotplug_mutex);
  51 }
  52
  53
  54 /* add this memory to iomem resource */
  55 static struct resource *register_memory_resource(u64 start, u64 size)
  56 {
  57         struct resource *res;
  58         res = kzalloc(sizeof(struct resource), GFP_KERNEL);
  59         BUG_ON(!res);
  60
  61         res->name = "System RAM";
  62         res->start = start;
  63         res->end = start + size - 1;
  64         res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
  65         if (request_resource(&iomem_resource, res) < 0) {
  66                 printk("System RAM resource %llx - %llx cannot be added\n",
  67                 (unsigned long long)res->start, (unsigned long long)res->end);
  68                 kfree(res);
  69                 res = NULL;
  70         }
  71         return res;
  72 }
  73
  74 static void release_memory_resource(struct resource *res)
  75 {
  76         if (!res)
  77                 return;
  78         release_resource(res);
  79         kfree(res);
  80         return;
  81 }
  82
  83 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
  84 #ifndef CONFIG_SPARSEMEM_VMEMMAP
  85 static void get_page_bootmem(unsigned long info,  struct page *page, int type)
  86 {
  87         atomic_set(&page->_mapcount, type);
  88         SetPagePrivate(page);
  89         set_page_private(page, info);
  90         atomic_inc(&page->_count);
  91 }
  92
  93 /* reference to __meminit __free_pages_bootmem is valid
  94  * so use __ref to tell modpost not to generate a warning */
  95 void __ref put_page_bootmem(struct page *page)
  96 {
  97         int type;
  98
  99         type = atomic_read(&page->_mapcount);
 100         BUG_ON(type >= -1);
 101
 102         if (atomic_dec_return(&page->_count) == 1) {
 103                 ClearPagePrivate(page);
 104                 set_page_private(page, 0);
 105                 reset_page_mapcount(page);
 106                 __free_pages_bootmem(page, 0);
 107         }
 108
 109 }
 110
 111 static void register_page_bootmem_info_section(unsigned long start_pfn)
 112 {
 113         unsigned long *usemap, mapsize, section_nr, i;
 114         struct mem_section *ms;
 115         struct page *page, *memmap;
 116
 117         if (!pfn_valid(start_pfn))
 118                 return;
 119
 120         section_nr = pfn_to_section_nr(start_pfn);
 121         ms = __nr_to_section(section_nr);
 122
 123         /* Get section's memmap address */
 124         memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
 125
 126         /*
 127          * Get page for the memmap's phys address
 128          * XXX: need more consideration for sparse_vmemmap...
 129          */
 130         page = virt_to_page(memmap);
 131         mapsize = sizeof(struct page) * PAGES_PER_SECTION;
 132         mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
 133
 134         /* remember memmap's page */
 135         for (i = 0; i < mapsize; i++, page++)
 136                 get_page_bootmem(section_nr, page, SECTION_INFO);
 137
 138         usemap = __nr_to_section(section_nr)->pageblock_flags;
 139         page = virt_to_page(usemap);
 140
 141         mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
 142
 143         for (i = 0; i < mapsize; i++, page++)
 144                 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
 145
 146 }
 147
 148 void register_page_bootmem_info_node(struct pglist_data *pgdat)
 149 {
 150         unsigned long i, pfn, end_pfn, nr_pages;
 151         int node = pgdat->node_id;
 152         struct page *page;
 153         struct zone *zone;
 154
 155         nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
 156         page = virt_to_page(pgdat);
 157
 158         for (i = 0; i < nr_pages; i++, page++)
 159                 get_page_bootmem(node, page, NODE_INFO);
 160
 161         zone = &pgdat->node_zones[0];
 162         for (; zone < pgdat->node_zones + MAX_NR_ZONES - 1; zone++) {
 163                 if (zone->wait_table) {
 164                         nr_pages = zone->wait_table_hash_nr_entries
 165                                 * sizeof(wait_queue_head_t);
 166                         nr_pages = PAGE_ALIGN(nr_pages) >> PAGE_SHIFT;
 167                         page = virt_to_page(zone->wait_table);
 168
 169                         for (i = 0; i < nr_pages; i++, page++)
 170                                 get_page_bootmem(node, page, NODE_INFO);
