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 <asm/dma.h>
  20 #include <asm/io.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 /* return the number of _pages_ that will be allocated for the boot bitmap */
  31 unsigned long __init bootmem_bootmap_pages (unsigned long pages)
  32 {
  33         unsigned long mapsize;
  34
  35         mapsize = (pages+7)/8;
  36         mapsize = (mapsize + ~PAGE_MASK) & PAGE_MASK;
  37         mapsize >>= PAGE_SHIFT;
  38
  39         return mapsize;
  40 }
  41
  42 /*
  43  * Called once to set up the allocator itself.
  44  */
  45 static unsigned long __init init_bootmem_core (pg_data_t *pgdat,
  46         unsigned long mapstart, unsigned long start, unsigned long end)
  47 {
  48         bootmem_data_t *bdata = pgdat->bdata;
  49         unsigned long mapsize = ((end - start)+7)/8;
  50
  51         pgdat->pgdat_next = pgdat_list;
  52         pgdat_list = pgdat;
  53
  54         mapsize = (mapsize + (sizeof(long) - 1UL)) & ~(sizeof(long) - 1UL);
  55         bdata->node_bootmem_map = phys_to_virt(mapstart << PAGE_SHIFT);
  56         bdata->node_boot_start = (start << PAGE_SHIFT);
  57         bdata->node_low_pfn = end;
  58
  59         /*
  60          * Initially all pages are reserved - setup_arch() has to
  61          * register free RAM areas explicitly.
  62          */
  63         memset(bdata->node_bootmem_map, 0xff, mapsize);
  64
  65         return mapsize;
  66 }
  67
  68 /*
  69  * Marks a particular physical memory range as unallocatable. Usable RAM
  70  * might be used for boot-time allocations - or it might get added
  71  * to the free page pool later on.
  72  */
  73 static void __init reserve_bootmem_core(bootmem_data_t *bdata, unsigned long addr, unsigned long size)
  74 {
  75         unsigned long i;
  76         /*
  77          * round up, partially reserved pages are considered
  78          * fully reserved.
  79          */
  80         unsigned long sidx = (addr - bdata->node_boot_start)/PAGE_SIZE;
  81         unsigned long eidx = (addr + size - bdata->node_boot_start +
  82                                                         PAGE_SIZE-1)/PAGE_SIZE;
  83         unsigned long end = (addr + size + PAGE_SIZE-1)/PAGE_SIZE;
  84
  85         if (!size) BUG();
  86
  87         if (sidx >= eidx)
  88                 BUG();
  89         if ((addr >> PAGE_SHIFT) >= bdata->node_low_pfn)
  90                 BUG();
  91         if (end > bdata->node_low_pfn)
  92                 BUG();
  93         for (i = sidx; i < eidx; i++)
  94                 if (test_and_set_bit(i, bdata->node_bootmem_map))
  95                         printk("hm, page %08lx reserved twice.\n", i*PAGE_SIZE);
  96 }
  97
  98 static void __init free_bootmem_core(bootmem_data_t *bdata, unsigned long addr, unsigned long size)
  99 {
 100         unsigned long i;
 101         unsigned long start;
 102         /*
 103          * round down end of usable mem, partially free pages are
 104          * considered reserved.
 105          */
 106         unsigned long sidx;
 107         unsigned long eidx = (addr + size - bdata->node_boot_start)/PAGE_SIZE;
 108         unsigned long end = (addr + size)/PAGE_SIZE;
 109
 110         if (!size) BUG();
 111         if (end > bdata->node_low_pfn)
 112                 BUG();
 113
 114         if (addr < bdata->last_success)
 115                 bdata->last_success = addr;
 116
 117         /*
 118          * Round up the beginning of the address.
 119          */
 120         start = (addr + PAGE_SIZE-1) / PAGE_SIZE;
 121         sidx = start - (bdata->node_boot_start/PAGE_SIZE);
 122
 123         for (i = sidx; i < eidx; i++) {
 124                 if (!test_and_clear_bit(i, bdata->node_bootmem_map))
 125                         BUG();
 126         }
 127 }
 128
 129 /*
 130  * We 'merge' subsequent allocations to save space. We might 'lose'
 131  * some fraction of a page if allocations cannot be satisfied due to
 132  * size constraints on boxes where there is physical RAM space
 133  * fragmentation - in these cases * (mostly large memory boxes) this
 134  * is not a problem.
 135  *
 136  * On low memory boxes we get it right in 100% of the cases.
 137  *
 138  * alignment has to be a power of 2 value.
 139  *
 140  * NOTE:  This function is _not_ reenetrant.
