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/module.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 unsigned long max_low_pfn;
  27 unsigned long min_low_pfn;
  28 unsigned long max_pfn;
  29
  30 #ifdef CONFIG_CRASH_DUMP
  31 /*
  32  * If we have booted due to a crash, max_pfn will be a very low value. We need
  33  * to know the amount of memory that the previous kernel used.
  34  */
  35 unsigned long saved_max_pfn;
  36 #endif
  37
  38 #ifndef CONFIG_NO_BOOTMEM
  39 bootmem_data_t bootmem_node_data[MAX_NUMNODES] __initdata;
  40
  41 static struct list_head bdata_list __initdata = LIST_HEAD_INIT(bdata_list);
  42
  43 static int bootmem_debug;
  44
  45 static int __init bootmem_debug_setup(char *buf)
  46 {
  47         bootmem_debug = 1;
  48         return 0;
  49 }
  50 early_param("bootmem_debug", bootmem_debug_setup);
  51
  52 #define bdebug(fmt, args...) ({                         \
  53         if (unlikely(bootmem_debug))                    \
  54                 printk(KERN_INFO                        \
  55                         "bootmem::%s " fmt,             \
  56                         __func__, ## args);             \
  57 })
  58
  59 static unsigned long __init bootmap_bytes(unsigned long pages)
  60 {
  61         unsigned long bytes = (pages + 7) / 8;
  62
  63         return ALIGN(bytes, sizeof(long));
  64 }
  65
  66 /**
  67  * bootmem_bootmap_pages - calculate bitmap size in pages
  68  * @pages: number of pages the bitmap has to represent
  69  */
  70 unsigned long __init bootmem_bootmap_pages(unsigned long pages)
  71 {
  72         unsigned long bytes = bootmap_bytes(pages);
  73
  74         return PAGE_ALIGN(bytes) >> PAGE_SHIFT;
  75 }
  76
  77 /*
  78  * link bdata in order
  79  */
  80 static void __init link_bootmem(bootmem_data_t *bdata)
  81 {
  82         struct list_head *iter;
  83
  84         list_for_each(iter, &bdata_list) {
  85                 bootmem_data_t *ent;
  86
  87                 ent = list_entry(iter, bootmem_data_t, list);
  88                 if (bdata->node_min_pfn < ent->node_min_pfn)
  89                         break;
  90         }
  91         list_add_tail(&bdata->list, iter);
  92 }
  93
  94 /*
  95  * Called once to set up the allocator itself.
  96  */
  97 static unsigned long __init init_bootmem_core(bootmem_data_t *bdata,
  98         unsigned long mapstart, unsigned long start, unsigned long end)
  99 {
 100         unsigned long mapsize;
 101
 102         mminit_validate_memmodel_limits(&start, &end);
 103         bdata->node_bootmem_map = phys_to_virt(PFN_PHYS(mapstart));
 104         bdata->node_min_pfn = start;
 105         bdata->node_low_pfn = end;
 106         link_bootmem(bdata);
 107
 108         /*
 109          * Initially all pages are reserved - setup_arch() has to
 110          * register free RAM areas explicitly.
 111          */
 112         mapsize = bootmap_bytes(end - start);
 113         memset(bdata->node_bootmem_map, 0xff, mapsize);
 114
 115         bdebug("nid=%td start=%lx map=%lx end=%lx mapsize=%lx\n",
 116                 bdata - bootmem_node_data, start, mapstart, end, mapsize);
 117
 118         return mapsize;
 119 }
 120
 121 /**
 122  * init_bootmem_node - register a node as boot memory
 123  * @pgdat: node to register
 124  * @freepfn: pfn where the bitmap for this node is to be placed
 125  * @startpfn: first pfn on the node
 126  * @endpfn: first pfn after the node
 127  *
 128  * Returns the number of bytes needed to hold the bitmap for this node.
 129  */
 130 unsigned long __init init_bootmem_node(pg_data_t *pgdat, unsigned long freepfn,
 131                                 unsigned long startpfn, unsigned long endpfn)
 132 {
 133         return init_bootmem_core(pgdat->bdata, freepfn, startpfn, endpfn);
 134 }
 135
 136 /**
 137  * init_bootmem - register boot memory
 138  * @start: pfn where the bitmap is to be placed
 139  * @pages: number of available physical pages
 140  *
 141  * Returns the number of bytes needed to hold the bitmap.
