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 100struct pglist_data __refdata contig_page_data; 101EXPORT_SYMBOL(contig_page_data); 102#endif 103 104unsigned long max_low_pfn; 105unsigned long min_low_pfn; 106unsigned long max_pfn; 107unsigned long long max_possible_pfn; 108 109static struct memblock_region memblock_memory_init_regions[INIT_MEMBLOCK_MEMORY_REGIONS] __initdata_memblock; 110static struct memblock_region memblock_reserved_init_regions[INIT_MEMBLOCK_RESERVED_REGIONS] __initdata_memblock; 111#ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP 112static struct memblock_region memblock_physmem_init_regions[INIT_PHYSMEM_REGIONS]; 113#endif 114 115struct 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 131struct 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 */ 145static __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 158static int memblock_debug __initdata_memblock; 159static bool system_has_some_mirror __initdata_memblock; 160static int memblock_can_resize __initdata_memblock; 161static int memblock_memory_in_slab __initdata_memblock; 162static int memblock_reserved_in_slab __initdata_memblock; 163 164bool __init_memblock memblock_has_mirror(void) 165{ 166 return system_has_some_mirror; 167} 168 169static 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 */ 175static 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 */ 183unsigned long __init_memblock 184memblock_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 190bool __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 */ 219static 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 */ 254static 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 */ 293static 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 */ 328static 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 335again: 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 349static 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 */ 371void __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 */ 414static 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 */ 513static 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 */ 554static 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 */ 588static 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 622repeat: 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 */ 706int __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 */ 728int __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 */ 749bool __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 (nid == NUMA_NO_NODE) 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 */ 788static 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 848static 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 863int __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 */ 881void __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 */ 895int __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 906int __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 917int __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 */ 940static 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 */ 969int __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 */ 981int __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 */ 993int __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 */ 1018int __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 */ 1030int __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 */ 1047int __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 1053static 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 (nid != NUMA_NO_NODE && 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 */ 1113void __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 if (WARN_ONCE(nid == MAX_NUMNODES, 1122 "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n")) 1123 nid = NUMA_NO_NODE; 1124 1125 for (; idx_a < type_a->cnt; idx_a++) { 1126 struct memblock_region *m = &type_a->regions[idx_a]; 1127 1128 phys_addr_t m_start = m->base; 1129 phys_addr_t m_end = m->base + m->size; 1130 int m_nid = memblock_get_region_node(m); 1131 1132 if (should_skip_region(type_a, m, nid, flags)) 1133 continue; 1134 1135 if (!type_b) { 1136 if (out_start) 1137 *out_start = m_start; 1138 if (out_end) 1139 *out_end = m_end; 1140 if (out_nid) 1141 *out_nid = m_nid; 1142 idx_a++; 1143 *idx = (u32)idx_a | (u64)idx_b << 32; 1144 return; 1145 } 1146 1147 /* scan areas before each reservation */ 1148 for (; idx_b < type_b->cnt + 1; idx_b++) { 1149 struct memblock_region *r; 1150 phys_addr_t r_start; 1151 phys_addr_t r_end; 1152 1153 r = &type_b->regions[idx_b]; 1154 r_start = idx_b ? r[-1].base + r[-1].size : 0; 1155 r_end = idx_b < type_b->cnt ? 