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 unsigned long __init_memblock 184 memblock_addrs_overlap(phys_addr_t base1, phys_addr_t size1, phys_addr_t base2, 185 phys_addr_t size2) 186 { 187 return ((base1 < (base2 + size2)) && (base2 < (base1 + size1))); 188 } 189 190 bool __init_memblock memblock_overlaps_region(struct memblock_type *type, 191 phys_addr_t base, phys_addr_t size) 192 { 193 unsigned long i; 194 195 memblock_cap_size(base, &size); 196 197 for (i = 0; i < type->cnt; i++) 198 if (memblock_addrs_overlap(base, size, type->regions[i].base, 199 type->regions[i].size)) 200 break; 201 return i < type->cnt; 202 } 203 204 /** 205 * __memblock_find_range_bottom_up - find free area utility in bottom-up 206 * @start: start of candidate range 207 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or 208 * %MEMBLOCK_ALLOC_ACCESSIBLE 209 * @size: size of free area to find 210 * @align: alignment of free area to find 211 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 212 * @flags: pick from blocks based on memory attributes 213 * 214 * Utility called from memblock_find_in_range_node(), find free area bottom-up. 215 * 216 * Return: 217 * Found address on success, 0 on failure. 218 */ 219 static phys_addr_t __init_memblock 220 __memblock_find_range_bottom_up(phys_addr_t start, phys_addr_t end, 221 phys_addr_t size, phys_addr_t align, int nid, 222 enum memblock_flags flags) 223 { 224 phys_addr_t this_start, this_end, cand; 225 u64 i; 226 227 for_each_free_mem_range(i, nid, flags, &this_start, &this_end, NULL) { 228 this_start = clamp(this_start, start, end); 229 this_end = clamp(this_end, start, end); 230 231 cand = round_up(this_start, align); 232 if (cand < this_end && this_end - cand >= size) 233 return cand; 234 } 235 236 return 0; 237 } 238 239 /** 240 * __memblock_find_range_top_down - find free area utility, in top-down 241 * @start: start of candidate range 242 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or 243 * %MEMBLOCK_ALLOC_ACCESSIBLE 244 * @size: size of free area to find 245 * @align: alignment of free area to find 246 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 247 * @flags: pick from blocks based on memory attributes 248 * 249 * Utility called from memblock_find_in_range_node(), find free area top-down. 250 * 251 * Return: 252 * Found address on success, 0 on failure. 253 */ 254 static phys_addr_t __init_memblock 255 __memblock_find_range_top_down(phys_addr_t start, phys_addr_t end, 256 phys_addr_t size, phys_addr_t align, int nid, 257 enum memblock_flags flags) 258 { 259 phys_addr_t this_start, this_end, cand; 260 u64 i; 261 262 for_each_free_mem_range_reverse(i, nid, flags, &this_start, &this_end, 263 NULL) { 264 this_start = clamp(this_start, start, end); 265 this_end = clamp(this_end, start, end); 266 267 if (this_end < size) 268 continue; 269 270 cand = round_down(this_end - size, align); 271 if (cand >= this_start) 272 return cand; 273 } 274 275 return 0; 276 } 277 278 /** 279 * memblock_find_in_range_node - find free area in given range and node 280 * @size: size of free area to find 281 * @align: alignment of free area to find 282 * @start: start of candidate range 283 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or 284 * %MEMBLOCK_ALLOC_ACCESSIBLE 285 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 286 * @flags: pick from blocks based on memory attributes 287 * 288 * Find @size free area aligned to @align in the specified range and node. 289 * 290 * Return: 291 * Found address on success, 0 on failure. 292 */ 293 static phys_addr_t __init_memblock memblock_find_in_range_node(phys_addr_t size, 294 phys_addr_t align, phys_addr_t start, 295 phys_addr_t end, int nid, 296 enum memblock_flags flags) 297 { 298 /* pump up @end */ 299 if (end == MEMBLOCK_ALLOC_ACCESSIBLE || 300 end == MEMBLOCK_ALLOC_NOLEAKTRACE) 301 end = memblock.current_limit; 302 303 /* avoid allocating the first page */ 304 start = max_t(phys_addr_t, start, PAGE_SIZE); 305 end = max(start, end); 306 307 if (memblock_bottom_up()) 308 return __memblock_find_range_bottom_up(start, end, size, align, 309 nid, flags); 310 else 311 return __memblock_find_range_top_down(start, end, size, align, 312 nid, flags); 313 } 314 315 /** 316 * memblock_find_in_range - find free area in given range 317 * @start: start of candidate range 318 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or 319 * %MEMBLOCK_ALLOC_ACCESSIBLE 320 * @size: size of free area to find 321 * @align: alignment of free area to find 322 * 323 * Find @size free area aligned to @align in the specified range. 324 * 325 * Return: 326 * Found address on success, 0 on failure. 327 */ 328 static phys_addr_t __init_memblock memblock_find_in_range(phys_addr_t start, 329 phys_addr_t end, phys_addr_t size, 330 phys_addr_t align) 331 { 332 phys_addr_t ret; 333 enum memblock_flags flags = choose_memblock_flags(); 334 335 again: 336 ret = memblock_find_in_range_node(size, align, start, end, 337 NUMA_NO_NODE, flags); 338 339 if (!ret && (flags & MEMBLOCK_MIRROR)) { 340 pr_warn_ratelimited("Could not allocate %pap bytes of mirrored memory\n", 341 &size); 342 flags &= ~MEMBLOCK_MIRROR; 343 goto again; 344 } 345 346 return ret; 347 } 348 349 static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r) 350 { 351 type->total_size -= type->regions[r].size; 352 memmove(&type->regions[r], &type->regions[r + 1], 353 (type->cnt - (r + 1)) * sizeof(type->regions[r])); 354 type->cnt--; 355 356 /* Special case for empty arrays */ 357 if (type->cnt == 0) { 358 WARN_ON(type->total_size != 0); 359 type->cnt = 1; 360 type->regions[0].base = 0; 361 type->regions[0].size = 0; 362 type->regions[0].flags = 0; 363 memblock_set_region_node(&type->regions[0], MAX_NUMNODES); 364 } 365 } 366 367 #ifndef CONFIG_ARCH_KEEP_MEMBLOCK 368 /** 369 * memblock_discard - discard memory and reserved arrays if they were allocated 370 */ 371 void __init memblock_discard(void) 372 { 373 phys_addr_t addr, size; 374 375 if (memblock.reserved.regions != memblock_reserved_init_regions) { 376 addr = __pa(memblock.reserved.regions); 377 size = PAGE_ALIGN(sizeof(struct memblock_region) * 378 memblock.reserved.max); 379 if (memblock_reserved_in_slab) 380 kfree(memblock.reserved.regions); 381 else 382 memblock_free_late(addr, size); 383 } 384 385 if (memblock.memory.regions != memblock_memory_init_regions) { 386 addr = __pa(memblock.memory.regions); 387 size = PAGE_ALIGN(sizeof(struct memblock_region) * 388 memblock.memory.max); 389 if (memblock_memory_in_slab) 390 kfree(memblock.memory.regions); 391 else 392 memblock_free_late(addr, size); 393 } 394 395 memblock_memory = NULL; 396 } 397 #endif 398 399 /** 400 * memblock_double_array - double the size of the memblock regions array 401 * @type: memblock type of the regions array being doubled 402 * @new_area_start: starting address of memory range to avoid overlap with 403 * @new_area_size: size of memory range to avoid overlap with 404 * 405 * Double the size of the @type regions array. If memblock is being used to 406 * allocate memory for a new reserved regions array and there is a previously 407 * allocated memory range [@new_area_start, @new_area_start + @new_area_size] 408 * waiting to be reserved, ensure the memory used by the new array does 409 * not overlap. 410 * 411 * Return: 412 * 0 on success, -1 on failure. 413 */ 414 static int __init_memblock memblock_double_array(struct memblock_type *type, 415 phys_addr_t new_area_start, 416 phys_addr_t new_area_size) 417 { 418 struct memblock_region *new_array, *old_array; 419 phys_addr_t old_alloc_size, new_alloc_size; 420 phys_addr_t old_size, new_size, addr, new_end; 421 int use_slab = slab_is_available(); 422 int *in_slab; 423 424 /* We don't allow resizing until we know about the reserved regions 425 * of memory that aren't suitable for allocation 426 */ 427 if (!memblock_can_resize) 428 panic("memblock: cannot resize %s array\n", type->name); 429 430 /* Calculate new doubled size */ 431 old_size = type->max * sizeof(struct memblock_region); 432 new_size = old_size << 1; 433 /* 434 * We need to allocated new one align to PAGE_SIZE, 435 * so we can free them completely later. 