 142  */
 143 unsigned long __init init_bootmem(unsigned long start, unsigned long pages)
 144 {
 145         max_low_pfn = pages;
 146         min_low_pfn = start;
 147         return init_bootmem_core(NODE_DATA(0)->bdata, start, 0, pages);
 148 }
 149 #endif
 150 /*
 151  * free_bootmem_late - free bootmem pages directly to page allocator
 152  * @addr: starting address of the range
 153  * @size: size of the range in bytes
 154  *
 155  * This is only useful when the bootmem allocator has already been torn
 156  * down, but we are still initializing the system.  Pages are given directly
 157  * to the page allocator, no bootmem metadata is updated because it is gone.
 158  */
 159 void __init free_bootmem_late(unsigned long addr, unsigned long size)
 160 {
 161         unsigned long cursor, end;
 162
 163         kmemleak_free_part(__va(addr), size);
 164
 165         cursor = PFN_UP(addr);
 166         end = PFN_DOWN(addr + size);
 167
 168         for (; cursor < end; cursor++) {
 169                 __free_pages_bootmem(pfn_to_page(cursor), 0);
 170                 totalram_pages++;
 171         }
 172 }
 173
 174 #ifdef CONFIG_NO_BOOTMEM
 175 static void __init __free_pages_memory(unsigned long start, unsigned long end)
 176 {
 177         int i;
 178         unsigned long start_aligned, end_aligned;
 179         int order = ilog2(BITS_PER_LONG);
 180
 181         start_aligned = (start + (BITS_PER_LONG - 1)) & ~(BITS_PER_LONG - 1);
 182         end_aligned = end & ~(BITS_PER_LONG - 1);
 183
 184         if (end_aligned <= start_aligned) {
 185                 for (i = start; i < end; i++)
 186                         __free_pages_bootmem(pfn_to_page(i), 0);
 187
 188                 return;
 189         }
 190
 191         for (i = start; i < start_aligned; i++)
 192                 __free_pages_bootmem(pfn_to_page(i), 0);
 193
 194         for (i = start_aligned; i < end_aligned; i += BITS_PER_LONG)
 195                 __free_pages_bootmem(pfn_to_page(i), order);
 196
 197         for (i = end_aligned; i < end; i++)
 198                 __free_pages_bootmem(pfn_to_page(i), 0);
 199 }
 200
 201 unsigned long __init free_all_memory_core_early(int nodeid)
 202 {
 203         int i;
 204         u64 start, end;
 205         unsigned long count = 0;
 206         struct range *range = NULL;
 207         int nr_range;
 208
 209         nr_range = get_free_all_memory_range(&range, nodeid);
 210
 211         for (i = 0; i < nr_range; i++) {
 212                 start = range[i].start;
 213                 end = range[i].end;
 214                 count += end - start;
 215                 __free_pages_memory(start, end);
 216         }
 217
 218         return count;
 219 }
 220 #else
 221 static unsigned long __init free_all_bootmem_core(bootmem_data_t *bdata)
 222 {
 223         int aligned;
 224         struct page *page;
 225         unsigned long start, end, pages, count = 0;
 226
 227         if (!bdata->node_bootmem_map)
 228                 return 0;
 229
 230         start = bdata->node_min_pfn;
 231         end = bdata->node_low_pfn;
 232
 233         /*
 234          * If the start is aligned to the machines wordsize, we might
 235          * be able to free pages in bulks of that order.
 236          */
 237         aligned = !(start & (BITS_PER_LONG - 1));
 238
 239         bdebug("nid=%td start=%lx end=%lx aligned=%d\n",
 240                 bdata - bootmem_node_data, start, end, aligned);
 241
 242         while (start < end) {
 243                 unsigned long *map, idx, vec;
 244
 245                 map = bdata->node_bootmem_map;
 246                 idx = start - bdata->node_min_pfn;
 247                 vec = ~map[idx / BITS_PER_LONG];
 248
 249                 if (aligned && vec == ~0UL && start + BITS_PER_LONG < end) {
 250                         int order = ilog2(BITS_PER_LONG);
 251
 252                         __free_pages_bootmem(pfn_to_page(start), order);
 253                         count += BITS_PER_LONG;
 254                 } else {
 255                         unsigned long off = 0;
 256
 257                         while (vec && off < BITS_PER_LONG) {
 258                                 if (vec & 1) {
 259                                         page = pfn_to_page(start + off);
 260                                         __free_pages_bootmem(page, 0);
 261                                         count++;
 262                                 }
 263                                 vec >>= 1;
 264                                 off++;
 265                         }
 266                 }
 267                 start += BITS_PER_LONG;
 268         }
 269
 270         page = virt_to_page(bdata->node_bootmem_map);
 271         pages = bdata->node_low_pfn - bdata->node_min_pfn;
 272         pages = bootmem_bootmap_pages(pages);
 273         count += pages;
 274         while (pages--)
 275                 __free_pages_bootmem(page++, 0);
 276
 277         bdebug("nid=%td released=%lx\n", bdata - bootmem_node_data, count);
 278
 279         return count;
 280 }
 281 #endif
 282
 283 /**
 284  * free_all_bootmem_node - release a node's free pages to the buddy allocator
 285  * @pgdat: node to be released
 286  *
 287  * Returns the number of pages actually released.
