| 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/export.h> |
| 15 | #include <linux/kmemleak.h> |
| 16 | #include <linux/range.h> |
| 17 | #include <linux/memblock.h> |
| 18 | #include <linux/bootmem.h> |
| 19 | |
| 20 | #include <asm/bug.h> |
| 21 | #include <asm/io.h> |
| 22 | |
| 23 | #include "internal.h" |
| 24 | |
| 25 | #ifndef CONFIG_HAVE_MEMBLOCK |
| 26 | #error CONFIG_HAVE_MEMBLOCK not defined |
| 27 | #endif |
| 28 | |
| 29 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
| 30 | struct pglist_data __refdata contig_page_data; |
| 31 | EXPORT_SYMBOL(contig_page_data); |
| 32 | #endif |
| 33 | |
| 34 | unsigned long max_low_pfn; |
| 35 | unsigned long min_low_pfn; |
| 36 | unsigned long max_pfn; |
| 37 | unsigned long long max_possible_pfn; |
| 38 | |
| 39 | static void * __init __alloc_memory_core_early(int nid, u64 size, u64 align, |
| 40 | u64 goal, u64 limit) |
| 41 | { |
| 42 | void *ptr; |
| 43 | u64 addr; |
| 44 | ulong flags = choose_memblock_flags(); |
| 45 | |
| 46 | if (limit > memblock.current_limit) |
| 47 | limit = memblock.current_limit; |
| 48 | |
| 49 | again: |
| 50 | addr = memblock_find_in_range_node(size, align, goal, limit, nid, |
| 51 | flags); |
| 52 | if (!addr && (flags & MEMBLOCK_MIRROR)) { |
| 53 | flags &= ~MEMBLOCK_MIRROR; |
| 54 | pr_warn("Could not allocate %pap bytes of mirrored memory\n", |
| 55 | &size); |
| 56 | goto again; |
| 57 | } |
| 58 | if (!addr) |
| 59 | return NULL; |
| 60 | |
| 61 | if (memblock_reserve(addr, size)) |
| 62 | return NULL; |
| 63 | |
| 64 | ptr = phys_to_virt(addr); |
| 65 | memset(ptr, 0, size); |
| 66 | /* |
| 67 | * The min_count is set to 0 so that bootmem allocated blocks |
| 68 | * are never reported as leaks. |
| 69 | */ |
| 70 | kmemleak_alloc(ptr, size, 0, 0); |
| 71 | return ptr; |
| 72 | } |
| 73 | |
| 74 | /* |
| 75 | * free_bootmem_late - free bootmem pages directly to page allocator |
| 76 | * @addr: starting address of the range |
| 77 | * @size: size of the range in bytes |
| 78 | * |
| 79 | * This is only useful when the bootmem allocator has already been torn |
| 80 | * down, but we are still initializing the system. Pages are given directly |
| 81 | * to the page allocator, no bootmem metadata is updated because it is gone. |
| 82 | */ |
| 83 | void __init free_bootmem_late(unsigned long addr, unsigned long size) |
| 84 | { |
| 85 | unsigned long cursor, end; |
| 86 | |
| 87 | kmemleak_free_part_phys(addr, size); |
| 88 | |
| 89 | cursor = PFN_UP(addr); |
| 90 | end = PFN_DOWN(addr + size); |
| 91 | |
| 92 | for (; cursor < end; cursor++) { |
| 93 | __free_pages_bootmem(pfn_to_page(cursor), cursor, 0); |
| 94 | totalram_pages++; |
| 95 | } |
| 96 | } |
| 97 | |
| 98 | static void __init __free_pages_memory(unsigned long start, unsigned long end) |
| 99 | { |
| 100 | int order; |
| 101 | |
| 102 | while (start < end) { |
| 103 | order = min(MAX_ORDER - 1UL, __ffs(start)); |
| 104 | |
| 105 | while (start + (1UL << order) > end) |
| 106 | order--; |
| 107 | |
| 108 | __free_pages_bootmem(pfn_to_page(start), start, order); |
| 109 | |
| 110 | start += (1UL << order); |
| 111 | } |
| 112 | } |
| 113 | |
| 114 | static unsigned long __init __free_memory_core(phys_addr_t start, |
| 115 | phys_addr_t end) |
| 116 | { |
| 117 | unsigned long start_pfn = PFN_UP(start); |
| 118 | unsigned long end_pfn = min_t(unsigned long, |
| 119 | PFN_DOWN(end), max_low_pfn); |
| 120 | |
| 121 | if (start_pfn > end_pfn) |
| 122 | return 0; |
| 123 | |
| 124 | __free_pages_memory(start_pfn, end_pfn); |
| 125 | |
| 126 | return end_pfn - start_pfn; |
| 127 | } |
| 128 | |
| 129 | static unsigned long __init free_low_memory_core_early(void) |
