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 panic("memblock: cannot resize %s array\n", type->name); 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_validate_numa_coverage - check if amount of memory with 739 * no node ID assigned is less than a threshold 740 * @threshold_bytes: maximal number of pages that can have unassigned node 741 * ID (in bytes). 742 * 743 * A buggy firmware may report memory that does not belong to any node. 744 * Check if amount of such memory is below @threshold_bytes. 745 * 746 * Return: true on success, false on failure. 747 */ 748 bool __init_memblock memblock_validate_numa_coverage(unsigned long threshold_bytes) 749 { 750 unsigned long nr_pages = 0; 751 unsigned long start_pfn, end_pfn, mem_size_mb; 752 int nid, i; 753 754 /* calculate lose page */ 755 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) { 756 if (nid == NUMA_NO_NODE) 757 nr_pages += end_pfn - start_pfn; 758 } 759 760 if ((nr_pages << PAGE_SHIFT) >= threshold_bytes) { 761 mem_size_mb = memblock_phys_mem_size() >> 20; 762 pr_err("NUMA: no nodes coverage for %luMB of %luMB RAM\n", 763 (nr_pages << PAGE_SHIFT) >> 20, mem_size_mb); 764 return false; 765 } 766 767 return true; 768 } 769 770 771 /** 772 * memblock_isolate_range - isolate given range into disjoint memblocks 773 * @type: memblock type to isolate range for 774 * @base: base of range to isolate 775 * @size: size of range to isolate 776 * @start_rgn: out parameter for the start of isolated region 777 * @end_rgn: out parameter for the end of isolated region 778 * 779 * Walk @type and ensure that regions don't cross the boundaries defined by 780 * [@base, @base + @size). Crossing regions are split at the boundaries, 781 * which may create at most two more regions. The index of the first 782 * region inside the range is returned in *@start_rgn and end in *@end_rgn. 783 * 784 * Return: 785 * 0 on success, -errno on failure. 786 */ 787 static int __init_memblock memblock_isolate_range(struct memblock_type *type, 788 phys_addr_t base, phys_addr_t size, 789 int *start_rgn, int *end_rgn) 790 { 791 phys_addr_t end = base + memblock_cap_size(base, &size); 792 int idx; 793 struct memblock_region *rgn; 794 795 *start_rgn = *end_rgn = 0; 796 797 if (!size) 798 return 0; 799 800 /* we'll create at most two more regions */ 801 while (type->cnt + 2 > type->max) 802 if (memblock_double_array(type, base, size) < 0) 803 return -ENOMEM; 804 805 for_each_memblock_type(idx, type, rgn) { 806 phys_addr_t rbase = rgn->base; 807 phys_addr_t rend = rbase + rgn->size; 808 809 if (rbase >= end) 810 break; 811 if (rend <= base) 812 continue; 813 814 if (rbase < base) { 815 /* 816 * @rgn intersects from below. Split and continue 817 * to process the next region - the new top half. 818 */ 819 rgn->base = base; 820 rgn->size -= base - rbase; 821 type->total_size -= base - rbase; 822 memblock_insert_region(type, idx, rbase, base - rbase, 823 memblock_get_region_node(rgn), 824 rgn->flags); 825 } else if (rend > end) { 826 /* 827 * @rgn intersects from above. Split and redo the 828 * current region - the new bottom half. 829 */ 830 rgn->base = end; 831 rgn->size -= end - rbase; 832 type->total_size -= end - rbase; 833 memblock_insert_region(type, idx--, rbase, end - rbase, 834 memblock_get_region_node(rgn), 835 rgn->flags); 836 } else { 837 /* @rgn is fully contained, record it */ 838 if (!*end_rgn) 839 *start_rgn = idx; 840 *end_rgn = idx + 1; 841 } 842 } 843 844 return 0; 845 } 846 847 static int __init_memblock memblock_remove_range(struct memblock_type *type, 848 phys_addr_t base, phys_addr_t size) 849 { 850 int start_rgn, end_rgn; 851 int i, ret; 852 853 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn); 854 if (ret) 855 return ret; 856 857 for (i = end_rgn - 1; i >= start_rgn; i--) 858 memblock_remove_region(type, i); 859 return 0; 860 } 861 862 int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size) 863 { 864 phys_addr_t end = base + size - 1; 865 866 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__, 867 &base, &end, (void *)_RET_IP_); 868 869 return memblock_remove_range(&memblock.memory, base, size); 870 } 871 872 /** 873 * memblock_free - free boot memory allocation 874 * @ptr: starting address of the boot memory allocation 875 * @size: size of the boot memory block in bytes 876 * 877 * Free boot memory block previously allocated by memblock_alloc_xx() API. 878 * The freeing memory will not be released to the buddy allocator. 879 */ 880 void __init_memblock memblock_free(void *ptr, size_t size) 881 { 882 if (ptr) 883 memblock_phys_free(__pa(ptr), size); 884 } 885 886 /** 887 * memblock_phys_free - free boot memory block 888 * @base: phys starting address of the boot memory block 889 * @size: size of the boot memory block in bytes 890 * 891 * Free boot memory block previously allocated by memblock_phys_alloc_xx() API. 892 * The freeing memory will not be released to the buddy allocator. 893 */ 894 int __init_memblock memblock_phys_free(phys_addr_t base, phys_addr_t size) 895 { 896 phys_addr_t end = base + size - 1; 897 898 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__, 899 &base, &end, (void *)_RET_IP_); 900 901 kmemleak_free_part_phys(base, size); 902 return memblock_remove_range(&memblock.reserved, base, size); 903 } 904 905 int __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size) 906 { 907 phys_addr_t end = base + size - 1; 908 909 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__, 910 &base, &end, (void *)_RET_IP_); 911 912 return memblock_add_range(&memblock.reserved, base, size, MAX_NUMNODES, 0); 913 } 914 915 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP 916 int __init_memblock memblock_physmem_add(phys_addr_t base, phys_addr_t size) 917 { 918 phys_addr_t end = base + size - 1; 919 920 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__, 921 &base, &end, (void *)_RET_IP_); 922 923 return memblock_add_range(&physmem, base, size, MAX_NUMNODES, 0); 924 } 925 #endif 926 927 /** 928 * memblock_setclr_flag - set or clear flag for a memory region 929 * @type: memblock type to set/clear flag for 930 * @base: base address of the region 931 * @size: size of the region 932 * @set: set or clear the flag 933 * @flag: the flag to update 934 * 935 * This function isolates region [@base, @base + @size), and sets/clears flag 936 * 937 * Return: 0 on success, -errno on failure. 