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