1156 r->base : PHYS_ADDR_MAX; 1157 1158 /* 1159 * if idx_b advanced past idx_a, 1160 * break out to advance idx_a 1161 */ 1162 if (r_start >= m_end) 1163 break; 1164 /* if the two regions intersect, we're done */ 1165 if (m_start < r_end) { 1166 if (out_start) 1167 *out_start = 1168 max(m_start, r_start); 1169 if (out_end) 1170 *out_end = min(m_end, r_end); 1171 if (out_nid) 1172 *out_nid = m_nid; 1173 /* 1174 * The region which ends first is 1175 * advanced for the next iteration. 1176 */ 1177 if (m_end <= r_end) 1178 idx_a++; 1179 else 1180 idx_b++; 1181 *idx = (u32)idx_a | (u64)idx_b << 32; 1182 return; 1183 } 1184 } 1185 } 1186 1187 /* signal end of iteration */ 1188 *idx = ULLONG_MAX; 1189} 1190 1191/** 1192 * __next_mem_range_rev - generic next function for for_each_*_range_rev() 1193 * 1194 * @idx: pointer to u64 loop variable 1195 * @nid: node selector, %NUMA_NO_NODE for all nodes 1196 * @flags: pick from blocks based on memory attributes 1197 * @type_a: pointer to memblock_type from where the range is taken 1198 * @type_b: pointer to memblock_type which excludes memory from being taken 1199 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL 1200 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL 1201 * @out_nid: ptr to int for nid of the range, can be %NULL 1202 * 1203 * Finds the next range from type_a which is not marked as unsuitable 1204 * in type_b. 1205 * 1206 * Reverse of __next_mem_range(). 1207 */ 1208void __init_memblock __next_mem_range_rev(u64 *idx, int nid, 1209 enum memblock_flags flags, 1210 struct memblock_type *type_a, 1211 struct memblock_type *type_b, 1212 phys_addr_t *out_start, 1213 phys_addr_t *out_end, int *out_nid) 1214{ 1215 int idx_a = *idx & 0xffffffff; 1216 int idx_b = *idx >> 32; 1217 1218 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n")) 1219 nid = NUMA_NO_NODE; 1220 1221 if (*idx == (u64)ULLONG_MAX) { 1222 idx_a = type_a->cnt - 1; 1223 if (type_b != NULL) 1224 idx_b = type_b->cnt; 1225 else 1226 idx_b = 0; 1227 } 1228 1229 for (; idx_a >= 0; idx_a--) { 1230 struct memblock_region *m = &type_a->regions[idx_a]; 1231 1232 phys_addr_t m_start = m->base; 1233 phys_addr_t m_end = m->base + m->size; 1234 int m_nid = memblock_get_region_node(m); 1235 1236 if (should_skip_region(type_a, m, nid, flags)) 1237 continue; 1238 1239 if (!type_b) { 1240 if (out_start) 1241 *out_start = m_start; 1242 if (out_end) 1243 *out_end = m_end; 1244 if (out_nid) 1245 *out_nid = m_nid; 1246 idx_a--; 1247 *idx = (u32)idx_a | (u64)idx_b << 32; 1248 return; 1249 } 1250 1251 /* scan areas before each reservation */ 1252 for (; idx_b >= 0; idx_b--) { 1253 struct memblock_region *r; 1254 phys_addr_t r_start; 1255 phys_addr_t r_end; 1256 1257 r = &type_b->regions[idx_b]; 1258 r_start = idx_b ? r[-1].base + r[-1].size : 0; 1259 r_end = idx_b < type_b->cnt ? 1260 r->base : PHYS_ADDR_MAX; 1261 /* 1262 * if idx_b advanced past idx_a, 1263 * break out to advance idx_a 1264 */ 1265 1266 if (r_end <= m_start) 1267 break; 1268 /* if the two regions intersect, we're done */ 1269 if (m_end > r_start) { 1270 if (out_start) 1271 *out_start = max(m_start, r_start); 1272 if (out_end) 1273 *out_end = min(m_end, r_end); 1274 if (out_nid) 1275 *out_nid = m_nid; 1276 if (m_start >= r_start) 1277 idx_a--; 1278 else 1279 idx_b--; 1280 *idx = (u32)idx_a | (u64)idx_b << 32; 1281 return; 1282 } 1283 } 1284 } 1285 /* signal end of iteration */ 1286 *idx = ULLONG_MAX; 1287} 1288 1289/* 1290 * Common iterator interface used to define for_each_mem_pfn_range(). 1291 */ 1292void __init_memblock __next_mem_pfn_range(int *idx, int nid, 1293 unsigned long *out_start_pfn, 1294 unsigned long *out_end_pfn, int *out_nid) 1295{ 1296 struct memblock_type *type = &memblock.memory; 1297 struct memblock_region *r; 1298 int r_nid; 1299 1300 while (++*idx < type->cnt) { 1301 r = &type->regions[*idx]; 1302 r_nid = memblock_get_region_node(r); 1303 1304 if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size)) 1305 continue; 1306 if (nid == MAX_NUMNODES || nid == r_nid) 1307 break; 1308 } 1309 if (*idx >= type->cnt) { 1310 *idx = -1; 1311 return; 1312 } 1313 1314 if (out_start_pfn) 1315 *out_start_pfn = PFN_UP(r->base); 1316 if (out_end_pfn) 1317 *out_end_pfn = PFN_DOWN(r->base + r->size); 1318 if (out_nid) 1319 *out_nid = r_nid; 1320} 1321 1322/** 1323 * memblock_set_node - set node ID on memblock regions 1324 * @base: base of area to set node ID for 1325 * @size: size of area to set node ID for 1326 * @type: memblock type to set node ID for 1327 * @nid: node ID to set 1328 * 1329 * Set the nid of memblock @type regions in [@base, @base + @size) to @nid. 1330 * Regions which cross the area boundaries are split as necessary. 1331 * 1332 * Return: 1333 * 0 on success, -errno on failure. 