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 = memblock_find_in_range_node(size, align, goal, limit, nid);
  45         if (!addr)
  46                 return NULL;
  47
  48         if (memblock_reserve(addr, size))
  49                 return NULL;
  50
  51         ptr = phys_to_virt(addr);
  52         memset(ptr, 0, size);
  53         /*
  54          * The min_count is set to 0 so that bootmem allocated blocks
  55          * are never reported as leaks.
  56          */
  57         kmemleak_alloc(ptr, size, 0, 0);
  58         return ptr;
  59 }
  60
  61 /*
  62  * free_bootmem_late - free bootmem pages directly to page allocator
  63  * @addr: starting address of the range
  64  * @size: size of the range in bytes
  65  *
  66  * This is only useful when the bootmem allocator has already been torn
  67  * down, but we are still initializing the system.  Pages are given directly
  68  * to the page allocator, no bootmem metadata is updated because it is gone.
  69  */
  70 void __init free_bootmem_late(unsigned long addr, unsigned long size)
  71 {
  72         unsigned long cursor, end;
  73
  74         kmemleak_free_part(__va(addr), size);
  75
  76         cursor = PFN_UP(addr);
  77         end = PFN_DOWN(addr + size);
  78
  79         for (; cursor < end; cursor++) {
  80                 __free_pages_bootmem(pfn_to_page(cursor), 0);
  81                 totalram_pages++;
  82         }
  83 }
  84
  85 static void __init __free_pages_memory(unsigned long start, unsigned long end)
  86 {
  87         int order;
  88
  89         while (start < end) {
  90                 order = min(MAX_ORDER - 1UL, __ffs(start));
  91
  92                 while (start + (1UL << order) > end)
  93                         order--;
  94
  95                 __free_pages_bootmem(pfn_to_page(start), order);
  96
  97                 start += (1UL << order);
  98         }
  99 }
 100
 101 static unsigned long __init __free_memory_core(phys_addr_t start,
 102                                  phys_addr_t end)
 103 {
 104         unsigned long start_pfn = PFN_UP(start);
 105         unsigned long end_pfn = min_t(unsigned long,
 106                                       PFN_DOWN(end), max_low_pfn);
 107
 108         if (start_pfn > end_pfn)
 109                 return 0;
 110
 111         __free_pages_memory(start_pfn, end_pfn);
 112
 113         return end_pfn - start_pfn;
 114 }
 115
 116 static unsigned long __init free_low_memory_core_early(void)
 117 {
 118         unsigned long count = 0;
 119         phys_addr_t start, end;
 120         u64 i;
 121
 122         for_each_free_mem_range(i, NUMA_NO_NODE, &start, &end, NULL)
 123                 count += __free_memory_core(start, end);
 124
 125 #ifdef CONFIG_ARCH_DISCARD_MEMBLOCK
 126         {
 127                 phys_addr_t size;
 128
 129                 /* Free memblock.reserved array if it was allocated */
 130                 size = get_allocated_memblock_reserved_regions_info(&start);
 131                 if (size)
 132                         count += __free_memory_core(start, start + size);
 133
 134                 /* Free memblock.memory array if it was allocated */
 135                 size = get_allocated_memblock_memory_regions_info(&start);
 136                 if (size)
 137                         count += __free_memory_core(start, start + size);
 138         }
 139 #endif
 140
 141         return count;
 142 }
 143
 144 static int reset_managed_pages_done __initdata;
 145
 146 void reset_node_managed_pages(pg_data_t *pgdat)
 147 {
 148         struct zone *z;
 149
 150         for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
 151                 z->managed_pages = 0;
 152 }
 153
 154 void __init reset_all_zones_managed_pages(void)
 155 {
 156         struct pglist_data *pgdat;
 157
 158         if (reset_managed_pages_done)
 159                 return;
 160
 161         for_each_online_pgdat(pgdat)
 162                 reset_node_managed_pages(pgdat);
 163
 164         reset_managed_pages_done = 1;
 165 }
 166
 167 /**
 168  * free_all_bootmem - release free pages to the buddy allocator
 169  *
 170  * Returns the number of pages actually released.
