1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Based on arch/arm/mm/init.c 4 * 5 * Copyright (C) 1995-2005 Russell King 6 * Copyright (C) 2012 ARM Ltd. 7 */ 8 9 #include <linux/kernel.h> 10 #include <linux/export.h> 11 #include <linux/errno.h> 12 #include <linux/swap.h> 13 #include <linux/init.h> 14 #include <linux/cache.h> 15 #include <linux/mman.h> 16 #include <linux/nodemask.h> 17 #include <linux/initrd.h> 18 #include <linux/gfp.h> 19 #include <linux/memblock.h> 20 #include <linux/sort.h> 21 #include <linux/of.h> 22 #include <linux/of_fdt.h> 23 #include <linux/dma-direct.h> 24 #include <linux/dma-map-ops.h> 25 #include <linux/efi.h> 26 #include <linux/swiotlb.h> 27 #include <linux/vmalloc.h> 28 #include <linux/mm.h> 29 #include <linux/kexec.h> 30 #include <linux/crash_dump.h> 31 #include <linux/hugetlb.h> 32 #include <linux/acpi_iort.h> 33 34 #include <asm/boot.h> 35 #include <asm/fixmap.h> 36 #include <asm/kasan.h> 37 #include <asm/kernel-pgtable.h> 38 #include <asm/memory.h> 39 #include <asm/numa.h> 40 #include <asm/sections.h> 41 #include <asm/setup.h> 42 #include <linux/sizes.h> 43 #include <asm/tlb.h> 44 #include <asm/alternative.h> 45 46 /* 47 * We need to be able to catch inadvertent references to memstart_addr 48 * that occur (potentially in generic code) before arm64_memblock_init() 49 * executes, which assigns it its actual value. So use a default value 50 * that cannot be mistaken for a real physical address. 51 */ 52 s64 memstart_addr __ro_after_init = -1; 53 EXPORT_SYMBOL(memstart_addr); 54 55 /* 56 * We create both ZONE_DMA and ZONE_DMA32. ZONE_DMA covers the first 1G of 57 * memory as some devices, namely the Raspberry Pi 4, have peripherals with 58 * this limited view of the memory. ZONE_DMA32 will cover the rest of the 32 59 * bit addressable memory area. 60 */ 61 phys_addr_t arm64_dma_phys_limit __ro_after_init; 62 static phys_addr_t arm64_dma32_phys_limit __ro_after_init; 63 64 #ifdef CONFIG_KEXEC_CORE 65 /* 66 * reserve_crashkernel() - reserves memory for crash kernel 67 * 68 * This function reserves memory area given in "crashkernel=" kernel command 69 * line parameter. The memory reserved is used by dump capture kernel when 70 * primary kernel is crashing. 71 */ 72 static void __init reserve_crashkernel(void) 73 { 74 unsigned long long crash_base, crash_size; 75 int ret; 76 77 ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(), 78 &crash_size, &crash_base); 79 /* no crashkernel= or invalid value specified */ 80 if (ret || !crash_size) 81 return; 82 83 crash_size = PAGE_ALIGN(crash_size); 84 85 if (crash_base == 0) { 86 /* Current arm64 boot protocol requires 2MB alignment */ 87 crash_base = memblock_find_in_range(0, arm64_dma32_phys_limit, 88 crash_size, SZ_2M); 89 if (crash_base == 0) { 90 pr_warn("cannot allocate crashkernel (size:0x%llx)\n", 91 crash_size); 92 return; 93 } 94 } else { 95 /* User specifies base address explicitly. */ 96 if (!memblock_is_region_memory(crash_base, crash_size)) { 97 pr_warn("cannot reserve crashkernel: region is not memory\n"); 98 return; 99 } 100 101 if (memblock_is_region_reserved(crash_base, crash_size)) { 102 pr_warn("cannot