mm/nobootmem.c

   1 /*
   2  *  bootmem - A boot-time physical memory allocator and configurator
   3  *
   4  *  Copyright (C) 1999 Ingo Molnar
   5  *                1999 Kanoj Sarcar, SGI
   6  *                2008 Johannes Weiner
   7  *
   8  * Access to this subsystem has to be serialized externally (which is true
   9  * for the boot process anyway).
  10  */
  11 #include <linux/init.h>
  12 #include <linux/pfn.h>
  13 #include <linux/slab.h>
  14 #include <linux/bootmem.h>
  15 #include <linux/export.h>
  16 #include <linux/kmemleak.h>
  17 #include <linux/range.h>
  18 #include <linux/memblock.h>
  19
  20 #include <asm/bug.h>
  21 #include <asm/io.h>
  22 #include <asm/processor.h>
  23
  24 #include "internal.h"
  25
  26 #ifndef CONFIG_NEED_MULTIPLE_NODES
  27 struct pglist_data __refdata contig_page_data;
  28 EXPORT_SYMBOL(contig_page_data);
  29 #endif
  30
  31 unsigned long max_low_pfn;
  32 unsigned long min_low_pfn;
  33 unsigned long max_pfn;
  34
  35 static void * __init __alloc_memory_core_early(int nid, u64 size, u64 align,
  36                                         u64 goal, u64 limit)
  37 {
  38         void *ptr;
  39         u64 addr;
  40
  41         if (limit > memblock.current_limit)
  42                 limit = memblock.current_limit;
  43
  44         addr = find_memory_core_early(nid, size, align, goal, limit);
  45
  46         if (addr == MEMBLOCK_ERROR)
  47                 return NULL;
  48
  49         ptr = phys_to_virt(addr);
  50         memset(ptr, 0, size);
  51         memblock_x86_reserve_range(addr, addr + size, "BOOTMEM");
  52         /*
  53          * The min_count is set to 0 so that bootmem allocated blocks
  54          * are never reported as leaks.
  55          */
  56         kmemleak_alloc(ptr, size, 0, 0);
  57         return ptr;
  58 }
  59
  60 /*
  61  * free_bootmem_late - free bootmem pages directly to page allocator
  62  * @addr: starting address of the range
  63  * @size: size of the range in bytes
  64  *
  65  * This is only useful when the bootmem allocator has already been torn
  66  * down, but we are still initializing the system.  Pages are given directly
  67  * to the page allocator, no bootmem metadata is updated because it is gone.
  68  */
  69 void __init free_bootmem_late(unsigned long addr, unsigned long size)
  70 {
  71         unsigned long cursor, end;
  72
  73         kmemleak_free_part(__va(addr), size);
  74
  75         cursor = PFN_UP(addr);
  76         end = PFN_DOWN(addr + size);
  77
  78         for (; cursor < end; cursor++) {
  79                 __free_pages_bootmem(pfn_to_page(cursor), 0);
  80                 totalram_pages++;
  81         }
  82 }
  83
  84 static void __init __free_pages_memory(unsigned long start, unsigned long end)
  85 {
  86         int i;
  87         unsigned long start_aligned, end_aligned;
  88         int order = ilog2(BITS_PER_LONG);
  89
  90         start_aligned = (start + (BITS_PER_LONG - 1)) & ~(BITS_PER_LONG - 1);
  91         end_aligned = end & ~(BITS_PER_LONG - 1);
  92
  93         if (end_aligned <= start_aligned) {
  94                 for (i = start; i < end; i++)
  95                         __free_pages_bootmem(pfn_to_page(i), 0);
  96
  97                 return;
  98         }
  99
 100         for (i = start; i < start_aligned; i++)
 101                 __free_pages_bootmem(pfn_to_page(i), 0);
 102
 103         for (i = start_aligned; i < end_aligned; i += BITS_PER_LONG)
 104                 __free_pages_bootmem(pfn_to_page(i), order);
 105
 106         for (i = end_aligned; i < end; i++)
 107                 __free_pages_bootmem(pfn_to_page(i), 0);
 108 }
 109
 110 unsigned long __init free_all_memory_core_early(int nodeid)
 111 {
 112         int i;
 113         u64 start, end;
 114         unsigned long count = 0;
 115         struct range *range = NULL;
 116         int nr_range;
 117
 118         nr_range = get_free_all_memory_range(&range, nodeid);
 119
 120         for (i = 0; i < nr_range; i++) {
 121                 start = range[i].start;
 122                 end = range[i].end;
 123                 count += end - start;
 124                 __free_pages_memory(start, end);
 125         }
 126
 127         return count;
 128 }
 129
 130 /**
 131  * free_all_bootmem_node - release a node's free pages to the buddy allocator
 132  * @pgdat: node to be released
 133  *
 134  * Returns the number of pages actually released.
