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 #include <linux/bootmem.h> 24 25 #include <asm/sections.h> 26 #include <linux/io.h> 27 28 #include "internal.h" 29 30 static struct memblock_region memblock_memory_init_regions[INIT_MEMBLOCK_REGIONS] __initdata_memblock; 31 static struct memblock_region memblock_reserved_init_regions[INIT_MEMBLOCK_REGIONS] __initdata_memblock; 32 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP 33 static struct memblock_region memblock_physmem_init_regions[INIT_PHYSMEM_REGIONS] __initdata_memblock; 34 #endif 35 36 struct memblock memblock __initdata_memblock = { 37 .memory.regions = memblock_memory_init_regions, 38 .memory.cnt = 1, /* empty dummy entry */ 39 .memory.max = INIT_MEMBLOCK_REGIONS, 40 .memory.name = "memory", 41 42 .reserved.regions = memblock_reserved_init_regions, 43 .reserved.cnt = 1, /* empty dummy entry */ 44 .reserved.max = INIT_MEMBLOCK_REGIONS, 45 .reserved.name = "reserved", 46 47 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP 48 .physmem.regions = memblock_physmem_init_regions, 49 .physmem.cnt = 1, /* empty dummy entry */ 50 .physmem.max = INIT_PHYSMEM_REGIONS, 51 .physmem.name = "physmem", 52 #endif 53 54 .bottom_up = false, 55 .current_limit = MEMBLOCK_ALLOC_ANYWHERE, 56 }; 57 58 int memblock_debug __initdata_memblock; 59 static bool system_has_some_mirror __initdata_memblock = false; 60 static int memblock_can_resize __initdata_memblock; 61 static int memblock_memory_in_slab __initdata_memblock = 0; 62 static int memblock_reserved_in_slab __initdata_memblock = 0; 63 64 ulong __init_memblock choose_memblock_flags(void) 65 { 66 return system_has_some_mirror ? MEMBLOCK_MIRROR : MEMBLOCK_NONE; 67 } 68 69 /* adjust *@size so that (@base + *@size) doesn't overflow, return new size */ 70 static inline phys_addr_t memblock_cap_size(phys_addr_t base, phys_addr_t *size) 71 { 72 return *size = min(*size, PHYS_ADDR_MAX - base); 73 } 74 75 /* 76 * Address comparison utilities 77 */ 78 static unsigned long __init_memblock memblock_addrs_overlap(phys_addr_t base1, phys_addr_t size1, 79 phys_addr_t base2, phys_addr_t size2) 80 { 81 return ((base1 < (base2 + size2)) && (base2 < (base1 + size1))); 82 } 83 84 bool __init_memblock memblock_overlaps_region(struct memblock_type *type, 85 phys_addr_t base, phys_addr_t size) 86 { 87 unsigned long i; 88 89 for (i = 0; i < type->cnt; i++) 90 if (memblock_addrs_overlap(base, size, type->regions[i].base, 91 type->regions[i].size)) 92 break; 93 return i < type->cnt; 94 } 95 96 /* 97 * __memblock_find_range_bottom_up - find free area utility in bottom-up 98 * @start: start of candidate range 99 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE} 100 * @size: size of free area to find 101 * @align: alignment of free area to find 102 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 103 * @flags: pick from blocks based on memory attributes 104 * 105 * Utility called from memblock_find_in_range_node(), find free area bottom-up. 106 * 107 * RETURNS: 108 * Found address on success, 0 on failure. 109 */ 110 static phys_addr_t __init_memblock 111 __memblock_find_range_bottom_up(phys_addr_t start, phys_addr_t end, 112 phys_addr_t size, phys_addr_t align, int nid, 113 ulong flags) 114 { 115 phys_addr_t this_start, this_end, cand; 116 u64 i; 117 118 for_each_free_mem_range(i, nid, flags, &this_start, &this_end, NULL) { 119 this_start = clamp(this_start, start, end); 120 this_end = clamp(this_end, start, end); 121 122 cand = round_up(this_start, align); 123 if (cand < this_end && this_end - cand >= size) 124 return cand; 125 } 126 127 return 0; 128 } 129 130 /** 131 * __memblock_find_range_top_down - find free area utility, in top-down 132 * @start: start of candidate range 133 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE} 134 * @size: size of free area to find 135 * @align: alignment of free area to find 136 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 137 * @flags: pick from blocks based on memory attributes 138 * 139 * Utility called from memblock_find_in_range_node(), find free area top-down. 140 * 141 * RETURNS: 142 * Found address on success, 0 on failure. 143 */ 144 static phys_addr_t __init_memblock 145 __memblock_find_range_top_down(phys_addr_t start, phys_addr_t end, 146 phys_addr_t size, phys_addr_t align, int nid, 147 ulong flags) 148 { 149 phys_addr_t this_start, this_end, cand; 150 u64 i; 151 152 for_each_free_mem_range_reverse(i, nid, flags, &this_start, &this_end, 153 NULL) { 154 this_start = clamp(this_start, start, end); 155 this_end = clamp(this_end, start, end); 156 157 if (this_end < size) 158 continue; 159 160 cand = round_down(this_end - size, align); 161 if (cand >= this_start) 162 return cand; 163 } 164 165 return 0; 166 } 167 168 /** 169 * memblock_find_in_range_node - find free area in given range and node 170 * @size: size of free area to find 171 * @align: alignment of free area to find 172 * @start: start of candidate range 173 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE} 174 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 175 * @flags: pick from blocks based on memory attributes 176 * 177 * Find @size free area aligned to @align in the specified range and node. 178 * 179 * When allocation direction is bottom-up, the @start should be greater 180 * than the end of the kernel image. Otherwise, it will be trimmed. The 181 * reason is that we want the bottom-up allocation just near the kernel 182 * image so it is highly likely that the allocated memory and the kernel 183 * will reside in the same node. 184 * 185 * If bottom-up allocation failed, will try to allocate memory top-down. 186 * 187 * RETURNS: 188 * Found address on success, 0 on failure. 189 */ 190 phys_addr_t __init_memblock memblock_find_in_range_node(phys_addr_t size, 191 phys_addr_t align, phys_addr_t start, 192 phys_addr_t end, int nid, ulong flags) 193 { 194 phys_addr_t kernel_end, ret; 195 196 /* pump up @end */ 197 if (end == MEMBLOCK_ALLOC_ACCESSIBLE) 198 end = memblock.current_limit; 199 200 /* avoid allocating the first page */ 201 start = max_t(phys_addr_t, start, PAGE_SIZE); 202 end = max(start, end); 203 kernel_end = __pa_symbol(_end); 204 205 /* 206 * try bottom-up allocation only when bottom-up mode 207 * is set and @end is above the kernel image. 208 */ 209 if (memblock_bottom_up() && end > kernel_end) { 210 phys_addr_t bottom_up_start; 211 212 /* make sure we will allocate above the kernel */ 213 bottom_up_start = max(start, kernel_end); 214 215 /* ok, try bottom-up allocation first */ 216 ret = __memblock_find_range_bottom_up(bottom_up_start, end, 217 size, align, nid, flags); 218 if (ret) 219 return ret; 220 221 /* 222 * we always limit bottom-up allocation above the kernel, 223 * but top-down allocation doesn't have the limit, so 224 * retrying top-down allocation may succeed when bottom-up 225 * allocation failed. 226 * 227 * bottom-up allocation is expected to be fail very rarely, 228 * so we use WARN_ONCE() here to see the stack trace if 229 * fail happens. 