mm/bootmem.c

   1 /*
   2  *  bootmem - A boot-time physical memory allocator and configurator
   3  *
   4  *  Copyright (C) 1999 Ingo Molnar
   5  *                1999 Kanoj Sarcar, SGI
   6  *                2008 Johannes Weiner
   7  *
   8  * Access to this subsystem has to be serialized externally (which is true
   9  * for the boot process anyway).
  10  */
  11 #include <linux/init.h>
  12 #include <linux/pfn.h>
  13 #include <linux/slab.h>
  14 #include <linux/bootmem.h>
  15 #include <linux/export.h>
  16 #include <linux/kmemleak.h>
  17 #include <linux/range.h>
  18 #include <linux/memblock.h>
  19
  20 #include <asm/bug.h>
  21 #include <asm/io.h>
  22 #include <asm/processor.h>
  23
  24 #include "internal.h"
  25
  26 #ifndef CONFIG_NEED_MULTIPLE_NODES
  27 struct pglist_data __refdata contig_page_data = {
  28         .bdata = &bootmem_node_data[0]
  29 };
  30 EXPORT_SYMBOL(contig_page_data);
  31 #endif
  32
  33 unsigned long max_low_pfn;
  34 unsigned long min_low_pfn;
  35 unsigned long max_pfn;
  36
  37 bootmem_data_t bootmem_node_data[MAX_NUMNODES] __initdata;
  38
  39 static struct list_head bdata_list __initdata = LIST_HEAD_INIT(bdata_list);
  40
  41 static int bootmem_debug;
  42
  43 static int __init bootmem_debug_setup(char *buf)
  44 {
  45         bootmem_debug = 1;
  46         return 0;
  47 }
  48 early_param("bootmem_debug", bootmem_debug_setup);
  49
  50 #define bdebug(fmt, args...) ({                         \
  51         if (unlikely(bootmem_debug))                    \
  52                 printk(KERN_INFO                        \
  53                         "bootmem::%s " fmt,             \
  54                         __func__, ## args);             \
  55 })
  56
  57 static unsigned long __init bootmap_bytes(unsigned long pages)
  58 {
  59         unsigned long bytes = DIV_ROUND_UP(pages, 8);
  60
  61         return ALIGN(bytes, sizeof(long));
  62 }
  63
  64 /**
  65  * bootmem_bootmap_pages - calculate bitmap size in pages
  66  * @pages: number of pages the bitmap has to represent
  67  */
  68 unsigned long __init bootmem_bootmap_pages(unsigned long pages)
  69 {
  70         unsigned long bytes = bootmap_bytes(pages);
  71
  72         return PAGE_ALIGN(bytes) >> PAGE_SHIFT;
  73 }
  74
  75 /*
  76  * link bdata in order
  77  */
  78 static void __init link_bootmem(bootmem_data_t *bdata)
  79 {
  80         bootmem_data_t *ent;
  81
  82         list_for_each_entry(ent, &bdata_list, list) {
  83                 if (bdata->node_min_pfn < ent->node_min_pfn) {
  84                         list_add_tail(&bdata->list, &ent->list);
  85                         return;
  86                 }
  87         }
  88
  89         list_add_tail(&bdata->list, &bdata_list);
  90 }
  91
  92 /*
  93  * Called once to set up the allocator itself.
  94  */
  95 static unsigned long __init init_bootmem_core(bootmem_data_t *bdata,
  96         unsigned long mapstart, unsigned long start, unsigned long end)
  97 {
  98         unsigned long mapsize;
  99
 100         mminit_validate_memmodel_limits(&start, &end);
 101         bdata->node_bootmem_map = phys_to_virt(PFN_PHYS(mapstart));
 102         bdata->node_min_pfn = start;
 103         bdata->node_low_pfn = end;
 104         link_bootmem(bdata);
 105
 106         /*
 107          * Initially all pages are reserved - setup_arch() has to
 108          * register free RAM areas explicitly.
