1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Copyright (C) 1995 Linus Torvalds 4 * 5 * This file contains the setup_arch() code, which handles the architecture-dependent 6 * parts of early kernel initialization. 7 */ 8 #include <linux/acpi.h> 9 #include <linux/console.h> 10 #include <linux/cpu.h> 11 #include <linux/crash_dump.h> 12 #include <linux/dma-map-ops.h> 13 #include <linux/efi.h> 14 #include <linux/hugetlb.h> 15 #include <linux/ima.h> 16 #include <linux/init_ohci1394_dma.h> 17 #include <linux/initrd.h> 18 #include <linux/iscsi_ibft.h> 19 #include <linux/memblock.h> 20 #include <linux/panic_notifier.h> 21 #include <linux/pci.h> 22 #include <linux/random.h> 23 #include <linux/root_dev.h> 24 #include <linux/static_call.h> 25 #include <linux/swiotlb.h> 26 #include <linux/tboot.h> 27 #include <linux/usb/xhci-dbgp.h> 28 #include <linux/vmalloc.h> 29 30 #include <uapi/linux/mount.h> 31 32 #include <xen/xen.h> 33 34 #include <asm/apic.h> 35 #include <asm/bios_ebda.h> 36 #include <asm/bugs.h> 37 #include <asm/cacheinfo.h> 38 #include <asm/coco.h> 39 #include <asm/cpu.h> 40 #include <asm/efi.h> 41 #include <asm/gart.h> 42 #include <asm/hypervisor.h> 43 #include <asm/io_apic.h> 44 #include <asm/kasan.h> 45 #include <asm/kaslr.h> 46 #include <asm/mce.h> 47 #include <asm/memtype.h> 48 #include <asm/mtrr.h> 49 #include <asm/nmi.h> 50 #include <asm/numa.h> 51 #include <asm/olpc_ofw.h> 52 #include <asm/pci-direct.h> 53 #include <asm/prom.h> 54 #include <asm/proto.h> 55 #include <asm/realmode.h> 56 #include <asm/thermal.h> 57 #include <asm/unwind.h> 58 #include <asm/vsyscall.h> 59 60 /* 61 * max_low_pfn_mapped: highest directly mapped pfn < 4 GB 62 * max_pfn_mapped: highest directly mapped pfn > 4 GB 63 * 64 * The direct mapping only covers E820_TYPE_RAM regions, so the ranges and gaps are 65 * represented by pfn_mapped[]. 66 */ 67 unsigned long max_low_pfn_mapped; 68 unsigned long max_pfn_mapped; 69 70 #ifdef CONFIG_DMI 71 RESERVE_BRK(dmi_alloc, 65536); 72 #endif 73 74 75 unsigned long _brk_start = (unsigned long)__brk_base; 76 unsigned long _brk_end = (unsigned long)__brk_base; 77 78 struct boot_params boot_params; 79 80 /* 81 * These are the four main kernel memory regions, we put them into 82 * the resource tree so that kdump tools and other debugging tools 83 * recover it: 84 */ 85 86 static struct resource rodata_resource = { 87 .name = "Kernel rodata", 88 .start = 0, 89 .end = 0, 90 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM 91 }; 92 93 static struct resource data_resource = { 94 .name = "Kernel data", 95 .start = 0, 96 .end = 0, 97 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM 98 }; 99 100 static struct resource code_resource = { 101 .name = "Kernel code", 102 .start = 0, 103 .end = 0, 104 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM 105 }; 106 107 static struct resource bss_resource = { 108 .name = "Kernel bss", 109 .start = 0, 110 .end = 0, 111 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM 112 }; 113 114 115 #ifdef CONFIG_X86_32 116 /* CPU data as detected by the assembly code in head_32.S */ 117 struct cpuinfo_x86 new_cpu_data; 118 119 struct apm_info apm_info; 120 EXPORT_SYMBOL(apm_info); 121 122 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \ 123 defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE) 124 struct ist_info ist_info; 125 EXPORT_SYMBOL(ist_info); 126 #else 127 struct ist_info ist_info; 128 #endif 129 130 #endif 131 132 struct cpuinfo_x86 boot_cpu_data __read_mostly; 133 EXPORT_SYMBOL(boot_cpu_data); 134 SYM_PIC_ALIAS(boot_cpu_data); 135 136 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64) 137 __visible unsigned long mmu_cr4_features __ro_after_init; 138 #else 139 __visible unsigned long mmu_cr4_features __ro_after_init = X86_CR4_PAE; 140 #endif 141 142 #ifdef CONFIG_IMA 143 static phys_addr_t ima_kexec_buffer_phys; 144 static size_t ima_kexec_buffer_size; 145 #endif 146 147 /* Boot loader ID and version as integers, for the benefit of proc_dointvec */ 148 int bootloader_type, bootloader_version; 149 150 static const struct ctl_table x86_sysctl_table[] = { 151 { 152 .procname = "unknown_nmi_panic", 153 .data = &unknown_nmi_panic, 154 .maxlen = sizeof(int), 155 .mode = 0644, 156 .proc_handler = proc_dointvec, 157 }, 158 { 159 .procname = "panic_on_unrecovered_nmi", 160 .data = &panic_on_unrecovered_nmi, 161 .maxlen = sizeof(int), 162 .mode = 0644, 163 .proc_handler = proc_dointvec, 164 }, 165 { 166 .procname = "panic_on_io_nmi", 167 .data = &panic_on_io_nmi, 168 .maxlen = sizeof(int), 169 .mode = 0644, 170 .proc_handler = proc_dointvec, 171 }, 172 { 173 .procname = "bootloader_type", 174 .data = &bootloader_type, 175 .maxlen = sizeof(int), 