230 */ 231 WARN_ONCE(IS_ENABLED(CONFIG_MEMORY_HOTREMOVE), 232 "memblock: bottom-up allocation failed, memory hotremove may be affected\n"); 233 } 234 235 return __memblock_find_range_top_down(start, end, size, align, nid, 236 flags); 237 } 238 239 /** 240 * memblock_find_in_range - find free area in given range 241 * @start: start of candidate range 242 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE} 243 * @size: size of free area to find 244 * @align: alignment of free area to find 245 * 246 * Find @size free area aligned to @align in the specified range. 247 * 248 * RETURNS: 249 * Found address on success, 0 on failure. 250 */ 251 phys_addr_t __init_memblock memblock_find_in_range(phys_addr_t start, 252 phys_addr_t end, phys_addr_t size, 253 phys_addr_t align) 254 { 255 phys_addr_t ret; 256 ulong flags = choose_memblock_flags(); 257 258 again: 259 ret = memblock_find_in_range_node(size, align, start, end, 260 NUMA_NO_NODE, flags); 261 262 if (!ret && (flags & MEMBLOCK_MIRROR)) { 263 pr_warn("Could not allocate %pap bytes of mirrored memory\n", 264 &size); 265 flags &= ~MEMBLOCK_MIRROR; 266 goto again; 267 } 268 269 return ret; 270 } 271 272 static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r) 273 { 274 type->total_size -= type->regions[r].size; 275 memmove(&type->regions[r], &type->regions[r + 1], 276 (type->cnt - (r + 1)) * sizeof(type->regions[r])); 277 type->cnt--; 278 279 /* Special case for empty arrays */ 280 if (type->cnt == 0) { 281 WARN_ON(type->total_size != 0); 282 type->cnt = 1; 283 type->regions[0].base = 0; 284 type->regions[0].size = 0; 285 type->regions[0].flags = 0; 286 memblock_set_region_node(&type->regions[0], MAX_NUMNODES); 287 } 288 } 289 290 #ifdef CONFIG_ARCH_DISCARD_MEMBLOCK 291 /** 292 * Discard memory and reserved arrays if they were allocated 293 */ 294 void __init memblock_discard(void) 295 { 296 phys_addr_t addr, size; 297 298 if (memblock.reserved.regions != memblock_reserved_init_regions) { 299 addr = __pa(memblock.reserved.regions); 300 size = PAGE_ALIGN(sizeof(struct memblock_region) * 301 memblock.reserved.max); 302 __memblock_free_late(addr, size); 303 } 304 305 if (memblock.memory.regions != memblock_memory_init_regions) { 306 addr = __pa(memblock.memory.regions); 307 size = PAGE_ALIGN(sizeof(struct memblock_region) * 308 memblock.memory.max); 309 __memblock_free_late(addr, size); 310 } 311 } 312 #endif 313 314 /** 315 * memblock_double_array - double the size of the memblock regions array 316 * @type: memblock type of the regions array being doubled 317 * @new_area_start: starting address of memory range to avoid overlap with 318 * @new_area_size: size of memory range to avoid overlap with 319 * 320 * Double the size of the @type regions array. If memblock is being used to 321 * allocate memory for a new reserved regions array and there is a previously 322 * allocated memory range [@new_area_start,@new_area_start+@new_area_size] 323 * waiting to be reserved, ensure the memory used by the new array does 324 * not overlap. 325 * 326 * RETURNS: 327 * 0 on success, -1 on failure. 328 */ 329 static int __init_memblock memblock_double_array(struct memblock_type *type, 330 phys_addr_t new_area_start, 331 phys_addr_t new_area_size) 332 { 333 struct memblock_region *new_array, *old_array; 334 phys_addr_t old_alloc_size, new_alloc_size; 335 phys_addr_t old_size, new_size, addr; 336 int use_slab = slab_is_available(); 337 int *in_slab; 338 339 /* We don't allow resizing until we know about the reserved regions 340 * of memory that aren't suitable for allocation 341 */ 342 if (!memblock_can_resize) 343 return -1; 344 345 /* Calculate new doubled size */ 346 old_size = type->max * sizeof(struct memblock_region); 347 new_size = old_size << 1; 348 /* 349 * We need to allocated new one align to PAGE_SIZE, 350 * so we can free them completely later. 351 */ 352 old_alloc_size = PAGE_ALIGN(old_size); 353 new_alloc_size = PAGE_ALIGN(new_size); 354 355 /* Retrieve the slab flag */ 356 if (type == &memblock.memory) 357 in_slab = &memblock_memory_in_slab; 358 else 359 in_slab = &memblock_reserved_in_slab; 360 361 /* Try to find some space for it. 362 * 363 * WARNING: We assume that either slab_is_available() and we use it or 364 * we use MEMBLOCK for allocations. That means that this is unsafe to 365 * use when bootmem is currently active (unless bootmem itself is 366 * implemented on top of MEMBLOCK which isn't the case yet) 367 * 368 * This should however not be an issue for now, as we currently only 369 * call into MEMBLOCK while it's still active, or much later when slab 370 * is active for memory hotplug operations 371 */ 372 if (use_slab) { 373 new_array = kmalloc(new_size, GFP_KERNEL); 374 addr = new_array ? __pa(new_array) : 0; 375 } else { 376 /* only exclude range when trying to double reserved.regions */ 377 if (type != &memblock.reserved) 378 new_area_start = new_area_size = 0; 379 380 addr = memblock_find_in_range(new_area_start + new_area_size, 381 memblock.current_limit, 382 new_alloc_size, PAGE_SIZE); 383 if (!addr && new_area_size) 384 addr = memblock_find_in_range(0, 385 min(new_area_start, memblock.current_limit), 386 new_alloc_size, PAGE_SIZE); 387 388 new_array = addr ? __va(addr) : NULL; 389 } 390 if (!addr) { 391 pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n", 392 type->name, type->max, type->max * 2); 393 return -1; 394 } 395 396 memblock_dbg("memblock: %s is doubled to %ld at [%#010llx-%#010llx]", 397 type->name, type->max * 2, (u64)addr, 398 (u64)addr + new_size - 1); 399 400 /* 401 * Found space, we now need to move the array over before we add the 402 * reserved region since it may be our reserved array itself that is 403 * full. 404 */ 405 memcpy(new_array, type->regions, old_size); 406 memset(new_array + type->max, 0, old_size); 407 old_array = type->regions; 408 type->regions = new_array; 409 type->max <<= 1; 410 411 /* Free old array. We needn't free it if the array is the static one */ 412 if (*in_slab) 413 kfree(old_array); 414 else if (old_array != memblock_memory_init_regions && 415 old_array != memblock_reserved_init_regions) 416 memblock_free(__pa(old_array), old_alloc_size); 417 418 /* 419 * Reserve the new array if that comes from the memblock. Otherwise, we 420 * needn't do it 421 */ 422 if (!use_slab) 423 BUG_ON(memblock_reserve(addr, new_alloc_size)); 424 425 /* Update slab flag */ 426 *in_slab = use_slab; 427 428 return 0; 429 } 430 431 /** 432 * memblock_merge_regions - merge neighboring compatible regions 433 * @type: memblock type to scan 434 * 435 * Scan @type and merge neighboring compatible regions. 