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         memblock_reserve(addr, size);
  49         ptr = phys_to_virt(addr);
  50         memset(ptr, 0, 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 static unsigned long __init free_low_memory_core_early(void)
 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 int reset_managed_pages_done __initdata;
 141
 142 static inline void __init reset_node_managed_pages(pg_data_t *pgdat)
 143 {
 144         struct zone *z;
 145
 146         if (reset_managed_pages_done)
 147                 return;
 148         for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
 149                 z->managed_pages = 0;
 150 }
 151
 152 void __init reset_all_zones_managed_pages(void)
 153 {
 154         struct pglist_data *pgdat;
 155
 156         for_each_online_pgdat(pgdat)
 157                 reset_node_managed_pages(pgdat);
 158         reset_managed_pages_done = 1;
 159 }
 160
 161 /**
 162  * free_all_bootmem - release free pages to the buddy allocator
 163  *
 164  * Returns the number of pages actually released.
 165  */
 166 unsigned long __init free_all_bootmem(void)
 167 {
 168         reset_all_zones_managed_pages();
 169
 170         /*
 171          * We need to use MAX_NUMNODES instead of NODE_DATA(0)->node_id
 172          *  because in some case like Node0 doesn't have RAM installed
 173          *  low ram will be on Node1
 174          */
 175         return free_low_memory_core_early();
 176 }
 177
 178 /**
 179  * free_bootmem_node - mark a page range as usable
 180  * @pgdat: node the range resides on
 181  * @physaddr: starting address of the range
 182  * @size: size of the range in bytes
 183  *
 184  * Partial pages will be considered reserved and left as they are.
 185  *
 186  * The range must reside completely on the specified node.
 187  */
 188 void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
 189                               unsigned long size)
 190 {
 191         kmemleak_free_part(__va(physaddr), size);
 192         memblock_free(physaddr, size);
 193 }
 194
 195 /**
 196  * free_bootmem - mark a page range as usable
 197  * @addr: starting address of the range
 198  * @size: size of the range in bytes
 199  *
 200  * Partial pages will be considered reserved and left as they are.
 201  *
 202  * The range must be contiguous but may span node boundaries.
 203  */
 204 void __init free_bootmem(unsigned long addr, unsigned long size)
 205 {
 206         kmemleak_free_part(__va(addr), size);
 207         memblock_free(addr, size);
 208 }
 209
 210 static void * __init ___alloc_bootmem_nopanic(unsigned long size,
 211                                         unsigned long align,
 212                                         unsigned long goal,
 213                                         unsigned long limit)
 214 {
 215         void *ptr;
 216
 217         if (WARN_ON_ONCE(slab_is_available()))
 218                 return kzalloc(size, GFP_NOWAIT);
 219
 220 restart:
 221
 222         ptr = __alloc_memory_core_early(MAX_NUMNODES, size, align, goal, limit);
 223
 224         if (ptr)
 225                 return ptr;
 226
 227         if (goal != 0) {
 228                 goal = 0;
 229                 goto restart;
 230         }
 231
 232         return NULL;
 233 }
 234
 235 /**
 236  * __alloc_bootmem_nopanic - allocate boot memory without panicking
 237  * @size: size of the request in bytes
 238  * @align: alignment of the region
 239  * @goal: preferred starting address of the region
 240  *
 241  * The goal is dropped if it can not be satisfied and the allocation will
 242  * fall back to memory below @goal.
 243  *
 244  * Allocation may happen on any node in the system.
 245  *
 246  * Returns NULL on failure.
 247  */
 248 void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align,
 249                                         unsigned long goal)
 250 {
 251         unsigned long limit = -1UL;
 252
 253         return ___alloc_bootmem_nopanic(size, align, goal, limit);
 254 }
 255
 256 static void * __init ___alloc_bootmem(unsigned long size, unsigned long align,
 257                                         unsigned long goal, unsigned long limit)
 258 {
 259         void *mem = ___alloc_bootmem_nopanic(size, align, goal, limit);
 260
 261         if (mem)
 262                 return mem;
 263         /*
 264          * Whoops, we cannot satisfy the allocation request.
 265          */
 266         printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size);
 267         panic("Out of memory");
 268         return NULL;
 269 }
 270
 271 /**
 272  * __alloc_bootmem - allocate boot memory
 273  * @size: size of the request in bytes
 274  * @align: alignment of the region
 275  * @goal: preferred starting address of the region
 276  *
 277  * The goal is dropped if it can not be satisfied and the allocation will
 278  * fall back to memory below @goal.
 279  *
 280  * Allocation may happen on any node in the system.
 281  *
 282  * The function panics if the request can not be satisfied.
