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 static void __init memblock_x86_reserve_range_setup_data(void) 499 { 500 struct setup_indirect *indirect; 501 struct setup_data *data; 502 u64 pa_data, pa_next; 503 u32 len; 504 505 pa_data = boot_params.hdr.setup_data; 506 while (pa_data) { 507 data = early_memremap(pa_data, sizeof(*data)); 508 if (!data) { 509 pr_warn("setup: failed to memremap setup_data entry\n"); 510 return; 511 } 512 513 len = sizeof(*data); 514 pa_next = data->next; 515 516 memblock_reserve(pa_data, sizeof(*data) + data->len); 517 518 if (data->type == SETUP_INDIRECT) { 519 len += data->len; 520 early_memunmap(data, sizeof(*data)); 521 data = early_memremap(pa_data, len); 522 if (!data) { 523 pr_warn("setup: failed to memremap indirect setup_data\n"); 524 return; 525 } 526 527 indirect = (struct setup_indirect *)data->data; 528 529 if (indirect->type != SETUP_INDIRECT) 530 memblock_reserve(indirect->addr, indirect->len); 531 } 532 533 pa_data = pa_next; 534 early_memunmap(data, len); 535 } 536 } 537 538 static void __init arch_reserve_crashkernel(void) 539 { 540 unsigned long long crash_base, crash_size, low_size = 0; 541 char *cmdline = boot_command_line; 542 bool high = false; 543 int ret; 544 545 if (!IS_ENABLED(CONFIG_CRASH_RESERVE)) 546 return; 547 548 ret = parse_crashkernel(cmdline, memblock_phys_mem_size(), 549 &crash_size, &crash_base, 550 &low_size, &high); 551 if (ret) 552 return; 553 554 if (xen_pv_domain()) { 555 pr_info("Ignoring crashkernel for a Xen PV domain\n"); 556 return; 557 } 558 559 reserve_crashkernel_generic(cmdline, crash_size, crash_base, 560 low_size, high); 561 } 562 563 static struct resource standard_io_resources[] = { 564 { .name = "dma1", .start = 0x00, .end = 0x1f, 565 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 566 { .name = "pic1", .start = 0x20, .end = 0x21, 567 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 568 { .name = "timer0", .start = 0x40, .end = 0x43, 569 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 570 { .name = "timer1", .start = 0x50, .end = 0x53, 571 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 572 { .name = "keyboard", .start = 0x60, .end = 0x60, 573 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 574 { .name = "keyboard", .start = 0x64, .end = 0x64, 575 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 576 { .name = "dma page reg", .start = 0x80, .end = 0x8f, 577 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 578 { .name = "pic2", .start = 0xa0, .end = 0xa1, 579 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 580 { .name = "dma2", .start = 0xc0, .end = 0xdf, 581 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 582 { .name = "fpu", .start = 0xf0, .end = 0xff, 583 .flags = IORESOURCE_BUSY | IORESOURCE_IO } 584 }; 585 586 void __init reserve_standard_io_resources(void) 587 { 588 int i; 589 590 /* request I/O space for devices used on all i[345]86 PCs */ 591 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++) 592 request_resource(&ioport_resource, &standard_io_resources[i]); 593 594 } 595 596 static bool __init snb_gfx_workaround_needed(void) 597 { 598 #ifdef CONFIG_PCI 599 int i; 600 u16 vendor, devid; 601 static const __initconst u16 snb_ids[] = { 602 0x0102, 603 0x0112, 604 0x0122, 605 0x0106, 606 0x0116, 607 0x0126, 608 0x010a, 609 }; 610 611 /* Assume no if something weird is going on with PCI */ 612 if (!early_pci_allowed()) 613 return false; 614 615 vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID); 616 if (vendor != 0x8086) 617 return false; 618 619 devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID); 620 for (i = 0; i < ARRAY_SIZE(snb_ids); i++) 621 if (devid == snb_ids[i]) 622 return true; 623 #endif 624 625 return false; 626 } 627 628 /* 629 * Sandy Bridge graphics has trouble with certain ranges, exclude 630 * them from allocation. 