1334 */ 1335int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size, 1336 struct memblock_type *type, int nid) 1337{ 1338#ifdef CONFIG_NUMA 1339 int start_rgn, end_rgn; 1340 int i, ret; 1341 1342 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn); 1343 if (ret) 1344 return ret; 1345 1346 for (i = start_rgn; i < end_rgn; i++) 1347 memblock_set_region_node(&type->regions[i], nid); 1348 1349 memblock_merge_regions(type, start_rgn, end_rgn); 1350#endif 1351 return 0; 1352} 1353 1354#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT 1355/** 1356 * __next_mem_pfn_range_in_zone - iterator for for_each_*_range_in_zone() 1357 * 1358 * @idx: pointer to u64 loop variable 1359 * @zone: zone in which all of the memory blocks reside 1360 * @out_spfn: ptr to ulong for start pfn of the range, can be %NULL 1361 * @out_epfn: ptr to ulong for end pfn of the range, can be %NULL 1362 * 1363 * This function is meant to be a zone/pfn specific wrapper for the 1364 * for_each_mem_range type iterators. Specifically they are used in the 1365 * deferred memory init routines and as such we were duplicating much of 1366 * this logic throughout the code. So instead of having it in multiple 1367 * locations it seemed like it would make more sense to centralize this to 1368 * one new iterator that does everything they need. 1369 */ 1370void __init_memblock 1371__next_mem_pfn_range_in_zone(u64 *idx, struct zone *zone, 1372 unsigned long *out_spfn, unsigned long *out_epfn) 1373{ 1374 int zone_nid = zone_to_nid(zone); 1375 phys_addr_t spa, epa; 1376 1377 __next_mem_range(idx, zone_nid, MEMBLOCK_NONE, 1378 &memblock.memory, &memblock.reserved, 1379 &spa, &epa, NULL); 1380 1381 while (*idx != U64_MAX) { 1382 unsigned long epfn = PFN_DOWN(epa); 1383 unsigned long spfn = PFN_UP(spa); 1384 1385 /* 1386 * Verify the end is at least past the start of the zone and 1387 * that we have at least one PFN to initialize. 1388 */ 1389 if (zone->zone_start_pfn < epfn && spfn < epfn) { 1390 /* if we went too far just stop searching */ 1391 if (zone_end_pfn(zone) <= spfn) { 1392 *idx = U64_MAX; 1393 break; 1394 } 1395 1396 if (out_spfn) 1397 *out_spfn = max(zone->zone_start_pfn, spfn); 1398 if (out_epfn) 1399 *out_epfn = min(zone_end_pfn(zone), epfn); 1400 1401 return; 1402 } 1403 1404 __next_mem_range(idx, zone_nid, MEMBLOCK_NONE, 1405 &memblock.memory, &memblock.reserved, 1406 &spa, &epa, NULL); 1407 } 1408 1409 /* signal end of iteration */ 1410 if (out_spfn) 1411 *out_spfn = ULONG_MAX; 1412 if (out_epfn) 1413 *out_epfn = 0; 1414} 1415 1416#endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ 1417 1418/** 1419 * memblock_alloc_range_nid - allocate boot memory block 1420 * @size: size of memory block to be allocated in bytes 1421 * @align: alignment of the region and block's size 1422 * @start: the lower bound of the memory region to allocate (phys address) 1423 * @end: the upper bound of the memory region to allocate (phys address) 1424 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 1425 * @exact_nid: control the allocation fall back to other nodes 1426 * 1427 * The allocation is performed from memory region limited by 1428 * memblock.current_limit if @end == %MEMBLOCK_ALLOC_ACCESSIBLE. 1429 * 1430 * If the specified node can not hold the requested memory and @exact_nid 1431 * is false, the allocation falls back to any node in the system. 1432 * 1433 * For systems with memory mirroring, the allocation is attempted first 1434 * from the regions with mirroring enabled and then retried from any 1435 * memory region. 1436 * 1437 * In addition, function using kmemleak_alloc_phys for allocated boot 1438 * memory block, it is never reported as leaks. 1439 * 1440 * Return: 1441 * Physical address of allocated memory block on success, %0 on failure. 1442 */ 1443phys_addr_t __init memblock_alloc_range_nid(phys_addr_t size, 1444 phys_addr_t align, phys_addr_t start, 1445 phys_addr_t end, int nid, 1446 bool exact_nid) 1447{ 1448 enum memblock_flags flags = choose_memblock_flags(); 1449 phys_addr_t found; 1450 1451 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n")) 1452 nid = NUMA_NO_NODE; 1453 1454 if (!align) { 1455 /* Can't use WARNs this early in boot on powerpc */ 1456 dump_stack(); 1457 align = SMP_CACHE_BYTES; 1458 } 1459 1460again: 1461 found = memblock_find_in_range_node(size, align, start, end, nid, 1462 flags); 1463 if (found && !memblock_reserve(found, size)) 1464 goto done; 1465 1466 if (nid != NUMA_NO_NODE && !exact_nid) { 1467 found = memblock_find_in_range_node(size, align, start, 1468 end, NUMA_NO_NODE, 1469 flags); 1470 if (found && !memblock_reserve(found, size)) 1471 goto done; 1472 } 1473 1474 if (flags & MEMBLOCK_MIRROR) { 1475 flags &= ~MEMBLOCK_MIRROR; 1476 pr_warn_ratelimited("Could not allocate %pap bytes of mirrored memory\n", 1477 &size); 1478 goto again; 1479 } 1480 1481 return 0; 1482 1483done: 1484 /* 1485 * Skip kmemleak for those places like kasan_init() and 1486 * early_pgtable_alloc() due to high volume. 1487 */ 1488 if (end != MEMBLOCK_ALLOC_NOLEAKTRACE) 1489 /* 1490 * Memblock allocated blocks are never reported as 1491 * leaks. This is because many of these blocks are 1492 * only referred via the physical address which is 1493 * not looked up by kmemleak. 