 283  */
 284 void * __init __alloc_bootmem(unsigned long size, unsigned long align,
 285                               unsigned long goal)
 286 {
 287         unsigned long limit = -1UL;
 288
 289         return ___alloc_bootmem(size, align, goal, limit);
 290 }
 291
 292 void * __init ___alloc_bootmem_node_nopanic(pg_data_t *pgdat,
 293                                                    unsigned long size,
 294                                                    unsigned long align,
 295                                                    unsigned long goal,
 296                                                    unsigned long limit)
 297 {
 298         void *ptr;
 299
 300 again:
 301         ptr = __alloc_memory_core_early(pgdat->node_id, size, align,
 302                                         goal, limit);
 303         if (ptr)
 304                 return ptr;
 305
 306         ptr = __alloc_memory_core_early(MAX_NUMNODES, size, align,
 307                                         goal, limit);
 308         if (ptr)
 309                 return ptr;
 310
 311         if (goal) {
 312                 goal = 0;
 313                 goto again;
 314         }
 315
 316         return NULL;
 317 }
 318
 319 void * __init __alloc_bootmem_node_nopanic(pg_data_t *pgdat, unsigned long size,
 320                                    unsigned long align, unsigned long goal)
 321 {
 322         if (WARN_ON_ONCE(slab_is_available()))
 323                 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
 324
 325         return ___alloc_bootmem_node_nopanic(pgdat, size, align, goal, 0);
 326 }
 327
 328 void * __init ___alloc_bootmem_node(pg_data_t *pgdat, unsigned long size,
 329                                     unsigned long align, unsigned long goal,
 330                                     unsigned long limit)
 331 {
 332         void *ptr;
 333
 334         ptr = ___alloc_bootmem_node_nopanic(pgdat, size, align, goal, limit);
 335         if (ptr)
 336                 return ptr;
 337
 338         printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size);
 339         panic("Out of memory");
 340         return NULL;
 341 }
 342
 343 /**
 344  * __alloc_bootmem_node - allocate boot memory from a specific node
 345  * @pgdat: node to allocate from
 346  * @size: size of the request in bytes
 347  * @align: alignment of the region
 348  * @goal: preferred starting address of the region
 349  *
 350  * The goal is dropped if it can not be satisfied and the allocation will
 351  * fall back to memory below @goal.
 352  *
 353  * Allocation may fall back to any node in the system if the specified node
 354  * can not hold the requested memory.
 355  *
 356  * The function panics if the request can not be satisfied.
 357  */
 358 void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size,
 359                                    unsigned long align, unsigned long goal)
 360 {
 361         if (WARN_ON_ONCE(slab_is_available()))
 362                 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
 363
 364         return ___alloc_bootmem_node(pgdat, size, align, goal, 0);
 365 }
 366
 367 void * __init __alloc_bootmem_node_high(pg_data_t *pgdat, unsigned long size,
 368                                    unsigned long align, unsigned long goal)
 369 {
 370         return __alloc_bootmem_node(pgdat, size, align, goal);
 371 }
 372
 373 #ifndef ARCH_LOW_ADDRESS_LIMIT
 374 #define ARCH_LOW_ADDRESS_LIMIT  0xffffffffUL
 375 #endif
 376
 377 /**
 378  * __alloc_bootmem_low - allocate low boot memory
 379  * @size: size of the request in bytes
 380  * @align: alignment of the region
 381  * @goal: preferred starting address of the region
 382  *
 383  * The goal is dropped if it can not be satisfied and the allocation will
 384  * fall back to memory below @goal.
 385  *
 386  * Allocation may happen on any node in the system.
 387  *
 388  * The function panics if the request can not be satisfied.
 389  */
 390 void * __init __alloc_bootmem_low(unsigned long size, unsigned long align,
 391                                   unsigned long goal)
 392 {
 393         return ___alloc_bootmem(size, align, goal, ARCH_LOW_ADDRESS_LIMIT);
 394 }
 395
 396 void * __init __alloc_bootmem_low_nopanic(unsigned long size,
 397                                           unsigned long align,
 398                                           unsigned long goal)
 399 {
 400         return ___alloc_bootmem_nopanic(size, align, goal,
 401                                         ARCH_LOW_ADDRESS_LIMIT);
 402 }
 403
 404 /**
 405  * __alloc_bootmem_low_node - allocate low boot memory from a specific node
 406  * @pgdat: node to allocate from
 407  * @size: size of the request in bytes
 408  * @align: alignment of the region
 409  * @goal: preferred starting address of the region
 410  *
 411  * The goal is dropped if it can not be satisfied and the allocation will
 412  * fall back to memory below @goal.
 413  *
 414  * Allocation may fall back to any node in the system if the specified node
 415  * can not hold the requested memory.
 416  *
 417  * The function panics if the request can not be satisfied.
 418  */
 419 void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size,
 420                                        unsigned long align, unsigned long goal)
 421 {
 422         if (WARN_ON_ONCE(slab_is_available()))
 423                 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
 424
 425         return ___alloc_bootmem_node(pgdat, size, align, goal,
 426                                      ARCH_LOW_ADDRESS_LIMIT);
 427 }