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 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); 134 135 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64) 136 __visible unsigned long mmu_cr4_features __ro_after_init; 137 #else 138 __visible unsigned long mmu_cr4_features __ro_after_init = X86_CR4_PAE; 139 #endif 140 141 #ifdef CONFIG_IMA 142 static phys_addr_t ima_kexec_buffer_phys; 143 static size_t ima_kexec_buffer_size; 144 #endif 145 146 /* Boot loader ID and version as integers, for the benefit of proc_dointvec */ 147 int bootloader_type, bootloader_version; 148 149 /* 150 * Setup options 151 */ 152 struct screen_info screen_info; 153 EXPORT_SYMBOL(screen_info); 154 struct edid_info edid_info; 155 EXPORT_SYMBOL_GPL(edid_info); 156 157 extern int root_mountflags; 158 159 unsigned long saved_video_mode; 160 161 #define RAMDISK_IMAGE_START_MASK 0x07FF 162 #define RAMDISK_PROMPT_FLAG 0x8000 163 #define RAMDISK_LOAD_FLAG 0x4000 164 165 static char __initdata command_line[COMMAND_LINE_SIZE]; 166 #ifdef CONFIG_CMDLINE_BOOL 167 char builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE; 168 bool builtin_cmdline_added __ro_after_init; 169 #endif 170 171 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE) 172 struct edd edd; 173 #ifdef CONFIG_EDD_MODULE 174 EXPORT_SYMBOL(edd); 175 #endif 176 /** 177 * copy_edd() - Copy the BIOS EDD information 178 * from boot_params into a safe place. 179 * 180 */ 181 static inline void __init copy_edd(void) 182 { 183 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer, 184 sizeof(edd.mbr_signature)); 185 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info)); 186 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries; 187 edd.edd_info_nr = boot_params.eddbuf_entries; 188 } 189 #else 190 static inline void __init copy_edd(void) 191 { 192 } 193 #endif 194 195 void * __init extend_brk(size_t size, size_t align) 196 { 197 size_t mask = align - 1; 198 void *ret; 199 200 BUG_ON(_brk_start == 0); 201 BUG_ON(align & mask); 202 203 _brk_end = (_brk_end + mask) & ~mask; 204 BUG_ON((char *)(_brk_end + size) > __brk_limit); 205 206 ret = (void *)_brk_end; 207 _brk_end += size; 208 209 memset(ret, 0, size); 210 211 return ret; 212 } 213 214 #ifdef CONFIG_X86_32 215 static void __init cleanup_highmap(void) 216 { 217 } 218 #endif 219 220 static void __init reserve_brk(void) 221 { 222 if (_brk_end > _brk_start) 223 memblock_reserve(__pa_symbol(_brk_start), 224 _brk_end - _brk_start); 225 226 /* Mark brk area as locked down and no longer taking any 227 new allocations */ 228 _brk_start = 0; 229 } 230 231 #ifdef CONFIG_BLK_DEV_INITRD 232 233 static u64 __init get_ramdisk_image(void) 234 { 235 u64 ramdisk_image = boot_params.hdr.ramdisk_image; 236 237 ramdisk_image |= (u64)boot_params.ext_ramdisk_image << 32; 238 239 if (ramdisk_image == 0) 240 ramdisk_image = phys_initrd_start; 241 242 return ramdisk_image; 243 } 244 static u64 __init get_ramdisk_size(void) 245 { 246 u64 ramdisk_size = boot_params.hdr.ramdisk_size; 247 248 ramdisk_size |= (u64)boot_params.ext_ramdisk_size << 32; 249 250 if (ramdisk_size == 0) 251 ramdisk_size = phys_initrd_size; 252 253 return ramdisk_size; 254 } 255 256 static void __init relocate_initrd(void) 257 { 258 /* Assume only end is not page aligned */ 259 u64 ramdisk_image = get_ramdisk_image(); 260 u64 ramdisk_size = get_ramdisk_size(); 261 u64 area_size = PAGE_ALIGN(ramdisk_size); 262 int ret = 0; 263 264 /* We need to move the initrd down into directly mapped mem */ 265 u64 relocated_ramdisk = memblock_phys_alloc_range(area_size, PAGE_SIZE, 0, 266 PFN_PHYS(max_pfn_mapped)); 267 if (!relocated_ramdisk) 268 panic("Cannot find place for new RAMDISK of size %lld\n", 269 ramdisk_size); 270 271 initrd_start = relocated_ramdisk + PAGE_OFFSET; 272 initrd_end = initrd_start + ramdisk_size; 273 printk(KERN_INFO "Allocated new RAMDISK: [mem %#010llx-%#010llx]\n", 274 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1); 275 276 ret = copy_from_early_mem((void *)initrd_start, ramdisk_image, ramdisk_size); 