1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * Procedures for maintaining information about logical memory blocks. 4 * 5 * Peter Bergner, IBM Corp. June 2001. 6 * Copyright (C) 2001 Peter Bergner. 7 */ 8 9 #include <linux/kernel.h> 10 #include <linux/slab.h> 11 #include <linux/init.h> 12 #include <linux/bitops.h> 13 #include <linux/poison.h> 14 #include <linux/pfn.h> 15 #include <linux/debugfs.h> 16 #include <linux/kmemleak.h> 17 #include <linux/seq_file.h> 18 #include <linux/memblock.h> 19 20 #include <asm/sections.h> 21 #include <linux/io.h> 22 23 #include "internal.h" 24 25 #define INIT_MEMBLOCK_REGIONS 128 26 #define INIT_PHYSMEM_REGIONS 4 27 28 #ifndef INIT_MEMBLOCK_RESERVED_REGIONS 29 # define INIT_MEMBLOCK_RESERVED_REGIONS INIT_MEMBLOCK_REGIONS 30 #endif 31 32 #ifndef INIT_MEMBLOCK_MEMORY_REGIONS 33 #define INIT_MEMBLOCK_MEMORY_REGIONS INIT_MEMBLOCK_REGIONS 34 #endif 35 36 /** 37 * DOC: memblock overview 38 * 39 * Memblock is a method of managing memory regions during the early 40 * boot period when the usual kernel memory allocators are not up and 41 * running. 42 * 43 * Memblock views the system memory as collections of contiguous 44 * regions. There are several types of these collections: 45 * 46 * * ``memory`` - describes the physical memory available to the 47 * kernel; this may differ from the actual physical memory installed 48 * in the system, for instance when the memory is restricted with 49 * ``mem=`` command line parameter 50 * * ``reserved`` - describes the regions that were allocated 51 * * ``physmem`` - describes the actual physical memory available during 52 * boot regardless of the possible restrictions and memory hot(un)plug; 53 * the ``physmem`` type is only available on some architectures. 54 * 55 * Each region is represented by struct memblock_region that 56 * defines the region extents, its attributes and NUMA node id on NUMA 57 * systems. Every memory type is described by the struct memblock_type 58 * which contains an array of memory regions along with 59 * the allocator metadata. The "memory" and "reserved" types are nicely 60 * wrapped with struct memblock. This structure is statically 61 * initialized at build time. The region arrays are initially sized to 62 * %INIT_MEMBLOCK_MEMORY_REGIONS for "memory" and 63 * %INIT_MEMBLOCK_RESERVED_REGIONS for "reserved". The region array 64 * for "physmem" is initially sized to %INIT_PHYSMEM_REGIONS. 65 * The memblock_allow_resize() enables automatic resizing of the region 66 * arrays during addition of new regions. This feature should be used 67 * with care so that memory allocated for the region array will not 68 * overlap with areas that should be reserved, for example initrd. 69 * 70 * The early architecture setup should tell memblock what the physical 71 * memory layout is by using memblock_add() or memblock_add_node() 72 * functions. The first function does not assign the region to a NUMA 73 * node and it is appropriate for UMA systems. Yet, it is possible to 74 * use it on NUMA systems as well and assign the region to a NUMA node 75 * later in the setup process using memblock_set_node(). The 76 * memblock_add_node() performs such an assignment directly. 77 * 78 * Once memblock is setup the memory can be allocated using one of the 79 * API variants: 80 * 81 * * memblock_phys_alloc*() - these functions return the **physical** 82 * address of the allocated memory 83 * * memblock_alloc*() - these functions return the **virtual** address 84 * of the allocated memory. 85 * 86 * Note, that both API variants use implicit assumptions about allowed 87 * memory ranges and the fallback methods. Consult the documentation 88 * of memblock_alloc_internal() and memblock_alloc_range_nid() 89 * functions for more elaborate description. 90 * 91 * As the system boot progresses, the architecture specific mem_init() 92 * function frees all the memory to the buddy page allocator. 93 * 94 * Unless an architecture enables %CONFIG_ARCH_KEEP_MEMBLOCK, the 95 * memblock data structures (except "physmem") will be discarded after the 96 * system initialization completes. 97 */ 98 99 #ifndef CONFIG_NUMA 100 struct pglist_data __refdata contig_page_data; 101 EXPORT_SYMBOL(contig_page_data); 102 #endif 103 104 unsigned long max_low_pfn; 105 unsigned long min_low_pfn; 106 unsigned long max_pfn; 107 unsigned long long max_possible_pfn; 108 109 static struct memblock_region memblock_memory_init_regions[INIT_MEMBLOCK_MEMORY_REGIONS] __initdata_memblock; 110 static struct memblock_region memblock_reserved_init_regions[INIT_MEMBLOCK_RESERVED_REGIONS] __initdata_memblock; 111 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP 112 static struct memblock_region memblock_physmem_init_regions[INIT_PHYSMEM_REGIONS]; 113 #endif 114 115 struct memblock memblock __initdata_memblock = { 116 .memory.regions = memblock_memory_init_regions, 117 .memory.cnt = 1, /* empty dummy entry */ 118 .memory.max = INIT_MEMBLOCK_MEMORY_REGIONS, 119 .memory.name = "memory", 120 121 .reserved.regions = memblock_reserved_init_regions, 122 .reserved.cnt = 1, /* empty dummy entry */ 123 .reserved.max = INIT_MEMBLOCK_RESERVED_REGIONS, 124 .reserved.name = "reserved", 125 126 .bottom_up = false, 127 .current_limit = MEMBLOCK_ALLOC_ANYWHERE, 128 }; 129 130 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP 131 struct memblock_type physmem = { 132 .regions = memblock_physmem_init_regions, 133 .cnt = 1, /* empty dummy entry */ 134 .max = INIT_PHYSMEM_REGIONS, 135 .name = "physmem", 136 }; 137 #endif 138 139 /* 140 * keep a pointer to &memblock.memory in the text section to use it in 141 * __next_mem_range() and its helpers. 142 * For architectures that do not keep memblock data after init, this 143 * pointer will be reset to NULL at memblock_discard() 144 */ 145 static __refdata struct memblock_type *memblock_memory = &memblock.memory; 146 147 #define for_each_memblock_type(i, memblock_type, rgn) \ 148 for (i = 0, rgn = &memblock_type->regions[0]; \ 149 i < memblock_type->cnt; \ 150 i++, rgn = &memblock_type->regions[i]) 151 152 #define memblock_dbg(fmt, ...) \ 153 do { \ 154 if (memblock_debug) \ 155 pr_info(fmt, ##__VA_ARGS__); \ 156 } while (0) 157 158 static int memblock_debug __initdata_memblock; 159 static bool system_has_some_mirror __initdata_memblock = false; 160 static int memblock_can_resize __initdata_memblock; 161 static int memblock_memory_in_slab __initdata_memblock = 0; 162 static int memblock_reserved_in_slab __initdata_memblock = 0; 163 164 static enum memblock_flags __init_memblock choose_memblock_flags(void) 165 { 166 return system_has_some_mirror ? MEMBLOCK_MIRROR : MEMBLOCK_NONE; 167 } 168 169 /* adjust *@size so that (@base + *@size) doesn't overflow, return new size */ 170 static inline phys_addr_t memblock_cap_size(phys_addr_t base, phys_addr_t *size) 171 { 172 return *size = min(*size, PHYS_ADDR_MAX - base); 173 } 174 175 /* 176 * Address comparison utilities 177 */ 178 static unsigned long __init_memblock memblock_addrs_overlap(phys_addr_t base1, phys_addr_t size1, 179 phys_addr_t base2, phys_addr_t size2) 180 { 181 return ((base1 < (base2 + size2)) && (base2 < (base1 + size1))); 182 } 183 184 bool __init_memblock memblock_overlaps_region(struct memblock_type *type, 185 phys_addr_t base, phys_addr_t size) 186 { 187 unsigned long i; 188 189 memblock_cap_size(base, &size); 190 191 for (i = 0; i < type->cnt; i++) 192 if (memblock_addrs_overlap(base, size, type->regions[i].base, 193 type->regions[i].size)) 194 break; 195 return i < type->cnt; 196 } 197 198 /** 199 * __memblock_find_range_bottom_up - find free area utility in bottom-up 200 * @start: start of candidate range 201 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or 202 * %MEMBLOCK_ALLOC_ACCESSIBLE 203 * @size: size of free area to find 204 * @align: alignment of free area to find 205 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 206 * @flags: pick from blocks based on memory attributes 207 * 208 * Utility called from memblock_find_in_range_node(), find free area bottom-up. 209 * 210 * Return: 211 * Found address on success, 0 on failure. 