176 .mode = 0444, 177 .proc_handler = proc_dointvec, 178 }, 179 { 180 .procname = "bootloader_version", 181 .data = &bootloader_version, 182 .maxlen = sizeof(int), 183 .mode = 0444, 184 .proc_handler = proc_dointvec, 185 }, 186 { 187 .procname = "io_delay_type", 188 .data = &io_delay_type, 189 .maxlen = sizeof(int), 190 .mode = 0644, 191 .proc_handler = proc_dointvec, 192 }, 193 #if defined(CONFIG_ACPI_SLEEP) 194 { 195 .procname = "acpi_video_flags", 196 .data = &acpi_realmode_flags, 197 .maxlen = sizeof(unsigned long), 198 .mode = 0644, 199 .proc_handler = proc_doulongvec_minmax, 200 }, 201 #endif 202 }; 203 204 static int __init init_x86_sysctl(void) 205 { 206 register_sysctl_init("kernel", x86_sysctl_table); 207 return 0; 208 } 209 arch_initcall(init_x86_sysctl); 210 211 /* 212 * Setup options 213 */ 214 struct screen_info screen_info; 215 EXPORT_SYMBOL(screen_info); 216 #if defined(CONFIG_FIRMWARE_EDID) 217 struct edid_info edid_info; 218 EXPORT_SYMBOL_GPL(edid_info); 219 #endif 220 221 extern int root_mountflags; 222 223 unsigned long saved_video_mode; 224 225 #define RAMDISK_IMAGE_START_MASK 0x07FF 226 #define RAMDISK_PROMPT_FLAG 0x8000 227 #define RAMDISK_LOAD_FLAG 0x4000 228 229 static char __initdata command_line[COMMAND_LINE_SIZE]; 230 #ifdef CONFIG_CMDLINE_BOOL 231 char builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE; 232 bool builtin_cmdline_added __ro_after_init; 233 #endif 234 235 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE) 236 struct edd edd; 237 #ifdef CONFIG_EDD_MODULE 238 EXPORT_SYMBOL(edd); 239 #endif 240 /** 241 * copy_edd() - Copy the BIOS EDD information 242 * from boot_params into a safe place. 243 * 244 */ 245 static inline void __init copy_edd(void) 246 { 247 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer, 248 sizeof(edd.mbr_signature)); 249 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info)); 250 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries; 251 edd.edd_info_nr = boot_params.eddbuf_entries; 252 } 253 #else 254 static inline void __init copy_edd(void) 255 { 256 } 257 #endif 258 259 void * __init extend_brk(size_t size, size_t align) 260 { 261 size_t mask = align - 1; 262 void *ret; 263 264 BUG_ON(_brk_start == 0); 265 BUG_ON(align & mask); 266 267 _brk_end = (_brk_end + mask) & ~mask; 268 BUG_ON((char *)(_brk_end + size) > __brk_limit); 269 270 ret = (void *)_brk_end; 271 _brk_end += size; 272 273 memset(ret, 0, size); 274 275 return ret; 276 } 277 278 #ifdef CONFIG_X86_32 279 static void __init cleanup_highmap(void) 280 { 281 } 282 #endif 283 284 static void __init reserve_brk(void) 285 { 286 if (_brk_end > _brk_start) 287 memblock_reserve_kern(__pa_symbol(_brk_start), 288 _brk_end - _brk_start); 289 290 /* Mark brk area as locked down and no longer taking any 291 new allocations */ 292 _brk_start = 0; 293 } 294 295 #ifdef CONFIG_BLK_DEV_INITRD 296 297 static u64 __init get_ramdisk_image(void) 298 { 299 u64 ramdisk_image = boot_params.hdr.ramdisk_image; 300 301 ramdisk_image |= (u64)boot_params.ext_ramdisk_image << 32; 302 303 if (ramdisk_image == 0) 304 ramdisk_image = phys_initrd_start; 305 306 return ramdisk_image; 307 } 308 static u64 __init get_ramdisk_size(void) 309 { 310 u64 ramdisk_size = boot_params.hdr.ramdisk_size; 311 312 ramdisk_size |= (u64)boot_params.ext_ramdisk_size << 32; 313 314 if (ramdisk_size == 0) 315 ramdisk_size = phys_initrd_size; 316 317 return ramdisk_size; 318 } 319 320 static void __init relocate_initrd(void) 321 { 322 /* Assume only end is not page aligned */ 323 u64 ramdisk_image = get_ramdisk_image(); 324 u64 ramdisk_size = get_ramdisk_size(); 325 u64 area_size = PAGE_ALIGN(ramdisk_size); 326 int ret = 0; 327 328 /* We need to move the initrd down into directly mapped mem */ 329 u64 relocated_ramdisk = memblock_phys_alloc_range(area_size, PAGE_SIZE, 0, 330 PFN_PHYS(max_pfn_mapped)); 331 if (!relocated_ramdisk) 332 panic("Cannot find place for new RAMDISK of size %lld\n", 333 ramdisk_size); 334 335 initrd_start = relocated_ramdisk + PAGE_OFFSET; 336 initrd_end = initrd_start + ramdisk_size; 337 printk(KERN_INFO "Allocated new RAMDISK: [mem %#010llx-%#010llx]\n", 338 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1); 339 340 ret = copy_from_early_mem((void *)initrd_start, ramdisk_image, ramdisk_size); 341 if (ret) 342 panic("Copy RAMDISK failed\n"); 343 344 printk(KERN_INFO "Move RAMDISK from [mem %#010llx-%#010llx] to" 345 " [mem %#010llx-%#010llx]\n", 346 ramdisk_image, ramdisk_image + ramdisk_size - 1, 347 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1); 348 } 349 350 static void __init early_reserve_initrd(void) 351 { 352 /* Assume only end is not page aligned */ 353 u64 ramdisk_image = get_ramdisk_image(); 354 u64 ramdisk_size = get_ramdisk_size(); 355 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size); 356 357 if (!boot_params.hdr.type_of_loader || 358 !ramdisk_image || !ramdisk_size) 359 return; /* No initrd provided by bootloader */ 360 361 memblock_reserve_kern(ramdisk_image, ramdisk_end - ramdisk_image); 362 } 363 364 static void __init reserve_initrd(void) 365 { 366 /* Assume only end is not page aligned */ 367 u64 ramdisk_image = get_ramdisk_image(); 368 u64 ramdisk_size = get_ramdisk_size(); 369 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size); 370 371 if (!boot_params.hdr.type_of_loader || 372 !ramdisk_image || !ramdisk_size) 373 return; /* No initrd provided by bootloader */ 374 375 initrd_start = 0; 376 377 printk(KERN_INFO "RAMDISK: [mem %#010llx-%#010llx]\n", ramdisk_image, 378 ramdisk_end - 1); 379 380 if (pfn_range_is_mapped(PFN_DOWN(ramdisk_image), 381 PFN_DOWN(ramdisk_end))) { 382 /* All are mapped, easy case */ 383 initrd_start = ramdisk_image + PAGE_OFFSET; 384 initrd_end = initrd_start + ramdisk_size; 385 return; 386 } 387 388 relocate_initrd(); 389 390 memblock_phys_free(ramdisk_image, ramdisk_end - ramdisk_image); 391 } 392 393 #else 394 static void __init early_reserve_initrd(void) 395 { 396 } 397 static void __init reserve_initrd(void) 398 { 399 } 400 #endif /* CONFIG_BLK_DEV_INITRD */ 401 402 static void __init add_early_ima_buffer(u64 phys_addr) 403 { 404 #ifdef CONFIG_IMA 405 struct ima_setup_data *data; 406 407 data = early_memremap(phys_addr + sizeof(struct setup_data), sizeof(*data)); 408 if (!data) { 409 pr_warn("setup: failed to memremap ima_setup_data entry\n"); 410 return; 411 } 412 413 if (data->size) { 414 memblock_reserve_kern(data->addr, data->size); 415 ima_kexec_buffer_phys = data->addr; 416 ima_kexec_buffer_size = data->size; 417 } 418 419 early_memunmap(data, sizeof(*data)); 420 #else 421 pr_warn("Passed IMA kexec data, but CONFIG_IMA not set. Ignoring.\n"); 422 #endif 423 } 424 425 #if defined(CONFIG_HAVE_IMA_KEXEC) && !defined(CONFIG_OF_FLATTREE) 426 int __init ima_free_kexec_buffer(void) 427 { 428 if (!ima_kexec_buffer_size) 429 return -ENOENT; 430 431 memblock_free_late(ima_kexec_buffer_phys, 432 ima_kexec_buffer_size); 433 434 ima_kexec_buffer_phys = 0; 435 ima_kexec_buffer_size = 0; 436 437 return 0; 438 } 439 440 int __init ima_get_kexec_buffer(void **addr, size_t *size) 441 { 442 int ret; 443 444 if (!ima_kexec_buffer_size) 445 return -ENOENT; 446 447 ret = ima_validate_range(ima_kexec_buffer_phys, ima_kexec_buffer_size); 448 if (ret) 449 return ret; 450 451 *addr = __va(ima_kexec_buffer_phys); 452 *size = ima_kexec_buffer_size; 453 454 return 0; 455 } 456 #endif 457 458 static void __init add_kho(u64 phys_addr, u32 data_len) 459 { 460 struct kho_data *kho; 461 u64 addr = phys_addr + sizeof(struct setup_data); 462 u64 size = data_len - sizeof(struct setup_data); 463 464 if (!IS_ENABLED(CONFIG_KEXEC_HANDOVER)) { 465 pr_warn("Passed KHO data, but CONFIG_KEXEC_HANDOVER not set. Ignoring.\n"); 466 return; 467 } 468 469 kho = early_memremap(addr, size); 470 if (!kho) { 471 pr_warn("setup: failed to memremap kho data (0x%llx, 0x%llx)\n", 472 addr, size); 473 return; 474 } 475 476 kho_populate(kho->fdt_addr, kho->fdt_size, kho->scratch_addr, kho->scratch_size); 477 478 early_memunmap(kho, size); 479 } 480 481 static void __init parse_setup_data(void) 482 { 483 struct setup_data *data; 484 u64 pa_data, pa_next; 485 486 pa_data = boot_params.hdr.setup_data; 487 while (pa_data) { 488 u32 data_len, data_type; 489 490 data = early_memremap(pa_data, sizeof(*data)); 491 data_len = data->len + sizeof(struct setup_data); 492 data_type = data->type; 493 pa_next = data->next; 494 early_memunmap(data, sizeof(*data)); 495 496 switch (data_type) { 497 case SETUP_E820_EXT: 498 e820__memory_setup_extended(pa_data, data_len); 499 break; 500 case SETUP_DTB: 501 add_dtb(pa_data); 502 break; 503 case SETUP_EFI: 504 parse_efi_setup(pa_data, data_len); 505 break; 506 case SETUP_IMA: 507 add_early_ima_buffer(pa_data); 508 break; 509 case SETUP_KEXEC_KHO: 510 add_kho(pa_data, data_len); 511 break; 512 case SETUP_RNG_SEED: 513 data = early_memremap(pa_data, data_len); 514 add_bootloader_randomness(data->data, data->len); 515 /* Zero seed for forward secrecy. */ 516 memzero_explicit(data->data, data->len); 517 /* Zero length in case we find ourselves back here by accident. */ 518 memzero_explicit(&data->len, sizeof(data->len)); 519 early_memunmap(data, data_len); 520 break; 521 default: 522 break; 523 } 524 pa_data = pa_next; 525 } 526 } 527 528 /* 529 * Translate the fields of 'struct boot_param' into global variables 530 * representing these parameters. 