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/bootmem.h>
  14 #include <linux/module.h>
  15 #include <linux/kmemleak.h>
  16 #include <linux/range.h>
  17
  18 #include <asm/bug.h>
  19 #include <asm/io.h>
  20 #include <asm/processor.h>
  21
  22 #include "internal.h"
  23
  24 unsigned long max_low_pfn;
  25 unsigned long min_low_pfn;
  26 unsigned long max_pfn;
  27
  28 #ifdef CONFIG_CRASH_DUMP
  29 /*
  30  * If we have booted due to a crash, max_pfn will be a very low value. We need
  31  * to know the amount of memory that the previous kernel used.
  32  */
  33 unsigned long saved_max_pfn;
  34 #endif
  35
  36 #ifndef CONFIG_NO_BOOTMEM
  37 bootmem_data_t bootmem_node_data[MAX_NUMNODES] __initdata;
  38
  39 static struct list_head bdata_list __initdata = LIST_HEAD_INIT(bdata_list);
  40
  41 static int bootmem_debug;
  42
  43 static int __init bootmem_debug_setup(char *buf)
  44 {
  45         bootmem_debug = 1;
  46         return 0;
  47 }
  48 early_param("bootmem_debug", bootmem_debug_setup);
  49
  50 #define bdebug(fmt, args...) ({                         \
  51         if (unlikely(bootmem_debug))                    \
  52                 printk(KERN_INFO                        \
  53                         "bootmem::%s " fmt,             \
  54                         __func__, ## args);             \
  55 })
  56
  57 static unsigned long __init bootmap_bytes(unsigned long pages)
  58 {
  59         unsigned long bytes = (pages + 7) / 8;
  60
  61         return ALIGN(bytes, sizeof(long));
  62 }
  63
  64 /**
  65  * bootmem_bootmap_pages - calculate bitmap size in pages
  66  * @pages: number of pages the bitmap has to represent
  67  */
  68 unsigned long __init bootmem_bootmap_pages(unsigned long pages)
  69 {
  70         unsigned long bytes = bootmap_bytes(pages);
  71
  72         return PAGE_ALIGN(bytes) >> PAGE_SHIFT;
  73 }
  74
  75 /*
  76  * link bdata in order
  77  */
  78 static void __init link_bootmem(bootmem_data_t *bdata)
  79 {
  80         struct list_head *iter;
  81
  82         list_for_each(iter, &bdata_list) {
  83                 bootmem_data_t *ent;
  84
  85                 ent = list_entry(iter, bootmem_data_t, list);
  86                 if (bdata->node_min_pfn < ent->node_min_pfn)
  87                         break;
  88         }
  89         list_add_tail(&bdata->list, iter);
  90 }
  91
  92 /*
  93  * Called once to set up the allocator itself.
  94  */
  95 static unsigned long __init init_bootmem_core(bootmem_data_t *bdata,
  96         unsigned long mapstart, unsigned long start, unsigned long end)
  97 {
  98         unsigned long mapsize;
  99
 100         mminit_validate_memmodel_limits(&start, &end);
 101         bdata->node_bootmem_map = phys_to_virt(PFN_PHYS(mapstart));
 102         bdata->node_min_pfn = start;
 103         bdata->node_low_pfn = end;
 104         link_bootmem(bdata);
 105
 106         /*
 107          * Initially all pages are reserved - setup_arch() has to
 108          * register free RAM areas explicitly.
 109          */
 110         mapsize = bootmap_bytes(end - start);
 111         memset(bdata->node_bootmem_map, 0xff, mapsize);
 112
 113         bdebug("nid=%td start=%lx map=%lx end=%lx mapsize=%lx\n",
 114                 bdata - bootmem_node_data, start, mapstart, end, mapsize);
 115
 116         return mapsize;
 117 }
 118
 119 /**
 120  * init_bootmem_node - register a node as boot memory
 121  * @pgdat: node to register
 122  * @freepfn: pfn where the bitmap for this node is to be placed
 123  * @startpfn: first pfn on the node
 124  * @endpfn: first pfn after the node
 125  *
 126  * Returns the number of bytes needed to hold the bitmap for this node.