212 */ 213 static phys_addr_t __init_memblock 214 __memblock_find_range_bottom_up(phys_addr_t start, phys_addr_t end, 215 phys_addr_t size, phys_addr_t align, int nid, 216 enum memblock_flags flags) 217 { 218 phys_addr_t this_start, this_end, cand; 219 u64 i; 220 221 for_each_free_mem_range(i, nid, flags, &this_start, &this_end, NULL) { 222 this_start = clamp(this_start, start, end); 223 this_end = clamp(this_end, start, end); 224 225 cand = round_up(this_start, align); 226 if (cand < this_end && this_end - cand >= size) 227 return cand; 228 } 229 230 return 0; 231 } 232 233 /** 234 * __memblock_find_range_top_down - find free area utility, in top-down 235 * @start: start of candidate range 236 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or 237 * %MEMBLOCK_ALLOC_ACCESSIBLE 238 * @size: size of free area to find 239 * @align: alignment of free area to find 240 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 241 * @flags: pick from blocks based on memory attributes 242 * 243 * Utility called from memblock_find_in_range_node(), find free area top-down. 244 * 245 * Return: 246 * Found address on success, 0 on failure. 247 */ 248 static phys_addr_t __init_memblock 249 __memblock_find_range_top_down(phys_addr_t start, phys_addr_t end, 250 phys_addr_t size, phys_addr_t align, int nid, 251 enum memblock_flags flags) 252 { 253 phys_addr_t this_start, this_end, cand; 254 u64 i; 255 256 for_each_free_mem_range_reverse(i, nid, flags, &this_start, &this_end, 257 NULL) { 258 this_start = clamp(this_start, start, end); 259 this_end = clamp(this_end, start, end); 260 261 if (this_end < size) 262 continue; 263 264 cand = round_down(this_end - size, align); 265 if (cand >= this_start) 266 return cand; 267 } 268 269 return 0; 270 } 271 272 /** 273 * memblock_find_in_range_node - find free area in given range and node 274 * @size: size of free area to find 275 * @align: alignment of free area to find 276 * @start: start of candidate range 277 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or 278 * %MEMBLOCK_ALLOC_ACCESSIBLE 279 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 280 * @flags: pick from blocks based on memory attributes 281 * 282 * Find @size free area aligned to @align in the specified range and node. 283 * 284 * Return: 285 * Found address on success, 0 on failure. 286 */ 287 static phys_addr_t __init_memblock memblock_find_in_range_node(phys_addr_t size, 288 phys_addr_t align, phys_addr_t start, 289 phys_addr_t end, int nid, 290 enum memblock_flags flags) 291 { 292 /* pump up @end */ 293 if (end == MEMBLOCK_ALLOC_ACCESSIBLE || 294 end == MEMBLOCK_ALLOC_NOLEAKTRACE) 295 end = memblock.current_limit; 296 297 /* avoid allocating the first page */ 298 start = max_t(phys_addr_t, start, PAGE_SIZE); 299 end = max(start, end); 300 301 if (memblock_bottom_up()) 302 return __memblock_find_range_bottom_up(start, end, size, align, 303 nid, flags); 304 else 305 return __memblock_find_range_top_down(start, end, size, align, 306 nid, flags); 307 } 308 309 /** 310 * memblock_find_in_range - find free area in given range 311 * @start: start of candidate range 312 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or 313 * %MEMBLOCK_ALLOC_ACCESSIBLE 314 * @size: size of free area to find 315 * @align: alignment of free area to find 316 * 317 * Find @size free area aligned to @align in the specified range. 318 * 319 * Return: 320 * Found address on success, 0 on failure. 321 */ 322 static phys_addr_t __init_memblock memblock_find_in_range(phys_addr_t start, 323 phys_addr_t end, phys_addr_t size, 324 phys_addr_t align) 325 { 326 phys_addr_t ret; 327 enum memblock_flags flags = choose_memblock_flags(); 328 329 again: 330 ret = memblock_find_in_range_node(size, align, start, end, 331 NUMA_NO_NODE, flags); 332 333 if (!ret && (flags & MEMBLOCK_MIRROR)) { 334 pr_warn_ratelimited("Could not allocate %pap bytes of mirrored memory\n", 335 &size); 336 flags &= ~MEMBLOCK_MIRROR; 337 goto again; 338 } 339 340 return ret; 341 } 342 343 static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r) 344 { 345 type->total_size -= type->regions[r].size; 346 memmove(&type->regions[r], &type->regions[r + 1], 347 (type->cnt - (r + 1)) * sizeof(type->regions[r])); 348 type->cnt--; 349 350 /* Special case for empty arrays */ 351 if (type->cnt == 0) { 352 WARN_ON(type->total_size != 0); 353 type->cnt = 1; 354 type->regions[0].base = 0; 355 type->regions[0].size = 0; 356 type->regions[0].flags = 0; 357 memblock_set_region_node(&type->regions[0], MAX_NUMNODES); 358 } 359 } 360 361 #ifndef CONFIG_ARCH_KEEP_MEMBLOCK 362 /** 363 * memblock_discard - discard memory and reserved arrays if they were allocated 364 */ 365 void __init memblock_discard(void) 366 { 367 phys_addr_t addr, size; 368 369 if (memblock.reserved.regions != memblock_reserved_init_regions) { 370 addr = __pa(memblock.reserved.regions); 371 size = PAGE_ALIGN(sizeof(struct memblock_region) * 372 memblock.reserved.max); 373 if (memblock_reserved_in_slab) 374 kfree(memblock.reserved.regions); 375 else 376 memblock_free_late(addr, size); 377 } 378 379 if (memblock.memory.regions != memblock_memory_init_regions) { 380 addr = __pa(memblock.memory.regions); 381 size = PAGE_ALIGN(sizeof(struct memblock_region) * 382 memblock.memory.max); 383 if (memblock_memory_in_slab) 384 kfree(memblock.memory.regions); 385 else 386 memblock_free_late(addr, size); 387 } 388 389 memblock_memory = NULL; 390 } 391 #endif 392 393 /** 394 * memblock_double_array - double the size of the memblock regions array 395 * @type: memblock type of the regions array being doubled 396 * @new_area_start: starting address of memory range to avoid overlap with 397 * @new_area_size: size of memory range to avoid overlap with 398 * 399 * Double the size of the @type regions array. If memblock is being used to 400 * allocate memory for a new reserved regions array and there is a previously 401 * allocated memory range [@new_area_start, @new_area_start + @new_area_size] 402 * waiting to be reserved, ensure the memory used by the new array does 403 * not overlap. 404 * 405 * Return: 406 * 0 on success, -1 on failure. 407 */ 408 static int __init_memblock memblock_double_array(struct memblock_type *type, 409 phys_addr_t new_area_start, 410 phys_addr_t new_area_size) 411 { 412 struct memblock_region *new_array, *old_array; 413 phys_addr_t old_alloc_size, new_alloc_size; 414 phys_addr_t old_size, new_size, addr, new_end; 415 int use_slab = slab_is_available(); 416 int *in_slab; 417 418 /* We don't allow resizing until we know about the reserved regions 419 * of memory that aren't suitable for allocation 420 */ 421 if (!memblock_can_resize) 422 return -1; 423 424 /* Calculate new doubled size */ 425 old_size = type->max * sizeof(struct memblock_region); 426 new_size = old_size << 1; 427 /* 428 * We need to allocated new one align to PAGE_SIZE, 429 * so we can free them completely later. 430 */ 431 old_alloc_size = PAGE_ALIGN(old_size); 432 new_alloc_size = PAGE_ALIGN(new_size); 433 434 /* Retrieve the slab flag */ 435 if (type == &memblock.memory) 436 in_slab = &memblock_memory_in_slab; 437 else 438 in_slab = &memblock_reserved_in_slab; 439 440 /* Try to find some space for it */ 441 if (use_slab) { 442 new_array = kmalloc(new_size, GFP_KERNEL); 443 addr = new_array ? __pa(new_array) : 0; 444 } else { 445 /* only exclude range when trying to double reserved.regions */ 446 if (type != &memblock.reserved) 447 new_area_start = new_area_size = 0; 448 449 addr = memblock_find_in_range(new_area_start + new_area_size, 450 memblock.current_limit, 451 new_alloc_size, PAGE_SIZE); 452 if (!addr && new_area_size) 453 addr = memblock_find_in_range(0, 454 min(new_area_start, memblock.current_limit), 455 new_alloc_size, PAGE_SIZE); 456 457 new_array = addr ? __va(addr) : NULL; 458 } 459 if (!addr) { 460 pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n", 461 type->name, type->max, type->max * 2); 462 return -1; 463 } 464 465 new_end = addr + new_size - 1; 466 memblock_dbg("memblock: %s is doubled to %ld at [%pa-%pa]", 467 type->name, type->max * 2, &addr, &new_end); 468 469 /* 470 * Found space, we now need to move the array over before we add the 471 * reserved region since it may be our reserved array itself that is 472 * full. 473 */ 474 memcpy(new_array, type->regions, old_size); 475 memset(new_array + type->max, 0, old_size); 476 old_array = type->regions; 477 type->regions = new_array; 478 type->max <<= 1; 479 480 /* Free old array. We needn't free it if the array is the static one */ 481 if (*in_slab) 482 kfree(old_array); 483 else if (old_array != memblock_memory_init_regions && 484 old_array != memblock_reserved_init_regions) 485 memblock_free(old_array, old_alloc_size); 486 487 /* 488 * Reserve the new array if that comes from the memblock. Otherwise, we 489 * needn't do it 490 */ 491 if (!use_slab) 492 BUG_ON(memblock_reserve(addr, new_alloc_size)); 493 494 /* Update slab flag */ 495 *in_slab = use_slab; 496 497 return 0; 498 } 499 500 /** 501 * memblock_merge_regions - merge neighboring compatible regions 502 * @type: memblock type to scan 503 * 504 * Scan @type and merge neighboring compatible regions. 