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/kmemleak.h> 21 #include <linux/seq_file.h> 22 #include <linux/memblock.h> 23 24 #include <asm/sections.h> 25 #include <linux/io.h> 26 27 #include "internal.h" 28 29 static struct memblock_region memblock_memory_init_regions[INIT_MEMBLOCK_REGIONS] __initdata_memblock; 30 static struct memblock_region memblock_reserved_init_regions[INIT_MEMBLOCK_REGIONS] __initdata_memblock; 31 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP 32 static struct memblock_region memblock_physmem_init_regions[INIT_PHYSMEM_REGIONS] __initdata_memblock; 33 #endif 34 35 struct memblock memblock __initdata_memblock = { 36 .memory.regions = memblock_memory_init_regions, 37 .memory.cnt = 1, /* empty dummy entry */ 38 .memory.max = INIT_MEMBLOCK_REGIONS, 39 .memory.name = "memory", 40 41 .reserved.regions = memblock_reserved_init_regions, 42 .reserved.cnt = 1, /* empty dummy entry */ 43 .reserved.max = INIT_MEMBLOCK_REGIONS, 44 .reserved.name = "reserved", 45 46 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP 47 .physmem.regions = memblock_physmem_init_regions, 48 .physmem.cnt = 1, /* empty dummy entry */ 49 .physmem.max = INIT_PHYSMEM_REGIONS, 50 .physmem.name = "physmem", 51 #endif 52 53 .bottom_up = false, 54 .current_limit = MEMBLOCK_ALLOC_ANYWHERE, 55 }; 56 57 int memblock_debug __initdata_memblock; 58 static bool system_has_some_mirror __initdata_memblock = false; 59 static int memblock_can_resize __initdata_memblock; 60 static int memblock_memory_in_slab __initdata_memblock = 0; 61 static int memblock_reserved_in_slab __initdata_memblock = 0; 62 63 ulong __init_memblock choose_memblock_flags(void) 64 { 65 return system_has_some_mirror ? MEMBLOCK_MIRROR : MEMBLOCK_NONE; 66 } 67 68 /* adjust *@size so that (@base + *@size) doesn't overflow, return new size */ 69 static inline phys_addr_t memblock_cap_size(phys_addr_t base, phys_addr_t *size) 70 { 71 return *size = min(*size, PHYS_ADDR_MAX - base); 72 } 73 74 /* 75 * Address comparison utilities 76 */ 77 static unsigned long __init_memblock memblock_addrs_overlap(phys_addr_t base1, phys_addr_t size1, 78 phys_addr_t base2, phys_addr_t size2) 79 { 80 return ((base1 < (base2 + size2)) && (base2 < (base1 + size1))); 81 } 82 83 bool __init_memblock memblock_overlaps_region(struct memblock_type *type, 84 phys_addr_t base, phys_addr_t size) 85 { 86 unsigned long i; 87 88 for (i = 0; i < type->cnt; i++) 89 if (memblock_addrs_overlap(base, size, type->regions[i].base, 90 type->regions[i].size)) 91 break; 92 return i < type->cnt; 93 } 94 95 /* 96 * __memblock_find_range_bottom_up - find free area utility in bottom-up 97 * @start: start of candidate range 98 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE} 99 * @size: size of free area to find 100 * @align: alignment of free area to find 101 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 102 * @flags: pick from blocks based on memory attributes 103 * 104 * Utility called from memblock_find_in_range_node(), find free area bottom-up. 105 * 106 * RETURNS: 107 * Found address on success, 0 on failure. 108 */ 109 static phys_addr_t __init_memblock 110 __memblock_find_range_bottom_up(phys_addr_t start, phys_addr_t end, 111 phys_addr_t size, phys_addr_t align, int nid, 112 ulong flags) 113 { 114 phys_addr_t this_start, this_end, cand; 115 u64 i; 116 117 for_each_free_mem_range(i, nid, flags, &this_start, &this_end, NULL) { 118 this_start = clamp(this_start, start, end); 119 this_end = clamp(this_end, start, end); 120 121 cand = round_up(this_start, align); 122 if (cand < this_end && this_end - cand >= size) 123 return cand; 124 } 125 126 return 0; 127 } 128 129 /** 130 * __memblock_find_range_top_down - find free area utility, in top-down 131 * @start: start of candidate range 132 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE} 133 * @size: size of free area to find 134 * @align: alignment of free area to find 135 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 136 * @flags: pick from blocks based on memory attributes 137 * 138 * Utility called from memblock_find_in_range_node(), find free area top-down. 139 * 140 * RETURNS: 141 * Found address on success, 0 on failure. 142 */ 143 static phys_addr_t __init_memblock 144 __memblock_find_range_top_down(phys_addr_t start, phys_addr_t end, 145 phys_addr_t size, phys_addr_t align, int nid, 146 ulong flags) 147 { 148 phys_addr_t this_start, this_end, cand; 149 u64 i; 150 151 for_each_free_mem_range_reverse(i, nid, flags, &this_start, &this_end, 152 NULL) { 153 this_start = clamp(this_start, start, end); 154 this_end = clamp(this_end, start, end); 155 156 if (this_end < size) 157 continue; 158 159 cand = round_down(this_end - size, align); 160 if (cand >= this_start) 161 return cand; 162 } 163 164 return 0; 165 } 166 167 /** 168 * memblock_find_in_range_node - find free area in given range and node 169 * @size: size of free area to find 170 * @align: alignment of free area to find 171 * @start: start of candidate range 172 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE} 173 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 174 * @flags: pick from blocks based on memory attributes 175 * 176 * Find @size free area aligned to @align in the specified range and node. 177 * 178 * When allocation direction is bottom-up, the @start should be greater 179 * than the end of the kernel image. Otherwise, it will be trimmed. The 180 * reason is that we want the bottom-up allocation just near the kernel 181 * image so it is highly likely that the allocated memory and the kernel 182 * will reside in the same node. 183 * 184 * If bottom-up allocation failed, will try to allocate memory top-down. 185 * 186 * RETURNS: 187 * Found address on success, 0 on failure. 188 */ 189 phys_addr_t __init_memblock memblock_find_in_range_node(phys_addr_t size, 190 phys_addr_t align, phys_addr_t start, 191 phys_addr_t end, int nid, ulong flags) 192 { 193 phys_addr_t kernel_end, ret; 194 195 /* pump up @end */ 196 if (end == MEMBLOCK_ALLOC_ACCESSIBLE) 197 end = memblock.current_limit; 198 199 /* avoid allocating the first page */ 200 start = max_t(phys_addr_t, start, PAGE_SIZE); 201 end = max(start, end); 202 kernel_end = __pa_symbol(_end); 203 204 /* 205 * try bottom-up allocation only when bottom-up mode 206 * is set and @end is above the kernel image. 207 */ 208 if (memblock_bottom_up() && end > kernel_end) { 209 phys_addr_t bottom_up_start; 210 211 /* make sure we will allocate above the kernel */ 212 bottom_up_start = max(start, kernel_end); 213 214 /* ok, try bottom-up allocation first */ 215 ret = __memblock_find_range_bottom_up(bottom_up_start, end, 216 size, align, nid, flags); 217 if (ret) 218 return ret; 219 220 /* 221 * we always limit bottom-up allocation above the kernel, 222 * but top-down allocation doesn't have the limit, so 223 * retrying top-down allocation may succeed when bottom-up 224 * allocation failed. 