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                                            MEMBLOCK_NONE);
  46         if (!addr)
  47                 return NULL;
  48
  49         if (memblock_reserve(addr, size))
  50                 return NULL;
  51
  52         ptr = phys_to_virt(addr);
  53         memset(ptr, 0, size);
  54         /*
  55          * The min_count is set to 0 so that bootmem allocated blocks
  56          * are never reported as leaks.
  57          */
  58         kmemleak_alloc(ptr, size, 0, 0);
  59         return ptr;
  60 }
  61
  62 /*
  63  * free_bootmem_late - free bootmem pages directly to page allocator
  64  * @addr: starting address of the range
  65  * @size: size of the range in bytes
  66  *
  67  * This is only useful when the bootmem allocator has already been torn
  68  * down, but we are still initializing the system.  Pages are given directly
  69  * to the page allocator, no bootmem metadata is updated because it is gone.
  70  */
  71 void __init free_bootmem_late(unsigned long addr, unsigned long size)
  72 {
  73         unsigned long cursor, end;
  74
  75         kmemleak_free_part(__va(addr), size);
  76
  77         cursor = PFN_UP(addr);
  78         end = PFN_DOWN(addr + size);
  79
  80         for (; cursor < end; cursor++) {
  81                 __free_pages_bootmem(pfn_to_page(cursor), 0);
  82                 totalram_pages++;
  83         }
  84 }
  85
  86 static void __init __free_pages_memory(unsigned long start, unsigned long end)
  87 {
  88         int order;
  89
  90         while (start < end) {
  91                 order = min(MAX_ORDER - 1UL, __ffs(start));
  92
  93                 while (start + (1UL << order) > end)
  94                         order--;
  95
  96                 __free_pages_bootmem(pfn_to_page(start), order);
  97
  98                 start += (1UL << order);
  99         }
 100 }
 101
 102 static unsigned long __init __free_memory_core(phys_addr_t start,
 103                                  phys_addr_t end)
 104 {
 105         unsigned long start_pfn = PFN_UP(start);
 106         unsigned long end_pfn = min_t(unsigned long,
 107                                       PFN_DOWN(end), max_low_pfn);
 108
 109         if (start_pfn > end_pfn)
 110                 return 0;
 111
 112         __free_pages_memory(start_pfn, end_pfn);
 113
 114         return end_pfn - start_pfn;
 115 }
 116
 117 static unsigned long __init free_low_memory_core_early(void)
 118 {
 119         unsigned long count = 0;
 120         phys_addr_t start, end;
 121         u64 i;
 122
 123         memblock_clear_hotplug(0, -1);
 124
 125         for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE, &start, &end,
 126                                 NULL)
 127                 count += __free_memory_core(start, end);
 128
 129 #ifdef CONFIG_ARCH_DISCARD_MEMBLOCK
 130         {
 131                 phys_addr_t size;
 132
 133                 /* Free memblock.reserved array if it was allocated */
 134                 size = get_allocated_memblock_reserved_regions_info(&start);
 135                 if (size)
 136                         count += __free_memory_core(start, start + size);
 137
 138                 /* Free memblock.memory array if it was allocated */
 139                 size = get_allocated_memblock_memory_regions_info(&start);
 140                 if (size)
 141                         count += __free_memory_core(start, start + size);
 142         }
 143 #endif
 144
 145         return count;
 146 }
 147
 148 static int reset_managed_pages_done __initdata;
 149
 150 void reset_node_managed_pages(pg_data_t *pgdat)
 151 {
 152         struct zone *z;
 153
 154         for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
 155                 z->managed_pages = 0;
 156 }
 157
 158 void __init reset_all_zones_managed_pages(void)
 159 {
 160         struct pglist_data *pgdat;
 161
 162         if (reset_managed_pages_done)
 163                 return;
 164
 165         for_each_online_pgdat(pgdat)
 166                 reset_node_managed_pages(pgdat);
 167
 168         reset_managed_pages_done = 1;
 169 }
 170
 171 /**
 172  * free_all_bootmem - release free pages to the buddy allocator
 173  *
 174  * Returns the number of pages actually released.