938 */ 939 static int __init_memblock memblock_setclr_flag(struct memblock_type *type, 940 phys_addr_t base, phys_addr_t size, int set, int flag) 941 { 942 int i, ret, start_rgn, end_rgn; 943 944 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn); 945 if (ret) 946 return ret; 947 948 for (i = start_rgn; i < end_rgn; i++) { 949 struct memblock_region *r = &type->regions[i]; 950 951 if (set) 952 r->flags |= flag; 953 else 954 r->flags &= ~flag; 955 } 956 957 memblock_merge_regions(type, start_rgn, end_rgn); 958 return 0; 959 } 960 961 /** 962 * memblock_mark_hotplug - Mark hotpluggable memory with flag MEMBLOCK_HOTPLUG. 963 * @base: the base phys addr of the region 964 * @size: the size of the region 965 * 966 * Return: 0 on success, -errno on failure. 967 */ 968 int __init_memblock memblock_mark_hotplug(phys_addr_t base, phys_addr_t size) 969 { 970 return memblock_setclr_flag(&memblock.memory, base, size, 1, MEMBLOCK_HOTPLUG); 971 } 972 973 /** 974 * memblock_clear_hotplug - Clear flag MEMBLOCK_HOTPLUG for a specified region. 975 * @base: the base phys addr of the region 976 * @size: the size of the region 977 * 978 * Return: 0 on success, -errno on failure. 979 */ 980 int __init_memblock memblock_clear_hotplug(phys_addr_t base, phys_addr_t size) 981 { 982 return memblock_setclr_flag(&memblock.memory, base, size, 0, MEMBLOCK_HOTPLUG); 983 } 984 985 /** 986 * memblock_mark_mirror - Mark mirrored memory with flag MEMBLOCK_MIRROR. 987 * @base: the base phys addr of the region 988 * @size: the size of the region 989 * 990 * Return: 0 on success, -errno on failure. 991 */ 992 int __init_memblock memblock_mark_mirror(phys_addr_t base, phys_addr_t size) 993 { 994 if (!mirrored_kernelcore) 995 return 0; 996 997 system_has_some_mirror = true; 998 999 return memblock_setclr_flag(&memblock.memory, base, size, 1, MEMBLOCK_MIRROR); 1000 } 1001 1002 /** 1003 * memblock_mark_nomap - Mark a memory region with flag MEMBLOCK_NOMAP. 1004 * @base: the base phys addr of the region 1005 * @size: the size of the region 1006 * 1007 * The memory regions marked with %MEMBLOCK_NOMAP will not be added to the 1008 * direct mapping of the physical memory. These regions will still be 1009 * covered by the memory map. The struct page representing NOMAP memory 1010 * frames in the memory map will be PageReserved() 1011 * 1012 * Note: if the memory being marked %MEMBLOCK_NOMAP was allocated from 1013 * memblock, the caller must inform kmemleak to ignore that memory 1014 * 1015 * Return: 0 on success, -errno on failure. 1016 */ 1017 int __init_memblock memblock_mark_nomap(phys_addr_t base, phys_addr_t size) 1018 { 1019 return memblock_setclr_flag(&memblock.memory, base, size, 1, MEMBLOCK_NOMAP); 1020 } 1021 1022 /** 1023 * memblock_clear_nomap - Clear flag MEMBLOCK_NOMAP for a specified region. 1024 * @base: the base phys addr of the region 1025 * @size: the size of the region 1026 * 1027 * Return: 0 on success, -errno on failure. 1028 */ 1029 int __init_memblock memblock_clear_nomap(phys_addr_t base, phys_addr_t size) 1030 { 1031 return memblock_setclr_flag(&memblock.memory, base, size, 0, MEMBLOCK_NOMAP); 1032 } 1033 1034 /** 1035 * memblock_reserved_mark_noinit - Mark a reserved memory region with flag 1036 * MEMBLOCK_RSRV_NOINIT which results in the struct pages not being initialized 1037 * for this region. 1038 * @base: the base phys addr of the region 1039 * @size: the size of the region 1040 * 1041 * struct pages will not be initialized for reserved memory regions marked with 1042 * %MEMBLOCK_RSRV_NOINIT. 1043 * 1044 * Return: 0 on success, -errno on failure. 1045 */ 1046 int __init_memblock memblock_reserved_mark_noinit(phys_addr_t base, phys_addr_t size) 1047 { 1048 return memblock_setclr_flag(&memblock.reserved, base, size, 1, 1049 MEMBLOCK_RSRV_NOINIT); 1050 } 1051 1052 static bool should_skip_region(struct memblock_type *type, 1053 struct memblock_region *m, 1054 int nid, int flags) 1055 { 1056 int m_nid = memblock_get_region_node(m); 1057 1058 /* we never skip regions when iterating memblock.reserved or physmem */ 1059 if (type != memblock_memory) 1060 return false; 1061 1062 /* only memory regions are associated with nodes, check it */ 1063 if (nid != NUMA_NO_NODE && nid != m_nid) 1064 return true; 1065 1066 /* skip hotpluggable memory regions if needed */ 1067 if (movable_node_is_enabled() && memblock_is_hotpluggable(m) && 1068 !(flags & MEMBLOCK_HOTPLUG)) 1069 return true; 1070 1071 /* if we want mirror memory skip non-mirror memory regions */ 1072 if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m)) 1073 return true; 1074 1075 /* skip nomap memory unless we were asked for it explicitly */ 1076 if (!(flags & MEMBLOCK_NOMAP) && memblock_is_nomap(m)) 1077 return true; 1078 1079 /* skip driver-managed memory unless we were asked for it explicitly */ 1080 if (!(flags & MEMBLOCK_DRIVER_MANAGED) && memblock_is_driver_managed(m)) 1081 return true; 1082 1083 return false; 1084 } 1085 1086 /** 1087 * __next_mem_range - next function for for_each_free_mem_range() etc. 1088 * @idx: pointer to u64 loop variable 1089 * @nid: node selector, %NUMA_NO_NODE for all nodes 1090 * @flags: pick from blocks based on memory attributes 1091 * @type_a: pointer to memblock_type from where the range is taken 1092 * @type_b: pointer to memblock_type which excludes memory from being taken 1093 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL 1094 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL 1095 * @out_nid: ptr to int for nid of the range, can be %NULL 1096 * 1097 * Find the first area from *@idx which matches @nid, fill the out 1098 * parameters, and update *@idx for the next iteration. The lower 32bit of 1099 * *@idx contains index into type_a and the upper 32bit indexes the 1100 * areas before each region in type_b. For example, if type_b regions 1101 * look like the following, 1102 * 1103 * 0:[0-16), 1:[32-48), 2:[128-130) 1104 * 1105 * The upper 32bit indexes the following regions. 