| 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 | .bdata = &bootmem_node_data[0] |
| 29 | }; |
| 30 | EXPORT_SYMBOL(contig_page_data); |
| 31 | #endif |
| 32 | |
| 33 | unsigned long max_low_pfn; |
| 34 | unsigned long min_low_pfn; |
| 35 | unsigned long max_pfn; |
| 36 | |
| 37 | bootmem_data_t bootmem_node_data[MAX_NUMNODES] __initdata; |
| 38 | |
| 39 | static struct list_head bdata_list __initdata = LIST_HEAD_INIT(bdata_list); |
| 40 | |
| 41 | static int bootmem_debug; |
| 42 | |
| 43 | static int __init bootmem_debug_setup(char *buf) |
| 44 | { |
| 45 | bootmem_debug = 1; |
| 46 | return 0; |
| 47 | } |
| 48 | early_param("bootmem_debug", bootmem_debug_setup); |
| 49 | |
| 50 | #define bdebug(fmt, args...) ({ \ |
| 51 | if (unlikely(bootmem_debug)) \ |
| 52 | printk(KERN_INFO \ |
| 53 | "bootmem::%s " fmt, \ |
| 54 | __func__, ## args); \ |
| 55 | }) |
| 56 | |
| 57 | static unsigned long __init bootmap_bytes(unsigned long pages) |
| 58 | { |
| 59 | unsigned long bytes = DIV_ROUND_UP(pages, 8); |
| 60 | |
| 61 | return ALIGN(bytes, sizeof(long)); |
| 62 | } |
| 63 | |
| 64 | /** |
| 65 | * bootmem_bootmap_pages - calculate bitmap size in pages |
| 66 | * @pages: number of pages the bitmap has to represent |
| 67 | */ |
| 68 | unsigned long __init bootmem_bootmap_pages(unsigned long pages) |
| 69 | { |
| 70 | unsigned long bytes = bootmap_bytes(pages); |
| 71 | |
| 72 | return PAGE_ALIGN(bytes) >> PAGE_SHIFT; |
| 73 | } |
| 74 | |
| 75 | /* |
| 76 | * link bdata in order |
| 77 | */ |
| 78 | static void __init link_bootmem(bootmem_data_t *bdata) |
| 79 | { |
| 80 | bootmem_data_t *ent; |
| 81 | |
| 82 | list_for_each_entry(ent, &bdata_list, list) { |
| 83 | if (bdata->node_min_pfn < ent->node_min_pfn) { |
| 84 | list_add_tail(&bdata->list, &ent->list); |
| 85 | return; |
| 86 | } |
| 87 | } |
| 88 | |
| 89 | list_add_tail(&bdata->list, &bdata_list); |
| 90 | } |
| 91 | |
| 92 | /* |
| 93 | * Called once to set up the allocator itself. |
| 94 | */ |
| 95 | static unsigned long __init init_bootmem_core(bootmem_data_t *bdata, |
| 96 | unsigned long mapstart, unsigned long start, unsigned long end) |
| 97 | { |
| 98 | unsigned long mapsize; |
| 99 | |
| 100 | mminit_validate_memmodel_limits(&start, &end); |
| 101 | bdata->node_bootmem_map = phys_to_virt(PFN_PHYS(mapstart)); |
| 102 | bdata->node_min_pfn = start; |
| 103 | bdata->node_low_pfn = end; |
| 104 | link_bootmem(bdata); |
| 105 | |
| 106 | /* |
| 107 | * Initially all pages are reserved - setup_arch() has to |
| 108 | * register free RAM areas explicitly. |
| 109 | */ |
| 110 | mapsize = bootmap_bytes(end - start); |
| 111 | memset(bdata->node_bootmem_map, 0xff, mapsize); |
| 112 | |
| 113 | bdebug("nid=%td start=%lx map=%lx end=%lx mapsize=%lx\n", |
| 114 | bdata - bootmem_node_data, start, mapstart, end, mapsize); |
| 115 | |
| 116 | return mapsize; |
| 117 | } |
| 118 | |
| 119 | /** |
| 120 | * init_bootmem_node - register a node as boot memory |
| 121 | * @pgdat: node to register |
| 122 | * @freepfn: pfn where the bitmap for this node is to be placed |
| 123 | * @startpfn: first pfn on the node |
| 124 | * @endpfn: first pfn after the node |
| 125 | * |
| 126 | * Returns the number of bytes needed to hold the bitmap for this node. |
| 127 | */ |
| 128 | unsigned long __init init_bootmem_node(pg_data_t *pgdat, unsigned long freepfn, |
| 129 | unsigned long startpfn, unsigned long endpfn) |
| 130 | { |
| 131 | return init_bootmem_core(pgdat->bdata, freepfn, startpfn, endpfn); |
| 132 | } |
| 133 | |
