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 #ifdef CONFIG_DEBUG_BOOTMEM
  96                         printk("hm, page %08lx reserved twice.\n", i*PAGE_SIZE);
  97 #endif
  98                 }
  99 }
 100
 101 static void __init free_bootmem_core(bootmem_data_t *bdata, unsigned long addr, unsigned long size)
 102 {
 103         unsigned long i;
 104         unsigned long start;
 105         /*
 106          * round down end of usable mem, partially free pages are
 107          * considered reserved.
 108          */
 109         unsigned long sidx;
 110         unsigned long eidx = (addr + size - bdata->node_boot_start)/PAGE_SIZE;
 111         unsigned long end = (addr + size)/PAGE_SIZE;
 112
 113         if (!size) BUG();
 114         if (end > bdata->node_low_pfn)
 115                 BUG();
 116
 117         if (addr < bdata->last_success)
 118                 bdata->last_success = addr;
 119
 120         /*
 121          * Round up the beginning of the address.
 122          */
 123         start = (addr + PAGE_SIZE-1) / PAGE_SIZE;
 124         sidx = start - (bdata->node_boot_start/PAGE_SIZE);
 125
 126         for (i = sidx; i < eidx; i++) {
 127                 if (!test_and_clear_bit(i, bdata->node_bootmem_map))
 128                         BUG();
 129         }
 130 }
 131
 132 /*
 133  * We 'merge' subsequent allocations to save space. We might 'lose'
 134  * some fraction of a page if allocations cannot be satisfied due to
 135  * size constraints on boxes where there is physical RAM space
 136  * fragmentation - in these cases (mostly large memory boxes) this
 137  * is not a problem.
 138  *
 139  * On low memory boxes we get it right in 100% of the cases.
 140  *
 141  * alignment has to be a power of 2 value.
 142  *
 143  * NOTE:  This function is _not_ reenetrant.
 144  */
 145 static void * __init
 146 __alloc_bootmem_core(struct bootmem_data *bdata, unsigned long size,
 147                 unsigned long align, unsigned long goal)
 148 {
 149         unsigned long offset, remaining_size, areasize, preferred;
 150         unsigned long i, start = 0, incr, eidx;
 151         void *ret;
 152
 153         if(!size) {
 154                 printk("__alloc_bootmem_core(): zero-sized request\n");
 155                 dump_stack();
 156                 BUG();
 157         }
 158         BUG_ON(align & (align-1));
 159
 160         eidx = bdata->node_low_pfn - (bdata->node_boot_start >> PAGE_SHIFT);
 161         offset = 0;
 162         if (align &&
 163             (bdata->node_boot_start & (align - 1UL)) != 0)
 164                 offset = (align - (bdata->node_boot_start & (align - 1UL)));
 165         offset >>= PAGE_SHIFT;
 166
 167         /*
 168          * We try to allocate bootmem pages above 'goal'
 169          * first, then we try to allocate lower pages.
 170          */
 171         if (goal && (goal >= bdata->node_boot_start) &&
 172             ((goal >> PAGE_SHIFT) < bdata->node_low_pfn)) {
 173                 preferred = goal - bdata->node_boot_start;
 174
 175                 if (bdata->last_success >= preferred)
 176                         preferred = bdata->last_success;
 177         } else
 178                 preferred = 0;
 179
 180         preferred = ((preferred + align - 1) & ~(align - 1)) >> PAGE_SHIFT;
 181         preferred += offset;
 182         areasize = (size+PAGE_SIZE-1)/PAGE_SIZE;
 183         incr = align >> PAGE_SHIFT ? : 1;
 184
 185 restart_scan:
 186         for (i = preferred; i < eidx; i += incr) {
 187                 unsigned long j;
 188                 i = find_next_zero_bit(bdata->node_bootmem_map, eidx, i);
 189                 i = ALIGN(i, incr);
 190                 if (test_bit(i, bdata->node_bootmem_map))
 191                         continue;
 192                 for (j = i + 1; j < i + areasize; ++j) {
 193                         if (j >= eidx)
 194                                 goto fail_block;
 195                         if (test_bit (j, bdata->node_bootmem_map))
 196                                 goto fail_block;
 197                 }
 198                 start = i;
 199                 goto found;
 200         fail_block:
 201                 i = ALIGN(j, incr);
 202         }
 203
 204         if (preferred > offset) {
 205                 preferred = offset;
 206                 goto restart_scan;
 207         }
 208         return NULL;
 209
 210 found:
 211         bdata->last_success = start << PAGE_SHIFT;
 212         BUG_ON(start >= eidx);
 213
 214         /*
 215          * Is the next page of the previous allocation-end the start
 216          * of this allocation's buffer? If yes then we can 'merge'
 217          * the previous partial page with this allocation.