631 */ 632 static void __init trim_snb_memory(void) 633 { 634 static const __initconst unsigned long bad_pages[] = { 635 0x20050000, 636 0x20110000, 637 0x20130000, 638 0x20138000, 639 0x40004000, 640 }; 641 int i; 642 643 if (!snb_gfx_workaround_needed()) 644 return; 645 646 printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n"); 647 648 /* 649 * SandyBridge integrated graphics devices have a bug that prevents 650 * them from accessing certain memory ranges, namely anything below 651 * 1M and in the pages listed in bad_pages[] above. 652 * 653 * To avoid these pages being ever accessed by SNB gfx devices reserve 654 * bad_pages that have not already been reserved at boot time. 655 * All memory below the 1 MB mark is anyway reserved later during 656 * setup_arch(), so there is no need to reserve it here. 657 */ 658 659 for (i = 0; i < ARRAY_SIZE(bad_pages); i++) { 660 if (memblock_reserve(bad_pages[i], PAGE_SIZE)) 661 printk(KERN_WARNING "failed to reserve 0x%08lx\n", 662 bad_pages[i]); 663 } 664 } 665 666 static void __init trim_bios_range(void) 667 { 668 /* 669 * A special case is the first 4Kb of memory; 670 * This is a BIOS owned area, not kernel ram, but generally 671 * not listed as such in the E820 table. 672 * 673 * This typically reserves additional memory (64KiB by default) 674 * since some BIOSes are known to corrupt low memory. See the 675 * Kconfig help text for X86_RESERVE_LOW. 676 */ 677 e820__range_update(0, PAGE_SIZE, E820_TYPE_RAM, E820_TYPE_RESERVED); 678 679 /* 680 * special case: Some BIOSes report the PC BIOS 681 * area (640Kb -> 1Mb) as RAM even though it is not. 682 * take them out. 683 */ 684 e820__range_remove(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_TYPE_RAM, 1); 685 686 e820__update_table(e820_table); 687 } 688 689 /* called before trim_bios_range() to spare extra sanitize */ 690 static void __init e820_add_kernel_range(void) 691 { 692 u64 start = __pa_symbol(_text); 693 u64 size = __pa_symbol(_end) - start; 694 695 /* 696 * Complain if .text .data and .bss are not marked as E820_TYPE_RAM and 697 * attempt to fix it by adding the range. We may have a confused BIOS, 698 * or the user may have used memmap=exactmap or memmap=xxM$yyM to 699 * exclude kernel range. If we really are running on top non-RAM, 700 * we will crash later anyways. 701 */ 702 if (e820__mapped_all(start, start + size, E820_TYPE_RAM)) 703 return; 704 705 pr_warn(".text .data .bss are not marked as E820_TYPE_RAM!\n"); 706 e820__range_remove(start, size, E820_TYPE_RAM, 0); 707 e820__range_add(start, size, E820_TYPE_RAM); 708 } 709 710 static void __init early_reserve_memory(void) 711 { 712 /* 713 * Reserve the memory occupied by the kernel between _text and 714 * __end_of_kernel_reserve symbols. Any kernel sections after the 715 * __end_of_kernel_reserve symbol must be explicitly reserved with a 716 * separate memblock_reserve() or they will be discarded. 717 */ 718 memblock_reserve(__pa_symbol(_text), 719 (unsigned long)__end_of_kernel_reserve - (unsigned long)_text); 720 721 /* 722 * The first 4Kb of memory is a BIOS owned area, but generally it is 723 * not listed as such in the E820 table. 