 127  */
 128 unsigned long __init init_bootmem_node(pg_data_t *pgdat, unsigned long freepfn,
 129                                 unsigned long startpfn, unsigned long endpfn)
 130 {
 131         return init_bootmem_core(pgdat->bdata, freepfn, startpfn, endpfn);
 132 }
 133
 134 /**
 135  * init_bootmem - register boot memory
 136  * @start: pfn where the bitmap is to be placed
 137  * @pages: number of available physical pages
 138  *
 139  * Returns the number of bytes needed to hold the bitmap.
 140  */
 141 unsigned long __init init_bootmem(unsigned long start, unsigned long pages)
 142 {
 143         max_low_pfn = pages;
 144         min_low_pfn = start;
 145         return init_bootmem_core(NODE_DATA(0)->bdata, start, 0, pages);
 146 }
 147 #endif
 148 /*
 149  * free_bootmem_late - free bootmem pages directly to page allocator
 150  * @addr: starting address of the range
 151  * @size: size of the range in bytes
 152  *
 153  * This is only useful when the bootmem allocator has already been torn
 154  * down, but we are still initializing the system.  Pages are given directly
 155  * to the page allocator, no bootmem metadata is updated because it is gone.
 156  */
 157 void __init free_bootmem_late(unsigned long addr, unsigned long size)
 158 {
 159         unsigned long cursor, end;
 160
 161         kmemleak_free_part(__va(addr), size);
 162
 163         cursor = PFN_UP(addr);
 164         end = PFN_DOWN(addr + size);
 165
 166         for (; cursor < end; cursor++) {
 167                 __free_pages_bootmem(pfn_to_page(cursor), 0);
 168                 totalram_pages++;
 169         }
 170 }
 171
 172 #ifdef CONFIG_NO_BOOTMEM
 173 static void __init __free_pages_memory(unsigned long start, unsigned long end)
 174 {
 175         int i;
 176         unsigned long start_aligned, end_aligned;
 177         int order = ilog2(BITS_PER_LONG);
 178
 179         start_aligned = (start + (BITS_PER_LONG - 1)) & ~(BITS_PER_LONG - 1);
 180         end_aligned = end & ~(BITS_PER_LONG - 1);
 181
 182         if (end_aligned <= start_aligned) {
 183                 for (i = start; i < end; i++)
 184                         __free_pages_bootmem(pfn_to_page(i), 0);
 185
 186                 return;
 187         }
 188
 189         for (i = start; i < start_aligned; i++)
 190                 __free_pages_bootmem(pfn_to_page(i), 0);
 191
 192         for (i = start_aligned; i < end_aligned; i += BITS_PER_LONG)
 193                 __free_pages_bootmem(pfn_to_page(i), order);
 194
 195         for (i = end_aligned; i < end; i++)
 196                 __free_pages_bootmem(pfn_to_page(i), 0);
 197 }
 198
 199 unsigned long __init free_all_memory_core_early(int nodeid)
 200 {
 201         int i;
 202         u64 start, end;
 203         unsigned long count = 0;
 204         struct range *range = NULL;
 205         int nr_range;
 206
 207         nr_range = get_free_all_memory_range(&range, nodeid);
 208
 209         for (i = 0; i < nr_range; i++) {
 210                 start = range[i].start;
 211                 end = range[i].end;
 212                 count += end - start;
 213                 __free_pages_memory(start, end);
 214         }
 215
 216         return count;
 217 }
 218 #else
 219 static unsigned long __init free_all_bootmem_core(bootmem_data_t *bdata)
 220 {
 221         int aligned;
 222         struct page *page;
 223         unsigned long start, end, pages, count = 0;
 224
 225         if (!bdata->node_bootmem_map)
 226                 return 0;
 227
 228         start = bdata->node_min_pfn;
 229         end = bdata->node_low_pfn;
 230
 231         /*
 232          * If the start is aligned to the machines wordsize, we might
 233          * be able to free pages in bulks of that order.
