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 #include <linux/bug.h>
  20 #include <linux/io.h>
  21
  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 unsigned long long max_possible_pfn;
  37
  38 bootmem_data_t bootmem_node_data[MAX_NUMNODES] __initdata;
  39
  40 static struct list_head bdata_list __initdata = LIST_HEAD_INIT(bdata_list);
  41
  42 static int bootmem_debug;
  43
  44 static int __init bootmem_debug_setup(char *buf)
  45 {
  46         bootmem_debug = 1;
  47         return 0;
  48 }
  49 early_param("bootmem_debug", bootmem_debug_setup);
  50
  51 #define bdebug(fmt, args...) ({                         \
  52         if (unlikely(bootmem_debug))                    \
  53                 printk(KERN_INFO                        \
  54                         "bootmem::%s " fmt,             \
  55                         __func__, ## args);             \
  56 })
  57
  58 static unsigned long __init bootmap_bytes(unsigned long pages)
  59 {
  60         unsigned long bytes = DIV_ROUND_UP(pages, 8);
  61
  62         return ALIGN(bytes, sizeof(long));
  63 }
  64
  65 /**
  66  * bootmem_bootmap_pages - calculate bitmap size in pages
  67  * @pages: number of pages the bitmap has to represent
  68  */
  69 unsigned long __init bootmem_bootmap_pages(unsigned long pages)
  70 {
  71         unsigned long bytes = bootmap_bytes(pages);
  72
  73         return PAGE_ALIGN(bytes) >> PAGE_SHIFT;
  74 }
  75
  76 /*
  77  * link bdata in order
  78  */
  79 static void __init link_bootmem(bootmem_data_t *bdata)
  80 {
  81         bootmem_data_t *ent;
  82
  83         list_for_each_entry(ent, &bdata_list, list) {
  84                 if (bdata->node_min_pfn < ent->node_min_pfn) {
  85                         list_add_tail(&bdata->list, &ent->list);
  86                         return;
  87                 }
  88         }
  89
  90         list_add_tail(&bdata->list, &bdata_list);
  91 }
  92
  93 /*
  94  * Called once to set up the allocator itself.
  95  */
  96 static unsigned long __init init_bootmem_core(bootmem_data_t *bdata,
  97         unsigned long mapstart, unsigned long start, unsigned long end)
  98 {
  99         unsigned long mapsize;
 100
 101         mminit_validate_memmodel_limits(&start, &end);
 102         bdata->node_bootmem_map = phys_to_virt(PFN_PHYS(mapstart));
 103         bdata->node_min_pfn = start;
 104         bdata->node_low_pfn = end;
 105         link_bootmem(bdata);
 106
 107         /*
 108          * Initially all pages are reserved - setup_arch() has to
 109          * register free RAM areas explicitly.
 110          */
 111         mapsize = bootmap_bytes(end - start);
 112         memset(bdata->node_bootmem_map, 0xff, mapsize);
 113
 114         bdebug("nid=%td start=%lx map=%lx end=%lx mapsize=%lx\n",
 115                 bdata - bootmem_node_data, start, mapstart, end, mapsize);
 116
 117         return mapsize;
 118 }
 119
 120 /**
 121  * init_bootmem_node - register a node as boot memory
 122  * @pgdat: node to register
 123  * @freepfn: pfn where the bitmap for this node is to be placed
 124  * @startpfn: first pfn on the node
 125  * @endpfn: first pfn after the node
 126  *
 127  * Returns the number of bytes needed to hold the bitmap for this node.
 128  */
 129 unsigned long __init init_bootmem_node(pg_data_t *pgdat, unsigned long freepfn,
 130                                 unsigned long startpfn, unsigned long endpfn)
 131 {
 132         return init_bootmem_core(pgdat->bdata, freepfn, startpfn, endpfn);
 133 }
 134
 135 /**
 136  * init_bootmem - register boot memory
 137  * @start: pfn where the bitmap is to be placed
 138  * @pages: number of available physical pages
 139  *
 140  * Returns the number of bytes needed to hold the bitmap.