 135  */
 136 unsigned long __init free_all_bootmem_node(pg_data_t *pgdat)
 137 {
 138         register_page_bootmem_info_node(pgdat);
 139
 140         /* free_all_memory_core_early(MAX_NUMNODES) will be called later */
 141         return 0;
 142 }
 143
 144 /**
 145  * free_all_bootmem - release free pages to the buddy allocator
 146  *
 147  * Returns the number of pages actually released.
 148  */
 149 unsigned long __init free_all_bootmem(void)
 150 {
 151         /*
 152          * We need to use MAX_NUMNODES instead of NODE_DATA(0)->node_id
 153          *  because in some case like Node0 doesn't have RAM installed
 154          *  low ram will be on Node1
 155          * Use MAX_NUMNODES will make sure all ranges in early_node_map[]
 156          *  will be used instead of only Node0 related
 157          */
 158         return free_all_memory_core_early(MAX_NUMNODES);
 159 }
 160
 161 /**
 162  * free_bootmem_node - mark a page range as usable
 163  * @pgdat: node the range resides on
 164  * @physaddr: starting address of the range
 165  * @size: size of the range in bytes
 166  *
 167  * Partial pages will be considered reserved and left as they are.
 168  *
 169  * The range must reside completely on the specified node.
 170  */
 171 void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
 172                               unsigned long size)
 173 {
 174         kmemleak_free_part(__va(physaddr), size);
 175         memblock_x86_free_range(physaddr, physaddr + size);
 176 }
 177
 178 /**
 179  * free_bootmem - mark a page range as usable
 180  * @addr: starting address of the range
 181  * @size: size of the range in bytes
 182  *
 183  * Partial pages will be considered reserved and left as they are.
 184  *
 185  * The range must be contiguous but may span node boundaries.
 186  */
 187 void __init free_bootmem(unsigned long addr, unsigned long size)
 188 {
 189         kmemleak_free_part(__va(addr), size);
 190         memblock_x86_free_range(addr, addr + size);
 191 }
 192
 193 static void * __init ___alloc_bootmem_nopanic(unsigned long size,
 194                                         unsigned long align,
 195                                         unsigned long goal,
 196                                         unsigned long limit)
 197 {
 198         void *ptr;
 199
 200         if (WARN_ON_ONCE(slab_is_available()))
 201                 return kzalloc(size, GFP_NOWAIT);
 202
 203 restart:
 204
 205         ptr = __alloc_memory_core_early(MAX_NUMNODES, size, align, goal, limit);
 206
 207         if (ptr)
 208                 return ptr;
 209
 210         if (goal != 0) {
 211                 goal = 0;
 212                 goto restart;
 213         }
 214
 215         return NULL;
 216 }
 217
 218 /**
 219  * __alloc_bootmem_nopanic - allocate boot memory without panicking
 220  * @size: size of the request in bytes
 221  * @align: alignment of the region
 222  * @goal: preferred starting address of the region
 223  *
 224  * The goal is dropped if it can not be satisfied and the allocation will
 225  * fall back to memory below @goal.
 226  *
 227  * Allocation may happen on any node in the system.
 228  *
 229  * Returns NULL on failure.
 230  */
 231 void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align,
 232                                         unsigned long goal)
 233 {
 234         unsigned long limit = -1UL;
 235
 236         return ___alloc_bootmem_nopanic(size, align, goal, limit);
 237 }
 238
 239 static void * __init ___alloc_bootmem(unsigned long size, unsigned long align,
 240                                         unsigned long goal, unsigned long limit)
 241 {
 242         void *mem = ___alloc_bootmem_nopanic(size, align, goal, limit);
 243
 244         if (mem)
 245                 return mem;
 246         /*
 247          * Whoops, we cannot satisfy the allocation request.
 248          */
 249         printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size);
 250         panic("Out of memory");
 251         return NULL;
 252 }
 253
 254 /**
 255  * __alloc_bootmem - allocate boot memory
 256  * @size: size of the request in bytes
 257  * @align: alignment of the region
 258  * @goal: preferred starting address of the region
 259  *
 260  * The goal is dropped if it can not be satisfied and the allocation will
 261  * fall back to memory below @goal.
 262  *
 263  * Allocation may happen on any node in the system.
 264  *
 265  * The function panics if the request can not be satisfied.
 266  */
 267 void * __init __alloc_bootmem(unsigned long size, unsigned long align,
 268                               unsigned long goal)
 269 {
 270         unsigned long limit = -1UL;
 271
 272         return ___alloc_bootmem(size, align, goal, limit);
 273 }
 274
 275 /**
 276  * __alloc_bootmem_node - allocate boot memory from a specific node
 277  * @pgdat: node to allocate from
 278  * @size: size of the request in bytes
 279  * @align: alignment of the region
 280  * @goal: preferred starting address of the region
 281  *
 282  * The goal is dropped if it can not be satisfied and the allocation will
 283  * fall back to memory below @goal.