531 */ 532 static void __init parse_boot_params(void) 533 { 534 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev); 535 screen_info = boot_params.screen_info; 536 #if defined(CONFIG_FIRMWARE_EDID) 537 edid_info = boot_params.edid_info; 538 #endif 539 #ifdef CONFIG_X86_32 540 apm_info.bios = boot_params.apm_bios_info; 541 ist_info = boot_params.ist_info; 542 #endif 543 saved_video_mode = boot_params.hdr.vid_mode; 544 bootloader_type = boot_params.hdr.type_of_loader; 545 if ((bootloader_type >> 4) == 0xe) { 546 bootloader_type &= 0xf; 547 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4; 548 } 549 bootloader_version = bootloader_type & 0xf; 550 bootloader_version |= boot_params.hdr.ext_loader_ver << 4; 551 552 #ifdef CONFIG_BLK_DEV_RAM 553 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK; 554 #endif 555 #ifdef CONFIG_EFI 556 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature, 557 EFI32_LOADER_SIGNATURE, 4)) { 558 set_bit(EFI_BOOT, &efi.flags); 559 } else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature, 560 EFI64_LOADER_SIGNATURE, 4)) { 561 set_bit(EFI_BOOT, &efi.flags); 562 set_bit(EFI_64BIT, &efi.flags); 563 } 564 #endif 565 566 if (!boot_params.hdr.root_flags) 567 root_mountflags &= ~MS_RDONLY; 568 } 569 570 static void __init memblock_x86_reserve_range_setup_data(void) 571 { 572 struct setup_indirect *indirect; 573 struct setup_data *data; 574 u64 pa_data, pa_next; 575 u32 len; 576 577 pa_data = boot_params.hdr.setup_data; 578 while (pa_data) { 579 data = early_memremap(pa_data, sizeof(*data)); 580 if (!data) { 581 pr_warn("setup: failed to memremap setup_data entry\n"); 582 return; 583 } 584 585 len = sizeof(*data); 586 pa_next = data->next; 587 588 memblock_reserve_kern(pa_data, sizeof(*data) + data->len); 589 590 if (data->type == SETUP_INDIRECT) { 591 len += data->len; 592 early_memunmap(data, sizeof(*data)); 593 data = early_memremap(pa_data, len); 594 if (!data) { 595 pr_warn("setup: failed to memremap indirect setup_data\n"); 596 return; 597 } 598 599 indirect = (struct setup_indirect *)data->data; 600 601 if (indirect->type != SETUP_INDIRECT) 602 memblock_reserve_kern(indirect->addr, indirect->len); 603 } 604 605 pa_data = pa_next; 606 early_memunmap(data, len); 607 } 608 } 609 610 static void __init arch_reserve_crashkernel(void) 611 { 612 unsigned long long crash_base, crash_size, low_size = 0, cma_size = 0; 613 bool high = false; 614 int ret; 615 616 if (!IS_ENABLED(CONFIG_CRASH_RESERVE)) 617 return; 618 619 ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(), 620 &crash_size, &crash_base, 621 &low_size, &cma_size, &high); 622 if (ret) 623 return; 624 625 if (xen_pv_domain()) { 626 pr_info("Ignoring crashkernel for a Xen PV domain\n"); 627 return; 628 } 629 630 reserve_crashkernel_generic(crash_size, crash_base, low_size, high); 631 reserve_crashkernel_cma(cma_size); 632 } 633 634 static struct resource standard_io_resources[] = { 635 { .name = "dma1", .start = 0x00, .end = 0x1f, 636 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 637 { .name = "pic1", .start = 0x20, .end = 0x21, 638 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 639 { .name = "timer0", .start = 0x40, .end = 0x43, 640 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 641 { .name = "timer1", .start = 0x50, .end = 0x53, 642 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 643 { .name = "keyboard", .start = 0x60, .end = 0x60, 644 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 645 { .name = "keyboard", .start = 0x64, .end = 0x64, 646 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 647 { .name = "dma page reg", .start = 0x80, .end = 0x8f, 648 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 649 { .name = "pic2", .start = 0xa0, .end = 0xa1, 650 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 651 { .name = "dma2", .start = 0xc0, .end = 0xdf, 652 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 653 { .name = "fpu", .start = 0xf0, .end = 0xff, 654 .flags = IORESOURCE_BUSY | IORESOURCE_IO } 655 }; 656 657 void __init reserve_standard_io_resources(void) 658 { 659 int i; 660 661 /* request I/O space for devices used on all i[345]86 PCs */ 662 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++) 663 request_resource(&ioport_resource, &standard_io_resources[i]); 664 665 } 666 667 static void __init setup_kernel_resources(void) 668 { 669 code_resource.start = __pa_symbol(_text); 670 code_resource.end = __pa_symbol(_etext)-1; 671 rodata_resource.start = __pa_symbol(__start_rodata); 672 rodata_resource.end = __pa_symbol(__end_rodata)-1; 673 data_resource.start = __pa_symbol(_sdata); 674 data_resource.end = __pa_symbol(_edata)-1; 675 bss_resource.start = __pa_symbol(__bss_start); 