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; 160 static int memblock_can_resize __initdata_memblock; 161 static int memblock_memory_in_slab __initdata_memblock; 162 static int memblock_reserved_in_slab __initdata_memblock; 163 164 bool __init_memblock memblock_has_mirror(void) 165 { 166 return system_has_some_mirror; 167 } 168 169 static enum memblock_flags __init_memblock choose_memblock_flags(void) 170 { 171 return system_has_some_mirror ? MEMBLOCK_MIRROR : MEMBLOCK_NONE; 172 } 173 174 /* adjust *@size so that (@base + *@size) doesn't overflow, return new size */ 175 static inline phys_addr_t memblock_cap_size(phys_addr_t base, phys_addr_t *size) 176 { 177 return *size = min(*size, PHYS_ADDR_MAX - base); 178 } 179 180 /* 181 * Address comparison utilities 182 */ 183 static unsigned long __init_memblock memblock_addrs_overlap(phys_addr_t base1, phys_addr_t size1, 184 phys_addr_t base2, phys_addr_t size2) 185 { 186 return ((base1 < (base2 + size2)) && (base2 < (base1 + size1))); 187 } 188 189 bool __init_memblock memblock_overlaps_region(struct memblock_type *type, 190 phys_addr_t base, phys_addr_t size) 191 { 192 unsigned long i; 193 194 memblock_cap_size(base, &size); 195 196 for (i = 0; i < type->cnt; i++) 197 if (memblock_addrs_overlap(base, size, type->regions[i].base, 198 type->regions[i].size)) 199 break; 200 return i < type->cnt; 201 } 202 203 /** 204 * __memblock_find_range_bottom_up - find free area utility in bottom-up 205 * @start: start of candidate range 206 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or 207 * %MEMBLOCK_ALLOC_ACCESSIBLE 208 * @size: size of free area to find 209 * @align: alignment of free area to find 210 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 211 * @flags: pick from blocks based on memory attributes 212 * 213 * Utility called from memblock_find_in_range_node(), find free area bottom-up. 214 * 215 * Return: 216 * Found address on success, 0 on failure. 217 */ 218 static phys_addr_t __init_memblock 219 __memblock_find_range_bottom_up(phys_addr_t start, phys_addr_t end, 220 phys_addr_t size, phys_addr_t align, int nid, 221 enum memblock_flags flags) 222 { 223 phys_addr_t this_start, this_end, cand; 224 u64 i; 225 226 for_each_free_mem_range(i, nid, flags, &this_start, &this_end, NULL) { 227 this_start = clamp(this_start, start, end); 228 this_end = clamp(this_end, start, end); 229 230 cand = round_up(this_start, align); 231 if (cand < this_end && this_end - cand >= size) 232 return cand; 233 } 234 235 return 0; 236 } 237 238 /** 239 * __memblock_find_range_top_down - find free area utility, in top-down 240 * @start: start of candidate range 241 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or 242 * %MEMBLOCK_ALLOC_ACCESSIBLE 243 * @size: size of free area to find 244 * @align: alignment of free area to find 245 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 246 * @flags: pick from blocks based on memory attributes 247 * 248 * Utility called from memblock_find_in_range_node(), find free area top-down. 249 * 250 * Return: 251 * Found address on success, 0 on failure. 252 */ 253 static phys_addr_t __init_memblock 254 __memblock_find_range_top_down(phys_addr_t start, phys_addr_t end, 255 phys_addr_t size, phys_addr_t align, int nid, 256 enum memblock_flags flags) 257 { 258 phys_addr_t this_start, this_end, cand; 259 u64 i; 260 261 for_each_free_mem_range_reverse(i, nid, flags, &this_start, &this_end, 262 NULL) { 263 this_start = clamp(this_start, start, end); 264 this_end = clamp(this_end, start, end); 265 266 if (this_end < size) 267 continue; 268 269 cand = round_down(this_end - size, align); 270 if (cand >= this_start) 271 return cand; 272 } 273 274 return 0; 275 } 276 277 /** 278 * memblock_find_in_range_node - find free area in given range and node 279 * @size: size of free area to find 280 * @align: alignment of free area to find 281 * @start: start of candidate range 282 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or 283 * %MEMBLOCK_ALLOC_ACCESSIBLE 284 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 285 * @flags: pick from blocks based on memory attributes 286 * 287 * Find @size free area aligned to @align in the specified range and node. 288 * 289 * Return: 290 * Found address on success, 0 on failure. 291 */ 292 static phys_addr_t __init_memblock memblock_find_in_range_node(phys_addr_t size, 293 phys_addr_t align, phys_addr_t start, 294 phys_addr_t end, int nid, 295 enum memblock_flags flags) 296 { 297 /* pump up @end */ 298 if (end == MEMBLOCK_ALLOC_ACCESSIBLE || 299 end == MEMBLOCK_ALLOC_NOLEAKTRACE) 300 end = memblock.current_limit; 301 302 /* avoid allocating the first page */ 303 start = max_t(phys_addr_t, start, PAGE_SIZE); 304 end = max(start, end); 305 306 if (memblock_bottom_up()) 307 return __memblock_find_range_bottom_up(start, end, size, align, 308 nid, flags); 309 else 310 return __memblock_find_range_top_down(start, end, size, align, 311 nid, flags); 312 } 313 314 /** 315 * memblock_find_in_range - find free area in given range 316 * @start: start of candidate range 317 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or 318 * %MEMBLOCK_ALLOC_ACCESSIBLE 319 * @size: size of free area to find 320 * @align: alignment of free area to find 321 * 322 * Find @size free area aligned to @align in the specified range. 323 * 324 * Return: 325 * Found address on success, 0 on failure. 326 */ 327 static phys_addr_t __init_memblock memblock_find_in_range(phys_addr_t start, 328 phys_addr_t end, phys_addr_t size, 329 phys_addr_t align) 330 { 331 phys_addr_t ret; 332 enum memblock_flags flags = choose_memblock_flags(); 333 334 again: 335 ret = memblock_find_in_range_node(size, align, start, end, 336 NUMA_NO_NODE, flags); 337 338 if (!ret && (flags & MEMBLOCK_MIRROR)) { 339 pr_warn_ratelimited("Could not allocate %pap bytes of mirrored memory\n", 340 &size); 341 flags &= ~MEMBLOCK_MIRROR; 342 goto again; 343 } 344 345 return ret; 346 } 347 348 static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r) 349 { 350 type->total_size -= type->regions[r].size; 351 memmove(&type->regions[r], &type->regions[r + 1], 352 (type->cnt - (r + 1)) * sizeof(type->regions[r])); 353 type->cnt--; 354 355 /* Special case for empty arrays */ 356 if (type->cnt == 0) { 357 WARN_ON(type->total_size != 0); 358 type->cnt = 1; 359 type->regions[0].base = 0; 360 type->regions[0].size = 0; 361 type->regions[0].flags = 0; 362 memblock_set_region_node(&type->regions[0], MAX_NUMNODES); 363 } 364 } 365 366 #ifndef CONFIG_ARCH_KEEP_MEMBLOCK 367 /** 368 * memblock_discard - discard memory and reserved arrays if they were allocated 369 */ 370 void __init memblock_discard(void) 371 { 372 phys_addr_t addr, size; 373 374 if (memblock.reserved.regions != memblock_reserved_init_regions) { 375 addr = __pa(memblock.reserved.regions); 376 size = PAGE_ALIGN(sizeof(struct memblock_region) * 377 memblock.reserved.max); 378 if (memblock_reserved_in_slab) 379 kfree(memblock.reserved.regions); 380 else 381 memblock_free_late(addr, size); 382 } 383 384 if (memblock.memory.regions != memblock_memory_init_regions) { 385 addr = __pa(memblock.memory.regions); 386 size = PAGE_ALIGN(sizeof(struct memblock_region) * 387 memblock.memory.max); 388 if (memblock_memory_in_slab) 389 kfree(memblock.memory.regions); 390 else 391 memblock_free_late(addr, size); 392 } 393 394 memblock_memory = NULL; 395 } 396 #endif 397 398 /** 399 * memblock_double_array - double the size of the memblock regions array 400 * @type: memblock type of the regions array being doubled 401 * @new_area_start: starting address of memory range to avoid overlap with 402 * @new_area_size: size of memory range to avoid overlap with 403 * 404 * Double the size of the @type regions array. If memblock is being used to 405 * allocate memory for a new reserved regions array and there is a previously 406 * allocated memory range [@new_area_start, @new_area_start + @new_area_size] 407 * waiting to be reserved, ensure the memory used by the new array does 408 * not overlap. 