1494 */ 1495 kmemleak_alloc_phys(found, size, 0); 1496 1497 /* 1498 * Some Virtual Machine platforms, such as Intel TDX or AMD SEV-SNP, 1499 * require memory to be accepted before it can be used by the 1500 * guest. 1501 * 1502 * Accept the memory of the allocated buffer. 1503 */ 1504 accept_memory(found, found + size); 1505 1506 return found; 1507} 1508 1509/** 1510 * memblock_phys_alloc_range - allocate a memory block inside specified range 1511 * @size: size of memory block to be allocated in bytes 1512 * @align: alignment of the region and block's size 1513 * @start: the lower bound of the memory region to allocate (physical address) 1514 * @end: the upper bound of the memory region to allocate (physical address) 1515 * 1516 * Allocate @size bytes in the between @start and @end. 1517 * 1518 * Return: physical address of the allocated memory block on success, 1519 * %0 on failure. 1520 */ 1521phys_addr_t __init memblock_phys_alloc_range(phys_addr_t size, 1522 phys_addr_t align, 1523 phys_addr_t start, 1524 phys_addr_t end) 1525{ 1526 memblock_dbg("%s: %llu bytes align=0x%llx from=%pa max_addr=%pa %pS\n", 1527 __func__, (u64)size, (u64)align, &start, &end, 1528 (void *)_RET_IP_); 1529 return memblock_alloc_range_nid(size, align, start, end, NUMA_NO_NODE, 1530 false); 1531} 1532 1533/** 1534 * memblock_phys_alloc_try_nid - allocate a memory block from specified NUMA node 1535 * @size: size of memory block to be allocated in bytes 1536 * @align: alignment of the region and block's size 1537 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 1538 * 1539 * Allocates memory block from the specified NUMA node. If the node 1540 * has no available memory, attempts to allocated from any node in the 1541 * system. 1542 * 1543 * Return: physical address of the allocated memory block on success, 1544 * %0 on failure. 1545 */ 1546phys_addr_t __init memblock_phys_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid) 1547{ 1548 return memblock_alloc_range_nid(size, align, 0, 1549 MEMBLOCK_ALLOC_ACCESSIBLE, nid, false); 1550} 1551 1552/** 1553 * memblock_alloc_internal - allocate boot memory block 1554 * @size: size of memory block to be allocated in bytes 1555 * @align: alignment of the region and block's size 1556 * @min_addr: the lower bound of the memory region to allocate (phys address) 1557 * @max_addr: the upper bound of the memory region to allocate (phys address) 1558 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 1559 * @exact_nid: control the allocation fall back to other nodes 1560 * 1561 * Allocates memory block using memblock_alloc_range_nid() and 1562 * converts the returned physical address to virtual. 1563 * 1564 * The @min_addr limit is dropped if it can not be satisfied and the allocation 1565 * will fall back to memory below @min_addr. Other constraints, such 1566 * as node and mirrored memory will be handled again in 1567 * memblock_alloc_range_nid(). 1568 * 1569 * Return: 1570 * Virtual address of allocated memory block on success, NULL on failure. 1571 */ 1572static void * __init memblock_alloc_internal( 1573 phys_addr_t size, phys_addr_t align, 1574 phys_addr_t min_addr, phys_addr_t max_addr, 1575 int nid, bool exact_nid) 1576{ 1577 phys_addr_t alloc; 1578 1579 /* 1580 * Detect any accidental use of these APIs after slab is ready, as at 1581 * this moment memblock may be deinitialized already and its 1582 * internal data may be destroyed (after execution of memblock_free_all) 1583 */ 1584 if (WARN_ON_ONCE(slab_is_available())) 1585 return kzalloc_node(size, GFP_NOWAIT, nid); 1586 1587 if (max_addr > memblock.current_limit) 1588 max_addr = memblock.current_limit; 1589 1590 alloc = memblock_alloc_range_nid(size, align, min_addr, max_addr, nid, 1591 exact_nid); 1592 1593 /* retry allocation without lower limit */ 1594 if (!alloc && min_addr) 1595 alloc = memblock_alloc_range_nid(size, align, 0, max_addr, nid, 1596 exact_nid); 1597 1598 if (!alloc) 1599 return NULL; 1600 1601 return phys_to_virt(alloc); 1602} 1603 1604/** 1605 * memblock_alloc_exact_nid_raw - allocate boot memory block on the exact node 1606 * without zeroing memory 1607 * @size: size of memory block to be allocated in bytes 1608 * @align: alignment of the region and block's size 1609 * @min_addr: the lower bound of the memory region from where the allocation 1610 * is preferred (phys address) 1611 * @max_addr: the upper bound of the memory region from where the allocation 1612 * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to 1613 * allocate only from memory limited by memblock.current_limit value 1614 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 1615 * 1616 * Public function, provides additional debug information (including caller 1617 * info), if enabled. Does not zero allocated memory. 1618 * 1619 * Return: 1620 * Virtual address of allocated memory block on success, NULL on failure. 