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         memblock_clear_hotplug(0, -1);
 123
 124         for_each_free_mem_range(i, NUMA_NO_NODE, &start, &end, NULL)
 125                 count += __free_memory_core(start, end);
 126
 127 #ifdef CONFIG_ARCH_DISCARD_MEMBLOCK
 128         {
 129                 phys_addr_t size;
 130
 131                 /* Free memblock.reserved array if it was allocated */
 132                 size = get_allocated_memblock_reserved_regions_info(&start);
 133                 if (size)
 134                         count += __free_memory_core(start, start + size);
 135
 136                 /* Free memblock.memory array if it was allocated */
 137                 size = get_allocated_memblock_memory_regions_info(&start);
 138                 if (size)
 139                         count += __free_memory_core(start, start + size);
 140         }
 141 #endif
 142
 143         return count;
 144 }
 145
 146 static int reset_managed_pages_done __initdata;
 147
 148 void reset_node_managed_pages(pg_data_t *pgdat)
 149 {
 150         struct zone *z;
 151
 152         for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
 153                 z->managed_pages = 0;
 154 }
 155
 156 void __init reset_all_zones_managed_pages(void)
 157 {
 158         struct pglist_data *pgdat;
 159
 160         if (reset_managed_pages_done)
 161                 return;
 162
 163         for_each_online_pgdat(pgdat)
 164                 reset_node_managed_pages(pgdat);
 165
 166         reset_managed_pages_done = 1;
 167 }
 168
 169 /**
 170  * free_all_bootmem - release free pages to the buddy allocator
 171  *
 172  * Returns the number of pages actually released.
 173  */
 174 unsigned long __init free_all_bootmem(void)
 175 {
 176         unsigned long pages;
 177
 178         reset_all_zones_managed_pages();
 179
 180         /*
 181          * We need to use NUMA_NO_NODE instead of NODE_DATA(0)->node_id
 182          *  because in some case like Node0 doesn't have RAM installed
 183          *  low ram will be on Node1
 184          */
 185         pages = free_low_memory_core_early();
 186         totalram_pages += pages;
 187
 188         return pages;
 189 }
 190
 191 /**
 192  * free_bootmem_node - mark a page range as usable
 193  * @pgdat: node the range resides on
 194  * @physaddr: starting address of the range
 195  * @size: size of the range in bytes
 196  *
 197  * Partial pages will be considered reserved and left as they are.
 198  *
 199  * The range must reside completely on the specified node.
 200  */
 201 void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
 202                               unsigned long size)
 203 {
 204         memblock_free(physaddr, size);
 205 }
 206
 207 /**
 208  * free_bootmem - mark a page range as usable
 209  * @addr: starting address of the range
 210  * @size: size of the range in bytes
 211  *
 212  * Partial pages will be considered reserved and left as they are.
 213  *
 214  * The range must be contiguous but may span node boundaries.
 215  */
 216 void __init free_bootmem(unsigned long addr, unsigned long size)
 217 {
 218         memblock_free(addr, size);
 219 }
 220
 221 static void * __init ___alloc_bootmem_nopanic(unsigned long size,
 222                                         unsigned long align,
 223                                         unsigned long goal,
 224                                         unsigned long limit)
 225 {
 226         void *ptr;
 227
 228         if (WARN_ON_ONCE(slab_is_available()))
 229                 return kzalloc(size, GFP_NOWAIT);
 230
 231 restart:
 232
 233         ptr = __alloc_memory_core_early(NUMA_NO_NODE, size, align, goal, limit);
 234
 235         if (ptr)
 236                 return ptr;
 237
 238         if (goal != 0) {
 239                 goal = 0;
 240                 goto restart;
 241         }
 242
 243         return NULL;
 244 }
 245
 246 /**
 247  * __alloc_bootmem_nopanic - allocate boot memory without panicking
 248  * @size: size of the request in bytes
 249  * @align: alignment of the region
 250  * @goal: preferred starting address of the region
 251  *
 252  * The goal is dropped if it can not be satisfied and the allocation will
 253  * fall back to memory below @goal.
 254  *
 255  * Allocation may happen on any node in the system.
 256  *
 257  * Returns NULL on failure.
 258  */
 259 void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align,
 260                                         unsigned long goal)
 261 {
 262         unsigned long limit = -1UL;
 263
 264         return ___alloc_bootmem_nopanic(size, align, goal, limit);
 265 }
 266
 267 static void * __init ___alloc_bootmem(unsigned long size, unsigned long align,
 268                                         unsigned long goal, unsigned long limit)
 269 {
 270         void *mem = ___alloc_bootmem_nopanic(size, align, goal, limit);
 271
 272         if (mem)
 273                 return mem;
 274         /*
 275          * Whoops, we cannot satisfy the allocation request.
 276          */
 277         printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size);
 278         panic("Out of memory");
 279         return NULL;
 280 }
 281
 282 /**
 283  * __alloc_bootmem - allocate boot memory
 284  * @size: size of the request in bytes
 285  * @align: alignment of the region
 286  * @goal: preferred starting address of the region
 287  *
 288  * The goal is dropped if it can not be satisfied and the allocation will
 289  * fall back to memory below @goal.