505 */ 506 static void __init_memblock memblock_merge_regions(struct memblock_type *type) 507 { 508 int i = 0; 509 510 /* cnt never goes below 1 */ 511 while (i < type->cnt - 1) { 512 struct memblock_region *this = &type->regions[i]; 513 struct memblock_region *next = &type->regions[i + 1]; 514 515 if (this->base + this->size != next->base || 516 memblock_get_region_node(this) != 517 memblock_get_region_node(next) || 518 this->flags != next->flags) { 519 BUG_ON(this->base + this->size > next->base); 520 i++; 521 continue; 522 } 523 524 this->size += next->size; 525 /* move forward from next + 1, index of which is i + 2 */ 526 memmove(next, next + 1, (type->cnt - (i + 2)) * sizeof(*next)); 527 type->cnt--; 528 } 529 } 530 531 /** 532 * memblock_insert_region - insert new memblock region 533 * @type: memblock type to insert into 534 * @idx: index for the insertion point 535 * @base: base address of the new region 536 * @size: size of the new region 537 * @nid: node id of the new region 538 * @flags: flags of the new region 539 * 540 * Insert new memblock region [@base, @base + @size) into @type at @idx. 541 * @type must already have extra room to accommodate the new region. 542 */ 543 static void __init_memblock memblock_insert_region(struct memblock_type *type, 544 int idx, phys_addr_t base, 545 phys_addr_t size, 546 int nid, 547 enum memblock_flags flags) 548 { 549 struct memblock_region *rgn = &type->regions[idx]; 550 551 BUG_ON(type->cnt >= type->max); 552 memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn)); 553 rgn->base = base; 554 rgn->size = size; 555 rgn->flags = flags; 556 memblock_set_region_node(rgn, nid); 557 type->cnt++; 558 type->total_size += size; 559 } 560 561 /** 562 * memblock_add_range - add new memblock region 563 * @type: memblock type to add new region into 564 * @base: base address of the new region 565 * @size: size of the new region 566 * @nid: nid of the new region 567 * @flags: flags of the new region 568 * 569 * Add new memblock region [@base, @base + @size) into @type. The new region 570 * is allowed to overlap with existing ones - overlaps don't affect already 571 * existing regions. @type is guaranteed to be minimal (all neighbouring 572 * compatible regions are merged) after the addition. 573 * 574 * Return: 575 * 0 on success, -errno on failure. 576 */ 577 static int __init_memblock memblock_add_range(struct memblock_type *type, 578 phys_addr_t base, phys_addr_t size, 579 int nid, enum memblock_flags flags) 580 { 581 bool insert = false; 582 phys_addr_t obase = base; 583 phys_addr_t end = base + memblock_cap_size(base, &size); 584 int idx, nr_new; 585 struct memblock_region *rgn; 586 587 if (!size) 588 return 0; 589 590 /* special case for empty array */ 591 if (type->regions[0].size == 0) { 592 WARN_ON(type->cnt != 1 || type->total_size); 593 type->regions[0].base = base; 594 type->regions[0].size = size; 595 type->regions[0].flags = flags; 596 memblock_set_region_node(&type->regions[0], nid); 597 type->total_size = size; 598 return 0; 599 } 600 601 /* 602 * The worst case is when new range overlaps all existing regions, 603 * then we'll need type->cnt + 1 empty regions in @type. So if 604 * type->cnt * 2 + 1 is less than type->max, we know 605 * that there is enough empty regions in @type, and we can insert 606 * regions directly. 607 */ 608 if (type->cnt * 2 + 1 < type->max) 609 insert = true; 610 611 repeat: 612 /* 613 * The following is executed twice. Once with %false @insert and 614 * then with %true. The first counts the number of regions needed 615 * to accommodate the new area. The second actually inserts them. 616 */ 617 base = obase; 618 nr_new = 0; 619 620 for_each_memblock_type(idx, type, rgn) { 621 phys_addr_t rbase = rgn->base; 622 phys_addr_t rend = rbase + rgn->size; 623 624 if (rbase >= end) 625 break; 626 if (rend <= base) 627 continue; 628 /* 629 * @rgn overlaps. If it separates the lower part of new 630 * area, insert that portion. 631 */ 632 if (rbase > base) { 633 #ifdef CONFIG_NUMA 634 WARN_ON(nid != memblock_get_region_node(rgn)); 635 #endif 636 WARN_ON(flags != rgn->flags); 637 nr_new++; 638 if (insert) 639 memblock_insert_region(type, idx++, base, 640 rbase - base, nid, 641 flags); 642 } 643 /* area below @rend is dealt with, forget about it */ 644 base = min(rend, end); 645 } 646 647 /* insert the remaining portion */ 648 if (base < end) { 649 nr_new++; 650 if (insert) 651 memblock_insert_region(type, idx, base, end - base, 652 nid, flags); 653 } 654 655 if (!nr_new) 656 return 0; 657 658 /* 659 * If this was the first round, resize array and repeat for actual 660 * insertions; otherwise, merge and return. 661 */ 662 if (!insert) { 663 while (type->cnt + nr_new > type->max) 664 if (memblock_double_array(type, obase, size) < 0) 665 return -ENOMEM; 666 insert = true; 667 goto repeat; 668 } else { 669 memblock_merge_regions(type); 670 return 0; 671 } 672 } 673 674 /** 675 * memblock_add_node - add new memblock region within a NUMA node 676 * @base: base address of the new region 677 * @size: size of the new region 678 * @nid: nid of the new region 679 * @flags: flags of the new region 680 * 681 * Add new memblock region [@base, @base + @size) to the "memory" 682 * type. See memblock_add_range() description for mode details 683 * 684 * Return: 685 * 0 on success, -errno on failure. 686 */ 687 int __init_memblock memblock_add_node(phys_addr_t base, phys_addr_t size, 688 int nid, enum memblock_flags flags) 689 { 690 phys_addr_t end = base + size - 1; 691 692 memblock_dbg("%s: [%pa-%pa] nid=%d flags=%x %pS\n", __func__, 693 &base, &end, nid, flags, (void *)_RET_IP_); 694 695 return memblock_add_range(&memblock.memory, base, size, nid, flags); 696 } 697 698 /** 699 * memblock_add - add new memblock region 700 * @base: base address of the new region 701 * @size: size of the new region 702 * 703 * Add new memblock region [@base, @base + @size) to the "memory" 704 * type. See memblock_add_range() description for mode details 705 * 706 * Return: 707 * 0 on success, -errno on failure. 708 */ 709 int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size) 710 { 711 phys_addr_t end = base + size - 1; 712 713 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__, 714 &base, &end, (void *)_RET_IP_); 715 716 return memblock_add_range(&memblock.memory, base, size, MAX_NUMNODES, 0); 717 } 718 719 /** 720 * memblock_isolate_range - isolate given range into disjoint memblocks 721 * @type: memblock type to isolate range for 722 * @base: base of range to isolate 723 * @size: size of range to isolate 724 * @start_rgn: out parameter for the start of isolated region 725 * @end_rgn: out parameter for the end of isolated region 726 * 727 * Walk @type and ensure that regions don't cross the boundaries defined by 728 * [@base, @base + @size). Crossing regions are split at the boundaries, 729 * which may create at most two more regions. The index of the first 730 * region inside the range is returned in *@start_rgn and end in *@end_rgn. 731 * 732 * Return: 733 * 0 on success, -errno on failure. 734 */ 735 static int __init_memblock memblock_isolate_range(struct memblock_type *type, 736 phys_addr_t base, phys_addr_t size, 737 int *start_rgn, int *end_rgn) 738 { 739 phys_addr_t end = base + memblock_cap_size(base, &size); 740 int idx; 741 struct memblock_region *rgn; 742 743 *start_rgn = *end_rgn = 0; 744 745 if (!size) 746 return 0; 747 748 /* we'll create at most two more regions */ 749 while (type->cnt + 2 > type->max) 750 if (memblock_double_array(type, base, size) < 0) 751 return -ENOMEM; 752 753 for_each_memblock_type(idx, type, rgn) { 754 phys_addr_t rbase = rgn->base; 755 phys_addr_t rend = rbase + rgn->size; 756 757 if (rbase >= end) 758 break; 759 if (rend <= base) 760 continue; 761 762 if (rbase < base) { 763 /* 764 * @rgn intersects from below. Split and continue 765 * to process the next region - the new top half. 766 */ 767 rgn->base = base; 768 rgn->size -= base - rbase; 769 type->total_size -= base - rbase; 770 memblock_insert_region(type, idx, rbase, base - rbase, 771 memblock_get_region_node(rgn), 772 rgn->flags); 773 } else if (rend > end) { 774 /* 775 * @rgn intersects from above. Split and redo the 776 * current region - the new bottom half. 777 */ 778 rgn->base = end; 779 rgn->size -= end - rbase; 780 type->total_size -= end - rbase; 781 memblock_insert_region(type, idx--, rbase, end - rbase, 782 memblock_get_region_node(rgn), 783 rgn->flags); 784 } else { 785 /* @rgn is fully contained, record it */ 786 if (!