 109          */
 110         mapsize = bootmap_bytes(end - start);
 111         memset(bdata->node_bootmem_map, 0xff, mapsize);
 112
 113         bdebug("nid=%td start=%lx map=%lx end=%lx mapsize=%lx\n",
 114                 bdata - bootmem_node_data, start, mapstart, end, mapsize);
 115
 116         return mapsize;
 117 }
 118
 119 /**
 120  * init_bootmem_node - register a node as boot memory
 121  * @pgdat: node to register
 122  * @freepfn: pfn where the bitmap for this node is to be placed
 123  * @startpfn: first pfn on the node
 124  * @endpfn: first pfn after the node
 125  *
 126  * Returns the number of bytes needed to hold the bitmap for this node.
 127  */
 128 unsigned long __init init_bootmem_node(pg_data_t *pgdat, unsigned long freepfn,
 129                                 unsigned long startpfn, unsigned long endpfn)
 130 {
 131         return init_bootmem_core(pgdat->bdata, freepfn, startpfn, endpfn);
 132 }
 133
 134 /**
 135  * init_bootmem - register boot memory
 136  * @start: pfn where the bitmap is to be placed
 137  * @pages: number of available physical pages
 138  *
 139  * Returns the number of bytes needed to hold the bitmap.
 140  */
 141 unsigned long __init init_bootmem(unsigned long start, unsigned long pages)
 142 {
 143         max_low_pfn = pages;
 144         min_low_pfn = start;
 145         return init_bootmem_core(NODE_DATA(0)->bdata, start, 0, pages);
 146 }
 147
 148 /*
 149  * free_bootmem_late - free bootmem pages directly to page allocator
 150  * @addr: starting address of the range
 151  * @size: size of the range in bytes
 152  *
 153  * This is only useful when the bootmem allocator has already been torn
 154  * down, but we are still initializing the system.  Pages are given directly
 155  * to the page allocator, no bootmem metadata is updated because it is gone.
 156  */
 157 void __init free_bootmem_late(unsigned long addr, unsigned long size)
 158 {
 159         unsigned long cursor, end;
 160
 161         kmemleak_free_part(__va(addr), size);
 162
 163         cursor = PFN_UP(addr);
 164         end = PFN_DOWN(addr + size);
 165
 166         for (; cursor < end; cursor++) {
 167                 __free_pages_bootmem(pfn_to_page(cursor), 0);
 168                 totalram_pages++;
 169         }
 170 }
 171
 172 static unsigned long __init free_all_bootmem_core(bootmem_data_t *bdata)
 173 {
 174         struct page *page;
 175         unsigned long start, end, pages, count = 0;
 176
 177         if (!bdata->node_bootmem_map)
 178                 return 0;
 179
 180         start = bdata->node_min_pfn;
 181         end = bdata->node_low_pfn;
 182
 183         bdebug("nid=%td start=%lx end=%lx\n",
 184                 bdata - bootmem_node_data, start, end);
 185
 186         while (start < end) {
 187                 unsigned long *map, idx, vec;
 188                 unsigned shift;
 189
 190                 map = bdata->node_bootmem_map;
 191                 idx = start - bdata->node_min_pfn;
 192                 shift = idx & (BITS_PER_LONG - 1);
 193                 /*
 194                  * vec holds at most BITS_PER_LONG map bits,
 195                  * bit 0 corresponds to start.
 196                  */
 197                 vec = ~map[idx / BITS_PER_LONG];
 198
 199                 if (shift) {
 200                         vec >>= shift;
 201                         if (end - start >= BITS_PER_LONG)
 202                                 vec |= ~map[idx / BITS_PER_LONG + 1] <<
 203                                         (BITS_PER_LONG - shift);
 204                 }
 205                 /*
 206                  * If we have a properly aligned and fully unreserved
 207                  * BITS_PER_LONG block of pages in front of us, free
 208                  * it in one go.