 171  */
 172 unsigned long __init free_all_bootmem(void)
 173 {
 174         unsigned long pages;
 175
 176         reset_all_zones_managed_pages();
 177
 178         /*
 179          * We need to use NUMA_NO_NODE instead of NODE_DATA(0)->node_id
 180          *  because in some case like Node0 doesn't have RAM installed
 181          *  low ram will be on Node1
 182          */
 183         pages = free_low_memory_core_early();
 184         totalram_pages += pages;
 185
 186         return pages;
 187 }
 188
 189 /**
 190  * free_bootmem_node - mark a page range as usable
 191  * @pgdat: node the range resides on
 192  * @physaddr: starting address of the range
 193  * @size: size of the range in bytes
 194  *
 195  * Partial pages will be considered reserved and left as they are.
 196  *
 197  * The range must reside completely on the specified node.
 198  */
 199 void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
 200                               unsigned long size)
 201 {
 202         memblock_free(physaddr, size);
 203 }
 204
 205 /**
 206  * free_bootmem - mark a page range as usable
 207  * @addr: starting address of the range
 208  * @size: size of the range in bytes
 209  *
 210  * Partial pages will be considered reserved and left as they are.
 211  *
 212  * The range must be contiguous but may span node boundaries.
 213  */
 214 void __init free_bootmem(unsigned long addr, unsigned long size)
 215 {
 216         memblock_free(addr, size);
 217 }
 218
 219 static void * __init ___alloc_bootmem_nopanic(unsigned long size,
 220                                         unsigned long align,
 221                                         unsigned long goal,
 222                                         unsigned long limit)
 223 {
 224         void *ptr;
 225
 226         if (WARN_ON_ONCE(slab_is_available()))
 227                 return kzalloc(size, GFP_NOWAIT);
 228
 229 restart:
 230
 231         ptr = __alloc_memory_core_early(NUMA_NO_NODE, size, align, goal, limit);
 232
 233         if (ptr)
 234                 return ptr;
 235
 236         if (goal != 0) {
 237                 goal = 0;
 238                 goto restart;
 239         }
 240
 241         return NULL;
 242 }
 243
 244 /**
 245  * __alloc_bootmem_nopanic - allocate boot memory without panicking
 246  * @size: size of the request in bytes
 247  * @align: alignment of the region
 248  * @goal: preferred starting address of the region
 249  *
 250  * The goal is dropped if it can not be satisfied and the allocation will
 251  * fall back to memory below @goal.
 252  *
 253  * Allocation may happen on any node in the system.
 254  *
 255  * Returns NULL on failure.
 256  */
 257 void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align,
 258                                         unsigned long goal)
 259 {
 260         unsigned long limit = -1UL;
 261
 262         return ___alloc_bootmem_nopanic(size, align, goal, limit);
 263 }
 264
 265 static void * __init ___alloc_bootmem(unsigned long size, unsigned long align,
 266                                         unsigned long goal, unsigned long limit)
 267 {
 268         void *mem = ___alloc_bootmem_nopanic(size, align, goal, limit);
 269
 270         if (mem)
 271                 return mem;
 272         /*
 273          * Whoops, we cannot satisfy the allocation request.
 274          */
 275         printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size);
 276         panic("Out of memory");
 277         return NULL;
 278 }
 279
 280 /**
 281  * __alloc_bootmem - allocate boot memory
 282  * @size: size of the request in bytes
 283  * @align: alignment of the region
 284  * @goal: preferred starting address of the region
 285  *
 286  * The goal is dropped if it can not be satisfied and the allocation will
 287  * fall back to memory below @goal.
 288  *
 289  * Allocation may happen on any node in the system.
 290  *
 291  * The function panics if the request can not be satisfied.