 171                 }
 172         }
 173
 174         pfn = pgdat->node_start_pfn;
 175         end_pfn = pfn + pgdat->node_spanned_pages;
 176
 177         /* register_section info */
 178         for (; pfn < end_pfn; pfn += PAGES_PER_SECTION)
 179                 register_page_bootmem_info_section(pfn);
 180
 181 }
 182 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
 183
 184 static void grow_zone_span(struct zone *zone, unsigned long start_pfn,
 185                            unsigned long end_pfn)
 186 {
 187         unsigned long old_zone_end_pfn;
 188
 189         zone_span_writelock(zone);
 190
 191         old_zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages;
 192         if (start_pfn < zone->zone_start_pfn)
 193                 zone->zone_start_pfn = start_pfn;
 194
 195         zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
 196                                 zone->zone_start_pfn;
 197
 198         zone_span_writeunlock(zone);
 199 }
 200
 201 static void grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn,
 202                             unsigned long end_pfn)
 203 {
 204         unsigned long old_pgdat_end_pfn =
 205                 pgdat->node_start_pfn + pgdat->node_spanned_pages;
 206
 207         if (start_pfn < pgdat->node_start_pfn)
 208                 pgdat->node_start_pfn = start_pfn;
 209
 210         pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
 211                                         pgdat->node_start_pfn;
 212 }
 213
 214 static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn)
 215 {
 216         struct pglist_data *pgdat = zone->zone_pgdat;
 217         int nr_pages = PAGES_PER_SECTION;
 218         int nid = pgdat->node_id;
 219         int zone_type;
 220         unsigned long flags;
 221
 222         zone_type = zone - pgdat->node_zones;
 223         if (!zone->wait_table) {
 224                 int ret;
 225
 226                 ret = init_currently_empty_zone(zone, phys_start_pfn,
 227                                                 nr_pages, MEMMAP_HOTPLUG);
 228                 if (ret)
 229                         return ret;
 230         }
 231         pgdat_resize_lock(zone->zone_pgdat, &flags);
 232         grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages);
 233         grow_pgdat_span(zone->zone_pgdat, phys_start_pfn,
 234                         phys_start_pfn + nr_pages);
 235         pgdat_resize_unlock(zone->zone_pgdat, &flags);
 236         memmap_init_zone(nr_pages, nid, zone_type,
 237                          phys_start_pfn, MEMMAP_HOTPLUG);
 238         return 0;
 239 }
 240
 241 static int __meminit __add_section(int nid, struct zone *zone,
 242                                         unsigned long phys_start_pfn)
 243 {
 244         int nr_pages = PAGES_PER_SECTION;
 245         int ret;
 246
 247         if (pfn_valid(phys_start_pfn))
 248                 return -EEXIST;
 249
 250         ret = sparse_add_one_section(zone, phys_start_pfn, nr_pages);
 251
 252         if (ret < 0)
 253                 return ret;
 254
 255         ret = __add_zone(zone, phys_start_pfn);
 256
 257         if (ret < 0)
 258                 return ret;
 259
 260         return register_new_memory(nid, __pfn_to_section(phys_start_pfn));
 261 }
 262
 263 #ifdef CONFIG_SPARSEMEM_VMEMMAP
 264 static int __remove_section(struct zone *zone, struct mem_section *ms)
 265 {
 266         /*
 267          * XXX: Freeing memmap with vmemmap is not implement yet.
 268          *      This should be removed later.
 269          */
 270         return -EBUSY;
 271 }
 272 #else
 273 static int __remove_section(struct zone *zone, struct mem_section *ms)
 274 {
 275         unsigned long flags;
 276         struct pglist_data *pgdat = zone->zone_pgdat;
 277         int ret = -EINVAL;
 278
 279         if (!valid_section(ms))
 280                 return ret;
 281
 282         ret = unregister_memory_section(ms);
 283         if (ret)
 284                 return ret;
 285
 286         pgdat_resize_lock(pgdat, &flags);
 287         sparse_remove_one_section(zone, ms);
 288         pgdat_resize_unlock(pgdat, &flags);
 289         return 0;
 290 }
 291 #endif
 292
 293 /*
 294  * Reasonably generic function for adding memory.  It is
 295  * expected that archs that support memory hotplug will
 296  * call this function after deciding the zone to which to
 297  * add the new pages.