*end_rgn) 787 *start_rgn = idx; 788 *end_rgn = idx + 1; 789 } 790 } 791 792 return 0; 793 } 794 795 static int __init_memblock memblock_remove_range(struct memblock_type *type, 796 phys_addr_t base, phys_addr_t size) 797 { 798 int start_rgn, end_rgn; 799 int i, ret; 800 801 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn); 802 if (ret) 803 return ret; 804 805 for (i = end_rgn - 1; i >= start_rgn; i--) 806 memblock_remove_region(type, i); 807 return 0; 808 } 809 810 int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size) 811 { 812 phys_addr_t end = base + size - 1; 813 814 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__, 815 &base, &end, (void *)_RET_IP_); 816 817 return memblock_remove_range(&memblock.memory, base, size); 818 } 819 820 /** 821 * memblock_free - free boot memory allocation 822 * @ptr: starting address of the boot memory allocation 823 * @size: size of the boot memory block in bytes 824 * 825 * Free boot memory block previously allocated by memblock_alloc_xx() API. 826 * The freeing memory will not be released to the buddy allocator. 827 */ 828 void __init_memblock memblock_free(void *ptr, size_t size) 829 { 830 if (ptr) 831 memblock_phys_free(__pa(ptr), size); 832 } 833 834 /** 835 * memblock_phys_free - free boot memory block 836 * @base: phys starting address of the boot memory block 837 * @size: size of the boot memory block in bytes 838 * 839 * Free boot memory block previously allocated by memblock_phys_alloc_xx() API. 840 * The freeing memory will not be released to the buddy allocator. 841 */ 842 int __init_memblock memblock_phys_free(phys_addr_t base, phys_addr_t size) 843 { 844 phys_addr_t end = base + size - 1; 845 846 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__, 847 &base, &end, (void *)_RET_IP_); 848 849 kmemleak_free_part_phys(base, size); 850 return memblock_remove_range(&memblock.reserved, base, size); 851 } 852 853 int __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size) 854 { 855 phys_addr_t end = base + size - 1; 856 857 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__, 858 &base, &end, (void *)_RET_IP_); 859 860 return memblock_add_range(&memblock.reserved, base, size, MAX_NUMNODES, 0); 861 } 862 863 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP 864 int __init_memblock memblock_physmem_add(phys_addr_t base, phys_addr_t size) 865 { 866 phys_addr_t end = base + size - 1; 867 868 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__, 869 &base, &end, (void *)_RET_IP_); 870 871 return memblock_add_range(&physmem, base, size, MAX_NUMNODES, 0); 872 } 873 #endif 874 875 /** 876 * memblock_setclr_flag - set or clear flag for a memory region 877 * @base: base address of the region 878 * @size: size of the region 879 * @set: set or clear the flag 880 * @flag: the flag to update 881 * 882 * This function isolates region [@base, @base + @size), and sets/clears flag 883 * 884 * Return: 0 on success, -errno on failure. 885 */ 886 static int __init_memblock memblock_setclr_flag(phys_addr_t base, 887 phys_addr_t size, int set, int flag) 888 { 889 struct memblock_type *type = &memblock.memory; 890 int i, ret, start_rgn, end_rgn; 891 892 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn); 893 if (ret) 894 return ret; 895 896 for (i = start_rgn; i < end_rgn; i++) { 897 struct memblock_region *r = &type->regions[i]; 898 899 if (set) 900 r->flags |= flag; 901 else 902 r->flags &= ~flag; 903 } 904 905 memblock_merge_regions(type); 906 return 0; 907 } 908 909 /** 910 * memblock_mark_hotplug - Mark hotpluggable memory with flag MEMBLOCK_HOTPLUG. 911 * @base: the base phys addr of the region 912 * @size: the size of the region 913 * 914 * Return: 0 on success, -errno on failure. 915 */ 916 int __init_memblock memblock_mark_hotplug(phys_addr_t base, phys_addr_t size) 917 { 918 return memblock_setclr_flag(base, size, 1, MEMBLOCK_HOTPLUG); 919 } 920 921 /** 922 * memblock_clear_hotplug - Clear flag MEMBLOCK_HOTPLUG for a specified region. 923 * @base: the base phys addr of the region 924 * @size: the size of the region 925 * 926 * Return: 0 on success, -errno on failure. 927 */ 928 int __init_memblock memblock_clear_hotplug(phys_addr_t base, phys_addr_t size) 929 { 930 return memblock_setclr_flag(base, size, 0, MEMBLOCK_HOTPLUG); 931 } 932 933 /** 934 * memblock_mark_mirror - Mark mirrored memory with flag MEMBLOCK_MIRROR. 935 * @base: the base phys addr of the region 936 * @size: the size of the region 937 * 938 * Return: 0 on success, -errno on failure. 939 */ 940 int __init_memblock memblock_mark_mirror(phys_addr_t base, phys_addr_t size) 941 { 942 if (!mirrored_kernelcore) 943 return 0; 944 945 system_has_some_mirror = true; 946 947 return memblock_setclr_flag(base, size, 1, MEMBLOCK_MIRROR); 948 } 949 950 /** 951 * memblock_mark_nomap - Mark a memory region with flag MEMBLOCK_NOMAP. 952 * @base: the base phys addr of the region 953 * @size: the size of the region 954 * 955 * The memory regions marked with %MEMBLOCK_NOMAP will not be added to the 956 * direct mapping of the physical memory. These regions will still be 957 * covered by the memory map. The struct page representing NOMAP memory 958 * frames in the memory map will be PageReserved() 959 * 960 * Note: if the memory being marked %MEMBLOCK_NOMAP was allocated from 961 * memblock, the caller must inform kmemleak to ignore that memory 962 * 963 * Return: 0 on success, -errno on failure. 964 */ 965 int __init_memblock memblock_mark_nomap(phys_addr_t base, phys_addr_t size) 966 { 967 return memblock_setclr_flag(base, size, 1, MEMBLOCK_NOMAP); 968 } 969 970 /** 971 * memblock_clear_nomap - Clear flag MEMBLOCK_NOMAP for a specified region. 972 * @base: the base phys addr of the region 973 * @size: the size of the region 974 * 975 * Return: 0 on success, -errno on failure. 976 */ 977 int __init_memblock memblock_clear_nomap(phys_addr_t base, phys_addr_t size) 978 { 979 return memblock_setclr_flag(base, size, 0, MEMBLOCK_NOMAP); 980 } 981 982 static bool should_skip_region(struct memblock_type *type, 983 struct memblock_region *m, 984 int nid, int flags) 985 { 986 int m_nid = memblock_get_region_node(m); 987 988 /* we never skip regions when iterating memblock.reserved or physmem */ 989 if (type != memblock_memory) 990 return false; 991 992 /* only memory regions are associated with nodes, check it */ 993 if (nid != NUMA_NO_NODE && nid != m_nid) 994 return true; 995 996 /* skip hotpluggable memory regions if needed */ 997 if (movable_node_is_enabled() && memblock_is_hotpluggable(m) && 998 !(flags & MEMBLOCK_HOTPLUG)) 999 return true; 1000 1001 /* if we want mirror memory skip non-mirror memory regions */ 1002 if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m)) 1003 return true; 1004 1005 /* skip nomap memory unless we were asked for it explicitly */ 1006 if (!(flags & MEMBLOCK_NOMAP) && memblock_is_nomap(m)) 1007 return true; 1008 1009 /* skip driver-managed memory unless we were asked for it explicitly */ 1010 if (!(flags & MEMBLOCK_DRIVER_MANAGED) && memblock_is_driver_managed(m)) 1011 return true; 1012 1013 return false; 1014 } 1015 1016 /** 1017 * __next_mem_range - next function for for_each_free_mem_range() etc. 1018 * @idx: pointer to u64 loop variable 1019 * @nid: node selector, %NUMA_NO_NODE for all nodes 1020 * @flags: pick from blocks based on memory attributes 1021 * @type_a: pointer to memblock_type from where the range is taken 1022 * @type_b: pointer to memblock_type which excludes memory from being taken 1023 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL 1024 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL 1025 * @out_nid: ptr to int for nid of the range, can be %NULL 1026 * 1027 * Find the first area from *@idx which matches @nid, fill the out 1028 * parameters, and update *@idx for the next iteration. The lower 32bit of 1029 * *@idx contains index into type_a and the upper 32bit indexes the 1030 * areas before each region in type_b. For example, if type_b regions 1031 * look like the following, 1032 * 1033 * 0:[0-16), 1:[32-48), 2:[128-130) 1034 * 1035 * The upper 32bit indexes the following regions. 1036 * 1037 * 0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX) 1038 * 1039 * As both region arrays are sorted, the function advances the two indices 1040 * in lockstep and returns each intersection. 