 234          */
 235         aligned = !(start & (BITS_PER_LONG - 1));
 236
 237         bdebug("nid=%td start=%lx end=%lx aligned=%d\n",
 238                 bdata - bootmem_node_data, start, end, aligned);
 239
 240         while (start < end) {
 241                 unsigned long *map, idx, vec;
 242
 243                 map = bdata->node_bootmem_map;
 244                 idx = start - bdata->node_min_pfn;
 245                 vec = ~map[idx / BITS_PER_LONG];
 246
 247                 if (aligned && vec == ~0UL && start + BITS_PER_LONG < end) {
 248                         int order = ilog2(BITS_PER_LONG);
 249
 250                         __free_pages_bootmem(pfn_to_page(start), order);
 251                         count += BITS_PER_LONG;
 252                 } else {
 253                         unsigned long off = 0;
 254
 255                         while (vec && off < BITS_PER_LONG) {
 256                                 if (vec & 1) {
 257                                         page = pfn_to_page(start + off);
 258                                         __free_pages_bootmem(page, 0);
 259                                         count++;
 260                                 }
 261                                 vec >>= 1;
 262                                 off++;
 263                         }
 264                 }
 265                 start += BITS_PER_LONG;
 266         }
 267
 268         page = virt_to_page(bdata->node_bootmem_map);
 269         pages = bdata->node_low_pfn - bdata->node_min_pfn;
 270         pages = bootmem_bootmap_pages(pages);
 271         count += pages;
 272         while (pages--)
 273                 __free_pages_bootmem(page++, 0);
 274
 275         bdebug("nid=%td released=%lx\n", bdata - bootmem_node_data, count);
 276
 277         return count;
 278 }
 279 #endif
 280
 281 /**
 282  * free_all_bootmem_node - release a node's free pages to the buddy allocator
 283  * @pgdat: node to be released
 284  *
 285  * Returns the number of pages actually released.
 286  */
 287 unsigned long __init free_all_bootmem_node(pg_data_t *pgdat)
 288 {
 289         register_page_bootmem_info_node(pgdat);
 290 #ifdef CONFIG_NO_BOOTMEM
 291         /* free_all_memory_core_early(MAX_NUMNODES) will be called later */
 292         return 0;
 293 #else
 294         return free_all_bootmem_core(pgdat->bdata);
 295 #endif
 296 }
 297
 298 /**
 299  * free_all_bootmem - release free pages to the buddy allocator
 300  *
 301  * Returns the number of pages actually released.
 302  */
 303 unsigned long __init free_all_bootmem(void)
 304 {
 305 #ifdef CONFIG_NO_BOOTMEM
 306         return free_all_memory_core_early(NODE_DATA(0)->node_id);
 307 #else
 308         return free_all_bootmem_core(NODE_DATA(0)->bdata);
 309 #endif
 310 }
 311
 312 #ifndef CONFIG_NO_BOOTMEM
 313 static void __init __free(bootmem_data_t *bdata,
 314                         unsigned long sidx, unsigned long eidx)
 315 {
 316         unsigned long idx;
 317
 318         bdebug("nid=%td start=%lx end=%lx\n", bdata - bootmem_node_data,
 319                 sidx + bdata->node_min_pfn,
 320                 eidx + bdata->node_min_pfn);
 321
 322         if (bdata->hint_idx > sidx)
 323                 bdata->hint_idx = sidx;
 324
 325         for (idx = sidx; idx < eidx; idx++)
 326                 if (!test_and_clear_bit(idx, bdata->node_bootmem_map))
 327                         BUG();
 328 }
 329
 330 static int __init __reserve(bootmem_data_t *bdata, unsigned long sidx,
 331                         unsigned long eidx, int flags)
 332 {
 333         unsigned long idx;
 334         int exclusive = flags & BOOTMEM_EXCLUSIVE;
 335
 336         bdebug("nid=%td start=%lx end=%lx flags=%x\n",
 337                 bdata - bootmem_node_data,
 338                 sidx + bdata->node_min_pfn,
 339                 eidx + bdata->node_min_pfn,
 340                 flags);
 341
 342         for (idx = sidx; idx < eidx; idx++)
 343                 if (test_and_set_bit(idx, bdata->node_bootmem_map)) {
 344                         if (exclusive) {
 345                                 __free(bdata, sidx, idx);
 346                                 return -EBUSY;
 347                         }
 348                         bdebug("silent double reserve of PFN %lx\n",
 349                                 idx + bdata->node_min_pfn);
 350                 }
 351         return 0;
 352 }
 353
 354 static int __init mark_bootmem_node(bootmem_data_t *bdata,
 355                                 unsigned long start, unsigned long end,
