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 EXPORT_UNUSED_SYMBOL(max_pfn);  /*  June 2006  */
  31
  32 static LIST_HEAD(bdata_list);
  33 #ifdef CONFIG_CRASH_DUMP
  34 /*
  35  * If we have booted due to a crash, max_pfn will be a very low value. We need
  36  * to know the amount of memory that the previous kernel used.
  37  */
  38 unsigned long saved_max_pfn;
  39 #endif
  40
  41 /* return the number of _pages_ that will be allocated for the boot bitmap */
  42 unsigned long __init bootmem_bootmap_pages(unsigned long pages)
  43 {
  44         unsigned long mapsize;
  45
  46         mapsize = (pages+7)/8;
  47         mapsize = (mapsize + ~PAGE_MASK) & PAGE_MASK;
  48         mapsize >>= PAGE_SHIFT;
  49
  50         return mapsize;
  51 }
  52
  53 /*
  54  * link bdata in order
  55  */
  56 static void __init link_bootmem(bootmem_data_t *bdata)
  57 {
  58         bootmem_data_t *ent;
  59
  60         if (list_empty(&bdata_list)) {
  61                 list_add(&bdata->list, &bdata_list);
  62                 return;
  63         }
  64         /* insert in order */
  65         list_for_each_entry(ent, &bdata_list, list) {
  66                 if (bdata->node_boot_start < ent->node_boot_start) {
  67                         list_add_tail(&bdata->list, &ent->list);
  68                         return;
  69                 }
  70         }
  71         list_add_tail(&bdata->list, &bdata_list);
  72 }
  73
  74 /*
  75  * Given an initialised bdata, it returns the size of the boot bitmap
  76  */
  77 static unsigned long __init get_mapsize(bootmem_data_t *bdata)
  78 {
  79         unsigned long mapsize;
  80         unsigned long start = PFN_DOWN(bdata->node_boot_start);
  81         unsigned long end = bdata->node_low_pfn;
  82
  83         mapsize = ((end - start) + 7) / 8;
  84         return ALIGN(mapsize, sizeof(long));
  85 }
  86
  87 /*
  88  * Called once to set up the allocator itself.
  89  */
  90 static unsigned long __init init_bootmem_core(pg_data_t *pgdat,
  91         unsigned long mapstart, unsigned long start, unsigned long end)
  92 {
  93         bootmem_data_t *bdata = pgdat->bdata;
  94         unsigned long mapsize;
  95
  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 /*
 112  * Marks a particular physical memory range as unallocatable. Usable RAM
 113  * might be used for boot-time allocations - or it might get added
 114  * to the free page pool later on.
 115  */
 116 static void __init reserve_bootmem_core(bootmem_data_t *bdata, unsigned long addr,
 117                                         unsigned long size)
 118 {
 119         unsigned long sidx, eidx;
 120         unsigned long i;
 121
 122         /*
 123          * round up, partially reserved pages are considered
 124          * fully reserved.
 125          */
 126         BUG_ON(!size);
 127         BUG_ON(PFN_DOWN(addr) >= bdata->node_low_pfn);
 128         BUG_ON(PFN_UP(addr + size) > bdata->node_low_pfn);
 129
 130         sidx = PFN_DOWN(addr - bdata->node_boot_start);
 131         eidx = PFN_UP(addr + size - bdata->node_boot_start);
 132
 133         for (i = sidx; i < eidx; i++)
 134                 if (test_and_set_bit(i, bdata->node_bootmem_map)) {
 135 #ifdef CONFIG_DEBUG_BOOTMEM
 136                         printk("hm, page %08lx reserved twice.\n", i*PAGE_SIZE);
 137 #endif
 138                 }
 139 }
 140
 141 static void __init free_bootmem_core(bootmem_data_t *bdata, unsigned long addr,
 142                                      unsigned long size)
 143 {
 144         unsigned long sidx, eidx;
 145         unsigned long i;
 146
 147         /*
 148          * round down end of usable mem, partially free pages are
 149          * considered reserved.
 150          */
 151         BUG_ON(!size);
 152         BUG_ON(PFN_DOWN(addr + size) > bdata->node_low_pfn);
 153
 154         if (addr < bdata->last_success)
 155                 bdata->last_success = addr;
 156
 157         /*
 158          * Round up the beginning of the address.
