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
  16 #include <asm/bug.h>
  17 #include <asm/io.h>
  18 #include <asm/processor.h>
  19
  20 #include "internal.h"
  21
  22 unsigned long max_low_pfn;
  23 unsigned long min_low_pfn;
  24 unsigned long max_pfn;
  25
  26 static LIST_HEAD(bdata_list);
  27 #ifdef CONFIG_CRASH_DUMP
  28 /*
  29  * If we have booted due to a crash, max_pfn will be a very low value. We need
  30  * to know the amount of memory that the previous kernel used.
  31  */
  32 unsigned long saved_max_pfn;
  33 #endif
  34
  35 bootmem_data_t bootmem_node_data[MAX_NUMNODES] __initdata;
  36
  37 /*
  38  * Given an initialised bdata, it returns the size of the boot bitmap
  39  */
  40 static unsigned long __init get_mapsize(bootmem_data_t *bdata)
  41 {
  42         unsigned long mapsize;
  43         unsigned long start = PFN_DOWN(bdata->node_boot_start);
  44         unsigned long end = bdata->node_low_pfn;
  45
  46         mapsize = ((end - start) + 7) / 8;
  47         return ALIGN(mapsize, sizeof(long));
  48 }
  49
  50 /* return the number of _pages_ that will be allocated for the boot bitmap */
  51 unsigned long __init bootmem_bootmap_pages(unsigned long pages)
  52 {
  53         unsigned long mapsize;
  54
  55         mapsize = (pages+7)/8;
  56         mapsize = (mapsize + ~PAGE_MASK) & PAGE_MASK;
  57         mapsize >>= PAGE_SHIFT;
  58
  59         return mapsize;
  60 }
  61
  62 /*
  63  * link bdata in order
  64  */
  65 static void __init link_bootmem(bootmem_data_t *bdata)
  66 {
  67         bootmem_data_t *ent;
  68
  69         if (list_empty(&bdata_list)) {
  70                 list_add(&bdata->list, &bdata_list);
  71                 return;
  72         }
  73         /* insert in order */
  74         list_for_each_entry(ent, &bdata_list, list) {
  75                 if (bdata->node_boot_start < ent->node_boot_start) {
  76                         list_add_tail(&bdata->list, &ent->list);
  77                         return;
  78                 }
  79         }
  80         list_add_tail(&bdata->list, &bdata_list);
  81 }
  82
  83 /*
  84  * Called once to set up the allocator itself.
  85  */
  86 static unsigned long __init init_bootmem_core(bootmem_data_t *bdata,
  87         unsigned long mapstart, unsigned long start, unsigned long end)
  88 {
  89         unsigned long mapsize;
  90
  91         mminit_validate_memmodel_limits(&start, &end);
  92         bdata->node_bootmem_map = phys_to_virt(PFN_PHYS(mapstart));
  93         bdata->node_boot_start = PFN_PHYS(start);
  94         bdata->node_low_pfn = end;
  95         link_bootmem(bdata);
  96
  97         /*
  98          * Initially all pages are reserved - setup_arch() has to
  99          * register free RAM areas explicitly.
 100          */
 101         mapsize = get_mapsize(bdata);
 102         memset(bdata->node_bootmem_map, 0xff, mapsize);
 103
 104         return mapsize;
 105 }
 106
 107 unsigned long __init init_bootmem_node(pg_data_t *pgdat, unsigned long freepfn,
 108                                 unsigned long startpfn, unsigned long endpfn)
 109 {
 110         return init_bootmem_core(pgdat->bdata, freepfn, startpfn, endpfn);
 111 }
 112
 113 unsigned long __init init_bootmem(unsigned long start, unsigned long pages)
 114 {
 115         max_low_pfn = pages;
 116         min_low_pfn = start;
 117         return init_bootmem_core(NODE_DATA(0)->bdata, start, 0, pages);
 118 }
 119
 120 static unsigned long __init free_all_bootmem_core(bootmem_data_t *bdata)
 121 {
 122         struct page *page;
 123         unsigned long pfn;
 124         unsigned long i, count;
 125         unsigned long idx;
 126         unsigned long *map;
 127         int gofast = 0;
 128
 129         BUG_ON(!bdata->node_bootmem_map);
 130
 131         count = 0;
 132         /* first extant page of the node */
 133         pfn = PFN_DOWN(bdata->node_boot_start);
 134         idx = bdata->node_low_pfn - pfn;
 135         map = bdata->node_bootmem_map;
 136         /*
 137          * Check if we are aligned to BITS_PER_LONG pages.  If so, we might
 138          * be able to free page orders of that size at once.