436 */ 437 old_alloc_size = PAGE_ALIGN(old_size); 438 new_alloc_size = PAGE_ALIGN(new_size); 439 440 /* Retrieve the slab flag */ 441 if (type == &memblock.memory) 442 in_slab = &memblock_memory_in_slab; 443 else 444 in_slab = &memblock_reserved_in_slab; 445 446 /* Try to find some space for it */ 447 if (use_slab) { 448 new_array = kmalloc(new_size, GFP_KERNEL); 449 addr = new_array ? __pa(new_array) : 0; 450 } else { 451 /* only exclude range when trying to double reserved.regions */ 452 if (type != &memblock.reserved) 453 new_area_start = new_area_size = 0; 454 455 addr = memblock_find_in_range(new_area_start + new_area_size, 456 memblock.current_limit, 457 new_alloc_size, PAGE_SIZE); 458 if (!addr && new_area_size) 459 addr = memblock_find_in_range(0, 460 min(new_area_start, memblock.current_limit), 461 new_alloc_size, PAGE_SIZE); 462 463 new_array = addr ? __va(addr) : NULL; 464 } 465 if (!addr) { 466 pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n", 467 type->name, type->max, type->max * 2); 468 return -1; 469 } 470 471 new_end = addr + new_size - 1; 472 memblock_dbg("memblock: %s is doubled to %ld at [%pa-%pa]", 473 type->name, type->max * 2, &addr, &new_end); 474 475 /* 476 * Found space, we now need to move the array over before we add the 477 * reserved region since it may be our reserved array itself that is 478 * full. 479 */ 480 memcpy(new_array, type->regions, old_size); 481 memset(new_array + type->max, 0, old_size); 482 old_array = type->regions; 483 type->regions = new_array; 484 type->max <<= 1; 485 486 /* Free old array. We needn't free it if the array is the static one */ 487 if (*in_slab) 488 kfree(old_array); 489 else if (old_array != memblock_memory_init_regions && 490 old_array != memblock_reserved_init_regions) 491 memblock_free(old_array, old_alloc_size); 492 493 /* 494 * Reserve the new array if that comes from the memblock. Otherwise, we 495 * needn't do it 496 */ 497 if (!use_slab) 498 BUG_ON(memblock_reserve(addr, new_alloc_size)); 499 500 /* Update slab flag */ 501 *in_slab = use_slab; 502 503 return 0; 504 } 505 506 /** 507 * memblock_merge_regions - merge neighboring compatible regions 508 * @type: memblock type to scan 509 * @start_rgn: start scanning from (@start_rgn - 1) 510 * @end_rgn: end scanning at (@end_rgn - 1) 511 * Scan @type and merge neighboring compatible regions in [@start_rgn - 1, @end_rgn) 512 */ 513 static void __init_memblock memblock_merge_regions(struct memblock_type *type, 514 unsigned long start_rgn, 515 unsigned long end_rgn) 516 { 517 int i = 0; 518 if (start_rgn) 519 i = start_rgn - 1; 520 end_rgn = min(end_rgn, type->cnt - 1); 521 while (i < end_rgn) { 522 struct memblock_region *this = &type->regions[i]; 523 struct memblock_region *next = &type->regions[i + 1]; 524 525 if (this->base + this->size != next->base || 526 memblock_get_region_node(this) != 527 memblock_get_region_node(next) || 528 this->flags != next->flags) { 529 BUG_ON(this->base + this->size > next->base); 530 i++; 531 continue; 532 } 533 534 this->size += next->size; 535 /* move forward from next + 1, index of which is i + 2 */ 536 memmove(next, next + 1, (type->cnt - (i + 2)) * sizeof(*next)); 537 type->cnt--; 538 end_rgn--; 539 } 540 } 541 542 /** 543 * memblock_insert_region - insert new memblock region 544 * @type: memblock type to insert into 545 * @idx: index for the insertion point 546 * @base: base address of the new region 547 * @size: size of the new region 548 * @nid: node id of the new region 549 * @flags: flags of the new region 550 * 551 * Insert new memblock region [@base, @base + @size) into @type at @idx. 552 * @type must already have extra room to accommodate the new region. 553 */ 554 static void __init_memblock memblock_insert_region(struct memblock_type *type, 555 int idx, phys_addr_t base, 556 phys_addr_t size, 557 int nid, 558 enum memblock_flags flags) 559 { 560 struct memblock_region *rgn = &type->regions[idx]; 561 562 BUG_ON(type->cnt >= type->max); 563 memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn)); 564 rgn->base = base; 565 rgn->size = size; 566 rgn->flags = flags; 567 memblock_set_region_node(rgn, nid); 568 type->cnt++; 569 type->total_size += size; 570 } 571 572 /** 573 * memblock_add_range - add new memblock region 574 * @type: memblock type to add new region into 575 * @base: base address of the new region 576 * @size: size of the new region 577 * @nid: nid of the new region 578 * @flags: flags of the new region 579 * 580 * Add new memblock region [@base, @base + @size) into @type. The new region 581 * is allowed to overlap with existing ones - overlaps don't affect already 582 * existing regions. @type is guaranteed to be minimal (all neighbouring 583 * compatible regions are merged) after the addition. 584 * 585 * Return: 586 * 0 on success, -errno on failure. 587 */ 588 static int __init_memblock memblock_add_range(struct memblock_type *type, 589 phys_addr_t base, phys_addr_t size, 590 int nid, enum memblock_flags flags) 591 { 592 bool insert = false; 593 phys_addr_t obase = base; 594 phys_addr_t end = base + memblock_cap_size(base, &size); 595 int idx, nr_new, start_rgn = -1, end_rgn; 596 struct memblock_region *rgn; 597 598 if (!size) 599 return 0; 600 601 /* special case for empty array */ 602 if (type->regions[0].size == 0) { 603 WARN_ON(type->cnt != 1 || type->total_size); 604 type->regions[0].base = base; 605 type->regions[0].size = size; 606 type->regions[0].flags = flags; 607 memblock_set_region_node(&type->regions[0], nid); 608 type->total_size = size; 609 return 0; 610 } 611 612 /* 613 * The worst case is when new range overlaps all existing regions, 614 * then we'll need type->cnt + 1 empty regions in @type. So if 615 * type->cnt * 2 + 1 is less than or equal to type->max, we know 616 * that there is enough empty regions in @type, and we can insert 617 * regions directly. 618 */ 619 if (type->cnt * 2 + 1 <= type->max) 620 insert = true; 621 622 repeat: 623 /* 624 * The following is executed twice. Once with %false @insert and 625 * then with %true. The first counts the number of regions needed 626 * to accommodate the new area. The second actually inserts them. 627 */ 628 base = obase; 629 nr_new = 0; 630 631 for_each_memblock_type(idx, type, rgn) { 632 phys_addr_t rbase = rgn->base; 633 phys_addr_t rend = rbase + rgn->size; 634 635 if (rbase >= end) 636 break; 637 if (rend <= base) 638 continue; 639 /* 640 * @rgn overlaps. If it separates the lower part of new 641 * area, insert that portion. 642 */ 643 if (rbase > base) { 644 #ifdef CONFIG_NUMA 645 WARN_ON(nid != memblock_get_region_node(rgn)); 646 #endif 647 WARN_ON(flags != rgn->flags); 648 nr_new++; 649 if (insert) { 650 if (start_rgn == -1) 651 start_rgn = idx; 652 end_rgn = idx + 1; 653 memblock_insert_region(type, idx++, base, 654 rbase - base, nid, 655 flags); 656 } 657 } 658 /* area below @rend is dealt with, forget about it */ 659 base = min(rend, end); 660 } 661 662 /* insert the remaining portion */ 663 if (base < end) { 664 nr_new++; 665 if (insert) { 666 if (start_rgn == -1) 667 start_rgn = idx; 668 end_rgn = idx + 1; 669 memblock_insert_region(type, idx, base, end - base, 670 nid, flags); 671 } 672 } 673 674 if (!nr_new) 675 return 0; 676 677 /* 678 * If this was the first round, resize array and repeat for actual 679 * insertions; otherwise, merge and return. 680 */ 681 if (!insert) { 682 while (type->cnt + nr_new > type->max) 683 if (memblock_double_array(type, obase, size) < 0) 684 return -ENOMEM; 685 insert = true; 686 goto repeat; 687 } else { 688 memblock_merge_regions(type, start_rgn, end_rgn); 689 return 0; 690 } 691 } 692 693 /** 694 * memblock_add_node - add new memblock region within a NUMA node 695 * @base: base address of the new region 696 * @size: size of the new region 697 * @nid: nid of the new region 698 * @flags: flags of the new region 699 * 700 * Add new memblock region [@base, @base + @size) to the "memory" 701 * type. See memblock_add_range() description for mode details 702 * 703 * Return: 704 * 0 on success, -errno on failure. 