 290  *
 291  * Allocation may happen on any node in the system.
 292  *
 293  * The function panics if the request can not be satisfied.
 294  */
 295 void * __init __alloc_bootmem(unsigned long size, unsigned long align,
 296                               unsigned long goal)
 297 {
 298         unsigned long limit = -1UL;
 299
 300         return ___alloc_bootmem(size, align, goal, limit);
 301 }
 302
 303 void * __init ___alloc_bootmem_node_nopanic(pg_data_t *pgdat,
 304                                                    unsigned long size,
 305                                                    unsigned long align,
 306                                                    unsigned long goal,
 307                                                    unsigned long limit)
 308 {
 309         void *ptr;
 310
 311 again:
 312         ptr = __alloc_memory_core_early(pgdat->node_id, size, align,
 313                                         goal, limit);
 314         if (ptr)
 315                 return ptr;
 316
 317         ptr = __alloc_memory_core_early(NUMA_NO_NODE, size, align,
 318                                         goal, limit);
 319         if (ptr)
 320                 return ptr;
 321
 322         if (goal) {
 323                 goal = 0;
 324                 goto again;
 325         }
 326
 327         return NULL;
 328 }
 329
 330 void * __init __alloc_bootmem_node_nopanic(pg_data_t *pgdat, unsigned long size,
 331                                    unsigned long align, unsigned long goal)
 332 {
 333         if (WARN_ON_ONCE(slab_is_available()))
 334                 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
 335
 336         return ___alloc_bootmem_node_nopanic(pgdat, size, align, goal, 0);
 337 }
 338
 339 static void * __init ___alloc_bootmem_node(pg_data_t *pgdat, unsigned long size,
 340                                     unsigned long align, unsigned long goal,
 341                                     unsigned long limit)
 342 {
 343         void *ptr;
 344
 345         ptr = ___alloc_bootmem_node_nopanic(pgdat, size, align, goal, limit);
 346         if (ptr)
 347                 return ptr;
 348
 349         printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size);
 350         panic("Out of memory");
 351         return NULL;
 352 }
 353
 354 /**
 355  * __alloc_bootmem_node - allocate boot memory from a specific node
 356  * @pgdat: node to allocate from
 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 fall back to any node in the system if the specified node
 365  * can not hold the requested memory.
 366  *
 367  * The function panics if the request can not be satisfied.
 368  */
 369 void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size,
 370                                    unsigned long align, unsigned long goal)
 371 {
 372         if (WARN_ON_ONCE(slab_is_available()))
 373                 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
 374
 375         return ___alloc_bootmem_node(pgdat, size, align, goal, 0);
 376 }
 377
 378 void * __init __alloc_bootmem_node_high(pg_data_t *pgdat, unsigned long size,
 379                                    unsigned long align, unsigned long goal)
 380 {
 381         return __alloc_bootmem_node(pgdat, size, align, goal);
 382 }
 383
 384 #ifndef ARCH_LOW_ADDRESS_LIMIT
 385 #define ARCH_LOW_ADDRESS_LIMIT  0xffffffffUL
 386 #endif
 387
 388 /**
 389  * __alloc_bootmem_low - allocate low boot memory
 390  * @size: size of the request in bytes
 391  * @align: alignment of the region
 392  * @goal: preferred starting address of the region
 393  *
 394  * The goal is dropped if it can not be satisfied and the allocation will
 395  * fall back to memory below @goal.
 396  *
 397  * Allocation may happen on any node in the system.
 398  *
 399  * The function panics if the request can not be satisfied.
 400  */
 401 void * __init __alloc_bootmem_low(unsigned long size, unsigned long align,
 402                                   unsigned long goal)
 403 {
 404         return ___alloc_bootmem(size, align, goal, ARCH_LOW_ADDRESS_LIMIT);
 405 }
 406
 407 void * __init __alloc_bootmem_low_nopanic(unsigned long size,
 408                                           unsigned long align,
 409                                           unsigned long goal)
 410 {
 411         return ___alloc_bootmem_nopanic(size, align, goal,
 412                                         ARCH_LOW_ADDRESS_LIMIT);
 413 }
 414
 415 /**
 416  * __alloc_bootmem_low_node - allocate low boot memory from a specific node
 417  * @pgdat: node to allocate from
 418  * @size: size of the request in bytes
 419  * @align: alignment of the region
 420  * @goal: preferred starting address of the region
 421  *
 422  * The goal is dropped if it can not be satisfied and the allocation will
 423  * fall back to memory below @goal.
 424  *
 425  * Allocation may fall back to any node in the system if the specified node
 426  * can not hold the requested memory.
 427  *
 428  * The function panics if the request can not be satisfied.
 429  */
 430 void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size,
 431                                        unsigned long align, unsigned long goal)
 432 {
 433         if (WARN_ON_ONCE(slab_is_available()))
 434                 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
 435
 436         return ___alloc_bootmem_node(pgdat, size, align, goal,
 437                                      ARCH_LOW_ADDRESS_LIMIT);
 438 }