1041 */ 1042 void __next_mem_range(u64 *idx, int nid, enum memblock_flags flags, 1043 struct memblock_type *type_a, 1044 struct memblock_type *type_b, phys_addr_t *out_start, 1045 phys_addr_t *out_end, int *out_nid) 1046 { 1047 int idx_a = *idx & 0xffffffff; 1048 int idx_b = *idx >> 32; 1049 1050 if (WARN_ONCE(nid == MAX_NUMNODES, 1051 "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n")) 1052 nid = NUMA_NO_NODE; 1053 1054 for (; idx_a < type_a->cnt; idx_a++) { 1055 struct memblock_region *m = &type_a->regions[idx_a]; 1056 1057 phys_addr_t m_start = m->base; 1058 phys_addr_t m_end = m->base + m->size; 1059 int m_nid = memblock_get_region_node(m); 1060 1061 if (should_skip_region(type_a, m, nid, flags)) 1062 continue; 1063 1064 if (!type_b) { 1065 if (out_start) 1066 *out_start = m_start; 1067 if (out_end) 1068 *out_end = m_end; 1069 if (out_nid) 1070 *out_nid = m_nid; 1071 idx_a++; 1072 *idx = (u32)idx_a | (u64)idx_b << 32; 1073 return; 1074 } 1075 1076 /* scan areas before each reservation */ 1077 for (; idx_b < type_b->cnt + 1; idx_b++) { 1078 struct memblock_region *r; 1079 phys_addr_t r_start; 1080 phys_addr_t r_end; 1081 1082 r = &type_b->regions[idx_b]; 1083 r_start = idx_b ? r[-1].base + r[-1].size : 0; 1084 r_end = idx_b < type_b->cnt ? 1085 r->base : PHYS_ADDR_MAX; 1086 1087 /* 1088 * if idx_b advanced past idx_a, 1089 * break out to advance idx_a 1090 */ 1091 if (r_start >= m_end) 1092 break; 1093 /* if the two regions intersect, we're done */ 1094 if (m_start < r_end) { 1095 if (out_start) 1096 *out_start = 1097 max(m_start, r_start); 1098 if (out_end) 1099 *out_end = min(m_end, r_end); 1100 if (out_nid) 1101 *out_nid = m_nid; 1102 /* 1103 * The region which ends first is 1104 * advanced for the next iteration. 1105 */ 1106 if (m_end <= r_end) 1107 idx_a++; 1108 else 1109 idx_b++; 1110 *idx = (u32)idx_a | (u64)idx_b << 32; 1111 return; 1112 } 1113 } 1114 } 1115 1116 /* signal end of iteration */ 1117 *idx = ULLONG_MAX; 1118 } 1119 1120 /** 1121 * __next_mem_range_rev - generic next function for for_each_*_range_rev() 1122 * 1123 * @idx: pointer to u64 loop variable 1124 * @nid: node selector, %NUMA_NO_NODE for all nodes 1125 * @flags: pick from blocks based on memory attributes 1126 * @type_a: pointer to memblock_type from where the range is taken 1127 * @type_b: pointer to memblock_type which excludes memory from being taken 1128 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL 1129 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL 1130 * @out_nid: ptr to int for nid of the range, can be %NULL 1131 * 1132 * Finds the next range from type_a which is not marked as unsuitable 1133 * in type_b. 1134 * 1135 * Reverse of __next_mem_range(). 1136 */ 1137 void __init_memblock __next_mem_range_rev(u64 *idx, int nid, 1138 enum memblock_flags flags, 1139 struct memblock_type *type_a, 1140 struct memblock_type *type_b, 1141 phys_addr_t *out_start, 1142 phys_addr_t *out_end, int *out_nid) 1143 { 1144 int idx_a = *idx & 0xffffffff; 1145 int idx_b = *idx >> 32; 1146 1147 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n")) 1148 nid = NUMA_NO_NODE; 1149 1150 if (*idx == (u64)ULLONG_MAX) { 1151 idx_a = type_a->cnt - 1; 1152 if (type_b != NULL) 1153 idx_b = type_b->cnt; 1154 else 1155 idx_b = 0; 1156 } 1157 1158 for (; idx_a >= 0; idx_a--) { 1159 struct memblock_region *m = &type_a->regions[idx_a]; 1160 1161 phys_addr_t m_start = m->base; 1162 phys_addr_t m_end = m->base + m->size; 1163 int m_nid = memblock_get_region_node(m); 1164 1165 if (should_skip_region(type_a, m, nid, flags)) 1166 continue; 1167 1168 if (!type_b) { 1169 if (out_start) 1170 *out_start = m_start; 1171 if (out_end) 1172 *out_end = m_end; 1173 if (out_nid) 1174 *out_nid = m_nid; 1175 idx_a--; 1176 *idx = (u32)idx_a | (u64)idx_b << 32; 1177 return; 1178 } 1179 1180 /* scan areas before each reservation */ 1181 for (; idx_b >= 0; idx_b--) { 1182 struct memblock_region *r; 1183 phys_addr_t r_start; 1184 phys_addr_t r_end; 1185 1186 r = &type_b->regions[idx_b]; 1187 r_start = idx_b ? r[-1].base + r[-1].size : 0; 1188 r_end = idx_b < type_b->cnt ? 1189 r->base : PHYS_ADDR_MAX; 1190 /* 1191 * if idx_b advanced past idx_a, 1192 * break out to advance idx_a 1193 */ 1194 1195 if (r_end <= m_start) 1196 break; 1197 /* if the two regions intersect, we're done */ 1198 if (m_end > r_start) { 1199 if (out_start) 1200 *out_start = max(m_start, r_start); 1201 if (out_end) 1202 *out_end = min(m_end, r_end); 1203 if (out_nid) 1204 *out_nid = m_nid; 1205 if (m_start >= r_start) 1206 idx_a--; 1207 else 1208 idx_b--; 1209 *idx = (u32)idx_a | (u64)idx_b << 32; 1210 return; 1211 } 1212 } 1213 } 1214 /* signal end of iteration */ 1215 *idx = ULLONG_MAX; 1216 } 1217 1218 /* 1219 * Common iterator interface used to define for_each_mem_pfn_range(). 1220 */ 1221 void __init_memblock __next_mem_pfn_range(int *idx, int nid, 1222 unsigned long *out_start_pfn, 1223 unsigned long *out_end_pfn, int *out_nid) 1224 { 1225 struct memblock_type *type = &memblock.memory; 1226 struct memblock_region *r; 1227 int r_nid; 1228 1229 while (++*idx < type->cnt) { 1230 r = &type->regions[*idx]; 1231 r_nid = memblock_get_region_node(r); 1232 1233 if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size)) 1234 continue; 1235 if (nid == MAX_NUMNODES || nid == r_nid) 1236 break; 1237 } 1238 if (*idx >= type->cnt) { 1239 *idx = -1; 1240 return; 1241 } 1242 1243 if (out_start_pfn) 1244 *out_start_pfn = PFN_UP(r->base); 1245 if (out_end_pfn) 1246 *out_end_pfn = PFN_DOWN(r->base + r->size); 1247 if (out_nid) 1248 *out_nid = r_nid; 1249 } 1250 1251 /** 1252 * memblock_set_node - set node ID on memblock regions 1253 * @base: base of area to set node ID for 1254 * @size: size of area to set node ID for 1255 * @type: memblock type to set node ID for 1256 * @nid: node ID to set 1257 * 1258 * Set the nid of memblock @type regions in [@base, @base + @size) to @nid. 1259 * Regions which cross the area boundaries are split as necessary. 1260 * 1261 * Return: 1262 * 0 on success, -errno on failure. 1263 */ 1264 int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size, 1265 struct memblock_type *type, int nid) 1266 { 1267 #ifdef CONFIG_NUMA 1268 int start_rgn, end_rgn; 1269 int i, ret; 1270 1271 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn); 1272 if (ret) 1273 return ret; 1274 1275 for (i = start_rgn; i < end_rgn; i++) 1276 memblock_set_region_node(&type->regions[i], nid); 1277 1278 memblock_merge_regions(type); 1279 #endif 1280 return 0; 1281 } 1282 1283 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT 1284 /** 1285 * __next_mem_pfn_range_in_zone - iterator for for_each_*_range_in_zone() 1286 * 1287 * @idx: pointer to u64 loop variable 1288 * @zone: zone in which all of the memory blocks reside 1289 * @out_spfn: ptr to ulong for start pfn of the range, can be %NULL 1290 * @out_epfn: ptr to ulong for end pfn of the range, can be %NULL 1291 * 1292 * This function is meant to be a zone/pfn specific wrapper for the 1293 * for_each_mem_range type iterators. Specifically they are used in the 1294 * deferred memory init routines and as such we were duplicating much of 1295 * this logic throughout the code. So instead of having it in multiple 1296 * locations it seemed like it would make more sense to centralize this to 1297 * one new iterator that does everything they need. 1298 */ 1299 void __init_memblock 1300 __next_mem_pfn_range_in_zone(u64 *idx, struct zone *zone, 1301 unsigned long *out_spfn, unsigned long *out_epfn) 1302 { 1303 int zone_nid = zone_to_nid(zone); 1304 phys_addr_t spa, epa; 1305 1306 __next_mem_range(idx, zone_nid, MEMBLOCK_NONE, 1307 &memblock.memory, &memblock.reserved, 1308 &spa, &epa, NULL); 1309 1310 while (*idx != U64_MAX) { 1311 unsigned long epfn = PFN_DOWN(epa); 1312 unsigned long spfn = PFN_UP(spa); 1313 1314 /* 1315 * Verify the end is at least past the start of the zone and 1316 * that we have at least one PFN to initialize. 1317 */ 1318 if (zone->zone_start_pfn < epfn && spfn < epfn) { 1319 /* if we went too far just stop searching */ 1320 if (zone_end_pfn(zone) <= spfn) { 1321 *idx = U64_MAX; 1322 break; 1323 } 1324 1325 if (out_spfn) 1326 *out_spfn = max(zone->zone_start_pfn, spfn); 1327 if (out_epfn) 1328 *out_epfn = min(zone_end_pfn(zone), epfn); 1329 1330 return; 1331 } 1332 1333 __next_mem_range(idx, zone_nid, MEMBLOCK_NONE, 1334 &memblock.memory, &memblock.reserved, 1335 &spa, &epa, NULL); 1336 } 1337 1338 /* signal end of iteration */ 1339 if (out_spfn) 1340 *out_spfn = ULONG_MAX; 1341 if (out_epfn) 1342 *out_epfn = 0; 1343 } 1344 1345 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ 1346 1347 /** 1348 * memblock_alloc_range_nid - allocate boot memory block 1349 * @size: size of memory block to be allocated in bytes 1350 * @align: alignment of the region and block's size 1351 * @start: the lower bound of the memory region to allocate (phys address) 1352 * @end: the upper bound of the memory region to allocate (phys address) 1353 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 1354 * @exact_nid: control the allocation fall back to other nodes 1355 * 1356 * The allocation is performed from memory region limited by 1357 * memblock.current_limit if @end == %MEMBLOCK_ALLOC_ACCESSIBLE. 