705 */ 706 int __init_memblock memblock_add_node(phys_addr_t base, phys_addr_t size, 707 int nid, enum memblock_flags flags) 708 { 709 phys_addr_t end = base + size - 1; 710 711 memblock_dbg("%s: [%pa-%pa] nid=%d flags=%x %pS\n", __func__, 712 &base, &end, nid, flags, (void *)_RET_IP_); 713 714 return memblock_add_range(&memblock.memory, base, size, nid, flags); 715 } 716 717 /** 718 * memblock_add - add new memblock region 719 * @base: base address of the new region 720 * @size: size of the new region 721 * 722 * Add new memblock region [@base, @base + @size) to the "memory" 723 * type. See memblock_add_range() description for mode details 724 * 725 * Return: 726 * 0 on success, -errno on failure. 727 */ 728 int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size) 729 { 730 phys_addr_t end = base + size - 1; 731 732 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__, 733 &base, &end, (void *)_RET_IP_); 734 735 return memblock_add_range(&memblock.memory, base, size, MAX_NUMNODES, 0); 736 } 737 738 /** 739 * memblock_validate_numa_coverage - check if amount of memory with 740 * no node ID assigned is less than a threshold 741 * @threshold_bytes: maximal number of pages that can have unassigned node 742 * ID (in bytes). 743 * 744 * A buggy firmware may report memory that does not belong to any node. 745 * Check if amount of such memory is below @threshold_bytes. 746 * 747 * Return: true on success, false on failure. 748 */ 749 bool __init_memblock memblock_validate_numa_coverage(unsigned long threshold_bytes) 750 { 751 unsigned long nr_pages = 0; 752 unsigned long start_pfn, end_pfn, mem_size_mb; 753 int nid, i; 754 755 /* calculate lose page */ 756 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) { 757 if (!numa_valid_node(nid)) 758 nr_pages += end_pfn - start_pfn; 759 } 760 761 if ((nr_pages << PAGE_SHIFT) >= threshold_bytes) { 762 mem_size_mb = memblock_phys_mem_size() >> 20; 763 pr_err("NUMA: no nodes coverage for %luMB of %luMB RAM\n", 764 (nr_pages << PAGE_SHIFT) >> 20, mem_size_mb); 765 return false; 766 } 767 768 return true; 769 } 770 771 772 /** 773 * memblock_isolate_range - isolate given range into disjoint memblocks 774 * @type: memblock type to isolate range for 775 * @base: base of range to isolate 776 * @size: size of range to isolate 777 * @start_rgn: out parameter for the start of isolated region 778 * @end_rgn: out parameter for the end of isolated region 779 * 780 * Walk @type and ensure that regions don't cross the boundaries defined by 781 * [@base, @base + @size). Crossing regions are split at the boundaries, 782 * which may create at most two more regions. The index of the first 783 * region inside the range is returned in *@start_rgn and end in *@end_rgn. 784 * 785 * Return: 786 * 0 on success, -errno on failure. 787 */ 788 static int __init_memblock memblock_isolate_range(struct memblock_type *type, 789 phys_addr_t base, phys_addr_t size, 790 int *start_rgn, int *end_rgn) 791 { 792 phys_addr_t end = base + memblock_cap_size(base, &size); 793 int idx; 794 struct memblock_region *rgn; 795 796 *start_rgn = *end_rgn = 0; 797 798 if (!size) 799 return 0; 800 801 /* we'll create at most two more regions */ 802 while (type->cnt + 2 > type->max) 803 if (memblock_double_array(type, base, size) < 0) 804 return -ENOMEM; 805 806 for_each_memblock_type(idx, type, rgn) { 807 phys_addr_t rbase = rgn->base; 808 phys_addr_t rend = rbase + rgn->size; 809 810 if (rbase >= end) 811 break; 812 if (rend <= base) 813 continue; 814 815 if (rbase < base) { 816 /* 817 * @rgn intersects from below. Split and continue 818 * to process the next region - the new top half. 819 */ 820 rgn->base = base; 821 rgn->size -= base - rbase; 822 type->total_size -= base - rbase; 823 memblock_insert_region(type, idx, rbase, base - rbase, 824 memblock_get_region_node(rgn), 825 rgn->flags); 826 } else if (rend > end) { 827 /* 828 * @rgn intersects from above. Split and redo the 829 * current region - the new bottom half. 830 */ 831 rgn->base = end; 832 rgn->size -= end - rbase; 833 type->total_size -= end - rbase; 834 memblock_insert_region(type, idx--, rbase, end - rbase, 835 memblock_get_region_node(rgn), 836 rgn->flags); 837 } else { 838 /* @rgn is fully contained, record it */ 839 if (!*end_rgn) 840 *start_rgn = idx; 841 *end_rgn = idx + 1; 842 } 843 } 844 845 return 0; 846 } 847 848 static int __init_memblock memblock_remove_range(struct memblock_type *type, 849 phys_addr_t base, phys_addr_t size) 850 { 851 int start_rgn, end_rgn; 852 int i, ret; 853 854 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn); 855 if (ret) 856 return ret; 857 858 for (i = end_rgn - 1; i >= start_rgn; i--) 859 memblock_remove_region(type, i); 860 return 0; 861 } 862 863 int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size) 864 { 865 phys_addr_t end = base + size - 1; 866 867 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__, 868 &base, &end, (void *)_RET_IP_); 869 870 return memblock_remove_range(&memblock.memory, base, size); 871 } 872 873 /** 874 * memblock_free - free boot memory allocation 875 * @ptr: starting address of the boot memory allocation 876 * @size: size of the boot memory block in bytes 877 * 878 * Free boot memory block previously allocated by memblock_alloc_xx() API. 879 * The freeing memory will not be released to the buddy allocator. 880 */ 881 void __init_memblock memblock_free(void *ptr, size_t size) 882 { 883 if (ptr) 884 memblock_phys_free(__pa(ptr), size); 885 } 886 887 /** 888 * memblock_phys_free - free boot memory block 889 * @base: phys starting address of the boot memory block 890 * @size: size of the boot memory block in bytes 891 * 892 * Free boot memory block previously allocated by memblock_phys_alloc_xx() API. 893 * The freeing memory will not be released to the buddy allocator. 894 */ 895 int __init_memblock memblock_phys_free(phys_addr_t base, phys_addr_t size) 896 { 897 phys_addr_t end = base + size - 1; 898 899 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__, 900 &base, &end, (void *)_RET_IP_); 901 902 kmemleak_free_part_phys(base, size); 903 return memblock_remove_range(&memblock.reserved, base, size); 904 } 905 906 int __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size) 907 { 908 phys_addr_t end = base + size - 1; 909 910 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__, 911 &base, &end, (void *)_RET_IP_); 912 913 return memblock_add_range(&memblock.reserved, base, size, MAX_NUMNODES, 0); 914 } 915 916 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP 917 int __init_memblock memblock_physmem_add(phys_addr_t base, phys_addr_t size) 918 { 919 phys_addr_t end = base + size - 1; 920 921 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__, 922 &base, &end, (void *)_RET_IP_); 923 924 return memblock_add_range(&physmem, base, size, MAX_NUMNODES, 0); 925 } 926 #endif 927 928 /** 929 * memblock_setclr_flag - set or clear flag for a memory region 930 * @type: memblock type to set/clear flag for 931 * @base: base address of the region 932 * @size: size of the region 933 * @set: set or clear the flag 934 * @flag: the flag to update 935 * 936 * This function isolates region [@base, @base + @size), and sets/clears flag 937 * 938 * Return: 0 on success, -errno on failure. 939 */ 940 static int __init_memblock memblock_setclr_flag(struct memblock_type *type, 941 phys_addr_t base, phys_addr_t size, int set, int flag) 942 { 943 int i, ret, start_rgn, end_rgn; 944 945 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn); 946 if (ret) 947 return ret; 948 949 for (i = start_rgn; i < end_rgn; i++) { 950 struct memblock_region *r = &type->regions[i]; 951 952 if (set) 953 r->flags |= flag; 954 else 955 r->flags &= ~flag; 956 } 957 958 memblock_merge_regions(type, start_rgn, end_rgn); 959 return 0; 960 } 961 962 /** 963 * memblock_mark_hotplug - Mark hotpluggable memory with flag MEMBLOCK_HOTPLUG. 964 * @base: the base phys addr of the region 965 * @size: the size of the region 966 * 967 * Return: 0 on success, -errno on failure. 968 */ 969 int __init_memblock memblock_mark_hotplug(phys_addr_t base, phys_addr_t size) 970 { 971 return memblock_setclr_flag(&memblock.memory, base, size, 1, MEMBLOCK_HOTPLUG); 972 } 973 974 /** 975 * memblock_clear_hotplug - Clear flag MEMBLOCK_HOTPLUG for a specified region. 