| 130 | { |
| 131 | unsigned long count = 0; |
| 132 | phys_addr_t start, end; |
| 133 | u64 i; |
| 134 | |
| 135 | memblock_clear_hotplug(0, -1); |
| 136 | |
| 137 | for_each_reserved_mem_region(i, &start, &end) |
| 138 | reserve_bootmem_region(start, end); |
| 139 | |
| 140 | /* |
| 141 | * We need to use NUMA_NO_NODE instead of NODE_DATA(0)->node_id |
| 142 | * because in some case like Node0 doesn't have RAM installed |
| 143 | * low ram will be on Node1 |
| 144 | */ |
| 145 | for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE, &start, &end, |
| 146 | NULL) |
| 147 | count += __free_memory_core(start, end); |
| 148 | |
| 149 | #ifdef CONFIG_ARCH_DISCARD_MEMBLOCK |
| 150 | { |
| 151 | phys_addr_t size; |
| 152 | |
| 153 | /* Free memblock.reserved array if it was allocated */ |
| 154 | size = get_allocated_memblock_reserved_regions_info(&start); |
| 155 | if (size) |
| 156 | count += __free_memory_core(start, start + size); |
| 157 | |
| 158 | /* Free memblock.memory array if it was allocated */ |
| 159 | size = get_allocated_memblock_memory_regions_info(&start); |
| 160 | if (size) |
| 161 | count += __free_memory_core(start, start + size); |
| 162 | } |
| 163 | #endif |
| 164 | |
| 165 | return count; |
| 166 | } |
| 167 | |
| 168 | static int reset_managed_pages_done __initdata; |
| 169 | |
| 170 | void reset_node_managed_pages(pg_data_t *pgdat) |
| 171 | { |
| 172 | struct zone *z; |
| 173 | |
| 174 | for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++) |
| 175 | z->managed_pages = 0; |
| 176 | } |
| 177 | |
| 178 | void __init reset_all_zones_managed_pages(void) |
| 179 | { |
| 180 | struct pglist_data *pgdat; |
| 181 | |
| 182 | if (reset_managed_pages_done) |
| 183 | return; |
| 184 | |
| 185 | for_each_online_pgdat(pgdat) |
| 186 | reset_node_managed_pages(pgdat); |
| 187 | |
| 188 | reset_managed_pages_done = 1; |
| 189 | } |
| 190 | |
| 191 | /** |
| 192 | * free_all_bootmem - release free pages to the buddy allocator |
| 193 | * |
| 194 | * Returns the number of pages actually released. |
| 195 | */ |
| 196 | unsigned long __init free_all_bootmem(void) |
| 197 | { |
| 198 | unsigned long pages; |
| 199 | |
| 200 | reset_all_zones_managed_pages(); |
| 201 | |
| 202 | pages = free_low_memory_core_early(); |
| 203 | totalram_pages += pages; |
| 204 | |
| 205 | return pages; |
| 206 | } |
| 207 | |
| 208 | /** |
| 209 | * free_bootmem_node - mark a page range as usable |
| 210 | * @pgdat: node the range resides on |
| 211 | * @physaddr: starting address of the range |
| 212 | * @size: size of the range in bytes |
| 213 | * |
| 214 | * Partial pages will be considered reserved and left as they are. |
| 215 | * |
| 216 | * The range must reside completely on the specified node. |
| 217 | */ |
| 218 | void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr, |
| 219 | unsigned long size) |
| 220 | { |
| 221 | memblock_free(physaddr, size); |
| 222 | } |
| 223 | |
| 224 | /** |
| 225 | * free_bootmem - mark a page range as usable |
| 226 | * @addr: starting address of the range |
| 227 | * @size: size of the range in bytes |
| 228 | * |
| 229 | * Partial pages will be considered reserved and left as they are. |
| 230 | * |
| 231 | * The range must be contiguous but may span node boundaries. |
| 232 | */ |
| 233 | void __init free_bootmem(unsigned long addr, unsigned long size) |
| 234 | { |
| 235 | memblock_free(addr, size); |
| 236 | } |
| 237 | |
| 238 | static void * __init ___alloc_bootmem_nopanic(unsigned long size, |
| 239 | unsigned long align, |
| 240 | unsigned long goal, |
| 241 | unsigned long limit) |
| 242 | { |
| 243 | void *ptr; |
| 244 | |
