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
  12 #include <linux/mm.h>
  13 #include <linux/kernel_stat.h>
  14 #include <linux/swap.h>
  15 #include <linux/interrupt.h>
  16 #include <linux/init.h>
  17 #include <linux/bootmem.h>
  18 #include <linux/mmzone.h>
  19 #include <linux/module.h>
  20 #include <asm/dma.h>
  21 #include <asm/io.h>
  22 #include "internal.h"
  23
  24 /*
  25  * Access to this subsystem has to be serialized externally. (this is
  26  * true for the boot process anyway)
  27  */
  28 unsigned long max_low_pfn;
  29 unsigned long min_low_pfn;
  30 unsigned long max_pfn;
  31
  32 EXPORT_SYMBOL(max_pfn);         /* This is exported so
  33                                  * dma_get_required_mask(), which uses
  34                                  * it, can be an inline function */
  35
  36 static LIST_HEAD(bdata_list);
  37 #ifdef CONFIG_CRASH_DUMP
  38 /*
  39  * If we have booted due to a crash, max_pfn will be a very low value. We need
  40  * to know the amount of memory that the previous kernel used.
  41  */
  42 unsigned long saved_max_pfn;
  43 #endif
  44
  45 /* return the number of _pages_ that will be allocated for the boot bitmap */
  46 unsigned long __init bootmem_bootmap_pages (unsigned long pages)
  47 {
  48         unsigned long mapsize;
  49
  50         mapsize = (pages+7)/8;
  51         mapsize = (mapsize + ~PAGE_MASK) & PAGE_MASK;
  52         mapsize >>= PAGE_SHIFT;
  53
  54         return mapsize;
  55 }
  56 /*
  57  * link bdata in order
  58  */
  59 static void link_bootmem(bootmem_data_t *bdata)
  60 {
  61         bootmem_data_t *ent;
  62         if (list_empty(&bdata_list)) {
  63                 list_add(&bdata->list, &bdata_list);
  64                 return;
  65         }
  66         /* insert in order */
  67         list_for_each_entry(ent, &bdata_list, list) {
  68                 if (bdata->node_boot_start < ent->node_boot_start) {
  69                         list_add_tail(&bdata->list, &ent->list);
  70                         return;
  71                 }
  72         }
  73         list_add_tail(&bdata->list, &bdata_list);
  74         return;
  75 }
  76
  77
  78 /*
  79  * Called once to set up the allocator itself.
  80  */
  81 static unsigned long __init init_bootmem_core (pg_data_t *pgdat,
  82         unsigned long mapstart, unsigned long start, unsigned long end)
  83 {
  84         bootmem_data_t *bdata = pgdat->bdata;
  85         unsigned long mapsize = ((end - start)+7)/8;
  86
  87         mapsize = ALIGN(mapsize, sizeof(long));
  88         bdata->node_bootmem_map = phys_to_virt(mapstart << PAGE_SHIFT);
  89         bdata->node_boot_start = (start << PAGE_SHIFT);
  90         bdata->node_low_pfn = end;
  91         link_bootmem(bdata);
  92
  93         /*
  94          * Initially all pages are reserved - setup_arch() has to
  95          * register free RAM areas explicitly.
  96          */
  97         memset(bdata->node_bootmem_map, 0xff, mapsize);
  98
  99         return mapsize;
 100 }
 101
 102 /*
 103  * Marks a particular physical memory range as unallocatable. Usable RAM
 104  * might be used for boot-time allocations - or it might get added
 105  * to the free page pool later on.
 106  */
 107 static void __init reserve_bootmem_core(bootmem_data_t *bdata, unsigned long addr, unsigned long size)
 108 {
 109         unsigned long i;
 110         /*
 111          * round up, partially reserved pages are considered
 112          * fully reserved.
 113          */
 114         unsigned long sidx = (addr - bdata->node_boot_start)/PAGE_SIZE;
 115         unsigned long eidx = (addr + size - bdata->node_boot_start +
 116                                                         PAGE_SIZE-1)/PAGE_SIZE;
 117         unsigned long end = (addr + size + PAGE_SIZE-1)/PAGE_SIZE;
 118
 119         BUG_ON(!size);
 120         BUG_ON(sidx >= eidx);
 121         BUG_ON((addr >> PAGE_SHIFT) >= bdata->node_low_pfn);
 122         BUG_ON(end > bdata->node_low_pfn);
 123
 124         for (i = sidx; i < eidx; i++)
 125                 if (test_and_set_bit(i, bdata->node_bootmem_map)) {
 126 #ifdef CONFIG_DEBUG_BOOTMEM
 127                         printk("hm, page %08lx reserved twice.\n", i*PAGE_SIZE);
 128 #endif
 129                 }
 130 }
 131
 132 static void __init free_bootmem_core(bootmem_data_t *bdata, unsigned long addr, unsigned long size)
 133 {
 134         unsigned long i;
 135         unsigned long start;
 136         /*
 137          * round down end of usable mem, partially free pages are
 138          * considered reserved.