976 * @base: the base phys addr of the region 977 * @size: the size of the region 978 * 979 * Return: 0 on success, -errno on failure. 980 */ 981 int __init_memblock memblock_clear_hotplug(phys_addr_t base, phys_addr_t size) 982 { 983 return memblock_setclr_flag(&memblock.memory, base, size, 0, MEMBLOCK_HOTPLUG); 984 } 985 986 /** 987 * memblock_mark_mirror - Mark mirrored memory with flag MEMBLOCK_MIRROR. 988 * @base: the base phys addr of the region 989 * @size: the size of the region 990 * 991 * Return: 0 on success, -errno on failure. 992 */ 993 int __init_memblock memblock_mark_mirror(phys_addr_t base, phys_addr_t size) 994 { 995 if (!mirrored_kernelcore) 996 return 0; 997 998 system_has_some_mirror = true; 999 1000 return memblock_setclr_flag(&memblock.memory, base, size, 1, MEMBLOCK_MIRROR); 1001 } 1002 1003 /** 1004 * memblock_mark_nomap - Mark a memory region with flag MEMBLOCK_NOMAP. 1005 * @base: the base phys addr of the region 1006 * @size: the size of the region 1007 * 1008 * The memory regions marked with %MEMBLOCK_NOMAP will not be added to the 1009 * direct mapping of the physical memory. These regions will still be 1010 * covered by the memory map. The struct page representing NOMAP memory 1011 * frames in the memory map will be PageReserved() 1012 * 1013 * Note: if the memory being marked %MEMBLOCK_NOMAP was allocated from 1014 * memblock, the caller must inform kmemleak to ignore that memory 1015 * 1016 * Return: 0 on success, -errno on failure. 1017 */ 1018 int __init_memblock memblock_mark_nomap(phys_addr_t base, phys_addr_t size) 1019 { 1020 return memblock_setclr_flag(&memblock.memory, base, size, 1, MEMBLOCK_NOMAP); 1021 } 1022 1023 /** 1024 * memblock_clear_nomap - Clear flag MEMBLOCK_NOMAP for a specified region. 1025 * @base: the base phys addr of the region 1026 * @size: the size of the region 1027 * 1028 * Return: 0 on success, -errno on failure. 1029 */ 1030 int __init_memblock memblock_clear_nomap(phys_addr_t base, phys_addr_t size) 1031 { 1032 return memblock_setclr_flag(&memblock.memory, base, size, 0, MEMBLOCK_NOMAP); 1033 } 1034 1035 /** 1036 * memblock_reserved_mark_noinit - Mark a reserved memory region with flag 1037 * MEMBLOCK_RSRV_NOINIT which results in the struct pages not being initialized 1038 * for this region. 1039 * @base: the base phys addr of the region 1040 * @size: the size of the region 1041 * 1042 * struct pages will not be initialized for reserved memory regions marked with 1043 * %MEMBLOCK_RSRV_NOINIT. 1044 * 1045 * Return: 0 on success, -errno on failure. 1046 */ 1047 int __init_memblock memblock_reserved_mark_noinit(phys_addr_t base, phys_addr_t size) 1048 { 1049 return memblock_setclr_flag(&memblock.reserved, base, size, 1, 1050 MEMBLOCK_RSRV_NOINIT); 1051 } 1052 1053 static bool should_skip_region(struct memblock_type *type, 1054 struct memblock_region *m, 1055 int nid, int flags) 1056 { 1057 int m_nid = memblock_get_region_node(m); 1058 1059 /* we never skip regions when iterating memblock.reserved or physmem */ 1060 if (type != memblock_memory) 1061 return false; 1062 1063 /* only memory regions are associated with nodes, check it */ 1064 if (numa_valid_node(nid) && nid != m_nid) 1065 return true; 1066 1067 /* skip hotpluggable memory regions if needed */ 1068 if (movable_node_is_enabled() && memblock_is_hotpluggable(m) && 1069 !(flags & MEMBLOCK_HOTPLUG)) 1070 return true; 1071 1072 /* if we want mirror memory skip non-mirror memory regions */ 1073 if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m)) 1074 return true; 1075 1076 /* skip nomap memory unless we were asked for it explicitly */ 1077 if (!(flags & MEMBLOCK_NOMAP) && memblock_is_nomap(m)) 1078 return true; 1079 1080 /* skip driver-managed memory unless we were asked for it explicitly */ 1081 if (!(flags & MEMBLOCK_DRIVER_MANAGED) && memblock_is_driver_managed(m)) 1082 return true; 1083 1084 return false; 1085 } 1086 1087 /** 1088 * __next_mem_range - next function for for_each_free_mem_range() etc. 1089 * @idx: pointer to u64 loop variable 1090 * @nid: node selector, %NUMA_NO_NODE for all nodes 1091 * @flags: pick from blocks based on memory attributes 1092 * @type_a: pointer to memblock_type from where the range is taken 1093 * @type_b: pointer to memblock_type which excludes memory from being taken 1094 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL 1095 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL 1096 * @out_nid: ptr to int for nid of the range, can be %NULL 1097 * 1098 * Find the first area from *@idx which matches @nid, fill the out 1099 * parameters, and update *@idx for the next iteration. The lower 32bit of 1100 * *@idx contains index into type_a and the upper 32bit indexes the 1101 * areas before each region in type_b. For example, if type_b regions 1102 * look like the following, 1103 * 1104 * 0:[0-16), 1:[32-48), 2:[128-130) 1105 * 1106 * The upper 32bit indexes the following regions. 1107 * 1108 * 0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX) 1109 * 1110 * As both region arrays are sorted, the function advances the two indices 1111 * in lockstep and returns each intersection. 1112 */ 1113 void __next_mem_range(u64 *idx, int nid, enum memblock_flags flags, 1114 struct memblock_type *type_a, 1115 struct memblock_type *type_b, phys_addr_t *out_start, 1116 phys_addr_t *out_end, int *out_nid) 1117 { 1118 int idx_a = *idx & 0xffffffff; 1119 int idx_b = *idx >> 32; 1120 1121 for (; idx_a < type_a->cnt; idx_a++) { 1122 struct memblock_region *m = &type_a->regions[idx_a]; 1123 1124 phys_addr_t m_start = m->base; 1125 phys_addr_t m_end = m->base + m->size; 1126 int m_nid = memblock_get_region_node(m); 1127 1128 if (should_skip_region(type_a, m, nid, flags)) 1129 continue; 1130 1131 if (!type_b) { 1132 if (out_start) 1133 *out_start = m_start; 1134 if (out_end) 1135 *out_end = m_end; 1136 if (out_nid) 1137 *out_nid = m_nid; 1138 idx_a++; 1139 *idx = (u32)idx_a | (u64)idx_b << 32; 1140 return; 1141 } 1142 1143 /* scan areas before each reservation */ 1144 for (; idx_b < type_b->cnt + 1; idx_b++) { 1145 struct memblock_region *r; 1146 phys_addr_t r_start; 1147 phys_addr_t r_end; 1148 1149 r = &type_b->regions[idx_b]; 1150 r_start = idx_b ? r[-1].base + r[-1].size : 0; 1151 r_end = idx_b < type_b->cnt ? 1152 r->base : PHYS_ADDR_MAX; 1153 1154 /* 1155 * if idx_b advanced past idx_a, 1156 * break out to advance idx_a 1157 */ 1158 if (r_start >= m_end) 1159 break; 1160 /* if the two regions intersect, we're done */ 1161 if (m_start < r_end) { 1162 if (out_start) 1163 *out_start = 1164 max(m_start, r_start); 1165 if (out_end) 1166 *out_end = min(m_end, r_end); 1167 if (out_nid) 1168 *out_nid = m_nid; 1169 /* 1170 * The region which ends first is 1171 * advanced for the next iteration. 1172 */ 1173 if (m_end <= r_end) 1174 idx_a++; 1175 else 1176 idx_b++; 1177 *idx = (u32)idx_a | (u64)idx_b << 32; 1178 return; 1179 } 1180 } 1181 } 1182 1183 /* signal end of iteration */ 1184 *idx = ULLONG_MAX; 1185 } 1186 1187 /** 1188 * __next_mem_range_rev - generic next function for for_each_*_range_rev() 1189 * 1190 * @idx: pointer to u64 loop variable 1191 * @nid: node selector, %NUMA_NO_NODE for all nodes 1192 * @flags: pick from blocks based on memory attributes 1193 * @type_a: pointer to memblock_type from where the range is taken 1194 * @type_b: pointer to memblock_type which excludes memory from being taken 1195 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL 1196 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL 1197 * @out_nid: ptr to int for nid of the range, can be %NULL 1198 * 1199 * Finds the next range from type_a which is not marked as unsuitable 1200 * in type_b. 1201 * 1202 * Reverse of __next_mem_range(). 