676 bss_resource.end = __pa_symbol(__bss_stop)-1; 677 678 insert_resource(&iomem_resource, &code_resource); 679 insert_resource(&iomem_resource, &rodata_resource); 680 insert_resource(&iomem_resource, &data_resource); 681 insert_resource(&iomem_resource, &bss_resource); 682 } 683 684 static bool __init snb_gfx_workaround_needed(void) 685 { 686 #ifdef CONFIG_PCI 687 int i; 688 u16 vendor, devid; 689 static const __initconst u16 snb_ids[] = { 690 0x0102, 691 0x0112, 692 0x0122, 693 0x0106, 694 0x0116, 695 0x0126, 696 0x010a, 697 }; 698 699 /* Assume no if something weird is going on with PCI */ 700 if (!early_pci_allowed()) 701 return false; 702 703 vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID); 704 if (vendor != 0x8086) 705 return false; 706 707 devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID); 708 for (i = 0; i < ARRAY_SIZE(snb_ids); i++) 709 if (devid == snb_ids[i]) 710 return true; 711 #endif 712 713 return false; 714 } 715 716 /* 717 * Sandy Bridge graphics has trouble with certain ranges, exclude 718 * them from allocation. 719 */ 720 static void __init trim_snb_memory(void) 721 { 722 static const __initconst unsigned long bad_pages[] = { 723 0x20050000, 724 0x20110000, 725 0x20130000, 726 0x20138000, 727 0x40004000, 728 }; 729 int i; 730 731 if (!snb_gfx_workaround_needed()) 732 return; 733 734 printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n"); 735 736 /* 737 * SandyBridge integrated graphics devices have a bug that prevents 738 * them from accessing certain memory ranges, namely anything below 739 * 1M and in the pages listed in bad_pages[] above. 740 * 741 * To avoid these pages being ever accessed by SNB gfx devices reserve 742 * bad_pages that have not already been reserved at boot time. 743 * All memory below the 1 MB mark is anyway reserved later during 744 * setup_arch(), so there is no need to reserve it here. 745 */ 746 747 for (i = 0; i < ARRAY_SIZE(bad_pages); i++) { 748 if (memblock_reserve(bad_pages[i], PAGE_SIZE)) 749 printk(KERN_WARNING "failed to reserve 0x%08lx\n", 750 bad_pages[i]); 751 } 752 } 753 754 static void __init trim_bios_range(void) 755 { 756 /* 757 * A special case is the first 4Kb of memory; 758 * This is a BIOS owned area, not kernel ram, but generally 759 * not listed as such in the E820 table. 760 * 761 * This typically reserves additional memory (64KiB by default) 762 * since some BIOSes are known to corrupt low memory. See the 763 * Kconfig help text for X86_RESERVE_LOW. 764 */ 765 e820__range_update(0, PAGE_SIZE, E820_TYPE_RAM, E820_TYPE_RESERVED); 766 767 /* 768 * special case: Some BIOSes report the PC BIOS 769 * area (640Kb -> 1Mb) as RAM even though it is not. 770 * take them out. 771 */ 772 e820__range_remove(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_TYPE_RAM, 1); 773 774 e820__update_table(e820_table); 775 } 776 777 /* called before trim_bios_range() to spare extra sanitize */ 778 static void __init e820_add_kernel_range(void) 779 { 780 u64 start = __pa_symbol(_text); 781 u64 size = __pa_symbol(_end) - start; 782 783 /* 784 * Complain if .text .data and .bss are not marked as E820_TYPE_RAM and 785 * attempt to fix it by adding the range. We may have a confused BIOS, 786 * or the user may have used memmap=exactmap or memmap=xxM$yyM to 787 * exclude kernel range. If we really are running on top non-RAM, 788 * we will crash later anyways. 789 */ 790 if (e820__mapped_all(start, start + size, E820_TYPE_RAM)) 791 return; 792 793 pr_warn(".text .data .bss are not marked as E820_TYPE_RAM!\n"); 794 e820__range_remove(start, size, E820_TYPE_RAM, 0); 795 e820__range_add(start, size, E820_TYPE_RAM); 796 } 797 798 static void __init early_reserve_memory(void) 799 { 800 /* 801 * Reserve the memory occupied by the kernel between _text and 802 * __end_of_kernel_reserve symbols. Any kernel sections after the 803 * __end_of_kernel_reserve symbol must be explicitly reserved with a 804 * separate memblock_reserve() or they will be discarded. 805 */ 806 memblock_reserve_kern(__pa_symbol(_text), 807 (unsigned long)__end_of_kernel_reserve - (unsigned long)_text); 808 809 /* 810 * The first 4Kb of memory is a BIOS owned area, but generally it is 811 * not listed as such in the E820 table. 812 * 813 * Reserve the first 64K of memory since some BIOSes are known to 814 * corrupt low memory. After the real mode trampoline is allocated the 815 * rest of the memory below 640k is reserved. 816 * 817 * In addition, make sure page 0 is always reserved because on 818 * systems with L1TF its contents can be leaked to user processes. 819 */ 820 memblock_reserve(0, SZ_64K); 821 822 early_reserve_initrd(); 823 824 memblock_x86_reserve_range_setup_data(); 825 826 reserve_bios_regions(); 827 trim_snb_memory(); 828 } 829 830 /* 831 * Dump out kernel offset information on panic. 