 356                                 int reserve, int flags)
 357 {
 358         unsigned long sidx, eidx;
 359
 360         bdebug("nid=%td start=%lx end=%lx reserve=%d flags=%x\n",
 361                 bdata - bootmem_node_data, start, end, reserve, flags);
 362
 363         BUG_ON(start < bdata->node_min_pfn);
 364         BUG_ON(end > bdata->node_low_pfn);
 365
 366         sidx = start - bdata->node_min_pfn;
 367         eidx = end - bdata->node_min_pfn;
 368
 369         if (reserve)
 370                 return __reserve(bdata, sidx, eidx, flags);
 371         else
 372                 __free(bdata, sidx, eidx);
 373         return 0;
 374 }
 375
 376 static int __init mark_bootmem(unsigned long start, unsigned long end,
 377                                 int reserve, int flags)
 378 {
 379         unsigned long pos;
 380         bootmem_data_t *bdata;
 381
 382         pos = start;
 383         list_for_each_entry(bdata, &bdata_list, list) {
 384                 int err;
 385                 unsigned long max;
 386
 387                 if (pos < bdata->node_min_pfn ||
 388                     pos >= bdata->node_low_pfn) {
 389                         BUG_ON(pos != start);
 390                         continue;
 391                 }
 392
 393                 max = min(bdata->node_low_pfn, end);
 394
 395                 err = mark_bootmem_node(bdata, pos, max, reserve, flags);
 396                 if (reserve && err) {
 397                         mark_bootmem(start, pos, 0, 0);
 398                         return err;
 399                 }
 400
 401                 if (max == end)
 402                         return 0;
 403                 pos = bdata->node_low_pfn;
 404         }
 405         BUG();
 406 }
 407 #endif
 408
 409 /**
 410  * free_bootmem_node - mark a page range as usable
 411  * @pgdat: node the range resides on
 412  * @physaddr: starting address of the range
 413  * @size: size of the range in bytes
 414  *
 415  * Partial pages will be considered reserved and left as they are.
 416  *
 417  * The range must reside completely on the specified node.
 418  */
 419 void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
 420                               unsigned long size)
 421 {
 422 #ifdef CONFIG_NO_BOOTMEM
 423         free_early(physaddr, physaddr + size);
 424 #else
 425         unsigned long start, end;
 426
 427         kmemleak_free_part(__va(physaddr), size);
 428
 429         start = PFN_UP(physaddr);
 430         end = PFN_DOWN(physaddr + size);
 431
 432         mark_bootmem_node(pgdat->bdata, start, end, 0, 0);
 433 #endif
 434 }
 435
 436 /**
 437  * free_bootmem - mark a page range as usable
 438  * @addr: starting address of the range
 439  * @size: size of the range in bytes
 440  *
 441  * Partial pages will be considered reserved and left as they are.
 442  *
 443  * The range must be contiguous but may span node boundaries.
 444  */
 445 void __init free_bootmem(unsigned long addr, unsigned long size)
 446 {
 447 #ifdef CONFIG_NO_BOOTMEM
 448         free_early(addr, addr + size);
 449 #else
 450         unsigned long start, end;
 451
 452         kmemleak_free_part(__va(addr), size);
 453
 454         start = PFN_UP(addr);
 455         end = PFN_DOWN(addr + size);
 456
 457         mark_bootmem(start, end, 0, 0);
 458 #endif
 459 }
 460
 461 /**
 462  * reserve_bootmem_node - mark a page range as reserved
 463  * @pgdat: node the range resides on
 464  * @physaddr: starting address of the range
 465  * @size: size of the range in bytes
 466  * @flags: reservation flags (see linux/bootmem.h)
 467  *
 468  * Partial pages will be reserved.
 469  *
 470  * The range must reside completely on the specified node.
 471  */
 472 int __init reserve_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
 473                                  unsigned long size, int flags)
 474 {
 475 #ifdef CONFIG_NO_BOOTMEM
 476         panic("no bootmem");
 477         return 0;
 478 #else
 479         unsigned long start, end;
 480
 481         start = PFN_DOWN(physaddr);
 482         end = PFN_UP(physaddr + size);
 483
 484         return mark_bootmem_node(pgdat->bdata, start, end, 1, flags);
 485 #endif
 486 }
 487
 488 /**
 489  * reserve_bootmem - mark a page range as usable
 490  * @addr: starting address of the range
 491  * @size: size of the range in bytes
 492  * @flags: reservation flags (see linux/bootmem.h)
 493  *
 494  * Partial pages will be reserved.
 495  *
 496  * The range must be contiguous but may span node boundaries.