| 134 | /** |
| 135 | * init_bootmem - register boot memory |
| 136 | * @start: pfn where the bitmap is to be placed |
| 137 | * @pages: number of available physical pages |
| 138 | * |
| 139 | * Returns the number of bytes needed to hold the bitmap. |
| 140 | */ |
| 141 | unsigned long __init init_bootmem(unsigned long start, unsigned long pages) |
| 142 | { |
| 143 | max_low_pfn = pages; |
| 144 | min_low_pfn = start; |
| 145 | return init_bootmem_core(NODE_DATA(0)->bdata, start, 0, pages); |
| 146 | } |
| 147 | |
| 148 | /* |
| 149 | * free_bootmem_late - free bootmem pages directly to page allocator |
| 150 | * @addr: starting physical address of the range |
| 151 | * @size: size of the range in bytes |
| 152 | * |
| 153 | * This is only useful when the bootmem allocator has already been torn |
| 154 | * down, but we are still initializing the system. Pages are given directly |
| 155 | * to the page allocator, no bootmem metadata is updated because it is gone. |
| 156 | */ |
| 157 | void __init free_bootmem_late(unsigned long physaddr, unsigned long size) |
| 158 | { |
| 159 | unsigned long cursor, end; |
| 160 | |
| 161 | kmemleak_free_part(__va(physaddr), size); |
| 162 | |
| 163 | cursor = PFN_UP(physaddr); |
| 164 | end = PFN_DOWN(physaddr + size); |
| 165 | |
| 166 | for (; cursor < end; cursor++) { |
| 167 | __free_pages_bootmem(pfn_to_page(cursor), 0); |
| 168 | totalram_pages++; |
| 169 | } |
| 170 | } |
| 171 | |
| 172 | static unsigned long __init free_all_bootmem_core(bootmem_data_t *bdata) |
| 173 | { |
| 174 | struct page *page; |
| 175 | unsigned long start, end, pages, count = 0; |
| 176 | |
| 177 | if (!bdata->node_bootmem_map) |
| 178 | return 0; |
| 179 | |
| 180 | start = bdata->node_min_pfn; |
| 181 | end = bdata->node_low_pfn; |
| 182 | |
| 183 | bdebug("nid=%td start=%lx end=%lx\n", |
| 184 | bdata - bootmem_node_data, start, end); |
| 185 | |
| 186 | while (start < end) { |
| 187 | unsigned long *map, idx, vec; |
| 188 | unsigned shift; |
| 189 | |
| 190 | map = bdata->node_bootmem_map; |
| 191 | idx = start - bdata->node_min_pfn; |
| 192 | shift = idx & (BITS_PER_LONG - 1); |
| 193 | /* |
| 194 | * vec holds at most BITS_PER_LONG map bits, |
| 195 | * bit 0 corresponds to start. |
| 196 | */ |
| 197 | vec = ~map[idx / BITS_PER_LONG]; |
| 198 | |
| 199 | if (shift) { |
| 200 | vec >>= shift; |
| 201 | if (end - start >= BITS_PER_LONG) |
| 202 | vec |= ~map[idx / BITS_PER_LONG + 1] << |
| 203 | (BITS_PER_LONG - shift); |
| 204 | } |
| 205 | /* |
| 206 | * If we have a properly aligned and fully unreserved |
| 207 | * BITS_PER_LONG block of pages in front of us, free |
| 208 | * it in one go. |
| 209 | */ |
| 210 | if (IS_ALIGNED(start, BITS_PER_LONG) && vec == ~0UL) { |
| 211 | int order = ilog2(BITS_PER_LONG); |
| 212 | |
| 213 | __free_pages_bootmem(pfn_to_page(start), order); |
| 214 | count += BITS_PER_LONG; |
| 215 | start += BITS_PER_LONG; |
| 216 | } else { |
| 217 | unsigned long cur = start; |
| 218 | |
| 219 | start = ALIGN(start + 1, BITS_PER_LONG); |
| 220 | while (vec && cur != start) { |
| 221 | if (vec & 1) { |
| 222 | page = pfn_to_page(cur); |
| 223 | __free_pages_bootmem(page, 0); |
| 224 | count++; |
| 225 | } |
| 226 | vec >>= 1; |
| 227 | ++cur; |
| 228 | } |
| 229 | } |
| 230 | } |
| 231 | |
| 232 | page = virt_to_page(bdata->node_bootmem_map); |
| 233 | pages = bdata->node_low_pfn - bdata->node_min_pfn; |
| 234 | pages = bootmem_bootmap_pages(pages); |
| 235 | count += pages; |
| 236 | while (pages--) |