409 * 410 * Return: 411 * 0 on success, -1 on failure. 412 */ 413 static int __init_memblock memblock_double_array(struct memblock_type *type, 414 phys_addr_t new_area_start, 415 phys_addr_t new_area_size) 416 { 417 struct memblock_region *new_array, *old_array; 418 phys_addr_t old_alloc_size, new_alloc_size; 419 phys_addr_t old_size, new_size, addr, new_end; 420 int use_slab = slab_is_available(); 421 int *in_slab; 422 423 /* We don't allow resizing until we know about the reserved regions 424 * of memory that aren't suitable for allocation 425 */ 426 if (!memblock_can_resize) 427 return -1; 428 429 /* Calculate new doubled size */ 430 old_size = type->max * sizeof(struct memblock_region); 431 new_size = old_size << 1; 432 /* 433 * We need to allocated new one align to PAGE_SIZE, 434 * so we can free them completely later. 435 */ 436 old_alloc_size = PAGE_ALIGN(old_size); 437 new_alloc_size = PAGE_ALIGN(new_size); 438 439 /* Retrieve the slab flag */ 440 if (type == &memblock.memory) 441 in_slab = &memblock_memory_in_slab; 442 else 443 in_slab = &memblock_reserved_in_slab; 444 445 /* Try to find some space for it */ 446 if (use_slab) { 447 new_array = kmalloc(new_size, GFP_KERNEL); 448 addr = new_array ? __pa(new_array) : 0; 449 } else { 450 /* only exclude range when trying to double reserved.regions */ 451 if (type != &memblock.reserved) 452 new_area_start = new_area_size = 0; 453 454 addr = memblock_find_in_range(new_area_start + new_area_size, 455 memblock.current_limit, 456 new_alloc_size, PAGE_SIZE); 457 if (!addr && new_area_size) 458 addr = memblock_find_in_range(0, 459 min(new_area_start, memblock.current_limit), 460 new_alloc_size, PAGE_SIZE); 461 462 new_array = addr ? __va(addr) : NULL; 463 } 464 if (!addr) { 465 pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n", 466 type->name, type->max, type->max * 2); 467 return -1; 468 } 469 470 new_end = addr + new_size - 1; 471 memblock_dbg("memblock: %s is doubled to %ld at [%pa-%pa]", 472 type->name, type->max * 2, &addr, &new_end); 473 474 /* 475 * Found space, we now need to move the array over before we add the 476 * reserved region since it may be our reserved array itself that is 477 * full. 478 */ 479 memcpy(new_array, type->regions, old_size); 480 memset(new_array + type->max, 0, old_size); 481 old_array = type->regions; 482 type->regions = new_array; 483 type->max <<= 1; 484 485 /* Free old array. We needn't free it if the array is the static one */ 486 if (*in_slab) 487 kfree(old_array); 488 else if (old_array != memblock_memory_init_regions && 489 old_array != memblock_reserved_init_regions) 490 memblock_free(old_array, old_alloc_size); 491 492 /* 493 * Reserve the new array if that comes from the memblock. Otherwise, we 494 * needn't do it 495 */ 496 if (!use_slab) 497 BUG_ON(memblock_reserve(addr, new_alloc_size)); 498 499 /* Update slab flag */ 500 *in_slab = use_slab; 501 502 return 0; 503 } 504 505 /** 506 * memblock_merge_regions - merge neighboring compatible regions 507 * @type: memblock type to scan 508 * @start_rgn: start scanning from (@start_rgn - 1) 509 * @end_rgn: end scanning at (@end_rgn - 1) 510 * Scan @type and merge neighboring compatible regions in [@start_rgn - 1, @end_rgn) 511 */ 512 static void __init_memblock memblock_merge_regions(struct memblock_type *type, 513 unsigned long start_rgn, 514 unsigned long end_rgn) 515 { 516 int i = 0; 517 if (start_rgn) 518 i = start_rgn - 1; 519 end_rgn = min(end_rgn, type->cnt - 1); 520 while (i < end_rgn) { 521 struct memblock_region *this = &type->regions[i]; 522 struct memblock_region *next = &type->regions[i + 1]; 523 524 if (this->base + this->size != next->base || 525 memblock_get_region_node(this) != 526 memblock_get_region_node(next) || 527 this->flags != next->flags) { 528 BUG_ON(this->base + this->size > next->base); 529 i++; 530 continue; 531 } 532 533 this->size += next->size; 534 /* move forward from next + 1, index of which is i + 2 */ 535 memmove(next, next + 1, (type->cnt - (i + 2)) * sizeof(*next)); 536 type->cnt--; 537 end_rgn--; 538 } 539 } 540 541 /** 542 * memblock_insert_region - insert new memblock region 543 * @type: memblock type to insert into 544 * @idx: index for the insertion point 545 * @base: base address of the new region 546 * @size: size of the new region 547 * @nid: node id of the new region 548 * @flags: flags of the new region 549 * 550 * Insert new memblock region [@base, @base + @size) into @type at @idx. 551 * @type must already have extra room to accommodate the new region. 552 */ 553 static void __init_memblock memblock_insert_region(struct memblock_type *type, 554 int idx, phys_addr_t base, 555 phys_addr_t size, 556 int nid, 557 enum memblock_flags flags) 558 { 559 struct memblock_region *rgn = &type->regions[idx]; 560 561 BUG_ON(type->cnt >= type->max); 562 memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn)); 563 rgn->base = base; 564 rgn->size = size; 565 rgn->flags = flags; 566 memblock_set_region_node(rgn, nid); 567 type->cnt++; 568 type->total_size += size; 569 } 570 571 /** 572 * memblock_add_range - add new memblock region 573 * @type: memblock type to add new region into 574 * @base: base address of the new region 575 * @size: size of the new region 576 * @nid: nid of the new region 577 * @flags: flags of the new region 578 * 579 * Add new memblock region [@base, @base + @size) into @type. The new region 580 * is allowed to overlap with existing ones - overlaps don't affect already 581 * existing regions. @type is guaranteed to be minimal (all neighbouring 582 * compatible regions are merged) after the addition. 583 * 584 * Return: 585 * 0 on success, -errno on failure. 586 */ 587 static int __init_memblock memblock_add_range(struct memblock_type *type, 588 phys_addr_t base, phys_addr_t size, 589 int nid, enum memblock_flags flags) 590 { 591 bool insert = false; 592 phys_addr_t obase = base; 593 phys_addr_t end = base + memblock_cap_size(base, &size); 594 int idx, nr_new, start_rgn = -1, end_rgn; 595 struct memblock_region *rgn; 596 597 if (!size) 598 return 0; 599 600 /* special case for empty array */ 601 if (type->regions[0].size == 0) { 602 WARN_ON(type->cnt != 1 || type->total_size); 603 type->regions[0].base = base; 604 type->regions[0].size = size; 605 type->regions[0].flags = flags; 606 memblock_set_region_node(&type->regions[0], nid); 607 type->total_size = size; 608 return 0; 609 } 610 611 /* 612 * The worst case is when new range overlaps all existing regions, 613 * then we'll need type->cnt + 1 empty regions in @type. So if 614 * type->cnt * 2 + 1 is less than or equal to type->max, we know 615 * that there is enough empty regions in @type, and we can insert 616 * regions directly. 617 */ 618 if (type->cnt * 2 + 1 <= type->max) 619 insert = true; 620 621 repeat: 622 /* 623 * The following is executed twice. Once with %false @insert and 624 * then with %true. The first counts the number of regions needed 625 * to accommodate the new area. The second actually inserts them. 626 */ 627 base = obase; 628 nr_new = 0; 629 630 for_each_memblock_type(idx, type, rgn) { 631 phys_addr_t rbase = rgn->base; 632 phys_addr_t rend = rbase + rgn->size; 633 634 if (rbase >= end) 635 break; 636 if (rend <= base) 637 continue; 638 /* 639 * @rgn overlaps. If it separates the lower part of new 640 * area, insert that portion. 641 */ 642 if (rbase > base) { 643 #ifdef CONFIG_NUMA 644 WARN_ON(nid != memblock_get_region_node(rgn)); 645 #endif 646 WARN_ON(flags != rgn->flags); 647 nr_new++; 648 if (insert) { 649 if (start_rgn == -1) 650 start_rgn = idx; 651 end_rgn = idx + 1; 652 memblock_insert_region(type, idx++, base, 653 rbase - base, nid, 654 flags); 655 } 656 } 657 /* area below @rend is dealt with, forget about it */ 658 base = min(rend, end); 659 } 660 661 /* insert the remaining portion */ 662 if (base < end) { 663 nr_new++; 664 if (insert) { 665 if (start_rgn == -1) 666 start_rgn = idx; 667 end_rgn = idx + 1; 668 memblock_insert_region(type, idx, base, end - base, 669 nid, flags); 670 } 671 } 672 673 if (!nr_new) 674 return 0; 675 676 /* 677 * If this was the first round, resize array and repeat for actual 678 * insertions; otherwise, merge and return. 