 497  */
 498 int __init reserve_bootmem(unsigned long addr, unsigned long size,
 499                             int flags)
 500 {
 501 #ifdef CONFIG_NO_BOOTMEM
 502         panic("no bootmem");
 503         return 0;
 504 #else
 505         unsigned long start, end;
 506
 507         start = PFN_DOWN(addr);
 508         end = PFN_UP(addr + size);
 509
 510         return mark_bootmem(start, end, 1, flags);
 511 #endif
 512 }
 513
 514 #ifndef CONFIG_NO_BOOTMEM
 515 static unsigned long __init align_idx(struct bootmem_data *bdata,
 516                                       unsigned long idx, unsigned long step)
 517 {
 518         unsigned long base = bdata->node_min_pfn;
 519
 520         /*
 521          * Align the index with respect to the node start so that the
 522          * combination of both satisfies the requested alignment.
 523          */
 524
 525         return ALIGN(base + idx, step) - base;
 526 }
 527
 528 static unsigned long __init align_off(struct bootmem_data *bdata,
 529                                       unsigned long off, unsigned long align)
 530 {
 531         unsigned long base = PFN_PHYS(bdata->node_min_pfn);
 532
 533         /* Same as align_idx for byte offsets */
 534
 535         return ALIGN(base + off, align) - base;
 536 }
 537
 538 static void * __init alloc_bootmem_core(struct bootmem_data *bdata,
 539                                         unsigned long size, unsigned long align,
 540                                         unsigned long goal, unsigned long limit)
 541 {
 542         unsigned long fallback = 0;
 543         unsigned long min, max, start, sidx, midx, step;
 544
 545         bdebug("nid=%td size=%lx [%lu pages] align=%lx goal=%lx limit=%lx\n",
 546                 bdata - bootmem_node_data, size, PAGE_ALIGN(size) >> PAGE_SHIFT,
 547                 align, goal, limit);
 548
 549         BUG_ON(!size);
 550         BUG_ON(align & (align - 1));
 551         BUG_ON(limit && goal + size > limit);
 552
 553         if (!bdata->node_bootmem_map)
 554                 return NULL;
 555
 556         min = bdata->node_min_pfn;
 557         max = bdata->node_low_pfn;
 558
 559         goal >>= PAGE_SHIFT;
 560         limit >>= PAGE_SHIFT;
 561
 562         if (limit && max > limit)
 563                 max = limit;
 564         if (max <= min)
 565                 return NULL;
 566
 567         step = max(align >> PAGE_SHIFT, 1UL);
 568
 569         if (goal && min < goal && goal < max)
 570                 start = ALIGN(goal, step);
 571         else
 572                 start = ALIGN(min, step);
 573
 574         sidx = start - bdata->node_min_pfn;
 575         midx = max - bdata->node_min_pfn;
 576
 577         if (bdata->hint_idx > sidx) {
 578                 /*
 579                  * Handle the valid case of sidx being zero and still
 580                  * catch the fallback below.
 581                  */
 582                 fallback = sidx + 1;
 583                 sidx = align_idx(bdata, bdata->hint_idx, step);
 584         }
 585
 586         while (1) {
 587                 int merge;
 588                 void *region;
 589                 unsigned long eidx, i, start_off, end_off;
 590 find_block:
 591                 sidx = find_next_zero_bit(bdata->node_bootmem_map, midx, sidx);
 592                 sidx = align_idx(bdata, sidx, step);
 593                 eidx = sidx + PFN_UP(size);
 594
 595                 if (sidx >= midx || eidx > midx)
 596                         break;
 597
 598                 for (i = sidx; i < eidx; i++)
 599                         if (test_bit(i, bdata->node_bootmem_map)) {
 600                                 sidx = align_idx(bdata, i, step);
 601                                 if (sidx == i)
 602                                         sidx += step;
 603                                 goto find_block;
 604                         }
 605
 606                 if (bdata->last_end_off & (PAGE_SIZE - 1) &&
 607                                 PFN_DOWN(bdata->last_end_off) + 1 == sidx)
 608                         start_off = align_off(bdata, bdata->last_end_off, align);
 609                 else
 610                         start_off = PFN_PHYS(sidx);
 611
 612                 merge = PFN_DOWN(start_off) < sidx;
 613                 end_off = start_off + size;
 614
 615                 bdata->last_end_off = end_off;
 616                 bdata->hint_idx = PFN_UP(end_off);
 617
 618                 /*
 619                  * Reserve the area now:
 620                  */
 621                 if (__reserve(bdata, PFN_DOWN(start_off) + merge,
 622                                 PFN_UP(end_off), BOOTMEM_EXCLUSIVE))
 623                         BUG();
 624
 625                 region = phys_to_virt(PFN_PHYS(bdata->node_min_pfn) +
 626                                 start_off);
 627                 memset(region, 0, size);
 628                 /*
 629                  * The min_count is set to 0 so that bootmem allocated blocks
 630                  * are never reported as leaks.