| 237 | __free_pages_bootmem(page++, 0); |
| 238 | |
| 239 | bdebug("nid=%td released=%lx\n", bdata - bootmem_node_data, count); |
| 240 | |
| 241 | return count; |
| 242 | } |
| 243 | |
| 244 | static void reset_node_lowmem_managed_pages(pg_data_t *pgdat) |
| 245 | { |
| 246 | struct zone *z; |
| 247 | |
| 248 | /* |
| 249 | * In free_area_init_core(), highmem zone's managed_pages is set to |
| 250 | * present_pages, and bootmem allocator doesn't allocate from highmem |
| 251 | * zones. So there's no need to recalculate managed_pages because all |
| 252 | * highmem pages will be managed by the buddy system. Here highmem |
| 253 | * zone also includes highmem movable zone. |
| 254 | */ |
| 255 | for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++) |
| 256 | if (!is_highmem(z)) |
| 257 | z->managed_pages = 0; |
| 258 | } |
| 259 | |
| 260 | /** |
| 261 | * free_all_bootmem_node - release a node's free pages to the buddy allocator |
| 262 | * @pgdat: node to be released |
| 263 | * |
| 264 | * Returns the number of pages actually released. |
| 265 | */ |
| 266 | unsigned long __init free_all_bootmem_node(pg_data_t *pgdat) |
| 267 | { |
| 268 | register_page_bootmem_info_node(pgdat); |
| 269 | reset_node_lowmem_managed_pages(pgdat); |
| 270 | return free_all_bootmem_core(pgdat->bdata); |
| 271 | } |
| 272 | |
| 273 | /** |
| 274 | * free_all_bootmem - release free pages to the buddy allocator |
| 275 | * |
| 276 | * Returns the number of pages actually released. |
| 277 | */ |
| 278 | unsigned long __init free_all_bootmem(void) |
| 279 | { |
| 280 | unsigned long total_pages = 0; |
| 281 | bootmem_data_t *bdata; |
| 282 | struct pglist_data *pgdat; |
| 283 | |
| 284 | for_each_online_pgdat(pgdat) |
| 285 | reset_node_lowmem_managed_pages(pgdat); |
| 286 | |
| 287 | list_for_each_entry(bdata, &bdata_list, list) |
| 288 | total_pages += free_all_bootmem_core(bdata); |
| 289 | |
| 290 | return total_pages; |
| 291 | } |
| 292 | |
| 293 | static void __init __free(bootmem_data_t *bdata, |
| 294 | unsigned long sidx, unsigned long eidx) |
| 295 | { |
| 296 | unsigned long idx; |
| 297 | |
| 298 | bdebug("nid=%td start=%lx end=%lx\n", bdata - bootmem_node_data, |
| 299 | sidx + bdata->node_min_pfn, |
| 300 | eidx + bdata->node_min_pfn); |
| 301 | |
| 302 | if (bdata->hint_idx > sidx) |
| 303 | bdata->hint_idx = sidx; |
| 304 | |
| 305 | for (idx = sidx; idx < eidx; idx++) |
| 306 | if (!test_and_clear_bit(idx, bdata->node_bootmem_map)) |
| 307 | BUG(); |
| 308 | } |
| 309 | |
| 310 | static int __init __reserve(bootmem_data_t *bdata, unsigned long sidx, |
| 311 | unsigned long eidx, int flags) |
| 312 | { |
| 313 | unsigned long idx; |
| 314 | int exclusive = flags & BOOTMEM_EXCLUSIVE; |
| 315 | |
| 316 | bdebug("nid=%td start=%lx end=%lx flags=%x\n", |
| 317 | bdata - bootmem_node_data, |
| 318 | sidx + bdata->node_min_pfn, |
| 319 | eidx + bdata->node_min_pfn, |
| 320 | flags); |
| 321 | |
| 322 | for (idx = sidx; idx < eidx; idx++) |
| 323 | if (test_and_set_bit(idx, bdata->node_bootmem_map)) { |
| 324 | if (exclusive) { |
| 325 | __free(bdata, sidx, idx); |
| 326 | return -EBUSY; |
| 327 | } |
| 328 | bdebug("silent double reserve of PFN %lx\n", |
| 329 | idx + bdata->node_min_pfn); |
| 330 | } |
| 331 | return 0; |
| 332 | } |
| 333 | |
| 334 | static int __init mark_bootmem_node(bootmem_data_t *bdata, |
| 335 | unsigned long start, unsigned long end, |
| 336 | int reserve, int flags) |
| 337 | { |
| 338 | unsigned long sidx, eidx; |
| 339 | |
| 340 | bdebug("nid=%td start=%lx end=%lx reserve=%d flags=%x\n", |