1106 * 1107 * 0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX) 1108 * 1109 * As both region arrays are sorted, the function advances the two indices 1110 * in lockstep and returns each intersection. 1111 */ 1112 void __next_mem_range(u64 *idx, int nid, enum memblock_flags flags, 1113 struct memblock_type *type_a, 1114 struct memblock_type *type_b, phys_addr_t *out_start, 1115 phys_addr_t *out_end, int *out_nid) 1116 { 1117 int idx_a = *idx & 0xffffffff; 1118 int idx_b = *idx >> 32; 1119 1120 if (WARN_ONCE(nid == MAX_NUMNODES, 1121 "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n")) 1122 nid = NUMA_NO_NODE; 1123 1124 for (; idx_a < type_a->cnt; idx_a++) { 1125 struct memblock_region *m = &type_a->regions[idx_a]; 1126 1127 phys_addr_t m_start = m->base; 1128 phys_addr_t m_end = m->base + m->size; 1129 int m_nid = memblock_get_region_node(m); 1130 1131 if (should_skip_region(type_a, m, nid, flags)) 1132 continue; 1133 1134 if (!type_b) { 1135 if (out_start) 1136 *out_start = m_start; 1137 if (out_end) 1138 *out_end = m_end; 1139 if (out_nid) 1140 *out_nid = m_nid; 1141 idx_a++; 1142 *idx = (u32)idx_a | (u64)idx_b << 32; 1143 return; 1144 } 1145 1146 /* scan areas before each reservation */ 1147 for (; idx_b < type_b->cnt + 1; idx_b++) { 1148 struct memblock_region *r; 1149 phys_addr_t r_start; 1150 phys_addr_t r_end; 1151 1152 r = &type_b->regions[idx_b]; 1153 r_start = idx_b ? r[-1].base + r[-1].size : 0; 1154 r_end = idx_b < type_b->cnt ? 1155 r->base : PHYS_ADDR_MAX; 1156 1157 /* 1158 * if idx_b advanced past idx_a, 1159 * break out to advance idx_a 1160 */ 1161 if (r_start >= m_end) 1162 break; 1163 /* if the two regions intersect, we're done */ 1164 if (m_start < r_end) { 1165 if (out_start) 1166 *out_start = 1167 max(m_start, r_start); 1168 if (out_end) 1169 *out_end = min(m_end, r_end); 1170 if (out_nid) 1171 *out_nid = m_nid; 1172 /* 1173 * The region which ends first is 1174 * advanced for the next iteration. 1175 */ 1176 if (m_end <= r_end) 1177 idx_a++; 1178 else 1179 idx_b++; 1180 *idx = (u32)idx_a | (u64)idx_b << 32; 1181 return; 1182 } 1183 } 1184 } 1185 1186 /* signal end of iteration */ 1187 *idx = ULLONG_MAX; 1188 } 1189 1190 /** 1191 * __next_mem_range_rev - generic next function for for_each_*_range_rev() 1192 * 1193 * @idx: pointer to u64 loop variable 1194 * @nid: node selector, %NUMA_NO_NODE for all nodes 1195 * @flags: pick from blocks based on memory attributes 1196 * @type_a: pointer to memblock_type from where the range is taken 1197 * @type_b: pointer to memblock_type which excludes memory from being taken 1198 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL 1199 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL 1200 * @out_nid: ptr to int for nid of the range, can be %NULL 1201 * 1202 * Finds the next range from type_a which is not marked as unsuitable 1203 * in type_b. 1204 * 1205 * Reverse of __next_mem_range(). 1206 */ 1207 void __init_memblock __next_mem_range_rev(u64 *idx, int nid, 1208 enum memblock_flags flags, 1209 struct memblock_type *type_a, 1210 struct memblock_type *type_b, 1211 phys_addr_t *out_start, 1212 phys_addr_t *out_end, int *out_nid) 1213 { 1214 int idx_a = *idx & 0xffffffff; 1215 int idx_b = *idx >> 32; 1216 1217 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n")) 1218 nid = NUMA_NO_NODE; 1219 1220 if (*idx == (u64)ULLONG_MAX) { 1221 idx_a = type_a->cnt - 1; 1222 if (type_b != NULL) 1223 idx_b = type_b->cnt; 1224 else 1225 idx_b = 0; 1226 } 1227 1228 for (; idx_a >= 0; idx_a--) { 1229 struct memblock_region *m = &type_a->regions[idx_a]; 1230 1231 phys_addr_t m_start = m->base; 1232 phys_addr_t m_end = m->base + m->size; 1233 int m_nid = memblock_get_region_node(m); 1234 1235 if (should_skip_region(type_a, m, nid, flags)) 1236 continue; 1237 1238 if (!type_b) { 1239 if (out_start) 1240 *out_start = m_start; 1241 if (out_end) 1242 *out_end = m_end; 1243 if (out_nid) 1244 *out_nid = m_nid; 1245 idx_a--; 1246 *idx = (u32)idx_a | (u64)idx_b << 32; 1247 return; 1248 } 1249 1250 /* scan areas before each reservation */ 1251 for (; idx_b >= 0; idx_b--) { 1252 struct memblock_region *r; 1253 phys_addr_t r_start; 1254 phys_addr_t r_end; 1255 1256 r = &type_b->regions[idx_b]; 1257 r_start = idx_b ? r[-1].base + r[-1].size : 0; 1258 r_end = idx_b < type_b->cnt ? 1259 r->base : PHYS_ADDR_MAX; 1260 /* 1261 * if idx_b advanced past idx_a, 1262 * break out to advance idx_a 1263 */ 1264 1265 if (r_end <= m_start) 1266 break; 1267 /* if the two regions intersect, we're done */ 1268 if (m_end > r_start) { 1269 if (out_start) 1270 *out_start = max(m_start, r_start); 1271 if (out_end) 1272 *out_end = min(m_end, r_end); 1273 if (out_nid) 1274 *out_nid = m_nid; 1275 if (m_start >= r_start) 1276 idx_a--; 1277 else 1278 idx_b--; 1279 *idx = (u32)idx_a | (u64)idx_b << 32; 1280 return; 1281 } 1282 } 1283 } 1284 /* signal end of iteration */ 1285 *idx = ULLONG_MAX; 1286 } 1287 1288 /* 1289 * Common iterator interface used to define for_each_mem_pfn_range(). 1290 */ 1291 void __init_memblock __next_mem_pfn_range(int *idx, int nid, 1292 unsigned long *out_start_pfn, 1293 unsigned long *out_end_pfn, int *out_nid) 1294 { 1295 struct memblock_type *type = &memblock.memory; 1296 struct memblock_region *r; 1297 int r_nid; 1298 1299 while (++*idx < type->cnt) { 1300 r = &type->regions[*idx]; 1301 r_nid = memblock_get_region_node(r); 1302 1303 if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size)) 1304 continue; 1305 if (nid == MAX_NUMNODES || nid == r_nid) 1306 break; 1307 } 1308 if (*idx >= type->cnt) { 1309 *idx = -1; 1310 return; 1311 } 1312 1313 if (out_start_pfn) 1314 *out_start_pfn = PFN_UP(r->base); 1315 if (out_end_pfn) 1316 *out_end_pfn = PFN_DOWN(r->base + r->size); 1317 if (out_nid) 1318 *out_nid = r_nid; 1319 } 1320 1321 /** 1322 * memblock_set_node - set node ID on memblock regions 1323 * @base: base of area to set node ID for 1324 * @size: size of area to set node ID for 1325 * @type: memblock type to set node ID for 1326 * @nid: node ID to set 1327 * 1328 * Set the nid of memblock @type regions in [@base, @base + @size) to @nid. 1329 * Regions which cross the area boundaries are split as necessary. 