 631                  */
 632                 kmemleak_alloc(region, size, 0, 0);
 633                 return region;
 634         }
 635
 636         if (fallback) {
 637                 sidx = align_idx(bdata, fallback - 1, step);
 638                 fallback = 0;
 639                 goto find_block;
 640         }
 641
 642         return NULL;
 643 }
 644
 645 static void * __init alloc_arch_preferred_bootmem(bootmem_data_t *bdata,
 646                                         unsigned long size, unsigned long align,
 647                                         unsigned long goal, unsigned long limit)
 648 {
 649         if (WARN_ON_ONCE(slab_is_available()))
 650                 return kzalloc(size, GFP_NOWAIT);
 651
 652 #ifdef CONFIG_HAVE_ARCH_BOOTMEM
 653         {
 654                 bootmem_data_t *p_bdata;
 655
 656                 p_bdata = bootmem_arch_preferred_node(bdata, size, align,
 657                                                         goal, limit);
 658                 if (p_bdata)
 659                         return alloc_bootmem_core(p_bdata, size, align,
 660                                                         goal, limit);
 661         }
 662 #endif
 663         return NULL;
 664 }
 665 #endif
 666
 667 static void * __init ___alloc_bootmem_nopanic(unsigned long size,
 668                                         unsigned long align,
 669                                         unsigned long goal,
 670                                         unsigned long limit)
 671 {
 672 #ifdef CONFIG_NO_BOOTMEM
 673         void *ptr;
 674
 675         if (WARN_ON_ONCE(slab_is_available()))
 676                 return kzalloc(size, GFP_NOWAIT);
 677
 678 restart:
 679
 680         ptr = __alloc_memory_core_early(MAX_NUMNODES, size, align, goal, limit);
 681
 682         if (ptr)
 683                 return ptr;
 684
 685         if (goal != 0) {
 686                 goal = 0;
 687                 goto restart;
 688         }
 689
 690         return NULL;
 691 #else
 692         bootmem_data_t *bdata;
 693         void *region;
 694
 695 restart:
 696         region = alloc_arch_preferred_bootmem(NULL, size, align, goal, limit);
 697         if (region)
 698                 return region;
 699
 700         list_for_each_entry(bdata, &bdata_list, list) {
 701                 if (goal && bdata->node_low_pfn <= PFN_DOWN(goal))
 702                         continue;
 703                 if (limit && bdata->node_min_pfn >= PFN_DOWN(limit))
 704                         break;
 705
 706                 region = alloc_bootmem_core(bdata, size, align, goal, limit);
 707                 if (region)
 708                         return region;
 709         }
 710
 711         if (goal) {
 712                 goal = 0;
 713                 goto restart;
 714         }
 715
 716         return NULL;
 717 #endif
 718 }
 719
 720 /**
 721  * __alloc_bootmem_nopanic - allocate boot memory without panicking
 722  * @size: size of the request in bytes
 723  * @align: alignment of the region
 724  * @goal: preferred starting address of the region
 725  *
 726  * The goal is dropped if it can not be satisfied and the allocation will
 727  * fall back to memory below @goal.
 728  *
 729  * Allocation may happen on any node in the system.
 730  *
 731  * Returns NULL on failure.
 732  */
 733 void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align,
 734                                         unsigned long goal)
 735 {
 736         unsigned long limit = 0;
 737
 738 #ifdef CONFIG_NO_BOOTMEM
 739         limit = -1UL;
 740 #endif
 741
 742         return ___alloc_bootmem_nopanic(size, align, goal, limit);
 743 }
 744
 745 static void * __init ___alloc_bootmem(unsigned long size, unsigned long align,
 746                                         unsigned long goal, unsigned long limit)
 747 {
 748         void *mem = ___alloc_bootmem_nopanic(size, align, goal, limit);
 749
 750         if (mem)
 751                 return mem;
 752         /*
 753          * Whoops, we cannot satisfy the allocation request.