| 341 | bdata - bootmem_node_data, start, end, reserve, flags); |
| 342 | |
| 343 | BUG_ON(start < bdata->node_min_pfn); |
| 344 | BUG_ON(end > bdata->node_low_pfn); |
| 345 | |
| 346 | sidx = start - bdata->node_min_pfn; |
| 347 | eidx = end - bdata->node_min_pfn; |
| 348 | |
| 349 | if (reserve) |
| 350 | return __reserve(bdata, sidx, eidx, flags); |
| 351 | else |
| 352 | __free(bdata, sidx, eidx); |
| 353 | return 0; |
| 354 | } |
| 355 | |
| 356 | static int __init mark_bootmem(unsigned long start, unsigned long end, |
| 357 | int reserve, int flags) |
| 358 | { |
| 359 | unsigned long pos; |
| 360 | bootmem_data_t *bdata; |
| 361 | |
| 362 | pos = start; |
| 363 | list_for_each_entry(bdata, &bdata_list, list) { |
| 364 | int err; |
| 365 | unsigned long max; |
| 366 | |
| 367 | if (pos < bdata->node_min_pfn || |
| 368 | pos >= bdata->node_low_pfn) { |
| 369 | BUG_ON(pos != start); |
| 370 | continue; |
| 371 | } |
| 372 | |
| 373 | max = min(bdata->node_low_pfn, end); |
| 374 | |
| 375 | err = mark_bootmem_node(bdata, pos, max, reserve, flags); |
| 376 | if (reserve && err) { |
| 377 | mark_bootmem(start, pos, 0, 0); |
| 378 | return err; |
| 379 | } |
| 380 | |
| 381 | if (max == end) |
| 382 | return 0; |
| 383 | pos = bdata->node_low_pfn; |
| 384 | } |
| 385 | BUG(); |
| 386 | } |
| 387 | |
| 388 | /** |
| 389 | * free_bootmem_node - mark a page range as usable |
| 390 | * @pgdat: node the range resides on |
| 391 | * @physaddr: starting address of the range |
| 392 | * @size: size of the range in bytes |
| 393 | * |
| 394 | * Partial pages will be considered reserved and left as they are. |
| 395 | * |
| 396 | * The range must reside completely on the specified node. |
| 397 | */ |
| 398 | void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr, |
| 399 | unsigned long size) |
| 400 | { |
| 401 | unsigned long start, end; |
| 402 | |
| 403 | kmemleak_free_part(__va(physaddr), size); |
| 404 | |
| 405 | start = PFN_UP(physaddr); |
| 406 | end = PFN_DOWN(physaddr + size); |
| 407 | |
| 408 | mark_bootmem_node(pgdat->bdata, start, end, 0, 0); |
| 409 | } |
| 410 | |
| 411 | /** |
| 412 | * free_bootmem - mark a page range as usable |
| 413 | * @addr: starting physical address of the range |
| 414 | * @size: size of the range in bytes |
| 415 | * |
| 416 | * Partial pages will be considered reserved and left as they are. |
| 417 | * |
| 418 | * The range must be contiguous but may span node boundaries. |
| 419 | */ |
| 420 | void __init free_bootmem(unsigned long physaddr, unsigned long size) |
| 421 | { |
| 422 | unsigned long start, end; |
| 423 | |
| 424 | kmemleak_free_part(__va(physaddr), size); |
| 425 | |
| 426 | start = PFN_UP(physaddr); |
| 427 | end = PFN_DOWN(physaddr + size); |
| 428 | |
| 429 | mark_bootmem(start, end, 0, 0); |
| 430 | } |
| 431 | |
| 432 | /** |
| 433 | * reserve_bootmem_node - mark a page range as reserved |
| 434 | * @pgdat: node the range resides on |
| 435 | * @physaddr: starting address of the range |
| 436 | * @size: size of the range in bytes |
| 437 | * @flags: reservation flags (see linux/bootmem.h) |
| 438 | * |
| 439 | * Partial pages will be reserved. |
| 440 | * |
| 441 | * The range must reside completely on the specified node. |
| 442 | */ |
| 443 | int __init reserve_bootmem_node(pg_data_t *pgdat, unsigned long physaddr, |
| 444 | unsigned long size, int flags) |
| 445 | { |
| 446 | unsigned long start, end; |
| 447 | |
| 448 | start = PFN_DOWN(physaddr); |
| 449 | end = PFN_UP(physaddr + size); |