 139          */
 140         if (!(pfn & (BITS_PER_LONG-1)))
 141                 gofast = 1;
 142
 143         for (i = 0; i < idx; ) {
 144                 unsigned long v = ~map[i / BITS_PER_LONG];
 145
 146                 if (gofast && v == ~0UL) {
 147                         int order;
 148
 149                         page = pfn_to_page(pfn);
 150                         count += BITS_PER_LONG;
 151                         order = ffs(BITS_PER_LONG) - 1;
 152                         __free_pages_bootmem(page, order);
 153                         i += BITS_PER_LONG;
 154                         page += BITS_PER_LONG;
 155                 } else if (v) {
 156                         unsigned long m;
 157
 158                         page = pfn_to_page(pfn);
 159                         for (m = 1; m && i < idx; m<<=1, page++, i++) {
 160                                 if (v & m) {
 161                                         count++;
 162                                         __free_pages_bootmem(page, 0);
 163                                 }
 164                         }
 165                 } else {
 166                         i += BITS_PER_LONG;
 167                 }
 168                 pfn += BITS_PER_LONG;
 169         }
 170
 171         /*
 172          * Now free the allocator bitmap itself, it's not
 173          * needed anymore:
 174          */
 175         page = virt_to_page(bdata->node_bootmem_map);
 176         idx = (get_mapsize(bdata) + PAGE_SIZE-1) >> PAGE_SHIFT;
 177         for (i = 0; i < idx; i++, page++)
 178                 __free_pages_bootmem(page, 0);
 179         count += i;
 180         bdata->node_bootmem_map = NULL;
 181
 182         return count;
 183 }
 184
 185 unsigned long __init free_all_bootmem_node(pg_data_t *pgdat)
 186 {
 187         register_page_bootmem_info_node(pgdat);
 188         return free_all_bootmem_core(pgdat->bdata);
 189 }
 190
 191 unsigned long __init free_all_bootmem(void)
 192 {
 193         return free_all_bootmem_core(NODE_DATA(0)->bdata);
 194 }
 195
 196 static void __init free_bootmem_core(bootmem_data_t *bdata, unsigned long addr,
 197                                      unsigned long size)
 198 {
 199         unsigned long sidx, eidx;
 200         unsigned long i;
 201
 202         BUG_ON(!size);
 203
 204         /* out range */
 205         if (addr + size < bdata->node_boot_start ||
 206                 PFN_DOWN(addr) > bdata->node_low_pfn)
 207                 return;
 208         /*
 209          * round down end of usable mem, partially free pages are
 210          * considered reserved.
 211          */
 212
 213         if (addr >= bdata->node_boot_start && addr < bdata->last_success)
 214                 bdata->last_success = addr;
 215
 216         /*
 217          * Round up to index to the range.
 218          */
 219         if (PFN_UP(addr) > PFN_DOWN(bdata->node_boot_start))
 220                 sidx = PFN_UP(addr) - PFN_DOWN(bdata->node_boot_start);
 221         else
 222                 sidx = 0;
 223
 224         eidx = PFN_DOWN(addr + size - bdata->node_boot_start);
 225         if (eidx > bdata->node_low_pfn - PFN_DOWN(bdata->node_boot_start))
 226                 eidx = bdata->node_low_pfn - PFN_DOWN(bdata->node_boot_start);
 227
 228         for (i = sidx; i < eidx; i++) {
 229                 if (unlikely(!test_and_clear_bit(i, bdata->node_bootmem_map)))
 230                         BUG();
 231         }
 232 }
 233
 234 void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
 235                               unsigned long size)
 236 {
 237         free_bootmem_core(pgdat->bdata, physaddr, size);
 238 }
 239
 240 void __init free_bootmem(unsigned long addr, unsigned long size)
 241 {
 242         bootmem_data_t *bdata;
 243         list_for_each_entry(bdata, &bdata_list, list)
 244                 free_bootmem_core(bdata, addr, size);
 245 }
 246
 247 /*
 248  * Marks a particular physical memory range as unallocatable. Usable RAM
 249  * might be used for boot-time allocations - or it might get added
 250  * to the free page pool later on.
 251  */
 252 static int __init can_reserve_bootmem_core(bootmem_data_t *bdata,
 253                         unsigned long addr, unsigned long size, int flags)
 254 {
 255         unsigned long sidx, eidx;
 256         unsigned long i;
 257
 258         BUG_ON(!size);
 259
 260         /* out of range, don't hold other */
 261         if (addr + size < bdata->node_boot_start ||
 262                 PFN_DOWN(addr) > bdata->node_low_pfn)
 263                 return 0;
 264
 265         /*
 266          * Round up to index to the range.
