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