1358 * 1359 * If the specified node can not hold the requested memory and @exact_nid 1360 * is false, the allocation falls back to any node in the system. 1361 * 1362 * For systems with memory mirroring, the allocation is attempted first 1363 * from the regions with mirroring enabled and then retried from any 1364 * memory region. 1365 * 1366 * In addition, function using kmemleak_alloc_phys for allocated boot 1367 * memory block, it is never reported as leaks. 1368 * 1369 * Return: 1370 * Physical address of allocated memory block on success, %0 on failure. 1371 */ 1372 phys_addr_t __init memblock_alloc_range_nid(phys_addr_t size, 1373 phys_addr_t align, phys_addr_t start, 1374 phys_addr_t end, int nid, 1375 bool exact_nid) 1376 { 1377 enum memblock_flags flags = choose_memblock_flags(); 1378 phys_addr_t found; 1379 1380 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n")) 1381 nid = NUMA_NO_NODE; 1382 1383 if (!align) { 1384 /* Can't use WARNs this early in boot on powerpc */ 1385 dump_stack(); 1386 align = SMP_CACHE_BYTES; 1387 } 1388 1389 again: 1390 found = memblock_find_in_range_node(size, align, start, end, nid, 1391 flags); 1392 if (found && !memblock_reserve(found, size)) 1393 goto done; 1394 1395 if (nid != NUMA_NO_NODE && !exact_nid) { 1396 found = memblock_find_in_range_node(size, align, start, 1397 end, NUMA_NO_NODE, 1398 flags); 1399 if (found && !memblock_reserve(found, size)) 1400 goto done; 1401 } 1402 1403 if (flags & MEMBLOCK_MIRROR) { 1404 flags &= ~MEMBLOCK_MIRROR; 1405 pr_warn_ratelimited("Could not allocate %pap bytes of mirrored memory\n", 1406 &size); 1407 goto again; 1408 } 1409 1410 return 0; 1411 1412 done: 1413 /* 1414 * Skip kmemleak for those places like kasan_init() and 1415 * early_pgtable_alloc() due to high volume. 1416 */ 1417 if (end != MEMBLOCK_ALLOC_NOLEAKTRACE) 1418 /* 1419 * Memblock allocated blocks are never reported as 1420 * leaks. This is because many of these blocks are 1421 * only referred via the physical address which is 1422 * not looked up by kmemleak. 1423 */ 1424 kmemleak_alloc_phys(found, size, 0); 1425 1426 return found; 1427 } 1428 1429 /** 1430 * memblock_phys_alloc_range - allocate a memory block inside specified range 1431 * @size: size of memory block to be allocated in bytes 1432 * @align: alignment of the region and block's size 1433 * @start: the lower bound of the memory region to allocate (physical address) 1434 * @end: the upper bound of the memory region to allocate (physical address) 1435 * 1436 * Allocate @size bytes in the between @start and @end. 1437 * 1438 * Return: physical address of the allocated memory block on success, 1439 * %0 on failure. 1440 */ 1441 phys_addr_t __init memblock_phys_alloc_range(phys_addr_t size, 1442 phys_addr_t align, 1443 phys_addr_t start, 1444 phys_addr_t end) 1445 { 1446 memblock_dbg("%s: %llu bytes align=0x%llx from=%pa max_addr=%pa %pS\n", 1447 __func__, (u64)size, (u64)align, &start, &end, 1448 (void *)_RET_IP_); 1449 return memblock_alloc_range_nid(size, align, start, end, NUMA_NO_NODE, 1450 false); 1451 } 1452 1453 /** 1454 * memblock_phys_alloc_try_nid - allocate a memory block from specified NUMA node 1455 * @size: size of memory block to be allocated in bytes 1456 * @align: alignment of the region and block's size 1457 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 1458 * 1459 * Allocates memory block from the specified NUMA node. If the node 1460 * has no available memory, attempts to allocated from any node in the 1461 * system. 1462 * 1463 * Return: physical address of the allocated memory block on success, 1464 * %0 on failure. 1465 */ 1466 phys_addr_t __init memblock_phys_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid) 1467 { 1468 return memblock_alloc_range_nid(size, align, 0, 1469 MEMBLOCK_ALLOC_ACCESSIBLE, nid, false); 1470 } 1471 1472 /** 1473 * memblock_alloc_internal - allocate boot memory block 1474 * @size: size of memory block to be allocated in bytes 1475 * @align: alignment of the region and block's size 1476 * @min_addr: the lower bound of the memory region to allocate (phys address) 1477 * @max_addr: the upper bound of the memory region to allocate (phys address) 1478 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 1479 * @exact_nid: control the allocation fall back to other nodes 1480 * 1481 * Allocates memory block using memblock_alloc_range_nid() and 1482 * converts the returned physical address to virtual. 1483 * 1484 * The @min_addr limit is dropped if it can not be satisfied and the allocation 1485 * will fall back to memory below @min_addr. Other constraints, such 1486 * as node and mirrored memory will be handled again in 1487 * memblock_alloc_range_nid(). 1488 * 1489 * Return: 1490 * Virtual address of allocated memory block on success, NULL on failure. 1491 */ 1492 static void * __init memblock_alloc_internal( 1493 phys_addr_t size, phys_addr_t align, 1494 phys_addr_t min_addr, phys_addr_t max_addr, 1495 int nid, bool exact_nid) 1496 { 1497 phys_addr_t alloc; 1498 1499 /* 1500 * Detect any accidental use of these APIs after slab is ready, as at 1501 * this moment memblock may be deinitialized already and its 1502 * internal data may be destroyed (after execution of memblock_free_all) 1503 */ 1504 if (WARN_ON_ONCE(slab_is_available())) 1505 return kzalloc_node(size, GFP_NOWAIT, nid); 1506 1507 if (max_addr > memblock.current_limit) 1508 max_addr = memblock.current_limit; 1509 1510 alloc = memblock_alloc_range_nid(size, align, min_addr, max_addr, nid, 1511 exact_nid); 1512 1513 /* retry allocation without lower limit */ 1514 if (!alloc && min_addr) 1515 alloc = memblock_alloc_range_nid(size, align, 0, max_addr, nid, 1516 exact_nid); 1517 1518 if (!alloc) 1519 return NULL; 1520 1521 return phys_to_virt(alloc); 1522 } 1523 1524 /** 1525 * memblock_alloc_exact_nid_raw - allocate boot memory block on the exact node 1526 * without zeroing memory 1527 * @size: size of memory block to be allocated in bytes 1528 * @align: alignment of the region and block's size 1529 * @min_addr: the lower bound of the memory region from where the allocation 1530 * is preferred (phys address) 1531 * @max_addr: the upper bound of the memory region from where the allocation 1532 * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to 1533 * allocate only from memory limited by memblock.current_limit value 1534 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 1535 * 1536 * Public function, provides additional debug information (including caller 1537 * info), if enabled. Does not zero allocated memory. 1538 * 1539 * Return: 1540 * Virtual address of allocated memory block on success, NULL on failure. 1541 */ 1542 void * __init memblock_alloc_exact_nid_raw( 1543 phys_addr_t size, phys_addr_t align, 1544 phys_addr_t min_addr, phys_addr_t max_addr, 1545 int nid) 1546 { 1547 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n", 1548 __func__, (u64)size, (u64)align, nid, &min_addr, 1549 &max_addr, (void *)_RET_IP_); 1550 1551 return memblock_alloc_internal(size, align, min_addr, max_addr, nid, 1552 true); 1553 } 1554 1555 /** 1556 * memblock_alloc_try_nid_raw - allocate boot memory block without zeroing 1557 * memory and without panicking 1558 * @size: size of memory block to be allocated in bytes 1559 * @align: alignment of the region and block's size 1560 * @min_addr: the lower bound of the memory region from where the allocation 1561 * is preferred (phys address) 1562 * @max_addr: the upper bound of the memory region from where the allocation 1563 * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to 1564 * allocate only from memory limited by memblock.current_limit value 1565 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 1566 * 1567 * Public function, provides additional debug information (including caller 1568 * info), if enabled. Does not zero allocated memory, does not panic if request 1569 * cannot be satisfied. 1570 * 1571 * Return: 1572 * Virtual address of allocated memory block on success, NULL on failure. 