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         end_pfn = bdata->node_low_pfn;
 200         limit = PFN_DOWN(limit);
 201         if (limit && end_pfn > limit)
 202                 end_pfn = limit;
 203
 204         eidx = end_pfn - PFN_DOWN(bdata->node_boot_start);
 205         offset = 0;
 206         if (align && (bdata->node_boot_start & (align - 1UL)) != 0)
 207                 offset = align - (bdata->node_boot_start & (align - 1UL));
 208         offset = PFN_DOWN(offset);
 209
 210         /*
 211          * We try to allocate bootmem pages above 'goal'
 212          * first, then we try to allocate lower pages.
 213          */
 214         if (goal && goal >= bdata->node_boot_start && PFN_DOWN(goal) < end_pfn) {
 215                 preferred = goal - bdata->node_boot_start;
 216
 217                 if (bdata->last_success >= preferred)
 218                         if (!limit || (limit && limit > bdata->last_success))
 219                                 preferred = bdata->last_success;
 220         } else
 221                 preferred = 0;
 222
 223         preferred = PFN_DOWN(ALIGN(preferred, align)) + offset;
 224         areasize = (size + PAGE_SIZE-1) / PAGE_SIZE;
 225         incr = align >> PAGE_SHIFT ? : 1;
 226
 227 restart_scan:
 228         for (i = preferred; i < eidx; i += incr) {
 229                 unsigned long j;
 230                 i = find_next_zero_bit(bdata->node_bootmem_map, eidx, i);
 231                 i = ALIGN(i, incr);
 232                 if (i >= eidx)
 233                         break;
 234                 if (test_bit(i, bdata->node_bootmem_map))
 235                         continue;
 236                 for (j = i + 1; j < i + areasize; ++j) {
 237                         if (j >= eidx)
 238                                 goto fail_block;
 239                         if (test_bit(j, bdata->node_bootmem_map))
 240                                 goto fail_block;
 241                 }
 242                 start = i;
 243                 goto found;
 244         fail_block:
 245                 i = ALIGN(j, incr);
 246         }
 247
 248         if (preferred > offset) {
 249                 preferred = offset;
 250                 goto restart_scan;
 251         }
 252         return NULL;
 253
 254 found:
 255         bdata->last_success = PFN_PHYS(start);
 256         BUG_ON(start >= eidx);
 257
 258         /*
 259          * Is the next page of the previous allocation-end the start
 260          * of this allocation's buffer? If yes then we can 'merge'
 261          * the previous partial page with this allocation.
 262          */
 263         if (align < PAGE_SIZE &&
 264             bdata->last_offset && bdata->last_pos+1 == start) {
 265                 offset = ALIGN(bdata->last_offset, align);
 266                 BUG_ON(offset > PAGE_SIZE);
 267                 remaining_size = PAGE_SIZE - offset;
 268                 if (size < remaining_size) {
 269                         areasize = 0;
 270                         /* last_pos unchanged */
 271                         bdata->last_offset = offset + size;
 272                         ret = phys_to_virt(bdata->last_pos * PAGE_SIZE +
 273                                            offset +
 274                                            bdata->node_boot_start);
 275                 } else {
 276                         remaining_size = size - remaining_size;
 277                         areasize = (remaining_size + PAGE_SIZE-1) / PAGE_SIZE;
 278                         ret = phys_to_virt(bdata->last_pos * PAGE_SIZE +
 279                                            offset +
 280                                            bdata->node_boot_start);
 281                         bdata->last_pos = start + areasize - 1;
 282                         bdata->last_offset = remaining_size;
 283                 }
 284                 bdata->last_offset &= ~PAGE_MASK;
 285         } else {
 286                 bdata->last_pos = start + areasize - 1;
 287                 bdata->last_offset = size & ~PAGE_MASK;
 288                 ret = phys_to_virt(start * PAGE_SIZE + bdata->node_boot_start);
 289         }
 290
 291         /*
 292          * Reserve the area now:
 293          */
 294         for (i = start; i < start + areasize; i++)
 295                 if (unlikely(test_and_set_bit(i, bdata->node_bootmem_map)))
 296                         BUG();
 297         memset(ret, 0, size);
 298         return ret;
 299 }
 300
 301 static unsigned long __init free_all_bootmem_core(pg_data_t *pgdat)
 302 {
 303         struct page *page;
 304         unsigned long pfn;
 305         bootmem_data_t *bdata = pgdat->bdata;
 306         unsigned long i, count, total = 0;
 307         unsigned long idx;
 308         unsigned long *map;
 309         int gofast = 0;
 310
 311         BUG_ON(!bdata->node_bootmem_map);
 312
 313         count = 0;
 314         /* first extant page of the node */
 315         pfn = PFN_DOWN(bdata->node_boot_start);
 316         idx = bdata->node_low_pfn - pfn;
 317         map = bdata->node_bootmem_map;
 318         /* Check physaddr is O(LOG2(BITS_PER_LONG)) page aligned */
 319         if (bdata->node_boot_start == 0 ||
 320             ffs(bdata->node_boot_start) - PAGE_SHIFT > ffs(BITS_PER_LONG))
 321                 gofast = 1;
 322         for (i = 0; i < idx; ) {
 323                 unsigned long v = ~map[i / BITS_PER_LONG];
 324
 325                 if (gofast && v == ~0UL) {
 326                         int order;
 327
 328                         page = pfn_to_page(pfn);
