| 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/slab.h> |
| 15 | #include <linux/init.h> |
| 16 | #include <linux/bitops.h> |
| 17 | #include <linux/poison.h> |
| 18 | #include <linux/pfn.h> |
| 19 | #include <linux/debugfs.h> |
| 20 | #include <linux/seq_file.h> |
| 21 | #include <linux/memblock.h> |
| 22 | |
| 23 | static struct memblock_region memblock_memory_init_regions[INIT_MEMBLOCK_REGIONS] __initdata_memblock; |
| 24 | static struct memblock_region memblock_reserved_init_regions[INIT_MEMBLOCK_REGIONS] __initdata_memblock; |
| 25 | |
| 26 | struct memblock memblock __initdata_memblock = { |
| 27 | .memory.regions = memblock_memory_init_regions, |
| 28 | .memory.cnt = 1, /* empty dummy entry */ |
| 29 | .memory.max = INIT_MEMBLOCK_REGIONS, |
| 30 | |
| 31 | .reserved.regions = memblock_reserved_init_regions, |
| 32 | .reserved.cnt = 1, /* empty dummy entry */ |
| 33 | .reserved.max = INIT_MEMBLOCK_REGIONS, |
| 34 | |
| 35 | .current_limit = MEMBLOCK_ALLOC_ANYWHERE, |
| 36 | }; |
| 37 | |
| 38 | int memblock_debug __initdata_memblock; |
| 39 | static int memblock_can_resize __initdata_memblock; |
| 40 | static int memblock_memory_in_slab __initdata_memblock = 0; |
| 41 | static int memblock_reserved_in_slab __initdata_memblock = 0; |
| 42 | |
| 43 | /* inline so we don't get a warning when pr_debug is compiled out */ |
| 44 | static __init_memblock const char * |
| 45 | memblock_type_name(struct memblock_type *type) |
| 46 | { |
| 47 | if (type == &memblock.memory) |
| 48 | return "memory"; |
| 49 | else if (type == &memblock.reserved) |
| 50 | return "reserved"; |
| 51 | else |
| 52 | return "unknown"; |
| 53 | } |
| 54 | |
| 55 | /* adjust *@size so that (@base + *@size) doesn't overflow, return new size */ |
| 56 | static inline phys_addr_t memblock_cap_size(phys_addr_t base, phys_addr_t *size) |
| 57 | { |
| 58 | return *size = min(*size, (phys_addr_t)ULLONG_MAX - base); |
| 59 | } |
| 60 | |
| 61 | /* |
| 62 | * Address comparison utilities |
| 63 | */ |
| 64 | static unsigned long __init_memblock memblock_addrs_overlap(phys_addr_t base1, phys_addr_t size1, |
| 65 | phys_addr_t base2, phys_addr_t size2) |
| 66 | { |
| 67 | return ((base1 < (base2 + size2)) && (base2 < (base1 + size1))); |
| 68 | } |
| 69 | |
| 70 | static long __init_memblock memblock_overlaps_region(struct memblock_type *type, |
| 71 | phys_addr_t base, phys_addr_t size) |
| 72 | { |
| 73 | unsigned long i; |
| 74 | |
| 75 | for (i = 0; i < type->cnt; i++) { |
| 76 | phys_addr_t rgnbase = type->regions[i].base; |
| 77 | phys_addr_t rgnsize = type->regions[i].size; |
| 78 | if (memblock_addrs_overlap(base, size, rgnbase, rgnsize)) |
| 79 | break; |
| 80 | } |
| 81 | |
| 82 | return (i < type->cnt) ? i : -1; |
| 83 | } |
| 84 | |
| 85 | /** |
| 86 | * memblock_find_in_range_node - find free area in given range and node |
| 87 | * @start: start of candidate range |
| 88 | * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE} |
| 89 | * @size: size of free area to find |
| 90 | * @align: alignment of free area to find |
| 91 | * @nid: nid of the free area to find, %MAX_NUMNODES for any node |
| 92 | * |
| 93 | * Find @size free area aligned to @align in the specified range and node. |
| 94 | * |
| 95 | * RETURNS: |
| 96 | * Found address on success, %0 on failure. |
| 97 | */ |
| 98 | phys_addr_t __init_memblock memblock_find_in_range_node(phys_addr_t start, |
| 99 | phys_addr_t end, phys_addr_t size, |
| 100 | phys_addr_t align, int nid) |
| 101 | { |
| 102 | phys_addr_t this_start, this_end, cand; |
| 103 | u64 i; |
| 104 | |
| 105 | /* pump up @end */ |
| 106 | if (end == MEMBLOCK_ALLOC_ACCESSIBLE) |
| 107 | end = memblock.current_limit; |
| 108 | |
| 109 | /* avoid allocating the first page */ |
| 110 | start = max_t(phys_addr_t, start, PAGE_SIZE); |
| 111 | end = max(start, end); |
| 112 | |
| 113 | for_each_free_mem_range_reverse(i, nid, &this_start, &this_end, NULL) { |
| 114 | this_start = clamp(this_start, start, end); |
| 115 | this_end = clamp(this_end, start, end); |
| 116 | |
| 117 | if (this_end < size) |
| 118 | continue; |
| 119 | |
| 120 | cand = round_down(this_end - size, align); |
| 121 | if (cand >= this_start) |
| 122 | return cand; |
| 123 | } |
| 124 | return 0; |
| 125 | } |
| 126 | |
| 127 | /** |
| 128 | * memblock_find_in_range - find free area in given range |
| 129 | * @start: start of candidate range |
| 130 | * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE} |
| 131 | * @size: size of free area to find |
| 132 | * @align: alignment of free area to find |
| 133 | * |
| 134 | * Find @size free area aligned to @align in the specified range. |
| 135 | * |
| 136 | * RETURNS: |
| 137 | * Found address on success, %0 on failure. |
| 138 | */ |
| 139 | phys_addr_t __init_memblock memblock_find_in_range(phys_addr_t start, |