 175  */
 176 unsigned long __init free_all_bootmem(void)
 177 {
 178         unsigned long pages;
 179
 180         reset_all_zones_managed_pages();
 181
 182         /*
 183          * We need to use NUMA_NO_NODE instead of NODE_DATA(0)->node_id
 184          *  because in some case like Node0 doesn't have RAM installed
 185          *  low ram will be on Node1
 186          */
 187         pages = free_low_memory_core_early();
 188         totalram_pages += pages;
 189
 190         return pages;
 191 }
 192
 193 /**
 194  * free_bootmem_node - mark a page range as usable
 195  * @pgdat: node the range resides on
 196  * @physaddr: starting address of the range
 197  * @size: size of the range in bytes
 198  *
 199  * Partial pages will be considered reserved and left as they are.
 200  *
 201  * The range must reside completely on the specified node.
 202  */
 203 void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
 204                               unsigned long size)
 205 {
 206         memblock_free(physaddr, size);
 207 }
 208
 209 /**
 210  * free_bootmem - mark a page range as usable
 211  * @addr: starting address of the range
 212  * @size: size of the range in bytes
 213  *
 214  * Partial pages will be considered reserved and left as they are.
 215  *
 216  * The range must be contiguous but may span node boundaries.
 217  */
 218 void __init free_bootmem(unsigned long addr, unsigned long size)
 219 {
 220         memblock_free(addr, size);
 221 }
 222
 223 static void * __init ___alloc_bootmem_nopanic(unsigned long size,
 224                                         unsigned long align,
 225                                         unsigned long goal,
 226                                         unsigned long limit)
 227 {
 228         void *ptr;
 229
 230         if (WARN_ON_ONCE(slab_is_available()))
 231                 return kzalloc(size, GFP_NOWAIT);
 232
 233 restart:
 234
 235         ptr = __alloc_memory_core_early(NUMA_NO_NODE, size, align, goal, limit);
 236
 237         if (ptr)
 238                 return ptr;
 239
 240         if (goal != 0) {
 241                 goal = 0;
 242                 goto restart;
 243         }
 244
 245         return NULL;
 246 }
 247
 248 /**
 249  * __alloc_bootmem_nopanic - allocate boot memory without panicking
 250  * @size: size of the request in bytes
 251  * @align: alignment of the region
 252  * @goal: preferred starting address of the region
 253  *
 254  * The goal is dropped if it can not be satisfied and the allocation will
 255  * fall back to memory below @goal.
 256  *
 257  * Allocation may happen on any node in the system.
 258  *
 259  * Returns NULL on failure.
 260  */
 261 void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align,
 262                                         unsigned long goal)
 263 {
 264         unsigned long limit = -1UL;
 265
 266         return ___alloc_bootmem_nopanic(size, align, goal, limit);
 267 }
 268
 269 static void * __init ___alloc_bootmem(unsigned long size, unsigned long align,
 270                                         unsigned long goal, unsigned long limit)
 271 {
 272         void *mem = ___alloc_bootmem_nopanic(size, align, goal, limit);
 273
 274         if (mem)
 275                 return mem;
 276         /*
 277          * Whoops, we cannot satisfy the allocation request.
 278          */
 279         printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size);
 280         panic("Out of memory");
 281         return NULL;
 282 }
 283
 284 /**
 285  * __alloc_bootmem - allocate boot memory
 286  * @size: size of the request in bytes
 287  * @align: alignment of the region
 288  * @goal: preferred starting address of the region
 289  *
 290  * The goal is dropped if it can not be satisfied and the allocation will
 291  * fall back to memory below @goal.
 292  *
 293  * Allocation may happen on any node in the system.
 294  *
 295  * The function panics if the request can not be satisfied.
