Yinghai Lu | 95f72d1 | 2010-07-12 14:36:09 +1000 | [diff] [blame^] | 1 | /* |
| 2 | * Procedures for maintaining information about logical memory blocks. |
| 3 | * |
| 4 | * Peter Bergner, IBM Corp. June 2001. |
| 5 | * Copyright (C) 2001 Peter Bergner. |
| 6 | * |
| 7 | * This program is free software; you can redistribute it and/or |
| 8 | * modify it under the terms of the GNU General Public License |
| 9 | * as published by the Free Software Foundation; either version |
| 10 | * 2 of the License, or (at your option) any later version. |
| 11 | */ |
| 12 | |
| 13 | #include <linux/kernel.h> |
| 14 | #include <linux/init.h> |
| 15 | #include <linux/bitops.h> |
| 16 | #include <linux/memblock.h> |
| 17 | |
| 18 | #define MEMBLOCK_ALLOC_ANYWHERE 0 |
| 19 | |
| 20 | struct memblock memblock; |
| 21 | |
| 22 | static int memblock_debug; |
| 23 | |
| 24 | static int __init early_memblock(char *p) |
| 25 | { |
| 26 | if (p && strstr(p, "debug")) |
| 27 | memblock_debug = 1; |
| 28 | return 0; |
| 29 | } |
| 30 | early_param("memblock", early_memblock); |
| 31 | |
| 32 | static void memblock_dump(struct memblock_region *region, char *name) |
| 33 | { |
| 34 | unsigned long long base, size; |
| 35 | int i; |
| 36 | |
| 37 | pr_info(" %s.cnt = 0x%lx\n", name, region->cnt); |
| 38 | |
| 39 | for (i = 0; i < region->cnt; i++) { |
| 40 | base = region->region[i].base; |
| 41 | size = region->region[i].size; |
| 42 | |
| 43 | pr_info(" %s[0x%x]\t0x%016llx - 0x%016llx, 0x%llx bytes\n", |
| 44 | name, i, base, base + size - 1, size); |
| 45 | } |
| 46 | } |
| 47 | |
| 48 | void memblock_dump_all(void) |
| 49 | { |
| 50 | if (!memblock_debug) |
| 51 | return; |
| 52 | |
| 53 | pr_info("MEMBLOCK configuration:\n"); |
| 54 | pr_info(" rmo_size = 0x%llx\n", (unsigned long long)memblock.rmo_size); |
| 55 | pr_info(" memory.size = 0x%llx\n", (unsigned long long)memblock.memory.size); |
| 56 | |
| 57 | memblock_dump(&memblock.memory, "memory"); |
| 58 | memblock_dump(&memblock.reserved, "reserved"); |
| 59 | } |
| 60 | |
| 61 | static unsigned long memblock_addrs_overlap(u64 base1, u64 size1, u64 base2, |
| 62 | u64 size2) |
| 63 | { |
| 64 | return ((base1 < (base2 + size2)) && (base2 < (base1 + size1))); |
| 65 | } |
| 66 | |
| 67 | static long memblock_addrs_adjacent(u64 base1, u64 size1, u64 base2, u64 size2) |
| 68 | { |
| 69 | if (base2 == base1 + size1) |
| 70 | return 1; |
| 71 | else if (base1 == base2 + size2) |
| 72 | return -1; |
| 73 | |
| 74 | return 0; |
| 75 | } |
| 76 | |
| 77 | static long memblock_regions_adjacent(struct memblock_region *rgn, |
| 78 | unsigned long r1, unsigned long r2) |
| 79 | { |
| 80 | u64 base1 = rgn->region[r1].base; |
| 81 | u64 size1 = rgn->region[r1].size; |
| 82 | u64 base2 = rgn->region[r2].base; |
| 83 | u64 size2 = rgn->region[r2].size; |
| 84 | |
| 85 | return memblock_addrs_adjacent(base1, size1, base2, size2); |
| 86 | } |
| 87 | |
| 88 | static void memblock_remove_region(struct memblock_region *rgn, unsigned long r) |
| 89 | { |
| 90 | unsigned long i; |
| 91 | |
| 92 | for (i = r; i < rgn->cnt - 1; i++) { |
| 93 | rgn->region[i].base = rgn->region[i + 1].base; |
| 94 | rgn->region[i].size = rgn->region[i + 1].size; |
| 95 | } |
| 96 | rgn->cnt--; |
| 97 | } |
| 98 | |
| 99 | /* Assumption: base addr of region 1 < base addr of region 2 */ |
| 100 | static void memblock_coalesce_regions(struct memblock_region *rgn, |
| 101 | unsigned long r1, unsigned long r2) |
| 102 | { |
