| 1 | |
| 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 | #include <linux/mutex.h>
|
| 20 | #include <linux/string_helpers.h>
|
| 21 |
|
| 22 | #include <linux/libfdt.h>
|
| 23 | #include <linux/kexec_handover.h>
|
| 24 | #include <linux/kho/abi/memblock.h>
|
| 25 |
|
| 26 | #include <asm/sections.h>
|
| 27 | #include <linux/io.h>
|
| 28 |
|
| 29 | #include "internal.h"
|
| 30 | #include "mm_init.h"
|
| 31 |
|
| 32 | #define INIT_MEMBLOCK_REGIONS 128
|
| 33 | #define INIT_PHYSMEM_REGIONS 4
|
| 34 |
|
| 35 | #ifndef INIT_MEMBLOCK_RESERVED_REGIONS
|
| 36 | # define INIT_MEMBLOCK_RESERVED_REGIONS INIT_MEMBLOCK_REGIONS
|
| 37 | #endif
|
| 38 |
|
| 39 | #ifndef INIT_MEMBLOCK_MEMORY_REGIONS
|
| 40 | #define INIT_MEMBLOCK_MEMORY_REGIONS INIT_MEMBLOCK_REGIONS
|
| 41 | #endif
|
| 42 |
|
| 43 | |
| 44 | * DOC: memblock overview
|
| 45 | *
|
| 46 | * Memblock is a method of managing memory regions during the early
|
| 47 | * boot period when the usual kernel memory allocators are not up and
|
| 48 | * running.
|
| 49 | *
|
| 50 | * Memblock views the system memory as collections of contiguous
|
| 51 | * regions. There are several types of these collections:
|
| 52 | *
|
| 53 | * * ``memory`` - describes the physical memory available to the
|
| 54 | * kernel; this may differ from the actual physical memory installed
|
| 55 | * in the system, for instance when the memory is restricted with
|
| 56 | * ``mem=`` command line parameter
|
| 57 | * * ``reserved`` - describes the regions that were allocated
|
| 58 | * * ``physmem`` - describes the actual physical memory available during
|
| 59 | * boot regardless of the possible restrictions and memory hot(un)plug;
|
| 60 | * the ``physmem`` type is only available on some architectures.
|
| 61 | *
|
| 62 | * Each region is represented by struct memblock_region that
|
| 63 | * defines the region extents, its attributes and NUMA node id on NUMA
|
| 64 | * systems. Every memory type is described by the struct memblock_type
|
| 65 | * which contains an array of memory regions along with
|
| 66 | * the allocator metadata. The "memory" and "reserved" types are nicely
|
| 67 | * wrapped with struct memblock. This structure is statically
|
| 68 | * initialized at build time. The region arrays are initially sized to
|
| 69 | * %INIT_MEMBLOCK_MEMORY_REGIONS for "memory" and
|
| 70 | * %INIT_MEMBLOCK_RESERVED_REGIONS for "reserved". The region array
|
| 71 | * for "physmem" is initially sized to %INIT_PHYSMEM_REGIONS.
|
| 72 | * The memblock_allow_resize() enables automatic resizing of the region
|
| 73 | * arrays during addition of new regions. This feature should be used
|
| 74 | * with care so that memory allocated for the region array will not
|
| 75 | * overlap with areas that should be reserved, for example initrd.
|
| 76 | *
|
| 77 | * The early architecture setup should tell memblock what the physical
|
| 78 | * memory layout is by using memblock_add() or memblock_add_node()
|
| 79 | * functions. The first function does not assign the region to a NUMA
|
| 80 | * node and it is appropriate for UMA systems. Yet, it is possible to
|
| 81 | * use it on NUMA systems as well and assign the region to a NUMA node
|
| 82 | * later in the setup process using memblock_set_node(). The
|
| 83 | * memblock_add_node() performs such an assignment directly.
|
| 84 | *
|
| 85 | * Once memblock is setup the memory can be allocated using one of the
|
| 86 | * API variants:
|
| 87 | *
|
| 88 | * * memblock_phys_alloc*() - these functions return the **physical**
|
| 89 | * address of the allocated memory
|
| 90 | * * memblock_alloc*() - these functions return the **virtual** address
|
| 91 | * of the allocated memory.
|
| 92 | *
|
| 93 | * Note, that both API variants use implicit assumptions about allowed
|
| 94 | * memory ranges and the fallback methods. Consult the documentation
|
| 95 | * of memblock_alloc_internal() and memblock_alloc_range_nid()
|
| 96 | * functions for more elaborate description.
|
| 97 | *
|
| 98 | * As the system boot progresses, the architecture specific mem_init()
|
| 99 | * function frees all the memory to the buddy page allocator.
|
| 100 | *
|
| 101 | * Unless an architecture enables %CONFIG_ARCH_KEEP_MEMBLOCK, the
|
| 102 | * memblock data structures (except "physmem") will be discarded after the
|
| 103 | * system initialization completes.
|
| 104 |
|
| 105 |
|
| 106 | #ifndef CONFIG_NUMA
|
| 107 | struct pglist_data __refdata contig_page_data;
|
| 108 | EXPORT_SYMBOL(contig_page_data);
|
| 109 | #endif
|
| 110 |
|
| 111 | unsigned long max_low_pfn;
|
| 112 | unsigned long min_low_pfn;
|
| 113 | unsigned long max_pfn;
|
| 114 | unsigned long long max_possible_pfn;
|
| 115 |
|
| 116 | #ifdef CONFIG_MEMBLOCK_KHO_SCRATCH
|
| 117 |
|
| 118 | static bool kho_scratch_only;
|
| 119 | #else
|
| 120 | #define kho_scratch_only false
|
| 121 | #endif
|
| 122 |
|
| 123 | static struct memblock_region memblock_memory_init_regions[INIT_MEMBLOCK_MEMORY_REGIONS] __initdata_memblock;
|
| 124 | static struct memblock_region memblock_reserved_init_regions[INIT_MEMBLOCK_RESERVED_REGIONS] __initdata_memblock;
|
| 125 | #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
|
| 126 | static struct memblock_region memblock_physmem_init_regions[INIT_PHYSMEM_REGIONS];
|
| 127 | #endif
|
| 128 |
|
| 129 | struct memblock memblock __initdata_memblock = {
|
| 130 | .memory.regions = memblock_memory_init_regions,
|
| 131 | .memory.max = INIT_MEMBLOCK_MEMORY_REGIONS,
|
| 132 | .memory.name = "memory",
|
| 133 |
|
| 134 | .reserved.regions = memblock_reserved_init_regions,
|
| 135 | .reserved.max = INIT_MEMBLOCK_RESERVED_REGIONS,
|
| 136 | .reserved.name = "reserved",
|
| 137 |
|
| 138 | .bottom_up = false,
|
| 139 | .current_limit = MEMBLOCK_ALLOC_ANYWHERE,
|
| 140 | };
|
| 141 |
|
| 142 | #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
|
| 143 | struct memblock_type physmem = {
|
| 144 | .regions = memblock_physmem_init_regions,
|
| 145 | .max = INIT_PHYSMEM_REGIONS,
|
| 146 | .name = "physmem",
|
| 147 | };
|
| 148 | #endif
|
| 149 |
|
| 150 | |
| 151 | * keep a pointer to &memblock.memory in the text section to use it in
|
| 152 | * __next_mem_range() and its helpers.
|
| 153 | * For architectures that do not keep memblock data after init, this
|
| 154 | * pointer will be reset to NULL at memblock_discard()
|
| 155 |
|
| 156 | static __refdata struct memblock_type *memblock_memory = &memblock.memory;
|
| 157 |
|
| 158 | #define for_each_memblock_type(i, memblock_type, rgn) \
|
| 159 | for (i = 0, rgn = &memblock_type->regions[0]; \
|
| 160 | i < memblock_type->cnt; \
|
| 161 | i++, rgn = &memblock_type->regions[i])
|
| 162 |
|
| 163 | #define memblock_dbg(fmt, ...) \
|
| 164 | do { \
|
| 165 | if (memblock_debug) \
|
| 166 | pr_info(fmt, ##__VA_ARGS__); \
|
| 167 | } while (0)
|
| 168 |
|
| 169 | static int memblock_debug __initdata_memblock;
|
| 170 | static bool system_has_some_mirror __initdata_memblock;
|
| 171 | static int memblock_can_resize __initdata_memblock;
|
| 172 | static int memblock_memory_in_slab __initdata_memblock;
|
| 173 | static int memblock_reserved_in_slab __initdata_memblock;
|
| 174 |
|
| 175 | bool __init_memblock memblock_has_mirror(void)
|
| 176 | {
|
| 177 | return system_has_some_mirror;
|
| 178 | }
|
| 179 |
|
| 180 | static enum memblock_flags __init_memblock choose_memblock_flags(void)
|
| 181 | {
|
| 182 |
|
| 183 | if (kho_scratch_only)
|
| 184 | return MEMBLOCK_KHO_SCRATCH;
|
| 185 |
|
| 186 | return system_has_some_mirror ? MEMBLOCK_MIRROR : MEMBLOCK_NONE;
|
| 187 | }
|
| 188 |
|
| 189 |
|
| 190 | static inline phys_addr_t memblock_cap_size(phys_addr_t base, phys_addr_t *size)
|
| 191 | {
|
| 192 | return *size = min(*size, PHYS_ADDR_MAX - base);
|
| 193 | }
|
| 194 |
|
| 195 | |
| 196 | * Address comparison utilities
|
| 197 |
|
| 198 | unsigned long __init_memblock
|
| 199 | memblock_addrs_overlap(phys_addr_t base1, phys_addr_t size1, phys_addr_t base2,
|
| 200 | phys_addr_t size2)
|
| 201 | {
|
| 202 | return ((base1 < (base2 + size2)) && (base2 < (base1 + size1)));
|
| 203 | }
|
| 204 |
|
| 205 | bool __init_memblock memblock_overlaps_region(struct memblock_type *type,
|
| 206 | phys_addr_t base, phys_addr_t size)
|
| 207 | {
|
| 208 | unsigned long i;
|
| 209 |
|
| 210 | memblock_cap_size(base, &size);
|
| 211 |
|
| 212 | for (i = 0; i < type->cnt; i++)
|
| 213 | if (memblock_addrs_overlap(base, size, type->regions[i].base,
|
| 214 | type->regions[i].size))
|
| 215 | return true;
|
| 216 | return false;
|
| 217 | }
|
| 218 |
|
| 219 | |
| 220 | * __memblock_find_range_bottom_up - find free area utility in bottom-up
|
| 221 | * @start: start of candidate range
|
| 222 | * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
|
| 223 | * %MEMBLOCK_ALLOC_ACCESSIBLE
|
| 224 | * @size: size of free area to find
|
| 225 | * @align: alignment of free area to find
|
| 226 | * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
|
| 227 | * @flags: pick from blocks based on memory attributes
|
| 228 | *
|
| 229 | * Utility called from memblock_find_in_range_node(), find free area bottom-up.
|
| 230 | *
|
| 231 | * Return:
|
| 232 | * Found address on success, 0 on failure.
|
| 233 |
|
| 234 | static phys_addr_t __init_memblock
|
| 235 | __memblock_find_range_bottom_up(phys_addr_t start, phys_addr_t end,
|
| 236 | phys_addr_t size, phys_addr_t align, int nid,
|
| 237 | enum memblock_flags flags)
|
| 238 | {
|
| 239 | phys_addr_t this_start, this_end, cand;
|
| 240 | u64 i;
|
| 241 |
|
| 242 | for_each_free_mem_range(i, nid, flags, &this_start, &this_end, NULL) {
|
| 243 | this_start = clamp(this_start, start, end);
|
| 244 | this_end = clamp(this_end, start, end);
|
| 245 |
|
| 246 | cand = round_up(this_start, align);
|
| 247 | if (cand < this_end && this_end - cand >= size)
|
| 248 | return cand;
|
| 249 | }
|
| 250 |
|
| 251 | return 0;
|
| 252 | }
|
| 253 |
|
| 254 | |
| 255 | * __memblock_find_range_top_down - find free area utility, in top-down
|
| 256 | * @start: start of candidate range
|
| 257 | * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
|
| 258 | * %MEMBLOCK_ALLOC_ACCESSIBLE
|
| 259 | * @size: size of free area to find
|
| 260 | * @align: alignment of free area to find
|
| 261 | * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
|
| 262 | * @flags: pick from blocks based on memory attributes
|
| 263 | *
|
| 264 | * Utility called from memblock_find_in_range_node(), find free area top-down.
|
| 265 | *
|
| 266 | * Return:
|
| 267 | * Found address on success, 0 on failure.
|
| 268 |
|
| 269 | static phys_addr_t __init_memblock
|
| 270 | __memblock_find_range_top_down(phys_addr_t start, phys_addr_t end,
|
| 271 | phys_addr_t size, phys_addr_t align, int nid,
|
| 272 | enum memblock_flags flags)
|
| 273 | {
|
| 274 | phys_addr_t this_start, this_end, cand;
|
| 275 | u64 i;
|
| 276 |
|
| 277 | for_each_free_mem_range_reverse(i, nid, flags, &this_start, &this_end,
|
| 278 | NULL) {
|
| 279 | this_start = clamp(this_start, start, end);
|
| 280 | this_end = clamp(this_end, start, end);
|
| 281 |
|
| 282 | if (this_end < size)
|
| 283 | continue;
|
| 284 |
|
| 285 | cand = round_down(this_end - size, align);
|
| 286 | if (cand >= this_start)
|
| 287 | return cand;
|
| 288 | }
|
| 289 |
|
| 290 | return 0;
|
| 291 | }
|
| 292 |
|
| 293 | |
| 294 | * memblock_find_in_range_node - find free area in given range and node
|
| 295 | * @size: size of free area to find
|
| 296 | * @align: alignment of free area to find
|
| 297 | * @start: start of candidate range
|
| 298 | * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
|
| 299 | * %MEMBLOCK_ALLOC_ACCESSIBLE
|
| 300 | * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
|
| 301 | * @flags: pick from blocks based on memory attributes
|
| 302 | *
|
| 303 | * Find @size free area aligned to @align in the specified range and node.
|
| 304 | *
|
| 305 | * Return:
|
| 306 | * Found address on success, 0 on failure.
|
| 307 |
|
| 308 | static phys_addr_t __init_memblock memblock_find_in_range_node(phys_addr_t size,
|
| 309 | phys_addr_t align, phys_addr_t start,
|
| 310 | phys_addr_t end, int nid,
|
| 311 | enum memblock_flags flags)
|
| 312 | {
|
| 313 |
|
| 314 | if (end == MEMBLOCK_ALLOC_ACCESSIBLE ||
|
| 315 | end == MEMBLOCK_ALLOC_NOLEAKTRACE)
|
| 316 | end = memblock.current_limit;
|
| 317 |
|
| 318 |
|
| 319 | start = max_t(phys_addr_t, start, PAGE_SIZE);
|
| 320 | end = max(start, end);
|
| 321 |
|
| 322 | if (memblock_bottom_up())
|
| 323 | return __memblock_find_range_bottom_up(start, end, size, align,
|
| 324 | nid, flags);
|
| 325 | else
|
| 326 | return __memblock_find_range_top_down(start, end, size, align,
|
| 327 | nid, flags);
|
| 328 | }
|
| 329 |
|
| 330 | |
| 331 | * memblock_find_in_range - find free area in given range
|
| 332 | * @start: start of candidate range
|
| 333 | * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
|
| 334 | * %MEMBLOCK_ALLOC_ACCESSIBLE
|
| 335 | * @size: size of free area to find
|
| 336 | * @align: alignment of free area to find
|
| 337 | *
|
| 338 | * Find @size free area aligned to @align in the specified range.
|
| 339 | *
|
| 340 | * Return:
|
| 341 | * Found address on success, 0 on failure.