| 450 | |
| 451 | return mark_bootmem_node(pgdat->bdata, start, end, 1, flags); |
| 452 | } |
| 453 | |
| 454 | /** |
| 455 | * reserve_bootmem - mark a page range as reserved |
| 456 | * @addr: starting address of the range |
| 457 | * @size: size of the range in bytes |
| 458 | * @flags: reservation flags (see linux/bootmem.h) |
| 459 | * |
| 460 | * Partial pages will be reserved. |
| 461 | * |
| 462 | * The range must be contiguous but may span node boundaries. |
| 463 | */ |
| 464 | int __init reserve_bootmem(unsigned long addr, unsigned long size, |
| 465 | int flags) |
| 466 | { |
| 467 | unsigned long start, end; |
| 468 | |
| 469 | start = PFN_DOWN(addr); |
| 470 | end = PFN_UP(addr + size); |
| 471 | |
| 472 | return mark_bootmem(start, end, 1, flags); |
| 473 | } |
| 474 | |
| 475 | static unsigned long __init align_idx(struct bootmem_data *bdata, |
| 476 | unsigned long idx, unsigned long step) |
| 477 | { |
| 478 | unsigned long base = bdata->node_min_pfn; |
| 479 | |
| 480 | /* |
| 481 | * Align the index with respect to the node start so that the |
| 482 | * combination of both satisfies the requested alignment. |
| 483 | */ |
| 484 | |
| 485 | return ALIGN(base + idx, step) - base; |
| 486 | } |
| 487 | |
| 488 | static unsigned long __init align_off(struct bootmem_data *bdata, |
| 489 | unsigned long off, unsigned long align) |
| 490 | { |
| 491 | unsigned long base = PFN_PHYS(bdata->node_min_pfn); |
| 492 | |
| 493 | /* Same as align_idx for byte offsets */ |
| 494 | |
| 495 | return ALIGN(base + off, align) - base; |
| 496 | } |
| 497 | |
| 498 | static void * __init alloc_bootmem_bdata(struct bootmem_data *bdata, |
| 499 | unsigned long size, unsigned long align, |
| 500 | unsigned long goal, unsigned long limit) |
| 501 | { |
| 502 | unsigned long fallback = 0; |
| 503 | unsigned long min, max, start, sidx, midx, step; |
| 504 | |
| 505 | bdebug("nid=%td size=%lx [%lu pages] align=%lx goal=%lx limit=%lx\n", |
| 506 | bdata - bootmem_node_data, size, PAGE_ALIGN(size) >> PAGE_SHIFT, |
| 507 | align, goal, limit); |
| 508 | |
| 509 | BUG_ON(!size); |
| 510 | BUG_ON(align & (align - 1)); |
| 511 | BUG_ON(limit && goal + size > limit); |
| 512 | |
| 513 | if (!bdata->node_bootmem_map) |
| 514 | return NULL; |
| 515 | |
| 516 | min = bdata->node_min_pfn; |
| 517 | max = bdata->node_low_pfn; |
| 518 | |
| 519 | goal >>= PAGE_SHIFT; |
| 520 | limit >>= PAGE_SHIFT; |
| 521 | |
| 522 | if (limit && max > limit) |
| 523 | max = limit; |
| 524 | if (max <= min) |
| 525 | return NULL; |
| 526 | |
| 527 | step = max(align >> PAGE_SHIFT, 1UL); |
| 528 | |
| 529 | if (goal && min < goal && goal < max) |
| 530 | start = ALIGN(goal, step); |
| 531 | else |
| 532 | start = ALIGN(min, step); |
| 533 | |
| 534 | sidx = start - bdata->node_min_pfn; |
| 535 | midx = max - bdata->node_min_pfn; |
| 536 | |
| 537 | if (bdata->hint_idx > sidx) { |
| 538 | /* |
| 539 | * Handle the valid case of sidx being zero and still |
| 540 | * catch the fallback below. |
| 541 | */ |
| 542 | fallback = sidx + 1; |
| 543 | sidx = align_idx(bdata, bdata->hint_idx, step); |
| 544 | } |
| 545 | |
| 546 | while (1) { |
| 547 | int merge; |
| 548 | void *region; |
| 549 | unsigned long eidx, i, start_off, end_off; |
| 550 | find_block: |
| 551 | sidx = find_next_zero_bit(bdata->node_bootmem_map, midx, sidx); |
| 552 | sidx = align_idx(bdata, sidx, step); |
| 553 | eidx = sidx + PFN_UP(size); |
| 554 | |
| 555 | if (sidx >= midx || eidx > midx) |
| 556 | break; |
| 557 | |