 139          */
 140         unsigned long sidx;
 141         unsigned long eidx = (addr + size - bdata->node_boot_start)/PAGE_SIZE;
 142         unsigned long end = (addr + size)/PAGE_SIZE;
 143
 144         BUG_ON(!size);
 145         BUG_ON(end > bdata->node_low_pfn);
 146
 147         if (addr < bdata->last_success)
 148                 bdata->last_success = addr;
 149
 150         /*
 151          * Round up the beginning of the address.
 152          */
 153         start = (addr + PAGE_SIZE-1) / PAGE_SIZE;
 154         sidx = start - (bdata->node_boot_start/PAGE_SIZE);
 155
 156         for (i = sidx; i < eidx; i++) {
 157                 if (unlikely(!test_and_clear_bit(i, bdata->node_bootmem_map)))
 158                         BUG();
 159         }
 160 }
 161
 162 /*
 163  * We 'merge' subsequent allocations to save space. We might 'lose'
 164  * some fraction of a page if allocations cannot be satisfied due to
 165  * size constraints on boxes where there is physical RAM space
 166  * fragmentation - in these cases (mostly large memory boxes) this
 167  * is not a problem.
 168  *
 169  * On low memory boxes we get it right in 100% of the cases.
 170  *
 171  * alignment has to be a power of 2 value.
 172  *
 173  * NOTE:  This function is _not_ reentrant.
 174  */
 175 void * __init
 176 __alloc_bootmem_core(struct bootmem_data *bdata, unsigned long size,
 177               unsigned long align, unsigned long goal, unsigned long limit)
 178 {
 179         unsigned long offset, remaining_size, areasize, preferred;
 180         unsigned long i, start = 0, incr, eidx, end_pfn = bdata->node_low_pfn;
 181         void *ret;
 182
 183         if(!size) {
 184                 printk("__alloc_bootmem_core(): zero-sized request\n");
 185                 BUG();
 186         }
 187         BUG_ON(align & (align-1));
 188
 189         if (limit && bdata->node_boot_start >= limit)
 190                 return NULL;
 191
 192         limit >>=PAGE_SHIFT;
 193         if (limit && end_pfn > limit)
 194                 end_pfn = limit;
 195
 196         eidx = end_pfn - (bdata->node_boot_start >> PAGE_SHIFT);
 197         offset = 0;
 198         if (align &&
 199             (bdata->node_boot_start & (align - 1UL)) != 0)
 200                 offset = (align - (bdata->node_boot_start & (align - 1UL)));
 201         offset >>= PAGE_SHIFT;
 202
 203         /*
 204          * We try to allocate bootmem pages above 'goal'
 205          * first, then we try to allocate lower pages.
 206          */
 207         if (goal && (goal >= bdata->node_boot_start) &&
 208             ((goal >> PAGE_SHIFT) < end_pfn)) {
 209                 preferred = goal - bdata->node_boot_start;
 210
 211                 if (bdata->last_success >= preferred)
 212                         if (!limit || (limit && limit > bdata->last_success))
 213                                 preferred = bdata->last_success;
 214         } else
 215                 preferred = 0;
 216
 217         preferred = ALIGN(preferred, align) >> PAGE_SHIFT;
 218         preferred += offset;
 219         areasize = (size+PAGE_SIZE-1)/PAGE_SIZE;
 220         incr = align >> PAGE_SHIFT ? : 1;
 221
 222 restart_scan:
 223         for (i = preferred; i < eidx; i += incr) {
 224                 unsigned long j;
 225                 i = find_next_zero_bit(bdata->node_bootmem_map, eidx, i);
 226                 i = ALIGN(i, incr);
 227                 if (i >= eidx)
 228                         break;
 229                 if (test_bit(i, bdata->node_bootmem_map))
 230                         continue;
 231                 for (j = i + 1; j < i + areasize; ++j) {
 232                         if (j >= eidx)
 233                                 goto fail_block;
 234                         if (test_bit (j, bdata->node_bootmem_map))
 235                                 goto fail_block;
 236                 }
 237                 start = i;
 238                 goto found;
 239         fail_block:
 240                 i = ALIGN(j, incr);
 241         }
 242
 243         if (preferred > offset) {
 244                 preferred = offset;
 245                 goto restart_scan;
 246         }
 247         return NULL;
 248
 249 found:
 250         bdata->last_success = start << PAGE_SHIFT;
 251         BUG_ON(start >= eidx);
 252
 253         /*
 254          * Is the next page of the previous allocation-end the start
 255          * of this allocation's buffer? If yes then we can 'merge'
 256          * the previous partial page with this allocation.