| 103 | rgn->region[r1].size += rgn->region[r2].size; |
| 104 | memblock_remove_region(rgn, r2); |
| 105 | } |
| 106 | |
| 107 | void __init memblock_init(void) |
| 108 | { |
| 109 | /* Create a dummy zero size MEMBLOCK which will get coalesced away later. |
| 110 | * This simplifies the memblock_add() code below... |
| 111 | */ |
| 112 | memblock.memory.region[0].base = 0; |
| 113 | memblock.memory.region[0].size = 0; |
| 114 | memblock.memory.cnt = 1; |
| 115 | |
| 116 | /* Ditto. */ |
| 117 | memblock.reserved.region[0].base = 0; |
| 118 | memblock.reserved.region[0].size = 0; |
| 119 | memblock.reserved.cnt = 1; |
| 120 | } |
| 121 | |
| 122 | void __init memblock_analyze(void) |
| 123 | { |
| 124 | int i; |
| 125 | |
| 126 | memblock.memory.size = 0; |
| 127 | |
| 128 | for (i = 0; i < memblock.memory.cnt; i++) |
| 129 | memblock.memory.size += memblock.memory.region[i].size; |
| 130 | } |
| 131 | |
| 132 | static long memblock_add_region(struct memblock_region *rgn, u64 base, u64 size) |
| 133 | { |
| 134 | unsigned long coalesced = 0; |
| 135 | long adjacent, i; |
| 136 | |
| 137 | if ((rgn->cnt == 1) && (rgn->region[0].size == 0)) { |
| 138 | rgn->region[0].base = base; |
| 139 | rgn->region[0].size = size; |
| 140 | return 0; |
| 141 | } |
| 142 | |
| 143 | /* First try and coalesce this MEMBLOCK with another. */ |
| 144 | for (i = 0; i < rgn->cnt; i++) { |
| 145 | u64 rgnbase = rgn->region[i].base; |
| 146 | u64 rgnsize = rgn->region[i].size; |
| 147 | |
| 148 | if ((rgnbase == base) && (rgnsize == size)) |
| 149 | /* Already have this region, so we're done */ |
| 150 | return 0; |
| 151 | |
| 152 | adjacent = memblock_addrs_adjacent(base, size, rgnbase, rgnsize); |
| 153 | if (adjacent > 0) { |
| 154 | rgn->region[i].base -= size; |
| 155 | rgn->region[i].size += size; |
| 156 | coalesced++; |
| 157 | break; |
| 158 | } else if (adjacent < 0) { |
| 159 | rgn->region[i].size += size; |
| 160 | coalesced++; |
| 161 | break; |
| 162 | } |
| 163 | } |
| 164 | |
| 165 | if ((i < rgn->cnt - 1) && memblock_regions_adjacent(rgn, i, i+1)) { |
| 166 | memblock_coalesce_regions(rgn, i, i+1); |
| 167 | coalesced++; |
| 168 | } |
| 169 | |
| 170 | if (coalesced) |
| 171 | return coalesced; |
| 172 | if (rgn->cnt >= MAX_MEMBLOCK_REGIONS) |
| 173 | return -1; |
| 174 | |
| 175 | /* Couldn't coalesce the MEMBLOCK, so add it to the sorted table. */ |
| 176 | for (i = rgn->cnt - 1; i >= 0; i--) { |
| 177 | if (base < rgn->region[i].base) { |
| 178 | rgn->region[i+1].base = rgn->region[i].base; |
| 179 | rgn->region[i+1].size = rgn->region[i].size; |
| 180 | } else { |
| 181 | rgn->region[i+1].base = base; |
| 182 | rgn->region[i+1].size = size; |
| 183 | break; |
| 184 | } |
| 185 | } |
| 186 | |
| 187 | if (base < rgn->region[0].base) { |
| 188 | rgn->region[0].base = base; |
| 189 | rgn->region[0].size = size; |
| 190 | } |
| 191 | rgn->cnt++; |
| 192 | |
| 193 | return 0; |
| 194 | } |
| 195 | |
| 196 | long memblock_add(u64 base, u64 size) |
| 197 | { |
| 198 | struct memblock_region *_rgn = &memblock.memory; |
| 199 | |
| 200 | /* On pSeries LPAR systems, the first MEMBLOCK is our RMO region. */ |
| 201 | if (base == 0) |
| 202 | memblock.rmo_size = size; |
| 203 | |
| 204 | return memblock_add_region(_rgn, base, size); |
| 205 | |
| 206 | } |
| 207 | |
| 208 | static long __memblock_remove(struct memblock_region *rgn, u64 base, u64 size) |
| 209 | { |
| 210 | u64 rgnbegin, rgnend; |
| 211 | u64 end = base + size; |
| 212 | int i; |