277 if (ret) 278 panic("Copy RAMDISK failed\n"); 279 280 printk(KERN_INFO "Move RAMDISK from [mem %#010llx-%#010llx] to" 281 " [mem %#010llx-%#010llx]\n", 282 ramdisk_image, ramdisk_image + ramdisk_size - 1, 283 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1); 284 } 285 286 static void __init early_reserve_initrd(void) 287 { 288 /* Assume only end is not page aligned */ 289 u64 ramdisk_image = get_ramdisk_image(); 290 u64 ramdisk_size = get_ramdisk_size(); 291 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size); 292 293 if (!boot_params.hdr.type_of_loader || 294 !ramdisk_image || !ramdisk_size) 295 return; /* No initrd provided by bootloader */ 296 297 memblock_reserve(ramdisk_image, ramdisk_end - ramdisk_image); 298 } 299 300 static void __init reserve_initrd(void) 301 { 302 /* Assume only end is not page aligned */ 303 u64 ramdisk_image = get_ramdisk_image(); 304 u64 ramdisk_size = get_ramdisk_size(); 305 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size); 306 307 if (!boot_params.hdr.type_of_loader || 308 !ramdisk_image || !ramdisk_size) 309 return; /* No initrd provided by bootloader */ 310 311 initrd_start = 0; 312 313 printk(KERN_INFO "RAMDISK: [mem %#010llx-%#010llx]\n", ramdisk_image, 314 ramdisk_end - 1); 315 316 if (pfn_range_is_mapped(PFN_DOWN(ramdisk_image), 317 PFN_DOWN(ramdisk_end))) { 318 /* All are mapped, easy case */ 319 initrd_start = ramdisk_image + PAGE_OFFSET; 320 initrd_end = initrd_start + ramdisk_size; 321 return; 322 } 323 324 relocate_initrd(); 325 326 memblock_phys_free(ramdisk_image, ramdisk_end - ramdisk_image); 327 } 328 329 #else 330 static void __init early_reserve_initrd(void) 331 { 332 } 333 static void __init reserve_initrd(void) 334 { 335 } 336 #endif /* CONFIG_BLK_DEV_INITRD */ 337 338 static void __init add_early_ima_buffer(u64 phys_addr) 339 { 340 #ifdef CONFIG_IMA 341 struct ima_setup_data *data; 342 343 data = early_memremap(phys_addr + sizeof(struct setup_data), sizeof(*data)); 344 if (!data) { 345 pr_warn("setup: failed to memremap ima_setup_data entry\n"); 346 return; 347 } 348 349 if (data->size) { 350 memblock_reserve(data->addr, data->size); 351 ima_kexec_buffer_phys = data->addr; 352 ima_kexec_buffer_size = data->size; 353 } 354 355 early_memunmap(data, sizeof(*data)); 356 #else 357 pr_warn("Passed IMA kexec data, but CONFIG_IMA not set. Ignoring.\n"); 358 #endif 359 } 360 361 #if defined(CONFIG_HAVE_IMA_KEXEC) && !defined(CONFIG_OF_FLATTREE) 362 int __init ima_free_kexec_buffer(void) 363 { 364 if (!ima_kexec_buffer_size) 365 return -ENOENT; 366 367 memblock_free_late(ima_kexec_buffer_phys, 368 ima_kexec_buffer_size); 369 370 ima_kexec_buffer_phys = 0; 371 ima_kexec_buffer_size = 0; 372 373 return 0; 374 } 375 376 int __init ima_get_kexec_buffer(void **addr, size_t *size) 377 { 378 if (!ima_kexec_buffer_size) 379 return -ENOENT; 380 381 *addr = __va(ima_kexec_buffer_phys); 382 *size = ima_kexec_buffer_size; 383 384 return 0; 385 } 386 #endif 387 388 static void __init parse_setup_data(void) 389 { 390 struct setup_data *data; 391 u64 pa_data, pa_next; 392 393 pa_data = boot_params.hdr.setup_data; 394 while (pa_data) { 395 u32 data_len, data_type; 396 397 data = early_memremap(pa_data, sizeof(*data)); 398 data_len = data->len + sizeof(struct setup_data); 399 data_type = data->type; 400 pa_next = data->next; 401 early_memunmap(data, sizeof(*data)); 402 403 switch (data_type) { 404 case SETUP_E820_EXT: 405 e820__memory_setup_extended(pa_data, data_len); 406 break; 407 case SETUP_DTB: 408 add_dtb(pa_data); 409 break; 410 case SETUP_EFI: 411 parse_efi_setup(pa_data, data_len); 412 break; 413 case SETUP_IMA: 414 add_early_ima_buffer(pa_data); 415 break; 416 case SETUP_RNG_SEED: 417 data = early_memremap(pa_data, data_len); 418 add_bootloader_randomness(data->data, data->len); 419 /* Zero seed for forward secrecy. */ 420 memzero_explicit(data->data, data->len); 421 /* Zero length in case we find ourselves back here by accident. */ 422 memzero_explicit(&data->len, sizeof(data->len)); 423 