 288  */
 289 unsigned long __init free_all_bootmem_node(pg_data_t *pgdat)
 290 {
 291         register_page_bootmem_info_node(pgdat);
 292 #ifdef CONFIG_NO_BOOTMEM
 293         /* free_all_memory_core_early(MAX_NUMNODES) will be called later */
 294         return 0;
 295 #else
 296         return free_all_bootmem_core(pgdat->bdata);
 297 #endif
 298 }
 299
 300 /**
 301  * free_all_bootmem - release free pages to the buddy allocator
 302  *
 303  * Returns the number of pages actually released.
 304  */
 305 unsigned long __init free_all_bootmem(void)
 306 {
 307 #ifdef CONFIG_NO_BOOTMEM
 308         /*
 309          * We need to use MAX_NUMNODES instead of NODE_DATA(0)->node_id
 310          *  because in some case like Node0 doesnt have RAM installed
 311          *  low ram will be on Node1
 312          * Use MAX_NUMNODES will make sure all ranges in early_node_map[]
 313          *  will be used instead of only Node0 related
 314          */
 315         return free_all_memory_core_early(MAX_NUMNODES);
 316 #else
 317         unsigned long total_pages = 0;
 318         bootmem_data_t *bdata;
 319
 320         list_for_each_entry(bdata, &bdata_list, list)
 321                 total_pages += free_all_bootmem_core(bdata);
 322
 323         return total_pages;
 324 #endif
 325 }
 326
 327 #ifndef CONFIG_NO_BOOTMEM
 328 static void __init __free(bootmem_data_t *bdata,
 329                         unsigned long sidx, unsigned long eidx)
 330 {
 331         unsigned long idx;
 332
 333         bdebug("nid=%td start=%lx end=%lx\n", bdata - bootmem_node_data,
 334                 sidx + bdata->node_min_pfn,
 335                 eidx + bdata->node_min_pfn);
 336
 337         if (bdata->hint_idx > sidx)
 338                 bdata->hint_idx = sidx;
 339
 340         for (idx = sidx; idx < eidx; idx++)
 341                 if (!test_and_clear_bit(idx, bdata->node_bootmem_map))
 342                         BUG();
 343 }
 344
 345 static int __init __reserve(bootmem_data_t *bdata, unsigned long sidx,
 346                         unsigned long eidx, int flags)
 347 {
 348         unsigned long idx;
 349         int exclusive = flags & BOOTMEM_EXCLUSIVE;
 350
 351         bdebug("nid=%td start=%lx end=%lx flags=%x\n",
 352                 bdata - bootmem_node_data,
 353                 sidx + bdata->node_min_pfn,
 354                 eidx + bdata->node_min_pfn,
 355                 flags);
 356
 357         for (idx = sidx; idx < eidx; idx++)
 358                 if (test_and_set_bit(idx, bdata->node_bootmem_map)) {
 359                         if (exclusive) {
 360                                 __free(bdata, sidx, idx);
 361                                 return -EBUSY;
 362                         }
 363                         bdebug("silent double reserve of PFN %lx\n",
 364                                 idx + bdata->node_min_pfn);
 365                 }
 366         return 0;
 367 }
 368
 369 static int __init mark_bootmem_node(bootmem_data_t *bdata,
 370                                 unsigned long start, unsigned long end,
 371                                 int reserve, int flags)
 372 {
 373         unsigned long sidx, eidx;
 374
 375         bdebug("nid=%td start=%lx end=%lx reserve=%d flags=%x\n",
 376                 bdata - bootmem_node_data, start, end, reserve, flags);
 377
 378         BUG_ON(start < bdata->node_min_pfn);
 379         BUG_ON(end > bdata->node_low_pfn);
 380
 381         sidx = start - bdata->node_min_pfn;
 382         eidx = end - bdata->node_min_pfn;
 383
 384         if (reserve)
 385                 return __reserve(bdata, sidx, eidx, flags);
 386         else
 387                 __free(bdata, sidx, eidx);
 388         return 0;
 389 }
 390
 391 static int __init mark_bootmem(unsigned long start, unsigned long end,
 392                                 int reserve, int flags)