 298  */
 299 int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn,
 300                         unsigned long nr_pages)
 301 {
 302         unsigned long i;
 303         int err = 0;
 304         int start_sec, end_sec;
 305         /* during initialize mem_map, align hot-added range to section */
 306         start_sec = pfn_to_section_nr(phys_start_pfn);
 307         end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
 308
 309         for (i = start_sec; i <= end_sec; i++) {
 310                 err = __add_section(nid, zone, i << PFN_SECTION_SHIFT);
 311
 312                 /*
 313                  * EEXIST is finally dealt with by ioresource collision
 314                  * check. see add_memory() => register_memory_resource()
 315                  * Warning will be printed if there is collision.
 316                  */
 317                 if (err && (err != -EEXIST))
 318                         break;
 319                 err = 0;
 320         }
 321
 322         return err;
 323 }
 324 EXPORT_SYMBOL_GPL(__add_pages);
 325
 326 /**
 327  * __remove_pages() - remove sections of pages from a zone
 328  * @zone: zone from which pages need to be removed
 329  * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
 330  * @nr_pages: number of pages to remove (must be multiple of section size)
 331  *
 332  * Generic helper function to remove section mappings and sysfs entries
 333  * for the section of the memory we are removing. Caller needs to make
 334  * sure that pages are marked reserved and zones are adjust properly by
 335  * calling offline_pages().
 336  */
 337 int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
 338                  unsigned long nr_pages)
 339 {
 340         unsigned long i, ret = 0;
 341         int sections_to_remove;
 342
 343         /*
 344          * We can only remove entire sections
 345          */
 346         BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
 347         BUG_ON(nr_pages % PAGES_PER_SECTION);
 348
 349         sections_to_remove = nr_pages / PAGES_PER_SECTION;
 350         for (i = 0; i < sections_to_remove; i++) {
 351                 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
 352                 release_mem_region(pfn << PAGE_SHIFT,
 353                                    PAGES_PER_SECTION << PAGE_SHIFT);
 354                 ret = __remove_section(zone, __pfn_to_section(pfn));
 355                 if (ret)
 356                         break;
 357         }
 358         return ret;
 359 }
 360 EXPORT_SYMBOL_GPL(__remove_pages);
 361
 362 void online_page(struct page *page)
 363 {
 364         unsigned long pfn = page_to_pfn(page);
 365
 366         totalram_pages++;
 367         if (pfn >= num_physpages)
 368                 num_physpages = pfn + 1;
 369
 370 #ifdef CONFIG_HIGHMEM
 371         if (PageHighMem(page))
 372                 totalhigh_pages++;
 373 #endif
 374
 375 #ifdef CONFIG_FLATMEM
 376         max_mapnr = max(page_to_pfn(page), max_mapnr);
 377 #endif
 378
 379         ClearPageReserved(page);
 380         init_page_count(page);
 381         __free_page(page);
 382 }
 383
 384 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
 385                         void *arg)
 386 {
 387         unsigned long i;
 388         unsigned long onlined_pages = *(unsigned long *)arg;
 389         struct page *page;
 390         if (PageReserved(pfn_to_page(start_pfn)))
 391                 for (i = 0; i < nr_pages; i++) {
 392                         page = pfn_to_page(start_pfn + i);
 393                         online_page(page);
 394                         onlined_pages++;
 395                 }
 396         *(unsigned long *)arg = onlined_pages;
 397         return 0;
 398 }
 399
 400
 401 int online_pages(unsigned long pfn, unsigned long nr_pages)
 402 {
 403         unsigned long onlined_pages = 0;
 404         struct zone *zone;
 405         int need_zonelists_rebuild = 0;
 406         int nid;
 407         int ret;
 408         struct memory_notify arg;
 409
 410         arg.start_pfn = pfn;
 411         arg.nr_pages = nr_pages;
 412         arg.status_change_nid = -1;
 413
 414         nid = page_to_nid(pfn_to_page(pfn));
 415         if (node_present_pages(nid) == 0)
 416                 arg.status_change_nid = nid;
 417
 418         ret = memory_notify(MEM_GOING_ONLINE, &arg);
 419         ret = notifier_to_errno(ret);
 420         if (ret) {
 421                 memory_notify(MEM_CANCEL_ONLINE, &arg);
 422                 return ret;
 423         }
 424         /*
 425          * This doesn't need a lock to do pfn_to_page().