| 213 | |
| 214 | rgnbegin = rgnend = 0; /* supress gcc warnings */ |
| 215 | |
| 216 | /* Find the region where (base, size) belongs to */ |
| 217 | for (i=0; i < rgn->cnt; i++) { |
| 218 | rgnbegin = rgn->region[i].base; |
| 219 | rgnend = rgnbegin + rgn->region[i].size; |
| 220 | |
| 221 | if ((rgnbegin <= base) && (end <= rgnend)) |
| 222 | break; |
| 223 | } |
| 224 | |
| 225 | /* Didn't find the region */ |
| 226 | if (i == rgn->cnt) |
| 227 | return -1; |
| 228 | |
| 229 | /* Check to see if we are removing entire region */ |
| 230 | if ((rgnbegin == base) && (rgnend == end)) { |
| 231 | memblock_remove_region(rgn, i); |
| 232 | return 0; |
| 233 | } |
| 234 | |
| 235 | /* Check to see if region is matching at the front */ |
| 236 | if (rgnbegin == base) { |
| 237 | rgn->region[i].base = end; |
| 238 | rgn->region[i].size -= size; |
| 239 | return 0; |
| 240 | } |
| 241 | |
| 242 | /* Check to see if the region is matching at the end */ |
| 243 | if (rgnend == end) { |
| 244 | rgn->region[i].size -= size; |
| 245 | return 0; |
| 246 | } |
| 247 | |
| 248 | /* |
| 249 | * We need to split the entry - adjust the current one to the |
| 250 | * beginging of the hole and add the region after hole. |
| 251 | */ |
| 252 | rgn->region[i].size = base - rgn->region[i].base; |
| 253 | return memblock_add_region(rgn, end, rgnend - end); |
| 254 | } |
| 255 | |
| 256 | long memblock_remove(u64 base, u64 size) |
| 257 | { |
| 258 | return __memblock_remove(&memblock.memory, base, size); |
| 259 | } |
| 260 | |
| 261 | long __init memblock_free(u64 base, u64 size) |
| 262 | { |
| 263 | return __memblock_remove(&memblock.reserved, base, size); |
| 264 | } |
| 265 | |
| 266 | long __init memblock_reserve(u64 base, u64 size) |
| 267 | { |
| 268 | struct memblock_region *_rgn = &memblock.reserved; |
| 269 | |
| 270 | BUG_ON(0 == size); |
| 271 | |
| 272 | return memblock_add_region(_rgn, base, size); |
| 273 | } |
| 274 | |
| 275 | long memblock_overlaps_region(struct memblock_region *rgn, u64 base, u64 size) |
| 276 | { |
| 277 | unsigned long i; |
| 278 | |
| 279 | for (i = 0; i < rgn->cnt; i++) { |
| 280 | u64 rgnbase = rgn->region[i].base; |
| 281 | u64 rgnsize = rgn->region[i].size; |
| 282 | if (memblock_addrs_overlap(base, size, rgnbase, rgnsize)) |
| 283 | break; |
| 284 | } |
| 285 | |
| 286 | return (i < rgn->cnt) ? i : -1; |
| 287 | } |
| 288 | |
| 289 | static u64 memblock_align_down(u64 addr, u64 size) |
| 290 | { |
| 291 | return addr & ~(size - 1); |
| 292 | } |
| 293 | |
| 294 | static u64 memblock_align_up(u64 addr, u64 size) |
| 295 | { |
| 296 | return (addr + (size - 1)) & ~(size - 1); |
| 297 | } |
| 298 | |
| 299 | static u64 __init memblock_alloc_nid_unreserved(u64 start, u64 end, |
| 300 | u64 size, u64 align) |
| 301 | { |
| 302 | u64 base, res_base; |
| 303 | long j; |
| 304 | |
| 305 | base = memblock_align_down((end - size), align); |
| 306 | while (start <= base) { |
| 307 | j = memblock_overlaps_region(&memblock.reserved, base, size); |
| 308 | if (j < 0) { |
| 309 | /* this area isn't reserved, take it */ |
| 310 | if (memblock_add_region(&memblock.reserved, base, size) < 0) |
| 311 | base = ~(u64)0; |
| 312 | return base; |
| 313 | } |
| 314 | res_base = memblock.reserved.region[j].base; |
| 315 | if (res_base < size) |
| 316 | break; |
| 317 | base = memblock_align_down(res_base - size, align); |
| 318 | } |
| 319 | |
| 320 | return ~(u64)0; |
| 321 | } |
| 322 | |