1330 * 1331 * Return: 1332 * 0 on success, -errno on failure. 1333 */ 1334 int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size, 1335 struct memblock_type *type, int nid) 1336 { 1337 #ifdef CONFIG_NUMA 1338 int start_rgn, end_rgn; 1339 int i, ret; 1340 1341 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn); 1342 if (ret) 1343 return ret; 1344 1345 for (i = start_rgn; i < end_rgn; i++) 1346 memblock_set_region_node(&type->regions[i], nid); 1347 1348 memblock_merge_regions(type, start_rgn, end_rgn); 1349 #endif 1350 return 0; 1351 } 1352 1353 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT 1354 /** 1355 * __next_mem_pfn_range_in_zone - iterator for for_each_*_range_in_zone() 1356 * 1357 * @idx: pointer to u64 loop variable 1358 * @zone: zone in which all of the memory blocks reside 1359 * @out_spfn: ptr to ulong for start pfn of the range, can be %NULL 1360 * @out_epfn: ptr to ulong for end pfn of the range, can be %NULL 1361 * 1362 * This function is meant to be a zone/pfn specific wrapper for the 1363 * for_each_mem_range type iterators. Specifically they are used in the 1364 * deferred memory init routines and as such we were duplicating much of 1365 * this logic throughout the code. So instead of having it in multiple 1366 * locations it seemed like it would make more sense to centralize this to 1367 * one new iterator that does everything they need. 1368 */ 1369 void __init_memblock 1370 __next_mem_pfn_range_in_zone(u64 *idx, struct zone *zone, 1371 unsigned long *out_spfn, unsigned long *out_epfn) 1372 { 1373 int zone_nid = zone_to_nid(zone); 1374 phys_addr_t spa, epa; 1375 1376 __next_mem_range(idx, zone_nid, MEMBLOCK_NONE, 1377 &memblock.memory, &memblock.reserved, 1378 &spa, &epa, NULL); 1379 1380 while (*idx != U64_MAX) { 1381 unsigned long epfn = PFN_DOWN(epa); 1382 unsigned long spfn = PFN_UP(spa); 1383 1384 /* 1385 * Verify the end is at least past the start of the zone and 1386 * that we have at least one PFN to initialize. 1387 */ 1388 if (zone->zone_start_pfn < epfn && spfn < epfn) { 1389 /* if we went too far just stop searching */ 1390 if (zone_end_pfn(zone) <= spfn) { 1391 *idx = U64_MAX; 1392 break; 1393 } 1394 1395 if (out_spfn) 1396 *out_spfn = max(zone->zone_start_pfn, spfn); 1397 if (out_epfn) 1398 *out_epfn = min(zone_end_pfn(zone), epfn); 1399 1400 return; 1401 } 1402 1403 __next_mem_range(idx, zone_nid, MEMBLOCK_NONE, 1404 &memblock.memory, &memblock.reserved, 1405 &spa, &epa, NULL); 1406 } 1407 1408 /* signal end of iteration */ 1409 if (out_spfn) 1410 *out_spfn = ULONG_MAX; 1411 if (out_epfn) 1412 *out_epfn = 0; 1413 } 1414 1415 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ 1416 1417 /** 1418 * memblock_alloc_range_nid - allocate boot memory block 1419 * @size: size of memory block to be allocated in bytes 1420 * @align: alignment of the region and block's size 1421 * @start: the lower bound of the memory region to allocate (phys address) 1422 * @end: the upper bound of the memory region to allocate (phys address) 1423 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 1424 * @exact_nid: control the allocation fall back to other nodes 1425 * 1426 * The allocation is performed from memory region limited by 1427 * memblock.current_limit if @end == %MEMBLOCK_ALLOC_ACCESSIBLE. 1428 * 1429 * If the specified node can not hold the requested memory and @exact_nid 1430 * is false, the allocation falls back to any node in the system. 1431 * 1432 * For systems with memory mirroring, the allocation is attempted first 1433 * from the regions with mirroring enabled and then retried from any 1434 * memory region. 1435 * 1436 * In addition, function using kmemleak_alloc_phys for allocated boot 1437 * memory block, it is never reported as leaks. 1438 * 1439 * Return: 1440 * Physical address of allocated memory block on success, %0 on failure. 1441 */ 1442 phys_addr_t __init memblock_alloc_range_nid(phys_addr_t size, 1443 phys_addr_t align, phys_addr_t start, 1444 phys_addr_t end, int nid, 1445 bool exact_nid) 1446 { 1447 enum memblock_flags flags = choose_memblock_flags(); 1448 phys_addr_t found; 1449 1450 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n")) 1451 nid = NUMA_NO_NODE; 1452 1453 if (!align) { 1454 /* Can't use WARNs this early in boot on powerpc */ 1455 dump_stack(); 1456 align = SMP_CACHE_BYTES; 1457 } 1458 1459 again: 1460 found = memblock_find_in_range_node(size, align, start, end, nid, 1461 flags); 1462 if (found && !memblock_reserve(found, size)) 1463 goto done; 1464 1465 if (nid != NUMA_NO_NODE && !exact_nid) { 1466 found = memblock_find_in_range_node(size, align, start, 1467 end, NUMA_NO_NODE, 1468 flags); 1469 if (found && !memblock_reserve(found, size)) 1470 goto done; 1471 } 1472 1473 if (flags & MEMBLOCK_MIRROR) { 1474 flags &= ~MEMBLOCK_MIRROR; 1475 pr_warn_ratelimited("Could not allocate %pap bytes of mirrored memory\n", 1476 &size); 1477 goto again; 1478 } 1479 1480 return 0; 1481 1482 done: 1483 /* 1484 * Skip kmemleak for those places like kasan_init() and 1485 * early_pgtable_alloc() due to high volume. 1486 */ 1487 if (end != MEMBLOCK_ALLOC_NOLEAKTRACE) 1488 /* 1489 * Memblock allocated blocks are never reported as 1490 * leaks. This is because many of these blocks are 1491 * only referred via the physical address which is 1492 * not looked up by kmemleak. 1493 */ 1494 kmemleak_alloc_phys(found, size, 0); 1495 1496 /* 1497 * Some Virtual Machine platforms, such as Intel TDX or AMD SEV-SNP, 1498 * require memory to be accepted before it can be used by the 1499 * guest. 1500 * 1501 * Accept the memory of the allocated buffer. 