1203 */ 1204 void __init_memblock __next_mem_range_rev(u64 *idx, int nid, 1205 enum memblock_flags flags, 1206 struct memblock_type *type_a, 1207 struct memblock_type *type_b, 1208 phys_addr_t *out_start, 1209 phys_addr_t *out_end, int *out_nid) 1210 { 1211 int idx_a = *idx & 0xffffffff; 1212 int idx_b = *idx >> 32; 1213 1214 if (*idx == (u64)ULLONG_MAX) { 1215 idx_a = type_a->cnt - 1; 1216 if (type_b != NULL) 1217 idx_b = type_b->cnt; 1218 else 1219 idx_b = 0; 1220 } 1221 1222 for (; idx_a >= 0; idx_a--) { 1223 struct memblock_region *m = &type_a->regions[idx_a]; 1224 1225 phys_addr_t m_start = m->base; 1226 phys_addr_t m_end = m->base + m->size; 1227 int m_nid = memblock_get_region_node(m); 1228 1229 if (should_skip_region(type_a, m, nid, flags)) 1230 continue; 1231 1232 if (!type_b) { 1233 if (out_start) 1234 *out_start = m_start; 1235 if (out_end) 1236 *out_end = m_end; 1237 if (out_nid) 1238 *out_nid = m_nid; 1239 idx_a--; 1240 *idx = (u32)idx_a | (u64)idx_b << 32; 1241 return; 1242 } 1243 1244 /* scan areas before each reservation */ 1245 for (; idx_b >= 0; idx_b--) { 1246 struct memblock_region *r; 1247 phys_addr_t r_start; 1248 phys_addr_t r_end; 1249 1250 r = &type_b->regions[idx_b]; 1251 r_start = idx_b ? r[-1].base + r[-1].size : 0; 1252 r_end = idx_b < type_b->cnt ? 1253 r->base : PHYS_ADDR_MAX; 1254 /* 1255 * if idx_b advanced past idx_a, 1256 * break out to advance idx_a 1257 */ 1258 1259 if (r_end <= m_start) 1260 break; 1261 /* if the two regions intersect, we're done */ 1262 if (m_end > r_start) { 1263 if (out_start) 1264 *out_start = max(m_start, r_start); 1265 if (out_end) 1266 *out_end = min(m_end, r_end); 1267 if (out_nid) 1268 *out_nid = m_nid; 1269 if (m_start >= r_start) 1270 idx_a--; 1271 else 1272 idx_b--; 1273 *idx = (u32)idx_a | (u64)idx_b << 32; 1274 return; 1275 } 1276 } 1277 } 1278 /* signal end of iteration */ 1279 *idx = ULLONG_MAX; 1280 } 1281 1282 /* 1283 * Common iterator interface used to define for_each_mem_pfn_range(). 1284 */ 1285 void __init_memblock __next_mem_pfn_range(int *idx, int nid, 1286 unsigned long *out_start_pfn, 1287 unsigned long *out_end_pfn, int *out_nid) 1288 { 1289 struct memblock_type *type = &memblock.memory; 1290 struct memblock_region *r; 1291 int r_nid; 1292 1293 while (++*idx < type->cnt) { 1294 r = &type->regions[*idx]; 1295 r_nid = memblock_get_region_node(r); 1296 1297 if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size)) 1298 continue; 1299 if (!numa_valid_node(nid) || nid == r_nid) 1300 break; 1301 } 1302 if (*idx >= type->cnt) { 1303 *idx = -1; 1304 return; 1305 } 1306 1307 if (out_start_pfn) 1308 *out_start_pfn = PFN_UP(r->base); 1309 if (out_end_pfn) 1310 *out_end_pfn = PFN_DOWN(r->base + r->size); 1311 if (out_nid) 1312 *out_nid = r_nid; 1313 } 1314 1315 /** 1316 * memblock_set_node - set node ID on memblock regions 1317 * @base: base of area to set node ID for 1318 * @size: size of area to set node ID for 1319 * @type: memblock type to set node ID for 1320 * @nid: node ID to set 1321 * 1322 * Set the nid of memblock @type regions in [@base, @base + @size) to @nid. 1323 * Regions which cross the area boundaries are split as necessary. 1324 * 1325 * Return: 1326 * 0 on success, -errno on failure. 1327 */ 1328 int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size, 1329 struct memblock_type *type, int nid) 1330 { 1331 #ifdef CONFIG_NUMA 1332 int start_rgn, end_rgn; 1333 int i, ret; 1334 1335 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn); 1336 if (ret) 1337 return ret; 1338 1339 for (i = start_rgn; i < end_rgn; i++) 1340 memblock_set_region_node(&type->regions[i], nid); 1341 1342 memblock_merge_regions(type, start_rgn, end_rgn); 1343 #endif 1344 return 0; 1345 } 1346 1347 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT 1348 /** 1349 * __next_mem_pfn_range_in_zone - iterator for for_each_*_range_in_zone() 1350 * 1351 * @idx: pointer to u64 loop variable 1352 * @zone: zone in which all of the memory blocks reside 1353 * @out_spfn: ptr to ulong for start pfn of the range, can be %NULL 1354 * @out_epfn: ptr to ulong for end pfn of the range, can be %NULL 1355 * 1356 * This function is meant to be a zone/pfn specific wrapper for the 1357 * for_each_mem_range type iterators. Specifically they are used in the 1358 * deferred memory init routines and as such we were duplicating much of 1359 * this logic throughout the code. So instead of having it in multiple 1360 * locations it seemed like it would make more sense to centralize this to 1361 * one new iterator that does everything they need. 1362 */ 1363 void __init_memblock 1364 __next_mem_pfn_range_in_zone(u64 *idx, struct zone *zone, 1365 unsigned long *out_spfn, unsigned long *out_epfn) 1366 { 1367 int zone_nid = zone_to_nid(zone); 1368 phys_addr_t spa, epa; 1369 1370 __next_mem_range(idx, zone_nid, MEMBLOCK_NONE, 1371 &memblock.memory, &memblock.reserved, 1372 &spa, &epa, NULL); 1373 1374 while (*idx != U64_MAX) { 1375 unsigned long epfn = PFN_DOWN(epa); 1376 unsigned long spfn = PFN_UP(spa); 1377 1378 /* 1379 * Verify the end is at least past the start of the zone and 1380 * that we have at least one PFN to initialize. 1381 */ 1382 if (zone->zone_start_pfn < epfn && spfn < epfn) { 1383 /* if we went too far just stop searching */ 1384 if (zone_end_pfn(zone) <= spfn) { 1385 *idx = U64_MAX; 1386 break; 1387 } 1388 1389 if (out_spfn) 1390 *out_spfn = max(zone->zone_start_pfn, spfn); 1391 if (out_epfn) 1392 *out_epfn = min(zone_end_pfn(zone), epfn); 1393 1394 return; 1395 } 1396 1397 __next_mem_range(idx, zone_nid, MEMBLOCK_NONE, 1398 &memblock.memory, &memblock.reserved, 1399 &spa, &epa, NULL); 1400 } 1401 1402 /* signal end of iteration */ 1403 if (out_spfn) 1404 *out_spfn = ULONG_MAX; 1405 if (out_epfn) 1406 *out_epfn = 0; 1407 } 1408 1409 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ 1410 1411 /** 1412 * memblock_alloc_range_nid - allocate boot memory block 1413 * @size: size of memory block to be allocated in bytes 1414 * @align: alignment of the region and block's size 1415 * @start: the lower bound of the memory region to allocate (phys address) 1416 * @end: the upper bound of the memory region to allocate (phys address) 1417 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 1418 * @exact_nid: control the allocation fall back to other nodes 1419 * 1420 * The allocation is performed from memory region limited by 1421 * memblock.current_limit if @end == %MEMBLOCK_ALLOC_ACCESSIBLE. 1422 * 1423 * If the specified node can not hold the requested memory and @exact_nid 1424 * is false, the allocation falls back to any node in the system. 1425 * 1426 * For systems with memory mirroring, the allocation is attempted first 1427 * from the regions with mirroring enabled and then retried from any 1428 * memory region. 1429 * 1430 * In addition, function using kmemleak_alloc_phys for allocated boot 1431 * memory block, it is never reported as leaks. 1432 * 1433 * Return: 1434 * Physical address of allocated memory block on success, %0 on failure. 1435 */ 1436 phys_addr_t __init memblock_alloc_range_nid(phys_addr_t size, 1437 phys_addr_t align, phys_addr_t start, 1438 phys_addr_t end, int nid, 1439 bool exact_nid) 1440 { 1441 enum memblock_flags flags = choose_memblock_flags(); 1442 phys_addr_t found; 1443 1444 if (!align) { 1445 /* Can't use WARNs this early in boot on powerpc */ 1446 dump_stack(); 1447 align = SMP_CACHE_BYTES; 1448 } 1449 1450 again: 1451 found = memblock_find_in_range_node(size, align, start, end, nid, 1452 flags); 1453 if (found && !memblock_reserve(found, size)) 1454 goto done; 1455 1456 if (numa_valid_node(nid) && !exact_nid) { 1457 found = memblock_find_in_range_node(size, align, start, 1458 end, NUMA_NO_NODE, 1459 flags); 1460 if (found && !memblock_reserve(found, size)) 1461 goto done; 1462 } 1463 1464 if (flags & MEMBLOCK_MIRROR) { 1465 flags &= ~MEMBLOCK_MIRROR; 1466 pr_warn_ratelimited("Could not allocate %pap bytes of mirrored memory\n", 1467 &size); 1468 goto again; 1469 } 1470 1471 return 0; 1472 1473 done: 1474 /* 1475 * Skip kmemleak for those places like kasan_init() and 1476 * early_pgtable_alloc() due to high volume. 1477 */ 1478 if (end != MEMBLOCK_ALLOC_NOLEAKTRACE) 1479 /* 1480 * Memblock allocated blocks are never reported as 1481 * leaks. This is because many of these blocks are 1482 * only referred via the physical address which is 1483 * not looked up by kmemleak. 1484 */ 1485 kmemleak_alloc_phys(found, size, 0); 1486 1487 /* 1488 * Some Virtual Machine platforms, such as Intel TDX or AMD SEV-SNP, 1489 * require memory to be accepted before it can be used by the 1490 * guest. 1491 * 1492 * Accept the memory of the allocated buffer. 