|
| 342 |
|
| 343 | static phys_addr_t __init_memblock memblock_find_in_range(phys_addr_t start,
|
| 344 | phys_addr_t end, phys_addr_t size,
|
| 345 | phys_addr_t align)
|
| 346 | {
|
| 347 | phys_addr_t ret;
|
| 348 | enum memblock_flags flags = choose_memblock_flags();
|
| 349 |
|
| 350 | again:
|
| 351 | ret = memblock_find_in_range_node(size, align, start, end,
|
| 352 | NUMA_NO_NODE, flags);
|
| 353 |
|
| 354 | if (!ret && (flags & MEMBLOCK_MIRROR)) {
|
| 355 | pr_warn_ratelimited("Could not allocate %pap bytes of mirrored memory\n",
|
| 356 | &size);
|
| 357 | flags &= ~MEMBLOCK_MIRROR;
|
| 358 | goto again;
|
| 359 | }
|
| 360 |
|
| 361 | return ret;
|
| 362 | }
|
| 363 |
|
| 364 | static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r)
|
| 365 | {
|
| 366 | type->total_size -= type->regions[r].size;
|
| 367 | memmove(&type->regions[r], &type->regions[r + 1],
|
| 368 | (type->cnt - (r + 1)) * sizeof(type->regions[r]));
|
| 369 | type->cnt--;
|
| 370 |
|
| 371 |
|
| 372 | if (type->cnt == 0) {
|
| 373 | WARN_ON(type->total_size != 0);
|
| 374 | type->regions[0].base = 0;
|
| 375 | type->regions[0].size = 0;
|
| 376 | type->regions[0].flags = 0;
|
| 377 | memblock_set_region_node(&type->regions[0], MAX_NUMNODES);
|
| 378 | }
|
| 379 | }
|
| 380 |
|
| 381 | #ifndef CONFIG_ARCH_KEEP_MEMBLOCK
|
| 382 | |
| 383 | * memblock_discard - discard memory and reserved arrays if they were allocated
|
| 384 |
|
| 385 | void __init memblock_discard(void)
|
| 386 | {
|
| 387 | phys_addr_t size;
|
| 388 | void *addr;
|
| 389 |
|
| 390 | if (memblock.reserved.regions != memblock_reserved_init_regions) {
|
| 391 | addr = memblock.reserved.regions;
|
| 392 | size = PAGE_ALIGN(sizeof(struct memblock_region) *
|
| 393 | memblock.reserved.max);
|
| 394 | if (memblock_reserved_in_slab)
|
| 395 | kfree(addr);
|
| 396 | else
|
| 397 | memblock_free(addr, size);
|
| 398 | }
|
| 399 |
|
| 400 | if (memblock.memory.regions != memblock_memory_init_regions) {
|
| 401 | addr = memblock.memory.regions;
|
| 402 | size = PAGE_ALIGN(sizeof(struct memblock_region) *
|
| 403 | memblock.memory.max);
|
| 404 | if (memblock_memory_in_slab)
|
| 405 | kfree(addr);
|
| 406 | else
|
| 407 | memblock_free(addr, size);
|
| 408 | }
|
| 409 |
|
| 410 | memblock_memory = NULL;
|
| 411 | }
|
| 412 | #endif
|
| 413 |
|
| 414 | |
| 415 | * memblock_double_array - double the size of the memblock regions array
|
| 416 | * @type: memblock type of the regions array being doubled
|
| 417 | * @new_area_start: starting address of memory range to avoid overlap with
|
| 418 | * @new_area_size: size of memory range to avoid overlap with
|
| 419 | *
|
| 420 | * Double the size of the @type regions array. If memblock is being used to
|
| 421 | * allocate memory for a new reserved regions array and there is a previously
|
| 422 | * allocated memory range [@new_area_start, @new_area_start + @new_area_size]
|
| 423 | * waiting to be reserved, ensure the memory used by the new array does
|
| 424 | * not overlap.
|
| 425 | *
|
| 426 | * Return:
|
| 427 | * 0 on success, -1 on failure.
|
| 428 |
|
| 429 | static int __init_memblock memblock_double_array(struct memblock_type *type,
|
| 430 | phys_addr_t new_area_start,
|
| 431 | phys_addr_t new_area_size)
|
| 432 | {
|
| 433 | struct memblock_region *new_array, *old_array;
|
| 434 | phys_addr_t old_alloc_size, new_alloc_size;
|
| 435 | phys_addr_t old_size, new_size, addr, new_end;
|
| 436 | int use_slab = slab_is_available();
|
| 437 | int *in_slab;
|
| 438 |
|
| 439 | |
| 440 | * of memory that aren't suitable for allocation
|
| 441 |
|
| 442 | if (!memblock_can_resize)
|
| 443 | panic("memblock: cannot resize %s array\n", type->name);
|
| 444 |
|
| 445 |
|
| 446 | old_size = type->max * sizeof(struct memblock_region);
|
| 447 | new_size = old_size << 1;
|
| 448 | |
| 449 | * We need to allocated new one align to PAGE_SIZE,
|
| 450 | * so we can free them completely later.
|
| 451 |
|
| 452 | old_alloc_size = PAGE_ALIGN(old_size);
|
| 453 | new_alloc_size = PAGE_ALIGN(new_size);
|
| 454 |
|
| 455 |
|
| 456 | if (type == &memblock.memory)
|
| 457 | in_slab = &memblock_memory_in_slab;
|
| 458 | else
|
| 459 | in_slab = &memblock_reserved_in_slab;
|
| 460 |
|
| 461 |
|
| 462 | if (use_slab) {
|
| 463 | new_array = kmalloc(new_size, GFP_KERNEL);
|
| 464 | addr = new_array ? __pa(new_array) : 0;
|
| 465 | } else {
|
| 466 |
|
| 467 | if (type != &memblock.reserved)
|
| 468 | new_area_start = new_area_size = 0;
|
| 469 |
|
| 470 | addr = memblock_find_in_range(new_area_start + new_area_size,
|
| 471 | memblock.current_limit,
|
| 472 | new_alloc_size, PAGE_SIZE);
|
| 473 | if (!addr && new_area_size)
|
| 474 | addr = memblock_find_in_range(0,
|
| 475 | min(new_area_start, memblock.current_limit),
|
| 476 | new_alloc_size, PAGE_SIZE);
|
| 477 |
|
| 478 | if (addr) {
|
| 479 |
|
| 480 | accept_memory(addr, new_alloc_size);
|
| 481 |
|
| 482 | new_array = __va(addr);
|
| 483 | } else {
|
| 484 | new_array = NULL;
|
| 485 | }
|
| 486 | }
|
| 487 | if (!addr) {
|
| 488 | pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n",
|
| 489 | type->name, type->max, type->max * 2);
|
| 490 | return -1;
|
| 491 | }
|
| 492 |
|
| 493 | new_end = addr + new_size - 1;
|
| 494 | memblock_dbg("memblock: %s is doubled to %ld at [%pa-%pa]",
|
| 495 | type->name, type->max * 2, &addr, &new_end);
|
| 496 |
|
| 497 | |
| 498 | * Found space, we now need to move the array over before we add the
|
| 499 | * reserved region since it may be our reserved array itself that is
|
| 500 | * full.
|
| 501 |
|
| 502 | memcpy(new_array, type->regions, old_size);
|
| 503 | memset(new_array + type->max, 0, old_size);
|
| 504 | old_array = type->regions;
|
| 505 | type->regions = new_array;
|
| 506 | type->max <<= 1;
|
| 507 |
|
| 508 |
|
| 509 | if (*in_slab)
|
| 510 | kfree(old_array);
|
| 511 | else if (old_array != memblock_memory_init_regions &&
|
| 512 | old_array != memblock_reserved_init_regions)
|
| 513 | memblock_free(old_array, old_alloc_size);
|
| 514 |
|
| 515 | |
| 516 | * Reserve the new array if that comes from the memblock. Otherwise, we
|
| 517 | * needn't do it
|
| 518 |
|
| 519 | if (!use_slab)
|
| 520 | BUG_ON(memblock_reserve_kern(addr, new_alloc_size));
|
| 521 |
|
| 522 |
|
| 523 | *in_slab = use_slab;
|
| 524 |
|
| 525 | return 0;
|
| 526 | }
|
| 527 |
|
| 528 | |
| 529 | * memblock_merge_regions - merge neighboring compatible regions
|
| 530 | * @type: memblock type to scan
|
| 531 | * @start_rgn: start scanning from (@start_rgn - 1)
|
| 532 | * @end_rgn: end scanning at (@end_rgn - 1)
|
| 533 | * Scan @type and merge neighboring compatible regions in [@start_rgn - 1, @end_rgn)
|
| 534 |
|
| 535 | static void __init_memblock memblock_merge_regions(struct memblock_type *type,
|
| 536 | unsigned long start_rgn,
|
| 537 | unsigned long end_rgn)
|
| 538 | {
|
| 539 | int i = 0;
|
| 540 | if (start_rgn)
|
| 541 | i = start_rgn - 1;
|
| 542 | end_rgn = min(end_rgn, type->cnt - 1);
|
| 543 | while (i < end_rgn) {
|
| 544 | struct memblock_region *this = &type->regions[i];
|
| 545 | struct memblock_region *next = &type->regions[i + 1];
|
| 546 |
|
| 547 | if (this->base + this->size != next->base ||
|
| 548 | memblock_get_region_node(this) !=
|
| 549 | memblock_get_region_node(next) ||
|
| 550 | this->flags != next->flags) {
|
| 551 | BUG_ON(this->base + this->size > next->base);
|
| 552 | i++;
|
| 553 | continue;
|
| 554 | }
|
| 555 |
|
| 556 | this->size += next->size;
|
| 557 |
|
| 558 | memmove(next, next + 1, (type->cnt - (i + 2)) * sizeof(*next));
|
| 559 | type->cnt--;
|
| 560 | end_rgn--;
|
| 561 | }
|
| 562 | }
|
| 563 |
|
| 564 | |
| 565 | * memblock_insert_region - insert new memblock region
|
| 566 | * @type: memblock type to insert into
|
| 567 | * @idx: index for the insertion point
|
| 568 | * @base: base address of the new region
|
| 569 | * @size: size of the new region
|
| 570 | * @nid: node id of the new region
|
| 571 | * @flags: flags of the new region
|
| 572 | *
|
| 573 | * Insert new memblock region [@base, @base + @size) into @type at @idx.
|
| 574 | * @type must already have extra room to accommodate the new region.
|
| 575 |
|
| 576 | static void __init_memblock memblock_insert_region(struct memblock_type *type,
|
| 577 | int idx, phys_addr_t base,
|
| 578 | phys_addr_t size,
|
| 579 | int nid,
|
| 580 | enum memblock_flags flags)
|
| 581 | {
|
| 582 | struct memblock_region *rgn = &type->regions[idx];
|
| 583 |
|
| 584 | BUG_ON(type->cnt >= type->max);
|
| 585 | memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn));
|
| 586 | rgn->base = base;
|
| 587 | rgn->size = size;
|
| 588 | rgn->flags = flags;
|
| 589 | memblock_set_region_node(rgn, nid);
|
| 590 | type->cnt++;
|
| 591 | type->total_size += size;
|
| 592 | }
|
| 593 |
|
| 594 | |
| 595 | * memblock_add_range - add new memblock region
|
| 596 | * @type: memblock type to add new region into
|
| 597 | * @base: base address of the new region
|
| 598 | * @size: size of the new region
|
| 599 | * @nid: nid of the new region
|
| 600 | * @flags: flags of the new region
|
| 601 | *
|
| 602 | * Add new memblock region [@base, @base + @size) into @type. The new region
|
| 603 | * is allowed to overlap with existing ones - overlaps don't affect already
|
| 604 | * existing regions. @type is guaranteed to be minimal (all neighbouring
|
| 605 | * compatible regions are merged) after the addition.
|
| 606 | *
|
| 607 | * Return:
|
| 608 | * 0 on success, -errno on failure.
|
| 609 |
|
| 610 | static int __init_memblock memblock_add_range(struct memblock_type *type,
|
| 611 | phys_addr_t base, phys_addr_t size,
|
| 612 | int nid, enum memblock_flags flags)
|
| 613 | {
|
| 614 | bool insert = false;
|
| 615 | phys_addr_t obase = base;
|
| 616 | phys_addr_t end = base + memblock_cap_size(base, &size);
|
| 617 | int idx, nr_new, start_rgn = -1, end_rgn;
|
| 618 | struct memblock_region *rgn;
|
| 619 |
|
| 620 | if (!size)
|
| 621 | return 0;
|
| 622 |
|
| 623 |
|
| 624 | if (type->regions[0].size == 0) {
|
| 625 | WARN_ON(type->cnt != 0 || type->total_size);
|
| 626 | type->regions[0].base = base;
|
| 627 | type->regions[0].size = size;
|
| 628 | type->regions[0].flags = flags;
|
| 629 | memblock_set_region_node(&type->regions[0], nid);
|
| 630 | type->total_size = size;
|
| 631 | type->cnt = 1;
|
| 632 | return 0;
|
| 633 | }
|
| 634 |
|
| 635 | |
| 636 | * The worst case is when new range overlaps all existing regions,
|
| 637 | * then we'll need type->cnt + 1 empty regions in @type. So if
|
| 638 | * type->cnt * 2 + 1 is less than or equal to type->max, we know
|
| 639 | * that there is enough empty regions in @type, and we can insert
|
| 640 | * regions directly.
|
| 641 |
|
| 642 | if (type->cnt * 2 + 1 <= type->max)
|
| 643 | insert = true;
|
| 644 |
|
| 645 | repeat:
|
| 646 | |
| 647 | * The following is executed twice. Once with %false @insert and
|
| 648 | * then with %true. The first counts the number of regions needed
|
| 649 | * to accommodate the new area. The second actually inserts them.
|
| 650 |
|
| 651 | base = obase;
|
| 652 | nr_new = 0;
|
| 653 |
|
| 654 | for_each_memblock_type(idx, type, rgn) {
|
| 655 | phys_addr_t rbase = rgn->base;
|
| 656 | phys_addr_t rend = rbase + rgn->size;
|
| 657 |
|
| 658 | if (rbase >= end)
|
| 659 | break;
|
| 660 | if (rend <= base)
|
| 661 | continue;
|
| 662 | |
| 663 | * @rgn overlaps. If it separates the lower part of new
|
| 664 | * area, insert that portion.
|
| 665 |
|
| 666 | if (rbase > base) {
|
| 667 | #ifdef CONFIG_NUMA
|
| 668 | WARN_ON(nid != memblock_get_region_node(rgn));
|
| 669 | #endif
|
| 670 | WARN_ON(flags != MEMBLOCK_NONE && flags != rgn->flags);
|
| 671 | nr_new++;
|
| 672 | if (insert) {
|
| 673 | if (start_rgn == -1)
|
| 674 | start_rgn = idx;
|
| 675 | end_rgn = idx + 1;
|
| 676 | memblock_insert_region(type, idx++, base,
|
| 677 | rbase - base, nid,
|
| 678 | flags);
|
| 679 | }
|
| 680 | }
|
| 681 |
|
| 682 | base = min(rend, end);
|
| 683 | }
|
| 684 |
|
| 685 |
|
| 686 | if (base < end) {
|
| 687 | nr_new++;
|
| 688 | if (insert) {
|
| 689 | if (start_rgn == -1)
|
| 690 | start_rgn = idx;
|
| 691 | end_rgn = idx + 1;
|
| 692 | memblock_insert_region(type, idx, base, end - base,
|
| 693 | nid, flags);
|
| 694 | }
|
| 695 | }
|
| 696 |
|
| 697 | if (!nr_new)
|
| 698 | return 0;
|
| 699 |
|
| 700 | |
| 701 | * If this was the first round, resize array and repeat for actual
|
| 702 | * insertions; otherwise, merge and return.
|
| 703 |
|
| 704 | if (!insert) {
|
| 705 | while (type->cnt + nr_new > type->max)
|
| 706 | if (memblock_double_array(type, obase, size) < 0)
|
| 707 | return -ENOMEM;
|
| 708 | insert = true;
|
| 709 | goto repeat;
|
| 710 | } else {
|
| 711 | memblock_merge_regions(type, start_rgn, end_rgn);
|
| 712 | return 0;
|
| 713 | }
|
| 714 | }
|
| 715 |
|
| 716 | |
| 717 | * memblock_add_node - add new memblock region within a NUMA node
|
| 718 | * @base: base address of the new region
|
| 719 | * @size: size of the new region
|
| 720 | * @nid: nid of the new region
|
| 721 | * @flags: flags of the new region
|
| 722 | *
|
| 723 | * Add new memblock region [@base, @base + @size) to the "memory"
|
| 724 | * type. See memblock_add_range() description for mode details
|
| 725 | *
|
| 726 | * Return:
|
| 727 | * 0 on success, -errno on failure.
|
| 728 |
|
| 729 | int __init_memblock memblock_add_node(phys_addr_t base, phys_addr_t size,
|
| 730 | int nid, enum memblock_flags flags)
|
| 731 | {
|
| 732 | phys_addr_t end = base + size - 1;
|
| 733 |
|
| 734 | memblock_dbg("%s: [%pa-%pa] nid=%d flags=%x %pS\n", __func__,
|
| 735 | &base, &end, nid, flags, (void *)_RET_IP_);
|
| 736 |
|
| 737 | return memblock_add_range(&memblock.memory, base, size, nid, flags);
|
| 738 | }
|
| 739 |
|
| 740 | |
| 741 | * memblock_add - add new memblock region
|
| 742 | * @base: base address of the new region
|
| 743 | * @size: size of the new region
|
| 744 | *
|
| 745 | * Add new memblock region [@base, @base + @size) to the "memory"
|
| 746 | * type. See memblock_add_range() description for mode details
|
| 747 | *
|
| 748 | * Return:
|
| 749 | * 0 on success, -errno on failure.
|
| 750 |
|
| 751 | int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size)
|
| 752 | {
|
| 753 | phys_addr_t end = base + size - 1;
|
| 754 |
|
| 755 | memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
|
| 756 | &base, &end, (void *)_RET_IP_);
|
| 757 |
|
| 758 | return memblock_add_range(&memblock.memory, base, size, MAX_NUMNODES, 0);
|
| 759 | }
|
| 760 |
|
| 761 | |
| 762 | * memblock_validate_numa_coverage - check if amount of memory with
|
| 763 | * no node ID assigned is less than a threshold
|
| 764 | * @threshold_bytes: maximal memory size that can have unassigned node
|
| 765 | * ID (in bytes).
|
| 766 | *
|
| 767 | * A buggy firmware may report memory that does not belong to any node.
|
| 768 | * Check if amount of such memory is below @threshold_bytes.
|
| 769 | *
|
| 770 | * Return: true on success, false on failure.