724 * 725 * Reserve the first 64K of memory since some BIOSes are known to 726 * corrupt low memory. After the real mode trampoline is allocated the 727 * rest of the memory below 640k is reserved. 728 * 729 * In addition, make sure page 0 is always reserved because on 730 * systems with L1TF its contents can be leaked to user processes. 731 */ 732 memblock_reserve(0, SZ_64K); 733 734 early_reserve_initrd(); 735 736 memblock_x86_reserve_range_setup_data(); 737 738 reserve_bios_regions(); 739 trim_snb_memory(); 740 } 741 742 /* 743 * Dump out kernel offset information on panic. 744 */ 745 static int 746 dump_kernel_offset(struct notifier_block *self, unsigned long v, void *p) 747 { 748 if (kaslr_enabled()) { 749 pr_emerg("Kernel Offset: 0x%lx from 0x%lx (relocation range: 0x%lx-0x%lx)\n", 750 kaslr_offset(), 751 __START_KERNEL, 752 __START_KERNEL_map, 753 MODULES_VADDR-1); 754 } else { 755 pr_emerg("Kernel Offset: disabled\n"); 756 } 757 758 return 0; 759 } 760 761 void x86_configure_nx(void) 762 { 763 if (boot_cpu_has(X86_FEATURE_NX)) 764 __supported_pte_mask |= _PAGE_NX; 765 else 766 __supported_pte_mask &= ~_PAGE_NX; 767 } 768 769 static void __init x86_report_nx(void) 770 { 771 if (!boot_cpu_has(X86_FEATURE_NX)) { 772 printk(KERN_NOTICE "Notice: NX (Execute Disable) protection " 773 "missing in CPU!\n"); 774 } else { 775 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE) 776 printk(KERN_INFO "NX (Execute Disable) protection: active\n"); 777 #else 778 /* 32bit non-PAE kernel, NX cannot be used */ 779 printk(KERN_NOTICE "Notice: NX (Execute Disable) protection " 780 "cannot be enabled: non-PAE kernel!\n"); 781 #endif 782 } 783 } 784 785 /* 786 * Determine if we were loaded by an EFI loader. If so, then we have also been 787 * passed the efi memmap, systab, etc., so we should use these data structures 788 * for initialization. Note, the efi init code path is determined by the 789 * global efi_enabled. This allows the same kernel image to be used on existing 790 * systems (with a traditional BIOS) as well as on EFI systems. 791 */ 792 /* 793 * setup_arch - architecture-specific boot-time initializations 794 * 795 * Note: On x86_64, fixmaps are ready for use even before this is called. 796 */ 797 798 void __init setup_arch(char **cmdline_p) 799 { 800 #ifdef CONFIG_X86_32 801 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data)); 802 803 /* 804 * copy kernel address range established so far and switch 805 * to the proper swapper page table 806 */ 807 clone_pgd_range(swapper_pg_dir + KERNEL_PGD_BOUNDARY, 808 initial_page_table + KERNEL_PGD_BOUNDARY, 809 KERNEL_PGD_PTRS); 810 811 load_cr3(swapper_pg_dir); 812 /* 813 * Note: Quark X1000 CPUs advertise PGE incorrectly and require 814 * a cr3 based tlb flush, so the following __flush_tlb_all() 815 * will not flush anything because the CPU quirk which clears 816 * X86_FEATURE_PGE has not been invoked yet. Though due to the 817 * load_cr3() above the TLB has been flushed already. The 818 * quirk is invoked before subsequent calls to __flush_tlb_all() 819 * so proper operation is guaranteed. 