1502 */ 1503 accept_memory(found, found + size); 1504 1505 return found; 1506 } 1507 1508 /** 1509 * memblock_phys_alloc_range - allocate a memory block inside specified range 1510 * @size: size of memory block to be allocated in bytes 1511 * @align: alignment of the region and block's size 1512 * @start: the lower bound of the memory region to allocate (physical address) 1513 * @end: the upper bound of the memory region to allocate (physical address) 1514 * 1515 * Allocate @size bytes in the between @start and @end. 1516 * 1517 * Return: physical address of the allocated memory block on success, 1518 * %0 on failure. 1519 */ 1520 phys_addr_t __init memblock_phys_alloc_range(phys_addr_t size, 1521 phys_addr_t align, 1522 phys_addr_t start, 1523 phys_addr_t end) 1524 { 1525 memblock_dbg("%s: %llu bytes align=0x%llx from=%pa max_addr=%pa %pS\n", 1526 __func__, (u64)size, (u64)align, &start, &end, 1527 (void *)_RET_IP_); 1528 return memblock_alloc_range_nid(size, align, start, end, NUMA_NO_NODE, 1529 false); 1530 } 1531 1532 /** 1533 * memblock_phys_alloc_try_nid - allocate a memory block from specified NUMA node 1534 * @size: size of memory block to be allocated in bytes 1535 * @align: alignment of the region and block's size 1536 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 1537 * 1538 * Allocates memory block from the specified NUMA node. If the node 1539 * has no available memory, attempts to allocated from any node in the 1540 * system. 1541 * 1542 * Return: physical address of the allocated memory block on success, 1543 * %0 on failure. 1544 */ 1545 phys_addr_t __init memblock_phys_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid) 1546 { 1547 return memblock_alloc_range_nid(size, align, 0, 1548 MEMBLOCK_ALLOC_ACCESSIBLE, nid, false); 1549 } 1550 1551 /** 1552 * memblock_alloc_internal - allocate boot memory block 1553 * @size: size of memory block to be allocated in bytes 1554 * @align: alignment of the region and block's size 1555 * @min_addr: the lower bound of the memory region to allocate (phys address) 1556 * @max_addr: the upper bound of the memory region to allocate (phys address) 1557 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 1558 * @exact_nid: control the allocation fall back to other nodes 1559 * 1560 * Allocates memory block using memblock_alloc_range_nid() and 1561 * converts the returned physical address to virtual. 1562 * 1563 * The @min_addr limit is dropped if it can not be satisfied and the allocation 1564 * will fall back to memory below @min_addr. Other constraints, such 1565 * as node and mirrored memory will be handled again in 1566 * memblock_alloc_range_nid(). 1567 * 1568 * Return: 1569 * Virtual address of allocated memory block on success, NULL on failure. 1570 */ 1571 static void * __init memblock_alloc_internal( 1572 phys_addr_t size, phys_addr_t align, 1573 phys_addr_t min_addr, phys_addr_t max_addr, 1574 int nid, bool exact_nid) 1575 { 1576 phys_addr_t alloc; 1577 1578 /* 1579 * Detect any accidental use of these APIs after slab is ready, as at 1580 * this moment memblock may be deinitialized already and its 1581 * internal data may be destroyed (after execution of memblock_free_all) 1582 */ 1583 if (WARN_ON_ONCE(slab_is_available())) 1584 return kzalloc_node(size, GFP_NOWAIT, nid); 1585 1586 if (max_addr > memblock.current_limit) 1587 max_addr = memblock.current_limit; 1588 1589 alloc = memblock_alloc_range_nid(size, align, min_addr, max_addr, nid, 1590 exact_nid); 1591 1592 /* retry allocation without lower limit */ 1593 if (!alloc && min_addr) 1594 alloc = memblock_alloc_range_nid(size, align, 0, max_addr, nid, 1595 exact_nid); 1596 1597 if (!alloc) 1598 return NULL; 1599 1600 return phys_to_virt(alloc); 1601 } 1602 1603 /** 1604 * memblock_alloc_exact_nid_raw - allocate boot memory block on the exact node 1605 * without zeroing memory 1606 * @size: size of memory block to be allocated in bytes 1607 * @align: alignment of the region and block's size 1608 * @min_addr: the lower bound of the memory region from where the allocation 1609 * is preferred (phys address) 1610 * @max_addr: the upper bound of the memory region from where the allocation 1611 * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to 1612 * allocate only from memory limited by memblock.current_limit value 1613 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 1614 * 1615 * Public function, provides additional debug information (including caller 1616 * info), if enabled. Does not zero allocated memory. 1617 * 1618 * Return: 1619 * Virtual address of allocated memory block on success, NULL on failure. 1620 */ 1621 void * __init memblock_alloc_exact_nid_raw( 1622 phys_addr_t size, phys_addr_t align, 1623 phys_addr_t min_addr, phys_addr_t max_addr, 1624 int nid) 1625 { 1626 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n", 1627 __func__, (u64)size, (u64)align, nid, &min_addr, 1628 &max_addr, (void *)_RET_IP_); 1629 1630 return memblock_alloc_internal(size, align, min_addr, max_addr, nid, 1631 true); 1632 } 1633 1634 /** 1635 * memblock_alloc_try_nid_raw - allocate boot memory block without zeroing 1636 * memory and without panicking 1637 * @size: size of memory block to be allocated in bytes 1638 * @align: alignment of the region and block's size 1639 * @min_addr: the lower bound of the memory region from where the allocation 1640 * is preferred (phys address) 1641 * @max_addr: the upper bound of the memory region from where the allocation 1642 * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to 1643 * allocate only from memory limited by memblock.current_limit value 1644 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 1645 * 1646 * Public function, provides additional debug information (including caller 1647 * info), if enabled. Does not zero allocated memory, does not panic if request 1648 * cannot be satisfied. 1649 * 1650 * Return: 1651 * Virtual address of allocated memory block on success, NULL on failure. 