832 */ 833 static int 834 dump_kernel_offset(struct notifier_block *self, unsigned long v, void *p) 835 { 836 if (kaslr_enabled()) { 837 pr_emerg("Kernel Offset: 0x%lx from 0x%lx (relocation range: 0x%lx-0x%lx)\n", 838 kaslr_offset(), 839 __START_KERNEL, 840 __START_KERNEL_map, 841 MODULES_VADDR-1); 842 } else { 843 pr_emerg("Kernel Offset: disabled\n"); 844 } 845 846 return 0; 847 } 848 849 void x86_configure_nx(void) 850 { 851 if (boot_cpu_has(X86_FEATURE_NX)) 852 __supported_pte_mask |= _PAGE_NX; 853 else 854 __supported_pte_mask &= ~_PAGE_NX; 855 } 856 857 static void __init x86_report_nx(void) 858 { 859 if (!boot_cpu_has(X86_FEATURE_NX)) { 860 printk(KERN_NOTICE "Notice: NX (Execute Disable) protection " 861 "missing in CPU!\n"); 862 } else { 863 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE) 864 printk(KERN_INFO "NX (Execute Disable) protection: active\n"); 865 #else 866 /* 32bit non-PAE kernel, NX cannot be used */ 867 printk(KERN_NOTICE "Notice: NX (Execute Disable) protection " 868 "cannot be enabled: non-PAE kernel!\n"); 869 #endif 870 } 871 } 872 873 /* 874 * Determine if we were loaded by an EFI loader. If so, then we have also been 875 * passed the efi memmap, systab, etc., so we should use these data structures 876 * for initialization. Note, the efi init code path is determined by the 877 * global efi_enabled. This allows the same kernel image to be used on existing 878 * systems (with a traditional BIOS) as well as on EFI systems. 879 */ 880 /* 881 * setup_arch - architecture-specific boot-time initializations 882 * 883 * Note: On x86_64, fixmaps are ready for use even before this is called. 884 */ 885 886 void __init setup_arch(char **cmdline_p) 887 { 888 #ifdef CONFIG_X86_32 889 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data)); 890 891 /* 892 * copy kernel address range established so far and switch 893 * to the proper swapper page table 894 */ 895 clone_pgd_range(swapper_pg_dir + KERNEL_PGD_BOUNDARY, 896 initial_page_table + KERNEL_PGD_BOUNDARY, 897 KERNEL_PGD_PTRS); 898 899 load_cr3(swapper_pg_dir); 900 /* 901 * Note: Quark X1000 CPUs advertise PGE incorrectly and require 902 * a cr3 based tlb flush, so the following __flush_tlb_all() 903 * will not flush anything because the CPU quirk which clears 904 * X86_FEATURE_PGE has not been invoked yet. Though due to the 905 * load_cr3() above the TLB has been flushed already. The 906 * quirk is invoked before subsequent calls to __flush_tlb_all() 907 * so proper operation is guaranteed. 908 */ 909 __flush_tlb_all(); 910 #else 911 printk(KERN_INFO "Command line: %s\n", boot_command_line); 912 boot_cpu_data.x86_phys_bits = MAX_PHYSMEM_BITS; 913 #endif 914 915 #ifdef CONFIG_CMDLINE_BOOL 916 #ifdef CONFIG_CMDLINE_OVERRIDE 917 strscpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 918 #else 919 if (builtin_cmdline[0]) { 920 /* append boot loader cmdline to builtin */ 921 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE); 922 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE); 923 strscpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 924 } 925 #endif 926 builtin_cmdline_added = true; 927 #endif 928 929 strscpy(command_line, boot_command_line, COMMAND_LINE_SIZE); 930 *cmdline_p = command_line; 931 932 /* 933 * If we have OLPC OFW, we might end up relocating the fixmap due to 934 * reserve_top(), so do this before touching the ioremap area. 935 */ 936 olpc_ofw_detect(); 937 938 idt_setup_early_traps(); 939 early_cpu_init(); 940 jump_label_init(); 941 static_call_init(); 942 early_ioremap_init(); 943 944 setup_olpc_ofw_pgd(); 945 946 parse_boot_params(); 947 948 x86_init.oem.arch_setup(); 949 950 /* 951 * Do some memory reservations *before* memory is added to memblock, so 952 * memblock allocations won't overwrite it. 953 * 954 * After this point, everything still needed from the boot loader or 955 * firmware or kernel text should be early reserved or marked not RAM in 956 * e820. All other memory is free game. 957 * 958 * This call needs to happen before e820__memory_setup() which calls the 959 * xen_memory_setup() on Xen dom0 which relies on the fact that those 960 * early reservations have happened already. 961 */ 962 early_reserve_memory(); 963 964 iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1; 965 e820__memory_setup(); 966 parse_setup_data(); 967 968 copy_edd(); 969 970 setup_initial_init_mm(_text, _etext, _edata, (void *)_brk_end); 971 972 /* 973 * x86_configure_nx() is called before parse_early_param() to detect 974 * whether hardware doesn't support NX (so that the early EHCI debug 975 * console setup can safely call set_fixmap()). 