 209                  */
 210                 if (IS_ALIGNED(start, BITS_PER_LONG) && vec == ~0UL) {
 211                         int order = ilog2(BITS_PER_LONG);
 212
 213                         __free_pages_bootmem(pfn_to_page(start), order);
 214                         count += BITS_PER_LONG;
 215                         start += BITS_PER_LONG;
 216                 } else {
 217                         unsigned long cur = start;
 218
 219                         start = ALIGN(start + 1, BITS_PER_LONG);
 220                         while (vec && cur != start) {
 221                                 if (vec & 1) {
 222                                         page = pfn_to_page(cur);
 223                                         __free_pages_bootmem(page, 0);
 224                                         count++;
 225                                 }
 226                                 vec >>= 1;
 227                                 ++cur;
 228                         }
 229                 }
 230         }
 231
 232         page = virt_to_page(bdata->node_bootmem_map);
 233         pages = bdata->node_low_pfn - bdata->node_min_pfn;
 234         pages = bootmem_bootmap_pages(pages);
 235         count += pages;
 236         while (pages--)
 237                 __free_pages_bootmem(page++, 0);
 238
 239         bdebug("nid=%td released=%lx\n", bdata - bootmem_node_data, count);
 240
 241         return count;
 242 }
 243
 244 /**
 245  * free_all_bootmem_node - release a node's free pages to the buddy allocator
 246  * @pgdat: node to be released
 247  *
 248  * Returns the number of pages actually released.
 249  */
 250 unsigned long __init free_all_bootmem_node(pg_data_t *pgdat)
 251 {
 252         register_page_bootmem_info_node(pgdat);
 253         return free_all_bootmem_core(pgdat->bdata);
 254 }
 255
 256 /**
 257  * free_all_bootmem - release free pages to the buddy allocator
 258  *
 259  * Returns the number of pages actually released.
 260  */
 261 unsigned long __init free_all_bootmem(void)
 262 {
 263         unsigned long total_pages = 0;
 264         bootmem_data_t *bdata;
 265
 266         list_for_each_entry(bdata, &bdata_list, list)
 267                 total_pages += free_all_bootmem_core(bdata);
 268
 269         return total_pages;
 270 }
 271
 272 static void __init __free(bootmem_data_t *bdata,
 273                         unsigned long sidx, unsigned long eidx)
 274 {
 275         unsigned long idx;
 276
 277         bdebug("nid=%td start=%lx end=%lx\n", bdata - bootmem_node_data,
 278                 sidx + bdata->node_min_pfn,
 279                 eidx + bdata->node_min_pfn);
 280
 281         if (bdata->hint_idx > sidx)
 282                 bdata->hint_idx = sidx;
 283
 284         for (idx = sidx; idx < eidx; idx++)
 285                 if (!test_and_clear_bit(idx, bdata->node_bootmem_map))
 286                         BUG();
 287 }
 288
 289 static int __init __reserve(bootmem_data_t *bdata, unsigned long sidx,
 290                         unsigned long eidx, int flags)
 291 {
 292         unsigned long idx;
 293         int exclusive = flags & BOOTMEM_EXCLUSIVE;
 294
 295         bdebug("nid=%td start=%lx end=%lx flags=%x\n",
 296                 bdata - bootmem_node_data,
 297                 sidx + bdata->node_min_pfn,
 298                 eidx + bdata->node_min_pfn,
 299                 flags);
 300
 301         for (idx = sidx; idx < eidx; idx++)
 302                 if (test_and_set_bit(idx, bdata->node_bootmem_map)) {
 303                         if (exclusive) {
 304                                 __free(bdata, sidx, idx);
 305                                 return -EBUSY;
 306                         }
 307                         bdebug("silent double reserve of PFN %lx\n",
