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
  19 #include <asm/bug.h>
  20 #include <asm/io.h>
  21 #include <asm/processor.h>
  22
  23 #include "internal.h"
  24
  25 unsigned long max_low_pfn;
  26 unsigned long min_low_pfn;
  27 unsigned long max_pfn;
  28
  29 #ifdef CONFIG_CRASH_DUMP
  30 /*
  31  * If we have booted due to a crash, max_pfn will be a very low value. We need
  32  * to know the amount of memory that the previous kernel used.
  33  */
  34 unsigned long saved_max_pfn;
  35 #endif
  36
  37 #ifndef CONFIG_NO_BOOTMEM
  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 = (pages + 7) / 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         struct list_head *iter;
  82
  83         list_for_each(iter, &bdata_list) {
  84                 bootmem_data_t *ent;
  85
  86                 ent = list_entry(iter, bootmem_data_t, list);
  87                 if (bdata->node_min_pfn < ent->node_min_pfn)
  88                         break;
  89         }
  90         list_add_tail(&bdata->list, iter);
  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 #endif
 149 /*
 150  * free_bootmem_late - free bootmem pages directly to page allocator
 151  * @addr: starting 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 addr, unsigned long size)
 159 {
 160         unsigned long cursor, end;
 161
 162         kmemleak_free_part(__va(addr), size);
 163
 164         cursor = PFN_UP(addr);
 165         end = PFN_DOWN(addr + size);
 166
 167         for (; cursor < end; cursor++) {
 168                 __free_pages_bootmem(pfn_to_page(cursor), 0);
 169                 totalram_pages++;
 170         }
 171 }
 172
 173 #ifdef CONFIG_NO_BOOTMEM
 174 static void __init __free_pages_memory(unsigned long start, unsigned long end)
 175 {
 176         int i;
 177         unsigned long start_aligned, end_aligned;
 178         int order = ilog2(BITS_PER_LONG);
 179
 180         start_aligned = (start + (BITS_PER_LONG - 1)) & ~(BITS_PER_LONG - 1);
 181         end_aligned = end & ~(BITS_PER_LONG - 1);
 182
 183         if (end_aligned <= start_aligned) {
 184                 for (i = start; i < end; i++)
 185                         __free_pages_bootmem(pfn_to_page(i), 0);
 186
 187                 return;
 188         }
 189
 190         for (i = start; i < start_aligned; i++)
 191                 __free_pages_bootmem(pfn_to_page(i), 0);
 192
 193         for (i = start_aligned; i < end_aligned; i += BITS_PER_LONG)
 194                 __free_pages_bootmem(pfn_to_page(i), order);
 195
 196         for (i = end_aligned; i < end; i++)
 197                 __free_pages_bootmem(pfn_to_page(i), 0);
 198 }
 199
 200 unsigned long __init free_all_memory_core_early(int nodeid)
 201 {
 202         int i;
 203         u64 start, end;
 204         unsigned long count = 0;
 205         struct range *range = NULL;
 206         int nr_range;
 207
 208         nr_range = get_free_all_memory_range(&range, nodeid);
 209
 210         for (i = 0; i < nr_range; i++) {
 211                 start = range[i].start;
 212                 end = range[i].end;
 213                 count += end - start;
 214                 __free_pages_memory(start, end);
 215         }
 216
 217         return count;
 218 }
 219 #else
 220 static unsigned long __init free_all_bootmem_core(bootmem_data_t *bdata)
 221 {
 222         int aligned;
 223         struct page *page;
 224         unsigned long start, end, pages, count = 0;
 225
 226         if (!bdata->node_bootmem_map)
 227                 return 0;
 228
 229         start = bdata->node_min_pfn;
 230         end = bdata->node_low_pfn;
 231
 232         /*
 233          * If the start is aligned to the machines wordsize, we might
 234          * be able to free pages in bulks of that order.