 267          */
 268         if (addr > bdata->node_boot_start)
 269                 sidx= PFN_DOWN(addr - bdata->node_boot_start);
 270         else
 271                 sidx = 0;
 272
 273         eidx = PFN_UP(addr + size - bdata->node_boot_start);
 274         if (eidx > bdata->node_low_pfn - PFN_DOWN(bdata->node_boot_start))
 275                 eidx = bdata->node_low_pfn - PFN_DOWN(bdata->node_boot_start);
 276
 277         for (i = sidx; i < eidx; i++) {
 278                 if (test_bit(i, bdata->node_bootmem_map)) {
 279                         if (flags & BOOTMEM_EXCLUSIVE)
 280                                 return -EBUSY;
 281                 }
 282         }
 283
 284         return 0;
 285
 286 }
 287
 288 static void __init reserve_bootmem_core(bootmem_data_t *bdata,
 289                         unsigned long addr, unsigned long size, int flags)
 290 {
 291         unsigned long sidx, eidx;
 292         unsigned long i;
 293
 294         BUG_ON(!size);
 295
 296         /* out of range */
 297         if (addr + size < bdata->node_boot_start ||
 298                 PFN_DOWN(addr) > bdata->node_low_pfn)
 299                 return;
 300
 301         /*
 302          * Round up to index to the range.
 303          */
 304         if (addr > bdata->node_boot_start)
 305                 sidx= PFN_DOWN(addr - bdata->node_boot_start);
 306         else
 307                 sidx = 0;
 308
 309         eidx = PFN_UP(addr + size - bdata->node_boot_start);
 310         if (eidx > bdata->node_low_pfn - PFN_DOWN(bdata->node_boot_start))
 311                 eidx = bdata->node_low_pfn - PFN_DOWN(bdata->node_boot_start);
 312
 313         for (i = sidx; i < eidx; i++) {
 314                 if (test_and_set_bit(i, bdata->node_bootmem_map)) {
 315 #ifdef CONFIG_DEBUG_BOOTMEM
 316                         printk("hm, page %08lx reserved twice.\n", i*PAGE_SIZE);
 317 #endif
 318                 }
 319         }
 320 }
 321
 322 int __init reserve_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
 323                                  unsigned long size, int flags)
 324 {
 325         int ret;
 326
 327         ret = can_reserve_bootmem_core(pgdat->bdata, physaddr, size, flags);
 328         if (ret < 0)
 329                 return -ENOMEM;
 330         reserve_bootmem_core(pgdat->bdata, physaddr, size, flags);
 331         return 0;
 332 }
 333
 334 #ifndef CONFIG_HAVE_ARCH_BOOTMEM_NODE
 335 int __init reserve_bootmem(unsigned long addr, unsigned long size,
 336                             int flags)
 337 {
 338         bootmem_data_t *bdata;
 339         int ret;
 340
 341         list_for_each_entry(bdata, &bdata_list, list) {
 342                 ret = can_reserve_bootmem_core(bdata, addr, size, flags);
 343                 if (ret < 0)
 344                         return ret;
 345         }
 346         list_for_each_entry(bdata, &bdata_list, list)
 347                 reserve_bootmem_core(bdata, addr, size, flags);
 348
 349         return 0;
 350 }
 351 #endif /* !CONFIG_HAVE_ARCH_BOOTMEM_NODE */
 352
 353 /*
 354  * We 'merge' subsequent allocations to save space. We might 'lose'
 355  * some fraction of a page if allocations cannot be satisfied due to
 356  * size constraints on boxes where there is physical RAM space
 357  * fragmentation - in these cases (mostly large memory boxes) this
 358  * is not a problem.
 359  *
 360  * On low memory boxes we get it right in 100% of the cases.
 361  *
 362  * alignment has to be a power of 2 value.
 363  *
 364  * NOTE:  This function is _not_ reentrant.