 141  */
 142 static void * __init
 143 __alloc_bootmem_core(struct bootmem_data *bdata, unsigned long size,
 144                 unsigned long align, unsigned long goal)
 145 {
 146         unsigned long offset, remaining_size, areasize, preferred;
 147         unsigned long i, start = 0, incr, eidx;
 148         void *ret;
 149
 150         if(!size) {
 151                 printk("__alloc_bootmem_core(): zero-sized request\n");
 152                 dump_stack();
 153                 BUG();
 154         }
 155         BUG_ON(align & (align-1));
 156
 157         eidx = bdata->node_low_pfn - (bdata->node_boot_start >> PAGE_SHIFT);
 158         offset = 0;
 159         if (align &&
 160             (bdata->node_boot_start & (align - 1UL)) != 0)
 161                 offset = (align - (bdata->node_boot_start & (align - 1UL)));
 162         offset >>= PAGE_SHIFT;
 163
 164         /*
 165          * We try to allocate bootmem pages above 'goal'
 166          * first, then we try to allocate lower pages.
 167          */
 168         if (goal && (goal >= bdata->node_boot_start) &&
 169             ((goal >> PAGE_SHIFT) < bdata->node_low_pfn)) {
 170                 preferred = goal - bdata->node_boot_start;
 171
 172                 if (bdata->last_success >= preferred)
 173                         preferred = bdata->last_success;
 174         } else
 175                 preferred = 0;
 176
 177         preferred = ((preferred + align - 1) & ~(align - 1)) >> PAGE_SHIFT;
 178         preferred += offset;
 179         areasize = (size+PAGE_SIZE-1)/PAGE_SIZE;
 180         incr = align >> PAGE_SHIFT ? : 1;
 181
 182 restart_scan:
 183         for (i = preferred; i < eidx; i += incr) {
 184                 unsigned long j;
 185                 i = find_next_zero_bit(bdata->node_bootmem_map, eidx, i);
 186                 i = (i + incr - 1) & -incr;
 187                 if (test_bit(i, bdata->node_bootmem_map))
 188                         continue;
 189                 for (j = i + 1; j < i + areasize; ++j) {
 190                         if (j >= eidx)
 191                                 goto fail_block;
 192                         if (test_bit (j, bdata->node_bootmem_map))
 193                                 goto fail_block;
 194                 }
 195                 start = i;
 196                 goto found;
 197         fail_block:
 198                 ;
 199         }
 200
 201         if (preferred > offset) {
 202                 preferred = offset;
 203                 goto restart_scan;
 204         }
 205         return NULL;
 206
 207 found:
 208         bdata->last_success = start << PAGE_SHIFT;
 209         BUG_ON(start >= eidx);
 210
 211         /*
 212          * Is the next page of the previous allocation-end the start
 213          * of this allocation's buffer? If yes then we can 'merge'
 214          * the previous partial page with this allocation.
 215          */
 216         if (align < PAGE_SIZE &&
 217             bdata->last_offset && bdata->last_pos+1 == start) {
 218                 offset = (bdata->last_offset+align-1) & ~(align-1);
 219                 BUG_ON(offset > PAGE_SIZE);
 220                 remaining_size = PAGE_SIZE-offset;
 221                 if (size < remaining_size) {
 222                         areasize = 0;
 223                         /* last_pos unchanged */
 224                         bdata->last_offset = offset+size;
 225                         ret = phys_to_virt(bdata->last_pos*PAGE_SIZE + offset +
 226                                                 bdata->node_boot_start);
 227                 } else {
 228                         remaining_size = size - remaining_size;
 229                         areasize = (remaining_size+PAGE_SIZE-1)/PAGE_SIZE;
 230                         ret = phys_to_virt(bdata->last_pos*PAGE_SIZE + offset +
 231                                                 bdata->node_boot_start);
 232                         bdata->last_pos = start+areasize-1;
 233                         bdata->last_offset = remaining_size;
 234                 }
 235                 bdata->last_offset &= ~PAGE_MASK;
 236         } else {
 237                 bdata->last_pos = start + areasize - 1;
 238                 bdata->last_offset = size & ~PAGE_MASK;
 239                 ret = phys_to_virt(start * PAGE_SIZE + bdata->node_boot_start);
 240         }
 241
 242         /*
 243          * Reserve the area now:
 244          */
 245         for (i = start; i < start+areasize; i++)
 246                 if (unlikely(test_and_set_bit(i, bdata->node_bootmem_map)))
 247                         BUG();
 248         memset(ret, 0, size);
 249         return ret;
 250 }
 251