| 323 | static u64 __init memblock_alloc_nid_region(struct memblock_property *mp, |
| 324 | u64 (*nid_range)(u64, u64, int *), |
| 325 | u64 size, u64 align, int nid) |
| 326 | { |
| 327 | u64 start, end; |
| 328 | |
| 329 | start = mp->base; |
| 330 | end = start + mp->size; |
| 331 | |
| 332 | start = memblock_align_up(start, align); |
| 333 | while (start < end) { |
| 334 | u64 this_end; |
| 335 | int this_nid; |
| 336 | |
| 337 | this_end = nid_range(start, end, &this_nid); |
| 338 | if (this_nid == nid) { |
| 339 | u64 ret = memblock_alloc_nid_unreserved(start, this_end, |
| 340 | size, align); |
| 341 | if (ret != ~(u64)0) |
| 342 | return ret; |
| 343 | } |
| 344 | start = this_end; |
| 345 | } |
| 346 | |
| 347 | return ~(u64)0; |
| 348 | } |
| 349 | |
| 350 | u64 __init memblock_alloc_nid(u64 size, u64 align, int nid, |
| 351 | u64 (*nid_range)(u64 start, u64 end, int *nid)) |
| 352 | { |
| 353 | struct memblock_region *mem = &memblock.memory; |
| 354 | int i; |
| 355 | |
| 356 | BUG_ON(0 == size); |
| 357 | |
| 358 | size = memblock_align_up(size, align); |
| 359 | |
| 360 | for (i = 0; i < mem->cnt; i++) { |
| 361 | u64 ret = memblock_alloc_nid_region(&mem->region[i], |
| 362 | nid_range, |
| 363 | size, align, nid); |
| 364 | if (ret != ~(u64)0) |
| 365 | return ret; |
| 366 | } |
| 367 | |
| 368 | return memblock_alloc(size, align); |
| 369 | } |
| 370 | |
| 371 | u64 __init memblock_alloc(u64 size, u64 align) |
| 372 | { |
| 373 | return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ANYWHERE); |
| 374 | } |
| 375 | |
| 376 | u64 __init memblock_alloc_base(u64 size, u64 align, u64 max_addr) |
| 377 | { |
| 378 | u64 alloc; |
| 379 | |
| 380 | alloc = __memblock_alloc_base(size, align, max_addr); |
| 381 | |
| 382 | if (alloc == 0) |
| 383 | panic("ERROR: Failed to allocate 0x%llx bytes below 0x%llx.\n", |
| 384 | (unsigned long long) size, (unsigned long long) max_addr); |
| 385 | |
| 386 | return alloc; |
| 387 | } |
| 388 | |
| 389 | u64 __init __memblock_alloc_base(u64 size, u64 align, u64 max_addr) |
| 390 | { |
| 391 | long i, j; |
| 392 | u64 base = 0; |
| 393 | u64 res_base; |
| 394 | |
| 395 | BUG_ON(0 == size); |
| 396 | |
| 397 | size = memblock_align_up(size, align); |
| 398 | |
| 399 | /* On some platforms, make sure we allocate lowmem */ |
| 400 | /* Note that MEMBLOCK_REAL_LIMIT may be MEMBLOCK_ALLOC_ANYWHERE */ |
| 401 | if (max_addr == MEMBLOCK_ALLOC_ANYWHERE) |
| 402 | max_addr = MEMBLOCK_REAL_LIMIT; |
| 403 | |
| 404 | for (i = memblock.memory.cnt - 1; i >= 0; i--) { |
| 405 | u64 memblockbase = memblock.memory.region[i].base; |
| 406 | u64 memblocksize = memblock.memory.region[i].size; |
| 407 | |
| 408 | if (memblocksize < size) |
| 409 | continue; |
| 410 | if (max_addr == MEMBLOCK_ALLOC_ANYWHERE) |
| 411 | base = memblock_align_down(memblockbase + memblocksize - size, align); |
| 412 | else if (memblockbase < max_addr) { |
| 413 | base = min(memblockbase + memblocksize, max_addr); |
| 414 | base = memblock_align_down(base - size, align); |
| 415 | } else |
| 416 | continue; |
| 417 | |
| 418 | while (base && memblockbase <= base) { |
| 419 | j = memblock_overlaps_region(&memblock.reserved, base, size); |
| 420 | if (j < 0) { |
| 421 | /* this area isn't reserved, take it */ |
| 422 | if (memblock_add_region(&memblock.reserved, base, size) < 0) |
| 423 | return 0; |
| 424 | return base; |
| 425 | } |
| 426 | res_base = memblock.reserved.region[j].base; |
| 427 | if (res_base < size) |
| 428 | break; |