 393 {
 394         unsigned long pos;
 395         bootmem_data_t *bdata;
 396
 397         pos = start;
 398         list_for_each_entry(bdata, &bdata_list, list) {
 399                 int err;
 400                 unsigned long max;
 401
 402                 if (pos < bdata->node_min_pfn ||
 403                     pos >= bdata->node_low_pfn) {
 404                         BUG_ON(pos != start);
 405                         continue;
 406                 }
 407
 408                 max = min(bdata->node_low_pfn, end);
 409
 410                 err = mark_bootmem_node(bdata, pos, max, reserve, flags);
 411                 if (reserve && err) {
 412                         mark_bootmem(start, pos, 0, 0);
 413                         return err;
 414                 }
 415
 416                 if (max == end)
 417                         return 0;
 418                 pos = bdata->node_low_pfn;
 419         }
 420         BUG();
 421 }
 422 #endif
 423
 424 /**
 425  * free_bootmem_node - mark a page range as usable
 426  * @pgdat: node the range resides on
 427  * @physaddr: starting address of the range
 428  * @size: size of the range in bytes
 429  *
 430  * Partial pages will be considered reserved and left as they are.
 431  *
 432  * The range must reside completely on the specified node.
 433  */
 434 void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
 435                               unsigned long size)
 436 {
 437 #ifdef CONFIG_NO_BOOTMEM
 438         kmemleak_free_part(__va(physaddr), size);
 439         memblock_x86_free_range(physaddr, physaddr + size);
 440 #else
 441         unsigned long start, end;
 442
 443         kmemleak_free_part(__va(physaddr), size);
 444
 445         start = PFN_UP(physaddr);
 446         end = PFN_DOWN(physaddr + size);
 447
 448         mark_bootmem_node(pgdat->bdata, start, end, 0, 0);
 449 #endif
 450 }
 451
 452 /**
 453  * free_bootmem - mark a page range as usable
 454  * @addr: starting address of the range
 455  * @size: size of the range in bytes
 456  *
 457  * Partial pages will be considered reserved and left as they are.
 458  *
 459  * The range must be contiguous but may span node boundaries.
 460  */
 461 void __init free_bootmem(unsigned long addr, unsigned long size)
 462 {
 463 #ifdef CONFIG_NO_BOOTMEM
 464         kmemleak_free_part(__va(addr), size);
 465         memblock_x86_free_range(addr, addr + size);
 466 #else
 467         unsigned long start, end;
 468
 469         kmemleak_free_part(__va(addr), size);
 470
 471         start = PFN_UP(addr);
 472         end = PFN_DOWN(addr + size);
 473
 474         mark_bootmem(start, end, 0, 0);
 475 #endif
 476 }
 477
 478 /**
 479  * reserve_bootmem_node - mark a page range as reserved
 480  * @pgdat: node the range resides on
 481  * @physaddr: starting address of the range
 482  * @size: size of the range in bytes
 483  * @flags: reservation flags (see linux/bootmem.h)
 484  *
 485  * Partial pages will be reserved.
 486  *
 487  * The range must reside completely on the specified node.
 488  */
 489 int __init reserve_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
 490                                  unsigned long size, int flags)
 491 {
 492 #ifdef CONFIG_NO_BOOTMEM
 493         panic("no bootmem");
 494         return 0;
 495 #else
 496         unsigned long start, end;
 497
 498         start = PFN_DOWN(physaddr);
 499         end = PFN_UP(physaddr + size);
 500
 501         return mark_bootmem_node(pgdat->bdata, start, end, 1, flags);
 502 #endif
 503 }
 504
 505 /**
 506  * reserve_bootmem - mark a page range as usable
 507  * @addr: starting address of the range
 508  * @size: size of the range in bytes
 509  * @flags: reservation flags (see linux/bootmem.h)
 510  *
 511  * Partial pages will be reserved.
 512  *
 513  * The range must be contiguous but may span node boundaries.