679 */ 680 if (!insert) { 681 while (type->cnt + nr_new > type->max) 682 if (memblock_double_array(type, obase, size) < 0) 683 return -ENOMEM; 684 insert = true; 685 goto repeat; 686 } else { 687 memblock_merge_regions(type, start_rgn, end_rgn); 688 return 0; 689 } 690 } 691 692 /** 693 * memblock_add_node - add new memblock region within a NUMA node 694 * @base: base address of the new region 695 * @size: size of the new region 696 * @nid: nid of the new region 697 * @flags: flags of the new region 698 * 699 * Add new memblock region [@base, @base + @size) to the "memory" 700 * type. See memblock_add_range() description for mode details 701 * 702 * Return: 703 * 0 on success, -errno on failure. 704 */ 705 int __init_memblock memblock_add_node(phys_addr_t base, phys_addr_t size, 706 int nid, enum memblock_flags flags) 707 { 708 phys_addr_t end = base + size - 1; 709 710 memblock_dbg("%s: [%pa-%pa] nid=%d flags=%x %pS\n", __func__, 711 &base, &end, nid, flags, (void *)_RET_IP_); 712 713 return memblock_add_range(&memblock.memory, base, size, nid, flags); 714 } 715 716 /** 717 * memblock_add - add new memblock region 718 * @base: base address of the new region 719 * @size: size of the new region 720 * 721 * Add new memblock region [@base, @base + @size) to the "memory" 722 * type. See memblock_add_range() description for mode details 723 * 724 * Return: 725 * 0 on success, -errno on failure. 726 */ 727 int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size) 728 { 729 phys_addr_t end = base + size - 1; 730 731 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__, 732 &base, &end, (void *)_RET_IP_); 733 734 return memblock_add_range(&memblock.memory, base, size, MAX_NUMNODES, 0); 735 } 736 737 /** 738 * memblock_isolate_range - isolate given range into disjoint memblocks 739 * @type: memblock type to isolate range for 740 * @base: base of range to isolate 741 * @size: size of range to isolate 742 * @start_rgn: out parameter for the start of isolated region 743 * @end_rgn: out parameter for the end of isolated region 744 * 745 * Walk @type and ensure that regions don't cross the boundaries defined by 746 * [@base, @base + @size). Crossing regions are split at the boundaries, 747 * which may create at most two more regions. The index of the first 748 * region inside the range is returned in *@start_rgn and end in *@end_rgn. 749 * 750 * Return: 751 * 0 on success, -errno on failure. 752 */ 753 static int __init_memblock memblock_isolate_range(struct memblock_type *type, 754 phys_addr_t base, phys_addr_t size, 755 int *start_rgn, int *end_rgn) 756 { 757 phys_addr_t end = base + memblock_cap_size(base, &size); 758 int idx; 759 struct memblock_region *rgn; 760 761 *start_rgn = *end_rgn = 0; 762 763 if (!size) 764 return 0; 765 766 /* we'll create at most two more regions */ 767 while (type->cnt + 2 > type->max) 768 if (memblock_double_array(type, base, size) < 0) 769 return -ENOMEM; 770 771 for_each_memblock_type(idx, type, rgn) { 772 phys_addr_t rbase = rgn->base; 773 phys_addr_t rend = rbase + rgn->size; 774 775 if (rbase >= end) 776 break; 777 if (rend <= base) 778 continue; 779 780 if (rbase < base) { 781 /* 782 * @rgn intersects from below. Split and continue 783 * to process the next region - the new top half. 784 */ 785 rgn->base = base; 786 rgn->size -= base - rbase; 787 type->total_size -= base - rbase; 788 memblock_insert_region(type, idx, rbase, base - rbase, 789 memblock_get_region_node(rgn), 790 rgn->flags); 791 } else if (rend > end) { 792 /* 793 * @rgn intersects from above. Split and redo the 794 * current region - the new bottom half. 795 */ 796 rgn->base = end; 797 rgn->size -= end - rbase; 798 type->total_size -= end - rbase; 799 memblock_insert_region(type, idx--, rbase, end - rbase, 800 memblock_get_region_node(rgn), 801 rgn->flags); 802 } else { 803 /* @rgn is fully contained, record it */ 804 if (!*end_rgn) 805 *start_rgn = idx; 806 *end_rgn = idx + 1; 807 } 808 } 809 810 return 0; 811 } 812 813 static int __init_memblock memblock_remove_range(struct memblock_type *type, 814 phys_addr_t base, phys_addr_t size) 815 { 816 int start_rgn, end_rgn; 817 int i, ret; 818 819 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn); 820 if (ret) 821 return ret; 822 823 for (i = end_rgn - 1; i >= start_rgn; i--) 824 memblock_remove_region(type, i); 825 return 0; 826 } 827 828 int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size) 829 { 830 phys_addr_t end = base + size - 1; 831 832 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__, 833 &base, &end, (void *)_RET_IP_); 834 835 return memblock_remove_range(&memblock.memory, base, size); 836 } 837 838 /** 839 * memblock_free - free boot memory allocation 840 * @ptr: starting address of the boot memory allocation 841 * @size: size of the boot memory block in bytes 842 * 843 * Free boot memory block previously allocated by memblock_alloc_xx() API. 844 * The freeing memory will not be released to the buddy allocator. 845 */ 846 void __init_memblock memblock_free(void *ptr, size_t size) 847 { 848 if (ptr) 849 memblock_phys_free(__pa(ptr), size); 850 } 851 852 /** 853 * memblock_phys_free - free boot memory block 854 * @base: phys starting address of the boot memory block 855 * @size: size of the boot memory block in bytes 856 * 857 * Free boot memory block previously allocated by memblock_phys_alloc_xx() API. 858 * The freeing memory will not be released to the buddy allocator. 859 */ 860 int __init_memblock memblock_phys_free(phys_addr_t base, phys_addr_t size) 861 { 862 phys_addr_t end = base + size - 1; 863 864 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__, 865 &base, &end, (void *)_RET_IP_); 866 867 kmemleak_free_part_phys(base, size); 868 return memblock_remove_range(&memblock.reserved, base, size); 869 } 870 871 int __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size) 872 { 873 phys_addr_t end = base + size - 1; 874 875 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__, 876 &base, &end, (void *)_RET_IP_); 877 878 return memblock_add_range(&memblock.reserved, base, size, MAX_NUMNODES, 0); 879 } 880 881 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP 882 int __init_memblock memblock_physmem_add(phys_addr_t base, phys_addr_t size) 883 { 884 phys_addr_t end = base + size - 1; 885 886 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__, 887 &base, &end, (void *)_RET_IP_); 888 889 return memblock_add_range(&physmem, base, size, MAX_NUMNODES, 0); 890 } 891 #endif 892 893 /** 894 * memblock_setclr_flag - set or clear flag for a memory region 895 * @base: base address of the region 896 * @size: size of the region 897 * @set: set or clear the flag 898 * @flag: the flag to update 899 * 900 * This function isolates region [@base, @base + @size), and sets/clears flag 901 * 902 * Return: 0 on success, -errno on failure. 903 */ 904 static int __init_memblock memblock_setclr_flag(phys_addr_t base, 905 phys_addr_t size, int set, int flag) 906 { 907 struct memblock_type *type = &memblock.memory; 908 int i, ret, start_rgn, end_rgn; 909 910 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn); 911 if (ret) 912 return ret; 913 914 for (i = start_rgn; i < end_rgn; i++) { 915 struct memblock_region *r = &type->regions[i]; 916 917 if (set) 918 r->flags |= flag; 919 else 920 r->flags &= ~flag; 921 } 922 923 memblock_merge_regions(type, start_rgn, end_rgn); 924 return 0; 925 } 926 927 /** 928 * memblock_mark_hotplug - Mark hotpluggable memory with flag MEMBLOCK_HOTPLUG. 929 * @base: the base phys addr of the region 930 * @size: the size of the region 931 * 932 * Return: 0 on success, -errno on failure. 933 */ 934 int __init_memblock memblock_mark_hotplug(phys_addr_t base, phys_addr_t size) 935 { 936 return memblock_setclr_flag(base, size, 1, MEMBLOCK_HOTPLUG); 937 } 938 939 /** 940 * memblock_clear_hotplug - Clear flag MEMBLOCK_HOTPLUG for a specified region. 