 235          */
 236         aligned = !(start & (BITS_PER_LONG - 1));
 237
 238         bdebug("nid=%td start=%lx end=%lx aligned=%d\n",
 239                 bdata - bootmem_node_data, start, end, aligned);
 240
 241         while (start < end) {
 242                 unsigned long *map, idx, vec;
 243
 244                 map = bdata->node_bootmem_map;
 245                 idx = start - bdata->node_min_pfn;
 246                 vec = ~map[idx / BITS_PER_LONG];
 247
 248                 if (aligned && vec == ~0UL && start + BITS_PER_LONG < end) {
 249                         int order = ilog2(BITS_PER_LONG);
 250
 251                         __free_pages_bootmem(pfn_to_page(start), order);
 252                         count += BITS_PER_LONG;
 253                 } else {
 254                         unsigned long off = 0;
 255
 256                         while (vec && off < BITS_PER_LONG) {
 257                                 if (vec & 1) {
 258                                         page = pfn_to_page(start + off);
 259                                         __free_pages_bootmem(page, 0);
 260                                         count++;
 261                                 }
 262                                 vec >>= 1;
 263                                 off++;
 264                         }
 265                 }
 266                 start += BITS_PER_LONG;
 267         }
 268
 269         page = virt_to_page(bdata->node_bootmem_map);
 270         pages = bdata->node_low_pfn - bdata->node_min_pfn;
 271         pages = bootmem_bootmap_pages(pages);
 272         count += pages;
 273         while (pages--)
 274                 __free_pages_bootmem(page++, 0);
 275
 276         bdebug("nid=%td released=%lx\n", bdata - bootmem_node_data, count);
 277
 278         return count;
 279 }
 280 #endif
 281
 282 /**
 283  * free_all_bootmem_node - release a node's free pages to the buddy allocator
 284  * @pgdat: node to be released
 285  *
 286  * Returns the number of pages actually released.
 287  */
 288 unsigned long __init free_all_bootmem_node(pg_data_t *pgdat)
 289 {
 290         register_page_bootmem_info_node(pgdat);
 291 #ifdef CONFIG_NO_BOOTMEM
 292         /* free_all_memory_core_early(MAX_NUMNODES) will be called later */
 293         return 0;
 294 #else
 295         return free_all_bootmem_core(pgdat->bdata);
 296 #endif
 297 }
 298
 299 /**
 300  * free_all_bootmem - release free pages to the buddy allocator
 301  *
 302  * Returns the number of pages actually released.
 303  */
 304 unsigned long __init free_all_bootmem(void)
 305 {
 306 #ifdef CONFIG_NO_BOOTMEM
 307         /*
 308          * We need to use MAX_NUMNODES instead of NODE_DATA(0)->node_id
 309          *  because in some case like Node0 doesnt have RAM installed
 310          *  low ram will be on Node1
 311          * Use MAX_NUMNODES will make sure all ranges in early_node_map[]
 312          *  will be used instead of only Node0 related
 313          */
 314         return free_all_memory_core_early(MAX_NUMNODES);
 315 #else
 316         unsigned long total_pages = 0;
 317         bootmem_data_t *bdata;
 318
 319         list_for_each_entry(bdata, &bdata_list, list)
 320                 total_pages += free_all_bootmem_core(bdata);
 321
 322         return total_pages;
 323 #endif
 324 }
 325
 326 #ifndef CONFIG_NO_BOOTMEM
 327 static void __init __free(bootmem_data_t *bdata,
 328                         unsigned long sidx, unsigned long eidx)
 329 {
 330         unsigned long idx;
 331
 332         bdebug("nid=%td start=%lx end=%lx\n", bdata - bootmem_node_data,
 333                 sidx + bdata->node_min_pfn,
 334                 eidx + bdata->node_min_pfn);
 335
 336         if (bdata->hint_idx > sidx)
 337                 bdata->hint_idx = sidx;
 338
 339         for (idx = sidx; idx < eidx; idx++)
 340                 if (!test_and_clear_bit(idx, bdata->node_bootmem_map))
 341                         BUG();
 342 }
