mm/bootmem.c

   1 /*
   2  *  linux/mm/bootmem.c
   3  *
   4  *  Copyright (C) 1999 Ingo Molnar
   5  *  Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
   6  *
   7  *  simple boot-time physical memory area allocator and
   8  *  free memory collector. It's used to deal with reserved
   9  *  system memory and memory holes as well.
  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 /*
  23  * Access to this subsystem has to be serialized externally. (this is
  24  * true for the boot process anyway)
  25  */
  26 unsigned long max_low_pfn;
  27 unsigned long min_low_pfn;
  28 unsigned long max_pfn;
  29
  30 static LIST_HEAD(bdata_list);
  31 #ifdef CONFIG_CRASH_DUMP
  32 /*
  33  * If we have booted due to a crash, max_pfn will be a very low value. We need
  34  * to know the amount of memory that the previous kernel used.
  35  */
  36 unsigned long saved_max_pfn;
  37 #endif
  38
  39 bootmem_data_t bootmem_node_data[MAX_NUMNODES] __initdata;
  40
  41 /*
  42  * Given an initialised bdata, it returns the size of the boot bitmap
  43  */
  44 static unsigned long __init get_mapsize(bootmem_data_t *bdata)
  45 {
  46         unsigned long mapsize;
  47         unsigned long start = PFN_DOWN(bdata->node_boot_start);
  48         unsigned long end = bdata->node_low_pfn;
  49
  50         mapsize = ((end - start) + 7) / 8;
  51         return ALIGN(mapsize, sizeof(long));
  52 }
  53
  54 /* return the number of _pages_ that will be allocated for the boot bitmap */
  55 unsigned long __init bootmem_bootmap_pages(unsigned long pages)
  56 {
  57         unsigned long mapsize;
  58
  59         mapsize = (pages+7)/8;
  60         mapsize = (mapsize + ~PAGE_MASK) & PAGE_MASK;
  61         mapsize >>= PAGE_SHIFT;
  62
  63         return mapsize;
  64 }
  65
  66 /*
  67  * link bdata in order
  68  */
  69 static void __init link_bootmem(bootmem_data_t *bdata)
  70 {
  71         bootmem_data_t *ent;
  72
  73         if (list_empty(&bdata_list)) {
  74                 list_add(&bdata->list, &bdata_list);
  75                 return;
  76         }
  77         /* insert in order */
  78         list_for_each_entry(ent, &bdata_list, list) {
  79                 if (bdata->node_boot_start < ent->node_boot_start) {
  80                         list_add_tail(&bdata->list, &ent->list);
  81                         return;
  82                 }
  83         }
  84         list_add_tail(&bdata->list, &bdata_list);
  85 }
  86
  87 /*
  88  * Called once to set up the allocator itself.
  89  */
  90 static unsigned long __init init_bootmem_core(bootmem_data_t *bdata,
  91         unsigned long mapstart, unsigned long start, unsigned long end)
  92 {
  93         unsigned long mapsize;
  94
  95         mminit_validate_memmodel_limits(&start, &end);
  96         bdata->node_bootmem_map = phys_to_virt(PFN_PHYS(mapstart));
  97         bdata->node_boot_start = PFN_PHYS(start);
  98         bdata->node_low_pfn = end;
  99         link_bootmem(bdata);
 100
 101         /*
 102          * Initially all pages are reserved - setup_arch() has to
 103          * register free RAM areas explicitly.