1621 */ 1622void * __init memblock_alloc_exact_nid_raw( 1623 phys_addr_t size, phys_addr_t align, 1624 phys_addr_t min_addr, phys_addr_t max_addr, 1625 int nid) 1626{ 1627 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n", 1628 __func__, (u64)size, (u64)align, nid, &min_addr, 1629 &max_addr, (void *)_RET_IP_); 1630 1631 return memblock_alloc_internal(size, align, min_addr, max_addr, nid, 1632 true); 1633} 1634 1635/** 1636 * memblock_alloc_try_nid_raw - allocate boot memory block without zeroing 1637 * memory and without panicking 1638 * @size: size of memory block to be allocated in bytes 1639 * @align: alignment of the region and block's size 1640 * @min_addr: the lower bound of the memory region from where the allocation 1641 * is preferred (phys address) 1642 * @max_addr: the upper bound of the memory region from where the allocation 1643 * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to 1644 * allocate only from memory limited by memblock.current_limit value 1645 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 1646 * 1647 * Public function, provides additional debug information (including caller 1648 * info), if enabled. Does not zero allocated memory, does not panic if request 1649 * cannot be satisfied. 1650 * 1651 * Return: 1652 * Virtual address of allocated memory block on success, NULL on failure. 1653 */ 1654void * __init memblock_alloc_try_nid_raw( 1655 phys_addr_t size, phys_addr_t align, 1656 phys_addr_t min_addr, phys_addr_t max_addr, 1657 int nid) 1658{ 1659 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n", 1660 __func__, (u64)size, (u64)align, nid, &min_addr, 1661 &max_addr, (void *)_RET_IP_); 1662 1663 return memblock_alloc_internal(size, align, min_addr, max_addr, nid, 1664 false); 1665} 1666 1667/** 1668 * memblock_alloc_try_nid - allocate boot memory block 1669 * @size: size of memory block to be allocated in bytes 1670 * @align: alignment of the region and block's size 1671 * @min_addr: the lower bound of the memory region from where the allocation 1672 * is preferred (phys address) 1673 * @max_addr: the upper bound of the memory region from where the allocation 1674 * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to 1675 * allocate only from memory limited by memblock.current_limit value 1676 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node 1677 * 1678 * Public function, provides additional debug information (including caller 1679 * info), if enabled. This function zeroes the allocated memory. 1680 * 1681 * Return: 1682 * Virtual address of allocated memory block on success, NULL on failure. 1683 */ 1684void * __init memblock_alloc_try_nid( 1685 phys_addr_t size, phys_addr_t align, 1686 phys_addr_t min_addr, phys_addr_t max_addr, 1687 int nid) 1688{ 1689 void *ptr; 1690 1691 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n", 1692 __func__, (u64)size, (u64)align, nid, &min_addr, 1693 &max_addr, (void *)_RET_IP_); 1694 ptr = memblock_alloc_internal(size, align, 1695 min_addr, max_addr, nid, false); 1696 if (ptr) 1697 memset(ptr, 0, size); 1698 1699 return ptr; 1700} 1701 1702/** 1703 * memblock_free_late - free pages directly to buddy allocator 1704 * @base: phys starting address of the boot memory block 1705 * @size: size of the boot memory block in bytes 1706 * 1707 * This is only useful when the memblock allocator has already been torn 1708 * down, but we are still initializing the system. Pages are released directly 1709 * to the buddy allocator. 1710 */ 1711void __init memblock_free_late(phys_addr_t base, phys_addr_t size) 1712{ 1713 phys_addr_t cursor, end; 1714 1715 end = base + size - 1; 1716 memblock_dbg("%s: [%pa-%pa] %pS\n", 1717 __func__, &base, &end, (void *)_RET_IP_); 1718 kmemleak_free_part_phys(base, size); 1719 cursor = PFN_UP(base); 1720 end = PFN_DOWN(base + size); 1721 1722 for (; cursor < end; cursor++) { 1723 memblock_free_pages(pfn_to_page(cursor), cursor, 0); 1724 totalram_pages_inc(); 1725 } 1726} 1727 1728/* 1729 * Remaining API functions 1730 */ 1731 1732phys_addr_t __init_memblock memblock_phys_mem_size(void) 1733{ 1734 return memblock.memory.total_size; 1735} 1736 1737phys_addr_t __init_memblock memblock_reserved_size(void) 1738{ 1739 return memblock.reserved.total_size; 1740} 1741 1742/* lowest address */ 1743phys_addr_t __init_memblock memblock_start_of_DRAM(void) 1744{ 1745 return memblock.memory.regions[0].base; 1746} 1747 1748phys_addr_t __init_memblock memblock_end_of_DRAM(void) 1749{ 1750 int idx = memblock.memory.cnt - 1; 1751 1752 return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size); 1753} 1754 1755static phys_addr_t __init_memblock __find_max_addr(phys_addr_t limit) 1756{ 1757 phys_addr_t max_addr = PHYS_ADDR_MAX; 1758 struct memblock_region *r; 1759 1760 /* 1761 * translate the memory @limit size into the max address within one of 1762 * the memory memblock regions, if the @limit exceeds the total size 1763 * of those regions, max_addr will keep original value PHYS_ADDR_MAX 1764 */ 1765 for_each_mem_region(r) { 