 343
 344 static int __init __reserve(bootmem_data_t *bdata, unsigned long sidx,
 345                         unsigned long eidx, int flags)
 346 {
 347         unsigned long idx;
 348         int exclusive = flags & BOOTMEM_EXCLUSIVE;
 349
 350         bdebug("nid=%td start=%lx end=%lx flags=%x\n",
 351                 bdata - bootmem_node_data,
 352                 sidx + bdata->node_min_pfn,
 353                 eidx + bdata->node_min_pfn,
 354                 flags);
 355
 356         for (idx = sidx; idx < eidx; idx++)
 357                 if (test_and_set_bit(idx, bdata->node_bootmem_map)) {
 358                         if (exclusive) {
 359                                 __free(bdata, sidx, idx);
 360                                 return -EBUSY;
 361                         }
 362                         bdebug("silent double reserve of PFN %lx\n",
 363                                 idx + bdata->node_min_pfn);
 364                 }
 365         return 0;
 366 }
 367
 368 static int __init mark_bootmem_node(bootmem_data_t *bdata,
 369                                 unsigned long start, unsigned long end,
 370                                 int reserve, int flags)
 371 {
 372         unsigned long sidx, eidx;
 373
 374         bdebug("nid=%td start=%lx end=%lx reserve=%d flags=%x\n",
 375                 bdata - bootmem_node_data, start, end, reserve, flags);
 376
 377         BUG_ON(start < bdata->node_min_pfn);
 378         BUG_ON(end > bdata->node_low_pfn);
 379
 380         sidx = start - bdata->node_min_pfn;
 381         eidx = end - bdata->node_min_pfn;
 382
 383         if (reserve)
 384                 return __reserve(bdata, sidx, eidx, flags);
 385         else
 386                 __free(bdata, sidx, eidx);
 387         return 0;
 388 }
 389
 390 static int __init mark_bootmem(unsigned long start, unsigned long end,
 391                                 int reserve, int flags)
 392 {
 393         unsigned long pos;
 394         bootmem_data_t *bdata;
 395
 396         pos = start;
 397         list_for_each_entry(bdata, &bdata_list, list) {
 398                 int err;
 399                 unsigned long max;
 400
 401                 if (pos < bdata->node_min_pfn ||
 402                     pos >= bdata->node_low_pfn) {
 403                         BUG_ON(pos != start);
 404                         continue;
 405                 }
 406
 407                 max = min(bdata->node_low_pfn, end);
 408
 409                 err = mark_bootmem_node(bdata, pos, max, reserve, flags);
 410                 if (reserve && err) {
 411                         mark_bootmem(start, pos, 0, 0);
 412                         return err;
 413                 }
 414
 415                 if (max == end)
 416                         return 0;
 417                 pos = bdata->node_low_pfn;
 418         }
 419         BUG();
 420 }
 421 #endif
 422
 423 /**
 424  * free_bootmem_node - mark a page range as usable
 425  * @pgdat: node the range resides on
 426  * @physaddr: starting address of the range
 427  * @size: size of the range in bytes
 428  *
 429  * Partial pages will be considered reserved and left as they are.
 430  *
 431  * The range must reside completely on the specified node.
 432  */
 433 void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
 434                               unsigned long size)
 435 {
 436 #ifdef CONFIG_NO_BOOTMEM
 437         free_early(physaddr, physaddr + size);
 438 #else
 439         unsigned long start, end;
 440
 441         kmemleak_free_part(__va(physaddr), size);
 442
 443         start = PFN_UP(physaddr);
 444         end = PFN_DOWN(physaddr + size);
 445
 446         mark_bootmem_node(pgdat->bdata, start, end, 0, 0);
 447 #endif
 448 }
 449
 450 /**
 451  * free_bootmem - mark a page range as usable
 452  * @addr: starting address of the range
 453  * @size: size of the range in bytes
 454  *
 455  * Partial pages will be considered reserved and left as they are.