976 */ 977 x86_configure_nx(); 978 979 parse_early_param(); 980 981 if (efi_enabled(EFI_BOOT)) 982 efi_memblock_x86_reserve_range(); 983 984 x86_report_nx(); 985 986 apic_setup_apic_calls(); 987 988 if (acpi_mps_check()) { 989 #ifdef CONFIG_X86_LOCAL_APIC 990 apic_is_disabled = true; 991 #endif 992 setup_clear_cpu_cap(X86_FEATURE_APIC); 993 } 994 995 e820__finish_early_params(); 996 997 if (efi_enabled(EFI_BOOT)) 998 efi_init(); 999 1000 reserve_ibft_region(); 1001 x86_init.resources.dmi_setup(); 1002 1003 /* 1004 * VMware detection requires dmi to be available, so this 1005 * needs to be done after dmi_setup(), for the boot CPU. 1006 * For some guest types (Xen PV, SEV-SNP, TDX) it is required to be 1007 * called before cache_bp_init() for setting up MTRR state. 1008 */ 1009 init_hypervisor_platform(); 1010 1011 tsc_early_init(); 1012 x86_init.resources.probe_roms(); 1013 1014 /* 1015 * Add resources for kernel text and data to the iomem_resource. 1016 * Do it after parse_early_param, so it can be debugged. 1017 */ 1018 setup_kernel_resources(); 1019 1020 e820_add_kernel_range(); 1021 trim_bios_range(); 1022 #ifdef CONFIG_X86_32 1023 if (ppro_with_ram_bug()) { 1024 e820__range_update(0x70000000ULL, 0x40000ULL, E820_TYPE_RAM, 1025 E820_TYPE_RESERVED); 1026 e820__update_table(e820_table); 1027 printk(KERN_INFO "fixed physical RAM map:\n"); 1028 e820__print_table("bad_ppro"); 1029 } 1030 #else 1031 early_gart_iommu_check(); 1032 #endif 1033 1034 /* 1035 * partially used pages are not usable - thus 1036 * we are rounding upwards: 1037 */ 1038 max_pfn = e820__end_of_ram_pfn(); 1039 1040 /* update e820 for memory not covered by WB MTRRs */ 1041 cache_bp_init(); 1042 if (mtrr_trim_uncached_memory(max_pfn)) 1043 max_pfn = e820__end_of_ram_pfn(); 1044 1045 max_possible_pfn = max_pfn; 1046 1047 /* 1048 * Define random base addresses for memory sections after max_pfn is 1049 * defined and before each memory section base is used. 1050 */ 1051 kernel_randomize_memory(); 1052 1053 #ifdef CONFIG_X86_32 1054 /* max_low_pfn get updated here */ 1055 find_low_pfn_range(); 1056 #else 1057 check_x2apic(); 1058 1059 /* How many end-of-memory variables you have, grandma! */ 1060 /* need this before calling reserve_initrd */ 1061 if (max_pfn > (1UL<<(32 - PAGE_SHIFT))) 1062 max_low_pfn = e820__end_of_low_ram_pfn(); 1063 else 1064 max_low_pfn = max_pfn; 1065 #endif 1066 1067 /* Find and reserve MPTABLE area */ 1068 x86_init.mpparse.find_mptable(); 1069 1070 early_alloc_pgt_buf(); 1071 1072 /* 1073 * Need to conclude brk, before e820__memblock_setup() 1074 * it could use memblock_find_in_range, could overlap with 1075 * brk area. 1076 */ 1077 reserve_brk(); 1078 1079 cleanup_highmap(); 1080 1081 e820__memblock_setup(); 1082 1083 /* 1084 * Needs to run after memblock setup because it needs the physical 1085 * memory size. 1086 */ 1087 mem_encrypt_setup_arch(); 1088 cc_random_init(); 1089 1090 efi_find_mirror(); 1091 efi_esrt_init(); 1092 efi_mokvar_table_init(); 1093 1094 /* 1095 * The EFI specification says that boot service code won't be 1096 * called after ExitBootServices(). This is, in fact, a lie. 1097 */ 1098 efi_reserve_boot_services(); 1099 1100 /* preallocate 4k for mptable mpc */ 1101 e820__memblock_alloc_reserved_mpc_new(); 1102 1103 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION 1104 setup_bios_corruption_check(); 1105 #endif 1106 1107 #ifdef CONFIG_X86_32 1108 printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n", 1109 (max_pfn_mapped<<PAGE_SHIFT) - 1); 1110 #endif 1111 1112 /* 1113 * Find free memory for the real mode trampoline and place it there. If 1114 * there is not enough free memory under 1M, on EFI-enabled systems 1115 * there will be additional attempt to reclaim the memory for the real 1116 * mode trampoline at efi_free_boot_services(). 1117 * 1118 * Unconditionally reserve the entire first 1M of RAM because BIOSes 1119 * are known to corrupt low memory and several hundred kilobytes are not 1120 * worth complex detection what memory gets clobbered. Windows does the 1121 * same thing for very similar reasons. 1122 * 1123 * Moreover, on machines with SandyBridge graphics or in setups that use 1124 * crashkernel the entire 1M is reserved anyway. 1125 * 1126 * Note the host kernel TDX also requires the first 1MB being reserved. 1127 */ 1128 x86_platform.realmode_reserve(); 1129 1130 init_mem_mapping(); 1131 1132 /* 1133 * init_mem_mapping() relies on the early IDT page fault handling. 