 159          */
 160         sidx = PFN_UP(addr) - PFN_DOWN(bdata->node_boot_start);
 161         eidx = PFN_DOWN(addr + size - bdata->node_boot_start);
 162
 163         for (i = sidx; i < eidx; i++) {
 164                 if (unlikely(!test_and_clear_bit(i, bdata->node_bootmem_map)))
 165                         BUG();
 166         }
 167 }
 168
 169 /*
 170  * We 'merge' subsequent allocations to save space. We might 'lose'
 171  * some fraction of a page if allocations cannot be satisfied due to
 172  * size constraints on boxes where there is physical RAM space
 173  * fragmentation - in these cases (mostly large memory boxes) this
 174  * is not a problem.
 175  *
 176  * On low memory boxes we get it right in 100% of the cases.
 177  *
 178  * alignment has to be a power of 2 value.
 179  *
 180  * NOTE:  This function is _not_ reentrant.
 181  */
 182 void * __init
 183 __alloc_bootmem_core(struct bootmem_data *bdata, unsigned long size,
 184               unsigned long align, unsigned long goal, unsigned long limit)
 185 {
 186         unsigned long offset, remaining_size, areasize, preferred;
 187         unsigned long i, start = 0, incr, eidx, end_pfn;
 188         void *ret;
 189
 190         if (!size) {
 191                 printk("__alloc_bootmem_core(): zero-sized request\n");
 192                 BUG();
 193         }
 194         BUG_ON(align & (align-1));
 195
 196         if (limit && bdata->node_boot_start >= limit)
 197                 return NULL;
 198
 199         /* on nodes without memory - bootmem_map is NULL */
 200         if (!bdata->node_bootmem_map)
 201                 return NULL;
 202
 203         end_pfn = bdata->node_low_pfn;
 204         limit = PFN_DOWN(limit);
 205         if (limit && end_pfn > limit)
 206                 end_pfn = limit;
 207
 208         eidx = end_pfn - PFN_DOWN(bdata->node_boot_start);
 209         offset = 0;
 210         if (align && (bdata->node_boot_start & (align - 1UL)) != 0)
 211                 offset = align - (bdata->node_boot_start & (align - 1UL));
 212         offset = PFN_DOWN(offset);
 213
 214         /*
 215          * We try to allocate bootmem pages above 'goal'
 216          * first, then we try to allocate lower pages.
 217          */
 218         if (goal && goal >= bdata->node_boot_start && PFN_DOWN(goal) < end_pfn) {
 219                 preferred = goal - bdata->node_boot_start;
 220
 221                 if (bdata->last_success >= preferred)
 222                         if (!limit || (limit && limit > bdata->last_success))
 223                                 preferred = bdata->last_success;
 224         } else
 225                 preferred = 0;
 226
 227         preferred = PFN_DOWN(ALIGN(preferred, align)) + offset;
 228         areasize = (size + PAGE_SIZE-1) / PAGE_SIZE;
 229         incr = align >> PAGE_SHIFT ? : 1;
 230
 231 restart_scan:
 232         for (i = preferred; i < eidx; i += incr) {
 233                 unsigned long j;
 234                 i = find_next_zero_bit(bdata->node_bootmem_map, eidx, i);
 235                 i = ALIGN(i, incr);
 236                 if (i >= eidx)
 237                         break;
 238                 if (test_bit(i, bdata->node_bootmem_map))
 239                         continue;
 240                 for (j = i + 1; j < i + areasize; ++j) {
 241                         if (j >= eidx)
 242                                 goto fail_block;
 243                         if (test_bit(j, bdata->node_bootmem_map))
 244                                 goto fail_block;
 245                 }
 246                 start = i;
 247                 goto found;
 248         fail_block:
 249                 i = ALIGN(j, incr);
 250         }
 251
 252         if (preferred > offset) {
 253                 preferred = offset;
 254                 goto restart_scan;
 255         }
 256         return NULL;
 257
 258 found:
 259         bdata->last_success = PFN_PHYS(start);
 260         BUG_ON(start >= eidx);
 261
 262         /*
 263          * Is the next page of the previous allocation-end the start
 264          * of this allocation's buffer? If yes then we can 'merge'
 265          * the previous partial page with this allocation.
