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