early_memunmap(data, data_len); 424 break; 425 default: 426 break; 427 } 428 pa_data = pa_next; 429 } 430 } 431 432 static void __init memblock_x86_reserve_range_setup_data(void) 433 { 434 struct setup_indirect *indirect; 435 struct setup_data *data; 436 u64 pa_data, pa_next; 437 u32 len; 438 439 pa_data = boot_params.hdr.setup_data; 440 while (pa_data) { 441 data = early_memremap(pa_data, sizeof(*data)); 442 if (!data) { 443 pr_warn("setup: failed to memremap setup_data entry\n"); 444 return; 445 } 446 447 len = sizeof(*data); 448 pa_next = data->next; 449 450 memblock_reserve(pa_data, sizeof(*data) + data->len); 451 452 if (data->type == SETUP_INDIRECT) { 453 len += data->len; 454 early_memunmap(data, sizeof(*data)); 455 data = early_memremap(pa_data, len); 456 if (!data) { 457 pr_warn("setup: failed to memremap indirect setup_data\n"); 458 return; 459 } 460 461 indirect = (struct setup_indirect *)data->data; 462 463 if (indirect->type != SETUP_INDIRECT) 464 memblock_reserve(indirect->addr, indirect->len); 465 } 466 467 pa_data = pa_next; 468 early_memunmap(data, len); 469 } 470 } 471 472 static void __init arch_reserve_crashkernel(void) 473 { 474 unsigned long long crash_base, crash_size, low_size = 0; 475 char *cmdline = boot_command_line; 476 bool high = false; 477 int ret; 478 479 if (!IS_ENABLED(CONFIG_CRASH_RESERVE)) 480 return; 481 482 ret = parse_crashkernel(cmdline, memblock_phys_mem_size(), 483 &crash_size, &crash_base, 484 &low_size, &high); 485 if (ret) 486 return; 487 488 if (xen_pv_domain()) { 489 pr_info("Ignoring crashkernel for a Xen PV domain\n"); 490 return; 491 } 492 493 reserve_crashkernel_generic(cmdline, crash_size, crash_base, 494 low_size, high); 495 } 496 497 static struct resource standard_io_resources[] = { 498 { .name = "dma1", .start = 0x00, .end = 0x1f, 499 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 500 { .name = "pic1", .start = 0x20, .end = 0x21, 501 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 502 { .name = "timer0", .start = 0x40, .end = 0x43, 503 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 504 { .name = "timer1", .start = 0x50, .end = 0x53, 505 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 506 { .name = "keyboard", .start = 0x60, .end = 0x60, 507 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 508 { .name = "keyboard", .start = 0x64, .end = 0x64, 509 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 510 { .name = "dma page reg", .start = 0x80, .end = 0x8f, 511 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 512 { .name = "pic2", .start = 0xa0, .end = 0xa1, 513 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 514 { .name = "dma2", .start = 0xc0, .end = 0xdf, 515 .flags = IORESOURCE_BUSY | IORESOURCE_IO }, 516 { .name = "fpu", .start = 0xf0, .end = 0xff, 517 .flags = IORESOURCE_BUSY | IORESOURCE_IO } 518 }; 519 520 void __init reserve_standard_io_resources(void) 521 { 522 int i; 523 524 /* request I/O space for devices used on all i[345]86 PCs */ 525 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++) 526 request_resource(&ioport_resource, &standard_io_resources[i]); 527 528 } 529 530 static bool __init snb_gfx_workaround_needed(void) 531 { 532 #ifdef CONFIG_PCI 533 int i; 534 u16 vendor, devid; 535 static const __initconst u16 snb_ids[] = { 536 0x0102, 537 0x0112, 538 0x0122, 539 0x0106, 540 0x0116, 541 0x0126, 542 0x010a, 543 }; 544 545 /* Assume no if something weird is going on with PCI */ 546 if (!early_pci_allowed()) 547 return false; 548 549 vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID); 550 if (vendor != 0x8086) 551 return false; 552 553 devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID); 554 for (i = 0; i < ARRAY_SIZE(snb_ids); i++) 555 if (devid == snb_ids[i]) 556 return true; 557 #endif 558 559 return false; 560 } 561 562 /* 563 * Sandy Bridge graphics has trouble with certain ranges, exclude 564 * them from allocation. 