 266          */
 267         if (align < PAGE_SIZE &&
 268             bdata->last_offset && bdata->last_pos+1 == start) {
 269                 offset = ALIGN(bdata->last_offset, align);
 270                 BUG_ON(offset > PAGE_SIZE);
 271                 remaining_size = PAGE_SIZE - offset;
 272                 if (size < remaining_size) {
 273                         areasize = 0;
 274                         /* last_pos unchanged */
 275                         bdata->last_offset = offset + size;
 276                         ret = phys_to_virt(bdata->last_pos * PAGE_SIZE +
 277                                            offset +
 278                                            bdata->node_boot_start);
 279                 } else {
 280                         remaining_size = size - remaining_size;
 281                         areasize = (remaining_size + PAGE_SIZE-1) / PAGE_SIZE;
 282                         ret = phys_to_virt(bdata->last_pos * PAGE_SIZE +
 283                                            offset +
 284                                            bdata->node_boot_start);
 285                         bdata->last_pos = start + areasize - 1;
 286                         bdata->last_offset = remaining_size;
 287                 }
 288                 bdata->last_offset &= ~PAGE_MASK;
 289         } else {
 290                 bdata->last_pos = start + areasize - 1;
 291                 bdata->last_offset = size & ~PAGE_MASK;
 292                 ret = phys_to_virt(start * PAGE_SIZE + bdata->node_boot_start);
 293         }
 294
 295         /*
 296          * Reserve the area now:
 297          */
 298         for (i = start; i < start + areasize; i++)
 299                 if (unlikely(test_and_set_bit(i, bdata->node_bootmem_map)))
 300                         BUG();
 301         memset(ret, 0, size);
 302         return ret;
 303 }
 304
 305 static unsigned long __init free_all_bootmem_core(pg_data_t *pgdat)
 306 {
 307         struct page *page;
 308         unsigned long pfn;
 309         bootmem_data_t *bdata = pgdat->bdata;
 310         unsigned long i, count, total = 0;
 311         unsigned long idx;
 312         unsigned long *map;
 313         int gofast = 0;
 314
 315         BUG_ON(!bdata->node_bootmem_map);
 316
 317         count = 0;
 318         /* first extant page of the node */
 319         pfn = PFN_DOWN(bdata->node_boot_start);
 320         idx = bdata->node_low_pfn - pfn;
 321         map = bdata->node_bootmem_map;
 322         /* Check physaddr is O(LOG2(BITS_PER_LONG)) page aligned */
 323         if (bdata->node_boot_start == 0 ||
 324             ffs(bdata->node_boot_start) - PAGE_SHIFT > ffs(BITS_PER_LONG))
 325                 gofast = 1;
 326         for (i = 0; i < idx; ) {
 327                 unsigned long v = ~map[i / BITS_PER_LONG];
 328
 329                 if (gofast && v == ~0UL) {
 330                         int order;
 331
 332                         page = pfn_to_page(pfn);
 333                         count += BITS_PER_LONG;
 334                         order = ffs(BITS_PER_LONG) - 1;
 335                         __free_pages_bootmem(page, order);
 336                         i += BITS_PER_LONG;
 337                         page += BITS_PER_LONG;
 338                 } else if (v) {
 339                         unsigned long m;
 340
 341                         page = pfn_to_page(pfn);
 342                         for (m = 1; m && i < idx; m<<=1, page++, i++) {
 343                                 if (v & m) {
 344                                         count++;
 345                                         __free_pages_bootmem(page, 0);
 346                                 }
 347                         }
 348                 } else {
 349                         i += BITS_PER_LONG;
 350                 }
 351                 pfn += BITS_PER_LONG;
 352         }
 353         total += count;
 354
 355         /*
 356          * Now free the allocator bitmap itself, it's not
 357          * needed anymore:
 358          */
 359         page = virt_to_page(bdata->node_bootmem_map);
 360         count = 0;
 361         idx = (get_mapsize(bdata) + PAGE_SIZE-1) >> PAGE_SHIFT;
 362         for (i = 0; i < idx; i++, page++) {
 363                 __free_pages_bootmem(page, 0);