 141  */
 142 unsigned long __init init_bootmem(unsigned long start, unsigned long pages)
 143 {
 144         max_low_pfn = pages;
 145         min_low_pfn = start;
 146         return init_bootmem_core(NODE_DATA(0)->bdata, start, 0, pages);
 147 }
 148
 149 /*
 150  * free_bootmem_late - free bootmem pages directly to page allocator
 151  * @addr: starting physical address of the range
 152  * @size: size of the range in bytes
 153  *
 154  * This is only useful when the bootmem allocator has already been torn
 155  * down, but we are still initializing the system.  Pages are given directly
 156  * to the page allocator, no bootmem metadata is updated because it is gone.
 157  */
 158 void __init free_bootmem_late(unsigned long physaddr, unsigned long size)
 159 {
 160         unsigned long cursor, end;
 161
 162         kmemleak_free_part(__va(physaddr), size);
 163
 164         cursor = PFN_UP(physaddr);
 165         end = PFN_DOWN(physaddr + size);
 166
 167         for (; cursor < end; cursor++) {
 168                 __free_pages_bootmem(pfn_to_page(cursor), cursor, 0);
 169                 totalram_pages++;
 170         }
 171 }
 172
 173 static unsigned long __init free_all_bootmem_core(bootmem_data_t *bdata)
 174 {
 175         struct page *page;
 176         unsigned long *map, start, end, pages, cur, count = 0;
 177
 178         if (!bdata->node_bootmem_map)
 179                 return 0;
 180
 181         map = bdata->node_bootmem_map;
 182         start = bdata->node_min_pfn;
 183         end = bdata->node_low_pfn;
 184
 185         bdebug("nid=%td start=%lx end=%lx\n",
 186                 bdata - bootmem_node_data, start, end);
 187
 188         while (start < end) {
 189                 unsigned long idx, vec;
 190                 unsigned shift;
 191
 192                 idx = start - bdata->node_min_pfn;
 193                 shift = idx & (BITS_PER_LONG - 1);
 194                 /*
 195                  * vec holds at most BITS_PER_LONG map bits,
 196                  * bit 0 corresponds to start.
 197                  */
 198                 vec = ~map[idx / BITS_PER_LONG];
 199
 200                 if (shift) {
 201                         vec >>= shift;
 202                         if (end - start >= BITS_PER_LONG)
 203                                 vec |= ~map[idx / BITS_PER_LONG + 1] <<
 204                                         (BITS_PER_LONG - shift);
 205                 }
 206                 /*
 207                  * If we have a properly aligned and fully unreserved
 208                  * BITS_PER_LONG block of pages in front of us, free
 209                  * it in one go.
 210                  */
 211                 if (IS_ALIGNED(start, BITS_PER_LONG) && vec == ~0UL) {
 212                         int order = ilog2(BITS_PER_LONG);
 213
 214                         __free_pages_bootmem(pfn_to_page(start), start, order);
 215                         count += BITS_PER_LONG;
 216                         start += BITS_PER_LONG;
 217                 } else {
 218                         cur = start;
 219
 220                         start = ALIGN(start + 1, BITS_PER_LONG);
 221                         while (vec && cur != start) {
 222                                 if (vec & 1) {
 223                                         page = pfn_to_page(cur);
 224                                         __free_pages_bootmem(page, cur, 0);
 225                                         count++;
 226                                 }
 227                                 vec >>= 1;
 228                                 ++cur;
 229                         }
 230                 }
 231         }
 232
 233         cur = bdata->node_min_pfn;
 234         page = virt_to_page(bdata->node_bootmem_map);
 235         pages = bdata->node_low_pfn - bdata->node_min_pfn;
 236         pages = bootmem_bootmap_pages(pages);
 237         count += pages;
 238         while (pages--)
 239                 __free_pages_bootmem(page++, cur++, 0);
 240         bdata->node_bootmem_map = NULL;
 241
 242         bdebug("nid=%td released=%lx\n", bdata - bootmem_node_data, count);
 243
 244         return count;
 245 }
 246
 247 static int reset_managed_pages_done __initdata;
 248
 249 void reset_node_managed_pages(pg_data_t *pgdat)
 250 {
 251         struct zone *z;
 252
 253         for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
 254                 z->managed_pages = 0;
 255 }
 256
 257 void __init reset_all_zones_managed_pages(void)
 258 {
 259         struct pglist_data *pgdat;
 260
 261         if (reset_managed_pages_done)
 262                 return;
 263
 264         for_each_online_pgdat(pgdat)
 265                 reset_node_managed_pages(pgdat);
 266
 267         reset_managed_pages_done = 1;
 268 }
 269
 270 /**
 271  * free_all_bootmem - release free pages to the buddy allocator
 272  *
 273  * Returns the number of pages actually released.