 104          */
 105         mapsize = get_mapsize(bdata);
 106         memset(bdata->node_bootmem_map, 0xff, mapsize);
 107
 108         return mapsize;
 109 }
 110
 111 unsigned long __init init_bootmem_node(pg_data_t *pgdat, unsigned long freepfn,
 112                                 unsigned long startpfn, unsigned long endpfn)
 113 {
 114         return init_bootmem_core(pgdat->bdata, freepfn, startpfn, endpfn);
 115 }
 116
 117 unsigned long __init init_bootmem(unsigned long start, unsigned long pages)
 118 {
 119         max_low_pfn = pages;
 120         min_low_pfn = start;
 121         return init_bootmem_core(NODE_DATA(0)->bdata, start, 0, pages);
 122 }
 123
 124 static unsigned long __init free_all_bootmem_core(bootmem_data_t *bdata)
 125 {
 126         struct page *page;
 127         unsigned long pfn;
 128         unsigned long i, count;
 129         unsigned long idx;
 130         unsigned long *map;
 131         int gofast = 0;
 132
 133         BUG_ON(!bdata->node_bootmem_map);
 134
 135         count = 0;
 136         /* first extant page of the node */
 137         pfn = PFN_DOWN(bdata->node_boot_start);
 138         idx = bdata->node_low_pfn - pfn;
 139         map = bdata->node_bootmem_map;
 140         /*
 141          * Check if we are aligned to BITS_PER_LONG pages.  If so, we might
 142          * be able to free page orders of that size at once.
 143          */
 144         if (!(pfn & (BITS_PER_LONG-1)))
 145                 gofast = 1;
 146
 147         for (i = 0; i < idx; ) {
 148                 unsigned long v = ~map[i / BITS_PER_LONG];
 149
 150                 if (gofast && v == ~0UL) {
 151                         int order;
 152
 153                         page = pfn_to_page(pfn);
 154                         count += BITS_PER_LONG;
 155                         order = ffs(BITS_PER_LONG) - 1;
 156                         __free_pages_bootmem(page, order);
 157                         i += BITS_PER_LONG;
 158                         page += BITS_PER_LONG;
 159                 } else if (v) {
 160                         unsigned long m;
 161
 162                         page = pfn_to_page(pfn);
 163                         for (m = 1; m && i < idx; m<<=1, page++, i++) {
 164                                 if (v & m) {
 165                                         count++;
 166                                         __free_pages_bootmem(page, 0);
 167                                 }
 168                         }
 169                 } else {
 170                         i += BITS_PER_LONG;
 171                 }
 172                 pfn += BITS_PER_LONG;
 173         }
 174
 175         /*
 176          * Now free the allocator bitmap itself, it's not
 177          * needed anymore:
 178          */
 179         page = virt_to_page(bdata->node_bootmem_map);
 180         idx = (get_mapsize(bdata) + PAGE_SIZE-1) >> PAGE_SHIFT;
 181         for (i = 0; i < idx; i++, page++)
 182                 __free_pages_bootmem(page, 0);
 183         count += i;
 184         bdata->node_bootmem_map = NULL;
 185
 186         return count;
 187 }
 188
 189 unsigned long __init free_all_bootmem_node(pg_data_t *pgdat)
 190 {
 191         register_page_bootmem_info_node(pgdat);
 192         return free_all_bootmem_core(pgdat->bdata);
 193 }
 194
 195 unsigned long __init free_all_bootmem(void)
 196 {
 197         return free_all_bootmem_core(NODE_DATA(0)->bdata);
 198 }
 199
 200 static void __init free_bootmem_core(bootmem_data_t *bdata, unsigned long addr,
 201                                      unsigned long size)
 202 {
 203         unsigned long sidx, eidx;
 204         unsigned long i;
 205
 206         BUG_ON(!size);
 207
 208         /* out range */
 209         if (addr + size < bdata->node_boot_start ||
 210                 PFN_DOWN(addr) > bdata->node_low_pfn)
 211                 return;
 212         /*
 213          * round down end of usable mem, partially free pages are
 214          * considered reserved.
 215          */
 216
 217         if (addr >= bdata->node_boot_start && addr < bdata->last_success)
 218                 bdata->last_success = addr;
 219
 220         /*
 221          * Round up to index to the range.