|
| 771 |
|
| 772 | bool __init_memblock memblock_validate_numa_coverage(unsigned long threshold_bytes)
|
| 773 | {
|
| 774 | unsigned long nr_pages = 0;
|
| 775 | unsigned long start_pfn, end_pfn, mem_size_mb;
|
| 776 | int nid, i;
|
| 777 |
|
| 778 |
|
| 779 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
|
| 780 | if (!numa_valid_node(nid))
|
| 781 | nr_pages += end_pfn - start_pfn;
|
| 782 | }
|
| 783 |
|
| 784 | if ((nr_pages << PAGE_SHIFT) > threshold_bytes) {
|
| 785 | mem_size_mb = memblock_phys_mem_size() / SZ_1M;
|
| 786 | pr_err("NUMA: no nodes coverage for %luMB of %luMB RAM\n",
|
| 787 | (nr_pages << PAGE_SHIFT) / SZ_1M, mem_size_mb);
|
| 788 | return false;
|
| 789 | }
|
| 790 |
|
| 791 | return true;
|
| 792 | }
|
| 793 |
|
| 794 |
|
| 795 | |
| 796 | * memblock_isolate_range - isolate given range into disjoint memblocks
|
| 797 | * @type: memblock type to isolate range for
|
| 798 | * @base: base of range to isolate
|
| 799 | * @size: size of range to isolate
|
| 800 | * @start_rgn: out parameter for the start of isolated region
|
| 801 | * @end_rgn: out parameter for the end of isolated region
|
| 802 | *
|
| 803 | * Walk @type and ensure that regions don't cross the boundaries defined by
|
| 804 | * [@base, @base + @size). Crossing regions are split at the boundaries,
|
| 805 | * which may create at most two more regions. The index of the first
|
| 806 | * region inside the range is returned in *@start_rgn and the index of the
|
| 807 | * first region after the range is returned in *@end_rgn.
|
| 808 | *
|
| 809 | * Return:
|
| 810 | * 0 on success, -errno on failure.
|
| 811 |
|
| 812 | static int __init_memblock memblock_isolate_range(struct memblock_type *type,
|
| 813 | phys_addr_t base, phys_addr_t size,
|
| 814 | int *start_rgn, int *end_rgn)
|
| 815 | {
|
| 816 | phys_addr_t end = base + memblock_cap_size(base, &size);
|
| 817 | int idx;
|
| 818 | struct memblock_region *rgn;
|
| 819 |
|
| 820 | *start_rgn = *end_rgn = 0;
|
| 821 |
|
| 822 | if (!size)
|
| 823 | return 0;
|
| 824 |
|
| 825 |
|
| 826 | while (type->cnt + 2 > type->max)
|
| 827 | if (memblock_double_array(type, base, size) < 0)
|
| 828 | return -ENOMEM;
|
| 829 |
|
| 830 | for_each_memblock_type(idx, type, rgn) {
|
| 831 | phys_addr_t rbase = rgn->base;
|
| 832 | phys_addr_t rend = rbase + rgn->size;
|
| 833 |
|
| 834 | if (rbase >= end)
|
| 835 | break;
|
| 836 | if (rend <= base)
|
| 837 | continue;
|
| 838 |
|
| 839 | if (rbase < base) {
|
| 840 | |
| 841 | * @rgn intersects from below. Split and continue
|
| 842 | * to process the next region - the new top half.
|
| 843 |
|
| 844 | rgn->base = base;
|
| 845 | rgn->size -= base - rbase;
|
| 846 | type->total_size -= base - rbase;
|
| 847 | memblock_insert_region(type, idx, rbase, base - rbase,
|
| 848 | memblock_get_region_node(rgn),
|
| 849 | rgn->flags);
|
| 850 | } else if (rend > end) {
|
| 851 | |
| 852 | * @rgn intersects from above. Split and redo the
|
| 853 | * current region - the new bottom half.
|
| 854 |
|
| 855 | rgn->base = end;
|
| 856 | rgn->size -= end - rbase;
|
| 857 | type->total_size -= end - rbase;
|
| 858 | memblock_insert_region(type, idx--, rbase, end - rbase,
|
| 859 | memblock_get_region_node(rgn),
|
| 860 | rgn->flags);
|
| 861 | } else {
|
| 862 |
|
| 863 | if (!*end_rgn)
|
| 864 | *start_rgn = idx;
|
| 865 | *end_rgn = idx + 1;
|
| 866 | }
|
| 867 | }
|
| 868 |
|
| 869 | return 0;
|
| 870 | }
|
| 871 |
|
| 872 | static int __init_memblock memblock_remove_range(struct memblock_type *type,
|
| 873 | phys_addr_t base, phys_addr_t size)
|
| 874 | {
|
| 875 | int start_rgn, end_rgn;
|
| 876 | int i, ret;
|
| 877 |
|
| 878 | ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
|
| 879 | if (ret)
|
| 880 | return ret;
|
| 881 |
|
| 882 | for (i = end_rgn - 1; i >= start_rgn; i--)
|
| 883 | memblock_remove_region(type, i);
|
| 884 | return 0;
|
| 885 | }
|
| 886 |
|
| 887 | int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size)
|
| 888 | {
|
| 889 | phys_addr_t end = base + size - 1;
|
| 890 |
|
| 891 | memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
|
| 892 | &base, &end, (void *)_RET_IP_);
|
| 893 |
|
| 894 | return memblock_remove_range(&memblock.memory, base, size);
|
| 895 | }
|
| 896 |
|
| 897 | static unsigned long __free_reserved_area(phys_addr_t start, phys_addr_t end,
|
| 898 | int poison)
|
| 899 | {
|
| 900 | unsigned long pages = 0, pfn;
|
| 901 |
|
| 902 | if (deferred_pages_enabled()) {
|
| 903 | WARN(1, "Cannot free reserved memory because of deferred initialization of the memory map");
|
| 904 | return 0;
|
| 905 | }
|
| 906 |
|
| 907 | for_each_valid_pfn(pfn, PFN_UP(start), PFN_DOWN(end)) {
|
| 908 | struct page *page = pfn_to_page(pfn);
|
| 909 | void *direct_map_addr;
|
| 910 |
|
| 911 | |
| 912 | * 'direct_map_addr' might be different from the kernel virtual
|
| 913 | * address because some architectures use aliases.
|
| 914 | * Going via physical address, pfn_to_page() and page_address()
|
| 915 | * ensures that we get a _writeable_ alias for the memset().
|
| 916 |
|
| 917 | direct_map_addr = page_address(page);
|
| 918 | |
| 919 | * Perform a kasan-unchecked memset() since this memory
|
| 920 | * has not been initialized.
|
| 921 |
|
| 922 | direct_map_addr = kasan_reset_tag(direct_map_addr);
|
| 923 | if ((unsigned int)poison <= 0xFF)
|
| 924 | memset(direct_map_addr, poison, PAGE_SIZE);
|
| 925 |
|
| 926 | free_reserved_page(page);
|
| 927 | pages++;
|
| 928 | }
|
| 929 | return pages;
|
| 930 | }
|
| 931 |
|
| 932 | unsigned long free_reserved_area(void *start, void *end, int poison, const char *s)
|
| 933 | {
|
| 934 | phys_addr_t start_pa, end_pa;
|
| 935 | unsigned long pages;
|
| 936 |
|
| 937 | |
| 938 | * end is the first address past the region and it may be beyond what
|
| 939 | * __pa() or __pa_symbol() can handle.
|
| 940 | * Use the address included in the range for the conversion and add back
|
| 941 | * 1 afterwards.
|
| 942 |
|
| 943 | if (__is_kernel((unsigned long)start)) {
|
| 944 | start_pa = __pa_symbol(start);
|
| 945 | end_pa = __pa_symbol(end - 1) + 1;
|
| 946 | } else {
|
| 947 | start_pa = __pa(start);
|
| 948 | end_pa = __pa(end - 1) + 1;
|
| 949 | }
|
| 950 |
|
| 951 | if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK)) {
|
| 952 | if (start_pa < end_pa)
|
| 953 | memblock_remove_range(&memblock.reserved,
|
| 954 | start_pa, end_pa - start_pa);
|
| 955 | }
|
| 956 |
|
| 957 | pages = __free_reserved_area(start_pa, end_pa, poison);
|
| 958 | if (pages && s)
|
| 959 | pr_info("Freeing %s memory: %ldK\n", s, K(pages));
|
| 960 |
|
| 961 | return pages;
|
| 962 | }
|
| 963 |
|
| 964 | |
| 965 | * memblock_free - free boot memory allocation
|
| 966 | * @ptr: starting address of the boot memory allocation
|
| 967 | * @size: size of the boot memory block in bytes
|
| 968 | *
|
| 969 | * Free boot memory block previously allocated by memblock_alloc_xx() API.
|
| 970 | * If called after the buddy allocator is available, the memory is released to
|
| 971 | * the buddy allocator.
|
| 972 |
|
| 973 | void __init_memblock memblock_free(void *ptr, size_t size)
|
| 974 | {
|
| 975 | if (ptr)
|
| 976 | memblock_phys_free(__pa(ptr), size);
|
| 977 | }
|
| 978 |
|
| 979 | |
| 980 | * memblock_phys_free - free boot memory block
|
| 981 | * @base: phys starting address of the boot memory block
|
| 982 | * @size: size of the boot memory block in bytes
|
| 983 | *
|
| 984 | * Free boot memory block previously allocated by memblock_phys_alloc_xx() API.
|
| 985 | * If called after the buddy allocator is available, the memory is released to
|
| 986 | * the buddy allocator.
|
| 987 |
|
| 988 | int __init_memblock memblock_phys_free(phys_addr_t base, phys_addr_t size)
|
| 989 | {
|
| 990 | phys_addr_t end = base + size - 1;
|
| 991 | int ret = 0;
|
| 992 |
|
| 993 | memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
|
| 994 | &base, &end, (void *)_RET_IP_);
|
| 995 |
|
| 996 | kmemleak_free_part_phys(base, size);
|
| 997 |
|
| 998 | if (!slab_is_available() || IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK))
|
| 999 | ret = memblock_remove_range(&memblock.reserved, base, size);
|
| 1000 |
|
| 1001 | if (slab_is_available())
|
| 1002 | __free_reserved_area(base, base + size, -1);
|
| 1003 |
|
| 1004 | return ret;
|
| 1005 | }
|
| 1006 |
|
| 1007 | int __init_memblock __memblock_reserve(phys_addr_t base, phys_addr_t size,
|
| 1008 | int nid, enum memblock_flags flags)
|
| 1009 | {
|
| 1010 | phys_addr_t end = base + size - 1;
|
| 1011 |
|
| 1012 | memblock_dbg("%s: [%pa-%pa] nid=%d flags=%x %pS\n", __func__,
|
| 1013 | &base, &end, nid, flags, (void *)_RET_IP_);
|
| 1014 |
|
| 1015 | return memblock_add_range(&memblock.reserved, base, size, nid, flags);
|
| 1016 | }
|
| 1017 |
|
| 1018 | #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
|
| 1019 | int __init_memblock memblock_physmem_add(phys_addr_t base, phys_addr_t size)
|
| 1020 | {
|
| 1021 | phys_addr_t end = base + size - 1;
|
| 1022 |
|
| 1023 | memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
|
| 1024 | &base, &end, (void *)_RET_IP_);
|
| 1025 |
|
| 1026 | return memblock_add_range(&physmem, base, size, MAX_NUMNODES, 0);
|
| 1027 | }
|
| 1028 | #endif
|
| 1029 |
|
| 1030 | |
| 1031 | * memblock_setclr_flag - set or clear flag for a memory region
|
| 1032 | * @type: memblock type to set/clear flag for
|
| 1033 | * @base: base address of the region
|
| 1034 | * @size: size of the region
|
| 1035 | * @set: set or clear the flag
|
| 1036 | * @flag: the flag to update
|
| 1037 | *
|
| 1038 | * This function isolates region [@base, @base + @size), and sets/clears flag
|
| 1039 | *
|
| 1040 | * Return: 0 on success, -errno on failure.
|
| 1041 |
|
| 1042 | static int __init_memblock memblock_setclr_flag(struct memblock_type *type,
|
| 1043 | phys_addr_t base, phys_addr_t size, int set, int flag)
|
| 1044 | {
|
| 1045 | int i, ret, start_rgn, end_rgn;
|
| 1046 |
|
| 1047 | ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
|
| 1048 | if (ret)
|
| 1049 | return ret;
|
| 1050 |
|
| 1051 | for (i = start_rgn; i < end_rgn; i++) {
|
| 1052 | struct memblock_region *r = &type->regions[i];
|
| 1053 |
|
| 1054 | if (set)
|
| 1055 | r->flags |= flag;
|
| 1056 | else
|
| 1057 | r->flags &= ~flag;
|
| 1058 | }
|
| 1059 |
|
| 1060 | memblock_merge_regions(type, start_rgn, end_rgn);
|
| 1061 | return 0;
|
| 1062 | }
|
| 1063 |
|
| 1064 | |
| 1065 | * memblock_mark_hotplug - Mark hotpluggable memory with flag MEMBLOCK_HOTPLUG.
|
| 1066 | * @base: the base phys addr of the region
|
| 1067 | * @size: the size of the region
|
| 1068 | *
|
| 1069 | * Return: 0 on success, -errno on failure.
|
| 1070 |
|
| 1071 | int __init_memblock memblock_mark_hotplug(phys_addr_t base, phys_addr_t size)
|
| 1072 | {
|
| 1073 | return memblock_setclr_flag(&memblock.memory, base, size, 1, MEMBLOCK_HOTPLUG);
|
| 1074 | }
|
| 1075 |
|
| 1076 | |
| 1077 | * memblock_clear_hotplug - Clear flag MEMBLOCK_HOTPLUG for a specified region.
|
| 1078 | * @base: the base phys addr of the region
|
| 1079 | * @size: the size of the region
|
| 1080 | *
|
| 1081 | * Return: 0 on success, -errno on failure.
|
| 1082 |
|
| 1083 | int __init_memblock memblock_clear_hotplug(phys_addr_t base, phys_addr_t size)
|
| 1084 | {
|
| 1085 | return memblock_setclr_flag(&memblock.memory, base, size, 0, MEMBLOCK_HOTPLUG);
|
| 1086 | }
|
| 1087 |
|
| 1088 | |
| 1089 | * memblock_mark_mirror - Mark mirrored memory with flag MEMBLOCK_MIRROR.
|
| 1090 | * @base: the base phys addr of the region
|
| 1091 | * @size: the size of the region
|
| 1092 | *
|
| 1093 | * Return: 0 on success, -errno on failure.
|
| 1094 |
|
| 1095 | int __init_memblock memblock_mark_mirror(phys_addr_t base, phys_addr_t size)
|
| 1096 | {
|
| 1097 | if (!mirrored_kernelcore)
|
| 1098 | return 0;
|
| 1099 |
|
| 1100 | system_has_some_mirror = true;
|
| 1101 |
|
| 1102 | return memblock_setclr_flag(&memblock.memory, base, size, 1, MEMBLOCK_MIRROR);
|
| 1103 | }
|
| 1104 |
|
| 1105 | |
| 1106 | * memblock_mark_nomap - Mark a memory region with flag MEMBLOCK_NOMAP.
|
| 1107 | * @base: the base phys addr of the region
|
| 1108 | * @size: the size of the region
|
| 1109 | *
|
| 1110 | * The memory regions marked with %MEMBLOCK_NOMAP will not be added to the
|
| 1111 | * direct mapping of the physical memory. These regions will still be
|
| 1112 | * covered by the memory map. The struct page representing NOMAP memory
|
| 1113 | * frames in the memory map will be PageReserved()
|
| 1114 | *
|
| 1115 | * Note: if the memory being marked %MEMBLOCK_NOMAP was allocated from
|
| 1116 | * memblock, the caller must inform kmemleak to ignore that memory
|
| 1117 | *
|
| 1118 | * Return: 0 on success, -errno on failure.
|
| 1119 |
|
| 1120 | int __init_memblock memblock_mark_nomap(phys_addr_t base, phys_addr_t size)
|
| 1121 | {
|
| 1122 | return memblock_setclr_flag(&memblock.memory, base, size, 1, MEMBLOCK_NOMAP);
|
| 1123 | }
|
| 1124 |
|
| 1125 | |
| 1126 | * memblock_clear_nomap - Clear flag MEMBLOCK_NOMAP for a specified region.
|
| 1127 | * @base: the base phys addr of the region
|
| 1128 | * @size: the size of the region
|
| 1129 | *
|
| 1130 | * Return: 0 on success, -errno on failure.
|
| 1131 |
|
| 1132 | int __init_memblock memblock_clear_nomap(phys_addr_t base, phys_addr_t size)
|
| 1133 | {
|
| 1134 | return memblock_setclr_flag(&memblock.memory, base, size, 0, MEMBLOCK_NOMAP);
|
| 1135 | }
|
| 1136 |
|
| 1137 | |
| 1138 | * memblock_reserved_mark_noinit - Mark a reserved memory region with flag
|
| 1139 | * MEMBLOCK_RSRV_NOINIT
|
| 1140 | *
|
| 1141 | * @base: the base phys addr of the region
|
| 1142 | * @size: the size of the region
|
| 1143 | *
|
| 1144 | * The struct pages for the reserved regions marked %MEMBLOCK_RSRV_NOINIT will
|
| 1145 | * not be fully initialized to allow the caller optimize their initialization.