reserve crashkernel: region overlaps reserved memory\n"); 103 return; 104 } 105 106 if (!IS_ALIGNED(crash_base, SZ_2M)) { 107 pr_warn("cannot reserve crashkernel: base address is not 2MB aligned\n"); 108 return; 109 } 110 } 111 memblock_reserve(crash_base, crash_size); 112 113 pr_info("crashkernel reserved: 0x%016llx - 0x%016llx (%lld MB)\n", 114 crash_base, crash_base + crash_size, crash_size >> 20); 115 116 crashk_res.start = crash_base; 117 crashk_res.end = crash_base + crash_size - 1; 118 } 119 #else 120 static void __init reserve_crashkernel(void) 121 { 122 } 123 #endif /* CONFIG_KEXEC_CORE */ 124 125 #ifdef CONFIG_CRASH_DUMP 126 static int __init early_init_dt_scan_elfcorehdr(unsigned long node, 127 const char *uname, int depth, void *data) 128 { 129 const __be32 *reg; 130 int len; 131 132 if (depth != 1 || strcmp(uname, "chosen") != 0) 133 return 0; 134 135 reg = of_get_flat_dt_prop(node, "linux,elfcorehdr", &len); 136 if (!reg || (len < (dt_root_addr_cells + dt_root_size_cells))) 137 return 1; 138 139 elfcorehdr_addr = dt_mem_next_cell(dt_root_addr_cells, ®); 140 elfcorehdr_size = dt_mem_next_cell(dt_root_size_cells, ®); 141 142 return 1; 143 } 144 145 /* 146 * reserve_elfcorehdr() - reserves memory for elf core header 147 * 148 * This function reserves the memory occupied by an elf core header 149 * described in the device tree. This region contains all the 150 * information about primary kernel's core image and is used by a dump 151 * capture kernel to access the system memory on primary kernel. 152 */ 153 static void __init reserve_elfcorehdr(void) 154 { 155 of_scan_flat_dt(early_init_dt_scan_elfcorehdr, NULL); 156 157 if (!elfcorehdr_size) 158 return; 159 160 if (memblock_is_region_reserved(elfcorehdr_addr, elfcorehdr_size)) { 161 pr_warn("elfcorehdr is overlapped\n"); 162 return; 163 } 164 165 memblock_reserve(elfcorehdr_addr, elfcorehdr_size); 166 167 pr_info("Reserving %lldKB of memory at 0x%llx for elfcorehdr\n", 168 elfcorehdr_size >> 10, elfcorehdr_addr); 169 } 170 #else 171 static void __init reserve_elfcorehdr(void) 172 { 173 } 174 #endif /* CONFIG_CRASH_DUMP */ 175 176 /* 177 * Return the maximum physical address for a zone accessible by the given bits 178 * limit. If DRAM starts above 32-bit, expand the zone to the maximum 179 * available memory, otherwise cap it at 32-bit. 180 */ 181 static phys_addr_t __init max_zone_phys(unsigned int zone_bits) 182 { 183 phys_addr_t zone_mask = DMA_BIT_MASK(zone_bits); 184 phys_addr_t phys_start = memblock_start_of_DRAM(); 185 186 if (phys_start > U32_MAX) 187 zone_mask = PHYS_ADDR_MAX; 188 else if (phys_start > zone_mask) 189 zone_mask = U32_MAX; 190 191 return min(zone_mask, memblock_end_of_DRAM() - 1) + 1; 192 } 193 194 static void __init zone_sizes_init(unsigned long min, unsigned long max) 195 { 196 unsigned long max_zone_pfns[MAX_NR_ZONES] = {0}; 197 unsigned int __maybe_unused acpi_zone_dma_bits; 198 unsigned int __maybe_unused dt_zone_dma_bits; 199 200 #ifdef CONFIG_ZONE_DMA 201 acpi_zone_dma_bits = fls64(acpi_iort_dma_get_max_cpu_address()); 202 dt_zone_dma_bits = fls64(of_dma_get_max_cpu_address(NULL)); 203 zone_dma_bits = min3(32U, dt_zone_dma_bits, acpi_zone_dma_bits); 204 arm64_dma_phys_limit = max_zone_phys(zone_dma_bits); 205 max_zone_pfns[ZONE_DMA] = PFN_DOWN(arm64_dma_phys_limit); 206 #endif 207 #ifdef CONFIG_ZONE_DMA32 208 max_zone_pfns[ZONE_DMA32] = PFN_DOWN(arm64_dma32_phys_limit); 209 #endif 210 max_zone_pfns[ZONE_NORMAL] = max; 211 212 free_area_init(max_zone_pfns); 213 } 214 215 int pfn_valid(unsigned long pfn) 216 { 217 phys_addr_t addr = pfn << PAGE_SHIFT; 218 219 if ((addr >> PAGE_SHIFT) != pfn) 220 return 0; 221 222 #ifdef CONFIG_SPARSEMEM 223 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS) 224 return 0; 225 226 if (!valid_section(__pfn_to_section(pfn))) 227 return 0; 228 #endif 229 return memblock_is_map_memory(addr); 230 } 231 EXPORT_SYMBOL(pfn_valid); 232 233 static phys_addr_t memory_limit = PHYS_ADDR_MAX; 234 235 /* 236 * Limit the memory size that was specified via FDT. 237 */ 238 static int __init early_mem(char *p) 239 { 240 if (!p) 241 return 1; 242 243 memory_limit = memparse(p, &p) & PAGE_MASK; 244 pr_notice("Memory limited to %lldMB\n", memory_limit >> 20); 245 246 return 0; 247 } 248 early_param("mem", early_mem); 249 250 static int __init early_init_dt_scan_usablemem(unsigned long node, 251 const char *uname, int depth, void *data) 252 { 253 struct memblock_region *usablemem = data; 254 const __be32 *reg; 255 int len; 256 257 if (depth != 1 || strcmp(uname, "chosen") != 0) 258 return 0; 259 260 reg = of_get_flat_dt_prop(node, "linux,usable-memory-range", &len); 261 if (!reg || (len < (dt_root_addr_cells + dt_root_size_cells))) 262 return 1; 263 264 usablemem->base = dt_mem_next_cell(dt_root_addr_cells, ®); 265 usablemem->size = dt_mem_next_cell(dt_root_size_cells, ®); 266 267 return 1; 268 } 269 270 static void __init fdt_enforce_memory_region(void) 271 { 272 struct memblock_region reg = { 273 .size = 0, 274 }; 275 276 of_scan_flat_dt(early_init_dt_scan_usablemem, ®); 277 278 if (reg.size) 279 memblock_cap_memory_range(reg.base, reg.size); 280 } 281 282 void __init arm64_memblock_init(void) 283 { 284 const s64 linear_region_size = PAGE_END - _PAGE_OFFSET(vabits_actual); 285 286 /* Handle linux,usable-memory-range property */ 287 fdt_enforce_memory_region(); 288 289 /* Remove memory above our supported physical address size */ 290 memblock_remove(1ULL << PHYS_MASK_SHIFT, ULLONG_MAX); 291 292 /* 293 * Select a suitable value for the base of physical memory. 294 */ 295 memstart_addr = round_down(memblock_start_of_DRAM(), 296 ARM64_MEMSTART_ALIGN); 297 298 if ((memblock_end_of_DRAM() - memstart_addr) > linear_region_size) 299 pr_warn("Memory doesn't fit in the linear mapping, VA_BITS too small\n"); 300 301 /* 302 * Remove the memory that we will not be able to cover with the 303 * linear mapping. Take care not to clip the kernel which may be 304 * high in memory. 