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