 274  */
 275 unsigned long __init free_all_bootmem(void)
 276 {
 277         unsigned long total_pages = 0;
 278         bootmem_data_t *bdata;
 279
 280         reset_all_zones_managed_pages();
 281
 282         list_for_each_entry(bdata, &bdata_list, list)
 283                 total_pages += free_all_bootmem_core(bdata);
 284
 285         totalram_pages += total_pages;
 286
 287         return total_pages;
 288 }
 289
 290 static void __init __free(bootmem_data_t *bdata,
 291                         unsigned long sidx, unsigned long eidx)
 292 {
 293         unsigned long idx;
 294
 295         bdebug("nid=%td start=%lx end=%lx\n", bdata - bootmem_node_data,
 296                 sidx + bdata->node_min_pfn,
 297                 eidx + bdata->node_min_pfn);
 298
 299         if (WARN_ON(bdata->node_bootmem_map == NULL))
 300                 return;
 301
 302         if (bdata->hint_idx > sidx)
 303                 bdata->hint_idx = sidx;
 304
 305         for (idx = sidx; idx < eidx; idx++)
 306                 if (!test_and_clear_bit(idx, bdata->node_bootmem_map))
 307                         BUG();
 308 }
 309
 310 static int __init __reserve(bootmem_data_t *bdata, unsigned long sidx,
 311                         unsigned long eidx, int flags)
 312 {
 313         unsigned long idx;
 314         int exclusive = flags & BOOTMEM_EXCLUSIVE;
 315
 316         bdebug("nid=%td start=%lx end=%lx flags=%x\n",
 317                 bdata - bootmem_node_data,
 318                 sidx + bdata->node_min_pfn,
 319                 eidx + bdata->node_min_pfn,
 320                 flags);
 321
 322         if (WARN_ON(bdata->node_bootmem_map == NULL))
 323                 return 0;
 324
 325         for (idx = sidx; idx < eidx; idx++)
 326                 if (test_and_set_bit(idx, bdata->node_bootmem_map)) {
 327                         if (exclusive) {
 328                                 __free(bdata, sidx, idx);
 329                                 return -EBUSY;
 330                         }
 331                         bdebug("silent double reserve of PFN %lx\n",
 332                                 idx + bdata->node_min_pfn);
 333                 }
 334         return 0;
 335 }
 336
 337 static int __init mark_bootmem_node(bootmem_data_t *bdata,
 338                                 unsigned long start, unsigned long end,
 339                                 int reserve, int flags)
 340 {
 341         unsigned long sidx, eidx;
 342
 343         bdebug("nid=%td start=%lx end=%lx reserve=%d flags=%x\n",
 344                 bdata - bootmem_node_data, start, end, reserve, flags);
 345
 346         BUG_ON(start < bdata->node_min_pfn);
 347         BUG_ON(end > bdata->node_low_pfn);
 348
 349         sidx = start - bdata->node_min_pfn;
 350         eidx = end - bdata->node_min_pfn;
 351
 352         if (reserve)
 353                 return __reserve(bdata, sidx, eidx, flags);
 354         else
 355                 __free(bdata, sidx, eidx);
 356         return 0;
 357 }
 358
 359 static int __init mark_bootmem(unsigned long start, unsigned long end,
 360                                 int reserve, int flags)
 361 {
 362         unsigned long pos;
 363         bootmem_data_t *bdata;
 364
 365         pos = start;
 366         list_for_each_entry(bdata, &bdata_list, list) {
 367                 int err;
 368                 unsigned long max;
 369
 370                 if (pos < bdata->node_min_pfn ||
 371                     pos >= bdata->node_low_pfn) {
 372                         BUG_ON(pos != start);
 373                         continue;
 374                 }
 375
 376                 max = min(bdata->node_low_pfn, end);
 377
 378                 err = mark_bootmem_node(bdata, pos, max, reserve, flags);
 379                 if (reserve && err) {
 380                         mark_bootmem(start, pos, 0, 0);
 381                         return err;
 382                 }
 383
 384                 if (max == end)
 385                         return 0;
 386                 pos = bdata->node_low_pfn;
 387         }
 388         BUG();
 389 }
 390
 391 /**
 392  * free_bootmem_node - mark a page range as usable
 393  * @pgdat: node the range resides on
 394  * @physaddr: starting address of the range
 395  * @size: size of the range in bytes
 396  *
 397  * Partial pages will be considered reserved and left as they are.
