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 /* return the number of _pages_ that will be allocated for the boot bitmap */
  40 unsigned long __init bootmem_bootmap_pages(unsigned long pages)
  41 {
  42         unsigned long mapsize;
  43
  44         mapsize = (pages+7)/8;
  45         mapsize = (mapsize + ~PAGE_MASK) & PAGE_MASK;
  46         mapsize >>= PAGE_SHIFT;
  47
  48         return mapsize;
  49 }
  50
  51 /*
  52  * link bdata in order
  53  */
  54 static void __init link_bootmem(bootmem_data_t *bdata)
  55 {
  56         bootmem_data_t *ent;
  57
  58         if (list_empty(&bdata_list)) {
  59                 list_add(&bdata->list, &bdata_list);
  60                 return;
  61         }
  62         /* insert in order */
  63         list_for_each_entry(ent, &bdata_list, list) {
  64                 if (bdata->node_boot_start < ent->node_boot_start) {
  65                         list_add_tail(&bdata->list, &ent->list);
  66                         return;
  67                 }
  68         }
  69         list_add_tail(&bdata->list, &bdata_list);
  70 }
  71
  72 /*
  73  * Given an initialised bdata, it returns the size of the boot bitmap
  74  */
  75 static unsigned long __init get_mapsize(bootmem_data_t *bdata)
  76 {
  77         unsigned long mapsize;
  78         unsigned long start = PFN_DOWN(bdata->node_boot_start);
  79         unsigned long end = bdata->node_low_pfn;
  80
  81         mapsize = ((end - start) + 7) / 8;
  82         return ALIGN(mapsize, sizeof(long));
  83 }
  84
  85 /*
  86  * Called once to set up the allocator itself.
  87  */
  88 static unsigned long __init init_bootmem_core(pg_data_t *pgdat,
  89         unsigned long mapstart, unsigned long start, unsigned long end)
  90 {
  91         bootmem_data_t *bdata = pgdat->bdata;
  92         unsigned long mapsize;
  93
  94         bdata->node_bootmem_map = phys_to_virt(PFN_PHYS(mapstart));
  95         bdata->node_boot_start = PFN_PHYS(start);
  96         bdata->node_low_pfn = end;
  97         link_bootmem(bdata);
  98
  99         /*
 100          * Initially all pages are reserved - setup_arch() has to
 101          * register free RAM areas explicitly.
 102          */
 103         mapsize = get_mapsize(bdata);
 104         memset(bdata->node_bootmem_map, 0xff, mapsize);
 105
 106         return mapsize;
 107 }
 108
 109 /*
 110  * Marks a particular physical memory range as unallocatable. Usable RAM
 111  * might be used for boot-time allocations - or it might get added
 112  * to the free page pool later on.
 113  */
 114 static void __init reserve_bootmem_core(bootmem_data_t *bdata, unsigned long addr,
 115                                         unsigned long size)
 116 {
 117         unsigned long sidx, eidx;
 118         unsigned long i;
 119
 120         /*
 121          * round up, partially reserved pages are considered
 122          * fully reserved.
 123          */
 124         BUG_ON(!size);
 125         BUG_ON(PFN_DOWN(addr) >= bdata->node_low_pfn);
 126         BUG_ON(PFN_UP(addr + size) > bdata->node_low_pfn);
 127
 128         sidx = PFN_DOWN(addr - bdata->node_boot_start);
 129         eidx = PFN_UP(addr + size - bdata->node_boot_start);
 130
 131         for (i = sidx; i < eidx; i++)
 132                 if (test_and_set_bit(i, bdata->node_bootmem_map)) {
 133 #ifdef CONFIG_DEBUG_BOOTMEM
 134                         printk("hm, page %08lx reserved twice.\n", i*PAGE_SIZE);
 135 #endif
 136                 }
 137 }
 138
 139 static void __init free_bootmem_core(bootmem_data_t *bdata, unsigned long addr,
 140                                      unsigned long size)
 141 {
 142         unsigned long sidx, eidx;
 143         unsigned long i;
 144
 145         /*
 146          * round down end of usable mem, partially free pages are
 147          * considered reserved.
 148          */
 149         BUG_ON(!size);
 150         BUG_ON(PFN_DOWN(addr + size) > bdata->node_low_pfn);
 151
 152         if (addr < bdata->last_success)
 153                 bdata->last_success = addr;
 154
 155         /*
 156          * Round up the beginning of the address.