225 * 226 * bottom-up allocation is expected to be fail very rarely, 227 * so we use WARN_ONCE() here to see the stack trace if 228 * fail happens. 229 */ 230 WARN_ONCE(1, "memblock: bottom-up allocation failed, memory hotunplug may be affected\n"); 231 } 232 233 return __memblock_find_range_top_down(start, end, size, align, nid, 234 flags); 235 } 236 237 /** 238 * memblock_find_in_range - find free area in given range 239 * @start: start of candidate range 240 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE} 241 * @size: size of free area to find 242 * @align: alignment of free area to find 243 * 244 * Find @size free area aligned to @align in the specified range. 245 * 246 * RETURNS: 247 * Found address on success, 0 on failure. 248 */ 249 phys_addr_t __init_memblock memblock_find_in_range(phys_addr_t start, 250 phys_addr_t end, phys_addr_t size, 251 phys_addr_t align) 252 { 253 phys_addr_t ret; 254 ulong flags = choose_memblock_flags(); 255 256 again: 257 ret = memblock_find_in_range_node(size, align, start, end, 258 NUMA_NO_NODE, flags); 259 260 if (!ret && (flags & MEMBLOCK_MIRROR)) { 261 pr_warn("Could not allocate %pap bytes of mirrored memory\n", 262 &size); 263 flags &= ~MEMBLOCK_MIRROR; 264 goto again; 265 } 266 267 return ret; 268 } 269 270 static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r) 271 { 272 type->total_size -= type->regions[r].size; 273 memmove(&type->regions[r], &type->regions[r + 1], 274 (type->cnt - (r + 1)) * sizeof(type->regions[r])); 275 type->cnt--; 276 277 /* Special case for empty arrays */ 278 if (type->cnt == 0) { 279 WARN_ON(type->total_size != 0); 280 type->cnt = 1; 281 type->regions[0].base = 0; 282 type->regions[0].size = 0; 283 type->regions[0].flags = 0; 284 memblock_set_region_node(&type->regions[0], MAX_NUMNODES); 285 } 286 } 287 288 #ifdef CONFIG_ARCH_DISCARD_MEMBLOCK 289 /** 290 * Discard memory and reserved arrays if they were allocated 291 */ 292 void __init memblock_discard(void) 293 { 294 phys_addr_t addr, size; 295 296 if (memblock.reserved.regions != memblock_reserved_init_regions) { 297 addr = __pa(memblock.reserved.regions); 298 size = PAGE_ALIGN(sizeof(struct memblock_region) * 299 memblock.reserved.max); 300 __memblock_free_late(addr, size); 301 } 302 303 if (memblock.memory.regions != memblock_memory_init_regions) { 304 addr = __pa(memblock.memory.regions); 305 size = PAGE_ALIGN(sizeof(struct memblock_region) * 306 memblock.memory.max); 307 __memblock_free_late(addr, size); 308 } 309 } 310 #endif 311 312 /** 313 * memblock_double_array - double the size of the memblock regions array 314 * @type: memblock type of the regions array being doubled 315 * @new_area_start: starting address of memory range to avoid overlap with 316 * @new_area_size: size of memory range to avoid overlap with 317 * 318 * Double the size of the @type regions array. If memblock is being used to 319 * allocate memory for a new reserved regions array and there is a previously 320 * allocated memory range [@new_area_start,@new_area_start+@new_area_size] 321 * waiting to be reserved, ensure the memory used by the new array does 322 * not overlap. 323 * 324 * RETURNS: 325 * 0 on success, -1 on failure. 326 */ 327 static int __init_memblock memblock_double_array(struct memblock_type *type, 328 phys_addr_t new_area_start, 329 phys_addr_t new_area_size) 330 { 331 struct memblock_region *new_array, *old_array; 332 phys_addr_t old_alloc_size, new_alloc_size; 333 phys_addr_t old_size, new_size, addr; 334 int use_slab = slab_is_available(); 335 int *in_slab; 336 337 /* We don't allow resizing until we know about the reserved regions 338 * of memory that aren't suitable for allocation 339 */ 340 if (!memblock_can_resize) 341 return -1; 342 343 /* Calculate new doubled size */ 344 old_size = type->max * sizeof(struct memblock_region); 345 new_size = old_size << 1; 346 /* 347 * We need to allocated new one align to PAGE_SIZE, 348 * so we can free them completely later. 349 */ 350 old_alloc_size = PAGE_ALIGN(old_size); 351 new_alloc_size = PAGE_ALIGN(new_size); 352 353 /* Retrieve the slab flag */ 354 if (type == &memblock.memory) 355 in_slab = &memblock_memory_in_slab; 356 else 357 in_slab = &memblock_reserved_in_slab; 358 359 /* Try to find some space for it. 360 * 361 * WARNING: We assume that either slab_is_available() and we use it or 362 * we use MEMBLOCK for allocations. That means that this is unsafe to 363 * use when bootmem is currently active (unless bootmem itself is 364 * implemented on top of MEMBLOCK which isn't the case yet) 365 * 366 * This should however not be an issue for now, as we currently only 367 * call into MEMBLOCK while it's still active, or much later when slab 368 * is active for memory hotplug operations 369 */ 370 if (use_slab) { 371 new_array = kmalloc(new_size, GFP_KERNEL); 372 addr = new_array ? __pa(new_array) : 0; 373 } else { 374 /* only exclude range when trying to double reserved.regions */ 375 if (type != &memblock.reserved) 376 new_area_start = new_area_size = 0; 377 378 addr = memblock_find_in_range(new_area_start + new_area_size, 379 memblock.current_limit, 380 new_alloc_size, PAGE_SIZE); 381 if (!addr && new_area_size) 382 addr = memblock_find_in_range(0, 383 min(new_area_start, memblock.current_limit), 384 new_alloc_size, PAGE_SIZE); 385 386 new_array = addr ? __va(addr) : NULL; 387 } 388 if (!addr) { 389 pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n", 390 type->name, type->max, type->max * 2); 391 return -1; 392 } 393 394 memblock_dbg("memblock: %s is doubled to %ld at [%#010llx-%#010llx]", 395 type->name, type->max * 2, (u64)addr, 396 (u64)addr + new_size - 1); 397 398 /* 399 * Found space, we now need to move the array over before we add the 400 * reserved region since it may be our reserved array itself that is 401 * full. 402 */ 403 memcpy(new_array, type->regions, old_size); 404 memset(new_array + type->max, 0, old_size); 405 old_array = type->regions; 406 type->regions = new_array; 407 type->max <<= 1; 408 409 /* Free old array. We needn't free it if the array is the static one */ 410 if (*in_slab) 411 kfree(old_array); 412 else if (old_array != memblock_memory_init_regions && 413 old_array != memblock_reserved_init_regions) 414 memblock_free(__pa(old_array), old_alloc_size); 415 416 /* 417 * Reserve the new array if that comes from the memblock. Otherwise, we 418 * needn't do it 419 */ 420 if (!use_slab) 421 BUG_ON(memblock_reserve(addr, new_alloc_size)); 422 423 /* Update slab flag */ 424 *in_slab = use_slab; 425 426 return 0; 427 } 428 429 /** 430 * memblock_merge_regions - merge neighboring compatible regions 431 * @type: memblock type to scan 432 * 433 * Scan @type and merge neighboring compatible regions. 