 398  *
 399  * The range must reside completely on the specified node.
 400  */
 401 void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
 402                               unsigned long size)
 403 {
 404         unsigned long start, end;
 405
 406         kmemleak_free_part(__va(physaddr), size);
 407
 408         start = PFN_UP(physaddr);
 409         end = PFN_DOWN(physaddr + size);
 410
 411         mark_bootmem_node(pgdat->bdata, start, end, 0, 0);
 412 }
 413
 414 /**
 415  * free_bootmem - mark a page range as usable
 416  * @addr: starting physical address of the range
 417  * @size: size of the range in bytes
 418  *
 419  * Partial pages will be considered reserved and left as they are.
 420  *
 421  * The range must be contiguous but may span node boundaries.
 422  */
 423 void __init free_bootmem(unsigned long physaddr, unsigned long size)
 424 {
 425         unsigned long start, end;
 426
 427         kmemleak_free_part(__va(physaddr), size);
 428
 429         start = PFN_UP(physaddr);
 430         end = PFN_DOWN(physaddr + size);
 431
 432         mark_bootmem(start, end, 0, 0);
 433 }
 434
 435 /**
 436  * reserve_bootmem_node - mark a page range as reserved
 437  * @pgdat: node the range resides on
 438  * @physaddr: starting address of the range
 439  * @size: size of the range in bytes
 440  * @flags: reservation flags (see linux/bootmem.h)
 441  *
 442  * Partial pages will be reserved.
 443  *
 444  * The range must reside completely on the specified node.
 445  */
 446 int __init reserve_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
 447                                  unsigned long size, int flags)
 448 {
 449         unsigned long start, end;
 450
 451         start = PFN_DOWN(physaddr);
 452         end = PFN_UP(physaddr + size);
 453
 454         return mark_bootmem_node(pgdat->bdata, start, end, 1, flags);
 455 }
 456
 457 /**
 458  * reserve_bootmem - mark a page range as reserved
 459  * @addr: starting address of the range
 460  * @size: size of the range in bytes
 461  * @flags: reservation flags (see linux/bootmem.h)
 462  *
 463  * Partial pages will be reserved.
 464  *
 465  * The range must be contiguous but may span node boundaries.
 466  */
 467 int __init reserve_bootmem(unsigned long addr, unsigned long size,
 468                             int flags)
 469 {
 470         unsigned long start, end;
 471
 472         start = PFN_DOWN(addr);
 473         end = PFN_UP(addr + size);
 474
 475         return mark_bootmem(start, end, 1, flags);
 476 }
 477
 478 static unsigned long __init align_idx(struct bootmem_data *bdata,
 479                                       unsigned long idx, unsigned long step)
 480 {
 481         unsigned long base = bdata->node_min_pfn;
 482
 483         /*
 484          * Align the index with respect to the node start so that the
 485          * combination of both satisfies the requested alignment.