| 140 | phys_addr_t end, phys_addr_t size, |
| 141 | phys_addr_t align) |
| 142 | { |
| 143 | return memblock_find_in_range_node(start, end, size, align, |
| 144 | MAX_NUMNODES); |
| 145 | } |
| 146 | |
| 147 | static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r) |
| 148 | { |
| 149 | type->total_size -= type->regions[r].size; |
| 150 | memmove(&type->regions[r], &type->regions[r + 1], |
| 151 | (type->cnt - (r + 1)) * sizeof(type->regions[r])); |
| 152 | type->cnt--; |
| 153 | |
| 154 | /* Special case for empty arrays */ |
| 155 | if (type->cnt == 0) { |
| 156 | WARN_ON(type->total_size != 0); |
| 157 | type->cnt = 1; |
| 158 | type->regions[0].base = 0; |
| 159 | type->regions[0].size = 0; |
| 160 | memblock_set_region_node(&type->regions[0], MAX_NUMNODES); |
| 161 | } |
| 162 | } |
| 163 | |
| 164 | phys_addr_t __init_memblock get_allocated_memblock_reserved_regions_info( |
| 165 | phys_addr_t *addr) |
| 166 | { |
| 167 | if (memblock.reserved.regions == memblock_reserved_init_regions) |
| 168 | return 0; |
| 169 | |
| 170 | *addr = __pa(memblock.reserved.regions); |
| 171 | |
| 172 | return PAGE_ALIGN(sizeof(struct memblock_region) * |
| 173 | memblock.reserved.max); |
| 174 | } |
| 175 | |
| 176 | /** |
| 177 | * memblock_double_array - double the size of the memblock regions array |
| 178 | * @type: memblock type of the regions array being doubled |
| 179 | * @new_area_start: starting address of memory range to avoid overlap with |
| 180 | * @new_area_size: size of memory range to avoid overlap with |
| 181 | * |
| 182 | * Double the size of the @type regions array. If memblock is being used to |
| 183 | * allocate memory for a new reserved regions array and there is a previously |
| 184 | * allocated memory range [@new_area_start,@new_area_start+@new_area_size] |
| 185 | * waiting to be reserved, ensure the memory used by the new array does |
| 186 | * not overlap. |
| 187 | * |
| 188 | * RETURNS: |
| 189 | * 0 on success, -1 on failure. |
| 190 | */ |
| 191 | static int __init_memblock memblock_double_array(struct memblock_type *type, |
| 192 | phys_addr_t new_area_start, |
| 193 | phys_addr_t new_area_size) |
| 194 | { |
| 195 | struct memblock_region *new_array, *old_array; |
| 196 | phys_addr_t old_alloc_size, new_alloc_size; |
| 197 | phys_addr_t old_size, new_size, addr; |
| 198 | int use_slab = slab_is_available(); |
| 199 | int *in_slab; |
| 200 | |
| 201 | /* We don't allow resizing until we know about the reserved regions |
| 202 | * of memory that aren't suitable for allocation |
| 203 | */ |
| 204 | if (!memblock_can_resize) |
| 205 | return -1; |
| 206 | |
| 207 | /* Calculate new doubled size */ |
| 208 | old_size = type->max * sizeof(struct memblock_region); |
| 209 | new_size = old_size << 1; |
| 210 | /* |
| 211 | * We need to allocated new one align to PAGE_SIZE, |
| 212 | * so we can free them completely later. |
| 213 | */ |
| 214 | old_alloc_size = PAGE_ALIGN(old_size); |
| 215 | new_alloc_size = PAGE_ALIGN(new_size); |
| 216 | |
| 217 | /* Retrieve the slab flag */ |
| 218 | if (type == &memblock.memory) |
| 219 | in_slab = &memblock_memory_in_slab; |
| 220 | else |
| 221 | in_slab = &memblock_reserved_in_slab; |
| 222 | |
| 223 | /* Try to find some space for it. |
| 224 | * |
| 225 | * WARNING: We assume that either slab_is_available() and we use it or |
| 226 | * we use MEMBLOCK for allocations. That means that this is unsafe to |
| 227 | * use when bootmem is currently active (unless bootmem itself is |
| 228 | * implemented on top of MEMBLOCK which isn't the case yet) |
| 229 | * |
| 230 | * This should however not be an issue for now, as we currently only |
| 231 | * call into MEMBLOCK while it's still active, or much later when slab |
| 232 | * is active for memory hotplug operations |
| 233 | */ |
| 234 | if (use_slab) { |
| 235 | new_array = kmalloc(new_size, GFP_KERNEL); |
| 236 | addr = new_array ? __pa(new_array) : 0; |
| 237 | } else { |
| 238 | /* only exclude range when trying to double reserved.regions */ |
| 239 | if (type != &memblock.reserved) |
| 240 | new_area_start = new_area_size = 0; |
| 241 | |
| 242 | addr = memblock_find_in_range(new_area_start + new_area_size, |
| 243 | memblock.current_limit, |
| 244 | new_alloc_size, PAGE_SIZE); |
| 245 | if (!addr && new_area_size) |
| 246 | addr = memblock_find_in_range(0, |
| 247 | min(new_area_start, memblock.current_limit), |
| 248 | new_alloc_size, PAGE_SIZE); |
| 249 | |
| 250 | new_array = addr ? __va(addr) : NULL; |
| 251 | } |
| 252 | if (!addr) { |
| 253 | pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n", |
| 254 | memblock_type_name(type), type->max, type->max * 2); |
| 255 | return -1; |
| 256 | } |
| 257 | |
| 258 | memblock_dbg("memblock: %s is doubled to %ld at [%#010llx-%#010llx]", |
| 259 | memblock_type_name(type), type->max * 2, (u64)addr, |