 218          */
 219         if (align < PAGE_SIZE &&
 220             bdata->last_offset && bdata->last_pos+1 == start) {
 221                 offset = (bdata->last_offset+align-1) & ~(align-1);
 222                 BUG_ON(offset > PAGE_SIZE);
 223                 remaining_size = PAGE_SIZE-offset;
 224                 if (size < remaining_size) {
 225                         areasize = 0;
 226                         /* last_pos unchanged */
 227                         bdata->last_offset = offset+size;
 228                         ret = phys_to_virt(bdata->last_pos*PAGE_SIZE + offset +
 229                                                 bdata->node_boot_start);
 230                 } else {
 231                         remaining_size = size - remaining_size;
 232                         areasize = (remaining_size+PAGE_SIZE-1)/PAGE_SIZE;
 233                         ret = phys_to_virt(bdata->last_pos*PAGE_SIZE + offset +
 234                                                 bdata->node_boot_start);
 235                         bdata->last_pos = start+areasize-1;
 236                         bdata->last_offset = remaining_size;
 237                 }
 238                 bdata->last_offset &= ~PAGE_MASK;
 239         } else {
 240                 bdata->last_pos = start + areasize - 1;
 241                 bdata->last_offset = size & ~PAGE_MASK;
 242                 ret = phys_to_virt(start * PAGE_SIZE + bdata->node_boot_start);
 243         }
 244
 245         /*
 246          * Reserve the area now:
 247          */
 248         for (i = start; i < start+areasize; i++)
 249                 if (unlikely(test_and_set_bit(i, bdata->node_bootmem_map)))
 250                         BUG();
 251         memset(ret, 0, size);
 252         return ret;
 253 }
 254
 255 static unsigned long __init free_all_bootmem_core(pg_data_t *pgdat)
 256 {
 257         struct page *page;
 258         bootmem_data_t *bdata = pgdat->bdata;
 259         unsigned long i, count, total = 0;
 260         unsigned long idx;
 261         unsigned long *map;
 262
 263         if (!bdata->node_bootmem_map) BUG();
 264
 265         count = 0;
 266         /* first extant page of the node */
 267         page = virt_to_page(phys_to_virt(bdata->node_boot_start));
 268         idx = bdata->node_low_pfn - (bdata->node_boot_start >> PAGE_SHIFT);
 269         map = bdata->node_bootmem_map;
 270         for (i = 0; i < idx; ) {
 271                 unsigned long v = ~map[i / BITS_PER_LONG];
 272                 if (v) {
 273                         unsigned long m;
 274                         for (m = 1; m && i < idx; m<<=1, page++, i++) {
 275                                 if (v & m) {
 276                                         count++;
 277                                         ClearPageReserved(page);
 278                                         set_page_count(page, 1);
 279                                         __free_page(page);
 280                                 }
 281                         }
 282                 } else {
 283                         i+=BITS_PER_LONG;
 284                         page += BITS_PER_LONG;
 285                 }
 286         }
 287         total += count;
 288
 289         /*
 290          * Now free the allocator bitmap itself, it's not
 291          * needed anymore:
 292          */
 293         page = virt_to_page(bdata->node_bootmem_map);
 294         count = 0;
 295         for (i = 0; i < ((bdata->node_low_pfn-(bdata->node_boot_start >> PAGE_SHIFT))/8 + PAGE_SIZE-1)/PAGE_SIZE; i++,page++) {
 296                 count++;
 297                 ClearPageReserved(page);
 298                 set_page_count(page, 1);
 299                 __free_page(page);
 300         }
 301         total += count;
 302         bdata->node_bootmem_map = NULL;
 303
 304         return total;
 305 }
 306
 307 unsigned long __init init_bootmem_node (pg_data_t *pgdat, unsigned long freepfn, unsigned long startpfn, unsigned long endpfn)
 308 {
 309         return(init_bootmem_core(pgdat, freepfn, startpfn, endpfn));
 310 }
 311
 312 void __init reserve_bootmem_node (pg_data_t *pgdat, unsigned long physaddr, unsigned long size)
 313 {
 314         reserve_bootmem_core(pgdat->bdata, physaddr, size);
 315 }
 316
 317 void __init free_bootmem_node (pg_data_t *pgdat, unsigned long physaddr, unsigned long size)
 318 {
 319         free_bootmem_core(pgdat->bdata, physaddr, size);
 320 }
 321
 322 unsigned long __init free_all_bootmem_node (pg_data_t *pgdat)
 323 {
 324         return(free_all_bootmem_core(pgdat));
 325 }
 326
 327 #ifndef CONFIG_DISCONTIGMEM
 328 unsigned long __init init_bootmem (unsigned long start, unsigned long pages)
 329 {
 330         max_low_pfn = pages;
 331         min_low_pfn = start;
 332         return(init_bootmem_core(&contig_page_data, start, 0, pages));
 333 }
 334
 335 #ifndef CONFIG_HAVE_ARCH_BOOTMEM_NODE
 336 void __init reserve_bootmem (unsigned long addr, unsigned long size)
 337 {
 338         reserve_bootmem_core(contig_page_data.bdata, addr, size);
 339 }
 340 #endif /* !CONFIG_HAVE_ARCH_BOOTMEM_NODE */
 341
 342 void __init free_bootmem (unsigned long addr, unsigned long size)
 343 {
 344         free_bootmem_core(contig_page_data.bdata, addr, size);
 345 }
 346
 347 unsigned long __init free_all_bootmem (void)
 348 {
 349         return(free_all_bootmem_core(&contig_page_data));
 350 }
 351 #endif /* !CONFIG_DISCONTIGMEM */
 352
 353 void * __init __alloc_bootmem (unsigned long size, unsigned long align, unsigned long goal)
 354 {
 355         pg_data_t *pgdat = pgdat_list;
 356         void *ptr;
 357
 358         for_each_pgdat(pgdat)
 359                 if ((ptr = __alloc_bootmem_core(pgdat->bdata, size,
 360                                                 align, goal)))
 361                         return(ptr);
 362
 363         /*
 364          * Whoops, we cannot satisfy the allocation request.
 365          */
 366         printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size);
 367         panic("Out of memory");
 368         return NULL;
 369 }
 370
 371 void * __init __alloc_bootmem_node (pg_data_t *pgdat, unsigned long size, unsigned long align, unsigned long goal)
 372 {
 373         void *ptr;
 374
 375         ptr = __alloc_bootmem_core(pgdat->bdata, size, align, goal);
 376         if (ptr)
 377                 return (ptr);
 378
 379         /*
 380          * Whoops, we cannot satisfy the allocation request.
 381          */
 382         printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size);
 383         panic("Out of memory");
 384         return NULL;
 385 }
 386