941 * @base: the base phys addr of the region 942 * @size: the size of the region 943 * 944 * Return: 0 on success, -errno on failure. 945 */ 946 int __init_memblock memblock_clear_hotplug(phys_addr_t base, phys_addr_t size) 947 { 948 return memblock_setclr_flag(base, size, 0, MEMBLOCK_HOTPLUG); 949 } 950 951 /** 952 * memblock_mark_mirror - Mark mirrored memory with flag MEMBLOCK_MIRROR. 953 * @base: the base phys addr of the region 954 * @size: the size of the region 955 * 956 * Return: 0 on success, -errno on failure. 957 */ 958 int __init_memblock memblock_mark_mirror(phys_addr_t base, phys_addr_t size) 959 { 960 if (!mirrored_kernelcore) 961 return 0; 962 963 system_has_some_mirror = true; 964 965 return memblock_setclr_flag(base, size, 1, MEMBLOCK_MIRROR); 966 } 967 968 /** 969 * memblock_mark_nomap - Mark a memory region with flag MEMBLOCK_NOMAP. 970 * @base: the base phys addr of the region 971 * @size: the size of the region 972 * 973 * The memory regions marked with %MEMBLOCK_NOMAP will not be added to the 974 * direct mapping of the physical memory. These regions will still be 975 * covered by the memory map. The struct page representing NOMAP memory 976 * frames in the memory map will be PageReserved() 977 * 978 * Note: if the memory being marked %MEMBLOCK_NOMAP was allocated from 979 * memblock, the caller must inform kmemleak to ignore that memory 980 * 981 * Return: 0 on success, -errno on failure. 982 */ 983 int __init_memblock memblock_mark_nomap(phys_addr_t base, phys_addr_t size) 984 { 985 return memblock_setclr_flag(base, size, 1, MEMBLOCK_NOMAP); 986 } 987 988 /** 989 * memblock_clear_nomap - Clear flag MEMBLOCK_NOMAP for a specified region. 990 * @base: the base phys addr of the region 991 * @size: the size of the region 992 * 993 * Return: 0 on success, -errno on failure. 994 */ 995 int __init_memblock memblock_clear_nomap(phys_addr_t base, phys_addr_t size) 996 { 997 return memblock_setclr_flag(base, size, 0, MEMBLOCK_NOMAP); 998 } 999 1000 static bool should_skip_region(struct memblock_type *type, 1001 struct memblock_region *m, 1002 int nid, int flags) 1003 { 1004 int m_nid = memblock_get_region_node(m); 1005 1006 /* we never skip regions when iterating memblock.reserved or physmem */ 1007 if (type != memblock_memory) 1008 return false; 1009 1010 /* only memory regions are associated with nodes, check it */ 1011 if (nid != NUMA_NO_NODE && nid != m_nid) 1012 return true; 1013 1014 /* skip hotpluggable memory regions if needed */ 1015 if (movable_node_is_enabled() && memblock_is_hotpluggable(m) && 1016 !(flags & MEMBLOCK_HOTPLUG)) 1017 return true; 1018 1019 /* if we want mirror memory skip non-mirror memory regions */ 1020 if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m)) 1021 return true; 1022 1023 /* skip nomap memory unless we were asked for it explicitly */ 1024 if (!(flags & MEMBLOCK_NOMAP) && memblock_is_nomap(m)) 1025 return true; 1026 1027 /* skip driver-managed memory unless we were asked for it explicitly */ 1028 if (!(flags & MEMBLOCK_DRIVER_MANAGED) && memblock_is_driver_managed(m)) 1029 return true; 1030 1031 return false; 1032 } 1033 1034 /** 1035 * __next_mem_range - next function for for_each_free_mem_range() etc. 1036 * @idx: pointer to u64 loop variable 1037 * @nid: node selector, %NUMA_NO_NODE for all nodes 1038 * @flags: pick from blocks based on memory attributes 1039 * @type_a: pointer to memblock_type from where the range is taken 1040 * @type_b: pointer to memblock_type which excludes memory from being taken 1041 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL 1042 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL 1043 * @out_nid: ptr to int for nid of the range, can be %NULL 1044 * 1045 * Find the first area from *@idx which matches @nid, fill the out 1046 * parameters, and update *@idx for the next iteration. The lower 32bit of 1047 * *@idx contains index into type_a and the upper 32bit indexes the 1048 * areas before each region in type_b. For example, if type_b regions 1049 * look like the following, 1050 * 1051 * 0:[0-16), 1:[32-48), 2:[128-130) 1052 * 1053 * The upper 32bit indexes the following regions. 1054 * 1055 * 0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX) 1056 * 1057 * As both region arrays are sorted, the function advances the two indices 1058 * in lockstep and returns each intersection. 1059 */ 1060 void __next_mem_range(u64 *idx, int nid, enum memblock_flags flags, 1061 struct memblock_type *type_a, 1062 struct memblock_type *type_b, phys_addr_t *out_start, 1063 phys_addr_t *out_end, int *out_nid) 1064 { 1065 int idx_a = *idx & 0xffffffff; 1066 int idx_b = *idx >> 32; 1067 1068 if (WARN_ONCE(nid == MAX_NUMNODES, 1069 "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n")) 1070 nid = NUMA_NO_NODE; 1071 1072 for (; idx_a < type_a->cnt; idx_a++) { 1073 struct memblock_region *m = &type_a->regions[idx_a]; 1074 1075 phys_addr_t m_start = m->base; 1076 phys_addr_t m_end = m->base + m->size; 1077 int m_nid = memblock_get_region_node(m); 1078 1079 if (should_skip_region(type_a, m, nid, flags)) 1080 continue; 1081 1082 if (!type_b) { 1083 if (out_start) 1084 *out_start = m_start; 1085 if (out_end) 1086 *out_end = m_end; 1087 if (out_nid) 1088 *out_nid = m_nid; 1089 idx_a++; 1090 *idx = (u32)idx_a | (u64)idx_b << 32; 1091 return; 1092 } 1093 1094 /* scan areas before each reservation */ 1095 for (; idx_b < type_b->cnt + 1; idx_b++) { 1096 struct memblock_region *r; 1097 phys_addr_t r_start; 1098 phys_addr_t r_end; 1099 1100 r = &type_b->regions[idx_b]; 1101 r_start = idx_b ? r[-1].base + r[-1].size : 0; 1102 r_end = idx_b < type_b->cnt ? 1103 r->base : PHYS_ADDR_MAX; 1104 1105 /* 1106 * if idx_b advanced past idx_a, 1107 * break out to advance idx_a 1108 */ 1109 if (r_start >= m_end) 1110 break; 1111 /* if the two regions intersect, we're done */ 1112 if (m_start < r_end) { 1113 if (out_start) 1114 *out_start = 1115 max(m_start, r_start); 1116 if (out_end) 1117 *out_end = min(m_end, r_end); 1118 if (out_nid) 1119 *out_nid = m_nid; 1120 /* 1121 * The region which ends first is 1122 * advanced for the next iteration. 1123 */ 1124 if (m_end <= r_end) 1125 idx_a++; 1126 else 1127 idx_b++; 1128 *idx = (u32)idx_a | (u64)idx_b << 32; 1129 return; 1130 } 1131 } 1132 } 1133 1134 /* signal end of iteration */ 1135 *idx = ULLONG_MAX; 1136 } 1137 1138 /** 1139 * __next_mem_range_rev - generic next function for for_each_*_range_rev() 1140 * 1141 * @idx: pointer to u64 loop variable 1142 * @nid: node selector, %NUMA_NO_NODE for all nodes 1143 * @flags: pick from blocks based on memory attributes 1144 * @type_a: pointer to memblock_type from where the range is taken 1145 * @type_b: pointer to memblock_type which excludes memory from being taken 1146 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL 1147 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL 1148 * @out_nid: ptr to int for nid of the range, can be %NULL 1149 * 1150 * Finds the next range from type_a which is not marked as unsuitable 1151 * in type_b. 1152 * 1153 * Reverse of __next_mem_range(). 1154 */ 1155 void __init_memblock __next_mem_range_rev(u64 *idx, int nid, 1156 enum memblock_flags flags, 1157 struct memblock_type *type_a, 1158 struct memblock_type *type_b, 1159 phys_addr_t *out_start, 1160 phys_addr_t *out_end, int *out_nid) 1161 { 1162 int idx_a = *idx & 0xffffffff; 1163 int idx_b = *idx >> 32; 1164 1165 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n")) 1166 nid = NUMA_NO_NODE; 1167 1168 if (*idx == (u64)ULLONG_MAX) { 1169 idx_a = type_a->cnt - 1; 1170 if (type_b != NULL) 1171 idx_b = type_b->cnt; 1172 else 1173 idx_b = 0; 1174 } 1175 1176 for (; idx_a >= 0; idx_a--) { 1177 struct memblock_region *m = &type_a->regions[idx_a]; 1178 1179 phys_addr_t m_start = m->base; 1180 phys_addr_t m_end = m->base + m->size; 1181 int m_nid = memblock_get_region_node(m); 1182 1183 if (should_skip_region(type_a, m, nid, flags)) 1184 continue; 1185 1186 if (!type_b) { 1187 if (out_start) 1188 *out_start = m_start; 1189 if (out_end) 1190 *out_end = m_end; 1191 if (out_nid) 1192 *out_nid = m_nid; 1193 idx_a--; 1194 *idx = (u32)idx_a | (u64)idx_b << 32; 1195 return; 1196 } 1197 1198 /* scan areas before each reservation */ 1199 for (; idx_b >= 0; idx_b--) { 1200 struct memblock_region *r; 1201 phys_addr_t r_start; 1202 phys_addr_t r_end; 1203 1204 r = &type_b->regions[idx_b]; 1205 r_start = idx_b ? r[-1].base + r[-1].size : 0; 1206 r_end = idx_b < type_b->cnt ? 