 365  */
 366 static void * __init
 367 alloc_bootmem_core(struct bootmem_data *bdata, unsigned long size,
 368                 unsigned long align, unsigned long goal, unsigned long limit)
 369 {
 370         unsigned long areasize, preferred;
 371         unsigned long i, start = 0, incr, eidx, end_pfn;
 372         void *ret;
 373         unsigned long node_boot_start;
 374         void *node_bootmem_map;
 375
 376         if (!size) {
 377                 printk("alloc_bootmem_core(): zero-sized request\n");
 378                 BUG();
 379         }
 380         BUG_ON(align & (align-1));
 381
 382         /* on nodes without memory - bootmem_map is NULL */
 383         if (!bdata->node_bootmem_map)
 384                 return NULL;
 385
 386         /* bdata->node_boot_start is supposed to be (12+6)bits alignment on x86_64 ? */
 387         node_boot_start = bdata->node_boot_start;
 388         node_bootmem_map = bdata->node_bootmem_map;
 389         if (align) {
 390                 node_boot_start = ALIGN(bdata->node_boot_start, align);
 391                 if (node_boot_start > bdata->node_boot_start)
 392                         node_bootmem_map = (unsigned long *)bdata->node_bootmem_map +
 393                             PFN_DOWN(node_boot_start - bdata->node_boot_start)/BITS_PER_LONG;
 394         }
 395
 396         if (limit && node_boot_start >= limit)
 397                 return NULL;
 398
 399         end_pfn = bdata->node_low_pfn;
 400         limit = PFN_DOWN(limit);
 401         if (limit && end_pfn > limit)
 402                 end_pfn = limit;
 403
 404         eidx = end_pfn - PFN_DOWN(node_boot_start);
 405
 406         /*
 407          * We try to allocate bootmem pages above 'goal'
 408          * first, then we try to allocate lower pages.
 409          */
 410         preferred = 0;
 411         if (goal && PFN_DOWN(goal) < end_pfn) {
 412                 if (goal > node_boot_start)
 413                         preferred = goal - node_boot_start;
 414
 415                 if (bdata->last_success > node_boot_start &&
 416                         bdata->last_success - node_boot_start >= preferred)
 417                         if (!limit || (limit && limit > bdata->last_success))
 418                                 preferred = bdata->last_success - node_boot_start;
 419         }
 420
 421         preferred = PFN_DOWN(ALIGN(preferred, align));
 422         areasize = (size + PAGE_SIZE-1) / PAGE_SIZE;
 423         incr = align >> PAGE_SHIFT ? : 1;
 424
 425 restart_scan:
 426         for (i = preferred; i < eidx;) {
 427                 unsigned long j;
 428
 429                 i = find_next_zero_bit(node_bootmem_map, eidx, i);
 430                 i = ALIGN(i, incr);
 431                 if (i >= eidx)
 432                         break;
 433                 if (test_bit(i, node_bootmem_map)) {
 434                         i += incr;
 435                         continue;
 436                 }
 437                 for (j = i + 1; j < i + areasize; ++j) {
 438                         if (j >= eidx)
 439                                 goto fail_block;
 440                         if (test_bit(j, node_bootmem_map))
 441                                 goto fail_block;
 442                 }
 443                 start = i;
 444                 goto found;
 445         fail_block:
 446                 i = ALIGN(j, incr);
 447                 if (i == j)
 448                         i += incr;
 449         }
 450
 451         if (preferred > 0) {
 452                 preferred = 0;
 453                 goto restart_scan;
 454         }
 455         return NULL;
 456
 457 found:
 458         bdata->last_success = PFN_PHYS(start) + node_boot_start;
 459         BUG_ON(start >= eidx);
 460
 461         /*
 462          * Is the next page of the previous allocation-end the start
 463          * of this allocation's buffer? If yes then we can 'merge'
 464          * the previous partial page with this allocation.
 465          */
 466         if (align < PAGE_SIZE &&
 467             bdata->last_offset && bdata->last_pos+1 == start) {
 468                 unsigned long offset, remaining_size;
 469                 offset = ALIGN(bdata->last_offset, align);
 470                 BUG_ON(offset > PAGE_SIZE);
 471                 remaining_size = PAGE_SIZE - offset;
 472                 if (size < remaining_size) {
 473                         areasize = 0;
 474                         /* last_pos unchanged */
 475                         bdata->last_offset = offset + size;
 476                         ret = phys_to_virt(bdata->last_pos * PAGE_SIZE +
 477                                            offset + node_boot_start);
 478                 } else {
 479                         remaining_size = size - remaining_size;
 480                         areasize = (remaining_size + PAGE_SIZE-1) / PAGE_SIZE;
 481                         ret = phys_to_virt(bdata->last_pos * PAGE_SIZE +
 482                                            offset + node_boot_start);
 483                         bdata->last_pos = start + areasize - 1;
 484                         bdata->last_offset = remaining_size;
 485                 }
 486                 bdata->last_offset &= ~PAGE_MASK;
 487         } else {
 488                 bdata->last_pos = start + areasize - 1;
 489                 bdata->last_offset = size & ~PAGE_MASK;
 490                 ret = phys_to_virt(start * PAGE_SIZE + node_boot_start);
 491         }
 492
 493         /*
 494          * Reserve the area now:
 495          */
 496         for (i = start; i < start + areasize; i++)
 497                 if (unlikely(test_and_set_bit(i, node_bootmem_map)))
 498                         BUG();
 499         memset(ret, 0, size);
 500         return ret;
 501 }
 502
 503 void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align,
 504                                       unsigned long goal)
 505 {
 506         bootmem_data_t *bdata;
 507         void *ptr;
 508
 509         list_for_each_entry(bdata, &bdata_list, list) {
 510                 ptr = alloc_bootmem_core(bdata, size, align, goal, 0);
 511                 if (ptr)
 512                         return ptr;
 513         }
 514         return NULL;
 515 }
 516
 517 void * __init __alloc_bootmem(unsigned long size, unsigned long align,
 518                               unsigned long goal)
 519 {
 520         void *mem = __alloc_bootmem_nopanic(size,align,goal);
 521
 522         if (mem)
 523                 return mem;
 524         /*
 525          * Whoops, we cannot satisfy the allocation request.