1134 * Now either enable FRED or install the real page fault handler 1135 * for 64-bit in the IDT. 1136 */ 1137 cpu_init_replace_early_idt(); 1138 1139 /* 1140 * Update mmu_cr4_features (and, indirectly, trampoline_cr4_features) 1141 * with the current CR4 value. This may not be necessary, but 1142 * auditing all the early-boot CR4 manipulation would be needed to 1143 * rule it out. 1144 * 1145 * Mask off features that don't work outside long mode (just 1146 * PCIDE for now). 1147 */ 1148 mmu_cr4_features = __read_cr4() & ~X86_CR4_PCIDE; 1149 1150 memblock_set_current_limit(get_max_mapped()); 1151 1152 /* 1153 * NOTE: On x86-32, only from this point on, fixmaps are ready for use. 1154 */ 1155 1156 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT 1157 if (init_ohci1394_dma_early) 1158 init_ohci1394_dma_on_all_controllers(); 1159 #endif 1160 /* Allocate bigger log buffer */ 1161 setup_log_buf(1); 1162 1163 if (efi_enabled(EFI_BOOT)) { 1164 switch (boot_params.secure_boot) { 1165 case efi_secureboot_mode_disabled: 1166 pr_info("Secure boot disabled\n"); 1167 break; 1168 case efi_secureboot_mode_enabled: 1169 pr_info("Secure boot enabled\n"); 1170 break; 1171 default: 1172 pr_info("Secure boot could not be determined\n"); 1173 break; 1174 } 1175 } 1176 1177 reserve_initrd(); 1178 1179 acpi_table_upgrade(); 1180 /* Look for ACPI tables and reserve memory occupied by them. */ 1181 acpi_boot_table_init(); 1182 1183 vsmp_init(); 1184 1185 io_delay_init(); 1186 1187 early_platform_quirks(); 1188 1189 /* Some platforms need the APIC registered for NUMA configuration */ 1190 early_acpi_boot_init(); 1191 x86_init.mpparse.early_parse_smp_cfg(); 1192 1193 x86_flattree_get_config(); 1194 1195 initmem_init(); 1196 dma_contiguous_reserve(max_pfn_mapped << PAGE_SHIFT); 1197 1198 if (boot_cpu_has(X86_FEATURE_GBPAGES)) { 1199 hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT); 1200 hugetlb_bootmem_alloc(); 1201 } 1202 1203 /* 1204 * Reserve memory for crash kernel after SRAT is parsed so that it 1205 * won't consume hotpluggable memory. 1206 */ 1207 arch_reserve_crashkernel(); 1208 1209 if (!early_xdbc_setup_hardware()) 1210 early_xdbc_register_console(); 1211 1212 x86_init.paging.pagetable_init(); 1213 1214 kasan_init(); 1215 1216 /* 1217 * Sync back kernel address range. 1218 * 1219 * FIXME: Can the later sync in setup_cpu_entry_areas() replace 1220 * this call? 1221 */ 1222 sync_initial_page_table(); 1223 1224 tboot_probe(); 1225 1226 map_vsyscall(); 1227 1228 x86_32_probe_apic(); 1229 1230 early_quirks(); 1231 1232 topology_apply_cmdline_limits_early(); 1233 1234 /* 1235 * Parse SMP configuration. Try ACPI first and then the platform 1236 * specific parser. 1237 */ 1238 acpi_boot_init(); 1239 x86_init.mpparse.parse_smp_cfg(); 1240 1241 /* Last opportunity to detect and map the local APIC */ 1242 init_apic_mappings(); 1243 1244 topology_init_possible_cpus(); 1245 1246 init_cpu_to_node(); 1247 init_gi_nodes(); 1248 1249 io_apic_init_mappings(); 1250 1251 x86_init.hyper.guest_late_init(); 1252 1253 e820__reserve_resources(); 1254 e820__register_nosave_regions(max_pfn); 1255 1256 x86_init.resources.reserve_resources(); 1257 1258 e820__setup_pci_gap(); 1259 1260 #ifdef CONFIG_VT 1261 #if defined(CONFIG_VGA_CONSOLE) 1262 if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY)) 1263 vgacon_register_screen(&screen_info); 1264 #endif 1265 #endif 1266 x86_init.oem.banner(); 1267 1268 x86_init.timers.wallclock_init(); 1269 1270 /* 1271 * This needs to run before setup_local_APIC() which soft-disables the 1272 * local APIC temporarily and that masks the thermal LVT interrupt, 1273 * leading to softlockups on machines which have configured SMI 1274 * interrupt delivery. 1275 */ 1276 therm_lvt_init(); 1277 1278 mcheck_init(); 1279 1280 register_refined_jiffies(CLOCK_TICK_RATE); 1281 1282 #ifdef CONFIG_EFI 1283 if (efi_enabled(EFI_BOOT)) 1284 efi_apply_memmap_quirks(); 1285 #endif 1286 1287 unwind_init(); 1288 } 1289 1290 #ifdef CONFIG_X86_32 1291 1292 static struct resource video_ram_resource = { 1293 .name = "Video RAM area", 1294 .start = 0xa0000, 1295 .end = 0xbffff, 1296 .flags = IORESOURCE_BUSY | IORESOURCE_MEM 1297 }; 1298 1299 void __init i386_reserve_resources(void) 1300 { 1301 request_resource(&iomem_resource, &video_ram_resource); 1302 reserve_standard_io_resources(); 1303 } 1304 1305 #endif /* CONFIG_X86_32 */ 1306 1307 static struct notifier_block kernel_offset_notifier = { 1308 .notifier_call = dump_kernel_offset 1309 }; 1310 1311 static int __init register_kernel_offset_dumper(void) 1312 { 1313 atomic_notifier_chain_register(&panic_notifier_list, 1314 &kernel_offset_notifier); 1315 return 0; 1316 } 1317 __initcall(register_kernel_offset_dumper); 1318 1319 #ifdef CONFIG_HOTPLUG_CPU 1320 bool arch_cpu_is_hotpluggable(int cpu) 1321 { 1322 return cpu > 0; 1323 } 1324 #endif /* CONFIG_HOTPLUG_CPU */ 1325