| 429 | base = memblock_align_down(res_base - size, align); |
| 430 | } |
| 431 | } |
| 432 | return 0; |
| 433 | } |
| 434 | |
| 435 | /* You must call memblock_analyze() before this. */ |
| 436 | u64 __init memblock_phys_mem_size(void) |
| 437 | { |
| 438 | return memblock.memory.size; |
| 439 | } |
| 440 | |
| 441 | u64 memblock_end_of_DRAM(void) |
| 442 | { |
| 443 | int idx = memblock.memory.cnt - 1; |
| 444 | |
| 445 | return (memblock.memory.region[idx].base + memblock.memory.region[idx].size); |
| 446 | } |
| 447 | |
| 448 | /* You must call memblock_analyze() after this. */ |
| 449 | void __init memblock_enforce_memory_limit(u64 memory_limit) |
| 450 | { |
| 451 | unsigned long i; |
| 452 | u64 limit; |
| 453 | struct memblock_property *p; |
| 454 | |
| 455 | if (!memory_limit) |
| 456 | return; |
| 457 | |
| 458 | /* Truncate the memblock regions to satisfy the memory limit. */ |
| 459 | limit = memory_limit; |
| 460 | for (i = 0; i < memblock.memory.cnt; i++) { |
| 461 | if (limit > memblock.memory.region[i].size) { |
| 462 | limit -= memblock.memory.region[i].size; |
| 463 | continue; |
| 464 | } |
| 465 | |
| 466 | memblock.memory.region[i].size = limit; |
| 467 | memblock.memory.cnt = i + 1; |
| 468 | break; |
| 469 | } |
| 470 | |
| 471 | if (memblock.memory.region[0].size < memblock.rmo_size) |
| 472 | memblock.rmo_size = memblock.memory.region[0].size; |
| 473 | |
| 474 | memory_limit = memblock_end_of_DRAM(); |
| 475 | |
| 476 | /* And truncate any reserves above the limit also. */ |
| 477 | for (i = 0; i < memblock.reserved.cnt; i++) { |
| 478 | p = &memblock.reserved.region[i]; |
| 479 | |
| 480 | if (p->base > memory_limit) |
| 481 | p->size = 0; |
| 482 | else if ((p->base + p->size) > memory_limit) |
| 483 | p->size = memory_limit - p->base; |
| 484 | |
| 485 | if (p->size == 0) { |
| 486 | memblock_remove_region(&memblock.reserved, i); |
| 487 | i--; |
| 488 | } |
| 489 | } |
| 490 | } |
| 491 | |
| 492 | int __init memblock_is_reserved(u64 addr) |
| 493 | { |
| 494 | int i; |
| 495 | |
| 496 | for (i = 0; i < memblock.reserved.cnt; i++) { |
| 497 | u64 upper = memblock.reserved.region[i].base + |
| 498 | memblock.reserved.region[i].size - 1; |
| 499 | if ((addr >= memblock.reserved.region[i].base) && (addr <= upper)) |
| 500 | return 1; |
| 501 | } |
| 502 | return 0; |
| 503 | } |
| 504 | |
| 505 | int memblock_is_region_reserved(u64 base, u64 size) |
| 506 | { |
| 507 | return memblock_overlaps_region(&memblock.reserved, base, size); |
| 508 | } |
| 509 | |
| 510 | /* |
| 511 | * Given a <base, len>, find which memory regions belong to this range. |
| 512 | * Adjust the request and return a contiguous chunk. |
| 513 | */ |
| 514 | int memblock_find(struct memblock_property *res) |
| 515 | { |
| 516 | int i; |
| 517 | u64 rstart, rend; |
| 518 | |
| 519 | rstart = res->base; |
| 520 | rend = rstart + res->size - 1; |
| 521 | |
| 522 | for (i = 0; i < memblock.memory.cnt; i++) { |
| 523 | u64 start = memblock.memory.region[i].base; |
| 524 | u64 end = start + memblock.memory.region[i].size - 1; |
| 525 | |
| 526 | if (start > rend) |
| 527 | return -1; |
| 528 | |
| 529 | if ((end >= rstart) && (start < rend)) { |
| 530 | /* adjust the request */ |
| 531 | if (rstart < start) |
| 532 | rstart = start; |
| 533 | if (rend > end) |
| 534 | rend = end; |
| 535 | res->base = rstart; |
| 536 | res->size = rend - rstart + 1; |
| 537 | return 0; |
| 538 | } |
| 539 | } |
| 540 | return -1; |
| 541 | } |