 252 static unsigned long __init free_all_bootmem_core(pg_data_t *pgdat)
 253 {
 254         struct page *page = pgdat->node_mem_map;
 255         bootmem_data_t *bdata = pgdat->bdata;
 256         unsigned long i, count, total = 0;
 257         unsigned long idx;
 258         unsigned long *map;
 259
 260         if (!bdata->node_bootmem_map) BUG();
 261
 262         count = 0;
 263         idx = bdata->node_low_pfn - (bdata->node_boot_start >> PAGE_SHIFT);
 264         map = bdata->node_bootmem_map;
 265         for (i = 0; i < idx; ) {
 266                 unsigned long v = ~map[i / BITS_PER_LONG];
 267                 if (v) {
 268                         unsigned long m;
 269                         for (m = 1; m && i < idx; m<<=1, page++, i++) {
 270                                 if (v & m) {
 271                                         count++;
 272                                         ClearPageReserved(page);
 273                                         set_page_count(page, 1);
 274                                         __free_page(page);
 275                                 }
 276                         }
 277                 } else {
 278                         i+=BITS_PER_LONG;
 279                         page += BITS_PER_LONG;
 280                 }
 281         }
 282         total += count;
 283
 284         /*
 285          * Now free the allocator bitmap itself, it's not
 286          * needed anymore:
 287          */
 288         page = virt_to_page(bdata->node_bootmem_map);
 289         count = 0;
 290         for (i = 0; i < ((bdata->node_low_pfn-(bdata->node_boot_start >> PAGE_SHIFT))/8 + PAGE_SIZE-1)/PAGE_SIZE; i++,page++) {
 291                 count++;
 292                 ClearPageReserved(page);
 293                 set_page_count(page, 1);
 294                 __free_page(page);
 295         }
 296         total += count;
 297         bdata->node_bootmem_map = NULL;
 298
 299         return total;
 300 }
 301
 302 unsigned long __init init_bootmem_node (pg_data_t *pgdat, unsigned long freepfn, unsigned long startpfn, unsigned long endpfn)
 303 {
 304         return(init_bootmem_core(pgdat, freepfn, startpfn, endpfn));
 305 }
 306
 307 void __init reserve_bootmem_node (pg_data_t *pgdat, unsigned long physaddr, unsigned long size)
 308 {
 309         reserve_bootmem_core(pgdat->bdata, physaddr, size);
 310 }
 311
 312 void __init free_bootmem_node (pg_data_t *pgdat, unsigned long physaddr, unsigned long size)
 313 {
 314         free_bootmem_core(pgdat->bdata, physaddr, size);
 315 }
 316
 317 unsigned long __init free_all_bootmem_node (pg_data_t *pgdat)
 318 {
 319         return(free_all_bootmem_core(pgdat));
 320 }
 321
 322 #ifndef CONFIG_DISCONTIGMEM
 323 unsigned long __init init_bootmem (unsigned long start, unsigned long pages)
 324 {
 325         max_low_pfn = pages;
 326         min_low_pfn = start;
 327         return(init_bootmem_core(&contig_page_data, start, 0, pages));
 328 }
 329
 330 #ifndef CONFIG_HAVE_ARCH_BOOTMEM_NODE
 331 void __init reserve_bootmem (unsigned long addr, unsigned long size)
 332 {
 333         reserve_bootmem_core(contig_page_data.bdata, addr, size);
 334 }
 335 #endif /* !CONFIG_HAVE_ARCH_BOOTMEM_NODE */
 336
 337 void __init free_bootmem (unsigned long addr, unsigned long size)
 338 {
 339         free_bootmem_core(contig_page_data.bdata, addr, size);
 340 }
 341
 342 unsigned long __init free_all_bootmem (void)
 343 {
 344         return(free_all_bootmem_core(&contig_page_data));
 345 }
 346 #endif /* !CONFIG_DISCONTIGMEM */
 347
 348 void * __init __alloc_bootmem (unsigned long size, unsigned long align, unsigned long goal)
 349 {
 350         pg_data_t *pgdat = pgdat_list;
 351         void *ptr;
 352
 353         for_each_pgdat(pgdat)
 354                 if ((ptr = __alloc_bootmem_core(pgdat->bdata, size,
 355                                                 align, goal)))
 356                         return(ptr);
 357
 358         /*
 359          * Whoops, we cannot satisfy the allocation request.
 360          */
 361         printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size);
 362         panic("Out of memory");
 363         return NULL;
 364 }
 365
 366 void * __init __alloc_bootmem_node (pg_data_t *pgdat, unsigned long size, unsigned long align, unsigned long goal)
 367 {
 368         void *ptr;
 369
 370         ptr = __alloc_bootmem_core(pgdat->bdata, size, align, goal);
 371         if (ptr)
 372                 return (ptr);
 373
 374         /*
 375          * Whoops, we cannot satisfy the allocation request.
 376          */
 377         printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size);
 378         panic("Out of memory");
 379         return NULL;
 380 }
 381