1766 if (limit <= r->size) { 1767 max_addr = r->base + limit; 1768 break; 1769 } 1770 limit -= r->size; 1771 } 1772 1773 return max_addr; 1774} 1775 1776void __init memblock_enforce_memory_limit(phys_addr_t limit) 1777{ 1778 phys_addr_t max_addr; 1779 1780 if (!limit) 1781 return; 1782 1783 max_addr = __find_max_addr(limit); 1784 1785 /* @limit exceeds the total size of the memory, do nothing */ 1786 if (max_addr == PHYS_ADDR_MAX) 1787 return; 1788 1789 /* truncate both memory and reserved regions */ 1790 memblock_remove_range(&memblock.memory, max_addr, 1791 PHYS_ADDR_MAX); 1792 memblock_remove_range(&memblock.reserved, max_addr, 1793 PHYS_ADDR_MAX); 1794} 1795 1796void __init memblock_cap_memory_range(phys_addr_t base, phys_addr_t size) 1797{ 1798 int start_rgn, end_rgn; 1799 int i, ret; 1800 1801 if (!size) 1802 return; 1803 1804 if (!memblock_memory->total_size) { 1805 pr_warn("%s: No memory registered yet\n", __func__); 1806 return; 1807 } 1808 1809 ret = memblock_isolate_range(&memblock.memory, base, size, 1810 &start_rgn, &end_rgn); 1811 if (ret) 1812 return; 1813 1814 /* remove all the MAP regions */ 1815 for (i = memblock.memory.cnt - 1; i >= end_rgn; i--) 1816 if (!memblock_is_nomap(&memblock.memory.regions[i])) 1817 memblock_remove_region(&memblock.memory, i); 1818 1819 for (i = start_rgn - 1; i >= 0; i--) 1820 if (!memblock_is_nomap(&memblock.memory.regions[i])) 1821 memblock_remove_region(&memblock.memory, i); 1822 1823 /* truncate the reserved regions */ 1824 memblock_remove_range(&memblock.reserved, 0, base); 1825 memblock_remove_range(&memblock.reserved, 1826 base + size, PHYS_ADDR_MAX); 1827} 1828 1829void __init memblock_mem_limit_remove_map(phys_addr_t limit) 1830{ 1831 phys_addr_t max_addr; 1832 1833 if (!limit) 1834 return; 1835 1836 max_addr = __find_max_addr(limit); 1837 1838 /* @limit exceeds the total size of the memory, do nothing */ 1839 if (max_addr == PHYS_ADDR_MAX) 1840 return; 1841 1842 memblock_cap_memory_range(0, max_addr); 1843} 1844 1845static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr) 1846{ 1847 unsigned int left = 0, right = type->cnt; 1848 1849 do { 1850 unsigned int mid = (right + left) / 2; 1851 1852 if (addr < type->regions[mid].base) 1853 right = mid; 1854 else if (addr >= (type->regions[mid].base + 1855 type->regions[mid].size)) 1856 left = mid + 1; 1857 else 1858 return mid; 1859 } while (left < right); 1860 return -1; 1861} 1862 1863bool __init_memblock memblock_is_reserved(phys_addr_t addr) 1864{ 1865 return memblock_search(&memblock.reserved, addr) != -1; 1866} 1867 1868bool __init_memblock memblock_is_memory(phys_addr_t addr) 1869{ 1870 return memblock_search(&memblock.memory, addr) != -1; 1871} 1872 1873bool __init_memblock memblock_is_map_memory(phys_addr_t addr) 1874{ 1875 int i = memblock_search(&memblock.memory, addr); 1876 1877 if (i == -1) 1878 return false; 1879 return !memblock_is_nomap(&memblock.memory.regions[i]); 1880} 1881 1882int __init_memblock memblock_search_pfn_nid(unsigned long pfn, 1883 unsigned long *start_pfn, unsigned long *end_pfn) 1884{ 1885 struct memblock_type *type = &memblock.memory; 1886 int mid = memblock_search(type, PFN_PHYS(pfn)); 1887 1888 if (mid == -1) 1889 return NUMA_NO_NODE; 1890 1891 *start_pfn = PFN_DOWN(type->regions[mid].base); 1892 *end_pfn = PFN_DOWN(type->regions[mid].base + type->regions[mid].size); 1893 1894 return memblock_get_region_node(&type->regions[mid]); 1895} 1896 1897/** 1898 * memblock_is_region_memory - check if a region is a subset of memory 1899 * @base: base of region to check 1900 * @size: size of region to check 1901 * 1902 * Check if the region [@base, @base + @size) is a subset of a memory block. 1903 * 1904 * Return: 1905 * 0 if false, non-zero if true 1906 */ 1907bool __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size) 1908{ 1909 int idx = memblock_search(&memblock.memory, base); 1910 phys_addr_t end = base + memblock_cap_size(base, &size); 1911 1912 if (idx == -1) 1913 return false; 1914 return (memblock.memory.regions[idx].base + 1915 memblock.memory.regions[idx].size) >= end; 1916} 1917 1918/** 1919 * memblock_is_region_reserved - check if a region intersects reserved memory 1920 * @base: base of region to check 1921 * @size: size of region to check 1922 * 1923 * Check if the region [@base, @base + @size) intersects a reserved 1924 * memory block. 