436 */ 437 static void __init_memblock memblock_merge_regions(struct memblock_type *type) 438 { 439 int i = 0; 440 441 /* cnt never goes below 1 */ 442 while (i < type->cnt - 1) { 443 struct memblock_region *this = &type->regions[i]; 444 struct memblock_region *next = &type->regions[i + 1]; 445 446 if (this->base + this->size != next->base || 447 memblock_get_region_node(this) != 448 memblock_get_region_node(next) || 449 this->flags != next->flags) { 450 BUG_ON(this->base + this->size > next->base); 451 i++; 452 continue; 453 } 454 455 this->size += next->size; 456 /* move forward from next + 1, index of which is i + 2 */ 457 memmove(next, next + 1, (type->cnt - (i + 2)) * sizeof(*next)); 458 type->cnt--; 459 } 460 } 461 462 /** 463 * memblock_insert_region - insert new memblock region 464 * @type: memblock type to insert into 465 * @idx: index for the insertion point 466 * @base: base address of the new region 467 * @size: size of the new region 468 * @nid: node id of the new region 469 * @flags: flags of the new region 470 * 471 * Insert new memblock region [@base,@base+@size) into @type at @idx. 472 * @type must already have extra room to accommodate the new region. 473 */ 474 static void __init_memblock memblock_insert_region(struct memblock_type *type, 475 int idx, phys_addr_t base, 476 phys_addr_t size, 477 int nid, unsigned long flags) 478 { 479 struct memblock_region *rgn = &type->regions[idx]; 480 481 BUG_ON(type->cnt >= type->max); 482 memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn)); 483 rgn->base = base; 484 rgn->size = size; 485 rgn->flags = flags; 486 memblock_set_region_node(rgn, nid); 487 type->cnt++; 488 type->total_size += size; 489 } 490 491 /** 492 * memblock_add_range - add new memblock region 493 * @type: memblock type to add new region into 494 * @base: base address of the new region 495 * @size: size of the new region 496 * @nid: nid of the new region 497 * @flags: flags of the new region 498 * 499 * Add new memblock region [@base,@base+@size) into @type. The new region 500 * is allowed to overlap with existing ones - overlaps don't affect already 501 * existing regions. @type is guaranteed to be minimal (all neighbouring 502 * compatible regions are merged) after the addition. 503 * 504 * RETURNS: 505 * 0 on success, -errno on failure. 506 */ 507 int __init_memblock memblock_add_range(struct memblock_type *type, 508 phys_addr_t base, phys_addr_t size, 509 int nid, unsigned long flags) 510 { 511 bool insert = false; 512 phys_addr_t obase = base; 513 phys_addr_t end = base + memblock_cap_size(base, &size); 514 int idx, nr_new; 515 struct memblock_region *rgn; 516 517 if (!size) 518 return 0; 519 520 /* special case for empty array */ 521 if (type->regions[0].size == 0) { 522 WARN_ON(type->cnt != 1 || type->total_size); 523 type->regions[0].base = base; 524 type->regions[0].size = size; 525 type->regions[0].flags = flags; 526 memblock_set_region_node(&type->regions[0], nid); 527 type->total_size = size; 528 return 0; 529 } 530 repeat: 531 /* 532 * The following is executed twice. Once with %false @insert and 533 * then with %true. The first counts the number of regions needed 534 * to accommodate the new area. The second actually inserts them. 535 */ 536 base = obase; 537 nr_new = 0; 538 539 for_each_memblock_type(idx, type, rgn) { 540 phys_addr_t rbase = rgn->base; 541 phys_addr_t rend = rbase + rgn->size; 542 543 if (rbase >= end) 544 break; 545 if (rend <= base) 546 continue; 547 /* 548 * @rgn overlaps. If it separates the lower part of new 549 * area, insert that portion. 550 */ 551 if (rbase > base) { 552 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP 553 WARN_ON(nid != memblock_get_region_node(rgn)); 554 #endif 555 WARN_ON(flags != rgn->flags); 556 nr_new++; 557 if (insert) 558 memblock_insert_region(type, idx++, base, 559 rbase - base, nid, 560 flags); 561 } 562 /* area below @rend is dealt with, forget about it */ 563 base = min(rend, end); 564 } 565 566 /* insert the remaining portion */ 567 if (base < end) { 568 nr_new++; 569 if (insert) 570 memblock_insert_region(type, idx, base, end - base, 571 nid, flags); 572 } 573 574 if (!nr_new) 575 return 0; 576 577 /* 578 * If this was the first round, resize array and repeat for actual 579 * insertions; otherwise, merge and return. 580 */ 581 if (!insert) { 582 while (type->cnt + nr_new > type->max) 583 if (memblock_double_array(type, obase, size) < 0) 584 return -ENOMEM; 585 insert = true; 586 goto repeat; 587 } else { 588 memblock_merge_regions(type); 589 return 0; 590 } 591 } 592 593 int __init_memblock memblock_add_node(phys_addr_t base, phys_addr_t size, 594 int nid) 595 { 596 return memblock_add_range(&memblock.memory, base, size, nid, 0); 597 } 598 599 int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size) 600 { 601 phys_addr_t end = base + size - 1; 602 603 memblock_dbg("memblock_add: [%pa-%pa] %pF\n", 604 &base, &end, (void *)_RET_IP_); 605 606 return memblock_add_range(&memblock.memory, base, size, MAX_NUMNODES, 0); 607 } 608 609 /** 610 * memblock_isolate_range - isolate given range into disjoint memblocks 611 * @type: memblock type to isolate range for 612 * @base: base of range to isolate 613 * @size: size of range to isolate 614 * @start_rgn: out parameter for the start of isolated region 615 * @end_rgn: out parameter for the end of isolated region 616 * 617 * Walk @type and ensure that regions don't cross the boundaries defined by 618 * [@base,@base+@size). Crossing regions are split at the boundaries, 619 * which may create at most two more regions. The index of the first 620 * region inside the range is returned in *@start_rgn and end in *@end_rgn. 621 * 622 * RETURNS: 623 * 0 on success, -errno on failure. 624 */ 625 static int __init_memblock memblock_isolate_range(struct memblock_type *type, 626 phys_addr_t base, phys_addr_t size, 627 int *start_rgn, int *end_rgn) 628 { 629 phys_addr_t end = base + memblock_cap_size(base, &size); 630 int idx; 631 struct memblock_region *rgn; 632 633 *start_rgn = *end_rgn = 0; 634 635 if (!size) 636 return 0; 637 638 /* we'll create at most two more regions */ 639 while (type->cnt + 2 > type->max) 640 if (memblock_double_array(type, base, size) < 0) 641 return -ENOMEM; 642 643 for_each_memblock_type(idx, type, rgn) { 644 phys_addr_t rbase = rgn->base; 645 phys_addr_t rend = rbase + rgn->size; 646 647 if (rbase >= end) 648 break; 649 if (rend <= base) 650 continue; 651 652 if (rbase < base) { 653 /* 654 * @rgn intersects from below. Split and continue 655 * to process the next region - the new top half. 656 */ 657 rgn->base = base; 658 rgn->size -= base - rbase; 659 type->total_size -= base - rbase; 660 memblock_insert_region(type, idx, rbase, base - rbase, 661 memblock_get_region_node(rgn), 662 rgn->flags); 663 } else if (rend > end) { 664 /* 665 * @rgn intersects from above. Split and redo the 666 * current region - the new bottom half. 667 */ 668 rgn->base = end; 669 rgn->size -= end - rbase; 670 type->total_size -= end - rbase; 671 memblock_insert_region(type, idx--, rbase, end - rbase, 672 memblock_get_region_node(rgn), 673 rgn->flags); 674 } else { 675 /* @rgn is fully contained, record it */ 676 if (!