434 */ 435 static void __init_memblock memblock_merge_regions(struct memblock_type *type) 436 { 437 int i = 0; 438 439 /* cnt never goes below 1 */ 440 while (i < type->cnt - 1) { 441 struct memblock_region *this = &type->regions[i]; 442 struct memblock_region *next = &type->regions[i + 1]; 443 444 if (this->base + this->size != next->base || 445 memblock_get_region_node(this) != 446 memblock_get_region_node(next) || 447 this->flags != next->flags) { 448 BUG_ON(this->base + this->size > next->base); 449 i++; 450 continue; 451 } 452 453 this->size += next->size; 454 /* move forward from next + 1, index of which is i + 2 */ 455 memmove(next, next + 1, (type->cnt - (i + 2)) * sizeof(*next)); 456 type->cnt--; 457 } 458 } 459 460 /** 461 * memblock_insert_region - insert new memblock region 462 * @type: memblock type to insert into 463 * @idx: index for the insertion point 464 * @base: base address of the new region 465 * @size: size of the new region 466 * @nid: node id of the new region 467 * @flags: flags of the new region 468 * 469 * Insert new memblock region [@base,@base+@size) into @type at @idx. 470 * @type must already have extra room to accommodate the new region. 471 */ 472 static void __init_memblock memblock_insert_region(struct memblock_type *type, 473 int idx, phys_addr_t base, 474 phys_addr_t size, 475 int nid, unsigned long flags) 476 { 477 struct memblock_region *rgn = &type->regions[idx]; 478 479 BUG_ON(type->cnt >= type->max); 480 memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn)); 481 rgn->base = base; 482 rgn->size = size; 483 rgn->flags = flags; 484 memblock_set_region_node(rgn, nid); 485 type->cnt++; 486 type->total_size += size; 487 } 488 489 /** 490 * memblock_add_range - add new memblock region 491 * @type: memblock type to add new region into 492 * @base: base address of the new region 493 * @size: size of the new region 494 * @nid: nid of the new region 495 * @flags: flags of the new region 496 * 497 * Add new memblock region [@base,@base+@size) into @type. The new region 498 * is allowed to overlap with existing ones - overlaps don't affect already 499 * existing regions. @type is guaranteed to be minimal (all neighbouring 500 * compatible regions are merged) after the addition. 501 * 502 * RETURNS: 503 * 0 on success, -errno on failure. 504 */ 505 int __init_memblock memblock_add_range(struct memblock_type *type, 506 phys_addr_t base, phys_addr_t size, 507 int nid, unsigned long flags) 508 { 509 bool insert = false; 510 phys_addr_t obase = base; 511 phys_addr_t end = base + memblock_cap_size(base, &size); 512 int idx, nr_new; 513 struct memblock_region *rgn; 514 515 if (!size) 516 return 0; 517 518 /* special case for empty array */ 519 if (type->regions[0].size == 0) { 520 WARN_ON(type->cnt != 1 || type->total_size); 521 type->regions[0].base = base; 522 type->regions[0].size = size; 523 type->regions[0].flags = flags; 524 memblock_set_region_node(&type->regions[0], nid); 525 type->total_size = size; 526 return 0; 527 } 528 repeat: 529 /* 530 * The following is executed twice. Once with %false @insert and 531 * then with %true. The first counts the number of regions needed 532 * to accommodate the new area. The second actually inserts them. 533 */ 534 base = obase; 535 nr_new = 0; 536 537 for_each_memblock_type(idx, type, rgn) { 538 phys_addr_t rbase = rgn->base; 539 phys_addr_t rend = rbase + rgn->size; 540 541 if (rbase >= end) 542 break; 543 if (rend <= base) 544 continue; 545 /* 546 * @rgn overlaps. If it separates the lower part of new 547 * area, insert that portion. 548 */ 549 if (rbase > base) { 550 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP 551 WARN_ON(nid != memblock_get_region_node(rgn)); 552 #endif 553 WARN_ON(flags != rgn->flags); 554 nr_new++; 555 if (insert) 556 memblock_insert_region(type, idx++, base, 557 rbase - base, nid, 558 flags); 559 } 560 /* area below @rend is dealt with, forget about it */ 561 base = min(rend, end); 562 } 563 564 /* insert the remaining portion */ 565 if (base < end) { 566 nr_new++; 567 if (insert) 568 memblock_insert_region(type, idx, base, end - base, 569 nid, flags); 570 } 571 572 if (!nr_new) 573 return 0; 574 575 /* 576 * If this was the first round, resize array and repeat for actual 577 * insertions; otherwise, merge and return. 578 */ 579 if (!insert) { 580 while (type->cnt + nr_new > type->max) 581 if (memblock_double_array(type, obase, size) < 0) 582 return -ENOMEM; 583 insert = true; 584 goto repeat; 585 } else { 586 memblock_merge_regions(type); 587 return 0; 588 } 589 } 590 591 int __init_memblock memblock_add_node(phys_addr_t base, phys_addr_t size, 592 int nid) 593 { 594 return memblock_add_range(&memblock.memory, base, size, nid, 0); 595 } 596 597 int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size) 598 { 599 phys_addr_t end = base + size - 1; 600 601 memblock_dbg("memblock_add: [%pa-%pa] %pF\n", 602 &base, &end, (void *)_RET_IP_); 603 604 return memblock_add_range(&memblock.memory, base, size, MAX_NUMNODES, 0); 605 } 606 607 /** 608 * memblock_isolate_range - isolate given range into disjoint memblocks 609 * @type: memblock type to isolate range for 610 * @base: base of range to isolate 611 * @size: size of range to isolate 612 * @start_rgn: out parameter for the start of isolated region 613 * @end_rgn: out parameter for the end of isolated region 614 * 615 * Walk @type and ensure that regions don't cross the boundaries defined by 616 * [@base,@base+@size). Crossing regions are split at the boundaries, 617 * which may create at most two more regions. The index of the first 618 * region inside the range is returned in *@start_rgn and end in *@end_rgn. 619 * 620 * RETURNS: 621 * 0 on success, -errno on failure. 622 */ 623 static int __init_memblock memblock_isolate_range(struct memblock_type *type, 624 phys_addr_t base, phys_addr_t size, 625 int *start_rgn, int *end_rgn) 626 { 627 phys_addr_t end = base + memblock_cap_size(base, &size); 628 int idx; 629 struct memblock_region *rgn; 630 631 *start_rgn = *end_rgn = 0; 632 633 if (!size) 634 return 0; 635 636 /* we'll create at most two more regions */ 637 while (type->cnt + 2 > type->max) 638 if (memblock_double_array(type, base, size) < 0) 639 return -ENOMEM; 640 641 for_each_memblock_type(idx, type, rgn) { 642 phys_addr_t rbase = rgn->base; 643 phys_addr_t rend = rbase + rgn->size; 644 645 if (rbase >= end) 646 break; 647 if (rend <= base) 648 continue; 649 650 if (rbase < base) { 651 /* 652 * @rgn intersects from below. Split and continue 653 * to process the next region - the new top half. 654 */ 655 rgn->base = base; 656 rgn->size -= base - rbase; 657 type->total_size -= base - rbase; 658 memblock_insert_region(type, idx, rbase, base - rbase, 659 memblock_get_region_node(rgn), 660 rgn->flags); 661 } else if (rend > end) { 662 /* 663 * @rgn intersects from above. Split and redo the 664 * current region - the new bottom half. 665 */ 666 rgn->base = end; 667 rgn->size -= end - rbase; 668 type->total_size -= end - rbase; 669 memblock_insert_region(type, idx--, rbase, end - rbase, 670 memblock_get_region_node(rgn), 671 rgn->flags); 672 } else { 673 /* @rgn is fully contained, record it */ 674 if (!