| 260 | (u64)addr + new_size - 1); |
| 261 | |
| 262 | /* |
| 263 | * Found space, we now need to move the array over before we add the |
| 264 | * reserved region since it may be our reserved array itself that is |
| 265 | * full. |
| 266 | */ |
| 267 | memcpy(new_array, type->regions, old_size); |
| 268 | memset(new_array + type->max, 0, old_size); |
| 269 | old_array = type->regions; |
| 270 | type->regions = new_array; |
| 271 | type->max <<= 1; |
| 272 | |
| 273 | /* Free old array. We needn't free it if the array is the static one */ |
| 274 | if (*in_slab) |
| 275 | kfree(old_array); |
| 276 | else if (old_array != memblock_memory_init_regions && |
| 277 | old_array != memblock_reserved_init_regions) |
| 278 | memblock_free(__pa(old_array), old_alloc_size); |
| 279 | |
| 280 | /* |
| 281 | * Reserve the new array if that comes from the memblock. Otherwise, we |
| 282 | * needn't do it |
| 283 | */ |
| 284 | if (!use_slab) |
| 285 | BUG_ON(memblock_reserve(addr, new_alloc_size)); |
| 286 | |
| 287 | /* Update slab flag */ |
| 288 | *in_slab = use_slab; |
| 289 | |
| 290 | return 0; |
| 291 | } |
| 292 | |
| 293 | /** |
| 294 | * memblock_merge_regions - merge neighboring compatible regions |
| 295 | * @type: memblock type to scan |
| 296 | * |
| 297 | * Scan @type and merge neighboring compatible regions. |
| 298 | */ |
| 299 | static void __init_memblock memblock_merge_regions(struct memblock_type *type) |
| 300 | { |
| 301 | int i = 0; |
| 302 | |
| 303 | /* cnt never goes below 1 */ |
| 304 | while (i < type->cnt - 1) { |
| 305 | struct memblock_region *this = &type->regions[i]; |
| 306 | struct memblock_region *next = &type->regions[i + 1]; |
| 307 | |
| 308 | if (this->base + this->size != next->base || |
| 309 | memblock_get_region_node(this) != |
| 310 | memblock_get_region_node(next)) { |
| 311 | BUG_ON(this->base + this->size > next->base); |
| 312 | i++; |
| 313 | continue; |
| 314 | } |
| 315 | |
| 316 | this->size += next->size; |
| 317 | /* move forward from next + 1, index of which is i + 2 */ |
| 318 | memmove(next, next + 1, (type->cnt - (i + 2)) * sizeof(*next)); |
| 319 | type->cnt--; |
| 320 | } |
| 321 | } |
| 322 | |
| 323 | /** |
| 324 | * memblock_insert_region - insert new memblock region |
| 325 | * @type: memblock type to insert into |
| 326 | * @idx: index for the insertion point |
| 327 | * @base: base address of the new region |
| 328 | * @size: size of the new region |
| 329 | * @nid: node id of the new region |
| 330 | * |
| 331 | * Insert new memblock region [@base,@base+@size) into @type at @idx. |
| 332 | * @type must already have extra room to accomodate the new region. |
| 333 | */ |
| 334 | static void __init_memblock memblock_insert_region(struct memblock_type *type, |
| 335 | int idx, phys_addr_t base, |
| 336 | phys_addr_t size, int nid) |
| 337 | { |
| 338 | struct memblock_region *rgn = &type->regions[idx]; |
| 339 | |
| 340 | BUG_ON(type->cnt >= type->max); |
| 341 | memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn)); |
| 342 | rgn->base = base; |
| 343 | rgn->size = size; |
| 344 | memblock_set_region_node(rgn, nid); |
| 345 | type->cnt++; |
| 346 | type->total_size += size; |
| 347 | } |
| 348 | |
| 349 | /** |
| 350 | * memblock_add_region - add new memblock region |
| 351 | * @type: memblock type to add new region into |
| 352 | * @base: base address of the new region |
| 353 | * @size: size of the new region |
| 354 | * @nid: nid of the new region |
| 355 | * |
| 356 | * Add new memblock region [@base,@base+@size) into @type. The new region |
| 357 | * is allowed to overlap with existing ones - overlaps don't affect already |
| 358 | * existing regions. @type is guaranteed to be minimal (all neighbouring |
| 359 | * compatible regions are merged) after the addition. |
| 360 | * |
| 361 | * RETURNS: |
| 362 | * 0 on success, -errno on failure. |
| 363 | */ |
| 364 | static int __init_memblock memblock_add_region(struct memblock_type *type, |
| 365 | phys_addr_t base, phys_addr_t size, int nid) |
| 366 | { |
| 367 | bool insert = false; |
| 368 | phys_addr_t obase = base; |
| 369 | phys_addr_t end = base + memblock_cap_size(base, &size); |
| 370 | int i, nr_new; |
| 371 | |
| 372 | if (!size) |
| 373 | return 0; |
| 374 | |
| 375 | /* special case for empty array */ |
| 376 | if (type->regions[0].size == 0) { |
| 377 | WARN_ON(type->cnt != 1 || type->total_size); |
| 378 | type->regions[0].base = base; |
| 379 | type->regions[0].size = size; |
| 380 | memblock_set_region_node(&type->regions[0], nid); |
| 381 | type->total_size = size; |
| 382 | return 0; |
| 383 | } |
| 384 | repeat: |
| 385 | /* |
| 386 | * The following is executed twice. Once with %false @insert and |
| 387 | * then with %true. The first counts the number of regions needed |