 486          */
 487
 488         return ALIGN(base + idx, step) - base;
 489 }
 490
 491 static unsigned long __init align_off(struct bootmem_data *bdata,
 492                                       unsigned long off, unsigned long align)
 493 {
 494         unsigned long base = PFN_PHYS(bdata->node_min_pfn);
 495
 496         /* Same as align_idx for byte offsets */
 497
 498         return ALIGN(base + off, align) - base;
 499 }
 500
 501 static void * __init alloc_bootmem_bdata(struct bootmem_data *bdata,
 502                                         unsigned long size, unsigned long align,
 503                                         unsigned long goal, unsigned long limit)
 504 {
 505         unsigned long fallback = 0;
 506         unsigned long min, max, start, sidx, midx, step;
 507
 508         bdebug("nid=%td size=%lx [%lu pages] align=%lx goal=%lx limit=%lx\n",
 509                 bdata - bootmem_node_data, size, PAGE_ALIGN(size) >> PAGE_SHIFT,
 510                 align, goal, limit);
 511
 512         BUG_ON(!size);
 513         BUG_ON(align & (align - 1));
 514         BUG_ON(limit && goal + size > limit);
 515
 516         if (!bdata->node_bootmem_map)
 517                 return NULL;
 518
 519         min = bdata->node_min_pfn;
 520         max = bdata->node_low_pfn;
 521
 522         goal >>= PAGE_SHIFT;
 523         limit >>= PAGE_SHIFT;
 524
 525         if (limit && max > limit)
 526                 max = limit;
 527         if (max <= min)
 528                 return NULL;
 529
 530         step = max(align >> PAGE_SHIFT, 1UL);
 531
 532         if (goal && min < goal && goal < max)
 533                 start = ALIGN(goal, step);
 534         else
 535                 start = ALIGN(min, step);
 536
 537         sidx = start - bdata->node_min_pfn;
 538         midx = max - bdata->node_min_pfn;
 539
 540         if (bdata->hint_idx > sidx) {
 541                 /*
 542                  * Handle the valid case of sidx being zero and still
 543                  * catch the fallback below.
 544                  */
 545                 fallback = sidx + 1;
 546                 sidx = align_idx(bdata, bdata->hint_idx, step);
 547         }
 548
 549         while (1) {
 550                 int merge;
 551                 void *region;
 552                 unsigned long eidx, i, start_off, end_off;
 553 find_block:
 554                 sidx = find_next_zero_bit(bdata->node_bootmem_map, midx, sidx);
 555                 sidx = align_idx(bdata, sidx, step);
 556                 eidx = sidx + PFN_UP(size);
 557
 558                 if (sidx >= midx || eidx > midx)
 559                         break;
 560
 561                 for (i = sidx; i < eidx; i++)
 562                         if (test_bit(i, bdata->node_bootmem_map)) {
 563                                 sidx = align_idx(bdata, i, step);
 564                                 if (sidx == i)
 565                                         sidx += step;
 566                                 goto find_block;
 567                         }
 568
 569                 if (bdata->last_end_off & (PAGE_SIZE - 1) &&
 570                                 PFN_DOWN(bdata->last_end_off) + 1 == sidx)
 571                         start_off = align_off(bdata, bdata->last_end_off, align);
 572                 else
 573                         start_off = PFN_PHYS(sidx);
 574
 575                 merge = PFN_DOWN(start_off) < sidx;
 576                 end_off = start_off + size;
 577
 578                 bdata->last_end_off = end_off;
 579                 bdata->hint_idx = PFN_UP(end_off);
 580
 581                 /*
 582                  * Reserve the area now:
 583                  */
 584                 if (__reserve(bdata, PFN_DOWN(start_off) + merge,
 585                                 PFN_UP(end_off), BOOTMEM_EXCLUSIVE))
 586                         BUG();
 587
 588                 region = phys_to_virt(PFN_PHYS(bdata->node_min_pfn) +
 589                                 start_off);
 590                 memset(region, 0, size);
 591                 /*
 592                  * The min_count is set to 0 so that bootmem allocated blocks
 593                  * are never reported as leaks.
 594                  */
 595                 kmemleak_alloc(region, size, 0, 0);
 596                 return region;
 597         }
 598
 599         if (fallback) {
 600                 sidx = align_idx(bdata, fallback - 1, step);
 601                 fallback = 0;
 602                 goto find_block;
 603         }
 604
 605         return NULL;
 606 }
 607
 608 static void * __init alloc_bootmem_core(unsigned long size,
 609                                         unsigned long align,
 610                                         unsigned long goal,
 611                                         unsigned long limit)
 612 {
 613         bootmem_data_t *bdata;
 614         void *region;
 615
 616         if (WARN_ON_ONCE(slab_is_available()))
 617                 return kzalloc(size, GFP_NOWAIT);
 618
 619         list_for_each_entry(bdata, &bdata_list, list) {
 620                 if (goal && bdata->node_low_pfn <= PFN_DOWN(goal))
 621                         continue;
 622                 if (limit && bdata->node_min_pfn >= PFN_DOWN(limit))
 623                         break;
 624
 625                 region = alloc_bootmem_bdata(bdata, size, align, goal, limit);
 626                 if (region)
 627                         return region;
 628         }
 629
 630         return NULL;
 631 }
 632
 633 static void * __init ___alloc_bootmem_nopanic(unsigned long size,
 634                                               unsigned long align,
 635                                               unsigned long goal,
 636                                               unsigned long limit)
 637 {
 638         void *ptr;
 639
 640 restart:
 641         ptr = alloc_bootmem_core(size, align, goal, limit);
 642         if (ptr)
 643                 return ptr;
 644         if (goal) {
 645                 goal = 0;
 646                 goto restart;
 647         }
 648
 649         return NULL;
 650 }
 651
 652 /**
 653  * __alloc_bootmem_nopanic - allocate boot memory without panicking
 654  * @size: size of the request in bytes
 655  * @align: alignment of the region
 656  * @goal: preferred starting address of the region
 657  *
 658  * The goal is dropped if it can not be satisfied and the allocation will
 659  * fall back to memory below @goal.