 456  *
 457  * The range must be contiguous but may span node boundaries.
 458  */
 459 void __init free_bootmem(unsigned long addr, unsigned long size)
 460 {
 461 #ifdef CONFIG_NO_BOOTMEM
 462         free_early(addr, addr + size);
 463 #else
 464         unsigned long start, end;
 465
 466         kmemleak_free_part(__va(addr), size);
 467
 468         start = PFN_UP(addr);
 469         end = PFN_DOWN(addr + size);
 470
 471         mark_bootmem(start, end, 0, 0);
 472 #endif
 473 }
 474
 475 /**
 476  * reserve_bootmem_node - mark a page range as reserved
 477  * @pgdat: node the range resides on
 478  * @physaddr: starting address of the range
 479  * @size: size of the range in bytes
 480  * @flags: reservation flags (see linux/bootmem.h)
 481  *
 482  * Partial pages will be reserved.
 483  *
 484  * The range must reside completely on the specified node.
 485  */
 486 int __init reserve_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
 487                                  unsigned long size, int flags)
 488 {
 489 #ifdef CONFIG_NO_BOOTMEM
 490         panic("no bootmem");
 491         return 0;
 492 #else
 493         unsigned long start, end;
 494
 495         start = PFN_DOWN(physaddr);
 496         end = PFN_UP(physaddr + size);
 497
 498         return mark_bootmem_node(pgdat->bdata, start, end, 1, flags);
 499 #endif
 500 }
 501
 502 /**
 503  * reserve_bootmem - mark a page range as usable
 504  * @addr: starting address of the range
 505  * @size: size of the range in bytes
 506  * @flags: reservation flags (see linux/bootmem.h)
 507  *
 508  * Partial pages will be reserved.
 509  *
 510  * The range must be contiguous but may span node boundaries.
 511  */
 512 int __init reserve_bootmem(unsigned long addr, unsigned long size,
 513                             int flags)
 514 {
 515 #ifdef CONFIG_NO_BOOTMEM
 516         panic("no bootmem");
 517         return 0;
 518 #else
 519         unsigned long start, end;
 520
 521         start = PFN_DOWN(addr);
 522         end = PFN_UP(addr + size);
 523
 524         return mark_bootmem(start, end, 1, flags);
 525 #endif
 526 }
 527
 528 #ifndef CONFIG_NO_BOOTMEM
 529 static unsigned long __init align_idx(struct bootmem_data *bdata,
 530                                       unsigned long idx, unsigned long step)
 531 {
 532         unsigned long base = bdata->node_min_pfn;
 533
 534         /*
 535          * Align the index with respect to the node start so that the
 536          * combination of both satisfies the requested alignment.
 537          */
 538
 539         return ALIGN(base + idx, step) - base;
 540 }
 541
 542 static unsigned long __init align_off(struct bootmem_data *bdata,
 543                                       unsigned long off, unsigned long align)
 544 {
 545         unsigned long base = PFN_PHYS(bdata->node_min_pfn);
 546
 547         /* Same as align_idx for byte offsets */
 548
 549         return ALIGN(base + off, align) - base;
 550 }
 551
 552 static void * __init alloc_bootmem_core(struct bootmem_data *bdata,
 553                                         unsigned long size, unsigned long align,
 554                                         unsigned long goal, unsigned long limit)
 555 {
 556         unsigned long fallback = 0;
 557         unsigned long min, max, start, sidx, midx, step;
 558
 559         bdebug("nid=%td size=%lx [%lu pages] align=%lx goal=%lx limit=%lx\n",
 560                 bdata - bootmem_node_data, size, PAGE_ALIGN(size) >> PAGE_SHIFT,
 561                 align, goal, limit);
 562
 563         BUG_ON(!size);
 564         BUG_ON(align & (align - 1));
 565         BUG_ON(limit && goal + size > limit);
 566
 567         if (!bdata->node_bootmem_map)
 568                 return NULL;
 569
 570         min = bdata->node_min_pfn;
 571         max = bdata->node_low_pfn;
 572
 573         goal >>= PAGE_SHIFT;
 574         limit >>= PAGE_SHIFT;
 575
 576         if (limit && max > limit)
 577                 max = limit;
 578         if (max <= min)
 579                 return NULL;
 580
 581         step = max(align >> PAGE_SHIFT, 1UL);
 582
 583         if (goal && min < goal && goal < max)
 584                 start = ALIGN(goal, step);
 585         else
 586                 start = ALIGN(min, step);
 587
 588         sidx = start - bdata->node_min_pfn;
 589         midx = max - bdata->node_min_pfn;
 590
 591         if (bdata->hint_idx > sidx) {
 592                 /*
 593                  * Handle the valid case of sidx being zero and still
 594                  * catch the fallback below.