 157          */
 158         sidx = PFN_UP(addr) - PFN_DOWN(bdata->node_boot_start);
 159         eidx = PFN_DOWN(addr + size - bdata->node_boot_start);
 160
 161         for (i = sidx; i < eidx; i++) {
 162                 if (unlikely(!test_and_clear_bit(i, bdata->node_bootmem_map)))
 163                         BUG();
 164         }
 165 }
 166
 167 /*
 168  * We 'merge' subsequent allocations to save space. We might 'lose'
 169  * some fraction of a page if allocations cannot be satisfied due to
 170  * size constraints on boxes where there is physical RAM space
 171  * fragmentation - in these cases (mostly large memory boxes) this
 172  * is not a problem.
 173  *
 174  * On low memory boxes we get it right in 100% of the cases.
 175  *
 176  * alignment has to be a power of 2 value.
 177  *
 178  * NOTE:  This function is _not_ reentrant.
 179  */
 180 void * __init
 181 __alloc_bootmem_core(struct bootmem_data *bdata, unsigned long size,
 182               unsigned long align, unsigned long goal, unsigned long limit)
 183 {
 184         unsigned long offset, remaining_size, areasize, preferred;
 185         unsigned long i, start = 0, incr, eidx, end_pfn;
 186         void *ret;
 187
 188         if (!size) {
 189                 printk("__alloc_bootmem_core(): zero-sized request\n");
 190                 BUG();
 191         }
 192         BUG_ON(align & (align-1));
 193
 194         if (limit && bdata->node_boot_start >= limit)
 195                 return NULL;
 196
 197         /* on nodes without memory - bootmem_map is NULL */
 198         if (!bdata->node_bootmem_map)
 199                 return NULL;
 200
 201         end_pfn = bdata->node_low_pfn;
 202         limit = PFN_DOWN(limit);
 203         if (limit && end_pfn > limit)
 204                 end_pfn = limit;
 205
 206         eidx = end_pfn - PFN_DOWN(bdata->node_boot_start);
 207         offset = 0;
 208         if (align && (bdata->node_boot_start & (align - 1UL)) != 0)
 209                 offset = align - (bdata->node_boot_start & (align - 1UL));
 210         offset = PFN_DOWN(offset);
 211
 212         /*
 213          * We try to allocate bootmem pages above 'goal'
 214          * first, then we try to allocate lower pages.
 215          */
 216         if (goal && goal >= bdata->node_boot_start && PFN_DOWN(goal) < end_pfn) {
 217                 preferred = goal - bdata->node_boot_start;
 218
 219                 if (bdata->last_success >= preferred)
 220                         if (!limit || (limit && limit > bdata->last_success))
 221                                 preferred = bdata->last_success;
 222         } else
 223                 preferred = 0;
 224
 225         preferred = PFN_DOWN(ALIGN(preferred, align)) + offset;
 226         areasize = (size + PAGE_SIZE-1) / PAGE_SIZE;
 227         incr = align >> PAGE_SHIFT ? : 1;
 228
 229 restart_scan:
 230         for (i = preferred; i < eidx; i += incr) {
 231                 unsigned long j;
 232                 i = find_next_zero_bit(bdata->node_bootmem_map, eidx, i);
 233                 i = ALIGN(i, incr);
 234                 if (i >= eidx)
 235                         break;
 236                 if (test_bit(i, bdata->node_bootmem_map))
 237                         continue;
 238                 for (j = i + 1; j < i + areasize; ++j) {
 239                         if (j >= eidx)
 240                                 goto fail_block;
 241                         if (test_bit(j, bdata->node_bootmem_map))
 242                                 goto fail_block;
 243                 }
 244                 start = i;
 245                 goto found;
 246         fail_block:
 247                 i = ALIGN(j, incr);
 248         }
 249
 250         if (preferred > offset) {
 251                 preferred = offset;
 252                 goto restart_scan;
 253         }
 254         return NULL;
 255
 256 found:
 257         bdata->last_success = PFN_PHYS(start);
 258         BUG_ON(start >= eidx);
 259
 260         /*
 261          * Is the next page of the previous allocation-end the start
 262          * of this allocation's buffer? If yes then we can 'merge'
 263          * the previous partial page with this allocation.