| 388 | * to accomodate the new area. The second actually inserts them. |
| 389 | */ |
| 390 | base = obase; |
| 391 | nr_new = 0; |
| 392 | |
| 393 | for (i = 0; i < type->cnt; i++) { |
| 394 | struct memblock_region *rgn = &type->regions[i]; |
| 395 | phys_addr_t rbase = rgn->base; |
| 396 | phys_addr_t rend = rbase + rgn->size; |
| 397 | |
| 398 | if (rbase >= end) |
| 399 | break; |
| 400 | if (rend <= base) |
| 401 | continue; |
| 402 | /* |
| 403 | * @rgn overlaps. If it separates the lower part of new |
| 404 | * area, insert that portion. |
| 405 | */ |
| 406 | if (rbase > base) { |
| 407 | nr_new++; |
| 408 | if (insert) |
| 409 | memblock_insert_region(type, i++, base, |
| 410 | rbase - base, nid); |
| 411 | } |
| 412 | /* area below @rend is dealt with, forget about it */ |
| 413 | base = min(rend, end); |
| 414 | } |
| 415 | |
| 416 | /* insert the remaining portion */ |
| 417 | if (base < end) { |
| 418 | nr_new++; |
| 419 | if (insert) |
| 420 | memblock_insert_region(type, i, base, end - base, nid); |
| 421 | } |
| 422 | |
| 423 | /* |
| 424 | * If this was the first round, resize array and repeat for actual |
| 425 | * insertions; otherwise, merge and return. |
| 426 | */ |
| 427 | if (!insert) { |
| 428 | while (type->cnt + nr_new > type->max) |
| 429 | if (memblock_double_array(type, obase, size) < 0) |
| 430 | return -ENOMEM; |
| 431 | insert = true; |
| 432 | goto repeat; |
| 433 | } else { |
| 434 | memblock_merge_regions(type); |
| 435 | return 0; |
| 436 | } |
| 437 | } |
| 438 | |
| 439 | int __init_memblock memblock_add_node(phys_addr_t base, phys_addr_t size, |
| 440 | int nid) |
| 441 | { |
| 442 | return memblock_add_region(&memblock.memory, base, size, nid); |
| 443 | } |
| 444 | |
| 445 | int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size) |
| 446 | { |
| 447 | return memblock_add_region(&memblock.memory, base, size, MAX_NUMNODES); |
| 448 | } |
| 449 | |
| 450 | /** |
| 451 | * memblock_isolate_range - isolate given range into disjoint memblocks |
| 452 | * @type: memblock type to isolate range for |
| 453 | * @base: base of range to isolate |
| 454 | * @size: size of range to isolate |
| 455 | * @start_rgn: out parameter for the start of isolated region |
| 456 | * @end_rgn: out parameter for the end of isolated region |
| 457 | * |
| 458 | * Walk @type and ensure that regions don't cross the boundaries defined by |
| 459 | * [@base,@base+@size). Crossing regions are split at the boundaries, |
| 460 | * which may create at most two more regions. The index of the first |
| 461 | * region inside the range is returned in *@start_rgn and end in *@end_rgn. |
| 462 | * |
| 463 | * RETURNS: |
| 464 | * 0 on success, -errno on failure. |
| 465 | */ |
| 466 | static int __init_memblock memblock_isolate_range(struct memblock_type *type, |
| 467 | phys_addr_t base, phys_addr_t size, |
| 468 | int *start_rgn, int *end_rgn) |
| 469 | { |
| 470 | phys_addr_t end = base + memblock_cap_size(base, &size); |
| 471 | int i; |
| 472 | |
| 473 | *start_rgn = *end_rgn = 0; |
| 474 | |
| 475 | if (!size) |
| 476 | return 0; |
| 477 | |
| 478 | /* we'll create at most two more regions */ |
| 479 | while (type->cnt + 2 > type->max) |
| 480 | if (memblock_double_array(type, base, size) < 0) |
| 481 | return -ENOMEM; |
| 482 | |
| 483 | for (i = 0; i < type->cnt; i++) { |
| 484 | struct memblock_region *rgn = &type->regions[i]; |
| 485 | phys_addr_t rbase = rgn->base; |
| 486 | phys_addr_t rend = rbase + rgn->size; |
| 487 | |
| 488 | if (rbase >= end) |
| 489 | break; |
| 490 | if (rend <= base) |
| 491 | continue; |
| 492 | |
| 493 | if (rbase < base) { |
| 494 | /* |
| 495 | * @rgn intersects from below. Split and continue |
| 496 | * to process the next region - the new top half. |
| 497 | */ |
| 498 | rgn->base = base; |
| 499 | rgn->size -= base - rbase; |
| 500 | type->total_size -= base - rbase; |
| 501 | memblock_insert_region(type, i, rbase, base - rbase, |
| 502 | memblock_get_region_node(rgn)); |
| 503 | } else if (rend > end) { |
| 504 | /* |
| 505 | * @rgn intersects from above. Split and redo the |
| 506 | * current region - the new bottom half. |
| 507 | */ |
| 508 | rgn->base = end; |
| 509 | rgn->size -= end - rbase; |
| 510 | type->total_size -= end - rbase; |
| 511 | memblock_insert_region(type, i--, rbase, end - rbase, |
| 512 | memblock_get_region_node(rgn)); |
| 513 | } else { |
| 514 | /* @rgn is fully contained, record it */ |
| 515 | if (!*end_rgn) |
| 516 | *start_rgn = i; |
| 517 | *end_rgn = i + 1; |
| 518 | } |
| 519 | } |
| 520 | |
| 521 | return 0; |
| 522 | } |
| 523 | |
| 524 | static int __init_memblock __memblock_remove(struct memblock_type *type, |
| 525 | phys_addr_t base, phys_addr_t size) |
| 526 | { |
| 527 | int start_rgn, end_rgn; |
| 528 | int i, ret; |
| 529 | |
