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