 284  *
 285  * Allocation may fall back to any node in the system if the specified node
 286  * can not hold the requested memory.
 287  *
 288  * The function panics if the request can not be satisfied.
 289  */
 290 void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size,
 291                                    unsigned long align, unsigned long goal)
 292 {
 293         void *ptr;
 294
 295         if (WARN_ON_ONCE(slab_is_available()))
 296                 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
 297
 298         ptr = __alloc_memory_core_early(pgdat->node_id, size, align,
 299                                          goal, -1ULL);
 300         if (ptr)
 301                 return ptr;
 302
 303         return __alloc_memory_core_early(MAX_NUMNODES, size, align,
 304                                          goal, -1ULL);
 305 }
 306
 307 void * __init __alloc_bootmem_node_high(pg_data_t *pgdat, unsigned long size,
 308                                    unsigned long align, unsigned long goal)
 309 {
 310         return __alloc_bootmem_node(pgdat, size, align, goal);
 311 }
 312
 313 #ifdef CONFIG_SPARSEMEM
 314 /**
 315  * alloc_bootmem_section - allocate boot memory from a specific section
 316  * @size: size of the request in bytes
 317  * @section_nr: sparse map section to allocate from
 318  *
 319  * Return NULL on failure.
 320  */
 321 void * __init alloc_bootmem_section(unsigned long size,
 322                                     unsigned long section_nr)
 323 {
 324         unsigned long pfn, goal, limit;
 325
 326         pfn = section_nr_to_pfn(section_nr);
 327         goal = pfn << PAGE_SHIFT;
 328         limit = section_nr_to_pfn(section_nr + 1) << PAGE_SHIFT;
 329
 330         return __alloc_memory_core_early(early_pfn_to_nid(pfn), size,
 331                                          SMP_CACHE_BYTES, goal, limit);
 332 }
 333 #endif
 334
 335 void * __init __alloc_bootmem_node_nopanic(pg_data_t *pgdat, unsigned long size,
 336                                    unsigned long align, unsigned long goal)
 337 {
 338         void *ptr;
 339
 340         if (WARN_ON_ONCE(slab_is_available()))
 341                 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
 342
 343         ptr =  __alloc_memory_core_early(pgdat->node_id, size, align,
 344                                                  goal, -1ULL);
 345         if (ptr)
 346                 return ptr;
 347
 348         return __alloc_bootmem_nopanic(size, align, goal);
 349 }
 350
 351 #ifndef ARCH_LOW_ADDRESS_LIMIT
 352 #define ARCH_LOW_ADDRESS_LIMIT  0xffffffffUL
 353 #endif
 354
 355 /**
 356  * __alloc_bootmem_low - allocate low boot memory
 357  * @size: size of the request in bytes
 358  * @align: alignment of the region
 359  * @goal: preferred starting address of the region
 360  *
 361  * The goal is dropped if it can not be satisfied and the allocation will
 362  * fall back to memory below @goal.
 363  *
 364  * Allocation may happen on any node in the system.
 365  *
 366  * The function panics if the request can not be satisfied.
 367  */
 368 void * __init __alloc_bootmem_low(unsigned long size, unsigned long align,
 369                                   unsigned long goal)
 370 {
 371         return ___alloc_bootmem(size, align, goal, ARCH_LOW_ADDRESS_LIMIT);
 372 }
 373
 374 /**
 375  * __alloc_bootmem_low_node - allocate low boot memory from a specific node
 376  * @pgdat: node to allocate from
 377  * @size: size of the request in bytes
 378  * @align: alignment of the region
 379  * @goal: preferred starting address of the region
 380  *
 381  * The goal is dropped if it can not be satisfied and the allocation will
 382  * fall back to memory below @goal.
 383  *
 384  * Allocation may fall back to any node in the system if the specified node
 385  * can not hold the requested memory.
 386  *
 387  * The function panics if the request can not be satisfied.
 388  */
 389 void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size,
 390                                        unsigned long align, unsigned long goal)
 391 {
 392         void *ptr;
 393
 394         if (WARN_ON_ONCE(slab_is_available()))
 395                 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
 396
 397         ptr = __alloc_memory_core_early(pgdat->node_id, size, align,
 398                                 goal, ARCH_LOW_ADDRESS_LIMIT);
 399         if (ptr)
 400                 return ptr;
 401
 402         return  __alloc_memory_core_early(MAX_NUMNODES, size, align,
 403                                 goal, ARCH_LOW_ADDRESS_LIMIT);
 404 }