1207 r->base : PHYS_ADDR_MAX; 1208 /* 1209 * if idx_b advanced past idx_a, 1210 * break out to advance idx_a 1211 */ 1212 1213 if (r_end <= m_start) 1214 break; 1215 /* if the two regions intersect, we're done */ 1216 if (m_end > r_start) { 1217 if (out_start) 1218 *out_start = max(m_start, r_start); 1219 if (out_end) 1220 *out_end = min(m_end, r_end); 1221 if (out_nid) 1222 *out_nid = m_nid; 1223 if (m_start >= r_start) 1224 idx_a--; 1225 else 1226 idx_b--; 1227 *idx = (u32)idx_a | (u64)idx_b << 32; 1228 return; 1229 } 1230 } 1231 } 1232 /* signal end of iteration */ 1233 *idx = ULLONG_MAX; 1234 } 1235 1236 /* 1237 * Common iterator interface used to define for_each_mem_pfn_range(). 1238 */ 1239 void __init_memblock __next_mem_pfn_range(int *idx, int nid, 1240 unsigned long *out_start_pfn, 1241 unsigned long *out_end_pfn, int *out_nid) 1242 { 1243 struct memblock_type *type = &memblock.memory; 1244 struct memblock_region *r; 1245 int r_nid; 1246 1247 while (++*idx < type->cnt) { 1248 r = &type->regions[*idx]; 1249 r_nid = memblock_get_region_node(r); 1250 1251 if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size)) 1252 continue; 1253 if (nid == MAX_NUMNODES || nid == r_nid) 1254 break; 1255 } 1256 if (*idx >= type->cnt) { 1257 *idx = -1; 1258 return; 1259 } 1260 1261 if (out_start_pfn) 1262 *out_start_pfn = PFN_UP(r->base); 1263 if (out_end_pfn) 1264 *out_end_pfn = PFN_DOWN(r->base + r->size); 1265 if (out_nid) 1266 *out_nid = r_nid; 1267 } 1268 1269 /** 1270 * memblock_set_node - set node ID on memblock regions 1271 * @base: base of area to set node ID for 1272 * @size: size of area to set node ID for 1273 * @type: memblock type to set node ID for 1274 * @nid: node ID to set 1275 * 1276 * Set the nid of memblock @type regions in [@base, @base + @size) to @nid. 1277 * Regions which cross the area boundaries are split as necessary. 1278 * 1279 * Return: 1280 * 0 on success, -errno on failure. 1281 */ 1282 int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size, 1283 struct memblock_type *type, int nid) 1284 { 1285 #ifdef CONFIG_NUMA 1286 int start_rgn, end_rgn; 1287 int i, ret; 1288 1289 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn); 1290 if (ret) 1291 return ret; 1292 1293 for (i = start_rgn; i < end_rgn; i++) 1294 memblock_set_region_node(&type->regions[i], nid); 1295 1296 memblock_merge_regions(type, start_rgn, end_rgn); 1297 #endif 1298 return 0; 1299 } 1300 1301 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT 1302 /** 1303 * __next_mem_pfn_range_in_zone - iterator for for_each_*_range_in_zone() 1304 * 1305 * @idx: pointer to u64 loop variable 1306 * @zone: zone in which all of the memory blocks reside 1307 * @out_spfn: ptr to ulong for start pfn of the range, can be %NULL 1308 * @out_epfn: ptr to ulong for end pfn of the range, can be %NULL 1309 * 1310 * This function is meant to be a zone/pfn specific wrapper for the 1311 * for_each_mem_range type iterators. Specifically they are used in the 1312 * deferred memory init routines and as such we were duplicating much of 1313 * this logic throughout the code. So instead of having it in multiple 1314 * locations it seemed like it would make more sense to centralize this to 1315 * one new iterator that does everything they need. 1316 */ 1317 void __init_memblock 1318 __next_mem_pfn_range_in_zone(u64 *idx, struct zone *zone, 1319 unsigned long *out_spfn, unsigned long *out_epfn) 1320 { 1321 int zone_nid = zone_to_nid(zone); 1322 phys_addr_t spa, epa; 1323 1324 __next_mem_range(idx, zone_nid, MEMBLOCK_NONE, 1325 &memblock.memory, &memblock.reserved, 1326 &spa, &epa, NULL); 1327 1328 while (*idx != U64_MAX) { 1329 unsigned long epfn = PFN_DOWN(epa); 1330 unsigned long spfn = PFN_UP(spa); 1331 1332 /* 1333 * Verify the end is at least past the start of the zone and 1334 * that we have at least one PFN to initialize. 1335 */ 1336 if (zone->zone_start_pfn < epfn && spfn < epfn) { 1337 /* if we went too far just stop searching */ 1338 if (zone_end_pfn(zone) <= spfn) { 1339 *idx = U64_MAX; 1340 break; 1341 } 1342 1343 if (out_spfn) 1344 *out_spfn = max(zone->zone_start_pfn, spfn); 1345 if (out_epfn) 1346 *out_epfn = min(zone_end_pfn(zone), epfn); 1347 1348 return; 1349 } 1350 1351 __next_mem_range(idx, zone_nid, MEMBLOCK_NONE, 1352 &memblock.memory, &memblock.reserved, 1353 &spa, &epa, NULL); 1354 } 1355 1356 /* signal end of iteration */ 1357 if (out_spfn) 1358 *out_spfn = ULONG_MAX; 1359 if (out_epfn) 1360 *out_epfn = 0; 1361 } 1362 1363 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ 1364 1365 /** 1366 * memblock_alloc_range_nid - allocate boot memory block 1367 * @size: size of memory block to be allocated in bytes 1368 * @align: alignment of the region and block's size 1369 * @start: the lower bound of the memory region to allocate (phys address) 1370 * @end: the upper bound of the memory region to allocate (phys address) 1371 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 1372 * @exact_nid: control the allocation fall back to other nodes 1373 * 1374 * The allocation is performed from memory region limited by 1375 * memblock.current_limit if @end == %MEMBLOCK_ALLOC_ACCESSIBLE. 1376 * 1377 * If the specified node can not hold the requested memory and @exact_nid 1378 * is false, the allocation falls back to any node in the system. 1379 * 1380 * For systems with memory mirroring, the allocation is attempted first 1381 * from the regions with mirroring enabled and then retried from any 1382 * memory region. 1383 * 1384 * In addition, function using kmemleak_alloc_phys for allocated boot 1385 * memory block, it is never reported as leaks. 1386 * 1387 * Return: 1388 * Physical address of allocated memory block on success, %0 on failure. 1389 */ 1390 phys_addr_t __init memblock_alloc_range_nid(phys_addr_t size, 1391 phys_addr_t align, phys_addr_t start, 1392 phys_addr_t end, int nid, 1393 bool exact_nid) 1394 { 1395 enum memblock_flags flags = choose_memblock_flags(); 1396 phys_addr_t found; 1397 1398 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n")) 1399 nid = NUMA_NO_NODE; 1400 1401 if (!align) { 1402 /* Can't use WARNs this early in boot on powerpc */ 1403 dump_stack(); 1404 align = SMP_CACHE_BYTES; 1405 } 1406 1407 again: 1408 found = memblock_find_in_range_node(size, align, start, end, nid, 1409 flags); 1410 if (found && !memblock_reserve(found, size)) 1411 goto done; 1412 1413 if (nid != NUMA_NO_NODE && !exact_nid) { 1414 found = memblock_find_in_range_node(size, align, start, 1415 end, NUMA_NO_NODE, 1416 flags); 1417 if (found && !memblock_reserve(found, size)) 1418 goto done; 1419 } 1420 1421 if (flags & MEMBLOCK_MIRROR) { 1422 flags &= ~MEMBLOCK_MIRROR; 1423 pr_warn_ratelimited("Could not allocate %pap bytes of mirrored memory\n", 1424 &size); 1425 goto again; 1426 } 1427 1428 return 0; 1429 1430 done: 1431 /* 1432 * Skip kmemleak for those places like kasan_init() and 1433 * early_pgtable_alloc() due to high volume. 1434 */ 1435 if (end != MEMBLOCK_ALLOC_NOLEAKTRACE) 1436 /* 1437 * Memblock allocated blocks are never reported as 1438 * leaks. This is because many of these blocks are 1439 * only referred via the physical address which is 1440 * not looked up by kmemleak. 1441 */ 1442 kmemleak_alloc_phys(found, size, 0); 1443 1444 /* 1445 * Some Virtual Machine platforms, such as Intel TDX or AMD SEV-SNP, 1446 * require memory to be accepted before it can be used by the 1447 * guest. 1448 * 1449 * Accept the memory of the allocated buffer. 