 595                  */
 596                 fallback = sidx + 1;
 597                 sidx = align_idx(bdata, bdata->hint_idx, step);
 598         }
 599
 600         while (1) {
 601                 int merge;
 602                 void *region;
 603                 unsigned long eidx, i, start_off, end_off;
 604 find_block:
 605                 sidx = find_next_zero_bit(bdata->node_bootmem_map, midx, sidx);
 606                 sidx = align_idx(bdata, sidx, step);
 607                 eidx = sidx + PFN_UP(size);
 608
 609                 if (sidx >= midx || eidx > midx)
 610                         break;
 611
 612                 for (i = sidx; i < eidx; i++)
 613                         if (test_bit(i, bdata->node_bootmem_map)) {
 614                                 sidx = align_idx(bdata, i, step);
 615                                 if (sidx == i)
 616                                         sidx += step;
 617                                 goto find_block;
 618                         }
 619
 620                 if (bdata->last_end_off & (PAGE_SIZE - 1) &&
 621                                 PFN_DOWN(bdata->last_end_off) + 1 == sidx)
 622                         start_off = align_off(bdata, bdata->last_end_off, align);
 623                 else
 624                         start_off = PFN_PHYS(sidx);
 625
 626                 merge = PFN_DOWN(start_off) < sidx;
 627                 end_off = start_off + size;
 628
 629                 bdata->last_end_off = end_off;
 630                 bdata->hint_idx = PFN_UP(end_off);
 631
 632                 /*
 633                  * Reserve the area now:
 634                  */
 635                 if (__reserve(bdata, PFN_DOWN(start_off) + merge,
 636                                 PFN_UP(end_off), BOOTMEM_EXCLUSIVE))
 637                         BUG();
 638
 639                 region = phys_to_virt(PFN_PHYS(bdata->node_min_pfn) +
 640                                 start_off);
 641                 memset(region, 0, size);
 642                 /*
 643                  * The min_count is set to 0 so that bootmem allocated blocks
 644                  * are never reported as leaks.