 222          */
 223         if (PFN_UP(addr) > PFN_DOWN(bdata->node_boot_start))
 224                 sidx = PFN_UP(addr) - PFN_DOWN(bdata->node_boot_start);
 225         else
 226                 sidx = 0;
 227
 228         eidx = PFN_DOWN(addr + size - bdata->node_boot_start);
 229         if (eidx > bdata->node_low_pfn - PFN_DOWN(bdata->node_boot_start))
 230                 eidx = bdata->node_low_pfn - PFN_DOWN(bdata->node_boot_start);
 231
 232         for (i = sidx; i < eidx; i++) {
 233                 if (unlikely(!test_and_clear_bit(i, bdata->node_bootmem_map)))
 234                         BUG();
 235         }
 236 }
 237
 238 void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
 239                               unsigned long size)
 240 {
 241         free_bootmem_core(pgdat->bdata, physaddr, size);
 242 }
 243
 244 void __init free_bootmem(unsigned long addr, unsigned long size)
 245 {
 246         bootmem_data_t *bdata;
 247         list_for_each_entry(bdata, &bdata_list, list)
 248                 free_bootmem_core(bdata, addr, size);
 249 }
 250
 251 /*
 252  * Marks a particular physical memory range as unallocatable. Usable RAM
 253  * might be used for boot-time allocations - or it might get added
 254  * to the free page pool later on.
 255  */
 256 static int __init can_reserve_bootmem_core(bootmem_data_t *bdata,
 257                         unsigned long addr, unsigned long size, int flags)
 258 {
 259         unsigned long sidx, eidx;
 260         unsigned long i;
 261
 262         BUG_ON(!size);
 263
 264         /* out of range, don't hold other */
 265         if (addr + size < bdata->node_boot_start ||
 266                 PFN_DOWN(addr) > bdata->node_low_pfn)
 267                 return 0;
 268
 269         /*
 270          * Round up to index to the range.
 271          */
 272         if (addr > bdata->node_boot_start)
 273                 sidx= PFN_DOWN(addr - bdata->node_boot_start);
 274         else
 275                 sidx = 0;
 276
 277         eidx = PFN_UP(addr + size - bdata->node_boot_start);
 278         if (eidx > bdata->node_low_pfn - PFN_DOWN(bdata->node_boot_start))
 279                 eidx = bdata->node_low_pfn - PFN_DOWN(bdata->node_boot_start);
 280
 281         for (i = sidx; i < eidx; i++) {
 282                 if (test_bit(i, bdata->node_bootmem_map)) {
 283                         if (flags & BOOTMEM_EXCLUSIVE)
 284                                 return -EBUSY;
 285                 }
 286         }
 287
 288         return 0;
 289
 290 }
 291
 292 static void __init reserve_bootmem_core(bootmem_data_t *bdata,
 293                         unsigned long addr, unsigned long size, int flags)
 294 {
 295         unsigned long sidx, eidx;
 296         unsigned long i;
 297
 298         BUG_ON(!size);
 299
 300         /* out of range */
 301         if (addr + size < bdata->node_boot_start ||
 302                 PFN_DOWN(addr) > bdata->node_low_pfn)
 303                 return;
 304
 305         /*
 306          * Round up to index to the range.