1450 */ 1451 accept_memory(found, found + size); 1452 1453 return found; 1454 } 1455 1456 /** 1457 * memblock_phys_alloc_range - allocate a memory block inside specified range 1458 * @size: size of memory block to be allocated in bytes 1459 * @align: alignment of the region and block's size 1460 * @start: the lower bound of the memory region to allocate (physical address) 1461 * @end: the upper bound of the memory region to allocate (physical address) 1462 * 1463 * Allocate @size bytes in the between @start and @end. 1464 * 1465 * Return: physical address of the allocated memory block on success, 1466 * %0 on failure. 1467 */ 1468 phys_addr_t __init memblock_phys_alloc_range(phys_addr_t size, 1469 phys_addr_t align, 1470 phys_addr_t start, 1471 phys_addr_t end) 1472 { 1473 memblock_dbg("%s: %llu bytes align=0x%llx from=%pa max_addr=%pa %pS\n", 1474 __func__, (u64)size, (u64)align, &start, &end, 1475 (void *)_RET_IP_); 1476 return memblock_alloc_range_nid(size, align, start, end, NUMA_NO_NODE, 1477 false); 1478 } 1479 1480 /** 1481 * memblock_phys_alloc_try_nid - allocate a memory block from specified NUMA node 1482 * @size: size of memory block to be allocated in bytes 1483 * @align: alignment of the region and block's size 1484 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 1485 * 1486 * Allocates memory block from the specified NUMA node. If the node 1487 * has no available memory, attempts to allocated from any node in the 1488 * system. 1489 * 1490 * Return: physical address of the allocated memory block on success, 1491 * %0 on failure. 1492 */ 1493 phys_addr_t __init memblock_phys_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid) 1494 { 1495 return memblock_alloc_range_nid(size, align, 0, 1496 MEMBLOCK_ALLOC_ACCESSIBLE, nid, false); 1497 } 1498 1499 /** 1500 * memblock_alloc_internal - allocate boot memory block 1501 * @size: size of memory block to be allocated in bytes 1502 * @align: alignment of the region and block's size 1503 * @min_addr: the lower bound of the memory region to allocate (phys address) 1504 * @max_addr: the upper bound of the memory region to allocate (phys address) 1505 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 1506 * @exact_nid: control the allocation fall back to other nodes 1507 * 1508 * Allocates memory block using memblock_alloc_range_nid() and 1509 * converts the returned physical address to virtual. 1510 * 1511 * The @min_addr limit is dropped if it can not be satisfied and the allocation 1512 * will fall back to memory below @min_addr. Other constraints, such 1513 * as node and mirrored memory will be handled again in 1514 * memblock_alloc_range_nid(). 1515 * 1516 * Return: 1517 * Virtual address of allocated memory block on success, NULL on failure. 1518 */ 1519 static void * __init memblock_alloc_internal( 1520 phys_addr_t size, phys_addr_t align, 1521 phys_addr_t min_addr, phys_addr_t max_addr, 1522 int nid, bool exact_nid) 1523 { 1524 phys_addr_t alloc; 1525 1526 /* 1527 * Detect any accidental use of these APIs after slab is ready, as at 1528 * this moment memblock may be deinitialized already and its 1529 * internal data may be destroyed (after execution of memblock_free_all) 1530 */ 1531 if (WARN_ON_ONCE(slab_is_available())) 1532 return kzalloc_node(size, GFP_NOWAIT, nid); 1533 1534 if (max_addr > memblock.current_limit) 1535 max_addr = memblock.current_limit; 1536 1537 alloc = memblock_alloc_range_nid(size, align, min_addr, max_addr, nid, 1538 exact_nid); 1539 1540 /* retry allocation without lower limit */ 1541 if (!alloc && min_addr) 1542 alloc = memblock_alloc_range_nid(size, align, 0, max_addr, nid, 1543 exact_nid); 1544 1545 if (!alloc) 1546 return NULL; 1547 1548 return phys_to_virt(alloc); 1549 } 1550 1551 /** 1552 * memblock_alloc_exact_nid_raw - allocate boot memory block on the exact node 1553 * without zeroing memory 1554 * @size: size of memory block to be allocated in bytes 1555 * @align: alignment of the region and block's size 1556 * @min_addr: the lower bound of the memory region from where the allocation 1557 * is preferred (phys address) 1558 * @max_addr: the upper bound of the memory region from where the allocation 1559 * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to 1560 * allocate only from memory limited by memblock.current_limit value 1561 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 1562 * 1563 * Public function, provides additional debug information (including caller 1564 * info), if enabled. Does not zero allocated memory. 1565 * 1566 * Return: 1567 * Virtual address of allocated memory block on success, NULL on failure. 1568 */ 1569 void * __init memblock_alloc_exact_nid_raw( 1570 phys_addr_t size, phys_addr_t align, 1571 phys_addr_t min_addr, phys_addr_t max_addr, 1572 int nid) 1573 { 1574 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n", 1575 __func__, (u64)size, (u64)align, nid, &min_addr, 1576 &max_addr, (void *)_RET_IP_); 1577 1578 return memblock_alloc_internal(size, align, min_addr, max_addr, nid, 1579 true); 1580 } 1581 1582 /** 1583 * memblock_alloc_try_nid_raw - allocate boot memory block without zeroing 1584 * memory and without panicking 1585 * @size: size of memory block to be allocated in bytes 1586 * @align: alignment of the region and block's size 1587 * @min_addr: the lower bound of the memory region from where the allocation 1588 * is preferred (phys address) 1589 * @max_addr: the upper bound of the memory region from where the allocation 1590 * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to 1591 * allocate only from memory limited by memblock.current_limit value 1592 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 1593 * 1594 * Public function, provides additional debug information (including caller 1595 * info), if enabled. Does not zero allocated memory, does not panic if request 1596 * cannot be satisfied. 1597 * 1598 * Return: 1599 * Virtual address of allocated memory block on success, NULL on failure. 1600 */ 1601 void * __init memblock_alloc_try_nid_raw( 1602 phys_addr_t size, phys_addr_t align, 1603 phys_addr_t min_addr, phys_addr_t max_addr, 1604 int nid) 1605 { 1606 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n", 1607 __func__, (u64)size, (u64)align, nid, &min_addr, 1608 &max_addr, (void *)_RET_IP_); 1609 1610 return memblock_alloc_internal(size, align, min_addr, max_addr, nid, 1611 false); 1612 } 1613 1614 /** 1615 * memblock_alloc_try_nid - allocate boot memory block 1616 * @size: size of memory block to be allocated in bytes 1617 * @align: alignment of the region and block's size 1618 * @min_addr: the lower bound of the memory region from where the allocation 1619 * is preferred (phys address) 1620 * @max_addr: the upper bound of the memory region from where the allocation 1621 * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to 1622 * allocate only from memory limited by memblock.current_limit value 1623 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 1624 * 1625 * Public function, provides additional debug information (including caller 1626 * info), if enabled. This function zeroes the allocated memory. 1627 * 1628 * Return: 1629 * Virtual address of allocated memory block on success, NULL on failure. 1630 */ 1631 void * __init memblock_alloc_try_nid( 1632 phys_addr_t size, phys_addr_t align, 1633 phys_addr_t min_addr, phys_addr_t max_addr, 1634 int nid) 1635 { 1636 void *ptr; 1637 1638 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n", 1639 __func__, (u64)size, (u64)align, nid, &min_addr, 1640 &max_addr, (void *)_RET_IP_); 1641 ptr = memblock_alloc_internal(size, align, 1642 min_addr, max_addr, nid, false); 1643 if (ptr) 1644 memset(ptr, 0, size); 1645 1646 return ptr; 1647 } 1648 1649 /** 1650 * memblock_free_late - free pages directly to buddy allocator 1651 * @base: phys starting address of the boot memory block 1652 * @size: size of the boot memory block in bytes 1653 * 1654 * This is only useful when the memblock allocator has already been torn 1655 * down, but we are still initializing the system. Pages are released directly 1656 * to the buddy allocator. 