 308                                 idx + bdata->node_min_pfn);
 309                 }
 310         return 0;
 311 }
 312
 313 static int __init mark_bootmem_node(bootmem_data_t *bdata,
 314                                 unsigned long start, unsigned long end,
 315                                 int reserve, int flags)
 316 {
 317         unsigned long sidx, eidx;
 318
 319         bdebug("nid=%td start=%lx end=%lx reserve=%d flags=%x\n",
 320                 bdata - bootmem_node_data, start, end, reserve, flags);
 321
 322         BUG_ON(start < bdata->node_min_pfn);
 323         BUG_ON(end > bdata->node_low_pfn);
 324
 325         sidx = start - bdata->node_min_pfn;
 326         eidx = end - bdata->node_min_pfn;
 327
 328         if (reserve)
 329                 return __reserve(bdata, sidx, eidx, flags);
 330         else
 331                 __free(bdata, sidx, eidx);
 332         return 0;
 333 }
 334
 335 static int __init mark_bootmem(unsigned long start, unsigned long end,
 336                                 int reserve, int flags)
 337 {
 338         unsigned long pos;
 339         bootmem_data_t *bdata;
 340
 341         pos = start;
 342         list_for_each_entry(bdata, &bdata_list, list) {
 343                 int err;
 344                 unsigned long max;
 345
 346                 if (pos < bdata->node_min_pfn ||
 347                     pos >= bdata->node_low_pfn) {
 348                         BUG_ON(pos != start);
 349                         continue;
 350                 }
 351
 352                 max = min(bdata->node_low_pfn, end);
 353
 354                 err = mark_bootmem_node(bdata, pos, max, reserve, flags);
 355                 if (reserve && err) {
 356                         mark_bootmem(start, pos, 0, 0);
 357                         return err;
 358                 }
 359
 360                 if (max == end)
 361                         return 0;
 362                 pos = bdata->node_low_pfn;
 363         }
 364         BUG();
 365 }
 366
 367 /**
 368  * free_bootmem_node - mark a page range as usable
 369  * @pgdat: node the range resides on
 370  * @physaddr: starting address of the range
 371  * @size: size of the range in bytes
 372  *
 373  * Partial pages will be considered reserved and left as they are.
 374  *
 375  * The range must reside completely on the specified node.
 376  */
 377 void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
 378                               unsigned long size)
 379 {
 380         unsigned long start, end;
 381
 382         kmemleak_free_part(__va(physaddr), size);
 383
 384         start = PFN_UP(physaddr);
 385         end = PFN_DOWN(physaddr + size);
 386
 387         mark_bootmem_node(pgdat->bdata, start, end, 0, 0);
 388 }
 389
 390 /**
 391  * free_bootmem - mark a page range as usable
 392  * @addr: starting address of the range
 393  * @size: size of the range in bytes
 394  *
 395  * Partial pages will be considered reserved and left as they are.
 396  *
 397  * The range must be contiguous but may span node boundaries.
 398  */
 399 void __init free_bootmem(unsigned long addr, unsigned long size)
 400 {
 401         unsigned long start, end;
 402
 403         kmemleak_free_part(__va(addr), size);
 404
 405         start = PFN_UP(addr);
 406         end = PFN_DOWN(addr + size);
 407
 408         mark_bootmem(start, end, 0, 0);
 409 }
 410
 411 /**
 412  * reserve_bootmem_node - mark a page range as reserved
 413  * @pgdat: node the range resides on
 414  * @physaddr: starting address of the range
 415  * @size: size of the range in bytes
 416  * @flags: reservation flags (see linux/bootmem.h)
 417  *
 418  * Partial pages will be reserved.
 419  *
 420  * The range must reside completely on the specified node.
