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 #ifdef CONFIG_CRASH_DUMP
  36 /*
  37  * If we have booted due to a crash, max_pfn will be a very low value. We need
  38  * to know the amount of memory that the previous kernel used.
  39  */
  40 unsigned long saved_max_pfn;
  41 #endif
  42
  43 static void * __init __alloc_memory_core_early(int nid, u64 size, u64 align,
  44                                         u64 goal, u64 limit)
  45 {
  46         void *ptr;
  47         u64 addr;
  48
  49         if (limit > memblock.current_limit)
  50                 limit = memblock.current_limit;
  51
  52         addr = find_memory_core_early(nid, size, align, goal, limit);
  53
  54         if (addr == MEMBLOCK_ERROR)
  55                 return NULL;
  56
  57         ptr = phys_to_virt(addr);
  58         memset(ptr, 0, size);
  59         memblock_x86_reserve_range(addr, addr + size, "BOOTMEM");
  60         /*
  61          * The min_count is set to 0 so that bootmem allocated blocks
  62          * are never reported as leaks.
  63          */
  64         kmemleak_alloc(ptr, size, 0, 0);
  65         return ptr;
  66 }
  67
  68 /*
  69  * free_bootmem_late - free bootmem pages directly to page allocator
  70  * @addr: starting address of the range
  71  * @size: size of the range in bytes
  72  *
  73  * This is only useful when the bootmem allocator has already been torn
  74  * down, but we are still initializing the system.  Pages are given directly
  75  * to the page allocator, no bootmem metadata is updated because it is gone.
  76  */
  77 void __init free_bootmem_late(unsigned long addr, unsigned long size)
  78 {
  79         unsigned long cursor, end;
  80
  81         kmemleak_free_part(__va(addr), size);
  82
  83         cursor = PFN_UP(addr);
  84         end = PFN_DOWN(addr + size);
  85
  86         for (; cursor < end; cursor++) {
  87                 __free_pages_bootmem(pfn_to_page(cursor), 0);
  88                 totalram_pages++;
  89         }
  90 }
  91
  92 static void __init __free_pages_memory(unsigned long start, unsigned long end)
  93 {
  94         int i;
  95         unsigned long start_aligned, end_aligned;
  96         int order = ilog2(BITS_PER_LONG);
  97
  98         start_aligned = (start + (BITS_PER_LONG - 1)) & ~(BITS_PER_LONG - 1);
  99         end_aligned = end & ~(BITS_PER_LONG - 1);
 100
 101         if (end_aligned <= start_aligned) {
 102                 for (i = start; i < end; i++)
 103                         __free_pages_bootmem(pfn_to_page(i), 0);
 104
 105                 return;
 106         }
 107
 108         for (i = start; i < start_aligned; i++)
 109                 __free_pages_bootmem(pfn_to_page(i), 0);
 110
 111         for (i = start_aligned; i < end_aligned; i += BITS_PER_LONG)
 112                 __free_pages_bootmem(pfn_to_page(i), order);
 113
 114         for (i = end_aligned; i < end; i++)
 115                 __free_pages_bootmem(pfn_to_page(i), 0);
 116 }
 117
 118 unsigned long __init free_all_memory_core_early(int nodeid)
 119 {
 120         int i;
 121         u64 start, end;
 122         unsigned long count = 0;
 123         struct range *range = NULL;
 124         int nr_range;
 125
 126         nr_range = get_free_all_memory_range(&range, nodeid);
 127
 128         for (i = 0; i < nr_range; i++) {
 129                 start = range[i].start;
 130                 end = range[i].end;
 131                 count += end - start;
 132                 __free_pages_memory(start, end);
 133         }
 134
 135         return count;
 136 }
 137
 138 /**
 139  * free_all_bootmem_node - release a node's free pages to the buddy allocator
 140  * @pgdat: node to be released
 141  *
 142  * Returns the number of pages actually released.
 143  */
 144 unsigned long __init free_all_bootmem_node(pg_data_t *pgdat)
 145 {
 146         register_page_bootmem_info_node(pgdat);
 147
 148         /* free_all_memory_core_early(MAX_NUMNODES) will be called later */
 149         return 0;
 150 }
 151
 152 /**
 153  * free_all_bootmem - release free pages to the buddy allocator
 154  *
 155  * Returns the number of pages actually released.