305 */ 306 memblock_remove(max_t(u64, memstart_addr + linear_region_size, 307 __pa_symbol(_end)), ULLONG_MAX); 308 if (memstart_addr + linear_region_size < memblock_end_of_DRAM()) { 309 /* ensure that memstart_addr remains sufficiently aligned */ 310 memstart_addr = round_up(memblock_end_of_DRAM() - linear_region_size, 311 ARM64_MEMSTART_ALIGN); 312 memblock_remove(0, memstart_addr); 313 } 314 315 /* 316 * If we are running with a 52-bit kernel VA config on a system that 317 * does not support it, we have to place the available physical 318 * memory in the 48-bit addressable part of the linear region, i.e., 319 * we have to move it upward. Since memstart_addr represents the 320 * physical address of PAGE_OFFSET, we have to *subtract* from it. 321 */ 322 if (IS_ENABLED(CONFIG_ARM64_VA_BITS_52) && (vabits_actual != 52)) 323 memstart_addr -= _PAGE_OFFSET(48) - _PAGE_OFFSET(52); 324 325 /* 326 * Apply the memory limit if it was set. Since the kernel may be loaded 327 * high up in memory, add back the kernel region that must be accessible 328 * via the linear mapping. 329 */ 330 if (memory_limit != PHYS_ADDR_MAX) { 331 memblock_mem_limit_remove_map(memory_limit); 332 memblock_add(__pa_symbol(_text), (u64)(_end - _text)); 333 } 334 335 if (IS_ENABLED(CONFIG_BLK_DEV_INITRD) && phys_initrd_size) { 336 /* 337 * Add back the memory we just removed if it results in the 338 * initrd to become inaccessible via the linear mapping. 339 * Otherwise, this is a no-op 340 */ 341 u64 base = phys_initrd_start & PAGE_MASK; 342 u64 size = PAGE_ALIGN(phys_initrd_start + phys_initrd_size) - base; 343 344 /* 345 * We can only add back the initrd memory if we don't end up 346 * with more memory than we can address via the linear mapping. 347 * It is up to the bootloader to position the kernel and the 348 * initrd reasonably close to each other (i.e., within 32 GB of 349 * each other) so that all granule/#levels combinations can 350 * always access both. 351 */ 352 if (WARN(base < memblock_start_of_DRAM() || 353 base + size > memblock_start_of_DRAM() + 354 linear_region_size, 355 "initrd not fully accessible via the linear mapping -- please check your bootloader ...\n")) { 356 phys_initrd_size = 0; 357 } else { 358 memblock_remove(base, size); /* clear MEMBLOCK_ flags */ 359 memblock_add(base, size); 360 memblock_reserve(base, size); 361 } 362 } 363 364 if (IS_ENABLED(CONFIG_RANDOMIZE_BASE)) { 365 extern u16 memstart_offset_seed; 366 u64 mmfr0 = read_cpuid(ID_AA64MMFR0_EL1); 367 int parange = cpuid_feature_extract_unsigned_field( 368 mmfr0, ID_AA64MMFR0_PARANGE_SHIFT); 369 s64 range = linear_region_size - 370 BIT(id_aa64mmfr0_parange_to_phys_shift(parange)); 371 372 /* 373 * If the size of the linear region exceeds, by a sufficient 374 * margin, the size of the region that the physical memory can 375 * span, randomize the linear region as well. 376 */ 377 if (memstart_offset_seed > 0 && range >= (s64)ARM64_MEMSTART_ALIGN) { 378 range /= ARM64_MEMSTART_ALIGN; 379 memstart_addr -= ARM64_MEMSTART_ALIGN * 380 ((range * memstart_offset_seed) >> 16); 381 } 382 } 383 384 /* 385 * Register the kernel text, kernel data, initrd, and initial 386 * pagetables with memblock. 