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         return free_all_memory_core_early(NODE_DATA(0)->node_id);
 308 #else
 309         return free_all_bootmem_core(NODE_DATA(0)->bdata);
 310 #endif
 311 }
 312
 313 #ifndef CONFIG_NO_BOOTMEM
 314 static void __init __free(bootmem_data_t *bdata,
 315                         unsigned long sidx, unsigned long eidx)
 316 {
 317         unsigned long idx;
 318
 319         bdebug("nid=%td start=%lx end=%lx\n", bdata - bootmem_node_data,
 320                 sidx + bdata->node_min_pfn,
 321                 eidx + bdata->node_min_pfn);
 322
 323         if (bdata->hint_idx > sidx)
 324                 bdata->hint_idx = sidx;
 325
 326         for (idx = sidx; idx < eidx; idx++)
 327                 if (!test_and_clear_bit(idx, bdata->node_bootmem_map))
 328                         BUG();
 329 }
 330
 331 static int __init __reserve(bootmem_data_t *bdata, unsigned long sidx,
 332                         unsigned long eidx, int flags)
 333 {
 334         unsigned long idx;
 335         int exclusive = flags & BOOTMEM_EXCLUSIVE;
 336
 337         bdebug("nid=%td start=%lx end=%lx flags=%x\n",
 338                 bdata - bootmem_node_data,
 339                 sidx + bdata->node_min_pfn,
 340                 eidx + bdata->node_min_pfn,
 341                 flags);
 342
 343         for (idx = sidx; idx < eidx; idx++)
 344                 if (test_and_set_bit(idx, bdata->node_bootmem_map)) {
 345                         if (exclusive) {
 346                                 __free(bdata, sidx, idx);
 347                                 return -EBUSY;
 348                         }
 349                         bdebug("silent double reserve of PFN %lx\n",
 350                                 idx + bdata->node_min_pfn);
 351                 }
 352         return 0;
 353 }
 354
 355 static int __init mark_bootmem_node(bootmem_data_t *bdata,
 356                                 unsigned long start, unsigned long end,
 357                                 int reserve, int flags)
 358 {
 359         unsigned long sidx, eidx;
 360
 361         bdebug("nid=%td start=%lx end=%lx reserve=%d flags=%x\n",
 362                 bdata - bootmem_node_data, start, end, reserve, flags);
 363
 364         BUG_ON(start < bdata->node_min_pfn);
 365         BUG_ON(end > bdata->node_low_pfn);
 366
 367         sidx = start - bdata->node_min_pfn;
 368         eidx = end - bdata->node_min_pfn;
 369
 370         if (reserve)
 371                 return __reserve(bdata, sidx, eidx, flags);
 372         else
 373                 __free(bdata, sidx, eidx);
 374         return 0;
 375 }
 376
 377 static int __init mark_bootmem(unsigned long start, unsigned long end,
 378                                 int reserve, int flags)
 379 {
 380         unsigned long pos;
 381         bootmem_data_t *bdata;
 382
 383         pos = start;
 384         list_for_each_entry(bdata, &bdata_list, list) {
 385                 int err;
 386                 unsigned long max;
 387
 388                 if (pos < bdata->node_min_pfn ||
 389                     pos >= bdata->node_low_pfn) {
 390                         BUG_ON(pos != start);
 391                         continue;
 392                 }
 393
 394                 max = min(bdata->node_low_pfn, end);
 395
 396                 err = mark_bootmem_node(bdata, pos, max, reserve, flags);
 397                 if (reserve && err) {
 398                         mark_bootmem(start, pos, 0, 0);
 399                         return err;
 400                 }
 401
 402                 if (max == end)
 403                         return 0;
 404                 pos = bdata->node_low_pfn;
 405         }
 406         BUG();
 407 }
 408 #endif
 409
 410 /**
 411  * free_bootmem_node - mark a page range as usable
 412  * @pgdat: node the range resides on
 413  * @physaddr: starting address of the range
 414  * @size: size of the range in bytes
 415  *
 416  * Partial pages will be considered reserved and left as they are.