1573 */ 1574 void * __init memblock_alloc_try_nid_raw( 1575 phys_addr_t size, phys_addr_t align, 1576 phys_addr_t min_addr, phys_addr_t max_addr, 1577 int nid) 1578 { 1579 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n", 1580 __func__, (u64)size, (u64)align, nid, &min_addr, 1581 &max_addr, (void *)_RET_IP_); 1582 1583 return memblock_alloc_internal(size, align, min_addr, max_addr, nid, 1584 false); 1585 } 1586 1587 /** 1588 * memblock_alloc_try_nid - allocate boot memory block 1589 * @size: size of memory block to be allocated in bytes 1590 * @align: alignment of the region and block's size 1591 * @min_addr: the lower bound of the memory region from where the allocation 1592 * is preferred (phys address) 1593 * @max_addr: the upper bound of the memory region from where the allocation 1594 * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to 1595 * allocate only from memory limited by memblock.current_limit value 1596 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 1597 * 1598 * Public function, provides additional debug information (including caller 1599 * info), if enabled. This function zeroes the allocated memory. 1600 * 1601 * Return: 1602 * Virtual address of allocated memory block on success, NULL on failure. 1603 */ 1604 void * __init memblock_alloc_try_nid( 1605 phys_addr_t size, phys_addr_t align, 1606 phys_addr_t min_addr, phys_addr_t max_addr, 1607 int nid) 1608 { 1609 void *ptr; 1610 1611 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n", 1612 __func__, (u64)size, (u64)align, nid, &min_addr, 1613 &max_addr, (void *)_RET_IP_); 1614 ptr = memblock_alloc_internal(size, align, 1615 min_addr, max_addr, nid, false); 1616 if (ptr) 1617 memset(ptr, 0, size); 1618 1619 return ptr; 1620 } 1621 1622 /** 1623 * memblock_free_late - free pages directly to buddy allocator 1624 * @base: phys starting address of the boot memory block 1625 * @size: size of the boot memory block in bytes 1626 * 1627 * This is only useful when the memblock allocator has already been torn 1628 * down, but we are still initializing the system. Pages are released directly 1629 * to the buddy allocator. 1630 */ 1631 void __init memblock_free_late(phys_addr_t base, phys_addr_t size) 1632 { 1633 phys_addr_t cursor, end; 1634 1635 end = base + size - 1; 1636 memblock_dbg("%s: [%pa-%pa] %pS\n", 1637 __func__, &base, &end, (void *)_RET_IP_); 1638 kmemleak_free_part_phys(base, size); 1639 cursor = PFN_UP(base); 1640 end = PFN_DOWN(base + size); 1641 1642 for (; cursor < end; cursor++) { 1643 /* 1644 * Reserved pages are always initialized by the end of 1645 * memblock_free_all() (by memmap_init() and, if deferred 1646 * initialization is enabled, memmap_init_reserved_pages()), so 1647 * these pages can be released directly to the buddy allocator. 1648 */ 1649 __free_pages_core(pfn_to_page(cursor), 0); 1650 totalram_pages_inc(); 1651 } 1652 } 1653 1654 /* 1655 * Remaining API functions 1656 */ 1657 1658 phys_addr_t __init_memblock memblock_phys_mem_size(void) 1659 { 1660 return memblock.memory.total_size; 1661 } 1662 1663 phys_addr_t __init_memblock memblock_reserved_size(void) 1664 { 1665 return memblock.reserved.total_size; 1666 } 1667 1668 /* lowest address */ 1669 phys_addr_t __init_memblock memblock_start_of_DRAM(void) 1670 { 1671 return memblock.memory.regions[0].base; 1672 } 1673 1674 phys_addr_t __init_memblock memblock_end_of_DRAM(void) 1675 { 1676 int idx = memblock.memory.cnt - 1; 1677 1678 return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size); 1679 } 1680 1681 static phys_addr_t __init_memblock __find_max_addr(phys_addr_t limit) 1682 { 1683 phys_addr_t max_addr = PHYS_ADDR_MAX; 1684 struct memblock_region *r; 1685 1686 /* 1687 * translate the memory @limit size into the max address within one of 1688 * the memory memblock regions, if the @limit exceeds the total size 1689 * of those regions, max_addr will keep original value PHYS_ADDR_MAX 1690 */ 1691 for_each_mem_region(r) { 1692 if (limit <= r->size) { 1693 max_addr = r->base + limit; 1694 break; 1695 } 1696 limit -= r->size; 1697 } 1698 1699 return max_addr; 1700 } 1701 1702 void __init memblock_enforce_memory_limit(phys_addr_t limit) 1703 { 1704 phys_addr_t max_addr; 1705 1706 if (!limit) 1707 return; 1708 1709 max_addr = __find_max_addr(limit); 1710 1711 /* @limit exceeds the total size of the memory, do nothing */ 1712 if (max_addr == PHYS_ADDR_MAX) 1713 return; 1714 1715 /* truncate both memory and reserved regions */ 1716 memblock_remove_range(&memblock.memory, max_addr, 1717 PHYS_ADDR_MAX); 1718 memblock_remove_range(&memblock.reserved, max_addr, 1719 PHYS_ADDR_MAX); 1720 } 1721 1722 void __init memblock_cap_memory_range(phys_addr_t base, phys_addr_t size) 1723 { 1724 int start_rgn, end_rgn; 1725 int i, ret; 1726 1727 if (!size) 1728 return; 1729 1730 if (!memblock_memory->total_size) { 1731 pr_warn("%s: No memory registered yet\n", __func__); 1732 return; 1733 } 1734 1735 ret = memblock_isolate_range(&memblock.memory, base, size, 1736 &start_rgn, &end_rgn); 1737 if (ret) 1738 return; 1739 1740 /* remove all the MAP regions */ 1741 for (i = memblock.memory.cnt - 1; i >= end_rgn; i--) 1742 if (!memblock_is_nomap(&memblock.memory.regions[i])) 1743 memblock_remove_region(&memblock.memory, i); 1744 1745 for (i = start_rgn - 1; i >= 0; i--) 1746 if (!memblock_is_nomap(&memblock.memory.regions[i])) 1747 memblock_remove_region(&memblock.memory, i); 1748 1749 /* truncate the reserved regions */ 1750 memblock_remove_range(&memblock.reserved, 0, base); 1751 memblock_remove_range(&memblock.reserved, 1752 base + size, PHYS_ADDR_MAX); 1753 } 1754 1755 void __init memblock_mem_limit_remove_map(phys_addr_t limit) 1756 { 1757 phys_addr_t max_addr; 1758 1759 if (!limit) 1760 return; 1761 1762 max_addr = __find_max_addr(limit); 1763 1764 /* @limit exceeds the total size of the memory, do nothing */ 1765 if (max_addr == PHYS_ADDR_MAX) 1766 return; 1767 1768 memblock_cap_memory_range(0, max_addr); 1769 } 1770 1771 static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr) 1772 { 1773 unsigned int left = 0, right = type->cnt; 1774 1775 do { 1776 unsigned int mid = (right + left) / 2; 1777 1778 if (addr < type->regions[mid].base) 1779 right = mid; 1780 else if (addr >= (type->regions[mid].base + 1781 type->regions[mid].size)) 1782 left = mid + 1; 1783 else 1784 return mid; 1785 } while (left < right); 1786 return -1; 1787 } 1788 1789 bool __init_memblock memblock_is_reserved(phys_addr_t addr) 1790 { 1791 return memblock_search(&memblock.reserved, addr) != -1; 1792 } 1793 1794 bool __init_memblock memblock_is_memory(phys_addr_t addr) 1795 { 1796 return memblock_search(&memblock.memory, addr) != -1; 1797 } 1798 1799 bool __init_memblock memblock_is_map_memory(phys_addr_t addr) 1800 { 1801 int i = memblock_search(&memblock.memory, addr); 1802 1803 if (i == -1) 1804 return false; 1805 return !memblock_is_nomap(&memblock.memory.regions[i]); 1806 } 1807 1808 int __init_memblock memblock_search_pfn_nid(unsigned long pfn, 1809 unsigned long *start_pfn, unsigned long *end_pfn) 1810 { 1811 struct memblock_type *type = &memblock.memory; 1812 int mid = memblock_search(type, PFN_PHYS(pfn)); 1813 1814 if (mid == -1) 1815 return -1; 1816 1817 *start_pfn = PFN_DOWN(type->regions[mid].base); 1818 *end_pfn = PFN_DOWN(type->regions[mid].base + type->regions[mid].size); 1819 1820 return memblock_get_region_node(&type->regions[mid]); 1821 } 1822 1823 /** 1824 * memblock_is_region_memory - check if a region is a subset of memory 1825 * @base: base of region to check 1826 * @size: size of region to check 1827 * 1828 * Check if the region [@base, @base + @size) is a subset of a memory block. 1829 * 1830 * Return: 1831 * 0 if false, non-zero if true 1832 */ 1833 bool __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size) 1834 { 1835 int idx = memblock_search(&memblock.memory, base); 1836 phys_addr_t end = base + memblock_cap_size(base, &size); 1837 1838 if (idx == -1) 1839 return false; 1840 return (memblock.memory.regions[idx].base + 1841 memblock.memory.regions[idx].size) >= end; 1842 } 1843 1844 /** 1845 * memblock_is_region_reserved - check if a region intersects reserved memory 1846 * @base: base of region to check 1847 * @size: size of region to check 1848 * 1849 * Check if the region [@base, @base + @size) intersects a reserved 1850 * memory block. 