565 */ 566 static void __init trim_snb_memory(void) 567 { 568 static const __initconst unsigned long bad_pages[] = { 569 0x20050000, 570 0x20110000, 571 0x20130000, 572 0x20138000, 573 0x40004000, 574 }; 575 int i; 576 577 if (!snb_gfx_workaround_needed()) 578 return; 579 580 printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n"); 581 582 /* 583 * SandyBridge integrated graphics devices have a bug that prevents 584 * them from accessing certain memory ranges, namely anything below 585 * 1M and in the pages listed in bad_pages[] above. 586 * 587 * To avoid these pages being ever accessed by SNB gfx devices reserve 588 * bad_pages that have not already been reserved at boot time. 589 * All memory below the 1 MB mark is anyway reserved later during 590 * setup_arch(), so there is no need to reserve it here. 591 */ 592 593 for (i = 0; i < ARRAY_SIZE(bad_pages); i++) { 594 if (memblock_reserve(bad_pages[i], PAGE_SIZE)) 595 printk(KERN_WARNING "failed to reserve 0x%08lx\n", 596 bad_pages[i]); 597 } 598 } 599 600 static void __init trim_bios_range(void) 601 { 602 /* 603 * A special case is the first 4Kb of memory; 604 * This is a BIOS owned area, not kernel ram, but generally 605 * not listed as such in the E820 table. 606 * 607 * This typically reserves additional memory (64KiB by default) 608 * since some BIOSes are known to corrupt low memory. See the 609 * Kconfig help text for X86_RESERVE_LOW. 610 */ 611 e820__range_update(0, PAGE_SIZE, E820_TYPE_RAM, E820_TYPE_RESERVED); 612 613 /* 614 * special case: Some BIOSes report the PC BIOS 615 * area (640Kb -> 1Mb) as RAM even though it is not. 616 * take them out. 617 */ 618 e820__range_remove(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_TYPE_RAM, 1); 619 620 e820__update_table(e820_table); 621 } 622 623 /* called before trim_bios_range() to spare extra sanitize */ 624 static void __init e820_add_kernel_range(void) 625 { 626 u64 start = __pa_symbol(_text); 627 u64 size = __pa_symbol(_end) - start; 628 629 /* 630 * Complain if .text .data and .bss are not marked as E820_TYPE_RAM and 631 * attempt to fix it by adding the range. We may have a confused BIOS, 632 * or the user may have used memmap=exactmap or memmap=xxM$yyM to 633 * exclude kernel range. If we really are running on top non-RAM, 634 * we will crash later anyways. 635 */ 636 if (e820__mapped_all(start, start + size, E820_TYPE_RAM)) 637 return; 638 639 pr_warn(".text .data .bss are not marked as E820_TYPE_RAM!\n"); 640 e820__range_remove(start, size, E820_TYPE_RAM, 0); 641 e820__range_add(start, size, E820_TYPE_RAM); 642 } 643 644 static void __init early_reserve_memory(void) 645 { 646 /* 647 * Reserve the memory occupied by the kernel between _text and 648 * __end_of_kernel_reserve symbols. Any kernel sections after the 649 * __end_of_kernel_reserve symbol must be explicitly reserved with a 650 * separate memblock_reserve() or they will be discarded. 651 */ 652 memblock_reserve(__pa_symbol(_text), 653 (unsigned long)__end_of_kernel_reserve - (unsigned long)_text); 654 655 /* 656 * The first 4Kb of memory is a BIOS owned area, but generally it is 657 * not listed as such in the E820 table. 658 * 659 * Reserve the first 64K of memory since some BIOSes are known to 660 * corrupt low memory. After the real mode trampoline is allocated the 661 * rest of the memory below 640k is reserved. 662 * 663 * In addition, make sure page 0 is always reserved because on 664 * systems with L1TF its contents can be leaked to user processes. 665 */ 666 memblock_reserve(0, SZ_64K); 667 668 early_reserve_initrd(); 669 670 memblock_x86_reserve_range_setup_data(); 671 672 reserve_bios_regions(); 673 trim_snb_memory(); 674 } 675 676 /* 677 * Dump out kernel offset information on panic. 678 */ 679 static int 680 dump_kernel_offset(struct notifier_block *self, unsigned long v, void *p) 681 { 682 if (kaslr_enabled()) { 683 pr_emerg("Kernel Offset: 0x%lx from 0x%lx (relocation range: 0x%lx-0x%lx)\n", 684 kaslr_offset(), 685 __START_KERNEL, 686 __START_KERNEL_map, 687 MODULES_VADDR-1); 688 } else { 689 pr_emerg("Kernel Offset: disabled\n"); 690 } 691 692 return 0; 693 } 694 695 void x86_configure_nx(void) 696 { 697 if (boot_cpu_has(X86_FEATURE_NX)) 698 __supported_pte_mask |= _PAGE_NX; 699 else 700 __supported_pte_mask &= ~_PAGE_NX; 701 } 702 703 static void __init x86_report_nx(void) 704 { 705 if (!boot_cpu_has(X86_FEATURE_NX)) { 706 printk(KERN_NOTICE "Notice: NX (Execute Disable) protection " 707 "missing in CPU!