|
| 1146 | *
|
| 1147 | * When %CONFIG_DEFERRED_STRUCT_PAGE_INIT is enabled, setting this flag
|
| 1148 | * completely bypasses the initialization of struct pages for such region.
|
| 1149 | *
|
| 1150 | * When %CONFIG_DEFERRED_STRUCT_PAGE_INIT is disabled, struct pages in this
|
| 1151 | * region will be initialized with default values but won't be marked as
|
| 1152 | * reserved.
|
| 1153 | *
|
| 1154 | * Return: 0 on success, -errno on failure.
|
| 1155 |
|
| 1156 | int __init_memblock memblock_reserved_mark_noinit(phys_addr_t base, phys_addr_t size)
|
| 1157 | {
|
| 1158 | return memblock_setclr_flag(&memblock.reserved, base, size, 1,
|
| 1159 | MEMBLOCK_RSRV_NOINIT);
|
| 1160 | }
|
| 1161 |
|
| 1162 | |
| 1163 | * memblock_reserved_mark_kern - Mark a reserved memory region with flag
|
| 1164 | * MEMBLOCK_RSRV_KERN
|
| 1165 | *
|
| 1166 | * @base: the base phys addr of the region
|
| 1167 | * @size: the size of the region
|
| 1168 | *
|
| 1169 | * Return: 0 on success, -errno on failure.
|
| 1170 |
|
| 1171 | int __init_memblock memblock_reserved_mark_kern(phys_addr_t base, phys_addr_t size)
|
| 1172 | {
|
| 1173 | return memblock_setclr_flag(&memblock.reserved, base, size, 1,
|
| 1174 | MEMBLOCK_RSRV_KERN);
|
| 1175 | }
|
| 1176 |
|
| 1177 | |
| 1178 | * memblock_mark_kho_scratch - Mark a memory region as MEMBLOCK_KHO_SCRATCH.
|
| 1179 | * @base: the base phys addr of the region
|
| 1180 | * @size: the size of the region
|
| 1181 | *
|
| 1182 | * Only memory regions marked with %MEMBLOCK_KHO_SCRATCH will be considered
|
| 1183 | * for allocations during early boot with kexec handover.
|
| 1184 | *
|
| 1185 | * Return: 0 on success, -errno on failure.
|
| 1186 |
|
| 1187 | __init int memblock_mark_kho_scratch(phys_addr_t base, phys_addr_t size)
|
| 1188 | {
|
| 1189 | return memblock_setclr_flag(&memblock.memory, base, size, 1,
|
| 1190 | MEMBLOCK_KHO_SCRATCH);
|
| 1191 | }
|
| 1192 |
|
| 1193 | |
| 1194 | * memblock_clear_kho_scratch - Clear MEMBLOCK_KHO_SCRATCH flag for a
|
| 1195 | * specified region.
|
| 1196 | * @base: the base phys addr of the region
|
| 1197 | * @size: the size of the region
|
| 1198 | *
|
| 1199 | * Return: 0 on success, -errno on failure.
|
| 1200 |
|
| 1201 | __init int memblock_clear_kho_scratch(phys_addr_t base, phys_addr_t size)
|
| 1202 | {
|
| 1203 | return memblock_setclr_flag(&memblock.memory, base, size, 0,
|
| 1204 | MEMBLOCK_KHO_SCRATCH);
|
| 1205 | }
|
| 1206 |
|
| 1207 | static bool should_skip_region(struct memblock_type *type,
|
| 1208 | struct memblock_region *m,
|
| 1209 | int nid, int flags)
|
| 1210 | {
|
| 1211 | int m_nid = memblock_get_region_node(m);
|
| 1212 |
|
| 1213 |
|
| 1214 | if (type != memblock_memory)
|
| 1215 | return false;
|
| 1216 |
|
| 1217 |
|
| 1218 | if (numa_valid_node(nid) && nid != m_nid)
|
| 1219 | return true;
|
| 1220 |
|
| 1221 |
|
| 1222 | if (movable_node_is_enabled() && memblock_is_hotpluggable(m) &&
|
| 1223 | !(flags & MEMBLOCK_HOTPLUG))
|
| 1224 | return true;
|
| 1225 |
|
| 1226 |
|
| 1227 | if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m))
|
| 1228 | return true;
|
| 1229 |
|
| 1230 |
|
| 1231 | if (!(flags & MEMBLOCK_NOMAP) && memblock_is_nomap(m))
|
| 1232 | return true;
|
| 1233 |
|
| 1234 |
|
| 1235 | if (!(flags & MEMBLOCK_DRIVER_MANAGED) && memblock_is_driver_managed(m))
|
| 1236 | return true;
|
| 1237 |
|
| 1238 | |
| 1239 | * In early alloc during kexec handover, we can only consider
|
| 1240 | * MEMBLOCK_KHO_SCRATCH regions for the allocations
|
| 1241 |
|
| 1242 | if ((flags & MEMBLOCK_KHO_SCRATCH) && !memblock_is_kho_scratch(m))
|
| 1243 | return true;
|
| 1244 |
|
| 1245 | return false;
|
| 1246 | }
|
| 1247 |
|
| 1248 | |
| 1249 | * __next_mem_range - next function for for_each_free_mem_range() etc.
|
| 1250 | * @idx: pointer to u64 loop variable
|
| 1251 | * @nid: node selector, %NUMA_NO_NODE for all nodes
|
| 1252 | * @flags: pick from blocks based on memory attributes
|
| 1253 | * @type_a: pointer to memblock_type from where the range is taken
|
| 1254 | * @type_b: pointer to memblock_type which excludes memory from being taken
|
| 1255 | * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
|
| 1256 | * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
|
| 1257 | * @out_nid: ptr to int for nid of the range, can be %NULL
|
| 1258 | *
|
| 1259 | * Find the first area from *@idx which matches @nid, fill the out
|
| 1260 | * parameters, and update *@idx for the next iteration. The lower 32bit of
|
| 1261 | * *@idx contains index into type_a and the upper 32bit indexes the
|
| 1262 | * areas before each region in type_b. For example, if type_b regions
|
| 1263 | * look like the following,
|
| 1264 | *
|
| 1265 | * 0:[0-16), 1:[32-48), 2:[128-130)
|
| 1266 | *
|
| 1267 | * The upper 32bit indexes the following regions.
|
| 1268 | *
|
| 1269 | * 0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX)
|
| 1270 | *
|
| 1271 | * As both region arrays are sorted, the function advances the two indices
|
| 1272 | * in lockstep and returns each intersection.
|
| 1273 |
|
| 1274 | void __next_mem_range(u64 *idx, int nid, enum memblock_flags flags,
|
| 1275 | struct memblock_type *type_a,
|
| 1276 | struct memblock_type *type_b, phys_addr_t *out_start,
|
| 1277 | phys_addr_t *out_end, int *out_nid)
|
| 1278 | {
|
| 1279 | int idx_a = *idx & 0xffffffff;
|
| 1280 | int idx_b = *idx >> 32;
|
| 1281 |
|
| 1282 | for (; idx_a < type_a->cnt; idx_a++) {
|
| 1283 | struct memblock_region *m = &type_a->regions[idx_a];
|
| 1284 |
|
| 1285 | phys_addr_t m_start = m->base;
|
| 1286 | phys_addr_t m_end = m->base + m->size;
|
| 1287 | int m_nid = memblock_get_region_node(m);
|
| 1288 |
|
| 1289 | if (should_skip_region(type_a, m, nid, flags))
|
| 1290 | continue;
|
| 1291 |
|
| 1292 | if (!type_b) {
|
| 1293 | if (out_start)
|
| 1294 | *out_start = m_start;
|
| 1295 | if (out_end)
|
| 1296 | *out_end = m_end;
|
| 1297 | if (out_nid)
|
| 1298 | *out_nid = m_nid;
|
| 1299 | idx_a++;
|
| 1300 | *idx = (u32)idx_a | (u64)idx_b << 32;
|
| 1301 | return;
|
| 1302 | }
|
| 1303 |
|
| 1304 |
|
| 1305 | for (; idx_b < type_b->cnt + 1; idx_b++) {
|
| 1306 | struct memblock_region *r;
|
| 1307 | phys_addr_t r_start;
|
| 1308 | phys_addr_t r_end;
|
| 1309 |
|
| 1310 | r = &type_b->regions[idx_b];
|
| 1311 | r_start = idx_b ? r[-1].base + r[-1].size : 0;
|
| 1312 | r_end = idx_b < type_b->cnt ?
|
| 1313 | r->base : PHYS_ADDR_MAX;
|
| 1314 |
|
| 1315 | |
| 1316 | * if idx_b advanced past idx_a,
|
| 1317 | * break out to advance idx_a
|
| 1318 |
|
| 1319 | if (r_start >= m_end)
|
| 1320 | break;
|
| 1321 |
|
| 1322 | if (m_start < r_end) {
|
| 1323 | if (out_start)
|
| 1324 | *out_start =
|
| 1325 | max(m_start, r_start);
|
| 1326 | if (out_end)
|
| 1327 | *out_end = min(m_end, r_end);
|
| 1328 | if (out_nid)
|
| 1329 | *out_nid = m_nid;
|
| 1330 | |
| 1331 | * The region which ends first is
|
| 1332 | * advanced for the next iteration.
|
| 1333 |
|
| 1334 | if (m_end <= r_end)
|
| 1335 | idx_a++;
|
| 1336 | else
|
| 1337 | idx_b++;
|
| 1338 | *idx = (u32)idx_a | (u64)idx_b << 32;
|
| 1339 | return;
|
| 1340 | }
|
| 1341 | }
|
| 1342 | }
|
| 1343 |
|
| 1344 |
|
| 1345 | *idx = ULLONG_MAX;
|
| 1346 | }
|
| 1347 |
|
| 1348 | |
| 1349 | * __next_mem_range_rev - generic next function for for_each_*_range_rev()
|
| 1350 | *
|
| 1351 | * @idx: pointer to u64 loop variable
|
| 1352 | * @nid: node selector, %NUMA_NO_NODE for all nodes
|
| 1353 | * @flags: pick from blocks based on memory attributes
|
| 1354 | * @type_a: pointer to memblock_type from where the range is taken
|
| 1355 | * @type_b: pointer to memblock_type which excludes memory from being taken
|
| 1356 | * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
|
| 1357 | * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
|
| 1358 | * @out_nid: ptr to int for nid of the range, can be %NULL
|
| 1359 | *
|
| 1360 | * Finds the next range from type_a which is not marked as unsuitable
|
| 1361 | * in type_b.
|
| 1362 | *
|
| 1363 | * Reverse of __next_mem_range().
|
| 1364 |
|
| 1365 | void __init_memblock __next_mem_range_rev(u64 *idx, int nid,
|
| 1366 | enum memblock_flags flags,
|
| 1367 | struct memblock_type *type_a,
|
| 1368 | struct memblock_type *type_b,
|
| 1369 | phys_addr_t *out_start,
|
| 1370 | phys_addr_t *out_end, int *out_nid)
|
| 1371 | {
|
| 1372 | int idx_a = *idx & 0xffffffff;
|
| 1373 | int idx_b = *idx >> 32;
|
| 1374 |
|
| 1375 | if (*idx == (u64)ULLONG_MAX) {
|
| 1376 | idx_a = type_a->cnt - 1;
|
| 1377 | if (type_b != NULL)
|
| 1378 | idx_b = type_b->cnt;
|
| 1379 | else
|
| 1380 | idx_b = 0;
|
| 1381 | }
|
| 1382 |
|
| 1383 | for (; idx_a >= 0; idx_a--) {
|
| 1384 | struct memblock_region *m = &type_a->regions[idx_a];
|
| 1385 |
|
| 1386 | phys_addr_t m_start = m->base;
|
| 1387 | phys_addr_t m_end = m->base + m->size;
|
| 1388 | int m_nid = memblock_get_region_node(m);
|
| 1389 |
|
| 1390 | if (should_skip_region(type_a, m, nid, flags))
|
| 1391 | continue;
|
| 1392 |
|
| 1393 | if (!type_b) {
|
| 1394 | if (out_start)
|
| 1395 | *out_start = m_start;
|
| 1396 | if (out_end)
|
| 1397 | *out_end = m_end;
|
| 1398 | if (out_nid)
|
| 1399 | *out_nid = m_nid;
|
| 1400 | idx_a--;
|
| 1401 | *idx = (u32)idx_a | (u64)idx_b << 32;
|
| 1402 | return;
|
| 1403 | }
|
| 1404 |
|
| 1405 |
|
| 1406 | for (; idx_b >= 0; idx_b--) {
|
| 1407 | struct memblock_region *r;
|
| 1408 | phys_addr_t r_start;
|
| 1409 | phys_addr_t r_end;
|
| 1410 |
|
| 1411 | r = &type_b->regions[idx_b];
|
| 1412 | r_start = idx_b ? r[-1].base + r[-1].size : 0;
|
| 1413 | r_end = idx_b < type_b->cnt ?
|
| 1414 | r->base : PHYS_ADDR_MAX;
|
| 1415 | |
| 1416 | * if idx_b advanced past idx_a,
|
| 1417 | * break out to advance idx_a
|
| 1418 |
|
| 1419 |
|
| 1420 | if (r_end <= m_start)
|
| 1421 | break;
|
| 1422 |
|
| 1423 | if (m_end > r_start) {
|
| 1424 | if (out_start)
|
| 1425 | *out_start = max(m_start, r_start);
|
| 1426 | if (out_end)
|
| 1427 | *out_end = min(m_end, r_end);
|
| 1428 | if (out_nid)
|
| 1429 | *out_nid = m_nid;
|
| 1430 | if (m_start >= r_start)
|
| 1431 | idx_a--;
|
| 1432 | else
|
| 1433 | idx_b--;
|
| 1434 | *idx = (u32)idx_a | (u64)idx_b << 32;
|
| 1435 | return;
|
| 1436 | }
|
| 1437 | }
|
| 1438 | }
|
| 1439 |
|
| 1440 | *idx = ULLONG_MAX;
|
| 1441 | }
|
| 1442 |
|
| 1443 | |
| 1444 | * Common iterator interface used to define for_each_mem_pfn_range().
|
| 1445 |
|
| 1446 | void __init_memblock __next_mem_pfn_range(int *idx, int nid,
|
| 1447 | unsigned long *out_start_pfn,
|
| 1448 | unsigned long *out_end_pfn, int *out_nid)
|
| 1449 | {
|
| 1450 | struct memblock_type *type = &memblock.memory;
|
| 1451 | struct memblock_region *r;
|
| 1452 | int r_nid;
|
| 1453 |
|
| 1454 | while (++*idx < type->cnt) {
|
| 1455 | r = &type->regions[*idx];
|
| 1456 | r_nid = memblock_get_region_node(r);
|
| 1457 |
|
| 1458 | if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size))
|
| 1459 | continue;
|
| 1460 | if (!numa_valid_node(nid) || nid == r_nid)
|
| 1461 | break;
|
| 1462 | }
|
| 1463 | if (*idx >= type->cnt) {
|
| 1464 | *idx = -1;
|
| 1465 | return;
|
| 1466 | }
|
| 1467 |
|
| 1468 | if (out_start_pfn)
|
| 1469 | *out_start_pfn = PFN_UP(r->base);
|
| 1470 | if (out_end_pfn)
|
| 1471 | *out_end_pfn = PFN_DOWN(r->base + r->size);
|
| 1472 | if (out_nid)
|
| 1473 | *out_nid = r_nid;
|
| 1474 | }
|
| 1475 |
|
| 1476 | |
| 1477 | * memblock_set_node - set node ID on memblock regions
|
| 1478 | * @base: base of area to set node ID for
|
| 1479 | * @size: size of area to set node ID for
|
| 1480 | * @type: memblock type to set node ID for
|
| 1481 | * @nid: node ID to set
|
| 1482 | *
|
| 1483 | * Set the nid of memblock @type regions in [@base, @base + @size) to @nid.
|
| 1484 | * Regions which cross the area boundaries are split as necessary.
|
| 1485 | *
|
| 1486 | * Return:
|
| 1487 | * 0 on success, -errno on failure.
|
| 1488 |
|
| 1489 | int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size,
|
| 1490 | struct memblock_type *type, int nid)
|
| 1491 | {
|
| 1492 | #ifdef CONFIG_NUMA
|
| 1493 | int start_rgn, end_rgn;
|
| 1494 | int i, ret;
|
| 1495 |
|
| 1496 | ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
|
| 1497 | if (ret)
|
| 1498 | return ret;
|
| 1499 |
|
| 1500 | for (i = start_rgn; i < end_rgn; i++)
|
| 1501 | memblock_set_region_node(&type->regions[i], nid);
|
| 1502 |
|
| 1503 | memblock_merge_regions(type, start_rgn, end_rgn);
|
| 1504 | #endif
|
| 1505 | return 0;
|
| 1506 | }
|
| 1507 |
|
| 1508 | static void memblock_prep_allocation(phys_addr_t start, phys_addr_t size,
|
| 1509 | bool kmemleak_trace)
|
| 1510 | {
|
| 1511 | |
| 1512 | * Skip kmemleak for those places like kasan_init() and
|
| 1513 | * early_pgtable_alloc() due to high volume.