 421  */
 422 int __init reserve_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
 423                                  unsigned long size, int flags)
 424 {
 425         unsigned long start, end;
 426
 427         start = PFN_DOWN(physaddr);
 428         end = PFN_UP(physaddr + size);
 429
 430         return mark_bootmem_node(pgdat->bdata, start, end, 1, flags);
 431 }
 432
 433 /**
 434  * reserve_bootmem - mark a page range as reserved
 435  * @addr: starting address of the range
 436  * @size: size of the range in bytes
 437  * @flags: reservation flags (see linux/bootmem.h)
 438  *
 439  * Partial pages will be reserved.
 440  *
 441  * The range must be contiguous but may span node boundaries.
 442  */
 443 int __init reserve_bootmem(unsigned long addr, unsigned long size,
 444                             int flags)
 445 {
 446         unsigned long start, end;
 447
 448         start = PFN_DOWN(addr);
 449         end = PFN_UP(addr + size);
 450
 451         return mark_bootmem(start, end, 1, flags);
 452 }
 453
 454 int __weak __init reserve_bootmem_generic(unsigned long phys, unsigned long len,
 455                                    int flags)
 456 {
 457         return reserve_bootmem(phys, len, flags);
 458 }
 459
 460 static unsigned long __init align_idx(struct bootmem_data *bdata,
 461                                       unsigned long idx, unsigned long step)
 462 {
 463         unsigned long base = bdata->node_min_pfn;
 464
 465         /*
 466          * Align the index with respect to the node start so that the
 467          * combination of both satisfies the requested alignment.
 468          */
 469
 470         return ALIGN(base + idx, step) - base;
 471 }
 472
 473 static unsigned long __init align_off(struct bootmem_data *bdata,
 474                                       unsigned long off, unsigned long align)
 475 {
 476         unsigned long base = PFN_PHYS(bdata->node_min_pfn);
 477
 478         /* Same as align_idx for byte offsets */
 479
 480         return ALIGN(base + off, align) - base;
 481 }
 482
 483 static void * __init alloc_bootmem_bdata(struct bootmem_data *bdata,
 484                                         unsigned long size, unsigned long align,
 485                                         unsigned long goal, unsigned long limit)
 486 {
 487         unsigned long fallback = 0;
 488         unsigned long min, max, start, sidx, midx, step;
 489
 490         bdebug("nid=%td size=%lx [%lu pages] align=%lx goal=%lx limit=%lx\n",
 491                 bdata - bootmem_node_data, size, PAGE_ALIGN(size) >> PAGE_SHIFT,
 492                 align, goal, limit);
 493
 494         BUG_ON(!size);
 495         BUG_ON(align & (align - 1));
 496         BUG_ON(limit && goal + size > limit);
 497
 498         if (!bdata->node_bootmem_map)
 499                 return NULL;
 500
 501         min = bdata->node_min_pfn;
 502         max = bdata->node_low_pfn;
 503
 504         goal >>= PAGE_SHIFT;
 505         limit >>= PAGE_SHIFT;
 506
 507         if (limit && max > limit)
 508                 max = limit;
 509         if (max <= min)
 510                 return NULL;
 511
 512         step = max(align >> PAGE_SHIFT, 1UL);
 513
 514         if (goal && min < goal && goal < max)
 515                 start = ALIGN(goal, step);
 516         else
 517                 start = ALIGN(min, step);
 518
 519         sidx = start - bdata->node_min_pfn;
 520         midx = max - bdata->node_min_pfn;
 521
 522         if (bdata->hint_idx > sidx) {
 523                 /*
 524                  * Handle the valid case of sidx being zero and still
 525                  * catch the fallback below.