 417  *
 418  * The range must reside completely on the specified node.
 419  */
 420 void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
 421                               unsigned long size)
 422 {
 423 #ifdef CONFIG_NO_BOOTMEM
 424         free_early(physaddr, physaddr + size);
 425 #else
 426         unsigned long start, end;
 427
 428         kmemleak_free_part(__va(physaddr), size);
 429
 430         start = PFN_UP(physaddr);
 431         end = PFN_DOWN(physaddr + size);
 432
 433         mark_bootmem_node(pgdat->bdata, start, end, 0, 0);
 434 #endif
 435 }
 436
 437 /**
 438  * free_bootmem - mark a page range as usable
 439  * @addr: starting address of the range
 440  * @size: size of the range in bytes
 441  *
 442  * Partial pages will be considered reserved and left as they are.
 443  *
 444  * The range must be contiguous but may span node boundaries.
 445  */
 446 void __init free_bootmem(unsigned long addr, unsigned long size)
 447 {
 448 #ifdef CONFIG_NO_BOOTMEM
 449         free_early(addr, addr + size);
 450 #else
 451         unsigned long start, end;
 452
 453         kmemleak_free_part(__va(addr), size);
 454
 455         start = PFN_UP(addr);
 456         end = PFN_DOWN(addr + size);
 457
 458         mark_bootmem(start, end, 0, 0);
 459 #endif
 460 }
 461
 462 /**
 463  * reserve_bootmem_node - mark a page range as reserved
 464  * @pgdat: node the range resides on
 465  * @physaddr: starting address of the range
 466  * @size: size of the range in bytes
 467  * @flags: reservation flags (see linux/bootmem.h)
 468  *
 469  * Partial pages will be reserved.
 470  *
 471  * The range must reside completely on the specified node.
 472  */
 473 int __init reserve_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
 474                                  unsigned long size, int flags)
 475 {
 476 #ifdef CONFIG_NO_BOOTMEM
 477         panic("no bootmem");
 478         return 0;
 479 #else
 480         unsigned long start, end;
 481
 482         start = PFN_DOWN(physaddr);
 483         end = PFN_UP(physaddr + size);
 484
 485         return mark_bootmem_node(pgdat->bdata, start, end, 1, flags);
 486 #endif
 487 }
 488
 489 /**
 490  * reserve_bootmem - mark a page range as usable
 491  * @addr: starting address of the range
 492  * @size: size of the range in bytes
 493  * @flags: reservation flags (see linux/bootmem.h)
 494  *
 495  * Partial pages will be reserved.
 496  *
 497  * The range must be contiguous but may span node boundaries.
 498  */
 499 int __init reserve_bootmem(unsigned long addr, unsigned long size,
 500                             int flags)
 501 {
 502 #ifdef CONFIG_NO_BOOTMEM
 503         panic("no bootmem");
 504         return 0;
 505 #else
 506         unsigned long start, end;
 507
 508         start = PFN_DOWN(addr);
 509         end = PFN_UP(addr + size);
 510
 511         return mark_bootmem(start, end, 1, flags);
 512 #endif
 513 }
 514
 515 #ifndef CONFIG_NO_BOOTMEM
 516 static unsigned long __init align_idx(struct bootmem_data *bdata,
 517                                       unsigned long idx, unsigned long step)
 518 {
 519         unsigned long base = bdata->node_min_pfn;
 520
 521         /*
 522          * Align the index with respect to the node start so that the
 523          * combination of both satisfies the requested alignment.