*end_rgn) 675 *start_rgn = idx; 676 *end_rgn = idx + 1; 677 } 678 } 679 680 return 0; 681 } 682 683 static int __init_memblock memblock_remove_range(struct memblock_type *type, 684 phys_addr_t base, phys_addr_t size) 685 { 686 int start_rgn, end_rgn; 687 int i, ret; 688 689 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn); 690 if (ret) 691 return ret; 692 693 for (i = end_rgn - 1; i >= start_rgn; i--) 694 memblock_remove_region(type, i); 695 return 0; 696 } 697 698 int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size) 699 { 700 phys_addr_t end = base + size - 1; 701 702 memblock_dbg("memblock_remove: [%pa-%pa] %pS\n", 703 &base, &end, (void *)_RET_IP_); 704 705 return memblock_remove_range(&memblock.memory, base, size); 706 } 707 708 709 int __init_memblock memblock_free(phys_addr_t base, phys_addr_t size) 710 { 711 phys_addr_t end = base + size - 1; 712 713 memblock_dbg(" memblock_free: [%pa-%pa] %pF\n", 714 &base, &end, (void *)_RET_IP_); 715 716 kmemleak_free_part_phys(base, size); 717 return memblock_remove_range(&memblock.reserved, base, size); 718 } 719 720 int __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size) 721 { 722 phys_addr_t end = base + size - 1; 723 724 memblock_dbg("memblock_reserve: [%pa-%pa] %pF\n", 725 &base, &end, (void *)_RET_IP_); 726 727 return memblock_add_range(&memblock.reserved, base, size, MAX_NUMNODES, 0); 728 } 729 730 /** 731 * 732 * This function isolates region [@base, @base + @size), and sets/clears flag 733 * 734 * Return 0 on success, -errno on failure. 735 */ 736 static int __init_memblock memblock_setclr_flag(phys_addr_t base, 737 phys_addr_t size, int set, int flag) 738 { 739 struct memblock_type *type = &memblock.memory; 740 int i, ret, start_rgn, end_rgn; 741 742 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn); 743 if (ret) 744 return ret; 745 746 for (i = start_rgn; i < end_rgn; i++) 747 if (set) 748 memblock_set_region_flags(&type->regions[i], flag); 749 else 750 memblock_clear_region_flags(&type->regions[i], flag); 751 752 memblock_merge_regions(type); 753 return 0; 754 } 755 756 /** 757 * memblock_mark_hotplug - Mark hotpluggable memory with flag MEMBLOCK_HOTPLUG. 758 * @base: the base phys addr of the region 759 * @size: the size of the region 760 * 761 * Return 0 on success, -errno on failure. 762 */ 763 int __init_memblock memblock_mark_hotplug(phys_addr_t base, phys_addr_t size) 764 { 765 return memblock_setclr_flag(base, size, 1, MEMBLOCK_HOTPLUG); 766 } 767 768 /** 769 * memblock_clear_hotplug - Clear flag MEMBLOCK_HOTPLUG for a specified region. 770 * @base: the base phys addr of the region 771 * @size: the size of the region 772 * 773 * Return 0 on success, -errno on failure. 774 */ 775 int __init_memblock memblock_clear_hotplug(phys_addr_t base, phys_addr_t size) 776 { 777 return memblock_setclr_flag(base, size, 0, MEMBLOCK_HOTPLUG); 778 } 779 780 /** 781 * memblock_mark_mirror - Mark mirrored memory with flag MEMBLOCK_MIRROR. 782 * @base: the base phys addr of the region 783 * @size: the size of the region 784 * 785 * Return 0 on success, -errno on failure. 786 */ 787 int __init_memblock memblock_mark_mirror(phys_addr_t base, phys_addr_t size) 788 { 789 system_has_some_mirror = true; 790 791 return memblock_setclr_flag(base, size, 1, MEMBLOCK_MIRROR); 792 } 793 794 /** 795 * memblock_mark_nomap - Mark a memory region with flag MEMBLOCK_NOMAP. 796 * @base: the base phys addr of the region 797 * @size: the size of the region 798 * 799 * Return 0 on success, -errno on failure. 800 */ 801 int __init_memblock memblock_mark_nomap(phys_addr_t base, phys_addr_t size) 802 { 803 return memblock_setclr_flag(base, size, 1, MEMBLOCK_NOMAP); 804 } 805 806 /** 807 * memblock_clear_nomap - Clear flag MEMBLOCK_NOMAP for a specified region. 808 * @base: the base phys addr of the region 809 * @size: the size of the region 810 * 811 * Return 0 on success, -errno on failure. 812 */ 813 int __init_memblock memblock_clear_nomap(phys_addr_t base, phys_addr_t size) 814 { 815 return memblock_setclr_flag(base, size, 0, MEMBLOCK_NOMAP); 816 } 817 818 /** 819 * __next_reserved_mem_region - next function for for_each_reserved_region() 820 * @idx: pointer to u64 loop variable 821 * @out_start: ptr to phys_addr_t for start address of the region, can be %NULL 822 * @out_end: ptr to phys_addr_t for end address of the region, can be %NULL 823 * 824 * Iterate over all reserved memory regions. 825 */ 826 void __init_memblock __next_reserved_mem_region(u64 *idx, 827 phys_addr_t *out_start, 828 phys_addr_t *out_end) 829 { 830 struct memblock_type *type = &memblock.reserved; 831 832 if (*idx < type->cnt) { 833 struct memblock_region *r = &type->regions[*idx]; 834 phys_addr_t base = r->base; 835 phys_addr_t size = r->size; 836 837 if (out_start) 838 *out_start = base; 839 if (out_end) 840 *out_end = base + size - 1; 841 842 *idx += 1; 843 return; 844 } 845 846 /* signal end of iteration */ 847 *idx = ULLONG_MAX; 848 } 849 850 /** 851 * __next__mem_range - next function for for_each_free_mem_range() etc. 852 * @idx: pointer to u64 loop variable 853 * @nid: node selector, %NUMA_NO_NODE for all nodes 854 * @flags: pick from blocks based on memory attributes 855 * @type_a: pointer to memblock_type from where the range is taken 856 * @type_b: pointer to memblock_type which excludes memory from being taken 857 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL 858 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL 859 * @out_nid: ptr to int for nid of the range, can be %NULL 860 * 861 * Find the first area from *@idx which matches @nid, fill the out 862 * parameters, and update *@idx for the next iteration. The lower 32bit of 863 * *@idx contains index into type_a and the upper 32bit indexes the 864 * areas before each region in type_b. For example, if type_b regions 865 * look like the following, 866 * 867 * 0:[0-16), 1:[32-48), 2:[128-130) 868 * 869 * The upper 32bit indexes the following regions. 870 * 871 * 0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX) 872 * 873 * As both region arrays are sorted, the function advances the two indices 874 * in lockstep and returns each intersection. 875 */ 876 void __init_memblock __next_mem_range(u64 *idx, int nid, ulong flags, 877 struct memblock_type *type_a, 878 struct memblock_type *type_b, 879 phys_addr_t *out_start, 880 phys_addr_t *out_end, int *out_nid) 881 { 882 int idx_a = *idx & 0xffffffff; 883 int idx_b = *idx >> 32; 884 885 if (WARN_ONCE(nid == MAX_NUMNODES, 886 "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n")) 887 nid = NUMA_NO_NODE; 888 889 for (; idx_a < type_a->cnt; idx_a++) { 890 struct memblock_region *m = &type_a->regions[idx_a]; 891 892 phys_addr_t m_start = m->base; 893 phys_addr_t m_end = m->base + m->size; 894 int m_nid = memblock_get_region_node(m); 895 896 /* only memory regions are associated with nodes, check it */ 897 if (nid != NUMA_NO_NODE && nid != m_nid) 898 continue; 899 900 /* skip hotpluggable memory regions if needed */ 901 if (movable_node_is_enabled() && memblock_is_hotpluggable(m)) 902 continue; 903 904 /* if we want mirror memory skip non-mirror memory regions */ 905 if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m)) 906 continue; 907 908 /* skip nomap memory unless we were asked for it explicitly */ 909 if (!