| 530 | ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn); |
| 531 | if (ret) |
| 532 | return ret; |
| 533 | |
| 534 | for (i = end_rgn - 1; i >= start_rgn; i--) |
| 535 | memblock_remove_region(type, i); |
| 536 | return 0; |
| 537 | } |
| 538 | |
| 539 | int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size) |
| 540 | { |
| 541 | return __memblock_remove(&memblock.memory, base, size); |
| 542 | } |
| 543 | |
| 544 | int __init_memblock memblock_free(phys_addr_t base, phys_addr_t size) |
| 545 | { |
| 546 | memblock_dbg(" memblock_free: [%#016llx-%#016llx] %pF\n", |
| 547 | (unsigned long long)base, |
| 548 | (unsigned long long)base + size, |
| 549 | (void *)_RET_IP_); |
| 550 | |
| 551 | return __memblock_remove(&memblock.reserved, base, size); |
| 552 | } |
| 553 | |
| 554 | int __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size) |
| 555 | { |
| 556 | struct memblock_type *_rgn = &memblock.reserved; |
| 557 | |
| 558 | memblock_dbg("memblock_reserve: [%#016llx-%#016llx] %pF\n", |
| 559 | (unsigned long long)base, |
| 560 | (unsigned long long)base + size, |
| 561 | (void *)_RET_IP_); |
| 562 | |
| 563 | return memblock_add_region(_rgn, base, size, MAX_NUMNODES); |
| 564 | } |
| 565 | |
| 566 | /** |
| 567 | * __next_free_mem_range - next function for for_each_free_mem_range() |
| 568 | * @idx: pointer to u64 loop variable |
| 569 | * @nid: nid: node selector, %MAX_NUMNODES for all nodes |
| 570 | * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL |
| 571 | * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL |
| 572 | * @out_nid: ptr to int for nid of the range, can be %NULL |
| 573 | * |
| 574 | * Find the first free area from *@idx which matches @nid, fill the out |
| 575 | * parameters, and update *@idx for the next iteration. The lower 32bit of |
| 576 | * *@idx contains index into memory region and the upper 32bit indexes the |
| 577 | * areas before each reserved region. For example, if reserved regions |
| 578 | * look like the following, |
| 579 | * |
| 580 | * 0:[0-16), 1:[32-48), 2:[128-130) |
| 581 | * |
| 582 | * The upper 32bit indexes the following regions. |
| 583 | * |
| 584 | * 0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX) |
| 585 | * |
| 586 | * As both region arrays are sorted, the function advances the two indices |
| 587 | * in lockstep and returns each intersection. |
| 588 | */ |
| 589 | void __init_memblock __next_free_mem_range(u64 *idx, int nid, |
| 590 | phys_addr_t *out_start, |
| 591 | phys_addr_t *out_end, int *out_nid) |
| 592 | { |
| 593 | struct memblock_type *mem = &memblock.memory; |
| 594 | struct memblock_type *rsv = &memblock.reserved; |
| 595 | int mi = *idx & 0xffffffff; |
| 596 | int ri = *idx >> 32; |
| 597 | |
| 598 | for ( ; mi < mem->cnt; mi++) { |
| 599 | struct memblock_region *m = &mem->regions[mi]; |
| 600 | phys_addr_t m_start = m->base; |
| 601 | phys_addr_t m_end = m->base + m->size; |
| 602 | |
| 603 | /* only memory regions are associated with nodes, check it */ |
| 604 | if (nid != MAX_NUMNODES && nid != memblock_get_region_node(m)) |
| 605 | continue; |
| 606 | |
| 607 | /* scan areas before each reservation for intersection */ |
| 608 | for ( ; ri < rsv->cnt + 1; ri++) { |
| 609 | struct memblock_region *r = &rsv->regions[ri]; |
| 610 | phys_addr_t r_start = ri ? r[-1].base + r[-1].size : 0; |
| 611 | phys_addr_t r_end = ri < rsv->cnt ? r->base : ULLONG_MAX; |
| 612 | |
| 613 | /* if ri advanced past mi, break out to advance mi */ |
| 614 | if (r_start >= m_end) |
| 615 | break; |
| 616 | /* if the two regions intersect, we're done */ |
| 617 | if (m_start < r_end) { |
| 618 | if (out_start) |
| 619 | *out_start = max(m_start, r_start); |
| 620 | if (out_end) |
| 621 | *out_end = min(m_end, r_end); |
| 622 | if (out_nid) |
| 623 | *out_nid = memblock_get_region_node(m); |
| 624 | /* |
| 625 | * The region which ends first is advanced |
| 626 | * for the next iteration. |
| 627 | */ |
| 628 | if (m_end <= r_end) |
| 629 | mi++; |
| 630 | else |
| 631 | ri++; |
| 632 | *idx = (u32)mi | (u64)ri << 32; |
| 633 | return; |
| 634 | } |
| 635 | } |
| 636 | } |
| 637 | |
| 638 | /* signal end of iteration */ |
| 639 | *idx = ULLONG_MAX; |
| 640 | } |
| 641 | |
| 642 | /** |
| 643 | * __next_free_mem_range_rev - next function for for_each_free_mem_range_reverse() |
| 644 | * @idx: pointer to u64 loop variable |
| 645 | * @nid: nid: node selector, %MAX_NUMNODES for all nodes |
| 646 | * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL |
| 647 | * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL |
| 648 | * @out_nid: ptr to int for nid of the range, can be %NULL |
| 649 | * |
| 650 | * Reverse of __next_free_mem_range(). |
| 651 | */ |
| 652 | void __init_memblock __next_free_mem_range_rev(u64 *idx, int nid, |
| 653 | phys_addr_t *out_start, |
| 654 | phys_addr_t *out_end, int *out_nid) |