1925 * 1926 * Return: 1927 * True if they intersect, false if not. 1928 */ 1929bool __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size) 1930{ 1931 return memblock_overlaps_region(&memblock.reserved, base, size); 1932} 1933 1934void __init_memblock memblock_trim_memory(phys_addr_t align) 1935{ 1936 phys_addr_t start, end, orig_start, orig_end; 1937 struct memblock_region *r; 1938 1939 for_each_mem_region(r) { 1940 orig_start = r->base; 1941 orig_end = r->base + r->size; 1942 start = round_up(orig_start, align); 1943 end = round_down(orig_end, align); 1944 1945 if (start == orig_start && end == orig_end) 1946 continue; 1947 1948 if (start < end) { 1949 r->base = start; 1950 r->size = end - start; 1951 } else { 1952 memblock_remove_region(&memblock.memory, 1953 r - memblock.memory.regions); 1954 r--; 1955 } 1956 } 1957} 1958 1959void __init_memblock memblock_set_current_limit(phys_addr_t limit) 1960{ 1961 memblock.current_limit = limit; 1962} 1963 1964phys_addr_t __init_memblock memblock_get_current_limit(void) 1965{ 1966 return memblock.current_limit; 1967} 1968 1969static void __init_memblock memblock_dump(struct memblock_type *type) 1970{ 1971 phys_addr_t base, end, size; 1972 enum memblock_flags flags; 1973 int idx; 1974 struct memblock_region *rgn; 1975 1976 pr_info(" %s.cnt = 0x%lx\n", type->name, type->cnt); 1977 1978 for_each_memblock_type(idx, type, rgn) { 1979 char nid_buf[32] = ""; 1980 1981 base = rgn->base; 1982 size = rgn->size; 1983 end = base + size - 1; 1984 flags = rgn->flags; 1985#ifdef CONFIG_NUMA 1986 if (memblock_get_region_node(rgn) != MAX_NUMNODES) 1987 snprintf(nid_buf, sizeof(nid_buf), " on node %d", 1988 memblock_get_region_node(rgn)); 1989#endif 1990 pr_info(" %s[%#x]\t[%pa-%pa], %pa bytes%s flags: %#x\n", 1991 type->name, idx, &base, &end, &size, nid_buf, flags); 1992 } 1993} 1994 1995static void __init_memblock __memblock_dump_all(void) 1996{ 1997 pr_info("MEMBLOCK configuration:\n"); 1998 pr_info(" memory size = %pa reserved size = %pa\n", 1999 &memblock.memory.total_size, 2000 &memblock.reserved.total_size); 2001 2002 memblock_dump(&memblock.memory); 2003 memblock_dump(&memblock.reserved); 2004#ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP 2005 memblock_dump(&physmem); 2006#endif 2007} 2008 2009void __init_memblock memblock_dump_all(void) 2010{ 2011 if (memblock_debug) 2012 __memblock_dump_all(); 2013} 2014 2015void __init memblock_allow_resize(void) 2016{ 2017 memblock_can_resize = 1; 2018} 2019 2020static int __init early_memblock(char *p) 2021{ 2022 if (p && strstr(p, "debug")) 2023 memblock_debug = 1; 2024 return 0; 2025} 2026early_param("memblock", early_memblock); 2027 2028static void __init free_memmap(unsigned long start_pfn, unsigned long end_pfn) 2029{ 2030 struct page *start_pg, *end_pg; 2031 phys_addr_t pg, pgend; 2032 2033 /* 2034 * Convert start_pfn/end_pfn to a struct page pointer. 2035 */ 2036 start_pg = pfn_to_page(start_pfn - 1) + 1; 2037 end_pg = pfn_to_page(end_pfn - 1) + 1; 2038 2039 /* 2040 * Convert to physical addresses, and round start upwards and end 2041 * downwards. 2042 */ 2043 pg = PAGE_ALIGN(__pa(start_pg)); 2044 pgend = __pa(end_pg) & PAGE_MASK; 2045 2046 /* 2047 * If there are free pages between these, free the section of the 2048 * memmap array. 2049 */ 2050 if (pg < pgend) 2051 memblock_phys_free(pg, pgend - pg); 2052} 2053 2054/* 2055 * The mem_map array can get very big. Free the unused area of the memory map. 2056 */ 2057static void __init free_unused_memmap(void) 2058{ 2059 unsigned long start, end, prev_end = 0; 2060 int i; 2061 2062 if (!IS_ENABLED(CONFIG_HAVE_ARCH_PFN_VALID) || 2063 IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP)) 2064 return; 2065 2066 /* 2067 * This relies on each bank being in address order. 2068 * The banks are sorted previously in bootmem_init(). 2069 */ 2070 for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, NULL) { 2071#ifdef CONFIG_SPARSEMEM 2072 /* 2073 * Take care not to free memmap entries that don't exist 2074 * due to SPARSEMEM sections which aren't present. 2075 */ 2076 start = min(start, ALIGN(prev_end, PAGES_PER_SECTION)); 2077#endif 2078 /* 2079 * Align down here since many operations in VM subsystem 2080 * presume that there are no holes in the memory map inside 2081 * a pageblock 2082 */ 2083 start = pageblock_start_pfn(start); 2084 2085 /* 2086 * If we had a previous bank, and there is a space 2087 * between the current bank and the previous, free it. 2088 */ 2089 if (prev_end && prev_end < start) 2090 free_memmap(prev_end, start); 2091 2092 /* 2093 * Align up here since many operations in VM subsystem 2094 * presume that there are no holes in the memory map inside 2095 * a pageblock 2096 */ 2097 prev_end = pageblock_align(end); 2098 } 2099 2100#ifdef CONFIG_SPARSEMEM 2101 if (!IS_ALIGNED(prev_end, PAGES_PER_SECTION)) { 2102 prev_end = pageblock_align(end); 2103 free_memmap(prev_end, ALIGN(prev_end, PAGES_PER_SECTION)); 2104 } 2105#endif 2106} 2107 2108static void __init __free_pages_memory(unsigned long start, unsigned long end) 2109{ 2110 int order; 2111 2112 while (start < end) { 2113 /* 2114 * Free the pages in the largest chunks alignment allows. 