| 245 | if (WARN_ON_ONCE(slab_is_available())) |
| 246 | return kzalloc(size, GFP_NOWAIT); |
| 247 | |
| 248 | restart: |
| 249 | |
| 250 | ptr = __alloc_memory_core_early(NUMA_NO_NODE, size, align, goal, limit); |
| 251 | |
| 252 | if (ptr) |
| 253 | return ptr; |
| 254 | |
| 255 | if (goal != 0) { |
| 256 | goal = 0; |
| 257 | goto restart; |
| 258 | } |
| 259 | |
| 260 | return NULL; |
| 261 | } |
| 262 | |
| 263 | /** |
| 264 | * __alloc_bootmem_nopanic - allocate boot memory without panicking |
| 265 | * @size: size of the request in bytes |
| 266 | * @align: alignment of the region |
| 267 | * @goal: preferred starting address of the region |
| 268 | * |
| 269 | * The goal is dropped if it can not be satisfied and the allocation will |
| 270 | * fall back to memory below @goal. |
| 271 | * |
| 272 | * Allocation may happen on any node in the system. |
| 273 | * |
| 274 | * Returns NULL on failure. |
| 275 | */ |
| 276 | void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align, |
| 277 | unsigned long goal) |
| 278 | { |
| 279 | unsigned long limit = -1UL; |
| 280 | |
| 281 | return ___alloc_bootmem_nopanic(size, align, goal, limit); |
| 282 | } |
| 283 | |
| 284 | static void * __init ___alloc_bootmem(unsigned long size, unsigned long align, |
| 285 | unsigned long goal, unsigned long limit) |
| 286 | { |
| 287 | void *mem = ___alloc_bootmem_nopanic(size, align, goal, limit); |
| 288 | |
| 289 | if (mem) |
| 290 | return mem; |
| 291 | /* |
| 292 | * Whoops, we cannot satisfy the allocation request. |
| 293 | */ |
| 294 | pr_alert("bootmem alloc of %lu bytes failed!\n", size); |
| 295 | panic("Out of memory"); |
| 296 | return NULL; |
| 297 | } |
| 298 | |
| 299 | /** |
| 300 | * __alloc_bootmem - allocate boot memory |
| 301 | * @size: size of the request in bytes |
| 302 | * @align: alignment of the region |
| 303 | * @goal: preferred starting address of the region |
| 304 | * |
| 305 | * The goal is dropped if it can not be satisfied and the allocation will |
| 306 | * fall back to memory below @goal. |
| 307 | * |
| 308 | * Allocation may happen on any node in the system. |
| 309 | * |
| 310 | * The function panics if the request can not be satisfied. |
| 311 | */ |
| 312 | void * __init __alloc_bootmem(unsigned long size, unsigned long align, |
| 313 | unsigned long goal) |
| 314 | { |
| 315 | unsigned long limit = -1UL; |
| 316 | |
| 317 | return ___alloc_bootmem(size, align, goal, limit); |
| 318 | } |
| 319 | |
| 320 | void * __init ___alloc_bootmem_node_nopanic(pg_data_t *pgdat, |
| 321 | unsigned long size, |
| 322 | unsigned long align, |
| 323 | unsigned long goal, |
| 324 | unsigned long limit) |
| 325 | { |
| 326 | void *ptr; |
| 327 | |
| 328 | again: |
| 329 | ptr = __alloc_memory_core_early(pgdat->node_id, size, align, |
| 330 | goal, limit); |
| 331 | if (ptr) |
| 332 | return ptr; |
| 333 | |
| 334 | ptr = __alloc_memory_core_early(NUMA_NO_NODE, size, align, |
| 335 | goal, limit); |
| 336 | if (ptr) |
| 337 | return ptr; |
| 338 | |
| 339 | if (goal) { |
| 340 | goal = 0; |
| 341 | goto again; |
| 342 | } |
| 343 | |
| 344 | return NULL; |
| 345 | } |
| 346 | |
| 347 | void * __init __alloc_bootmem_node_nopanic(pg_data_t *pgdat, unsigned long size, |
| 348 | unsigned long align, unsigned long goal) |
| 349 | { |
| 350 | if (WARN_ON_ONCE(slab_is_available())) |
| 351 | return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); |
| 352 | |
| 353 | return ___alloc_bootmem_node_nopanic(pgdat, size, align, goal, 0); |
| 354 | } |
| 355 | |
| 356 | static void * __init ___alloc_bootmem_node(pg_data_t *pgdat, unsigned long size, |
| 357 | unsigned long align, unsigned long goal, |
| 358 | unsigned long limit) |