1652 */ 1653 void * __init memblock_alloc_try_nid_raw( 1654 phys_addr_t size, phys_addr_t align, 1655 phys_addr_t min_addr, phys_addr_t max_addr, 1656 int nid) 1657 { 1658 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n", 1659 __func__, (u64)size, (u64)align, nid, &min_addr, 1660 &max_addr, (void *)_RET_IP_); 1661 1662 return memblock_alloc_internal(size, align, min_addr, max_addr, nid, 1663 false); 1664 } 1665 1666 /** 1667 * memblock_alloc_try_nid - allocate boot memory block 1668 * @size: size of memory block to be allocated in bytes 1669 * @align: alignment of the region and block's size 1670 * @min_addr: the lower bound of the memory region from where the allocation 1671 * is preferred (phys address) 1672 * @max_addr: the upper bound of the memory region from where the allocation 1673 * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to 1674 * allocate only from memory limited by memblock.current_limit value 1675 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 1676 * 1677 * Public function, provides additional debug information (including caller 1678 * info), if enabled. This function zeroes the allocated memory. 1679 * 1680 * Return: 1681 * Virtual address of allocated memory block on success, NULL on failure. 1682 */ 1683 void * __init memblock_alloc_try_nid( 1684 phys_addr_t size, phys_addr_t align, 1685 phys_addr_t min_addr, phys_addr_t max_addr, 1686 int nid) 1687 { 1688 void *ptr; 1689 1690 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n", 1691 __func__, (u64)size, (u64)align, nid, &min_addr, 1692 &max_addr, (void *)_RET_IP_); 1693 ptr = memblock_alloc_internal(size, align, 1694 min_addr, max_addr, nid, false); 1695 if (ptr) 1696 memset(ptr, 0, size); 1697 1698 return ptr; 1699 } 1700 1701 /** 1702 * memblock_free_late - free pages directly to buddy allocator 1703 * @base: phys starting address of the boot memory block 1704 * @size: size of the boot memory block in bytes 1705 * 1706 * This is only useful when the memblock allocator has already been torn 1707 * down, but we are still initializing the system. Pages are released directly 1708 * to the buddy allocator. 1709 */ 1710 void __init memblock_free_late(phys_addr_t base, phys_addr_t size) 1711 { 1712 phys_addr_t cursor, end; 1713 1714 end = base + size - 1; 1715 memblock_dbg("%s: [%pa-%pa] %pS\n", 1716 __func__, &base, &end, (void *)_RET_IP_); 1717 kmemleak_free_part_phys(base, size); 1718 cursor = PFN_UP(base); 1719 end = PFN_DOWN(base + size); 1720 1721 for (; cursor < end; cursor++) { 1722 memblock_free_pages(pfn_to_page(cursor), cursor, 0); 1723 totalram_pages_inc(); 1724 } 1725 } 1726 1727 /* 1728 * Remaining API functions 1729 */ 1730 1731 phys_addr_t __init_memblock memblock_phys_mem_size(void) 1732 { 1733 return memblock.memory.total_size; 1734 } 1735 1736 phys_addr_t __init_memblock memblock_reserved_size(void) 1737 { 1738 return memblock.reserved.total_size; 1739 } 1740 1741 /* lowest address */ 1742 phys_addr_t __init_memblock memblock_start_of_DRAM(void) 1743 { 1744 return memblock.memory.regions[0].base; 1745 } 1746 1747 phys_addr_t __init_memblock memblock_end_of_DRAM(void) 1748 { 1749 int idx = memblock.memory.cnt - 1; 1750 1751 return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size); 1752 } 1753 1754 static phys_addr_t __init_memblock __find_max_addr(phys_addr_t limit) 1755 { 1756 phys_addr_t max_addr = PHYS_ADDR_MAX; 1757 struct memblock_region *r; 1758 1759 /* 1760 * translate the memory @limit size into the max address within one of 1761 * the memory memblock regions, if the @limit exceeds the total size 1762 * of those regions, max_addr will keep original value PHYS_ADDR_MAX 1763 */ 1764 for_each_mem_region(r) { 1765 if (limit <= r->size) { 1766 max_addr = r->base + limit; 1767 break; 1768 } 1769 limit -= r->size; 1770 } 1771 1772 return max_addr; 1773 } 1774 1775 void __init memblock_enforce_memory_limit(phys_addr_t limit) 1776 { 1777 phys_addr_t max_addr; 1778 1779 if (!limit) 1780 return; 1781 1782 max_addr = __find_max_addr(limit); 1783 1784 /* @limit exceeds the total size of the memory, do nothing */ 1785 if (max_addr == PHYS_ADDR_MAX) 1786 return; 1787 1788 /* truncate both memory and reserved regions */ 1789 memblock_remove_range(&memblock.memory, max_addr, 1790 PHYS_ADDR_MAX); 1791 memblock_remove_range(&memblock.reserved, max_addr, 1792 PHYS_ADDR_MAX); 1793 } 1794 1795 void __init memblock_cap_memory_range(phys_addr_t base, phys_addr_t size) 1796 { 1797 int start_rgn, end_rgn; 1798 int i, ret; 1799 1800 if (!size) 1801 return; 1802 1803 if (!memblock_memory->total_size) { 1804 pr_warn("%s: No memory registered yet\n", __func__); 1805 return; 1806 } 1807 1808 ret = memblock_isolate_range(&memblock.memory, base, size, 1809 &start_rgn, &end_rgn); 1810 if (ret) 1811 return; 1812 1813 /* remove all the MAP regions */ 1814 for (i = memblock.memory.cnt - 1; i >= end_rgn; i--) 1815 if (!memblock_is_nomap(&memblock.memory.regions[i])) 1816 memblock_remove_region(&memblock.memory, i); 1817 1818 for (i = start_rgn - 1; i >= 0; i--) 1819 if (!memblock_is_nomap(&memblock.memory.regions[i])) 1820 memblock_remove_region(&memblock.memory, i); 1821 1822 /* truncate the reserved regions */ 1823 memblock_remove_range(&memblock.reserved, 0, base); 1824 memblock_remove_range(&memblock.reserved, 1825 base + size, PHYS_ADDR_MAX); 1826 } 1827 1828 void __init memblock_mem_limit_remove_map(phys_addr_t limit) 1829 { 1830 phys_addr_t max_addr; 1831 1832 if (!limit) 1833 return; 1834 1835 max_addr = __find_max_addr(limit); 1836 1837 /* @limit exceeds the total size of the memory, do nothing */ 1838 if (max_addr == PHYS_ADDR_MAX) 1839 return; 1840 1841 memblock_cap_memory_range(0, max_addr); 1842 } 1843 1844 static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr) 1845 { 1846 unsigned int left = 0, right = type->cnt; 1847 1848 do { 1849 unsigned int mid = (right + left) / 2; 1850 1851 if (addr < type->regions[mid].base) 1852 right = mid; 1853 else if (addr >= (type->regions[mid].base + 1854 type->regions[mid].size)) 1855 left = mid + 1; 1856 else 1857 return mid; 1858 } while (left < right); 1859 return -1; 1860 } 1861 1862 bool __init_memblock memblock_is_reserved(phys_addr_t addr) 1863 { 1864 return memblock_search(&memblock.reserved, addr) != -1; 1865 } 1866 1867 bool __init_memblock memblock_is_memory(phys_addr_t addr) 1868 { 1869 return memblock_search(&memblock.memory, addr) != -1; 1870 } 1871 1872 bool __init_memblock memblock_is_map_memory(phys_addr_t addr) 1873 { 1874 int i = memblock_search(&memblock.memory, addr); 1875 1876 if (i == -1) 1877 return false; 1878 return !memblock_is_nomap(&memblock.memory.regions[i]); 1879 } 1880 1881 int __init_memblock memblock_search_pfn_nid(unsigned long pfn, 1882 unsigned long *start_pfn, unsigned long *end_pfn) 1883 { 1884 struct memblock_type *type = &memblock.memory; 1885 int mid = memblock_search(type, PFN_PHYS(pfn)); 1886 1887 if (mid == -1) 1888 return NUMA_NO_NODE; 1889 1890 *start_pfn = PFN_DOWN(type->regions[mid].base); 1891 *end_pfn = PFN_DOWN(type->regions[mid].base + type->regions[mid].size); 1892 1893 return memblock_get_region_node(&type->regions[mid]); 1894 } 1895 1896 /** 1897 * memblock_is_region_memory - check if a region is a subset of memory 1898 * @base: base of region to check 1899 * @size: size of region to check 1900 * 1901 * Check if the region [@base, @base + @size) is a subset of a memory block. 1902 * 1903 * Return: 1904 * 0 if false, non-zero if true 1905 */ 1906 bool __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size) 1907 { 1908 int idx = memblock_search(&memblock.memory, base); 1909 phys_addr_t end = base + memblock_cap_size(base, &size); 1910 1911 if (idx == -1) 1912 return false; 1913 return (memblock.memory.regions[idx].base + 1914 memblock.memory.regions[idx].size) >= end; 1915 } 1916 1917 /** 1918 * memblock_is_region_reserved - check if a region intersects reserved memory 1919 * @base: base of region to check 1920 * @size: size of region to check 1921 * 1922 * Check if the region [@base, @base + @size) intersects a reserved 1923 * memory block. 1924 * 1925 * Return: 1926 * True if they intersect, false if not. 1927 */ 1928 bool __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size) 1929 { 1930 return memblock_overlaps_region(&memblock.reserved, base, size); 1931 } 1932 1933 void __init_memblock memblock_trim_memory(phys_addr_t align) 1934 { 1935 phys_addr_t start, end, orig_start, orig_end; 1936 struct memblock_region *r; 1937 1938 for_each_mem_region(r) { 1939 orig_start = r->base; 1940 orig_end = r->base + r->size; 1941 start = round_up(orig_start, align); 1942 end = round_down(orig_end, align); 1943 1944 if (start == orig_start && end == orig_end) 1945 continue; 1946 1947 if (start < end) { 1948 r->base = start; 1949 r->size = end - start; 1950 } else { 1951 memblock_remove_region(&memblock.memory, 1952 r - memblock.memory.regions); 1953 r--; 1954 } 1955 } 1956 } 1957 1958 void __init_memblock memblock_set_current_limit(phys_addr_t limit) 1959 { 1960 memblock.current_limit = limit; 1961 } 1962 1963 phys_addr_t __init_memblock memblock_get_current_limit(void) 1964 { 1965 return memblock.current_limit; 1966 } 1967 1968 static void __init_memblock memblock_dump(struct memblock_type *type) 1969 { 1970 phys_addr_t base, end, size; 1971 enum memblock_flags flags; 1972 int idx; 1973 struct memblock_region *rgn; 1974 1975 pr_info(" %s.cnt = 0x%lx\n", type->name, type->cnt); 1976 1977 for_each_memblock_type(idx, type, rgn) { 1978 char nid_buf[32] = ""; 1979 1980 base = rgn->base; 1981 size = rgn->size; 1982 end = base + size - 1; 1983 flags = rgn->flags; 1984 #ifdef CONFIG_NUMA 1985 if (memblock_get_region_node(rgn) != MAX_NUMNODES) 1986 snprintf(nid_buf, sizeof(nid_buf), " on node %d", 1987 memblock_get_region_node(rgn)); 1988 #endif 1989 pr_info(" %s[%#x]\t[%pa-%pa], %pa bytes%s flags: %#x\n", 1990 type->name, idx, &base, &end, &size, nid_buf, flags); 1991 } 1992 } 1993 1994 static void __init_memblock __memblock_dump_all(void) 1995 { 1996 pr_info("MEMBLOCK configuration:\n"); 1997 pr_info(" memory size = %pa reserved size = %pa\n", 1998 &memblock.memory.total_size, 1999 &memblock.reserved.total_size); 2000 2001 memblock_dump(&memblock.memory); 2002 memblock_dump(&memblock.reserved); 2003 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP 2004 memblock_dump(&physmem); 2005 #endif 2006 } 2007 2008 void __init_memblock memblock_dump_all(void) 2009 { 2010 if (memblock_debug) 2011 __memblock_dump_all(); 2012 } 2013 2014 void __init memblock_allow_resize(void) 2015 { 2016 memblock_can_resize = 1; 2017 } 2018 2019 static int __init early_memblock(char *p) 2020 { 2021 if (p && strstr(p, "debug")) 2022 memblock_debug = 1; 2023 return 0; 2024 } 2025 early_param("memblock", early_memblock); 2026 2027 static void __init free_memmap(unsigned long start_pfn, unsigned long end_pfn) 2028 { 2029 struct page *start_pg, *end_pg; 2030 phys_addr_t pg, pgend; 2031 2032 /* 2033 * Convert start_pfn/end_pfn to a struct page pointer. 2034 */ 2035 start_pg = pfn_to_page(start_pfn - 1) + 1; 2036 end_pg = pfn_to_page(end_pfn - 1) + 1; 2037 2038 /* 2039 * Convert to physical addresses, and round start upwards and end 2040 * downwards. 2041 */ 2042 pg = PAGE_ALIGN(__pa(start_pg)); 2043 pgend = __pa(end_pg) & PAGE_MASK; 2044 2045 /* 2046 * If there are free pages between these, free the section of the 2047 * memmap array. 2048 */ 2049 if (pg < pgend) 2050 memblock_phys_free(pg, pgend - pg); 2051 } 2052 2053 /* 2054 * The mem_map array can get very big. Free the unused area of the memory map. 2055 */ 2056 static void __init free_unused_memmap(void) 2057 { 2058 unsigned long start, end, prev_end = 0; 2059 int i; 2060 2061 if (!IS_ENABLED(CONFIG_HAVE_ARCH_PFN_VALID) || 2062 IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP)) 2063 return; 2064 2065 /* 2066 * This relies on each bank being in address order. 2067 * The banks are sorted previously in bootmem_init(). 2068 */ 2069 for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, NULL) { 2070 #ifdef CONFIG_SPARSEMEM 2071 /* 2072 * Take care not to free memmap entries that don't exist 2073 * due to SPARSEMEM sections which aren't present. 