 426          * The section can't be removed here because of the
 427          * memory_block->state_mutex.
 428          */
 429         zone = page_zone(pfn_to_page(pfn));
 430         /*
 431          * If this zone is not populated, then it is not in zonelist.
 432          * This means the page allocator ignores this zone.
 433          * So, zonelist must be updated after online.
 434          */
 435         mutex_lock(&zonelists_mutex);
 436         if (!populated_zone(zone))
 437                 need_zonelists_rebuild = 1;
 438
 439         ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
 440                 online_pages_range);
 441         if (ret) {
 442                 mutex_unlock(&zonelists_mutex);
 443                 printk(KERN_DEBUG "online_pages %lx at %lx failed\n",
 444                         nr_pages, pfn);
 445                 memory_notify(MEM_CANCEL_ONLINE, &arg);
 446                 return ret;
 447         }
 448
 449         zone->present_pages += onlined_pages;
 450         zone->zone_pgdat->node_present_pages += onlined_pages;
 451         if (need_zonelists_rebuild)
 452                 build_all_zonelists(zone);
 453         else
 454                 zone_pcp_update(zone);
 455
 456         mutex_unlock(&zonelists_mutex);
 457         setup_per_zone_wmarks();
 458         calculate_zone_inactive_ratio(zone);
 459         if (onlined_pages) {
 460                 kswapd_run(zone_to_nid(zone));
 461                 node_set_state(zone_to_nid(zone), N_HIGH_MEMORY);
 462         }
 463
 464         vm_total_pages = nr_free_pagecache_pages();
 465
 466         writeback_set_ratelimit();
 467
 468         if (onlined_pages)
 469                 memory_notify(MEM_ONLINE, &arg);
 470
 471         return 0;
 472 }
 473 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
 474
 475 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
 476 static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
 477 {
 478         struct pglist_data *pgdat;
 479         unsigned long zones_size[MAX_NR_ZONES] = {0};
 480         unsigned long zholes_size[MAX_NR_ZONES] = {0};
 481         unsigned long start_pfn = start >> PAGE_SHIFT;
 482
 483         pgdat = arch_alloc_nodedata(nid);
 484         if (!pgdat)
 485                 return NULL;
 486
 487         arch_refresh_nodedata(nid, pgdat);
 488
 489         /* we can use NODE_DATA(nid) from here */
 490
 491         /* init node's zones as empty zones, we don't have any present pages.*/
 492         free_area_init_node(nid, zones_size, start_pfn, zholes_size);
 493
 494         return pgdat;
 495 }
 496
 497 static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
 498 {
 499         arch_refresh_nodedata(nid, NULL);
 500         arch_free_nodedata(pgdat);
 501         return;
 502 }
 503
 504
 505 /*
 506  * called by cpu_up() to online a node without onlined memory.
 507  */
 508 int mem_online_node(int nid)
 509 {
 510         pg_data_t       *pgdat;
 511         int     ret;
 512
 513         lock_memory_hotplug();
 514         pgdat = hotadd_new_pgdat(nid, 0);
 515         if (pgdat) {
 516                 ret = -ENOMEM;
 517                 goto out;
 518         }
 519         node_set_online(nid);
 520         ret = register_one_node(nid);
 521         BUG_ON(ret);
 522
 523 out:
 524         unlock_memory_hotplug();
 525         return ret;
 526 }
 527
 528 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
 529 int __ref add_memory(int nid, u64 start, u64 size)
 530 {
 531         pg_data_t *pgdat = NULL;
 532         int new_pgdat = 0;
 533         struct resource *res;
 534         int ret;
 535
 536         lock_memory_hotplug();
 537
 538         res = register_memory_resource(start, size);
 539         ret = -EEXIST;
 540         if (!res)
 541                 goto out;
 542
 543         if (!node_online(nid)) {
 544                 pgdat = hotadd_new_pgdat(nid, start);
 545                 ret = -ENOMEM;
 546                 if (!pgdat)
 547                         goto out;
 548                 new_pgdat = 1;
 549         }
 550
 551         /* call arch's memory hotadd */
 552         ret = arch_add_memory(nid, start, size);
 553
 554         if (ret < 0)
 555                 goto error;
 556
 557         /* we online node here. we can't roll back from here. */
 558         node_set_online(nid);
 559
 560         if (new_pgdat) {
 561                 ret = register_one_node(nid);
 562                 /*
 563                  * If sysfs file of new node can't create, cpu on the node
 564                  * can't be hot-added. There is no rollback way now.