(flags & MEMBLOCK_NOMAP) && memblock_is_nomap(m)) 910 continue; 911 912 if (!type_b) { 913 if (out_start) 914 *out_start = m_start; 915 if (out_end) 916 *out_end = m_end; 917 if (out_nid) 918 *out_nid = m_nid; 919 idx_a++; 920 *idx = (u32)idx_a | (u64)idx_b << 32; 921 return; 922 } 923 924 /* scan areas before each reservation */ 925 for (; idx_b < type_b->cnt + 1; idx_b++) { 926 struct memblock_region *r; 927 phys_addr_t r_start; 928 phys_addr_t r_end; 929 930 r = &type_b->regions[idx_b]; 931 r_start = idx_b ? r[-1].base + r[-1].size : 0; 932 r_end = idx_b < type_b->cnt ? 933 r->base : PHYS_ADDR_MAX; 934 935 /* 936 * if idx_b advanced past idx_a, 937 * break out to advance idx_a 938 */ 939 if (r_start >= m_end) 940 break; 941 /* if the two regions intersect, we're done */ 942 if (m_start < r_end) { 943 if (out_start) 944 *out_start = 945 max(m_start, r_start); 946 if (out_end) 947 *out_end = min(m_end, r_end); 948 if (out_nid) 949 *out_nid = m_nid; 950 /* 951 * The region which ends first is 952 * advanced for the next iteration. 953 */ 954 if (m_end <= r_end) 955 idx_a++; 956 else 957 idx_b++; 958 *idx = (u32)idx_a | (u64)idx_b << 32; 959 return; 960 } 961 } 962 } 963 964 /* signal end of iteration */ 965 *idx = ULLONG_MAX; 966 } 967 968 /** 969 * __next_mem_range_rev - generic next function for for_each_*_range_rev() 970 * 971 * Finds the next range from type_a which is not marked as unsuitable 972 * in type_b. 973 * 974 * @idx: pointer to u64 loop variable 975 * @nid: node selector, %NUMA_NO_NODE for all nodes 976 * @flags: pick from blocks based on memory attributes 977 * @type_a: pointer to memblock_type from where the range is taken 978 * @type_b: pointer to memblock_type which excludes memory from being taken 979 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL 980 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL 981 * @out_nid: ptr to int for nid of the range, can be %NULL 982 * 983 * Reverse of __next_mem_range(). 984 */ 985 void __init_memblock __next_mem_range_rev(u64 *idx, int nid, ulong flags, 986 struct memblock_type *type_a, 987 struct memblock_type *type_b, 988 phys_addr_t *out_start, 989 phys_addr_t *out_end, int *out_nid) 990 { 991 int idx_a = *idx & 0xffffffff; 992 int idx_b = *idx >> 32; 993 994 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n")) 995 nid = NUMA_NO_NODE; 996 997 if (*idx == (u64)ULLONG_MAX) { 998 idx_a = type_a->cnt - 1; 999 if (type_b != NULL) 1000 idx_b = type_b->cnt; 1001 else 1002 idx_b = 0; 1003 } 1004 1005 for (; idx_a >= 0; idx_a--) { 1006 struct memblock_region *m = &type_a->regions[idx_a]; 1007 1008 phys_addr_t m_start = m->base; 1009 phys_addr_t m_end = m->base + m->size; 1010 int m_nid = memblock_get_region_node(m); 1011 1012 /* only memory regions are associated with nodes, check it */ 1013 if (nid != NUMA_NO_NODE && nid != m_nid) 1014 continue; 1015 1016 /* skip hotpluggable memory regions if needed */ 1017 if (movable_node_is_enabled() && memblock_is_hotpluggable(m)) 1018 continue; 1019 1020 /* if we want mirror memory skip non-mirror memory regions */ 1021 if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m)) 1022 continue; 1023 1024 /* skip nomap memory unless we were asked for it explicitly */ 1025 if (!(flags & MEMBLOCK_NOMAP) && memblock_is_nomap(m)) 1026 continue; 1027 1028 if (!type_b) { 1029 if (out_start) 1030 *out_start = m_start; 1031 if (out_end) 1032 *out_end = m_end; 1033 if (out_nid) 1034 *out_nid = m_nid; 1035 idx_a--; 1036 *idx = (u32)idx_a | (u64)idx_b << 32; 1037 return; 1038 } 1039 1040 /* scan areas before each reservation */ 1041 for (; idx_b >= 0; idx_b--) { 1042 struct memblock_region *r; 1043 phys_addr_t r_start; 1044 phys_addr_t r_end; 1045 1046 r = &type_b->regions[idx_b]; 1047 r_start = idx_b ? r[-1].base + r[-1].size : 0; 1048 r_end = idx_b < type_b->cnt ? 1049 r->base : PHYS_ADDR_MAX; 1050 /* 1051 * if idx_b advanced past idx_a, 1052 * break out to advance idx_a 1053 */ 1054 1055 if (r_end <= m_start) 1056 break; 1057 /* if the two regions intersect, we're done */ 1058 if (m_end > r_start) { 1059 if (out_start) 1060 *out_start = max(m_start, r_start); 1061 if (out_end) 1062 *out_end = min(m_end, r_end); 1063 if (out_nid) 1064 *out_nid = m_nid; 1065 if (m_start >= r_start) 1066 idx_a--; 1067 else 1068 idx_b--; 1069 *idx = (u32)idx_a | (u64)idx_b << 32; 1070 return; 1071 } 1072 } 1073 } 1074 /* signal end of iteration */ 1075 *idx = ULLONG_MAX; 1076 } 1077 1078 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP 1079 /* 1080 * Common iterator interface used to define for_each_mem_range(). 1081 */ 1082 void __init_memblock __next_mem_pfn_range(int *idx, int nid, 1083 unsigned long *out_start_pfn, 1084 unsigned long *out_end_pfn, int *out_nid) 1085 { 1086 struct memblock_type *type = &memblock.memory; 1087 struct memblock_region *r; 1088 1089 while (++*idx < type->cnt) { 1090 r = &type->regions[*idx]; 1091 1092 if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size)) 1093 continue; 1094 if (nid == MAX_NUMNODES || nid == r->nid) 1095 break; 1096 } 1097 if (*idx >= type->cnt) { 1098 *idx = -1; 1099 return; 1100 } 1101 1102 if (out_start_pfn) 1103 *out_start_pfn = PFN_UP(r->base); 1104 if (out_end_pfn) 1105 *out_end_pfn = PFN_DOWN(r->base + r->size); 1106 if (out_nid) 1107 *out_nid = r->nid; 1108 } 1109 1110 /** 1111 * memblock_set_node - set node ID on memblock regions 1112 * @base: base of area to set node ID for 1113 * @size: size of area to set node ID for 1114 * @type: memblock type to set node ID for 1115 * @nid: node ID to set 1116 * 1117 * Set the nid of memblock @type regions in [@base,@base+@size) to @nid. 1118 * Regions which cross the area boundaries are split as necessary. 1119 * 1120 * RETURNS: 1121 * 0 on success, -errno on failure. 1122 */ 1123 int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size, 1124 struct memblock_type *type, int nid) 1125 { 1126 int start_rgn, end_rgn; 1127 int i, ret; 1128 1129 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn); 1130 if (ret) 1131 return ret; 1132 1133 for (i = start_rgn; i < end_rgn; i++) 1134 memblock_set_region_node(&type->regions[i], nid); 1135 1136 memblock_merge_regions(type); 1137 return 0; 1138 } 1139 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ 1140 1141 static phys_addr_t __init memblock_alloc_range_nid(phys_addr_t size, 1142 phys_addr_t align, phys_addr_t start, 1143 phys_addr_t end, int nid, ulong flags) 1144 { 1145 phys_addr_t found; 1146 1147 if (!align) 1148 align = SMP_CACHE_BYTES; 1149 1150 found = memblock_find_in_range_node(size, align, start, end, nid, 1151 flags); 1152 if (found && !memblock_reserve(found, size)) { 1153 /* 1154 * The min_count is set to 0 so that memblock allocations are 1155 * never reported as leaks. 