mm/bootmem.c

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