 524          */
 525
 526         return ALIGN(base + idx, step) - base;
 527 }
 528
 529 static unsigned long __init align_off(struct bootmem_data *bdata,
 530                                       unsigned long off, unsigned long align)
 531 {
 532         unsigned long base = PFN_PHYS(bdata->node_min_pfn);
 533
 534         /* Same as align_idx for byte offsets */
 535
 536         return ALIGN(base + off, align) - base;
 537 }
 538
 539 static void * __init alloc_bootmem_core(struct bootmem_data *bdata,
 540                                         unsigned long size, unsigned long align,
 541                                         unsigned long goal, unsigned long limit)
 542 {
 543         unsigned long fallback = 0;
 544         unsigned long min, max, start, sidx, midx, step;
 545
 546         bdebug("nid=%td size=%lx [%lu pages] align=%lx goal=%lx limit=%lx\n",
 547                 bdata - bootmem_node_data, size, PAGE_ALIGN(size) >> PAGE_SHIFT,
 548                 align, goal, limit);
 549
 550         BUG_ON(!size);
 551         BUG_ON(align & (align - 1));
 552         BUG_ON(limit && goal + size > limit);
 553
 554         if (!bdata->node_bootmem_map)
 555                 return NULL;
 556
 557         min = bdata->node_min_pfn;
 558         max = bdata->node_low_pfn;
 559
 560         goal >>= PAGE_SHIFT;
 561         limit >>= PAGE_SHIFT;
 562
 563         if (limit && max > limit)
 564                 max = limit;
 565         if (max <= min)
 566                 return NULL;
 567
 568         step = max(align >> PAGE_SHIFT, 1UL);
 569
 570         if (goal && min < goal && goal < max)
 571                 start = ALIGN(goal, step);
 572         else
 573                 start = ALIGN(min, step);
 574
 575         sidx = start - bdata->node_min_pfn;
 576         midx = max - bdata->node_min_pfn;
 577
 578         if (bdata->hint_idx > sidx) {
 579                 /*
 580                  * Handle the valid case of sidx being zero and still
 581                  * catch the fallback below.
 582                  */
 583                 fallback = sidx + 1;
 584                 sidx = align_idx(bdata, bdata->hint_idx, step);
 585         }
 586
 587         while (1) {
 588                 int merge;
 589                 void *region;
 590                 unsigned long eidx, i, start_off, end_off;
 591 find_block:
 592                 sidx = find_next_zero_bit(bdata->node_bootmem_map, midx, sidx);
 593                 sidx = align_idx(bdata, sidx, step);
 594                 eidx = sidx + PFN_UP(size);
 595
 596                 if (sidx >= midx || eidx > midx)
 597                         break;
 598
 599                 for (i = sidx; i < eidx; i++)
 600                         if (test_bit(i, bdata->node_bootmem_map)) {
 601                                 sidx = align_idx(bdata, i, step);
 602                                 if (sidx == i)
 603                                         sidx += step;
 604                                 goto find_block;
 605                         }
 606
 607                 if (bdata->last_end_off & (PAGE_SIZE - 1) &&
 608                                 PFN_DOWN(bdata->last_end_off) + 1 == sidx)
 609                         start_off = align_off(bdata, bdata->last_end_off, align);
 610                 else
 611                         start_off = PFN_PHYS(sidx);
 612
 613                 merge = PFN_DOWN(start_off) < sidx;
 614                 end_off = start_off + size;
 615
 616                 bdata->last_end_off = end_off;
 617                 bdata->hint_idx = PFN_UP(end_off);
 618
 619                 /*
 620                  * Reserve the area now:
 621                  */
 622                 if (__reserve(bdata, PFN_DOWN(start_off) + merge,
 623                                 PFN_UP(end_off), BOOTMEM_EXCLUSIVE))
 624                         BUG();
 625
 626                 region = phys_to_virt(PFN_PHYS(bdata->node_min_pfn) +
 627                                 start_off);
 628                 memset(region, 0, size);
 629                 /*
 630                  * The min_count is set to 0 so that bootmem allocated blocks
 631                  * are never reported as leaks.