*end_rgn) 677 *start_rgn = idx; 678 *end_rgn = idx + 1; 679 } 680 } 681 682 return 0; 683 } 684 685 static int __init_memblock memblock_remove_range(struct memblock_type *type, 686 phys_addr_t base, phys_addr_t size) 687 { 688 int start_rgn, end_rgn; 689 int i, ret; 690 691 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn); 692 if (ret) 693 return ret; 694 695 for (i = end_rgn - 1; i >= start_rgn; i--) 696 memblock_remove_region(type, i); 697 return 0; 698 } 699 700 int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size) 701 { 702 phys_addr_t end = base + size - 1; 703 704 memblock_dbg("memblock_remove: [%pa-%pa] %pS\n", 705 &base, &end, (void *)_RET_IP_); 706 707 return memblock_remove_range(&memblock.memory, base, size); 708 } 709 710 711 int __init_memblock memblock_free(phys_addr_t base, phys_addr_t size) 712 { 713 phys_addr_t end = base + size - 1; 714 715 memblock_dbg(" memblock_free: [%pa-%pa] %pF\n", 716 &base, &end, (void *)_RET_IP_); 717 718 kmemleak_free_part_phys(base, size); 719 return memblock_remove_range(&memblock.reserved, base, size); 720 } 721 722 int __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size) 723 { 724 phys_addr_t end = base + size - 1; 725 726 memblock_dbg("memblock_reserve: [%pa-%pa] %pF\n", 727 &base, &end, (void *)_RET_IP_); 728 729 return memblock_add_range(&memblock.reserved, base, size, MAX_NUMNODES, 0); 730 } 731 732 /** 733 * 734 * This function isolates region [@base, @base + @size), and sets/clears flag 735 * 736 * Return 0 on success, -errno on failure. 737 */ 738 static int __init_memblock memblock_setclr_flag(phys_addr_t base, 739 phys_addr_t size, int set, int flag) 740 { 741 struct memblock_type *type = &memblock.memory; 742 int i, ret, start_rgn, end_rgn; 743 744 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn); 745 if (ret) 746 return ret; 747 748 for (i = start_rgn; i < end_rgn; i++) 749 if (set) 750 memblock_set_region_flags(&type->regions[i], flag); 751 else 752 memblock_clear_region_flags(&type->regions[i], flag); 753 754 memblock_merge_regions(type); 755 return 0; 756 } 757 758 /** 759 * memblock_mark_hotplug - Mark hotpluggable memory with flag MEMBLOCK_HOTPLUG. 760 * @base: the base phys addr of the region 761 * @size: the size of the region 762 * 763 * Return 0 on success, -errno on failure. 764 */ 765 int __init_memblock memblock_mark_hotplug(phys_addr_t base, phys_addr_t size) 766 { 767 return memblock_setclr_flag(base, size, 1, MEMBLOCK_HOTPLUG); 768 } 769 770 /** 771 * memblock_clear_hotplug - Clear flag MEMBLOCK_HOTPLUG for a specified region. 772 * @base: the base phys addr of the region 773 * @size: the size of the region 774 * 775 * Return 0 on success, -errno on failure. 776 */ 777 int __init_memblock memblock_clear_hotplug(phys_addr_t base, phys_addr_t size) 778 { 779 return memblock_setclr_flag(base, size, 0, MEMBLOCK_HOTPLUG); 780 } 781 782 /** 783 * memblock_mark_mirror - Mark mirrored memory with flag MEMBLOCK_MIRROR. 784 * @base: the base phys addr of the region 785 * @size: the size of the region 786 * 787 * Return 0 on success, -errno on failure. 788 */ 789 int __init_memblock memblock_mark_mirror(phys_addr_t base, phys_addr_t size) 790 { 791 system_has_some_mirror = true; 792 793 return memblock_setclr_flag(base, size, 1, MEMBLOCK_MIRROR); 794 } 795 796 /** 797 * memblock_mark_nomap - Mark a memory region with flag MEMBLOCK_NOMAP. 798 * @base: the base phys addr of the region 799 * @size: the size of the region 800 * 801 * Return 0 on success, -errno on failure. 802 */ 803 int __init_memblock memblock_mark_nomap(phys_addr_t base, phys_addr_t size) 804 { 805 return memblock_setclr_flag(base, size, 1, MEMBLOCK_NOMAP); 806 } 807 808 /** 809 * memblock_clear_nomap - Clear flag MEMBLOCK_NOMAP for a specified region. 810 * @base: the base phys addr of the region 811 * @size: the size of the region 812 * 813 * Return 0 on success, -errno on failure. 814 */ 815 int __init_memblock memblock_clear_nomap(phys_addr_t base, phys_addr_t size) 816 { 817 return memblock_setclr_flag(base, size, 0, MEMBLOCK_NOMAP); 818 } 819 820 /** 821 * __next_reserved_mem_region - next function for for_each_reserved_region() 822 * @idx: pointer to u64 loop variable 823 * @out_start: ptr to phys_addr_t for start address of the region, can be %NULL 824 * @out_end: ptr to phys_addr_t for end address of the region, can be %NULL 825 * 826 * Iterate over all reserved memory regions. 827 */ 828 void __init_memblock __next_reserved_mem_region(u64 *idx, 829 phys_addr_t *out_start, 830 phys_addr_t *out_end) 831 { 832 struct memblock_type *type = &memblock.reserved; 833 834 if (*idx < type->cnt) { 835 struct memblock_region *r = &type->regions[*idx]; 836 phys_addr_t base = r->base; 837 phys_addr_t size = r->size; 838 839 if (out_start) 840 *out_start = base; 841 if (out_end) 842 *out_end = base + size - 1; 843 844 *idx += 1; 845 return; 846 } 847 848 /* signal end of iteration */ 849 *idx = ULLONG_MAX; 850 } 851 852 /** 853 * __next__mem_range - next function for for_each_free_mem_range() etc. 854 * @idx: pointer to u64 loop variable 855 * @nid: node selector, %NUMA_NO_NODE for all nodes 856 * @flags: pick from blocks based on memory attributes 857 * @type_a: pointer to memblock_type from where the range is taken 858 * @type_b: pointer to memblock_type which excludes memory from being taken 859 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL 860 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL 861 * @out_nid: ptr to int for nid of the range, can be %NULL 862 * 863 * Find the first area from *@idx which matches @nid, fill the out 864 * parameters, and update *@idx for the next iteration. The lower 32bit of 865 * *@idx contains index into type_a and the upper 32bit indexes the 866 * areas before each region in type_b. For example, if type_b regions 867 * look like the following, 868 * 869 * 0:[0-16), 1:[32-48), 2:[128-130) 870 * 871 * The upper 32bit indexes the following regions. 872 * 873 * 0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX) 874 * 875 * As both region arrays are sorted, the function advances the two indices 876 * in lockstep and returns each intersection. 877 */ 878 void __init_memblock __next_mem_range(u64 *idx, int nid, ulong flags, 879 struct memblock_type *type_a, 880 struct memblock_type *type_b, 881 phys_addr_t *out_start, 882 phys_addr_t *out_end, int *out_nid) 883 { 884 int idx_a = *idx & 0xffffffff; 885 int idx_b = *idx >> 32; 886 887 if (WARN_ONCE(nid == MAX_NUMNODES, 888 "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n")) 889 nid = NUMA_NO_NODE; 890 891 for (; idx_a < type_a->cnt; idx_a++) { 892 struct memblock_region *m = &type_a->regions[idx_a]; 893 894 phys_addr_t m_start = m->base; 895 phys_addr_t m_end = m->base + m->size; 896 int m_nid = memblock_get_region_node(m); 897 898 /* only memory regions are associated with nodes, check it */ 899 if (nid != NUMA_NO_NODE && nid != m_nid) 900 continue; 901 902 /* skip hotpluggable memory regions if needed */ 903 if (movable_node_is_enabled() && memblock_is_hotpluggable(m)) 904 continue; 905 906 /* if we want mirror memory skip non-mirror memory regions */ 907 if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m)) 908 continue; 909 910 /* skip nomap memory unless we were asked for it explicitly */ 911 if (!