| 655 | { |
| 656 | struct memblock_type *mem = &memblock.memory; |
| 657 | struct memblock_type *rsv = &memblock.reserved; |
| 658 | int mi = *idx & 0xffffffff; |
| 659 | int ri = *idx >> 32; |
| 660 | |
| 661 | if (*idx == (u64)ULLONG_MAX) { |
| 662 | mi = mem->cnt - 1; |
| 663 | ri = rsv->cnt; |
| 664 | } |
| 665 | |
| 666 | for ( ; mi >= 0; mi--) { |
| 667 | struct memblock_region *m = &mem->regions[mi]; |
| 668 | phys_addr_t m_start = m->base; |
| 669 | phys_addr_t m_end = m->base + m->size; |
| 670 | |
| 671 | /* only memory regions are associated with nodes, check it */ |
| 672 | if (nid != MAX_NUMNODES && nid != memblock_get_region_node(m)) |
| 673 | continue; |
| 674 | |
| 675 | /* scan areas before each reservation for intersection */ |
| 676 | for ( ; ri >= 0; ri--) { |
| 677 | struct memblock_region *r = &rsv->regions[ri]; |
| 678 | phys_addr_t r_start = ri ? r[-1].base + r[-1].size : 0; |
| 679 | phys_addr_t r_end = ri < rsv->cnt ? r->base : ULLONG_MAX; |
| 680 | |
| 681 | /* if ri advanced past mi, break out to advance mi */ |
| 682 | if (r_end <= m_start) |
| 683 | break; |
| 684 | /* if the two regions intersect, we're done */ |
| 685 | if (m_end > r_start) { |
| 686 | if (out_start) |
| 687 | *out_start = max(m_start, r_start); |
| 688 | if (out_end) |
| 689 | *out_end = min(m_end, r_end); |
| 690 | if (out_nid) |
| 691 | *out_nid = memblock_get_region_node(m); |
| 692 | |
| 693 | if (m_start >= r_start) |
| 694 | mi--; |
| 695 | else |
| 696 | ri--; |
| 697 | *idx = (u32)mi | (u64)ri << 32; |
| 698 | return; |
| 699 | } |
| 700 | } |
| 701 | } |
| 702 | |
| 703 | *idx = ULLONG_MAX; |
| 704 | } |
| 705 | |
| 706 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
| 707 | /* |
| 708 | * Common iterator interface used to define for_each_mem_range(). |
| 709 | */ |
| 710 | void __init_memblock __next_mem_pfn_range(int *idx, int nid, |
| 711 | unsigned long *out_start_pfn, |
| 712 | unsigned long *out_end_pfn, int *out_nid) |
| 713 | { |
| 714 | struct memblock_type *type = &memblock.memory; |
| 715 | struct memblock_region *r; |
| 716 | |
| 717 | while (++*idx < type->cnt) { |
| 718 | r = &type->regions[*idx]; |
| 719 | |
| 720 | if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size)) |
| 721 | continue; |
| 722 | if (nid == MAX_NUMNODES || nid == r->nid) |
| 723 | break; |
| 724 | } |
| 725 | if (*idx >= type->cnt) { |
| 726 | *idx = -1; |
| 727 | return; |
| 728 | } |
| 729 | |
| 730 | if (out_start_pfn) |
| 731 | *out_start_pfn = PFN_UP(r->base); |
| 732 | if (out_end_pfn) |
| 733 | *out_end_pfn = PFN_DOWN(r->base + r->size); |
| 734 | if (out_nid) |
| 735 | *out_nid = r->nid; |
| 736 | } |
| 737 | |
| 738 | /** |
| 739 | * memblock_set_node - set node ID on memblock regions |
| 740 | * @base: base of area to set node ID for |
| 741 | * @size: size of area to set node ID for |
| 742 | * @nid: node ID to set |
| 743 | * |
| 744 | * Set the nid of memblock memory regions in [@base,@base+@size) to @nid. |
| 745 | * Regions which cross the area boundaries are split as necessary. |
| 746 | * |
| 747 | * RETURNS: |
| 748 | * 0 on success, -errno on failure. |
| 749 | */ |
| 750 | int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size, |
| 751 | int nid) |
| 752 | { |
| 753 | struct memblock_type *type = &memblock.memory; |
| 754 | int start_rgn, end_rgn; |
| 755 | int i, ret; |
| 756 | |
| 757 | ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn); |
| 758 | if (ret) |
| 759 | return ret; |
| 760 | |
| 761 | for (i = start_rgn; i < end_rgn; i++) |
| 762 | memblock_set_region_node(&type->regions[i], nid); |
| 763 | |
| 764 | memblock_merge_regions(type); |
| 765 | return 0; |
| 766 | } |
| 767 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
| 768 | |
| 769 | static phys_addr_t __init memblock_alloc_base_nid(phys_addr_t size, |
| 770 | phys_addr_t align, phys_addr_t max_addr, |
| 771 | int nid) |
| 772 | { |
| 773 | phys_addr_t found; |
| 774 | |
| 775 | if (WARN_ON(!align)) |
| 776 | align = __alignof__(long long); |
| 777 | |
| 778 | /* align @size to avoid excessive fragmentation on reserved array */ |
| 779 | size = round_up(size, align); |
| 780 | |
| 781 | found = memblock_find_in_range_node(0, max_addr, size, align, nid); |
| 782 | if (found && !memblock_reserve(found, size)) |
| 783 | return found; |
| 784 | |
| 785 | return 0; |
| 786 | } |
| 787 | |
| 788 | phys_addr_t __init memblock_alloc_nid(phys_addr_t size, phys_addr_t align, int nid) |
| 789 | { |
| 790 | return memblock_alloc_base_nid(size, align, MEMBLOCK_ALLOC_ACCESSIBLE, nid); |
| 791 | } |
| 792 | |
| 793 | phys_addr_t __init __memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr) |
| 794 | { |
| 795 | return memblock_alloc_base_nid(size, align, max_addr, MAX_NUMNODES); |
| 796 | } |
| 797 | |