 364                 count++;
 365         }
 366         total += count;
 367         bdata->node_bootmem_map = NULL;
 368
 369         return total;
 370 }
 371
 372 unsigned long __init init_bootmem_node(pg_data_t *pgdat, unsigned long freepfn,
 373                                 unsigned long startpfn, unsigned long endpfn)
 374 {
 375         return init_bootmem_core(pgdat, freepfn, startpfn, endpfn);
 376 }
 377
 378 void __init reserve_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
 379                                  unsigned long size)
 380 {
 381         reserve_bootmem_core(pgdat->bdata, physaddr, size);
 382 }
 383
 384 void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
 385                               unsigned long size)
 386 {
 387         free_bootmem_core(pgdat->bdata, physaddr, size);
 388 }
 389
 390 unsigned long __init free_all_bootmem_node(pg_data_t *pgdat)
 391 {
 392         return free_all_bootmem_core(pgdat);
 393 }
 394
 395 unsigned long __init init_bootmem(unsigned long start, unsigned long pages)
 396 {
 397         max_low_pfn = pages;
 398         min_low_pfn = start;
 399         return init_bootmem_core(NODE_DATA(0), start, 0, pages);
 400 }
 401
 402 #ifndef CONFIG_HAVE_ARCH_BOOTMEM_NODE
 403 void __init reserve_bootmem(unsigned long addr, unsigned long size)
 404 {
 405         reserve_bootmem_core(NODE_DATA(0)->bdata, addr, size);
 406 }
 407 #endif /* !CONFIG_HAVE_ARCH_BOOTMEM_NODE */
 408
 409 void __init free_bootmem(unsigned long addr, unsigned long size)
 410 {
 411         free_bootmem_core(NODE_DATA(0)->bdata, addr, size);
 412 }
 413
 414 unsigned long __init free_all_bootmem(void)
 415 {
 416         return free_all_bootmem_core(NODE_DATA(0));
 417 }
 418
 419 void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align,
 420                                       unsigned long goal)
 421 {
 422         bootmem_data_t *bdata;
 423         void *ptr;
 424
 425         list_for_each_entry(bdata, &bdata_list, list) {
 426                 ptr = __alloc_bootmem_core(bdata, size, align, goal, 0);
 427                 if (ptr)
 428                         return ptr;
 429         }
 430         return NULL;
 431 }
 432
 433 void * __init __alloc_bootmem(unsigned long size, unsigned long align,
 434                               unsigned long goal)
 435 {
 436         void *mem = __alloc_bootmem_nopanic(size,align,goal);
 437
 438         if (mem)
 439                 return mem;
 440         /*
 441          * Whoops, we cannot satisfy the allocation request.
 442          */
 443         printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size);
 444         panic("Out of memory");
 445         return NULL;
 446 }
 447
 448
 449 void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size,
 450                                    unsigned long align, unsigned long goal)
 451 {
 452         void *ptr;
 453
 454         ptr = __alloc_bootmem_core(pgdat->bdata, size, align, goal, 0);
 455         if (ptr)
 456                 return ptr;
 457
 458         return __alloc_bootmem(size, align, goal);
 459 }
 460
 461 #ifndef ARCH_LOW_ADDRESS_LIMIT
 462 #define ARCH_LOW_ADDRESS_LIMIT  0xffffffffUL
 463 #endif
 464
 465 void * __init __alloc_bootmem_low(unsigned long size, unsigned long align,
 466                                   unsigned long goal)
 467 {
 468         bootmem_data_t *bdata;
 469         void *ptr;
 470
 471         list_for_each_entry(bdata, &bdata_list, list) {
 472                 ptr = __alloc_bootmem_core(bdata, size, align, goal,
 473                                                 ARCH_LOW_ADDRESS_LIMIT);
 474                 if (ptr)
 475                         return ptr;
 476         }
 477
 478         /*
 479          * Whoops, we cannot satisfy the allocation request.
 480          */
 481         printk(KERN_ALERT "low bootmem alloc of %lu bytes failed!\n", size);
 482         panic("Out of low memory");
 483         return NULL;
 484 }
 485
 486 void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size,
 487                                        unsigned long align, unsigned long goal)
 488 {
 489         return __alloc_bootmem_core(pgdat->bdata, size, align, goal,
 490                                     ARCH_LOW_ADDRESS_LIMIT);
 491 }