 565                  * So, check by BUG_ON() to catch it reluctantly..
 566                  */
 567                 BUG_ON(ret);
 568         }
 569
 570         /* create new memmap entry */
 571         firmware_map_add_hotplug(start, start + size, "System RAM");
 572
 573         goto out;
 574
 575 error:
 576         /* rollback pgdat allocation and others */
 577         if (new_pgdat)
 578                 rollback_node_hotadd(nid, pgdat);
 579         if (res)
 580                 release_memory_resource(res);
 581
 582 out:
 583         unlock_memory_hotplug();
 584         return ret;
 585 }
 586 EXPORT_SYMBOL_GPL(add_memory);
 587
 588 #ifdef CONFIG_MEMORY_HOTREMOVE
 589 /*
 590  * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
 591  * set and the size of the free page is given by page_order(). Using this,
 592  * the function determines if the pageblock contains only free pages.
 593  * Due to buddy contraints, a free page at least the size of a pageblock will
 594  * be located at the start of the pageblock
 595  */
 596 static inline int pageblock_free(struct page *page)
 597 {
 598         return PageBuddy(page) && page_order(page) >= pageblock_order;
 599 }
 600
 601 /* Return the start of the next active pageblock after a given page */
 602 static struct page *next_active_pageblock(struct page *page)
 603 {
 604         /* Ensure the starting page is pageblock-aligned */
 605         BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
 606
 607         /* If the entire pageblock is free, move to the end of free page */
 608         if (pageblock_free(page)) {
 609                 int order;
 610                 /* be careful. we don't have locks, page_order can be changed.*/
 611                 order = page_order(page);
 612                 if ((order < MAX_ORDER) && (order >= pageblock_order))
 613                         return page + (1 << order);
 614         }
 615
 616         return page + pageblock_nr_pages;
 617 }
 618
 619 /* Checks if this range of memory is likely to be hot-removable. */
 620 int is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
 621 {
 622         struct page *page = pfn_to_page(start_pfn);
 623         struct page *end_page = page + nr_pages;
 624
 625         /* Check the starting page of each pageblock within the range */
 626         for (; page < end_page; page = next_active_pageblock(page)) {
 627                 if (!is_pageblock_removable_nolock(page))
 628                         return 0;
 629                 cond_resched();
 630         }
 631
 632         /* All pageblocks in the memory block are likely to be hot-removable */
 633         return 1;
 634 }
 635
 636 /*
 637  * Confirm all pages in a range [start, end) is belongs to the same zone.
 638  */
 639 static int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn)
 640 {
 641         unsigned long pfn;
 642         struct zone *zone = NULL;
 643         struct page *page;
 644         int i;
 645         for (pfn = start_pfn;
 646              pfn < end_pfn;
 647              pfn += MAX_ORDER_NR_PAGES) {
 648                 i = 0;
 649                 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
 650                 while ((i < MAX_ORDER_NR_PAGES) && !pfn_valid_within(pfn + i))
 651                         i++;
 652                 if (i == MAX_ORDER_NR_PAGES)
 653                         continue;
 654                 page = pfn_to_page(pfn + i);
 655                 if (zone && page_zone(page) != zone)
 656                         return 0;
 657                 zone = page_zone(page);
 658         }
 659         return 1;
 660 }
 661
 662 /*
 663  * Scanning pfn is much easier than scanning lru list.
 664  * Scan pfn from start to end and Find LRU page.