 645                  */
 646                 kmemleak_alloc(region, size, 0, 0);
 647                 return region;
 648         }
 649
 650         if (fallback) {
 651                 sidx = align_idx(bdata, fallback - 1, step);
 652                 fallback = 0;
 653                 goto find_block;
 654         }
 655
 656         return NULL;
 657 }
 658
 659 static void * __init alloc_arch_preferred_bootmem(bootmem_data_t *bdata,
 660                                         unsigned long size, unsigned long align,
 661                                         unsigned long goal, unsigned long limit)
 662 {
 663         if (WARN_ON_ONCE(slab_is_available()))
 664                 return kzalloc(size, GFP_NOWAIT);
 665
 666 #ifdef CONFIG_HAVE_ARCH_BOOTMEM
 667         {
 668                 bootmem_data_t *p_bdata;
 669
 670                 p_bdata = bootmem_arch_preferred_node(bdata, size, align,
 671                                                         goal, limit);
 672                 if (p_bdata)
 673                         return alloc_bootmem_core(p_bdata, size, align,
 674                                                         goal, limit);
 675         }
 676 #endif
 677         return NULL;
 678 }
 679 #endif
 680
 681 static void * __init ___alloc_bootmem_nopanic(unsigned long size,
 682                                         unsigned long align,
 683                                         unsigned long goal,
 684                                         unsigned long limit)
 685 {
 686 #ifdef CONFIG_NO_BOOTMEM
 687         void *ptr;
 688
 689         if (WARN_ON_ONCE(slab_is_available()))
 690                 return kzalloc(size, GFP_NOWAIT);
 691
 692 restart:
 693
 694         ptr = __alloc_memory_core_early(MAX_NUMNODES, size, align, goal, limit);
 695
 696         if (ptr)
 697                 return ptr;
 698
 699         if (goal != 0) {
 700                 goal = 0;
 701                 goto restart;
 702         }
 703
 704         return NULL;
 705 #else
 706         bootmem_data_t *bdata;
 707         void *region;
 708
 709 restart:
 710         region = alloc_arch_preferred_bootmem(NULL, size, align, goal, limit);
 711         if (region)
 712                 return region;
 713
 714         list_for_each_entry(bdata, &bdata_list, list) {
 715                 if (goal && bdata->node_low_pfn <= PFN_DOWN(goal))
 716                         continue;
 717                 if (limit && bdata->node_min_pfn >= PFN_DOWN(limit))
 718                         break;
 719
 720                 region = alloc_bootmem_core(bdata, size, align, goal, limit);
 721                 if (region)
 722                         return region;
 723         }
 724
 725         if (goal) {
 726                 goal = 0;
 727                 goto restart;
 728         }
 729
 730         return NULL;
 731 #endif
 732 }
 733
 734 /**
 735  * __alloc_bootmem_nopanic - allocate boot memory without panicking
 736  * @size: size of the request in bytes
 737  * @align: alignment of the region
 738  * @goal: preferred starting address of the region
 739  *
 740  * The goal is dropped if it can not be satisfied and the allocation will
 741  * fall back to memory below @goal.
 742  *
 743  * Allocation may happen on any node in the system.
 744  *
 745  * Returns NULL on failure.
 746  */
 747 void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align,
 748                                         unsigned long goal)
 749 {
 750         unsigned long limit = 0;
 751
 752 #ifdef CONFIG_NO_BOOTMEM
 753         limit = -1UL;
 754 #endif
 755
 756         return ___alloc_bootmem_nopanic(size, align, goal, limit);
 757 }
 758
 759 static void * __init ___alloc_bootmem(unsigned long size, unsigned long align,
 760                                         unsigned long goal, unsigned long limit)
 761 {
 762         void *mem = ___alloc_bootmem_nopanic(size, align, goal, limit);
 763
 764         if (mem)
 765                 return mem;
 766         /*
 767          * Whoops, we cannot satisfy the allocation request.
 768          */
 769         printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size);
 770         panic("Out of memory");
 771         return NULL;
 772 }
 773
 774 /**
 775  * __alloc_bootmem - allocate boot memory
 776  * @size: size of the request in bytes
 777  * @align: alignment of the region
 778  * @goal: preferred starting address of the region
 779  *
 780  * The goal is dropped if it can not be satisfied and the allocation will
 781  * fall back to memory below @goal.
 782  *
 783  * Allocation may happen on any node in the system.
 784  *
 785  * The function panics if the request can not be satisfied.