1156 */ 1157 kmemleak_alloc_phys(found, size, 0, 0); 1158 return found; 1159 } 1160 return 0; 1161 } 1162 1163 phys_addr_t __init memblock_alloc_range(phys_addr_t size, phys_addr_t align, 1164 phys_addr_t start, phys_addr_t end, 1165 ulong flags) 1166 { 1167 return memblock_alloc_range_nid(size, align, start, end, NUMA_NO_NODE, 1168 flags); 1169 } 1170 1171 phys_addr_t __init memblock_alloc_base_nid(phys_addr_t size, 1172 phys_addr_t align, phys_addr_t max_addr, 1173 int nid, ulong flags) 1174 { 1175 return memblock_alloc_range_nid(size, align, 0, max_addr, nid, flags); 1176 } 1177 1178 phys_addr_t __init memblock_alloc_nid(phys_addr_t size, phys_addr_t align, int nid) 1179 { 1180 ulong flags = choose_memblock_flags(); 1181 phys_addr_t ret; 1182 1183 again: 1184 ret = memblock_alloc_base_nid(size, align, MEMBLOCK_ALLOC_ACCESSIBLE, 1185 nid, flags); 1186 1187 if (!ret && (flags & MEMBLOCK_MIRROR)) { 1188 flags &= ~MEMBLOCK_MIRROR; 1189 goto again; 1190 } 1191 return ret; 1192 } 1193 1194 phys_addr_t __init __memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr) 1195 { 1196 return memblock_alloc_base_nid(size, align, max_addr, NUMA_NO_NODE, 1197 MEMBLOCK_NONE); 1198 } 1199 1200 phys_addr_t __init memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr) 1201 { 1202 phys_addr_t alloc; 1203 1204 alloc = __memblock_alloc_base(size, align, max_addr); 1205 1206 if (alloc == 0) 1207 panic("ERROR: Failed to allocate %pa bytes below %pa.\n", 1208 &size, &max_addr); 1209 1210 return alloc; 1211 } 1212 1213 phys_addr_t __init memblock_alloc(phys_addr_t size, phys_addr_t align) 1214 { 1215 return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE); 1216 } 1217 1218 phys_addr_t __init memblock_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid) 1219 { 1220 phys_addr_t res = memblock_alloc_nid(size, align, nid); 1221 1222 if (res) 1223 return res; 1224 return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE); 1225 } 1226 1227 /** 1228 * memblock_virt_alloc_internal - allocate boot memory block 1229 * @size: size of memory block to be allocated in bytes 1230 * @align: alignment of the region and block's size 1231 * @min_addr: the lower bound of the memory region to allocate (phys address) 1232 * @max_addr: the upper bound of the memory region to allocate (phys address) 1233 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 1234 * 1235 * The @min_addr limit is dropped if it can not be satisfied and the allocation 1236 * will fall back to memory below @min_addr. Also, allocation may fall back 1237 * to any node in the system if the specified node can not 1238 * hold the requested memory. 1239 * 1240 * The allocation is performed from memory region limited by 1241 * memblock.current_limit if @max_addr == %BOOTMEM_ALLOC_ACCESSIBLE. 1242 * 1243 * The memory block is aligned on SMP_CACHE_BYTES if @align == 0. 1244 * 1245 * The phys address of allocated boot memory block is converted to virtual and 1246 * allocated memory is reset to 0. 1247 * 1248 * In addition, function sets the min_count to 0 using kmemleak_alloc for 1249 * allocated boot memory block, so that it is never reported as leaks. 1250 * 1251 * RETURNS: 1252 * Virtual address of allocated memory block on success, NULL on failure. 1253 */ 1254 static void * __init memblock_virt_alloc_internal( 1255 phys_addr_t size, phys_addr_t align, 1256 phys_addr_t min_addr, phys_addr_t max_addr, 1257 int nid) 1258 { 1259 phys_addr_t alloc; 1260 void *ptr; 1261 ulong flags = choose_memblock_flags(); 1262 1263 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n")) 1264 nid = NUMA_NO_NODE; 1265 1266 /* 1267 * Detect any accidental use of these APIs after slab is ready, as at 1268 * this moment memblock may be deinitialized already and its 1269 * internal data may be destroyed (after execution of free_all_bootmem) 1270 */ 1271 if (WARN_ON_ONCE(slab_is_available())) 1272 return kzalloc_node(size, GFP_NOWAIT, nid); 1273 1274 if (!align) 1275 align = SMP_CACHE_BYTES; 1276 1277 if (max_addr > memblock.current_limit) 1278 max_addr = memblock.current_limit; 1279 again: 1280 alloc = memblock_find_in_range_node(size, align, min_addr, max_addr, 1281 nid, flags); 1282 if (alloc && !memblock_reserve(alloc, size)) 1283 goto done; 1284 1285 if (nid != NUMA_NO_NODE) { 1286 alloc = memblock_find_in_range_node(size, align, min_addr, 1287 max_addr, NUMA_NO_NODE, 1288 flags); 1289 if (alloc && !memblock_reserve(alloc, size)) 1290 goto done; 1291 } 1292 1293 if (min_addr) { 1294 min_addr = 0; 1295 goto again; 1296 } 1297 1298 if (flags & MEMBLOCK_MIRROR) { 1299 flags &= ~MEMBLOCK_MIRROR; 1300 pr_warn("Could not allocate %pap bytes of mirrored memory\n", 1301 &size); 1302 goto again; 1303 } 1304 1305 return NULL; 1306 done: 1307 ptr = phys_to_virt(alloc); 1308 1309 /* 1310 * The min_count is set to 0 so that bootmem allocated blocks 1311 * are never reported as leaks. This is because many of these blocks 1312 * are only referred via the physical address which is not 1313 * looked up by kmemleak. 1314 */ 1315 kmemleak_alloc(ptr, size, 0, 0); 1316 1317 return ptr; 1318 } 1319 1320 /** 1321 * memblock_virt_alloc_try_nid_raw - allocate boot memory block without zeroing 1322 * memory and without panicking 1323 * @size: size of memory block to be allocated in bytes 1324 * @align: alignment of the region and block's size 1325 * @min_addr: the lower bound of the memory region from where the allocation 1326 * is preferred (phys address) 1327 * @max_addr: the upper bound of the memory region from where the allocation 1328 * is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to 1329 * allocate only from memory limited by memblock.current_limit value 1330 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 1331 * 1332 * Public function, provides additional debug information (including caller 1333 * info), if enabled. Does not zero allocated memory, does not panic if request 1334 * cannot be satisfied. 1335 * 1336 * RETURNS: 1337 * Virtual address of allocated memory block on success, NULL on failure. 1338 */ 1339 void * __init memblock_virt_alloc_try_nid_raw( 1340 phys_addr_t size, phys_addr_t align, 1341 phys_addr_t min_addr, phys_addr_t max_addr, 1342 int nid) 1343 { 1344 void *ptr; 1345 1346 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n", 1347 __func__, (u64)size, (u64)align, nid, (u64)min_addr, 1348 (u64)max_addr, (void *)_RET_IP_); 1349 1350 ptr = memblock_virt_alloc_internal(size, align, 1351 min_addr, max_addr, nid); 1352 #ifdef CONFIG_DEBUG_VM 1353 if (ptr && size > 0) 1354 memset(ptr, PAGE_POISON_PATTERN, size); 1355 #endif 1356 return ptr; 1357 } 1358 1359 /** 1360 * memblock_virt_alloc_try_nid_nopanic - allocate boot memory block 1361 * @size: size of memory block to be allocated in bytes 1362 * @align: alignment of the region and block's size 1363 * @min_addr: the lower bound of the memory region from where the allocation 1364 * is preferred (phys address) 1365 * @max_addr: the upper bound of the memory region from where the allocation 1366 * is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to 1367 * allocate only from memory limited by memblock.current_limit value 1368 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 1369 * 1370 * Public function, provides additional debug information (including caller 1371 * info), if enabled. This function zeroes the allocated memory. 