| 798 | phys_addr_t __init memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr) |
| 799 | { |
| 800 | phys_addr_t alloc; |
| 801 | |
| 802 | alloc = __memblock_alloc_base(size, align, max_addr); |
| 803 | |
| 804 | if (alloc == 0) |
| 805 | panic("ERROR: Failed to allocate 0x%llx bytes below 0x%llx.\n", |
| 806 | (unsigned long long) size, (unsigned long long) max_addr); |
| 807 | |
| 808 | return alloc; |
| 809 | } |
| 810 | |
| 811 | phys_addr_t __init memblock_alloc(phys_addr_t size, phys_addr_t align) |
| 812 | { |
| 813 | return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE); |
| 814 | } |
| 815 | |
| 816 | phys_addr_t __init memblock_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid) |
| 817 | { |
| 818 | phys_addr_t res = memblock_alloc_nid(size, align, nid); |
| 819 | |
| 820 | if (res) |
| 821 | return res; |
| 822 | return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE); |
| 823 | } |
| 824 | |
| 825 | |
| 826 | /* |
| 827 | * Remaining API functions |
| 828 | */ |
| 829 | |
| 830 | phys_addr_t __init memblock_phys_mem_size(void) |
| 831 | { |
| 832 | return memblock.memory.total_size; |
| 833 | } |
| 834 | |
| 835 | phys_addr_t __init memblock_mem_size(unsigned long limit_pfn) |
| 836 | { |
| 837 | unsigned long pages = 0; |
| 838 | struct memblock_region *r; |
| 839 | unsigned long start_pfn, end_pfn; |
| 840 | |
| 841 | for_each_memblock(memory, r) { |
| 842 | start_pfn = memblock_region_memory_base_pfn(r); |
| 843 | end_pfn = memblock_region_memory_end_pfn(r); |
| 844 | start_pfn = min_t(unsigned long, start_pfn, limit_pfn); |
| 845 | end_pfn = min_t(unsigned long, end_pfn, limit_pfn); |
| 846 | pages += end_pfn - start_pfn; |
| 847 | } |
| 848 | |
| 849 | return (phys_addr_t)pages << PAGE_SHIFT; |
| 850 | } |
| 851 | |
| 852 | /* lowest address */ |
| 853 | phys_addr_t __init_memblock memblock_start_of_DRAM(void) |
| 854 | { |
| 855 | return memblock.memory.regions[0].base; |
| 856 | } |
| 857 | |
| 858 | phys_addr_t __init_memblock memblock_end_of_DRAM(void) |
| 859 | { |
| 860 | int idx = memblock.memory.cnt - 1; |
| 861 | |
| 862 | return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size); |
| 863 | } |
| 864 | |
| 865 | void __init memblock_enforce_memory_limit(phys_addr_t limit) |
| 866 | { |
| 867 | unsigned long i; |
| 868 | phys_addr_t max_addr = (phys_addr_t)ULLONG_MAX; |
| 869 | |
| 870 | if (!limit) |
| 871 | return; |
| 872 | |
| 873 | /* find out max address */ |
| 874 | for (i = 0; i < memblock.memory.cnt; i++) { |
| 875 | struct memblock_region *r = &memblock.memory.regions[i]; |
| 876 | |
| 877 | if (limit <= r->size) { |
| 878 | max_addr = r->base + limit; |
| 879 | break; |
| 880 | } |
| 881 | limit -= r->size; |
| 882 | } |
| 883 | |
| 884 | /* truncate both memory and reserved regions */ |
| 885 | __memblock_remove(&memblock.memory, max_addr, (phys_addr_t)ULLONG_MAX); |
| 886 | __memblock_remove(&memblock.reserved, max_addr, (phys_addr_t)ULLONG_MAX); |
| 887 | } |
| 888 | |
| 889 | static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr) |
| 890 | { |
| 891 | unsigned int left = 0, right = type->cnt; |
| 892 | |
| 893 | do { |
| 894 | unsigned int mid = (right + left) / 2; |
| 895 | |
| 896 | if (addr < type->regions[mid].base) |
| 897 | right = mid; |
| 898 | else if (addr >= (type->regions[mid].base + |
| 899 | type->regions[mid].size)) |
| 900 | left = mid + 1; |
| 901 | else |
| 902 | return mid; |
| 903 | } while (left < right); |
| 904 | return -1; |
| 905 | } |
| 906 | |
| 907 | int __init memblock_is_reserved(phys_addr_t addr) |
| 908 | { |
| 909 | return memblock_search(&memblock.reserved, addr) != -1; |
| 910 | } |
| 911 | |
| 912 | int __init_memblock memblock_is_memory(phys_addr_t addr) |
| 913 | { |
| 914 | return memblock_search(&memblock.memory, addr) != -1; |
| 915 | } |
| 916 | |
| 917 | /** |
| 918 | * memblock_is_region_memory - check if a region is a subset of memory |
| 919 | * @base: base of region to check |
| 920 | * @size: size of region to check |
| 921 | * |
| 922 | * Check if the region [@base, @base+@size) is a subset of a memory block. |
| 923 | * |
| 924 | * RETURNS: |
| 925 | * 0 if false, non-zero if true |
| 926 | */ |
| 927 | int __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size) |
| 928 | { |
| 929 | int idx = memblock_search(&memblock.memory, base); |
| 930 | phys_addr_t end = base + memblock_cap_size(base, &size); |
| 931 | |
| 932 | if (idx == -1) |
| 933 | return 0; |
| 934 | return memblock.memory.regions[idx].base <= base && |
| 935 | (memblock.memory.regions[idx].base + |
| 936 | memblock.memory.regions[idx].size) >= end; |
| 937 | } |
| 938 | |
| 939 | /** |
| 940 | * memblock_is_region_reserved - check if a region intersects reserved memory |
| 941 | * @base: base of region to check |
| 942 | * @size: size of region to check |