 257          */
 258         if (align < PAGE_SIZE &&
 259             bdata->last_offset && bdata->last_pos+1 == start) {
 260                 offset = ALIGN(bdata->last_offset, align);
 261                 BUG_ON(offset > PAGE_SIZE);
 262                 remaining_size = PAGE_SIZE-offset;
 263                 if (size < remaining_size) {
 264                         areasize = 0;
 265                         /* last_pos unchanged */
 266                         bdata->last_offset = offset+size;
 267                         ret = phys_to_virt(bdata->last_pos*PAGE_SIZE + offset +
 268                                                 bdata->node_boot_start);
 269                 } else {
 270                         remaining_size = size - remaining_size;
 271                         areasize = (remaining_size+PAGE_SIZE-1)/PAGE_SIZE;
 272                         ret = phys_to_virt(bdata->last_pos*PAGE_SIZE + offset +
 273                                                 bdata->node_boot_start);
 274                         bdata->last_pos = start+areasize-1;
 275                         bdata->last_offset = remaining_size;
 276                 }
 277                 bdata->last_offset &= ~PAGE_MASK;
 278         } else {
 279                 bdata->last_pos = start + areasize - 1;
 280                 bdata->last_offset = size & ~PAGE_MASK;
 281                 ret = phys_to_virt(start * PAGE_SIZE + bdata->node_boot_start);
 282         }
 283
 284         /*
 285          * Reserve the area now:
 286          */
 287         for (i = start; i < start+areasize; i++)
 288                 if (unlikely(test_and_set_bit(i, bdata->node_bootmem_map)))
 289                         BUG();
 290         memset(ret, 0, size);
 291         return ret;
 292 }
 293
 294 static unsigned long __init free_all_bootmem_core(pg_data_t *pgdat)
 295 {
 296         struct page *page;
 297         unsigned long pfn;
 298         bootmem_data_t *bdata = pgdat->bdata;
 299         unsigned long i, count, total = 0;
 300         unsigned long idx;
 301         unsigned long *map;
 302         int gofast = 0;
 303
 304         BUG_ON(!bdata->node_bootmem_map);
 305
 306         count = 0;
 307         /* first extant page of the node */
 308         pfn = bdata->node_boot_start >> PAGE_SHIFT;
 309         idx = bdata->node_low_pfn - (bdata->node_boot_start >> PAGE_SHIFT);
 310         map = bdata->node_bootmem_map;
 311         /* Check physaddr is O(LOG2(BITS_PER_LONG)) page aligned */
 312         if (bdata->node_boot_start == 0 ||
 313             ffs(bdata->node_boot_start) - PAGE_SHIFT > ffs(BITS_PER_LONG))
 314                 gofast = 1;
 315         for (i = 0; i < idx; ) {
 316                 unsigned long v = ~map[i / BITS_PER_LONG];
 317
 318                 if (gofast && v == ~0UL) {
 319                         int order;
 320
 321                         page = pfn_to_page(pfn);
 322                         count += BITS_PER_LONG;
 323                         order = ffs(BITS_PER_LONG) - 1;
 324                         __free_pages_bootmem(page, order);
 325                         i += BITS_PER_LONG;
 326                         page += BITS_PER_LONG;
 327                 } else if (v) {
 328                         unsigned long m;
 329
 330                         page = pfn_to_page(pfn);
 331                         for (m = 1; m && i < idx; m<<=1, page++, i++) {
 332                                 if (v & m) {
 333                                         count++;
 334                                         __free_pages_bootmem(page, 0);
 335                                 }
 336                         }
 337                 } else {
 338                         i+=BITS_PER_LONG;
 339                 }
 340                 pfn += BITS_PER_LONG;
 341         }
 342         total += count;
 343
 344         /*
 345          * Now free the allocator bitmap itself, it's not