\n"); 708 } else { 709 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE) 710 printk(KERN_INFO "NX (Execute Disable) protection: active\n"); 711 #else 712 /* 32bit non-PAE kernel, NX cannot be used */ 713 printk(KERN_NOTICE "Notice: NX (Execute Disable) protection " 714 "cannot be enabled: non-PAE kernel!\n"); 715 #endif 716 } 717 } 718 719 /* 720 * Determine if we were loaded by an EFI loader. If so, then we have also been 721 * passed the efi memmap, systab, etc., so we should use these data structures 722 * for initialization. Note, the efi init code path is determined by the 723 * global efi_enabled. This allows the same kernel image to be used on existing 724 * systems (with a traditional BIOS) as well as on EFI systems. 725 */ 726 /* 727 * setup_arch - architecture-specific boot-time initializations 728 * 729 * Note: On x86_64, fixmaps are ready for use even before this is called. 730 */ 731 732 void __init setup_arch(char **cmdline_p) 733 { 734 #ifdef CONFIG_X86_32 735 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data)); 736 737 /* 738 * copy kernel address range established so far and switch 739 * to the proper swapper page table 740 */ 741 clone_pgd_range(swapper_pg_dir + KERNEL_PGD_BOUNDARY, 742 initial_page_table + KERNEL_PGD_BOUNDARY, 743 KERNEL_PGD_PTRS); 744 745 load_cr3(swapper_pg_dir); 746 /* 747 * Note: Quark X1000 CPUs advertise PGE incorrectly and require 748 * a cr3 based tlb flush, so the following __flush_tlb_all() 749 * will not flush anything because the CPU quirk which clears 750 * X86_FEATURE_PGE has not been invoked yet. Though due to the 751 * load_cr3() above the TLB has been flushed already. The 752 * quirk is invoked before subsequent calls to __flush_tlb_all() 753 * so proper operation is guaranteed. 754 */ 755 __flush_tlb_all(); 756 #else 757 printk(KERN_INFO "Command line: %s\n", boot_command_line); 758 boot_cpu_data.x86_phys_bits = MAX_PHYSMEM_BITS; 759 #endif 760 761 #ifdef CONFIG_CMDLINE_BOOL 762 #ifdef CONFIG_CMDLINE_OVERRIDE 763 strscpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 764 #else 765 if (builtin_cmdline[0]) { 766 /* append boot loader cmdline to builtin */ 767 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE); 768 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE); 769 strscpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE); 770 } 771 #endif 772 builtin_cmdline_added = true; 773 #endif 774 775 strscpy(command_line, boot_command_line, COMMAND_LINE_SIZE); 776 *cmdline_p = command_line; 777 778 /* 779 * If we have OLPC OFW, we might end up relocating the fixmap due to 780 * reserve_top(), so do this before touching the ioremap area. 781 */ 782 olpc_ofw_detect(); 783 784 idt_setup_early_traps(); 785 early_cpu_init(); 786 jump_label_init(); 787 static_call_init(); 788 early_ioremap_init(); 789 790 setup_olpc_ofw_pgd(); 791 792 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev); 793 screen_info = boot_params.screen_info; 794 edid_info = boot_params.edid_info; 795 #ifdef CONFIG_X86_32 796 apm_info.bios = boot_params.apm_bios_info; 797 ist_info = boot_params.ist_info; 798 #endif 799 saved_video_mode = boot_params.hdr.vid_mode; 800 bootloader_type = boot_params.hdr.type_of_loader; 801 if ((bootloader_type >> 4) == 0xe) { 802 bootloader_type &= 0xf; 803 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4; 804 } 805 bootloader_version = bootloader_type & 0xf; 806 bootloader_version |= boot_params.hdr.ext_loader_ver << 4; 807 808 #ifdef CONFIG_BLK_DEV_RAM 809 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK; 810 #endif 811 #ifdef CONFIG_EFI 812 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature, 813 EFI32_LOADER_SIGNATURE, 4)) { 814 set_bit(EFI_BOOT, &efi.flags); 815 } else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature, 816 EFI64_LOADER_SIGNATURE, 4)) { 817 set_bit(EFI_BOOT, &efi.flags); 818 set_bit(EFI_64BIT, &efi.flags); 819 } 820 #endif 821 822 x86_init.oem.arch_setup(); 823 824 /* 825 * Do some memory reservations *before* memory is added to memblock, so 826 * memblock allocations won't overwrite it. 827 * 828 * After this point, everything still needed from the boot loader or 829 * firmware or kernel text should be early reserved or marked not RAM in 830 * e820. All other memory is free game. 831 * 832 * This call needs to happen before e820__memory_setup() which calls the 833 * xen_memory_setup() on Xen dom0 which relies on the fact that those 834 * early reservations have happened already. 835 */ 836 early_reserve_memory(); 837 838 iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1; 839 e820__memory_setup(); 840 parse_setup_data(); 841 842 copy_edd(); 843 844 if (!boot_params.hdr.root_flags) 845 root_mountflags &= ~MS_RDONLY; 846 setup_initial_init_mm(_text, _etext, _edata, (void *)_brk_end); 847 848 code_resource.start = __pa_symbol(_text); 849 code_resource.end = __pa_symbol(_etext)-1; 850 rodata_resource.start = __pa_symbol(__start_rodata); 851 rodata_resource.end = __pa_symbol(__end_rodata)-1; 852 data_resource.start = __pa_symbol(_sdata); 853 data_resource.end = __pa_symbol(_edata)-1; 854 bss_resource.start = __pa_symbol(__bss_start); 855 bss_resource.end = __pa_symbol(__bss_stop)-1; 856 857 /* 858 * x86_configure_nx() is called before parse_early_param() to detect 859 * whether hardware doesn't support NX (so that the early EHCI debug 860 * console setup can safely call set_fixmap()). 861 */ 862 x86_configure_nx(); 863 864 parse_early_param(); 865 866 if (efi_enabled(EFI_BOOT)) 867 efi_memblock_x86_reserve_range(); 868 869 #ifdef CONFIG_MEMORY_HOTPLUG 870 /* 871 * Memory used by the kernel cannot be hot-removed because Linux 872 * cannot migrate the kernel pages. When memory hotplug is 873 * enabled, we should prevent memblock from allocating memory 874 * for the kernel. 875 * 876 * ACPI SRAT records all hotpluggable memory ranges. But before 877 * SRAT is parsed, we don't know about it. 878 * 879 * The kernel image is loaded into memory at very early time. We 880 * cannot prevent this anyway. So on NUMA system, we set any 881 * node the kernel resides in as un-hotpluggable. 882 * 883 * Since on modern servers, one node could have double-digit 884 * gigabytes memory, we can assume the memory around the kernel 885 * image is also un-hotpluggable. So before SRAT is parsed, just 886 * allocate memory near the kernel image to try the best to keep 887 * the kernel away from hotpluggable memory. 888 */ 889 if (movable_node_is_enabled()) 890 memblock_set_bottom_up(true); 891 #endif 892 893 x86_report_nx(); 894 895 apic_setup_apic_calls(); 896 897 if (acpi_mps_check()) { 898 #ifdef CONFIG_X86_LOCAL_APIC 899 apic_is_disabled = true; 900 #endif 901 setup_clear_cpu_cap(X86_FEATURE_APIC); 902 } 903 904 e820__reserve_setup_data(); 905 e820__finish_early_params(); 906 907 if (efi_enabled(EFI_BOOT)) 908 efi_init(); 909 910 reserve_ibft_region(); 911 x86_init.resources.dmi_setup(); 912 913 /* 914 * VMware detection requires dmi to be available, so this 915 * needs to be done after dmi_setup(), for the boot CPU. 916 * For some guest types (Xen PV, SEV-SNP, TDX) it is required to be 917 * called before cache_bp_init() for setting up MTRR state. 