 632                  */
 633                 kmemleak_alloc(region, size, 0, 0);
 634                 return region;
 635         }
 636
 637         if (fallback) {
 638                 sidx = align_idx(bdata, fallback - 1, step);
 639                 fallback = 0;
 640                 goto find_block;
 641         }
 642
 643         return NULL;
 644 }
 645
 646 static void * __init alloc_arch_preferred_bootmem(bootmem_data_t *bdata,
 647                                         unsigned long size, unsigned long align,
 648                                         unsigned long goal, unsigned long limit)
 649 {
 650         if (WARN_ON_ONCE(slab_is_available()))
 651                 return kzalloc(size, GFP_NOWAIT);
 652
 653 #ifdef CONFIG_HAVE_ARCH_BOOTMEM
 654         {
 655                 bootmem_data_t *p_bdata;
 656
 657                 p_bdata = bootmem_arch_preferred_node(bdata, size, align,
 658                                                         goal, limit);
 659                 if (p_bdata)
 660                         return alloc_bootmem_core(p_bdata, size, align,
 661                                                         goal, limit);
 662         }
 663 #endif
 664         return NULL;
 665 }
 666 #endif
 667
 668 static void * __init ___alloc_bootmem_nopanic(unsigned long size,
 669                                         unsigned long align,
 670                                         unsigned long goal,
 671                                         unsigned long limit)
 672 {
 673 #ifdef CONFIG_NO_BOOTMEM
 674         void *ptr;
 675
 676         if (WARN_ON_ONCE(slab_is_available()))
 677                 return kzalloc(size, GFP_NOWAIT);
 678
 679 restart:
 680
 681         ptr = __alloc_memory_core_early(MAX_NUMNODES, size, align, goal, limit);
 682
 683         if (ptr)
 684                 return ptr;
 685
 686         if (goal != 0) {
 687                 goal = 0;
 688                 goto restart;
 689         }
 690
 691         return NULL;
 692 #else
 693         bootmem_data_t *bdata;
 694         void *region;
 695
 696 restart:
 697         region = alloc_arch_preferred_bootmem(NULL, size, align, goal, limit);
 698         if (region)
 699                 return region;
 700
 701         list_for_each_entry(bdata, &bdata_list, list) {
 702                 if (goal && bdata->node_low_pfn <= PFN_DOWN(goal))
 703                         continue;
 704                 if (limit && bdata->node_min_pfn >= PFN_DOWN(limit))
 705                         break;
 706
 707                 region = alloc_bootmem_core(bdata, size, align, goal, limit);
 708                 if (region)
 709                         return region;
 710         }
 711
 712         if (goal) {
 713                 goal = 0;
 714                 goto restart;
 715         }
 716
 717         return NULL;
 718 #endif
 719 }
 720
 721 /**
 722  * __alloc_bootmem_nopanic - allocate boot memory without panicking
 723  * @size: size of the request in bytes
 724  * @align: alignment of the region
 725  * @goal: preferred starting address of the region
 726  *
 727  * The goal is dropped if it can not be satisfied and the allocation will
 728  * fall back to memory below @goal.
 729  *
 730  * Allocation may happen on any node in the system.
 731  *
 732  * Returns NULL on failure.
 733  */
 734 void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align,
 735                                         unsigned long goal)
 736 {
 737         unsigned long limit = 0;
 738
 739 #ifdef CONFIG_NO_BOOTMEM
 740         limit = -1UL;
 741 #endif
 742
 743         return ___alloc_bootmem_nopanic(size, align, goal, limit);
 744 }
 745
 746 static void * __init ___alloc_bootmem(unsigned long size, unsigned long align,
 747                                         unsigned long goal, unsigned long limit)
 748 {
 749         void *mem = ___alloc_bootmem_nopanic(size, align, goal, limit);
 750
 751         if (mem)
 752                 return mem;
 753         /*
 754          * Whoops, we cannot satisfy the allocation request.
 755          */
 756         printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size);
 757         panic("Out of memory");
 758         return NULL;
 759 }
 760
 761 /**
 762  * __alloc_bootmem - allocate boot memory
 763  * @size: size of the request in bytes
 764  * @align: alignment of the region
 765  * @goal: preferred starting address of the region
 766  *
 767  * The goal is dropped if it can not be satisfied and the allocation will
 768  * fall back to memory below @goal.
 769  *
 770  * Allocation may happen on any node in the system.
 771  *
 772  * The function panics if the request can not be satisfied.