1493 */ 1494 accept_memory(found, found + size); 1495 1496 return found; 1497 } 1498 1499 /** 1500 * memblock_phys_alloc_range - allocate a memory block inside specified range 1501 * @size: size of memory block to be allocated in bytes 1502 * @align: alignment of the region and block's size 1503 * @start: the lower bound of the memory region to allocate (physical address) 1504 * @end: the upper bound of the memory region to allocate (physical address) 1505 * 1506 * Allocate @size bytes in the between @start and @end. 1507 * 1508 * Return: physical address of the allocated memory block on success, 1509 * %0 on failure. 1510 */ 1511 phys_addr_t __init memblock_phys_alloc_range(phys_addr_t size, 1512 phys_addr_t align, 1513 phys_addr_t start, 1514 phys_addr_t end) 1515 { 1516 memblock_dbg("%s: %llu bytes align=0x%llx from=%pa max_addr=%pa %pS\n", 1517 __func__, (u64)size, (u64)align, &start, &end, 1518 (void *)_RET_IP_); 1519 return memblock_alloc_range_nid(size, align, start, end, NUMA_NO_NODE, 1520 false); 1521 } 1522 1523 /** 1524 * memblock_phys_alloc_try_nid - allocate a memory block from specified NUMA node 1525 * @size: size of memory block to be allocated in bytes 1526 * @align: alignment of the region and block's size 1527 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 1528 * 1529 * Allocates memory block from the specified NUMA node. If the node 1530 * has no available memory, attempts to allocated from any node in the 1531 * system. 1532 * 1533 * Return: physical address of the allocated memory block on success, 1534 * %0 on failure. 1535 */ 1536 phys_addr_t __init memblock_phys_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid) 1537 { 1538 return memblock_alloc_range_nid(size, align, 0, 1539 MEMBLOCK_ALLOC_ACCESSIBLE, nid, false); 1540 } 1541 1542 /** 1543 * memblock_alloc_internal - allocate boot memory block 1544 * @size: size of memory block to be allocated in bytes 1545 * @align: alignment of the region and block's size 1546 * @min_addr: the lower bound of the memory region to allocate (phys address) 1547 * @max_addr: the upper bound of the memory region to allocate (phys address) 1548 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 1549 * @exact_nid: control the allocation fall back to other nodes 1550 * 1551 * Allocates memory block using memblock_alloc_range_nid() and 1552 * converts the returned physical address to virtual. 1553 * 1554 * The @min_addr limit is dropped if it can not be satisfied and the allocation 1555 * will fall back to memory below @min_addr. Other constraints, such 1556 * as node and mirrored memory will be handled again in 1557 * memblock_alloc_range_nid(). 1558 * 1559 * Return: 1560 * Virtual address of allocated memory block on success, NULL on failure. 1561 */ 1562 static void * __init memblock_alloc_internal( 1563 phys_addr_t size, phys_addr_t align, 1564 phys_addr_t min_addr, phys_addr_t max_addr, 1565 int nid, bool exact_nid) 1566 { 1567 phys_addr_t alloc; 1568 1569 /* 1570 * Detect any accidental use of these APIs after slab is ready, as at 1571 * this moment memblock may be deinitialized already and its 1572 * internal data may be destroyed (after execution of memblock_free_all) 1573 */ 1574 if (WARN_ON_ONCE(slab_is_available())) 1575 return kzalloc_node(size, GFP_NOWAIT, nid); 1576 1577 if (max_addr > memblock.current_limit) 1578 max_addr = memblock.current_limit; 1579 1580 alloc = memblock_alloc_range_nid(size, align, min_addr, max_addr, nid, 1581 exact_nid); 1582 1583 /* retry allocation without lower limit */ 1584 if (!alloc && min_addr) 1585 alloc = memblock_alloc_range_nid(size, align, 0, max_addr, nid, 1586 exact_nid); 1587 1588 if (!alloc) 1589 return NULL; 1590 1591 return phys_to_virt(alloc); 1592 } 1593 1594 /** 1595 * memblock_alloc_exact_nid_raw - allocate boot memory block on the exact node 1596 * without zeroing memory 1597 * @size: size of memory block to be allocated in bytes 1598 * @align: alignment of the region and block's size 1599 * @min_addr: the lower bound of the memory region from where the allocation 1600 * is preferred (phys address) 1601 * @max_addr: the upper bound of the memory region from where the allocation 1602 * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to 1603 * allocate only from memory limited by memblock.current_limit value 1604 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 1605 * 1606 * Public function, provides additional debug information (including caller 1607 * info), if enabled. Does not zero allocated memory. 1608 * 1609 * Return: 1610 * Virtual address of allocated memory block on success, NULL on failure. 1611 */ 1612 void * __init memblock_alloc_exact_nid_raw( 1613 phys_addr_t size, phys_addr_t align, 1614 phys_addr_t min_addr, phys_addr_t max_addr, 1615 int nid) 1616 { 1617 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n", 1618 __func__, (u64)size, (u64)align, nid, &min_addr, 1619 &max_addr, (void *)_RET_IP_); 1620 1621 return memblock_alloc_internal(size, align, min_addr, max_addr, nid, 1622 true); 1623 } 1624 1625 /** 1626 * memblock_alloc_try_nid_raw - allocate boot memory block without zeroing 1627 * memory and without panicking 1628 * @size: size of memory block to be allocated in bytes 1629 * @align: alignment of the region and block's size 1630 * @min_addr: the lower bound of the memory region from where the allocation 1631 * is preferred (phys address) 1632 * @max_addr: the upper bound of the memory region from where the allocation 1633 * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to 1634 * allocate only from memory limited by memblock.current_limit value 1635 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 1636 * 1637 * Public function, provides additional debug information (including caller 1638 * info), if enabled. Does not zero allocated memory, does not panic if request 1639 * cannot be satisfied. 1640 * 1641 * Return: 1642 * Virtual address of allocated memory block on success, NULL on failure. 1643 */ 1644 void * __init memblock_alloc_try_nid_raw( 1645 phys_addr_t size, phys_addr_t align, 1646 phys_addr_t min_addr, phys_addr_t max_addr, 1647 int nid) 1648 { 1649 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n", 1650 __func__, (u64)size, (u64)align, nid, &min_addr, 1651 &max_addr, (void *)_RET_IP_); 1652 1653 return memblock_alloc_internal(size, align, min_addr, max_addr, nid, 1654 false); 1655 } 1656 1657 /** 1658 * memblock_alloc_try_nid - allocate boot memory block 1659 * @size: size of memory block to be allocated in bytes 1660 * @align: alignment of the region and block's size 1661 * @min_addr: the lower bound of the memory region from where the allocation 1662 * is preferred (phys address) 1663 * @max_addr: the upper bound of the memory region from where the allocation 1664 * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to 1665 * allocate only from memory limited by memblock.current_limit value 1666 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 1667 * 1668 * Public function, provides additional debug information (including caller 1669 * info), if enabled. This function zeroes the allocated memory. 1670 * 1671 * Return: 1672 * Virtual address of allocated memory block on success, NULL on failure. 1673 */ 1674 void * __init memblock_alloc_try_nid( 1675 phys_addr_t size, phys_addr_t align, 1676 phys_addr_t min_addr, phys_addr_t max_addr, 1677 int nid) 1678 { 1679 void *ptr; 1680 1681 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n", 1682 __func__, (u64)size, (u64)align, nid, &min_addr, 1683 &max_addr, (void *)_RET_IP_); 1684 ptr = memblock_alloc_internal(size, align, 1685 min_addr, max_addr, nid, false); 1686 if (ptr) 1687 memset(ptr, 0, size); 1688 1689 return ptr; 1690 } 1691 1692 /** 1693 * memblock_free_late - free pages directly to buddy allocator 1694 * @base: phys starting address of the boot memory block 1695 * @size: size of the boot memory block in bytes 1696 * 1697 * This is only useful when the memblock allocator has already been torn 1698 * down, but we are still initializing the system. Pages are released directly 1699 * to the buddy allocator. 