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