 526          */
 527         printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size);
 528         panic("Out of memory");
 529         return NULL;
 530 }
 531
 532 void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size,
 533                                    unsigned long align, unsigned long goal)
 534 {
 535         void *ptr;
 536
 537         ptr = alloc_bootmem_core(pgdat->bdata, size, align, goal, 0);
 538         if (ptr)
 539                 return ptr;
 540
 541         return __alloc_bootmem(size, align, goal);
 542 }
 543
 544 #ifdef CONFIG_SPARSEMEM
 545 void * __init alloc_bootmem_section(unsigned long size,
 546                                     unsigned long section_nr)
 547 {
 548         void *ptr;
 549         unsigned long limit, goal, start_nr, end_nr, pfn;
 550         struct pglist_data *pgdat;
 551
 552         pfn = section_nr_to_pfn(section_nr);
 553         goal = PFN_PHYS(pfn);
 554         limit = PFN_PHYS(section_nr_to_pfn(section_nr + 1)) - 1;
 555         pgdat = NODE_DATA(early_pfn_to_nid(pfn));
 556         ptr = alloc_bootmem_core(pgdat->bdata, size, SMP_CACHE_BYTES, goal,
 557                                 limit);
 558
 559         if (!ptr)
 560                 return NULL;
 561
 562         start_nr = pfn_to_section_nr(PFN_DOWN(__pa(ptr)));
 563         end_nr = pfn_to_section_nr(PFN_DOWN(__pa(ptr) + size));
 564         if (start_nr != section_nr || end_nr != section_nr) {
 565                 printk(KERN_WARNING "alloc_bootmem failed on section %ld.\n",
 566                        section_nr);
 567                 free_bootmem_core(pgdat->bdata, __pa(ptr), size);
 568                 ptr = NULL;
 569         }
 570
 571         return ptr;
 572 }
 573 #endif
 574
 575 void * __init __alloc_bootmem_node_nopanic(pg_data_t *pgdat, unsigned long size,
 576                                    unsigned long align, unsigned long goal)
 577 {
 578         void *ptr;
 579
 580         ptr = alloc_bootmem_core(pgdat->bdata, size, align, goal, 0);
 581         if (ptr)
 582                 return ptr;
 583
 584         return __alloc_bootmem_nopanic(size, align, goal);
 585 }
 586
 587 #ifndef ARCH_LOW_ADDRESS_LIMIT
 588 #define ARCH_LOW_ADDRESS_LIMIT  0xffffffffUL
 589 #endif
 590
 591 void * __init __alloc_bootmem_low(unsigned long size, unsigned long align,
 592                                   unsigned long goal)
 593 {
 594         bootmem_data_t *bdata;
 595         void *ptr;
 596
 597         list_for_each_entry(bdata, &bdata_list, list) {
 598                 ptr = alloc_bootmem_core(bdata, size, align, goal,
 599                                         ARCH_LOW_ADDRESS_LIMIT);
 600                 if (ptr)
 601                         return ptr;
 602         }
 603
 604         /*
 605          * Whoops, we cannot satisfy the allocation request.
 606          */
 607         printk(KERN_ALERT "low bootmem alloc of %lu bytes failed!\n", size);
 608         panic("Out of low memory");
 609         return NULL;
 610 }
 611
 612 void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size,
 613                                        unsigned long align, unsigned long goal)
 614 {
 615         return alloc_bootmem_core(pgdat->bdata, size, align, goal,
 616                                 ARCH_LOW_ADDRESS_LIMIT);
 617 }