 526                  */
 527                 fallback = sidx + 1;
 528                 sidx = align_idx(bdata, bdata->hint_idx, step);
 529         }
 530
 531         while (1) {
 532                 int merge;
 533                 void *region;
 534                 unsigned long eidx, i, start_off, end_off;
 535 find_block:
 536                 sidx = find_next_zero_bit(bdata->node_bootmem_map, midx, sidx);
 537                 sidx = align_idx(bdata, sidx, step);
 538                 eidx = sidx + PFN_UP(size);
 539
 540                 if (sidx >= midx || eidx > midx)
 541                         break;
 542
 543                 for (i = sidx; i < eidx; i++)
 544                         if (test_bit(i, bdata->node_bootmem_map)) {
 545                                 sidx = align_idx(bdata, i, step);
 546                                 if (sidx == i)
 547                                         sidx += step;
 548                                 goto find_block;
 549                         }
 550
 551                 if (bdata->last_end_off & (PAGE_SIZE - 1) &&
 552                                 PFN_DOWN(bdata->last_end_off) + 1 == sidx)
 553                         start_off = align_off(bdata, bdata->last_end_off, align);
 554                 else
 555                         start_off = PFN_PHYS(sidx);
 556
 557                 merge = PFN_DOWN(start_off) < sidx;
 558                 end_off = start_off + size;
 559
 560                 bdata->last_end_off = end_off;
 561                 bdata->hint_idx = PFN_UP(end_off);
 562
 563                 /*
 564                  * Reserve the area now:
 565                  */
 566                 if (__reserve(bdata, PFN_DOWN(start_off) + merge,
 567                                 PFN_UP(end_off), BOOTMEM_EXCLUSIVE))
 568                         BUG();
 569
 570                 region = phys_to_virt(PFN_PHYS(bdata->node_min_pfn) +
 571                                 start_off);
 572                 memset(region, 0, size);
 573                 /*
 574                  * The min_count is set to 0 so that bootmem allocated blocks
 575                  * are never reported as leaks.
 576                  */
 577                 kmemleak_alloc(region, size, 0, 0);
 578                 return region;
 579         }
 580
 581         if (fallback) {
 582                 sidx = align_idx(bdata, fallback - 1, step);
 583                 fallback = 0;
 584                 goto find_block;
 585         }
 586
 587         return NULL;
 588 }
 589
 590 static void * __init alloc_arch_preferred_bootmem(bootmem_data_t *bdata,
 591                                         unsigned long size, unsigned long align,
 592                                         unsigned long goal, unsigned long limit)
 593 {
 594         if (WARN_ON_ONCE(slab_is_available()))
 595                 return kzalloc(size, GFP_NOWAIT);
 596
 597 #ifdef CONFIG_HAVE_ARCH_BOOTMEM
 598         {
 599                 bootmem_data_t *p_bdata;
 600
 601                 p_bdata = bootmem_arch_preferred_node(bdata, size, align,
 602                                                         goal, limit);
 603                 if (p_bdata)
 604                         return alloc_bootmem_bdata(p_bdata, size, align,
 605                                                         goal, limit);
 606         }
 607 #endif
 608         return NULL;
 609 }
 610
 611 static void * __init alloc_bootmem_core(unsigned long size,
 612                                         unsigned long align,
 613                                         unsigned long goal,
 614                                         unsigned long limit)
 615 {
 616         bootmem_data_t *bdata;
 617         void *region;
 618
 619         region = alloc_arch_preferred_bootmem(NULL, size, align, goal, limit);
 620         if (region)
 621                 return region;
 622
 623         list_for_each_entry(bdata, &bdata_list, list) {
 624                 if (goal && bdata->node_low_pfn <= PFN_DOWN(goal))
 625                         continue;
 626                 if (limit && bdata->node_min_pfn >= PFN_DOWN(limit))
 627                         break;
 628
 629                 region = alloc_bootmem_bdata(bdata, size, align, goal, limit);
 630                 if (region)
 631                         return region;
 632         }
 633
 634         return NULL;
 635 }
 636
 637 static void * __init ___alloc_bootmem_nopanic(unsigned long size,
 638                                               unsigned long align,
 639                                               unsigned long goal,
 640                                               unsigned long limit)
 641 {
 642         void *ptr;
 643
 644 restart:
 645         ptr = alloc_bootmem_core(size, align, goal, limit);
 646         if (ptr)
 647                 return ptr;
 648         if (goal) {
 649                 goal = 0;
 650                 goto restart;
 651         }
 652
 653         return NULL;
 654 }
 655
 656 /**
 657  * __alloc_bootmem_nopanic - allocate boot memory without panicking
 658  * @size: size of the request in bytes
 659  * @align: alignment of the region
 660  * @goal: preferred starting address of the region
 661  *
 662  * The goal is dropped if it can not be satisfied and the allocation will
 663  * fall back to memory below @goal.