 346          * needed anymore:
 347          */
 348         page = virt_to_page(bdata->node_bootmem_map);
 349         count = 0;
 350         for (i = 0; i < ((bdata->node_low_pfn-(bdata->node_boot_start >> PAGE_SHIFT))/8 + PAGE_SIZE-1)/PAGE_SIZE; i++,page++) {
 351                 count++;
 352                 __free_pages_bootmem(page, 0);
 353         }
 354         total += count;
 355         bdata->node_bootmem_map = NULL;
 356
 357         return total;
 358 }
 359
 360 unsigned long __init init_bootmem_node (pg_data_t *pgdat, unsigned long freepfn, unsigned long startpfn, unsigned long endpfn)
 361 {
 362         return(init_bootmem_core(pgdat, freepfn, startpfn, endpfn));
 363 }
 364
 365 void __init reserve_bootmem_node (pg_data_t *pgdat, unsigned long physaddr, unsigned long size)
 366 {
 367         reserve_bootmem_core(pgdat->bdata, physaddr, size);
 368 }
 369
 370 void __init free_bootmem_node (pg_data_t *pgdat, unsigned long physaddr, unsigned long size)
 371 {
 372         free_bootmem_core(pgdat->bdata, physaddr, size);
 373 }
 374
 375 unsigned long __init free_all_bootmem_node (pg_data_t *pgdat)
 376 {
 377         return(free_all_bootmem_core(pgdat));
 378 }
 379
 380 unsigned long __init init_bootmem (unsigned long start, unsigned long pages)
 381 {
 382         max_low_pfn = pages;
 383         min_low_pfn = start;
 384         return(init_bootmem_core(NODE_DATA(0), start, 0, pages));
 385 }
 386
 387 #ifndef CONFIG_HAVE_ARCH_BOOTMEM_NODE
 388 void __init reserve_bootmem (unsigned long addr, unsigned long size)
 389 {
 390         reserve_bootmem_core(NODE_DATA(0)->bdata, addr, size);
 391 }
 392 #endif /* !CONFIG_HAVE_ARCH_BOOTMEM_NODE */
 393
 394 void __init free_bootmem (unsigned long addr, unsigned long size)
 395 {
 396         free_bootmem_core(NODE_DATA(0)->bdata, addr, size);
 397 }
 398
 399 unsigned long __init free_all_bootmem (void)
 400 {
 401         return(free_all_bootmem_core(NODE_DATA(0)));
 402 }
 403
 404 void * __init __alloc_bootmem(unsigned long size, unsigned long align, unsigned long goal)
 405 {
 406         bootmem_data_t *bdata;
 407         void *ptr;
 408
 409         list_for_each_entry(bdata, &bdata_list, list)
 410                 if ((ptr = __alloc_bootmem_core(bdata, size, align, goal, 0)))
 411                         return(ptr);
 412
 413         /*
 414          * Whoops, we cannot satisfy the allocation request.
 415          */
 416         printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size);
 417         panic("Out of memory");
 418         return NULL;
 419 }
 420
 421
 422 void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size, unsigned long align,
 423                                    unsigned long goal)
 424 {
 425         void *ptr;
 426
 427         ptr = __alloc_bootmem_core(pgdat->bdata, size, align, goal, 0);
 428         if (ptr)
 429                 return (ptr);
 430
 431         return __alloc_bootmem(size, align, goal);
 432 }
 433
 434 #define LOW32LIMIT 0xffffffff
 435
 436 void * __init __alloc_bootmem_low(unsigned long size, unsigned long align, unsigned long goal)
 437 {
 438         bootmem_data_t *bdata;
 439         void *ptr;
 440
 441         list_for_each_entry(bdata, &bdata_list, list)
 442                 if ((ptr = __alloc_bootmem_core(bdata, size,
 443                                                  align, goal, LOW32LIMIT)))
 444                         return(ptr);
 445
 446         /*
 447          * Whoops, we cannot satisfy the allocation request.
 448          */
 449         printk(KERN_ALERT "low bootmem alloc of %lu bytes failed!\n", size);
 450         panic("Out of low memory");
 451         return NULL;
 452 }
 453
 454 void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size,
 455                                        unsigned long align, unsigned long goal)
 456 {
 457         return __alloc_bootmem_core(pgdat->bdata, size, align, goal, LOW32LIMIT);
 458 }