 665  */
 666 static unsigned long scan_lru_pages(unsigned long start, unsigned long end)
 667 {
 668         unsigned long pfn;
 669         struct page *page;
 670         for (pfn = start; pfn < end; pfn++) {
 671                 if (pfn_valid(pfn)) {
 672                         page = pfn_to_page(pfn);
 673                         if (PageLRU(page))
 674                                 return pfn;
 675                 }
 676         }
 677         return 0;
 678 }
 679
 680 static struct page *
 681 hotremove_migrate_alloc(struct page *page, unsigned long private, int **x)
 682 {
 683         /* This should be improooooved!! */
 684         return alloc_page(GFP_HIGHUSER_MOVABLE);
 685 }
 686
 687 #define NR_OFFLINE_AT_ONCE_PAGES        (256)
 688 static int
 689 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
 690 {
 691         unsigned long pfn;
 692         struct page *page;
 693         int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
 694         int not_managed = 0;
 695         int ret = 0;
 696         LIST_HEAD(source);
 697
 698         for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
 699                 if (!pfn_valid(pfn))
 700                         continue;
 701                 page = pfn_to_page(pfn);
 702                 if (!page_count(page))
 703                         continue;
 704                 /*
 705                  * We can skip free pages. And we can only deal with pages on
 706                  * LRU.
 707                  */
 708                 ret = isolate_lru_page(page);
 709                 if (!ret) { /* Success */
 710                         list_add_tail(&page->lru, &source);
 711                         move_pages--;
 712                         inc_zone_page_state(page, NR_ISOLATED_ANON +
 713                                             page_is_file_cache(page));
 714
 715                 } else {
 716 #ifdef CONFIG_DEBUG_VM
 717                         printk(KERN_ALERT "removing pfn %lx from LRU failed\n",
 718                                pfn);
 719                         dump_page(page);
 720 #endif
 721                         /* Becasue we don't have big zone->lock. we should
 722                            check this again here. */
 723                         if (page_count(page)) {
 724                                 not_managed++;
 725                                 ret = -EBUSY;
 726                                 break;
 727                         }
 728                 }
 729         }
 730         if (!list_empty(&source)) {
 731                 if (not_managed) {
 732                         putback_lru_pages(&source);
 733                         goto out;
 734                 }
 735                 /* this function returns # of failed pages */
 736                 ret = migrate_pages(&source, hotremove_migrate_alloc, 0, 1);
 737                 if (ret)
 738                         putback_lru_pages(&source);
 739         }
 740 out:
 741         return ret;
 742 }
 743
 744 /*
 745  * remove from free_area[] and mark all as Reserved.
 746  */
 747 static int
 748 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
 749                         void *data)
 750 {
 751         __offline_isolated_pages(start, start + nr_pages);
 752         return 0;
 753 }
 754
 755 static void
 756 offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
 757 {
 758         walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
 759                                 offline_isolated_pages_cb);
 760 }
 761
 762 /*
 763  * Check all pages in range, recoreded as memory resource, are isolated.
 764  */
 765 static int
 766 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
 767                         void *data)
 768 {
 769         int ret;
 770         long offlined = *(long *)data;
 771         ret = test_pages_isolated(start_pfn, start_pfn + nr_pages);
 772         offlined = nr_pages;
 773         if (!ret)
 774                 *(long *)data += offlined;
 775         return ret;
 776 }
 777
 778 static long
 779 check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
 780 {
 781         long offlined = 0;
 782         int ret;
 783
 784         ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
 785                         check_pages_isolated_cb);
 786         if (ret < 0)
 787                 offlined = (long)ret;
 788         return offlined;
 789 }
 790
 791 static int offline_pages(unsigned long start_pfn,
 792                   unsigned long end_pfn, unsigned long timeout)
 793 {
 794         unsigned long pfn, nr_pages, expire;
 795         long offlined_pages;
 796         int ret, drain, retry_max, node;
 797         struct zone *zone;
 798         struct memory_notify arg;
 799
 800         BUG_ON(start_pfn >= end_pfn);
 801         /* at least, alignment against pageblock is necessary */
 802         if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
 803                 return -EINVAL;
 804         if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
 805                 return -EINVAL;
 806         /* This makes hotplug much easier...and readable.