 307          */
 308         if (addr > bdata->node_boot_start)
 309                 sidx= PFN_DOWN(addr - bdata->node_boot_start);
 310         else
 311                 sidx = 0;
 312
 313         eidx = PFN_UP(addr + size - bdata->node_boot_start);
 314         if (eidx > bdata->node_low_pfn - PFN_DOWN(bdata->node_boot_start))
 315                 eidx = bdata->node_low_pfn - PFN_DOWN(bdata->node_boot_start);
 316
 317         for (i = sidx; i < eidx; i++) {
 318                 if (test_and_set_bit(i, bdata->node_bootmem_map)) {
 319 #ifdef CONFIG_DEBUG_BOOTMEM
 320                         printk("hm, page %08lx reserved twice.\n", i*PAGE_SIZE);
 321 #endif
 322                 }
 323         }
 324 }
 325
 326 int __init reserve_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
 327                                  unsigned long size, int flags)
 328 {
 329         int ret;
 330
 331         ret = can_reserve_bootmem_core(pgdat->bdata, physaddr, size, flags);
 332         if (ret < 0)
 333                 return -ENOMEM;
 334         reserve_bootmem_core(pgdat->bdata, physaddr, size, flags);
 335         return 0;
 336 }
 337
 338 #ifndef CONFIG_HAVE_ARCH_BOOTMEM_NODE
 339 int __init reserve_bootmem(unsigned long addr, unsigned long size,
 340                             int flags)
 341 {
 342         bootmem_data_t *bdata;
 343         int ret;
 344
 345         list_for_each_entry(bdata, &bdata_list, list) {
 346                 ret = can_reserve_bootmem_core(bdata, addr, size, flags);
 347                 if (ret < 0)
 348                         return ret;
 349         }
 350         list_for_each_entry(bdata, &bdata_list, list)
 351                 reserve_bootmem_core(bdata, addr, size, flags);
 352
 353         return 0;
 354 }
 355 #endif /* !CONFIG_HAVE_ARCH_BOOTMEM_NODE */
 356
 357 /*
 358  * We 'merge' subsequent allocations to save space. We might 'lose'
 359  * some fraction of a page if allocations cannot be satisfied due to
 360  * size constraints on boxes where there is physical RAM space
 361  * fragmentation - in these cases (mostly large memory boxes) this
 362  * is not a problem.
 363  *
 364  * On low memory boxes we get it right in 100% of the cases.
 365  *
 366  * alignment has to be a power of 2 value.
 367  *
 368  * NOTE:  This function is _not_ reentrant.
 369  */
 370 static void * __init
 371 alloc_bootmem_core(struct bootmem_data *bdata, unsigned long size,
 372                 unsigned long align, unsigned long goal, unsigned long limit)
 373 {
 374         unsigned long areasize, preferred;
 375         unsigned long i, start = 0, incr, eidx, end_pfn;
 376         void *ret;
 377         unsigned long node_boot_start;
 378         void *node_bootmem_map;
 379
 380         if (!size) {
 381                 printk("alloc_bootmem_core(): zero-sized request\n");
 382                 BUG();
 383         }
 384         BUG_ON(align & (align-1));
 385
 386         /* on nodes without memory - bootmem_map is NULL */
 387         if (!bdata->node_bootmem_map)
 388                 return NULL;
 389
 390         /* bdata->node_boot_start is supposed to be (12+6)bits alignment on x86_64 ? */
 391         node_boot_start = bdata->node_boot_start;
 392         node_bootmem_map = bdata->node_bootmem_map;
 393         if (align) {
 394                 node_boot_start = ALIGN(bdata->node_boot_start, align);
 395                 if (node_boot_start > bdata->node_boot_start)
 396                         node_bootmem_map = (unsigned long *)bdata->node_bootmem_map +
 397                             PFN_DOWN(node_boot_start - bdata->node_boot_start)/BITS_PER_LONG;
 398         }
 399
 400         if (limit && node_boot_start >= limit)
 401                 return NULL;
 402
 403         end_pfn = bdata->node_low_pfn;
 404         limit = PFN_DOWN(limit);
 405         if (limit && end_pfn > limit)
 406                 end_pfn = limit;
 407
 408         eidx = end_pfn - PFN_DOWN(node_boot_start);
 409
 410         /*
 411          * We try to allocate bootmem pages above 'goal'
 412          * first, then we try to allocate lower pages.