 786  */
 787 void * __init __alloc_bootmem(unsigned long size, unsigned long align,
 788                               unsigned long goal)
 789 {
 790         unsigned long limit = 0;
 791
 792 #ifdef CONFIG_NO_BOOTMEM
 793         limit = -1UL;
 794 #endif
 795
 796         return ___alloc_bootmem(size, align, goal, limit);
 797 }
 798
 799 #ifndef CONFIG_NO_BOOTMEM
 800 static void * __init ___alloc_bootmem_node(bootmem_data_t *bdata,
 801                                 unsigned long size, unsigned long align,
 802                                 unsigned long goal, unsigned long limit)
 803 {
 804         void *ptr;
 805
 806         ptr = alloc_arch_preferred_bootmem(bdata, size, align, goal, limit);
 807         if (ptr)
 808                 return ptr;
 809
 810         ptr = alloc_bootmem_core(bdata, size, align, goal, limit);
 811         if (ptr)
 812                 return ptr;
 813
 814         return ___alloc_bootmem(size, align, goal, limit);
 815 }
 816 #endif
 817
 818 /**
 819  * __alloc_bootmem_node - allocate boot memory from a specific node
 820  * @pgdat: node to allocate from
 821  * @size: size of the request in bytes
 822  * @align: alignment of the region
 823  * @goal: preferred starting address of the region
 824  *
 825  * The goal is dropped if it can not be satisfied and the allocation will
 826  * fall back to memory below @goal.
 827  *
 828  * Allocation may fall back to any node in the system if the specified node
 829  * can not hold the requested memory.
 830  *
 831  * The function panics if the request can not be satisfied.
 832  */
 833 void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size,
 834                                    unsigned long align, unsigned long goal)
 835 {
 836         void *ptr;
 837
 838         if (WARN_ON_ONCE(slab_is_available()))
 839                 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
 840
 841 #ifdef CONFIG_NO_BOOTMEM
 842         ptr = __alloc_memory_core_early(pgdat->node_id, size, align,
 843                                          goal, -1ULL);
 844         if (ptr)
 845                 return ptr;
 846
 847         ptr = __alloc_memory_core_early(MAX_NUMNODES, size, align,
 848                                          goal, -1ULL);
 849 #else
 850         ptr = ___alloc_bootmem_node(pgdat->bdata, size, align, goal, 0);
 851 #endif
 852
 853         return ptr;
 854 }
 855
 856 void * __init __alloc_bootmem_node_high(pg_data_t *pgdat, unsigned long size,
 857                                    unsigned long align, unsigned long goal)
 858 {
 859 #ifdef MAX_DMA32_PFN
 860         unsigned long end_pfn;
 861
 862         if (WARN_ON_ONCE(slab_is_available()))
 863                 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
 864
 865         /* update goal according ...MAX_DMA32_PFN */
 866         end_pfn = pgdat->node_start_pfn + pgdat->node_spanned_pages;
 867
 868         if (end_pfn > MAX_DMA32_PFN + (128 >> (20 - PAGE_SHIFT)) &&
 869             (goal >> PAGE_SHIFT) < MAX_DMA32_PFN) {
 870                 void *ptr;
 871                 unsigned long new_goal;
 872
 873                 new_goal = MAX_DMA32_PFN << PAGE_SHIFT;
 874 #ifdef CONFIG_NO_BOOTMEM
 875                 ptr =  __alloc_memory_core_early(pgdat->node_id, size, align,
 876                                                  new_goal, -1ULL);
 877 #else
 878                 ptr = alloc_bootmem_core(pgdat->bdata, size, align,
 879                                                  new_goal, 0);
 880 #endif
 881                 if (ptr)
 882                         return ptr;
 883         }
 884 #endif
 885
 886         return __alloc_bootmem_node(pgdat, size, align, goal);
 887
 888 }
 889
 890 #ifdef CONFIG_SPARSEMEM
 891 /**
 892  * alloc_bootmem_section - allocate boot memory from a specific section
 893  * @size: size of the request in bytes
 894  * @section_nr: sparse map section to allocate from
 895  *
 896  * Return NULL on failure.