 514  */
 515 int __init reserve_bootmem(unsigned long addr, unsigned long size,
 516                             int flags)
 517 {
 518 #ifdef CONFIG_NO_BOOTMEM
 519         panic("no bootmem");
 520         return 0;
 521 #else
 522         unsigned long start, end;
 523
 524         start = PFN_DOWN(addr);
 525         end = PFN_UP(addr + size);
 526
 527         return mark_bootmem(start, end, 1, flags);
 528 #endif
 529 }
 530
 531 #ifndef CONFIG_NO_BOOTMEM
 532 int __weak __init reserve_bootmem_generic(unsigned long phys, unsigned long len,
 533                                    int flags)
 534 {
 535         return reserve_bootmem(phys, len, flags);
 536 }
 537
 538 static unsigned long __init align_idx(struct bootmem_data *bdata,
 539                                       unsigned long idx, unsigned long step)
 540 {
 541         unsigned long base = bdata->node_min_pfn;
 542
 543         /*
 544          * Align the index with respect to the node start so that the
 545          * combination of both satisfies the requested alignment.
 546          */
 547
 548         return ALIGN(base + idx, step) - base;
 549 }
 550
 551 static unsigned long __init align_off(struct bootmem_data *bdata,
 552                                       unsigned long off, unsigned long align)
 553 {
 554         unsigned long base = PFN_PHYS(bdata->node_min_pfn);
 555
 556         /* Same as align_idx for byte offsets */
 557
 558         return ALIGN(base + off, align) - base;
 559 }
 560
 561 static void * __init alloc_bootmem_core(struct bootmem_data *bdata,
 562                                         unsigned long size, unsigned long align,
 563                                         unsigned long goal, unsigned long limit)
 564 {
 565         unsigned long fallback = 0;
 566         unsigned long min, max, start, sidx, midx, step;
 567
 568         bdebug("nid=%td size=%lx [%lu pages] align=%lx goal=%lx limit=%lx\n",
 569                 bdata - bootmem_node_data, size, PAGE_ALIGN(size) >> PAGE_SHIFT,
 570                 align, goal, limit);
 571
 572         BUG_ON(!size);
 573         BUG_ON(align & (align - 1));
 574         BUG_ON(limit && goal + size > limit);
 575
 576         if (!bdata->node_bootmem_map)
 577                 return NULL;
 578
 579         min = bdata->node_min_pfn;
 580         max = bdata->node_low_pfn;
 581
 582         goal >>= PAGE_SHIFT;
 583         limit >>= PAGE_SHIFT;
 584
 585         if (limit && max > limit)
 586                 max = limit;
 587         if (max <= min)
 588                 return NULL;
 589
 590         step = max(align >> PAGE_SHIFT, 1UL);
 591
 592         if (goal && min < goal && goal < max)
 593                 start = ALIGN(goal, step);
 594         else
 595                 start = ALIGN(min, step);
 596
 597         sidx = start - bdata->node_min_pfn;
 598         midx = max - bdata->node_min_pfn;
 599
 600         if (bdata->hint_idx > sidx) {
 601                 /*
 602                  * Handle the valid case of sidx being zero and still
 603                  * catch the fallback below.
 604                  */
 605                 fallback = sidx + 1;
 606                 sidx = align_idx(bdata, bdata->hint_idx, step);
 607         }
 608
 609         while (1) {
 610                 int merge;
 611                 void *region;
 612                 unsigned long eidx, i, start_off, end_off;
 613 find_block:
 614                 sidx = find_next_zero_bit(bdata->node_bootmem_map, midx, sidx);
 615                 sidx = align_idx(bdata, sidx, step);
 616                 eidx = sidx + PFN_UP(size);
 617
 618                 if (sidx >= midx || eidx > midx)
 619                         break;
 620
 621                 for (i = sidx; i < eidx; i++)
 622                         if (test_bit(i, bdata->node_bootmem_map)) {
 623                                 sidx = align_idx(bdata, i, step);
 624                                 if (sidx == i)
 625                                         sidx += step;
 626                                 goto find_block;
 627                         }
 628
 629                 if (bdata->last_end_off & (PAGE_SIZE - 1) &&
 630                                 PFN_DOWN(bdata->last_end_off) + 1 == sidx)
 631                         start_off = align_off(bdata, bdata->last_end_off, align);
 632                 else
 633                         start_off = PFN_PHYS(sidx);
 634
 635                 merge = PFN_DOWN(start_off) < sidx;
 636                 end_off = start_off + size;
 637
 638                 bdata->last_end_off = end_off;
 639                 bdata->hint_idx = PFN_UP(end_off);
 640
 641                 /*
 642                  * Reserve the area now:
 643                  */
 644                 if (__reserve(bdata, PFN_DOWN(start_off) + merge,
 645                                 PFN_UP(end_off), BOOTMEM_EXCLUSIVE))
 646                         BUG();
 647
 648                 region = phys_to_virt(PFN_PHYS(bdata->node_min_pfn) +
 649                                 start_off);
 650                 memset(region, 0, size);
 651                 /*
 652                  * The min_count is set to 0 so that bootmem allocated blocks
 653                  * are never reported as leaks.