 754          */
 755         printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size);
 756         panic("Out of memory");
 757         return NULL;
 758 }
 759
 760 /**
 761  * __alloc_bootmem - allocate boot memory
 762  * @size: size of the request in bytes
 763  * @align: alignment of the region
 764  * @goal: preferred starting address of the region
 765  *
 766  * The goal is dropped if it can not be satisfied and the allocation will
 767  * fall back to memory below @goal.
 768  *
 769  * Allocation may happen on any node in the system.
 770  *
 771  * The function panics if the request can not be satisfied.
 772  */
 773 void * __init __alloc_bootmem(unsigned long size, unsigned long align,
 774                               unsigned long goal)
 775 {
 776         unsigned long limit = 0;
 777
 778 #ifdef CONFIG_NO_BOOTMEM
 779         limit = -1UL;
 780 #endif
 781
 782         return ___alloc_bootmem(size, align, goal, limit);
 783 }
 784
 785 #ifndef CONFIG_NO_BOOTMEM
 786 static void * __init ___alloc_bootmem_node(bootmem_data_t *bdata,
 787                                 unsigned long size, unsigned long align,
 788                                 unsigned long goal, unsigned long limit)
 789 {
 790         void *ptr;
 791
 792         ptr = alloc_arch_preferred_bootmem(bdata, size, align, goal, limit);
 793         if (ptr)
 794                 return ptr;
 795
 796         ptr = alloc_bootmem_core(bdata, size, align, goal, limit);
 797         if (ptr)
 798                 return ptr;
 799
 800         return ___alloc_bootmem(size, align, goal, limit);
 801 }
 802 #endif
 803
 804 /**
 805  * __alloc_bootmem_node - allocate boot memory from a specific node
 806  * @pgdat: node to allocate from
 807  * @size: size of the request in bytes
 808  * @align: alignment of the region
 809  * @goal: preferred starting address of the region
 810  *
 811  * The goal is dropped if it can not be satisfied and the allocation will
 812  * fall back to memory below @goal.
 813  *
 814  * Allocation may fall back to any node in the system if the specified node
 815  * can not hold the requested memory.
 816  *
 817  * The function panics if the request can not be satisfied.
 818  */
 819 void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size,
 820                                    unsigned long align, unsigned long goal)
 821 {
 822         if (WARN_ON_ONCE(slab_is_available()))
 823                 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
 824
 825 #ifdef CONFIG_NO_BOOTMEM
 826         return __alloc_memory_core_early(pgdat->node_id, size, align,
 827                                          goal, -1ULL);
 828 #else
 829         return ___alloc_bootmem_node(pgdat->bdata, size, align, goal, 0);
 830 #endif
 831 }
 832
 833 void * __init __alloc_bootmem_node_high(pg_data_t *pgdat, unsigned long size,
 834                                    unsigned long align, unsigned long goal)
 835 {
 836 #ifdef MAX_DMA32_PFN
 837         unsigned long end_pfn;
 838
 839         if (WARN_ON_ONCE(slab_is_available()))
 840                 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
 841
 842         /* update goal according ...MAX_DMA32_PFN */
 843         end_pfn = pgdat->node_start_pfn + pgdat->node_spanned_pages;
 844
 845         if (end_pfn > MAX_DMA32_PFN + (128 >> (20 - PAGE_SHIFT)) &&
 846             (goal >> PAGE_SHIFT) < MAX_DMA32_PFN) {
 847                 void *ptr;
 848                 unsigned long new_goal;
 849
 850                 new_goal = MAX_DMA32_PFN << PAGE_SHIFT;
 851 #ifdef CONFIG_NO_BOOTMEM
 852                 ptr =  __alloc_memory_core_early(pgdat->node_id, size, align,
 853                                                  new_goal, -1ULL);
 854 #else
 855                 ptr = alloc_bootmem_core(pgdat->bdata, size, align,
 856                                                  new_goal, 0);
 857 #endif
 858                 if (ptr)
 859                         return ptr;
 860         }
 861 #endif
 862
 863         return __alloc_bootmem_node(pgdat, size, align, goal);
 864
 865 }
 866
 867 #ifdef CONFIG_SPARSEMEM
 868 /**
 869  * alloc_bootmem_section - allocate boot memory from a specific section
 870  * @size: size of the request in bytes
 871  * @section_nr: sparse map section to allocate from
 872  *
 873  * Return NULL on failure.