|
| 1514 |
|
| 1515 | if (kmemleak_trace)
|
| 1516 | |
| 1517 | * Memblock allocated blocks are never reported as
|
| 1518 | * leaks. This is because many of these blocks are
|
| 1519 | * only referred via the physical address which is
|
| 1520 | * not looked up by kmemleak.
|
| 1521 |
|
| 1522 | kmemleak_alloc_phys(start, size, 0);
|
| 1523 |
|
| 1524 | |
| 1525 | * Some Virtual Machine platforms, such as Intel TDX or AMD SEV-SNP,
|
| 1526 | * require memory to be accepted before it can be used by the
|
| 1527 | * guest.
|
| 1528 | *
|
| 1529 | * Accept the memory of the allocated buffer.
|
| 1530 |
|
| 1531 | accept_memory(start, size);
|
| 1532 | }
|
| 1533 |
|
| 1534 | |
| 1535 | * memblock_alloc_range_nid - allocate boot memory block
|
| 1536 | * @size: size of memory block to be allocated in bytes
|
| 1537 | * @align: alignment of the region and block's size
|
| 1538 | * @start: the lower bound of the memory region to allocate (phys address)
|
| 1539 | * @end: the upper bound of the memory region to allocate (phys address)
|
| 1540 | * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
|
| 1541 | * @exact_nid: control the allocation fall back to other nodes
|
| 1542 | *
|
| 1543 | * The allocation is performed from memory region limited by
|
| 1544 | * memblock.current_limit if @end == %MEMBLOCK_ALLOC_ACCESSIBLE.
|
| 1545 | *
|
| 1546 | * If the specified node can not hold the requested memory and @exact_nid
|
| 1547 | * is false, the allocation falls back to any node in the system.
|
| 1548 | *
|
| 1549 | * For systems with memory mirroring, the allocation is attempted first
|
| 1550 | * from the regions with mirroring enabled and then retried from any
|
| 1551 | * memory region.
|
| 1552 | *
|
| 1553 | * In addition, function using kmemleak_alloc_phys for allocated boot
|
| 1554 | * memory block, it is never reported as leaks.
|
| 1555 | *
|
| 1556 | * Return:
|
| 1557 | * Physical address of allocated memory block on success, %0 on failure.
|
| 1558 |
|
| 1559 | phys_addr_t __init memblock_alloc_range_nid(phys_addr_t size,
|
| 1560 | phys_addr_t align, phys_addr_t start,
|
| 1561 | phys_addr_t end, int nid,
|
| 1562 | bool exact_nid)
|
| 1563 | {
|
| 1564 | enum memblock_flags flags = choose_memblock_flags();
|
| 1565 | phys_addr_t found;
|
| 1566 |
|
| 1567 | |
| 1568 | * Detect any accidental use of these APIs after slab is ready, as at
|
| 1569 | * this moment memblock may be deinitialized already and its
|
| 1570 | * internal data may be destroyed (after execution of memblock_free_all)
|
| 1571 |
|
| 1572 | if (WARN_ON_ONCE(slab_is_available())) {
|
| 1573 | void *vaddr = kzalloc_node(size, GFP_NOWAIT, nid);
|
| 1574 |
|
| 1575 | return vaddr ? virt_to_phys(vaddr) : 0;
|
| 1576 | }
|
| 1577 |
|
| 1578 | if (!align) {
|
| 1579 |
|
| 1580 | dump_stack();
|
| 1581 | align = SMP_CACHE_BYTES;
|
| 1582 | }
|
| 1583 |
|
| 1584 | again:
|
| 1585 | found = memblock_find_in_range_node(size, align, start, end, nid,
|
| 1586 | flags);
|
| 1587 | if (found && !__memblock_reserve(found, size, nid, MEMBLOCK_RSRV_KERN))
|
| 1588 | goto done;
|
| 1589 |
|
| 1590 | if (numa_valid_node(nid) && !exact_nid) {
|
| 1591 | found = memblock_find_in_range_node(size, align, start,
|
| 1592 | end, NUMA_NO_NODE,
|
| 1593 | flags);
|
| 1594 | if (found && !memblock_reserve_kern(found, size))
|
| 1595 | goto done;
|
| 1596 | }
|
| 1597 |
|
| 1598 | if (flags & MEMBLOCK_MIRROR) {
|
| 1599 | flags &= ~MEMBLOCK_MIRROR;
|
| 1600 | pr_warn_ratelimited("Could not allocate %pap bytes of mirrored memory\n",
|
| 1601 | &size);
|
| 1602 | goto again;
|
| 1603 | }
|
| 1604 |
|
| 1605 | return 0;
|
| 1606 |
|
| 1607 | done:
|
| 1608 | memblock_prep_allocation(found, size, end != MEMBLOCK_ALLOC_NOLEAKTRACE);
|
| 1609 | return found;
|
| 1610 | }
|
| 1611 |
|
| 1612 | |
| 1613 | * memblock_phys_alloc_range - allocate a memory block inside specified range
|
| 1614 | * @size: size of memory block to be allocated in bytes
|
| 1615 | * @align: alignment of the region and block's size
|
| 1616 | * @start: the lower bound of the memory region to allocate (physical address)
|
| 1617 | * @end: the upper bound of the memory region to allocate (physical address)
|
| 1618 | *
|
| 1619 | * Allocate @size bytes in the between @start and @end.
|
| 1620 | *
|
| 1621 | * Return: physical address of the allocated memory block on success,
|
| 1622 | * %0 on failure.
|
| 1623 |
|
| 1624 | phys_addr_t __init memblock_phys_alloc_range(phys_addr_t size,
|
| 1625 | phys_addr_t align,
|
| 1626 | phys_addr_t start,
|
| 1627 | phys_addr_t end)
|
| 1628 | {
|
| 1629 | memblock_dbg("%s: %llu bytes align=0x%llx from=%pa max_addr=%pa %pS\n",
|
| 1630 | __func__, (u64)size, (u64)align, &start, &end,
|
| 1631 | (void *)_RET_IP_);
|
| 1632 | return memblock_alloc_range_nid(size, align, start, end, NUMA_NO_NODE,
|
| 1633 | false);
|
| 1634 | }
|
| 1635 |
|
| 1636 | |
| 1637 | * memblock_phys_alloc_try_nid - allocate a memory block from specified NUMA node
|
| 1638 | * @size: size of memory block to be allocated in bytes
|
| 1639 | * @align: alignment of the region and block's size
|
| 1640 | * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
|
| 1641 | *
|
| 1642 | * Allocates memory block from the specified NUMA node. If the node
|
| 1643 | * has no available memory, attempts to allocated from any node in the
|
| 1644 | * system.
|
| 1645 | *
|
| 1646 | * Return: physical address of the allocated memory block on success,
|
| 1647 | * %0 on failure.
|
| 1648 |
|
| 1649 | phys_addr_t __init memblock_phys_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid)
|
| 1650 | {
|
| 1651 | return memblock_alloc_range_nid(size, align, 0,
|
| 1652 | MEMBLOCK_ALLOC_ACCESSIBLE, nid, false);
|
| 1653 | }
|
| 1654 |
|
| 1655 | |
| 1656 | * memblock_alloc_internal - allocate boot memory block
|
| 1657 | * @size: size of memory block to be allocated in bytes
|
| 1658 | * @align: alignment of the region and block's size
|
| 1659 | * @min_addr: the lower bound of the memory region to allocate (phys address)
|
| 1660 | * @max_addr: the upper bound of the memory region to allocate (phys address)
|
| 1661 | * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
|
| 1662 | * @exact_nid: control the allocation fall back to other nodes
|
| 1663 | *
|
| 1664 | * Allocates memory block using memblock_alloc_range_nid() and
|
| 1665 | * converts the returned physical address to virtual.
|
| 1666 | *
|
| 1667 | * The @min_addr limit is dropped if it can not be satisfied and the allocation
|
| 1668 | * will fall back to memory below @min_addr. Other constraints, such
|
| 1669 | * as node and mirrored memory will be handled again in
|
| 1670 | * memblock_alloc_range_nid().
|
| 1671 | *
|
| 1672 | * Return:
|
| 1673 | * Virtual address of allocated memory block on success, NULL on failure.
|
| 1674 |
|
| 1675 | static void * __init memblock_alloc_internal(
|
| 1676 | phys_addr_t size, phys_addr_t align,
|
| 1677 | phys_addr_t min_addr, phys_addr_t max_addr,
|
| 1678 | int nid, bool exact_nid)
|
| 1679 | {
|
| 1680 | phys_addr_t alloc;
|
| 1681 |
|
| 1682 |
|
| 1683 | if (max_addr > memblock.current_limit)
|
| 1684 | max_addr = memblock.current_limit;
|
| 1685 |
|
| 1686 | alloc = memblock_alloc_range_nid(size, align, min_addr, max_addr, nid,
|
| 1687 | exact_nid);
|
| 1688 |
|
| 1689 |
|
| 1690 | if (!alloc && min_addr)
|
| 1691 | alloc = memblock_alloc_range_nid(size, align, 0, max_addr, nid,
|
| 1692 | exact_nid);
|
| 1693 |
|
| 1694 | if (!alloc)
|
| 1695 | return NULL;
|
| 1696 |
|
| 1697 | return phys_to_virt(alloc);
|
| 1698 | }
|
| 1699 |
|
| 1700 | |
| 1701 | * memblock_alloc_exact_nid_raw - allocate boot memory block on the exact node
|
| 1702 | * without zeroing memory
|
| 1703 | * @size: size of memory block to be allocated in bytes
|
| 1704 | * @align: alignment of the region and block's size
|
| 1705 | * @min_addr: the lower bound of the memory region from where the allocation
|
| 1706 | * is preferred (phys address)
|
| 1707 | * @max_addr: the upper bound of the memory region from where the allocation
|
| 1708 | * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to
|
| 1709 | * allocate only from memory limited by memblock.current_limit value
|
| 1710 | * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
|
| 1711 | *
|
| 1712 | * Public function, provides additional debug information (including caller
|
| 1713 | * info), if enabled. Does not zero allocated memory.
|
| 1714 | *
|
| 1715 | * Return:
|
| 1716 | * Virtual address of allocated memory block on success, NULL on failure.
|
| 1717 |
|
| 1718 | void * __init memblock_alloc_exact_nid_raw(
|
| 1719 | phys_addr_t size, phys_addr_t align,
|
| 1720 | phys_addr_t min_addr, phys_addr_t max_addr,
|
| 1721 | int nid)
|
| 1722 | {
|
| 1723 | memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n",
|
| 1724 | __func__, (u64)size, (u64)align, nid, &min_addr,
|
| 1725 | &max_addr, (void *)_RET_IP_);
|
| 1726 |
|
| 1727 | return memblock_alloc_internal(size, align, min_addr, max_addr, nid,
|
| 1728 | true);
|
| 1729 | }
|
| 1730 |
|
| 1731 | |
| 1732 | * memblock_alloc_try_nid_raw - allocate boot memory block without zeroing
|
| 1733 | * memory and without panicking
|
| 1734 | * @size: size of memory block to be allocated in bytes
|
| 1735 | * @align: alignment of the region and block's size
|
| 1736 | * @min_addr: the lower bound of the memory region from where the allocation
|
| 1737 | * is preferred (phys address)
|
| 1738 | * @max_addr: the upper bound of the memory region from where the allocation
|
| 1739 | * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to
|
| 1740 | * allocate only from memory limited by memblock.current_limit value
|
| 1741 | * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
|
| 1742 | *
|
| 1743 | * Public function, provides additional debug information (including caller
|
| 1744 | * info), if enabled. Does not zero allocated memory, does not panic if request
|
| 1745 | * cannot be satisfied.
|
| 1746 | *
|
| 1747 | * Return:
|
| 1748 | * Virtual address of allocated memory block on success, NULL on failure.
|
| 1749 |
|
| 1750 | void * __init memblock_alloc_try_nid_raw(
|
| 1751 | phys_addr_t size, phys_addr_t align,
|
| 1752 | phys_addr_t min_addr, phys_addr_t max_addr,
|
| 1753 | int nid)
|
| 1754 | {
|
| 1755 | memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n",
|
| 1756 | __func__, (u64)size, (u64)align, nid, &min_addr,
|
| 1757 | &max_addr, (void *)_RET_IP_);
|
| 1758 |
|
| 1759 | return memblock_alloc_internal(size, align, min_addr, max_addr, nid,
|
| 1760 | false);
|
| 1761 | }
|
| 1762 |
|
| 1763 | |
| 1764 | * memblock_alloc_hugetlb - allocate boot memory for HugeTLB pages
|
| 1765 | * @size: size of the memory to be allocated in bytes
|
| 1766 | * @nid: nid of the free memory to find, %NUMA_NO_NODE for any node
|
| 1767 | * @exact_nid: only allocate from the specified nid. If %false, the specified
|
| 1768 | * nid is tried first, and then all nodes are tried as fallback.
|
| 1769 | *
|
| 1770 | * HugeTLB pages are always aligned by their size, so the alignment matches
|
| 1771 | * @size. Since the memory is for userspace, mirrored memory is not used. The
|
| 1772 | * memory is not zeroed. Does not panic if request cannot be satisfied.
|
| 1773 | *
|
| 1774 | * Return:
|
| 1775 | * Virtual address of allocated memory block on success, %NULL on failure.
|
| 1776 |
|
| 1777 | void * __init memblock_alloc_hugetlb(phys_addr_t size, int nid, bool exact_nid)
|
| 1778 | {
|
| 1779 | enum memblock_flags flags = choose_memblock_flags();
|
| 1780 | phys_addr_t addr, start = 0, end = MEMBLOCK_ALLOC_ACCESSIBLE;
|
| 1781 |
|
| 1782 | memblock_dbg("%s: %llu bytes, nid=%d, exact_nid=%d %pS\n", __func__,
|
| 1783 | (u64)size, nid, exact_nid, (void *)_RET_IP_);
|
| 1784 |
|
| 1785 |
|
| 1786 | flags &= ~MEMBLOCK_MIRROR;
|
| 1787 | retry:
|
| 1788 |
|
| 1789 | addr = memblock_find_in_range_node(size, size, start, end, nid, flags);
|
| 1790 | if (addr)
|
| 1791 | goto found;
|
| 1792 |
|
| 1793 |
|
| 1794 | if (numa_valid_node(nid) && !exact_nid) {
|
| 1795 | nid = NUMA_NO_NODE;
|
| 1796 | |
| 1797 | * If a previous candidate overlapped with KHO scratch, it would
|
| 1798 | * update start or end. Now that the search is opening to all
|
| 1799 | * nodes, reset them.
|
| 1800 |
|
| 1801 | start = 0;
|
| 1802 | end = MEMBLOCK_ALLOC_ACCESSIBLE;
|
| 1803 |
|
| 1804 | goto retry;
|
| 1805 | }
|
| 1806 |
|
| 1807 |
|
| 1808 | return NULL;
|
| 1809 |
|
| 1810 | found:
|
| 1811 | |
| 1812 | * HugeTLB pages can be preserved with KHO and no preserved memory can
|
| 1813 | * be in scratch. So retry if found address overlaps with scratch.
|
| 1814 | *
|
| 1815 | * Scratch areas are normally not very large, so this shouldn't take too
|
| 1816 | * many retries.
|
| 1817 |
|
| 1818 | if (kho_scratch_overlap(addr, size)) {
|
| 1819 | if (memblock_bottom_up())
|
| 1820 | start = addr + size;
|
| 1821 | else
|
| 1822 | end = addr;
|
| 1823 |
|
| 1824 | goto retry;
|
| 1825 | }
|
| 1826 |
|
| 1827 | if (__memblock_reserve(addr, size, nid, MEMBLOCK_RSRV_KERN | MEMBLOCK_RSRV_HUGETLB))
|
| 1828 | return NULL;
|
| 1829 |
|
| 1830 | memblock_prep_allocation(addr, size, true);
|
| 1831 | return phys_to_virt(addr);
|
| 1832 | }
|
| 1833 |
|
| 1834 | |
| 1835 | * memblock_alloc_try_nid - allocate boot memory block
|
| 1836 | * @size: size of memory block to be allocated in bytes
|
| 1837 | * @align: alignment of the region and block's size
|
| 1838 | * @min_addr: the lower bound of the memory region from where the allocation
|
| 1839 | * is preferred (phys address)
|
| 1840 | * @max_addr: the upper bound of the memory region from where the allocation
|
| 1841 | * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to
|
| 1842 | * allocate only from memory limited by memblock.current_limit value
|
| 1843 | * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
|
| 1844 | *
|
| 1845 | * Public function, provides additional debug information (including caller
|
| 1846 | * info), if enabled. This function zeroes the allocated memory.
|
| 1847 | *
|
| 1848 | * Return:
|
| 1849 | * Virtual address of allocated memory block on success, NULL on failure.