918 */ 919 init_hypervisor_platform(); 920 921 tsc_early_init(); 922 x86_init.resources.probe_roms(); 923 924 /* after parse_early_param, so could debug it */ 925 insert_resource(&iomem_resource, &code_resource); 926 insert_resource(&iomem_resource, &rodata_resource); 927 insert_resource(&iomem_resource, &data_resource); 928 insert_resource(&iomem_resource, &bss_resource); 929 930 e820_add_kernel_range(); 931 trim_bios_range(); 932 #ifdef CONFIG_X86_32 933 if (ppro_with_ram_bug()) { 934 e820__range_update(0x70000000ULL, 0x40000ULL, E820_TYPE_RAM, 935 E820_TYPE_RESERVED); 936 e820__update_table(e820_table); 937 printk(KERN_INFO "fixed physical RAM map:\n"); 938 e820__print_table("bad_ppro"); 939 } 940 #else 941 early_gart_iommu_check(); 942 #endif 943 944 /* 945 * partially used pages are not usable - thus 946 * we are rounding upwards: 947 */ 948 max_pfn = e820__end_of_ram_pfn(); 949 950 /* update e820 for memory not covered by WB MTRRs */ 951 cache_bp_init(); 952 if (mtrr_trim_uncached_memory(max_pfn)) 953 max_pfn = e820__end_of_ram_pfn(); 954 955 max_possible_pfn = max_pfn; 956 957 /* 958 * Define random base addresses for memory sections after max_pfn is 959 * defined and before each memory section base is used. 960 */ 961 kernel_randomize_memory(); 962 963 #ifdef CONFIG_X86_32 964 /* max_low_pfn get updated here */ 965 find_low_pfn_range(); 966 #else 967 check_x2apic(); 968 969 /* How many end-of-memory variables you have, grandma! */ 970 /* need this before calling reserve_initrd */ 971 if (max_pfn > (1UL<<(32 - PAGE_SHIFT))) 972 max_low_pfn = e820__end_of_low_ram_pfn(); 973 else 974 max_low_pfn = max_pfn; 975 976 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1; 977 #endif 978 979 /* Find and reserve MPTABLE area */ 980 x86_init.mpparse.find_mptable(); 981 982 early_alloc_pgt_buf(); 983 984 /* 985 * Need to conclude brk, before e820__memblock_setup() 986 * it could use memblock_find_in_range, could overlap with 987 * brk area. 988 */ 989 reserve_brk(); 990 991 cleanup_highmap(); 992 993 memblock_set_current_limit(ISA_END_ADDRESS); 994 e820__memblock_setup(); 995 996 /* 997 * Needs to run after memblock setup because it needs the physical 998 * memory size. 999 */ 1000 mem_encrypt_setup_arch(); 1001 cc_random_init(); 1002 1003 efi_find_mirror(); 1004 efi_esrt_init(); 1005 efi_mokvar_table_init(); 1006 1007 /* 1008 * The EFI specification says that boot service code won't be 1009 * called after ExitBootServices(). This is, in fact, a lie. 1010 */ 1011 efi_reserve_boot_services(); 1012 1013 /* preallocate 4k for mptable mpc */ 1014 e820__memblock_alloc_reserved_mpc_new(); 1015 1016 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION 1017 setup_bios_corruption_check(); 1018 #endif 1019 1020 #ifdef CONFIG_X86_32 1021 printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n", 1022 (max_pfn_mapped<<PAGE_SHIFT) - 1); 1023 #endif 1024 1025 /* 1026 * Find free memory for the real mode trampoline and place it there. If 1027 * there is not enough free memory under 1M, on EFI-enabled systems 1028 * there will be additional attempt to reclaim the memory for the real 1029 * mode trampoline at efi_free_boot_services(). 1030 * 1031 * Unconditionally reserve the entire first 1M of RAM because BIOSes 1032 * are known to corrupt low memory and several hundred kilobytes are not 1033 * worth complex detection what memory gets clobbered. Windows does the 1034 * same thing for very similar reasons. 1035 * 1036 * Moreover, on machines with SandyBridge graphics or in setups that use 1037 * crashkernel the entire 1M is reserved anyway. 1038 * 1039 * Note the host kernel TDX also requires the first 1MB being reserved. 1040 */ 1041 x86_platform.realmode_reserve(); 1042 1043 init_mem_mapping(); 1044 1045 /* 1046 * init_mem_mapping() relies on the early IDT page fault handling. 1047 * Now either enable FRED or install the real page fault handler 1048 * for 64-bit in the IDT. 1049 */ 1050 cpu_init_replace_early_idt(); 1051 1052 /* 1053 * Update mmu_cr4_features (and, indirectly, trampoline_cr4_features) 1054 * with the current CR4 value. This may not be necessary, but 1055 * auditing all the early-boot CR4 manipulation would be needed to 1056 * rule it out. 1057 * 1058 * Mask off features that don't work outside long mode (just 1059 * PCIDE for now). 