(flags & MEMBLOCK_NOMAP) && memblock_is_nomap(m)) 912 continue; 913 914 if (!type_b) { 915 if (out_start) 916 *out_start = m_start; 917 if (out_end) 918 *out_end = m_end; 919 if (out_nid) 920 *out_nid = m_nid; 921 idx_a++; 922 *idx = (u32)idx_a | (u64)idx_b << 32; 923 return; 924 } 925 926 /* scan areas before each reservation */ 927 for (; idx_b < type_b->cnt + 1; idx_b++) { 928 struct memblock_region *r; 929 phys_addr_t r_start; 930 phys_addr_t r_end; 931 932 r = &type_b->regions[idx_b]; 933 r_start = idx_b ? r[-1].base + r[-1].size : 0; 934 r_end = idx_b < type_b->cnt ? 935 r->base : PHYS_ADDR_MAX; 936 937 /* 938 * if idx_b advanced past idx_a, 939 * break out to advance idx_a 940 */ 941 if (r_start >= m_end) 942 break; 943 /* if the two regions intersect, we're done */ 944 if (m_start < r_end) { 945 if (out_start) 946 *out_start = 947 max(m_start, r_start); 948 if (out_end) 949 *out_end = min(m_end, r_end); 950 if (out_nid) 951 *out_nid = m_nid; 952 /* 953 * The region which ends first is 954 * advanced for the next iteration. 955 */ 956 if (m_end <= r_end) 957 idx_a++; 958 else 959 idx_b++; 960 *idx = (u32)idx_a | (u64)idx_b << 32; 961 return; 962 } 963 } 964 } 965 966 /* signal end of iteration */ 967 *idx = ULLONG_MAX; 968 } 969 970 /** 971 * __next_mem_range_rev - generic next function for for_each_*_range_rev() 972 * 973 * Finds the next range from type_a which is not marked as unsuitable 974 * in type_b. 975 * 976 * @idx: pointer to u64 loop variable 977 * @nid: node selector, %NUMA_NO_NODE for all nodes 978 * @flags: pick from blocks based on memory attributes 979 * @type_a: pointer to memblock_type from where the range is taken 980 * @type_b: pointer to memblock_type which excludes memory from being taken 981 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL 982 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL 983 * @out_nid: ptr to int for nid of the range, can be %NULL 984 * 985 * Reverse of __next_mem_range(). 986 */ 987 void __init_memblock __next_mem_range_rev(u64 *idx, int nid, ulong flags, 988 struct memblock_type *type_a, 989 struct memblock_type *type_b, 990 phys_addr_t *out_start, 991 phys_addr_t *out_end, int *out_nid) 992 { 993 int idx_a = *idx & 0xffffffff; 994 int idx_b = *idx >> 32; 995 996 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n")) 997 nid = NUMA_NO_NODE; 998 999 if (*idx == (u64)ULLONG_MAX) { 1000 idx_a = type_a->cnt - 1; 1001 if (type_b != NULL) 1002 idx_b = type_b->cnt; 1003 else 1004 idx_b = 0; 1005 } 1006 1007 for (; idx_a >= 0; idx_a--) { 1008 struct memblock_region *m = &type_a->regions[idx_a]; 1009 1010 phys_addr_t m_start = m->base; 1011 phys_addr_t m_end = m->base + m->size; 1012 int m_nid = memblock_get_region_node(m); 1013 1014 /* only memory regions are associated with nodes, check it */ 1015 if (nid != NUMA_NO_NODE && nid != m_nid) 1016 continue; 1017 1018 /* skip hotpluggable memory regions if needed */ 1019 if (movable_node_is_enabled() && memblock_is_hotpluggable(m)) 1020 continue; 1021 1022 /* if we want mirror memory skip non-mirror memory regions */ 1023 if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m)) 1024 continue; 1025 1026 /* skip nomap memory unless we were asked for it explicitly */ 1027 if (!(flags & MEMBLOCK_NOMAP) && memblock_is_nomap(m)) 1028 continue; 1029 1030 if (!type_b) { 1031 if (out_start) 1032 *out_start = m_start; 1033 if (out_end) 1034 *out_end = m_end; 1035 if (out_nid) 1036 *out_nid = m_nid; 1037 idx_a--; 1038 *idx = (u32)idx_a | (u64)idx_b << 32; 1039 return; 1040 } 1041 1042 /* scan areas before each reservation */ 1043 for (; idx_b >= 0; idx_b--) { 1044 struct memblock_region *r; 1045 phys_addr_t r_start; 1046 phys_addr_t r_end; 1047 1048 r = &type_b->regions[idx_b]; 1049 r_start = idx_b ? r[-1].base + r[-1].size : 0; 1050 r_end = idx_b < type_b->cnt ? 1051 r->base : PHYS_ADDR_MAX; 1052 /* 1053 * if idx_b advanced past idx_a, 1054 * break out to advance idx_a 1055 */ 1056 1057 if (r_end <= m_start) 1058 break; 1059 /* if the two regions intersect, we're done */ 1060 if (m_end > r_start) { 1061 if (out_start) 1062 *out_start = max(m_start, r_start); 1063 if (out_end) 1064 *out_end = min(m_end, r_end); 1065 if (out_nid) 1066 *out_nid = m_nid; 1067 if (m_start >= r_start) 1068 idx_a--; 1069 else 1070 idx_b--; 1071 *idx = (u32)idx_a | (u64)idx_b << 32; 1072 return; 1073 } 1074 } 1075 } 1076 /* signal end of iteration */ 1077 *idx = ULLONG_MAX; 1078 } 1079 1080 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP 1081 /* 1082 * Common iterator interface used to define for_each_mem_range(). 1083 */ 1084 void __init_memblock __next_mem_pfn_range(int *idx, int nid, 1085 unsigned long *out_start_pfn, 1086 unsigned long *out_end_pfn, int *out_nid) 1087 { 1088 struct memblock_type *type = &memblock.memory; 1089 struct memblock_region *r; 1090 1091 while (++*idx < type->cnt) { 1092 r = &type->regions[*idx]; 1093 1094 if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size)) 1095 continue; 1096 if (nid == MAX_NUMNODES || nid == r->nid) 1097 break; 1098 } 1099 if (*idx >= type->cnt) { 1100 *idx = -1; 1101 return; 1102 } 1103 1104 if (out_start_pfn) 1105 *out_start_pfn = PFN_UP(r->base); 1106 if (out_end_pfn) 1107 *out_end_pfn = PFN_DOWN(r->base + r->size); 1108 if (out_nid) 1109 *out_nid = r->nid; 1110 } 1111 1112 /** 1113 * memblock_set_node - set node ID on memblock regions 1114 * @base: base of area to set node ID for 1115 * @size: size of area to set node ID for 1116 * @type: memblock type to set node ID for 1117 * @nid: node ID to set 1118 * 1119 * Set the nid of memblock @type regions in [@base,@base+@size) to @nid. 1120 * Regions which cross the area boundaries are split as necessary. 1121 * 1122 * RETURNS: 1123 * 0 on success, -errno on failure. 1124 */ 1125 int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size, 1126 struct memblock_type *type, int nid) 1127 { 1128 int start_rgn, end_rgn; 1129 int i, ret; 1130 1131 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn); 1132 if (ret) 1133 return ret; 1134 1135 for (i = start_rgn; i < end_rgn; i++) 1136 memblock_set_region_node(&type->regions[i], nid); 1137 1138 memblock_merge_regions(type); 1139 return 0; 1140 } 1141 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ 1142 1143 static phys_addr_t __init memblock_alloc_range_nid(phys_addr_t size, 1144 phys_addr_t align, phys_addr_t start, 1145 phys_addr_t end, int nid, ulong flags) 1146 { 1147 phys_addr_t found; 1148 1149 if (!align) 1150 align = SMP_CACHE_BYTES; 1151 1152 found = memblock_find_in_range_node(size, align, start, end, nid, 1153 flags); 1154 if (found && !memblock_reserve(found, size)) { 1155 /* 1156 * The min_count is set to 0 so that memblock allocations are 1157 * never reported as leaks. 