| 558 | for (i = sidx; i < eidx; i++) |
| 559 | if (test_bit(i, bdata->node_bootmem_map)) { |
| 560 | sidx = align_idx(bdata, i, step); |
| 561 | if (sidx == i) |
| 562 | sidx += step; |
| 563 | goto find_block; |
| 564 | } |
| 565 | |
| 566 | if (bdata->last_end_off & (PAGE_SIZE - 1) && |
| 567 | PFN_DOWN(bdata->last_end_off) + 1 == sidx) |
| 568 | start_off = align_off(bdata, bdata->last_end_off, align); |
| 569 | else |
| 570 | start_off = PFN_PHYS(sidx); |
| 571 | |
| 572 | merge = PFN_DOWN(start_off) < sidx; |
| 573 | end_off = start_off + size; |
| 574 | |
| 575 | bdata->last_end_off = end_off; |
| 576 | bdata->hint_idx = PFN_UP(end_off); |
| 577 | |
| 578 | /* |
| 579 | * Reserve the area now: |
| 580 | */ |
| 581 | if (__reserve(bdata, PFN_DOWN(start_off) + merge, |
| 582 | PFN_UP(end_off), BOOTMEM_EXCLUSIVE)) |
| 583 | BUG(); |
| 584 | |
| 585 | region = phys_to_virt(PFN_PHYS(bdata->node_min_pfn) + |
| 586 | start_off); |
| 587 | memset(region, 0, size); |
| 588 | /* |
| 589 | * The min_count is set to 0 so that bootmem allocated blocks |
| 590 | * are never reported as leaks. |
| 591 | */ |
| 592 | kmemleak_alloc(region, size, 0, 0); |
| 593 | return region; |
| 594 | } |
| 595 | |
| 596 | if (fallback) { |
| 597 | sidx = align_idx(bdata, fallback - 1, step); |
| 598 | fallback = 0; |
| 599 | goto find_block; |
| 600 | } |
| 601 | |
| 602 | return NULL; |
| 603 | } |
| 604 | |
| 605 | static void * __init alloc_bootmem_core(unsigned long size, |
| 606 | unsigned long align, |
| 607 | unsigned long goal, |
| 608 | unsigned long limit) |
| 609 | { |
| 610 | bootmem_data_t *bdata; |
| 611 | void *region; |
| 612 | |
| 613 | if (WARN_ON_ONCE(slab_is_available())) |
| 614 | return kzalloc(size, GFP_NOWAIT); |
| 615 | |
| 616 | list_for_each_entry(bdata, &bdata_list, list) { |
| 617 | if (goal && bdata->node_low_pfn <= PFN_DOWN(goal)) |
| 618 | continue; |
| 619 | if (limit && bdata->node_min_pfn >= PFN_DOWN(limit)) |
| 620 | break; |
| 621 | |
| 622 | region = alloc_bootmem_bdata(bdata, size, align, goal, limit); |
| 623 | if (region) |
| 624 | return region; |
| 625 | } |
| 626 | |
| 627 | return NULL; |
| 628 | } |
| 629 | |
| 630 | static void * __init ___alloc_bootmem_nopanic(unsigned long size, |
| 631 | unsigned long align, |
| 632 | unsigned long goal, |
| 633 | unsigned long limit) |
| 634 | { |
| 635 | void *ptr; |
| 636 | |
| 637 | restart: |
| 638 | ptr = alloc_bootmem_core(size, align, goal, limit); |
| 639 | if (ptr) |
| 640 | return ptr; |
| 641 | if (goal) { |
| 642 | goal = 0; |
| 643 | goto restart; |
| 644 | } |
| 645 | |
| 646 | return NULL; |
| 647 | } |
| 648 | |
| 649 | /** |
| 650 | * __alloc_bootmem_nopanic - allocate boot memory without panicking |
| 651 | * @size: size of the request in bytes |
| 652 | * @align: alignment of the region |
| 653 | * @goal: preferred starting address of the region |
| 654 | * |
| 655 | * The goal is dropped if it can not be satisfied and the allocation will |
| 656 | * fall back to memory below @goal. |
| 657 | * |
| 658 | * Allocation may happen on any node in the system. |
| 659 | * |
| 660 | * Returns NULL on failure. |
| 661 | */ |
| 662 | void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align, |
| 663 | unsigned long goal) |
| 664 | { |
| 665 | unsigned long limit = 0; |
| 666 | |
| 667 | return ___alloc_bootmem_nopanic(size, align, goal, limit); |
| 668 | } |
| 669 | |
| 670 | static void * __init ___alloc_bootmem(unsigned long size, unsigned long align, |
| 671 | unsigned long goal, unsigned long limit) |