 413          */
 414         preferred = 0;
 415         if (goal && PFN_DOWN(goal) < end_pfn) {
 416                 if (goal > node_boot_start)
 417                         preferred = goal - node_boot_start;
 418
 419                 if (bdata->last_success > node_boot_start &&
 420                         bdata->last_success - node_boot_start >= preferred)
 421                         if (!limit || (limit && limit > bdata->last_success))
 422                                 preferred = bdata->last_success - node_boot_start;
 423         }
 424
 425         preferred = PFN_DOWN(ALIGN(preferred, align));
 426         areasize = (size + PAGE_SIZE-1) / PAGE_SIZE;
 427         incr = align >> PAGE_SHIFT ? : 1;
 428
 429 restart_scan:
 430         for (i = preferred; i < eidx;) {
 431                 unsigned long j;
 432
 433                 i = find_next_zero_bit(node_bootmem_map, eidx, i);
 434                 i = ALIGN(i, incr);
 435                 if (i >= eidx)
 436                         break;
 437                 if (test_bit(i, node_bootmem_map)) {
 438                         i += incr;
 439                         continue;
 440                 }
 441                 for (j = i + 1; j < i + areasize; ++j) {
 442                         if (j >= eidx)
 443                                 goto fail_block;
 444                         if (test_bit(j, node_bootmem_map))
 445                                 goto fail_block;
 446                 }
 447                 start = i;
 448                 goto found;
 449         fail_block:
 450                 i = ALIGN(j, incr);
 451                 if (i == j)
 452                         i += incr;
 453         }
 454
 455         if (preferred > 0) {
 456                 preferred = 0;
 457                 goto restart_scan;
 458         }
 459         return NULL;
 460
 461 found:
 462         bdata->last_success = PFN_PHYS(start) + node_boot_start;
 463         BUG_ON(start >= eidx);
 464
 465         /*
 466          * Is the next page of the previous allocation-end the start
 467          * of this allocation's buffer? If yes then we can 'merge'
 468          * the previous partial page with this allocation.
 469          */
 470         if (align < PAGE_SIZE &&
 471             bdata->last_offset && bdata->last_pos+1 == start) {
 472                 unsigned long offset, remaining_size;
 473                 offset = ALIGN(bdata->last_offset, align);
 474                 BUG_ON(offset > PAGE_SIZE);
 475                 remaining_size = PAGE_SIZE - offset;
 476                 if (size < remaining_size) {
 477                         areasize = 0;
 478                         /* last_pos unchanged */
 479                         bdata->last_offset = offset + size;
 480                         ret = phys_to_virt(bdata->last_pos * PAGE_SIZE +
 481                                            offset + node_boot_start);
 482                 } else {
 483                         remaining_size = size - remaining_size;
 484                         areasize = (remaining_size + PAGE_SIZE-1) / PAGE_SIZE;
 485                         ret = phys_to_virt(bdata->last_pos * PAGE_SIZE +
 486                                            offset + node_boot_start);
 487                         bdata->last_pos = start + areasize - 1;
 488                         bdata->last_offset = remaining_size;
 489                 }
 490                 bdata->last_offset &= ~PAGE_MASK;
 491         } else {
 492                 bdata->last_pos = start + areasize - 1;
 493                 bdata->last_offset = size & ~PAGE_MASK;
 494                 ret = phys_to_virt(start * PAGE_SIZE + node_boot_start);
 495         }
 496
 497         /*
 498          * Reserve the area now:
 499          */
 500         for (i = start; i < start + areasize; i++)
 501                 if (unlikely(test_and_set_bit(i, node_bootmem_map)))
 502                         BUG();
 503         memset(ret, 0, size);
 504         return ret;
 505 }
 506
 507 void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align,
 508                                       unsigned long goal)
 509 {
 510         bootmem_data_t *bdata;
 511         void *ptr;
 512
 513         list_for_each_entry(bdata, &bdata_list, list) {
 514                 ptr = alloc_bootmem_core(bdata, size, align, goal, 0);
 515                 if (ptr)
 516                         return ptr;
 517         }
 518         return NULL;
 519 }
 520
 521 void * __init __alloc_bootmem(unsigned long size, unsigned long align,
 522                               unsigned long goal)
 523 {
 524         void *mem = __alloc_bootmem_nopanic(size,align,goal);
 525
 526         if (mem)
 527                 return mem;
 528         /*
 529          * Whoops, we cannot satisfy the allocation request.