2074 */ 2075 start = min(start, ALIGN(prev_end, PAGES_PER_SECTION)); 2076 #endif 2077 /* 2078 * Align down here since many operations in VM subsystem 2079 * presume that there are no holes in the memory map inside 2080 * a pageblock 2081 */ 2082 start = pageblock_start_pfn(start); 2083 2084 /* 2085 * If we had a previous bank, and there is a space 2086 * between the current bank and the previous, free it. 2087 */ 2088 if (prev_end && prev_end < start) 2089 free_memmap(prev_end, start); 2090 2091 /* 2092 * Align up here since many operations in VM subsystem 2093 * presume that there are no holes in the memory map inside 2094 * a pageblock 2095 */ 2096 prev_end = pageblock_align(end); 2097 } 2098 2099 #ifdef CONFIG_SPARSEMEM 2100 if (!IS_ALIGNED(prev_end, PAGES_PER_SECTION)) { 2101 prev_end = pageblock_align(end); 2102 free_memmap(prev_end, ALIGN(prev_end, PAGES_PER_SECTION)); 2103 } 2104 #endif 2105 } 2106 2107 static void __init __free_pages_memory(unsigned long start, unsigned long end) 2108 { 2109 int order; 2110 2111 while (start < end) { 2112 /* 2113 * Free the pages in the largest chunks alignment allows. 2114 * 2115 * __ffs() behaviour is undefined for 0. start == 0 is 2116 * MAX_PAGE_ORDER-aligned, set order to MAX_PAGE_ORDER for 2117 * the case. 2118 */ 2119 if (start) 2120 order = min_t(int, MAX_PAGE_ORDER, __ffs(start)); 2121 else 2122 order = MAX_PAGE_ORDER; 2123 2124 while (start + (1UL << order) > end) 2125 order--; 2126 2127 memblock_free_pages(pfn_to_page(start), start, order); 2128 2129 start += (1UL << order); 2130 } 2131 } 2132 2133 static unsigned long __init __free_memory_core(phys_addr_t start, 2134 phys_addr_t end) 2135 { 2136 unsigned long start_pfn = PFN_UP(start); 2137 unsigned long end_pfn = min_t(unsigned long, 2138 PFN_DOWN(end), max_low_pfn); 2139 2140 if (start_pfn >= end_pfn) 2141 return 0; 2142 2143 __free_pages_memory(start_pfn, end_pfn); 2144 2145 return end_pfn - start_pfn; 2146 } 2147 2148 static void __init memmap_init_reserved_pages(void) 2149 { 2150 struct memblock_region *region; 2151 phys_addr_t start, end; 2152 int nid; 2153 2154 /* 2155 * set nid on all reserved pages and also treat struct 2156 * pages for the NOMAP regions as PageReserved 2157 */ 2158 for_each_mem_region(region) { 2159 nid = memblock_get_region_node(region); 2160 start = region->base; 2161 end = start + region->size; 2162 2163 if (memblock_is_nomap(region)) 2164 reserve_bootmem_region(start, end, nid); 2165 2166 memblock_set_node(start, end, &memblock.reserved, nid); 2167 } 2168 2169 /* 2170 * initialize struct pages for reserved regions that don't have 2171 * the MEMBLOCK_RSRV_NOINIT flag set 2172 */ 2173 for_each_reserved_mem_region(region) { 2174 if (!memblock_is_reserved_noinit(region)) { 2175 nid = memblock_get_region_node(region); 2176 start = region->base; 2177 end = start + region->size; 2178 2179 reserve_bootmem_region(start, end, nid); 2180 } 2181 } 2182 } 2183 2184 static unsigned long __init free_low_memory_core_early(void) 2185 { 2186 unsigned long count = 0; 2187 phys_addr_t start, end; 2188 u64 i; 2189 2190 memblock_clear_hotplug(0, -1); 2191 2192 memmap_init_reserved_pages(); 2193 2194 /* 2195 * We need to use NUMA_NO_NODE instead of NODE_DATA(0)->node_id 2196 * because in some case like Node0 doesn't have RAM installed 2197 * low ram will be on Node1 2198 */ 2199 for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE, &start, &end, 2200 NULL) 2201 count += __free_memory_core(start, end); 2202 2203 return count; 2204 } 2205 2206 static int reset_managed_pages_done __initdata; 2207 2208 static void __init reset_node_managed_pages(pg_data_t *pgdat) 2209 { 2210 struct zone *z; 2211 2212 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++) 2213 atomic_long_set(&z->managed_pages, 0); 2214 } 2215 2216 void __init reset_all_zones_managed_pages(void) 2217 { 2218 struct pglist_data *pgdat; 2219 2220 if (reset_managed_pages_done) 2221 return; 2222 2223 for_each_online_pgdat(pgdat) 2224 reset_node_managed_pages(pgdat); 2225 2226 reset_managed_pages_done = 1; 2227 } 2228 2229 /** 2230 * memblock_free_all - release free pages to the buddy allocator 2231 */ 2232 void __init memblock_free_all(void) 2233 { 2234 unsigned long pages; 2235 2236 free_unused_memmap(); 2237 reset_all_zones_managed_pages(); 2238 2239 pages = free_low_memory_core_early(); 2240 totalram_pages_add(pages); 2241 } 2242 2243 #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_ARCH_KEEP_MEMBLOCK) 2244 static const char * const flagname[] = { 2245 [ilog2(MEMBLOCK_HOTPLUG)] = "HOTPLUG", 2246 [ilog2(MEMBLOCK_MIRROR)] = "MIRROR", 2247 [ilog2(MEMBLOCK_NOMAP)] = "NOMAP", 2248 [ilog2(MEMBLOCK_DRIVER_MANAGED)] = "DRV_MNG", 2249 }; 2250 2251 static int memblock_debug_show(struct seq_file *m, void *private) 2252 { 2253 struct memblock_type *type = m->private; 2254 struct memblock_region *reg; 2255 int i, j, nid; 2256 unsigned int count = ARRAY_SIZE(flagname); 2257 phys_addr_t end; 2258 2259 for (i = 0; i < type->cnt; i++) { 2260 reg = &type->regions[i]; 2261 end = reg->base + reg->size - 1; 2262 nid = memblock_get_region_node(reg); 2263 2264 seq_printf(m, "%4d: ", i); 2265 seq_printf(m, "%pa..%pa ", ®->base, &end); 2266 if (nid != MAX_NUMNODES) 2267 seq_printf(m, "%4d ", nid); 2268 else 2269 seq_printf(m, "%4c ", 'x'); 2270 if (reg->flags) { 2271 for (j = 0; j < count; j++) { 2272 if (reg->flags & (1U << j)) { 2273 seq_printf(m, "%s\n", flagname[j]); 2274 break; 2275 } 2276 } 2277 if (j == count) 2278 seq_printf(m, "%s\n", "UNKNOWN"); 2279 } else { 2280 seq_printf(m, "%s\n", "NONE"); 2281 } 2282 } 2283 return 0; 2284 } 2285 DEFINE_SHOW_ATTRIBUTE(memblock_debug); 2286 2287 static int __init memblock_init_debugfs(void) 2288 { 2289 struct dentry *root = debugfs_create_dir("memblock", NULL); 2290 2291 debugfs_create_file("memory", 0444, root, 2292 &memblock.memory, &memblock_debug_fops); 2293 debugfs_create_file("reserved", 0444, root, 2294 &memblock.reserved, &memblock_debug_fops); 2295 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP 2296 debugfs_create_file("physmem", 0444, root, &physmem, 2297 &memblock_debug_fops); 2298 #endif 2299 2300 return 0; 2301 } 2302 __initcall(memblock_init_debugfs); 2303 2304 #endif /* CONFIG_DEBUG_FS */ 2305