| 672 | { |
| 673 | void *mem = ___alloc_bootmem_nopanic(size, align, goal, limit); |
| 674 | |
| 675 | if (mem) |
| 676 | return mem; |
| 677 | /* |
| 678 | * Whoops, we cannot satisfy the allocation request. |
| 679 | */ |
| 680 | printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size); |
| 681 | panic("Out of memory"); |
| 682 | return NULL; |
| 683 | } |
| 684 | |
| 685 | /** |
| 686 | * __alloc_bootmem - allocate boot memory |
| 687 | * @size: size of the request in bytes |
| 688 | * @align: alignment of the region |
| 689 | * @goal: preferred starting address of the region |
| 690 | * |
| 691 | * The goal is dropped if it can not be satisfied and the allocation will |
| 692 | * fall back to memory below @goal. |
| 693 | * |
| 694 | * Allocation may happen on any node in the system. |
| 695 | * |
| 696 | * The function panics if the request can not be satisfied. |
| 697 | */ |
| 698 | void * __init __alloc_bootmem(unsigned long size, unsigned long align, |
| 699 | unsigned long goal) |
| 700 | { |
| 701 | unsigned long limit = 0; |
| 702 | |
| 703 | return ___alloc_bootmem(size, align, goal, limit); |
| 704 | } |
| 705 | |
| 706 | void * __init ___alloc_bootmem_node_nopanic(pg_data_t *pgdat, |
| 707 | unsigned long size, unsigned long align, |
| 708 | unsigned long goal, unsigned long limit) |
| 709 | { |
| 710 | void *ptr; |
| 711 | |
| 712 | if (WARN_ON_ONCE(slab_is_available())) |
| 713 | return kzalloc(size, GFP_NOWAIT); |
| 714 | again: |
| 715 | |
| 716 | /* do not panic in alloc_bootmem_bdata() */ |
| 717 | if (limit && goal + size > limit) |
| 718 | limit = 0; |
| 719 | |
| 720 | ptr = alloc_bootmem_bdata(pgdat->bdata, size, align, goal, limit); |
| 721 | if (ptr) |
| 722 | return ptr; |
| 723 | |
| 724 | ptr = alloc_bootmem_core(size, align, goal, limit); |
| 725 | if (ptr) |
| 726 | return ptr; |
| 727 | |
| 728 | if (goal) { |
| 729 | goal = 0; |
| 730 | goto again; |
| 731 | } |
| 732 | |
| 733 | return NULL; |
| 734 | } |
| 735 | |
| 736 | void * __init __alloc_bootmem_node_nopanic(pg_data_t *pgdat, unsigned long size, |
| 737 | unsigned long align, unsigned long goal) |
| 738 | { |
| 739 | if (WARN_ON_ONCE(slab_is_available())) |
| 740 | return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); |
| 741 | |
| 742 | return ___alloc_bootmem_node_nopanic(pgdat, size, align, goal, 0); |
| 743 | } |
| 744 | |
| 745 | void * __init ___alloc_bootmem_node(pg_data_t *pgdat, unsigned long size, |
| 746 | unsigned long align, unsigned long goal, |
| 747 | unsigned long limit) |
| 748 | { |
| 749 | void *ptr; |
| 750 | |
| 751 | ptr = ___alloc_bootmem_node_nopanic(pgdat, size, align, goal, 0); |
| 752 | if (ptr) |
| 753 | return ptr; |
| 754 | |
| 755 | printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size); |
| 756 | panic("Out of memory"); |
| 757 | return NULL; |
| 758 | } |
| 759 | |
| 760 | /** |
| 761 | * __alloc_bootmem_node - allocate boot memory from a specific node |
| 762 | * @pgdat: node to allocate from |
| 763 | * @size: size of the request in bytes |
| 764 | * @align: alignment of the region |
| 765 | * @goal: preferred starting address of the region |
| 766 | * |
| 767 | * The goal is dropped if it can not be satisfied and the allocation will |
| 768 | * fall back to memory below @goal. |
| 769 | * |
| 770 | * Allocation may fall back to any node in the system if the specified node |
| 771 | * can not hold the requested memory. |
| 772 | * |
| 773 | * The function panics if the request can not be satisfied. |
| 774 | */ |