2115 * 2116 * __ffs() behaviour is undefined for 0. start == 0 is 2117 * MAX_PAGE_ORDER-aligned, set order to MAX_PAGE_ORDER for 2118 * the case. 2119 */ 2120 if (start) 2121 order = min_t(int, MAX_PAGE_ORDER, __ffs(start)); 2122 else 2123 order = MAX_PAGE_ORDER; 2124 2125 while (start + (1UL << order) > end) 2126 order--; 2127 2128 memblock_free_pages(pfn_to_page(start), start, order); 2129 2130 start += (1UL << order); 2131 } 2132} 2133 2134static unsigned long __init __free_memory_core(phys_addr_t start, 2135 phys_addr_t end) 2136{ 2137 unsigned long start_pfn = PFN_UP(start); 2138 unsigned long end_pfn = min_t(unsigned long, 2139 PFN_DOWN(end), max_low_pfn); 2140 2141 if (start_pfn >= end_pfn) 2142 return 0; 2143 2144 __free_pages_memory(start_pfn, end_pfn); 2145 2146 return end_pfn - start_pfn; 2147} 2148 2149static void __init memmap_init_reserved_pages(void) 2150{ 2151 struct memblock_region *region; 2152 phys_addr_t start, end; 2153 int nid; 2154 2155 /* 2156 * set nid on all reserved pages and also treat struct 2157 * pages for the NOMAP regions as PageReserved 2158 */ 2159 for_each_mem_region(region) { 2160 nid = memblock_get_region_node(region); 2161 start = region->base; 2162 end = start + region->size; 2163 2164 if (memblock_is_nomap(region)) 2165 reserve_bootmem_region(start, end, nid); 2166 2167 memblock_set_node(start, end, &memblock.reserved, nid); 2168 } 2169 2170 /* 2171 * initialize struct pages for reserved regions that don't have 2172 * the MEMBLOCK_RSRV_NOINIT flag set 2173 */ 2174 for_each_reserved_mem_region(region) { 2175 if (!memblock_is_reserved_noinit(region)) { 2176 nid = memblock_get_region_node(region); 2177 start = region->base; 2178 end = start + region->size; 2179 2180 if (nid == NUMA_NO_NODE || nid >= MAX_NUMNODES) 2181 nid = early_pfn_to_nid(PFN_DOWN(start)); 2182 2183 reserve_bootmem_region(start, end, nid); 2184 } 2185 } 2186} 2187 2188static unsigned long __init free_low_memory_core_early(void) 2189{ 2190 unsigned long count = 0; 2191 phys_addr_t start, end; 2192 u64 i; 2193 2194 memblock_clear_hotplug(0, -1); 2195 2196 memmap_init_reserved_pages(); 2197 2198 /* 2199 * We need to use NUMA_NO_NODE instead of NODE_DATA(0)->node_id 2200 * because in some case like Node0 doesn't have RAM installed 2201 * low ram will be on Node1 2202 */ 2203 for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE, &start, &end, 2204 NULL) 2205 count += __free_memory_core(start, end); 2206 2207 return count; 2208} 2209 2210static int reset_managed_pages_done __initdata; 2211 2212static void __init reset_node_managed_pages(pg_data_t *pgdat) 2213{ 2214 struct zone *z; 2215 2216 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++) 2217 atomic_long_set(&z->managed_pages, 0); 2218} 2219 2220void __init reset_all_zones_managed_pages(void) 2221{ 2222 struct pglist_data *pgdat; 2223 2224 if (reset_managed_pages_done) 2225 return; 2226 2227 for_each_online_pgdat(pgdat) 2228 reset_node_managed_pages(pgdat); 2229 2230 reset_managed_pages_done = 1; 2231} 2232 2233/** 2234 * memblock_free_all - release free pages to the buddy allocator 2235 */ 2236void __init memblock_free_all(void) 2237{ 2238 unsigned long pages; 2239 2240 free_unused_memmap(); 2241 reset_all_zones_managed_pages(); 2242 2243 pages = free_low_memory_core_early(); 2244 totalram_pages_add(pages); 2245} 2246 2247#if defined(CONFIG_DEBUG_FS) && defined(CONFIG_ARCH_KEEP_MEMBLOCK) 2248static const char * const flagname[] = { 2249 [ilog2(MEMBLOCK_HOTPLUG)] = "HOTPLUG", 2250 [ilog2(MEMBLOCK_MIRROR)] = "MIRROR", 2251 [ilog2(MEMBLOCK_NOMAP)] = "NOMAP", 2252 [ilog2(MEMBLOCK_DRIVER_MANAGED)] = "DRV_MNG", 2253 [ilog2(MEMBLOCK_RSRV_NOINIT)] = "RSV_NIT", 2254}; 2255 2256static int memblock_debug_show(struct seq_file *m, void *private) 2257{ 2258 struct memblock_type *type = m->private; 2259 struct memblock_region *reg; 2260 int i, j, nid; 2261 unsigned int count = ARRAY_SIZE(flagname); 2262 phys_addr_t end; 2263 2264 for (i = 0; i < type->cnt; i++) { 2265 reg = &type->regions[i]; 2266 end = reg->base + reg->size - 1; 2267 nid = memblock_get_region_node(reg); 2268 2269 seq_printf(m, "%4d: ", i); 2270 seq_printf(m, "%pa..%pa ", ®->base, &end); 2271 if (nid != MAX_NUMNODES) 2272 seq_printf(m, "%4d ", nid); 2273 else 2274 seq_printf(m, "%4c ", 'x'); 2275 if (reg->flags) { 2276 for (j = 0; j < count; j++) { 2277 if (reg->flags & (1U << j)) { 2278 seq_printf(m, "%s\n", flagname[j]); 2279 break; 2280 } 2281 } 2282 if (j == count) 2283 seq_printf(m, "%s\n", "UNKNOWN"); 2284 } else { 2285 seq_printf(m, "%s\n", "NONE"); 2286 } 2287 } 2288 return 0; 2289} 2290DEFINE_SHOW_ATTRIBUTE(memblock_debug); 2291 2292static int __init memblock_init_debugfs(void) 2293{ 2294 struct dentry *root = debugfs_create_dir("memblock", NULL); 2295 2296 debugfs_create_file("memory", 0444, root, 2297 &memblock.memory, &memblock_debug_fops); 2298 debugfs_create_file("reserved", 0444, root, 2299 &memblock.reserved, &memblock_debug_fops); 2300#ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP 2301 debugfs_create_file("physmem", 0444, root, &physmem, 2302 &memblock_debug_fops); 2303#endif 2304 2305 return 0; 2306} 2307__initcall(memblock_init_debugfs); 2308 2309#endif /* CONFIG_DEBUG_FS */ 2310