 660  *
 661  * Allocation may happen on any node in the system.
 662  *
 663  * Returns NULL on failure.
 664  */
 665 void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align,
 666                                         unsigned long goal)
 667 {
 668         unsigned long limit = 0;
 669
 670         return ___alloc_bootmem_nopanic(size, align, goal, limit);
 671 }
 672
 673 static void * __init ___alloc_bootmem(unsigned long size, unsigned long align,
 674                                         unsigned long goal, unsigned long limit)
 675 {
 676         void *mem = ___alloc_bootmem_nopanic(size, align, goal, limit);
 677
 678         if (mem)
 679                 return mem;
 680         /*
 681          * Whoops, we cannot satisfy the allocation request.
 682          */
 683         printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size);
 684         panic("Out of memory");
 685         return NULL;
 686 }
 687
 688 /**
 689  * __alloc_bootmem - allocate boot memory
 690  * @size: size of the request in bytes
 691  * @align: alignment of the region
 692  * @goal: preferred starting address of the region
 693  *
 694  * The goal is dropped if it can not be satisfied and the allocation will
 695  * fall back to memory below @goal.
 696  *
 697  * Allocation may happen on any node in the system.
 698  *
 699  * The function panics if the request can not be satisfied.
 700  */
 701 void * __init __alloc_bootmem(unsigned long size, unsigned long align,
 702                               unsigned long goal)
 703 {
 704         unsigned long limit = 0;
 705
 706         return ___alloc_bootmem(size, align, goal, limit);
 707 }
 708
 709 void * __init ___alloc_bootmem_node_nopanic(pg_data_t *pgdat,
 710                                 unsigned long size, unsigned long align,
 711                                 unsigned long goal, unsigned long limit)
 712 {
 713         void *ptr;
 714
 715         if (WARN_ON_ONCE(slab_is_available()))
 716                 return kzalloc(size, GFP_NOWAIT);
 717 again:
 718
 719         /* do not panic in alloc_bootmem_bdata() */
 720         if (limit && goal + size > limit)
 721                 limit = 0;
 722
 723         ptr = alloc_bootmem_bdata(pgdat->bdata, size, align, goal, limit);
 724         if (ptr)
 725                 return ptr;
 726
 727         ptr = alloc_bootmem_core(size, align, goal, limit);
 728         if (ptr)
 729                 return ptr;
 730
 731         if (goal) {
 732                 goal = 0;
 733                 goto again;
 734         }
 735
 736         return NULL;
 737 }
 738
 739 void * __init __alloc_bootmem_node_nopanic(pg_data_t *pgdat, unsigned long size,
 740                                    unsigned long align, unsigned long goal)
 741 {
 742         if (WARN_ON_ONCE(slab_is_available()))
 743                 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
 744
 745         return ___alloc_bootmem_node_nopanic(pgdat, size, align, goal, 0);
 746 }
 747
 748 void * __init ___alloc_bootmem_node(pg_data_t *pgdat, unsigned long size,
 749                                     unsigned long align, unsigned long goal,
 750                                     unsigned long limit)
 751 {
 752         void *ptr;
 753
 754         ptr = ___alloc_bootmem_node_nopanic(pgdat, size, align, goal, 0);
 755         if (ptr)
 756                 return ptr;
 757
 758         printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size);
 759         panic("Out of memory");
 760         return NULL;
 761 }
 762
 763 /**
 764  * __alloc_bootmem_node - allocate boot memory from a specific node
 765  * @pgdat: node to allocate from
 766  * @size: size of the request in bytes
 767  * @align: alignment of the region
 768  * @goal: preferred starting address of the region
 769  *
 770  * The goal is dropped if it can not be satisfied and the allocation will
 771  * fall back to memory below @goal.