 874  */
 875 void * __init alloc_bootmem_section(unsigned long size,
 876                                     unsigned long section_nr)
 877 {
 878 #ifdef CONFIG_NO_BOOTMEM
 879         unsigned long pfn, goal, limit;
 880
 881         pfn = section_nr_to_pfn(section_nr);
 882         goal = pfn << PAGE_SHIFT;
 883         limit = section_nr_to_pfn(section_nr + 1) << PAGE_SHIFT;
 884
 885         return __alloc_memory_core_early(early_pfn_to_nid(pfn), size,
 886                                          SMP_CACHE_BYTES, goal, limit);
 887 #else
 888         bootmem_data_t *bdata;
 889         unsigned long pfn, goal, limit;
 890
 891         pfn = section_nr_to_pfn(section_nr);
 892         goal = pfn << PAGE_SHIFT;
 893         limit = section_nr_to_pfn(section_nr + 1) << PAGE_SHIFT;
 894         bdata = &bootmem_node_data[early_pfn_to_nid(pfn)];
 895
 896         return alloc_bootmem_core(bdata, size, SMP_CACHE_BYTES, goal, limit);
 897 #endif
 898 }
 899 #endif
 900
 901 void * __init __alloc_bootmem_node_nopanic(pg_data_t *pgdat, unsigned long size,
 902                                    unsigned long align, unsigned long goal)
 903 {
 904         void *ptr;
 905
 906         if (WARN_ON_ONCE(slab_is_available()))
 907                 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
 908
 909 #ifdef CONFIG_NO_BOOTMEM
 910         ptr =  __alloc_memory_core_early(pgdat->node_id, size, align,
 911                                                  goal, -1ULL);
 912 #else
 913         ptr = alloc_arch_preferred_bootmem(pgdat->bdata, size, align, goal, 0);
 914         if (ptr)
 915                 return ptr;
 916
 917         ptr = alloc_bootmem_core(pgdat->bdata, size, align, goal, 0);
 918 #endif
 919         if (ptr)
 920                 return ptr;
 921
 922         return __alloc_bootmem_nopanic(size, align, goal);
 923 }
 924
 925 #ifndef ARCH_LOW_ADDRESS_LIMIT
 926 #define ARCH_LOW_ADDRESS_LIMIT  0xffffffffUL
 927 #endif
 928
 929 /**
 930  * __alloc_bootmem_low - allocate low boot memory
 931  * @size: size of the request in bytes
 932  * @align: alignment of the region
 933  * @goal: preferred starting address of the region
 934  *
 935  * The goal is dropped if it can not be satisfied and the allocation will
 936  * fall back to memory below @goal.
 937  *
 938  * Allocation may happen on any node in the system.
 939  *
 940  * The function panics if the request can not be satisfied.
 941  */
 942 void * __init __alloc_bootmem_low(unsigned long size, unsigned long align,
 943                                   unsigned long goal)
 944 {
 945         return ___alloc_bootmem(size, align, goal, ARCH_LOW_ADDRESS_LIMIT);
 946 }
 947
 948 /**
 949  * __alloc_bootmem_low_node - allocate low boot memory from a specific node
 950  * @pgdat: node to allocate from
 951  * @size: size of the request in bytes
 952  * @align: alignment of the region
 953  * @goal: preferred starting address of the region
 954  *
 955  * The goal is dropped if it can not be satisfied and the allocation will
 956  * fall back to memory below @goal.
 957  *
 958  * Allocation may fall back to any node in the system if the specified node
 959  * can not hold the requested memory.
 960  *
 961  * The function panics if the request can not be satisfied.
 962  */
 963 void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size,
 964                                        unsigned long align, unsigned long goal)
 965 {
 966         if (WARN_ON_ONCE(slab_is_available()))
 967                 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
 968
 969 #ifdef CONFIG_NO_BOOTMEM
 970         return __alloc_memory_core_early(pgdat->node_id, size, align,
 971                                 goal, ARCH_LOW_ADDRESS_LIMIT);
 972 #else
 973         return ___alloc_bootmem_node(pgdat->bdata, size, align,
 974                                 goal, ARCH_LOW_ADDRESS_LIMIT);
 975 #endif
 976 }