1372 * 1373 * RETURNS: 1374 * Virtual address of allocated memory block on success, NULL on failure. 1375 */ 1376 void * __init memblock_virt_alloc_try_nid_nopanic( 1377 phys_addr_t size, phys_addr_t align, 1378 phys_addr_t min_addr, phys_addr_t max_addr, 1379 int nid) 1380 { 1381 void *ptr; 1382 1383 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n", 1384 __func__, (u64)size, (u64)align, nid, (u64)min_addr, 1385 (u64)max_addr, (void *)_RET_IP_); 1386 1387 ptr = memblock_virt_alloc_internal(size, align, 1388 min_addr, max_addr, nid); 1389 if (ptr) 1390 memset(ptr, 0, size); 1391 return ptr; 1392 } 1393 1394 /** 1395 * memblock_virt_alloc_try_nid - allocate boot memory block with panicking 1396 * @size: size of memory block to be allocated in bytes 1397 * @align: alignment of the region and block's size 1398 * @min_addr: the lower bound of the memory region from where the allocation 1399 * is preferred (phys address) 1400 * @max_addr: the upper bound of the memory region from where the allocation 1401 * is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to 1402 * allocate only from memory limited by memblock.current_limit value 1403 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 1404 * 1405 * Public panicking version of memblock_virt_alloc_try_nid_nopanic() 1406 * which provides debug information (including caller info), if enabled, 1407 * and panics if the request can not be satisfied. 1408 * 1409 * RETURNS: 1410 * Virtual address of allocated memory block on success, NULL on failure. 1411 */ 1412 void * __init memblock_virt_alloc_try_nid( 1413 phys_addr_t size, phys_addr_t align, 1414 phys_addr_t min_addr, phys_addr_t max_addr, 1415 int nid) 1416 { 1417 void *ptr; 1418 1419 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n", 1420 __func__, (u64)size, (u64)align, nid, (u64)min_addr, 1421 (u64)max_addr, (void *)_RET_IP_); 1422 ptr = memblock_virt_alloc_internal(size, align, 1423 min_addr, max_addr, nid); 1424 if (ptr) { 1425 memset(ptr, 0, size); 1426 return ptr; 1427 } 1428 1429 panic("%s: Failed to allocate %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx\n", 1430 __func__, (u64)size, (u64)align, nid, (u64)min_addr, 1431 (u64)max_addr); 1432 return NULL; 1433 } 1434 1435 /** 1436 * __memblock_free_early - free boot memory block 1437 * @base: phys starting address of the boot memory block 1438 * @size: size of the boot memory block in bytes 1439 * 1440 * Free boot memory block previously allocated by memblock_virt_alloc_xx() API. 1441 * The freeing memory will not be released to the buddy allocator. 1442 */ 1443 void __init __memblock_free_early(phys_addr_t base, phys_addr_t size) 1444 { 1445 memblock_dbg("%s: [%#016llx-%#016llx] %pF\n", 1446 __func__, (u64)base, (u64)base + size - 1, 1447 (void *)_RET_IP_); 1448 kmemleak_free_part_phys(base, size); 1449 memblock_remove_range(&memblock.reserved, base, size); 1450 } 1451 1452 /* 1453 * __memblock_free_late - free bootmem block pages directly to buddy allocator 1454 * @addr: phys starting address of the boot memory block 1455 * @size: size of the boot memory block in bytes 1456 * 1457 * This is only useful when the bootmem allocator has already been torn 1458 * down, but we are still initializing the system. Pages are released directly 1459 * to the buddy allocator, no bootmem metadata is updated because it is gone. 1460 */ 1461 void __init __memblock_free_late(phys_addr_t base, phys_addr_t size) 1462 { 1463 u64 cursor, end; 1464 1465 memblock_dbg("%s: [%#016llx-%#016llx] %pF\n", 1466 __func__, (u64)base, (u64)base + size - 1, 1467 (void *)_RET_IP_); 1468 kmemleak_free_part_phys(base, size); 1469 cursor = PFN_UP(base); 1470 end = PFN_DOWN(base + size); 1471 1472 for (; cursor < end; cursor++) { 1473 __free_pages_bootmem(pfn_to_page(cursor), cursor, 0); 1474 totalram_pages++; 1475 } 1476 } 1477 1478 /* 1479 * Remaining API functions 1480 */ 1481 1482 phys_addr_t __init_memblock memblock_phys_mem_size(void) 1483 { 1484 return memblock.memory.total_size; 1485 } 1486 1487 phys_addr_t __init_memblock memblock_reserved_size(void) 1488 { 1489 return memblock.reserved.total_size; 1490 } 1491 1492 phys_addr_t __init memblock_mem_size(unsigned long limit_pfn) 1493 { 1494 unsigned long pages = 0; 1495 struct memblock_region *r; 1496 unsigned long start_pfn, end_pfn; 1497 1498 for_each_memblock(memory, r) { 1499 start_pfn = memblock_region_memory_base_pfn(r); 1500 end_pfn = memblock_region_memory_end_pfn(r); 1501 start_pfn = min_t(unsigned long, start_pfn, limit_pfn); 1502 end_pfn = min_t(unsigned long, end_pfn, limit_pfn); 1503 pages += end_pfn - start_pfn; 1504 } 1505 1506 return PFN_PHYS(pages); 1507 } 1508 1509 /* lowest address */ 1510 phys_addr_t __init_memblock memblock_start_of_DRAM(void) 1511 { 1512 return memblock.memory.regions[0].base; 1513 } 1514 1515 phys_addr_t __init_memblock memblock_end_of_DRAM(void) 1516 { 1517 int idx = memblock.memory.cnt - 1; 1518 1519 return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size); 1520 } 1521 1522 static phys_addr_t __init_memblock __find_max_addr(phys_addr_t limit) 1523 { 1524 phys_addr_t max_addr = PHYS_ADDR_MAX; 1525 struct memblock_region *r; 1526 1527 /* 1528 * translate the memory @limit size into the max address within one of 1529 * the memory memblock regions, if the @limit exceeds the total size 1530 * of those regions, max_addr will keep original value PHYS_ADDR_MAX 1531 */ 1532 for_each_memblock(memory, r) { 1533 if (limit <= r->size) { 1534 max_addr = r->base + limit; 1535 break; 1536 } 1537 limit -= r->size; 1538 } 1539 1540 return max_addr; 1541 } 1542 1543 void __init memblock_enforce_memory_limit(phys_addr_t limit) 1544 { 1545 phys_addr_t max_addr = PHYS_ADDR_MAX; 1546 1547 if (!limit) 1548 return; 1549 1550 max_addr = __find_max_addr(limit); 1551 1552 /* @limit exceeds the total size of the memory, do nothing */ 1553 if (max_addr == PHYS_ADDR_MAX) 1554 return; 1555 1556 /* truncate both memory and reserved regions */ 1557 memblock_remove_range(&memblock.memory, max_addr, 1558 PHYS_ADDR_MAX); 1559 memblock_remove_range(&memblock.reserved, max_addr, 1560 PHYS_ADDR_MAX); 1561 } 1562 1563 void __init memblock_cap_memory_range(phys_addr_t base, phys_addr_t size) 1564 { 1565 int start_rgn, end_rgn; 1566 int i, ret; 1567 1568 if (!size) 1569 return; 1570 1571 ret = memblock_isolate_range(&memblock.memory, base, size, 1572 &start_rgn, &end_rgn); 1573 if (ret) 1574 return; 1575 1576 /* remove all the MAP regions */ 1577 for (i = memblock.memory.cnt - 1; i >= end_rgn; i--) 1578 if (!memblock_is_nomap(&memblock.memory.regions[i])) 1579 