| 359 | { |
| 360 | void *ptr; |
| 361 | |
| 362 | ptr = ___alloc_bootmem_node_nopanic(pgdat, size, align, goal, limit); |
| 363 | if (ptr) |
| 364 | return ptr; |
| 365 | |
| 366 | pr_alert("bootmem alloc of %lu bytes failed!\n", size); |
| 367 | panic("Out of memory"); |
| 368 | return NULL; |
| 369 | } |
| 370 | |
| 371 | /** |
| 372 | * __alloc_bootmem_node - allocate boot memory from a specific node |
| 373 | * @pgdat: node to allocate from |
| 374 | * @size: size of the request in bytes |
| 375 | * @align: alignment of the region |
| 376 | * @goal: preferred starting address of the region |
| 377 | * |
| 378 | * The goal is dropped if it can not be satisfied and the allocation will |
| 379 | * fall back to memory below @goal. |
| 380 | * |
| 381 | * Allocation may fall back to any node in the system if the specified node |
| 382 | * can not hold the requested memory. |
| 383 | * |
| 384 | * The function panics if the request can not be satisfied. |
| 385 | */ |
| 386 | void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size, |
| 387 | unsigned long align, unsigned long goal) |
| 388 | { |
| 389 | if (WARN_ON_ONCE(slab_is_available())) |
| 390 | return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); |
| 391 | |
| 392 | return ___alloc_bootmem_node(pgdat, size, align, goal, 0); |
| 393 | } |
| 394 | |
| 395 | void * __init __alloc_bootmem_node_high(pg_data_t *pgdat, unsigned long size, |
| 396 | unsigned long align, unsigned long goal) |
| 397 | { |
| 398 | return __alloc_bootmem_node(pgdat, size, align, goal); |
| 399 | } |
| 400 | |
| 401 | |
| 402 | /** |
| 403 | * __alloc_bootmem_low - allocate low boot memory |
| 404 | * @size: size of the request in bytes |
| 405 | * @align: alignment of the region |
| 406 | * @goal: preferred starting address of the region |
| 407 | * |
| 408 | * The goal is dropped if it can not be satisfied and the allocation will |
| 409 | * fall back to memory below @goal. |
| 410 | * |
| 411 | * Allocation may happen on any node in the system. |
| 412 | * |
| 413 | * The function panics if the request can not be satisfied. |
| 414 | */ |
| 415 | void * __init __alloc_bootmem_low(unsigned long size, unsigned long align, |
| 416 | unsigned long goal) |
| 417 | { |
| 418 | return ___alloc_bootmem(size, align, goal, ARCH_LOW_ADDRESS_LIMIT); |
| 419 | } |
| 420 | |
| 421 | void * __init __alloc_bootmem_low_nopanic(unsigned long size, |
| 422 | unsigned long align, |
| 423 | unsigned long goal) |
| 424 | { |
| 425 | return ___alloc_bootmem_nopanic(size, align, goal, |
| 426 | ARCH_LOW_ADDRESS_LIMIT); |
| 427 | } |
| 428 | |
| 429 | /** |
| 430 | * __alloc_bootmem_low_node - allocate low boot memory from a specific node |
| 431 | * @pgdat: node to allocate from |
| 432 | * @size: size of the request in bytes |
| 433 | * @align: alignment of the region |
| 434 | * @goal: preferred starting address of the region |
| 435 | * |
| 436 | * The goal is dropped if it can not be satisfied and the allocation will |
| 437 | * fall back to memory below @goal. |
| 438 | * |
| 439 | * Allocation may fall back to any node in the system if the specified node |
| 440 | * can not hold the requested memory. |
| 441 | * |
| 442 | * The function panics if the request can not be satisfied. |
| 443 | */ |
| 444 | void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size, |
| 445 | unsigned long align, unsigned long goal) |
| 446 | { |
| 447 | if (WARN_ON_ONCE(slab_is_available())) |
| 448 | return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); |
| 449 | |
| 450 | return ___alloc_bootmem_node(pgdat, size, align, goal, |
| 451 | ARCH_LOW_ADDRESS_LIMIT); |
| 452 | } |