|
| 1850 |
|
| 1851 | void * __init memblock_alloc_try_nid(
|
| 1852 | phys_addr_t size, phys_addr_t align,
|
| 1853 | phys_addr_t min_addr, phys_addr_t max_addr,
|
| 1854 | int nid)
|
| 1855 | {
|
| 1856 | void *ptr;
|
| 1857 |
|
| 1858 | memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n",
|
| 1859 | __func__, (u64)size, (u64)align, nid, &min_addr,
|
| 1860 | &max_addr, (void *)_RET_IP_);
|
| 1861 | ptr = memblock_alloc_internal(size, align,
|
| 1862 | min_addr, max_addr, nid, false);
|
| 1863 | if (ptr)
|
| 1864 | memset(ptr, 0, size);
|
| 1865 |
|
| 1866 | return ptr;
|
| 1867 | }
|
| 1868 |
|
| 1869 | |
| 1870 | * __memblock_alloc_or_panic - Try to allocate memory and panic on failure
|
| 1871 | * @size: size of memory block to be allocated in bytes
|
| 1872 | * @align: alignment of the region and block's size
|
| 1873 | * @func: caller func name
|
| 1874 | *
|
| 1875 | * This function attempts to allocate memory using memblock_alloc,
|
| 1876 | * and in case of failure, it calls panic with the formatted message.
|
| 1877 | * This function should not be used directly, please use the macro memblock_alloc_or_panic.
|
| 1878 |
|
| 1879 | void *__init __memblock_alloc_or_panic(phys_addr_t size, phys_addr_t align,
|
| 1880 | const char *func)
|
| 1881 | {
|
| 1882 | void *addr = memblock_alloc(size, align);
|
| 1883 |
|
| 1884 | if (unlikely(!addr))
|
| 1885 | panic("%s: Failed to allocate %pap bytes\n", func, &size);
|
| 1886 | return addr;
|
| 1887 | }
|
| 1888 |
|
| 1889 | |
| 1890 | * Remaining API functions
|
| 1891 |
|
| 1892 |
|
| 1893 | phys_addr_t __init_memblock memblock_phys_mem_size(void)
|
| 1894 | {
|
| 1895 | return memblock.memory.total_size;
|
| 1896 | }
|
| 1897 |
|
| 1898 | phys_addr_t __init_memblock memblock_reserved_size(void)
|
| 1899 | {
|
| 1900 | return memblock.reserved.total_size;
|
| 1901 | }
|
| 1902 |
|
| 1903 | phys_addr_t __init_memblock memblock_reserved_hugetlb_size(phys_addr_t limit, int nid)
|
| 1904 | {
|
| 1905 | struct memblock_region *r;
|
| 1906 | phys_addr_t total = 0;
|
| 1907 |
|
| 1908 | for_each_reserved_mem_region(r) {
|
| 1909 | phys_addr_t size = r->size;
|
| 1910 |
|
| 1911 | if (r->base > limit)
|
| 1912 | break;
|
| 1913 |
|
| 1914 | if (r->base + r->size > limit)
|
| 1915 | size = limit - r->base;
|
| 1916 |
|
| 1917 | if (nid == memblock_get_region_node(r) || !numa_valid_node(nid))
|
| 1918 | if (r->flags & MEMBLOCK_RSRV_HUGETLB)
|
| 1919 | total += size;
|
| 1920 | }
|
| 1921 |
|
| 1922 | return total;
|
| 1923 | }
|
| 1924 |
|
| 1925 | phys_addr_t __init_memblock memblock_reserved_kern_size(phys_addr_t limit, int nid)
|
| 1926 | {
|
| 1927 | struct memblock_region *r;
|
| 1928 | phys_addr_t total = 0;
|
| 1929 |
|
| 1930 | for_each_reserved_mem_region(r) {
|
| 1931 | phys_addr_t size = r->size;
|
| 1932 |
|
| 1933 | if (r->base > limit)
|
| 1934 | break;
|
| 1935 |
|
| 1936 | if (r->base + r->size > limit)
|
| 1937 | size = limit - r->base;
|
| 1938 |
|
| 1939 | if (nid == memblock_get_region_node(r) || !numa_valid_node(nid))
|
| 1940 | if (r->flags & MEMBLOCK_RSRV_KERN)
|
| 1941 | total += size;
|
| 1942 | }
|
| 1943 |
|
| 1944 | return total;
|
| 1945 | }
|
| 1946 |
|
| 1947 | |
| 1948 | * memblock_estimated_nr_free_pages - return estimated number of free pages
|
| 1949 | * from memblock point of view
|
| 1950 | *
|
| 1951 | * During bootup, subsystems might need a rough estimate of the number of free
|
| 1952 | * pages in the whole system, before precise numbers are available from the
|
| 1953 | * buddy. Especially with CONFIG_DEFERRED_STRUCT_PAGE_INIT, the numbers
|
| 1954 | * obtained from the buddy might be very imprecise during bootup.
|
| 1955 | *
|
| 1956 | * Return:
|
| 1957 | * An estimated number of free pages from memblock point of view.
|
| 1958 |
|
| 1959 | unsigned long __init memblock_estimated_nr_free_pages(void)
|
| 1960 | {
|
| 1961 | return PHYS_PFN(memblock_phys_mem_size() -
|
| 1962 | memblock_reserved_kern_size(MEMBLOCK_ALLOC_ANYWHERE, NUMA_NO_NODE));
|
| 1963 | }
|
| 1964 |
|
| 1965 |
|
| 1966 | phys_addr_t __init_memblock memblock_start_of_DRAM(void)
|
| 1967 | {
|
| 1968 | return memblock.memory.regions[0].base;
|
| 1969 | }
|
| 1970 |
|
| 1971 | phys_addr_t __init_memblock memblock_end_of_DRAM(void)
|
| 1972 | {
|
| 1973 | int idx = memblock.memory.cnt - 1;
|
| 1974 |
|
| 1975 | return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size);
|
| 1976 | }
|
| 1977 |
|
| 1978 | static phys_addr_t __init_memblock __find_max_addr(phys_addr_t limit)
|
| 1979 | {
|
| 1980 | phys_addr_t max_addr = PHYS_ADDR_MAX;
|
| 1981 | struct memblock_region *r;
|
| 1982 |
|
| 1983 | |
| 1984 | * translate the memory @limit size into the max address within one of
|
| 1985 | * the memory memblock regions, if the @limit exceeds the total size
|
| 1986 | * of those regions, max_addr will keep original value PHYS_ADDR_MAX
|
| 1987 |
|
| 1988 | for_each_mem_region(r) {
|
| 1989 | if (limit <= r->size) {
|
| 1990 | max_addr = r->base + limit;
|
| 1991 | break;
|
| 1992 | }
|
| 1993 | limit -= r->size;
|
| 1994 | }
|
| 1995 |
|
| 1996 | return max_addr;
|
| 1997 | }
|
| 1998 |
|
| 1999 | void __init memblock_enforce_memory_limit(phys_addr_t limit)
|
| 2000 | {
|
| 2001 | phys_addr_t max_addr;
|
| 2002 |
|
| 2003 | if (!limit)
|
| 2004 | return;
|
| 2005 |
|
| 2006 | max_addr = __find_max_addr(limit);
|
| 2007 |
|
| 2008 |
|
| 2009 | if (max_addr == PHYS_ADDR_MAX)
|
| 2010 | return;
|
| 2011 |
|
| 2012 |
|
| 2013 | memblock_remove_range(&memblock.memory, max_addr,
|
| 2014 | PHYS_ADDR_MAX);
|
| 2015 | memblock_remove_range(&memblock.reserved, max_addr,
|
| 2016 | PHYS_ADDR_MAX);
|
| 2017 | }
|
| 2018 |
|
| 2019 | void __init memblock_cap_memory_range(phys_addr_t base, phys_addr_t size)
|
| 2020 | {
|
| 2021 | int start_rgn, end_rgn;
|
| 2022 | int i, ret;
|
| 2023 |
|
| 2024 | if (!size)
|
| 2025 | return;
|
| 2026 |
|
| 2027 | if (!memblock_memory->total_size) {
|
| 2028 | pr_warn("%s: No memory registered yet\n", __func__);
|
| 2029 | return;
|
| 2030 | }
|
| 2031 |
|
| 2032 | ret = memblock_isolate_range(&memblock.memory, base, size,
|
| 2033 | &start_rgn, &end_rgn);
|
| 2034 | if (ret)
|
| 2035 | return;
|
| 2036 |
|
| 2037 |
|
| 2038 | for (i = memblock.memory.cnt - 1; i >= end_rgn; i--)
|
| 2039 | if (!memblock_is_nomap(&memblock.memory.regions[i]))
|
| 2040 | memblock_remove_region(&memblock.memory, i);
|
| 2041 |
|
| 2042 | for (i = start_rgn - 1; i >= 0; i--)
|
| 2043 | if (!memblock_is_nomap(&memblock.memory.regions[i]))
|
| 2044 | memblock_remove_region(&memblock.memory, i);
|
| 2045 |
|
| 2046 |
|
| 2047 | memblock_remove_range(&memblock.reserved, 0, base);
|
| 2048 | memblock_remove_range(&memblock.reserved,
|
| 2049 | base + size, PHYS_ADDR_MAX);
|
| 2050 | }
|
| 2051 |
|
| 2052 | void __init memblock_mem_limit_remove_map(phys_addr_t limit)
|
| 2053 | {
|
| 2054 | phys_addr_t max_addr;
|
| 2055 |
|
| 2056 | if (!limit)
|
| 2057 | return;
|
| 2058 |
|
| 2059 | max_addr = __find_max_addr(limit);
|
| 2060 |
|
| 2061 |
|
| 2062 | if (max_addr == PHYS_ADDR_MAX)
|
| 2063 | return;
|
| 2064 |
|
| 2065 | memblock_cap_memory_range(0, max_addr);
|
| 2066 | }
|
| 2067 |
|
| 2068 | static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr)
|
| 2069 | {
|
| 2070 | unsigned int left = 0, right = type->cnt;
|
| 2071 |
|
| 2072 | do {
|
| 2073 | unsigned int mid = (right + left) / 2;
|
| 2074 |
|
| 2075 | if (addr < type->regions[mid].base)
|
| 2076 | right = mid;
|
| 2077 | else if (addr >= (type->regions[mid].base +
|
| 2078 | type->regions[mid].size))
|
| 2079 | left = mid + 1;
|
| 2080 | else
|
| 2081 | return mid;
|
| 2082 | } while (left < right);
|
| 2083 | return -1;
|
| 2084 | }
|
| 2085 |
|
| 2086 | bool __init_memblock memblock_is_reserved(phys_addr_t addr)
|
| 2087 | {
|
| 2088 | return memblock_search(&memblock.reserved, addr) != -1;
|
| 2089 | }
|
| 2090 |
|
| 2091 | bool __init_memblock memblock_is_memory(phys_addr_t addr)
|
| 2092 | {
|
| 2093 | return memblock_search(&memblock.memory, addr) != -1;
|
| 2094 | }
|
| 2095 |
|
| 2096 | bool __init_memblock memblock_is_map_memory(phys_addr_t addr)
|
| 2097 | {
|
| 2098 | int i = memblock_search(&memblock.memory, addr);
|
| 2099 |
|
| 2100 | if (i == -1)
|
| 2101 | return false;
|
| 2102 | return !memblock_is_nomap(&memblock.memory.regions[i]);
|
| 2103 | }
|
| 2104 |
|
| 2105 | int __init_memblock memblock_search_pfn_nid(unsigned long pfn,
|
| 2106 | unsigned long *start_pfn, unsigned long *end_pfn)
|
| 2107 | {
|
| 2108 | struct memblock_type *type = &memblock.memory;
|
| 2109 | int mid = memblock_search(type, PFN_PHYS(pfn));
|
| 2110 |
|
| 2111 | if (mid == -1)
|
| 2112 | return NUMA_NO_NODE;
|
| 2113 |
|
| 2114 | *start_pfn = PFN_DOWN(type->regions[mid].base);
|
| 2115 | *end_pfn = PFN_DOWN(type->regions[mid].base + type->regions[mid].size);
|
| 2116 |
|
| 2117 | return memblock_get_region_node(&type->regions[mid]);
|
| 2118 | }
|
| 2119 |
|
| 2120 | |
| 2121 | * memblock_is_region_memory - check if a region is a subset of memory
|
| 2122 | * @base: base of region to check
|
| 2123 | * @size: size of region to check
|
| 2124 | *
|
| 2125 | * Check if the region [@base, @base + @size) is a subset of a memory block.
|
| 2126 | *
|
| 2127 | * Return:
|
| 2128 | * 0 if false, non-zero if true
|
| 2129 |
|
| 2130 | bool __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size)
|
| 2131 | {
|
| 2132 | int idx = memblock_search(&memblock.memory, base);
|
| 2133 | phys_addr_t end = base + memblock_cap_size(base, &size);
|
| 2134 |
|
| 2135 | if (idx == -1)
|
| 2136 | return false;
|
| 2137 | return (memblock.memory.regions[idx].base +
|
| 2138 | memblock.memory.regions[idx].size) >= end;
|
| 2139 | }
|
| 2140 |
|
| 2141 | |
| 2142 | * memblock_is_region_reserved - check if a region intersects reserved memory
|
| 2143 | * @base: base of region to check
|
| 2144 | * @size: size of region to check
|
| 2145 | *
|
| 2146 | * Check if the region [@base, @base + @size) intersects a reserved
|
| 2147 | * memory block.
|
| 2148 | *
|
| 2149 | * Return:
|
| 2150 | * True if they intersect, false if not.
|
| 2151 |
|
| 2152 | bool __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size)
|
| 2153 | {
|
| 2154 | return memblock_overlaps_region(&memblock.reserved, base, size);
|
| 2155 | }
|
| 2156 |
|
| 2157 | void __init_memblock memblock_trim_memory(phys_addr_t align)
|
| 2158 | {
|
| 2159 | phys_addr_t start, end, orig_start, orig_end;
|
| 2160 | struct memblock_region *r;
|
| 2161 |
|
| 2162 | for_each_mem_region(r) {
|
| 2163 | orig_start = r->base;
|
| 2164 | orig_end = r->base + r->size;
|
| 2165 | start = round_up(orig_start, align);
|
| 2166 | end = round_down(orig_end, align);
|
| 2167 |
|
| 2168 | if (start == orig_start && end == orig_end)
|
| 2169 | continue;
|
| 2170 |
|
| 2171 | if (start < end) {
|
| 2172 | r->base = start;
|
| 2173 | r->size = end - start;
|
| 2174 | } else {
|
| 2175 | memblock_remove_region(&memblock.memory,
|
| 2176 | r - memblock.memory.regions);
|
| 2177 | r--;
|
| 2178 | }
|
| 2179 | }
|
| 2180 | }
|
| 2181 |
|
| 2182 | void __init_memblock memblock_set_current_limit(phys_addr_t limit)
|
| 2183 | {
|
| 2184 | memblock.current_limit = limit;
|
| 2185 | }
|
| 2186 |
|
| 2187 | phys_addr_t __init_memblock memblock_get_current_limit(void)
|
| 2188 | {
|
| 2189 | return memblock.current_limit;
|
| 2190 | }
|
| 2191 |
|
| 2192 | static void __init_memblock memblock_dump(struct memblock_type *type)
|
| 2193 | {
|
| 2194 | phys_addr_t base, end, size;
|
| 2195 | enum memblock_flags flags;
|
| 2196 | int idx;
|
| 2197 | struct memblock_region *rgn;
|
| 2198 |
|
| 2199 | pr_info(" %s.cnt = 0x%lx\n", type->name, type->cnt);
|
| 2200 |
|
| 2201 | for_each_memblock_type(idx, type, rgn) {
|
| 2202 | char nid_buf[32] = "";
|
| 2203 |
|
| 2204 | base = rgn->base;
|
| 2205 | size = rgn->size;
|
| 2206 | end = base + size - 1;
|
| 2207 | flags = rgn->flags;
|
| 2208 | #ifdef CONFIG_NUMA
|
| 2209 | if (numa_valid_node(memblock_get_region_node(rgn)))
|
| 2210 | snprintf(nid_buf, sizeof(nid_buf), " on node %d",
|
| 2211 | memblock_get_region_node(rgn));
|
| 2212 | #endif
|
| 2213 | pr_info(" %s[%#x]\t[%pa-%pa], %pa bytes%s flags: %#x\n",
|
| 2214 | type->name, idx, &base, &end, &size, nid_buf, flags);
|
| 2215 | }
|
| 2216 | }
|
| 2217 |
|
| 2218 | static void __init_memblock __memblock_dump_all(void)
|
| 2219 | {
|
| 2220 | pr_info("MEMBLOCK configuration:\n");
|
| 2221 | pr_info(" memory size = %pa reserved size = %pa\n",
|
| 2222 | &memblock.memory.total_size,
|
| 2223 | &memblock.reserved.total_size);
|
| 2224 |
|
| 2225 | memblock_dump(&memblock.memory);
|
| 2226 | memblock_dump(&memblock.reserved);
|
| 2227 | #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
|
| 2228 | memblock_dump(&physmem);
|
| 2229 | #endif
|
| 2230 | }
|
| 2231 |
|
| 2232 | void __init_memblock memblock_dump_all(void)
|
| 2233 | {
|
| 2234 | if (memblock_debug)
|
| 2235 | __memblock_dump_all();
|
| 2236 | }
|
| 2237 |
|
| 2238 | void __init memblock_allow_resize(void)
|
| 2239 | {
|
| 2240 | memblock_can_resize = 1;
|
| 2241 | }
|
| 2242 |
|
| 2243 | static int __init early_memblock(char *p)
|
| 2244 | {
|
| 2245 | if (p && strstr(p, "debug"))
|
| 2246 | memblock_debug = 1;
|
| 2247 | return 0;
|
| 2248 | }
|
| 2249 | early_param("memblock", early_memblock);
|
| 2250 |
|
| 2251 | static void __init free_memmap(unsigned long start_pfn, unsigned long end_pfn)
|
| 2252 | {
|
| 2253 | struct page *start_pg, *end_pg;
|
| 2254 | phys_addr_t pg, pgend;
|
| 2255 |
|
| 2256 | |
| 2257 | * Convert start_pfn/end_pfn to a struct page pointer.