 897  */
 898 void * __init alloc_bootmem_section(unsigned long size,
 899                                     unsigned long section_nr)
 900 {
 901 #ifdef CONFIG_NO_BOOTMEM
 902         unsigned long pfn, goal, limit;
 903
 904         pfn = section_nr_to_pfn(section_nr);
 905         goal = pfn << PAGE_SHIFT;
 906         limit = section_nr_to_pfn(section_nr + 1) << PAGE_SHIFT;
 907
 908         return __alloc_memory_core_early(early_pfn_to_nid(pfn), size,
 909                                          SMP_CACHE_BYTES, goal, limit);
 910 #else
 911         bootmem_data_t *bdata;
 912         unsigned long pfn, goal, limit;
 913
 914         pfn = section_nr_to_pfn(section_nr);
 915         goal = pfn << PAGE_SHIFT;
 916         limit = section_nr_to_pfn(section_nr + 1) << PAGE_SHIFT;
 917         bdata = &bootmem_node_data[early_pfn_to_nid(pfn)];
 918
 919         return alloc_bootmem_core(bdata, size, SMP_CACHE_BYTES, goal, limit);
 920 #endif
 921 }
 922 #endif
 923
 924 void * __init __alloc_bootmem_node_nopanic(pg_data_t *pgdat, unsigned long size,
 925                                    unsigned long align, unsigned long goal)
 926 {
 927         void *ptr;
 928
 929         if (WARN_ON_ONCE(slab_is_available()))
 930                 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
 931
 932 #ifdef CONFIG_NO_BOOTMEM
 933         ptr =  __alloc_memory_core_early(pgdat->node_id, size, align,
 934                                                  goal, -1ULL);
 935 #else
 936         ptr = alloc_arch_preferred_bootmem(pgdat->bdata, size, align, goal, 0);
 937         if (ptr)
 938                 return ptr;
 939
 940         ptr = alloc_bootmem_core(pgdat->bdata, size, align, goal, 0);
 941 #endif
 942         if (ptr)
 943                 return ptr;
 944
 945         return __alloc_bootmem_nopanic(size, align, goal);
 946 }
 947
 948 #ifndef ARCH_LOW_ADDRESS_LIMIT
 949 #define ARCH_LOW_ADDRESS_LIMIT  0xffffffffUL
 950 #endif
 951
 952 /**
 953  * __alloc_bootmem_low - allocate low boot memory
 954  * @size: size of the request in bytes
 955  * @align: alignment of the region
 956  * @goal: preferred starting address of the region
 957  *
 958  * The goal is dropped if it can not be satisfied and the allocation will
 959  * fall back to memory below @goal.
 960  *
 961  * Allocation may happen on any node in the system.
 962  *
 963  * The function panics if the request can not be satisfied.
 964  */
 965 void * __init __alloc_bootmem_low(unsigned long size, unsigned long align,
 966                                   unsigned long goal)
 967 {
 968         return ___alloc_bootmem(size, align, goal, ARCH_LOW_ADDRESS_LIMIT);
 969 }
 970
 971 /**
 972  * __alloc_bootmem_low_node - allocate low boot memory from a specific node
 973  * @pgdat: node to allocate from
 974  * @size: size of the request in bytes
 975  * @align: alignment of the region
 976  * @goal: preferred starting address of the region
 977  *
 978  * The goal is dropped if it can not be satisfied and the allocation will
 979  * fall back to memory below @goal.
 980  *
 981  * Allocation may fall back to any node in the system if the specified node
 982  * can not hold the requested memory.
 983  *
 984  * The function panics if the request can not be satisfied.
 985  */
 986 void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size,
 987                                        unsigned long align, unsigned long goal)
 988 {
 989         void *ptr;
 990
 991         if (WARN_ON_ONCE(slab_is_available()))
 992                 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
 993
 994 #ifdef CONFIG_NO_BOOTMEM
 995         ptr = __alloc_memory_core_early(pgdat->node_id, size, align,
 996                                 goal, ARCH_LOW_ADDRESS_LIMIT);
 997         if (ptr)
 998                 return ptr;
 999         ptr = __alloc_memory_core_early(MAX_NUMNODES, size, align,
1000                                 goal, ARCH_LOW_ADDRESS_LIMIT);
1001 #else
1002         ptr = ___alloc_bootmem_node(pgdat->bdata, size, align,
1003                                 goal, ARCH_LOW_ADDRESS_LIMIT);
1004 #endif
1005         return ptr;
1006 }