 264          */
 265         if (align < PAGE_SIZE &&
 266             bdata->last_offset && bdata->last_pos+1 == start) {
 267                 offset = ALIGN(bdata->last_offset, align);
 268                 BUG_ON(offset > PAGE_SIZE);
 269                 remaining_size = PAGE_SIZE - offset;
 270                 if (size < remaining_size) {
 271                         areasize = 0;
 272                         /* last_pos unchanged */
 273                         bdata->last_offset = offset + size;
 274                         ret = phys_to_virt(bdata->last_pos * PAGE_SIZE +
 275                                            offset +
 276                                            bdata->node_boot_start);
 277                 } else {
 278                         remaining_size = size - remaining_size;
 279                         areasize = (remaining_size + PAGE_SIZE-1) / PAGE_SIZE;
 280                         ret = phys_to_virt(bdata->last_pos * PAGE_SIZE +
 281                                            offset +
 282                                            bdata->node_boot_start);
 283                         bdata->last_pos = start + areasize - 1;
 284                         bdata->last_offset = remaining_size;
 285                 }
 286                 bdata->last_offset &= ~PAGE_MASK;
 287         } else {
 288                 bdata->last_pos = start + areasize - 1;
 289                 bdata->last_offset = size & ~PAGE_MASK;
 290                 ret = phys_to_virt(start * PAGE_SIZE + bdata->node_boot_start);
 291         }
 292
 293         /*
 294          * Reserve the area now:
 295          */
 296         for (i = start; i < start + areasize; i++)
 297                 if (unlikely(test_and_set_bit(i, bdata->node_bootmem_map)))
 298                         BUG();
 299         memset(ret, 0, size);
 300         return ret;
 301 }
 302
 303 static unsigned long __init free_all_bootmem_core(pg_data_t *pgdat)
 304 {
 305         struct page *page;
 306         unsigned long pfn;
 307         bootmem_data_t *bdata = pgdat->bdata;
 308         unsigned long i, count, total = 0;
 309         unsigned long idx;
 310         unsigned long *map;
 311         int gofast = 0;
 312
 313         BUG_ON(!bdata->node_bootmem_map);
 314
 315         count = 0;
 316         /* first extant page of the node */
 317         pfn = PFN_DOWN(bdata->node_boot_start);
 318         idx = bdata->node_low_pfn - pfn;
 319         map = bdata->node_bootmem_map;
 320         /* Check physaddr is O(LOG2(BITS_PER_LONG)) page aligned */
 321         if (bdata->node_boot_start == 0 ||
 322             ffs(bdata->node_boot_start) - PAGE_SHIFT > ffs(BITS_PER_LONG))
 323                 gofast = 1;
 324         for (i = 0; i < idx; ) {
 325                 unsigned long v = ~map[i / BITS_PER_LONG];
 326
 327                 if (gofast && v == ~0UL) {
 328                         int order;
 329
 330                         page = pfn_to_page(pfn);
 331                         count += BITS_PER_LONG;
 332                         order = ffs(BITS_PER_LONG) - 1;
 333                         __free_pages_bootmem(page, order);
 334                         i += BITS_PER_LONG;
 335                         page += BITS_PER_LONG;
 336                 } else if (v) {
 337                         unsigned long m;
 338
 339                         page = pfn_to_page(pfn);
 340                         for (m = 1; m && i < idx; m<<=1, page++, i++) {
 341                                 if (v & m) {
 342                                         count++;
 343                                         __free_pages_bootmem(page, 0);
 344                                 }
 345                         }
 346                 } else {
 347                         i += BITS_PER_LONG;
 348                 }
 349                 pfn += BITS_PER_LONG;
 350         }
 351         total += count;
 352
 353         /*
 354          * Now free the allocator bitmap itself, it's not
 355          * needed anymore:
 356          */
 357         page = virt_to_page(bdata->node_bootmem_map);
 358         count = 0;
 359         idx = (get_mapsize(bdata) + PAGE_SIZE-1) >> PAGE_SHIFT;
 360         for (i = 0; i < idx; i++, page++) {
 361                 __free_pages_bootmem(page, 0);