1700 */ 1701 void __init memblock_free_late(phys_addr_t base, phys_addr_t size) 1702 { 1703 phys_addr_t cursor, end; 1704 1705 end = base + size - 1; 1706 memblock_dbg("%s: [%pa-%pa] %pS\n", 1707 __func__, &base, &end, (void *)_RET_IP_); 1708 kmemleak_free_part_phys(base, size); 1709 cursor = PFN_UP(base); 1710 end = PFN_DOWN(base + size); 1711 1712 for (; cursor < end; cursor++) { 1713 memblock_free_pages(pfn_to_page(cursor), cursor, 0); 1714 totalram_pages_inc(); 1715 } 1716 } 1717 1718 /* 1719 * Remaining API functions 1720 */ 1721 1722 phys_addr_t __init_memblock memblock_phys_mem_size(void) 1723 { 1724 return memblock.memory.total_size; 1725 } 1726 1727 phys_addr_t __init_memblock memblock_reserved_size(void) 1728 { 1729 return memblock.reserved.total_size; 1730 } 1731 1732 /* lowest address */ 1733 phys_addr_t __init_memblock memblock_start_of_DRAM(void) 1734 { 1735 return memblock.memory.regions[0].base; 1736 } 1737 1738 phys_addr_t __init_memblock memblock_end_of_DRAM(void) 1739 { 1740 int idx = memblock.memory.cnt - 1; 1741 1742 return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size); 1743 } 1744 1745 static phys_addr_t __init_memblock __find_max_addr(phys_addr_t limit) 1746 { 1747 phys_addr_t max_addr = PHYS_ADDR_MAX; 1748 struct memblock_region *r; 1749 1750 /* 1751 * translate the memory @limit size into the max address within one of 1752 * the memory memblock regions, if the @limit exceeds the total size 1753 * of those regions, max_addr will keep original value PHYS_ADDR_MAX 1754 */ 1755 for_each_mem_region(r) { 1756 if (limit <= r->size) { 1757 max_addr = r->base + limit; 1758 break; 1759 } 1760 limit -= r->size; 1761 } 1762 1763 return max_addr; 1764 } 1765 1766 void __init memblock_enforce_memory_limit(phys_addr_t limit) 1767 { 1768 phys_addr_t max_addr; 1769 1770 if (!limit) 1771 return; 1772 1773 max_addr = __find_max_addr(limit); 1774 1775 /* @limit exceeds the total size of the memory, do nothing */ 1776 if (max_addr == PHYS_ADDR_MAX) 1777 return; 1778 1779 /* truncate both memory and reserved regions */ 1780 memblock_remove_range(&memblock.memory, max_addr, 1781 PHYS_ADDR_MAX); 1782 memblock_remove_range(&memblock.reserved, max_addr, 1783 PHYS_ADDR_MAX); 1784 } 1785 1786 void __init memblock_cap_memory_range(phys_addr_t base, phys_addr_t size) 1787 { 1788 int start_rgn, end_rgn; 1789 int i, ret; 1790 1791 if (!size) 1792 return; 1793 1794 if (!memblock_memory->total_size) { 1795 pr_warn("%s: No memory registered yet\n", __func__); 1796 return; 1797 } 1798 1799 ret = memblock_isolate_range(&memblock.memory, base, size, 1800 &start_rgn, &end_rgn); 1801 if (ret) 1802 return; 1803 1804 /* remove all the MAP regions */ 1805 for (i = memblock.memory.cnt - 1; i >= end_rgn; i--) 1806 if (!memblock_is_nomap(&memblock.memory.regions[i])) 1807 memblock_remove_region(&memblock.memory, i); 1808 1809 for (i = start_rgn - 1; i >= 0; i--) 1810 if (!memblock_is_nomap(&memblock.memory.regions[i])) 1811 memblock_remove_region(&memblock.memory, i); 1812 1813 /* truncate the reserved regions */ 1814 memblock_remove_range(&memblock.reserved, 0, base); 1815 memblock_remove_range(&memblock.reserved, 1816 base + size, PHYS_ADDR_MAX); 1817 } 1818 1819 void __init memblock_mem_limit_remove_map(phys_addr_t limit) 1820 { 1821 phys_addr_t max_addr; 1822 1823 if (!limit) 1824 return; 1825 1826 max_addr = __find_max_addr(limit); 1827 1828 /* @limit exceeds the total size of the memory, do nothing */ 1829 if (max_addr == PHYS_ADDR_MAX) 1830 return; 1831 1832 memblock_cap_memory_range(0, max_addr); 1833 } 1834 1835 static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr) 1836 { 1837 unsigned int left = 0, right = type->cnt; 1838 1839 do { 1840 unsigned int mid = (right + left) / 2; 1841 1842 if (addr < type->regions[mid].base) 1843 right = mid; 1844 else if (addr >= (type->regions[mid].base + 1845 type->regions[mid].size)) 1846 left = mid + 1; 1847 else 1848 return mid; 1849 } while (left < right); 1850 return -1; 1851 } 1852 1853 bool __init_memblock memblock_is_reserved(phys_addr_t addr) 1854 { 1855 return memblock_search(&memblock.reserved, addr) != -1; 1856 } 1857 1858 bool __init_memblock memblock_is_memory(phys_addr_t addr) 1859 { 1860 return memblock_search(&memblock.memory, addr) != -1; 1861 } 1862 1863 bool __init_memblock memblock_is_map_memory(phys_addr_t addr) 1864 { 1865 int i = memblock_search(&memblock.memory, addr); 1866 1867 if (i == -1) 1868 return false; 1869 return !memblock_is_nomap(&memblock.memory.regions[i]); 1870 } 1871 1872 int __init_memblock memblock_search_pfn_nid(unsigned long pfn, 1873 unsigned long *start_pfn, unsigned long *end_pfn) 1874 { 1875 struct memblock_type *type = &memblock.memory; 1876 int mid = memblock_search(type, PFN_PHYS(pfn)); 1877 1878 if (mid == -1) 1879 return NUMA_NO_NODE; 1880 1881 *start_pfn = PFN_DOWN(type->regions[mid].base); 1882 *end_pfn = PFN_DOWN(type->regions[mid].base + type->regions[mid].size); 1883 1884 return memblock_get_region_node(&type->regions[mid]); 1885 } 1886 1887 /** 1888 * memblock_is_region_memory - check if a region is a subset of memory 1889 * @base: base of region to check 1890 * @size: size of region to check 1891 * 1892 * Check if the region [@base, @base + @size) is a subset of a memory block. 1893 * 1894 * Return: 1895 * 0 if false, non-zero if true 1896 */ 1897 bool __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size) 1898 { 1899 int idx = memblock_search(&memblock.memory, base); 1900 phys_addr_t end = base + memblock_cap_size(base, &size); 1901 1902 if (idx == -1) 1903 return false; 1904 return (memblock.memory.regions[idx].base + 1905 memblock.memory.regions[idx].size) >= end; 1906 } 1907 1908 /** 1909 * memblock_is_region_reserved - check if a region intersects reserved memory 1910 * @base: base of region to check 1911 * @size: size of region to check 1912 * 1913 * Check if the region [@base, @base + @size) intersects a reserved 1914 * memory block. 1915 * 1916 * Return: 1917 * True if they intersect, false if not. 1918 */ 1919 bool __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size) 1920 { 1921 return memblock_overlaps_region(&memblock.reserved, base, size); 1922 } 1923 1924 void __init_memblock memblock_trim_memory(phys_addr_t align) 1925 { 1926 phys_addr_t start, end, orig_start, orig_end; 1927 struct memblock_region *r; 1928 1929 for_each_mem_region(r) { 1930 orig_start = r->base; 1931 orig_end = r->base + r->size; 1932 start = round_up(orig_start, align); 1933 end = round_down(orig_end, align); 1934 1935 if (start == orig_start && end == orig_end) 1936 continue; 1937 1938 if (start < end) { 1939 r->base = start; 1940 r->size = end - start; 1941 } else { 1942 memblock_remove_region(&memblock.memory, 1943 r - memblock.memory.regions); 1944 r--; 1945 } 1946 } 1947 } 1948 1949 void __init_memblock memblock_set_current_limit(phys_addr_t limit) 1950 { 1951 memblock.current_limit = limit; 1952 } 1953 1954 phys_addr_t __init_memblock memblock_get_current_limit(void) 1955 { 1956 return memblock.current_limit; 1957 } 1958 1959 static void __init_memblock memblock_dump(struct memblock_type *type) 1960 { 1961 phys_addr_t base, end, size; 1962 enum memblock_flags flags; 1963 int idx; 1964 struct memblock_region *rgn; 1965 1966 pr_info(" %s.cnt = 0x%lx\n", type->name, type->cnt); 1967 1968 for_each_memblock_type(idx, type, rgn) { 1969 char nid_buf[32] = ""; 1970 1971 base = rgn->base; 1972 size = rgn->size; 1973 end = base + size - 1; 1974 flags = rgn->flags; 1975 #ifdef CONFIG_NUMA 1976 if (numa_valid_node(memblock_get_region_node(rgn))) 1977 snprintf(nid_buf, sizeof(nid_buf), " on node %d", 1978 memblock_get_region_node(rgn)); 1979 #endif 1980 pr_info(" %s[%#x]\t[%pa-%pa], %pa bytes%s flags: %#x\n", 1981 type->name, idx, &base, &end, &size, nid_buf, flags); 1982 } 1983 } 1984 1985 static void __init_memblock __memblock_dump_all(void) 1986 { 1987 pr_info("MEMBLOCK configuration:\n"); 1988 pr_info(" memory size = %pa reserved size = %pa\n", 1989 &memblock.memory.total_size, 1990 &memblock.reserved.total_size); 1991 1992 memblock_dump(&memblock.memory); 1993 memblock_dump(&memblock.reserved); 1994 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP 1995 memblock_dump(&physmem); 1996 #endif 1997 } 1998 1999 void __init_memblock memblock_dump_all(void) 2000 { 2001 if (memblock_debug) 2002 __memblock_dump_all(); 2003 } 2004 2005 void __init memblock_allow_resize(void) 2006 { 2007 memblock_can_resize = 1; 2008 } 2009 2010 static int __init early_memblock(char *p) 2011 { 2012 if (p && strstr(p, "debug")) 2013 memblock_debug = 1; 2014 return 0; 2015 } 2016 early_param("memblock", early_memblock); 2017 2018 static void __init free_memmap(unsigned long start_pfn, unsigned long end_pfn) 2019 { 2020 struct page *start_pg, *end_pg; 2021 phys_addr_t pg, pgend; 2022 2023 /* 2024 * Convert start_pfn/end_pfn to a struct page pointer. 