1158 */ 1159 kmemleak_alloc_phys(found, size, 0, 0); 1160 return found; 1161 } 1162 return 0; 1163 } 1164 1165 phys_addr_t __init memblock_alloc_range(phys_addr_t size, phys_addr_t align, 1166 phys_addr_t start, phys_addr_t end, 1167 ulong flags) 1168 { 1169 return memblock_alloc_range_nid(size, align, start, end, NUMA_NO_NODE, 1170 flags); 1171 } 1172 1173 phys_addr_t __init memblock_alloc_base_nid(phys_addr_t size, 1174 phys_addr_t align, phys_addr_t max_addr, 1175 int nid, ulong flags) 1176 { 1177 return memblock_alloc_range_nid(size, align, 0, max_addr, nid, flags); 1178 } 1179 1180 phys_addr_t __init memblock_alloc_nid(phys_addr_t size, phys_addr_t align, int nid) 1181 { 1182 ulong flags = choose_memblock_flags(); 1183 phys_addr_t ret; 1184 1185 again: 1186 ret = memblock_alloc_base_nid(size, align, MEMBLOCK_ALLOC_ACCESSIBLE, 1187 nid, flags); 1188 1189 if (!ret && (flags & MEMBLOCK_MIRROR)) { 1190 flags &= ~MEMBLOCK_MIRROR; 1191 goto again; 1192 } 1193 return ret; 1194 } 1195 1196 phys_addr_t __init __memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr) 1197 { 1198 return memblock_alloc_base_nid(size, align, max_addr, NUMA_NO_NODE, 1199 MEMBLOCK_NONE); 1200 } 1201 1202 phys_addr_t __init memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr) 1203 { 1204 phys_addr_t alloc; 1205 1206 alloc = __memblock_alloc_base(size, align, max_addr); 1207 1208 if (alloc == 0) 1209 panic("ERROR: Failed to allocate %pa bytes below %pa.\n", 1210 &size, &max_addr); 1211 1212 return alloc; 1213 } 1214 1215 phys_addr_t __init memblock_alloc(phys_addr_t size, phys_addr_t align) 1216 { 1217 return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE); 1218 } 1219 1220 phys_addr_t __init memblock_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid) 1221 { 1222 phys_addr_t res = memblock_alloc_nid(size, align, nid); 1223 1224 if (res) 1225 return res; 1226 return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE); 1227 } 1228 1229 #if defined(CONFIG_NO_BOOTMEM) 1230 /** 1231 * memblock_virt_alloc_internal - allocate boot memory block 1232 * @size: size of memory block to be allocated in bytes 1233 * @align: alignment of the region and block's size 1234 * @min_addr: the lower bound of the memory region to allocate (phys address) 1235 * @max_addr: the upper bound of the memory region to allocate (phys address) 1236 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 1237 * 1238 * The @min_addr limit is dropped if it can not be satisfied and the allocation 1239 * will fall back to memory below @min_addr. Also, allocation may fall back 1240 * to any node in the system if the specified node can not 1241 * hold the requested memory. 1242 * 1243 * The allocation is performed from memory region limited by 1244 * memblock.current_limit if @max_addr == %BOOTMEM_ALLOC_ACCESSIBLE. 1245 * 1246 * The memory block is aligned on SMP_CACHE_BYTES if @align == 0. 1247 * 1248 * The phys address of allocated boot memory block is converted to virtual and 1249 * allocated memory is reset to 0. 1250 * 1251 * In addition, function sets the min_count to 0 using kmemleak_alloc for 1252 * allocated boot memory block, so that it is never reported as leaks. 1253 * 1254 * RETURNS: 1255 * Virtual address of allocated memory block on success, NULL on failure. 1256 */ 1257 static void * __init memblock_virt_alloc_internal( 1258 phys_addr_t size, phys_addr_t align, 1259 phys_addr_t min_addr, phys_addr_t max_addr, 1260 int nid) 1261 { 1262 phys_addr_t alloc; 1263 void *ptr; 1264 ulong flags = choose_memblock_flags(); 1265 1266 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n")) 1267 nid = NUMA_NO_NODE; 1268 1269 /* 1270 * Detect any accidental use of these APIs after slab is ready, as at 1271 * this moment memblock may be deinitialized already and its 1272 * internal data may be destroyed (after execution of free_all_bootmem) 1273 */ 1274 if (WARN_ON_ONCE(slab_is_available())) 1275 return kzalloc_node(size, GFP_NOWAIT, nid); 1276 1277 if (!align) 1278 align = SMP_CACHE_BYTES; 1279 1280 if (max_addr > memblock.current_limit) 1281 max_addr = memblock.current_limit; 1282 again: 1283 alloc = memblock_find_in_range_node(size, align, min_addr, max_addr, 1284 nid, flags); 1285 if (alloc && !memblock_reserve(alloc, size)) 1286 goto done; 1287 1288 if (nid != NUMA_NO_NODE) { 1289 alloc = memblock_find_in_range_node(size, align, min_addr, 1290 max_addr, NUMA_NO_NODE, 1291 flags); 1292 if (alloc && !memblock_reserve(alloc, size)) 1293 goto done; 1294 } 1295 1296 if (min_addr) { 1297 min_addr = 0; 1298 goto again; 1299 } 1300 1301 if (flags & MEMBLOCK_MIRROR) { 1302 flags &= ~MEMBLOCK_MIRROR; 1303 pr_warn("Could not allocate %pap bytes of mirrored memory\n", 1304 &size); 1305 goto again; 1306 } 1307 1308 return NULL; 1309 done: 1310 ptr = phys_to_virt(alloc); 1311 1312 /* 1313 * The min_count is set to 0 so that bootmem allocated blocks 1314 * are never reported as leaks. This is because many of these blocks 1315 * are only referred via the physical address which is not 1316 * looked up by kmemleak. 1317 */ 1318 kmemleak_alloc(ptr, size, 0, 0); 1319 1320 return ptr; 1321 } 1322 1323 /** 1324 * memblock_virt_alloc_try_nid_raw - allocate boot memory block without zeroing 1325 * memory and without panicking 1326 * @size: size of memory block to be allocated in bytes 1327 * @align: alignment of the region and block's size 1328 * @min_addr: the lower bound of the memory region from where the allocation 1329 * is preferred (phys address) 1330 * @max_addr: the upper bound of the memory region from where the allocation 1331 * is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to 1332 * allocate only from memory limited by memblock.current_limit value 1333 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 1334 * 1335 * Public function, provides additional debug information (including caller 1336 * info), if enabled. Does not zero allocated memory, does not panic if request 1337 * cannot be satisfied. 1338 * 1339 * RETURNS: 1340 * Virtual address of allocated memory block on success, NULL on failure. 1341 */ 1342 void * __init memblock_virt_alloc_try_nid_raw( 1343 phys_addr_t size, phys_addr_t align, 1344 phys_addr_t min_addr, phys_addr_t max_addr, 1345 int nid) 1346 { 1347 void *ptr; 1348 1349 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n", 1350 __func__, (u64)size, (u64)align, nid, (u64)min_addr, 1351 (u64)max_addr, (void *)_RET_IP_); 1352 1353 ptr = memblock_virt_alloc_internal(size, align, 1354 min_addr, max_addr, nid); 1355 #ifdef CONFIG_DEBUG_VM 1356 if (ptr && size > 0) 1357 memset(ptr, PAGE_POISON_PATTERN, size); 1358 #endif 1359 return ptr; 1360 } 1361 1362 /** 1363 * memblock_virt_alloc_try_nid_nopanic - allocate boot memory block 1364 * @size: size of memory block to be allocated in bytes 1365 * @align: alignment of the region and block's size 1366 * @min_addr: the lower bound of the memory region from where the allocation 1367 * is preferred (phys address) 1368 * @max_addr: the upper bound of the memory region from where the allocation 1369 * is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to 1370 * allocate only from memory limited by memblock.current_limit value 1371 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 1372 * 1373 * Public function, provides additional debug information (including caller 1374 * info), if enabled. This function zeroes the allocated memory. 