 772  *
 773  * Allocation may fall back to any node in the system if the specified node
 774  * can not hold the requested memory.
 775  *
 776  * The function panics if the request can not be satisfied.
 777  */
 778 void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size,
 779                                    unsigned long align, unsigned long goal)
 780 {
 781         if (WARN_ON_ONCE(slab_is_available()))
 782                 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
 783
 784         return  ___alloc_bootmem_node(pgdat, size, align, goal, 0);
 785 }
 786
 787 void * __init __alloc_bootmem_node_high(pg_data_t *pgdat, unsigned long size,
 788                                    unsigned long align, unsigned long goal)
 789 {
 790 #ifdef MAX_DMA32_PFN
 791         unsigned long end_pfn;
 792
 793         if (WARN_ON_ONCE(slab_is_available()))
 794                 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
 795
 796         /* update goal according ...MAX_DMA32_PFN */
 797         end_pfn = pgdat_end_pfn(pgdat);
 798
 799         if (end_pfn > MAX_DMA32_PFN + (128 >> (20 - PAGE_SHIFT)) &&
 800             (goal >> PAGE_SHIFT) < MAX_DMA32_PFN) {
 801                 void *ptr;
 802                 unsigned long new_goal;
 803
 804                 new_goal = MAX_DMA32_PFN << PAGE_SHIFT;
 805                 ptr = alloc_bootmem_bdata(pgdat->bdata, size, align,
 806                                                  new_goal, 0);
 807                 if (ptr)
 808                         return ptr;
 809         }
 810 #endif
 811
 812         return __alloc_bootmem_node(pgdat, size, align, goal);
 813
 814 }
 815
 816 #ifndef ARCH_LOW_ADDRESS_LIMIT
 817 #define ARCH_LOW_ADDRESS_LIMIT  0xffffffffUL
 818 #endif
 819
 820 /**
 821  * __alloc_bootmem_low - allocate low boot memory
 822  * @size: size of the request in bytes
 823  * @align: alignment of the region
 824  * @goal: preferred starting address of the region
 825  *
 826  * The goal is dropped if it can not be satisfied and the allocation will
 827  * fall back to memory below @goal.
 828  *
 829  * Allocation may happen on any node in the system.
 830  *
 831  * The function panics if the request can not be satisfied.
 832  */
 833 void * __init __alloc_bootmem_low(unsigned long size, unsigned long align,
 834                                   unsigned long goal)
 835 {
 836         return ___alloc_bootmem(size, align, goal, ARCH_LOW_ADDRESS_LIMIT);
 837 }
 838
 839 void * __init __alloc_bootmem_low_nopanic(unsigned long size,
 840                                           unsigned long align,
 841                                           unsigned long goal)
 842 {
 843         return ___alloc_bootmem_nopanic(size, align, goal,
 844                                         ARCH_LOW_ADDRESS_LIMIT);
 845 }
 846
 847 /**
 848  * __alloc_bootmem_low_node - allocate low boot memory from a specific node
 849  * @pgdat: node to allocate from
 850  * @size: size of the request in bytes
 851  * @align: alignment of the region
 852  * @goal: preferred starting address of the region
 853  *
 854  * The goal is dropped if it can not be satisfied and the allocation will
 855  * fall back to memory below @goal.
 856  *
 857  * Allocation may fall back to any node in the system if the specified node
 858  * can not hold the requested memory.
 859  *
 860  * The function panics if the request can not be satisfied.
 861  */
 862 void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size,
 863                                        unsigned long align, unsigned long goal)
 864 {
 865         if (WARN_ON_ONCE(slab_is_available()))
 866                 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
 867
 868         return ___alloc_bootmem_node(pgdat, size, align,
 869                                      goal, ARCH_LOW_ADDRESS_LIMIT);
 870 }