 773  */
 774 void * __init __alloc_bootmem(unsigned long size, unsigned long align,
 775                               unsigned long goal)
 776 {
 777         unsigned long limit = 0;
 778
 779 #ifdef CONFIG_NO_BOOTMEM
 780         limit = -1UL;
 781 #endif
 782
 783         return ___alloc_bootmem(size, align, goal, limit);
 784 }
 785
 786 #ifndef CONFIG_NO_BOOTMEM
 787 static void * __init ___alloc_bootmem_node(bootmem_data_t *bdata,
 788                                 unsigned long size, unsigned long align,
 789                                 unsigned long goal, unsigned long limit)
 790 {
 791         void *ptr;
 792
 793         ptr = alloc_arch_preferred_bootmem(bdata, size, align, goal, limit);
 794         if (ptr)
 795                 return ptr;
 796
 797         ptr = alloc_bootmem_core(bdata, size, align, goal, limit);
 798         if (ptr)
 799                 return ptr;
 800
 801         return ___alloc_bootmem(size, align, goal, limit);
 802 }
 803 #endif
 804
 805 /**
 806  * __alloc_bootmem_node - allocate boot memory from a specific node
 807  * @pgdat: node to allocate from
 808  * @size: size of the request in bytes
 809  * @align: alignment of the region
 810  * @goal: preferred starting address of the region
 811  *
 812  * The goal is dropped if it can not be satisfied and the allocation will
 813  * fall back to memory below @goal.
 814  *
 815  * Allocation may fall back to any node in the system if the specified node
 816  * can not hold the requested memory.
 817  *
 818  * The function panics if the request can not be satisfied.
 819  */
 820 void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size,
 821                                    unsigned long align, unsigned long goal)
 822 {
 823         if (WARN_ON_ONCE(slab_is_available()))
 824                 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
 825
 826 #ifdef CONFIG_NO_BOOTMEM
 827         return __alloc_memory_core_early(pgdat->node_id, size, align,
 828                                          goal, -1ULL);
 829 #else
 830         return ___alloc_bootmem_node(pgdat->bdata, size, align, goal, 0);
 831 #endif
 832 }
 833
 834 void * __init __alloc_bootmem_node_high(pg_data_t *pgdat, unsigned long size,
 835                                    unsigned long align, unsigned long goal)
 836 {
 837 #ifdef MAX_DMA32_PFN
 838         unsigned long end_pfn;
 839
 840         if (WARN_ON_ONCE(slab_is_available()))
 841                 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
 842
 843         /* update goal according ...MAX_DMA32_PFN */
 844         end_pfn = pgdat->node_start_pfn + pgdat->node_spanned_pages;
 845
 846         if (end_pfn > MAX_DMA32_PFN + (128 >> (20 - PAGE_SHIFT)) &&
 847             (goal >> PAGE_SHIFT) < MAX_DMA32_PFN) {
 848                 void *ptr;
 849                 unsigned long new_goal;
 850
 851                 new_goal = MAX_DMA32_PFN << PAGE_SHIFT;
 852 #ifdef CONFIG_NO_BOOTMEM
 853                 ptr =  __alloc_memory_core_early(pgdat->node_id, size, align,
 854                                                  new_goal, -1ULL);
 855 #else
 856                 ptr = alloc_bootmem_core(pgdat->bdata, size, align,
 857                                                  new_goal, 0);
 858 #endif
 859                 if (ptr)
 860                         return ptr;
 861         }
 862 #endif
 863
 864         return __alloc_bootmem_node(pgdat, size, align, goal);
 865
 866 }
 867
 868 #ifdef CONFIG_SPARSEMEM
 869 /**
 870  * alloc_bootmem_section - allocate boot memory from a specific section
 871  * @size: size of the request in bytes
 872  * @section_nr: sparse map section to allocate from
 873  *
 874  * Return NULL on failure.