1375 * 1376 * RETURNS: 1377 * Virtual address of allocated memory block on success, NULL on failure. 1378 */ 1379 void * __init memblock_virt_alloc_try_nid_nopanic( 1380 phys_addr_t size, phys_addr_t align, 1381 phys_addr_t min_addr, phys_addr_t max_addr, 1382 int nid) 1383 { 1384 void *ptr; 1385 1386 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n", 1387 __func__, (u64)size, (u64)align, nid, (u64)min_addr, 1388 (u64)max_addr, (void *)_RET_IP_); 1389 1390 ptr = memblock_virt_alloc_internal(size, align, 1391 min_addr, max_addr, nid); 1392 if (ptr) 1393 memset(ptr, 0, size); 1394 return ptr; 1395 } 1396 1397 /** 1398 * memblock_virt_alloc_try_nid - allocate boot memory block with panicking 1399 * @size: size of memory block to be allocated in bytes 1400 * @align: alignment of the region and block's size 1401 * @min_addr: the lower bound of the memory region from where the allocation 1402 * is preferred (phys address) 1403 * @max_addr: the upper bound of the memory region from where the allocation 1404 * is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to 1405 * allocate only from memory limited by memblock.current_limit value 1406 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 1407 * 1408 * Public panicking version of memblock_virt_alloc_try_nid_nopanic() 1409 * which provides debug information (including caller info), if enabled, 1410 * and panics if the request can not be satisfied. 1411 * 1412 * RETURNS: 1413 * Virtual address of allocated memory block on success, NULL on failure. 1414 */ 1415 void * __init memblock_virt_alloc_try_nid( 1416 phys_addr_t size, phys_addr_t align, 1417 phys_addr_t min_addr, phys_addr_t max_addr, 1418 int nid) 1419 { 1420 void *ptr; 1421 1422 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n", 1423 __func__, (u64)size, (u64)align, nid, (u64)min_addr, 1424 (u64)max_addr, (void *)_RET_IP_); 1425 ptr = memblock_virt_alloc_internal(size, align, 1426 min_addr, max_addr, nid); 1427 if (ptr) { 1428 memset(ptr, 0, size); 1429 return ptr; 1430 } 1431 1432 panic("%s: Failed to allocate %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx\n", 1433 __func__, (u64)size, (u64)align, nid, (u64)min_addr, 1434 (u64)max_addr); 1435 return NULL; 1436 } 1437 #endif 1438 1439 /** 1440 * __memblock_free_early - free boot memory block 1441 * @base: phys starting address of the boot memory block 1442 * @size: size of the boot memory block in bytes 1443 * 1444 * Free boot memory block previously allocated by memblock_virt_alloc_xx() API. 1445 * The freeing memory will not be released to the buddy allocator. 1446 */ 1447 void __init __memblock_free_early(phys_addr_t base, phys_addr_t size) 1448 { 1449 memblock_dbg("%s: [%#016llx-%#016llx] %pF\n", 1450 __func__, (u64)base, (u64)base + size - 1, 1451 (void *)_RET_IP_); 1452 kmemleak_free_part_phys(base, size); 1453 memblock_remove_range(&memblock.reserved, base, size); 1454 } 1455 1456 /* 1457 * __memblock_free_late - free bootmem block pages directly to buddy allocator 1458 * @addr: phys starting address of the boot memory block 1459 * @size: size of the boot memory block in bytes 1460 * 1461 * This is only useful when the bootmem allocator has already been torn 1462 * down, but we are still initializing the system. Pages are released directly 1463 * to the buddy allocator, no bootmem metadata is updated because it is gone. 1464 */ 1465 void __init __memblock_free_late(phys_addr_t base, phys_addr_t size) 1466 { 1467 u64 cursor, end; 1468 1469 memblock_dbg("%s: [%#016llx-%#016llx] %pF\n", 1470 __func__, (u64)base, (u64)base + size - 1, 1471 (void *)_RET_IP_); 1472 kmemleak_free_part_phys(base, size); 1473 cursor = PFN_UP(base); 1474 end = PFN_DOWN(base + size); 1475 1476 for (; cursor < end; cursor++) { 1477 __free_pages_bootmem(pfn_to_page(cursor), cursor, 0); 1478 totalram_pages++; 1479 } 1480 } 1481 1482 /* 1483 * Remaining API functions 1484 */ 1485 1486 phys_addr_t __init_memblock memblock_phys_mem_size(void) 1487 { 1488 return memblock.memory.total_size; 1489 } 1490 1491 phys_addr_t __init_memblock memblock_reserved_size(void) 1492 { 1493 return memblock.reserved.total_size; 1494 } 1495 1496 phys_addr_t __init memblock_mem_size(unsigned long limit_pfn) 1497 { 1498 unsigned long pages = 0; 1499 struct memblock_region *r; 1500 unsigned long start_pfn, end_pfn; 1501 1502 for_each_memblock(memory, r) { 1503 start_pfn = memblock_region_memory_base_pfn(r); 1504 end_pfn = memblock_region_memory_end_pfn(r); 1505 start_pfn = min_t(unsigned long, start_pfn, limit_pfn); 1506 end_pfn = min_t(unsigned long, end_pfn, limit_pfn); 1507 pages += end_pfn - start_pfn; 1508 } 1509 1510 return PFN_PHYS(pages); 1511 } 1512 1513 /* lowest address */ 1514 phys_addr_t __init_memblock memblock_start_of_DRAM(void) 1515 { 1516 return memblock.memory.regions[0].base; 1517 } 1518 1519 phys_addr_t __init_memblock memblock_end_of_DRAM(void) 1520 { 1521 int idx = memblock.memory.cnt - 1; 1522 1523 return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size); 1524 } 1525 1526 static phys_addr_t __init_memblock __find_max_addr(phys_addr_t limit) 1527 { 1528 phys_addr_t max_addr = PHYS_ADDR_MAX; 1529 struct memblock_region *r; 1530 1531 /* 1532 * translate the memory @limit size into the max address within one of 1533 * the memory memblock regions, if the @limit exceeds the total size 1534 * of those regions, max_addr will keep original value PHYS_ADDR_MAX 1535 */ 1536 for_each_memblock(memory, r) { 1537 if (limit <= r->size) { 1538 max_addr = r->base + limit; 1539 break; 1540 } 1541 limit -= r->size; 1542 } 1543 1544 return max_addr; 1545 } 1546 1547 void __init memblock_enforce_memory_limit(phys_addr_t limit) 1548 { 1549 phys_addr_t max_addr = PHYS_ADDR_MAX; 1550 1551 if (!limit) 1552 return; 1553 1554 max_addr = __find_max_addr(limit); 1555 1556 /* @limit exceeds the total size of the memory, do nothing */ 1557 if (max_addr == PHYS_ADDR_MAX) 1558 return; 1559 1560 /* truncate both memory and reserved regions */ 1561 memblock_remove_range(&memblock.memory, max_addr, 1562 PHYS_ADDR_MAX); 1563 memblock_remove_range(&memblock.reserved, max_addr, 1564 PHYS_ADDR_MAX); 1565 } 1566 1567 void __init memblock_cap_memory_range(phys_addr_t base, phys_addr_t size) 1568 { 1569 int start_rgn, end_rgn; 1570 int i, ret; 1571 1572 if (!size) 1573 return; 1574 1575 ret = memblock_isolate_range(&memblock.memory, base, size, 1576 &start_rgn, &end_rgn); 1577 if (ret) 1578 return; 1579 1580 /* remove all the MAP regions */ 1581 for (i = memblock.memory.cnt - 1; i >= end_rgn; i--) 1582 if (!memblock_is_nomap(&memblock.memory.regions[i])) 1583 