memblock_remove_region(&memblock.memory, i); 1580 1581 for (i = start_rgn - 1; i >= 0; i--) 1582 if (!memblock_is_nomap(&memblock.memory.regions[i])) 1583 memblock_remove_region(&memblock.memory, i); 1584 1585 /* truncate the reserved regions */ 1586 memblock_remove_range(&memblock.reserved, 0, base); 1587 memblock_remove_range(&memblock.reserved, 1588 base + size, PHYS_ADDR_MAX); 1589 } 1590 1591 void __init memblock_mem_limit_remove_map(phys_addr_t limit) 1592 { 1593 phys_addr_t max_addr; 1594 1595 if (!limit) 1596 return; 1597 1598 max_addr = __find_max_addr(limit); 1599 1600 /* @limit exceeds the total size of the memory, do nothing */ 1601 if (max_addr == PHYS_ADDR_MAX) 1602 return; 1603 1604 memblock_cap_memory_range(0, max_addr); 1605 } 1606 1607 static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr) 1608 { 1609 unsigned int left = 0, right = type->cnt; 1610 1611 do { 1612 unsigned int mid = (right + left) / 2; 1613 1614 if (addr < type->regions[mid].base) 1615 right = mid; 1616 else if (addr >= (type->regions[mid].base + 1617 type->regions[mid].size)) 1618 left = mid + 1; 1619 else 1620 return mid; 1621 } while (left < right); 1622 return -1; 1623 } 1624 1625 bool __init memblock_is_reserved(phys_addr_t addr) 1626 { 1627 return memblock_search(&memblock.reserved, addr) != -1; 1628 } 1629 1630 bool __init_memblock memblock_is_memory(phys_addr_t addr) 1631 { 1632 return memblock_search(&memblock.memory, addr) != -1; 1633 } 1634 1635 bool __init_memblock memblock_is_map_memory(phys_addr_t addr) 1636 { 1637 int i = memblock_search(&memblock.memory, addr); 1638 1639 if (i == -1) 1640 return false; 1641 return !memblock_is_nomap(&memblock.memory.regions[i]); 1642 } 1643 1644 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP 1645 int __init_memblock memblock_search_pfn_nid(unsigned long pfn, 1646 unsigned long *start_pfn, unsigned long *end_pfn) 1647 { 1648 struct memblock_type *type = &memblock.memory; 1649 int mid = memblock_search(type, PFN_PHYS(pfn)); 1650 1651 if (mid == -1) 1652 return -1; 1653 1654 *start_pfn = PFN_DOWN(type->regions[mid].base); 1655 *end_pfn = PFN_DOWN(type->regions[mid].base + type->regions[mid].size); 1656 1657 return type->regions[mid].nid; 1658 } 1659 #endif 1660 1661 /** 1662 * memblock_is_region_memory - check if a region is a subset of memory 1663 * @base: base of region to check 1664 * @size: size of region to check 1665 * 1666 * Check if the region [@base, @base+@size) is a subset of a memory block. 1667 * 1668 * RETURNS: 1669 * 0 if false, non-zero if true 1670 */ 1671 bool __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size) 1672 { 1673 int idx = memblock_search(&memblock.memory, base); 1674 phys_addr_t end = base + memblock_cap_size(base, &size); 1675 1676 if (idx == -1) 1677 return false; 1678 return (memblock.memory.regions[idx].base + 1679 memblock.memory.regions[idx].size) >= end; 1680 } 1681 1682 /** 1683 * memblock_is_region_reserved - check if a region intersects reserved memory 1684 * @base: base of region to check 1685 * @size: size of region to check 1686 * 1687 * Check if the region [@base, @base+@size) intersects a reserved memory block. 1688 * 1689 * RETURNS: 1690 * True if they intersect, false if not. 1691 */ 1692 bool __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size) 1693 { 1694 memblock_cap_size(base, &size); 1695 return memblock_overlaps_region(&memblock.reserved, base, size); 1696 } 1697 1698 void __init_memblock memblock_trim_memory(phys_addr_t align) 1699 { 1700 phys_addr_t start, end, orig_start, orig_end; 1701 struct memblock_region *r; 1702 1703 for_each_memblock(memory, r) { 1704 orig_start = r->base; 1705 orig_end = r->base + r->size; 1706 start = round_up(orig_start, align); 1707 end = round_down(orig_end, align); 1708 1709 if (start == orig_start && end == orig_end) 1710 continue; 1711 1712 if (start < end) { 1713 r->base = start; 1714 r->size = end - start; 1715 } else { 1716 memblock_remove_region(&memblock.memory, 1717 r - memblock.memory.regions); 1718 r--; 1719 } 1720 } 1721 } 1722 1723 void __init_memblock memblock_set_current_limit(phys_addr_t limit) 1724 { 1725 memblock.current_limit = limit; 1726 } 1727 1728 phys_addr_t __init_memblock memblock_get_current_limit(void) 1729 { 1730 return memblock.current_limit; 1731 } 1732 1733 static void __init_memblock memblock_dump(struct memblock_type *type) 1734 { 1735 phys_addr_t base, end, size; 1736 unsigned long flags; 1737 int idx; 1738 struct memblock_region *rgn; 1739 1740 pr_info(" %s.cnt = 0x%lx\n", type->name, type->cnt); 1741 1742 for_each_memblock_type(idx, type, rgn) { 1743 char nid_buf[32] = ""; 1744 1745 base = rgn->base; 1746 size = rgn->size; 1747 end = base + size - 1; 1748 flags = rgn->flags; 1749 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP 1750 if (memblock_get_region_node(rgn) != MAX_NUMNODES) 1751 snprintf(nid_buf, sizeof(nid_buf), " on node %d", 1752 memblock_get_region_node(rgn)); 1753 #endif 1754 pr_info(" %s[%#x]\t[%pa-%pa], %pa bytes%s flags: %#lx\n", 1755 type->name, idx, &base, &end, &size, nid_buf, flags); 1756 } 1757 } 1758 1759 void __init_memblock __memblock_dump_all(void) 1760 { 1761 pr_info("MEMBLOCK configuration:\n"); 1762 pr_info(" memory size = %pa reserved size = %pa\n", 1763 &memblock.memory.total_size, 1764 &memblock.reserved.total_size); 1765 1766 memblock_dump(&memblock.memory); 1767 memblock_dump(&memblock.reserved); 1768 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP 1769 memblock_dump(&memblock.physmem); 1770 #endif 1771 } 1772 1773 void __init memblock_allow_resize(void) 1774 { 1775 memblock_can_resize = 1; 1776 } 1777 1778 static int __init early_memblock(char *p) 1779 { 1780 if (p && strstr(p, "debug")) 1781 memblock_debug = 1; 1782 return 0; 1783 } 1784 early_param("memblock", early_memblock); 1785 1786 #if defined(CONFIG_DEBUG_FS) && !defined(CONFIG_ARCH_DISCARD_MEMBLOCK) 1787 1788 static int memblock_debug_show(struct seq_file *m, void *private) 1789 { 1790 struct memblock_type *type = m->private; 1791 struct memblock_region *reg; 1792 int i; 1793 phys_addr_t end; 1794 1795 for (i = 0; i < type->cnt; i++) { 1796 reg = &type->regions[i]; 1797 end = reg->base + reg->size - 1; 1798 1799 seq_printf(m, "%4d: ", i); 1800 seq_printf(m, "%pa..%pa\n", ®->base, &end); 1801 } 1802 return 0; 1803 } 1804 DEFINE_SHOW_ATTRIBUTE(memblock_debug); 1805 1806 static int __init memblock_init_debugfs(void) 1807 { 1808 struct dentry *root = debugfs_create_dir("memblock", NULL); 1809 if (!root) 1810 return -ENXIO; 1811 debugfs_create_file("memory", 0444, root, 1812 &memblock.memory, &memblock_debug_fops); 1813 debugfs_create_file("reserved", 0444, root, 1814 &memblock.reserved, &memblock_debug_fops); 1815 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP 1816 debugfs_create_file("physmem", 0444, root, 1817 &memblock.physmem, &memblock_debug_fops); 1818 #endif 1819 1820 return 0; 1821 } 1822 __initcall(memblock_init_debugfs); 1823 1824 #endif /* CONFIG_DEBUG_FS */ 1825