 875  */
 876 void * __init alloc_bootmem_section(unsigned long size,
 877                                     unsigned long section_nr)
 878 {
 879 #ifdef CONFIG_NO_BOOTMEM
 880         unsigned long pfn, goal, limit;
 881
 882         pfn = section_nr_to_pfn(section_nr);
 883         goal = pfn << PAGE_SHIFT;
 884         limit = section_nr_to_pfn(section_nr + 1) << PAGE_SHIFT;
 885
 886         return __alloc_memory_core_early(early_pfn_to_nid(pfn), size,
 887                                          SMP_CACHE_BYTES, goal, limit);
 888 #else
 889         bootmem_data_t *bdata;
 890         unsigned long pfn, goal, limit;
 891
 892         pfn = section_nr_to_pfn(section_nr);
 893         goal = pfn << PAGE_SHIFT;
 894         limit = section_nr_to_pfn(section_nr + 1) << PAGE_SHIFT;
 895         bdata = &bootmem_node_data[early_pfn_to_nid(pfn)];
 896
 897         return alloc_bootmem_core(bdata, size, SMP_CACHE_BYTES, goal, limit);
 898 #endif
 899 }
 900 #endif
 901
 902 void * __init __alloc_bootmem_node_nopanic(pg_data_t *pgdat, unsigned long size,
 903                                    unsigned long align, unsigned long goal)
 904 {
 905         void *ptr;
 906
 907         if (WARN_ON_ONCE(slab_is_available()))
 908                 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
 909
 910 #ifdef CONFIG_NO_BOOTMEM
 911         ptr =  __alloc_memory_core_early(pgdat->node_id, size, align,
 912                                                  goal, -1ULL);
 913 #else
 914         ptr = alloc_arch_preferred_bootmem(pgdat->bdata, size, align, goal, 0);
 915         if (ptr)
 916                 return ptr;
 917
 918         ptr = alloc_bootmem_core(pgdat->bdata, size, align, goal, 0);
 919 #endif
 920         if (ptr)
 921                 return ptr;
 922
 923         return __alloc_bootmem_nopanic(size, align, goal);
 924 }
 925
 926 #ifndef ARCH_LOW_ADDRESS_LIMIT
 927 #define ARCH_LOW_ADDRESS_LIMIT  0xffffffffUL
 928 #endif
 929
 930 /**
 931  * __alloc_bootmem_low - allocate low boot memory
 932  * @size: size of the request in bytes
 933  * @align: alignment of the region
 934  * @goal: preferred starting address of the region
 935  *
 936  * The goal is dropped if it can not be satisfied and the allocation will
 937  * fall back to memory below @goal.
 938  *
 939  * Allocation may happen on any node in the system.
 940  *
 941  * The function panics if the request can not be satisfied.
 942  */
 943 void * __init __alloc_bootmem_low(unsigned long size, unsigned long align,
 944                                   unsigned long goal)
 945 {
 946         return ___alloc_bootmem(size, align, goal, ARCH_LOW_ADDRESS_LIMIT);
 947 }
 948
 949 /**
 950  * __alloc_bootmem_low_node - allocate low boot memory from a specific node
 951  * @pgdat: node to allocate from
 952  * @size: size of the request in bytes
 953  * @align: alignment of the region
 954  * @goal: preferred starting address of the region
 955  *
 956  * The goal is dropped if it can not be satisfied and the allocation will
 957  * fall back to memory below @goal.
 958  *
 959  * Allocation may fall back to any node in the system if the specified node
 960  * can not hold the requested memory.
 961  *
 962  * The function panics if the request can not be satisfied.
 963  */
 964 void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size,
 965                                        unsigned long align, unsigned long goal)
 966 {
 967         if (WARN_ON_ONCE(slab_is_available()))
 968                 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
 969
 970 #ifdef CONFIG_NO_BOOTMEM
 971         return __alloc_memory_core_early(pgdat->node_id, size, align,
 972                                 goal, ARCH_LOW_ADDRESS_LIMIT);
 973 #else
 974         return ___alloc_bootmem_node(pgdat->bdata, size, align,
 975                                 goal, ARCH_LOW_ADDRESS_LIMIT);
 976 #endif
 977 }