 292  */
 293 void * __init __alloc_bootmem(unsigned long size, unsigned long align,
 294                               unsigned long goal)
 295 {
 296         unsigned long limit = -1UL;
 297
 298         return ___alloc_bootmem(size, align, goal, limit);
 299 }
 300
 301 void * __init ___alloc_bootmem_node_nopanic(pg_data_t *pgdat,
 302                                                    unsigned long size,
 303                                                    unsigned long align,
 304                                                    unsigned long goal,
 305                                                    unsigned long limit)
 306 {
 307         void *ptr;
 308
 309 again:
 310         ptr = __alloc_memory_core_early(pgdat->node_id, size, align,
 311                                         goal, limit);
 312         if (ptr)
 313                 return ptr;
 314
 315         ptr = __alloc_memory_core_early(NUMA_NO_NODE, size, align,
 316                                         goal, limit);
 317         if (ptr)
 318                 return ptr;
 319
 320         if (goal) {
 321                 goal = 0;
 322                 goto again;
 323         }
 324
 325         return NULL;
 326 }
 327
 328 void * __init __alloc_bootmem_node_nopanic(pg_data_t *pgdat, unsigned long size,
 329                                    unsigned long align, unsigned long goal)
 330 {
 331         if (WARN_ON_ONCE(slab_is_available()))
 332                 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
 333
 334         return ___alloc_bootmem_node_nopanic(pgdat, size, align, goal, 0);
 335 }
 336
 337 static void * __init ___alloc_bootmem_node(pg_data_t *pgdat, unsigned long size,
 338                                     unsigned long align, unsigned long goal,
 339                                     unsigned long limit)
 340 {
 341         void *ptr;
 342
 343         ptr = ___alloc_bootmem_node_nopanic(pgdat, size, align, goal, limit);
 344         if (ptr)
 345                 return ptr;
 346
 347         printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size);
 348         panic("Out of memory");
 349         return NULL;
 350 }
 351
 352 /**
 353  * __alloc_bootmem_node - allocate boot memory from a specific node
 354  * @pgdat: node to allocate from
 355  * @size: size of the request in bytes
 356  * @align: alignment of the region
 357  * @goal: preferred starting address of the region
 358  *
 359  * The goal is dropped if it can not be satisfied and the allocation will
 360  * fall back to memory below @goal.
 361  *
 362  * Allocation may fall back to any node in the system if the specified node
 363  * can not hold the requested memory.
 364  *
 365  * The function panics if the request can not be satisfied.
 366  */
 367 void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size,
 368                                    unsigned long align, unsigned long goal)
 369 {
 370         if (WARN_ON_ONCE(slab_is_available()))
 371                 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
 372
 373         return ___alloc_bootmem_node(pgdat, size, align, goal, 0);
 374 }
 375
 376 void * __init __alloc_bootmem_node_high(pg_data_t *pgdat, unsigned long size,
 377                                    unsigned long align, unsigned long goal)
 378 {
 379         return __alloc_bootmem_node(pgdat, size, align, goal);
 380 }
 381
 382 #ifndef ARCH_LOW_ADDRESS_LIMIT
 383 #define ARCH_LOW_ADDRESS_LIMIT  0xffffffffUL
 384 #endif
 385
 386 /**
 387  * __alloc_bootmem_low - allocate low boot memory
 388  * @size: size of the request in bytes
 389  * @align: alignment of the region
 390  * @goal: preferred starting address of the region
 391  *
 392  * The goal is dropped if it can not be satisfied and the allocation will
 393  * fall back to memory below @goal.
 394  *
 395  * Allocation may happen on any node in the system.
 396  *
 397  * The function panics if the request can not be satisfied.
 398  */
 399 void * __init __alloc_bootmem_low(unsigned long size, unsigned long align,
 400                                   unsigned long goal)
 401 {
 402         return ___alloc_bootmem(size, align, goal, ARCH_LOW_ADDRESS_LIMIT);
 403 }
 404
 405 void * __init __alloc_bootmem_low_nopanic(unsigned long size,
 406                                           unsigned long align,
 407                                           unsigned long goal)
 408 {
 409         return ___alloc_bootmem_nopanic(size, align, goal,
 410                                         ARCH_LOW_ADDRESS_LIMIT);
 411 }
 412
 413 /**
 414  * __alloc_bootmem_low_node - allocate low boot memory from a specific node
 415  * @pgdat: node to allocate from
 416  * @size: size of the request in bytes
 417  * @align: alignment of the region
 418  * @goal: preferred starting address of the region
 419  *
 420  * The goal is dropped if it can not be satisfied and the allocation will
 421  * fall back to memory below @goal.
 422  *
 423  * Allocation may fall back to any node in the system if the specified node
 424  * can not hold the requested memory.
 425  *
 426  * The function panics if the request can not be satisfied.
 427  */
 428 void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size,
 429                                        unsigned long align, unsigned long goal)
 430 {
 431         if (WARN_ON_ONCE(slab_is_available()))
 432                 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
 433
 434         return ___alloc_bootmem_node(pgdat, size, align, goal,
 435                                      ARCH_LOW_ADDRESS_LIMIT);
 436 }