 664  *
 665  * Allocation may happen on any node in the system.
 666  *
 667  * Returns NULL on failure.
 668  */
 669 void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align,
 670                                         unsigned long goal)
 671 {
 672         unsigned long limit = 0;
 673
 674         return ___alloc_bootmem_nopanic(size, align, goal, limit);
 675 }
 676
 677 static void * __init ___alloc_bootmem(unsigned long size, unsigned long align,
 678                                         unsigned long goal, unsigned long limit)
 679 {
 680         void *mem = ___alloc_bootmem_nopanic(size, align, goal, limit);
 681
 682         if (mem)
 683                 return mem;
 684         /*
 685          * Whoops, we cannot satisfy the allocation request.
 686          */
 687         printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size);
 688         panic("Out of memory");
 689         return NULL;
 690 }
 691
 692 /**
 693  * __alloc_bootmem - allocate boot memory
 694  * @size: size of the request in bytes
 695  * @align: alignment of the region
 696  * @goal: preferred starting address of the region
 697  *
 698  * The goal is dropped if it can not be satisfied and the allocation will
 699  * fall back to memory below @goal.
 700  *
 701  * Allocation may happen on any node in the system.
 702  *
 703  * The function panics if the request can not be satisfied.
 704  */
 705 void * __init __alloc_bootmem(unsigned long size, unsigned long align,
 706                               unsigned long goal)
 707 {
 708         unsigned long limit = 0;
 709
 710         return ___alloc_bootmem(size, align, goal, limit);
 711 }
 712
 713 void * __init ___alloc_bootmem_node_nopanic(pg_data_t *pgdat,
 714                                 unsigned long size, unsigned long align,
 715                                 unsigned long goal, unsigned long limit)
 716 {
 717         void *ptr;
 718
 719 again:
 720         ptr = alloc_arch_preferred_bootmem(pgdat->bdata, size,
 721                                            align, goal, limit);
 722         if (ptr)
 723                 return ptr;
 724
 725         /* do not panic in alloc_bootmem_bdata() */
 726         if (limit && goal + size > limit)
 727                 limit = 0;
 728
 729         ptr = alloc_bootmem_bdata(pgdat->bdata, size, align, goal, limit);
 730         if (ptr)
 731                 return ptr;
 732
 733         ptr = alloc_bootmem_core(size, align, goal, limit);
 734         if (ptr)
 735                 return ptr;
 736
 737         if (goal) {
 738                 goal = 0;
 739                 goto again;
 740         }
 741
 742         return NULL;
 743 }
 744
 745 void * __init __alloc_bootmem_node_nopanic(pg_data_t *pgdat, unsigned long size,
 746                                    unsigned long align, unsigned long goal)
 747 {
 748         if (WARN_ON_ONCE(slab_is_available()))
 749                 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
 750
 751         return ___alloc_bootmem_node_nopanic(pgdat, size, align, goal, 0);
 752 }
 753
 754 void * __init ___alloc_bootmem_node(pg_data_t *pgdat, unsigned long size,
 755                                     unsigned long align, unsigned long goal,
 756                                     unsigned long limit)
 757 {
 758         void *ptr;
 759
 760         ptr = ___alloc_bootmem_node_nopanic(pgdat, size, align, goal, 0);
 761         if (ptr)
 762                 return ptr;
 763
 764         printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size);
 765         panic("Out of memory");
 766         return NULL;
 767 }
 768
 769 /**
 770  * __alloc_bootmem_node - allocate boot memory from a specific node
 771  * @pgdat: node to allocate from
 772  * @size: size of the request in bytes
 773  * @align: alignment of the region
 774  * @goal: preferred starting address of the region
 775  *
 776  * The goal is dropped if it can not be satisfied and the allocation will
 777  * fall back to memory below @goal.