 296  */
 297 void * __init __alloc_bootmem(unsigned long size, unsigned long align,
 298                               unsigned long goal)
 299 {
 300         unsigned long limit = -1UL;
 301
 302         return ___alloc_bootmem(size, align, goal, limit);
 303 }
 304
 305 void * __init ___alloc_bootmem_node_nopanic(pg_data_t *pgdat,
 306                                                    unsigned long size,
 307                                                    unsigned long align,
 308                                                    unsigned long goal,
 309                                                    unsigned long limit)
 310 {
 311         void *ptr;
 312
 313 again:
 314         ptr = __alloc_memory_core_early(pgdat->node_id, size, align,
 315                                         goal, limit);
 316         if (ptr)
 317                 return ptr;
 318
 319         ptr = __alloc_memory_core_early(NUMA_NO_NODE, size, align,
 320                                         goal, limit);
 321         if (ptr)
 322                 return ptr;
 323
 324         if (goal) {
 325                 goal = 0;
 326                 goto again;
 327         }
 328
 329         return NULL;
 330 }
 331
 332 void * __init __alloc_bootmem_node_nopanic(pg_data_t *pgdat, unsigned long size,
 333                                    unsigned long align, unsigned long goal)
 334 {
 335         if (WARN_ON_ONCE(slab_is_available()))
 336                 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
 337
 338         return ___alloc_bootmem_node_nopanic(pgdat, size, align, goal, 0);
 339 }
 340
 341 static void * __init ___alloc_bootmem_node(pg_data_t *pgdat, unsigned long size,
 342                                     unsigned long align, unsigned long goal,
 343                                     unsigned long limit)
 344 {
 345         void *ptr;
 346
 347         ptr = ___alloc_bootmem_node_nopanic(pgdat, size, align, goal, limit);
 348         if (ptr)
 349                 return ptr;
 350
 351         printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size);
 352         panic("Out of memory");
 353         return NULL;
 354 }
 355
 356 /**
 357  * __alloc_bootmem_node - allocate boot memory from a specific node
 358  * @pgdat: node to allocate from
 359  * @size: size of the request in bytes
 360  * @align: alignment of the region
 361  * @goal: preferred starting address of the region
 362  *
 363  * The goal is dropped if it can not be satisfied and the allocation will
 364  * fall back to memory below @goal.
 365  *
 366  * Allocation may fall back to any node in the system if the specified node
 367  * can not hold the requested memory.
 368  *
 369  * The function panics if the request can not be satisfied.
 370  */
 371 void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size,
 372                                    unsigned long align, unsigned long goal)
 373 {
 374         if (WARN_ON_ONCE(slab_is_available()))
 375                 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
 376
 377         return ___alloc_bootmem_node(pgdat, size, align, goal, 0);
 378 }
 379
 380 void * __init __alloc_bootmem_node_high(pg_data_t *pgdat, unsigned long size,
 381                                    unsigned long align, unsigned long goal)
 382 {
 383         return __alloc_bootmem_node(pgdat, size, align, goal);
 384 }
 385
 386 #ifndef ARCH_LOW_ADDRESS_LIMIT
 387 #define ARCH_LOW_ADDRESS_LIMIT  0xffffffffUL
 388 #endif
 389
 390 /**
 391  * __alloc_bootmem_low - allocate low boot memory
 392  * @size: size of the request in bytes
 393  * @align: alignment of the region
 394  * @goal: preferred starting address of the region
 395  *
 396  * The goal is dropped if it can not be satisfied and the allocation will
 397  * fall back to memory below @goal.
 398  *
 399  * Allocation may happen on any node in the system.
 400  *
 401  * The function panics if the request can not be satisfied.
 402  */
 403 void * __init __alloc_bootmem_low(unsigned long size, unsigned long align,
 404                                   unsigned long goal)
 405 {
 406         return ___alloc_bootmem(size, align, goal, ARCH_LOW_ADDRESS_LIMIT);
 407 }
 408
 409 void * __init __alloc_bootmem_low_nopanic(unsigned long size,
 410                                           unsigned long align,
 411                                           unsigned long goal)
 412 {
 413         return ___alloc_bootmem_nopanic(size, align, goal,
 414                                         ARCH_LOW_ADDRESS_LIMIT);
 415 }
 416
 417 /**
 418  * __alloc_bootmem_low_node - allocate low boot memory from a specific node
 419  * @pgdat: node to allocate from
 420  * @size: size of the request in bytes
 421  * @align: alignment of the region
 422  * @goal: preferred starting address of the region
 423  *
 424  * The goal is dropped if it can not be satisfied and the allocation will
 425  * fall back to memory below @goal.
 426  *
 427  * Allocation may fall back to any node in the system if the specified node
 428  * can not hold the requested memory.
 429  *
 430  * The function panics if the request can not be satisfied.
 431  */
 432 void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size,
 433                                        unsigned long align, unsigned long goal)
 434 {
 435         if (WARN_ON_ONCE(slab_is_available()))
 436                 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
 437
 438         return ___alloc_bootmem_node(pgdat, size, align, goal,
 439                                      ARCH_LOW_ADDRESS_LIMIT);
 440 }