 654                  */
 655                 kmemleak_alloc(region, size, 0, 0);
 656                 return region;
 657         }
 658
 659         if (fallback) {
 660                 sidx = align_idx(bdata, fallback - 1, step);
 661                 fallback = 0;
 662                 goto find_block;
 663         }
 664
 665         return NULL;
 666 }
 667
 668 static void * __init alloc_arch_preferred_bootmem(bootmem_data_t *bdata,
 669                                         unsigned long size, unsigned long align,
 670                                         unsigned long goal, unsigned long limit)
 671 {
 672         if (WARN_ON_ONCE(slab_is_available()))
 673                 return kzalloc(size, GFP_NOWAIT);
 674
 675 #ifdef CONFIG_HAVE_ARCH_BOOTMEM
 676         {
 677                 bootmem_data_t *p_bdata;
 678
 679                 p_bdata = bootmem_arch_preferred_node(bdata, size, align,
 680                                                         goal, limit);
 681                 if (p_bdata)
 682                         return alloc_bootmem_core(p_bdata, size, align,
 683                                                         goal, limit);
 684         }
 685 #endif
 686         return NULL;
 687 }
 688 #endif
 689
 690 static void * __init ___alloc_bootmem_nopanic(unsigned long size,
 691                                         unsigned long align,
 692                                         unsigned long goal,
 693                                         unsigned long limit)
 694 {
 695 #ifdef CONFIG_NO_BOOTMEM
 696         void *ptr;
 697
 698         if (WARN_ON_ONCE(slab_is_available()))
 699                 return kzalloc(size, GFP_NOWAIT);
 700
 701 restart:
 702
 703         ptr = __alloc_memory_core_early(MAX_NUMNODES, size, align, goal, limit);
 704
 705         if (ptr)
 706                 return ptr;
 707
 708         if (goal != 0) {
 709                 goal = 0;
 710                 goto restart;
 711         }
 712
 713         return NULL;
 714 #else
 715         bootmem_data_t *bdata;
 716         void *region;
 717
 718 restart:
 719         region = alloc_arch_preferred_bootmem(NULL, size, align, goal, limit);
 720         if (region)
 721                 return region;
 722
 723         list_for_each_entry(bdata, &bdata_list, list) {
 724                 if (goal && bdata->node_low_pfn <= PFN_DOWN(goal))
 725                         continue;
 726                 if (limit && bdata->node_min_pfn >= PFN_DOWN(limit))
 727                         break;
 728
 729                 region = alloc_bootmem_core(bdata, size, align, goal, limit);
 730                 if (region)
 731                         return region;
 732         }
 733
 734         if (goal) {
 735                 goal = 0;
 736                 goto restart;
 737         }
 738
 739         return NULL;
 740 #endif
 741 }
 742
 743 /**
 744  * __alloc_bootmem_nopanic - allocate boot memory without panicking
 745  * @size: size of the request in bytes
 746  * @align: alignment of the region
 747  * @goal: preferred starting address of the region
 748  *
 749  * The goal is dropped if it can not be satisfied and the allocation will
 750  * fall back to memory below @goal.
 751  *
 752  * Allocation may happen on any node in the system.
 753  *
 754  * Returns NULL on failure.
 755  */
 756 void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align,
 757                                         unsigned long goal)
 758 {
 759         unsigned long limit = 0;
 760
 761 #ifdef CONFIG_NO_BOOTMEM
 762         limit = -1UL;
 763 #endif
 764
 765         return ___alloc_bootmem_nopanic(size, align, goal, limit);
 766 }
 767
 768 static void * __init ___alloc_bootmem(unsigned long size, unsigned long align,
 769                                         unsigned long goal, unsigned long limit)
 770 {
 771         void *mem = ___alloc_bootmem_nopanic(size, align, goal, limit);
 772
 773         if (mem)
 774                 return mem;
 775         /*
 776          * Whoops, we cannot satisfy the allocation request.