|
| 2258 |
|
| 2259 | start_pg = pfn_to_page(start_pfn - 1) + 1;
|
| 2260 | end_pg = pfn_to_page(end_pfn - 1) + 1;
|
| 2261 |
|
| 2262 | |
| 2263 | * Convert to physical addresses, and round start upwards and end
|
| 2264 | * downwards.
|
| 2265 |
|
| 2266 | pg = PAGE_ALIGN(__pa(start_pg));
|
| 2267 | pgend = PAGE_ALIGN_DOWN(__pa(end_pg));
|
| 2268 |
|
| 2269 | |
| 2270 | * If there are free pages between these, free the section of the
|
| 2271 | * memmap array.
|
| 2272 |
|
| 2273 | if (pg < pgend)
|
| 2274 | memblock_phys_free(pg, pgend - pg);
|
| 2275 | }
|
| 2276 |
|
| 2277 | |
| 2278 | * The mem_map array can get very big. Free the unused area of the memory map.
|
| 2279 |
|
| 2280 | static void __init free_unused_memmap(void)
|
| 2281 | {
|
| 2282 | unsigned long start, end, prev_end = 0;
|
| 2283 | int i;
|
| 2284 |
|
| 2285 | if (!IS_ENABLED(CONFIG_HAVE_ARCH_PFN_VALID) ||
|
| 2286 | IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP))
|
| 2287 | return;
|
| 2288 |
|
| 2289 | |
| 2290 | * This relies on each bank being in address order.
|
| 2291 | * The banks are sorted previously in bootmem_init().
|
| 2292 |
|
| 2293 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, NULL) {
|
| 2294 | #ifdef CONFIG_SPARSEMEM
|
| 2295 | |
| 2296 | * Take care not to free memmap entries that don't exist
|
| 2297 | * due to SPARSEMEM sections which aren't present.
|
| 2298 |
|
| 2299 | start = min(start, ALIGN(prev_end, PAGES_PER_SECTION));
|
| 2300 | #endif
|
| 2301 | |
| 2302 | * Align down here since many operations in VM subsystem
|
| 2303 | * presume that there are no holes in the memory map inside
|
| 2304 | * a pageblock
|
| 2305 |
|
| 2306 | start = pageblock_start_pfn(start);
|
| 2307 |
|
| 2308 | |
| 2309 | * If we had a previous bank, and there is a space
|
| 2310 | * between the current bank and the previous, free it.
|
| 2311 |
|
| 2312 | if (prev_end && prev_end < start)
|
| 2313 | free_memmap(prev_end, start);
|
| 2314 |
|
| 2315 | |
| 2316 | * Align up here since many operations in VM subsystem
|
| 2317 | * presume that there are no holes in the memory map inside
|
| 2318 | * a pageblock
|
| 2319 |
|
| 2320 | prev_end = pageblock_align(end);
|
| 2321 | }
|
| 2322 |
|
| 2323 | #ifdef CONFIG_SPARSEMEM
|
| 2324 | if (!IS_ALIGNED(prev_end, PAGES_PER_SECTION))
|
| 2325 | free_memmap(prev_end, ALIGN(prev_end, PAGES_PER_SECTION));
|
| 2326 | #endif
|
| 2327 | }
|
| 2328 |
|
| 2329 | static void __init __free_pages_memory(unsigned long start, unsigned long end)
|
| 2330 | {
|
| 2331 | int order;
|
| 2332 |
|
| 2333 | while (start < end) {
|
| 2334 | |
| 2335 | * Free the pages in the largest chunks alignment allows.
|
| 2336 | *
|
| 2337 | * __ffs() behaviour is undefined for 0. start == 0 is
|
| 2338 | * MAX_PAGE_ORDER-aligned, set order to MAX_PAGE_ORDER for
|
| 2339 | * the case.
|
| 2340 |
|
| 2341 | if (start)
|
| 2342 | order = min_t(int, MAX_PAGE_ORDER, __ffs(start));
|
| 2343 | else
|
| 2344 | order = MAX_PAGE_ORDER;
|
| 2345 |
|
| 2346 | while (start + (1UL << order) > end)
|
| 2347 | order--;
|
| 2348 |
|
| 2349 | memblock_free_pages(start, order);
|
| 2350 |
|
| 2351 | start += (1UL << order);
|
| 2352 | }
|
| 2353 | }
|
| 2354 |
|
| 2355 | static unsigned long __init __free_memory_core(phys_addr_t start,
|
| 2356 | phys_addr_t end)
|
| 2357 | {
|
| 2358 | unsigned long start_pfn = PFN_UP(start);
|
| 2359 | unsigned long end_pfn = PFN_DOWN(end);
|
| 2360 |
|
| 2361 | if (!IS_ENABLED(CONFIG_HIGHMEM) && end_pfn > max_low_pfn)
|
| 2362 | end_pfn = max_low_pfn;
|
| 2363 |
|
| 2364 | if (start_pfn >= end_pfn)
|
| 2365 | return 0;
|
| 2366 |
|
| 2367 | __free_pages_memory(start_pfn, end_pfn);
|
| 2368 |
|
| 2369 | return end_pfn - start_pfn;
|
| 2370 | }
|
| 2371 |
|
| 2372 | |
| 2373 | * Initialised pages do not have PageReserved set. This function is called
|
| 2374 | * for each reserved range and marks the pages PageReserved.
|
| 2375 | * When deferred initialization of struct pages is enabled it also ensures
|
| 2376 | * that struct pages are properly initialised.
|
| 2377 |
|
| 2378 | static void __init memmap_init_reserved_range(phys_addr_t start,
|
| 2379 | phys_addr_t end, int nid)
|
| 2380 | {
|
| 2381 | unsigned long pfn;
|
| 2382 |
|
| 2383 | for_each_valid_pfn(pfn, PFN_DOWN(start), PFN_UP(end)) {
|
| 2384 | struct page *page = pfn_to_page(pfn);
|
| 2385 |
|
| 2386 | init_deferred_page(pfn, nid);
|
| 2387 |
|
| 2388 | |
| 2389 | * no need for atomic set_bit because the struct
|
| 2390 | * page is not visible yet so nobody should
|
| 2391 | * access it yet.
|
| 2392 |
|
| 2393 | __SetPageReserved(page);
|
| 2394 | }
|
| 2395 | }
|
| 2396 |
|
| 2397 | static void __init memmap_init_reserved_pages(void)
|
| 2398 | {
|
| 2399 | struct memblock_region *region;
|
| 2400 | phys_addr_t start, end;
|
| 2401 | int nid;
|
| 2402 | unsigned long max_reserved;
|
| 2403 |
|
| 2404 | |
| 2405 | * set nid on all reserved pages and also treat struct
|
| 2406 | * pages for the NOMAP regions as PageReserved
|
| 2407 |
|
| 2408 | repeat:
|
| 2409 | max_reserved = memblock.reserved.max;
|
| 2410 | for_each_mem_region(region) {
|
| 2411 | nid = memblock_get_region_node(region);
|
| 2412 | start = region->base;
|
| 2413 | end = start + region->size;
|
| 2414 |
|
| 2415 | if (memblock_is_nomap(region))
|
| 2416 | memmap_init_reserved_range(start, end, nid);
|
| 2417 |
|
| 2418 | memblock_set_node(start, region->size, &memblock.reserved, nid);
|
| 2419 | }
|
| 2420 | |
| 2421 | * 'max' is changed means memblock.reserved has been doubled its
|
| 2422 | * array, which may result a new reserved region before current
|
| 2423 | * 'start'. Now we should repeat the procedure to set its node id.
|
| 2424 |
|
| 2425 | if (max_reserved != memblock.reserved.max)
|
| 2426 | goto repeat;
|
| 2427 |
|
| 2428 | |
| 2429 | * initialize struct pages for reserved regions that don't have
|
| 2430 | * the MEMBLOCK_RSRV_NOINIT flag set
|
| 2431 |
|
| 2432 | for_each_reserved_mem_region(region) {
|
| 2433 | if (!memblock_is_reserved_noinit(region)) {
|
| 2434 | nid = memblock_get_region_node(region);
|
| 2435 | start = region->base;
|
| 2436 | end = start + region->size;
|
| 2437 |
|
| 2438 | if (!numa_valid_node(nid))
|
| 2439 | nid = early_pfn_to_nid(PFN_DOWN(start));
|
| 2440 |
|
| 2441 | memmap_init_reserved_range(start, end, nid);
|
| 2442 | }
|
| 2443 | }
|
| 2444 | }
|
| 2445 |
|
| 2446 | static unsigned long __init free_low_memory_core_early(void)
|
| 2447 | {
|
| 2448 | unsigned long count = 0;
|
| 2449 | phys_addr_t start, end;
|
| 2450 | u64 i;
|
| 2451 |
|
| 2452 | memblock_clear_hotplug(0, -1);
|
| 2453 |
|
| 2454 | memmap_init_reserved_pages();
|
| 2455 |
|
| 2456 | |
| 2457 | * We need to use NUMA_NO_NODE instead of NODE_DATA(0)->node_id
|
| 2458 | * because in some case like Node0 doesn't have RAM installed
|
| 2459 | * low ram will be on Node1
|
| 2460 |
|
| 2461 | for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE, &start, &end,
|
| 2462 | NULL)
|
| 2463 | count += __free_memory_core(start, end);
|
| 2464 |
|
| 2465 | return count;
|
| 2466 | }
|
| 2467 |
|
| 2468 | static int reset_managed_pages_done __initdata;
|
| 2469 |
|
| 2470 | static void __init reset_node_managed_pages(pg_data_t *pgdat)
|
| 2471 | {
|
| 2472 | struct zone *z;
|
| 2473 |
|
| 2474 | for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
|
| 2475 | atomic_long_set(&z->managed_pages, 0);
|
| 2476 | }
|
| 2477 |
|
| 2478 | void __init reset_all_zones_managed_pages(void)
|
| 2479 | {
|
| 2480 | struct pglist_data *pgdat;
|
| 2481 |
|
| 2482 | if (reset_managed_pages_done)
|
| 2483 | return;
|
| 2484 |
|
| 2485 | for_each_online_pgdat(pgdat)
|
| 2486 | reset_node_managed_pages(pgdat);
|
| 2487 |
|
| 2488 | reset_managed_pages_done = 1;
|
| 2489 | }
|
| 2490 |
|
| 2491 | |
| 2492 | * memblock_free_all - release free pages to the buddy allocator
|
| 2493 |
|
| 2494 | void __init memblock_free_all(void)
|
| 2495 | {
|
| 2496 | unsigned long pages;
|
| 2497 |
|
| 2498 | free_unused_memmap();
|
| 2499 | reset_all_zones_managed_pages();
|
| 2500 |
|
| 2501 | memblock_clear_kho_scratch_only();
|
| 2502 | pages = free_low_memory_core_early();
|
| 2503 | totalram_pages_add(pages);
|
| 2504 | }
|
| 2505 |
|
| 2506 |
|
| 2507 | #define RESERVE_MEM_MAX_ENTRIES 8
|
| 2508 | #define RESERVE_MEM_NAME_SIZE 16
|
| 2509 | struct reserve_mem_table {
|
| 2510 | char name[RESERVE_MEM_NAME_SIZE];
|
| 2511 | phys_addr_t start;
|
| 2512 | phys_addr_t size;
|
| 2513 | };
|
| 2514 | static struct reserve_mem_table reserved_mem_table[RESERVE_MEM_MAX_ENTRIES];
|
| 2515 | static int reserved_mem_count;
|
| 2516 | static DEFINE_MUTEX(reserve_mem_lock);
|
| 2517 |
|
| 2518 |
|
| 2519 | static void __init reserved_mem_add(phys_addr_t start, phys_addr_t size,
|
| 2520 | const char *name)
|
| 2521 | {
|
| 2522 | struct reserve_mem_table *map;
|
| 2523 |
|
| 2524 | map = &reserved_mem_table[reserved_mem_count++];
|
| 2525 | map->start = start;
|
| 2526 | map->size = size;
|
| 2527 | strscpy(map->name, name);
|
| 2528 | }
|
| 2529 |
|
| 2530 | static struct reserve_mem_table *reserve_mem_find_by_name_nolock(const char *name)
|
| 2531 | {
|
| 2532 | struct reserve_mem_table *map;
|
| 2533 | int i;
|
| 2534 |
|
| 2535 | for (i = 0; i < reserved_mem_count; i++) {
|
| 2536 | map = &reserved_mem_table[i];
|
| 2537 | if (!map->size)
|
| 2538 | continue;
|
| 2539 | if (strcmp(name, map->name) == 0)
|
| 2540 | return map;
|
| 2541 | }
|
| 2542 | return NULL;
|
| 2543 | }
|
| 2544 |
|
| 2545 | |
| 2546 | * reserve_mem_find_by_name - Find reserved memory region with a given name
|
| 2547 | * @name: The name that is attached to a reserved memory region
|
| 2548 | * @start: If found, holds the start address
|
| 2549 | * @size: If found, holds the size of the address.
|
| 2550 | *
|
| 2551 | * @start and @size are only updated if @name is found.
|
| 2552 | *
|
| 2553 | * Returns: 1 if found or 0 if not found.
|
| 2554 |
|
| 2555 | int reserve_mem_find_by_name(const char *name, phys_addr_t *start, phys_addr_t *size)
|
| 2556 | {
|
| 2557 | struct reserve_mem_table *map;
|
| 2558 |
|
| 2559 | guard(mutex)(&reserve_mem_lock);
|
| 2560 | map = reserve_mem_find_by_name_nolock(name);
|
| 2561 | if (!map)
|
| 2562 | return 0;
|
| 2563 |
|
| 2564 | *start = map->start;
|
| 2565 | *size = map->size;
|
| 2566 | return 1;
|
| 2567 | }
|
| 2568 | EXPORT_SYMBOL_GPL(reserve_mem_find_by_name);
|
| 2569 |
|
| 2570 | |
| 2571 | * reserve_mem_release_by_name - Release reserved memory region with a given name
|
| 2572 | * @name: The name that is attached to a reserved memory region
|
| 2573 | *
|
| 2574 | * Forcibly release the pages in the reserved memory region so that those memory
|
| 2575 | * can be used as free memory. After released the reserved region size becomes 0.
|
| 2576 | *
|
| 2577 | * Returns: 1 if released or 0 if not found.
