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(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_reserve(addr, size);
  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         unsigned long i, start_aligned, end_aligned;
  86         int order = ilog2(BITS_PER_LONG);
  87
  88         start_aligned = (start + (BITS_PER_LONG - 1)) & ~(BITS_PER_LONG - 1);
  89         end_aligned = end & ~(BITS_PER_LONG - 1);
  90
  91         if (end_aligned <= start_aligned) {
  92                 for (i = start; i < end; i++)
  93                         __free_pages_bootmem(pfn_to_page(i), 0);
  94
  95                 return;
  96         }
  97
  98         for (i = start; i < start_aligned; i++)
  99                 __free_pages_bootmem(pfn_to_page(i), 0);
 100
 101         for (i = start_aligned; i < end_aligned; i += BITS_PER_LONG)
 102                 __free_pages_bootmem(pfn_to_page(i), order);
 103
 104         for (i = end_aligned; i < end; i++)
 105                 __free_pages_bootmem(pfn_to_page(i), 0);
 106 }
 107
 108 static unsigned long __init __free_memory_core(phys_addr_t start,
 109                                  phys_addr_t end)
 110 {
 111         unsigned long start_pfn = PFN_UP(start);
 112         unsigned long end_pfn = min_t(unsigned long,
 113                                       PFN_DOWN(end), max_low_pfn);
 114
 115         if (start_pfn > end_pfn)
 116                 return 0;
 117
 118         __free_pages_memory(start_pfn, end_pfn);
 119
 120         return end_pfn - start_pfn;
 121 }
 122
 123 unsigned long __init free_low_memory_core_early(int nodeid)
 124 {
 125         unsigned long count = 0;
 126         phys_addr_t start, end, size;
 127         u64 i;
 128
 129         for_each_free_mem_range(i, MAX_NUMNODES, &start, &end, NULL)
 130                 count += __free_memory_core(start, end);
 131
 132         /* free range that is used for reserved array if we allocate it */
 133         size = get_allocated_memblock_reserved_regions_info(&start);
 134         if (size)
 135                 count += __free_memory_core(start, start + size);
 136
 137         return count;
 138 }
 139
 140 static void reset_node_lowmem_managed_pages(pg_data_t *pgdat)
 141 {
 142         struct zone *z;
 143
 144         /*
 145          * In free_area_init_core(), highmem zone's managed_pages is set to
 146          * present_pages, and bootmem allocator doesn't allocate from highmem
 147          * zones. So there's no need to recalculate managed_pages because all
 148          * highmem pages will be managed by the buddy system. Here highmem
 149          * zone also includes highmem movable zone.
 150          */
 151         for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
 152                 if (!is_highmem(z))
 153                         z->managed_pages = 0;
 154 }
 155
 156 /**
 157  * free_all_bootmem_node - release a node's free pages to the buddy allocator
 158  * @pgdat: node to be released
 159  *
 160  * Returns the number of pages actually released.
 161  */
 162 unsigned long __init free_all_bootmem_node(pg_data_t *pgdat)
 163 {
 164         register_page_bootmem_info_node(pgdat);
 165         reset_node_lowmem_managed_pages(pgdat);
 166
 167         /* free_low_memory_core_early(MAX_NUMNODES) will be called later */
 168         return 0;
 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         struct pglist_data *pgdat;
 179
 180         for_each_online_pgdat(pgdat)
 181                 reset_node_lowmem_managed_pages(pgdat);
 182
 183         /*
 184          * We need to use MAX_NUMNODES instead of NODE_DATA(0)->node_id
 185          *  because in some case like Node0 doesn't have RAM installed
 186          *  low ram will be on Node1
 187          */
 188         return free_low_memory_core_early(MAX_NUMNODES);
 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         kmemleak_free_part(__va(physaddr), size);
 205         memblock_free(physaddr, size);
 206 }
 207
 208 /**
 209  * free_bootmem - mark a page range as usable
 210  * @addr: starting address of the range
 211  * @size: size of the range in bytes
 212  *
 213  * Partial pages will be considered reserved and left as they are.
 214  *
 215  * The range must be contiguous but may span node boundaries.
 216  */
 217 void __init free_bootmem(unsigned long addr, unsigned long size)
 218 {
 219         kmemleak_free_part(__va(addr), size);
 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(MAX_NUMNODES, 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(MAX_NUMNODES, 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 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 /**
 410  * __alloc_bootmem_low_node - allocate low boot memory from a specific node
 411  * @pgdat: node to allocate from
 412  * @size: size of the request in bytes
 413  * @align: alignment of the region
 414  * @goal: preferred starting address of the region
 415  *
 416  * The goal is dropped if it can not be satisfied and the allocation will
 417  * fall back to memory below @goal.
 418  *
 419  * Allocation may fall back to any node in the system if the specified node
 420  * can not hold the requested memory.
 421  *
 422  * The function panics if the request can not be satisfied.
 423  */
 424 void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size,
 425                                        unsigned long align, unsigned long goal)
 426 {
 427         if (WARN_ON_ONCE(slab_is_available()))
 428                 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
 429
 430         return ___alloc_bootmem_node(pgdat, size, align, goal,
 431                                      ARCH_LOW_ADDRESS_LIMIT);
 432 }