 777          */
 778         printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size);
 779         panic("Out of memory");
 780         return NULL;
 781 }
 782
 783 /**
 784  * __alloc_bootmem - allocate boot memory
 785  * @size: size of the request in bytes
 786  * @align: alignment of the region
 787  * @goal: preferred starting address of the region
 788  *
 789  * The goal is dropped if it can not be satisfied and the allocation will
 790  * fall back to memory below @goal.
 791  *
 792  * Allocation may happen on any node in the system.
 793  *
 794  * The function panics if the request can not be satisfied.
 795  */
 796 void * __init __alloc_bootmem(unsigned long size, unsigned long align,
 797                               unsigned long goal)
 798 {
 799         unsigned long limit = 0;
 800
 801 #ifdef CONFIG_NO_BOOTMEM
 802         limit = -1UL;
 803 #endif
 804
 805         return ___alloc_bootmem(size, align, goal, limit);
 806 }
 807
 808 #ifndef CONFIG_NO_BOOTMEM
 809 static void * __init ___alloc_bootmem_node(bootmem_data_t *bdata,
 810                                 unsigned long size, unsigned long align,
 811                                 unsigned long goal, unsigned long limit)
 812 {
 813         void *ptr;
 814
 815         ptr = alloc_arch_preferred_bootmem(bdata, size, align, goal, limit);
 816         if (ptr)
 817                 return ptr;
 818
 819         ptr = alloc_bootmem_core(bdata, size, align, goal, limit);
 820         if (ptr)
 821                 return ptr;
 822
 823         return ___alloc_bootmem(size, align, goal, limit);
 824 }
 825 #endif
 826
 827 /**
 828  * __alloc_bootmem_node - allocate boot memory from a specific node
 829  * @pgdat: node to allocate from
 830  * @size: size of the request in bytes
 831  * @align: alignment of the region
 832  * @goal: preferred starting address of the region
 833  *
 834  * The goal is dropped if it can not be satisfied and the allocation will
 835  * fall back to memory below @goal.
 836  *
 837  * Allocation may fall back to any node in the system if the specified node
 838  * can not hold the requested memory.
 839  *
 840  * The function panics if the request can not be satisfied.
 841  */
 842 void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size,
 843                                    unsigned long align, unsigned long goal)
 844 {
 845         void *ptr;
 846
 847         if (WARN_ON_ONCE(slab_is_available()))
 848                 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
 849
 850 #ifdef CONFIG_NO_BOOTMEM
 851         ptr = __alloc_memory_core_early(pgdat->node_id, size, align,
 852                                          goal, -1ULL);
 853         if (ptr)
 854                 return ptr;
 855
 856         ptr = __alloc_memory_core_early(MAX_NUMNODES, size, align,
 857                                          goal, -1ULL);
 858 #else
 859         ptr = ___alloc_bootmem_node(pgdat->bdata, size, align, goal, 0);
 860 #endif
 861
 862         return ptr;
 863 }
 864
 865 void * __init __alloc_bootmem_node_high(pg_data_t *pgdat, unsigned long size,
 866                                    unsigned long align, unsigned long goal)
 867 {
 868 #ifdef MAX_DMA32_PFN
 869         unsigned long end_pfn;
 870
 871         if (WARN_ON_ONCE(slab_is_available()))
 872                 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
 873
 874         /* update goal according ...MAX_DMA32_PFN */
 875         end_pfn = pgdat->node_start_pfn + pgdat->node_spanned_pages;
 876
 877         if (end_pfn > MAX_DMA32_PFN + (128 >> (20 - PAGE_SHIFT)) &&
 878             (goal >> PAGE_SHIFT) < MAX_DMA32_PFN) {
 879                 void *ptr;
 880                 unsigned long new_goal;
 881
 882                 new_goal = MAX_DMA32_PFN << PAGE_SHIFT;
 883 #ifdef CONFIG_NO_BOOTMEM
 884                 ptr =  __alloc_memory_core_early(pgdat->node_id, size, align,
 885                                                  new_goal, -1ULL);
 886 #else
 887                 ptr = alloc_bootmem_core(pgdat->bdata, size, align,
 888                                                  new_goal, 0);
 889 #endif
 890                 if (ptr)
 891                         return ptr;
 892         }
 893 #endif
 894
 895         return __alloc_bootmem_node(pgdat, size, align, goal);
 896
 897 }
 898
 899 #ifdef CONFIG_SPARSEMEM
 900 /**
 901  * alloc_bootmem_section - allocate boot memory from a specific section
 902  * @size: size of the request in bytes
 903  * @section_nr: sparse map section to allocate from
 904  *
 905  * Return NULL on failure.