2025 */ 2026 start_pg = pfn_to_page(start_pfn - 1) + 1; 2027 end_pg = pfn_to_page(end_pfn - 1) + 1; 2028 2029 /* 2030 * Convert to physical addresses, and round start upwards and end 2031 * downwards. 2032 */ 2033 pg = PAGE_ALIGN(__pa(start_pg)); 2034 pgend = __pa(end_pg) & PAGE_MASK; 2035 2036 /* 2037 * If there are free pages between these, free the section of the 2038 * memmap array. 2039 */ 2040 if (pg < pgend) 2041 memblock_phys_free(pg, pgend - pg); 2042 } 2043 2044 /* 2045 * The mem_map array can get very big. Free the unused area of the memory map. 2046 */ 2047 static void __init free_unused_memmap(void) 2048 { 2049 unsigned long start, end, prev_end = 0; 2050 int i; 2051 2052 if (!IS_ENABLED(CONFIG_HAVE_ARCH_PFN_VALID) || 2053 IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP)) 2054 return; 2055 2056 /* 2057 * This relies on each bank being in address order. 2058 * The banks are sorted previously in bootmem_init(). 2059 */ 2060 for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, NULL) { 2061 #ifdef CONFIG_SPARSEMEM 2062 /* 2063 * Take care not to free memmap entries that don't exist 2064 * due to SPARSEMEM sections which aren't present. 2065 */ 2066 start = min(start, ALIGN(prev_end, PAGES_PER_SECTION)); 2067 #endif 2068 /* 2069 * Align down here since many operations in VM subsystem 2070 * presume that there are no holes in the memory map inside 2071 * a pageblock 2072 */ 2073 start = pageblock_start_pfn(start); 2074 2075 /* 2076 * If we had a previous bank, and there is a space 2077 * between the current bank and the previous, free it. 2078 */ 2079 if (prev_end && prev_end < start) 2080 free_memmap(prev_end, start); 2081 2082 /* 2083 * Align up here since many operations in VM subsystem 2084 * presume that there are no holes in the memory map inside 2085 * a pageblock 2086 */ 2087 prev_end = pageblock_align(end); 2088 } 2089 2090 #ifdef CONFIG_SPARSEMEM 2091 if (!IS_ALIGNED(prev_end, PAGES_PER_SECTION)) { 2092 prev_end = pageblock_align(end); 2093 free_memmap(prev_end, ALIGN(prev_end, PAGES_PER_SECTION)); 2094 } 2095 #endif 2096 } 2097 2098 static void __init __free_pages_memory(unsigned long start, unsigned long end) 2099 { 2100 int order; 2101 2102 while (start < end) { 2103 /* 2104 * Free the pages in the largest chunks alignment allows. 2105 * 2106 * __ffs() behaviour is undefined for 0. start == 0 is 2107 * MAX_PAGE_ORDER-aligned, set order to MAX_PAGE_ORDER for 2108 * the case. 2109 */ 2110 if (start) 2111 order = min_t(int, MAX_PAGE_ORDER, __ffs(start)); 2112 else 2113 order = MAX_PAGE_ORDER; 2114 2115 while (start + (1UL << order) > end) 2116 order--; 2117 2118 memblock_free_pages(pfn_to_page(start), start, order); 2119 2120 start += (1UL << order); 2121 } 2122 } 2123 2124 static unsigned long __init __free_memory_core(phys_addr_t start, 2125 phys_addr_t end) 2126 { 2127 unsigned long start_pfn = PFN_UP(start); 2128 unsigned long end_pfn = min_t(unsigned long, 2129 PFN_DOWN(end), max_low_pfn); 2130 2131 if (start_pfn >= end_pfn) 2132 return 0; 2133 2134 __free_pages_memory(start_pfn, end_pfn); 2135 2136 return end_pfn - start_pfn; 2137 } 2138 2139 static void __init memmap_init_reserved_pages(void) 2140 { 2141 struct memblock_region *region; 2142 phys_addr_t start, end; 2143 int nid; 2144 2145 /* 2146 * set nid on all reserved pages and also treat struct 2147 * pages for the NOMAP regions as PageReserved 2148 */ 2149 for_each_mem_region(region) { 2150 nid = memblock_get_region_node(region); 2151 start = region->base; 2152 end = start + region->size; 2153 2154 if (memblock_is_nomap(region)) 2155 reserve_bootmem_region(start, end, nid); 2156 2157 memblock_set_node(start, end, &memblock.reserved, nid); 2158 } 2159 2160 /* 2161 * initialize struct pages for reserved regions that don't have 2162 * the MEMBLOCK_RSRV_NOINIT flag set 2163 */ 2164 for_each_reserved_mem_region(region) { 2165 if (!memblock_is_reserved_noinit(region)) { 2166 nid = memblock_get_region_node(region); 2167 start = region->base; 2168 end = start + region->size; 2169 2170 if (!numa_valid_node(nid)) 2171 nid = early_pfn_to_nid(PFN_DOWN(start)); 2172 2173 reserve_bootmem_region(start, end, nid); 2174 } 2175 } 2176 } 2177 2178 static unsigned long __init free_low_memory_core_early(void) 2179 { 2180 unsigned long count = 0; 2181 phys_addr_t start, end; 2182 u64 i; 2183 2184 memblock_clear_hotplug(0, -1); 2185 2186 memmap_init_reserved_pages(); 2187 2188 /* 2189 * We need to use NUMA_NO_NODE instead of NODE_DATA(0)->node_id 2190 * because in some case like Node0 doesn't have RAM installed 2191 * low ram will be on Node1 2192 */ 2193 for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE, &start, &end, 2194 NULL) 2195 count += __free_memory_core(start, end); 2196 2197 return count; 2198 } 2199 2200 static int reset_managed_pages_done __initdata; 2201 2202 static void __init reset_node_managed_pages(pg_data_t *pgdat) 2203 { 2204 struct zone *z; 2205 2206 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++) 2207 atomic_long_set(&z->managed_pages, 0); 2208 } 2209 2210 void __init reset_all_zones_managed_pages(void) 2211 { 2212 struct pglist_data *pgdat; 2213 2214 if (reset_managed_pages_done) 2215 return; 2216 2217 for_each_online_pgdat(pgdat) 2218 reset_node_managed_pages(pgdat); 2219 2220 reset_managed_pages_done = 1; 2221 } 2222 2223 /** 2224 * memblock_free_all - release free pages to the buddy allocator 2225 */ 2226 void __init memblock_free_all(void) 2227 { 2228 unsigned long pages; 2229 2230 free_unused_memmap(); 2231 reset_all_zones_managed_pages(); 2232 2233 pages = free_low_memory_core_early(); 2234 totalram_pages_add(pages); 2235 } 2236 2237 #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_ARCH_KEEP_MEMBLOCK) 2238 static const char * const flagname[] = { 2239 [ilog2(MEMBLOCK_HOTPLUG)] = "HOTPLUG", 2240 [ilog2(MEMBLOCK_MIRROR)] = "MIRROR", 2241 [ilog2(MEMBLOCK_NOMAP)] = "NOMAP", 2242 [ilog2(MEMBLOCK_DRIVER_MANAGED)] = "DRV_MNG", 2243 [ilog2(MEMBLOCK_RSRV_NOINIT)] = "RSV_NIT", 2244 }; 2245 2246 static int memblock_debug_show(struct seq_file *m, void *private) 2247 { 2248 struct memblock_type *type = m->private; 2249 struct memblock_region *reg; 2250 int i, j, nid; 2251 unsigned int count = ARRAY_SIZE(flagname); 2252 phys_addr_t end; 2253 2254 for (i = 0; i < type->cnt; i++) { 2255 reg = &type->regions[i]; 2256 end = reg->base + reg->size - 1; 2257 nid = memblock_get_region_node(reg); 2258 2259 seq_printf(m, "%4d: ", i); 2260 seq_printf(m, "%pa..%pa ", ®->base, &end); 2261 if (numa_valid_node(nid)) 2262 seq_printf(m, "%4d ", nid); 2263 else 2264 seq_printf(m, "%4c ", 'x'); 2265 if (reg->flags) { 2266 for (j = 0; j < count; j++) { 2267 if (reg->flags & (1U << j)) { 2268 seq_printf(m, "%s\n", flagname[j]); 2269 break; 2270 } 2271 } 2272 if (j == count) 2273 seq_printf(m, "%s\n", "UNKNOWN"); 2274 } else { 2275 seq_printf(m, "%s\n", "NONE"); 2276 } 2277 } 2278 return 0; 2279 } 2280 DEFINE_SHOW_ATTRIBUTE(memblock_debug); 2281 2282 static int __init memblock_init_debugfs(void) 2283 { 2284 struct dentry *root = debugfs_create_dir("memblock", NULL); 2285 2286 debugfs_create_file("memory", 0444, root, 2287 &memblock.memory, &memblock_debug_fops); 2288 debugfs_create_file("reserved", 0444, root, 2289 &memblock.reserved, &memblock_debug_fops); 2290 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP 2291 debugfs_create_file("physmem", 0444, root, &physmem, 2292 &memblock_debug_fops); 2293 #endif 2294 2295 return 0; 2296 } 2297 __initcall(memblock_init_debugfs); 2298 2299 #endif /* CONFIG_DEBUG_FS */ 2300