1851 * 1852 * Return: 1853 * True if they intersect, false if not. 1854 */ 1855 bool __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size) 1856 { 1857 return memblock_overlaps_region(&memblock.reserved, base, size); 1858 } 1859 1860 void __init_memblock memblock_trim_memory(phys_addr_t align) 1861 { 1862 phys_addr_t start, end, orig_start, orig_end; 1863 struct memblock_region *r; 1864 1865 for_each_mem_region(r) { 1866 orig_start = r->base; 1867 orig_end = r->base + r->size; 1868 start = round_up(orig_start, align); 1869 end = round_down(orig_end, align); 1870 1871 if (start == orig_start && end == orig_end) 1872 continue; 1873 1874 if (start < end) { 1875 r->base = start; 1876 r->size = end - start; 1877 } else { 1878 memblock_remove_region(&memblock.memory, 1879 r - memblock.memory.regions); 1880 r--; 1881 } 1882 } 1883 } 1884 1885 void __init_memblock memblock_set_current_limit(phys_addr_t limit) 1886 { 1887 memblock.current_limit = limit; 1888 } 1889 1890 phys_addr_t __init_memblock memblock_get_current_limit(void) 1891 { 1892 return memblock.current_limit; 1893 } 1894 1895 static void __init_memblock memblock_dump(struct memblock_type *type) 1896 { 1897 phys_addr_t base, end, size; 1898 enum memblock_flags flags; 1899 int idx; 1900 struct memblock_region *rgn; 1901 1902 pr_info(" %s.cnt = 0x%lx\n", type->name, type->cnt); 1903 1904 for_each_memblock_type(idx, type, rgn) { 1905 char nid_buf[32] = ""; 1906 1907 base = rgn->base; 1908 size = rgn->size; 1909 end = base + size - 1; 1910 flags = rgn->flags; 1911 #ifdef CONFIG_NUMA 1912 if (memblock_get_region_node(rgn) != MAX_NUMNODES) 1913 snprintf(nid_buf, sizeof(nid_buf), " on node %d", 1914 memblock_get_region_node(rgn)); 1915 #endif 1916 pr_info(" %s[%#x]\t[%pa-%pa], %pa bytes%s flags: %#x\n", 1917 type->name, idx, &base, &end, &size, nid_buf, flags); 1918 } 1919 } 1920 1921 static void __init_memblock __memblock_dump_all(void) 1922 { 1923 pr_info("MEMBLOCK configuration:\n"); 1924 pr_info(" memory size = %pa reserved size = %pa\n", 1925 &memblock.memory.total_size, 1926 &memblock.reserved.total_size); 1927 1928 memblock_dump(&memblock.memory); 1929 memblock_dump(&memblock.reserved); 1930 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP 1931 memblock_dump(&physmem); 1932 #endif 1933 } 1934 1935 void __init_memblock memblock_dump_all(void) 1936 { 1937 if (memblock_debug) 1938 __memblock_dump_all(); 1939 } 1940 1941 void __init memblock_allow_resize(void) 1942 { 1943 memblock_can_resize = 1; 1944 } 1945 1946 static int __init early_memblock(char *p) 1947 { 1948 if (p && strstr(p, "debug")) 1949 memblock_debug = 1; 1950 return 0; 1951 } 1952 early_param("memblock", early_memblock); 1953 1954 static void __init free_memmap(unsigned long start_pfn, unsigned long end_pfn) 1955 { 1956 struct page *start_pg, *end_pg; 1957 phys_addr_t pg, pgend; 1958 1959 /* 1960 * Convert start_pfn/end_pfn to a struct page pointer. 1961 */ 1962 start_pg = pfn_to_page(start_pfn - 1) + 1; 1963 end_pg = pfn_to_page(end_pfn - 1) + 1; 1964 1965 /* 1966 * Convert to physical addresses, and round start upwards and end 1967 * downwards. 1968 */ 1969 pg = PAGE_ALIGN(__pa(start_pg)); 1970 pgend = __pa(end_pg) & PAGE_MASK; 1971 1972 /* 1973 * If there are free pages between these, free the section of the 1974 * memmap array. 1975 */ 1976 if (pg < pgend) 1977 memblock_phys_free(pg, pgend - pg); 1978 } 1979 1980 /* 1981 * The mem_map array can get very big. Free the unused area of the memory map. 1982 */ 1983 static void __init free_unused_memmap(void) 1984 { 1985 unsigned long start, end, prev_end = 0; 1986 int i; 1987 1988 if (!IS_ENABLED(CONFIG_HAVE_ARCH_PFN_VALID) || 1989 IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP)) 1990 return; 1991 1992 /* 1993 * This relies on each bank being in address order. 1994 * The banks are sorted previously in bootmem_init(). 1995 */ 1996 for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, NULL) { 1997 #ifdef CONFIG_SPARSEMEM 1998 /* 1999 * Take care not to free memmap entries that don't exist 2000 * due to SPARSEMEM sections which aren't present. 2001 */ 2002 start = min(start, ALIGN(prev_end, PAGES_PER_SECTION)); 2003 #endif 2004 /* 2005 * Align down here since many operations in VM subsystem 2006 * presume that there are no holes in the memory map inside 2007 * a pageblock 2008 */ 2009 start = pageblock_start_pfn(start); 2010 2011 /* 2012 * If we had a previous bank, and there is a space 2013 * between the current bank and the previous, free it. 2014 */ 2015 if (prev_end && prev_end < start) 2016 free_memmap(prev_end, start); 2017 2018 /* 2019 * Align up here since many operations in VM subsystem 2020 * presume that there are no holes in the memory map inside 2021 * a pageblock 2022 */ 2023 prev_end = pageblock_align(end); 2024 } 2025 2026 #ifdef CONFIG_SPARSEMEM 2027 if (!IS_ALIGNED(prev_end, PAGES_PER_SECTION)) { 2028 prev_end = pageblock_align(end); 2029 free_memmap(prev_end, ALIGN(prev_end, PAGES_PER_SECTION)); 2030 } 2031 #endif 2032 } 2033 2034 static void __init __free_pages_memory(unsigned long start, unsigned long end) 2035 { 2036 int order; 2037 2038 while (start < end) { 2039 order = min(MAX_ORDER - 1UL, __ffs(start)); 2040 2041 while (start + (1UL << order) > end) 2042 order--; 2043 2044 memblock_free_pages(pfn_to_page(start), start, order); 2045 2046 start += (1UL << order); 2047 } 2048 } 2049 2050 static unsigned long __init __free_memory_core(phys_addr_t start, 2051 phys_addr_t end) 2052 { 2053 unsigned long start_pfn = PFN_UP(start); 2054 unsigned long end_pfn = min_t(unsigned long, 2055 PFN_DOWN(end), max_low_pfn); 2056 2057 if (start_pfn >= end_pfn) 2058 return 0; 2059 2060 __free_pages_memory(start_pfn, end_pfn); 2061 2062 return end_pfn - start_pfn; 2063 } 2064 2065 static void __init memmap_init_reserved_pages(void) 2066 { 2067 struct memblock_region *region; 2068 phys_addr_t start, end; 2069 u64 i; 2070 2071 /* initialize struct pages for the reserved regions */ 2072 for_each_reserved_mem_range(i, &start, &end) 2073 reserve_bootmem_region(start, end); 2074 2075 /* and also treat struct pages for the NOMAP regions as PageReserved */ 2076 for_each_mem_region(region) { 2077 if (memblock_is_nomap(region)) { 2078 start = region->base; 2079 end = start + region->size; 2080 reserve_bootmem_region(start, end); 2081 } 2082 } 2083 } 2084 2085 static unsigned long __init free_low_memory_core_early(void) 2086 { 2087 unsigned long count = 0; 2088 phys_addr_t start, end; 2089 u64 i; 2090 2091 memblock_clear_hotplug(0, -1); 2092 2093 memmap_init_reserved_pages(); 2094 2095 /* 2096 * We need to use NUMA_NO_NODE instead of NODE_DATA(0)->node_id 2097 * because in some case like Node0 doesn't have RAM installed 2098 * low ram will be on Node1 2099 */ 2100 for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE, &start, &end, 2101 NULL) 2102 count += __free_memory_core(start, end); 2103 2104 return count; 2105 } 2106 2107 static int reset_managed_pages_done __initdata; 2108 2109 void reset_node_managed_pages(pg_data_t *pgdat) 2110 { 2111 struct zone *z; 2112 2113 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++) 2114 atomic_long_set(&z->managed_pages, 0); 2115 } 2116 2117 void __init reset_all_zones_managed_pages(void) 2118 { 2119 struct pglist_data *pgdat; 2120 2121 if (reset_managed_pages_done) 2122 return; 2123 2124 for_each_online_pgdat(pgdat) 2125 reset_node_managed_pages(pgdat); 2126 2127 reset_managed_pages_done = 1; 2128 } 2129 2130 /** 2131 * memblock_free_all - release free pages to the buddy allocator 2132 */ 2133 void __init memblock_free_all(void) 2134 { 2135 unsigned long pages; 2136 2137 free_unused_memmap(); 2138 reset_all_zones_managed_pages(); 2139 2140 pages = free_low_memory_core_early(); 2141 totalram_pages_add(pages); 2142 } 2143 2144 #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_ARCH_KEEP_MEMBLOCK) 2145 2146 static int memblock_debug_show(struct seq_file *m, void *private) 2147 { 2148 struct memblock_type *type = m->private; 2149 struct memblock_region *reg; 2150 int i; 2151 phys_addr_t end; 2152 2153 for (i = 0; i < type->cnt; i++) { 2154 reg = &type->regions[i]; 2155 end = reg->base + reg->size - 1; 2156 2157 seq_printf(m, "%4d: ", i); 2158 seq_printf(m, "%pa..%pa\n", ®->base, &end); 2159 } 2160 return 0; 2161 } 2162 DEFINE_SHOW_ATTRIBUTE(memblock_debug); 2163 2164 static int __init memblock_init_debugfs(void) 2165 { 2166 struct dentry *root = debugfs_create_dir("memblock", NULL); 2167 2168 debugfs_create_file("memory", 0444, root, 2169 &memblock.memory, &memblock_debug_fops); 2170 debugfs_create_file("reserved", 0444, root, 2171 &memblock.reserved, &memblock_debug_fops); 2172 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP 2173 debugfs_create_file("physmem", 0444, root, &physmem, 2174 &memblock_debug_fops); 2175 #endif 2176 2177 return 0; 2178 } 2179 __initcall(memblock_init_debugfs); 2180 2181 #endif /* CONFIG_DEBUG_FS */ 2182