 906  */
 907 void * __init alloc_bootmem_section(unsigned long size,
 908                                     unsigned long section_nr)
 909 {
 910 #ifdef CONFIG_NO_BOOTMEM
 911         unsigned long pfn, goal, limit;
 912
 913         pfn = section_nr_to_pfn(section_nr);
 914         goal = pfn << PAGE_SHIFT;
 915         limit = section_nr_to_pfn(section_nr + 1) << PAGE_SHIFT;
 916
 917         return __alloc_memory_core_early(early_pfn_to_nid(pfn), size,
 918                                          SMP_CACHE_BYTES, goal, limit);
 919 #else
 920         bootmem_data_t *bdata;
 921         unsigned long pfn, goal, limit;
 922
 923         pfn = section_nr_to_pfn(section_nr);
 924         goal = pfn << PAGE_SHIFT;
 925         limit = section_nr_to_pfn(section_nr + 1) << PAGE_SHIFT;
 926         bdata = &bootmem_node_data[early_pfn_to_nid(pfn)];
 927
 928         return alloc_bootmem_core(bdata, size, SMP_CACHE_BYTES, goal, limit);
 929 #endif
 930 }
 931 #endif
 932
 933 void * __init __alloc_bootmem_node_nopanic(pg_data_t *pgdat, unsigned long size,
 934                                    unsigned long align, unsigned long goal)
 935 {
 936         void *ptr;
 937
 938         if (WARN_ON_ONCE(slab_is_available()))
 939                 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
 940
 941 #ifdef CONFIG_NO_BOOTMEM
 942         ptr =  __alloc_memory_core_early(pgdat->node_id, size, align,
 943                                                  goal, -1ULL);
 944 #else
 945         ptr = alloc_arch_preferred_bootmem(pgdat->bdata, size, align, goal, 0);
 946         if (ptr)
 947                 return ptr;
 948
 949         ptr = alloc_bootmem_core(pgdat->bdata, size, align, goal, 0);
 950 #endif
 951         if (ptr)
 952                 return ptr;
 953
 954         return __alloc_bootmem_nopanic(size, align, goal);
 955 }
 956
 957 #ifndef ARCH_LOW_ADDRESS_LIMIT
 958 #define ARCH_LOW_ADDRESS_LIMIT  0xffffffffUL
 959 #endif
 960
 961 /**
 962  * __alloc_bootmem_low - allocate low boot memory
 963  * @size: size of the request in bytes
 964  * @align: alignment of the region
 965  * @goal: preferred starting address of the region
 966  *
 967  * The goal is dropped if it can not be satisfied and the allocation will
 968  * fall back to memory below @goal.
 969  *
 970  * Allocation may happen on any node in the system.
 971  *
 972  * The function panics if the request can not be satisfied.
 973  */
 974 void * __init __alloc_bootmem_low(unsigned long size, unsigned long align,
 975                                   unsigned long goal)
 976 {
 977         return ___alloc_bootmem(size, align, goal, ARCH_LOW_ADDRESS_LIMIT);
 978 }
 979
 980 /**
 981  * __alloc_bootmem_low_node - allocate low boot memory from a specific node
 982  * @pgdat: node to allocate from
 983  * @size: size of the request in bytes
 984  * @align: alignment of the region
 985  * @goal: preferred starting address of the region
 986  *
 987  * The goal is dropped if it can not be satisfied and the allocation will
 988  * fall back to memory below @goal.
 989  *
 990  * Allocation may fall back to any node in the system if the specified node
 991  * can not hold the requested memory.
 992  *
 993  * The function panics if the request can not be satisfied.
 994  */
 995 void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size,
 996                                        unsigned long align, unsigned long goal)
 997 {
 998         void *ptr;
 999
1000         if (WARN_ON_ONCE(slab_is_available()))
1001                 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
1002
1003 #ifdef CONFIG_NO_BOOTMEM
1004         ptr = __alloc_memory_core_early(pgdat->node_id, size, align,
1005                                 goal, ARCH_LOW_ADDRESS_LIMIT);
1006         if (ptr)
1007                 return ptr;
1008         ptr = __alloc_memory_core_early(MAX_NUMNODES, size, align,
1009                                 goal, ARCH_LOW_ADDRESS_LIMIT);
1010 #else
1011         ptr = ___alloc_bootmem_node(pgdat->bdata, size, align,
1012                                 goal, ARCH_LOW_ADDRESS_LIMIT);
1013 #endif
1014         return ptr;
1015 }