1060 */ 1061 mmu_cr4_features = __read_cr4() & ~X86_CR4_PCIDE; 1062 1063 memblock_set_current_limit(get_max_mapped()); 1064 1065 /* 1066 * NOTE: On x86-32, only from this point on, fixmaps are ready for use. 1067 */ 1068 1069 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT 1070 if (init_ohci1394_dma_early) 1071 init_ohci1394_dma_on_all_controllers(); 1072 #endif 1073 /* Allocate bigger log buffer */ 1074 setup_log_buf(1); 1075 1076 if (efi_enabled(EFI_BOOT)) { 1077 switch (boot_params.secure_boot) { 1078 case efi_secureboot_mode_disabled: 1079 pr_info("Secure boot disabled\n"); 1080 break; 1081 case efi_secureboot_mode_enabled: 1082 pr_info("Secure boot enabled\n"); 1083 break; 1084 default: 1085 pr_info("Secure boot could not be determined\n"); 1086 break; 1087 } 1088 } 1089 1090 reserve_initrd(); 1091 1092 acpi_table_upgrade(); 1093 /* Look for ACPI tables and reserve memory occupied by them. */ 1094 acpi_boot_table_init(); 1095 1096 vsmp_init(); 1097 1098 io_delay_init(); 1099 1100 early_platform_quirks(); 1101 1102 /* Some platforms need the APIC registered for NUMA configuration */ 1103 early_acpi_boot_init(); 1104 x86_init.mpparse.early_parse_smp_cfg(); 1105 1106 x86_flattree_get_config(); 1107 1108 initmem_init(); 1109 dma_contiguous_reserve(max_pfn_mapped << PAGE_SHIFT); 1110 1111 if (boot_cpu_has(X86_FEATURE_GBPAGES)) 1112 hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT); 1113 1114 /* 1115 * Reserve memory for crash kernel after SRAT is parsed so that it 1116 * won't consume hotpluggable memory. 1117 */ 1118 arch_reserve_crashkernel(); 1119 1120 if (!early_xdbc_setup_hardware()) 1121 early_xdbc_register_console(); 1122 1123 x86_init.paging.pagetable_init(); 1124 1125 kasan_init(); 1126 1127 /* 1128 * Sync back kernel address range. 1129 * 1130 * FIXME: Can the later sync in setup_cpu_entry_areas() replace 1131 * this call? 1132 */ 1133 sync_initial_page_table(); 1134 1135 tboot_probe(); 1136 1137 map_vsyscall(); 1138 1139 x86_32_probe_apic(); 1140 1141 early_quirks(); 1142 1143 topology_apply_cmdline_limits_early(); 1144 1145 /* 1146 * Parse SMP configuration. Try ACPI first and then the platform 1147 * specific parser. 1148 */ 1149 acpi_boot_init(); 1150 x86_init.mpparse.parse_smp_cfg(); 1151 1152 /* Last opportunity to detect and map the local APIC */ 1153 init_apic_mappings(); 1154 1155 topology_init_possible_cpus(); 1156 1157 init_cpu_to_node(); 1158 init_gi_nodes(); 1159 1160 io_apic_init_mappings(); 1161 1162 x86_init.hyper.guest_late_init(); 1163 1164 e820__reserve_resources(); 1165 e820__register_nosave_regions(max_pfn); 1166 1167 x86_init.resources.reserve_resources(); 1168 1169 e820__setup_pci_gap(); 1170 1171 #ifdef CONFIG_VT 1172 #if defined(CONFIG_VGA_CONSOLE) 1173 if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY)) 1174 vgacon_register_screen(&screen_info); 1175 #endif 1176 #endif 1177 x86_init.oem.banner(); 1178 1179 x86_init.timers.wallclock_init(); 1180 1181 /* 1182 * This needs to run before setup_local_APIC() which soft-disables the 1183 * local APIC temporarily and that masks the thermal LVT interrupt, 1184 * leading to softlockups on machines which have configured SMI 1185 * interrupt delivery. 1186 */ 1187 therm_lvt_init(); 1188 1189 mcheck_init(); 1190 1191 register_refined_jiffies(CLOCK_TICK_RATE); 1192 1193 #ifdef CONFIG_EFI 1194 if (efi_enabled(EFI_BOOT)) 1195 efi_apply_memmap_quirks(); 1196 #endif 1197 1198 unwind_init(); 1199 } 1200 1201 #ifdef CONFIG_X86_32 1202 1203 static struct resource video_ram_resource = { 1204 .name = "Video RAM area", 1205 .start = 0xa0000, 1206 .end = 0xbffff, 1207 .flags = IORESOURCE_BUSY | IORESOURCE_MEM 1208 }; 1209 1210 void __init i386_reserve_resources(void) 1211 { 1212 request_resource(&iomem_resource, &video_ram_resource); 1213 reserve_standard_io_resources(); 1214 } 1215 1216 #endif /* CONFIG_X86_32 */ 1217 1218 static struct notifier_block kernel_offset_notifier = { 1219 .notifier_call = dump_kernel_offset 1220 }; 1221 1222 static int __init register_kernel_offset_dumper(void) 1223 { 1224 atomic_notifier_chain_register(&panic_notifier_list, 1225 &kernel_offset_notifier); 1226 return 0; 1227 } 1228 __initcall(register_kernel_offset_dumper); 1229 1230 #ifdef CONFIG_HOTPLUG_CPU 1231 bool arch_cpu_is_hotpluggable(int cpu) 1232 { 1233 return cpu > 0; 1234 } 1235 #endif /* CONFIG_HOTPLUG_CPU */ 1236