| 943 | * |
| 944 | * Check if the region [@base, @base+@size) intersects a reserved memory block. |
| 945 | * |
| 946 | * RETURNS: |
| 947 | * 0 if false, non-zero if true |
| 948 | */ |
| 949 | int __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size) |
| 950 | { |
| 951 | memblock_cap_size(base, &size); |
| 952 | return memblock_overlaps_region(&memblock.reserved, base, size) >= 0; |
| 953 | } |
| 954 | |
| 955 | void __init_memblock memblock_trim_memory(phys_addr_t align) |
| 956 | { |
| 957 | int i; |
| 958 | phys_addr_t start, end, orig_start, orig_end; |
| 959 | struct memblock_type *mem = &memblock.memory; |
| 960 | |
| 961 | for (i = 0; i < mem->cnt; i++) { |
| 962 | orig_start = mem->regions[i].base; |
| 963 | orig_end = mem->regions[i].base + mem->regions[i].size; |
| 964 | start = round_up(orig_start, align); |
| 965 | end = round_down(orig_end, align); |
| 966 | |
| 967 | if (start == orig_start && end == orig_end) |
| 968 | continue; |
| 969 | |
| 970 | if (start < end) { |
| 971 | mem->regions[i].base = start; |
| 972 | mem->regions[i].size = end - start; |
| 973 | } else { |
| 974 | memblock_remove_region(mem, i); |
| 975 | i--; |
| 976 | } |
| 977 | } |
| 978 | } |
| 979 | |
| 980 | void __init_memblock memblock_set_current_limit(phys_addr_t limit) |
| 981 | { |
| 982 | memblock.current_limit = limit; |
| 983 | } |
| 984 | |
| 985 | static void __init_memblock memblock_dump(struct memblock_type *type, char *name) |
| 986 | { |
| 987 | unsigned long long base, size; |
| 988 | int i; |
| 989 | |
| 990 | pr_info(" %s.cnt = 0x%lx\n", name, type->cnt); |
| 991 | |
| 992 | for (i = 0; i < type->cnt; i++) { |
| 993 | struct memblock_region *rgn = &type->regions[i]; |
| 994 | char nid_buf[32] = ""; |
| 995 | |
| 996 | base = rgn->base; |
| 997 | size = rgn->size; |
| 998 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
| 999 | if (memblock_get_region_node(rgn) != MAX_NUMNODES) |
| 1000 | snprintf(nid_buf, sizeof(nid_buf), " on node %d", |
| 1001 | memblock_get_region_node(rgn)); |
| 1002 | #endif |
| 1003 | pr_info(" %s[%#x]\t[%#016llx-%#016llx], %#llx bytes%s\n", |
| 1004 | name, i, base, base + size - 1, size, nid_buf); |
| 1005 | } |
| 1006 | } |
| 1007 | |
| 1008 | void __init_memblock __memblock_dump_all(void) |
| 1009 | { |
| 1010 | pr_info("MEMBLOCK configuration:\n"); |
| 1011 | pr_info(" memory size = %#llx reserved size = %#llx\n", |
| 1012 | (unsigned long long)memblock.memory.total_size, |
| 1013 | (unsigned long long)memblock.reserved.total_size); |
| 1014 | |
| 1015 | memblock_dump(&memblock.memory, "memory"); |
| 1016 | memblock_dump(&memblock.reserved, "reserved"); |
| 1017 | } |
| 1018 | |
| 1019 | void __init memblock_allow_resize(void) |
| 1020 | { |
| 1021 | memblock_can_resize = 1; |
| 1022 | } |
| 1023 | |
| 1024 | static int __init early_memblock(char *p) |
| 1025 | { |
| 1026 | if (p && strstr(p, "debug")) |
| 1027 | memblock_debug = 1; |
| 1028 | return 0; |
| 1029 | } |
| 1030 | early_param("memblock", early_memblock); |
| 1031 | |
| 1032 | #if defined(CONFIG_DEBUG_FS) && !defined(CONFIG_ARCH_DISCARD_MEMBLOCK) |
| 1033 | |
| 1034 | static int memblock_debug_show(struct seq_file *m, void *private) |
| 1035 | { |
| 1036 | struct memblock_type *type = m->private; |
| 1037 | struct memblock_region *reg; |
| 1038 | int i; |
| 1039 | |
| 1040 | for (i = 0; i < type->cnt; i++) { |
| 1041 | reg = &type->regions[i]; |
| 1042 | seq_printf(m, "%4d: ", i); |
| 1043 | if (sizeof(phys_addr_t) == 4) |
| 1044 | seq_printf(m, "0x%08lx..0x%08lx\n", |
| 1045 | (unsigned long)reg->base, |
| 1046 | (unsigned long)(reg->base + reg->size - 1)); |
| 1047 | else |
| 1048 | seq_printf(m, "0x%016llx..0x%016llx\n", |
| 1049 | (unsigned long long)reg->base, |
| 1050 | (unsigned long long)(reg->base + reg->size - 1)); |
| 1051 | |
| 1052 | } |
| 1053 | return 0; |
| 1054 | } |
| 1055 | |
| 1056 | static int memblock_debug_open(struct inode *inode, struct file *file) |
| 1057 | { |
| 1058 | return single_open(file, memblock_debug_show, inode->i_private); |
| 1059 | } |
| 1060 | |
| 1061 | static const struct file_operations memblock_debug_fops = { |
| 1062 | .open = memblock_debug_open, |
| 1063 | .read = seq_read, |
| 1064 | .llseek = seq_lseek, |
| 1065 | .release = single_release, |
| 1066 | }; |
| 1067 | |
| 1068 | static int __init memblock_init_debugfs(void) |
| 1069 | { |
| 1070 | struct dentry *root = debugfs_create_dir("memblock", NULL); |
| 1071 | if (!root) |
| 1072 | return -ENXIO; |
| 1073 | debugfs_create_file("memory", S_IRUGO, root, &memblock.memory, &memblock_debug_fops); |
| 1074 | debugfs_create_file("reserved", S_IRUGO, root, &memblock.reserved, &memblock_debug_fops); |
| 1075 | |
| 1076 | return 0; |
| 1077 | } |
| 1078 | __initcall(memblock_init_debugfs); |
| 1079 | |
| 1080 | #endif /* CONFIG_DEBUG_FS */ |