820 */ 821 __flush_tlb_all(); 822 #else 823 printk(KERN_INFO "Command line: %s\n", boot_command_line); 824 boot_cpu_data.x86_phys_bits = MAX_PHYSMEM_BITS; 825 #endif 826 827 #ifdef CONFIG_CMDLINE_BOOL 828 #ifdef CONFIG_CMDLINE_OVERRIDE 829 strscpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 830 #else 831 if (builtin_cmdline[0]) { 832 /* append boot loader cmdline to builtin */ 833 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE); 834 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE); 835 strscpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 836 } 837 #endif 838 builtin_cmdline_added = true; 839 #endif 840 841 strscpy(command_line, boot_command_line, COMMAND_LINE_SIZE); 842 *cmdline_p = command_line; 843 844 /* 845 * If we have OLPC OFW, we might end up relocating the fixmap due to 846 * reserve_top(), so do this before touching the ioremap area. 847 */ 848 olpc_ofw_detect(); 849 850 idt_setup_early_traps(); 851 early_cpu_init(); 852 jump_label_init(); 853 static_call_init(); 854 early_ioremap_init(); 855 856 setup_olpc_ofw_pgd(); 857 858 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev); 859 screen_info = boot_params.screen_info; 860 edid_info = boot_params.edid_info; 861 #ifdef CONFIG_X86_32 862 apm_info.bios = boot_params.apm_bios_info; 863 ist_info = boot_params.ist_info; 864 #endif 865 saved_video_mode = boot_params.hdr.vid_mode; 866 bootloader_type = boot_params.hdr.type_of_loader; 867 if ((bootloader_type >> 4) == 0xe) { 868 bootloader_type &= 0xf; 869 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4; 870 } 871 bootloader_version = bootloader_type & 0xf; 872 bootloader_version |= boot_params.hdr.ext_loader_ver << 4; 873 874 #ifdef CONFIG_BLK_DEV_RAM 875 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK; 876 #endif 877 #ifdef CONFIG_EFI 878 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature, 879 EFI32_LOADER_SIGNATURE, 4)) { 880 set_bit(EFI_BOOT, &efi.flags); 881 } else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature, 882 EFI64_LOADER_SIGNATURE, 4)) { 883 set_bit(EFI_BOOT, &efi.flags); 884 set_bit(EFI_64BIT, &efi.flags); 885 } 886 #endif 887 888 x86_init.oem.arch_setup(); 889 890 /* 891 * Do some memory reservations *before* memory is added to memblock, so 892 * memblock allocations won't overwrite it. 893 * 894 * After this point, everything still needed from the boot loader or 895 * firmware or kernel text should be early reserved or marked not RAM in 896 * e820. All other memory is free game. 897 * 898 * This call needs to happen before e820__memory_setup() which calls the 899 * xen_memory_setup() on Xen dom0 which relies on the fact that those 900 * early reservations have happened already. 901 */ 902 early_reserve_memory(); 903 904 iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1; 905 e820__memory_setup(); 906 parse_setup_data(); 907 908 copy_edd(); 909 910 if (!boot_params.hdr.root_flags) 911 root_mountflags &= ~MS_RDONLY; 912 setup_initial_init_mm(_text, _etext, _edata, (void *)_brk_end); 913 914 code_resource.start = __pa_symbol(_text); 915 code_resource.end = __pa_symbol(_etext)-1; 916 rodata_resource.start = __pa_symbol(__start_rodata); 917 rodata_resource.end = __pa_symbol(__end_rodata)-1; 918 data_resource.start = __pa_symbol(_sdata); 919 data_resource.end = __pa_symbol(_edata)-1; 920 bss_resource.start = __pa_symbol(__bss_start); 921 bss_resource.end = __pa_symbol(__bss_stop)-1; 922 923 /* 924 * x86_configure_nx() is called before parse_early_param() to detect 925 * whether hardware doesn't support NX (so that the early EHCI debug 926 * console setup can safely call set_fixmap()). 