|
| 2578 |
|
| 2579 | int reserve_mem_release_by_name(const char *name)
|
| 2580 | {
|
| 2581 | char buf[RESERVE_MEM_NAME_SIZE + 12];
|
| 2582 | struct reserve_mem_table *map;
|
| 2583 | void *start, *end;
|
| 2584 |
|
| 2585 | guard(mutex)(&reserve_mem_lock);
|
| 2586 | map = reserve_mem_find_by_name_nolock(name);
|
| 2587 | if (!map)
|
| 2588 | return 0;
|
| 2589 |
|
| 2590 | start = phys_to_virt(map->start);
|
| 2591 | end = start + map->size;
|
| 2592 | snprintf(buf, sizeof(buf), "reserve_mem:%s", name);
|
| 2593 | free_reserved_area(start, end, 0, buf);
|
| 2594 | map->size = 0;
|
| 2595 |
|
| 2596 | return 1;
|
| 2597 | }
|
| 2598 |
|
| 2599 | #ifdef CONFIG_MEMBLOCK_KHO_SCRATCH
|
| 2600 | __init void memblock_set_kho_scratch_only(void)
|
| 2601 | {
|
| 2602 | kho_scratch_only = true;
|
| 2603 | }
|
| 2604 |
|
| 2605 | __init void memblock_clear_kho_scratch_only(void)
|
| 2606 | {
|
| 2607 | kho_scratch_only = false;
|
| 2608 | }
|
| 2609 | #endif
|
| 2610 |
|
| 2611 | #ifdef CONFIG_KEXEC_HANDOVER
|
| 2612 |
|
| 2613 | static int __init reserved_mem_preserve(void)
|
| 2614 | {
|
| 2615 | unsigned int nr_preserved = 0;
|
| 2616 | int err;
|
| 2617 |
|
| 2618 | for (unsigned int i = 0; i < reserved_mem_count; i++, nr_preserved++) {
|
| 2619 | struct reserve_mem_table *map = &reserved_mem_table[i];
|
| 2620 | struct page *page = phys_to_page(map->start);
|
| 2621 | unsigned int nr_pages = map->size >> PAGE_SHIFT;
|
| 2622 |
|
| 2623 | err = kho_preserve_pages(page, nr_pages);
|
| 2624 | if (err)
|
| 2625 | goto err_unpreserve;
|
| 2626 | }
|
| 2627 |
|
| 2628 | return 0;
|
| 2629 |
|
| 2630 | err_unpreserve:
|
| 2631 | for (unsigned int i = 0; i < nr_preserved; i++) {
|
| 2632 | struct reserve_mem_table *map = &reserved_mem_table[i];
|
| 2633 | struct page *page = phys_to_page(map->start);
|
| 2634 | unsigned int nr_pages = map->size >> PAGE_SHIFT;
|
| 2635 |
|
| 2636 | kho_unpreserve_pages(page, nr_pages);
|
| 2637 | }
|
| 2638 |
|
| 2639 | return err;
|
| 2640 | }
|
| 2641 |
|
| 2642 | static int __init prepare_kho_fdt(void)
|
| 2643 | {
|
| 2644 | struct page *fdt_page;
|
| 2645 | void *fdt;
|
| 2646 | int err;
|
| 2647 |
|
| 2648 | fdt_page = alloc_page(GFP_KERNEL);
|
| 2649 | if (!fdt_page) {
|
| 2650 | err = -ENOMEM;
|
| 2651 | goto err_report;
|
| 2652 | }
|
| 2653 |
|
| 2654 | fdt = page_to_virt(fdt_page);
|
| 2655 | err = kho_preserve_pages(fdt_page, 1);
|
| 2656 | if (err)
|
| 2657 | goto err_free_fdt;
|
| 2658 |
|
| 2659 | err |= fdt_create(fdt, PAGE_SIZE);
|
| 2660 | err |= fdt_finish_reservemap(fdt);
|
| 2661 | err |= fdt_begin_node(fdt, "");
|
| 2662 | err |= fdt_property_string(fdt, "compatible", MEMBLOCK_KHO_NODE_COMPATIBLE);
|
| 2663 |
|
| 2664 | for (unsigned int i = 0; !err && i < reserved_mem_count; i++) {
|
| 2665 | struct reserve_mem_table *map = &reserved_mem_table[i];
|
| 2666 |
|
| 2667 | err |= fdt_begin_node(fdt, map->name);
|
| 2668 | err |= fdt_property_string(fdt, "compatible", RESERVE_MEM_KHO_NODE_COMPATIBLE);
|
| 2669 | err |= fdt_property(fdt, "start", &map->start, sizeof(map->start));
|
| 2670 | err |= fdt_property(fdt, "size", &map->size, sizeof(map->size));
|
| 2671 | err |= fdt_end_node(fdt);
|
| 2672 | }
|
| 2673 | err |= fdt_end_node(fdt);
|
| 2674 | err |= fdt_finish(fdt);
|
| 2675 |
|
| 2676 | if (err)
|
| 2677 | goto err_unpreserve_fdt;
|
| 2678 |
|
| 2679 | err = kho_add_subtree(MEMBLOCK_KHO_FDT, fdt, fdt_totalsize(fdt));
|
| 2680 | if (err)
|
| 2681 | goto err_unpreserve_fdt;
|
| 2682 |
|
| 2683 | err = reserved_mem_preserve();
|
| 2684 | if (err)
|
| 2685 | goto err_remove_subtree;
|
| 2686 |
|
| 2687 | return 0;
|
| 2688 |
|
| 2689 | err_remove_subtree:
|
| 2690 | kho_remove_subtree(fdt);
|
| 2691 | err_unpreserve_fdt:
|
| 2692 | kho_unpreserve_pages(fdt_page, 1);
|
| 2693 | err_free_fdt:
|
| 2694 | put_page(fdt_page);
|
| 2695 | err_report:
|
| 2696 | pr_err("failed to prepare memblock FDT for KHO: %d\n", err);
|
| 2697 |
|
| 2698 | return err;
|
| 2699 | }
|
| 2700 |
|
| 2701 | static int __init reserve_mem_init(void)
|
| 2702 | {
|
| 2703 | int err;
|
| 2704 |
|
| 2705 | if (!kho_is_enabled() || !reserved_mem_count)
|
| 2706 | return 0;
|
| 2707 |
|
| 2708 | err = prepare_kho_fdt();
|
| 2709 | if (err)
|
| 2710 | return err;
|
| 2711 | return err;
|
| 2712 | }
|
| 2713 | late_initcall(reserve_mem_init);
|
| 2714 |
|
| 2715 | static void *__init reserve_mem_kho_retrieve_fdt(void)
|
| 2716 | {
|
| 2717 | phys_addr_t fdt_phys;
|
| 2718 | static void *fdt;
|
| 2719 | int err;
|
| 2720 |
|
| 2721 | if (fdt)
|
| 2722 | return fdt;
|
| 2723 |
|
| 2724 | err = kho_retrieve_subtree(MEMBLOCK_KHO_FDT, &fdt_phys, NULL);
|
| 2725 | if (err) {
|
| 2726 | if (err != -ENOENT)
|
| 2727 | pr_warn("failed to retrieve FDT '%s' from KHO: %d\n",
|
| 2728 | MEMBLOCK_KHO_FDT, err);
|
| 2729 | return NULL;
|
| 2730 | }
|
| 2731 |
|
| 2732 | fdt = phys_to_virt(fdt_phys);
|
| 2733 |
|
| 2734 | err = fdt_node_check_compatible(fdt, 0, MEMBLOCK_KHO_NODE_COMPATIBLE);
|
| 2735 | if (err) {
|
| 2736 | pr_warn("FDT '%s' is incompatible with '%s': %d\n",
|
| 2737 | MEMBLOCK_KHO_FDT, MEMBLOCK_KHO_NODE_COMPATIBLE, err);
|
| 2738 | fdt = NULL;
|
| 2739 | }
|
| 2740 |
|
| 2741 | return fdt;
|
| 2742 | }
|
| 2743 |
|
| 2744 | static bool __init reserve_mem_kho_revive(const char *name, phys_addr_t size,
|
| 2745 | phys_addr_t align)
|
| 2746 | {
|
| 2747 | int err, len_start, len_size, offset;
|
| 2748 | const phys_addr_t *p_start, *p_size;
|
| 2749 | const void *fdt;
|
| 2750 |
|
| 2751 | fdt = reserve_mem_kho_retrieve_fdt();
|
| 2752 | if (!fdt)
|
| 2753 | return false;
|
| 2754 |
|
| 2755 | offset = fdt_subnode_offset(fdt, 0, name);
|
| 2756 | if (offset < 0) {
|
| 2757 | pr_warn("FDT '%s' has no child '%s': %d\n",
|
| 2758 | MEMBLOCK_KHO_FDT, name, offset);
|
| 2759 | return false;
|
| 2760 | }
|
| 2761 | err = fdt_node_check_compatible(fdt, offset, RESERVE_MEM_KHO_NODE_COMPATIBLE);
|
| 2762 | if (err) {
|
| 2763 | pr_warn("Node '%s' is incompatible with '%s': %d\n",
|
| 2764 | name, RESERVE_MEM_KHO_NODE_COMPATIBLE, err);
|
| 2765 | return false;
|
| 2766 | }
|
| 2767 |
|
| 2768 | p_start = fdt_getprop(fdt, offset, "start", &len_start);
|
| 2769 | p_size = fdt_getprop(fdt, offset, "size", &len_size);
|
| 2770 | if (!p_start || len_start != sizeof(*p_start) || !p_size ||
|
| 2771 | len_size != sizeof(*p_size)) {
|
| 2772 | return false;
|
| 2773 | }
|
| 2774 |
|
| 2775 | if (*p_start & (align - 1)) {
|
| 2776 | pr_warn("KHO reserve-mem '%s' has wrong alignment (0x%lx, 0x%lx)\n",
|
| 2777 | name, (long)align, (long)*p_start);
|
| 2778 | return false;
|
| 2779 | }
|
| 2780 |
|
| 2781 | if (*p_size != size) {
|
| 2782 | pr_warn("KHO reserve-mem '%s' has wrong size (0x%lx != 0x%lx)\n",
|
| 2783 | name, (long)*p_size, (long)size);
|
| 2784 | return false;
|
| 2785 | }
|
| 2786 |
|
| 2787 | reserved_mem_add(*p_start, size, name);
|
| 2788 | pr_info("Revived memory reservation '%s' from KHO\n", name);
|
| 2789 |
|
| 2790 | return true;
|
| 2791 | }
|
| 2792 | #else
|
| 2793 | static bool __init reserve_mem_kho_revive(const char *name, phys_addr_t size,
|
| 2794 | phys_addr_t align)
|
| 2795 | {
|
| 2796 | return false;
|
| 2797 | }
|
| 2798 | #endif
|
| 2799 |
|
| 2800 | |
| 2801 | * Parse reserve_mem=nn:align:name
|
| 2802 |
|
| 2803 | static int __init reserve_mem(char *p)
|
| 2804 | {
|
| 2805 | phys_addr_t start, size, align, tmp;
|
| 2806 | char *name;
|
| 2807 | char *oldp;
|
| 2808 | int len;
|
| 2809 |
|
| 2810 | if (!p)
|
| 2811 | goto err_param;
|
| 2812 |
|
| 2813 |
|
| 2814 | if (reserved_mem_count >= RESERVE_MEM_MAX_ENTRIES) {
|
| 2815 | pr_err("reserve_mem: no more room for reserved memory\n");
|
| 2816 | return -EBUSY;
|
| 2817 | }
|
| 2818 |
|
| 2819 | oldp = p;
|
| 2820 | size = memparse(p, &p);
|
| 2821 | if (!size || p == oldp)
|
| 2822 | goto err_param;
|
| 2823 |
|
| 2824 | if (*p != ':')
|
| 2825 | goto err_param;
|
| 2826 |
|
| 2827 | align = memparse(p+1, &p);
|
| 2828 | if (*p != ':')
|
| 2829 | goto err_param;
|
| 2830 |
|
| 2831 | |
| 2832 | * memblock_phys_alloc() doesn't like a zero size align,
|
| 2833 | * but it is OK for this command to have it.
|
| 2834 |
|
| 2835 | if (align < SMP_CACHE_BYTES)
|
| 2836 | align = SMP_CACHE_BYTES;
|
| 2837 |
|
| 2838 | name = p + 1;
|
| 2839 | len = strlen(name);
|
| 2840 |
|
| 2841 |
|
| 2842 | if (!len || len >= RESERVE_MEM_NAME_SIZE)
|
| 2843 | goto err_param;
|
| 2844 |
|
| 2845 |
|
| 2846 | for (p = name; *p; p++) {
|
| 2847 | if (!isspace(*p))
|
| 2848 | break;
|
| 2849 | }
|
| 2850 | if (!*p)
|
| 2851 | goto err_param;
|
| 2852 |
|
| 2853 |
|
| 2854 | if (reserve_mem_find_by_name(name, &start, &tmp)) {
|
| 2855 | pr_err("reserve_mem: name \"%s\" was already used\n", name);
|
| 2856 | return -EBUSY;
|
| 2857 | }
|
| 2858 |
|
| 2859 |
|
| 2860 | if (reserve_mem_kho_revive(name, size, align))
|
| 2861 | return 1;
|
| 2862 |
|
| 2863 |
|
| 2864 | start = memblock_phys_alloc(size, align);
|
| 2865 | if (!start) {
|
| 2866 | pr_err("reserve_mem: memblock allocation failed\n");
|
| 2867 | return -ENOMEM;
|
| 2868 | }
|
| 2869 |
|
| 2870 | reserved_mem_add(start, size, name);
|
| 2871 |
|
| 2872 | return 1;
|
| 2873 | err_param:
|
| 2874 | pr_err("reserve_mem: empty or malformed parameter\n");
|
| 2875 | return -EINVAL;
|
| 2876 | }
|
| 2877 | __setup("reserve_mem=", reserve_mem);
|
| 2878 |
|
| 2879 | #ifdef CONFIG_DEBUG_FS
|
| 2880 | #ifdef CONFIG_ARCH_KEEP_MEMBLOCK
|
| 2881 | static const char * const flagname[] = {
|
| 2882 | [ilog2(MEMBLOCK_HOTPLUG)] = "HOTPLUG",
|
| 2883 | [ilog2(MEMBLOCK_MIRROR)] = "MIRROR",
|
| 2884 | [ilog2(MEMBLOCK_NOMAP)] = "NOMAP",
|
| 2885 | [ilog2(MEMBLOCK_DRIVER_MANAGED)] = "DRV_MNG",
|
| 2886 | [ilog2(MEMBLOCK_RSRV_NOINIT)] = "RSV_NIT",
|
| 2887 | [ilog2(MEMBLOCK_RSRV_KERN)] = "RSV_KERN",
|
| 2888 | [ilog2(MEMBLOCK_KHO_SCRATCH)] = "KHO_SCRATCH",
|
| 2889 | [ilog2(MEMBLOCK_RSRV_HUGETLB)] = "RSV_HUGETLB",
|
| 2890 | };
|
| 2891 |
|
| 2892 | static int memblock_debug_show(struct seq_file *m, void *private)
|
| 2893 | {
|
| 2894 | struct memblock_type *type = m->private;
|
| 2895 | struct memblock_region *reg;
|
| 2896 | int i, j, nid;
|
| 2897 | unsigned int count = ARRAY_SIZE(flagname);
|
| 2898 | phys_addr_t end;
|
| 2899 |
|
| 2900 | for (i = 0; i < type->cnt; i++) {
|
| 2901 | reg = &type->regions[i];
|
| 2902 | end = reg->base + reg->size - 1;
|
| 2903 | nid = memblock_get_region_node(reg);
|
| 2904 |
|
| 2905 | seq_printf(m, "%4d: ", i);
|
| 2906 | seq_printf(m, "%pa..%pa ", ®->base, &end);
|
| 2907 | if (numa_valid_node(nid))
|
| 2908 | seq_printf(m, "%4d ", nid);
|
| 2909 | else
|
| 2910 | seq_printf(m, "%4c ", 'x');
|
| 2911 | if (reg->flags) {
|
| 2912 | unsigned int flags = reg->flags;
|
| 2913 | bool first = true;
|
| 2914 |
|
| 2915 | for (j = 0; flags; j++, flags >>= 1) {
|
| 2916 | if (!(flags & 1))
|
| 2917 | continue;
|
| 2918 | if (!first)
|
| 2919 | seq_putc(m, '|');
|
| 2920 | seq_puts(m, j < count ? flagname[j] : "UNKNOWN");
|
| 2921 | first = false;
|
| 2922 | }
|
| 2923 | seq_putc(m, '\n');
|
| 2924 | } else {
|
| 2925 | seq_printf(m, "%s\n", "NONE");
|
| 2926 | }
|
| 2927 | }
|
| 2928 | return 0;
|
| 2929 | }
|
| 2930 | DEFINE_SHOW_ATTRIBUTE(memblock_debug);
|
| 2931 |
|
| 2932 | static inline void memblock_debugfs_expose_arrays(struct dentry *root)
|
| 2933 | {
|
| 2934 | debugfs_create_file("memory", 0444, root,
|
| 2935 | &memblock.memory, &memblock_debug_fops);
|
| 2936 | debugfs_create_file("reserved", 0444, root,
|
| 2937 | &memblock.reserved, &memblock_debug_fops);
|
| 2938 | #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
|
| 2939 | debugfs_create_file("physmem", 0444, root, &physmem,
|
| 2940 | &memblock_debug_fops);
|
| 2941 | #endif
|
| 2942 | }
|
| 2943 |
|
| 2944 | #else
|
| 2945 |
|
| 2946 | static inline void memblock_debugfs_expose_arrays(struct dentry *root) { }
|
| 2947 |
|
| 2948 | #endif
|
| 2949 |
|
| 2950 | static int memblock_reserve_mem_show(struct seq_file *m, void *private)
|
| 2951 | {
|
| 2952 | struct reserve_mem_table *map;
|
| 2953 | char txtsz[16];
|
| 2954 |
|
| 2955 | guard(mutex)(&reserve_mem_lock);
|
| 2956 | for (int i = 0; i < reserved_mem_count; i++) {
|
| 2957 | map = &reserved_mem_table[i];
|
| 2958 | if (!map->size)
|
| 2959 | continue;
|
| 2960 |
|
| 2961 | memset(txtsz, 0, sizeof(txtsz));
|
| 2962 | string_get_size(map->size, 1, STRING_UNITS_2, txtsz, sizeof(txtsz));
|
| 2963 | seq_printf(m, "%s\t\t(%s)\n", map->name, txtsz);
|
| 2964 | }
|
| 2965 |
|
| 2966 | return 0;
|
| 2967 | }
|
| 2968 | DEFINE_SHOW_ATTRIBUTE(memblock_reserve_mem);
|
| 2969 |
|
| 2970 | static int __init memblock_init_debugfs(void)
|
| 2971 | {
|
| 2972 | struct dentry *root;
|
| 2973 |
|
| 2974 | if (!IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK) && !reserved_mem_count)
|
| 2975 | return 0;
|
| 2976 |
|
| 2977 | root = debugfs_create_dir("memblock", NULL);
|
| 2978 |
|
| 2979 | if (reserved_mem_count)
|
| 2980 | debugfs_create_file("reserve_mem_param", 0444, root, NULL,
|
| 2981 | &memblock_reserve_mem_fops);
|
| 2982 |
|
| 2983 | memblock_debugfs_expose_arrays(root);
|
| 2984 | return 0;
|
| 2985 | }
|
| 2986 | __initcall(memblock_init_debugfs);
|
| 2987 |
|
| 2988 | #endif
|