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 static inline void __init reset_node_managed_pages(pg_data_t *pgdat)
 147 {
 148         struct zone *z;
 149
 150         if (reset_managed_pages_done)
 151                 return;
 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         for_each_online_pgdat(pgdat)
 161                 reset_node_managed_pages(pgdat);
 162         reset_managed_pages_done = 1;
 163 }
 164
 165 /**
 166  * free_all_bootmem - release free pages to the buddy allocator
 167  *
 168  * Returns the number of pages actually released.
 169  */
 170 unsigned long __init free_all_bootmem(void)
 171 {
 172         unsigned long pages;
 173
 174         reset_all_zones_managed_pages();
 175
 176         /*
 177          * We need to use NUMA_NO_NODE instead of NODE_DATA(0)->node_id
 178          *  because in some case like Node0 doesn't have RAM installed
 179          *  low ram will be on Node1
 180          */
 181         pages = free_low_memory_core_early();
 182         totalram_pages += pages;
 183
 184         return pages;
 185 }
 186
 187 /**
 188  * free_bootmem_node - mark a page range as usable
 189  * @pgdat: node the range resides on
 190  * @physaddr: starting address of the range
 191  * @size: size of the range in bytes
 192  *
 193  * Partial pages will be considered reserved and left as they are.
 194  *
 195  * The range must reside completely on the specified node.
 196  */
 197 void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
 198                               unsigned long size)
 199 {
 200         memblock_free(physaddr, size);
 201 }
 202
 203 /**
 204  * free_bootmem - mark a page range as usable
 205  * @addr: starting address of the range
 206  * @size: size of the range in bytes
 207  *
 208  * Partial pages will be considered reserved and left as they are.
 209  *
 210  * The range must be contiguous but may span node boundaries.
 211  */
 212 void __init free_bootmem(unsigned long addr, unsigned long size)
 213 {
 214         memblock_free(addr, size);
 215 }
 216
 217 static void * __init ___alloc_bootmem_nopanic(unsigned long size,
 218                                         unsigned long align,
 219                                         unsigned long goal,
 220                                         unsigned long limit)
 221 {
 222         void *ptr;
 223
 224         if (WARN_ON_ONCE(slab_is_available()))
 225                 return kzalloc(size, GFP_NOWAIT);
 226
 227 restart:
 228
 229         ptr = __alloc_memory_core_early(NUMA_NO_NODE, size, align, goal, limit);
 230
 231         if (ptr)
 232                 return ptr;
 233
 234         if (goal != 0) {
 235                 goal = 0;
 236                 goto restart;
 237         }
 238
 239         return NULL;
 240 }
 241
 242 /**
 243  * __alloc_bootmem_nopanic - allocate boot memory without panicking
 244  * @size: size of the request in bytes
 245  * @align: alignment of the region
 246  * @goal: preferred starting address of the region
 247  *
 248  * The goal is dropped if it can not be satisfied and the allocation will
 249  * fall back to memory below @goal.
 250  *
 251  * Allocation may happen on any node in the system.
 252  *
 253  * Returns NULL on failure.
 254  */
 255 void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align,
 256                                         unsigned long goal)
 257 {
 258         unsigned long limit = -1UL;
 259
 260         return ___alloc_bootmem_nopanic(size, align, goal, limit);
 261 }
 262
 263 static void * __init ___alloc_bootmem(unsigned long size, unsigned long align,
 264                                         unsigned long goal, unsigned long limit)
 265 {
 266         void *mem = ___alloc_bootmem_nopanic(size, align, goal, limit);
 267
 268         if (mem)
 269                 return mem;
 270         /*
 271          * Whoops, we cannot satisfy the allocation request.
 272          */
 273         printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size);
 274         panic("Out of memory");
 275         return NULL;
 276 }
 277
 278 /**
 279  * __alloc_bootmem - allocate boot memory
 280  * @size: size of the request in bytes
 281  * @align: alignment of the region
 282  * @goal: preferred starting address of the region
 283  *
 284  * The goal is dropped if it can not be satisfied and the allocation will
 285  * fall back to memory below @goal.
 286  *
 287  * Allocation may happen on any node in the system.
 288  *
 289  * The function panics if the request can not be satisfied.