927 */ 928 x86_configure_nx(); 929 930 parse_early_param(); 931 932 if (efi_enabled(EFI_BOOT)) 933 efi_memblock_x86_reserve_range(); 934 935 #ifdef CONFIG_MEMORY_HOTPLUG 936 /* 937 * Memory used by the kernel cannot be hot-removed because Linux 938 * cannot migrate the kernel pages. When memory hotplug is 939 * enabled, we should prevent memblock from allocating memory 940 * for the kernel. 941 * 942 * ACPI SRAT records all hotpluggable memory ranges. But before 943 * SRAT is parsed, we don't know about it. 944 * 945 * The kernel image is loaded into memory at very early time. We 946 * cannot prevent this anyway. So on NUMA system, we set any 947 * node the kernel resides in as un-hotpluggable. 948 * 949 * Since on modern servers, one node could have double-digit 950 * gigabytes memory, we can assume the memory around the kernel 951 * image is also un-hotpluggable. So before SRAT is parsed, just 952 * allocate memory near the kernel image to try the best to keep 953 * the kernel away from hotpluggable memory. 954 */ 955 if (movable_node_is_enabled()) 956 memblock_set_bottom_up(true); 957 #endif 958 959 x86_report_nx(); 960 961 apic_setup_apic_calls(); 962 963 if (acpi_mps_check()) { 964 #ifdef CONFIG_X86_LOCAL_APIC 965 apic_is_disabled = true; 966 #endif 967 setup_clear_cpu_cap(X86_FEATURE_APIC); 968 } 969 970 e820__reserve_setup_data(); 971 e820__finish_early_params(); 972 973 if (efi_enabled(EFI_BOOT)) 974 efi_init(); 975 976 reserve_ibft_region(); 977 x86_init.resources.dmi_setup(); 978 979 /* 980 * VMware detection requires dmi to be available, so this 981 * needs to be done after dmi_setup(), for the boot CPU. 982 * For some guest types (Xen PV, SEV-SNP, TDX) it is required to be 983 * called before cache_bp_init() for setting up MTRR state. 984 */ 985 init_hypervisor_platform(); 986 987 tsc_early_init(); 988 x86_init.resources.probe_roms(); 989 990 /* after parse_early_param, so could debug it */ 991 insert_resource(&iomem_resource, &code_resource); 992 insert_resource(&iomem_resource, &rodata_resource); 993 insert_resource(&iomem_resource, &data_resource); 994 insert_resource(&iomem_resource, &bss_resource); 995 996 e820_add_kernel_range(); 997 trim_bios_range(); 998 #ifdef CONFIG_X86_32 999 if (ppro_with_ram_bug()) { 1000 e820__range_update(0x70000000ULL, 0x40000ULL, E820_TYPE_RAM, 1001 E820_TYPE_RESERVED); 1002 e820__update_table(e820_table); 1003 printk(KERN_INFO "fixed physical RAM map:\n"); 1004 e820__print_table("bad_ppro"); 1005 } 1006 #else 1007 early_gart_iommu_check(); 1008 #endif 1009 1010 /* 1011 * partially used pages are not usable - thus 1012 * we are rounding upwards: 1013 */ 1014 max_pfn = e820__end_of_ram_pfn(); 1015 1016 /* update e820 for memory not covered by WB MTRRs */ 1017 cache_bp_init(); 1018 if (mtrr_trim_uncached_memory(max_pfn)) 1019 max_pfn = e820__end_of_ram_pfn(); 1020 1021 max_possible_pfn = max_pfn; 1022 1023 /* 1024 * Define random base addresses for memory sections after max_pfn is 1025 * defined and before each memory section base is used. 1026 */ 1027 kernel_randomize_memory(); 1028 1029 #ifdef CONFIG_X86_32 1030 /* max_low_pfn get updated here */ 1031 find_low_pfn_range(); 1032 #else 1033 check_x2apic(); 1034 1035 /* How many end-of-memory variables you have, grandma! */ 1036 /* need this before calling reserve_initrd */ 1037 if (max_pfn > (1UL<<(32 - PAGE_SHIFT))) 1038 max_low_pfn = e820__end_of_low_ram_pfn(); 1039 else 1040 max_low_pfn = max_pfn; 1041 1042 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1; 1043 #endif 1044 1045 /* Find and reserve MPTABLE area */ 1046 