 530          */
 531         printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size);
 532         panic("Out of memory");
 533         return NULL;
 534 }
 535
 536 void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size,
 537                                    unsigned long align, unsigned long goal)
 538 {
 539         void *ptr;
 540
 541         ptr = alloc_bootmem_core(pgdat->bdata, size, align, goal, 0);
 542         if (ptr)
 543                 return ptr;
 544
 545         return __alloc_bootmem(size, align, goal);
 546 }
 547
 548 #ifdef CONFIG_SPARSEMEM
 549 void * __init alloc_bootmem_section(unsigned long size,
 550                                     unsigned long section_nr)
 551 {
 552         void *ptr;
 553         unsigned long limit, goal, start_nr, end_nr, pfn;
 554         struct pglist_data *pgdat;
 555
 556         pfn = section_nr_to_pfn(section_nr);
 557         goal = PFN_PHYS(pfn);
 558         limit = PFN_PHYS(section_nr_to_pfn(section_nr + 1)) - 1;
 559         pgdat = NODE_DATA(early_pfn_to_nid(pfn));
 560         ptr = alloc_bootmem_core(pgdat->bdata, size, SMP_CACHE_BYTES, goal,
 561                                 limit);
 562
 563         if (!ptr)
 564                 return NULL;
 565
 566         start_nr = pfn_to_section_nr(PFN_DOWN(__pa(ptr)));
 567         end_nr = pfn_to_section_nr(PFN_DOWN(__pa(ptr) + size));
 568         if (start_nr != section_nr || end_nr != section_nr) {
 569                 printk(KERN_WARNING "alloc_bootmem failed on section %ld.\n",
 570                        section_nr);
 571                 free_bootmem_core(pgdat->bdata, __pa(ptr), size);
 572                 ptr = NULL;
 573         }
 574
 575         return ptr;
 576 }
 577 #endif
 578
 579 void * __init __alloc_bootmem_node_nopanic(pg_data_t *pgdat, unsigned long size,
 580                                    unsigned long align, unsigned long goal)
 581 {
 582         void *ptr;
 583
 584         ptr = alloc_bootmem_core(pgdat->bdata, size, align, goal, 0);
 585         if (ptr)
 586                 return ptr;
 587
 588         return __alloc_bootmem_nopanic(size, align, goal);
 589 }
 590
 591 #ifndef ARCH_LOW_ADDRESS_LIMIT
 592 #define ARCH_LOW_ADDRESS_LIMIT  0xffffffffUL
 593 #endif
 594
 595 void * __init __alloc_bootmem_low(unsigned long size, unsigned long align,
 596                                   unsigned long goal)
 597 {
 598         bootmem_data_t *bdata;
 599         void *ptr;
 600
 601         list_for_each_entry(bdata, &bdata_list, list) {
 602                 ptr = alloc_bootmem_core(bdata, size, align, goal,
 603                                         ARCH_LOW_ADDRESS_LIMIT);
 604                 if (ptr)
 605                         return ptr;
 606         }
 607
 608         /*
 609          * Whoops, we cannot satisfy the allocation request.
 610          */
 611         printk(KERN_ALERT "low bootmem alloc of %lu bytes failed!\n", size);
 612         panic("Out of low memory");
 613         return NULL;
 614 }
 615
 616 void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size,
 617                                        unsigned long align, unsigned long goal)
 618 {
 619         return alloc_bootmem_core(pgdat->bdata, size, align, goal,
 620                                 ARCH_LOW_ADDRESS_LIMIT);
 621 }