 290  */
 291 void * __init __alloc_bootmem(unsigned long size, unsigned long align,
 292                               unsigned long goal)
 293 {
 294         unsigned long limit = -1UL;
 295
 296         return ___alloc_bootmem(size, align, goal, limit);
 297 }
 298
 299 void * __init ___alloc_bootmem_node_nopanic(pg_data_t *pgdat,
 300                                                    unsigned long size,
 301                                                    unsigned long align,
 302                                                    unsigned long goal,
 303                                                    unsigned long limit)
 304 {
 305         void *ptr;
 306
 307 again:
 308         ptr = __alloc_memory_core_early(pgdat->node_id, size, align,
 309                                         goal, limit);
 310         if (ptr)
 311                 return ptr;
 312
 313         ptr = __alloc_memory_core_early(NUMA_NO_NODE, size, align,
 314                                         goal, limit);
 315         if (ptr)
 316                 return ptr;
 317
 318         if (goal) {
 319                 goal = 0;
 320                 goto again;
 321         }
 322
 323         return NULL;
 324 }
 325
 326 void * __init __alloc_bootmem_node_nopanic(pg_data_t *pgdat, unsigned long size,
 327                                    unsigned long align, unsigned long goal)
 328 {
 329         if (WARN_ON_ONCE(slab_is_available()))
 330                 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
 331
 332         return ___alloc_bootmem_node_nopanic(pgdat, size, align, goal, 0);
 333 }
 334
 335 static void * __init ___alloc_bootmem_node(pg_data_t *pgdat, unsigned long size,
 336                                     unsigned long align, unsigned long goal,
 337                                     unsigned long limit)
 338 {
 339         void *ptr;
 340
 341         ptr = ___alloc_bootmem_node_nopanic(pgdat, size, align, goal, limit);
 342         if (ptr)
 343                 return ptr;
 344
 345         printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size);
 346         panic("Out of memory");
 347         return NULL;
 348 }
 349
 350 /**
 351  * __alloc_bootmem_node - allocate boot memory from a specific node
 352  * @pgdat: node to allocate from
 353  * @size: size of the request in bytes
 354  * @align: alignment of the region
 355  * @goal: preferred starting address of the region
 356  *
 357  * The goal is dropped if it can not be satisfied and the allocation will
 358  * fall back to memory below @goal.
 359  *
 360  * Allocation may fall back to any node in the system if the specified node
 361  * can not hold the requested memory.
 362  *
 363  * The function panics if the request can not be satisfied.
 364  */
 365 void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size,
 366                                    unsigned long align, unsigned long goal)
 367 {
 368         if (WARN_ON_ONCE(slab_is_available()))
 369                 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
 370
 371         return ___alloc_bootmem_node(pgdat, size, align, goal, 0);
 372 }
 373
 374 void * __init __alloc_bootmem_node_high(pg_data_t *pgdat, unsigned long size,
 375                                    unsigned long align, unsigned long goal)
 376 {
 377         return __alloc_bootmem_node(pgdat, size, align, goal);
 378 }
 379
 380 #ifndef ARCH_LOW_ADDRESS_LIMIT
 381 #define ARCH_LOW_ADDRESS_LIMIT  0xffffffffUL
 382 #endif
 383
 384 /**
 385  * __alloc_bootmem_low - allocate low boot memory
 386  * @size: size of the request in bytes
 387  * @align: alignment of the region
 388  * @goal: preferred starting address of the region
 389  *
 390  * The goal is dropped if it can not be satisfied and the allocation will
 391  * fall back to memory below @goal.
 392  *
 393  * Allocation may happen on any node in the system.
 394  *
 395  * The function panics if the request can not be satisfied.
 396  */
 397 void * __init __alloc_bootmem_low(unsigned long size, unsigned long align,
 398                                   unsigned long goal)
 399 {
 400         return ___alloc_bootmem(size, align, goal, ARCH_LOW_ADDRESS_LIMIT);
 401 }
 402
 403 void * __init __alloc_bootmem_low_nopanic(unsigned long size,
 404                                           unsigned long align,
 405                                           unsigned long goal)
 406 {
 407         return ___alloc_bootmem_nopanic(size, align, goal,
 408                                         ARCH_LOW_ADDRESS_LIMIT);
 409 }
 410
 411 /**
 412  * __alloc_bootmem_low_node - allocate low boot memory from a specific node
 413  * @pgdat: node to allocate from
 414  * @size: size of the request in bytes
 415  * @align: alignment of the region
 416  * @goal: preferred starting address of the region
 417  *
 418  * The goal is dropped if it can not be satisfied and the allocation will
 419  * fall back to memory below @goal.
 420  *
 421  * Allocation may fall back to any node in the system if the specified node
 422  * can not hold the requested memory.
 423  *
 424  * The function panics if the request can not be satisfied.
 425  */
 426 void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size,
 427                                        unsigned long align, unsigned long goal)
 428 {
 429         if (WARN_ON_ONCE(slab_is_available()))
 430                 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
 431
 432         return ___alloc_bootmem_node(pgdat, size, align, goal,
 433                                      ARCH_LOW_ADDRESS_LIMIT);
 434 }