 329                         count += BITS_PER_LONG;
 330                         order = ffs(BITS_PER_LONG) - 1;
 331                         __free_pages_bootmem(page, order);
 332                         i += BITS_PER_LONG;
 333                         page += BITS_PER_LONG;
 334                 } else if (v) {
 335                         unsigned long m;
 336
 337                         page = pfn_to_page(pfn);
 338                         for (m = 1; m && i < idx; m<<=1, page++, i++) {
 339                                 if (v & m) {
 340                                         count++;
 341                                         __free_pages_bootmem(page, 0);
 342                                 }
 343                         }
 344                 } else {
 345                         i += BITS_PER_LONG;
 346                 }
 347                 pfn += BITS_PER_LONG;
 348         }
 349         total += count;
 350
 351         /*
 352          * Now free the allocator bitmap itself, it's not
 353          * needed anymore:
 354          */
 355         page = virt_to_page(bdata->node_bootmem_map);
 356         count = 0;
 357         idx = (get_mapsize(bdata) + PAGE_SIZE-1) >> PAGE_SHIFT;
 358         for (i = 0; i < idx; i++, page++) {
 359                 __free_pages_bootmem(page, 0);
 360                 count++;
 361         }
 362         total += count;
 363         bdata->node_bootmem_map = NULL;
 364
 365         return total;
 366 }
 367
 368 unsigned long __init init_bootmem_node(pg_data_t *pgdat, unsigned long freepfn,
 369                                 unsigned long startpfn, unsigned long endpfn)
 370 {
 371         return init_bootmem_core(pgdat, freepfn, startpfn, endpfn);
 372 }
 373
 374 void __init reserve_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
 375                                  unsigned long size)
 376 {
 377         reserve_bootmem_core(pgdat->bdata, physaddr, size);
 378 }
 379
 380 void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
 381                               unsigned long size)
 382 {
 383         free_bootmem_core(pgdat->bdata, physaddr, size);
 384 }
 385
 386 unsigned long __init free_all_bootmem_node(pg_data_t *pgdat)
 387 {
 388         return free_all_bootmem_core(pgdat);
 389 }
 390
 391 unsigned long __init init_bootmem(unsigned long start, unsigned long pages)
 392 {
 393         max_low_pfn = pages;
 394         min_low_pfn = start;
 395         return init_bootmem_core(NODE_DATA(0), start, 0, pages);
 396 }
 397
 398 #ifndef CONFIG_HAVE_ARCH_BOOTMEM_NODE
 399 void __init reserve_bootmem(unsigned long addr, unsigned long size)
 400 {
 401         reserve_bootmem_core(NODE_DATA(0)->bdata, addr, size);
 402 }
 403 #endif /* !CONFIG_HAVE_ARCH_BOOTMEM_NODE */
 404
 405 void __init free_bootmem(unsigned long addr, unsigned long size)
 406 {
 407         free_bootmem_core(NODE_DATA(0)->bdata, addr, size);
 408 }
 409
 410 unsigned long __init free_all_bootmem(void)
 411 {
 412         return free_all_bootmem_core(NODE_DATA(0));
 413 }
 414
 415 void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align,
 416                                       unsigned long goal)
 417 {
 418         bootmem_data_t *bdata;
 419         void *ptr;
 420
 421         list_for_each_entry(bdata, &bdata_list, list) {
 422                 ptr = __alloc_bootmem_core(bdata, size, align, goal, 0);
 423                 if (ptr)
 424                         return ptr;
 425         }
 426         return NULL;
 427 }
 428
 429 void * __init __alloc_bootmem(unsigned long size, unsigned long align,
 430                               unsigned long goal)
 431 {
 432         void *mem = __alloc_bootmem_nopanic(size,align,goal);
 433
 434         if (mem)
 435                 return mem;
 436         /*
 437          * Whoops, we cannot satisfy the allocation request.
 438          */
 439         printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size);
 440         panic("Out of memory");
 441         return NULL;
 442 }
 443
 444
 445 void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size,
 446                                    unsigned long align, unsigned long goal)
 447 {
 448         void *ptr;
 449
 450         ptr = __alloc_bootmem_core(pgdat->bdata, size, align, goal, 0);
 451         if (ptr)
 452                 return ptr;
 453
 454         return __alloc_bootmem(size, align, goal);
 455 }
 456
 457 #ifndef ARCH_LOW_ADDRESS_LIMIT
 458 #define ARCH_LOW_ADDRESS_LIMIT  0xffffffffUL
 459 #endif
 460
 461 void * __init __alloc_bootmem_low(unsigned long size, unsigned long align,
 462                                   unsigned long goal)
 463 {
 464         bootmem_data_t *bdata;
 465         void *ptr;
 466
 467         list_for_each_entry(bdata, &bdata_list, list) {
 468                 ptr = __alloc_bootmem_core(bdata, size, align, goal,
 469                                                 ARCH_LOW_ADDRESS_LIMIT);
 470                 if (ptr)
 471                         return ptr;
 472         }
 473
 474         /*
 475          * Whoops, we cannot satisfy the allocation request.
 476          */
 477         printk(KERN_ALERT "low bootmem alloc of %lu bytes failed!\n", size);
 478         panic("Out of low memory");
 479         return NULL;
 480 }
 481
 482 void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size,
 483                                        unsigned long align, unsigned long goal)
 484 {
 485         return __alloc_bootmem_core(pgdat->bdata, size, align, goal,
 486                                     ARCH_LOW_ADDRESS_LIMIT);
 487 }