x86_init.mpparse.find_mptable(); 1047 1048 early_alloc_pgt_buf(); 1049 1050 /* 1051 * Need to conclude brk, before e820__memblock_setup() 1052 * it could use memblock_find_in_range, could overlap with 1053 * brk area. 1054 */ 1055 reserve_brk(); 1056 1057 cleanup_highmap(); 1058 1059 memblock_set_current_limit(ISA_END_ADDRESS); 1060 e820__memblock_setup(); 1061 1062 /* 1063 * Needs to run after memblock setup because it needs the physical 1064 * memory size. 1065 */ 1066 mem_encrypt_setup_arch(); 1067 cc_random_init(); 1068 1069 efi_find_mirror(); 1070 efi_esrt_init(); 1071 efi_mokvar_table_init(); 1072 1073 /* 1074 * The EFI specification says that boot service code won't be 1075 * called after ExitBootServices(). This is, in fact, a lie. 1076 */ 1077 efi_reserve_boot_services(); 1078 1079 /* preallocate 4k for mptable mpc */ 1080 e820__memblock_alloc_reserved_mpc_new(); 1081 1082 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION 1083 setup_bios_corruption_check(); 1084 #endif 1085 1086 #ifdef CONFIG_X86_32 1087 printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n", 1088 (max_pfn_mapped<<PAGE_SHIFT) - 1); 1089 #endif 1090 1091 /* 1092 * Find free memory for the real mode trampoline and place it there. If 1093 * there is not enough free memory under 1M, on EFI-enabled systems 1094 * there will be additional attempt to reclaim the memory for the real 1095 * mode trampoline at efi_free_boot_services(). 1096 * 1097 * Unconditionally reserve the entire first 1M of RAM because BIOSes 1098 * are known to corrupt low memory and several hundred kilobytes are not 1099 * worth complex detection what memory gets clobbered. Windows does the 1100 * same thing for very similar reasons. 1101 * 1102 * Moreover, on machines with SandyBridge graphics or in setups that use 1103 * crashkernel the entire 1M is reserved anyway. 1104 * 1105 * Note the host kernel TDX also requires the first 1MB being reserved. 1106 */ 1107 x86_platform.realmode_reserve(); 1108 1109 init_mem_mapping(); 1110 1111 /* 1112 * init_mem_mapping() relies on the early IDT page fault handling. 1113 * Now either enable FRED or install the real page fault handler 1114 * for 64-bit in the IDT. 1115 */ 1116 cpu_init_replace_early_idt(); 1117 1118 /* 1119 * Update mmu_cr4_features (and, indirectly, trampoline_cr4_features) 1120 * with the current CR4 value. This may not be necessary, but 1121 * auditing all the early-boot CR4 manipulation would be needed to 1122 * rule it out. 1123 * 1124 * Mask off features that don't work outside long mode (just 1125 * PCIDE for now). 1126 */ 1127 mmu_cr4_features = __read_cr4() & ~X86_CR4_PCIDE; 1128 1129 memblock_set_current_limit(get_max_mapped()); 1130 1131 /* 1132 * NOTE: On x86-32, only from this point on, fixmaps are ready for use. 1133 */ 1134 1135 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT 1136 if (init_ohci1394_dma_early) 1137 init_ohci1394_dma_on_all_controllers(); 1138 #endif 1139 /* Allocate bigger log buffer */ 1140 setup_log_buf(1); 1141 1142 if (efi_enabled(EFI_BOOT)) { 1143 switch (boot_params.secure_boot) { 1144 case efi_secureboot_mode_disabled: 1145 pr_info("Secure boot disabled\n"); 1146 break; 1147 case efi_secureboot_mode_enabled: 1148 pr_info("Secure boot enabled\n"); 1149 break; 1150 default: 1151 pr_info("Secure boot could not be determined\n"); 1152 break; 1153 } 1154 } 1155 1156 reserve_initrd(); 1157 1158 acpi_table_upgrade(); 1159 /* Look for ACPI tables and reserve memory