memblock_remove_region(&memblock.memory, i); 1584 1585 for (i = start_rgn - 1; i >= 0; i--) 1586 if (!memblock_is_nomap(&memblock.memory.regions[i])) 1587 memblock_remove_region(&memblock.memory, i); 1588 1589 /* truncate the reserved regions */ 1590 memblock_remove_range(&memblock.reserved, 0, base); 1591 memblock_remove_range(&memblock.reserved, 1592 base + size, PHYS_ADDR_MAX); 1593 } 1594 1595 void __init memblock_mem_limit_remove_map(phys_addr_t limit) 1596 { 1597 phys_addr_t max_addr; 1598 1599 if (!limit) 1600 return; 1601 1602 max_addr = __find_max_addr(limit); 1603 1604 /* @limit exceeds the total size of the memory, do nothing */ 1605 if (max_addr == PHYS_ADDR_MAX) 1606 return; 1607 1608 memblock_cap_memory_range(0, max_addr); 1609 } 1610 1611 static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr) 1612 { 1613 unsigned int left = 0, right = type->cnt; 1614 1615 do { 1616 unsigned int mid = (right + left) / 2; 1617 1618 if (addr < type->regions[mid].base) 1619 right = mid; 1620 else if (addr >= (type->regions[mid].base + 1621 type->regions[mid].size)) 1622 left = mid + 1; 1623 else 1624 return mid; 1625 } while (left < right); 1626 return -1; 1627 } 1628 1629 bool __init memblock_is_reserved(phys_addr_t addr) 1630 { 1631 return memblock_search(&memblock.reserved, addr) != -1; 1632 } 1633 1634 bool __init_memblock memblock_is_memory(phys_addr_t addr) 1635 { 1636 return memblock_search(&memblock.memory, addr) != -1; 1637 } 1638 1639 bool __init_memblock memblock_is_map_memory(phys_addr_t addr) 1640 { 1641 int i = memblock_search(&memblock.memory, addr); 1642 1643 if (i == -1) 1644 return false; 1645 return !memblock_is_nomap(&memblock.memory.regions[i]); 1646 } 1647 1648 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP 1649 int __init_memblock memblock_search_pfn_nid(unsigned long pfn, 1650 unsigned long *start_pfn, unsigned long *end_pfn) 1651 { 1652 struct memblock_type *type = &memblock.memory; 1653 int mid = memblock_search(type, PFN_PHYS(pfn)); 1654 1655 if (mid == -1) 1656 return -1; 1657 1658 *start_pfn = PFN_DOWN(type->regions[mid].base); 1659 *end_pfn = PFN_DOWN(type->regions[mid].base + type->regions[mid].size); 1660 1661 return type->regions[mid].nid; 1662 } 1663 #endif 1664 1665 /** 1666 * memblock_is_region_memory - check if a region is a subset of memory 1667 * @base: base of region to check 1668 * @size: size of region to check 1669 * 1670 * Check if the region [@base, @base+@size) is a subset of a memory block. 1671 * 1672 * RETURNS: 1673 * 0 if false, non-zero if true 1674 */ 1675 bool __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size) 1676 { 1677 int idx = memblock_search(&memblock.memory, base); 1678 phys_addr_t end = base + memblock_cap_size(base, &size); 1679 1680 if (idx == -1) 1681 return false; 1682 return (memblock.memory.regions[idx].base + 1683 memblock.memory.regions[idx].size) >= end; 1684 } 1685 1686 /** 1687 * memblock_is_region_reserved - check if a region intersects reserved memory 1688 * @base: base of region to check 1689 * @size: size of region to check 1690 * 1691 * Check if the region [@base, @base+@size) intersects a reserved memory block. 1692 * 1693 * RETURNS: 1694 * True if they intersect, false if not. 1695 */ 1696 bool __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size) 1697 { 1698 memblock_cap_size(base, &size); 1699 return memblock_overlaps_region(&memblock.reserved, base, size); 1700 } 1701 1702 void __init_memblock memblock_trim_memory(phys_addr_t align) 1703 { 1704 phys_addr_t start, end, orig_start, orig_end; 1705 struct memblock_region *r; 1706 1707 for_each_memblock(memory, r) { 1708 orig_start = r->base; 1709 orig_end = r->base + r->size; 1710 start = round_up(orig_start, align); 1711 end = round_down(orig_end, align); 1712 1713 if (start == orig_start && end == orig_end) 1714 continue; 1715 1716 if (start < end) { 1717 r->base = start; 1718 r->size = end - start; 1719 } else { 1720 memblock_remove_region(&memblock.memory, 1721 r - memblock.memory.regions); 1722 r--; 1723 } 1724 } 1725 } 1726 1727 void __init_memblock memblock_set_current_limit(phys_addr_t limit) 1728 { 1729 memblock.current_limit = limit; 1730 } 1731 1732 phys_addr_t __init_memblock memblock_get_current_limit(void) 1733 { 1734 return memblock.current_limit; 1735 } 1736 1737 static void __init_memblock memblock_dump(struct memblock_type *type) 1738 { 1739 phys_addr_t base, end, size; 1740 unsigned long flags; 1741 int idx; 1742 struct memblock_region *rgn; 1743 1744 pr_info(" %s.cnt = 0x%lx\n", type->name, type->cnt); 1745 1746 for_each_memblock_type(idx, type, rgn) { 1747 char nid_buf[32] = ""; 1748 1749 base = rgn->base; 1750 size = rgn->size; 1751 end = base + size - 1; 1752 flags = rgn->flags; 1753 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP 1754 if (memblock_get_region_node(rgn) != MAX_NUMNODES) 1755 snprintf(nid_buf, sizeof(nid_buf), " on node %d", 1756 memblock_get_region_node(rgn)); 1757 #endif 1758 pr_info(" %s[%#x]\t[%pa-%pa], %pa bytes%s flags: %#lx\n", 1759 type->name, idx, &base, &end, &size, nid_buf, flags); 1760 } 1761 } 1762 1763 void __init_memblock __memblock_dump_all(void) 1764 { 1765 pr_info("MEMBLOCK configuration:\n"); 1766 pr_info(" memory size = %pa reserved size = %pa\n", 1767 &memblock.memory.total_size, 1768 &memblock.reserved.total_size); 1769 1770 memblock_dump(&memblock.memory); 1771 memblock_dump(&memblock.reserved); 1772 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP 1773 memblock_dump(&memblock.physmem); 1774 #endif 1775 } 1776 1777 void __init memblock_allow_resize(void) 1778 { 1779 memblock_can_resize = 1; 1780 } 1781 1782 static int __init early_memblock(char *p) 1783 { 1784 if (p && strstr(p, "debug")) 1785 memblock_debug = 1; 1786 return 0; 1787 } 1788 early_param("memblock", early_memblock); 1789 1790 #if defined(CONFIG_DEBUG_FS) && !defined(CONFIG_ARCH_DISCARD_MEMBLOCK) 1791 1792 static int memblock_debug_show(struct seq_file *m, void *private) 1793 { 1794 struct memblock_type *type = m->private; 1795 struct memblock_region *reg; 1796 int i; 1797 phys_addr_t end; 1798 1799 for (i = 0; i < type->cnt; i++) { 1800 reg = &type->regions[i]; 1801 end = reg->base + reg->size - 1; 1802 1803 seq_printf(m, "%4d: ", i); 1804 seq_printf(m, "%pa..%pa\n", ®->base, &end); 1805 } 1806 return 0; 1807 } 1808 DEFINE_SHOW_ATTRIBUTE(memblock_debug); 1809 1810 static int __init memblock_init_debugfs(void) 1811 { 1812 struct dentry *root = debugfs_create_dir("memblock", NULL); 1813 if (!root) 1814 return -ENXIO; 1815 debugfs_create_file("memory", 0444, root, 1816 &memblock.memory, &memblock_debug_fops); 1817 debugfs_create_file("reserved", 0444, root, 1818 &memblock.reserved, &memblock_debug_fops); 1819 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP 1820 debugfs_create_file("physmem", 0444, root, 1821 &memblock.physmem, &memblock_debug_fops); 1822 #endif 1823 1824 return 0; 1825 } 1826 __initcall(memblock_init_debugfs); 1827 1828 #endif /* CONFIG_DEBUG_FS */ 1829