 807            we assume this for now. .*/
 808         if (!test_pages_in_a_zone(start_pfn, end_pfn))
 809                 return -EINVAL;
 810
 811         lock_memory_hotplug();
 812
 813         zone = page_zone(pfn_to_page(start_pfn));
 814         node = zone_to_nid(zone);
 815         nr_pages = end_pfn - start_pfn;
 816
 817         /* set above range as isolated */
 818         ret = start_isolate_page_range(start_pfn, end_pfn);
 819         if (ret)
 820                 goto out;
 821
 822         arg.start_pfn = start_pfn;
 823         arg.nr_pages = nr_pages;
 824         arg.status_change_nid = -1;
 825         if (nr_pages >= node_present_pages(node))
 826                 arg.status_change_nid = node;
 827
 828         ret = memory_notify(MEM_GOING_OFFLINE, &arg);
 829         ret = notifier_to_errno(ret);
 830         if (ret)
 831                 goto failed_removal;
 832
 833         pfn = start_pfn;
 834         expire = jiffies + timeout;
 835         drain = 0;
 836         retry_max = 5;
 837 repeat:
 838         /* start memory hot removal */
 839         ret = -EAGAIN;
 840         if (time_after(jiffies, expire))
 841                 goto failed_removal;
 842         ret = -EINTR;
 843         if (signal_pending(current))
 844                 goto failed_removal;
 845         ret = 0;
 846         if (drain) {
 847                 lru_add_drain_all();
 848                 cond_resched();
 849                 drain_all_pages();
 850         }
 851
 852         pfn = scan_lru_pages(start_pfn, end_pfn);
 853         if (pfn) { /* We have page on LRU */
 854                 ret = do_migrate_range(pfn, end_pfn);
 855                 if (!ret) {
 856                         drain = 1;
 857                         goto repeat;
 858                 } else {
 859                         if (ret < 0)
 860                                 if (--retry_max == 0)
 861                                         goto failed_removal;
 862                         yield();
 863                         drain = 1;
 864                         goto repeat;
 865                 }
 866         }
 867         /* drain all zone's lru pagevec, this is asyncronous... */
 868         lru_add_drain_all();
 869         yield();
 870         /* drain pcp pages , this is synchrouns. */
 871         drain_all_pages();
 872         /* check again */
 873         offlined_pages = check_pages_isolated(start_pfn, end_pfn);
 874         if (offlined_pages < 0) {
 875                 ret = -EBUSY;
 876                 goto failed_removal;
 877         }
 878         printk(KERN_INFO "Offlined Pages %ld\n", offlined_pages);
 879         /* Ok, all of our target is islaoted.
 880            We cannot do rollback at this point. */
 881         offline_isolated_pages(start_pfn, end_pfn);
 882         /* reset pagetype flags and makes migrate type to be MOVABLE */
 883         undo_isolate_page_range(start_pfn, end_pfn);
 884         /* removal success */
 885         zone->present_pages -= offlined_pages;
 886         zone->zone_pgdat->node_present_pages -= offlined_pages;
 887         totalram_pages -= offlined_pages;
 888
 889         setup_per_zone_wmarks();
 890         calculate_zone_inactive_ratio(zone);
 891         if (!node_present_pages(node)) {
 892                 node_clear_state(node, N_HIGH_MEMORY);
 893                 kswapd_stop(node);
 894         }
 895
 896         vm_total_pages = nr_free_pagecache_pages();
 897         writeback_set_ratelimit();
 898
 899         memory_notify(MEM_OFFLINE, &arg);
 900         unlock_memory_hotplug();
 901         return 0;
 902
 903 failed_removal:
 904         printk(KERN_INFO "memory offlining %lx to %lx failed\n",
 905                 start_pfn, end_pfn);
 906         memory_notify(MEM_CANCEL_OFFLINE, &arg);
 907         /* pushback to free area */
 908         undo_isolate_page_range(start_pfn, end_pfn);
 909
 910 out:
 911         unlock_memory_hotplug();
 912         return ret;
 913 }
 914
 915 int remove_memory(u64 start, u64 size)
 916 {
 917         unsigned long start_pfn, end_pfn;
 918
 919         start_pfn = PFN_DOWN(start);
 920         end_pfn = start_pfn + PFN_DOWN(size);
 921         return offline_pages(start_pfn, end_pfn, 120 * HZ);
 922 }
 923 #else
 924 int remove_memory(u64 start, u64 size)
 925 {
 926         return -EINVAL;
 927 }
 928 #endif /* CONFIG_MEMORY_HOTREMOVE */
 929 EXPORT_SYMBOL_GPL(remove_memory);