 362                 count++;
 363         }
 364         total += count;
 365         bdata->node_bootmem_map = NULL;
 366
 367         return total;
 368 }
 369
 370 unsigned long __init init_bootmem_node(pg_data_t *pgdat, unsigned long freepfn,
 371                                 unsigned long startpfn, unsigned long endpfn)
 372 {
 373         return init_bootmem_core(pgdat, freepfn, startpfn, endpfn);
 374 }
 375
 376 void __init reserve_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
 377                                  unsigned long size)
 378 {
 379         reserve_bootmem_core(pgdat->bdata, physaddr, size);
 380 }
 381
 382 void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
 383                               unsigned long size)
 384 {
 385         free_bootmem_core(pgdat->bdata, physaddr, size);
 386 }
 387
 388 unsigned long __init free_all_bootmem_node(pg_data_t *pgdat)
 389 {
 390         return free_all_bootmem_core(pgdat);
 391 }
 392
 393 unsigned long __init init_bootmem(unsigned long start, unsigned long pages)
 394 {
 395         max_low_pfn = pages;
 396         min_low_pfn = start;
 397         return init_bootmem_core(NODE_DATA(0), start, 0, pages);
 398 }
 399
 400 #ifndef CONFIG_HAVE_ARCH_BOOTMEM_NODE
 401 void __init reserve_bootmem(unsigned long addr, unsigned long size)
 402 {
 403         reserve_bootmem_core(NODE_DATA(0)->bdata, addr, size);
 404 }
 405 #endif /* !CONFIG_HAVE_ARCH_BOOTMEM_NODE */
 406
 407 void __init free_bootmem(unsigned long addr, unsigned long size)
 408 {
 409         free_bootmem_core(NODE_DATA(0)->bdata, addr, size);
 410 }
 411
 412 unsigned long __init free_all_bootmem(void)
 413 {
 414         return free_all_bootmem_core(NODE_DATA(0));
 415 }
 416
 417 void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align,
 418                                       unsigned long goal)
 419 {
 420         bootmem_data_t *bdata;
 421         void *ptr;
 422
 423         list_for_each_entry(bdata, &bdata_list, list) {
 424                 ptr = __alloc_bootmem_core(bdata, size, align, goal, 0);
 425                 if (ptr)
 426                         return ptr;
 427         }
 428         return NULL;
 429 }
 430
 431 void * __init __alloc_bootmem(unsigned long size, unsigned long align,
 432                               unsigned long goal)
 433 {
 434         void *mem = __alloc_bootmem_nopanic(size,align,goal);
 435
 436         if (mem)
 437                 return mem;
 438         /*
 439          * Whoops, we cannot satisfy the allocation request.
 440          */
 441         printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size);
 442         panic("Out of memory");
 443         return NULL;
 444 }
 445
 446
 447 void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size,
 448                                    unsigned long align, unsigned long goal)
 449 {
 450         void *ptr;
 451
 452         ptr = __alloc_bootmem_core(pgdat->bdata, size, align, goal, 0);
 453         if (ptr)
 454                 return ptr;
 455
 456         return __alloc_bootmem(size, align, goal);
 457 }
 458
 459 #ifndef ARCH_LOW_ADDRESS_LIMIT
 460 #define ARCH_LOW_ADDRESS_LIMIT  0xffffffffUL
 461 #endif
 462
 463 void * __init __alloc_bootmem_low(unsigned long size, unsigned long align,
 464                                   unsigned long goal)
 465 {
 466         bootmem_data_t *bdata;
 467         void *ptr;
 468
 469         list_for_each_entry(bdata, &bdata_list, list) {
 470                 ptr = __alloc_bootmem_core(bdata, size, align, goal,
 471                                                 ARCH_LOW_ADDRESS_LIMIT);
 472                 if (ptr)
 473                         return ptr;
 474         }
 475
 476         /*
 477          * Whoops, we cannot satisfy the allocation request.
 478          */
 479         printk(KERN_ALERT "low bootmem alloc of %lu bytes failed!\n", size);
 480         panic("Out of low memory");
 481         return NULL;
 482 }
 483
 484 void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size,
 485                                        unsigned long align, unsigned long goal)
 486 {
 487         return __alloc_bootmem_core(pgdat->bdata, size, align, goal,
 488                                     ARCH_LOW_ADDRESS_LIMIT);
 489 }