387 */ 388 memblock_reserve(__pa_symbol(_stext), _end - _stext); 389 if (IS_ENABLED(CONFIG_BLK_DEV_INITRD) && phys_initrd_size) { 390 /* the generic initrd code expects virtual addresses */ 391 initrd_start = __phys_to_virt(phys_initrd_start); 392 initrd_end = initrd_start + phys_initrd_size; 393 } 394 395 early_init_fdt_scan_reserved_mem(); 396 397 if (IS_ENABLED(CONFIG_ZONE_DMA32)) 398 arm64_dma32_phys_limit = max_zone_phys(32); 399 else 400 arm64_dma32_phys_limit = PHYS_MASK + 1; 401 402 reserve_elfcorehdr(); 403 404 high_memory = __va(memblock_end_of_DRAM() - 1) + 1; 405 406 dma_contiguous_reserve(arm64_dma32_phys_limit); 407 } 408 409 void __init bootmem_init(void) 410 { 411 unsigned long min, max; 412 413 min = PFN_UP(memblock_start_of_DRAM()); 414 max = PFN_DOWN(memblock_end_of_DRAM()); 415 416 early_memtest(min << PAGE_SHIFT, max << PAGE_SHIFT); 417 418 max_pfn = max_low_pfn = max; 419 min_low_pfn = min; 420 421 arch_numa_init(); 422 423 /* 424 * must be done after arch_numa_init() which calls numa_init() to 425 * initialize node_online_map that gets used in hugetlb_cma_reserve() 426 * while allocating required CMA size across online nodes. 427 */ 428 #if defined(CONFIG_HUGETLB_PAGE) && defined(CONFIG_CMA) 429 arm64_hugetlb_cma_reserve(); 430 #endif 431 432 dma_pernuma_cma_reserve(); 433 434 /* 435 * sparse_init() tries to allocate memory from memblock, so must be 436 * done after the fixed reservations 437 */ 438 sparse_init(); 439 zone_sizes_init(min, max); 440 441 /* 442 * request_standard_resources() depends on crashkernel's memory being 443 * reserved, so do it here. 444 */ 445 reserve_crashkernel(); 446 447 memblock_dump_all(); 448 } 449 450 /* 451 * mem_init() marks the free areas in the mem_map and tells us how much memory 452 * is free. This is done after various parts of the system have claimed their 453 * memory after the kernel image. 454 */ 455 void __init mem_init(void) 456 { 457 if (swiotlb_force == SWIOTLB_FORCE || 458 max_pfn > PFN_DOWN(arm64_dma_phys_limit ? : arm64_dma32_phys_limit)) 459 swiotlb_init(1); 460 else 461 swiotlb_force = SWIOTLB_NO_FORCE; 462 463 set_max_mapnr(max_pfn - PHYS_PFN_OFFSET); 464 465 /* this will put all unused low memory onto the freelists */ 466 memblock_free_all(); 467 468 mem_init_print_info(NULL); 469 470 /* 471 * Check boundaries twice: Some fundamental inconsistencies can be 472 * detected at build time already. 473 */ 474 #ifdef CONFIG_COMPAT 475 BUILD_BUG_ON(TASK_SIZE_32 > DEFAULT_MAP_WINDOW_64); 476 #endif 477 478 if (PAGE_SIZE >= 16384 && get_num_physpages() <= 128) { 479 extern int sysctl_overcommit_memory; 480 /* 481 * On a machine this small we won't get anywhere without 482 * overcommit, so turn it on by default. 483 */ 484 sysctl_overcommit_memory = OVERCOMMIT_ALWAYS; 485 } 486 } 487 488 void free_initmem(void) 489 { 490 free_reserved_area(lm_alias(__init_begin), 491 lm_alias(__init_end), 492 POISON_FREE_INITMEM, "unused kernel"); 493 /* 494 * Unmap the __init region but leave the VM area in place. This 495 * prevents the region from being reused for kernel modules, which 496 * is not supported by kallsyms. 497 */ 498 unmap_kernel_range((u64)__init_begin, (u64)(__init_end - __init_begin)); 499 } 500 501 void dump_mem_limit(void) 502 { 503 if (memory_limit != PHYS_ADDR_MAX) { 504 pr_emerg("Memory Limit: %llu MB\n", memory_limit >> 20); 505 } else { 506 pr_emerg("Memory Limit: none\n"); 507 } 508 } 509