 156  */
 157 unsigned long __init free_all_bootmem(void)
 158 {
 159         /*
 160          * We need to use MAX_NUMNODES instead of NODE_DATA(0)->node_id
 161          *  because in some case like Node0 doesnt have RAM installed
 162          *  low ram will be on Node1
 163          * Use MAX_NUMNODES will make sure all ranges in early_node_map[]
 164          *  will be used instead of only Node0 related
 165          */
 166         return free_all_memory_core_early(MAX_NUMNODES);
 167 }
 168
 169 /**
 170  * free_bootmem_node - mark a page range as usable
 171  * @pgdat: node the range resides on
 172  * @physaddr: starting address of the range
 173  * @size: size of the range in bytes
 174  *
 175  * Partial pages will be considered reserved and left as they are.
 176  *
 177  * The range must reside completely on the specified node.
 178  */
 179 void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
 180                               unsigned long size)
 181 {
 182         kmemleak_free_part(__va(physaddr), size);
 183         memblock_x86_free_range(physaddr, physaddr + size);
 184 }
 185
 186 /**
 187  * free_bootmem - mark a page range as usable
 188  * @addr: starting address of the range
 189  * @size: size of the range in bytes
 190  *
 191  * Partial pages will be considered reserved and left as they are.
 192  *
 193  * The range must be contiguous but may span node boundaries.
 194  */
 195 void __init free_bootmem(unsigned long addr, unsigned long size)
 196 {
 197         kmemleak_free_part(__va(addr), size);
 198         memblock_x86_free_range(addr, addr + size);
 199 }
 200
 201 static void * __init ___alloc_bootmem_nopanic(unsigned long size,
 202                                         unsigned long align,
 203                                         unsigned long goal,
 204                                         unsigned long limit)
 205 {
 206         void *ptr;
 207
 208         if (WARN_ON_ONCE(slab_is_available()))
 209                 return kzalloc(size, GFP_NOWAIT);
 210
 211 restart:
 212
 213         ptr = __alloc_memory_core_early(MAX_NUMNODES, size, align, goal, limit);
 214
 215         if (ptr)
 216                 return ptr;
 217
 218         if (goal != 0) {
 219                 goal = 0;
 220                 goto restart;
 221         }
 222
 223         return NULL;
 224 }
 225
 226 /**
 227  * __alloc_bootmem_nopanic - allocate boot memory without panicking
 228  * @size: size of the request in bytes
 229  * @align: alignment of the region
 230  * @goal: preferred starting address of the region
 231  *
 232  * The goal is dropped if it can not be satisfied and the allocation will
 233  * fall back to memory below @goal.
 234  *
 235  * Allocation may happen on any node in the system.
 236  *
 237  * Returns NULL on failure.
 238  */
 239 void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align,
 240                                         unsigned long goal)
 241 {
 242         unsigned long limit = -1UL;
 243
 244         return ___alloc_bootmem_nopanic(size, align, goal, limit);
 245 }
 246
 247 static void * __init ___alloc_bootmem(unsigned long size, unsigned long align,
 248                                         unsigned long goal, unsigned long limit)
 249 {
 250         void *mem = ___alloc_bootmem_nopanic(size, align, goal, limit);
 251
 252         if (mem)
 253                 return mem;
 254         /*
 255          * Whoops, we cannot satisfy the allocation request.
 256          */
 257         printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size);
 258         panic("Out of memory");
 259         return NULL;
 260 }
 261
 262 /**
 263  * __alloc_bootmem - allocate boot memory
 264  * @size: size of the request in bytes
 265  * @align: alignment of the region
 266  * @goal: preferred starting address of the region
 267  *
 268  * The goal is dropped if it can not be satisfied and the allocation will
 269  * fall back to memory below @goal.
 270  *
 271  * Allocation may happen on any node in the system.
 272  *
 273  * The function panics if the request can not be satisfied.
 274  */
 275 void * __init __alloc_bootmem(unsigned long size, unsigned long align,
 276                               unsigned long goal)
 277 {
 278         unsigned long limit = -1UL;
 279
 280         return ___alloc_bootmem(size, align, goal, limit);
 281 }
 282
 283 /**
 284  * __alloc_bootmem_node - allocate boot memory from a specific node
 285  * @pgdat: node to allocate from
 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 fall back to any node in the system if the specified node
 294  * can not hold the requested memory.