 778  *
 779  * Allocation may fall back to any node in the system if the specified node
 780  * can not hold the requested memory.
 781  *
 782  * The function panics if the request can not be satisfied.
 783  */
 784 void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size,
 785                                    unsigned long align, unsigned long goal)
 786 {
 787         if (WARN_ON_ONCE(slab_is_available()))
 788                 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
 789
 790         return  ___alloc_bootmem_node(pgdat, size, align, goal, 0);
 791 }
 792
 793 void * __init __alloc_bootmem_node_high(pg_data_t *pgdat, unsigned long size,
 794                                    unsigned long align, unsigned long goal)
 795 {
 796 #ifdef MAX_DMA32_PFN
 797         unsigned long end_pfn;
 798
 799         if (WARN_ON_ONCE(slab_is_available()))
 800                 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
 801
 802         /* update goal according ...MAX_DMA32_PFN */
 803         end_pfn = pgdat->node_start_pfn + pgdat->node_spanned_pages;
 804
 805         if (end_pfn > MAX_DMA32_PFN + (128 >> (20 - PAGE_SHIFT)) &&
 806             (goal >> PAGE_SHIFT) < MAX_DMA32_PFN) {
 807                 void *ptr;
 808                 unsigned long new_goal;
 809
 810                 new_goal = MAX_DMA32_PFN << PAGE_SHIFT;
 811                 ptr = alloc_bootmem_bdata(pgdat->bdata, size, align,
 812                                                  new_goal, 0);
 813                 if (ptr)
 814                         return ptr;
 815         }
 816 #endif
 817
 818         return __alloc_bootmem_node(pgdat, size, align, goal);
 819
 820 }
 821
 822 #ifndef ARCH_LOW_ADDRESS_LIMIT
 823 #define ARCH_LOW_ADDRESS_LIMIT  0xffffffffUL
 824 #endif
 825
 826 /**
 827  * __alloc_bootmem_low - allocate low boot memory
 828  * @size: size of the request in bytes
 829  * @align: alignment of the region
 830  * @goal: preferred starting address of the region
 831  *
 832  * The goal is dropped if it can not be satisfied and the allocation will
 833  * fall back to memory below @goal.
 834  *
 835  * Allocation may happen on any node in the system.
 836  *
 837  * The function panics if the request can not be satisfied.
 838  */
 839 void * __init __alloc_bootmem_low(unsigned long size, unsigned long align,
 840                                   unsigned long goal)
 841 {
 842         return ___alloc_bootmem(size, align, goal, ARCH_LOW_ADDRESS_LIMIT);
 843 }
 844
 845 /**
 846  * __alloc_bootmem_low_node - allocate low boot memory from a specific node
 847  * @pgdat: node to allocate from
 848  * @size: size of the request in bytes
 849  * @align: alignment of the region
 850  * @goal: preferred starting address of the region
 851  *
 852  * The goal is dropped if it can not be satisfied and the allocation will
 853  * fall back to memory below @goal.
 854  *
 855  * Allocation may fall back to any node in the system if the specified node
 856  * can not hold the requested memory.
 857  *
 858  * The function panics if the request can not be satisfied.
 859  */
 860 void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size,
 861                                        unsigned long align, unsigned long goal)
 862 {
 863         if (WARN_ON_ONCE(slab_is_available()))
 864                 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
 865
 866         return ___alloc_bootmem_node(pgdat, size, align,
 867                                      goal, ARCH_LOW_ADDRESS_LIMIT);
 868 }