1657 */ 1658 void __init memblock_free_late(phys_addr_t base, phys_addr_t size) 1659 { 1660 phys_addr_t cursor, end; 1661 1662 end = base + size - 1; 1663 memblock_dbg("%s: [%pa-%pa] %pS\n", 1664 __func__, &base, &end, (void *)_RET_IP_); 1665 kmemleak_free_part_phys(base, size); 1666 cursor = PFN_UP(base); 1667 end = PFN_DOWN(base + size); 1668 1669 for (; cursor < end; cursor++) { 1670 memblock_free_pages(pfn_to_page(cursor), cursor, 0); 1671 totalram_pages_inc(); 1672 } 1673 } 1674 1675 /* 1676 * Remaining API functions 1677 */ 1678 1679 phys_addr_t __init_memblock memblock_phys_mem_size(void) 1680 { 1681 return memblock.memory.total_size; 1682 } 1683 1684 phys_addr_t __init_memblock memblock_reserved_size(void) 1685 { 1686 return memblock.reserved.total_size; 1687 } 1688 1689 /* lowest address */ 1690 phys_addr_t __init_memblock memblock_start_of_DRAM(void) 1691 { 1692 return memblock.memory.regions[0].base; 1693 } 1694 1695 phys_addr_t __init_memblock memblock_end_of_DRAM(void) 1696 { 1697 int idx = memblock.memory.cnt - 1; 1698 1699 return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size); 1700 } 1701 1702 static phys_addr_t __init_memblock __find_max_addr(phys_addr_t limit) 1703 { 1704 phys_addr_t max_addr = PHYS_ADDR_MAX; 1705 struct memblock_region *r; 1706 1707 /* 1708 * translate the memory @limit size into the max address within one of 1709 * the memory memblock regions, if the @limit exceeds the total size 1710 * of those regions, max_addr will keep original value PHYS_ADDR_MAX 1711 */ 1712 for_each_mem_region(r) { 1713 if (limit <= r->size) { 1714 max_addr = r->base + limit; 1715 break; 1716 } 1717 limit -= r->size; 1718 } 1719 1720 return max_addr; 1721 } 1722 1723 void __init memblock_enforce_memory_limit(phys_addr_t limit) 1724 { 1725 phys_addr_t max_addr; 1726 1727 if (!limit) 1728 return; 1729 1730 max_addr = __find_max_addr(limit); 1731 1732 /* @limit exceeds the total size of the memory, do nothing */ 1733 if (max_addr == PHYS_ADDR_MAX) 1734 return; 1735 1736 /* truncate both memory and reserved regions */ 1737 memblock_remove_range(&memblock.memory, max_addr, 1738 PHYS_ADDR_MAX); 1739 memblock_remove_range(&memblock.reserved, max_addr, 1740 PHYS_ADDR_MAX); 1741 } 1742 1743 void __init memblock_cap_memory_range(phys_addr_t base, phys_addr_t size) 1744 { 1745 int start_rgn, end_rgn; 1746 int i, ret; 1747 1748 if (!size) 1749 return; 1750 1751 if (!memblock_memory->total_size) { 1752 pr_warn("%s: No memory registered yet\n", __func__); 1753 return; 1754 } 1755 1756 ret = memblock_isolate_range(&memblock.memory, base, size, 1757 &start_rgn, &end_rgn); 1758 if (ret) 1759 return; 1760 1761 /* remove all the MAP regions */ 1762 for (i = memblock.memory.cnt - 1; i >= end_rgn; i--) 1763 if (!memblock_is_nomap(&memblock.memory.regions[i])) 1764 memblock_remove_region(&memblock.memory, i); 1765 1766 for (i = start_rgn - 1; i >= 0; i--) 1767 if (!memblock_is_nomap(&memblock.memory.regions[i])) 1768 memblock_remove_region(&memblock.memory, i); 1769 1770 /* truncate the reserved regions */ 1771 memblock_remove_range(&memblock.reserved, 0, base); 1772 memblock_remove_range(&memblock.reserved, 1773 base + size, PHYS_ADDR_MAX); 1774 } 1775 1776 void __init memblock_mem_limit_remove_map(phys_addr_t limit) 1777 { 1778 phys_addr_t max_addr; 1779 1780 if (!limit) 1781 return; 1782 1783 max_addr = __find_max_addr(limit); 1784 1785 /* @limit exceeds the total size of the memory, do nothing */ 1786 if (max_addr == PHYS_ADDR_MAX) 1787 return; 1788 1789 memblock_cap_memory_range(0, max_addr); 1790 } 1791 1792 static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr) 1793 { 1794 unsigned int left = 0, right = type->cnt; 1795 1796 do { 1797 unsigned int mid = (right + left) / 2; 1798 1799 if (addr < type->regions[mid].base) 1800 right = mid; 1801 else if (addr >= (type->regions[mid].base + 1802 type->regions[mid].size)) 1803 left = mid + 1; 1804 else 1805 return mid; 1806 } while (left < right); 1807 return -1; 1808 } 1809 1810 bool __init_memblock memblock_is_reserved(phys_addr_t addr) 1811 { 1812 return memblock_search(&memblock.reserved, addr) != -1; 1813 } 1814 1815 bool __init_memblock memblock_is_memory(phys_addr_t addr) 1816 { 1817 return memblock_search(&memblock.memory, addr) != -1; 1818 } 1819 1820 bool __init_memblock memblock_is_map_memory(phys_addr_t addr) 1821 { 1822 int i = memblock_search(&memblock.memory, addr); 1823 1824 if (i == -1) 1825 return false; 1826 return !memblock_is_nomap(&memblock.memory.regions[i]); 1827 } 1828 1829 int __init_memblock memblock_search_pfn_nid(unsigned long pfn, 1830 unsigned long *start_pfn, unsigned long *end_pfn) 1831 { 1832 struct memblock_type *type = &memblock.memory; 1833 int mid = memblock_search(type, PFN_PHYS(pfn)); 1834 1835 if (mid == -1) 1836 return -1; 1837 1838 *start_pfn = PFN_DOWN(type->regions[mid].base); 1839 *end_pfn = PFN_DOWN(type->regions[mid].base + type->regions[mid].size); 1840 1841 return memblock_get_region_node(&type->regions[mid]); 1842 } 1843 1844 /** 1845 * memblock_is_region_memory - check if a region is a subset of memory 1846 * @base: base of region to check 1847 * @size: size of region to check 1848 * 1849 * Check if the region [@base, @base + @size) is a subset of a memory block. 1850 * 1851 * Return: 1852 * 0 if false, non-zero if true 1853 */ 1854 bool __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size) 1855 { 1856 int idx = memblock_search(&memblock.memory, base); 1857 phys_addr_t end = base + memblock_cap_size(base, &size); 1858 1859 if (idx == -1) 1860 return false; 1861 return (memblock.memory.regions[idx].base + 1862 memblock.memory.regions[idx].size) >= end; 1863 } 1864 1865 /** 1866 * memblock_is_region_reserved - check if a region intersects reserved memory 1867 * @base: base of region to check 1868 * @size: size of region to check 1869 * 1870 * Check if the region [@base, @base + @size) intersects a reserved 1871 * memory block. 1872 * 1873 * Return: 1874 * True if they intersect, false if not. 1875 */ 1876 bool __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size) 1877 { 1878 return memblock_overlaps_region(&memblock.reserved, base, size); 1879 } 1880 1881 void __init_memblock memblock_trim_memory(phys_addr_t align) 1882 { 1883 phys_addr_t start, end, orig_start, orig_end; 1884 struct memblock_region *r; 1885 1886 for_each_mem_region(r) { 1887 orig_start = r->base; 1888 orig_end = r->base + r->size; 1889 start = round_up(orig_start, align); 1890 end = round_down(orig_end, align); 1891 1892 if (start == orig_start && end == orig_end) 1893 continue; 1894 1895 if (start < end) { 1896 r->base = start; 1897 r->size = end - start; 1898 } else { 1899 memblock_remove_region(&memblock.memory, 1900 r - memblock.memory.regions); 1901 r--; 1902 } 1903 } 1904 } 1905 1906 void __init_memblock memblock_set_current_limit(phys_addr_t limit) 1907 { 1908 memblock.current_limit = limit; 1909 } 1910 1911 phys_addr_t __init_memblock memblock_get_current_limit(void) 1912 { 1913 return memblock.current_limit; 1914 } 1915 1916 static void __init_memblock memblock_dump(struct memblock_type *type) 1917 { 1918 phys_addr_t base, end, size; 1919 enum memblock_flags flags; 1920 int idx; 1921 struct memblock_region *rgn; 1922 1923 pr_info(" %s.cnt = 0x%lx\n", type->name, type->cnt); 1924 1925 for_each_memblock_type(idx, type, rgn) { 1926 char nid_buf[32] = ""; 1927 1928 base = rgn->base; 1929 size = rgn->size; 1930 end = base + size - 1; 1931 flags = rgn->flags; 1932 #ifdef CONFIG_NUMA 1933 if (memblock_get_region_node(rgn) != MAX_NUMNODES) 1934 snprintf(nid_buf, sizeof(nid_buf), " on node %d", 1935 memblock_get_region_node(rgn)); 1936 #endif 1937 pr_info(" %s[%#x]\t[%pa-%pa], %pa bytes%s flags: %#x\n", 1938 type->name, idx, &base, &end, &size, nid_buf, flags); 1939 } 1940 } 1941 1942 static void __init_memblock __memblock_dump_all(void) 1943 { 1944 pr_info("MEMBLOCK configuration:\n"); 1945 pr_info(" memory size = %pa reserved size = %pa\n", 1946 &memblock.memory.total_size, 1947 &memblock.reserved.total_size); 1948 1949 memblock_dump(&memblock.memory); 1950 memblock_dump(&memblock.reserved); 1951 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP 1952 memblock_dump(&physmem); 1953 #endif 1954 } 1955 1956 void __init_memblock memblock_dump_all(void) 1957 { 1958 if (memblock_debug) 1959 __memblock_dump_all(); 1960 } 1961 1962 void __init memblock_allow_resize(void) 1963 { 1964 memblock_can_resize = 1; 1965 } 1966 1967 static int __init early_memblock(char *p) 1968 { 1969 if (p && strstr(p, "debug")) 1970 memblock_debug = 1; 1971 return 0; 1972 } 1973 early_param("memblock", early_memblock); 1974 1975 static void __init free_memmap(unsigned long start_pfn, unsigned long end_pfn) 1976 { 1977 struct page *start_pg, *end_pg; 1978 phys_addr_t pg, pgend; 1979 1980 /* 1981 * Convert start_pfn/end_pfn to a struct page pointer. 