 295  *
 296  * The function panics if the request can not be satisfied.
 297  */
 298 void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size,
 299                                    unsigned long align, unsigned long goal)
 300 {
 301         void *ptr;
 302
 303         if (WARN_ON_ONCE(slab_is_available()))
 304                 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
 305
 306         ptr = __alloc_memory_core_early(pgdat->node_id, size, align,
 307                                          goal, -1ULL);
 308         if (ptr)
 309                 return ptr;
 310
 311         return __alloc_memory_core_early(MAX_NUMNODES, size, align,
 312                                          goal, -1ULL);
 313 }
 314
 315 void * __init __alloc_bootmem_node_high(pg_data_t *pgdat, unsigned long size,
 316                                    unsigned long align, unsigned long goal)
 317 {
 318 #ifdef MAX_DMA32_PFN
 319         unsigned long end_pfn;
 320
 321         if (WARN_ON_ONCE(slab_is_available()))
 322                 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
 323
 324         /* update goal according ...MAX_DMA32_PFN */
 325         end_pfn = pgdat->node_start_pfn + pgdat->node_spanned_pages;
 326
 327         if (end_pfn > MAX_DMA32_PFN + (128 >> (20 - PAGE_SHIFT)) &&
 328             (goal >> PAGE_SHIFT) < MAX_DMA32_PFN) {
 329                 void *ptr;
 330                 unsigned long new_goal;
 331
 332                 new_goal = MAX_DMA32_PFN << PAGE_SHIFT;
 333                 ptr =  __alloc_memory_core_early(pgdat->node_id, size, align,
 334                                                  new_goal, -1ULL);
 335                 if (ptr)
 336                         return ptr;
 337         }
 338 #endif
 339
 340         return __alloc_bootmem_node(pgdat, size, align, goal);
 341
 342 }
 343
 344 #ifdef CONFIG_SPARSEMEM
 345 /**
 346  * alloc_bootmem_section - allocate boot memory from a specific section
 347  * @size: size of the request in bytes
 348  * @section_nr: sparse map section to allocate from
 349  *
 350  * Return NULL on failure.
 351  */
 352 void * __init alloc_bootmem_section(unsigned long size,
 353                                     unsigned long section_nr)
 354 {
 355         unsigned long pfn, goal, limit;
 356
 357         pfn = section_nr_to_pfn(section_nr);
 358         goal = pfn << PAGE_SHIFT;
 359         limit = section_nr_to_pfn(section_nr + 1) << PAGE_SHIFT;
 360
 361         return __alloc_memory_core_early(early_pfn_to_nid(pfn), size,
 362                                          SMP_CACHE_BYTES, goal, limit);
 363 }
 364 #endif
 365
 366 void * __init __alloc_bootmem_node_nopanic(pg_data_t *pgdat, unsigned long size,
 367                                    unsigned long align, unsigned long goal)
 368 {
 369         void *ptr;
 370
 371         if (WARN_ON_ONCE(slab_is_available()))
 372                 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
 373
 374         ptr =  __alloc_memory_core_early(pgdat->node_id, size, align,
 375                                                  goal, -1ULL);
 376         if (ptr)
 377                 return ptr;
 378
 379         return __alloc_bootmem_nopanic(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 /**
 406  * __alloc_bootmem_low_node - allocate low boot memory from a specific node
 407  * @pgdat: node to allocate from
 408  * @size: size of the request in bytes
 409  * @align: alignment of the region
 410  * @goal: preferred starting address of the region
 411  *
 412  * The goal is dropped if it can not be satisfied and the allocation will
 413  * fall back to memory below @goal.
 414  *
 415  * Allocation may fall back to any node in the system if the specified node
 416  * can not hold the requested memory.
 417  *
 418  * The function panics if the request can not be satisfied.
 419  */
 420 void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size,
 421                                        unsigned long align, unsigned long goal)
 422 {
 423         void *ptr;
 424
 425         if (WARN_ON_ONCE(slab_is_available()))
 426                 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
 427
 428         ptr = __alloc_memory_core_early(pgdat->node_id, size, align,
 429                                 goal, ARCH_LOW_ADDRESS_LIMIT);
 430         if (ptr)
 431                 return ptr;
 432
 433         return  __alloc_memory_core_early(MAX_NUMNODES, size, align,
 434                                 goal, ARCH_LOW_ADDRESS_LIMIT);
 435 }