| 775 | void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size, |
| 776 | unsigned long align, unsigned long goal) |
| 777 | { |
| 778 | if (WARN_ON_ONCE(slab_is_available())) |
| 779 | return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); |
| 780 | |
| 781 | return ___alloc_bootmem_node(pgdat, size, align, goal, 0); |
| 782 | } |
| 783 | |
| 784 | void * __init __alloc_bootmem_node_high(pg_data_t *pgdat, unsigned long size, |
| 785 | unsigned long align, unsigned long goal) |
| 786 | { |
| 787 | #ifdef MAX_DMA32_PFN |
| 788 | unsigned long end_pfn; |
| 789 | |
| 790 | if (WARN_ON_ONCE(slab_is_available())) |
| 791 | return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); |
| 792 | |
| 793 | /* update goal according ...MAX_DMA32_PFN */ |
| 794 | end_pfn = pgdat->node_start_pfn + pgdat->node_spanned_pages; |
| 795 | |
| 796 | if (end_pfn > MAX_DMA32_PFN + (128 >> (20 - PAGE_SHIFT)) && |
| 797 | (goal >> PAGE_SHIFT) < MAX_DMA32_PFN) { |
| 798 | void *ptr; |
| 799 | unsigned long new_goal; |
| 800 | |
| 801 | new_goal = MAX_DMA32_PFN << PAGE_SHIFT; |
| 802 | ptr = alloc_bootmem_bdata(pgdat->bdata, size, align, |
| 803 | new_goal, 0); |
| 804 | if (ptr) |
| 805 | return ptr; |
| 806 | } |
| 807 | #endif |
| 808 | |
| 809 | return __alloc_bootmem_node(pgdat, size, align, goal); |
| 810 | |
| 811 | } |
| 812 | |
| 813 | #ifndef ARCH_LOW_ADDRESS_LIMIT |
| 814 | #define ARCH_LOW_ADDRESS_LIMIT 0xffffffffUL |
| 815 | #endif |
| 816 | |
| 817 | /** |
| 818 | * __alloc_bootmem_low - allocate low boot memory |
| 819 | * @size: size of the request in bytes |
| 820 | * @align: alignment of the region |
| 821 | * @goal: preferred starting address of the region |
| 822 | * |
| 823 | * The goal is dropped if it can not be satisfied and the allocation will |
| 824 | * fall back to memory below @goal. |
| 825 | * |
| 826 | * Allocation may happen on any node in the system. |
| 827 | * |
| 828 | * The function panics if the request can not be satisfied. |
| 829 | */ |
| 830 | void * __init __alloc_bootmem_low(unsigned long size, unsigned long align, |
| 831 | unsigned long goal) |
| 832 | { |
| 833 | return ___alloc_bootmem(size, align, goal, ARCH_LOW_ADDRESS_LIMIT); |
| 834 | } |
| 835 | |
| 836 | void * __init __alloc_bootmem_low_nopanic(unsigned long size, |
| 837 | unsigned long align, |
| 838 | unsigned long goal) |
| 839 | { |
| 840 | return ___alloc_bootmem_nopanic(size, align, goal, |
| 841 | ARCH_LOW_ADDRESS_LIMIT); |
| 842 | } |
| 843 | |
| 844 | /** |
| 845 | * __alloc_bootmem_low_node - allocate low boot memory from a specific node |
| 846 | * @pgdat: node to allocate from |
| 847 | * @size: size of the request in bytes |
| 848 | * @align: alignment of the region |
| 849 | * @goal: preferred starting address of the region |
| 850 | * |
| 851 | * The goal is dropped if it can not be satisfied and the allocation will |
| 852 | * fall back to memory below @goal. |
| 853 | * |
| 854 | * Allocation may fall back to any node in the system if the specified node |
| 855 | * can not hold the requested memory. |
| 856 | * |
| 857 | * The function panics if the request can not be satisfied. |
| 858 | */ |
| 859 | void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size, |
| 860 | unsigned long align, unsigned long goal) |
| 861 | { |
| 862 | if (WARN_ON_ONCE(slab_is_available())) |
| 863 | return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); |
| 864 | |
| 865 | return ___alloc_bootmem_node(pgdat, size, align, |
| 866 | goal, ARCH_LOW_ADDRESS_LIMIT); |
| 867 | } |