1982 */ 1983 start_pg = pfn_to_page(start_pfn - 1) + 1; 1984 end_pg = pfn_to_page(end_pfn - 1) + 1; 1985 1986 /* 1987 * Convert to physical addresses, and round start upwards and end 1988 * downwards. 1989 */ 1990 pg = PAGE_ALIGN(__pa(start_pg)); 1991 pgend = __pa(end_pg) & PAGE_MASK; 1992 1993 /* 1994 * If there are free pages between these, free the section of the 1995 * memmap array. 1996 */ 1997 if (pg < pgend) 1998 memblock_phys_free(pg, pgend - pg); 1999 } 2000 2001 /* 2002 * The mem_map array can get very big. Free the unused area of the memory map. 2003 */ 2004 static void __init free_unused_memmap(void) 2005 { 2006 unsigned long start, end, prev_end = 0; 2007 int i; 2008 2009 if (!IS_ENABLED(CONFIG_HAVE_ARCH_PFN_VALID) || 2010 IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP)) 2011 return; 2012 2013 /* 2014 * This relies on each bank being in address order. 2015 * The banks are sorted previously in bootmem_init(). 2016 */ 2017 for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, NULL) { 2018 #ifdef CONFIG_SPARSEMEM 2019 /* 2020 * Take care not to free memmap entries that don't exist 2021 * due to SPARSEMEM sections which aren't present. 2022 */ 2023 start = min(start, ALIGN(prev_end, PAGES_PER_SECTION)); 2024 #endif 2025 /* 2026 * Align down here since many operations in VM subsystem 2027 * presume that there are no holes in the memory map inside 2028 * a pageblock 2029 */ 2030 start = pageblock_start_pfn(start); 2031 2032 /* 2033 * If we had a previous bank, and there is a space 2034 * between the current bank and the previous, free it. 2035 */ 2036 if (prev_end && prev_end < start) 2037 free_memmap(prev_end, start); 2038 2039 /* 2040 * Align up here since many operations in VM subsystem 2041 * presume that there are no holes in the memory map inside 2042 * a pageblock 2043 */ 2044 prev_end = pageblock_align(end); 2045 } 2046 2047 #ifdef CONFIG_SPARSEMEM 2048 if (!IS_ALIGNED(prev_end, PAGES_PER_SECTION)) { 2049 prev_end = pageblock_align(end); 2050 free_memmap(prev_end, ALIGN(prev_end, PAGES_PER_SECTION)); 2051 } 2052 #endif 2053 } 2054 2055 static void __init __free_pages_memory(unsigned long start, unsigned long end) 2056 { 2057 int order; 2058 2059 while (start < end) { 2060 /* 2061 * Free the pages in the largest chunks alignment allows. 2062 * 2063 * __ffs() behaviour is undefined for 0. start == 0 is 2064 * MAX_ORDER-aligned, set order to MAX_ORDER for the case. 2065 */ 2066 if (start) 2067 order = min_t(int, MAX_ORDER, __ffs(start)); 2068 else 2069 order = MAX_ORDER; 2070 2071 while (start + (1UL << order) > end) 2072 order--; 2073 2074 memblock_free_pages(pfn_to_page(start), start, order); 2075 2076 start += (1UL << order); 2077 } 2078 } 2079 2080 static unsigned long __init __free_memory_core(phys_addr_t start, 2081 phys_addr_t end) 2082 { 2083 unsigned long start_pfn = PFN_UP(start); 2084 unsigned long end_pfn = min_t(unsigned long, 2085 PFN_DOWN(end), max_low_pfn); 2086 2087 if (start_pfn >= end_pfn) 2088 return 0; 2089 2090 __free_pages_memory(start_pfn, end_pfn); 2091 2092 return end_pfn - start_pfn; 2093 } 2094 2095 static void __init memmap_init_reserved_pages(void) 2096 { 2097 struct memblock_region *region; 2098 phys_addr_t start, end; 2099 int nid; 2100 2101 /* 2102 * set nid on all reserved pages and also treat struct 2103 * pages for the NOMAP regions as PageReserved 2104 */ 2105 for_each_mem_region(region) { 2106 nid = memblock_get_region_node(region); 2107 start = region->base; 2108 end = start + region->size; 2109 2110 if (memblock_is_nomap(region)) 2111 reserve_bootmem_region(start, end, nid); 2112 2113 memblock_set_node(start, end, &memblock.reserved, nid); 2114 } 2115 2116 /* initialize struct pages for the reserved regions */ 2117 for_each_reserved_mem_region(region) { 2118 nid = memblock_get_region_node(region); 2119 start = region->base; 2120 end = start + region->size; 2121 2122 reserve_bootmem_region(start, end, nid); 2123 } 2124 } 2125 2126 static unsigned long __init free_low_memory_core_early(void) 2127 { 2128 unsigned long count = 0; 2129 phys_addr_t start, end; 2130 u64 i; 2131 2132 memblock_clear_hotplug(0, -1); 2133 2134 memmap_init_reserved_pages(); 2135 2136 /* 2137 * We need to use NUMA_NO_NODE instead of NODE_DATA(0)->node_id 2138 * because in some case like Node0 doesn't have RAM installed 2139 * low ram will be on Node1 2140 */ 2141 for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE, &start, &end, 2142 NULL) 2143 count += __free_memory_core(start, end); 2144 2145 return count; 2146 } 2147 2148 static int reset_managed_pages_done __initdata; 2149 2150 static void __init reset_node_managed_pages(pg_data_t *pgdat) 2151 { 2152 struct zone *z; 2153 2154 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++) 2155 atomic_long_set(&z->managed_pages, 0); 2156 } 2157 2158 void __init reset_all_zones_managed_pages(void) 2159 { 2160 struct pglist_data *pgdat; 2161 2162 if (reset_managed_pages_done) 2163 return; 2164 2165 for_each_online_pgdat(pgdat) 2166 reset_node_managed_pages(pgdat); 2167 2168 reset_managed_pages_done = 1; 2169 } 2170 2171 /** 2172 * memblock_free_all - release free pages to the buddy allocator 2173 */ 2174 void __init memblock_free_all(void) 2175 { 2176 unsigned long pages; 2177 2178 free_unused_memmap(); 2179 reset_all_zones_managed_pages(); 2180 2181 pages = free_low_memory_core_early(); 2182 totalram_pages_add(pages); 2183 } 2184 2185 #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_ARCH_KEEP_MEMBLOCK) 2186 static const char * const flagname[] = { 2187 [ilog2(MEMBLOCK_HOTPLUG)] = "HOTPLUG", 2188 [ilog2(MEMBLOCK_MIRROR)] = "MIRROR", 2189 [ilog2(MEMBLOCK_NOMAP)] = "NOMAP", 2190 [ilog2(MEMBLOCK_DRIVER_MANAGED)] = "DRV_MNG", 2191 }; 2192 2193 static int memblock_debug_show(struct seq_file *m, void *private) 2194 { 2195 struct memblock_type *type = m->private; 2196 struct memblock_region *reg; 2197 int i, j, nid; 2198 unsigned int count = ARRAY_SIZE(flagname); 2199 phys_addr_t end; 2200 2201 for (i = 0; i < type->cnt; i++) { 2202 reg = &type->regions[i]; 2203 end = reg->base + reg->size - 1; 2204 nid = memblock_get_region_node(reg); 2205 2206 seq_printf(m, "%4d: ", i); 2207 seq_printf(m, "%pa..%pa ", ®->base, &end); 2208 if (nid != MAX_NUMNODES) 2209 seq_printf(m, "%4d ", nid); 2210 else 2211 seq_printf(m, "%4c ", 'x'); 2212 if (reg->flags) { 2213 for (j = 0; j < count; j++) { 2214 if (reg->flags & (1U << j)) { 2215 seq_printf(m, "%s\n", flagname[j]); 2216 break; 2217 } 2218 } 2219 if (j == count) 2220 seq_printf(m, "%s\n", "UNKNOWN"); 2221 } else { 2222 seq_printf(m, "%s\n", "NONE"); 2223 } 2224 } 2225 return 0; 2226 } 2227 DEFINE_SHOW_ATTRIBUTE(memblock_debug); 2228 2229 static int __init memblock_init_debugfs(void) 2230 { 2231 struct dentry *root = debugfs_create_dir("memblock", NULL); 2232 2233 debugfs_create_file("memory", 0444, root, 2234 &memblock.memory, &memblock_debug_fops); 2235 debugfs_create_file("reserved", 0444, root, 2236 &memblock.reserved, &memblock_debug_fops); 2237 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP 2238 debugfs_create_file("physmem", 0444, root, &physmem, 2239 &memblock_debug_fops); 2240 #endif 2241 2242 return 0; 2243 } 2244 __initcall(memblock_init_debugfs); 2245 2246 #endif /* CONFIG_DEBUG_FS */ 2247