occupied by them. */ 1160 acpi_boot_table_init(); 1161 1162 vsmp_init(); 1163 1164 io_delay_init(); 1165 1166 early_platform_quirks(); 1167 1168 /* Some platforms need the APIC registered for NUMA configuration */ 1169 early_acpi_boot_init(); 1170 x86_init.mpparse.early_parse_smp_cfg(); 1171 1172 x86_flattree_get_config(); 1173 1174 initmem_init(); 1175 dma_contiguous_reserve(max_pfn_mapped << PAGE_SHIFT); 1176 1177 if (boot_cpu_has(X86_FEATURE_GBPAGES)) 1178 hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT); 1179 1180 /* 1181 * Reserve memory for crash kernel after SRAT is parsed so that it 1182 * won't consume hotpluggable memory. 1183 */ 1184 arch_reserve_crashkernel(); 1185 1186 if (!early_xdbc_setup_hardware()) 1187 early_xdbc_register_console(); 1188 1189 x86_init.paging.pagetable_init(); 1190 1191 kasan_init(); 1192 1193 /* 1194 * Sync back kernel address range. 1195 * 1196 * FIXME: Can the later sync in setup_cpu_entry_areas() replace 1197 * this call? 1198 */ 1199 sync_initial_page_table(); 1200 1201 tboot_probe(); 1202 1203 map_vsyscall(); 1204 1205 x86_32_probe_apic(); 1206 1207 early_quirks(); 1208 1209 topology_apply_cmdline_limits_early(); 1210 1211 /* 1212 * Parse SMP configuration. Try ACPI first and then the platform 1213 * specific parser. 1214 */ 1215 acpi_boot_init(); 1216 x86_init.mpparse.parse_smp_cfg(); 1217 1218 /* Last opportunity to detect and map the local APIC */ 1219 init_apic_mappings(); 1220 1221 topology_init_possible_cpus(); 1222 1223 init_cpu_to_node(); 1224 init_gi_nodes(); 1225 1226 io_apic_init_mappings(); 1227 1228 x86_init.hyper.guest_late_init(); 1229 1230 e820__reserve_resources(); 1231 e820__register_nosave_regions(max_pfn); 1232 1233 x86_init.resources.reserve_resources(); 1234 1235 e820__setup_pci_gap(); 1236 1237 #ifdef CONFIG_VT 1238 #if defined(CONFIG_VGA_CONSOLE) 1239 if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY)) 1240 vgacon_register_screen(&screen_info); 1241 #endif 1242 #endif 1243 x86_init.oem.banner(); 1244 1245 x86_init.timers.wallclock_init(); 1246 1247 /* 1248 * This needs to run before setup_local_APIC() which soft-disables the 1249 * local APIC temporarily and that masks the thermal LVT interrupt, 1250 * leading to softlockups on machines which have configured SMI 1251 * interrupt delivery. 1252 */ 1253 therm_lvt_init(); 1254 1255 mcheck_init(); 1256 1257 register_refined_jiffies(CLOCK_TICK_RATE); 1258 1259 #ifdef CONFIG_EFI 1260 if (efi_enabled(EFI_BOOT)) 1261 efi_apply_memmap_quirks(); 1262 #endif 1263 1264 unwind_init(); 1265 } 1266 1267 #ifdef CONFIG_X86_32 1268 1269 static struct resource video_ram_resource = { 1270 .name = "Video RAM area", 1271 .start = 0xa0000, 1272 .end = 0xbffff, 1273 .flags = IORESOURCE_BUSY | IORESOURCE_MEM 1274 }; 1275 1276 void __init i386_reserve_resources(void) 1277 { 1278 request_resource(&iomem_resource, &video_ram_resource); 1279 reserve_standard_io_resources(); 1280 } 1281 1282 #endif /* CONFIG_X86_32 */ 1283 1284 static struct notifier_block kernel_offset_notifier = { 1285 .notifier_call = dump_kernel_offset 1286 }; 1287 1288 static int __init register_kernel_offset_dumper(void) 1289 { 1290 atomic_notifier_chain_register(&panic_notifier_list, 1291 &kernel_offset_notifier); 1292 return 0; 1293 } 1294 __initcall(register_kernel_offset_dumper); 1295 1296 #ifdef CONFIG_HOTPLUG_CPU 1297 bool arch_cpu_is_hotpluggable(int cpu) 1298 { 1299 return cpu > 0; 1300 } 1301 #endif /* CONFIG_HOTPLUG_CPU */ 1302